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>
63 #include <linux/atomic.h>
64 #include <linux/refcount.h>
66 #include <net/checksum.h>
67 #include <net/tcp_states.h>
68 #include <linux/net_tstamp.h>
69 #include <net/l3mdev.h>
72 * This structure really needs to be cleaned up.
73 * Most of it is for TCP, and not used by any of
74 * the other protocols.
77 /* Define this to get the SOCK_DBG debugging facility. */
78 #define SOCK_DEBUGGING
80 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
81 printk(KERN_DEBUG msg); } while (0)
83 /* Validate arguments and do nothing */
84 static inline __printf(2, 3)
85 void SOCK_DEBUG(const struct sock
*sk
, const char *msg
, ...)
90 /* This is the per-socket lock. The spinlock provides a synchronization
91 * between user contexts and software interrupt processing, whereas the
92 * mini-semaphore synchronizes multiple users amongst themselves.
99 * We express the mutex-alike socket_lock semantics
100 * to the lock validator by explicitly managing
101 * the slock as a lock variant (in addition to
104 #ifdef CONFIG_DEBUG_LOCK_ALLOC
105 struct lockdep_map dep_map
;
113 typedef __u32 __bitwise __portpair
;
114 typedef __u64 __bitwise __addrpair
;
117 * struct sock_common - minimal network layer representation of sockets
118 * @skc_daddr: Foreign IPv4 addr
119 * @skc_rcv_saddr: Bound local IPv4 addr
120 * @skc_hash: hash value used with various protocol lookup tables
121 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
122 * @skc_dport: placeholder for inet_dport/tw_dport
123 * @skc_num: placeholder for inet_num/tw_num
124 * @skc_family: network address family
125 * @skc_state: Connection state
126 * @skc_reuse: %SO_REUSEADDR setting
127 * @skc_reuseport: %SO_REUSEPORT setting
128 * @skc_bound_dev_if: bound device index if != 0
129 * @skc_bind_node: bind hash linkage for various protocol lookup tables
130 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
131 * @skc_prot: protocol handlers inside a network family
132 * @skc_net: reference to the network namespace of this socket
133 * @skc_node: main hash linkage for various protocol lookup tables
134 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
135 * @skc_tx_queue_mapping: tx queue number for this connection
136 * @skc_rx_queue_mapping: rx queue number for this connection
137 * @skc_flags: place holder for sk_flags
138 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
139 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
140 * @skc_incoming_cpu: record/match cpu processing incoming packets
141 * @skc_refcnt: reference count
143 * This is the minimal network layer representation of sockets, the header
144 * for struct sock and struct inet_timewait_sock.
147 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
148 * address on 64bit arches : cf INET_MATCH()
151 __addrpair skc_addrpair
;
154 __be32 skc_rcv_saddr
;
158 unsigned int skc_hash
;
159 __u16 skc_u16hashes
[2];
161 /* skc_dport && skc_num must be grouped as well */
163 __portpair skc_portpair
;
170 unsigned short skc_family
;
171 volatile unsigned char skc_state
;
172 unsigned char skc_reuse
:4;
173 unsigned char skc_reuseport
:1;
174 unsigned char skc_ipv6only
:1;
175 unsigned char skc_net_refcnt
:1;
176 int skc_bound_dev_if
;
178 struct hlist_node skc_bind_node
;
179 struct hlist_node skc_portaddr_node
;
181 struct proto
*skc_prot
;
182 possible_net_t skc_net
;
184 #if IS_ENABLED(CONFIG_IPV6)
185 struct in6_addr skc_v6_daddr
;
186 struct in6_addr skc_v6_rcv_saddr
;
189 atomic64_t skc_cookie
;
191 /* following fields are padding to force
192 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
193 * assuming IPV6 is enabled. We use this padding differently
194 * for different kind of 'sockets'
197 unsigned long skc_flags
;
198 struct sock
*skc_listener
; /* request_sock */
199 struct inet_timewait_death_row
*skc_tw_dr
; /* inet_timewait_sock */
202 * fields between dontcopy_begin/dontcopy_end
203 * are not copied in sock_copy()
206 int skc_dontcopy_begin
[0];
209 struct hlist_node skc_node
;
210 struct hlist_nulls_node skc_nulls_node
;
212 unsigned short skc_tx_queue_mapping
;
214 unsigned short skc_rx_queue_mapping
;
217 int skc_incoming_cpu
;
219 u32 skc_tw_rcv_nxt
; /* struct tcp_timewait_sock */
222 refcount_t skc_refcnt
;
224 int skc_dontcopy_end
[0];
227 u32 skc_window_clamp
;
228 u32 skc_tw_snd_nxt
; /* struct tcp_timewait_sock */
233 struct bpf_sk_storage
;
236 * struct sock - network layer representation of sockets
237 * @__sk_common: shared layout with inet_timewait_sock
238 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
239 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
240 * @sk_lock: synchronizer
241 * @sk_kern_sock: True if sock is using kernel lock classes
242 * @sk_rcvbuf: size of receive buffer in bytes
243 * @sk_wq: sock wait queue and async head
244 * @sk_rx_dst: receive input route used by early demux
245 * @sk_dst_cache: destination cache
246 * @sk_dst_pending_confirm: need to confirm neighbour
247 * @sk_policy: flow policy
248 * @sk_receive_queue: incoming packets
249 * @sk_wmem_alloc: transmit queue bytes committed
250 * @sk_tsq_flags: TCP Small Queues flags
251 * @sk_write_queue: Packet sending queue
252 * @sk_omem_alloc: "o" is "option" or "other"
253 * @sk_wmem_queued: persistent queue size
254 * @sk_forward_alloc: space allocated forward
255 * @sk_napi_id: id of the last napi context to receive data for sk
256 * @sk_ll_usec: usecs to busypoll when there is no data
257 * @sk_allocation: allocation mode
258 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
259 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
260 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
261 * @sk_sndbuf: size of send buffer in bytes
262 * @__sk_flags_offset: empty field used to determine location of bitfield
263 * @sk_padding: unused element for alignment
264 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
265 * @sk_no_check_rx: allow zero checksum in RX packets
266 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
267 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
268 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
269 * @sk_gso_max_size: Maximum GSO segment size to build
270 * @sk_gso_max_segs: Maximum number of GSO segments
271 * @sk_pacing_shift: scaling factor for TCP Small Queues
272 * @sk_lingertime: %SO_LINGER l_linger setting
273 * @sk_backlog: always used with the per-socket spinlock held
274 * @sk_callback_lock: used with the callbacks in the end of this struct
275 * @sk_error_queue: rarely used
276 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
277 * IPV6_ADDRFORM for instance)
278 * @sk_err: last error
279 * @sk_err_soft: errors that don't cause failure but are the cause of a
280 * persistent failure not just 'timed out'
281 * @sk_drops: raw/udp drops counter
282 * @sk_ack_backlog: current listen backlog
283 * @sk_max_ack_backlog: listen backlog set in listen()
284 * @sk_uid: user id of owner
285 * @sk_priority: %SO_PRIORITY setting
286 * @sk_type: socket type (%SOCK_STREAM, etc)
287 * @sk_protocol: which protocol this socket belongs in this network family
288 * @sk_peer_pid: &struct pid for this socket's peer
289 * @sk_peer_cred: %SO_PEERCRED setting
290 * @sk_rcvlowat: %SO_RCVLOWAT setting
291 * @sk_rcvtimeo: %SO_RCVTIMEO setting
292 * @sk_sndtimeo: %SO_SNDTIMEO setting
293 * @sk_txhash: computed flow hash for use on transmit
294 * @sk_filter: socket filtering instructions
295 * @sk_timer: sock cleanup timer
296 * @sk_stamp: time stamp of last packet received
297 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
298 * @sk_tsflags: SO_TIMESTAMPING socket options
299 * @sk_tskey: counter to disambiguate concurrent tstamp requests
300 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
301 * @sk_socket: Identd and reporting IO signals
302 * @sk_user_data: RPC layer private data
303 * @sk_frag: cached page frag
304 * @sk_peek_off: current peek_offset value
305 * @sk_send_head: front of stuff to transmit
306 * @sk_security: used by security modules
307 * @sk_mark: generic packet mark
308 * @sk_cgrp_data: cgroup data for this cgroup
309 * @sk_memcg: this socket's memory cgroup association
310 * @sk_write_pending: a write to stream socket waits to start
311 * @sk_state_change: callback to indicate change in the state of the sock
312 * @sk_data_ready: callback to indicate there is data to be processed
313 * @sk_write_space: callback to indicate there is bf sending space available
314 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
315 * @sk_backlog_rcv: callback to process the backlog
316 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
317 * @sk_reuseport_cb: reuseport group container
318 * @sk_rcu: used during RCU grace period
319 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
320 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
321 * @sk_txtime_unused: unused txtime flags
325 * Now struct inet_timewait_sock also uses sock_common, so please just
326 * don't add nothing before this first member (__sk_common) --acme
328 struct sock_common __sk_common
;
329 #define sk_node __sk_common.skc_node
330 #define sk_nulls_node __sk_common.skc_nulls_node
331 #define sk_refcnt __sk_common.skc_refcnt
332 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
334 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
337 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
338 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
339 #define sk_hash __sk_common.skc_hash
340 #define sk_portpair __sk_common.skc_portpair
341 #define sk_num __sk_common.skc_num
342 #define sk_dport __sk_common.skc_dport
343 #define sk_addrpair __sk_common.skc_addrpair
344 #define sk_daddr __sk_common.skc_daddr
345 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
346 #define sk_family __sk_common.skc_family
347 #define sk_state __sk_common.skc_state
348 #define sk_reuse __sk_common.skc_reuse
349 #define sk_reuseport __sk_common.skc_reuseport
350 #define sk_ipv6only __sk_common.skc_ipv6only
351 #define sk_net_refcnt __sk_common.skc_net_refcnt
352 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
353 #define sk_bind_node __sk_common.skc_bind_node
354 #define sk_prot __sk_common.skc_prot
355 #define sk_net __sk_common.skc_net
356 #define sk_v6_daddr __sk_common.skc_v6_daddr
357 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
358 #define sk_cookie __sk_common.skc_cookie
359 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
360 #define sk_flags __sk_common.skc_flags
361 #define sk_rxhash __sk_common.skc_rxhash
363 socket_lock_t sk_lock
;
366 struct sk_buff_head sk_error_queue
;
367 struct sk_buff
*sk_rx_skb_cache
;
368 struct sk_buff_head sk_receive_queue
;
370 * The backlog queue is special, it is always used with
371 * the per-socket spinlock held and requires low latency
372 * access. Therefore we special case it's implementation.
