1 /* SPDX-License-Identifier: GPL-2.0-or-later */
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.
7 * Definitions for the AF_INET socket handler.
9 * Version: @(#)sock.h 1.0.4 05/13/93
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>
17 * Alan Cox : Volatiles in skbuff pointers. See
18 * skbuff comments. May be overdone,
19 * better to prove they can be removed
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
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 */
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>
64 #include <linux/atomic.h>
65 #include <linux/refcount.h>
67 #include <net/checksum.h>
68 #include <net/tcp_states.h>
69 #include <linux/net_tstamp.h>
70 #include <net/l3mdev.h>
73 * This structure really needs to be cleaned up.
74 * Most of it is for TCP, and not used by any of
75 * the other protocols.
78 /* Define this to get the SOCK_DBG debugging facility. */
79 #define SOCK_DEBUGGING
81 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
82 printk(KERN_DEBUG msg); } while (0)
84 /* Validate arguments and do nothing */
85 static inline __printf(2, 3)
86 void SOCK_DEBUG(const struct sock
*sk
, const char *msg
, ...)
91 /* This is the per-socket lock. The spinlock provides a synchronization
92 * between user contexts and software interrupt processing, whereas the
93 * mini-semaphore synchronizes multiple users amongst themselves.
100 * We express the mutex-alike socket_lock semantics
101 * to the lock validator by explicitly managing
102 * the slock as a lock variant (in addition to
105 #ifdef CONFIG_DEBUG_LOCK_ALLOC
106 struct lockdep_map dep_map
;
114 typedef __u32 __bitwise __portpair
;
115 typedef __u64 __bitwise __addrpair
;
118 * struct sock_common - minimal network layer representation of sockets
119 * @skc_daddr: Foreign IPv4 addr
120 * @skc_rcv_saddr: Bound local IPv4 addr
121 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
122 * @skc_hash: hash value used with various protocol lookup tables
123 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
124 * @skc_dport: placeholder for inet_dport/tw_dport
125 * @skc_num: placeholder for inet_num/tw_num
126 * @skc_portpair: __u32 union of @skc_dport & @skc_num
127 * @skc_family: network address family
128 * @skc_state: Connection state
129 * @skc_reuse: %SO_REUSEADDR setting
130 * @skc_reuseport: %SO_REUSEPORT setting
131 * @skc_ipv6only: socket is IPV6 only
132 * @skc_net_refcnt: socket is using net ref counting
133 * @skc_bound_dev_if: bound device index if != 0
134 * @skc_bind_node: bind hash linkage for various protocol lookup tables
135 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
136 * @skc_prot: protocol handlers inside a network family
137 * @skc_net: reference to the network namespace of this socket
138 * @skc_v6_daddr: IPV6 destination address
139 * @skc_v6_rcv_saddr: IPV6 source address
140 * @skc_cookie: socket's cookie value
141 * @skc_node: main hash linkage for various protocol lookup tables
142 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
143 * @skc_tx_queue_mapping: tx queue number for this connection
144 * @skc_rx_queue_mapping: rx queue number for this connection
145 * @skc_flags: place holder for sk_flags
146 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
147 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
148 * @skc_listener: connection request listener socket (aka rsk_listener)
149 * [union with @skc_flags]
150 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
151 * [union with @skc_flags]
152 * @skc_incoming_cpu: record/match cpu processing incoming packets
153 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
154 * [union with @skc_incoming_cpu]
155 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
156 * [union with @skc_incoming_cpu]
157 * @skc_refcnt: reference count
159 * This is the minimal network layer representation of sockets, the header
160 * for struct sock and struct inet_timewait_sock.
163 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
164 * address on 64bit arches : cf INET_MATCH()
167 __addrpair skc_addrpair
;
170 __be32 skc_rcv_saddr
;
174 unsigned int skc_hash
;
175 __u16 skc_u16hashes
[2];
177 /* skc_dport && skc_num must be grouped as well */
179 __portpair skc_portpair
;
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
;
194 struct hlist_node skc_bind_node
;
195 struct hlist_node skc_portaddr_node
;
197 struct proto
*skc_prot
;
198 possible_net_t skc_net
;
200 #if IS_ENABLED(CONFIG_IPV6)
201 struct in6_addr skc_v6_daddr
;
202 struct in6_addr skc_v6_rcv_saddr
;
205 atomic64_t skc_cookie
;
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'
213 unsigned long skc_flags
;
214 struct sock
*skc_listener
; /* request_sock */
215 struct inet_timewait_death_row
*skc_tw_dr
; /* inet_timewait_sock */
218 * fields between dontcopy_begin/dontcopy_end
219 * are not copied in sock_copy()
222 int skc_dontcopy_begin
[0];
225 struct hlist_node skc_node
;
226 struct hlist_nulls_node skc_nulls_node
;
228 unsigned short skc_tx_queue_mapping
;
230 unsigned short skc_rx_queue_mapping
;
233 int skc_incoming_cpu
;
235 u32 skc_tw_rcv_nxt
; /* struct tcp_timewait_sock */
238 refcount_t skc_refcnt
;
240 int skc_dontcopy_end
[0];
243 u32 skc_window_clamp
;
244 u32 skc_tw_snd_nxt
; /* struct tcp_timewait_sock */
249 struct bpf_sk_storage
;
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_dst_cache: destination cache
262 * @sk_dst_pending_confirm: need to confirm neighbour
263 * @sk_policy: flow policy
264 * @sk_rx_skb_cache: cache copy of recently accessed RX skb
265 * @sk_receive_queue: incoming packets
266 * @sk_wmem_alloc: transmit queue bytes committed
267 * @sk_tsq_flags: TCP Small Queues flags
268 * @sk_write_queue: Packet sending queue
269 * @sk_omem_alloc: "o" is "option" or "other"
270 * @sk_wmem_queued: persistent queue size
271 * @sk_forward_alloc: space allocated forward
272 * @sk_napi_id: id of the last napi context to receive data for sk
273 * @sk_ll_usec: usecs to busypoll when there is no data
274 * @sk_allocation: allocation mode
275 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
276 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
277 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
278 * @sk_sndbuf: size of send buffer in bytes
279 * @__sk_flags_offset: empty field used to determine location of bitfield
280 * @sk_padding: unused element for alignment
281 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
282 * @sk_no_check_rx: allow zero checksum in RX packets
283 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
284 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
285 * @sk_route_forced_caps: static, forced route capabilities
286 * (set in tcp_init_sock())
287 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
288 * @sk_gso_max_size: Maximum GSO segment size to build
289 * @sk_gso_max_segs: Maximum number of GSO segments
290 * @sk_pacing_shift: scaling factor for TCP Small Queues
291 * @sk_lingertime: %SO_LINGER l_linger setting
292 * @sk_backlog: always used with the per-socket spinlock held
293 * @sk_callback_lock: used with the callbacks in the end of this struct
294 * @sk_error_queue: rarely used
295 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
296 * IPV6_ADDRFORM for instance)
297 * @sk_err: last error
298 * @sk_err_soft: errors that don't cause failure but are the cause of a
299 * persistent failure not just 'timed out'
300 * @sk_drops: raw/udp drops counter
301 * @sk_ack_backlog: current listen backlog
302 * @sk_max_ack_backlog: listen backlog set in listen()
303 * @sk_uid: user id of owner
304 * @sk_priority: %SO_PRIORITY setting
305 * @sk_type: socket type (%SOCK_STREAM, etc)
306 * @sk_protocol: which protocol this socket belongs in this network family
307 * @sk_peer_pid: &struct pid for this socket's peer
308 * @sk_peer_cred: %SO_PEERCRED setting
309 * @sk_rcvlowat: %SO_RCVLOWAT setting
310 * @sk_rcvtimeo: %SO_RCVTIMEO setting
311 * @sk_sndtimeo: %SO_SNDTIMEO setting
312 * @sk_txhash: computed flow hash for use on transmit
313 * @sk_filter: socket filtering instructions
314 * @sk_timer: sock cleanup timer
315 * @sk_stamp: time stamp of last packet received
316 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
317 * @sk_tsflags: SO_TIMESTAMPING socket options
318 * @sk_tskey: counter to disambiguate concurrent tstamp requests
319 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
320 * @sk_socket: Identd and reporting IO signals
321 * @sk_user_data: RPC layer private data
322 * @sk_frag: cached page frag
323 * @sk_peek_off: current peek_offset value
324 * @sk_send_head: front of stuff to transmit
325 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
326 * @sk_tx_skb_cache: cache copy of recently accessed TX skb
327 * @sk_security: used by security modules
328 * @sk_mark: generic packet mark
329 * @sk_cgrp_data: cgroup data for this cgroup
330 * @sk_memcg: this socket's memory cgroup association
331 * @sk_write_pending: a write to stream socket waits to start
332 * @sk_state_change: callback to indicate change in the state of the sock
333 * @sk_data_ready: callback to indicate there is data to be processed
334 * @sk_write_space: callback to indicate there is bf sending space available
335 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
336 * @sk_backlog_rcv: callback to process the backlog
337 * @sk_validate_xmit_skb: ptr to an optional validate function
338 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
339 * @sk_reuseport_cb: reuseport group container
340 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
341 * @sk_rcu: used during RCU grace period
342 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
343 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
344 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
345 * @sk_txtime_unused: unused txtime flags
349 * Now struct inet_timewait_sock also uses sock_common, so please just
350 * don't add nothing before this first member (__sk_common) --acme
352 struct sock_common __sk_common
;
353 #define sk_node __sk_common.skc_node
354 #define sk_nulls_node __sk_common.skc_nulls_node
355 #define sk_refcnt __sk_common.skc_refcnt
356 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
358 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
361 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
362 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
363 #define sk_hash __sk_common.skc_hash
364 #define sk_portpair __sk_common.skc_portpair
365 #define sk_num __sk_common.skc_num
366 #define sk_dport __sk_common.skc_dport
367 #define sk_addrpair __sk_common.skc_addrpair
368 #define sk_daddr __sk_common.skc_daddr
369 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
370 #define sk_family __sk_common.skc_family
371 #define sk_state __sk_common.skc_state
372 #define sk_reuse __sk_common.skc_reuse
373 #define sk_reuseport __sk_common.skc_reuseport
374 #define sk_ipv6only __sk_common.skc_ipv6only
375 #define sk_net_refcnt __sk_common.skc_net_refcnt
376 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
377 #define sk_bind_node __sk_common.skc_bind_node
378 #define sk_prot __sk_common.skc_prot
379 #define sk_net __sk_common.skc_net
380 #define sk_v6_daddr __sk_common.skc_v6_daddr
381 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
382 #define sk_cookie __sk_common.skc_cookie
383 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
384 #define sk_flags __sk_common.skc_flags
385 #define sk_rxhash __sk_common.skc_rxhash
387 socket_lock_t sk_lock
;
390 struct sk_buff_head sk_error_queue
;
391 struct sk_buff
*sk_rx_skb_cache
;
392 struct sk_buff_head sk_receive_queue
;
394 * The backlog queue is special, it is always used with
395 * the per-socket spinlock held and requires low latency
396 * access. Therefore we special case it's implementation.
