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>
64 #include <linux/filter.h>
65 #include <linux/rculist_nulls.h>
66 #include <linux/poll.h>
68 #include <linux/atomic.h>
69 #include <linux/refcount.h>
71 #include <net/checksum.h>
72 #include <net/tcp_states.h>
73 #include <linux/net_tstamp.h>
77 * This structure really needs to be cleaned up.
78 * Most of it is for TCP, and not used by any of
79 * the other protocols.
82 /* Define this to get the SOCK_DBG debugging facility. */
83 #define SOCK_DEBUGGING
85 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
86 printk(KERN_DEBUG msg); } while (0)
88 /* Validate arguments and do nothing */
89 static inline __printf(2, 3)
90 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
95 /* This is the per-socket lock. The spinlock provides a synchronization
96 * between user contexts and software interrupt processing, whereas the
97 * mini-semaphore synchronizes multiple users amongst themselves.
102 wait_queue_head_t wq;
104 * We express the mutex-alike socket_lock semantics
105 * to the lock validator by explicitly managing
106 * the slock as a lock variant (in addition to
109 #ifdef CONFIG_DEBUG_LOCK_ALLOC
110 struct lockdep_map dep_map;
118 typedef __u32 __bitwise __portpair;
119 typedef __u64 __bitwise __addrpair;
122 * struct sock_common - minimal network layer representation of sockets
123 * @skc_daddr: Foreign IPv4 addr
124 * @skc_rcv_saddr: Bound local IPv4 addr
125 * @skc_hash: hash value used with various protocol lookup tables
126 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
127 * @skc_dport: placeholder for inet_dport/tw_dport
128 * @skc_num: placeholder for inet_num/tw_num
129 * @skc_family: network address family
130 * @skc_state: Connection state
131 * @skc_reuse: %SO_REUSEADDR setting
132 * @skc_reuseport: %SO_REUSEPORT setting
133 * @skc_bound_dev_if: bound device index if != 0
134 * @skc_bind_node: bind hash linkage for various protocol lookup tables
135 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
136 * @skc_prot: protocol handlers inside a network family
137 * @skc_net: reference to the network namespace of this socket
138 * @skc_node: main hash linkage for various protocol lookup tables
139 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
140 * @skc_tx_queue_mapping: tx queue number for this connection
141 * @skc_flags: place holder for sk_flags
142 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
143 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
144 * @skc_incoming_cpu: record/match cpu processing incoming packets
145 * @skc_refcnt: reference count
147 * This is the minimal network layer representation of sockets, the header
148 * for struct sock and struct inet_timewait_sock.
151 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
152 * address on 64bit arches : cf INET_MATCH()
155 __addrpair skc_addrpair;
158 __be32 skc_rcv_saddr;
162 unsigned int skc_hash;
163 __u16 skc_u16hashes[2];
165 /* skc_dport && skc_num must be grouped as well */
167 __portpair skc_portpair;
174 unsigned short skc_family;
175 volatile unsigned char skc_state;
176 unsigned char skc_reuse:4;
177 unsigned char skc_reuseport:1;
178 unsigned char skc_ipv6only:1;
179 unsigned char skc_net_refcnt:1;
180 int skc_bound_dev_if;
182 struct hlist_node skc_bind_node;
183 struct hlist_node skc_portaddr_node;
185 struct proto *skc_prot;
186 possible_net_t skc_net;
188 #if IS_ENABLED(CONFIG_IPV6)
189 struct in6_addr skc_v6_daddr;
190 struct in6_addr skc_v6_rcv_saddr;
193 atomic64_t skc_cookie;
195 /* following fields are padding to force
196 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
197 * assuming IPV6 is enabled. We use this padding differently
198 * for different kind of 'sockets'
201 unsigned long skc_flags;
202 struct sock *skc_listener; /* request_sock */
203 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
206 * fields between dontcopy_begin/dontcopy_end
207 * are not copied in sock_copy()
210 int skc_dontcopy_begin[0];
213 struct hlist_node skc_node;
214 struct hlist_nulls_node skc_nulls_node;
216 int skc_tx_queue_mapping;
218 int skc_incoming_cpu;
220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
223 refcount_t skc_refcnt;
225 int skc_dontcopy_end[0];
228 u32 skc_window_clamp;
229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
235 * struct sock - network layer representation of sockets
236 * @__sk_common: shared layout with inet_timewait_sock
237 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
238 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
239 * @sk_lock: synchronizer
240 * @sk_kern_sock: True if sock is using kernel lock classes
241 * @sk_rcvbuf: size of receive buffer in bytes
242 * @sk_wq: sock wait queue and async head
243 * @sk_rx_dst: receive input route used by early demux
244 * @sk_dst_cache: destination cache
245 * @sk_dst_pending_confirm: need to confirm neighbour
246 * @sk_policy: flow policy
247 * @sk_receive_queue: incoming packets
248 * @sk_wmem_alloc: transmit queue bytes committed
249 * @sk_tsq_flags: TCP Small Queues flags
250 * @sk_write_queue: Packet sending queue
251 * @sk_omem_alloc: "o" is "option" or "other"
252 * @sk_wmem_queued: persistent queue size
253 * @sk_forward_alloc: space allocated forward
254 * @sk_napi_id: id of the last napi context to receive data for sk
255 * @sk_ll_usec: usecs to busypoll when there is no data
256 * @sk_allocation: allocation mode
257 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
258 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
259 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
260 * @sk_sndbuf: size of send buffer in bytes
261 * @__sk_flags_offset: empty field used to determine location of bitfield
262 * @sk_padding: unused element for alignment
263 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
264 * @sk_no_check_rx: allow zero checksum in RX packets
265 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
266 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
267 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
268 * @sk_gso_max_size: Maximum GSO segment size to build
269 * @sk_gso_max_segs: Maximum number of GSO segments
270 * @sk_lingertime: %SO_LINGER l_linger setting
271 * @sk_backlog: always used with the per-socket spinlock held
272 * @sk_callback_lock: used with the callbacks in the end of this struct
273 * @sk_error_queue: rarely used
274 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
275 * IPV6_ADDRFORM for instance)
276 * @sk_err: last error
277 * @sk_err_soft: errors that don't cause failure but are the cause of a
278 * persistent failure not just 'timed out'
279 * @sk_drops: raw/udp drops counter