373 * Note : rmem_alloc is in this structure to fill a hole
374 * on 64bit arches, not because its logically part of
380 struct sk_buff
*head
;
381 struct sk_buff
*tail
;
383 #define sk_rmem_alloc sk_backlog.rmem_alloc
385 int sk_forward_alloc
;
386 #ifdef CONFIG_NET_RX_BUSY_POLL
387 unsigned int sk_ll_usec
;
388 /* ===== mostly read cache line ===== */
389 unsigned int sk_napi_id
;
393 struct sk_filter __rcu
*sk_filter
;
395 struct socket_wq __rcu
*sk_wq
;
396 struct socket_wq
*sk_wq_raw
;
399 struct xfrm_policy __rcu
*sk_policy
[2];
401 struct dst_entry
*sk_rx_dst
;
402 struct dst_entry __rcu
*sk_dst_cache
;
403 atomic_t sk_omem_alloc
;
406 /* ===== cache line for TX ===== */
408 refcount_t sk_wmem_alloc
;
409 unsigned long sk_tsq_flags
;
411 struct sk_buff
*sk_send_head
;
412 struct rb_root tcp_rtx_queue
;
414 struct sk_buff
*sk_tx_skb_cache
;
415 struct sk_buff_head sk_write_queue
;
417 int sk_write_pending
;
418 __u32 sk_dst_pending_confirm
;
419 u32 sk_pacing_status
; /* see enum sk_pacing */
421 struct timer_list sk_timer
;
424 unsigned long sk_pacing_rate
; /* bytes per second */
425 unsigned long sk_max_pacing_rate
;
426 struct page_frag sk_frag
;
427 netdev_features_t sk_route_caps
;
428 netdev_features_t sk_route_nocaps
;
429 netdev_features_t sk_route_forced_caps
;
431 unsigned int sk_gso_max_size
;
436 * Because of non atomicity rules, all
437 * changes are protected by socket lock.
439 unsigned int __sk_flags_offset
[0];
440 #ifdef __BIG_ENDIAN_BITFIELD
441 #define SK_FL_PROTO_SHIFT 16
442 #define SK_FL_PROTO_MASK 0x00ff0000
444 #define SK_FL_TYPE_SHIFT 0
445 #define SK_FL_TYPE_MASK 0x0000ffff
447 #define SK_FL_PROTO_SHIFT 8
448 #define SK_FL_PROTO_MASK 0x0000ff00
450 #define SK_FL_TYPE_SHIFT 16
451 #define SK_FL_TYPE_MASK 0xffff0000
454 unsigned int sk_padding
: 1,
461 #define SK_PROTOCOL_MAX U8_MAX
464 unsigned long sk_lingertime
;
465 struct proto
*sk_prot_creator
;
466 rwlock_t sk_callback_lock
;
470 u32 sk_max_ack_backlog
;
472 struct pid
*sk_peer_pid
;
473 const struct cred
*sk_peer_cred
;
476 #if BITS_PER_LONG==32
477 seqlock_t sk_stamp_seq
;
485 u8 sk_txtime_deadline_mode
: 1,
486 sk_txtime_report_errors
: 1,
487 sk_txtime_unused
: 6;
489 struct socket
*sk_socket
;
491 #ifdef CONFIG_SECURITY
494 struct sock_cgroup_data sk_cgrp_data
;
495 struct mem_cgroup
*sk_memcg
;
496 void (*sk_state_change
)(struct sock
*sk
);
497 void (*sk_data_ready
)(struct sock
*sk
);
498 void (*sk_write_space
)(struct sock
*sk
);
499 void (*sk_error_report
)(struct sock
*sk
);
500 int (*sk_backlog_rcv
)(struct sock
*sk
,
501 struct sk_buff
*skb
);
502 #ifdef CONFIG_SOCK_VALIDATE_XMIT
503 struct sk_buff
* (*sk_validate_xmit_skb
)(struct sock
*sk
,
504 struct net_device
*dev
,
505 struct sk_buff
*skb
);
507 void (*sk_destruct
)(struct sock
*sk
);
508 struct sock_reuseport __rcu
*sk_reuseport_cb
;
509 #ifdef CONFIG_BPF_SYSCALL
510 struct bpf_sk_storage __rcu
*sk_bpf_storage
;
512 struct rcu_head sk_rcu
;
517 SK_PACING_NEEDED
= 1,
521 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
523 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
524 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
527 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
528 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
529 * on a socket means that the socket will reuse everybody else's port
530 * without looking at the other's sk_reuse value.
533 #define SK_NO_REUSE 0
534 #define SK_CAN_REUSE 1
535 #define SK_FORCE_REUSE 2
537 int sk_set_peek_off(struct sock
*sk
, int val
);
539 static inline int sk_peek_offset(struct sock
*sk
, int flags
)
541 if (unlikely(flags
& MSG_PEEK
)) {
542 return READ_ONCE(sk
->sk_peek_off
);
548 static inline void sk_peek_offset_bwd(struct sock
*sk
, int val
)
550 s32 off
= READ_ONCE(sk
->sk_peek_off
);
552 if (unlikely(off
>= 0)) {
553 off
= max_t(s32
, off
- val
, 0);
554 WRITE_ONCE(sk
->sk_peek_off
, off
);
558 static inline void sk_peek_offset_fwd(struct sock
*sk
, int val
)
560 sk_peek_offset_bwd(sk
, -val
);
564 * Hashed lists helper routines
566 static inline struct sock
*sk_entry(const struct hlist_node
*node
)
568 return hlist_entry(node
, struct sock
, sk_node
);
571 static inline struct sock
*__sk_head(const struct hlist_head
*head
)
573 return hlist_entry(head
->first
, struct sock
, sk_node
);
576 static inline struct sock
*sk_head(const struct hlist_head
*head
)
578 return hlist_empty(head
) ? NULL
: __sk_head(head
);
581 static inline struct sock
*__sk_nulls_head(const struct hlist_nulls_head
*head
)
583 return hlist_nulls_entry(head
->first
, struct sock
, sk_nulls_node
);
586 static inline struct sock
*sk_nulls_head(const struct hlist_nulls_head
*head
)
588 return hlist_nulls_empty(head
) ? NULL
: __sk_nulls_head(head
);
591 static inline struct sock
*sk_next(const struct sock
*sk
)
593 return hlist_entry_safe(sk
->sk_node
.next
, struct sock
, sk_node
);
596 static inline struct sock
*sk_nulls_next(const struct sock
*sk
)
598 return (!is_a_nulls(sk
->sk_nulls_node
.next
)) ?
599 hlist_nulls_entry(sk
->sk_nulls_node
.next
,
600 struct sock
, sk_nulls_node
) :
604 static inline bool sk_unhashed(const struct sock
*sk
)
606 return hlist_unhashed(&sk
->sk_node
);
609 static inline bool sk_hashed(const struct sock
*sk
)
611 return !sk_unhashed(sk
);
614 static inline void sk_node_init(struct hlist_node
*node
)
619 static inline void sk_nulls_node_init(struct hlist_nulls_node
*node
)
624 static inline void __sk_del_node(struct sock
*sk
)
626 __hlist_del(&sk
->sk_node
);
629 /* NB: equivalent to hlist_del_init_rcu */
630 static inline bool __sk_del_node_init(struct sock
*sk
)
634 sk_node_init(&sk
->sk_node
);
640 /* Grab socket reference count. This operation is valid only
641 when sk is ALREADY grabbed f.e. it is found in hash table
642 or a list and the lookup is made under lock preventing hash table
646 static __always_inline
void sock_hold(struct sock
*sk
)
648 refcount_inc(&sk
->sk_refcnt
);
651 /* Ungrab socket in the context, which assumes that socket refcnt
652 cannot hit zero, f.e. it is true in context of any socketcall.
654 static __always_inline
void __sock_put(struct sock
*sk
)
656 refcount_dec(&sk
->sk_refcnt
);
659 static inline bool sk_del_node_init(struct sock
*sk
)
661 bool rc
= __sk_del_node_init(sk
);
664 /* paranoid for a while -acme */
665 WARN_ON(refcount_read(&sk
->sk_refcnt
) == 1);
670 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
672 static inline bool __sk_nulls_del_node_init_rcu(struct sock
*sk
)
675 hlist_nulls_del_init_rcu(&sk
->sk_nulls_node
);
681 static inline bool sk_nulls_del_node_init_rcu(struct sock
*sk
)
683 bool rc
= __sk_nulls_del_node_init_rcu(sk
);
686 /* paranoid for a while -acme */
687 WARN_ON(refcount_read(&sk
->sk_refcnt
) == 1);
693 static inline void __sk_add_node(struct sock
*sk
, struct hlist_head
*list
)
695 hlist_add_head(&sk
->sk_node
, list
);
698 static inline void sk_add_node(struct sock
*sk
, struct hlist_head
*list
)
701 __sk_add_node(sk
, list
);
704 static inline void sk_add_node_rcu(struct sock
*sk
, struct hlist_head
*list
)
707 if (IS_ENABLED(CONFIG_IPV6
) && sk
->sk_reuseport
&&
708 sk
->sk_family
== AF_INET6
)
709 hlist_add_tail_rcu(&sk
->sk_node
, list
);
711 hlist_add_head_rcu(&sk
->sk_node
, list
);
714 static inline void sk_add_node_tail_rcu(struct sock
*sk
, struct hlist_head
*list
)
717 hlist_add_tail_rcu(&sk
->sk_node
, list
);
720 static inline void __sk_nulls_add_node_rcu(struct sock
*sk
, struct hlist_nulls_head
*list
)
722 hlist_nulls_add_head_rcu(&sk
->sk_nulls_node
, list
);
725 static inline void sk_nulls_add_node_rcu(struct sock
*sk
, struct hlist_nulls_head
*list
)
728 __sk_nulls_add_node_rcu(sk
, list
);
731 static inline void __sk_del_bind_node(struct sock
*sk
)
733 __hlist_del(&sk
->sk_bind_node
);
736 static inline void sk_add_bind_node(struct sock
*sk
,
737 struct hlist_head
*list
)
739 hlist_add_head(&sk
->sk_bind_node
, list
);
742 #define sk_for_each(__sk, list) \
743 hlist_for_each_entry(__sk, list, sk_node)
744 #define sk_for_each_rcu(__sk, list) \
745 hlist_for_each_entry_rcu(__sk, list, sk_node)
746 #define sk_nulls_for_each(__sk, node, list) \
747 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
748 #define sk_nulls_for_each_rcu(__sk, node, list) \
749 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
750 #define sk_for_each_from(__sk) \
751 hlist_for_each_entry_from(__sk, sk_node)
752 #define sk_nulls_for_each_from(__sk, node) \
753 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
754 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
755 #define sk_for_each_safe(__sk, tmp, list) \
756 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
757 #define sk_for_each_bound(__sk, list) \
758 hlist_for_each_entry(__sk, list, sk_bind_node)
761 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
762 * @tpos: the type * to use as a loop cursor.