397 * Note : rmem_alloc is in this structure to fill a hole
398 * on 64bit arches, not because its logically part of
404 struct sk_buff
*head
;
405 struct sk_buff
*tail
;
407 #define sk_rmem_alloc sk_backlog.rmem_alloc
409 int sk_forward_alloc
;
410 #ifdef CONFIG_NET_RX_BUSY_POLL
411 unsigned int sk_ll_usec
;
412 /* ===== mostly read cache line ===== */
413 unsigned int sk_napi_id
;
417 struct sk_filter __rcu
*sk_filter
;
419 struct socket_wq __rcu
*sk_wq
;
421 struct socket_wq
*sk_wq_raw
;
425 struct xfrm_policy __rcu
*sk_policy
[2];
427 struct dst_entry
*sk_rx_dst
;
428 struct dst_entry __rcu
*sk_dst_cache
;
429 atomic_t sk_omem_alloc
;
432 /* ===== cache line for TX ===== */
434 refcount_t sk_wmem_alloc
;
435 unsigned long sk_tsq_flags
;
437 struct sk_buff
*sk_send_head
;
438 struct rb_root tcp_rtx_queue
;
440 struct sk_buff
*sk_tx_skb_cache
;
441 struct sk_buff_head sk_write_queue
;
443 int sk_write_pending
;
444 __u32 sk_dst_pending_confirm
;
445 u32 sk_pacing_status
; /* see enum sk_pacing */
447 struct timer_list sk_timer
;
450 unsigned long sk_pacing_rate
; /* bytes per second */
451 unsigned long sk_max_pacing_rate
;
452 struct page_frag sk_frag
;
453 netdev_features_t sk_route_caps
;
454 netdev_features_t sk_route_nocaps
;
455 netdev_features_t sk_route_forced_caps
;
457 unsigned int sk_gso_max_size
;
462 * Because of non atomicity rules, all
463 * changes are protected by socket lock.
474 unsigned long sk_lingertime
;
475 struct proto
*sk_prot_creator
;
476 rwlock_t sk_callback_lock
;
480 u32 sk_max_ack_backlog
;
482 struct pid
*sk_peer_pid
;
483 const struct cred
*sk_peer_cred
;
486 #if BITS_PER_LONG==32
487 seqlock_t sk_stamp_seq
;
495 u8 sk_txtime_deadline_mode
: 1,
496 sk_txtime_report_errors
: 1,
497 sk_txtime_unused
: 6;
499 struct socket
*sk_socket
;
501 #ifdef CONFIG_SECURITY
504 struct sock_cgroup_data sk_cgrp_data
;
505 struct mem_cgroup
*sk_memcg
;
506 void (*sk_state_change
)(struct sock
*sk
);
507 void (*sk_data_ready
)(struct sock
*sk
);
508 void (*sk_write_space
)(struct sock
*sk
);
509 void (*sk_error_report
)(struct sock
*sk
);
510 int (*sk_backlog_rcv
)(struct sock
*sk
,
511 struct sk_buff
*skb
);
512 #ifdef CONFIG_SOCK_VALIDATE_XMIT
513 struct sk_buff
* (*sk_validate_xmit_skb
)(struct sock
*sk
,
514 struct net_device
*dev
,
515 struct sk_buff
*skb
);
517 void (*sk_destruct
)(struct sock
*sk
);
518 struct sock_reuseport __rcu
*sk_reuseport_cb
;
519 #ifdef CONFIG_BPF_SYSCALL
520 struct bpf_sk_storage __rcu
*sk_bpf_storage
;
522 struct rcu_head sk_rcu
;
527 SK_PACING_NEEDED
= 1,
531 /* Pointer stored in sk_user_data might not be suitable for copying
532 * when cloning the socket. For instance, it can point to a reference
533 * counted object. sk_user_data bottom bit is set if pointer must not
536 #define SK_USER_DATA_NOCOPY 1UL
537 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */
538 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF)
541 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
544 static inline bool sk_user_data_is_nocopy(const struct sock
*sk
)
546 return ((uintptr_t)sk
->sk_user_data
& SK_USER_DATA_NOCOPY
);
549 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
551 #define rcu_dereference_sk_user_data(sk) \
553 void *__tmp = rcu_dereference(__sk_user_data((sk))); \
554 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \
556 #define rcu_assign_sk_user_data(sk, ptr) \
558 uintptr_t __tmp = (uintptr_t)(ptr); \
559 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
560 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \
562 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \
564 uintptr_t __tmp = (uintptr_t)(ptr); \
565 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
566 rcu_assign_pointer(__sk_user_data((sk)), \
567 __tmp | SK_USER_DATA_NOCOPY); \
571 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
572 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
573 * on a socket means that the socket will reuse everybody else's port
574 * without looking at the other's sk_reuse value.
577 #define SK_NO_REUSE 0
578 #define SK_CAN_REUSE 1
579 #define SK_FORCE_REUSE 2
581 int sk_set_peek_off(struct sock
*sk
, int val
);
583 static inline int sk_peek_offset(struct sock
*sk
, int flags
)
585 if (unlikely(flags
& MSG_PEEK
)) {
586 return READ_ONCE(sk
->sk_peek_off
);
592 static inline void sk_peek_offset_bwd(struct sock
*sk
, int val
)
594 s32 off
= READ_ONCE(sk
->sk_peek_off
);
596 if (unlikely(off
>= 0)) {
597 off
= max_t(s32
, off
- val
, 0);
598 WRITE_ONCE(sk
->sk_peek_off
, off
);
602 static inline void sk_peek_offset_fwd(struct sock
*sk
, int val
)
604 sk_peek_offset_bwd(sk
, -val
);
608 * Hashed lists helper routines
610 static inline struct sock
*sk_entry(const struct hlist_node
*node
)
612 return hlist_entry(node
, struct sock
, sk_node
);
615 static inline struct sock
*__sk_head(const struct hlist_head
*head
)
617 return hlist_entry(head
->first
, struct sock
, sk_node
);
620 static inline struct sock
*sk_head(const struct hlist_head
*head
)
622 return hlist_empty(head
) ? NULL
: __sk_head(head
);
625 static inline struct sock
*__sk_nulls_head(const struct hlist_nulls_head
*head
)
627 return hlist_nulls_entry(head
->first
, struct sock
, sk_nulls_node
);
630 static inline struct sock
*sk_nulls_head(const struct hlist_nulls_head
*head
)
632 return hlist_nulls_empty(head
) ? NULL
: __sk_nulls_head(head
);
635 static inline struct sock
*sk_next(const struct sock
*sk
)
637 return hlist_entry_safe(sk
->sk_node
.next
, struct sock
, sk_node
);
640 static inline struct sock
*sk_nulls_next(const struct sock
*sk
)
642 return (!is_a_nulls(sk
->sk_nulls_node
.next
)) ?
643 hlist_nulls_entry(sk
->sk_nulls_node
.next
,
644 struct sock
, sk_nulls_node
) :
648 static inline bool sk_unhashed(const struct sock
*sk
)
650 return hlist_unhashed(&sk
->sk_node
);
653 static inline bool sk_hashed(const struct sock
*sk
)
655 return !sk_unhashed(sk
);
658 static inline void sk_node_init(struct hlist_node
*node
)
663 static inline void sk_nulls_node_init(struct hlist_nulls_node
*node
)
668 static inline void __sk_del_node(struct sock
*sk
)
670 __hlist_del(&sk
->sk_node
);
673 /* NB: equivalent to hlist_del_init_rcu */
674 static inline bool __sk_del_node_init(struct sock
*sk
)
678 sk_node_init(&sk
->sk_node
);
684 /* Grab socket reference count. This operation is valid only
685 when sk is ALREADY grabbed f.e. it is found in hash table
686 or a list and the lookup is made under lock preventing hash table
690 static __always_inline
void sock_hold(struct sock
*sk
)
692 refcount_inc(&sk
->sk_refcnt
);
695 /* Ungrab socket in the context, which assumes that socket refcnt
696 cannot hit zero, f.e. it is true in context of any socketcall.
698 static __always_inline
void __sock_put(struct sock
*sk
)
700 refcount_dec(&sk
->sk_refcnt
);
703 static inline bool sk_del_node_init(struct sock
*sk
)
705 bool rc
= __sk_del_node_init(sk
);
708 /* paranoid for a while -acme */
709 WARN_ON(refcount_read(&sk
->sk_refcnt
) == 1);
714 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
716 static inline bool __sk_nulls_del_node_init_rcu(struct sock
*sk
)
719 hlist_nulls_del_init_rcu(&sk
->sk_nulls_node
);
725 static inline bool sk_nulls_del_node_init_rcu(struct sock
*sk
)
727 bool rc
= __sk_nulls_del_node_init_rcu(sk
);
730 /* paranoid for a while -acme */
731 WARN_ON(refcount_read(&sk
->sk_refcnt
) == 1);
737 static inline void __sk_add_node(struct sock
*sk
, struct hlist_head
*list
)
739 hlist_add_head(&sk
->sk_node
, list
);
742 static inline void sk_add_node(struct sock
*sk
, struct hlist_head
*list
)
745 __sk_add_node(sk
, list
);
748 static inline void sk_add_node_rcu(struct sock
*sk
, struct hlist_head
*list
)
751 if (IS_ENABLED(CONFIG_IPV6
) && sk
->sk_reuseport
&&
752 sk
->sk_family
== AF_INET6
)
753 hlist_add_tail_rcu(&sk
->sk_node
, list
);
755 hlist_add_head_rcu(&sk
->sk_node
, list
);
758 static inline void sk_add_node_tail_rcu(struct sock
*sk
, struct hlist_head
*list
)
761 hlist_add_tail_rcu(&sk
->sk_node
, list
);
764 static inline void __sk_nulls_add_node_rcu(struct sock
*sk
, struct hlist_nulls_head
*list
)
766 hlist_nulls_add_head_rcu(&sk
->sk_nulls_node
, list
);
769 static inline void __sk_nulls_add_node_tail_rcu(struct sock
*sk
, struct hlist_nulls_head
*list
)
771 hlist_nulls_add_tail_rcu(&sk
->sk_nulls_node
, list
);
774 static inline void sk_nulls_add_node_rcu(struct sock
*sk
, struct hlist_nulls_head
*list
)
777 __sk_nulls_add_node_rcu(sk
, list
);
780 static inline void __sk_del_bind_node(struct sock
*sk
)
782 __hlist_del(&sk
->sk_bind_node
);
785 static inline void sk_add_bind_node(struct sock
*sk
,
786 struct hlist_head
*list
)
788 hlist_add_head(&sk
->sk_bind_node
, list
);
791 #define sk_for_each(__sk, list) \
792 hlist_for_each_entry(__sk, list, sk_node)
793 #define sk_for_each_rcu(__sk, list) \
794 hlist_for_each_entry_rcu(__sk, list, sk_node)
795 #define sk_nulls_for_each(__sk, node, list) \
796 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
797 #define sk_nulls_for_each_rcu(__sk, node, list) \
798 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
799 #define sk_for_each_from(__sk) \
800 hlist_for_each_entry_from(__sk, sk_node)
801 #define sk_nulls_for_each_from(__sk, node) \
802 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
803 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
804 #define sk_for_each_safe(__sk, tmp, list) \
805 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
806 #define sk_for_each_bound(__sk, list) \
807 hlist_for_each_entry(__sk, list, sk_bind_node)
810 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
811 * @tpos: the type * to use as a loop cursor.
