1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Definitions for the AF_INET socket handler.
9 * Version: @(#)sock.h 1.0.4 05/13/93
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche <flla@stud.uni-sb.de>
17 * Alan Cox : Volatiles in skbuff pointers. See
18 * skbuff comments. May be overdone,
19 * better to prove they can be removed
21 * Alan Cox : Added a zapped field for tcp to note
22 * a socket is reset and must stay shut up
23 * Alan Cox : New fields for options
24 * Pauline Middelink : identd support
25 * Alan Cox : Eliminate low level recv/recvfrom
26 * David S. Miller : New socket lookup architecture.
27 * Steve Whitehouse: Default routines for sock_ops
28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
29 * protinfo be just a void pointer, as the
30 * protocol specific parts were moved to
31 * respective headers and ipv4/v6, etc now
32 * use private slabcaches for its socks
33 * Pedro Hortas : New flags field for socket options
38 #include <linux/hardirq.h>
39 #include <linux/kernel.h>
40 #include <linux/list.h>
41 #include <linux/list_nulls.h>
42 #include <linux/timer.h>
43 #include <linux/cache.h>
44 #include <linux/bitops.h>
45 #include <linux/lockdep.h>
46 #include <linux/netdevice.h>
47 #include <linux/skbuff.h> /* struct sk_buff */
49 #include <linux/security.h>
50 #include <linux/slab.h>
51 #include <linux/uaccess.h>
52 #include <linux/page_counter.h>
53 #include <linux/memcontrol.h>
54 #include <linux/static_key.h>
55 #include <linux/sched.h>
56 #include <linux/wait.h>
57 #include <linux/cgroup-defs.h>
58 #include <linux/rbtree.h>
59 #include <linux/filter.h>
60 #include <linux/rculist_nulls.h>
61 #include <linux/poll.h>
63 #include <linux/atomic.h>
64 #include <linux/refcount.h>
66 #include <net/checksum.h>
67 #include <net/tcp_states.h>
68 #include <linux/net_tstamp.h>
69 #include <net/l3mdev.h>
72 * This structure really needs to be cleaned up.
73 * Most of it is for TCP, and not used by any of
74 * the other protocols.
77 /* Define this to get the SOCK_DBG debugging facility. */
78 #define SOCK_DEBUGGING
80 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
81 printk(KERN_DEBUG msg); } while (0)
83 /* Validate arguments and do nothing */
84 static inline __printf(2, 3)
85 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
90 /* This is the per-socket lock. The spinlock provides a synchronization
91 * between user contexts and software interrupt processing, whereas the
92 * mini-semaphore synchronizes multiple users amongst themselves.
99 * We express the mutex-alike socket_lock semantics
100 * to the lock validator by explicitly managing
101 * the slock as a lock variant (in addition to
104 #ifdef CONFIG_DEBUG_LOCK_ALLOC
105 struct lockdep_map dep_map;
113 typedef __u32 __bitwise __portpair;
114 typedef __u64 __bitwise __addrpair;
117 * struct sock_common - minimal network layer representation of sockets
118 * @skc_daddr: Foreign IPv4 addr
119 * @skc_rcv_saddr: Bound local IPv4 addr
120 * @skc_hash: hash value used with various protocol lookup tables
121 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
122 * @skc_dport: placeholder for inet_dport/tw_dport
123 * @skc_num: placeholder for inet_num/tw_num
124 * @skc_family: network address family
125 * @skc_state: Connection state
126 * @skc_reuse: %SO_REUSEADDR setting
127 * @skc_reuseport: %SO_REUSEPORT setting
128 * @skc_bound_dev_if: bound device index if != 0
129 * @skc_bind_node: bind hash linkage for various protocol lookup tables
130 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
131 * @skc_prot: protocol handlers inside a network family
132 * @skc_net: reference to the network namespace of this socket
133 * @skc_node: main hash linkage for various protocol lookup tables
134 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
135 * @skc_tx_queue_mapping: tx queue number for this connection
136 * @skc_rx_queue_mapping: rx queue number for this connection
137 * @skc_flags: place holder for sk_flags
138 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
139 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
140 * @skc_incoming_cpu: record/match cpu processing incoming packets
141 * @skc_refcnt: reference count
143 * This is the minimal network layer representation of sockets, the header
144 * for struct sock and struct inet_timewait_sock.
