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>
75 #include <net/l3mdev.h>
78 * This structure really needs to be cleaned up.
79 * Most of it is for TCP, and not used by any of
80 * the other protocols.
83 /* Define this to get the SOCK_DBG debugging facility. */
84 #define SOCK_DEBUGGING
86 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
87 printk(KERN_DEBUG msg); } while (0)
89 /* Validate arguments and do nothing */
90 static inline __printf(2, 3)
91 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
96 /* This is the per-socket lock. The spinlock provides a synchronization
97 * between user contexts and software interrupt processing, whereas the
98 * mini-semaphore synchronizes multiple users amongst themselves.
103 wait_queue_head_t wq;
105 * We express the mutex-alike socket_lock semantics
106 * to the lock validator by explicitly managing
107 * the slock as a lock variant (in addition to
110 #ifdef CONFIG_DEBUG_LOCK_ALLOC
111 struct lockdep_map dep_map;
119 typedef __u32 __bitwise __portpair;
120 typedef __u64 __bitwise __addrpair;
123 * struct sock_common - minimal network layer representation of sockets
124 * @skc_daddr: Foreign IPv4 addr
125 * @skc_rcv_saddr: Bound local IPv4 addr
126 * @skc_hash: hash value used with various protocol lookup tables
127 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
128 * @skc_dport: placeholder for inet_dport/tw_dport
129 * @skc_num: placeholder for inet_num/tw_num
130 * @skc_family: network address family
131 * @skc_state: Connection state
132 * @skc_reuse: %SO_REUSEADDR setting
133 * @skc_reuseport: %SO_REUSEPORT setting
134 * @skc_bound_dev_if: bound device index if != 0
135 * @skc_bind_node: bind hash linkage for various protocol lookup tables
136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137 * @skc_prot: protocol handlers inside a network family
138 * @skc_net: reference to the network namespace of this socket
139 * @skc_node: main hash linkage for various protocol lookup tables
140 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
141 * @skc_tx_queue_mapping: tx queue number for this connection
142 * @skc_rx_queue_mapping: rx queue number for this connection
143 * @skc_flags: place holder for sk_flags
144 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
145 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
146 * @skc_incoming_cpu: record/match cpu processing incoming packets
147 * @skc_refcnt: reference count
149 * This is the minimal network layer representation of sockets, the header
150 * for struct sock and struct inet_timewait_sock.
153 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
154 * address on 64bit arches : cf INET_MATCH()
157 __addrpair skc_addrpair;
160 __be32 skc_rcv_saddr;
164 unsigned int skc_hash;
165 __u16 skc_u16hashes[2];
167 /* skc_dport && skc_num must be grouped as well */
169 __portpair skc_portpair;
176 unsigned short skc_family;
177 volatile unsigned char skc_state;
178 unsigned char skc_reuse:4;
179 unsigned char skc_reuseport:1;
180 unsigned char skc_ipv6only:1;
181 unsigned char skc_net_refcnt:1;
182 int skc_bound_dev_if;
184 struct hlist_node skc_bind_node;
185 struct hlist_node skc_portaddr_node;
187 struct proto *skc_prot;
188 possible_net_t skc_net;
190 #if IS_ENABLED(CONFIG_IPV6)
191 struct in6_addr skc_v6_daddr;
192 struct in6_addr skc_v6_rcv_saddr;
195 atomic64_t skc_cookie;
197 /* following fields are padding to force
198 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
199 * assuming IPV6 is enabled. We use this padding differently
200 * for different kind of 'sockets'
203 unsigned long skc_flags;
204 struct sock *skc_listener; /* request_sock */
205 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
208 * fields between dontcopy_begin/dontcopy_end
209 * are not copied in sock_copy()
212 int skc_dontcopy_begin[0];
215 struct hlist_node skc_node;
216 struct hlist_nulls_node skc_nulls_node;
218 unsigned short skc_tx_queue_mapping;
220 unsigned short skc_rx_queue_mapping;
223 int skc_incoming_cpu;
225 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
228 refcount_t skc_refcnt;
230 int skc_dontcopy_end[0];
233 u32 skc_window_clamp;
234 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
240 * struct sock - network layer representation of sockets
241 * @__sk_common: shared layout with inet_timewait_sock
242 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
243 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
244 * @sk_lock: synchronizer
245 * @sk_kern_sock: True if sock is using kernel lock classes
246 * @sk_rcvbuf: size of receive buffer in bytes
247 * @sk_wq: sock wait queue and async head
248 * @sk_rx_dst: receive input route used by early demux
249 * @sk_dst_cache: destination cache
250 * @sk_dst_pending_confirm: need to confirm neighbour
251 * @sk_policy: flow policy
252 * @sk_receive_queue: incoming packets
253 * @sk_wmem_alloc: transmit queue bytes committed
254 * @sk_tsq_flags: TCP Small Queues flags
255 * @sk_write_queue: Packet sending queue
256 * @sk_omem_alloc: "o" is "option" or "other"
257 * @sk_wmem_queued: persistent queue size
258 * @sk_forward_alloc: space allocated forward
259 * @sk_napi_id: id of the last napi context to receive data for sk
260 * @sk_ll_usec: usecs to busypoll when there is no data
261 * @sk_allocation: allocation mode
262 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
263 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
264 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
265 * @sk_sndbuf: size of send buffer in bytes
266 * @__sk_flags_offset: empty field used to determine location of bitfield
267 * @sk_padding: unused element for alignment
268 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
269 * @sk_no_check_rx: allow zero checksum in RX packets
270 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
271 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
272 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
273 * @sk_gso_max_size: Maximum GSO segment size to build
274 * @sk_gso_max_segs: Maximum number of GSO segments
275 * @sk_pacing_shift: scaling factor for TCP Small Queues
276 * @sk_lingertime: %SO_LINGER l_linger setting
277 * @sk_backlog: always used with the per-socket spinlock held
278 * @sk_callback_lock: used with the callbacks in the end of this struct
279 * @sk_error_queue: rarely used
280 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
281 * IPV6_ADDRFORM for instance)
282 * @sk_err: last error
283 * @sk_err_soft: errors that don't cause failure but are the cause of a
284 * persistent failure not just 'timed out'
285 * @sk_drops: raw/udp drops counter
286 * @sk_ack_backlog: current listen backlog
287 * @sk_max_ack_backlog: listen backlog set in listen()
288 * @sk_uid: user id of owner
289 * @sk_priority: %SO_PRIORITY setting
290 * @sk_type: socket type (%SOCK_STREAM, etc)
291 * @sk_protocol: which protocol this socket belongs in this network family
292 * @sk_peer_pid: &struct pid for this socket's peer