812 * @pos: the &struct hlist_node to use as a loop cursor.
813 * @head: the head for your list.
814 * @offset: offset of hlist_node within the struct.
817 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
818 for (pos = rcu_dereference(hlist_first_rcu(head)); \
820 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
821 pos = rcu_dereference(hlist_next_rcu(pos)))
823 static inline struct user_namespace
*sk_user_ns(struct sock
*sk
)
825 /* Careful only use this in a context where these parameters
826 * can not change and must all be valid, such as recvmsg from
829 return sk
->sk_socket
->file
->f_cred
->user_ns
;
843 SOCK_USE_WRITE_QUEUE
, /* whether to call sk->sk_write_space in sock_wfree */
844 SOCK_DBG
, /* %SO_DEBUG setting */
845 SOCK_RCVTSTAMP
, /* %SO_TIMESTAMP setting */
846 SOCK_RCVTSTAMPNS
, /* %SO_TIMESTAMPNS setting */
847 SOCK_LOCALROUTE
, /* route locally only, %SO_DONTROUTE setting */
848 SOCK_QUEUE_SHRUNK
, /* write queue has been shrunk recently */
849 SOCK_MEMALLOC
, /* VM depends on this socket for swapping */
850 SOCK_TIMESTAMPING_RX_SOFTWARE
, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
851 SOCK_FASYNC
, /* fasync() active */
853 SOCK_ZEROCOPY
, /* buffers from userspace */
854 SOCK_WIFI_STATUS
, /* push wifi status to userspace */
855 SOCK_NOFCS
, /* Tell NIC not to do the Ethernet FCS.
856 * Will use last 4 bytes of packet sent from
857 * user-space instead.
859 SOCK_FILTER_LOCKED
, /* Filter cannot be changed anymore */
860 SOCK_SELECT_ERR_QUEUE
, /* Wake select on error queue */
861 SOCK_RCU_FREE
, /* wait rcu grace period in sk_destruct() */
863 SOCK_XDP
, /* XDP is attached */
864 SOCK_TSTAMP_NEW
, /* Indicates 64 bit timestamps always */
867 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
869 static inline void sock_copy_flags(struct sock
*nsk
, struct sock
*osk
)
871 nsk
->sk_flags
= osk
->sk_flags
;
874 static inline void sock_set_flag(struct sock
*sk
, enum sock_flags flag
)
876 __set_bit(flag
, &sk
->sk_flags
);
879 static inline void sock_reset_flag(struct sock
*sk
, enum sock_flags flag
)
881 __clear_bit(flag
, &sk
->sk_flags
);
884 static inline void sock_valbool_flag(struct sock
*sk
, enum sock_flags bit
,
888 sock_set_flag(sk
, bit
);
890 sock_reset_flag(sk
, bit
);
893 static inline bool sock_flag(const struct sock
*sk
, enum sock_flags flag
)
895 return test_bit(flag
, &sk
->sk_flags
);
899 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key
);
900 static inline int sk_memalloc_socks(void)
902 return static_branch_unlikely(&memalloc_socks_key
);
905 void __receive_sock(struct file
*file
);
908 static inline int sk_memalloc_socks(void)
913 static inline void __receive_sock(struct file
*file
)
917 static inline gfp_t
sk_gfp_mask(const struct sock
*sk
, gfp_t gfp_mask
)
919 return gfp_mask
| (sk
->sk_allocation
& __GFP_MEMALLOC
);
922 static inline void sk_acceptq_removed(struct sock
*sk
)
924 WRITE_ONCE(sk
->sk_ack_backlog
, sk
->sk_ack_backlog
- 1);
927 static inline void sk_acceptq_added(struct sock
*sk
)
929 WRITE_ONCE(sk
->sk_ack_backlog
, sk
->sk_ack_backlog
+ 1);
932 static inline bool sk_acceptq_is_full(const struct sock
*sk
)
934 return READ_ONCE(sk
->sk_ack_backlog
) > READ_ONCE(sk
->sk_max_ack_backlog
);
938 * Compute minimal free write space needed to queue new packets.
940 static inline int sk_stream_min_wspace(const struct sock
*sk
)
942 return READ_ONCE(sk
->sk_wmem_queued
) >> 1;
945 static inline int sk_stream_wspace(const struct sock
*sk
)
947 return READ_ONCE(sk
->sk_sndbuf
) - READ_ONCE(sk
->sk_wmem_queued
);
950 static inline void sk_wmem_queued_add(struct sock
*sk
, int val
)
952 WRITE_ONCE(sk
->sk_wmem_queued
, sk
->sk_wmem_queued
+ val
);
955 void sk_stream_write_space(struct sock
*sk
);
957 /* OOB backlog add */
958 static inline void __sk_add_backlog(struct sock
*sk
, struct sk_buff
*skb
)
960 /* dont let skb dst not refcounted, we are going to leave rcu lock */
963 if (!sk
->sk_backlog
.tail
)
964 WRITE_ONCE(sk
->sk_backlog
.head
, skb
);
966 sk
->sk_backlog
.tail
->next
= skb
;
968 WRITE_ONCE(sk
->sk_backlog
.tail
, skb
);
973 * Take into account size of receive queue and backlog queue
974 * Do not take into account this skb truesize,
975 * to allow even a single big packet to come.
977 static inline bool sk_rcvqueues_full(const struct sock
*sk
, unsigned int limit
)
979 unsigned int qsize
= sk
->sk_backlog
.len
+ atomic_read(&sk
->sk_rmem_alloc
);
981 return qsize
> limit
;
984 /* The per-socket spinlock must be held here. */
985 static inline __must_check
int sk_add_backlog(struct sock
*sk
, struct sk_buff
*skb
,
988 if (sk_rcvqueues_full(sk
, limit
))
992 * If the skb was allocated from pfmemalloc reserves, only
993 * allow SOCK_MEMALLOC sockets to use it as this socket is
994 * helping free memory
996 if (skb_pfmemalloc(skb
) && !sock_flag(sk
, SOCK_MEMALLOC
))
999 __sk_add_backlog(sk
, skb
);
1000 sk
->sk_backlog
.len
+= skb
->truesize
;
1004 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
);
1006 static inline int sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1008 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
1009 return __sk_backlog_rcv(sk
, skb
);
1011 return sk
->sk_backlog_rcv(sk
, skb
);
1014 static inline void sk_incoming_cpu_update(struct sock
*sk
)
1016 int cpu
= raw_smp_processor_id();
1018 if (unlikely(READ_ONCE(sk
->sk_incoming_cpu
) != cpu
))
1019 WRITE_ONCE(sk
->sk_incoming_cpu
, cpu
);
1022 static inline void sock_rps_record_flow_hash(__u32 hash
)
1025 struct rps_sock_flow_table
*sock_flow_table
;
1028 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
1029 rps_record_sock_flow(sock_flow_table
, hash
);
1034 static inline void sock_rps_record_flow(const struct sock
*sk
)
1037 if (static_branch_unlikely(&rfs_needed
)) {
1038 /* Reading sk->sk_rxhash might incur an expensive cache line
1041 * TCP_ESTABLISHED does cover almost all states where RFS
1042 * might be useful, and is cheaper [1] than testing :
1043 * IPv4: inet_sk(sk)->inet_daddr
1044 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1045 * OR an additional socket flag
1046 * [1] : sk_state and sk_prot are in the same cache line.
1048 if (sk
->sk_state
== TCP_ESTABLISHED
)
1049 sock_rps_record_flow_hash(sk
->sk_rxhash
);
1054 static inline void sock_rps_save_rxhash(struct sock
*sk
,
1055 const struct sk_buff
*skb
)
1058 if (unlikely(sk
->sk_rxhash
!= skb
->hash
))
1059 sk
->sk_rxhash
= skb
->hash
;
1063 static inline void sock_rps_reset_rxhash(struct sock
*sk
)
1070 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1072 release_sock(__sk); \
1073 __rc = __condition; \
1075 *(__timeo) = wait_woken(__wait, \
1076 TASK_INTERRUPTIBLE, \
1079 sched_annotate_sleep(); \
1081 __rc = __condition; \
1085 int sk_stream_wait_connect(struct sock
*sk
, long *timeo_p
);
1086 int sk_stream_wait_memory(struct sock
*sk
, long *timeo_p
);
1087 void sk_stream_wait_close(struct sock
*sk
, long timeo_p
);
1088 int sk_stream_error(struct sock
*sk
, int flags
, int err
);
1089 void sk_stream_kill_queues(struct sock
*sk
);
1090 void sk_set_memalloc(struct sock
*sk
);
1091 void sk_clear_memalloc(struct sock
*sk
);
1093 void __sk_flush_backlog(struct sock
*sk
);
1095 static inline bool sk_flush_backlog(struct sock
*sk
)
1097 if (unlikely(READ_ONCE(sk
->sk_backlog
.tail
))) {
1098 __sk_flush_backlog(sk
);
1104 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
);
1106 struct request_sock_ops
;
1107 struct timewait_sock_ops
;
1108 struct inet_hashinfo
;
1109 struct raw_hashinfo
;
1110 struct smc_hashinfo
;
1114 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1115 * un-modified. Special care is taken when initializing object to zero.