147 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
148 * address on 64bit arches : cf INET_MATCH()
151 __addrpair skc_addrpair;
154 __be32 skc_rcv_saddr;
158 unsigned int skc_hash;
159 __u16 skc_u16hashes[2];
161 /* skc_dport && skc_num must be grouped as well */
163 __portpair skc_portpair;
170 unsigned short skc_family;
171 volatile unsigned char skc_state;
172 unsigned char skc_reuse:4;
173 unsigned char skc_reuseport:1;
174 unsigned char skc_ipv6only:1;
175 unsigned char skc_net_refcnt:1;
176 int skc_bound_dev_if;
178 struct hlist_node skc_bind_node;
179 struct hlist_node skc_portaddr_node;
181 struct proto *skc_prot;
182 possible_net_t skc_net;
184 #if IS_ENABLED(CONFIG_IPV6)
185 struct in6_addr skc_v6_daddr;
186 struct in6_addr skc_v6_rcv_saddr;
189 atomic64_t skc_cookie;
191 /* following fields are padding to force
192 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
193 * assuming IPV6 is enabled. We use this padding differently
194 * for different kind of 'sockets'
197 unsigned long skc_flags;
198 struct sock *skc_listener; /* request_sock */
199 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
202 * fields between dontcopy_begin/dontcopy_end
203 * are not copied in sock_copy()
206 int skc_dontcopy_begin[0];
209 struct hlist_node skc_node;
210 struct hlist_nulls_node skc_nulls_node;
212 unsigned short skc_tx_queue_mapping;
214 unsigned short skc_rx_queue_mapping;
217 int skc_incoming_cpu;
219 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
222 refcount_t skc_refcnt;
224 int skc_dontcopy_end[0];
227 u32 skc_window_clamp;
228 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
233 struct bpf_sk_storage;
236 * struct sock - network layer representation of sockets
237 * @__sk_common: shared layout with inet_timewait_sock
238 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
239 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
240 * @sk_lock: synchronizer
241 * @sk_kern_sock: True if sock is using kernel lock classes
242 * @sk_rcvbuf: size of receive buffer in bytes
243 * @sk_wq: sock wait queue and async head
244 * @sk_rx_dst: receive input route used by early demux
245 * @sk_dst_cache: destination cache
246 * @sk_dst_pending_confirm: need to confirm neighbour
247 * @sk_policy: flow policy
248 * @sk_receive_queue: incoming packets
249 * @sk_wmem_alloc: transmit queue bytes committed
250 * @sk_tsq_flags: TCP Small Queues flags
251 * @sk_write_queue: Packet sending queue
252 * @sk_omem_alloc: "o" is "option" or "other"
253 * @sk_wmem_queued: persistent queue size
254 * @sk_forward_alloc: space allocated forward
255 * @sk_napi_id: id of the last napi context to receive data for sk
256 * @sk_ll_usec: usecs to busypoll when there is no data
257 * @sk_allocation: allocation mode
258 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
259 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
260 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
261 * @sk_sndbuf: size of send buffer in bytes
262 * @__sk_flags_offset: empty field used to determine location of bitfield
263 * @sk_padding: unused element for alignment
264 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
265 * @sk_no_check_rx: allow zero checksum in RX packets
266 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
267 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
268 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
269 * @sk_gso_max_size: Maximum GSO segment size to build
270 * @sk_gso_max_segs: Maximum number of GSO segments
271 * @sk_pacing_shift: scaling factor for TCP Small Queues
272 * @sk_lingertime: %SO_LINGER l_linger setting
273 * @sk_backlog: always used with the per-socket spinlock held
274 * @sk_callback_lock: used with the callbacks in the end of this struct
275 * @sk_error_queue: rarely used
276 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
277 * IPV6_ADDRFORM for instance)
278 * @sk_err: last error
279 * @sk_err_soft: errors that don't cause failure but are the cause of a
280 * persistent failure not just 'timed out'
281 * @sk_drops: raw/udp drops counter
282 * @sk_ack_backlog: current listen backlog
283 * @sk_max_ack_backlog: listen backlog set in listen()
284 * @sk_uid: user id of owner
285 * @sk_priority: %SO_PRIORITY setting
286 * @sk_type: socket type (%SOCK_STREAM, etc)
287 * @sk_protocol: which protocol this socket belongs in this network family
288 * @sk_peer_pid: &struct pid for this socket's peer
289 * @sk_peer_cred: %SO_PEERCRED setting
290 * @sk_rcvlowat: %SO_RCVLOWAT setting
291 * @sk_rcvtimeo: %SO_RCVTIMEO setting
292 * @sk_sndtimeo: %SO_SNDTIMEO setting
293 * @sk_txhash: computed flow hash for use on transmit
294 * @sk_filter: socket filtering instructions
295 * @sk_timer: sock cleanup timer
296 * @sk_stamp: time stamp of last packet received
297 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
298 * @sk_tsflags: SO_TIMESTAMPING socket options
299 * @sk_tskey: counter to disambiguate concurrent tstamp requests
300 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
301 * @sk_socket: Identd and reporting IO signals
302 * @sk_user_data: RPC layer private data
303 * @sk_frag: cached page frag
304 * @sk_peek_off: current peek_offset value
305 * @sk_send_head: front of stuff to transmit
306 * @sk_security: used by security modules
307 * @sk_mark: generic packet mark
308 * @sk_cgrp_data: cgroup data for this cgroup
309 * @sk_memcg: this socket's memory cgroup association
310 * @sk_write_pending: a write to stream socket waits to start
311 * @sk_state_change: callback to indicate change in the state of the sock
312 * @sk_data_ready: callback to indicate there is data to be processed
313 * @sk_write_space: callback to indicate there is bf sending space available
314 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
315 * @sk_backlog_rcv: callback to process the backlog
316 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
317 * @sk_reuseport_cb: reuseport group container
318 * @sk_rcu: used during RCU grace period
319 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
320 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
321 * @sk_txtime_unused: unused txtime flags
325 * Now struct inet_timewait_sock also uses sock_common, so please just
326 * don't add nothing before this first member (__sk_common) --acme
328 struct sock_common __sk_common;
329 #define sk_node __sk_common.skc_node
330 #define sk_nulls_node __sk_common.skc_nulls_node
331 #define sk_refcnt __sk_common.skc_refcnt
332 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
334 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
337 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
338 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
339 #define sk_hash __sk_common.skc_hash
340 #define sk_portpair __sk_common.skc_portpair
341 #define sk_num __sk_common.skc_num
342 #define sk_dport __sk_common.skc_dport
343 #define sk_addrpair __sk_common.skc_addrpair
344 #define sk_daddr __sk_common.skc_daddr
345 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
346 #define sk_family __sk_common.skc_family
347 #define sk_state __sk_common.skc_state
348 #define sk_reuse __sk_common.skc_reuse
349 #define sk_reuseport __sk_common.skc_reuseport
350 #define sk_ipv6only __sk_common.skc_ipv6only
351 #define sk_net_refcnt __sk_common.skc_net_refcnt
352 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
353 #define sk_bind_node __sk_common.skc_bind_node
354 #define sk_prot __sk_common.skc_prot
355 #define sk_net __sk_common.skc_net
356 #define sk_v6_daddr __sk_common.skc_v6_daddr
357 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
358 #define sk_cookie __sk_common.skc_cookie
359 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
360 #define sk_flags __sk_common.skc_flags
361 #define sk_rxhash __sk_common.skc_rxhash
363 socket_lock_t sk_lock;
366 struct sk_buff_head sk_error_queue;
367 struct sk_buff *sk_rx_skb_cache;
368 struct sk_buff_head sk_receive_queue;
370 * The backlog queue is special, it is always used with
371 * the per-socket spinlock held and requires low latency
372 * access. Therefore we special case it's implementation.