293 * @sk_peer_cred: %SO_PEERCRED setting
294 * @sk_rcvlowat: %SO_RCVLOWAT setting
295 * @sk_rcvtimeo: %SO_RCVTIMEO setting
296 * @sk_sndtimeo: %SO_SNDTIMEO setting
297 * @sk_txhash: computed flow hash for use on transmit
298 * @sk_filter: socket filtering instructions
299 * @sk_timer: sock cleanup timer
300 * @sk_stamp: time stamp of last packet received
301 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
302 * @sk_tsflags: SO_TIMESTAMPING socket options
303 * @sk_tskey: counter to disambiguate concurrent tstamp requests
304 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
305 * @sk_socket: Identd and reporting IO signals
306 * @sk_user_data: RPC layer private data
307 * @sk_frag: cached page frag
308 * @sk_peek_off: current peek_offset value
309 * @sk_send_head: front of stuff to transmit
310 * @sk_security: used by security modules
311 * @sk_mark: generic packet mark
312 * @sk_cgrp_data: cgroup data for this cgroup
313 * @sk_memcg: this socket's memory cgroup association
314 * @sk_write_pending: a write to stream socket waits to start
315 * @sk_wait_pending: number of threads blocked on this socket
316 * @sk_state_change: callback to indicate change in the state of the sock
317 * @sk_data_ready: callback to indicate there is data to be processed
318 * @sk_write_space: callback to indicate there is bf sending space available
319 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
320 * @sk_backlog_rcv: callback to process the backlog
321 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
322 * @sk_reuseport_cb: reuseport group container
323 * @sk_rcu: used during RCU grace period
324 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
325 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
326 * @sk_txtime_unused: unused txtime flags
330 * Now struct inet_timewait_sock also uses sock_common, so please just
331 * don't add nothing before this first member (__sk_common) --acme
333 struct sock_common __sk_common;
334 #define sk_node __sk_common.skc_node
335 #define sk_nulls_node __sk_common.skc_nulls_node
336 #define sk_refcnt __sk_common.skc_refcnt
337 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
339 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
342 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
343 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
344 #define sk_hash __sk_common.skc_hash
345 #define sk_portpair __sk_common.skc_portpair
346 #define sk_num __sk_common.skc_num
347 #define sk_dport __sk_common.skc_dport
348 #define sk_addrpair __sk_common.skc_addrpair
349 #define sk_daddr __sk_common.skc_daddr
350 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
351 #define sk_family __sk_common.skc_family
352 #define sk_state __sk_common.skc_state
353 #define sk_reuse __sk_common.skc_reuse
354 #define sk_reuseport __sk_common.skc_reuseport
355 #define sk_ipv6only __sk_common.skc_ipv6only
356 #define sk_net_refcnt __sk_common.skc_net_refcnt
357 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
358 #define sk_bind_node __sk_common.skc_bind_node
359 #define sk_prot __sk_common.skc_prot
360 #define sk_net __sk_common.skc_net
361 #define sk_v6_daddr __sk_common.skc_v6_daddr
362 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
363 #define sk_cookie __sk_common.skc_cookie
364 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
365 #define sk_flags __sk_common.skc_flags
366 #define sk_rxhash __sk_common.skc_rxhash
368 socket_lock_t sk_lock;
371 struct sk_buff_head sk_error_queue;
372 struct sk_buff_head sk_receive_queue;
374 * The backlog queue is special, it is always used with
375 * the per-socket spinlock held and requires low latency
376 * access. Therefore we special case it's implementation.
377 * Note : rmem_alloc is in this structure to fill a hole
378 * on 64bit arches, not because its logically part of
384 struct sk_buff *head;
385 struct sk_buff *tail;
387 #define sk_rmem_alloc sk_backlog.rmem_alloc
389 int sk_forward_alloc;
390 #ifdef CONFIG_NET_RX_BUSY_POLL
391 unsigned int sk_ll_usec;
392 /* ===== mostly read cache line ===== */
393 unsigned int sk_napi_id;
398 struct sk_filter __rcu *sk_filter;
400 struct socket_wq __rcu *sk_wq;
401 struct socket_wq *sk_wq_raw;
404 struct xfrm_policy __rcu *sk_policy[2];
406 struct dst_entry __rcu *sk_rx_dst;
407 struct dst_entry __rcu *sk_dst_cache;
408 atomic_t sk_omem_alloc;
411 /* ===== cache line for TX ===== */
413 refcount_t sk_wmem_alloc;
414 unsigned long sk_tsq_flags;
416 struct sk_buff *sk_send_head;
417 struct rb_root tcp_rtx_queue;
419 struct sk_buff_head sk_write_queue;
421 int sk_write_pending;
422 __u32 sk_dst_pending_confirm;
423 u32 sk_pacing_status; /* see enum sk_pacing */
425 struct timer_list sk_timer;
428 u32 sk_pacing_rate; /* bytes per second */
429 u32 sk_max_pacing_rate;
430 struct page_frag sk_frag;
431 netdev_features_t sk_route_caps;
432 netdev_features_t sk_route_nocaps;
433 netdev_features_t sk_route_forced_caps;
435 unsigned int sk_gso_max_size;
440 * Because of non atomicity rules, all
441 * changes are protected by socket lock.
443 unsigned int __sk_flags_offset[0];
444 #ifdef __BIG_ENDIAN_BITFIELD
445 #define SK_FL_PROTO_SHIFT 16
446 #define SK_FL_PROTO_MASK 0x00ff0000
448 #define SK_FL_TYPE_SHIFT 0
449 #define SK_FL_TYPE_MASK 0x0000ffff
451 #define SK_FL_PROTO_SHIFT 8
452 #define SK_FL_PROTO_MASK 0x0000ff00
454 #define SK_FL_TYPE_SHIFT 16
455 #define SK_FL_TYPE_MASK 0xffff0000
458 unsigned int sk_padding : 1,
465 #define SK_PROTOCOL_MAX U8_MAX
468 unsigned long sk_lingertime;
469 struct proto *sk_prot_creator;
470 rwlock_t sk_callback_lock;
474 u32 sk_max_ack_backlog;
476 spinlock_t sk_peer_lock;
477 struct pid *sk_peer_pid;
478 const struct cred *sk_peer_cred;
482 #if BITS_PER_LONG==32
483 seqlock_t sk_stamp_seq;
491 u8 sk_txtime_deadline_mode : 1,
492 sk_txtime_report_errors : 1,
493 sk_txtime_unused : 6;
495 struct socket *sk_socket;
497 #ifdef CONFIG_SECURITY
500 struct sock_cgroup_data sk_cgrp_data;
501 struct mem_cgroup *sk_memcg;
502 void (*sk_state_change)(struct sock *sk);
503 void (*sk_data_ready)(struct sock *sk);
504 void (*sk_write_space)(struct sock *sk);
505 void (*sk_error_report)(struct sock *sk);
506 int (*sk_backlog_rcv)(struct sock *sk,
507 struct sk_buff *skb);
508 #ifdef CONFIG_SOCK_VALIDATE_XMIT
509 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
510 struct net_device *dev,
511 struct sk_buff *skb);
513 void (*sk_destruct)(struct sock *sk);
514 struct sock_reuseport __rcu *sk_reuseport_cb;
515 struct rcu_head sk_rcu;
520 SK_PACING_NEEDED = 1,
524 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
526 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
527 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
530 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
531 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
532 * on a socket means that the socket will reuse everybody else's port
533 * without looking at the other's sk_reuse value.
