2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the AF_INET socket handler.
8 * Version: @(#)sock.h 1.0.4 05/13/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche <flla@stud.uni-sb.de>
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
20 * Alan Cox : Added a zapped field for tcp to note
21 * a socket is reset and must stay shut up
22 * Alan Cox : New fields for options
23 * Pauline Middelink : identd support
24 * Alan Cox : Eliminate low level recv/recvfrom
25 * David S. Miller : New socket lookup architecture.
26 * Steve Whitehouse: Default routines for sock_ops
27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
28 * protinfo be just a void pointer, as the
29 * protocol specific parts were moved to
30 * respective headers and ipv4/v6, etc now
31 * use private slabcaches for its socks
32 * Pedro Hortas : New flags field for socket options
35 * This program is free software; you can redistribute it and/or
36 * modify it under the terms of the GNU General Public License
37 * as published by the Free Software Foundation; either version
38 * 2 of the License, or (at your option) any later version.
43 #include <linux/hardirq.h>
44 #include <linux/kernel.h>
45 #include <linux/list.h>
46 #include <linux/list_nulls.h>
47 #include <linux/timer.h>
48 #include <linux/cache.h>
49 #include <linux/bitops.h>
50 #include <linux/lockdep.h>
51 #include <linux/netdevice.h>
52 #include <linux/skbuff.h> /* struct sk_buff */
54 #include <linux/security.h>
55 #include <linux/slab.h>
56 #include <linux/uaccess.h>
57 #include <linux/page_counter.h>
58 #include <linux/memcontrol.h>
59 #include <linux/static_key.h>
60 #include <linux/sched.h>
61 #include <linux/wait.h>
62 #include <linux/cgroup-defs.h>
63 #include <linux/rbtree.h>
64 #include <linux/filter.h>
65 #include <linux/rculist_nulls.h>
66 #include <linux/poll.h>
68 #include <linux/atomic.h>
69 #include <linux/refcount.h>
71 #include <net/checksum.h>
72 #include <net/tcp_states.h>
73 #include <linux/net_tstamp.h>
77 * This structure really needs to be cleaned up.
78 * Most of it is for TCP, and not used by any of
79 * the other protocols.
82 /* Define this to get the SOCK_DBG debugging facility. */
83 #define SOCK_DEBUGGING
85 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
86 printk(KERN_DEBUG msg); } while (0)
88 /* Validate arguments and do nothing */
89 static inline __printf(2, 3)
90 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
95 /* This is the per-socket lock. The spinlock provides a synchronization
96 * between user contexts and software interrupt processing, whereas the
97 * mini-semaphore synchronizes multiple users amongst themselves.
102 wait_queue_head_t wq;
104 * We express the mutex-alike socket_lock semantics
105 * to the lock validator by explicitly managing
106 * the slock as a lock variant (in addition to
109 #ifdef CONFIG_DEBUG_LOCK_ALLOC
110 struct lockdep_map dep_map;
118 typedef __u32 __bitwise __portpair;
119 typedef __u64 __bitwise __addrpair;
122 * struct sock_common - minimal network layer representation of sockets
123 * @skc_daddr: Foreign IPv4 addr
124 * @skc_rcv_saddr: Bound local IPv4 addr
125 * @skc_hash: hash value used with various protocol lookup tables
126 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
127 * @skc_dport: placeholder for inet_dport/tw_dport
128 * @skc_num: placeholder for inet_num/tw_num
129 * @skc_family: network address family
130 * @skc_state: Connection state
131 * @skc_reuse: %SO_REUSEADDR setting
132 * @skc_reuseport: %SO_REUSEPORT setting
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_node: main hash linkage for various protocol lookup tables
139 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
140 * @skc_tx_queue_mapping: tx queue number for this connection
141 * @skc_flags: place holder for sk_flags
142 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
143 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
144 * @skc_incoming_cpu: record/match cpu processing incoming packets
145 * @skc_refcnt: reference count
147 * This is the minimal network layer representation of sockets, the header
148 * for struct sock and struct inet_timewait_sock.
151 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
152 * address on 64bit arches : cf INET_MATCH()
155 __addrpair skc_addrpair;
158 __be32 skc_rcv_saddr;
162 unsigned int skc_hash;
163 __u16 skc_u16hashes[2];
165 /* skc_dport && skc_num must be grouped as well */
167 __portpair skc_portpair;
174 unsigned short skc_family;
175 volatile unsigned char skc_state;
176 unsigned char skc_reuse:4;
177 unsigned char skc_reuseport:1;
178 unsigned char skc_ipv6only:1;
179 unsigned char skc_net_refcnt:1;
180 int skc_bound_dev_if;
182 struct hlist_node skc_bind_node;
183 struct hlist_node skc_portaddr_node;
185 struct proto *skc_prot;
186 possible_net_t skc_net;
188 #if IS_ENABLED(CONFIG_IPV6)
189 struct in6_addr skc_v6_daddr;
190 struct in6_addr skc_v6_rcv_saddr;
193 atomic64_t skc_cookie;
195 /* following fields are padding to force
196 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
197 * assuming IPV6 is enabled. We use this padding differently
198 * for different kind of 'sockets'
201 unsigned long skc_flags;
202 struct sock *skc_listener; /* request_sock */
203 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
206 * fields between dontcopy_begin/dontcopy_end
207 * are not copied in sock_copy()
210 int skc_dontcopy_begin[0];
213 struct hlist_node skc_node;
214 struct hlist_nulls_node skc_nulls_node;
216 int skc_tx_queue_mapping;
218 int skc_incoming_cpu;
220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
223 refcount_t skc_refcnt;
225 int skc_dontcopy_end[0];
228 u32 skc_window_clamp;
229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
235 * struct sock - network layer representation of sockets
236 * @__sk_common: shared layout with inet_timewait_sock
237 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
238 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
239 * @sk_lock: synchronizer
240 * @sk_kern_sock: True if sock is using kernel lock classes
241 * @sk_rcvbuf: size of receive buffer in bytes
242 * @sk_wq: sock wait queue and async head
243 * @sk_rx_dst: receive input route used by early demux
244 * @sk_dst_cache: destination cache
245 * @sk_dst_pending_confirm: need to confirm neighbour
246 * @sk_policy: flow policy
247 * @sk_receive_queue: incoming packets
248 * @sk_wmem_alloc: transmit queue bytes committed
249 * @sk_tsq_flags: TCP Small Queues flags
250 * @sk_write_queue: Packet sending queue
251 * @sk_omem_alloc: "o" is "option" or "other"
252 * @sk_wmem_queued: persistent queue size
253 * @sk_forward_alloc: space allocated forward
254 * @sk_napi_id: id of the last napi context to receive data for sk
255 * @sk_ll_usec: usecs to busypoll when there is no data
256 * @sk_allocation: allocation mode
257 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
258 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
259 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
260 * @sk_sndbuf: size of send buffer in bytes
261 * @__sk_flags_offset: empty field used to determine location of bitfield
262 * @sk_padding: unused element for alignment
263 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
264 * @sk_no_check_rx: allow zero checksum in RX packets
265 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
266 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