373 * Note : rmem_alloc is in this structure to fill a hole
374 * on 64bit arches, not because its logically part of
380 struct sk_buff *head;
381 struct sk_buff *tail;
383 #define sk_rmem_alloc sk_backlog.rmem_alloc
385 int sk_forward_alloc;
386 #ifdef CONFIG_NET_RX_BUSY_POLL
387 unsigned int sk_ll_usec;
388 /* ===== mostly read cache line ===== */
389 unsigned int sk_napi_id;
393 struct sk_filter __rcu *sk_filter;
395 struct socket_wq __rcu *sk_wq;
396 struct socket_wq *sk_wq_raw;
399 struct xfrm_policy __rcu *sk_policy[2];
401 struct dst_entry *sk_rx_dst;
402 struct dst_entry __rcu *sk_dst_cache;
403 atomic_t sk_omem_alloc;
406 /* ===== cache line for TX ===== */
408 refcount_t sk_wmem_alloc;
409 unsigned long sk_tsq_flags;
411 struct sk_buff *sk_send_head;
412 struct rb_root tcp_rtx_queue;
414 struct sk_buff *sk_tx_skb_cache;
415 struct sk_buff_head sk_write_queue;
417 int sk_write_pending;
418 __u32 sk_dst_pending_confirm;
419 u32 sk_pacing_status; /* see enum sk_pacing */
421 struct timer_list sk_timer;
424 unsigned long sk_pacing_rate; /* bytes per second */
425 unsigned long sk_max_pacing_rate;
426 struct page_frag sk_frag;
427 netdev_features_t sk_route_caps;
428 netdev_features_t sk_route_nocaps;
429 netdev_features_t sk_route_forced_caps;
431 unsigned int sk_gso_max_size;
436 * Because of non atomicity rules, all
437 * changes are protected by socket lock.
439 unsigned int __sk_flags_offset[0];
440 #ifdef __BIG_ENDIAN_BITFIELD
441 #define SK_FL_PROTO_SHIFT 16
442 #define SK_FL_PROTO_MASK 0x00ff0000
444 #define SK_FL_TYPE_SHIFT 0
445 #define SK_FL_TYPE_MASK 0x0000ffff
447 #define SK_FL_PROTO_SHIFT 8
448 #define SK_FL_PROTO_MASK 0x0000ff00
450 #define SK_FL_TYPE_SHIFT 16
451 #define SK_FL_TYPE_MASK 0xffff0000
454 unsigned int sk_padding : 1,
461 #define SK_PROTOCOL_MAX U8_MAX
464 unsigned long sk_lingertime;
465 struct proto *sk_prot_creator;
466 rwlock_t sk_callback_lock;
470 u32 sk_max_ack_backlog;
472 struct pid *sk_peer_pid;
473 const struct cred *sk_peer_cred;
476 #if BITS_PER_LONG==32
477 seqlock_t sk_stamp_seq;
485 u8 sk_txtime_deadline_mode : 1,
486 sk_txtime_report_errors : 1,
487 sk_txtime_unused : 6;
489 struct socket *sk_socket;
491 #ifdef CONFIG_SECURITY
494 struct sock_cgroup_data sk_cgrp_data;
495 struct mem_cgroup *sk_memcg;
496 void (*sk_state_change)(struct sock *sk);
497 void (*sk_data_ready)(struct sock *sk);
498 void (*sk_write_space)(struct sock *sk);
499 void (*sk_error_report)(struct sock *sk);
500 int (*sk_backlog_rcv)(struct sock *sk,
501 struct sk_buff *skb);
502 #ifdef CONFIG_SOCK_VALIDATE_XMIT
503 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
504 struct net_device *dev,
505 struct sk_buff *skb);
507 void (*sk_destruct)(struct sock *sk);
508 struct sock_reuseport __rcu *sk_reuseport_cb;
509 #ifdef CONFIG_BPF_SYSCALL
510 struct bpf_sk_storage __rcu *sk_bpf_storage;
512 struct rcu_head sk_rcu;
517 SK_PACING_NEEDED = 1,
521 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
523 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
524 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
527 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
528 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
529 * on a socket means that the socket will reuse everybody else's port
530 * without looking at the other's sk_reuse value.