763 * @pos: the &struct hlist_node to use as a loop cursor.
764 * @head: the head for your list.
765 * @offset: offset of hlist_node within the struct.
768 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
769 for (pos = rcu_dereference(hlist_first_rcu(head)); \
771 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
772 pos = rcu_dereference(hlist_next_rcu(pos)))
774 static inline struct user_namespace
*sk_user_ns(struct sock
*sk
)
776 /* Careful only use this in a context where these parameters
777 * can not change and must all be valid, such as recvmsg from
780 return sk
->sk_socket
->file
->f_cred
->user_ns
;
794 SOCK_USE_WRITE_QUEUE
, /* whether to call sk->sk_write_space in sock_wfree */
795 SOCK_DBG
, /* %SO_DEBUG setting */
796 SOCK_RCVTSTAMP
, /* %SO_TIMESTAMP setting */
797 SOCK_RCVTSTAMPNS
, /* %SO_TIMESTAMPNS setting */
798 SOCK_LOCALROUTE
, /* route locally only, %SO_DONTROUTE setting */
799 SOCK_QUEUE_SHRUNK
, /* write queue has been shrunk recently */
800 SOCK_MEMALLOC
, /* VM depends on this socket for swapping */
801 SOCK_TIMESTAMPING_RX_SOFTWARE
, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
802 SOCK_FASYNC
, /* fasync() active */
804 SOCK_ZEROCOPY
, /* buffers from userspace */
805 SOCK_WIFI_STATUS
, /* push wifi status to userspace */
806 SOCK_NOFCS
, /* Tell NIC not to do the Ethernet FCS.
807 * Will use last 4 bytes of packet sent from
808 * user-space instead.
810 SOCK_FILTER_LOCKED
, /* Filter cannot be changed anymore */
811 SOCK_SELECT_ERR_QUEUE
, /* Wake select on error queue */
812 SOCK_RCU_FREE
, /* wait rcu grace period in sk_destruct() */
814 SOCK_XDP
, /* XDP is attached */
815 SOCK_TSTAMP_NEW
, /* Indicates 64 bit timestamps always */
818 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
820 static inline void sock_copy_flags(struct sock
*nsk
, struct sock
*osk
)
822 nsk
->sk_flags
= osk
->sk_flags
;
825 static inline void sock_set_flag(struct sock
*sk
, enum sock_flags flag
)
827 __set_bit(flag
, &sk
->sk_flags
);
830 static inline void sock_reset_flag(struct sock
*sk
, enum sock_flags flag
)
832 __clear_bit(flag
, &sk
->sk_flags
);
835 static inline bool sock_flag(const struct sock
*sk
, enum sock_flags flag
)
837 return test_bit(flag
, &sk
->sk_flags
);
841 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key
);
842 static inline int sk_memalloc_socks(void)
844 return static_branch_unlikely(&memalloc_socks_key
);
848 static inline int sk_memalloc_socks(void)
855 static inline gfp_t
sk_gfp_mask(const struct sock
*sk
, gfp_t gfp_mask
)
857 return gfp_mask
| (sk
->sk_allocation
& __GFP_MEMALLOC
);
860 static inline void sk_acceptq_removed(struct sock
*sk
)
862 WRITE_ONCE(sk
->sk_ack_backlog
, sk
->sk_ack_backlog
- 1);
865 static inline void sk_acceptq_added(struct sock
*sk
)
867 WRITE_ONCE(sk
->sk_ack_backlog
, sk
->sk_ack_backlog
+ 1);
870 static inline bool sk_acceptq_is_full(const struct sock
*sk
)
872 return READ_ONCE(sk
->sk_ack_backlog
) > READ_ONCE(sk
->sk_max_ack_backlog
);
876 * Compute minimal free write space needed to queue new packets.
878 static inline int sk_stream_min_wspace(const struct sock
*sk
)
880 return READ_ONCE(sk
->sk_wmem_queued
) >> 1;
883 static inline int sk_stream_wspace(const struct sock
*sk
)
885 return READ_ONCE(sk
->sk_sndbuf
) - READ_ONCE(sk
->sk_wmem_queued
);
888 static inline void sk_wmem_queued_add(struct sock
*sk
, int val
)
890 WRITE_ONCE(sk
->sk_wmem_queued
, sk
->sk_wmem_queued
+ val
);
893 void sk_stream_write_space(struct sock
*sk
);
895 /* OOB backlog add */
896 static inline void __sk_add_backlog(struct sock
*sk
, struct sk_buff
*skb
)
898 /* dont let skb dst not refcounted, we are going to leave rcu lock */
901 if (!sk
->sk_backlog
.tail
)
902 WRITE_ONCE(sk
->sk_backlog
.head
, skb
);
904 sk
->sk_backlog
.tail
->next
= skb
;
906 WRITE_ONCE(sk
->sk_backlog
.tail
, skb
);
911 * Take into account size of receive queue and backlog queue
912 * Do not take into account this skb truesize,
913 * to allow even a single big packet to come.
915 static inline bool sk_rcvqueues_full(const struct sock
*sk
, unsigned int limit
)
917 unsigned int qsize
= sk
->sk_backlog
.len
+ atomic_read(&sk
->sk_rmem_alloc
);
919 return qsize
> limit
;
922 /* The per-socket spinlock must be held here. */
923 static inline __must_check
int sk_add_backlog(struct sock
*sk
, struct sk_buff
*skb
,
926 if (sk_rcvqueues_full(sk
, limit
))
930 * If the skb was allocated from pfmemalloc reserves, only
931 * allow SOCK_MEMALLOC sockets to use it as this socket is
932 * helping free memory
934 if (skb_pfmemalloc(skb
) && !sock_flag(sk
, SOCK_MEMALLOC
))
937 __sk_add_backlog(sk
, skb
);
938 sk
->sk_backlog
.len
+= skb
->truesize
;
942 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
);
944 static inline int sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
946 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
947 return __sk_backlog_rcv(sk
, skb
);
949 return sk
->sk_backlog_rcv(sk
, skb
);
952 static inline void sk_incoming_cpu_update(struct sock
*sk
)
954 int cpu
= raw_smp_processor_id();
956 if (unlikely(READ_ONCE(sk
->sk_incoming_cpu
) != cpu
))
957 WRITE_ONCE(sk
->sk_incoming_cpu
, cpu
);
960 static inline void sock_rps_record_flow_hash(__u32 hash
)
963 struct rps_sock_flow_table
*sock_flow_table
;
966 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
967 rps_record_sock_flow(sock_flow_table
, hash
);
972 static inline void sock_rps_record_flow(const struct sock
*sk
)
975 if (static_branch_unlikely(&rfs_needed
)) {
976 /* Reading sk->sk_rxhash might incur an expensive cache line
979 * TCP_ESTABLISHED does cover almost all states where RFS
980 * might be useful, and is cheaper [1] than testing :
981 * IPv4: inet_sk(sk)->inet_daddr
982 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
983 * OR an additional socket flag
984 * [1] : sk_state and sk_prot are in the same cache line.
986 if (sk
->sk_state
== TCP_ESTABLISHED
)
987 sock_rps_record_flow_hash(sk
->sk_rxhash
);
992 static inline void sock_rps_save_rxhash(struct sock
*sk
,
993 const struct sk_buff
*skb
)
996 if (unlikely(sk
->sk_rxhash
!= skb
->hash
))
997 sk
->sk_rxhash
= skb
->hash
;
1001 static inline void sock_rps_reset_rxhash(struct sock
*sk
)
1008 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1010 release_sock(__sk); \
1011 __rc = __condition; \
1013 *(__timeo) = wait_woken(__wait, \
1014 TASK_INTERRUPTIBLE, \
1017 sched_annotate_sleep(); \
1019 __rc = __condition; \
1023 int sk_stream_wait_connect(struct sock
*sk
, long *timeo_p
);
1024 int sk_stream_wait_memory(struct sock
*sk
, long *timeo_p
);
1025 void sk_stream_wait_close(struct sock
*sk
, long timeo_p
);
1026 int sk_stream_error(struct sock
*sk
, int flags
, int err
);
1027 void sk_stream_kill_queues(struct sock
*sk
);
1028 void sk_set_memalloc(struct sock
*sk
);
1029 void sk_clear_memalloc(struct sock
*sk
);
1031 void __sk_flush_backlog(struct sock
*sk
);
1033 static inline bool sk_flush_backlog(struct sock
*sk
)
1035 if (unlikely(READ_ONCE(sk
->sk_backlog
.tail
))) {
1036 __sk_flush_backlog(sk
);
1042 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
);
1044 struct request_sock_ops
;
1045 struct timewait_sock_ops
;
1046 struct inet_hashinfo
;
1047 struct raw_hashinfo
;
1048 struct smc_hashinfo
;
1052 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1053 * un-modified. Special care is taken when initializing object to zero.