536 #define SK_NO_REUSE 0
537 #define SK_CAN_REUSE 1
538 #define SK_FORCE_REUSE 2
540 int sk_set_peek_off(struct sock *sk, int val);
542 static inline int sk_peek_offset(struct sock *sk, int flags)
544 if (unlikely(flags & MSG_PEEK)) {
545 return READ_ONCE(sk->sk_peek_off);
551 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
553 s32 off = READ_ONCE(sk->sk_peek_off);
555 if (unlikely(off >= 0)) {
556 off = max_t(s32, off - val, 0);
557 WRITE_ONCE(sk->sk_peek_off, off);
561 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
563 sk_peek_offset_bwd(sk, -val);
567 * Hashed lists helper routines
569 static inline struct sock *sk_entry(const struct hlist_node *node)
571 return hlist_entry(node, struct sock, sk_node);
574 static inline struct sock *__sk_head(const struct hlist_head *head)
576 return hlist_entry(head->first, struct sock, sk_node);
579 static inline struct sock *sk_head(const struct hlist_head *head)
581 return hlist_empty(head) ? NULL : __sk_head(head);
584 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
586 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
589 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
591 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
594 static inline struct sock *sk_next(const struct sock *sk)
596 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
599 static inline struct sock *sk_nulls_next(const struct sock *sk)
601 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
602 hlist_nulls_entry(sk->sk_nulls_node.next,
603 struct sock, sk_nulls_node) :
607 static inline bool sk_unhashed(const struct sock *sk)
609 return hlist_unhashed(&sk->sk_node);
612 static inline bool sk_hashed(const struct sock *sk)
614 return !sk_unhashed(sk);
617 static inline void sk_node_init(struct hlist_node *node)
622 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
627 static inline void __sk_del_node(struct sock *sk)
629 __hlist_del(&sk->sk_node);
632 /* NB: equivalent to hlist_del_init_rcu */
633 static inline bool __sk_del_node_init(struct sock *sk)
637 sk_node_init(&sk->sk_node);
643 /* Grab socket reference count. This operation is valid only
644 when sk is ALREADY grabbed f.e. it is found in hash table
645 or a list and the lookup is made under lock preventing hash table
649 static __always_inline void sock_hold(struct sock *sk)
651 refcount_inc(&sk->sk_refcnt);
654 /* Ungrab socket in the context, which assumes that socket refcnt
655 cannot hit zero, f.e. it is true in context of any socketcall.
657 static __always_inline void __sock_put(struct sock *sk)
659 refcount_dec(&sk->sk_refcnt);
662 static inline bool sk_del_node_init(struct sock *sk)
664 bool rc = __sk_del_node_init(sk);
667 /* paranoid for a while -acme */
668 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
673 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
675 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
678 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
684 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
686 bool rc = __sk_nulls_del_node_init_rcu(sk);
689 /* paranoid for a while -acme */
690 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
696 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
698 hlist_add_head(&sk->sk_node, list);
701 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
704 __sk_add_node(sk, list);
707 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
710 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
711 sk->sk_family == AF_INET6)
712 hlist_add_tail_rcu(&sk->sk_node, list);
714 hlist_add_head_rcu(&sk->sk_node, list);
717 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
720 hlist_add_tail_rcu(&sk->sk_node, list);
723 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
725 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
728 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
730 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
733 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
736 __sk_nulls_add_node_rcu(sk, list);
739 static inline void __sk_del_bind_node(struct sock *sk)
741 __hlist_del(&sk->sk_bind_node);
744 static inline void sk_add_bind_node(struct sock *sk,
745 struct hlist_head *list)
747 hlist_add_head(&sk->sk_bind_node, list);
750 #define sk_for_each(__sk, list) \
751 hlist_for_each_entry(__sk, list, sk_node)
752 #define sk_for_each_rcu(__sk, list) \
753 hlist_for_each_entry_rcu(__sk, list, sk_node)
754 #define sk_nulls_for_each(__sk, node, list) \
755 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
756 #define sk_nulls_for_each_rcu(__sk, node, list) \
757 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
758 #define sk_for_each_from(__sk) \
759 hlist_for_each_entry_from(__sk, sk_node)
760 #define sk_nulls_for_each_from(__sk, node) \
761 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
762 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
763 #define sk_for_each_safe(__sk, tmp, list) \
764 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
765 #define sk_for_each_bound(__sk, list) \
766 hlist_for_each_entry(__sk, list, sk_bind_node)
769 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
770 * @tpos: the type * to use as a loop cursor.
771 * @pos: the &struct hlist_node to use as a loop cursor.
772 * @head: the head for your list.
773 * @offset: offset of hlist_node within the struct.
776 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
777 for (pos = rcu_dereference(hlist_first_rcu(head)); \
779 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
780 pos = rcu_dereference(hlist_next_rcu(pos)))
782 static inline struct user_namespace *sk_user_ns(struct sock *sk)
784 /* Careful only use this in a context where these parameters
785 * can not change and must all be valid, such as recvmsg from
788 return sk->sk_socket->file->f_cred->user_ns;
802 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
803 SOCK_DBG, /* %SO_DEBUG setting */
804 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
805 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
806 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
807 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
808 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
809 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
810 SOCK_FASYNC, /* fasync() active */
812 SOCK_ZEROCOPY, /* buffers from userspace */
813 SOCK_WIFI_STATUS, /* push wifi status to userspace */
814 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
815 * Will use last 4 bytes of packet sent from
816 * user-space instead.
818 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
819 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
820 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
824 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
826 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
828 nsk->sk_flags = osk->sk_flags;
831 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
833 __set_bit(flag, &sk->sk_flags);
836 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
838 __clear_bit(flag, &sk->sk_flags);
841 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
843 return test_bit(flag, &sk->sk_flags);
847 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
848 static inline int sk_memalloc_socks(void)
850 return static_branch_unlikely(&memalloc_socks_key);
853 void __receive_sock(struct file *file);
856 static inline int sk_memalloc_socks(void)
861 static inline void __receive_sock(struct file *file)
865 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
867 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
870 static inline void sk_acceptq_removed(struct sock *sk)
872 sk->sk_ack_backlog--;
875 static inline void sk_acceptq_added(struct sock *sk)
877 sk->sk_ack_backlog++;
880 static inline bool sk_acceptq_is_full(const struct sock *sk)
882 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
886 * Compute minimal free write space needed to queue new packets.
888 static inline int sk_stream_min_wspace(const struct sock *sk)
890 return sk->sk_wmem_queued >> 1;
893 static inline int sk_stream_wspace(const struct sock *sk)
895 return sk->sk_sndbuf - sk->sk_wmem_queued;
898 void sk_stream_write_space(struct sock *sk);
900 /* OOB backlog add */
901 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
903 /* dont let skb dst not refcounted, we are going to leave rcu lock */
906 if (!sk->sk_backlog.tail)
907 WRITE_ONCE(sk->sk_backlog.head, skb);
909 sk->sk_backlog.tail->next = skb;
911 WRITE_ONCE(sk->sk_backlog.tail, skb);
916 * Take into account size of receive queue and backlog queue
917 * Do not take into account this skb truesize,
918 * to allow even a single big packet to come.
920 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
922 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
924 return qsize > limit;
927 /* The per-socket spinlock must be held here. */
928 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
931 if (sk_rcvqueues_full(sk, limit))
935 * If the skb was allocated from pfmemalloc reserves, only
936 * allow SOCK_MEMALLOC sockets to use it as this socket is
937 * helping free memory
939 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
942 __sk_add_backlog(sk, skb);
943 sk->sk_backlog.len += skb->truesize;
947 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
949 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
951 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
952 return __sk_backlog_rcv(sk, skb);
954 return sk->sk_backlog_rcv(sk, skb);
957 static inline void sk_incoming_cpu_update(struct sock *sk)
959 int cpu = raw_smp_processor_id();
961 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
962 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
965 static inline void sock_rps_record_flow_hash(__u32 hash)
968 struct rps_sock_flow_table *sock_flow_table;
971 sock_flow_table = rcu_dereference(rps_sock_flow_table);
972 rps_record_sock_flow(sock_flow_table, hash);
977 static inline void sock_rps_record_flow(const struct sock *sk)
980 if (static_key_false(&rfs_needed)) {
981 /* Reading sk->sk_rxhash might incur an expensive cache line
984 * TCP_ESTABLISHED does cover almost all states where RFS
985 * might be useful, and is cheaper [1] than testing :
986 * IPv4: inet_sk(sk)->inet_daddr
987 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
988 * OR an additional socket flag
989 * [1] : sk_state and sk_prot are in the same cache line.