1117 static inline void sk_prot_clear_nulls(struct sock
*sk
, int size
)
1119 if (offsetof(struct sock
, sk_node
.next
) != 0)
1120 memset(sk
, 0, offsetof(struct sock
, sk_node
.next
));
1121 memset(&sk
->sk_node
.pprev
, 0,
1122 size
- offsetof(struct sock
, sk_node
.pprev
));
1125 /* Networking protocol blocks we attach to sockets.
1126 * socket layer -> transport layer interface
1129 void (*close
)(struct sock
*sk
,
1131 int (*pre_connect
)(struct sock
*sk
,
1132 struct sockaddr
*uaddr
,
1134 int (*connect
)(struct sock
*sk
,
1135 struct sockaddr
*uaddr
,
1137 int (*disconnect
)(struct sock
*sk
, int flags
);
1139 struct sock
* (*accept
)(struct sock
*sk
, int flags
, int *err
,
1142 int (*ioctl
)(struct sock
*sk
, int cmd
,
1144 int (*init
)(struct sock
*sk
);
1145 void (*destroy
)(struct sock
*sk
);
1146 void (*shutdown
)(struct sock
*sk
, int how
);
1147 int (*setsockopt
)(struct sock
*sk
, int level
,
1148 int optname
, sockptr_t optval
,
1149 unsigned int optlen
);
1150 int (*getsockopt
)(struct sock
*sk
, int level
,
1151 int optname
, char __user
*optval
,
1152 int __user
*option
);
1153 void (*keepalive
)(struct sock
*sk
, int valbool
);
1154 #ifdef CONFIG_COMPAT
1155 int (*compat_ioctl
)(struct sock
*sk
,
1156 unsigned int cmd
, unsigned long arg
);
1158 int (*sendmsg
)(struct sock
*sk
, struct msghdr
*msg
,
1160 int (*recvmsg
)(struct sock
*sk
, struct msghdr
*msg
,
1161 size_t len
, int noblock
, int flags
,
1163 int (*sendpage
)(struct sock
*sk
, struct page
*page
,
1164 int offset
, size_t size
, int flags
);
1165 int (*bind
)(struct sock
*sk
,
1166 struct sockaddr
*addr
, int addr_len
);
1167 int (*bind_add
)(struct sock
*sk
,
1168 struct sockaddr
*addr
, int addr_len
);
1170 int (*backlog_rcv
) (struct sock
*sk
,
1171 struct sk_buff
*skb
);
1173 void (*release_cb
)(struct sock
*sk
);
1175 /* Keeping track of sk's, looking them up, and port selection methods. */
1176 int (*hash
)(struct sock
*sk
);
1177 void (*unhash
)(struct sock
*sk
);
1178 void (*rehash
)(struct sock
*sk
);
1179 int (*get_port
)(struct sock
*sk
, unsigned short snum
);
1181 /* Keeping track of sockets in use */
1182 #ifdef CONFIG_PROC_FS
1183 unsigned int inuse_idx
;
1186 bool (*stream_memory_free
)(const struct sock
*sk
, int wake
);
1187 bool (*stream_memory_read
)(const struct sock
*sk
);
1188 /* Memory pressure */
1189 void (*enter_memory_pressure
)(struct sock
*sk
);
1190 void (*leave_memory_pressure
)(struct sock
*sk
);
1191 atomic_long_t
*memory_allocated
; /* Current allocated memory. */
1192 struct percpu_counter
*sockets_allocated
; /* Current number of sockets. */
1194 * Pressure flag: try to collapse.
1195 * Technical note: it is used by multiple contexts non atomically.
1196 * All the __sk_mem_schedule() is of this nature: accounting
1197 * is strict, actions are advisory and have some latency.
1199 unsigned long *memory_pressure
;
1204 u32 sysctl_wmem_offset
;
1205 u32 sysctl_rmem_offset
;
1210 struct kmem_cache
*slab
;
1211 unsigned int obj_size
;
1212 slab_flags_t slab_flags
;
1213 unsigned int useroffset
; /* Usercopy region offset */
1214 unsigned int usersize
; /* Usercopy region size */
1216 struct percpu_counter
*orphan_count
;
1218 struct request_sock_ops
*rsk_prot
;
1219 struct timewait_sock_ops
*twsk_prot
;
1222 struct inet_hashinfo
*hashinfo
;
1223 struct udp_table
*udp_table
;
1224 struct raw_hashinfo
*raw_hash
;
1225 struct smc_hashinfo
*smc_hash
;
1228 struct module
*owner
;
1232 struct list_head node
;
1233 #ifdef SOCK_REFCNT_DEBUG
1236 int (*diag_destroy
)(struct sock
*sk
, int err
);
1237 } __randomize_layout
;
1239 int proto_register(struct proto
*prot
, int alloc_slab
);
1240 void proto_unregister(struct proto
*prot
);
1241 int sock_load_diag_module(int family
, int protocol
);
1243 #ifdef SOCK_REFCNT_DEBUG
1244 static inline void sk_refcnt_debug_inc(struct sock
*sk
)
1246 atomic_inc(&sk
->sk_prot
->socks
);
1249 static inline void sk_refcnt_debug_dec(struct sock
*sk
)
1251 atomic_dec(&sk
->sk_prot
->socks
);
1252 printk(KERN_DEBUG
"%s socket %p released, %d are still alive\n",
1253 sk
->sk_prot
->name
, sk
, atomic_read(&sk
->sk_prot
->socks
));
1256 static inline void sk_refcnt_debug_release(const struct sock
*sk
)
1258 if (refcount_read(&sk
->sk_refcnt
) != 1)
1259 printk(KERN_DEBUG
"Destruction of the %s socket %p delayed, refcnt=%d\n",
1260 sk
->sk_prot
->name
, sk
, refcount_read(&sk
->sk_refcnt
));
1262 #else /* SOCK_REFCNT_DEBUG */
1263 #define sk_refcnt_debug_inc(sk) do { } while (0)
1264 #define sk_refcnt_debug_dec(sk) do { } while (0)
1265 #define sk_refcnt_debug_release(sk) do { } while (0)
1266 #endif /* SOCK_REFCNT_DEBUG */
1268 static inline bool __sk_stream_memory_free(const struct sock
*sk
, int wake
)
1270 if (READ_ONCE(sk
->sk_wmem_queued
) >= READ_ONCE(sk
->sk_sndbuf
))
1273 return sk
->sk_prot
->stream_memory_free
?
1274 sk
->sk_prot
->stream_memory_free(sk
, wake
) : true;
1277 static inline bool sk_stream_memory_free(const struct sock
*sk
)
1279 return __sk_stream_memory_free(sk
, 0);
1282 static inline bool __sk_stream_is_writeable(const struct sock
*sk
, int wake
)
1284 return sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
) &&
1285 __sk_stream_memory_free(sk
, wake
);
1288 static inline bool sk_stream_is_writeable(const struct sock
*sk
)
1290 return __sk_stream_is_writeable(sk
, 0);
1293 static inline int sk_under_cgroup_hierarchy(struct sock
*sk
,
1294 struct cgroup
*ancestor
)
1296 #ifdef CONFIG_SOCK_CGROUP_DATA
1297 return cgroup_is_descendant(sock_cgroup_ptr(&sk
->sk_cgrp_data
),
1304 static inline bool sk_has_memory_pressure(const struct sock
*sk
)
1306 return sk
->sk_prot
->memory_pressure
!= NULL
;
1309 static inline bool sk_under_memory_pressure(const struct sock
*sk
)
1311 if (!sk
->sk_prot
->memory_pressure
)
1314 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
1315 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
1318 return !!*sk
->sk_prot
->memory_pressure
;
1322 sk_memory_allocated(const struct sock
*sk
)
1324 return atomic_long_read(sk
->sk_prot
->memory_allocated
);
1328 sk_memory_allocated_add(struct sock
*sk
, int amt
)
1330 return atomic_long_add_return(amt
, sk
->sk_prot
->memory_allocated
);
1334 sk_memory_allocated_sub(struct sock
*sk
, int amt
)
1336 atomic_long_sub(amt
, sk
->sk_prot
->memory_allocated
);
1339 static inline void sk_sockets_allocated_dec(struct sock
*sk
)
1341 percpu_counter_dec(sk
->sk_prot
->sockets_allocated
);
1344 static inline void sk_sockets_allocated_inc(struct sock
*sk
)
1346 percpu_counter_inc(sk
->sk_prot
->sockets_allocated
);
1350 sk_sockets_allocated_read_positive(struct sock
*sk
)
1352 return percpu_counter_read_positive(sk
->sk_prot
->sockets_allocated
);
1356 proto_sockets_allocated_sum_positive(struct proto
*prot
)
1358 return percpu_counter_sum_positive(prot
->sockets_allocated
);
1362 proto_memory_allocated(struct proto
*prot
)
1364 return atomic_long_read(prot
->memory_allocated
);
1368 proto_memory_pressure(struct proto
*prot
)
1370 if (!prot
->memory_pressure
)
1372 return !!*prot
->memory_pressure
;
1376 #ifdef CONFIG_PROC_FS
1377 /* Called with local bh disabled */
1378 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int inc
);
1379 int sock_prot_inuse_get(struct net
*net
, struct proto
*proto
);
1380 int sock_inuse_get(struct net
*net
);
1382 static inline void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
,
1389 /* With per-bucket locks this operation is not-atomic, so that
1390 * this version is not worse.