267 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
268 * @sk_gso_max_size: Maximum GSO segment size to build
269 * @sk_gso_max_segs: Maximum number of GSO segments
270 * @sk_pacing_shift: scaling factor for TCP Small Queues
271 * @sk_lingertime: %SO_LINGER l_linger setting
272 * @sk_backlog: always used with the per-socket spinlock held
273 * @sk_callback_lock: used with the callbacks in the end of this struct
274 * @sk_error_queue: rarely used
275 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
276 * IPV6_ADDRFORM for instance)
277 * @sk_err: last error
278 * @sk_err_soft: errors that don't cause failure but are the cause of a
279 * persistent failure not just 'timed out'
280 * @sk_drops: raw/udp drops counter
281 * @sk_ack_backlog: current listen backlog
282 * @sk_max_ack_backlog: listen backlog set in listen()
283 * @sk_uid: user id of owner
284 * @sk_priority: %SO_PRIORITY setting
285 * @sk_type: socket type (%SOCK_STREAM, etc)
286 * @sk_protocol: which protocol this socket belongs in this network family
287 * @sk_peer_pid: &struct pid for this socket's peer
288 * @sk_peer_cred: %SO_PEERCRED setting
289 * @sk_rcvlowat: %SO_RCVLOWAT setting
290 * @sk_rcvtimeo: %SO_RCVTIMEO setting
291 * @sk_sndtimeo: %SO_SNDTIMEO setting
292 * @sk_txhash: computed flow hash for use on transmit
293 * @sk_filter: socket filtering instructions
294 * @sk_timer: sock cleanup timer
295 * @sk_stamp: time stamp of last packet received
296 * @sk_tsflags: SO_TIMESTAMPING socket options
297 * @sk_tskey: counter to disambiguate concurrent tstamp requests
298 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
299 * @sk_socket: Identd and reporting IO signals
300 * @sk_user_data: RPC layer private data
301 * @sk_frag: cached page frag
302 * @sk_peek_off: current peek_offset value
303 * @sk_send_head: front of stuff to transmit
304 * @sk_security: used by security modules
305 * @sk_mark: generic packet mark
306 * @sk_cgrp_data: cgroup data for this cgroup
307 * @sk_memcg: this socket's memory cgroup association
308 * @sk_write_pending: a write to stream socket waits to start
309 * @sk_state_change: callback to indicate change in the state of the sock
310 * @sk_data_ready: callback to indicate there is data to be processed
311 * @sk_write_space: callback to indicate there is bf sending space available
312 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
313 * @sk_backlog_rcv: callback to process the backlog
314 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
315 * @sk_reuseport_cb: reuseport group container
316 * @sk_rcu: used during RCU grace period
320 * Now struct inet_timewait_sock also uses sock_common, so please just
321 * don't add nothing before this first member (__sk_common) --acme
323 struct sock_common __sk_common;
324 #define sk_node __sk_common.skc_node
325 #define sk_nulls_node __sk_common.skc_nulls_node
326 #define sk_refcnt __sk_common.skc_refcnt
327 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
329 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
330 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
331 #define sk_hash __sk_common.skc_hash
332 #define sk_portpair __sk_common.skc_portpair
333 #define sk_num __sk_common.skc_num
334 #define sk_dport __sk_common.skc_dport
335 #define sk_addrpair __sk_common.skc_addrpair
336 #define sk_daddr __sk_common.skc_daddr
337 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
338 #define sk_family __sk_common.skc_family
339 #define sk_state __sk_common.skc_state
340 #define sk_reuse __sk_common.skc_reuse
341 #define sk_reuseport __sk_common.skc_reuseport
342 #define sk_ipv6only __sk_common.skc_ipv6only
343 #define sk_net_refcnt __sk_common.skc_net_refcnt
344 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
345 #define sk_bind_node __sk_common.skc_bind_node
346 #define sk_prot __sk_common.skc_prot
347 #define sk_net __sk_common.skc_net
348 #define sk_v6_daddr __sk_common.skc_v6_daddr
349 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
350 #define sk_cookie __sk_common.skc_cookie
351 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
352 #define sk_flags __sk_common.skc_flags
353 #define sk_rxhash __sk_common.skc_rxhash
355 socket_lock_t sk_lock;
358 struct sk_buff_head sk_error_queue;
359 struct sk_buff_head sk_receive_queue;
361 * The backlog queue is special, it is always used with
362 * the per-socket spinlock held and requires low latency
363 * access. Therefore we special case it's implementation.
364 * Note : rmem_alloc is in this structure to fill a hole
365 * on 64bit arches, not because its logically part of
371 struct sk_buff *head;
372 struct sk_buff *tail;
374 #define sk_rmem_alloc sk_backlog.rmem_alloc
376 int sk_forward_alloc;
377 #ifdef CONFIG_NET_RX_BUSY_POLL
378 unsigned int sk_ll_usec;
379 /* ===== mostly read cache line ===== */
380 unsigned int sk_napi_id;
384 struct sk_filter __rcu *sk_filter;
386 struct socket_wq __rcu *sk_wq;
387 struct socket_wq *sk_wq_raw;
390 struct xfrm_policy __rcu *sk_policy[2];
392 struct dst_entry *sk_rx_dst;
393 struct dst_entry __rcu *sk_dst_cache;
394 atomic_t sk_omem_alloc;
397 /* ===== cache line for TX ===== */
399 refcount_t sk_wmem_alloc;
400 unsigned long sk_tsq_flags;
402 struct sk_buff *sk_send_head;
403 struct rb_root tcp_rtx_queue;
405 struct sk_buff_head sk_write_queue;
407 int sk_write_pending;
408 __u32 sk_dst_pending_confirm;
409 u32 sk_pacing_status; /* see enum sk_pacing */
411 struct timer_list sk_timer;
414 u32 sk_pacing_rate; /* bytes per second */
415 u32 sk_max_pacing_rate;
416 struct page_frag sk_frag;
417 netdev_features_t sk_route_caps;
418 netdev_features_t sk_route_nocaps;
420 unsigned int sk_gso_max_size;
425 * Because of non atomicity rules, all
426 * changes are protected by socket lock.
428 unsigned int __sk_flags_offset[0];
429 #ifdef __BIG_ENDIAN_BITFIELD
430 #define SK_FL_PROTO_SHIFT 16
431 #define SK_FL_PROTO_MASK 0x00ff0000
433 #define SK_FL_TYPE_SHIFT 0
434 #define SK_FL_TYPE_MASK 0x0000ffff
436 #define SK_FL_PROTO_SHIFT 8
437 #define SK_FL_PROTO_MASK 0x0000ff00
439 #define SK_FL_TYPE_SHIFT 16
440 #define SK_FL_TYPE_MASK 0xffff0000
443 kmemcheck_bitfield_begin(flags);
444 unsigned int sk_padding : 1,
451 #define SK_PROTOCOL_MAX U8_MAX
452 kmemcheck_bitfield_end(flags);
456 unsigned long sk_lingertime;
457 struct proto *sk_prot_creator;
458 rwlock_t sk_callback_lock;
462 u32 sk_max_ack_backlog;
464 struct pid *sk_peer_pid;
465 const struct cred *sk_peer_cred;
472 struct socket *sk_socket;
474 #ifdef CONFIG_SECURITY
477 struct sock_cgroup_data sk_cgrp_data;
478 struct mem_cgroup *sk_memcg;
479 void (*sk_state_change)(struct sock *sk);
480 void (*sk_data_ready)(struct sock *sk);
481 void (*sk_write_space)(struct sock *sk);
482 void (*sk_error_report)(struct sock *sk);
483 int (*sk_backlog_rcv)(struct sock *sk,
484 struct sk_buff *skb);
485 void (*sk_destruct)(struct sock *sk);
486 struct sock_reuseport __rcu *sk_reuseport_cb;
487 struct rcu_head sk_rcu;
492 SK_PACING_NEEDED = 1,
496 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
498 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
499 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
502 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
503 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
504 * on a socket means that the socket will reuse everybody else's port
505 * without looking at the other's sk_reuse value.