991 if (sk->sk_state == TCP_ESTABLISHED) {
992 /* This READ_ONCE() is paired with the WRITE_ONCE()
993 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
995 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1001 static inline void sock_rps_save_rxhash(struct sock *sk,
1002 const struct sk_buff *skb)
1005 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1006 * here, and another one in sock_rps_record_flow().
1008 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1009 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1013 static inline void sock_rps_reset_rxhash(struct sock *sk)
1016 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1017 WRITE_ONCE(sk->sk_rxhash, 0);
1021 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1023 __sk->sk_wait_pending++; \
1024 release_sock(__sk); \
1025 __rc = __condition; \
1027 *(__timeo) = wait_woken(__wait, \
1028 TASK_INTERRUPTIBLE, \
1031 sched_annotate_sleep(); \
1033 __sk->sk_wait_pending--; \
1034 __rc = __condition; \
1038 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1039 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1040 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1041 int sk_stream_error(struct sock *sk, int flags, int err);
1042 void sk_stream_kill_queues(struct sock *sk);
1043 void sk_set_memalloc(struct sock *sk);
1044 void sk_clear_memalloc(struct sock *sk);
1046 void __sk_flush_backlog(struct sock *sk);
1048 static inline bool sk_flush_backlog(struct sock *sk)
1050 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1051 __sk_flush_backlog(sk);
1057 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1059 struct request_sock_ops;
1060 struct timewait_sock_ops;
1061 struct inet_hashinfo;
1062 struct raw_hashinfo;
1063 struct smc_hashinfo;
1067 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1068 * un-modified. Special care is taken when initializing object to zero.
1070 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1072 if (offsetof(struct sock, sk_node.next) != 0)
1073 memset(sk, 0, offsetof(struct sock, sk_node.next));
1074 memset(&sk->sk_node.pprev, 0,
1075 size - offsetof(struct sock, sk_node.pprev));
1078 /* Networking protocol blocks we attach to sockets.
1079 * socket layer -> transport layer interface
1082 void (*close)(struct sock *sk,
1084 int (*pre_connect)(struct sock *sk,
1085 struct sockaddr *uaddr,
1087 int (*connect)(struct sock *sk,
1088 struct sockaddr *uaddr,
1090 int (*disconnect)(struct sock *sk, int flags);
1092 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1095 int (*ioctl)(struct sock *sk, int cmd,
1097 int (*init)(struct sock *sk);
1098 void (*destroy)(struct sock *sk);
1099 void (*shutdown)(struct sock *sk, int how);
1100 int (*setsockopt)(struct sock *sk, int level,
1101 int optname, char __user *optval,
1102 unsigned int optlen);
1103 int (*getsockopt)(struct sock *sk, int level,
1104 int optname, char __user *optval,
1105 int __user *option);
1106 void (*keepalive)(struct sock *sk, int valbool);
1107 #ifdef CONFIG_COMPAT
1108 int (*compat_setsockopt)(struct sock *sk,
1110 int optname, char __user *optval,
1111 unsigned int optlen);
1112 int (*compat_getsockopt)(struct sock *sk,
1114 int optname, char __user *optval,
1115 int __user *option);
1116 int (*compat_ioctl)(struct sock *sk,
1117 unsigned int cmd, unsigned long arg);
1119 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1121 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1122 size_t len, int noblock, int flags,
1124 int (*sendpage)(struct sock *sk, struct page *page,
1125 int offset, size_t size, int flags);
1126 int (*bind)(struct sock *sk,
1127 struct sockaddr *uaddr, int addr_len);
1129 int (*backlog_rcv) (struct sock *sk,
1130 struct sk_buff *skb);
1132 void (*release_cb)(struct sock *sk);
1134 /* Keeping track of sk's, looking them up, and port selection methods. */
1135 int (*hash)(struct sock *sk);
1136 void (*unhash)(struct sock *sk);
1137 void (*rehash)(struct sock *sk);
1138 int (*get_port)(struct sock *sk, unsigned short snum);
1140 /* Keeping track of sockets in use */
1141 #ifdef CONFIG_PROC_FS
1142 unsigned int inuse_idx;
1145 bool (*stream_memory_free)(const struct sock *sk, int wake);
1146 bool (*stream_memory_read)(const struct sock *sk);
1147 /* Memory pressure */
1148 void (*enter_memory_pressure)(struct sock *sk);
1149 void (*leave_memory_pressure)(struct sock *sk);
1150 atomic_long_t *memory_allocated; /* Current allocated memory. */
1151 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1153 * Pressure flag: try to collapse.
1154 * Technical note: it is used by multiple contexts non atomically.
1155 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1156 * All the __sk_mem_schedule() is of this nature: accounting
1157 * is strict, actions are advisory and have some latency.
1159 unsigned long *memory_pressure;
1164 u32 sysctl_wmem_offset;
1165 u32 sysctl_rmem_offset;
1170 struct kmem_cache *slab;
1171 unsigned int obj_size;
1172 slab_flags_t slab_flags;
1173 unsigned int useroffset; /* Usercopy region offset */
1174 unsigned int usersize; /* Usercopy region size */
1176 struct percpu_counter *orphan_count;
1178 struct request_sock_ops *rsk_prot;
1179 struct timewait_sock_ops *twsk_prot;
1182 struct inet_hashinfo *hashinfo;
1183 struct udp_table *udp_table;
1184 struct raw_hashinfo *raw_hash;
1185 struct smc_hashinfo *smc_hash;
1188 struct module *owner;
1192 struct list_head node;
1193 #ifdef SOCK_REFCNT_DEBUG
1196 int (*diag_destroy)(struct sock *sk, int err);
1197 } __randomize_layout;
1199 int proto_register(struct proto *prot, int alloc_slab);
1200 void proto_unregister(struct proto *prot);
1201 int sock_load_diag_module(int family, int protocol);
1203 #ifdef SOCK_REFCNT_DEBUG
1204 static inline void sk_refcnt_debug_inc(struct sock *sk)
1206 atomic_inc(&sk->sk_prot->socks);
1209 static inline void sk_refcnt_debug_dec(struct sock *sk)
1211 atomic_dec(&sk->sk_prot->socks);
1212 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1213 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1216 static inline void sk_refcnt_debug_release(const struct sock *sk)
1218 if (refcount_read(&sk->sk_refcnt) != 1)
1219 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1220 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1222 #else /* SOCK_REFCNT_DEBUG */
1223 #define sk_refcnt_debug_inc(sk) do { } while (0)
1224 #define sk_refcnt_debug_dec(sk) do { } while (0)
1225 #define sk_refcnt_debug_release(sk) do { } while (0)
1226 #endif /* SOCK_REFCNT_DEBUG */
1228 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1230 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1233 return sk->sk_prot->stream_memory_free ?