1055 static inline void sk_prot_clear_nulls(struct sock
*sk
, int size
)
1057 if (offsetof(struct sock
, sk_node
.next
) != 0)
1058 memset(sk
, 0, offsetof(struct sock
, sk_node
.next
));
1059 memset(&sk
->sk_node
.pprev
, 0,
1060 size
- offsetof(struct sock
, sk_node
.pprev
));
1063 /* Networking protocol blocks we attach to sockets.
1064 * socket layer -> transport layer interface
1067 void (*close
)(struct sock
*sk
,
1069 int (*pre_connect
)(struct sock
*sk
,
1070 struct sockaddr
*uaddr
,
1072 int (*connect
)(struct sock
*sk
,
1073 struct sockaddr
*uaddr
,
1075 int (*disconnect
)(struct sock
*sk
, int flags
);
1077 struct sock
* (*accept
)(struct sock
*sk
, int flags
, int *err
,
1080 int (*ioctl
)(struct sock
*sk
, int cmd
,
1082 int (*init
)(struct sock
*sk
);
1083 void (*destroy
)(struct sock
*sk
);
1084 void (*shutdown
)(struct sock
*sk
, int how
);
1085 int (*setsockopt
)(struct sock
*sk
, int level
,
1086 int optname
, char __user
*optval
,
1087 unsigned int optlen
);
1088 int (*getsockopt
)(struct sock
*sk
, int level
,
1089 int optname
, char __user
*optval
,
1090 int __user
*option
);
1091 void (*keepalive
)(struct sock
*sk
, int valbool
);
1092 #ifdef CONFIG_COMPAT
1093 int (*compat_setsockopt
)(struct sock
*sk
,
1095 int optname
, char __user
*optval
,
1096 unsigned int optlen
);
1097 int (*compat_getsockopt
)(struct sock
*sk
,
1099 int optname
, char __user
*optval
,
1100 int __user
*option
);
1101 int (*compat_ioctl
)(struct sock
*sk
,
1102 unsigned int cmd
, unsigned long arg
);
1104 int (*sendmsg
)(struct sock
*sk
, struct msghdr
*msg
,
1106 int (*recvmsg
)(struct sock
*sk
, struct msghdr
*msg
,
1107 size_t len
, int noblock
, int flags
,
1109 int (*sendpage
)(struct sock
*sk
, struct page
*page
,
1110 int offset
, size_t size
, int flags
);
1111 int (*bind
)(struct sock
*sk
,
1112 struct sockaddr
*uaddr
, int addr_len
);
1114 int (*backlog_rcv
) (struct sock
*sk
,
1115 struct sk_buff
*skb
);
1117 void (*release_cb
)(struct sock
*sk
);
1119 /* Keeping track of sk's, looking them up, and port selection methods. */
1120 int (*hash
)(struct sock
*sk
);
1121 void (*unhash
)(struct sock
*sk
);
1122 void (*rehash
)(struct sock
*sk
);
1123 int (*get_port
)(struct sock
*sk
, unsigned short snum
);
1125 /* Keeping track of sockets in use */
1126 #ifdef CONFIG_PROC_FS
1127 unsigned int inuse_idx
;
1130 bool (*stream_memory_free
)(const struct sock
*sk
, int wake
);
1131 bool (*stream_memory_read
)(const struct sock
*sk
);
1132 /* Memory pressure */
1133 void (*enter_memory_pressure
)(struct sock
*sk
);
1134 void (*leave_memory_pressure
)(struct sock
*sk
);
1135 atomic_long_t
*memory_allocated
; /* Current allocated memory. */
1136 struct percpu_counter
*sockets_allocated
; /* Current number of sockets. */
1138 * Pressure flag: try to collapse.
1139 * Technical note: it is used by multiple contexts non atomically.
1140 * All the __sk_mem_schedule() is of this nature: accounting
1141 * is strict, actions are advisory and have some latency.
1143 unsigned long *memory_pressure
;
1148 u32 sysctl_wmem_offset
;
1149 u32 sysctl_rmem_offset
;
1154 struct kmem_cache
*slab
;
1155 unsigned int obj_size
;
1156 slab_flags_t slab_flags
;
1157 unsigned int useroffset
; /* Usercopy region offset */
1158 unsigned int usersize
; /* Usercopy region size */
1160 struct percpu_counter
*orphan_count
;
1162 struct request_sock_ops
*rsk_prot
;
1163 struct timewait_sock_ops
*twsk_prot
;
1166 struct inet_hashinfo
*hashinfo
;
1167 struct udp_table
*udp_table
;
1168 struct raw_hashinfo
*raw_hash
;
1169 struct smc_hashinfo
*smc_hash
;
1172 struct module
*owner
;
1176 struct list_head node
;
1177 #ifdef SOCK_REFCNT_DEBUG
1180 int (*diag_destroy
)(struct sock
*sk
, int err
);
1181 } __randomize_layout
;
1183 int proto_register(struct proto
*prot
, int alloc_slab
);
1184 void proto_unregister(struct proto
*prot
);
1185 int sock_load_diag_module(int family
, int protocol
);
1187 #ifdef SOCK_REFCNT_DEBUG
1188 static inline void sk_refcnt_debug_inc(struct sock
*sk
)
1190 atomic_inc(&sk
->sk_prot
->socks
);
1193 static inline void sk_refcnt_debug_dec(struct sock
*sk
)
1195 atomic_dec(&sk
->sk_prot
->socks
);
1196 printk(KERN_DEBUG
"%s socket %p released, %d are still alive\n",
1197 sk
->sk_prot
->name
, sk
, atomic_read(&sk
->sk_prot
->socks
));
1200 static inline void sk_refcnt_debug_release(const struct sock
*sk
)
1202 if (refcount_read(&sk
->sk_refcnt
) != 1)
1203 printk(KERN_DEBUG
"Destruction of the %s socket %p delayed, refcnt=%d\n",
1204 sk
->sk_prot
->name
, sk
, refcount_read(&sk
->sk_refcnt
));
1206 #else /* SOCK_REFCNT_DEBUG */
1207 #define sk_refcnt_debug_inc(sk) do { } while (0)
1208 #define sk_refcnt_debug_dec(sk) do { } while (0)
1209 #define sk_refcnt_debug_release(sk) do { } while (0)
1210 #endif /* SOCK_REFCNT_DEBUG */
1212 static inline bool __sk_stream_memory_free(const struct sock
*sk
, int wake
)
1214 if (READ_ONCE(sk
->sk_wmem_queued
) >= READ_ONCE(sk
->sk_sndbuf
))
1217 return sk
->sk_prot
->stream_memory_free
?
1218 sk
->sk_prot
->stream_memory_free(sk
, wake
) : true;
1221 static inline bool sk_stream_memory_free(const struct sock
*sk
)
1223 return __sk_stream_memory_free(sk
, 0);
1226 static inline bool __sk_stream_is_writeable(const struct sock
*sk
, int wake
)
1228 return sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
) &&
1229 __sk_stream_memory_free(sk
, wake
);
1232 static inline bool sk_stream_is_writeable(const struct sock
*sk
)
1234 return __sk_stream_is_writeable(sk
, 0);
1237 static inline int sk_under_cgroup_hierarchy(struct sock
*sk
,
1238 struct cgroup
*ancestor
)
1240 #ifdef CONFIG_SOCK_CGROUP_DATA
1241 return cgroup_is_descendant(sock_cgroup_ptr(&sk
->sk_cgrp_data
),
1248 static inline bool sk_has_memory_pressure(const struct sock
*sk
)
1250 return sk
->sk_prot
->memory_pressure
!= NULL
;
1253 static inline bool sk_under_memory_pressure(const struct sock
*sk
)
1255 if (!sk
->sk_prot
->memory_pressure
)
1258 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
1259 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
1262 return !!*sk
->sk_prot
->memory_pressure
;
1266 sk_memory_allocated(const struct sock
*sk
)
1268 return atomic_long_read(sk
->sk_prot
->memory_allocated
);
1272 sk_memory_allocated_add(struct sock
*sk
, int amt
)
1274 return atomic_long_add_return(amt
, sk
->sk_prot
->memory_allocated
);
1278 sk_memory_allocated_sub(struct sock
*sk
, int amt
)
1280 atomic_long_sub(amt
, sk
->sk_prot
->memory_allocated
);
1283 static inline void sk_sockets_allocated_dec(struct sock
*sk
)
1285 percpu_counter_dec(sk
->sk_prot
->sockets_allocated
);
1288 static inline void sk_sockets_allocated_inc(struct sock
*sk
)
1290 percpu_counter_inc(sk
->sk_prot
->sockets_allocated
);
1294 sk_sockets_allocated_read_positive(struct sock
*sk
)
1296 return percpu_counter_read_positive(sk
->sk_prot
->sockets_allocated
);
1300 proto_sockets_allocated_sum_positive(struct proto
*prot
)
1302 return percpu_counter_sum_positive(prot
->sockets_allocated
);
1306 proto_memory_allocated(struct proto
*prot
)
1308 return atomic_long_read(prot
->memory_allocated
);
1312 proto_memory_pressure(struct proto
*prot
)
1314 if (!prot
->memory_pressure
)
1316 return !!*prot
->memory_pressure
;
1320 #ifdef CONFIG_PROC_FS
1321 /* Called with local bh disabled */
1322 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int inc
);
1323 int sock_prot_inuse_get(struct net
*net
, struct proto
*proto
);
1324 int sock_inuse_get(struct net
*net
);
1326 static inline void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
,
1333 /* With per-bucket locks this operation is not-atomic, so that
1334 * this version is not worse.