280 * @sk_ack_backlog: current listen backlog
281 * @sk_max_ack_backlog: listen backlog set in listen()
282 * @sk_uid: user id of owner
283 * @sk_priority: %SO_PRIORITY setting
284 * @sk_type: socket type (%SOCK_STREAM, etc)
285 * @sk_protocol: which protocol this socket belongs in this network family
286 * @sk_peer_pid: &struct pid for this socket's peer
287 * @sk_peer_cred: %SO_PEERCRED setting
288 * @sk_rcvlowat: %SO_RCVLOWAT setting
289 * @sk_rcvtimeo: %SO_RCVTIMEO setting
290 * @sk_sndtimeo: %SO_SNDTIMEO setting
291 * @sk_txhash: computed flow hash for use on transmit
292 * @sk_filter: socket filtering instructions
293 * @sk_timer: sock cleanup timer
294 * @sk_stamp: time stamp of last packet received
295 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
296 * @sk_tsflags: SO_TIMESTAMPING socket options
297 * @sk_tskey: counter to disambiguate concurrent tstamp requests
298 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
299 * @sk_socket: Identd and reporting IO signals
300 * @sk_user_data: RPC layer private data
301 * @sk_frag: cached page frag
302 * @sk_peek_off: current peek_offset value
303 * @sk_send_head: front of stuff to transmit
304 * @sk_security: used by security modules
305 * @sk_mark: generic packet mark
306 * @sk_cgrp_data: cgroup data for this cgroup
307 * @sk_memcg: this socket's memory cgroup association
308 * @sk_write_pending: a write to stream socket waits to start
309 * @sk_state_change: callback to indicate change in the state of the sock
310 * @sk_data_ready: callback to indicate there is data to be processed
311 * @sk_write_space: callback to indicate there is bf sending space available
312 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
313 * @sk_backlog_rcv: callback to process the backlog
314 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
315 * @sk_reuseport_cb: reuseport group container
316 * @sk_rcu: used during RCU grace period
320 * Now struct inet_timewait_sock also uses sock_common, so please just
321 * don't add nothing before this first member (__sk_common) --acme
323 struct sock_common __sk_common;
324 #define sk_node __sk_common.skc_node
325 #define sk_nulls_node __sk_common.skc_nulls_node
326 #define sk_refcnt __sk_common.skc_refcnt
327 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
329 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
330 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
331 #define sk_hash __sk_common.skc_hash
332 #define sk_portpair __sk_common.skc_portpair
333 #define sk_num __sk_common.skc_num
334 #define sk_dport __sk_common.skc_dport
335 #define sk_addrpair __sk_common.skc_addrpair
336 #define sk_daddr __sk_common.skc_daddr
337 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
338 #define sk_family __sk_common.skc_family
339 #define sk_state __sk_common.skc_state
340 #define sk_reuse __sk_common.skc_reuse
341 #define sk_reuseport __sk_common.skc_reuseport
342 #define sk_ipv6only __sk_common.skc_ipv6only
343 #define sk_net_refcnt __sk_common.skc_net_refcnt
344 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
345 #define sk_bind_node __sk_common.skc_bind_node
346 #define sk_prot __sk_common.skc_prot
347 #define sk_net __sk_common.skc_net
348 #define sk_v6_daddr __sk_common.skc_v6_daddr
349 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
350 #define sk_cookie __sk_common.skc_cookie
351 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
352 #define sk_flags __sk_common.skc_flags
353 #define sk_rxhash __sk_common.skc_rxhash
355 socket_lock_t sk_lock;
358 struct sk_buff_head sk_error_queue;
359 struct sk_buff_head sk_receive_queue;
361 * The backlog queue is special, it is always used with
362 * the per-socket spinlock held and requires low latency
363 * access. Therefore we special case it's implementation.
364 * Note : rmem_alloc is in this structure to fill a hole
365 * on 64bit arches, not because its logically part of
371 struct sk_buff *head;
372 struct sk_buff *tail;
374 #define sk_rmem_alloc sk_backlog.rmem_alloc
376 int sk_forward_alloc;
377 #ifdef CONFIG_NET_RX_BUSY_POLL
378 unsigned int sk_ll_usec;
379 /* ===== mostly read cache line ===== */
380 unsigned int sk_napi_id;
384 struct sk_filter __rcu *sk_filter;
386 struct socket_wq __rcu *sk_wq;
387 struct socket_wq *sk_wq_raw;
390 struct xfrm_policy __rcu *sk_policy[2];
392 struct dst_entry *sk_rx_dst;
393 struct dst_entry __rcu *sk_dst_cache;
394 atomic_t sk_omem_alloc;
397 /* ===== cache line for TX ===== */
399 refcount_t sk_wmem_alloc;
400 unsigned long sk_tsq_flags;
401 struct sk_buff *sk_send_head;
402 struct sk_buff_head sk_write_queue;
404 int sk_write_pending;
405 __u32 sk_dst_pending_confirm;
406 u32 sk_pacing_status; /* see enum sk_pacing */
408 struct timer_list sk_timer;
411 u32 sk_pacing_rate; /* bytes per second */
412 u32 sk_max_pacing_rate;
413 struct page_frag sk_frag;
414 netdev_features_t sk_route_caps;
415 netdev_features_t sk_route_nocaps;
417 unsigned int sk_gso_max_size;
422 * Because of non atomicity rules, all
423 * changes are protected by socket lock.
425 unsigned int __sk_flags_offset[0];
426 #ifdef __BIG_ENDIAN_BITFIELD
427 #define SK_FL_PROTO_SHIFT 16
428 #define SK_FL_PROTO_MASK 0x00ff0000
430 #define SK_FL_TYPE_SHIFT 0
431 #define SK_FL_TYPE_MASK 0x0000ffff
433 #define SK_FL_PROTO_SHIFT 8
434 #define SK_FL_PROTO_MASK 0x0000ff00
436 #define SK_FL_TYPE_SHIFT 16
437 #define SK_FL_TYPE_MASK 0xffff0000
440 unsigned int sk_padding : 1,
447 #define SK_PROTOCOL_MAX U8_MAX
449 unsigned long sk_lingertime;
450 struct proto *sk_prot_creator;
451 rwlock_t sk_callback_lock;
455 u32 sk_max_ack_backlog;
457 spinlock_t sk_peer_lock;
458 struct pid *sk_peer_pid;
459 const struct cred *sk_peer_cred;
463 #if BITS_PER_LONG==32
464 seqlock_t sk_stamp_seq;
470 struct socket *sk_socket;
472 #ifdef CONFIG_SECURITY
475 struct sock_cgroup_data sk_cgrp_data;
476 struct mem_cgroup *sk_memcg;
477 void (*sk_state_change)(struct sock *sk);
478 void (*sk_data_ready)(struct sock *sk);
479 void (*sk_write_space)(struct sock *sk);
480 void (*sk_error_report)(struct sock *sk);
481 int (*sk_backlog_rcv)(struct sock *sk,
482 struct sk_buff *skb);
483 void (*sk_destruct)(struct sock *sk);
484 struct sock_reuseport __rcu *sk_reuseport_cb;
485 struct rcu_head sk_rcu;
490 SK_PACING_NEEDED = 1,
494 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
496 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
497 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
500 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
501 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
502 * on a socket means that the socket will reuse everybody else's port
503 * without looking at the other's sk_reuse value.