533 #define SK_NO_REUSE 0
534 #define SK_CAN_REUSE 1
535 #define SK_FORCE_REUSE 2
537 int sk_set_peek_off(struct sock *sk, int val);
539 static inline int sk_peek_offset(struct sock *sk, int flags)
541 if (unlikely(flags & MSG_PEEK)) {
542 return READ_ONCE(sk->sk_peek_off);
548 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
550 s32 off = READ_ONCE(sk->sk_peek_off);
552 if (unlikely(off >= 0)) {
553 off = max_t(s32, off - val, 0);
554 WRITE_ONCE(sk->sk_peek_off, off);
558 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
560 sk_peek_offset_bwd(sk, -val);
564 * Hashed lists helper routines
566 static inline struct sock *sk_entry(const struct hlist_node *node)
568 return hlist_entry(node, struct sock, sk_node);
571 static inline struct sock *__sk_head(const struct hlist_head *head)
573 return hlist_entry(head->first, struct sock, sk_node);
576 static inline struct sock *sk_head(const struct hlist_head *head)
578 return hlist_empty(head) ? NULL : __sk_head(head);
581 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
583 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
586 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
588 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
591 static inline struct sock *sk_next(const struct sock *sk)
593 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
596 static inline struct sock *sk_nulls_next(const struct sock *sk)
598 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
599 hlist_nulls_entry(sk->sk_nulls_node.next,
600 struct sock, sk_nulls_node) :
604 static inline bool sk_unhashed(const struct sock *sk)
606 return hlist_unhashed(&sk->sk_node);
609 static inline bool sk_hashed(const struct sock *sk)
611 return !sk_unhashed(sk);
614 static inline void sk_node_init(struct hlist_node *node)
619 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
624 static inline void __sk_del_node(struct sock *sk)
626 __hlist_del(&sk->sk_node);
629 /* NB: equivalent to hlist_del_init_rcu */
630 static inline bool __sk_del_node_init(struct sock *sk)
634 sk_node_init(&sk->sk_node);
640 /* Grab socket reference count. This operation is valid only
641 when sk is ALREADY grabbed f.e. it is found in hash table
642 or a list and the lookup is made under lock preventing hash table
646 static __always_inline void sock_hold(struct sock *sk)
648 refcount_inc(&sk->sk_refcnt);
651 /* Ungrab socket in the context, which assumes that socket refcnt
652 cannot hit zero, f.e. it is true in context of any socketcall.
654 static __always_inline void __sock_put(struct sock *sk)
656 refcount_dec(&sk->sk_refcnt);
659 static inline bool sk_del_node_init(struct sock *sk)
661 bool rc = __sk_del_node_init(sk);
664 /* paranoid for a while -acme */
665 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
670 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
672 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
675 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
681 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
683 bool rc = __sk_nulls_del_node_init_rcu(sk);
686 /* paranoid for a while -acme */
687 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
693 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
695 hlist_add_head(&sk->sk_node, list);
698 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
701 __sk_add_node(sk, list);
704 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
707 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
708 sk->sk_family == AF_INET6)
709 hlist_add_tail_rcu(&sk->sk_node, list);
711 hlist_add_head_rcu(&sk->sk_node, list);
714 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
717 hlist_add_tail_rcu(&sk->sk_node, list);
720 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
722 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
725 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
727 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
730 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
733 __sk_nulls_add_node_rcu(sk, list);
736 static inline void __sk_del_bind_node(struct sock *sk)
738 __hlist_del(&sk->sk_bind_node);
741 static inline void sk_add_bind_node(struct sock *sk,
742 struct hlist_head *list)
744 hlist_add_head(&sk->sk_bind_node, list);
747 #define sk_for_each(__sk, list) \
748 hlist_for_each_entry(__sk, list, sk_node)
749 #define sk_for_each_rcu(__sk, list) \
750 hlist_for_each_entry_rcu(__sk, list, sk_node)
751 #define sk_nulls_for_each(__sk, node, list) \
752 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
753 #define sk_nulls_for_each_rcu(__sk, node, list) \
754 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
755 #define sk_for_each_from(__sk) \
756 hlist_for_each_entry_from(__sk, sk_node)
757 #define sk_nulls_for_each_from(__sk, node) \
758 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
759 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
760 #define sk_for_each_safe(__sk, tmp, list) \
761 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
762 #define sk_for_each_bound(__sk, list) \
763 hlist_for_each_entry(__sk, list, sk_bind_node)
766 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
767 * @tpos: the type * to use as a loop cursor.
768 * @pos: the &struct hlist_node to use as a loop cursor.
769 * @head: the head for your list.
770 * @offset: offset of hlist_node within the struct.
773 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
774 for (pos = rcu_dereference(hlist_first_rcu(head)); \
776 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
777 pos = rcu_dereference(hlist_next_rcu(pos)))
779 static inline struct user_namespace *sk_user_ns(struct sock *sk)
781 /* Careful only use this in a context where these parameters
782 * can not change and must all be valid, such as recvmsg from
785 return sk->sk_socket->file->f_cred->user_ns;
799 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
800 SOCK_DBG, /* %SO_DEBUG setting */
801 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
802 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
803 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
804 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
805 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
806 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
807 SOCK_FASYNC, /* fasync() active */
809 SOCK_ZEROCOPY, /* buffers from userspace */
810 SOCK_WIFI_STATUS, /* push wifi status to userspace */
811 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
812 * Will use last 4 bytes of packet sent from
813 * user-space instead.