1336 static inline int __sk_prot_rehash(struct sock
*sk
)
1338 sk
->sk_prot
->unhash(sk
);
1339 return sk
->sk_prot
->hash(sk
);
1342 /* About 10 seconds */
1343 #define SOCK_DESTROY_TIME (10*HZ)
1345 /* Sockets 0-1023 can't be bound to unless you are superuser */
1346 #define PROT_SOCK 1024
1348 #define SHUTDOWN_MASK 3
1349 #define RCV_SHUTDOWN 1
1350 #define SEND_SHUTDOWN 2
1352 #define SOCK_SNDBUF_LOCK 1
1353 #define SOCK_RCVBUF_LOCK 2
1354 #define SOCK_BINDADDR_LOCK 4
1355 #define SOCK_BINDPORT_LOCK 8
1357 struct socket_alloc
{
1358 struct socket socket
;
1359 struct inode vfs_inode
;
1362 static inline struct socket
*SOCKET_I(struct inode
*inode
)
1364 return &container_of(inode
, struct socket_alloc
, vfs_inode
)->socket
;
1367 static inline struct inode
*SOCK_INODE(struct socket
*socket
)
1369 return &container_of(socket
, struct socket_alloc
, socket
)->vfs_inode
;
1373 * Functions for memory accounting
1375 int __sk_mem_raise_allocated(struct sock
*sk
, int size
, int amt
, int kind
);
1376 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
);
1377 void __sk_mem_reduce_allocated(struct sock
*sk
, int amount
);
1378 void __sk_mem_reclaim(struct sock
*sk
, int amount
);
1380 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1381 * do not necessarily have 16x time more memory than 4KB ones.
1383 #define SK_MEM_QUANTUM 4096
1384 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1385 #define SK_MEM_SEND 0
1386 #define SK_MEM_RECV 1
1388 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1389 static inline long sk_prot_mem_limits(const struct sock
*sk
, int index
)
1391 long val
= sk
->sk_prot
->sysctl_mem
[index
];
1393 #if PAGE_SIZE > SK_MEM_QUANTUM
1394 val
<<= PAGE_SHIFT
- SK_MEM_QUANTUM_SHIFT
;
1395 #elif PAGE_SIZE < SK_MEM_QUANTUM
1396 val
>>= SK_MEM_QUANTUM_SHIFT
- PAGE_SHIFT
;
1401 static inline int sk_mem_pages(int amt
)
1403 return (amt
+ SK_MEM_QUANTUM
- 1) >> SK_MEM_QUANTUM_SHIFT
;
1406 static inline bool sk_has_account(struct sock
*sk
)
1408 /* return true if protocol supports memory accounting */
1409 return !!sk
->sk_prot
->memory_allocated
;
1412 static inline bool sk_wmem_schedule(struct sock
*sk
, int size
)
1414 if (!sk_has_account(sk
))
1416 return size
<= sk
->sk_forward_alloc
||
1417 __sk_mem_schedule(sk
, size
, SK_MEM_SEND
);
1421 sk_rmem_schedule(struct sock
*sk
, struct sk_buff
*skb
, int size
)
1423 if (!sk_has_account(sk
))
1425 return size
<= sk
->sk_forward_alloc
||
1426 __sk_mem_schedule(sk
, size
, SK_MEM_RECV
) ||
1427 skb_pfmemalloc(skb
);
1430 static inline void sk_mem_reclaim(struct sock
*sk
)
1432 if (!sk_has_account(sk
))
1434 if (sk
->sk_forward_alloc
>= SK_MEM_QUANTUM
)
1435 __sk_mem_reclaim(sk
, sk
->sk_forward_alloc
);
1438 static inline void sk_mem_reclaim_partial(struct sock
*sk
)
1440 if (!sk_has_account(sk
))
1442 if (sk
->sk_forward_alloc
> SK_MEM_QUANTUM
)
1443 __sk_mem_reclaim(sk
, sk
->sk_forward_alloc
- 1);
1446 static inline void sk_mem_charge(struct sock
*sk
, int size
)
1448 if (!sk_has_account(sk
))
1450 sk
->sk_forward_alloc
-= size
;
1453 static inline void sk_mem_uncharge(struct sock
*sk
, int size
)
1455 if (!sk_has_account(sk
))
1457 sk
->sk_forward_alloc
+= size
;
1459 /* Avoid a possible overflow.
1460 * TCP send queues can make this happen, if sk_mem_reclaim()
1461 * is not called and more than 2 GBytes are released at once.
1463 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1464 * no need to hold that much forward allocation anyway.
1466 if (unlikely(sk
->sk_forward_alloc
>= 1 << 21))
1467 __sk_mem_reclaim(sk
, 1 << 20);
1470 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key
);
1471 static inline void sk_wmem_free_skb(struct sock
*sk
, struct sk_buff
*skb
)
1473 sock_set_flag(sk
, SOCK_QUEUE_SHRUNK
);
1474 sk_wmem_queued_add(sk
, -skb
->truesize
);
1475 sk_mem_uncharge(sk
, skb
->truesize
);
1476 if (static_branch_unlikely(&tcp_tx_skb_cache_key
) &&
1477 !sk
->sk_tx_skb_cache
&& !skb_cloned(skb
)) {
1478 skb_zcopy_clear(skb
, true);
1479 sk
->sk_tx_skb_cache
= skb
;
1485 static inline void sock_release_ownership(struct sock
*sk
)
1487 if (sk
->sk_lock
.owned
) {
1488 sk
->sk_lock
.owned
= 0;
1490 /* The sk_lock has mutex_unlock() semantics: */
1491 mutex_release(&sk
->sk_lock
.dep_map
, _RET_IP_
);
1496 * Macro so as to not evaluate some arguments when
1497 * lockdep is not enabled.
1499 * Mark both the sk_lock and the sk_lock.slock as a
1500 * per-address-family lock class.
1502 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1504 sk->sk_lock.owned = 0; \
1505 init_waitqueue_head(&sk->sk_lock.wq); \
1506 spin_lock_init(&(sk)->sk_lock.slock); \
1507 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1508 sizeof((sk)->sk_lock)); \
1509 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1511 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1514 #ifdef CONFIG_LOCKDEP
1515 static inline bool lockdep_sock_is_held(const struct sock
*sk
)
1517 return lockdep_is_held(&sk
->sk_lock
) ||
1518 lockdep_is_held(&sk
->sk_lock
.slock
);
1522 void lock_sock_nested(struct sock
*sk
, int subclass
);
1524 static inline void lock_sock(struct sock
*sk
)
1526 lock_sock_nested(sk
, 0);
1529 void __release_sock(struct sock
*sk
);
1530 void release_sock(struct sock
*sk
);
1532 /* BH context may only use the following locking interface. */
1533 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1534 #define bh_lock_sock_nested(__sk) \
1535 spin_lock_nested(&((__sk)->sk_lock.slock), \
1536 SINGLE_DEPTH_NESTING)
1537 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1539 bool lock_sock_fast(struct sock
*sk
);
1541 * unlock_sock_fast - complement of lock_sock_fast
1545 * fast unlock socket for user context.
1546 * If slow mode is on, we call regular release_sock()
1548 static inline void unlock_sock_fast(struct sock
*sk
, bool slow
)
1553 spin_unlock_bh(&sk
->sk_lock
.slock
);
1556 /* Used by processes to "lock" a socket state, so that
1557 * interrupts and bottom half handlers won't change it
1558 * from under us. It essentially blocks any incoming
1559 * packets, so that we won't get any new data or any
1560 * packets that change the state of the socket.
1562 * While locked, BH processing will add new packets to
1563 * the backlog queue. This queue is processed by the
1564 * owner of the socket lock right before it is released.
1566 * Since ~2.3.5 it is also exclusive sleep lock serializing
1567 * accesses from user process context.
1570 static inline void sock_owned_by_me(const struct sock
*sk
)
1572 #ifdef CONFIG_LOCKDEP
1573 WARN_ON_ONCE(!lockdep_sock_is_held(sk
) && debug_locks
);
1577 static inline bool sock_owned_by_user(const struct sock
*sk
)
1579 sock_owned_by_me(sk
);
1580 return sk
->sk_lock
.owned
;
1583 static inline bool sock_owned_by_user_nocheck(const struct sock
*sk
)
1585 return sk
->sk_lock
.owned
;
1588 /* no reclassification while locks are held */
1589 static inline bool sock_allow_reclassification(const struct sock
*csk
)
1591 struct sock
*sk
= (struct sock
*)csk
;
1593 return !sk
->sk_lock
.owned
&& !spin_is_locked(&sk
->sk_lock
.slock
);
1596 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1597 struct proto
*prot
, int kern
);
1598 void sk_free(struct sock
*sk
);
1599 void sk_destruct(struct sock
*sk
);
1600 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
);
1601 void sk_free_unlock_clone(struct sock
*sk
);
1603 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1605 void __sock_wfree(struct sk_buff
*skb
);
1606 void sock_wfree(struct sk_buff
*skb
);
1607 struct sk_buff
*sock_omalloc(struct sock
*sk
, unsigned long size
,
1609 void skb_orphan_partial(struct sk_buff
*skb
);
1610 void sock_rfree(struct sk_buff
*skb
);
1611 void sock_efree(struct sk_buff
*skb
);
1613 void sock_edemux(struct sk_buff
*skb
);
1615 #define sock_edemux sock_efree
1618 int sock_setsockopt(struct socket
*sock
, int level
, int op
,
1619 char __user
*optval
, unsigned int optlen
);
1621 int sock_getsockopt(struct socket
*sock
, int level
, int op
,
1622 char __user
*optval
, int __user
*optlen
);
1623 int sock_gettstamp(struct socket
*sock
, void __user
*userstamp
,
1624 bool timeval
, bool time32
);
1625 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1626 int noblock
, int *errcode
);
1627 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1628 unsigned long data_len
, int noblock
,
1629 int *errcode
, int max_page_order
);
1630 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
);
1631 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
);
1632 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
);
1633 void sk_send_sigurg(struct sock
*sk
);
1635 struct sockcm_cookie
{
1641 static inline void sockcm_init(struct sockcm_cookie
*sockc
,
1642 const struct sock
*sk
)
1644 *sockc
= (struct sockcm_cookie
) { .tsflags
= sk
->sk_tsflags
};
1647 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
1648 struct sockcm_cookie
*sockc
);
1649 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
1650 struct sockcm_cookie
*sockc
);
1653 * Functions to fill in entries in struct proto_ops when a protocol
1654 * does not implement a particular function.
