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_tsflags: SO_TIMESTAMPING socket options
302 * @sk_tskey: counter to disambiguate concurrent tstamp requests
303 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
304 * @sk_socket: Identd and reporting IO signals
305 * @sk_user_data: RPC layer private data
306 * @sk_frag: cached page frag
307 * @sk_peek_off: current peek_offset value
308 * @sk_send_head: front of stuff to transmit
309 * @sk_security: used by security modules
310 * @sk_mark: generic packet mark
311 * @sk_cgrp_data: cgroup data for this cgroup
312 * @sk_memcg: this socket's memory cgroup association
313 * @sk_write_pending: a write to stream socket waits to start
314 * @sk_state_change: callback to indicate change in the state of the sock
315 * @sk_data_ready: callback to indicate there is data to be processed
316 * @sk_write_space: callback to indicate there is bf sending space available
317 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
318 * @sk_backlog_rcv: callback to process the backlog
319 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
320 * @sk_reuseport_cb: reuseport group container
321 * @sk_rcu: used during RCU grace period
322 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
323 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
324 * @sk_txtime_unused: unused txtime flags
328 * Now struct inet_timewait_sock also uses sock_common, so please just
329 * don't add nothing before this first member (__sk_common) --acme
331 struct sock_common __sk_common;
332 #define sk_node __sk_common.skc_node
333 #define sk_nulls_node __sk_common.skc_nulls_node
334 #define sk_refcnt __sk_common.skc_refcnt
335 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
337 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
340 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
341 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
342 #define sk_hash __sk_common.skc_hash
343 #define sk_portpair __sk_common.skc_portpair
344 #define sk_num __sk_common.skc_num
345 #define sk_dport __sk_common.skc_dport
346 #define sk_addrpair __sk_common.skc_addrpair
347 #define sk_daddr __sk_common.skc_daddr
348 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
349 #define sk_family __sk_common.skc_family
350 #define sk_state __sk_common.skc_state
351 #define sk_reuse __sk_common.skc_reuse
352 #define sk_reuseport __sk_common.skc_reuseport
353 #define sk_ipv6only __sk_common.skc_ipv6only
354 #define sk_net_refcnt __sk_common.skc_net_refcnt
355 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
356 #define sk_bind_node __sk_common.skc_bind_node
357 #define sk_prot __sk_common.skc_prot
358 #define sk_net __sk_common.skc_net
359 #define sk_v6_daddr __sk_common.skc_v6_daddr
360 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
361 #define sk_cookie __sk_common.skc_cookie
362 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
363 #define sk_flags __sk_common.skc_flags
364 #define sk_rxhash __sk_common.skc_rxhash
366 socket_lock_t sk_lock;
369 struct sk_buff_head sk_error_queue;
370 struct sk_buff_head sk_receive_queue;
372 * The backlog queue is special, it is always used with
373 * the per-socket spinlock held and requires low latency
374 * access. Therefore we special case it's implementation.
375 * Note : rmem_alloc is in this structure to fill a hole
376 * on 64bit arches, not because its logically part of
382 struct sk_buff *head;
383 struct sk_buff *tail;
385 #define sk_rmem_alloc sk_backlog.rmem_alloc
387 int sk_forward_alloc;
388 #ifdef CONFIG_NET_RX_BUSY_POLL
389 unsigned int sk_ll_usec;
390 /* ===== mostly read cache line ===== */
391 unsigned int sk_napi_id;
395 struct sk_filter __rcu *sk_filter;
397 struct socket_wq __rcu *sk_wq;
398 struct socket_wq *sk_wq_raw;
401 struct xfrm_policy __rcu *sk_policy[2];
403 struct dst_entry *sk_rx_dst;
404 struct dst_entry __rcu *sk_dst_cache;
405 atomic_t sk_omem_alloc;
408 /* ===== cache line for TX ===== */
410 refcount_t sk_wmem_alloc;
411 unsigned long sk_tsq_flags;
413 struct sk_buff *sk_send_head;
414 struct rb_root tcp_rtx_queue;
416 struct sk_buff_head sk_write_queue;
418 int sk_write_pending;
419 __u32 sk_dst_pending_confirm;
420 u32 sk_pacing_status; /* see enum sk_pacing */
422 struct timer_list sk_timer;
425 unsigned long sk_pacing_rate; /* bytes per second */
426 unsigned long sk_max_pacing_rate;
427 struct page_frag sk_frag;
428 netdev_features_t sk_route_caps;
429 netdev_features_t sk_route_nocaps;
430 netdev_features_t sk_route_forced_caps;
432 unsigned int sk_gso_max_size;
437 * Because of non atomicity rules, all
438 * changes are protected by socket lock.
440 unsigned int __sk_flags_offset[0];
441 #ifdef __BIG_ENDIAN_BITFIELD
442 #define SK_FL_PROTO_SHIFT 16
443 #define SK_FL_PROTO_MASK 0x00ff0000
445 #define SK_FL_TYPE_SHIFT 0
446 #define SK_FL_TYPE_MASK 0x0000ffff
448 #define SK_FL_PROTO_SHIFT 8
449 #define SK_FL_PROTO_MASK 0x0000ff00
451 #define SK_FL_TYPE_SHIFT 16
452 #define SK_FL_TYPE_MASK 0xffff0000
455 unsigned int sk_padding : 1,
462 #define SK_PROTOCOL_MAX U8_MAX
465 unsigned long sk_lingertime;
466 struct proto *sk_prot_creator;
467 rwlock_t sk_callback_lock;
471 u32 sk_max_ack_backlog;
473 struct pid *sk_peer_pid;
474 const struct cred *sk_peer_cred;
483 u8 sk_txtime_deadline_mode : 1,
484 sk_txtime_report_errors : 1,
485 sk_txtime_unused : 6;
487 struct socket *sk_socket;
489 #ifdef CONFIG_SECURITY
492 struct sock_cgroup_data sk_cgrp_data;
493 struct mem_cgroup *sk_memcg;
494 void (*sk_state_change)(struct sock *sk);
495 void (*sk_data_ready)(struct sock *sk);
496 void (*sk_write_space)(struct sock *sk);
497 void (*sk_error_report)(struct sock *sk);
498 int (*sk_backlog_rcv)(struct sock *sk,
499 struct sk_buff *skb);
500 #ifdef CONFIG_SOCK_VALIDATE_XMIT
501 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
502 struct net_device *dev,
503 struct sk_buff *skb);
505 void (*sk_destruct)(struct sock *sk);
506 struct sock_reuseport __rcu *sk_reuseport_cb;
507 struct rcu_head sk_rcu;
512 SK_PACING_NEEDED = 1,
516 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
518 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
519 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
522 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
523 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
524 * on a socket means that the socket will reuse everybody else's port
525 * without looking at the other's sk_reuse value.