815 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
816 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
817 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
819 SOCK_XDP, /* XDP is attached */
820 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
823 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
825 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
827 nsk->sk_flags = osk->sk_flags;
830 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
832 __set_bit(flag, &sk->sk_flags);
835 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
837 __clear_bit(flag, &sk->sk_flags);
840 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
842 return test_bit(flag, &sk->sk_flags);
846 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
847 static inline int sk_memalloc_socks(void)
849 return static_branch_unlikely(&memalloc_socks_key);
853 static inline int sk_memalloc_socks(void)
860 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
862 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
865 static inline void sk_acceptq_removed(struct sock *sk)
867 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
870 static inline void sk_acceptq_added(struct sock *sk)
872 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
875 static inline bool sk_acceptq_is_full(const struct sock *sk)
877 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
881 * Compute minimal free write space needed to queue new packets.
883 static inline int sk_stream_min_wspace(const struct sock *sk)
885 return READ_ONCE(sk->sk_wmem_queued) >> 1;
888 static inline int sk_stream_wspace(const struct sock *sk)
890 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
893 static inline void sk_wmem_queued_add(struct sock *sk, int val)
895 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
898 void sk_stream_write_space(struct sock *sk);
900 /* OOB backlog add */
901 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
903 /* dont let skb dst not refcounted, we are going to leave rcu lock */
906 if (!sk->sk_backlog.tail)
907 WRITE_ONCE(sk->sk_backlog.head, skb);
909 sk->sk_backlog.tail->next = skb;
911 WRITE_ONCE(sk->sk_backlog.tail, skb);
916 * Take into account size of receive queue and backlog queue
917 * Do not take into account this skb truesize,
918 * to allow even a single big packet to come.
920 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
922 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
924 return qsize > limit;
927 /* The per-socket spinlock must be held here. */
928 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
931 if (sk_rcvqueues_full(sk, limit))
935 * If the skb was allocated from pfmemalloc reserves, only
936 * allow SOCK_MEMALLOC sockets to use it as this socket is
937 * helping free memory
939 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
942 __sk_add_backlog(sk, skb);
943 sk->sk_backlog.len += skb->truesize;
947 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
949 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
951 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
952 return __sk_backlog_rcv(sk, skb);
954 return sk->sk_backlog_rcv(sk, skb);
957 static inline void sk_incoming_cpu_update(struct sock *sk)
959 int cpu = raw_smp_processor_id();
961 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
962 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
965 static inline void sock_rps_record_flow_hash(__u32 hash)
968 struct rps_sock_flow_table *sock_flow_table;
971 sock_flow_table = rcu_dereference(rps_sock_flow_table);
972 rps_record_sock_flow(sock_flow_table, hash);
977 static inline void sock_rps_record_flow(const struct sock *sk)
980 if (static_branch_unlikely(&rfs_needed)) {
981 /* Reading sk->sk_rxhash might incur an expensive cache line
984 * TCP_ESTABLISHED does cover almost all states where RFS
985 * might be useful, and is cheaper [1] than testing :
986 * IPv4: inet_sk(sk)->inet_daddr
987 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
988 * OR an additional socket flag
989 * [1] : sk_state and sk_prot are in the same cache line.
991 if (sk->sk_state == TCP_ESTABLISHED)
992 sock_rps_record_flow_hash(sk->sk_rxhash);
997 static inline void sock_rps_save_rxhash(struct sock *sk,
998 const struct sk_buff *skb)
1001 if (unlikely(sk->sk_rxhash != skb->hash))
1002 sk->sk_rxhash = skb->hash;
1006 static inline void sock_rps_reset_rxhash(struct sock *sk)
1013 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1015 release_sock(__sk); \
1016 __rc = __condition; \
1018 *(__timeo) = wait_woken(__wait, \
1019 TASK_INTERRUPTIBLE, \
1022 sched_annotate_sleep(); \
1024 __rc = __condition; \
1028 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1029 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1030 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1031 int sk_stream_error(struct sock *sk, int flags, int err);
1032 void sk_stream_kill_queues(struct sock *sk);
1033 void sk_set_memalloc(struct sock *sk);
1034 void sk_clear_memalloc(struct sock *sk);
1036 void __sk_flush_backlog(struct sock *sk);
1038 static inline bool sk_flush_backlog(struct sock *sk)
1040 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1041 __sk_flush_backlog(sk);
1047 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1049 struct request_sock_ops;
1050 struct timewait_sock_ops;
1051 struct inet_hashinfo;
1052 struct raw_hashinfo;
1053 struct smc_hashinfo;
1057 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1058 * un-modified. Special care is taken when initializing object to zero.
1060 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1062 if (offsetof(struct sock, sk_node.next) != 0)
1063 memset(sk, 0, offsetof(struct sock, sk_node.next));
1064 memset(&sk->sk_node.pprev, 0,
1065 size - offsetof(struct sock, sk_node.pprev));
1068 /* Networking protocol blocks we attach to sockets.