506 #define SK_NO_REUSE 0
507 #define SK_CAN_REUSE 1
508 #define SK_FORCE_REUSE 2
510 int sk_set_peek_off(struct sock *sk, int val);
512 static inline int sk_peek_offset(struct sock *sk, int flags)
514 if (unlikely(flags & MSG_PEEK)) {
515 return READ_ONCE(sk->sk_peek_off);
521 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
523 s32 off = READ_ONCE(sk->sk_peek_off);
525 if (unlikely(off >= 0)) {
526 off = max_t(s32, off - val, 0);
527 WRITE_ONCE(sk->sk_peek_off, off);
531 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
533 sk_peek_offset_bwd(sk, -val);
537 * Hashed lists helper routines
539 static inline struct sock *sk_entry(const struct hlist_node *node)
541 return hlist_entry(node, struct sock, sk_node);
544 static inline struct sock *__sk_head(const struct hlist_head *head)
546 return hlist_entry(head->first, struct sock, sk_node);
549 static inline struct sock *sk_head(const struct hlist_head *head)
551 return hlist_empty(head) ? NULL : __sk_head(head);
554 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
556 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
559 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
561 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
564 static inline struct sock *sk_next(const struct sock *sk)
566 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
569 static inline struct sock *sk_nulls_next(const struct sock *sk)
571 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
572 hlist_nulls_entry(sk->sk_nulls_node.next,
573 struct sock, sk_nulls_node) :
577 static inline bool sk_unhashed(const struct sock *sk)
579 return hlist_unhashed(&sk->sk_node);
582 static inline bool sk_hashed(const struct sock *sk)
584 return !sk_unhashed(sk);
587 static inline void sk_node_init(struct hlist_node *node)
592 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
597 static inline void __sk_del_node(struct sock *sk)
599 __hlist_del(&sk->sk_node);
602 /* NB: equivalent to hlist_del_init_rcu */
603 static inline bool __sk_del_node_init(struct sock *sk)
607 sk_node_init(&sk->sk_node);
613 /* Grab socket reference count. This operation is valid only
614 when sk is ALREADY grabbed f.e. it is found in hash table
615 or a list and the lookup is made under lock preventing hash table
619 static __always_inline void sock_hold(struct sock *sk)
621 refcount_inc(&sk->sk_refcnt);
624 /* Ungrab socket in the context, which assumes that socket refcnt
625 cannot hit zero, f.e. it is true in context of any socketcall.
627 static __always_inline void __sock_put(struct sock *sk)
629 refcount_dec(&sk->sk_refcnt);
632 static inline bool sk_del_node_init(struct sock *sk)
634 bool rc = __sk_del_node_init(sk);
637 /* paranoid for a while -acme */
638 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
643 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
645 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
648 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
654 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
656 bool rc = __sk_nulls_del_node_init_rcu(sk);
659 /* paranoid for a while -acme */
660 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
666 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
668 hlist_add_head(&sk->sk_node, list);
671 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
674 __sk_add_node(sk, list);
677 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
680 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
681 sk->sk_family == AF_INET6)
682 hlist_add_tail_rcu(&sk->sk_node, list);
684 hlist_add_head_rcu(&sk->sk_node, list);
687 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
690 hlist_add_tail_rcu(&sk->sk_node, list);
693 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
695 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
698 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
700 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
703 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
706 __sk_nulls_add_node_rcu(sk, list);
709 static inline void __sk_del_bind_node(struct sock *sk)
711 __hlist_del(&sk->sk_bind_node);
714 static inline void sk_add_bind_node(struct sock *sk,
715 struct hlist_head *list)
717 hlist_add_head(&sk->sk_bind_node, list);
720 #define sk_for_each(__sk, list) \
721 hlist_for_each_entry(__sk, list, sk_node)
722 #define sk_for_each_rcu(__sk, list) \
723 hlist_for_each_entry_rcu(__sk, list, sk_node)
724 #define sk_nulls_for_each(__sk, node, list) \
725 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
726 #define sk_nulls_for_each_rcu(__sk, node, list) \
727 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
728 #define sk_for_each_from(__sk) \
729 hlist_for_each_entry_from(__sk, sk_node)
730 #define sk_nulls_for_each_from(__sk, node) \
731 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
732 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
733 #define sk_for_each_safe(__sk, tmp, list) \
734 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
735 #define sk_for_each_bound(__sk, list) \
736 hlist_for_each_entry(__sk, list, sk_bind_node)
739 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
740 * @tpos: the type * to use as a loop cursor.
741 * @pos: the &struct hlist_node to use as a loop cursor.
742 * @head: the head for your list.
743 * @offset: offset of hlist_node within the struct.
746 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
747 for (pos = rcu_dereference((head)->first); \
749 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
750 pos = rcu_dereference(pos->next))
752 static inline struct user_namespace *sk_user_ns(struct sock *sk)
754 /* Careful only use this in a context where these parameters
755 * can not change and must all be valid, such as recvmsg from
758 return sk->sk_socket->file->f_cred->user_ns;
772 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
773 SOCK_DBG, /* %SO_DEBUG setting */
774 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
775 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
776 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
777 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
778 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
779 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
780 SOCK_FASYNC, /* fasync() active */
782 SOCK_ZEROCOPY, /* buffers from userspace */
783 SOCK_WIFI_STATUS, /* push wifi status to userspace */
784 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
785 * Will use last 4 bytes of packet sent from
786 * user-space instead.
788 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
789 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
790 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
793 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
795 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
797 nsk->sk_flags = osk->sk_flags;
800 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
802 __set_bit(flag, &sk->sk_flags);
805 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
807 __clear_bit(flag, &sk->sk_flags);
810 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
812 return test_bit(flag, &sk->sk_flags);
816 extern struct static_key memalloc_socks;
817 static inline int sk_memalloc_socks(void)
819 return static_key_false(&memalloc_socks);
822 void __receive_sock(struct file *file);
825 static inline int sk_memalloc_socks(void)
830 static inline void __receive_sock(struct file *file)
834 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
836 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
839 static inline void sk_acceptq_removed(struct sock *sk)
841 sk->sk_ack_backlog--;
844 static inline void sk_acceptq_added(struct sock *sk)
846 sk->sk_ack_backlog++;
849 static inline bool sk_acceptq_is_full(const struct sock *sk)
851 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
855 * Compute minimal free write space needed to queue new packets.
857 static inline int sk_stream_min_wspace(const struct sock *sk)
859 return sk->sk_wmem_queued >> 1;
862 static inline int sk_stream_wspace(const struct sock *sk)
864 return sk->sk_sndbuf - sk->sk_wmem_queued;
867 void sk_stream_write_space(struct sock *sk);
869 /* OOB backlog add */
870 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
872 /* dont let skb dst not refcounted, we are going to leave rcu lock */
875 if (!sk->sk_backlog.tail)
876 sk->sk_backlog.head = skb;
878 sk->sk_backlog.tail->next = skb;
880 sk->sk_backlog.tail = skb;
885 * Take into account size of receive queue and backlog queue
886 * Do not take into account this skb truesize,
887 * to allow even a single big packet to come.