508 #define SK_NO_REUSE 0
509 #define SK_CAN_REUSE 1
510 #define SK_FORCE_REUSE 2
512 int sk_set_peek_off(struct sock *sk, int val);
514 static inline int sk_peek_offset(struct sock *sk, int flags)
516 if (unlikely(flags & MSG_PEEK)) {
517 return READ_ONCE(sk->sk_peek_off);
523 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
525 s32 off = READ_ONCE(sk->sk_peek_off);
527 if (unlikely(off >= 0)) {
528 off = max_t(s32, off - val, 0);
529 WRITE_ONCE(sk->sk_peek_off, off);
533 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
535 sk_peek_offset_bwd(sk, -val);
539 * Hashed lists helper routines
541 static inline struct sock *sk_entry(const struct hlist_node *node)
543 return hlist_entry(node, struct sock, sk_node);
546 static inline struct sock *__sk_head(const struct hlist_head *head)
548 return hlist_entry(head->first, struct sock, sk_node);
551 static inline struct sock *sk_head(const struct hlist_head *head)
553 return hlist_empty(head) ? NULL : __sk_head(head);
556 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
558 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
561 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
563 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
566 static inline struct sock *sk_next(const struct sock *sk)
568 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
571 static inline struct sock *sk_nulls_next(const struct sock *sk)
573 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
574 hlist_nulls_entry(sk->sk_nulls_node.next,
575 struct sock, sk_nulls_node) :
579 static inline bool sk_unhashed(const struct sock *sk)
581 return hlist_unhashed(&sk->sk_node);
584 static inline bool sk_hashed(const struct sock *sk)
586 return !sk_unhashed(sk);
589 static inline void sk_node_init(struct hlist_node *node)
594 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
599 static inline void __sk_del_node(struct sock *sk)
601 __hlist_del(&sk->sk_node);
604 /* NB: equivalent to hlist_del_init_rcu */
605 static inline bool __sk_del_node_init(struct sock *sk)
609 sk_node_init(&sk->sk_node);
615 /* Grab socket reference count. This operation is valid only
616 when sk is ALREADY grabbed f.e. it is found in hash table
617 or a list and the lookup is made under lock preventing hash table
621 static __always_inline void sock_hold(struct sock *sk)
623 refcount_inc(&sk->sk_refcnt);
626 /* Ungrab socket in the context, which assumes that socket refcnt
627 cannot hit zero, f.e. it is true in context of any socketcall.
629 static __always_inline void __sock_put(struct sock *sk)
631 refcount_dec(&sk->sk_refcnt);
634 static inline bool sk_del_node_init(struct sock *sk)
636 bool rc = __sk_del_node_init(sk);
639 /* paranoid for a while -acme */
640 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
645 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
647 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
650 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
656 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
658 bool rc = __sk_nulls_del_node_init_rcu(sk);
661 /* paranoid for a while -acme */
662 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
668 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
670 hlist_add_head(&sk->sk_node, list);
673 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
676 __sk_add_node(sk, list);
679 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
682 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
683 sk->sk_family == AF_INET6)
684 hlist_add_tail_rcu(&sk->sk_node, list);
686 hlist_add_head_rcu(&sk->sk_node, list);
689 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
691 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
692 sk->sk_family == AF_INET6)
693 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
695 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
698 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
701 __sk_nulls_add_node_rcu(sk, list);
704 static inline void __sk_del_bind_node(struct sock *sk)
706 __hlist_del(&sk->sk_bind_node);
709 static inline void sk_add_bind_node(struct sock *sk,
710 struct hlist_head *list)
712 hlist_add_head(&sk->sk_bind_node, list);
715 #define sk_for_each(__sk, list) \
716 hlist_for_each_entry(__sk, list, sk_node)
717 #define sk_for_each_rcu(__sk, list) \
718 hlist_for_each_entry_rcu(__sk, list, sk_node)
719 #define sk_nulls_for_each(__sk, node, list) \
720 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
721 #define sk_nulls_for_each_rcu(__sk, node, list) \
722 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
723 #define sk_for_each_from(__sk) \
724 hlist_for_each_entry_from(__sk, sk_node)
725 #define sk_nulls_for_each_from(__sk, node) \
726 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
727 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
728 #define sk_for_each_safe(__sk, tmp, list) \
729 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
730 #define sk_for_each_bound(__sk, list) \
731 hlist_for_each_entry(__sk, list, sk_bind_node)
734 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
735 * @tpos: the type * to use as a loop cursor.
736 * @pos: the &struct hlist_node to use as a loop cursor.
737 * @head: the head for your list.
738 * @offset: offset of hlist_node within the struct.
741 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
742 for (pos = rcu_dereference(hlist_first_rcu(head)); \
744 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
745 pos = rcu_dereference(hlist_next_rcu(pos)))
747 static inline struct user_namespace *sk_user_ns(struct sock *sk)
749 /* Careful only use this in a context where these parameters
750 * can not change and must all be valid, such as recvmsg from
753 return sk->sk_socket->file->f_cred->user_ns;
767 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
768 SOCK_DBG, /* %SO_DEBUG setting */
769 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
770 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
771 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
772 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
773 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
774 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
775 SOCK_FASYNC, /* fasync() active */
777 SOCK_ZEROCOPY, /* buffers from userspace */
778 SOCK_WIFI_STATUS, /* push wifi status to userspace */
779 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
780 * Will use last 4 bytes of packet sent from
781 * user-space instead.
783 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
784 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
785 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
788 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
790 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
792 nsk->sk_flags = osk->sk_flags;
795 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
797 __set_bit(flag, &sk->sk_flags);
800 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
802 __clear_bit(flag, &sk->sk_flags);
805 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
807 return test_bit(flag, &sk->sk_flags);
811 extern struct static_key memalloc_socks;
812 static inline int sk_memalloc_socks(void)
814 return static_key_false(&memalloc_socks);
818 static inline int sk_memalloc_socks(void)
825 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
827 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
830 static inline void sk_acceptq_removed(struct sock *sk)
832 sk->sk_ack_backlog--;
835 static inline void sk_acceptq_added(struct sock *sk)
837 sk->sk_ack_backlog++;
840 static inline bool sk_acceptq_is_full(const struct sock *sk)
842 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
846 * Compute minimal free write space needed to queue new packets.