1069 * socket layer -> transport layer interface
1072 void (*close)(struct sock *sk,
1074 int (*pre_connect)(struct sock *sk,
1075 struct sockaddr *uaddr,
1077 int (*connect)(struct sock *sk,
1078 struct sockaddr *uaddr,
1080 int (*disconnect)(struct sock *sk, int flags);
1082 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1085 int (*ioctl)(struct sock *sk, int cmd,
1087 int (*init)(struct sock *sk);
1088 void (*destroy)(struct sock *sk);
1089 void (*shutdown)(struct sock *sk, int how);
1090 int (*setsockopt)(struct sock *sk, int level,
1091 int optname, char __user *optval,
1092 unsigned int optlen);
1093 int (*getsockopt)(struct sock *sk, int level,
1094 int optname, char __user *optval,
1095 int __user *option);
1096 void (*keepalive)(struct sock *sk, int valbool);
1097 #ifdef CONFIG_COMPAT
1098 int (*compat_setsockopt)(struct sock *sk,
1100 int optname, char __user *optval,
1101 unsigned int optlen);
1102 int (*compat_getsockopt)(struct sock *sk,
1104 int optname, char __user *optval,
1105 int __user *option);
1106 int (*compat_ioctl)(struct sock *sk,
1107 unsigned int cmd, unsigned long arg);
1109 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1111 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1112 size_t len, int noblock, int flags,
1114 int (*sendpage)(struct sock *sk, struct page *page,
1115 int offset, size_t size, int flags);
1116 int (*bind)(struct sock *sk,
1117 struct sockaddr *uaddr, int addr_len);
1119 int (*backlog_rcv) (struct sock *sk,
1120 struct sk_buff *skb);
1122 void (*release_cb)(struct sock *sk);
1124 /* Keeping track of sk's, looking them up, and port selection methods. */
1125 int (*hash)(struct sock *sk);
1126 void (*unhash)(struct sock *sk);
1127 void (*rehash)(struct sock *sk);
1128 int (*get_port)(struct sock *sk, unsigned short snum);
1130 /* Keeping track of sockets in use */
1131 #ifdef CONFIG_PROC_FS
1132 unsigned int inuse_idx;
1135 bool (*stream_memory_free)(const struct sock *sk, int wake);
1136 bool (*stream_memory_read)(const struct sock *sk);
1137 /* Memory pressure */
1138 void (*enter_memory_pressure)(struct sock *sk);
1139 void (*leave_memory_pressure)(struct sock *sk);
1140 atomic_long_t *memory_allocated; /* Current allocated memory. */
1141 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1143 * Pressure flag: try to collapse.
1144 * Technical note: it is used by multiple contexts non atomically.
1145 * All the __sk_mem_schedule() is of this nature: accounting
1146 * is strict, actions are advisory and have some latency.
1148 unsigned long *memory_pressure;
1153 u32 sysctl_wmem_offset;
1154 u32 sysctl_rmem_offset;
1159 struct kmem_cache *slab;
1160 unsigned int obj_size;
1161 slab_flags_t slab_flags;
1162 unsigned int useroffset; /* Usercopy region offset */
1163 unsigned int usersize; /* Usercopy region size */
1165 struct percpu_counter *orphan_count;
1167 struct request_sock_ops *rsk_prot;
1168 struct timewait_sock_ops *twsk_prot;
1171 struct inet_hashinfo *hashinfo;
1172 struct udp_table *udp_table;
1173 struct raw_hashinfo *raw_hash;
1174 struct smc_hashinfo *smc_hash;
1177 struct module *owner;
1181 struct list_head node;
1182 #ifdef SOCK_REFCNT_DEBUG
1185 int (*diag_destroy)(struct sock *sk, int err);
1186 } __randomize_layout;
1188 int proto_register(struct proto *prot, int alloc_slab);
1189 void proto_unregister(struct proto *prot);
1190 int sock_load_diag_module(int family, int protocol);
1192 #ifdef SOCK_REFCNT_DEBUG
1193 static inline void sk_refcnt_debug_inc(struct sock *sk)
1195 atomic_inc(&sk->sk_prot->socks);
1198 static inline void sk_refcnt_debug_dec(struct sock *sk)
1200 atomic_dec(&sk->sk_prot->socks);
1201 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1202 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1205 static inline void sk_refcnt_debug_release(const struct sock *sk)
1207 if (refcount_read(&sk->sk_refcnt) != 1)
1208 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1209 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1211 #else /* SOCK_REFCNT_DEBUG */
1212 #define sk_refcnt_debug_inc(sk) do { } while (0)
1213 #define sk_refcnt_debug_dec(sk) do { } while (0)
1214 #define sk_refcnt_debug_release(sk) do { } while (0)
1215 #endif /* SOCK_REFCNT_DEBUG */
1217 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1219 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1222 return sk->sk_prot->stream_memory_free ?