1656 int sock_no_bind(struct socket
*, struct sockaddr
*, int);
1657 int sock_no_connect(struct socket
*, struct sockaddr
*, int, int);
1658 int sock_no_socketpair(struct socket
*, struct socket
*);
1659 int sock_no_accept(struct socket
*, struct socket
*, int, bool);
1660 int sock_no_getname(struct socket
*, struct sockaddr
*, int);
1661 int sock_no_ioctl(struct socket
*, unsigned int, unsigned long);
1662 int sock_no_listen(struct socket
*, int);
1663 int sock_no_shutdown(struct socket
*, int);
1664 int sock_no_getsockopt(struct socket
*, int , int, char __user
*, int __user
*);
1665 int sock_no_setsockopt(struct socket
*, int, int, char __user
*, unsigned int);
1666 int sock_no_sendmsg(struct socket
*, struct msghdr
*, size_t);
1667 int sock_no_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t len
);
1668 int sock_no_recvmsg(struct socket
*, struct msghdr
*, size_t, int);
1669 int sock_no_mmap(struct file
*file
, struct socket
*sock
,
1670 struct vm_area_struct
*vma
);
1671 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
1672 size_t size
, int flags
);
1673 ssize_t
sock_no_sendpage_locked(struct sock
*sk
, struct page
*page
,
1674 int offset
, size_t size
, int flags
);
1677 * Functions to fill in entries in struct proto_ops when a protocol
1678 * uses the inet style.
1680 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1681 char __user
*optval
, int __user
*optlen
);
1682 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
1684 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1685 char __user
*optval
, unsigned int optlen
);
1686 int compat_sock_common_getsockopt(struct socket
*sock
, int level
,
1687 int optname
, char __user
*optval
, int __user
*optlen
);
1688 int compat_sock_common_setsockopt(struct socket
*sock
, int level
,
1689 int optname
, char __user
*optval
, unsigned int optlen
);
1691 void sk_common_release(struct sock
*sk
);
1694 * Default socket callbacks and setup code
1697 /* Initialise core socket variables */
1698 void sock_init_data(struct socket
*sock
, struct sock
*sk
);
1701 * Socket reference counting postulates.
1703 * * Each user of socket SHOULD hold a reference count.
1704 * * Each access point to socket (an hash table bucket, reference from a list,
1705 * running timer, skb in flight MUST hold a reference count.
1706 * * When reference count hits 0, it means it will never increase back.
1707 * * When reference count hits 0, it means that no references from
1708 * outside exist to this socket and current process on current CPU
1709 * is last user and may/should destroy this socket.
1710 * * sk_free is called from any context: process, BH, IRQ. When
1711 * it is called, socket has no references from outside -> sk_free
1712 * may release descendant resources allocated by the socket, but
1713 * to the time when it is called, socket is NOT referenced by any
1714 * hash tables, lists etc.
1715 * * Packets, delivered from outside (from network or from another process)
1716 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1717 * when they sit in queue. Otherwise, packets will leak to hole, when
1718 * socket is looked up by one cpu and unhasing is made by another CPU.
1719 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1720 * (leak to backlog). Packet socket does all the processing inside
1721 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1722 * use separate SMP lock, so that they are prone too.
1725 /* Ungrab socket and destroy it, if it was the last reference. */
1726 static inline void sock_put(struct sock
*sk
)
1728 if (refcount_dec_and_test(&sk
->sk_refcnt
))
1731 /* Generic version of sock_put(), dealing with all sockets
1732 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1734 void sock_gen_put(struct sock
*sk
);
1736 int __sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
,
1737 unsigned int trim_cap
, bool refcounted
);
1738 static inline int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
,
1741 return __sk_receive_skb(sk
, skb
, nested
, 1, true);
1744 static inline void sk_tx_queue_set(struct sock
*sk
, int tx_queue
)
1746 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1747 if (WARN_ON_ONCE((unsigned short)tx_queue
>= USHRT_MAX
))
1749 sk
->sk_tx_queue_mapping
= tx_queue
;
1752 #define NO_QUEUE_MAPPING USHRT_MAX
1754 static inline void sk_tx_queue_clear(struct sock
*sk
)
1756 sk
->sk_tx_queue_mapping
= NO_QUEUE_MAPPING
;
1759 static inline int sk_tx_queue_get(const struct sock
*sk
)
1761 if (sk
&& sk
->sk_tx_queue_mapping
!= NO_QUEUE_MAPPING
)
1762 return sk
->sk_tx_queue_mapping
;
1767 static inline void sk_rx_queue_set(struct sock
*sk
, const struct sk_buff
*skb
)
1770 if (skb_rx_queue_recorded(skb
)) {
1771 u16 rx_queue
= skb_get_rx_queue(skb
);
1773 if (WARN_ON_ONCE(rx_queue
== NO_QUEUE_MAPPING
))
1776 sk
->sk_rx_queue_mapping
= rx_queue
;
1781 static inline void sk_rx_queue_clear(struct sock
*sk
)
1784 sk
->sk_rx_queue_mapping
= NO_QUEUE_MAPPING
;
1789 static inline int sk_rx_queue_get(const struct sock
*sk
)
1791 if (sk
&& sk
->sk_rx_queue_mapping
!= NO_QUEUE_MAPPING
)
1792 return sk
->sk_rx_queue_mapping
;
1798 static inline void sk_set_socket(struct sock
*sk
, struct socket
*sock
)
1800 sk_tx_queue_clear(sk
);
1801 sk
->sk_socket
= sock
;
1804 static inline wait_queue_head_t
*sk_sleep(struct sock
*sk
)
1806 BUILD_BUG_ON(offsetof(struct socket_wq
, wait
) != 0);
1807 return &rcu_dereference_raw(sk
->sk_wq
)->wait
;
1809 /* Detach socket from process context.
1810 * Announce socket dead, detach it from wait queue and inode.
1811 * Note that parent inode held reference count on this struct sock,
1812 * we do not release it in this function, because protocol
1813 * probably wants some additional cleanups or even continuing
1814 * to work with this socket (TCP).
1816 static inline void sock_orphan(struct sock
*sk
)
1818 write_lock_bh(&sk
->sk_callback_lock
);
1819 sock_set_flag(sk
, SOCK_DEAD
);
1820 sk_set_socket(sk
, NULL
);
1822 write_unlock_bh(&sk
->sk_callback_lock
);
1825 static inline void sock_graft(struct sock
*sk
, struct socket
*parent
)
1827 WARN_ON(parent
->sk
);
1828 write_lock_bh(&sk
->sk_callback_lock
);
1829 rcu_assign_pointer(sk
->sk_wq
, &parent
->wq
);
1831 sk_set_socket(sk
, parent
);
1832 sk
->sk_uid
= SOCK_INODE(parent
)->i_uid
;
1833 security_sock_graft(sk
, parent
);
1834 write_unlock_bh(&sk
->sk_callback_lock
);
1837 kuid_t
sock_i_uid(struct sock
*sk
);
1838 unsigned long sock_i_ino(struct sock
*sk
);
1840 static inline kuid_t
sock_net_uid(const struct net
*net
, const struct sock
*sk
)
1842 return sk
? sk
->sk_uid
: make_kuid(net
->user_ns
, 0);
1845 static inline u32
net_tx_rndhash(void)
1847 u32 v
= prandom_u32();
1852 static inline void sk_set_txhash(struct sock
*sk
)
1854 sk
->sk_txhash
= net_tx_rndhash();
1857 static inline void sk_rethink_txhash(struct sock
*sk
)
1863 static inline struct dst_entry
*
1864 __sk_dst_get(struct sock
*sk
)
1866 return rcu_dereference_check(sk
->sk_dst_cache
,
1867 lockdep_sock_is_held(sk
));
1870 static inline struct dst_entry
*
1871 sk_dst_get(struct sock
*sk
)
1873 struct dst_entry
*dst
;
1876 dst
= rcu_dereference(sk
->sk_dst_cache
);
1877 if (dst
&& !atomic_inc_not_zero(&dst
->__refcnt
))
1883 static inline void dst_negative_advice(struct sock
*sk
)
1885 struct dst_entry
*ndst
, *dst
= __sk_dst_get(sk
);
1887 sk_rethink_txhash(sk
);
1889 if (dst
&& dst
->ops
->negative_advice
) {
1890 ndst
= dst
->ops
->negative_advice(dst
);
1893 rcu_assign_pointer(sk
->sk_dst_cache
, ndst
);
1894 sk_tx_queue_clear(sk
);
1895 sk
->sk_dst_pending_confirm
= 0;
1901 __sk_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
1903 struct dst_entry
*old_dst
;
1905 sk_tx_queue_clear(sk
);
1906 sk
->sk_dst_pending_confirm
= 0;
1907 old_dst
= rcu_dereference_protected(sk
->sk_dst_cache
,
1908 lockdep_sock_is_held(sk
));
1909 rcu_assign_pointer(sk
->sk_dst_cache
, dst
);
1910 dst_release(old_dst
);
1914 sk_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
1916 struct dst_entry
*old_dst
;
1918 sk_tx_queue_clear(sk
);
1919 sk
->sk_dst_pending_confirm
= 0;
1920 old_dst
= xchg((__force
struct dst_entry
**)&sk
->sk_dst_cache
, dst
);
1921 dst_release(old_dst
);
1925 __sk_dst_reset(struct sock
*sk
)
1927 __sk_dst_set(sk
, NULL
);
1931 sk_dst_reset(struct sock
*sk
)
1933 sk_dst_set(sk
, NULL
);
1936 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
);
1938 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
);
1940 static inline void sk_dst_confirm(struct sock
*sk
)
1942 if (!READ_ONCE(sk
->sk_dst_pending_confirm
))
1943 WRITE_ONCE(sk
->sk_dst_pending_confirm
, 1);
1946 static inline void sock_confirm_neigh(struct sk_buff
*skb
, struct neighbour
*n
)
1948 if (skb_get_dst_pending_confirm(skb
)) {
1949 struct sock
*sk
= skb
->sk
;
1950 unsigned long now
= jiffies
;
1952 /* avoid dirtying neighbour */
1953 if (READ_ONCE(n
->confirmed
) != now
)
1954 WRITE_ONCE(n
->confirmed
, now
);
1955 if (sk
&& READ_ONCE(sk
->sk_dst_pending_confirm
))
1956 WRITE_ONCE(sk
->sk_dst_pending_confirm
, 0);
1960 bool sk_mc_loop(struct sock
*sk
);
1962 static inline bool sk_can_gso(const struct sock
*sk
)
1964 return net_gso_ok(sk
->sk_route_caps
, sk
->sk_gso_type
);
1967 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
);
1969 static inline void sk_nocaps_add(struct sock
*sk
, netdev_features_t flags
)
1971 sk
->sk_route_nocaps
|= flags
;
1972 sk
->sk_route_caps
&= ~flags
;
1975 static inline int skb_do_copy_data_nocache(struct sock
*sk
, struct sk_buff
*skb
,
1976 struct iov_iter
*from
, char *to
,
1977 int copy
, int offset
)
1979 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1981 if (!csum_and_copy_from_iter_full(to
, copy
, &csum
, from
))
1983 skb
->csum
= csum_block_add(skb
->csum
, csum
, offset
);
1984 } else if (sk
->sk_route_caps
& NETIF_F_NOCACHE_COPY
) {
1985 if (!copy_from_iter_full_nocache(to
, copy
, from
))
1987 } else if (!copy_from_iter_full(to
, copy
, from
))
1993 static inline int skb_add_data_nocache(struct sock
*sk
, struct sk_buff
*skb
,
1994 struct iov_iter
*from
, int copy
)
1996 int err
, offset
= skb
->len
;
1998 err
= skb_do_copy_data_nocache(sk
, skb
, from
, skb_put(skb
, copy
),
2001 __skb_trim(skb
, offset
);
2006 static inline int skb_copy_to_page_nocache(struct sock
*sk
, struct iov_iter
*from
,
2007 struct sk_buff
*skb
,
2013 err
= skb_do_copy_data_nocache(sk
, skb
, from
, page_address(page
) + off
,
2019 skb
->data_len
+= copy
;
2020 skb
->truesize
+= copy
;
2021 sk_wmem_queued_add(sk
, copy
);
2022 sk_mem_charge(sk
, copy
);
2027 * sk_wmem_alloc_get - returns write allocations
2030 * Returns sk_wmem_alloc minus initial offset of one
2032 static inline int sk_wmem_alloc_get(const struct sock
*sk
)
2034 return refcount_read(&sk
->sk_wmem_alloc
) - 1;
2038 * sk_rmem_alloc_get - returns read allocations
2041 * Returns sk_rmem_alloc
2043 static inline int sk_rmem_alloc_get(const struct sock
*sk
)
2045 return atomic_read(&sk
->sk_rmem_alloc
);
2049 * sk_has_allocations - check if allocations are outstanding
2052 * Returns true if socket has write or read allocations
2054 static inline bool sk_has_allocations(const struct sock
*sk
)
2056 return sk_wmem_alloc_get(sk
) || sk_rmem_alloc_get(sk
);
2060 * skwq_has_sleeper - check if there are any waiting processes
2061 * @wq: struct socket_wq
2063 * Returns true if socket_wq has waiting processes
2065 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2066 * barrier call. They were added due to the race found within the tcp code.