848 static inline int sk_stream_min_wspace(const struct sock *sk)
850 return sk->sk_wmem_queued >> 1;
853 static inline int sk_stream_wspace(const struct sock *sk)
855 return sk->sk_sndbuf - sk->sk_wmem_queued;
858 void sk_stream_write_space(struct sock *sk);
860 /* OOB backlog add */
861 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
863 /* dont let skb dst not refcounted, we are going to leave rcu lock */
866 if (!sk->sk_backlog.tail)
867 sk->sk_backlog.head = skb;
869 sk->sk_backlog.tail->next = skb;
871 sk->sk_backlog.tail = skb;
876 * Take into account size of receive queue and backlog queue
877 * Do not take into account this skb truesize,
878 * to allow even a single big packet to come.
880 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
882 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
884 return qsize > limit;
887 /* The per-socket spinlock must be held here. */
888 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
891 if (sk_rcvqueues_full(sk, limit))
895 * If the skb was allocated from pfmemalloc reserves, only
896 * allow SOCK_MEMALLOC sockets to use it as this socket is
897 * helping free memory
899 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
902 __sk_add_backlog(sk, skb);
903 sk->sk_backlog.len += skb->truesize;
907 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
909 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
911 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
912 return __sk_backlog_rcv(sk, skb);
914 return sk->sk_backlog_rcv(sk, skb);
917 static inline void sk_incoming_cpu_update(struct sock *sk)
919 int cpu = raw_smp_processor_id();
921 if (unlikely(sk->sk_incoming_cpu != cpu))
922 sk->sk_incoming_cpu = cpu;
925 static inline void sock_rps_record_flow_hash(__u32 hash)
928 struct rps_sock_flow_table *sock_flow_table;
931 sock_flow_table = rcu_dereference(rps_sock_flow_table);
932 rps_record_sock_flow(sock_flow_table, hash);
937 static inline void sock_rps_record_flow(const struct sock *sk)
940 if (static_key_false(&rfs_needed)) {
941 /* Reading sk->sk_rxhash might incur an expensive cache line
944 * TCP_ESTABLISHED does cover almost all states where RFS
945 * might be useful, and is cheaper [1] than testing :
946 * IPv4: inet_sk(sk)->inet_daddr
947 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
948 * OR an additional socket flag
949 * [1] : sk_state and sk_prot are in the same cache line.
951 if (sk->sk_state == TCP_ESTABLISHED)
952 sock_rps_record_flow_hash(sk->sk_rxhash);
957 static inline void sock_rps_save_rxhash(struct sock *sk,
958 const struct sk_buff *skb)
961 if (unlikely(sk->sk_rxhash != skb->hash))
962 sk->sk_rxhash = skb->hash;
966 static inline void sock_rps_reset_rxhash(struct sock *sk)
973 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
975 release_sock(__sk); \
976 __rc = __condition; \
978 *(__timeo) = wait_woken(__wait, \
979 TASK_INTERRUPTIBLE, \
982 sched_annotate_sleep(); \
984 __rc = __condition; \
988 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
989 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
990 void sk_stream_wait_close(struct sock *sk, long timeo_p);
991 int sk_stream_error(struct sock *sk, int flags, int err);
992 void sk_stream_kill_queues(struct sock *sk);
993 void sk_set_memalloc(struct sock *sk);
994 void sk_clear_memalloc(struct sock *sk);
996 void __sk_flush_backlog(struct sock *sk);
998 static inline bool sk_flush_backlog(struct sock *sk)
1000 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1001 __sk_flush_backlog(sk);
1007 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1009 struct request_sock_ops;
1010 struct timewait_sock_ops;
1011 struct inet_hashinfo;
1012 struct raw_hashinfo;
1013 struct smc_hashinfo;
1017 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1018 * un-modified. Special care is taken when initializing object to zero.
1020 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1022 if (offsetof(struct sock, sk_node.next) != 0)
1023 memset(sk, 0, offsetof(struct sock, sk_node.next));
1024 memset(&sk->sk_node.pprev, 0,
1025 size - offsetof(struct sock, sk_node.pprev));
1028 /* Networking protocol blocks we attach to sockets.
1029 * socket layer -> transport layer interface
1032 void (*close)(struct sock *sk,
1034 int (*connect)(struct sock *sk,
1035 struct sockaddr *uaddr,
1037 int (*disconnect)(struct sock *sk, int flags);
1039 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1042 int (*ioctl)(struct sock *sk, int cmd,
1044 int (*init)(struct sock *sk);
1045 void (*destroy)(struct sock *sk);
1046 void (*shutdown)(struct sock *sk, int how);
1047 int (*setsockopt)(struct sock *sk, int level,
1048 int optname, char __user *optval,
1049 unsigned int optlen);
1050 int (*getsockopt)(struct sock *sk, int level,
1051 int optname, char __user *optval,
1052 int __user *option);
1053 void (*keepalive)(struct sock *sk, int valbool);
1054 #ifdef CONFIG_COMPAT
1055 int (*compat_setsockopt)(struct sock *sk,
1057 int optname, char __user *optval,
1058 unsigned int optlen);
1059 int (*compat_getsockopt)(struct sock *sk,
1061 int optname, char __user *optval,
1062 int __user *option);
1063 int (*compat_ioctl)(struct sock *sk,
1064 unsigned int cmd, unsigned long arg);
1066 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1068 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1069 size_t len, int noblock, int flags,
1071 int (*sendpage)(struct sock *sk, struct page *page,
1072 int offset, size_t size, int flags);
1073 int (*bind)(struct sock *sk,
1074 struct sockaddr *uaddr, int addr_len);
1076 int (*backlog_rcv) (struct sock *sk,
1077 struct sk_buff *skb);
1079 void (*release_cb)(struct sock *sk);
1081 /* Keeping track of sk's, looking them up, and port selection methods. */
1082 int (*hash)(struct sock *sk);
1083 void (*unhash)(struct sock *sk);
1084 void (*rehash)(struct sock *sk);
1085 int (*get_port)(struct sock *sk, unsigned short snum);
1087 /* Keeping track of sockets in use */
1088 #ifdef CONFIG_PROC_FS
1089 unsigned int inuse_idx;
1092 bool (*stream_memory_free)(const struct sock *sk);
1093 /* Memory pressure */
1094 void (*enter_memory_pressure)(struct sock *sk);
1095 void (*leave_memory_pressure)(struct sock *sk);
1096 atomic_long_t *memory_allocated; /* Current allocated memory. */
1097 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1099 * Pressure flag: try to collapse.
1100 * Technical note: it is used by multiple contexts non atomically.
1101 * All the __sk_mem_schedule() is of this nature: accounting
1102 * is strict, actions are advisory and have some latency.