1392 static inline int __sk_prot_rehash(struct sock
*sk
)
1394 sk
->sk_prot
->unhash(sk
);
1395 return sk
->sk_prot
->hash(sk
);
1398 /* About 10 seconds */
1399 #define SOCK_DESTROY_TIME (10*HZ)
1401 /* Sockets 0-1023 can't be bound to unless you are superuser */
1402 #define PROT_SOCK 1024
1404 #define SHUTDOWN_MASK 3
1405 #define RCV_SHUTDOWN 1
1406 #define SEND_SHUTDOWN 2
1408 #define SOCK_SNDBUF_LOCK 1
1409 #define SOCK_RCVBUF_LOCK 2
1410 #define SOCK_BINDADDR_LOCK 4
1411 #define SOCK_BINDPORT_LOCK 8
1413 struct socket_alloc
{
1414 struct socket socket
;
1415 struct inode vfs_inode
;
1418 static inline struct socket
*SOCKET_I(struct inode
*inode
)
1420 return &container_of(inode
, struct socket_alloc
, vfs_inode
)->socket
;
1423 static inline struct inode
*SOCK_INODE(struct socket
*socket
)
1425 return &container_of(socket
, struct socket_alloc
, socket
)->vfs_inode
;
1429 * Functions for memory accounting
1431 int __sk_mem_raise_allocated(struct sock
*sk
, int size
, int amt
, int kind
);
1432 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
);
1433 void __sk_mem_reduce_allocated(struct sock
*sk
, int amount
);
1434 void __sk_mem_reclaim(struct sock
*sk
, int amount
);
1436 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1437 * do not necessarily have 16x time more memory than 4KB ones.
1439 #define SK_MEM_QUANTUM 4096
1440 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1441 #define SK_MEM_SEND 0
1442 #define SK_MEM_RECV 1
1444 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1445 static inline long sk_prot_mem_limits(const struct sock
*sk
, int index
)
1447 long val
= sk
->sk_prot
->sysctl_mem
[index
];
1449 #if PAGE_SIZE > SK_MEM_QUANTUM
1450 val
<<= PAGE_SHIFT
- SK_MEM_QUANTUM_SHIFT
;
1451 #elif PAGE_SIZE < SK_MEM_QUANTUM
1452 val
>>= SK_MEM_QUANTUM_SHIFT
- PAGE_SHIFT
;
1457 static inline int sk_mem_pages(int amt
)
1459 return (amt
+ SK_MEM_QUANTUM
- 1) >> SK_MEM_QUANTUM_SHIFT
;
1462 static inline bool sk_has_account(struct sock
*sk
)
1464 /* return true if protocol supports memory accounting */
1465 return !!sk
->sk_prot
->memory_allocated
;
1468 static inline bool sk_wmem_schedule(struct sock
*sk
, int size
)
1470 if (!sk_has_account(sk
))
1472 return size
<= sk
->sk_forward_alloc
||
1473 __sk_mem_schedule(sk
, size
, SK_MEM_SEND
);
1477 sk_rmem_schedule(struct sock
*sk
, struct sk_buff
*skb
, int size
)
1479 if (!sk_has_account(sk
))
1481 return size
<= sk
->sk_forward_alloc
||
1482 __sk_mem_schedule(sk
, size
, SK_MEM_RECV
) ||
1483 skb_pfmemalloc(skb
);
1486 static inline void sk_mem_reclaim(struct sock
*sk
)
1488 if (!sk_has_account(sk
))
1490 if (sk
->sk_forward_alloc
>= SK_MEM_QUANTUM
)
1491 __sk_mem_reclaim(sk
, sk
->sk_forward_alloc
);
1494 static inline void sk_mem_reclaim_partial(struct sock
*sk
)
1496 if (!sk_has_account(sk
))
1498 if (sk
->sk_forward_alloc
> SK_MEM_QUANTUM
)
1499 __sk_mem_reclaim(sk
, sk
->sk_forward_alloc
- 1);
1502 static inline void sk_mem_charge(struct sock
*sk
, int size
)
1504 if (!sk_has_account(sk
))
1506 sk
->sk_forward_alloc
-= size
;
1509 static inline void sk_mem_uncharge(struct sock
*sk
, int size
)
1511 if (!sk_has_account(sk
))
1513 sk
->sk_forward_alloc
+= size
;
1515 /* Avoid a possible overflow.
1516 * TCP send queues can make this happen, if sk_mem_reclaim()
1517 * is not called and more than 2 GBytes are released at once.
1519 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1520 * no need to hold that much forward allocation anyway.
1522 if (unlikely(sk
->sk_forward_alloc
>= 1 << 21))
1523 __sk_mem_reclaim(sk
, 1 << 20);
1526 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key
);
1527 static inline void sk_wmem_free_skb(struct sock
*sk
, struct sk_buff
*skb
)
1529 sock_set_flag(sk
, SOCK_QUEUE_SHRUNK
);
1530 sk_wmem_queued_add(sk
, -skb
->truesize
);
1531 sk_mem_uncharge(sk
, skb
->truesize
);
1532 if (static_branch_unlikely(&tcp_tx_skb_cache_key
) &&
1533 !sk
->sk_tx_skb_cache
&& !skb_cloned(skb
)) {
1535 skb_zcopy_clear(skb
, true);
1536 sk
->sk_tx_skb_cache
= skb
;
1542 static inline void sock_release_ownership(struct sock
*sk
)
1544 if (sk
->sk_lock
.owned
) {
1545 sk
->sk_lock
.owned
= 0;
1547 /* The sk_lock has mutex_unlock() semantics: */
1548 mutex_release(&sk
->sk_lock
.dep_map
, _RET_IP_
);
1553 * Macro so as to not evaluate some arguments when
1554 * lockdep is not enabled.
1556 * Mark both the sk_lock and the sk_lock.slock as a
1557 * per-address-family lock class.
1559 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1561 sk->sk_lock.owned = 0; \
1562 init_waitqueue_head(&sk->sk_lock.wq); \
1563 spin_lock_init(&(sk)->sk_lock.slock); \
1564 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1565 sizeof((sk)->sk_lock)); \
1566 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1568 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1571 #ifdef CONFIG_LOCKDEP
1572 static inline bool lockdep_sock_is_held(const struct sock
*sk
)
1574 return lockdep_is_held(&sk
->sk_lock
) ||
1575 lockdep_is_held(&sk
->sk_lock
.slock
);
1579 void lock_sock_nested(struct sock
*sk
, int subclass
);
1581 static inline void lock_sock(struct sock
*sk
)
1583 lock_sock_nested(sk
, 0);
1586 void __release_sock(struct sock
*sk
);
1587 void release_sock(struct sock
*sk
);
1589 /* BH context may only use the following locking interface. */
1590 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1591 #define bh_lock_sock_nested(__sk) \
1592 spin_lock_nested(&((__sk)->sk_lock.slock), \
1593 SINGLE_DEPTH_NESTING)
1594 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1596 bool lock_sock_fast(struct sock
*sk
);
1598 * unlock_sock_fast - complement of lock_sock_fast
1602 * fast unlock socket for user context.
1603 * If slow mode is on, we call regular release_sock()
1605 static inline void unlock_sock_fast(struct sock
*sk
, bool slow
)
1610 spin_unlock_bh(&sk
->sk_lock
.slock
);
1613 /* Used by processes to "lock" a socket state, so that
1614 * interrupts and bottom half handlers won't change it
1615 * from under us. It essentially blocks any incoming
1616 * packets, so that we won't get any new data or any
1617 * packets that change the state of the socket.
1619 * While locked, BH processing will add new packets to
1620 * the backlog queue. This queue is processed by the
1621 * owner of the socket lock right before it is released.
1623 * Since ~2.3.5 it is also exclusive sleep lock serializing
1624 * accesses from user process context.
1627 static inline void sock_owned_by_me(const struct sock
*sk
)
1629 #ifdef CONFIG_LOCKDEP
1630 WARN_ON_ONCE(!lockdep_sock_is_held(sk
) && debug_locks
);
1634 static inline bool sock_owned_by_user(const struct sock
*sk
)
1636 sock_owned_by_me(sk
);
1637 return sk
->sk_lock
.owned
;
1640 static inline bool sock_owned_by_user_nocheck(const struct sock
*sk
)
1642 return sk
->sk_lock
.owned
;
1645 /* no reclassification while locks are held */
1646 static inline bool sock_allow_reclassification(const struct sock
*csk
)
1648 struct sock
*sk
= (struct sock
*)csk
;
1650 return !sk
->sk_lock
.owned
&& !spin_is_locked(&sk
->sk_lock
.slock
);
1653 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1654 struct proto
*prot
, int kern
);
1655 void sk_free(struct sock
*sk
);
1656 void sk_destruct(struct sock
*sk
);
1657 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
);
1658 void sk_free_unlock_clone(struct sock
*sk
);
1660 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1662 void __sock_wfree(struct sk_buff
*skb
);
1663 void sock_wfree(struct sk_buff
*skb
);
1664 struct sk_buff
*sock_omalloc(struct sock
*sk
, unsigned long size
,
1666 void skb_orphan_partial(struct sk_buff
*skb
);
1667 void sock_rfree(struct sk_buff
*skb
);
1668 void sock_efree(struct sk_buff
*skb
);
1670 void sock_edemux(struct sk_buff
*skb
);
1671 void sock_pfree(struct sk_buff
*skb
);
1673 #define sock_edemux sock_efree
1676 int sock_setsockopt(struct socket
*sock
, int level
, int op
,
1677 sockptr_t optval
, unsigned int optlen
);
1679 int sock_getsockopt(struct socket
*sock
, int level
, int op
,
1680 char __user
*optval
, int __user
*optlen
);
1681 int sock_gettstamp(struct socket
*sock
, void __user
*userstamp
,
1682 bool timeval
, bool time32
);
1683 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1684 int noblock
, int *errcode
);
1685 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1686 unsigned long data_len
, int noblock
,
1687 int *errcode
, int max_page_order
);
1688 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
);
1689 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
);
1690 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
);
1691 void sk_send_sigurg(struct sock
*sk
);
1693 struct sockcm_cookie
{
1699 static inline void sockcm_init(struct sockcm_cookie
*sockc
,
1700 const struct sock
*sk
)
1702 *sockc
= (struct sockcm_cookie
) { .tsflags
= sk
->sk_tsflags
};
1705 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
1706 struct sockcm_cookie
*sockc
);
1707 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
1708 struct sockcm_cookie
*sockc
);
1711 * Functions to fill in entries in struct proto_ops when a protocol
1712 * does not implement a particular function.