1234 sk->sk_prot->stream_memory_free(sk, wake) : true;
1237 static inline bool sk_stream_memory_free(const struct sock *sk)
1239 return __sk_stream_memory_free(sk, 0);
1242 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1244 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1245 __sk_stream_memory_free(sk, wake);
1248 static inline bool sk_stream_is_writeable(const struct sock *sk)
1250 return __sk_stream_is_writeable(sk, 0);
1253 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1254 struct cgroup *ancestor)
1256 #ifdef CONFIG_SOCK_CGROUP_DATA
1257 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1264 static inline bool sk_has_memory_pressure(const struct sock *sk)
1266 return sk->sk_prot->memory_pressure != NULL;
1269 static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1271 return sk->sk_prot->memory_pressure &&
1272 !!READ_ONCE(*sk->sk_prot->memory_pressure);
1275 static inline bool sk_under_memory_pressure(const struct sock *sk)
1277 if (!sk->sk_prot->memory_pressure)
1280 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1281 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1284 return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1288 sk_memory_allocated(const struct sock *sk)
1290 return atomic_long_read(sk->sk_prot->memory_allocated);
1294 sk_memory_allocated_add(struct sock *sk, int amt)
1296 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1300 sk_memory_allocated_sub(struct sock *sk, int amt)
1302 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1305 static inline void sk_sockets_allocated_dec(struct sock *sk)
1307 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1310 static inline void sk_sockets_allocated_inc(struct sock *sk)
1312 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1316 sk_sockets_allocated_read_positive(struct sock *sk)
1318 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1322 proto_sockets_allocated_sum_positive(struct proto *prot)
1324 return percpu_counter_sum_positive(prot->sockets_allocated);
1328 proto_memory_allocated(struct proto *prot)
1330 return atomic_long_read(prot->memory_allocated);
1334 proto_memory_pressure(struct proto *prot)
1336 if (!prot->memory_pressure)
1338 return !!READ_ONCE(*prot->memory_pressure);
1342 #ifdef CONFIG_PROC_FS
1343 /* Called with local bh disabled */
1344 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1345 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1346 int sock_inuse_get(struct net *net);
1348 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1355 /* With per-bucket locks this operation is not-atomic, so that
1356 * this version is not worse.
1358 static inline int __sk_prot_rehash(struct sock *sk)
1360 sk->sk_prot->unhash(sk);
1361 return sk->sk_prot->hash(sk);
1364 /* About 10 seconds */
1365 #define SOCK_DESTROY_TIME (10*HZ)
1367 /* Sockets 0-1023 can't be bound to unless you are superuser */
1368 #define PROT_SOCK 1024
1370 #define SHUTDOWN_MASK 3
1371 #define RCV_SHUTDOWN 1
1372 #define SEND_SHUTDOWN 2
1374 #define SOCK_SNDBUF_LOCK 1
1375 #define SOCK_RCVBUF_LOCK 2
1376 #define SOCK_BINDADDR_LOCK 4
1377 #define SOCK_BINDPORT_LOCK 8
1379 struct socket_alloc {
1380 struct socket socket;
1381 struct inode vfs_inode;
1384 static inline struct socket *SOCKET_I(struct inode *inode)
1386 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1389 static inline struct inode *SOCK_INODE(struct socket *socket)
1391 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1395 * Functions for memory accounting
1397 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1398 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1399 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1400 void __sk_mem_reclaim(struct sock *sk, int amount);
1402 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1403 * do not necessarily have 16x time more memory than 4KB ones.
1405 #define SK_MEM_QUANTUM 4096
1406 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1407 #define SK_MEM_SEND 0
1408 #define SK_MEM_RECV 1
1410 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1411 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1413 long val = READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1415 #if PAGE_SIZE > SK_MEM_QUANTUM
1416 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1417 #elif PAGE_SIZE < SK_MEM_QUANTUM
1418 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1423 static inline int sk_mem_pages(int amt)
1425 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1428 static inline bool sk_has_account(struct sock *sk)
1430 /* return true if protocol supports memory accounting */
1431 return !!sk->sk_prot->memory_allocated;
1434 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1436 if (!sk_has_account(sk))
1438 return size <= sk->sk_forward_alloc ||
1439 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1443 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1445 if (!sk_has_account(sk))
1447 return size<= sk->sk_forward_alloc ||
1448 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1449 skb_pfmemalloc(skb);
1452 static inline void sk_mem_reclaim(struct sock *sk)
1454 if (!sk_has_account(sk))
1456 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1457 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1460 static inline void sk_mem_reclaim_partial(struct sock *sk)
1462 if (!sk_has_account(sk))
1464 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1465 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1468 static inline void sk_mem_charge(struct sock *sk, int size)
1470 if (!sk_has_account(sk))
1472 sk->sk_forward_alloc -= size;
1475 static inline void sk_mem_uncharge(struct sock *sk, int size)
1477 if (!sk_has_account(sk))
1479 sk->sk_forward_alloc += size;
1481 /* Avoid a possible overflow.
1482 * TCP send queues can make this happen, if sk_mem_reclaim()
1483 * is not called and more than 2 GBytes are released at once.
1485 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1486 * no need to hold that much forward allocation anyway.
1488 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1489 __sk_mem_reclaim(sk, 1 << 20);
1492 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1494 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1495 sk->sk_wmem_queued -= skb->truesize;
1496 sk_mem_uncharge(sk, skb->truesize);
1500 static inline void sock_release_ownership(struct sock *sk)
1502 if (sk->sk_lock.owned) {
1503 sk->sk_lock.owned = 0;
1505 /* The sk_lock has mutex_unlock() semantics: */
1506 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1511 * Macro so as to not evaluate some arguments when
1512 * lockdep is not enabled.
1514 * Mark both the sk_lock and the sk_lock.slock as a
1515 * per-address-family lock class.
1517 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1519 sk->sk_lock.owned = 0; \
1520 init_waitqueue_head(&sk->sk_lock.wq); \
1521 spin_lock_init(&(sk)->sk_lock.slock); \
1522 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1523 sizeof((sk)->sk_lock)); \
1524 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1526 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1529 #ifdef CONFIG_LOCKDEP
1530 static inline bool lockdep_sock_is_held(const struct sock *sk)
1532 return lockdep_is_held(&sk->sk_lock) ||
1533 lockdep_is_held(&sk->sk_lock.slock);
1537 void lock_sock_nested(struct sock *sk, int subclass);
1539 static inline void lock_sock(struct sock *sk)
1541 lock_sock_nested(sk, 0);
1544 void __release_sock(struct sock *sk);
1545 void release_sock(struct sock *sk);
1547 /* BH context may only use the following locking interface. */
1548 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1549 #define bh_lock_sock_nested(__sk) \
1550 spin_lock_nested(&((__sk)->sk_lock.slock), \
1551 SINGLE_DEPTH_NESTING)
1552 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1554 bool lock_sock_fast(struct sock *sk);
1556 * unlock_sock_fast - complement of lock_sock_fast
1560 * fast unlock socket for user context.
1561 * If slow mode is on, we call regular release_sock()
1563 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1568 spin_unlock_bh(&sk->sk_lock.slock);
1571 /* Used by processes to "lock" a socket state, so that
1572 * interrupts and bottom half handlers won't change it
1573 * from under us. It essentially blocks any incoming
1574 * packets, so that we won't get any new data or any
1575 * packets that change the state of the socket.