1104 unsigned long *memory_pressure;
1109 u32 sysctl_wmem_offset;
1110 u32 sysctl_rmem_offset;
1115 struct kmem_cache *slab;
1116 unsigned int obj_size;
1119 struct percpu_counter *orphan_count;
1121 struct request_sock_ops *rsk_prot;
1122 struct timewait_sock_ops *twsk_prot;
1125 struct inet_hashinfo *hashinfo;
1126 struct udp_table *udp_table;
1127 struct raw_hashinfo *raw_hash;
1128 struct smc_hashinfo *smc_hash;
1131 struct module *owner;
1135 struct list_head node;
1136 #ifdef SOCK_REFCNT_DEBUG
1139 int (*diag_destroy)(struct sock *sk, int err);
1140 } __randomize_layout;
1142 int proto_register(struct proto *prot, int alloc_slab);
1143 void proto_unregister(struct proto *prot);
1145 #ifdef SOCK_REFCNT_DEBUG
1146 static inline void sk_refcnt_debug_inc(struct sock *sk)
1148 atomic_inc(&sk->sk_prot->socks);
1151 static inline void sk_refcnt_debug_dec(struct sock *sk)
1153 atomic_dec(&sk->sk_prot->socks);
1154 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1155 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1158 static inline void sk_refcnt_debug_release(const struct sock *sk)
1160 if (refcount_read(&sk->sk_refcnt) != 1)
1161 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1162 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1164 #else /* SOCK_REFCNT_DEBUG */
1165 #define sk_refcnt_debug_inc(sk) do { } while (0)
1166 #define sk_refcnt_debug_dec(sk) do { } while (0)
1167 #define sk_refcnt_debug_release(sk) do { } while (0)
1168 #endif /* SOCK_REFCNT_DEBUG */
1170 static inline bool sk_stream_memory_free(const struct sock *sk)
1172 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1175 return sk->sk_prot->stream_memory_free ?
1176 sk->sk_prot->stream_memory_free(sk) : true;
1179 static inline bool sk_stream_is_writeable(const struct sock *sk)
1181 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1182 sk_stream_memory_free(sk);
1185 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1186 struct cgroup *ancestor)
1188 #ifdef CONFIG_SOCK_CGROUP_DATA
1189 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1196 static inline bool sk_has_memory_pressure(const struct sock *sk)
1198 return sk->sk_prot->memory_pressure != NULL;
1201 static inline bool sk_under_memory_pressure(const struct sock *sk)
1203 if (!sk->sk_prot->memory_pressure)
1206 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1207 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1210 return !!*sk->sk_prot->memory_pressure;
1214 sk_memory_allocated(const struct sock *sk)
1216 return atomic_long_read(sk->sk_prot->memory_allocated);
1220 sk_memory_allocated_add(struct sock *sk, int amt)
1222 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1226 sk_memory_allocated_sub(struct sock *sk, int amt)
1228 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1231 static inline void sk_sockets_allocated_dec(struct sock *sk)
1233 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1236 static inline void sk_sockets_allocated_inc(struct sock *sk)
1238 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1242 sk_sockets_allocated_read_positive(struct sock *sk)
1244 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1248 proto_sockets_allocated_sum_positive(struct proto *prot)
1250 return percpu_counter_sum_positive(prot->sockets_allocated);
1254 proto_memory_allocated(struct proto *prot)
1256 return atomic_long_read(prot->memory_allocated);
1260 proto_memory_pressure(struct proto *prot)
1262 if (!prot->memory_pressure)
1264 return !!*prot->memory_pressure;
1268 #ifdef CONFIG_PROC_FS
1269 /* Called with local bh disabled */
1270 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1271 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1273 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1280 /* With per-bucket locks this operation is not-atomic, so that
1281 * this version is not worse.
1283 static inline int __sk_prot_rehash(struct sock *sk)
1285 sk->sk_prot->unhash(sk);
1286 return sk->sk_prot->hash(sk);
1289 /* About 10 seconds */
1290 #define SOCK_DESTROY_TIME (10*HZ)
1292 /* Sockets 0-1023 can't be bound to unless you are superuser */
1293 #define PROT_SOCK 1024
1295 #define SHUTDOWN_MASK 3
1296 #define RCV_SHUTDOWN 1
1297 #define SEND_SHUTDOWN 2
1299 #define SOCK_SNDBUF_LOCK 1
1300 #define SOCK_RCVBUF_LOCK 2
1301 #define SOCK_BINDADDR_LOCK 4
1302 #define SOCK_BINDPORT_LOCK 8
1304 struct socket_alloc {
1305 struct socket socket;
1306 struct inode vfs_inode;
1309 static inline struct socket *SOCKET_I(struct inode *inode)
1311 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1314 static inline struct inode *SOCK_INODE(struct socket *socket)
1316 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1320 * Functions for memory accounting
1322 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1323 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1324 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1325 void __sk_mem_reclaim(struct sock *sk, int amount);
1327 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1328 * do not necessarily have 16x time more memory than 4KB ones.
1330 #define SK_MEM_QUANTUM 4096
1331 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1332 #define SK_MEM_SEND 0
1333 #define SK_MEM_RECV 1
1335 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1336 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1338 long val = sk->sk_prot->sysctl_mem[index];
1340 #if PAGE_SIZE > SK_MEM_QUANTUM
1341 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1342 #elif PAGE_SIZE < SK_MEM_QUANTUM
1343 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1348 static inline int sk_mem_pages(int amt)
1350 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1353 static inline bool sk_has_account(struct sock *sk)
1355 /* return true if protocol supports memory accounting */
1356 return !!sk->sk_prot->memory_allocated;
1359 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1361 if (!sk_has_account(sk))
1363 return size <= sk->sk_forward_alloc ||
1364 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1368 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1370 if (!sk_has_account(sk))
1372 return size<= sk->sk_forward_alloc ||
1373 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1374 skb_pfmemalloc(skb);
1377 static inline void sk_mem_reclaim(struct sock *sk)
1379 if (!sk_has_account(sk))
1381 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1382 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1385 static inline void sk_mem_reclaim_partial(struct sock *sk)
1387 if (!sk_has_account(sk))
1389 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1390 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1393 static inline void sk_mem_charge(struct sock *sk, int size)
1395 if (!sk_has_account(sk))
1397 sk->sk_forward_alloc -= size;
1400 static inline void sk_mem_uncharge(struct sock *sk, int size)
1402 if (!sk_has_account(sk))
1404 sk->sk_forward_alloc += size;
1406 /* Avoid a possible overflow.
1407 * TCP send queues can make this happen, if sk_mem_reclaim()
1408 * is not called and more than 2 GBytes are released at once.
1410 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1411 * no need to hold that much forward allocation anyway.
1413 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1414 __sk_mem_reclaim(sk, 1 << 20);
1417 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1419 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1420 sk->sk_wmem_queued -= skb->truesize;
1421 sk_mem_uncharge(sk, skb->truesize);
1425 static inline void sock_release_ownership(struct sock *sk)
1427 if (sk->sk_lock.owned) {
1428 sk->sk_lock.owned = 0;
1430 /* The sk_lock has mutex_unlock() semantics: */
1431 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1436 * Macro so as to not evaluate some arguments when
1437 * lockdep is not enabled.
1439 * Mark both the sk_lock and the sk_lock.slock as a
1440 * per-address-family lock class.