1714 int sock_no_bind(struct socket
*, struct sockaddr
*, int);
1715 int sock_no_connect(struct socket
*, struct sockaddr
*, int, int);
1716 int sock_no_socketpair(struct socket
*, struct socket
*);
1717 int sock_no_accept(struct socket
*, struct socket
*, int, bool);
1718 int sock_no_getname(struct socket
*, struct sockaddr
*, int);
1719 int sock_no_ioctl(struct socket
*, unsigned int, unsigned long);
1720 int sock_no_listen(struct socket
*, int);
1721 int sock_no_shutdown(struct socket
*, int);
1722 int sock_no_sendmsg(struct socket
*, struct msghdr
*, size_t);
1723 int sock_no_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t len
);
1724 int sock_no_recvmsg(struct socket
*, struct msghdr
*, size_t, int);
1725 int sock_no_mmap(struct file
*file
, struct socket
*sock
,
1726 struct vm_area_struct
*vma
);
1727 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
1728 size_t size
, int flags
);
1729 ssize_t
sock_no_sendpage_locked(struct sock
*sk
, struct page
*page
,
1730 int offset
, size_t size
, int flags
);
1733 * Functions to fill in entries in struct proto_ops when a protocol
1734 * uses the inet style.
1736 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1737 char __user
*optval
, int __user
*optlen
);
1738 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
1740 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1741 sockptr_t optval
, unsigned int optlen
);
1743 void sk_common_release(struct sock
*sk
);
1746 * Default socket callbacks and setup code
1749 /* Initialise core socket variables */
1750 void sock_init_data(struct socket
*sock
, struct sock
*sk
);
1753 * Socket reference counting postulates.
1755 * * Each user of socket SHOULD hold a reference count.
1756 * * Each access point to socket (an hash table bucket, reference from a list,
1757 * running timer, skb in flight MUST hold a reference count.
1758 * * When reference count hits 0, it means it will never increase back.
1759 * * When reference count hits 0, it means that no references from
1760 * outside exist to this socket and current process on current CPU
1761 * is last user and may/should destroy this socket.
1762 * * sk_free is called from any context: process, BH, IRQ. When
1763 * it is called, socket has no references from outside -> sk_free
1764 * may release descendant resources allocated by the socket, but
1765 * to the time when it is called, socket is NOT referenced by any
1766 * hash tables, lists etc.
1767 * * Packets, delivered from outside (from network or from another process)
1768 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1769 * when they sit in queue. Otherwise, packets will leak to hole, when
1770 * socket is looked up by one cpu and unhasing is made by another CPU.
1771 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1772 * (leak to backlog). Packet socket does all the processing inside
1773 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1774 * use separate SMP lock, so that they are prone too.
1777 /* Ungrab socket and destroy it, if it was the last reference. */
1778 static inline void sock_put(struct sock
*sk
)
1780 if (refcount_dec_and_test(&sk
->sk_refcnt
))
1783 /* Generic version of sock_put(), dealing with all sockets
1784 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1786 void sock_gen_put(struct sock
*sk
);
1788 int __sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
,
1789 unsigned int trim_cap
, bool refcounted
);
1790 static inline int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
,
1793 return __sk_receive_skb(sk
, skb
, nested
, 1, true);
1796 static inline void sk_tx_queue_set(struct sock
*sk
, int tx_queue
)
1798 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1799 if (WARN_ON_ONCE((unsigned short)tx_queue
>= USHRT_MAX
))
1801 sk
->sk_tx_queue_mapping
= tx_queue
;
1804 #define NO_QUEUE_MAPPING USHRT_MAX
1806 static inline void sk_tx_queue_clear(struct sock
*sk
)
1808 sk
->sk_tx_queue_mapping
= NO_QUEUE_MAPPING
;
1811 static inline int sk_tx_queue_get(const struct sock
*sk
)
1813 if (sk
&& sk
->sk_tx_queue_mapping
!= NO_QUEUE_MAPPING
)
1814 return sk
->sk_tx_queue_mapping
;
1819 static inline void sk_rx_queue_set(struct sock
*sk
, const struct sk_buff
*skb
)
1822 if (skb_rx_queue_recorded(skb
)) {
1823 u16 rx_queue
= skb_get_rx_queue(skb
);
1825 if (WARN_ON_ONCE(rx_queue
== NO_QUEUE_MAPPING
))
1828 sk
->sk_rx_queue_mapping
= rx_queue
;
1833 static inline void sk_rx_queue_clear(struct sock
*sk
)
1836 sk
->sk_rx_queue_mapping
= NO_QUEUE_MAPPING
;
1841 static inline int sk_rx_queue_get(const struct sock
*sk
)
1843 if (sk
&& sk
->sk_rx_queue_mapping
!= NO_QUEUE_MAPPING
)
1844 return sk
->sk_rx_queue_mapping
;
1850 static inline void sk_set_socket(struct sock
*sk
, struct socket
*sock
)
1852 sk
->sk_socket
= sock
;
1855 static inline wait_queue_head_t
*sk_sleep(struct sock
*sk
)
1857 BUILD_BUG_ON(offsetof(struct socket_wq
, wait
) != 0);
1858 return &rcu_dereference_raw(sk
->sk_wq
)->wait
;
1860 /* Detach socket from process context.
1861 * Announce socket dead, detach it from wait queue and inode.
1862 * Note that parent inode held reference count on this struct sock,
1863 * we do not release it in this function, because protocol
1864 * probably wants some additional cleanups or even continuing
1865 * to work with this socket (TCP).
1867 static inline void sock_orphan(struct sock
*sk
)
1869 write_lock_bh(&sk
->sk_callback_lock
);
1870 sock_set_flag(sk
, SOCK_DEAD
);
1871 sk_set_socket(sk
, NULL
);
1873 write_unlock_bh(&sk
->sk_callback_lock
);
1876 static inline void sock_graft(struct sock
*sk
, struct socket
*parent
)
1878 WARN_ON(parent
->sk
);
1879 write_lock_bh(&sk
->sk_callback_lock
);
1880 rcu_assign_pointer(sk
->sk_wq
, &parent
->wq
);
1882 sk_set_socket(sk
, parent
);
1883 sk
->sk_uid
= SOCK_INODE(parent
)->i_uid
;
1884 security_sock_graft(sk
, parent
);
1885 write_unlock_bh(&sk
->sk_callback_lock
);
1888 kuid_t
sock_i_uid(struct sock
*sk
);
1889 unsigned long sock_i_ino(struct sock
*sk
);
1891 static inline kuid_t
sock_net_uid(const struct net
*net
, const struct sock
*sk
)
1893 return sk
? sk
->sk_uid
: make_kuid(net
->user_ns
, 0);
1896 static inline u32
net_tx_rndhash(void)
1898 u32 v
= prandom_u32();
1903 static inline void sk_set_txhash(struct sock
*sk
)
1905 sk
->sk_txhash
= net_tx_rndhash();
1908 static inline void sk_rethink_txhash(struct sock
*sk
)
1914 static inline struct dst_entry
*
1915 __sk_dst_get(struct sock
*sk
)
1917 return rcu_dereference_check(sk
->sk_dst_cache
,
1918 lockdep_sock_is_held(sk
));
1921 static inline struct dst_entry
*
1922 sk_dst_get(struct sock
*sk
)
1924 struct dst_entry
*dst
;
1927 dst
= rcu_dereference(sk
->sk_dst_cache
);
1928 if (dst
&& !atomic_inc_not_zero(&dst
->__refcnt
))
1934 static inline void dst_negative_advice(struct sock
*sk
)
1936 struct dst_entry
*ndst
, *dst
= __sk_dst_get(sk
);
1938 sk_rethink_txhash(sk
);
1940 if (dst
&& dst
->ops
->negative_advice
) {
1941 ndst
= dst
->ops
->negative_advice(dst
);
1944 rcu_assign_pointer(sk
->sk_dst_cache
, ndst
);
1945 sk_tx_queue_clear(sk
);
1946 sk
->sk_dst_pending_confirm
= 0;
1952 __sk_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
1954 struct dst_entry
*old_dst
;
1956 sk_tx_queue_clear(sk
);
1957 sk
->sk_dst_pending_confirm
= 0;
1958 old_dst
= rcu_dereference_protected(sk
->sk_dst_cache
,
1959 lockdep_sock_is_held(sk
));
1960 rcu_assign_pointer(sk
->sk_dst_cache
, dst
);
1961 dst_release(old_dst
);
1965 sk_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
1967 struct dst_entry
*old_dst
;
1969 sk_tx_queue_clear(sk
);
1970 sk
->sk_dst_pending_confirm
= 0;
1971 old_dst
= xchg((__force
struct dst_entry
**)&sk
->sk_dst_cache
, dst
);
1972 dst_release(old_dst
);
1976 __sk_dst_reset(struct sock
*sk
)
1978 __sk_dst_set(sk
, NULL
);
1982 sk_dst_reset(struct sock
*sk
)
1984 sk_dst_set(sk
, NULL
);
1987 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
);
1989 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
);
1991 static inline void sk_dst_confirm(struct sock
*sk
)
1993 if (!READ_ONCE(sk
->sk_dst_pending_confirm
))
1994 WRITE_ONCE(sk
->sk_dst_pending_confirm
, 1);
1997 static inline void sock_confirm_neigh(struct sk_buff
*skb
, struct neighbour
*n
)
1999 if (skb_get_dst_pending_confirm(skb
)) {
2000 struct sock
*sk
= skb
->sk
;
2001 unsigned long now
= jiffies
;
2003 /* avoid dirtying neighbour */
2004 if (READ_ONCE(n
->confirmed
) != now
)
2005 WRITE_ONCE(n
->confirmed
, now
);
2006 if (sk
&& READ_ONCE(sk
->sk_dst_pending_confirm
))
2007 WRITE_ONCE(sk
->sk_dst_pending_confirm
, 0);
2011 bool sk_mc_loop(struct sock
*sk
);
2013 static inline bool sk_can_gso(const struct sock
*sk
)
2015 return net_gso_ok(sk
->sk_route_caps
, sk
->sk_gso_type
);
2018 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
);
2020 static inline void sk_nocaps_add(struct sock
*sk
, netdev_features_t flags
)
2022 sk
->sk_route_nocaps
|= flags
;
2023 sk
->sk_route_caps
&= ~flags
;
2026 static inline int skb_do_copy_data_nocache(struct sock
*sk
, struct sk_buff
*skb
,
2027 struct iov_iter
*from
, char *to
,
2028 int copy
, int offset
)
2030 if (skb
->ip_summed
== CHECKSUM_NONE
) {
2032 if (!csum_and_copy_from_iter_full(to
, copy
, &csum
, from
))
2034 skb
->csum
= csum_block_add(skb
->csum
, csum
, offset
);
2035 } else if (sk
->sk_route_caps
& NETIF_F_NOCACHE_COPY
) {
2036 if (!copy_from_iter_full_nocache(to
, copy
, from
))
2038 } else if (!copy_from_iter_full(to
, copy
, from
))
2044 static inline int skb_add_data_nocache(struct sock
*sk
, struct sk_buff
*skb
,
2045 struct iov_iter
*from
, int copy
)
2047 int err
, offset
= skb
->len
;
2049 err
= skb_do_copy_data_nocache(sk
, skb
, from
, skb_put(skb
, copy
),
2052 __skb_trim(skb
, offset
);
2057 static inline int skb_copy_to_page_nocache(struct sock
*sk
, struct iov_iter
*from
,
2058 struct sk_buff
*skb
,
2064 err
= skb_do_copy_data_nocache(sk
, skb
, from
, page_address(page
) + off
,
2070 skb
->data_len
+= copy
;
2071 skb
->truesize
+= copy
;
2072 sk_wmem_queued_add(sk
, copy
);
2073 sk_mem_charge(sk
, copy
);
2078 * sk_wmem_alloc_get - returns write allocations
2081 * Return: sk_wmem_alloc minus initial offset of one
2083 static inline int sk_wmem_alloc_get(const struct sock
*sk
)
2085 return refcount_read(&sk
->sk_wmem_alloc
) - 1;
2089 * sk_rmem_alloc_get - returns read allocations
2092 * Return: sk_rmem_alloc
2094 static inline int sk_rmem_alloc_get(const struct sock
*sk
)
2096 return atomic_read(&sk
->sk_rmem_alloc
);
2100 * sk_has_allocations - check if allocations are outstanding
2103 * Return: true if socket has write or read allocations
2105 static inline bool sk_has_allocations(const struct sock
*sk
)
2107 return sk_wmem_alloc_get(sk
) || sk_rmem_alloc_get(sk
);
2111 * skwq_has_sleeper - check if there are any waiting processes
2112 * @wq: struct socket_wq
2114 * Return: true if socket_wq has waiting processes
2116 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2117 * barrier call. They were added due to the race found within the tcp code.