889 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
891 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
893 return qsize > limit;
896 /* The per-socket spinlock must be held here. */
897 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
900 if (sk_rcvqueues_full(sk, limit))
904 * If the skb was allocated from pfmemalloc reserves, only
905 * allow SOCK_MEMALLOC sockets to use it as this socket is
906 * helping free memory
908 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
911 __sk_add_backlog(sk, skb);
912 sk->sk_backlog.len += skb->truesize;
916 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
918 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
920 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
921 return __sk_backlog_rcv(sk, skb);
923 return sk->sk_backlog_rcv(sk, skb);
926 static inline void sk_incoming_cpu_update(struct sock *sk)
928 int cpu = raw_smp_processor_id();
930 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
931 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
934 static inline void sock_rps_record_flow_hash(__u32 hash)
937 struct rps_sock_flow_table *sock_flow_table;
940 sock_flow_table = rcu_dereference(rps_sock_flow_table);
941 rps_record_sock_flow(sock_flow_table, hash);
946 static inline void sock_rps_record_flow(const struct sock *sk)
949 if (static_key_false(&rfs_needed)) {
950 /* Reading sk->sk_rxhash might incur an expensive cache line
953 * TCP_ESTABLISHED does cover almost all states where RFS
954 * might be useful, and is cheaper [1] than testing :
955 * IPv4: inet_sk(sk)->inet_daddr
956 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
957 * OR an additional socket flag
958 * [1] : sk_state and sk_prot are in the same cache line.
960 if (sk->sk_state == TCP_ESTABLISHED)
961 sock_rps_record_flow_hash(sk->sk_rxhash);
966 static inline void sock_rps_save_rxhash(struct sock *sk,
967 const struct sk_buff *skb)
970 if (unlikely(sk->sk_rxhash != skb->hash))
971 sk->sk_rxhash = skb->hash;
975 static inline void sock_rps_reset_rxhash(struct sock *sk)
982 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
984 release_sock(__sk); \
985 __rc = __condition; \
987 *(__timeo) = wait_woken(__wait, \
988 TASK_INTERRUPTIBLE, \
991 sched_annotate_sleep(); \
993 __rc = __condition; \
997 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
998 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
999 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1000 int sk_stream_error(struct sock *sk, int flags, int err);
1001 void sk_stream_kill_queues(struct sock *sk);
1002 void sk_set_memalloc(struct sock *sk);
1003 void sk_clear_memalloc(struct sock *sk);
1005 void __sk_flush_backlog(struct sock *sk);
1007 static inline bool sk_flush_backlog(struct sock *sk)
1009 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1010 __sk_flush_backlog(sk);
1016 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1018 struct request_sock_ops;
1019 struct timewait_sock_ops;
1020 struct inet_hashinfo;
1021 struct raw_hashinfo;
1022 struct smc_hashinfo;
1026 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1027 * un-modified. Special care is taken when initializing object to zero.
1029 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1031 if (offsetof(struct sock, sk_node.next) != 0)
1032 memset(sk, 0, offsetof(struct sock, sk_node.next));
1033 memset(&sk->sk_node.pprev, 0,
1034 size - offsetof(struct sock, sk_node.pprev));
1037 /* Networking protocol blocks we attach to sockets.
1038 * socket layer -> transport layer interface
1041 void (*close)(struct sock *sk,
1043 int (*connect)(struct sock *sk,
1044 struct sockaddr *uaddr,
1046 int (*disconnect)(struct sock *sk, int flags);
1048 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1051 int (*ioctl)(struct sock *sk, int cmd,
1053 int (*init)(struct sock *sk);
1054 void (*destroy)(struct sock *sk);
1055 void (*shutdown)(struct sock *sk, int how);
1056 int (*setsockopt)(struct sock *sk, int level,
1057 int optname, char __user *optval,
1058 unsigned int optlen);
1059 int (*getsockopt)(struct sock *sk, int level,
1060 int optname, char __user *optval,
1061 int __user *option);
1062 void (*keepalive)(struct sock *sk, int valbool);
1063 #ifdef CONFIG_COMPAT
1064 int (*compat_setsockopt)(struct sock *sk,
1066 int optname, char __user *optval,
1067 unsigned int optlen);
1068 int (*compat_getsockopt)(struct sock *sk,
1070 int optname, char __user *optval,
1071 int __user *option);
1072 int (*compat_ioctl)(struct sock *sk,
1073 unsigned int cmd, unsigned long arg);
1075 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1077 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1078 size_t len, int noblock, int flags,
1080 int (*sendpage)(struct sock *sk, struct page *page,
1081 int offset, size_t size, int flags);
1082 int (*bind)(struct sock *sk,
1083 struct sockaddr *uaddr, int addr_len);
1085 int (*backlog_rcv) (struct sock *sk,
1086 struct sk_buff *skb);
1088 void (*release_cb)(struct sock *sk);
1090 /* Keeping track of sk's, looking them up, and port selection methods. */
1091 int (*hash)(struct sock *sk);
1092 void (*unhash)(struct sock *sk);
1093 void (*rehash)(struct sock *sk);
1094 int (*get_port)(struct sock *sk, unsigned short snum);
1096 /* Keeping track of sockets in use */
1097 #ifdef CONFIG_PROC_FS
1098 unsigned int inuse_idx;
1101 bool (*stream_memory_free)(const struct sock *sk);
1102 /* Memory pressure */
1103 void (*enter_memory_pressure)(struct sock *sk);
1104 void (*leave_memory_pressure)(struct sock *sk);
1105 atomic_long_t *memory_allocated; /* Current allocated memory. */
1106 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1108 * Pressure flag: try to collapse.
1109 * Technical note: it is used by multiple contexts non atomically.
1110 * All the __sk_mem_schedule() is of this nature: accounting
1111 * is strict, actions are advisory and have some latency.
1113 unsigned long *memory_pressure;
1120 struct kmem_cache *slab;
1121 unsigned int obj_size;
1124 struct percpu_counter *orphan_count;
1126 struct request_sock_ops *rsk_prot;
1127 struct timewait_sock_ops *twsk_prot;
1130 struct inet_hashinfo *hashinfo;
1131 struct udp_table *udp_table;
1132 struct raw_hashinfo *raw_hash;
1133 struct smc_hashinfo *smc_hash;
1136 struct module *owner;
1140 struct list_head node;
1141 #ifdef SOCK_REFCNT_DEBUG
1144 int (*diag_destroy)(struct sock *sk, int err);
1145 } __randomize_layout;
1147 int proto_register(struct proto *prot, int alloc_slab);
1148 void proto_unregister(struct proto *prot);
1150 #ifdef SOCK_REFCNT_DEBUG
1151 static inline void sk_refcnt_debug_inc(struct sock *sk)
1153 atomic_inc(&sk->sk_prot->socks);
1156 static inline void sk_refcnt_debug_dec(struct sock *sk)
1158 atomic_dec(&sk->sk_prot->socks);
1159 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1160 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1163 static inline void sk_refcnt_debug_release(const struct sock *sk)
1165 if (refcount_read(&sk->sk_refcnt) != 1)
1166 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1167 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1169 #else /* SOCK_REFCNT_DEBUG */
1170 #define sk_refcnt_debug_inc(sk) do { } while (0)
1171 #define sk_refcnt_debug_dec(sk) do { } while (0)
1172 #define sk_refcnt_debug_release(sk) do { } while (0)
1173 #endif /* SOCK_REFCNT_DEBUG */
1175 static inline bool sk_stream_memory_free(const struct sock *sk)
1177 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1180 return sk->sk_prot->stream_memory_free ?