528 #define SK_NO_REUSE 0
529 #define SK_CAN_REUSE 1
530 #define SK_FORCE_REUSE 2
532 int sk_set_peek_off(struct sock *sk, int val);
534 static inline int sk_peek_offset(struct sock *sk, int flags)
536 if (unlikely(flags & MSG_PEEK)) {
537 return READ_ONCE(sk->sk_peek_off);
543 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
545 s32 off = READ_ONCE(sk->sk_peek_off);
547 if (unlikely(off >= 0)) {
548 off = max_t(s32, off - val, 0);
549 WRITE_ONCE(sk->sk_peek_off, off);
553 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
555 sk_peek_offset_bwd(sk, -val);
559 * Hashed lists helper routines
561 static inline struct sock *sk_entry(const struct hlist_node *node)
563 return hlist_entry(node, struct sock, sk_node);
566 static inline struct sock *__sk_head(const struct hlist_head *head)
568 return hlist_entry(head->first, struct sock, sk_node);
571 static inline struct sock *sk_head(const struct hlist_head *head)
573 return hlist_empty(head) ? NULL : __sk_head(head);
576 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
578 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
581 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
583 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
586 static inline struct sock *sk_next(const struct sock *sk)
588 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
591 static inline struct sock *sk_nulls_next(const struct sock *sk)
593 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
594 hlist_nulls_entry(sk->sk_nulls_node.next,
595 struct sock, sk_nulls_node) :
599 static inline bool sk_unhashed(const struct sock *sk)
601 return hlist_unhashed(&sk->sk_node);
604 static inline bool sk_hashed(const struct sock *sk)
606 return !sk_unhashed(sk);
609 static inline void sk_node_init(struct hlist_node *node)
614 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
619 static inline void __sk_del_node(struct sock *sk)
621 __hlist_del(&sk->sk_node);
624 /* NB: equivalent to hlist_del_init_rcu */
625 static inline bool __sk_del_node_init(struct sock *sk)
629 sk_node_init(&sk->sk_node);
635 /* Grab socket reference count. This operation is valid only
636 when sk is ALREADY grabbed f.e. it is found in hash table
637 or a list and the lookup is made under lock preventing hash table
641 static __always_inline void sock_hold(struct sock *sk)
643 refcount_inc(&sk->sk_refcnt);
646 /* Ungrab socket in the context, which assumes that socket refcnt
647 cannot hit zero, f.e. it is true in context of any socketcall.
649 static __always_inline void __sock_put(struct sock *sk)
651 refcount_dec(&sk->sk_refcnt);
654 static inline bool sk_del_node_init(struct sock *sk)
656 bool rc = __sk_del_node_init(sk);
659 /* paranoid for a while -acme */
660 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
665 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
667 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
670 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
676 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
678 bool rc = __sk_nulls_del_node_init_rcu(sk);
681 /* paranoid for a while -acme */
682 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
688 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
690 hlist_add_head(&sk->sk_node, list);
693 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
696 __sk_add_node(sk, list);
699 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
702 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
703 sk->sk_family == AF_INET6)
704 hlist_add_tail_rcu(&sk->sk_node, list);
706 hlist_add_head_rcu(&sk->sk_node, list);
709 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
711 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
714 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
717 __sk_nulls_add_node_rcu(sk, list);
720 static inline void __sk_del_bind_node(struct sock *sk)
722 __hlist_del(&sk->sk_bind_node);
725 static inline void sk_add_bind_node(struct sock *sk,
726 struct hlist_head *list)
728 hlist_add_head(&sk->sk_bind_node, list);
731 #define sk_for_each(__sk, list) \
732 hlist_for_each_entry(__sk, list, sk_node)
733 #define sk_for_each_rcu(__sk, list) \
734 hlist_for_each_entry_rcu(__sk, list, sk_node)
735 #define sk_nulls_for_each(__sk, node, list) \
736 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
737 #define sk_nulls_for_each_rcu(__sk, node, list) \
738 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
739 #define sk_for_each_from(__sk) \
740 hlist_for_each_entry_from(__sk, sk_node)
741 #define sk_nulls_for_each_from(__sk, node) \
742 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
743 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
744 #define sk_for_each_safe(__sk, tmp, list) \
745 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
746 #define sk_for_each_bound(__sk, list) \
747 hlist_for_each_entry(__sk, list, sk_bind_node)
750 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
751 * @tpos: the type * to use as a loop cursor.
752 * @pos: the &struct hlist_node to use as a loop cursor.
753 * @head: the head for your list.
754 * @offset: offset of hlist_node within the struct.
757 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
758 for (pos = rcu_dereference(hlist_first_rcu(head)); \
760 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
761 pos = rcu_dereference(hlist_next_rcu(pos)))
763 static inline struct user_namespace *sk_user_ns(struct sock *sk)
765 /* Careful only use this in a context where these parameters
766 * can not change and must all be valid, such as recvmsg from
769 return sk->sk_socket->file->f_cred->user_ns;
783 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
784 SOCK_DBG, /* %SO_DEBUG setting */
785 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
786 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
787 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
788 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
789 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
790 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
791 SOCK_FASYNC, /* fasync() active */
793 SOCK_ZEROCOPY, /* buffers from userspace */
794 SOCK_WIFI_STATUS, /* push wifi status to userspace */
795 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
796 * Will use last 4 bytes of packet sent from
797 * user-space instead.