2119 * Consider following tcp code paths::
2122 * sys_select receive packet
2124 * __add_wait_queue update tp->rcv_nxt
2126 * tp->rcv_nxt check sock_def_readable
2128 * schedule rcu_read_lock();
2129 * wq = rcu_dereference(sk->sk_wq);
2130 * if (wq && waitqueue_active(&wq->wait))
2131 * wake_up_interruptible(&wq->wait)
2135 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2136 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2137 * could then endup calling schedule and sleep forever if there are no more
2138 * data on the socket.
2141 static inline bool skwq_has_sleeper(struct socket_wq
*wq
)
2143 return wq
&& wq_has_sleeper(&wq
->wait
);
2147 * sock_poll_wait - place memory barrier behind the poll_wait call.
2149 * @sock: socket to wait on
2152 * See the comments in the wq_has_sleeper function.
2154 static inline void sock_poll_wait(struct file
*filp
, struct socket
*sock
,
2157 if (!poll_does_not_wait(p
)) {
2158 poll_wait(filp
, &sock
->wq
.wait
, p
);
2159 /* We need to be sure we are in sync with the
2160 * socket flags modification.
2162 * This memory barrier is paired in the wq_has_sleeper.
2168 static inline void skb_set_hash_from_sk(struct sk_buff
*skb
, struct sock
*sk
)
2170 if (sk
->sk_txhash
) {
2172 skb
->hash
= sk
->sk_txhash
;
2176 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
);
2179 * Queue a received datagram if it will fit. Stream and sequenced
2180 * protocols can't normally use this as they need to fit buffers in
2181 * and play with them.
2183 * Inlined as it's very short and called for pretty much every
2184 * packet ever received.
2186 static inline void skb_set_owner_r(struct sk_buff
*skb
, struct sock
*sk
)
2190 skb
->destructor
= sock_rfree
;
2191 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
2192 sk_mem_charge(sk
, skb
->truesize
);
2195 void sk_reset_timer(struct sock
*sk
, struct timer_list
*timer
,
2196 unsigned long expires
);
2198 void sk_stop_timer(struct sock
*sk
, struct timer_list
*timer
);
2200 int __sk_queue_drop_skb(struct sock
*sk
, struct sk_buff_head
*sk_queue
,
2201 struct sk_buff
*skb
, unsigned int flags
,
2202 void (*destructor
)(struct sock
*sk
,
2203 struct sk_buff
*skb
));
2204 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
);
2205 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
);
2207 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
);
2208 struct sk_buff
*sock_dequeue_err_skb(struct sock
*sk
);
2211 * Recover an error report and clear atomically
2214 static inline int sock_error(struct sock
*sk
)
2217 if (likely(!sk
->sk_err
))
2219 err
= xchg(&sk
->sk_err
, 0);
2223 static inline unsigned long sock_wspace(struct sock
*sk
)
2227 if (!(sk
->sk_shutdown
& SEND_SHUTDOWN
)) {
2228 amt
= sk
->sk_sndbuf
- refcount_read(&sk
->sk_wmem_alloc
);
2236 * We use sk->sk_wq_raw, from contexts knowing this
2237 * pointer is not NULL and cannot disappear/change.
2239 static inline void sk_set_bit(int nr
, struct sock
*sk
)
2241 if ((nr
== SOCKWQ_ASYNC_NOSPACE
|| nr
== SOCKWQ_ASYNC_WAITDATA
) &&
2242 !sock_flag(sk
, SOCK_FASYNC
))
2245 set_bit(nr
, &sk
->sk_wq_raw
->flags
);
2248 static inline void sk_clear_bit(int nr
, struct sock
*sk
)
2250 if ((nr
== SOCKWQ_ASYNC_NOSPACE
|| nr
== SOCKWQ_ASYNC_WAITDATA
) &&
2251 !sock_flag(sk
, SOCK_FASYNC
))
2254 clear_bit(nr
, &sk
->sk_wq_raw
->flags
);
2257 static inline void sk_wake_async(const struct sock
*sk
, int how
, int band
)
2259 if (sock_flag(sk
, SOCK_FASYNC
)) {
2261 sock_wake_async(rcu_dereference(sk
->sk_wq
), how
, band
);
2266 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2267 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2268 * Note: for send buffers, TCP works better if we can build two skbs at
2271 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2273 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2274 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2276 static inline void sk_stream_moderate_sndbuf(struct sock
*sk
)
2280 if (sk
->sk_userlocks
& SOCK_SNDBUF_LOCK
)
2283 val
= min(sk
->sk_sndbuf
, sk
->sk_wmem_queued
>> 1);
2285 WRITE_ONCE(sk
->sk_sndbuf
, max_t(u32
, val
, SOCK_MIN_SNDBUF
));
2288 struct sk_buff
*sk_stream_alloc_skb(struct sock
*sk
, int size
, gfp_t gfp
,
2289 bool force_schedule
);
2292 * sk_page_frag - return an appropriate page_frag
2295 * Use the per task page_frag instead of the per socket one for
2296 * optimization when we know that we're in the normal context and owns
2297 * everything that's associated with %current.
2299 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2300 * inside other socket operations and end up recursing into sk_page_frag()
2301 * while it's already in use.
2303 * Return: a per task page_frag if context allows that,
2304 * otherwise a per socket one.
2306 static inline struct page_frag
*sk_page_frag(struct sock
*sk
)
2308 if (gfpflags_normal_context(sk
->sk_allocation
))
2309 return ¤t
->task_frag
;
2311 return &sk
->sk_frag
;
2314 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
);
2317 * Default write policy as shown to user space via poll/select/SIGIO
2319 static inline bool sock_writeable(const struct sock
*sk
)
2321 return refcount_read(&sk
->sk_wmem_alloc
) < (READ_ONCE(sk
->sk_sndbuf
) >> 1);
2324 static inline gfp_t
gfp_any(void)
2326 return in_softirq() ? GFP_ATOMIC
: GFP_KERNEL
;
2329 static inline long sock_rcvtimeo(const struct sock
*sk
, bool noblock
)
2331 return noblock
? 0 : sk
->sk_rcvtimeo
;
2334 static inline long sock_sndtimeo(const struct sock
*sk
, bool noblock
)
2336 return noblock
? 0 : sk
->sk_sndtimeo
;
2339 static inline int sock_rcvlowat(const struct sock
*sk
, int waitall
, int len
)
2341 int v
= waitall
? len
: min_t(int, READ_ONCE(sk
->sk_rcvlowat
), len
);
2346 /* Alas, with timeout socket operations are not restartable.
2347 * Compare this to poll().
2349 static inline int sock_intr_errno(long timeo
)
2351 return timeo
== MAX_SCHEDULE_TIMEOUT
? -ERESTARTSYS
: -EINTR
;
2354 struct sock_skb_cb
{
2358 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2359 * using skb->cb[] would keep using it directly and utilize its
2360 * alignement guarantee.
2362 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2363 sizeof(struct sock_skb_cb)))
2365 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2366 SOCK_SKB_CB_OFFSET))
2368 #define sock_skb_cb_check_size(size) \
2369 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2372 sock_skb_set_dropcount(const struct sock
*sk
, struct sk_buff
*skb
)
2374 SOCK_SKB_CB(skb
)->dropcount
= sock_flag(sk
, SOCK_RXQ_OVFL
) ?