1181 sk->sk_prot->stream_memory_free(sk) : true;
1184 static inline bool sk_stream_is_writeable(const struct sock *sk)
1186 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1187 sk_stream_memory_free(sk);
1190 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1191 struct cgroup *ancestor)
1193 #ifdef CONFIG_SOCK_CGROUP_DATA
1194 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1201 static inline bool sk_has_memory_pressure(const struct sock *sk)
1203 return sk->sk_prot->memory_pressure != NULL;
1206 static inline bool sk_under_memory_pressure(const struct sock *sk)
1208 if (!sk->sk_prot->memory_pressure)
1211 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1212 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1215 return !!*sk->sk_prot->memory_pressure;
1219 sk_memory_allocated(const struct sock *sk)
1221 return atomic_long_read(sk->sk_prot->memory_allocated);
1225 sk_memory_allocated_add(struct sock *sk, int amt)
1227 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1231 sk_memory_allocated_sub(struct sock *sk, int amt)
1233 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1236 static inline void sk_sockets_allocated_dec(struct sock *sk)
1238 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1241 static inline void sk_sockets_allocated_inc(struct sock *sk)
1243 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1247 sk_sockets_allocated_read_positive(struct sock *sk)
1249 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1253 proto_sockets_allocated_sum_positive(struct proto *prot)
1255 return percpu_counter_sum_positive(prot->sockets_allocated);
1259 proto_memory_allocated(struct proto *prot)
1261 return atomic_long_read(prot->memory_allocated);
1265 proto_memory_pressure(struct proto *prot)
1267 if (!prot->memory_pressure)
1269 return !!*prot->memory_pressure;
1273 #ifdef CONFIG_PROC_FS
1274 /* Called with local bh disabled */
1275 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1276 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1278 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1285 /* With per-bucket locks this operation is not-atomic, so that
1286 * this version is not worse.
1288 static inline int __sk_prot_rehash(struct sock *sk)
1290 sk->sk_prot->unhash(sk);
1291 return sk->sk_prot->hash(sk);
1294 /* About 10 seconds */
1295 #define SOCK_DESTROY_TIME (10*HZ)
1297 /* Sockets 0-1023 can't be bound to unless you are superuser */
1298 #define PROT_SOCK 1024
1300 #define SHUTDOWN_MASK 3
1301 #define RCV_SHUTDOWN 1
1302 #define SEND_SHUTDOWN 2
1304 #define SOCK_SNDBUF_LOCK 1
1305 #define SOCK_RCVBUF_LOCK 2
1306 #define SOCK_BINDADDR_LOCK 4
1307 #define SOCK_BINDPORT_LOCK 8
1309 struct socket_alloc {
1310 struct socket socket;
1311 struct inode vfs_inode;
1314 static inline struct socket *SOCKET_I(struct inode *inode)
1316 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1319 static inline struct inode *SOCK_INODE(struct socket *socket)
1321 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1325 * Functions for memory accounting
1327 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1328 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1329 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1330 void __sk_mem_reclaim(struct sock *sk, int amount);
1332 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1333 * do not necessarily have 16x time more memory than 4KB ones.
1335 #define SK_MEM_QUANTUM 4096
1336 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1337 #define SK_MEM_SEND 0
1338 #define SK_MEM_RECV 1
1340 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1341 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1343 long val = READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1345 #if PAGE_SIZE > SK_MEM_QUANTUM
1346 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1347 #elif PAGE_SIZE < SK_MEM_QUANTUM
1348 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1353 static inline int sk_mem_pages(int amt)
1355 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1358 static inline bool sk_has_account(struct sock *sk)
1360 /* return true if protocol supports memory accounting */
1361 return !!sk->sk_prot->memory_allocated;
1364 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1366 if (!sk_has_account(sk))
1368 return size <= sk->sk_forward_alloc ||
1369 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1373 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1375 if (!sk_has_account(sk))
1377 return size<= sk->sk_forward_alloc ||
1378 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1379 skb_pfmemalloc(skb);
1382 static inline void sk_mem_reclaim(struct sock *sk)
1384 if (!sk_has_account(sk))
1386 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1387 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1390 static inline void sk_mem_reclaim_partial(struct sock *sk)
1392 if (!sk_has_account(sk))
1394 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1395 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1398 static inline void sk_mem_charge(struct sock *sk, int size)
1400 if (!sk_has_account(sk))
1402 sk->sk_forward_alloc -= size;
1405 static inline void sk_mem_uncharge(struct sock *sk, int size)
1407 if (!sk_has_account(sk))
1409 sk->sk_forward_alloc += size;
1411 /* Avoid a possible overflow.
1412 * TCP send queues can make this happen, if sk_mem_reclaim()
1413 * is not called and more than 2 GBytes are released at once.
1415 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1416 * no need to hold that much forward allocation anyway.
1418 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1419 __sk_mem_reclaim(sk, 1 << 20);
1422 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1424 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1425 sk->sk_wmem_queued -= skb->truesize;
1426 sk_mem_uncharge(sk, skb->truesize);
1430 static inline void sock_release_ownership(struct sock *sk)
1432 if (sk->sk_lock.owned) {
1433 sk->sk_lock.owned = 0;
1435 /* The sk_lock has mutex_unlock() semantics: */
1436 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1441 * Macro so as to not evaluate some arguments when
1442 * lockdep is not enabled.
1444 * Mark both the sk_lock and the sk_lock.slock as a
1445 * per-address-family lock class.
1447 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1449 sk->sk_lock.owned = 0; \
1450 init_waitqueue_head(&sk->sk_lock.wq); \
1451 spin_lock_init(&(sk)->sk_lock.slock); \
1452 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1453 sizeof((sk)->sk_lock)); \
1454 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1456 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1459 #ifdef CONFIG_LOCKDEP
1460 static inline bool lockdep_sock_is_held(const struct sock *csk)
1462 struct sock *sk = (struct sock *)csk;
1464 return lockdep_is_held(&sk->sk_lock) ||
1465 lockdep_is_held(&sk->sk_lock.slock);
1469 void lock_sock_nested(struct sock *sk, int subclass);
1471 static inline void lock_sock(struct sock *sk)
1473 lock_sock_nested(sk, 0);
1476 void __release_sock(struct sock *sk);
1477 void release_sock(struct sock *sk);
1479 /* BH context may only use the following locking interface. */
1480 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1481 #define bh_lock_sock_nested(__sk) \
1482 spin_lock_nested(&((__sk)->sk_lock.slock), \
1483 SINGLE_DEPTH_NESTING)
1484 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1486 bool lock_sock_fast(struct sock *sk);
1488 * unlock_sock_fast - complement of lock_sock_fast
1492 * fast unlock socket for user context.