799 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
800 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
801 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
803 SOCK_XDP, /* XDP is attached */
806 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
808 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
810 nsk->sk_flags = osk->sk_flags;
813 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
815 __set_bit(flag, &sk->sk_flags);
818 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
820 __clear_bit(flag, &sk->sk_flags);
823 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
825 return test_bit(flag, &sk->sk_flags);
829 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
830 static inline int sk_memalloc_socks(void)
832 return static_branch_unlikely(&memalloc_socks_key);
836 static inline int sk_memalloc_socks(void)
843 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
845 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
848 static inline void sk_acceptq_removed(struct sock *sk)
850 sk->sk_ack_backlog--;
853 static inline void sk_acceptq_added(struct sock *sk)
855 sk->sk_ack_backlog++;
858 static inline bool sk_acceptq_is_full(const struct sock *sk)
860 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
864 * Compute minimal free write space needed to queue new packets.
866 static inline int sk_stream_min_wspace(const struct sock *sk)
868 return sk->sk_wmem_queued >> 1;
871 static inline int sk_stream_wspace(const struct sock *sk)
873 return sk->sk_sndbuf - sk->sk_wmem_queued;
876 void sk_stream_write_space(struct sock *sk);
878 /* OOB backlog add */
879 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
881 /* dont let skb dst not refcounted, we are going to leave rcu lock */
884 if (!sk->sk_backlog.tail)
885 sk->sk_backlog.head = skb;
887 sk->sk_backlog.tail->next = skb;
889 sk->sk_backlog.tail = skb;
894 * Take into account size of receive queue and backlog queue
895 * Do not take into account this skb truesize,
896 * to allow even a single big packet to come.
898 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
900 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
902 return qsize > limit;
905 /* The per-socket spinlock must be held here. */
906 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
909 if (sk_rcvqueues_full(sk, limit))
913 * If the skb was allocated from pfmemalloc reserves, only
914 * allow SOCK_MEMALLOC sockets to use it as this socket is
915 * helping free memory
917 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
920 __sk_add_backlog(sk, skb);
921 sk->sk_backlog.len += skb->truesize;
925 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
927 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
929 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
930 return __sk_backlog_rcv(sk, skb);
932 return sk->sk_backlog_rcv(sk, skb);
935 static inline void sk_incoming_cpu_update(struct sock *sk)
937 int cpu = raw_smp_processor_id();
939 if (unlikely(sk->sk_incoming_cpu != cpu))
940 sk->sk_incoming_cpu = cpu;
943 static inline void sock_rps_record_flow_hash(__u32 hash)
946 struct rps_sock_flow_table *sock_flow_table;
949 sock_flow_table = rcu_dereference(rps_sock_flow_table);
950 rps_record_sock_flow(sock_flow_table, hash);
955 static inline void sock_rps_record_flow(const struct sock *sk)
958 if (static_key_false(&rfs_needed)) {
959 /* Reading sk->sk_rxhash might incur an expensive cache line
962 * TCP_ESTABLISHED does cover almost all states where RFS
963 * might be useful, and is cheaper [1] than testing :
964 * IPv4: inet_sk(sk)->inet_daddr
965 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
966 * OR an additional socket flag
967 * [1] : sk_state and sk_prot are in the same cache line.
969 if (sk->sk_state == TCP_ESTABLISHED)
970 sock_rps_record_flow_hash(sk->sk_rxhash);
975 static inline void sock_rps_save_rxhash(struct sock *sk,
976 const struct sk_buff *skb)
979 if (unlikely(sk->sk_rxhash != skb->hash))
980 sk->sk_rxhash = skb->hash;
984 static inline void sock_rps_reset_rxhash(struct sock *sk)
991 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
993 release_sock(__sk); \
994 __rc = __condition; \
996 *(__timeo) = wait_woken(__wait, \
997 TASK_INTERRUPTIBLE, \
1000 sched_annotate_sleep(); \
1002 __rc = __condition; \
1006 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1007 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1008 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1009 int sk_stream_error(struct sock *sk, int flags, int err);
1010 void sk_stream_kill_queues(struct sock *sk);
1011 void sk_set_memalloc(struct sock *sk);
1012 void sk_clear_memalloc(struct sock *sk);
1014 void __sk_flush_backlog(struct sock *sk);
1016 static inline bool sk_flush_backlog(struct sock *sk)
1018 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1019 __sk_flush_backlog(sk);
1025 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1027 struct request_sock_ops;
1028 struct timewait_sock_ops;
1029 struct inet_hashinfo;
1030 struct raw_hashinfo;
1031 struct smc_hashinfo;
1035 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1036 * un-modified. Special care is taken when initializing object to zero.
1038 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1040 if (offsetof(struct sock, sk_node.next) != 0)
1041 memset(sk, 0, offsetof(struct sock, sk_node.next));
1042 memset(&sk->sk_node.pprev, 0,
1043 size - offsetof(struct sock, sk_node.pprev));
1046 /* Networking protocol blocks we attach to sockets.
1047 * socket layer -> transport layer interface
1050 void (*close)(struct sock *sk,
1052 int (*pre_connect)(struct sock *sk,
1053 struct sockaddr *uaddr,
1055 int (*connect)(struct sock *sk,
1056 struct sockaddr *uaddr,
1058 int (*disconnect)(struct sock *sk, int flags);
1060 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1063 int (*ioctl)(struct sock *sk, int cmd,
1065 int (*init)(struct sock *sk);
1066 void (*destroy)(struct sock *sk);
1067 void (*shutdown)(struct sock *sk, int how);
1068 int (*setsockopt)(struct sock *sk, int level,
1069 int optname, char __user *optval,
1070 unsigned int optlen);
1071 int (*getsockopt)(struct sock *sk, int level,
1072 int optname, char __user *optval,
1073 int __user *option);
1074 void (*keepalive)(struct sock *sk, int valbool);
1075 #ifdef CONFIG_COMPAT
1076 int (*compat_setsockopt)(struct sock *sk,
1078 int optname, char __user *optval,
1079 unsigned int optlen);
1080 int (*compat_getsockopt)(struct sock *sk,
1082 int optname, char __user *optval,
1083 int __user *option);
1084 int (*compat_ioctl)(struct sock *sk,
1085 unsigned int cmd, unsigned long arg);
1087 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1089 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1090 size_t len, int noblock, int flags,
1092 int (*sendpage)(struct sock *sk, struct page *page,
1093 int offset, size_t size, int flags);
1094 int (*bind)(struct sock *sk,
1095 struct sockaddr *uaddr, int addr_len);
1097 int (*backlog_rcv) (struct sock *sk,
1098 struct sk_buff *skb);
1100 void (*release_cb)(struct sock *sk);
1102 /* Keeping track of sk's, looking them up, and port selection methods. */
1103 int (*hash)(struct sock *sk);
1104 void (*unhash)(struct sock *sk);
1105 void (*rehash)(struct sock *sk);
1106 int (*get_port)(struct sock *sk, unsigned short snum);
1108 /* Keeping track of sockets in use */
1109 #ifdef CONFIG_PROC_FS
1110 unsigned int inuse_idx;
1113 bool (*stream_memory_free)(const struct sock *sk, int wake);
1114 bool (*stream_memory_read)(const struct sock *sk);
1115 /* Memory pressure */
1116 void (*enter_memory_pressure)(struct sock *sk);
1117 void (*leave_memory_pressure)(struct sock *sk);
1118 atomic_long_t *memory_allocated; /* Current allocated memory. */
1119 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1121 * Pressure flag: try to collapse.