1577 * While locked, BH processing will add new packets to
1578 * the backlog queue. This queue is processed by the
1579 * owner of the socket lock right before it is released.
1581 * Since ~2.3.5 it is also exclusive sleep lock serializing
1582 * accesses from user process context.
1585 static inline void sock_owned_by_me(const struct sock *sk)
1587 #ifdef CONFIG_LOCKDEP
1588 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1592 static inline bool sock_owned_by_user(const struct sock *sk)
1594 sock_owned_by_me(sk);
1595 return sk->sk_lock.owned;
1598 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1600 return sk->sk_lock.owned;
1603 /* no reclassification while locks are held */
1604 static inline bool sock_allow_reclassification(const struct sock *csk)
1606 struct sock *sk = (struct sock *)csk;
1608 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1611 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1612 struct proto *prot, int kern);
1613 void sk_free(struct sock *sk);
1614 void sk_destruct(struct sock *sk);
1615 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1616 void sk_free_unlock_clone(struct sock *sk);
1618 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1620 void __sock_wfree(struct sk_buff *skb);
1621 void sock_wfree(struct sk_buff *skb);
1622 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1624 void skb_orphan_partial(struct sk_buff *skb);
1625 void sock_rfree(struct sk_buff *skb);
1626 void sock_efree(struct sk_buff *skb);
1628 void sock_edemux(struct sk_buff *skb);
1630 #define sock_edemux sock_efree
1633 int sock_setsockopt(struct socket *sock, int level, int op,
1634 char __user *optval, unsigned int optlen);
1636 int sock_getsockopt(struct socket *sock, int level, int op,
1637 char __user *optval, int __user *optlen);
1638 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1639 int noblock, int *errcode);
1640 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1641 unsigned long data_len, int noblock,
1642 int *errcode, int max_page_order);
1643 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1644 void sock_kfree_s(struct sock *sk, void *mem, int size);
1645 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1646 void sk_send_sigurg(struct sock *sk);
1648 struct sockcm_cookie {
1654 static inline void sockcm_init(struct sockcm_cookie *sockc,
1655 const struct sock *sk)
1657 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1660 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1661 struct sockcm_cookie *sockc);
1662 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1663 struct sockcm_cookie *sockc);
1666 * Functions to fill in entries in struct proto_ops when a protocol
1667 * does not implement a particular function.
1669 int sock_no_bind(struct socket *, struct sockaddr *, int);
1670 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1671 int sock_no_socketpair(struct socket *, struct socket *);
1672 int sock_no_accept(struct socket *, struct socket *, int, bool);
1673 int sock_no_getname(struct socket *, struct sockaddr *, int);
1674 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1675 int sock_no_listen(struct socket *, int);
1676 int sock_no_shutdown(struct socket *, int);
1677 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1678 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1679 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1680 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1681 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1682 int sock_no_mmap(struct file *file, struct socket *sock,
1683 struct vm_area_struct *vma);
1684 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1685 size_t size, int flags);
1686 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1687 int offset, size_t size, int flags);
1690 * Functions to fill in entries in struct proto_ops when a protocol
1691 * uses the inet style.
1693 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1694 char __user *optval, int __user *optlen);
1695 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1697 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1698 char __user *optval, unsigned int optlen);
1699 int compat_sock_common_getsockopt(struct socket *sock, int level,
1700 int optname, char __user *optval, int __user *optlen);
1701 int compat_sock_common_setsockopt(struct socket *sock, int level,
1702 int optname, char __user *optval, unsigned int optlen);
1704 void sk_common_release(struct sock *sk);
1707 * Default socket callbacks and setup code
1710 /* Initialise core socket variables */
1711 void sock_init_data(struct socket *sock, struct sock *sk);
1714 * Socket reference counting postulates.
1716 * * Each user of socket SHOULD hold a reference count.
1717 * * Each access point to socket (an hash table bucket, reference from a list,
1718 * running timer, skb in flight MUST hold a reference count.
1719 * * When reference count hits 0, it means it will never increase back.
1720 * * When reference count hits 0, it means that no references from
1721 * outside exist to this socket and current process on current CPU
1722 * is last user and may/should destroy this socket.
1723 * * sk_free is called from any context: process, BH, IRQ. When
1724 * it is called, socket has no references from outside -> sk_free
1725 * may release descendant resources allocated by the socket, but
1726 * to the time when it is called, socket is NOT referenced by any
1727 * hash tables, lists etc.
1728 * * Packets, delivered from outside (from network or from another process)
1729 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1730 * when they sit in queue. Otherwise, packets will leak to hole, when
1731 * socket is looked up by one cpu and unhasing is made by another CPU.
1732 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1733 * (leak to backlog). Packet socket does all the processing inside
1734 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1735 * use separate SMP lock, so that they are prone too.
1738 /* Ungrab socket and destroy it, if it was the last reference. */
1739 static inline void sock_put(struct sock *sk)
1741 if (refcount_dec_and_test(&sk->sk_refcnt))
1744 /* Generic version of sock_put(), dealing with all sockets
1745 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1747 void sock_gen_put(struct sock *sk);
1749 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1750 unsigned int trim_cap, bool refcounted);
1751 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1754 return __sk_receive_skb(sk, skb, nested, 1, true);
1757 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1759 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1760 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1762 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1763 * other WRITE_ONCE() because socket lock might be not held.
1765 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1768 #define NO_QUEUE_MAPPING USHRT_MAX
1770 static inline void sk_tx_queue_clear(struct sock *sk)
1772 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1773 * other WRITE_ONCE() because socket lock might be not held.
1775 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1778 static inline int sk_tx_queue_get(const struct sock *sk)
1781 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
1782 * and sk_tx_queue_set().
1784 int val = READ_ONCE(sk->sk_tx_queue_mapping);
1786 if (val != NO_QUEUE_MAPPING)
1792 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1795 if (skb_rx_queue_recorded(skb)) {
1796 u16 rx_queue = skb_get_rx_queue(skb);
1798 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1801 sk->sk_rx_queue_mapping = rx_queue;
1806 static inline void sk_rx_queue_clear(struct sock *sk)
1809 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1814 static inline int sk_rx_queue_get(const struct sock *sk)
1816 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1817 return sk->sk_rx_queue_mapping;
1823 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1825 sk->sk_socket = sock;
1828 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1830 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1831 return &rcu_dereference_raw(sk->sk_wq)->wait;
1833 /* Detach socket from process context.
1834 * Announce socket dead, detach it from wait queue and inode.
1835 * Note that parent inode held reference count on this struct sock,
1836 * we do not release it in this function, because protocol
1837 * probably wants some additional cleanups or even continuing
1838 * to work with this socket (TCP).