1493 * If slow mode is on, we call regular release_sock()
1495 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1500 spin_unlock_bh(&sk->sk_lock.slock);
1503 /* Used by processes to "lock" a socket state, so that
1504 * interrupts and bottom half handlers won't change it
1505 * from under us. It essentially blocks any incoming
1506 * packets, so that we won't get any new data or any
1507 * packets that change the state of the socket.
1509 * While locked, BH processing will add new packets to
1510 * the backlog queue. This queue is processed by the
1511 * owner of the socket lock right before it is released.
1513 * Since ~2.3.5 it is also exclusive sleep lock serializing
1514 * accesses from user process context.
1517 static inline void sock_owned_by_me(const struct sock *sk)
1519 #ifdef CONFIG_LOCKDEP
1520 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1524 static inline bool sock_owned_by_user(const struct sock *sk)
1526 sock_owned_by_me(sk);
1527 return sk->sk_lock.owned;
1530 /* no reclassification while locks are held */
1531 static inline bool sock_allow_reclassification(const struct sock *csk)
1533 struct sock *sk = (struct sock *)csk;
1535 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1538 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1539 struct proto *prot, int kern);
1540 void sk_free(struct sock *sk);
1541 void sk_destruct(struct sock *sk);
1542 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1543 void sk_free_unlock_clone(struct sock *sk);
1545 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1547 void __sock_wfree(struct sk_buff *skb);
1548 void sock_wfree(struct sk_buff *skb);
1549 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1551 void skb_orphan_partial(struct sk_buff *skb);
1552 void sock_rfree(struct sk_buff *skb);
1553 void sock_efree(struct sk_buff *skb);
1555 void sock_edemux(struct sk_buff *skb);
1557 #define sock_edemux sock_efree
1560 int sock_setsockopt(struct socket *sock, int level, int op,
1561 char __user *optval, unsigned int optlen);
1563 int sock_getsockopt(struct socket *sock, int level, int op,
1564 char __user *optval, int __user *optlen);
1565 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1566 int noblock, int *errcode);
1567 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1568 unsigned long data_len, int noblock,
1569 int *errcode, int max_page_order);
1570 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1571 void sock_kfree_s(struct sock *sk, void *mem, int size);
1572 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1573 void sk_send_sigurg(struct sock *sk);
1575 struct sockcm_cookie {
1580 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1581 struct sockcm_cookie *sockc);
1582 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1583 struct sockcm_cookie *sockc);
1586 * Functions to fill in entries in struct proto_ops when a protocol
1587 * does not implement a particular function.
1589 int sock_no_bind(struct socket *, struct sockaddr *, int);
1590 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1591 int sock_no_socketpair(struct socket *, struct socket *);
1592 int sock_no_accept(struct socket *, struct socket *, int, bool);
1593 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1594 unsigned int sock_no_poll(struct file *, struct socket *,
1595 struct poll_table_struct *);
1596 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1597 int sock_no_listen(struct socket *, int);
1598 int sock_no_shutdown(struct socket *, int);
1599 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1600 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1601 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1602 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1603 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1604 int sock_no_mmap(struct file *file, struct socket *sock,
1605 struct vm_area_struct *vma);
1606 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1607 size_t size, int flags);
1608 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1609 int offset, size_t size, int flags);
1612 * Functions to fill in entries in struct proto_ops when a protocol
1613 * uses the inet style.
1615 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1616 char __user *optval, int __user *optlen);
1617 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1619 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1620 char __user *optval, unsigned int optlen);
1621 int compat_sock_common_getsockopt(struct socket *sock, int level,
1622 int optname, char __user *optval, int __user *optlen);
1623 int compat_sock_common_setsockopt(struct socket *sock, int level,
1624 int optname, char __user *optval, unsigned int optlen);
1626 void sk_common_release(struct sock *sk);
1629 * Default socket callbacks and setup code
1632 /* Initialise core socket variables */
1633 void sock_init_data(struct socket *sock, struct sock *sk);
1636 * Socket reference counting postulates.
1638 * * Each user of socket SHOULD hold a reference count.
1639 * * Each access point to socket (an hash table bucket, reference from a list,
1640 * running timer, skb in flight MUST hold a reference count.
1641 * * When reference count hits 0, it means it will never increase back.
1642 * * When reference count hits 0, it means that no references from
1643 * outside exist to this socket and current process on current CPU
1644 * is last user and may/should destroy this socket.
1645 * * sk_free is called from any context: process, BH, IRQ. When
1646 * it is called, socket has no references from outside -> sk_free
1647 * may release descendant resources allocated by the socket, but
1648 * to the time when it is called, socket is NOT referenced by any
1649 * hash tables, lists etc.
1650 * * Packets, delivered from outside (from network or from another process)
1651 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1652 * when they sit in queue. Otherwise, packets will leak to hole, when
1653 * socket is looked up by one cpu and unhasing is made by another CPU.
1654 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1655 * (leak to backlog). Packet socket does all the processing inside
1656 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1657 * use separate SMP lock, so that they are prone too.
1660 /* Ungrab socket and destroy it, if it was the last reference. */
1661 static inline void sock_put(struct sock *sk)
1663 if (refcount_dec_and_test(&sk->sk_refcnt))
1666 /* Generic version of sock_put(), dealing with all sockets
1667 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1669 void sock_gen_put(struct sock *sk);
1671 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1672 unsigned int trim_cap, bool refcounted);
1673 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1676 return __sk_receive_skb(sk, skb, nested, 1, true);
1679 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1681 sk->sk_tx_queue_mapping = tx_queue;
1684 static inline void sk_tx_queue_clear(struct sock *sk)
1686 sk->sk_tx_queue_mapping = -1;
1689 static inline int sk_tx_queue_get(const struct sock *sk)
1691 return sk ? sk->sk_tx_queue_mapping : -1;
1694 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1696 sk->sk_socket = sock;
1699 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1701 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1702 return &rcu_dereference_raw(sk->sk_wq)->wait;
1704 /* Detach socket from process context.
1705 * Announce socket dead, detach it from wait queue and inode.
1706 * Note that parent inode held reference count on this struct sock,
1707 * we do not release it in this function, because protocol
1708 * probably wants some additional cleanups or even continuing
1709 * to work with this socket (TCP).