1122 * Technical note: it is used by multiple contexts non atomically.
1123 * All the __sk_mem_schedule() is of this nature: accounting
1124 * is strict, actions are advisory and have some latency.
1126 unsigned long *memory_pressure;
1131 u32 sysctl_wmem_offset;
1132 u32 sysctl_rmem_offset;
1137 struct kmem_cache *slab;
1138 unsigned int obj_size;
1139 slab_flags_t slab_flags;
1140 unsigned int useroffset; /* Usercopy region offset */
1141 unsigned int usersize; /* Usercopy region size */
1143 struct percpu_counter *orphan_count;
1145 struct request_sock_ops *rsk_prot;
1146 struct timewait_sock_ops *twsk_prot;
1149 struct inet_hashinfo *hashinfo;
1150 struct udp_table *udp_table;
1151 struct raw_hashinfo *raw_hash;
1152 struct smc_hashinfo *smc_hash;
1155 struct module *owner;
1159 struct list_head node;
1160 #ifdef SOCK_REFCNT_DEBUG
1163 int (*diag_destroy)(struct sock *sk, int err);
1164 } __randomize_layout;
1166 int proto_register(struct proto *prot, int alloc_slab);
1167 void proto_unregister(struct proto *prot);
1168 int sock_load_diag_module(int family, int protocol);
1170 #ifdef SOCK_REFCNT_DEBUG
1171 static inline void sk_refcnt_debug_inc(struct sock *sk)
1173 atomic_inc(&sk->sk_prot->socks);
1176 static inline void sk_refcnt_debug_dec(struct sock *sk)
1178 atomic_dec(&sk->sk_prot->socks);
1179 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1180 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1183 static inline void sk_refcnt_debug_release(const struct sock *sk)
1185 if (refcount_read(&sk->sk_refcnt) != 1)
1186 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1187 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1189 #else /* SOCK_REFCNT_DEBUG */
1190 #define sk_refcnt_debug_inc(sk) do { } while (0)
1191 #define sk_refcnt_debug_dec(sk) do { } while (0)
1192 #define sk_refcnt_debug_release(sk) do { } while (0)
1193 #endif /* SOCK_REFCNT_DEBUG */
1195 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1197 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1200 return sk->sk_prot->stream_memory_free ?
1201 sk->sk_prot->stream_memory_free(sk, wake) : true;
1204 static inline bool sk_stream_memory_free(const struct sock *sk)
1206 return __sk_stream_memory_free(sk, 0);
1209 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1211 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1212 __sk_stream_memory_free(sk, wake);
1215 static inline bool sk_stream_is_writeable(const struct sock *sk)
1217 return __sk_stream_is_writeable(sk, 0);
1220 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1221 struct cgroup *ancestor)
1223 #ifdef CONFIG_SOCK_CGROUP_DATA
1224 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1231 static inline bool sk_has_memory_pressure(const struct sock *sk)
1233 return sk->sk_prot->memory_pressure != NULL;
1236 static inline bool sk_under_memory_pressure(const struct sock *sk)
1238 if (!sk->sk_prot->memory_pressure)
1241 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1242 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1245 return !!*sk->sk_prot->memory_pressure;
1249 sk_memory_allocated(const struct sock *sk)
1251 return atomic_long_read(sk->sk_prot->memory_allocated);
1255 sk_memory_allocated_add(struct sock *sk, int amt)
1257 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1261 sk_memory_allocated_sub(struct sock *sk, int amt)
1263 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1266 static inline void sk_sockets_allocated_dec(struct sock *sk)
1268 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1271 static inline void sk_sockets_allocated_inc(struct sock *sk)
1273 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1277 sk_sockets_allocated_read_positive(struct sock *sk)
1279 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1283 proto_sockets_allocated_sum_positive(struct proto *prot)
1285 return percpu_counter_sum_positive(prot->sockets_allocated);
1289 proto_memory_allocated(struct proto *prot)
1291 return atomic_long_read(prot->memory_allocated);
1295 proto_memory_pressure(struct proto *prot)
1297 if (!prot->memory_pressure)
1299 return !!*prot->memory_pressure;
1303 #ifdef CONFIG_PROC_FS
1304 /* Called with local bh disabled */
1305 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1306 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1307 int sock_inuse_get(struct net *net);
1309 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1316 /* With per-bucket locks this operation is not-atomic, so that
1317 * this version is not worse.
1319 static inline int __sk_prot_rehash(struct sock *sk)
1321 sk->sk_prot->unhash(sk);
1322 return sk->sk_prot->hash(sk);
1325 /* About 10 seconds */
1326 #define SOCK_DESTROY_TIME (10*HZ)
1328 /* Sockets 0-1023 can't be bound to unless you are superuser */
1329 #define PROT_SOCK 1024
1331 #define SHUTDOWN_MASK 3
1332 #define RCV_SHUTDOWN 1
1333 #define SEND_SHUTDOWN 2
1335 #define SOCK_SNDBUF_LOCK 1
1336 #define SOCK_RCVBUF_LOCK 2
1337 #define SOCK_BINDADDR_LOCK 4
1338 #define SOCK_BINDPORT_LOCK 8
1340 struct socket_alloc {
1341 struct socket socket;
1342 struct inode vfs_inode;
1345 static inline struct socket *SOCKET_I(struct inode *inode)
1347 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1350 static inline struct inode *SOCK_INODE(struct socket *socket)
1352 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1356 * Functions for memory accounting
1358 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1359 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1360 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1361 void __sk_mem_reclaim(struct sock *sk, int amount);
1363 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1364 * do not necessarily have 16x time more memory than 4KB ones.