1840 static inline void sock_orphan(struct sock *sk)
1842 write_lock_bh(&sk->sk_callback_lock);
1843 sock_set_flag(sk, SOCK_DEAD);
1844 sk_set_socket(sk, NULL);
1846 write_unlock_bh(&sk->sk_callback_lock);
1849 static inline void sock_graft(struct sock *sk, struct socket *parent)
1851 WARN_ON(parent->sk);
1852 write_lock_bh(&sk->sk_callback_lock);
1853 rcu_assign_pointer(sk->sk_wq, parent->wq);
1855 sk_set_socket(sk, parent);
1856 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1857 security_sock_graft(sk, parent);
1858 write_unlock_bh(&sk->sk_callback_lock);
1861 kuid_t sock_i_uid(struct sock *sk);
1862 unsigned long __sock_i_ino(struct sock *sk);
1863 unsigned long sock_i_ino(struct sock *sk);
1865 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1867 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1870 static inline u32 net_tx_rndhash(void)
1872 u32 v = prandom_u32();
1877 static inline void sk_set_txhash(struct sock *sk)
1879 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
1880 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
1883 static inline void sk_rethink_txhash(struct sock *sk)
1889 static inline struct dst_entry *
1890 __sk_dst_get(struct sock *sk)
1892 return rcu_dereference_check(sk->sk_dst_cache,
1893 lockdep_sock_is_held(sk));
1896 static inline struct dst_entry *
1897 sk_dst_get(struct sock *sk)
1899 struct dst_entry *dst;
1902 dst = rcu_dereference(sk->sk_dst_cache);
1903 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1909 static inline void dst_negative_advice(struct sock *sk)
1911 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1913 sk_rethink_txhash(sk);
1915 if (dst && dst->ops->negative_advice) {
1916 ndst = dst->ops->negative_advice(dst);
1919 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1920 sk_tx_queue_clear(sk);
1921 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
1927 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1929 struct dst_entry *old_dst;
1931 sk_tx_queue_clear(sk);
1932 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
1933 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1934 lockdep_sock_is_held(sk));
1935 rcu_assign_pointer(sk->sk_dst_cache, dst);
1936 dst_release(old_dst);
1940 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1942 struct dst_entry *old_dst;
1944 sk_tx_queue_clear(sk);
1945 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
1946 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1947 dst_release(old_dst);
1951 __sk_dst_reset(struct sock *sk)
1953 __sk_dst_set(sk, NULL);
1957 sk_dst_reset(struct sock *sk)
1959 sk_dst_set(sk, NULL);
1962 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1964 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1966 static inline void sk_dst_confirm(struct sock *sk)
1968 if (!sk->sk_dst_pending_confirm)
1969 sk->sk_dst_pending_confirm = 1;
1972 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1974 if (skb_get_dst_pending_confirm(skb)) {
1975 struct sock *sk = skb->sk;
1976 unsigned long now = jiffies;
1978 /* avoid dirtying neighbour */
1979 if (n->confirmed != now)
1981 if (sk && sk->sk_dst_pending_confirm)
1982 sk->sk_dst_pending_confirm = 0;
1986 bool sk_mc_loop(struct sock *sk);
1988 static inline bool sk_can_gso(const struct sock *sk)
1990 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1993 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1995 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1997 sk->sk_route_nocaps |= flags;
1998 sk->sk_route_caps &= ~flags;
2001 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2002 struct iov_iter *from, char *to,
2003 int copy, int offset)
2005 if (skb->ip_summed == CHECKSUM_NONE) {
2007 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2009 skb->csum = csum_block_add(skb->csum, csum, offset);
2010 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2011 if (!copy_from_iter_full_nocache(to, copy, from))
2013 } else if (!copy_from_iter_full(to, copy, from))
2019 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2020 struct iov_iter *from, int copy)
2022 int err, offset = skb->len;
2024 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2027 __skb_trim(skb, offset);
2032 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2033 struct sk_buff *skb,
2039 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2045 skb->data_len += copy;
2046 skb->truesize += copy;
2047 sk->sk_wmem_queued += copy;
2048 sk_mem_charge(sk, copy);
2053 * sk_wmem_alloc_get - returns write allocations
2056 * Returns sk_wmem_alloc minus initial offset of one
2058 static inline int sk_wmem_alloc_get(const struct sock *sk)
2060 return refcount_read(&sk->sk_wmem_alloc) - 1;
2064 * sk_rmem_alloc_get - returns read allocations
2067 * Returns sk_rmem_alloc
2069 static inline int sk_rmem_alloc_get(const struct sock *sk)
2071 return atomic_read(&sk->sk_rmem_alloc);
2075 * sk_has_allocations - check if allocations are outstanding
2078 * Returns true if socket has write or read allocations
2080 static inline bool sk_has_allocations(const struct sock *sk)
2082 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2086 * skwq_has_sleeper - check if there are any waiting processes
2087 * @wq: struct socket_wq
2089 * Returns true if socket_wq has waiting processes
2091 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2092 * barrier call. They were added due to the race found within the tcp code.
2094 * Consider following tcp code paths::
2097 * sys_select receive packet
2099 * __add_wait_queue update tp->rcv_nxt
2101 * tp->rcv_nxt check sock_def_readable
2103 * schedule rcu_read_lock();
2104 * wq = rcu_dereference(sk->sk_wq);
2105 * if (wq && waitqueue_active(&wq->wait))
2106 * wake_up_interruptible(&wq->wait)
2110 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2111 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2112 * could then endup calling schedule and sleep forever if there are no more
2113 * data on the socket.
2116 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2118 return wq && wq_has_sleeper(&wq->wait);
2122 * sock_poll_wait - place memory barrier behind the poll_wait call.
2124 * @sock: socket to wait on
2127 * See the comments in the wq_has_sleeper function.
2129 * Do not derive sock from filp->private_data here. An SMC socket establishes
2130 * an internal TCP socket that is used in the fallback case. All socket
2131 * operations on the SMC socket are then forwarded to the TCP socket. In case of
2132 * poll, the filp->private_data pointer references the SMC socket because the
2133 * TCP socket has no file assigned.
2135 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2138 if (!poll_does_not_wait(p)) {
2139 poll_wait(filp, &sock->wq->wait, p);
2140 /* We need to be sure we are in sync with the
2141 * socket flags modification.
2143 * This memory barrier is paired in the wq_has_sleeper.
2149 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2151 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2152 u32 txhash = READ_ONCE(sk->sk_txhash);
2160 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2163 * Queue a received datagram if it will fit. Stream and sequenced
2164 * protocols can't normally use this as they need to fit buffers in
2165 * and play with them.
2167 * Inlined as it's very short and called for pretty much every
2168 * packet ever received.
2170 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2174 skb->destructor = sock_rfree;
2175 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2176 sk_mem_charge(sk, skb->truesize);
2179 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2181 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2183 if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2184 skb_set_owner_r(skb, sk);
2192 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2193 unsigned long expires);
2195 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2197 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2198 struct sk_buff *skb, unsigned int flags,
2199 void (*destructor)(struct sock *sk,
2200 struct sk_buff *skb));
2201 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2202 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2204 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2205 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2208 * Recover an error report and clear atomically
2211 static inline int sock_error(struct sock *sk)
2214 if (likely(!sk->sk_err))
2216 err = xchg(&sk->sk_err, 0);
2220 static inline unsigned long sock_wspace(struct sock *sk)
2224 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2225 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2233 * We use sk->sk_wq_raw, from contexts knowing this
2234 * pointer is not NULL and cannot disappear/change.