1442 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1444 sk->sk_lock.owned = 0; \
1445 init_waitqueue_head(&sk->sk_lock.wq); \
1446 spin_lock_init(&(sk)->sk_lock.slock); \
1447 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1448 sizeof((sk)->sk_lock)); \
1449 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1451 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1454 #ifdef CONFIG_LOCKDEP
1455 static inline bool lockdep_sock_is_held(const struct sock *csk)
1457 struct sock *sk = (struct sock *)csk;
1459 return lockdep_is_held(&sk->sk_lock) ||
1460 lockdep_is_held(&sk->sk_lock.slock);
1464 void lock_sock_nested(struct sock *sk, int subclass);
1466 static inline void lock_sock(struct sock *sk)
1468 lock_sock_nested(sk, 0);
1471 void release_sock(struct sock *sk);
1473 /* BH context may only use the following locking interface. */
1474 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1475 #define bh_lock_sock_nested(__sk) \
1476 spin_lock_nested(&((__sk)->sk_lock.slock), \
1477 SINGLE_DEPTH_NESTING)
1478 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1480 bool lock_sock_fast(struct sock *sk);
1482 * unlock_sock_fast - complement of lock_sock_fast
1486 * fast unlock socket for user context.
1487 * If slow mode is on, we call regular release_sock()
1489 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1494 spin_unlock_bh(&sk->sk_lock.slock);
1497 /* Used by processes to "lock" a socket state, so that
1498 * interrupts and bottom half handlers won't change it
1499 * from under us. It essentially blocks any incoming
1500 * packets, so that we won't get any new data or any
1501 * packets that change the state of the socket.
1503 * While locked, BH processing will add new packets to
1504 * the backlog queue. This queue is processed by the
1505 * owner of the socket lock right before it is released.
1507 * Since ~2.3.5 it is also exclusive sleep lock serializing
1508 * accesses from user process context.
1511 static inline void sock_owned_by_me(const struct sock *sk)
1513 #ifdef CONFIG_LOCKDEP
1514 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1518 static inline bool sock_owned_by_user(const struct sock *sk)
1520 sock_owned_by_me(sk);
1521 return sk->sk_lock.owned;
1524 /* no reclassification while locks are held */
1525 static inline bool sock_allow_reclassification(const struct sock *csk)
1527 struct sock *sk = (struct sock *)csk;
1529 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1532 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1533 struct proto *prot, int kern);
1534 void sk_free(struct sock *sk);
1535 void sk_destruct(struct sock *sk);
1536 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1537 void sk_free_unlock_clone(struct sock *sk);
1539 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1541 void __sock_wfree(struct sk_buff *skb);
1542 void sock_wfree(struct sk_buff *skb);
1543 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1545 void skb_orphan_partial(struct sk_buff *skb);
1546 void sock_rfree(struct sk_buff *skb);
1547 void sock_efree(struct sk_buff *skb);
1549 void sock_edemux(struct sk_buff *skb);
1551 #define sock_edemux sock_efree
1554 int sock_setsockopt(struct socket *sock, int level, int op,
1555 char __user *optval, unsigned int optlen);
1557 int sock_getsockopt(struct socket *sock, int level, int op,
1558 char __user *optval, int __user *optlen);
1559 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1560 int noblock, int *errcode);
1561 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1562 unsigned long data_len, int noblock,
1563 int *errcode, int max_page_order);
1564 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1565 void sock_kfree_s(struct sock *sk, void *mem, int size);
1566 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1567 void sk_send_sigurg(struct sock *sk);
1569 struct sockcm_cookie {
1574 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1575 struct sockcm_cookie *sockc);
1576 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1577 struct sockcm_cookie *sockc);
1580 * Functions to fill in entries in struct proto_ops when a protocol
1581 * does not implement a particular function.
1583 int sock_no_bind(struct socket *, struct sockaddr *, int);
1584 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1585 int sock_no_socketpair(struct socket *, struct socket *);
1586 int sock_no_accept(struct socket *, struct socket *, int, bool);
1587 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1588 unsigned int sock_no_poll(struct file *, struct socket *,
1589 struct poll_table_struct *);
1590 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1591 int sock_no_listen(struct socket *, int);
1592 int sock_no_shutdown(struct socket *, int);
1593 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1594 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1595 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1596 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1597 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1598 int sock_no_mmap(struct file *file, struct socket *sock,
1599 struct vm_area_struct *vma);
1600 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1601 size_t size, int flags);
1602 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1603 int offset, size_t size, int flags);
1606 * Functions to fill in entries in struct proto_ops when a protocol
1607 * uses the inet style.
1609 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1610 char __user *optval, int __user *optlen);
1611 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1613 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1614 char __user *optval, unsigned int optlen);
1615 int compat_sock_common_getsockopt(struct socket *sock, int level,
1616 int optname, char __user *optval, int __user *optlen);
1617 int compat_sock_common_setsockopt(struct socket *sock, int level,
1618 int optname, char __user *optval, unsigned int optlen);
1620 void sk_common_release(struct sock *sk);
1623 * Default socket callbacks and setup code
1626 /* Initialise core socket variables */
1627 void sock_init_data(struct socket *sock, struct sock *sk);
1630 * Socket reference counting postulates.
1632 * * Each user of socket SHOULD hold a reference count.
1633 * * Each access point to socket (an hash table bucket, reference from a list,
1634 * running timer, skb in flight MUST hold a reference count.
1635 * * When reference count hits 0, it means it will never increase back.
1636 * * When reference count hits 0, it means that no references from
1637 * outside exist to this socket and current process on current CPU
1638 * is last user and may/should destroy this socket.
1639 * * sk_free is called from any context: process, BH, IRQ. When
1640 * it is called, socket has no references from outside -> sk_free
1641 * may release descendant resources allocated by the socket, but
1642 * to the time when it is called, socket is NOT referenced by any
1643 * hash tables, lists etc.
1644 * * Packets, delivered from outside (from network or from another process)
1645 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1646 * when they sit in queue. Otherwise, packets will leak to hole, when
1647 * socket is looked up by one cpu and unhasing is made by another CPU.
1648 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1649 * (leak to backlog). Packet socket does all the processing inside
1650 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1651 * use separate SMP lock, so that they are prone too.
1654 /* Ungrab socket and destroy it, if it was the last reference. */
1655 static inline void sock_put(struct sock *sk)
1657 if (refcount_dec_and_test(&sk->sk_refcnt))
1660 /* Generic version of sock_put(), dealing with all sockets
1661 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1663 void sock_gen_put(struct sock *sk);
1665 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1666 unsigned int trim_cap, bool refcounted);
1667 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1670 return __sk_receive_skb(sk, skb, nested, 1, true);
1673 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1675 sk->sk_tx_queue_mapping = tx_queue;
1678 static inline void sk_tx_queue_clear(struct sock *sk)
1680 sk->sk_tx_queue_mapping = -1;
1683 static inline int sk_tx_queue_get(const struct sock *sk)
1685 return sk ? sk->sk_tx_queue_mapping : -1;
1688 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1690 sk_tx_queue_clear(sk);
1691 sk->sk_socket = sock;
1694 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1696 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1697 return &rcu_dereference_raw(sk->sk_wq)->wait;
1699 /* Detach socket from process context.
1700 * Announce socket dead, detach it from wait queue and inode.