1366 #define SK_MEM_QUANTUM 4096
1367 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1368 #define SK_MEM_SEND 0
1369 #define SK_MEM_RECV 1
1371 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1372 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1374 long val = sk->sk_prot->sysctl_mem[index];
1376 #if PAGE_SIZE > SK_MEM_QUANTUM
1377 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1378 #elif PAGE_SIZE < SK_MEM_QUANTUM
1379 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1384 static inline int sk_mem_pages(int amt)
1386 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1389 static inline bool sk_has_account(struct sock *sk)
1391 /* return true if protocol supports memory accounting */
1392 return !!sk->sk_prot->memory_allocated;
1395 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1397 if (!sk_has_account(sk))
1399 return size <= sk->sk_forward_alloc ||
1400 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1404 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1406 if (!sk_has_account(sk))
1408 return size<= sk->sk_forward_alloc ||
1409 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1410 skb_pfmemalloc(skb);
1413 static inline void sk_mem_reclaim(struct sock *sk)
1415 if (!sk_has_account(sk))
1417 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1418 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1421 static inline void sk_mem_reclaim_partial(struct sock *sk)
1423 if (!sk_has_account(sk))
1425 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1426 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1429 static inline void sk_mem_charge(struct sock *sk, int size)
1431 if (!sk_has_account(sk))
1433 sk->sk_forward_alloc -= size;
1436 static inline void sk_mem_uncharge(struct sock *sk, int size)
1438 if (!sk_has_account(sk))
1440 sk->sk_forward_alloc += size;
1442 /* Avoid a possible overflow.
1443 * TCP send queues can make this happen, if sk_mem_reclaim()
1444 * is not called and more than 2 GBytes are released at once.
1446 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1447 * no need to hold that much forward allocation anyway.
1449 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1450 __sk_mem_reclaim(sk, 1 << 20);
1453 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1455 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1456 sk->sk_wmem_queued -= skb->truesize;
1457 sk_mem_uncharge(sk, skb->truesize);
1461 static inline void sock_release_ownership(struct sock *sk)
1463 if (sk->sk_lock.owned) {
1464 sk->sk_lock.owned = 0;
1466 /* The sk_lock has mutex_unlock() semantics: */
1467 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1472 * Macro so as to not evaluate some arguments when
1473 * lockdep is not enabled.
1475 * Mark both the sk_lock and the sk_lock.slock as a
1476 * per-address-family lock class.
1478 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1480 sk->sk_lock.owned = 0; \
1481 init_waitqueue_head(&sk->sk_lock.wq); \
1482 spin_lock_init(&(sk)->sk_lock.slock); \
1483 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1484 sizeof((sk)->sk_lock)); \
1485 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1487 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1490 #ifdef CONFIG_LOCKDEP
1491 static inline bool lockdep_sock_is_held(const struct sock *sk)
1493 return lockdep_is_held(&sk->sk_lock) ||
1494 lockdep_is_held(&sk->sk_lock.slock);
1498 void lock_sock_nested(struct sock *sk, int subclass);
1500 static inline void lock_sock(struct sock *sk)
1502 lock_sock_nested(sk, 0);
1505 void __release_sock(struct sock *sk);
1506 void release_sock(struct sock *sk);
1508 /* BH context may only use the following locking interface. */
1509 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1510 #define bh_lock_sock_nested(__sk) \
1511 spin_lock_nested(&((__sk)->sk_lock.slock), \
1512 SINGLE_DEPTH_NESTING)
1513 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1515 bool lock_sock_fast(struct sock *sk);
1517 * unlock_sock_fast - complement of lock_sock_fast
1521 * fast unlock socket for user context.
1522 * If slow mode is on, we call regular release_sock()
1524 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1529 spin_unlock_bh(&sk->sk_lock.slock);
1532 /* Used by processes to "lock" a socket state, so that
1533 * interrupts and bottom half handlers won't change it
1534 * from under us. It essentially blocks any incoming
1535 * packets, so that we won't get any new data or any
1536 * packets that change the state of the socket.
1538 * While locked, BH processing will add new packets to
1539 * the backlog queue. This queue is processed by the
1540 * owner of the socket lock right before it is released.
1542 * Since ~2.3.5 it is also exclusive sleep lock serializing
1543 * accesses from user process context.
1546 static inline void sock_owned_by_me(const struct sock *sk)
1548 #ifdef CONFIG_LOCKDEP
1549 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1553 static inline bool sock_owned_by_user(const struct sock *sk)
1555 sock_owned_by_me(sk);
1556 return sk->sk_lock.owned;
1559 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1561 return sk->sk_lock.owned;
1564 /* no reclassification while locks are held */
1565 static inline bool sock_allow_reclassification(const struct sock *csk)
1567 struct sock *sk = (struct sock *)csk;
1569 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1572 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1573 struct proto *prot, int kern);
1574 void sk_free(struct sock *sk);
1575 void sk_destruct(struct sock *sk);
1576 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1577 void sk_free_unlock_clone(struct sock *sk);
1579 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1581 void __sock_wfree(struct sk_buff *skb);
1582 void sock_wfree(struct sk_buff *skb);
1583 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1585 void skb_orphan_partial(struct sk_buff *skb);
1586 void sock_rfree(struct sk_buff *skb);
1587 void sock_efree(struct sk_buff *skb);
1589 void sock_edemux(struct sk_buff *skb);
1591 #define sock_edemux sock_efree
1594 int sock_setsockopt(struct socket *sock, int level, int op,
1595 char __user *optval, unsigned int optlen);
1597 int sock_getsockopt(struct socket *sock, int level, int op,
1598 char __user *optval, int __user *optlen);
1599 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1600 int noblock, int *errcode);
1601 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1602 unsigned long data_len, int noblock,
1603 int *errcode, int max_page_order);
1604 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1605 void sock_kfree_s(struct sock *sk, void *mem, int size);
1606 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1607 void sk_send_sigurg(struct sock *sk);
1609 struct sockcm_cookie {
1615 static inline void sockcm_init(struct sockcm_cookie *sockc,
1616 const struct sock *sk)
1618 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1621 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1622 struct sockcm_cookie *sockc);
1623 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1624 struct sockcm_cookie *sockc);
1627 * Functions to fill in entries in struct proto_ops when a protocol
1628 * does not implement a particular function.