1711 static inline void sock_orphan(struct sock *sk)
1713 write_lock_bh(&sk->sk_callback_lock);
1714 sock_set_flag(sk, SOCK_DEAD);
1715 sk_set_socket(sk, NULL);
1717 write_unlock_bh(&sk->sk_callback_lock);
1720 static inline void sock_graft(struct sock *sk, struct socket *parent)
1722 WARN_ON(parent->sk);
1723 write_lock_bh(&sk->sk_callback_lock);
1724 sk->sk_wq = parent->wq;
1726 sk_set_socket(sk, parent);
1727 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1728 security_sock_graft(sk, parent);
1729 write_unlock_bh(&sk->sk_callback_lock);
1732 kuid_t sock_i_uid(struct sock *sk);
1733 unsigned long sock_i_ino(struct sock *sk);
1735 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1737 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1740 static inline u32 net_tx_rndhash(void)
1742 u32 v = prandom_u32();
1747 static inline void sk_set_txhash(struct sock *sk)
1749 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
1750 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
1753 static inline void sk_rethink_txhash(struct sock *sk)
1759 static inline struct dst_entry *
1760 __sk_dst_get(struct sock *sk)
1762 return rcu_dereference_check(sk->sk_dst_cache,
1763 lockdep_sock_is_held(sk));
1766 static inline struct dst_entry *
1767 sk_dst_get(struct sock *sk)
1769 struct dst_entry *dst;
1772 dst = rcu_dereference(sk->sk_dst_cache);
1773 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1779 static inline void dst_negative_advice(struct sock *sk)
1781 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1783 sk_rethink_txhash(sk);
1785 if (dst && dst->ops->negative_advice) {
1786 ndst = dst->ops->negative_advice(dst);
1789 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1790 sk_tx_queue_clear(sk);
1791 sk->sk_dst_pending_confirm = 0;
1797 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1799 struct dst_entry *old_dst;
1801 sk_tx_queue_clear(sk);
1802 sk->sk_dst_pending_confirm = 0;
1803 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1804 lockdep_sock_is_held(sk));
1805 rcu_assign_pointer(sk->sk_dst_cache, dst);
1806 dst_release(old_dst);
1810 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1812 struct dst_entry *old_dst;
1814 sk_tx_queue_clear(sk);
1815 sk->sk_dst_pending_confirm = 0;
1816 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1817 dst_release(old_dst);
1821 __sk_dst_reset(struct sock *sk)
1823 __sk_dst_set(sk, NULL);
1827 sk_dst_reset(struct sock *sk)
1829 sk_dst_set(sk, NULL);
1832 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1834 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1836 static inline void sk_dst_confirm(struct sock *sk)
1838 if (!sk->sk_dst_pending_confirm)
1839 sk->sk_dst_pending_confirm = 1;
1842 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1844 if (skb_get_dst_pending_confirm(skb)) {
1845 struct sock *sk = skb->sk;
1846 unsigned long now = jiffies;
1848 /* avoid dirtying neighbour */
1849 if (n->confirmed != now)
1851 if (sk && sk->sk_dst_pending_confirm)
1852 sk->sk_dst_pending_confirm = 0;
1856 bool sk_mc_loop(struct sock *sk);
1858 static inline bool sk_can_gso(const struct sock *sk)
1860 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1863 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1865 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1867 sk->sk_route_nocaps |= flags;
1868 sk->sk_route_caps &= ~flags;
1871 static inline bool sk_check_csum_caps(struct sock *sk)
1873 return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1874 (sk->sk_family == PF_INET &&
1875 (sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1876 (sk->sk_family == PF_INET6 &&
1877 (sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1880 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1881 struct iov_iter *from, char *to,
1882 int copy, int offset)
1884 if (skb->ip_summed == CHECKSUM_NONE) {
1886 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1888 skb->csum = csum_block_add(skb->csum, csum, offset);
1889 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1890 if (!copy_from_iter_full_nocache(to, copy, from))
1892 } else if (!copy_from_iter_full(to, copy, from))
1898 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1899 struct iov_iter *from, int copy)
1901 int err, offset = skb->len;
1903 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1906 __skb_trim(skb, offset);
1911 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1912 struct sk_buff *skb,
1918 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1924 skb->data_len += copy;
1925 skb->truesize += copy;
1926 sk->sk_wmem_queued += copy;
1927 sk_mem_charge(sk, copy);
1932 * sk_wmem_alloc_get - returns write allocations
1935 * Returns sk_wmem_alloc minus initial offset of one
1937 static inline int sk_wmem_alloc_get(const struct sock *sk)
1939 return refcount_read(&sk->sk_wmem_alloc) - 1;
1943 * sk_rmem_alloc_get - returns read allocations
1946 * Returns sk_rmem_alloc
1948 static inline int sk_rmem_alloc_get(const struct sock *sk)
1950 return atomic_read(&sk->sk_rmem_alloc);
1954 * sk_has_allocations - check if allocations are outstanding
1957 * Returns true if socket has write or read allocations
1959 static inline bool sk_has_allocations(const struct sock *sk)
1961 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1965 * skwq_has_sleeper - check if there are any waiting processes
1966 * @wq: struct socket_wq
1968 * Returns true if socket_wq has waiting processes
1970 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1971 * barrier call. They were added due to the race found within the tcp code.
1973 * Consider following tcp code paths::
1976 * sys_select receive packet
1978 * __add_wait_queue update tp->rcv_nxt
1980 * tp->rcv_nxt check sock_def_readable
1982 * schedule rcu_read_lock();
1983 * wq = rcu_dereference(sk->sk_wq);
1984 * if (wq && waitqueue_active(&wq->wait))
1985 * wake_up_interruptible(&wq->wait)
1989 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1990 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1991 * could then endup calling schedule and sleep forever if there are no more
1992 * data on the socket.
1995 static inline bool skwq_has_sleeper(struct socket_wq *wq)
1997 return wq && wq_has_sleeper(&wq->wait);
2001 * sock_poll_wait - place memory barrier behind the poll_wait call.
2003 * @wait_address: socket wait queue
2006 * See the comments in the wq_has_sleeper function.
2008 static inline void sock_poll_wait(struct file *filp,
2009 wait_queue_head_t *wait_address, poll_table *p)
2011 if (!poll_does_not_wait(p) && wait_address) {
2012 poll_wait(filp, wait_address, p);
2013 /* We need to be sure we are in sync with the
2014 * socket flags modification.
2016 * This memory barrier is paired in the wq_has_sleeper.
2022 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2024 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2025 u32 txhash = READ_ONCE(sk->sk_txhash);
2033 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2036 * Queue a received datagram if it will fit. Stream and sequenced
2037 * protocols can't normally use this as they need to fit buffers in
2038 * and play with them.
2040 * Inlined as it's very short and called for pretty much every
2041 * packet ever received.
2043 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2047 skb->destructor = sock_rfree;
2048 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2049 sk_mem_charge(sk, skb->truesize);
2052 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2053 unsigned long expires);
2055 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2057 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2058 struct sk_buff *skb, unsigned int flags,
2059 void (*destructor)(struct sock *sk,
2060 struct sk_buff *skb));
2061 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2062 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2064 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2065 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2068 * Recover an error report and clear atomically
2071 static inline int sock_error(struct sock *sk)
2074 if (likely(!sk->sk_err))
2076 err = xchg(&sk->sk_err, 0);
2080 static inline unsigned long sock_wspace(struct sock *sk)
2084 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2085 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2093 * We use sk->sk_wq_raw, from contexts knowing this
2094 * pointer is not NULL and cannot disappear/change.