1701 * Note that parent inode held reference count on this struct sock,
1702 * we do not release it in this function, because protocol
1703 * probably wants some additional cleanups or even continuing
1704 * to work with this socket (TCP).
1706 static inline void sock_orphan(struct sock *sk)
1708 write_lock_bh(&sk->sk_callback_lock);
1709 sock_set_flag(sk, SOCK_DEAD);
1710 sk_set_socket(sk, NULL);
1712 write_unlock_bh(&sk->sk_callback_lock);
1715 static inline void sock_graft(struct sock *sk, struct socket *parent)
1717 WARN_ON(parent->sk);
1718 write_lock_bh(&sk->sk_callback_lock);
1719 sk->sk_wq = parent->wq;
1721 sk_set_socket(sk, parent);
1722 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1723 security_sock_graft(sk, parent);
1724 write_unlock_bh(&sk->sk_callback_lock);
1727 kuid_t sock_i_uid(struct sock *sk);
1728 unsigned long sock_i_ino(struct sock *sk);
1730 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1732 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1735 static inline u32 net_tx_rndhash(void)
1737 u32 v = prandom_u32();
1742 static inline void sk_set_txhash(struct sock *sk)
1744 sk->sk_txhash = net_tx_rndhash();
1747 static inline void sk_rethink_txhash(struct sock *sk)
1753 static inline struct dst_entry *
1754 __sk_dst_get(struct sock *sk)
1756 return rcu_dereference_check(sk->sk_dst_cache,
1757 lockdep_sock_is_held(sk));
1760 static inline struct dst_entry *
1761 sk_dst_get(struct sock *sk)
1763 struct dst_entry *dst;
1766 dst = rcu_dereference(sk->sk_dst_cache);
1767 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1773 static inline void dst_negative_advice(struct sock *sk)
1775 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1777 sk_rethink_txhash(sk);
1779 if (dst && dst->ops->negative_advice) {
1780 ndst = dst->ops->negative_advice(dst);
1783 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1784 sk_tx_queue_clear(sk);
1785 sk->sk_dst_pending_confirm = 0;
1791 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1793 struct dst_entry *old_dst;
1795 sk_tx_queue_clear(sk);
1796 sk->sk_dst_pending_confirm = 0;
1797 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1798 lockdep_sock_is_held(sk));
1799 rcu_assign_pointer(sk->sk_dst_cache, dst);
1800 dst_release(old_dst);
1804 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1806 struct dst_entry *old_dst;
1808 sk_tx_queue_clear(sk);
1809 sk->sk_dst_pending_confirm = 0;
1810 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1811 dst_release(old_dst);
1815 __sk_dst_reset(struct sock *sk)
1817 __sk_dst_set(sk, NULL);
1821 sk_dst_reset(struct sock *sk)
1823 sk_dst_set(sk, NULL);
1826 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1828 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1830 static inline void sk_dst_confirm(struct sock *sk)
1832 if (!sk->sk_dst_pending_confirm)
1833 sk->sk_dst_pending_confirm = 1;
1836 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1838 if (skb_get_dst_pending_confirm(skb)) {
1839 struct sock *sk = skb->sk;
1840 unsigned long now = jiffies;
1842 /* avoid dirtying neighbour */
1843 if (n->confirmed != now)
1845 if (sk && sk->sk_dst_pending_confirm)
1846 sk->sk_dst_pending_confirm = 0;
1850 bool sk_mc_loop(struct sock *sk);
1852 static inline bool sk_can_gso(const struct sock *sk)
1854 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1857 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1859 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1861 sk->sk_route_nocaps |= flags;
1862 sk->sk_route_caps &= ~flags;
1865 static inline bool sk_check_csum_caps(struct sock *sk)
1867 return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1868 (sk->sk_family == PF_INET &&
1869 (sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1870 (sk->sk_family == PF_INET6 &&
1871 (sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1874 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1875 struct iov_iter *from, char *to,
1876 int copy, int offset)
1878 if (skb->ip_summed == CHECKSUM_NONE) {
1880 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1882 skb->csum = csum_block_add(skb->csum, csum, offset);
1883 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1884 if (!copy_from_iter_full_nocache(to, copy, from))
1886 } else if (!copy_from_iter_full(to, copy, from))
1892 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1893 struct iov_iter *from, int copy)
1895 int err, offset = skb->len;
1897 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1900 __skb_trim(skb, offset);
1905 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1906 struct sk_buff *skb,
1912 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1918 skb->data_len += copy;
1919 skb->truesize += copy;
1920 sk->sk_wmem_queued += copy;
1921 sk_mem_charge(sk, copy);
1926 * sk_wmem_alloc_get - returns write allocations
1929 * Returns sk_wmem_alloc minus initial offset of one
1931 static inline int sk_wmem_alloc_get(const struct sock *sk)
1933 return refcount_read(&sk->sk_wmem_alloc) - 1;
1937 * sk_rmem_alloc_get - returns read allocations
1940 * Returns sk_rmem_alloc
1942 static inline int sk_rmem_alloc_get(const struct sock *sk)
1944 return atomic_read(&sk->sk_rmem_alloc);
1948 * sk_has_allocations - check if allocations are outstanding
1951 * Returns true if socket has write or read allocations
1953 static inline bool sk_has_allocations(const struct sock *sk)
1955 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1959 * skwq_has_sleeper - check if there are any waiting processes
1960 * @wq: struct socket_wq
1962 * Returns true if socket_wq has waiting processes
1964 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1965 * barrier call. They were added due to the race found within the tcp code.
1967 * Consider following tcp code paths::
1970 * sys_select receive packet
1972 * __add_wait_queue update tp->rcv_nxt
1974 * tp->rcv_nxt check sock_def_readable
1976 * schedule rcu_read_lock();
1977 * wq = rcu_dereference(sk->sk_wq);
1978 * if (wq && waitqueue_active(&wq->wait))
1979 * wake_up_interruptible(&wq->wait)
1983 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1984 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1985 * could then endup calling schedule and sleep forever if there are no more
1986 * data on the socket.
1989 static inline bool skwq_has_sleeper(struct socket_wq *wq)
1991 return wq && wq_has_sleeper(&wq->wait);
1995 * sock_poll_wait - place memory barrier behind the poll_wait call.
1997 * @wait_address: socket wait queue
2000 * See the comments in the wq_has_sleeper function.
2002 static inline void sock_poll_wait(struct file *filp,
2003 wait_queue_head_t *wait_address, poll_table *p)
2005 if (!poll_does_not_wait(p) && wait_address) {
2006 poll_wait(filp, wait_address, p);
2007 /* We need to be sure we are in sync with the
2008 * socket flags modification.
2010 * This memory barrier is paired in the wq_has_sleeper.
2016 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2018 if (sk->sk_txhash) {
2020 skb->hash = sk->sk_txhash;
2024 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2027 * Queue a received datagram if it will fit. Stream and sequenced
2028 * protocols can't normally use this as they need to fit buffers in
2029 * and play with them.
2031 * Inlined as it's very short and called for pretty much every
2032 * packet ever received.