1630 int sock_no_bind(struct socket *, struct sockaddr *, int);
1631 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1632 int sock_no_socketpair(struct socket *, struct socket *);
1633 int sock_no_accept(struct socket *, struct socket *, int, bool);
1634 int sock_no_getname(struct socket *, struct sockaddr *, int);
1635 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1636 int sock_no_listen(struct socket *, int);
1637 int sock_no_shutdown(struct socket *, int);
1638 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1639 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1640 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1641 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1642 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1643 int sock_no_mmap(struct file *file, struct socket *sock,
1644 struct vm_area_struct *vma);
1645 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1646 size_t size, int flags);
1647 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1648 int offset, size_t size, int flags);
1651 * Functions to fill in entries in struct proto_ops when a protocol
1652 * uses the inet style.
1654 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1655 char __user *optval, int __user *optlen);
1656 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1658 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1659 char __user *optval, unsigned int optlen);
1660 int compat_sock_common_getsockopt(struct socket *sock, int level,
1661 int optname, char __user *optval, int __user *optlen);
1662 int compat_sock_common_setsockopt(struct socket *sock, int level,
1663 int optname, char __user *optval, unsigned int optlen);
1665 void sk_common_release(struct sock *sk);
1668 * Default socket callbacks and setup code
1671 /* Initialise core socket variables */
1672 void sock_init_data(struct socket *sock, struct sock *sk);
1675 * Socket reference counting postulates.
1677 * * Each user of socket SHOULD hold a reference count.
1678 * * Each access point to socket (an hash table bucket, reference from a list,
1679 * running timer, skb in flight MUST hold a reference count.
1680 * * When reference count hits 0, it means it will never increase back.
1681 * * When reference count hits 0, it means that no references from
1682 * outside exist to this socket and current process on current CPU
1683 * is last user and may/should destroy this socket.
1684 * * sk_free is called from any context: process, BH, IRQ. When
1685 * it is called, socket has no references from outside -> sk_free
1686 * may release descendant resources allocated by the socket, but
1687 * to the time when it is called, socket is NOT referenced by any
1688 * hash tables, lists etc.
1689 * * Packets, delivered from outside (from network or from another process)
1690 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1691 * when they sit in queue. Otherwise, packets will leak to hole, when
1692 * socket is looked up by one cpu and unhasing is made by another CPU.
1693 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1694 * (leak to backlog). Packet socket does all the processing inside
1695 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1696 * use separate SMP lock, so that they are prone too.
1699 /* Ungrab socket and destroy it, if it was the last reference. */
1700 static inline void sock_put(struct sock *sk)
1702 if (refcount_dec_and_test(&sk->sk_refcnt))
1705 /* Generic version of sock_put(), dealing with all sockets
1706 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1708 void sock_gen_put(struct sock *sk);
1710 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1711 unsigned int trim_cap, bool refcounted);
1712 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1715 return __sk_receive_skb(sk, skb, nested, 1, true);
1718 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1720 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1721 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1723 sk->sk_tx_queue_mapping = tx_queue;
1726 #define NO_QUEUE_MAPPING USHRT_MAX
1728 static inline void sk_tx_queue_clear(struct sock *sk)
1730 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1733 static inline int sk_tx_queue_get(const struct sock *sk)
1735 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1736 return sk->sk_tx_queue_mapping;
1741 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1744 if (skb_rx_queue_recorded(skb)) {
1745 u16 rx_queue = skb_get_rx_queue(skb);
1747 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1750 sk->sk_rx_queue_mapping = rx_queue;
1755 static inline void sk_rx_queue_clear(struct sock *sk)
1758 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1763 static inline int sk_rx_queue_get(const struct sock *sk)
1765 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1766 return sk->sk_rx_queue_mapping;
1772 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1774 sk_tx_queue_clear(sk);
1775 sk->sk_socket = sock;
1778 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1780 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1781 return &rcu_dereference_raw(sk->sk_wq)->wait;
1783 /* Detach socket from process context.
1784 * Announce socket dead, detach it from wait queue and inode.
1785 * Note that parent inode held reference count on this struct sock,
1786 * we do not release it in this function, because protocol
1787 * probably wants some additional cleanups or even continuing
1788 * to work with this socket (TCP).
1790 static inline void sock_orphan(struct sock *sk)
1792 write_lock_bh(&sk->sk_callback_lock);
1793 sock_set_flag(sk, SOCK_DEAD);
1794 sk_set_socket(sk, NULL);
1796 write_unlock_bh(&sk->sk_callback_lock);
1799 static inline void sock_graft(struct sock *sk, struct socket *parent)
1801 WARN_ON(parent->sk);
1802 write_lock_bh(&sk->sk_callback_lock);
1803 rcu_assign_pointer(sk->sk_wq, parent->wq);
1805 sk_set_socket(sk, parent);
1806 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1807 security_sock_graft(sk, parent);
1808 write_unlock_bh(&sk->sk_callback_lock);
1811 kuid_t sock_i_uid(struct sock *sk);
1812 unsigned long sock_i_ino(struct sock *sk);
1814 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1816 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1819 static inline u32 net_tx_rndhash(void)
1821 u32 v = prandom_u32();
1826 static inline void sk_set_txhash(struct sock *sk)
1828 sk->sk_txhash = net_tx_rndhash();
1831 static inline void sk_rethink_txhash(struct sock *sk)
1837 static inline struct dst_entry *
1838 __sk_dst_get(struct sock *sk)
1840 return rcu_dereference_check(sk->sk_dst_cache,
1841 lockdep_sock_is_held(sk));
1844 static inline struct dst_entry *
1845 sk_dst_get(struct sock *sk)
1847 struct dst_entry *dst;
1850 dst = rcu_dereference(sk->sk_dst_cache);
1851 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1857 static inline void dst_negative_advice(struct sock *sk)
1859 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1861 sk_rethink_txhash(sk);
1863 if (dst && dst->ops->negative_advice) {
1864 ndst = dst->ops->negative_advice(dst);
1867 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1868 sk_tx_queue_clear(sk);
1869 sk->sk_dst_pending_confirm = 0;
1875 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1877 struct dst_entry *old_dst;
1879 sk_tx_queue_clear(sk);
1880 sk->sk_dst_pending_confirm = 0;
1881 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1882 lockdep_sock_is_held(sk));
1883 rcu_assign_pointer(sk->sk_dst_cache, dst);
1884 dst_release(old_dst);
1888 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1890 struct dst_entry *old_dst;