2068 * Consider following tcp code paths::
2071 * sys_select receive packet
2073 * __add_wait_queue update tp->rcv_nxt
2075 * tp->rcv_nxt check sock_def_readable
2077 * schedule rcu_read_lock();
2078 * wq = rcu_dereference(sk->sk_wq);
2079 * if (wq && waitqueue_active(&wq->wait))
2080 * wake_up_interruptible(&wq->wait)
2084 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2085 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2086 * could then endup calling schedule and sleep forever if there are no more
2087 * data on the socket.
2090 static inline bool skwq_has_sleeper(struct socket_wq
*wq
)
2092 return wq
&& wq_has_sleeper(&wq
->wait
);
2096 * sock_poll_wait - place memory barrier behind the poll_wait call.
2098 * @sock: socket to wait on
2101 * See the comments in the wq_has_sleeper function.
2103 static inline void sock_poll_wait(struct file
*filp
, struct socket
*sock
,
2106 if (!poll_does_not_wait(p
)) {
2107 poll_wait(filp
, &sock
->wq
.wait
, p
);
2108 /* We need to be sure we are in sync with the
2109 * socket flags modification.
2111 * This memory barrier is paired in the wq_has_sleeper.
2117 static inline void skb_set_hash_from_sk(struct sk_buff
*skb
, struct sock
*sk
)
2119 if (sk
->sk_txhash
) {
2121 skb
->hash
= sk
->sk_txhash
;
2125 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
);
2128 * Queue a received datagram if it will fit. Stream and sequenced
2129 * protocols can't normally use this as they need to fit buffers in
2130 * and play with them.
2132 * Inlined as it's very short and called for pretty much every
2133 * packet ever received.
2135 static inline void skb_set_owner_r(struct sk_buff
*skb
, struct sock
*sk
)
2139 skb
->destructor
= sock_rfree
;
2140 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
2141 sk_mem_charge(sk
, skb
->truesize
);
2144 void sk_reset_timer(struct sock
*sk
, struct timer_list
*timer
,
2145 unsigned long expires
);
2147 void sk_stop_timer(struct sock
*sk
, struct timer_list
*timer
);
2149 int __sk_queue_drop_skb(struct sock
*sk
, struct sk_buff_head
*sk_queue
,
2150 struct sk_buff
*skb
, unsigned int flags
,
2151 void (*destructor
)(struct sock
*sk
,
2152 struct sk_buff
*skb
));
2153 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
);
2154 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
);
2156 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
);
2157 struct sk_buff
*sock_dequeue_err_skb(struct sock
*sk
);
2160 * Recover an error report and clear atomically
2163 static inline int sock_error(struct sock
*sk
)
2166 if (likely(!sk
->sk_err
))
2168 err
= xchg(&sk
->sk_err
, 0);
2172 static inline unsigned long sock_wspace(struct sock
*sk
)
2176 if (!(sk
->sk_shutdown
& SEND_SHUTDOWN
)) {
2177 amt
= sk
->sk_sndbuf
- refcount_read(&sk
->sk_wmem_alloc
);
2185 * We use sk->sk_wq_raw, from contexts knowing this
2186 * pointer is not NULL and cannot disappear/change.
2188 static inline void sk_set_bit(int nr
, struct sock
*sk
)
2190 if ((nr
== SOCKWQ_ASYNC_NOSPACE
|| nr
== SOCKWQ_ASYNC_WAITDATA
) &&
2191 !sock_flag(sk
, SOCK_FASYNC
))
2194 set_bit(nr
, &sk
->sk_wq_raw
->flags
);
2197 static inline void sk_clear_bit(int nr
, struct sock
*sk
)
2199 if ((nr
== SOCKWQ_ASYNC_NOSPACE
|| nr
== SOCKWQ_ASYNC_WAITDATA
) &&
2200 !sock_flag(sk
, SOCK_FASYNC
))
2203 clear_bit(nr
, &sk
->sk_wq_raw
->flags
);
2206 static inline void sk_wake_async(const struct sock
*sk
, int how
, int band
)
2208 if (sock_flag(sk
, SOCK_FASYNC
)) {
2210 sock_wake_async(rcu_dereference(sk
->sk_wq
), how
, band
);
2215 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2216 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2217 * Note: for send buffers, TCP works better if we can build two skbs at
2220 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2222 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2223 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2225 static inline void sk_stream_moderate_sndbuf(struct sock
*sk
)
2229 if (sk
->sk_userlocks
& SOCK_SNDBUF_LOCK
)
2232 val
= min(sk
->sk_sndbuf
, sk
->sk_wmem_queued
>> 1);
2234 WRITE_ONCE(sk
->sk_sndbuf
, max_t(u32
, val
, SOCK_MIN_SNDBUF
));
2237 struct sk_buff
*sk_stream_alloc_skb(struct sock
*sk
, int size
, gfp_t gfp
,
2238 bool force_schedule
);
2241 * sk_page_frag - return an appropriate page_frag
2244 * Use the per task page_frag instead of the per socket one for
2245 * optimization when we know that we're in the normal context and owns
2246 * everything that's associated with %current.
2248 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2249 * inside other socket operations and end up recursing into sk_page_frag()
2250 * while it's already in use.
2252 static inline struct page_frag
*sk_page_frag(struct sock
*sk
)
2254 if (gfpflags_normal_context(sk
->sk_allocation
))
2255 return ¤t
->task_frag
;
2257 return &sk
->sk_frag
;
2260 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
);
2263 * Default write policy as shown to user space via poll/select/SIGIO
2265 static inline bool sock_writeable(const struct sock
*sk
)
2267 return refcount_read(&sk
->sk_wmem_alloc
) < (READ_ONCE(sk
->sk_sndbuf
) >> 1);
2270 static inline gfp_t
gfp_any(void)
2272 return in_softirq() ? GFP_ATOMIC
: GFP_KERNEL
;
2275 static inline long sock_rcvtimeo(const struct sock
*sk
, bool noblock
)
2277 return noblock
? 0 : sk
->sk_rcvtimeo
;
2280 static inline long sock_sndtimeo(const struct sock
*sk
, bool noblock
)
2282 return noblock
? 0 : sk
->sk_sndtimeo
;
2285 static inline int sock_rcvlowat(const struct sock
*sk
, int waitall
, int len
)
2287 int v
= waitall
? len
: min_t(int, READ_ONCE(sk
->sk_rcvlowat
), len
);
2292 /* Alas, with timeout socket operations are not restartable.
2293 * Compare this to poll().
2295 static inline int sock_intr_errno(long timeo
)
2297 return timeo
== MAX_SCHEDULE_TIMEOUT
? -ERESTARTSYS
: -EINTR
;
2300 struct sock_skb_cb
{
2304 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2305 * using skb->cb[] would keep using it directly and utilize its
2306 * alignement guarantee.
2308 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2309 sizeof(struct sock_skb_cb)))
2311 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2312 SOCK_SKB_CB_OFFSET))
2314 #define sock_skb_cb_check_size(size) \
2315 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2318 sock_skb_set_dropcount(const struct sock
*sk
, struct sk_buff
*skb
)
2320 SOCK_SKB_CB(skb
)->dropcount
= sock_flag(sk
, SOCK_RXQ_OVFL
) ?