2375 atomic_read(&sk
->sk_drops
) : 0;
2378 static inline void sk_drops_add(struct sock
*sk
, const struct sk_buff
*skb
)
2380 int segs
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
2382 atomic_add(segs
, &sk
->sk_drops
);
2385 static inline ktime_t
sock_read_timestamp(struct sock
*sk
)
2387 #if BITS_PER_LONG==32
2392 seq
= read_seqbegin(&sk
->sk_stamp_seq
);
2394 } while (read_seqretry(&sk
->sk_stamp_seq
, seq
));
2398 return READ_ONCE(sk
->sk_stamp
);
2402 static inline void sock_write_timestamp(struct sock
*sk
, ktime_t kt
)
2404 #if BITS_PER_LONG==32
2405 write_seqlock(&sk
->sk_stamp_seq
);
2407 write_sequnlock(&sk
->sk_stamp_seq
);
2409 WRITE_ONCE(sk
->sk_stamp
, kt
);
2413 void __sock_recv_timestamp(struct msghdr
*msg
, struct sock
*sk
,
2414 struct sk_buff
*skb
);
2415 void __sock_recv_wifi_status(struct msghdr
*msg
, struct sock
*sk
,
2416 struct sk_buff
*skb
);
2419 sock_recv_timestamp(struct msghdr
*msg
, struct sock
*sk
, struct sk_buff
*skb
)
2421 ktime_t kt
= skb
->tstamp
;
2422 struct skb_shared_hwtstamps
*hwtstamps
= skb_hwtstamps(skb
);
2425 * generate control messages if
2426 * - receive time stamping in software requested
2427 * - software time stamp available and wanted
2428 * - hardware time stamps available and wanted
2430 if (sock_flag(sk
, SOCK_RCVTSTAMP
) ||
2431 (sk
->sk_tsflags
& SOF_TIMESTAMPING_RX_SOFTWARE
) ||
2432 (kt
&& sk
->sk_tsflags
& SOF_TIMESTAMPING_SOFTWARE
) ||
2433 (hwtstamps
->hwtstamp
&&
2434 (sk
->sk_tsflags
& SOF_TIMESTAMPING_RAW_HARDWARE
)))
2435 __sock_recv_timestamp(msg
, sk
, skb
);
2437 sock_write_timestamp(sk
, kt
);
2439 if (sock_flag(sk
, SOCK_WIFI_STATUS
) && skb
->wifi_acked_valid
)
2440 __sock_recv_wifi_status(msg
, sk
, skb
);
2443 void __sock_recv_ts_and_drops(struct msghdr
*msg
, struct sock
*sk
,
2444 struct sk_buff
*skb
);
2446 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2447 static inline void sock_recv_ts_and_drops(struct msghdr
*msg
, struct sock
*sk
,
2448 struct sk_buff
*skb
)
2450 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2451 (1UL << SOCK_RCVTSTAMP))
2452 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2453 SOF_TIMESTAMPING_RAW_HARDWARE)
2455 if (sk
->sk_flags
& FLAGS_TS_OR_DROPS
|| sk
->sk_tsflags
& TSFLAGS_ANY
)
2456 __sock_recv_ts_and_drops(msg
, sk
, skb
);
2457 else if (unlikely(sock_flag(sk
, SOCK_TIMESTAMP
)))
2458 sock_write_timestamp(sk
, skb
->tstamp
);
2459 else if (unlikely(sk
->sk_stamp
== SK_DEFAULT_STAMP
))
2460 sock_write_timestamp(sk
, 0);
2463 void __sock_tx_timestamp(__u16 tsflags
, __u8
*tx_flags
);
2466 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2467 * @sk: socket sending this packet
2468 * @tsflags: timestamping flags to use
2469 * @tx_flags: completed with instructions for time stamping
2470 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2472 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2474 static inline void _sock_tx_timestamp(struct sock
*sk
, __u16 tsflags
,
2475 __u8
*tx_flags
, __u32
*tskey
)
2477 if (unlikely(tsflags
)) {
2478 __sock_tx_timestamp(tsflags
, tx_flags
);
2479 if (tsflags
& SOF_TIMESTAMPING_OPT_ID
&& tskey
&&
2480 tsflags
& SOF_TIMESTAMPING_TX_RECORD_MASK
)
2481 *tskey
= sk
->sk_tskey
++;
2483 if (unlikely(sock_flag(sk
, SOCK_WIFI_STATUS
)))
2484 *tx_flags
|= SKBTX_WIFI_STATUS
;
2487 static inline void sock_tx_timestamp(struct sock
*sk
, __u16 tsflags
,
2490 _sock_tx_timestamp(sk
, tsflags
, tx_flags
, NULL
);
2493 static inline void skb_setup_tx_timestamp(struct sk_buff
*skb
, __u16 tsflags
)
2495 _sock_tx_timestamp(skb
->sk
, tsflags
, &skb_shinfo(skb
)->tx_flags
,
2496 &skb_shinfo(skb
)->tskey
);
2499 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key
);
2501 * sk_eat_skb - Release a skb if it is no longer needed
2502 * @sk: socket to eat this skb from
2503 * @skb: socket buffer to eat
2505 * This routine must be called with interrupts disabled or with the socket
2506 * locked so that the sk_buff queue operation is ok.
2508 static inline void sk_eat_skb(struct sock
*sk
, struct sk_buff
*skb
)
2510 __skb_unlink(skb
, &sk
->sk_receive_queue
);
2511 if (static_branch_unlikely(&tcp_rx_skb_cache_key
) &&
2512 !sk
->sk_rx_skb_cache
) {
2513 sk
->sk_rx_skb_cache
= skb
;
2521 struct net
*sock_net(const struct sock
*sk
)
2523 return read_pnet(&sk
->sk_net
);
2527 void sock_net_set(struct sock
*sk
, struct net
*net
)
2529 write_pnet(&sk
->sk_net
, net
);
2533 skb_sk_is_prefetched(struct sk_buff
*skb
)
2536 return skb
->destructor
== sock_pfree
;
2539 #endif /* CONFIG_INET */
2542 /* This helper checks if a socket is a full socket,
2543 * ie _not_ a timewait or request socket.
2545 static inline bool sk_fullsock(const struct sock
*sk
)
2547 return (1 << sk
->sk_state
) & ~(TCPF_TIME_WAIT
| TCPF_NEW_SYN_RECV
);
2551 sk_is_refcounted(struct sock
*sk
)
2553 /* Only full sockets have sk->sk_flags. */
2554 return !sk_fullsock(sk
) || !sock_flag(sk
, SOCK_RCU_FREE
);
2558 * skb_steal_sock - steal a socket from an sk_buff
2559 * @skb: sk_buff to steal the socket from
2560 * @refcounted: is set to true if the socket is reference-counted
2562 static inline struct sock
*
2563 skb_steal_sock(struct sk_buff
*skb
, bool *refcounted
)
2566 struct sock
*sk
= skb
->sk
;
2569 if (skb_sk_is_prefetched(skb
))
2570 *refcounted
= sk_is_refcounted(sk
);
2571 skb
->destructor
= NULL
;
2575 *refcounted
= false;
2579 /* Checks if this SKB belongs to an HW offloaded socket
2580 * and whether any SW fallbacks are required based on dev.
2581 * Check decrypted mark in case skb_orphan() cleared socket.
2583 static inline struct sk_buff
*sk_validate_xmit_skb(struct sk_buff
*skb
,
2584 struct net_device
*dev
)
2586 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2587 struct sock
*sk
= skb
->sk
;
2589 if (sk
&& sk_fullsock(sk
) && sk
->sk_validate_xmit_skb
) {
2590 skb
= sk
->sk_validate_xmit_skb(sk
, dev
, skb
);
2591 #ifdef CONFIG_TLS_DEVICE
2592 } else if (unlikely(skb
->decrypted
)) {
2593 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2603 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2604 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2606 static inline bool sk_listener(const struct sock
*sk
)
2608 return (1 << sk
->sk_state
) & (TCPF_LISTEN
| TCPF_NEW_SYN_RECV
);
2611 void sock_enable_timestamp(struct sock
*sk
, enum sock_flags flag
);
2612 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
, int level
,
2615 bool sk_ns_capable(const struct sock
*sk
,
2616 struct user_namespace
*user_ns
, int cap
);
2617 bool sk_capable(const struct sock
*sk
, int cap
);
2618 bool sk_net_capable(const struct sock
*sk
, int cap
);
2620 void sk_get_meminfo(const struct sock
*sk
, u32
*meminfo
);
2622 /* Take into consideration the size of the struct sk_buff overhead in the
2623 * determination of these values, since that is non-constant across
2624 * platforms. This makes socket queueing behavior and performance
2625 * not depend upon such differences.
2627 #define _SK_MEM_PACKETS 256
2628 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2629 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2630 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2632 extern __u32 sysctl_wmem_max
;
2633 extern __u32 sysctl_rmem_max
;
2635 extern int sysctl_tstamp_allow_data
;
2636 extern int sysctl_optmem_max
;
2638 extern __u32 sysctl_wmem_default
;
2639 extern __u32 sysctl_rmem_default
;
2641 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key
);
2643 static inline int sk_get_wmem0(const struct sock
*sk
, const struct proto
*proto
)
2645 /* Does this proto have per netns sysctl_wmem ? */
2646 if (proto
->sysctl_wmem_offset
)
2647 return *(int *)((void *)sock_net(sk
) + proto
->sysctl_wmem_offset
);
2649 return *proto
->sysctl_wmem
;
2652 static inline int sk_get_rmem0(const struct sock
*sk
, const struct proto
*proto
)
2654 /* Does this proto have per netns sysctl_rmem ? */
2655 if (proto
->sysctl_rmem_offset
)
2656 return *(int *)((void *)sock_net(sk
) + proto
->sysctl_rmem_offset
);
2658 return *proto
->sysctl_rmem
;
2661 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2662 * Some wifi drivers need to tweak it to get more chunks.
2663 * They can use this helper from their ndo_start_xmit()
2665 static inline void sk_pacing_shift_update(struct sock
*sk
, int val
)
2667 if (!sk
|| !sk_fullsock(sk
) || READ_ONCE(sk
->sk_pacing_shift
) == val
)
2669 WRITE_ONCE(sk
->sk_pacing_shift
, val
);
2672 /* if a socket is bound to a device, check that the given device
2673 * index is either the same or that the socket is bound to an L3
2674 * master device and the given device index is also enslaved to
2677 static inline bool sk_dev_equal_l3scope(struct sock
*sk
, int dif
)
2681 if (!sk
->sk_bound_dev_if
|| sk
->sk_bound_dev_if
== dif
)
2684 mdif
= l3mdev_master_ifindex_by_index(sock_net(sk
), dif
);
2685 if (mdif
&& mdif
== sk
->sk_bound_dev_if
)
2691 void sock_def_readable(struct sock
*sk
);
2693 int sock_bindtoindex(struct sock
*sk
, int ifindex
, bool lock_sk
);
2694 void sock_enable_timestamps(struct sock
*sk
);
2695 void sock_no_linger(struct sock
*sk
);
2696 void sock_set_keepalive(struct sock
*sk
);
2697 void sock_set_priority(struct sock
*sk
, u32 priority
);
2698 void sock_set_rcvbuf(struct sock
*sk
, int val
);
2699 void sock_set_mark(struct sock
*sk
, u32 val
);
2700 void sock_set_reuseaddr(struct sock
*sk
);
2701 void sock_set_reuseport(struct sock
*sk
);
2702 void sock_set_sndtimeo(struct sock
*sk
, s64 secs
);
2704 int sock_bind_add(struct sock
*sk
, struct sockaddr
*addr
, int addr_len
);
2706 #endif /* _SOCK_H */