1223 sk->sk_prot->stream_memory_free(sk, wake) : true;
1226 static inline bool sk_stream_memory_free(const struct sock *sk)
1228 return __sk_stream_memory_free(sk, 0);
1231 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1233 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1234 __sk_stream_memory_free(sk, wake);
1237 static inline bool sk_stream_is_writeable(const struct sock *sk)
1239 return __sk_stream_is_writeable(sk, 0);
1242 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1243 struct cgroup *ancestor)
1245 #ifdef CONFIG_SOCK_CGROUP_DATA
1246 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1253 static inline bool sk_has_memory_pressure(const struct sock *sk)
1255 return sk->sk_prot->memory_pressure != NULL;
1258 static inline bool sk_under_memory_pressure(const struct sock *sk)
1260 if (!sk->sk_prot->memory_pressure)
1263 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1264 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1267 return !!*sk->sk_prot->memory_pressure;
1271 sk_memory_allocated(const struct sock *sk)
1273 return atomic_long_read(sk->sk_prot->memory_allocated);
1277 sk_memory_allocated_add(struct sock *sk, int amt)
1279 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1283 sk_memory_allocated_sub(struct sock *sk, int amt)
1285 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1288 static inline void sk_sockets_allocated_dec(struct sock *sk)
1290 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1293 static inline void sk_sockets_allocated_inc(struct sock *sk)
1295 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1299 sk_sockets_allocated_read_positive(struct sock *sk)
1301 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1305 proto_sockets_allocated_sum_positive(struct proto *prot)
1307 return percpu_counter_sum_positive(prot->sockets_allocated);
1311 proto_memory_allocated(struct proto *prot)
1313 return atomic_long_read(prot->memory_allocated);
1317 proto_memory_pressure(struct proto *prot)
1319 if (!prot->memory_pressure)
1321 return !!*prot->memory_pressure;
1325 #ifdef CONFIG_PROC_FS
1326 /* Called with local bh disabled */
1327 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1328 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1329 int sock_inuse_get(struct net *net);
1331 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1338 /* With per-bucket locks this operation is not-atomic, so that
1339 * this version is not worse.
1341 static inline int __sk_prot_rehash(struct sock *sk)
1343 sk->sk_prot->unhash(sk);
1344 return sk->sk_prot->hash(sk);
1347 /* About 10 seconds */
1348 #define SOCK_DESTROY_TIME (10*HZ)
1350 /* Sockets 0-1023 can't be bound to unless you are superuser */
1351 #define PROT_SOCK 1024
1353 #define SHUTDOWN_MASK 3
1354 #define RCV_SHUTDOWN 1
1355 #define SEND_SHUTDOWN 2
1357 #define SOCK_SNDBUF_LOCK 1
1358 #define SOCK_RCVBUF_LOCK 2
1359 #define SOCK_BINDADDR_LOCK 4
1360 #define SOCK_BINDPORT_LOCK 8
1362 struct socket_alloc {
1363 struct socket socket;
1364 struct inode vfs_inode;
1367 static inline struct socket *SOCKET_I(struct inode *inode)
1369 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1372 static inline struct inode *SOCK_INODE(struct socket *socket)
1374 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1378 * Functions for memory accounting
1380 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1381 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1382 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1383 void __sk_mem_reclaim(struct sock *sk, int amount);
1385 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1386 * do not necessarily have 16x time more memory than 4KB ones.
1388 #define SK_MEM_QUANTUM 4096
1389 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1390 #define SK_MEM_SEND 0
1391 #define SK_MEM_RECV 1
1393 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1394 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1396 long val = sk->sk_prot->sysctl_mem[index];
1398 #if PAGE_SIZE > SK_MEM_QUANTUM
1399 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1400 #elif PAGE_SIZE < SK_MEM_QUANTUM
1401 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1406 static inline int sk_mem_pages(int amt)
1408 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1411 static inline bool sk_has_account(struct sock *sk)
1413 /* return true if protocol supports memory accounting */
1414 return !!sk->sk_prot->memory_allocated;
1417 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1419 if (!sk_has_account(sk))
1421 return size <= sk->sk_forward_alloc ||
1422 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1426 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1428 if (!sk_has_account(sk))
1430 return size<= sk->sk_forward_alloc ||
1431 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1432 skb_pfmemalloc(skb);
1435 static inline void sk_mem_reclaim(struct sock *sk)
1437 if (!sk_has_account(sk))
1439 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1440 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1443 static inline void sk_mem_reclaim_partial(struct sock *sk)
1445 if (!sk_has_account(sk))
1447 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1448 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1451 static inline void sk_mem_charge(struct sock *sk, int size)
1453 if (!sk_has_account(sk))
1455 sk->sk_forward_alloc -= size;
1458 static inline void sk_mem_uncharge(struct sock *sk, int size)
1460 if (!sk_has_account(sk))
1462 sk->sk_forward_alloc += size;
1464 /* Avoid a possible overflow.
1465 * TCP send queues can make this happen, if sk_mem_reclaim()
1466 * is not called and more than 2 GBytes are released at once.
1468 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1469 * no need to hold that much forward allocation anyway.
1471 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1472 __sk_mem_reclaim(sk, 1 << 20);
1475 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
1476 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1478 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1479 sk_wmem_queued_add(sk, -skb->truesize);
1480 sk_mem_uncharge(sk, skb->truesize);
1481 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1482 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1483 skb_zcopy_clear(skb, true);
1484 sk->sk_tx_skb_cache = skb;
1490 static inline void sock_release_ownership(struct sock *sk)
1492 if (sk->sk_lock.owned) {
1493 sk->sk_lock.owned = 0;
1495 /* The sk_lock has mutex_unlock() semantics: */
1496 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1501 * Macro so as to not evaluate some arguments when
1502 * lockdep is not enabled.
1504 * Mark both the sk_lock and the sk_lock.slock as a
1505 * per-address-family lock class.