1892 sk_tx_queue_clear(sk);
1893 sk->sk_dst_pending_confirm = 0;
1894 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1895 dst_release(old_dst);
1899 __sk_dst_reset(struct sock *sk)
1901 __sk_dst_set(sk, NULL);
1905 sk_dst_reset(struct sock *sk)
1907 sk_dst_set(sk, NULL);
1910 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1912 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1914 static inline void sk_dst_confirm(struct sock *sk)
1916 if (!sk->sk_dst_pending_confirm)
1917 sk->sk_dst_pending_confirm = 1;
1920 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1922 if (skb_get_dst_pending_confirm(skb)) {
1923 struct sock *sk = skb->sk;
1924 unsigned long now = jiffies;
1926 /* avoid dirtying neighbour */
1927 if (n->confirmed != now)
1929 if (sk && sk->sk_dst_pending_confirm)
1930 sk->sk_dst_pending_confirm = 0;
1934 bool sk_mc_loop(struct sock *sk);
1936 static inline bool sk_can_gso(const struct sock *sk)
1938 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1941 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1943 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1945 sk->sk_route_nocaps |= flags;
1946 sk->sk_route_caps &= ~flags;
1949 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1950 struct iov_iter *from, char *to,
1951 int copy, int offset)
1953 if (skb->ip_summed == CHECKSUM_NONE) {
1955 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1957 skb->csum = csum_block_add(skb->csum, csum, offset);
1958 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1959 if (!copy_from_iter_full_nocache(to, copy, from))
1961 } else if (!copy_from_iter_full(to, copy, from))
1967 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1968 struct iov_iter *from, int copy)
1970 int err, offset = skb->len;
1972 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1975 __skb_trim(skb, offset);
1980 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1981 struct sk_buff *skb,
1987 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1993 skb->data_len += copy;
1994 skb->truesize += copy;
1995 sk->sk_wmem_queued += copy;
1996 sk_mem_charge(sk, copy);
2001 * sk_wmem_alloc_get - returns write allocations
2004 * Returns sk_wmem_alloc minus initial offset of one
2006 static inline int sk_wmem_alloc_get(const struct sock *sk)
2008 return refcount_read(&sk->sk_wmem_alloc) - 1;
2012 * sk_rmem_alloc_get - returns read allocations
2015 * Returns sk_rmem_alloc
2017 static inline int sk_rmem_alloc_get(const struct sock *sk)
2019 return atomic_read(&sk->sk_rmem_alloc);
2023 * sk_has_allocations - check if allocations are outstanding
2026 * Returns true if socket has write or read allocations
2028 static inline bool sk_has_allocations(const struct sock *sk)
2030 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2034 * skwq_has_sleeper - check if there are any waiting processes
2035 * @wq: struct socket_wq
2037 * Returns true if socket_wq has waiting processes
2039 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2040 * barrier call. They were added due to the race found within the tcp code.
2042 * Consider following tcp code paths::
2045 * sys_select receive packet
2047 * __add_wait_queue update tp->rcv_nxt
2049 * tp->rcv_nxt check sock_def_readable
2051 * schedule rcu_read_lock();
2052 * wq = rcu_dereference(sk->sk_wq);
2053 * if (wq && waitqueue_active(&wq->wait))
2054 * wake_up_interruptible(&wq->wait)
2058 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2059 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2060 * could then endup calling schedule and sleep forever if there are no more
2061 * data on the socket.
2064 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2066 return wq && wq_has_sleeper(&wq->wait);
2070 * sock_poll_wait - place memory barrier behind the poll_wait call.
2072 * @sock: socket to wait on
2075 * See the comments in the wq_has_sleeper function.
2077 * Do not derive sock from filp->private_data here. An SMC socket establishes
2078 * an internal TCP socket that is used in the fallback case. All socket
2079 * operations on the SMC socket are then forwarded to the TCP socket. In case of
2080 * poll, the filp->private_data pointer references the SMC socket because the
2081 * TCP socket has no file assigned.
2083 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2086 if (!poll_does_not_wait(p)) {
2087 poll_wait(filp, &sock->wq->wait, p);
2088 /* We need to be sure we are in sync with the
2089 * socket flags modification.
2091 * This memory barrier is paired in the wq_has_sleeper.
2097 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2099 if (sk->sk_txhash) {
2101 skb->hash = sk->sk_txhash;
2105 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2108 * Queue a received datagram if it will fit. Stream and sequenced
2109 * protocols can't normally use this as they need to fit buffers in
2110 * and play with them.
2112 * Inlined as it's very short and called for pretty much every
2113 * packet ever received.
2115 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2119 skb->destructor = sock_rfree;
2120 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2121 sk_mem_charge(sk, skb->truesize);
2124 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2125 unsigned long expires);
2127 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2129 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2130 struct sk_buff *skb, unsigned int flags,
2131 void (*destructor)(struct sock *sk,
2132 struct sk_buff *skb));
2133 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2134 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2136 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2137 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2140 * Recover an error report and clear atomically
2143 static inline int sock_error(struct sock *sk)
2146 if (likely(!sk->sk_err))
2148 err = xchg(&sk->sk_err, 0);
2152 static inline unsigned long sock_wspace(struct sock *sk)
2156 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2157 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2165 * We use sk->sk_wq_raw, from contexts knowing this
2166 * pointer is not NULL and cannot disappear/change.
2168 static inline void sk_set_bit(int nr, struct sock *sk)
2170 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2171 !sock_flag(sk, SOCK_FASYNC))
2174 set_bit(nr, &sk->sk_wq_raw->flags);
2177 static inline void sk_clear_bit(int nr, struct sock *sk)
2179 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2180 !sock_flag(sk, SOCK_FASYNC))
2183 clear_bit(nr, &sk->sk_wq_raw->flags);
2186 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2188 if (sock_flag(sk, SOCK_FASYNC)) {
2190 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2195 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2196 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2197 * Note: for send buffers, TCP works better if we can build two skbs at
2200 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2202 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2203 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2205 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2207 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2208 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2209 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2213 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2214 bool force_schedule);
2217 * sk_page_frag - return an appropriate page_frag
2220 * If socket allocation mode allows current thread to sleep, it means its
2221 * safe to use the per task page_frag instead of the per socket one.