2034 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2038 skb->destructor = sock_rfree;
2039 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2040 sk_mem_charge(sk, skb->truesize);
2043 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2044 unsigned long expires);
2046 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2048 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2049 struct sk_buff *skb, unsigned int flags,
2050 void (*destructor)(struct sock *sk,
2051 struct sk_buff *skb));
2052 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2053 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2055 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2056 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2059 * Recover an error report and clear atomically
2062 static inline int sock_error(struct sock *sk)
2065 if (likely(!sk->sk_err))
2067 err = xchg(&sk->sk_err, 0);
2071 static inline unsigned long sock_wspace(struct sock *sk)
2075 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2076 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2084 * We use sk->sk_wq_raw, from contexts knowing this
2085 * pointer is not NULL and cannot disappear/change.
2087 static inline void sk_set_bit(int nr, struct sock *sk)
2089 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2090 !sock_flag(sk, SOCK_FASYNC))
2093 set_bit(nr, &sk->sk_wq_raw->flags);
2096 static inline void sk_clear_bit(int nr, struct sock *sk)
2098 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2099 !sock_flag(sk, SOCK_FASYNC))
2102 clear_bit(nr, &sk->sk_wq_raw->flags);
2105 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2107 if (sock_flag(sk, SOCK_FASYNC)) {
2109 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2114 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2115 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2116 * Note: for send buffers, TCP works better if we can build two skbs at
2119 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2121 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2122 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2124 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2126 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2127 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2128 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2132 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2133 bool force_schedule);
2136 * sk_page_frag - return an appropriate page_frag
2139 * If socket allocation mode allows current thread to sleep, it means its
2140 * safe to use the per task page_frag instead of the per socket one.
2142 static inline struct page_frag *sk_page_frag(struct sock *sk)
2144 if (gfpflags_allow_blocking(sk->sk_allocation))
2145 return ¤t->task_frag;
2147 return &sk->sk_frag;
2150 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2153 * Default write policy as shown to user space via poll/select/SIGIO
2155 static inline bool sock_writeable(const struct sock *sk)
2157 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2160 static inline gfp_t gfp_any(void)
2162 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2165 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2167 return noblock ? 0 : sk->sk_rcvtimeo;
2170 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2172 return noblock ? 0 : sk->sk_sndtimeo;
2175 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2177 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2180 /* Alas, with timeout socket operations are not restartable.
2181 * Compare this to poll().
2183 static inline int sock_intr_errno(long timeo)
2185 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2188 struct sock_skb_cb {
2192 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2193 * using skb->cb[] would keep using it directly and utilize its
2194 * alignement guarantee.
2196 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2197 sizeof(struct sock_skb_cb)))
2199 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2200 SOCK_SKB_CB_OFFSET))
2202 #define sock_skb_cb_check_size(size) \
2203 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2206 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2208 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2209 atomic_read(&sk->sk_drops) : 0;
2212 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2214 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2216 atomic_add(segs, &sk->sk_drops);
2219 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2220 struct sk_buff *skb);
2221 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2222 struct sk_buff *skb);
2225 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2227 ktime_t kt = skb->tstamp;
2228 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2231 * generate control messages if
2232 * - receive time stamping in software requested
2233 * - software time stamp available and wanted
2234 * - hardware time stamps available and wanted
2236 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2237 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2238 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2239 (hwtstamps->hwtstamp &&
2240 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2241 __sock_recv_timestamp(msg, sk, skb);
2245 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2246 __sock_recv_wifi_status(msg, sk, skb);
2249 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2250 struct sk_buff *skb);
2252 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2253 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2254 struct sk_buff *skb)
2256 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2257 (1UL << SOCK_RCVTSTAMP))
2258 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2259 SOF_TIMESTAMPING_RAW_HARDWARE)
2261 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2262 __sock_recv_ts_and_drops(msg, sk, skb);
2263 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2264 sk->sk_stamp = skb->tstamp;
2265 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2269 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2272 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2273 * @sk: socket sending this packet
2274 * @tsflags: timestamping flags to use
2275 * @tx_flags: completed with instructions for time stamping
2277 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2279 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2282 if (unlikely(tsflags))
2283 __sock_tx_timestamp(tsflags, tx_flags);
2284 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2285 *tx_flags |= SKBTX_WIFI_STATUS;
2289 * sk_eat_skb - Release a skb if it is no longer needed
2290 * @sk: socket to eat this skb from
2291 * @skb: socket buffer to eat
2293 * This routine must be called with interrupts disabled or with the socket
2294 * locked so that the sk_buff queue operation is ok.
2296 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2298 __skb_unlink(skb, &sk->sk_receive_queue);
2303 struct net *sock_net(const struct sock *sk)
2305 return read_pnet(&sk->sk_net);
2309 void sock_net_set(struct sock *sk, struct net *net)
2311 write_pnet(&sk->sk_net, net);
2314 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2317 struct sock *sk = skb->sk;
2319 skb->destructor = NULL;
2326 /* This helper checks if a socket is a full socket,
2327 * ie _not_ a timewait or request socket.
2329 static inline bool sk_fullsock(const struct sock *sk)
2331 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2334 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2335 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2337 static inline bool sk_listener(const struct sock *sk)
2339 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2343 * sk_state_load - read sk->sk_state for lockless contexts
2344 * @sk: socket pointer
2346 * Paired with sk_state_store(). Used in places we do not hold socket lock :
2347 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2349 static inline int sk_state_load(const struct sock *sk)
2351 return smp_load_acquire(&sk->sk_state);
2355 * sk_state_store - update sk->sk_state
2356 * @sk: socket pointer
2357 * @newstate: new state
2359 * Paired with sk_state_load(). Should be used in contexts where
2360 * state change might impact lockless readers.
2362 static inline void sk_state_store(struct sock *sk, int newstate)
2364 smp_store_release(&sk->sk_state, newstate);
2367 void sock_enable_timestamp(struct sock *sk, int flag);
2368 int sock_get_timestamp(struct sock *, struct timeval __user *);
2369 int sock_get_timestampns(struct sock *, struct timespec __user *);
2370 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2373 bool sk_ns_capable(const struct sock *sk,
2374 struct user_namespace *user_ns, int cap);
2375 bool sk_capable(const struct sock *sk, int cap);
2376 bool sk_net_capable(const struct sock *sk, int cap);
2378 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2380 /* Take into consideration the size of the struct sk_buff overhead in the
2381 * determination of these values, since that is non-constant across
2382 * platforms. This makes socket queueing behavior and performance
2383 * not depend upon such differences.
2385 #define _SK_MEM_PACKETS 256
2386 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2387 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2388 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2390 extern __u32 sysctl_wmem_max;
2391 extern __u32 sysctl_rmem_max;
2393 extern int sysctl_tstamp_allow_data;
2394 extern int sysctl_optmem_max;
2396 extern __u32 sysctl_wmem_default;
2397 extern __u32 sysctl_rmem_default;
2399 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2401 /* Does this proto have per netns sysctl_wmem ? */
2402 if (proto->sysctl_wmem_offset)
2403 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2405 return *proto->sysctl_wmem;
2408 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2410 /* Does this proto have per netns sysctl_rmem ? */
2411 if (proto->sysctl_rmem_offset)
2412 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2414 return *proto->sysctl_rmem;
2417 #endif /* _SOCK_H */