2236 static inline void sk_set_bit(int nr, struct sock *sk)
2238 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2239 !sock_flag(sk, SOCK_FASYNC))
2242 set_bit(nr, &sk->sk_wq_raw->flags);
2245 static inline void sk_clear_bit(int nr, struct sock *sk)
2247 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2248 !sock_flag(sk, SOCK_FASYNC))
2251 clear_bit(nr, &sk->sk_wq_raw->flags);
2254 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2256 if (sock_flag(sk, SOCK_FASYNC)) {
2258 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2263 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2264 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2265 * Note: for send buffers, TCP works better if we can build two skbs at
2268 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2270 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2271 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2273 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2275 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2276 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2277 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2281 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2282 bool force_schedule);
2285 * sk_page_frag - return an appropriate page_frag
2288 * Use the per task page_frag instead of the per socket one for
2289 * optimization when we know that we're in the normal context and owns
2290 * everything that's associated with %current.
2292 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2293 * inside other socket operations and end up recursing into sk_page_frag()
2294 * while it's already in use.
2296 static inline struct page_frag *sk_page_frag(struct sock *sk)
2298 if (gfpflags_normal_context(sk->sk_allocation))
2299 return ¤t->task_frag;
2301 return &sk->sk_frag;
2304 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2306 int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2307 int sg_start, int *sg_curr, unsigned int *sg_size,
2308 int first_coalesce);
2311 * Default write policy as shown to user space via poll/select/SIGIO
2313 static inline bool sock_writeable(const struct sock *sk)
2315 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2318 static inline gfp_t gfp_any(void)
2320 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2323 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2325 return noblock ? 0 : sk->sk_rcvtimeo;
2328 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2330 return noblock ? 0 : sk->sk_sndtimeo;
2333 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2335 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2338 /* Alas, with timeout socket operations are not restartable.
2339 * Compare this to poll().
2341 static inline int sock_intr_errno(long timeo)
2343 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2346 struct sock_skb_cb {
2350 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2351 * using skb->cb[] would keep using it directly and utilize its
2352 * alignement guarantee.
2354 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2355 sizeof(struct sock_skb_cb)))
2357 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2358 SOCK_SKB_CB_OFFSET))
2360 #define sock_skb_cb_check_size(size) \
2361 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2364 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2366 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2367 atomic_read(&sk->sk_drops) : 0;
2370 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2372 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2374 atomic_add(segs, &sk->sk_drops);
2377 static inline ktime_t sock_read_timestamp(struct sock *sk)
2379 #if BITS_PER_LONG==32
2384 seq = read_seqbegin(&sk->sk_stamp_seq);
2386 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2390 return READ_ONCE(sk->sk_stamp);
2394 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2396 #if BITS_PER_LONG==32
2397 write_seqlock(&sk->sk_stamp_seq);
2399 write_sequnlock(&sk->sk_stamp_seq);
2401 WRITE_ONCE(sk->sk_stamp, kt);
2405 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2406 struct sk_buff *skb);
2407 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2408 struct sk_buff *skb);
2411 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2413 ktime_t kt = skb->tstamp;
2414 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2417 * generate control messages if
2418 * - receive time stamping in software requested
2419 * - software time stamp available and wanted
2420 * - hardware time stamps available and wanted
2422 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2423 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2424 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2425 (hwtstamps->hwtstamp &&
2426 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2427 __sock_recv_timestamp(msg, sk, skb);
2429 sock_write_timestamp(sk, kt);
2431 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2432 __sock_recv_wifi_status(msg, sk, skb);
2435 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2436 struct sk_buff *skb);
2438 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2439 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2440 struct sk_buff *skb)
2442 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2443 (1UL << SOCK_RCVTSTAMP))
2444 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2445 SOF_TIMESTAMPING_RAW_HARDWARE)
2447 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2448 __sock_recv_ts_and_drops(msg, sk, skb);
2449 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2450 sock_write_timestamp(sk, skb->tstamp);
2451 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2452 sock_write_timestamp(sk, 0);
2455 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2458 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2459 * @sk: socket sending this packet
2460 * @tsflags: timestamping flags to use
2461 * @tx_flags: completed with instructions for time stamping
2462 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2464 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2466 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2467 __u8 *tx_flags, __u32 *tskey)
2469 if (unlikely(tsflags)) {
2470 __sock_tx_timestamp(tsflags, tx_flags);
2471 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2472 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2473 *tskey = sk->sk_tskey++;
2475 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2476 *tx_flags |= SKBTX_WIFI_STATUS;
2479 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2482 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2485 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2487 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2488 &skb_shinfo(skb)->tskey);
2492 * sk_eat_skb - Release a skb if it is no longer needed
2493 * @sk: socket to eat this skb from
2494 * @skb: socket buffer to eat
2496 * This routine must be called with interrupts disabled or with the socket
2497 * locked so that the sk_buff queue operation is ok.
2499 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2501 __skb_unlink(skb, &sk->sk_receive_queue);
2506 struct net *sock_net(const struct sock *sk)
2508 return read_pnet(&sk->sk_net);
2512 void sock_net_set(struct sock *sk, struct net *net)
2514 write_pnet(&sk->sk_net, net);
2517 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2520 struct sock *sk = skb->sk;
2522 skb->destructor = NULL;
2529 /* This helper checks if a socket is a full socket,
2530 * ie _not_ a timewait or request socket.
2532 static inline bool sk_fullsock(const struct sock *sk)
2534 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2537 /* Checks if this SKB belongs to an HW offloaded socket
2538 * and whether any SW fallbacks are required based on dev.
2540 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2541 struct net_device *dev)
2543 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2544 struct sock *sk = skb->sk;
2546 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb)
2547 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2553 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2554 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2556 static inline bool sk_listener(const struct sock *sk)
2558 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2561 void sock_enable_timestamp(struct sock *sk, int flag);
2562 int sock_get_timestamp(struct sock *, struct timeval __user *);
2563 int sock_get_timestampns(struct sock *, struct timespec __user *);
2564 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2567 bool sk_ns_capable(const struct sock *sk,
2568 struct user_namespace *user_ns, int cap);
2569 bool sk_capable(const struct sock *sk, int cap);
2570 bool sk_net_capable(const struct sock *sk, int cap);
2572 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2574 /* Take into consideration the size of the struct sk_buff overhead in the
2575 * determination of these values, since that is non-constant across
2576 * platforms. This makes socket queueing behavior and performance
2577 * not depend upon such differences.
2579 #define _SK_MEM_PACKETS 256
2580 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2581 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2582 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2584 extern __u32 sysctl_wmem_max;
2585 extern __u32 sysctl_rmem_max;
2587 extern int sysctl_tstamp_allow_data;
2588 extern int sysctl_optmem_max;
2590 extern __u32 sysctl_wmem_default;
2591 extern __u32 sysctl_rmem_default;
2593 /* On 32bit arches, an skb frag is limited to 2^15 */
2594 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2596 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2598 /* Does this proto have per netns sysctl_wmem ? */
2599 if (proto->sysctl_wmem_offset)
2600 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2602 return *proto->sysctl_wmem;
2605 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2607 /* Does this proto have per netns sysctl_rmem ? */
2608 if (proto->sysctl_rmem_offset)
2609 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2611 return *proto->sysctl_rmem;
2614 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2615 * Some wifi drivers need to tweak it to get more chunks.
2616 * They can use this helper from their ndo_start_xmit()
2618 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2620 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2622 sk->sk_pacing_shift = val;
2625 /* if a socket is bound to a device, check that the given device
2626 * index is either the same or that the socket is bound to an L3
2627 * master device and the given device index is also enslaved to
2630 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2634 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2637 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2638 if (mdif && mdif == sk->sk_bound_dev_if)
2644 #endif /* _SOCK_H */