2321 atomic_read(&sk
->sk_drops
) : 0;
2324 static inline void sk_drops_add(struct sock
*sk
, const struct sk_buff
*skb
)
2326 int segs
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
2328 atomic_add(segs
, &sk
->sk_drops
);
2331 static inline ktime_t
sock_read_timestamp(struct sock
*sk
)
2333 #if BITS_PER_LONG==32
2338 seq
= read_seqbegin(&sk
->sk_stamp_seq
);
2340 } while (read_seqretry(&sk
->sk_stamp_seq
, seq
));
2344 return READ_ONCE(sk
->sk_stamp
);
2348 static inline void sock_write_timestamp(struct sock
*sk
, ktime_t kt
)
2350 #if BITS_PER_LONG==32
2351 write_seqlock(&sk
->sk_stamp_seq
);
2353 write_sequnlock(&sk
->sk_stamp_seq
);
2355 WRITE_ONCE(sk
->sk_stamp
, kt
);
2359 void __sock_recv_timestamp(struct msghdr
*msg
, struct sock
*sk
,
2360 struct sk_buff
*skb
);
2361 void __sock_recv_wifi_status(struct msghdr
*msg
, struct sock
*sk
,
2362 struct sk_buff
*skb
);
2365 sock_recv_timestamp(struct msghdr
*msg
, struct sock
*sk
, struct sk_buff
*skb
)
2367 ktime_t kt
= skb
->tstamp
;
2368 struct skb_shared_hwtstamps
*hwtstamps
= skb_hwtstamps(skb
);
2371 * generate control messages if
2372 * - receive time stamping in software requested
2373 * - software time stamp available and wanted
2374 * - hardware time stamps available and wanted
2376 if (sock_flag(sk
, SOCK_RCVTSTAMP
) ||
2377 (sk
->sk_tsflags
& SOF_TIMESTAMPING_RX_SOFTWARE
) ||
2378 (kt
&& sk
->sk_tsflags
& SOF_TIMESTAMPING_SOFTWARE
) ||
2379 (hwtstamps
->hwtstamp
&&
2380 (sk
->sk_tsflags
& SOF_TIMESTAMPING_RAW_HARDWARE
)))
2381 __sock_recv_timestamp(msg
, sk
, skb
);
2383 sock_write_timestamp(sk
, kt
);
2385 if (sock_flag(sk
, SOCK_WIFI_STATUS
) && skb
->wifi_acked_valid
)
2386 __sock_recv_wifi_status(msg
, sk
, skb
);
2389 void __sock_recv_ts_and_drops(struct msghdr
*msg
, struct sock
*sk
,
2390 struct sk_buff
*skb
);
2392 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2393 static inline void sock_recv_ts_and_drops(struct msghdr
*msg
, struct sock
*sk
,
2394 struct sk_buff
*skb
)
2396 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2397 (1UL << SOCK_RCVTSTAMP))
2398 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2399 SOF_TIMESTAMPING_RAW_HARDWARE)
2401 if (sk
->sk_flags
& FLAGS_TS_OR_DROPS
|| sk
->sk_tsflags
& TSFLAGS_ANY
)
2402 __sock_recv_ts_and_drops(msg
, sk
, skb
);
2403 else if (unlikely(sock_flag(sk
, SOCK_TIMESTAMP
)))
2404 sock_write_timestamp(sk
, skb
->tstamp
);
2405 else if (unlikely(sk
->sk_stamp
== SK_DEFAULT_STAMP
))
2406 sock_write_timestamp(sk
, 0);
2409 void __sock_tx_timestamp(__u16 tsflags
, __u8
*tx_flags
);
2412 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2413 * @sk: socket sending this packet
2414 * @tsflags: timestamping flags to use
2415 * @tx_flags: completed with instructions for time stamping
2416 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2418 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2420 static inline void _sock_tx_timestamp(struct sock
*sk
, __u16 tsflags
,
2421 __u8
*tx_flags
, __u32
*tskey
)
2423 if (unlikely(tsflags
)) {
2424 __sock_tx_timestamp(tsflags
, tx_flags
);
2425 if (tsflags
& SOF_TIMESTAMPING_OPT_ID
&& tskey
&&
2426 tsflags
& SOF_TIMESTAMPING_TX_RECORD_MASK
)
2427 *tskey
= sk
->sk_tskey
++;
2429 if (unlikely(sock_flag(sk
, SOCK_WIFI_STATUS
)))
2430 *tx_flags
|= SKBTX_WIFI_STATUS
;
2433 static inline void sock_tx_timestamp(struct sock
*sk
, __u16 tsflags
,
2436 _sock_tx_timestamp(sk
, tsflags
, tx_flags
, NULL
);
2439 static inline void skb_setup_tx_timestamp(struct sk_buff
*skb
, __u16 tsflags
)
2441 _sock_tx_timestamp(skb
->sk
, tsflags
, &skb_shinfo(skb
)->tx_flags
,
2442 &skb_shinfo(skb
)->tskey
);
2446 * sk_eat_skb - Release a skb if it is no longer needed
2447 * @sk: socket to eat this skb from
2448 * @skb: socket buffer to eat
2450 * This routine must be called with interrupts disabled or with the socket
2451 * locked so that the sk_buff queue operation is ok.
2453 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key
);
2454 static inline void sk_eat_skb(struct sock
*sk
, struct sk_buff
*skb
)
2456 __skb_unlink(skb
, &sk
->sk_receive_queue
);
2457 if (static_branch_unlikely(&tcp_rx_skb_cache_key
) &&
2458 !sk
->sk_rx_skb_cache
) {
2459 sk
->sk_rx_skb_cache
= skb
;
2467 struct net
*sock_net(const struct sock
*sk
)
2469 return read_pnet(&sk
->sk_net
);
2473 void sock_net_set(struct sock
*sk
, struct net
*net
)
2475 write_pnet(&sk
->sk_net
, net
);
2478 static inline struct sock
*skb_steal_sock(struct sk_buff
*skb
)
2481 struct sock
*sk
= skb
->sk
;
2483 skb
->destructor
= NULL
;
2490 /* This helper checks if a socket is a full socket,
2491 * ie _not_ a timewait or request socket.
2493 static inline bool sk_fullsock(const struct sock
*sk
)
2495 return (1 << sk
->sk_state
) & ~(TCPF_TIME_WAIT
| TCPF_NEW_SYN_RECV
);
2498 /* Checks if this SKB belongs to an HW offloaded socket
2499 * and whether any SW fallbacks are required based on dev.
2500 * Check decrypted mark in case skb_orphan() cleared socket.
2502 static inline struct sk_buff
*sk_validate_xmit_skb(struct sk_buff
*skb
,
2503 struct net_device
*dev
)
2505 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2506 struct sock
*sk
= skb
->sk
;
2508 if (sk
&& sk_fullsock(sk
) && sk
->sk_validate_xmit_skb
) {
2509 skb
= sk
->sk_validate_xmit_skb(sk
, dev
, skb
);
2510 #ifdef CONFIG_TLS_DEVICE
2511 } else if (unlikely(skb
->decrypted
)) {
2512 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2522 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2523 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2525 static inline bool sk_listener(const struct sock
*sk
)
2527 return (1 << sk
->sk_state
) & (TCPF_LISTEN
| TCPF_NEW_SYN_RECV
);
2530 void sock_enable_timestamp(struct sock
*sk
, enum sock_flags flag
);
2531 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
, int level
,
2534 bool sk_ns_capable(const struct sock
*sk
,
2535 struct user_namespace
*user_ns
, int cap
);
2536 bool sk_capable(const struct sock
*sk
, int cap
);
2537 bool sk_net_capable(const struct sock
*sk
, int cap
);
2539 void sk_get_meminfo(const struct sock
*sk
, u32
*meminfo
);
2541 /* Take into consideration the size of the struct sk_buff overhead in the
2542 * determination of these values, since that is non-constant across
2543 * platforms. This makes socket queueing behavior and performance
2544 * not depend upon such differences.
2546 #define _SK_MEM_PACKETS 256
2547 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2548 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2549 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2551 extern __u32 sysctl_wmem_max
;
2552 extern __u32 sysctl_rmem_max
;
2554 extern int sysctl_tstamp_allow_data
;
2555 extern int sysctl_optmem_max
;
2557 extern __u32 sysctl_wmem_default
;
2558 extern __u32 sysctl_rmem_default
;
2560 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key
);
2562 static inline int sk_get_wmem0(const struct sock
*sk
, const struct proto
*proto
)
2564 /* Does this proto have per netns sysctl_wmem ? */
2565 if (proto
->sysctl_wmem_offset
)
2566 return *(int *)((void *)sock_net(sk
) + proto
->sysctl_wmem_offset
);
2568 return *proto
->sysctl_wmem
;
2571 static inline int sk_get_rmem0(const struct sock
*sk
, const struct proto
*proto
)
2573 /* Does this proto have per netns sysctl_rmem ? */
2574 if (proto
->sysctl_rmem_offset
)
2575 return *(int *)((void *)sock_net(sk
) + proto
->sysctl_rmem_offset
);
2577 return *proto
->sysctl_rmem
;
2580 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2581 * Some wifi drivers need to tweak it to get more chunks.
2582 * They can use this helper from their ndo_start_xmit()
2584 static inline void sk_pacing_shift_update(struct sock
*sk
, int val
)
2586 if (!sk
|| !sk_fullsock(sk
) || sk
->sk_pacing_shift
== val
)
2588 sk
->sk_pacing_shift
= val
;
2591 /* if a socket is bound to a device, check that the given device
2592 * index is either the same or that the socket is bound to an L3
2593 * master device and the given device index is also enslaved to
2596 static inline bool sk_dev_equal_l3scope(struct sock
*sk
, int dif
)
2600 if (!sk
->sk_bound_dev_if
|| sk
->sk_bound_dev_if
== dif
)
2603 mdif
= l3mdev_master_ifindex_by_index(sock_net(sk
), dif
);
2604 if (mdif
&& mdif
== sk
->sk_bound_dev_if
)
2610 #endif /* _SOCK_H */