2223 static inline struct page_frag *sk_page_frag(struct sock *sk)
2225 if (gfpflags_allow_blocking(sk->sk_allocation))
2226 return ¤t->task_frag;
2228 return &sk->sk_frag;
2231 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2234 * Default write policy as shown to user space via poll/select/SIGIO
2236 static inline bool sock_writeable(const struct sock *sk)
2238 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2241 static inline gfp_t gfp_any(void)
2243 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2246 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2248 return noblock ? 0 : sk->sk_rcvtimeo;
2251 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2253 return noblock ? 0 : sk->sk_sndtimeo;
2256 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2258 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2261 /* Alas, with timeout socket operations are not restartable.
2262 * Compare this to poll().
2264 static inline int sock_intr_errno(long timeo)
2266 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2269 struct sock_skb_cb {
2273 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2274 * using skb->cb[] would keep using it directly and utilize its
2275 * alignement guarantee.
2277 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2278 sizeof(struct sock_skb_cb)))
2280 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2281 SOCK_SKB_CB_OFFSET))
2283 #define sock_skb_cb_check_size(size) \
2284 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2287 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2289 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2290 atomic_read(&sk->sk_drops) : 0;
2293 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2295 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2297 atomic_add(segs, &sk->sk_drops);
2300 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2301 struct sk_buff *skb);
2302 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2303 struct sk_buff *skb);
2306 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2308 ktime_t kt = skb->tstamp;
2309 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2312 * generate control messages if
2313 * - receive time stamping in software requested
2314 * - software time stamp available and wanted
2315 * - hardware time stamps available and wanted
2317 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2318 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2319 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2320 (hwtstamps->hwtstamp &&
2321 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2322 __sock_recv_timestamp(msg, sk, skb);
2326 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2327 __sock_recv_wifi_status(msg, sk, skb);
2330 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2331 struct sk_buff *skb);
2333 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2334 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2335 struct sk_buff *skb)
2337 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2338 (1UL << SOCK_RCVTSTAMP))
2339 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2340 SOF_TIMESTAMPING_RAW_HARDWARE)
2342 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2343 __sock_recv_ts_and_drops(msg, sk, skb);
2344 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2345 sk->sk_stamp = skb->tstamp;
2346 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2350 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2353 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2354 * @sk: socket sending this packet
2355 * @tsflags: timestamping flags to use
2356 * @tx_flags: completed with instructions for time stamping
2357 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2359 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2361 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2362 __u8 *tx_flags, __u32 *tskey)
2364 if (unlikely(tsflags)) {
2365 __sock_tx_timestamp(tsflags, tx_flags);
2366 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2367 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2368 *tskey = sk->sk_tskey++;
2370 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2371 *tx_flags |= SKBTX_WIFI_STATUS;
2374 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2377 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2380 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2382 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2383 &skb_shinfo(skb)->tskey);
2387 * sk_eat_skb - Release a skb if it is no longer needed
2388 * @sk: socket to eat this skb from
2389 * @skb: socket buffer to eat
2391 * This routine must be called with interrupts disabled or with the socket
2392 * locked so that the sk_buff queue operation is ok.
2394 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2396 __skb_unlink(skb, &sk->sk_receive_queue);
2401 struct net *sock_net(const struct sock *sk)
2403 return read_pnet(&sk->sk_net);
2407 void sock_net_set(struct sock *sk, struct net *net)
2409 write_pnet(&sk->sk_net, net);
2412 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2415 struct sock *sk = skb->sk;
2417 skb->destructor = NULL;
2424 /* This helper checks if a socket is a full socket,
2425 * ie _not_ a timewait or request socket.
2427 static inline bool sk_fullsock(const struct sock *sk)
2429 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2432 /* Checks if this SKB belongs to an HW offloaded socket
2433 * and whether any SW fallbacks are required based on dev.
2435 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2436 struct net_device *dev)
2438 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2439 struct sock *sk = skb->sk;
2441 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb)
2442 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2448 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2449 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2451 static inline bool sk_listener(const struct sock *sk)
2453 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2456 void sock_enable_timestamp(struct sock *sk, int flag);
2457 int sock_get_timestamp(struct sock *, struct timeval __user *);
2458 int sock_get_timestampns(struct sock *, struct timespec __user *);
2459 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2462 bool sk_ns_capable(const struct sock *sk,
2463 struct user_namespace *user_ns, int cap);
2464 bool sk_capable(const struct sock *sk, int cap);
2465 bool sk_net_capable(const struct sock *sk, int cap);
2467 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2469 /* Take into consideration the size of the struct sk_buff overhead in the
2470 * determination of these values, since that is non-constant across
2471 * platforms. This makes socket queueing behavior and performance
2472 * not depend upon such differences.
2474 #define _SK_MEM_PACKETS 256
2475 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2476 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2477 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2479 extern __u32 sysctl_wmem_max;
2480 extern __u32 sysctl_rmem_max;
2482 extern int sysctl_tstamp_allow_data;
2483 extern int sysctl_optmem_max;
2485 extern __u32 sysctl_wmem_default;
2486 extern __u32 sysctl_rmem_default;
2488 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2490 /* Does this proto have per netns sysctl_wmem ? */
2491 if (proto->sysctl_wmem_offset)
2492 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2494 return *proto->sysctl_wmem;
2497 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2499 /* Does this proto have per netns sysctl_rmem ? */
2500 if (proto->sysctl_rmem_offset)
2501 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2503 return *proto->sysctl_rmem;
2506 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2507 * Some wifi drivers need to tweak it to get more chunks.
2508 * They can use this helper from their ndo_start_xmit()
2510 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2512 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2514 sk->sk_pacing_shift = val;
2517 /* if a socket is bound to a device, check that the given device
2518 * index is either the same or that the socket is bound to an L3
2519 * master device and the given device index is also enslaved to
2522 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2526 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2529 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2530 if (mdif && mdif == sk->sk_bound_dev_if)
2536 #endif /* _SOCK_H */