2096 static inline void sk_set_bit(int nr, struct sock *sk)
2098 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2099 !sock_flag(sk, SOCK_FASYNC))
2102 set_bit(nr, &sk->sk_wq_raw->flags);
2105 static inline void sk_clear_bit(int nr, struct sock *sk)
2107 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2108 !sock_flag(sk, SOCK_FASYNC))
2111 clear_bit(nr, &sk->sk_wq_raw->flags);
2114 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2116 if (sock_flag(sk, SOCK_FASYNC)) {
2118 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2123 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2124 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2125 * Note: for send buffers, TCP works better if we can build two skbs at
2128 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2130 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2131 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2133 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2135 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2136 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2137 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2141 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2142 bool force_schedule);
2145 * sk_page_frag - return an appropriate page_frag
2148 * Use the per task page_frag instead of the per socket one for
2149 * optimization when we know that we're in the normal context and owns
2150 * everything that's associated with %current.
2152 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2153 * inside other socket operations and end up recursing into sk_page_frag()
2154 * while it's already in use.
2156 static inline struct page_frag *sk_page_frag(struct sock *sk)
2158 if (gfpflags_normal_context(sk->sk_allocation))
2159 return ¤t->task_frag;
2161 return &sk->sk_frag;
2164 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2167 * Default write policy as shown to user space via poll/select/SIGIO
2169 static inline bool sock_writeable(const struct sock *sk)
2171 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2174 static inline gfp_t gfp_any(void)
2176 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2179 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2181 return noblock ? 0 : sk->sk_rcvtimeo;
2184 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2186 return noblock ? 0 : sk->sk_sndtimeo;
2189 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2191 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2194 /* Alas, with timeout socket operations are not restartable.
2195 * Compare this to poll().
2197 static inline int sock_intr_errno(long timeo)
2199 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2202 struct sock_skb_cb {
2206 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2207 * using skb->cb[] would keep using it directly and utilize its
2208 * alignement guarantee.
2210 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2211 sizeof(struct sock_skb_cb)))
2213 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2214 SOCK_SKB_CB_OFFSET))
2216 #define sock_skb_cb_check_size(size) \
2217 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2220 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2222 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2223 atomic_read(&sk->sk_drops) : 0;
2226 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2228 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2230 atomic_add(segs, &sk->sk_drops);
2233 static inline ktime_t sock_read_timestamp(struct sock *sk)
2235 #if BITS_PER_LONG==32
2240 seq = read_seqbegin(&sk->sk_stamp_seq);
2242 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2246 return READ_ONCE(sk->sk_stamp);
2250 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2252 #if BITS_PER_LONG==32
2253 write_seqlock(&sk->sk_stamp_seq);
2255 write_sequnlock(&sk->sk_stamp_seq);
2257 WRITE_ONCE(sk->sk_stamp, kt);
2261 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2262 struct sk_buff *skb);
2263 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2264 struct sk_buff *skb);
2267 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2269 ktime_t kt = skb->tstamp;
2270 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2273 * generate control messages if
2274 * - receive time stamping in software requested
2275 * - software time stamp available and wanted
2276 * - hardware time stamps available and wanted
2278 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2279 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2280 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2281 (hwtstamps->hwtstamp &&
2282 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2283 __sock_recv_timestamp(msg, sk, skb);
2285 sock_write_timestamp(sk, kt);
2287 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2288 __sock_recv_wifi_status(msg, sk, skb);
2291 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2292 struct sk_buff *skb);
2294 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2295 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2296 struct sk_buff *skb)
2298 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2299 (1UL << SOCK_RCVTSTAMP))
2300 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2301 SOF_TIMESTAMPING_RAW_HARDWARE)
2303 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2304 __sock_recv_ts_and_drops(msg, sk, skb);
2305 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2306 sock_write_timestamp(sk, skb->tstamp);
2307 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2308 sock_write_timestamp(sk, 0);
2311 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2314 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2315 * @sk: socket sending this packet
2316 * @tsflags: timestamping flags to use
2317 * @tx_flags: completed with instructions for time stamping
2318 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2320 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2322 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2323 __u8 *tx_flags, __u32 *tskey)
2325 if (unlikely(tsflags)) {
2326 __sock_tx_timestamp(tsflags, tx_flags);
2327 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2328 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2329 *tskey = sk->sk_tskey++;
2331 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2332 *tx_flags |= SKBTX_WIFI_STATUS;
2335 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2338 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2341 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2343 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2344 &skb_shinfo(skb)->tskey);
2348 * sk_eat_skb - Release a skb if it is no longer needed
2349 * @sk: socket to eat this skb from
2350 * @skb: socket buffer to eat
2352 * This routine must be called with interrupts disabled or with the socket
2353 * locked so that the sk_buff queue operation is ok.
2355 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2357 __skb_unlink(skb, &sk->sk_receive_queue);
2362 struct net *sock_net(const struct sock *sk)
2364 return read_pnet(&sk->sk_net);
2368 void sock_net_set(struct sock *sk, struct net *net)
2370 write_pnet(&sk->sk_net, net);
2373 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2376 struct sock *sk = skb->sk;
2378 skb->destructor = NULL;
2385 /* This helper checks if a socket is a full socket,
2386 * ie _not_ a timewait or request socket.
2388 static inline bool sk_fullsock(const struct sock *sk)
2390 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2393 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2394 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2396 static inline bool sk_listener(const struct sock *sk)
2398 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2402 * sk_state_load - read sk->sk_state for lockless contexts
2403 * @sk: socket pointer
2405 * Paired with sk_state_store(). Used in places we do not hold socket lock :
2406 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2408 static inline int sk_state_load(const struct sock *sk)
2410 return smp_load_acquire(&sk->sk_state);
2414 * sk_state_store - update sk->sk_state
2415 * @sk: socket pointer
2416 * @newstate: new state
2418 * Paired with sk_state_load(). Should be used in contexts where
2419 * state change might impact lockless readers.
2421 static inline void sk_state_store(struct sock *sk, int newstate)
2423 smp_store_release(&sk->sk_state, newstate);
2426 void sock_enable_timestamp(struct sock *sk, int flag);
2427 int sock_get_timestamp(struct sock *, struct timeval __user *);
2428 int sock_get_timestampns(struct sock *, struct timespec __user *);
2429 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2432 bool sk_ns_capable(const struct sock *sk,
2433 struct user_namespace *user_ns, int cap);
2434 bool sk_capable(const struct sock *sk, int cap);
2435 bool sk_net_capable(const struct sock *sk, int cap);
2437 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2439 /* Take into consideration the size of the struct sk_buff overhead in the
2440 * determination of these values, since that is non-constant across
2441 * platforms. This makes socket queueing behavior and performance
2442 * not depend upon such differences.
2444 #define _SK_MEM_PACKETS 256
2445 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2446 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2447 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2449 extern __u32 sysctl_wmem_max;
2450 extern __u32 sysctl_rmem_max;
2452 extern int sysctl_tstamp_allow_data;
2453 extern int sysctl_optmem_max;
2455 extern __u32 sysctl_wmem_default;
2456 extern __u32 sysctl_rmem_default;
2458 /* On 32bit arches, an skb frag is limited to 2^15 */
2459 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2461 #endif /* _SOCK_H */