1507 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1509 sk->sk_lock.owned = 0; \
1510 init_waitqueue_head(&sk->sk_lock.wq); \
1511 spin_lock_init(&(sk)->sk_lock.slock); \
1512 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1513 sizeof((sk)->sk_lock)); \
1514 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1516 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1519 #ifdef CONFIG_LOCKDEP
1520 static inline bool lockdep_sock_is_held(const struct sock *sk)
1522 return lockdep_is_held(&sk->sk_lock) ||
1523 lockdep_is_held(&sk->sk_lock.slock);
1527 void lock_sock_nested(struct sock *sk, int subclass);
1529 static inline void lock_sock(struct sock *sk)
1531 lock_sock_nested(sk, 0);
1534 void __release_sock(struct sock *sk);
1535 void release_sock(struct sock *sk);
1537 /* BH context may only use the following locking interface. */
1538 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1539 #define bh_lock_sock_nested(__sk) \
1540 spin_lock_nested(&((__sk)->sk_lock.slock), \
1541 SINGLE_DEPTH_NESTING)
1542 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1544 bool lock_sock_fast(struct sock *sk);
1546 * unlock_sock_fast - complement of lock_sock_fast
1550 * fast unlock socket for user context.
1551 * If slow mode is on, we call regular release_sock()
1553 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1558 spin_unlock_bh(&sk->sk_lock.slock);
1561 /* Used by processes to "lock" a socket state, so that
1562 * interrupts and bottom half handlers won't change it
1563 * from under us. It essentially blocks any incoming
1564 * packets, so that we won't get any new data or any
1565 * packets that change the state of the socket.
1567 * While locked, BH processing will add new packets to
1568 * the backlog queue. This queue is processed by the
1569 * owner of the socket lock right before it is released.
1571 * Since ~2.3.5 it is also exclusive sleep lock serializing
1572 * accesses from user process context.
1575 static inline void sock_owned_by_me(const struct sock *sk)
1577 #ifdef CONFIG_LOCKDEP
1578 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1582 static inline bool sock_owned_by_user(const struct sock *sk)
1584 sock_owned_by_me(sk);
1585 return sk->sk_lock.owned;
1588 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1590 return sk->sk_lock.owned;
1593 /* no reclassification while locks are held */
1594 static inline bool sock_allow_reclassification(const struct sock *csk)
1596 struct sock *sk = (struct sock *)csk;
1598 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1601 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1602 struct proto *prot, int kern);
1603 void sk_free(struct sock *sk);
1604 void sk_destruct(struct sock *sk);
1605 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1606 void sk_free_unlock_clone(struct sock *sk);
1608 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1610 void __sock_wfree(struct sk_buff *skb);
1611 void sock_wfree(struct sk_buff *skb);
1612 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1614 void skb_orphan_partial(struct sk_buff *skb);
1615 void sock_rfree(struct sk_buff *skb);
1616 void sock_efree(struct sk_buff *skb);
1618 void sock_edemux(struct sk_buff *skb);
1620 #define sock_edemux sock_efree
1623 int sock_setsockopt(struct socket *sock, int level, int op,
1624 char __user *optval, unsigned int optlen);
1626 int sock_getsockopt(struct socket *sock, int level, int op,
1627 char __user *optval, int __user *optlen);
1628 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1629 bool timeval, bool time32);
1630 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1631 int noblock, int *errcode);
1632 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1633 unsigned long data_len, int noblock,
1634 int *errcode, int max_page_order);
1635 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1636 void sock_kfree_s(struct sock *sk, void *mem, int size);
1637 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1638 void sk_send_sigurg(struct sock *sk);
1640 struct sockcm_cookie {
1646 static inline void sockcm_init(struct sockcm_cookie *sockc,
1647 const struct sock *sk)
1649 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1652 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1653 struct sockcm_cookie *sockc);
1654 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1655 struct sockcm_cookie *sockc);
1658 * Functions to fill in entries in struct proto_ops when a protocol
1659 * does not implement a particular function.
1661 int sock_no_bind(struct socket *, struct sockaddr *, int);
1662 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1663 int sock_no_socketpair(struct socket *, struct socket *);
1664 int sock_no_accept(struct socket *, struct socket *, int, bool);
1665 int sock_no_getname(struct socket *, struct sockaddr *, int);
1666 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1667 int sock_no_listen(struct socket *, int);
1668 int sock_no_shutdown(struct socket *, int);
1669 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1670 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1671 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1672 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1673 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1674 int sock_no_mmap(struct file *file, struct socket *sock,
1675 struct vm_area_struct *vma);
1676 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1677 size_t size, int flags);
1678 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1679 int offset, size_t size, int flags);
1682 * Functions to fill in entries in struct proto_ops when a protocol
1683 * uses the inet style.
1685 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1686 char __user *optval, int __user *optlen);
1687 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1689 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1690 char __user *optval, unsigned int optlen);
1691 int compat_sock_common_getsockopt(struct socket *sock, int level,
1692 int optname, char __user *optval, int __user *optlen);
1693 int compat_sock_common_setsockopt(struct socket *sock, int level,
1694 int optname, char __user *optval, unsigned int optlen);
1696 void sk_common_release(struct sock *sk);
1699 * Default socket callbacks and setup code
1702 /* Initialise core socket variables */
1703 void sock_init_data(struct socket *sock, struct sock *sk);
1706 * Socket reference counting postulates.
1708 * * Each user of socket SHOULD hold a reference count.
1709 * * Each access point to socket (an hash table bucket, reference from a list,
1710 * running timer, skb in flight MUST hold a reference count.
1711 * * When reference count hits 0, it means it will never increase back.
1712 * * When reference count hits 0, it means that no references from
1713 * outside exist to this socket and current process on current CPU
1714 * is last user and may/should destroy this socket.
1715 * * sk_free is called from any context: process, BH, IRQ. When
1716 * it is called, socket has no references from outside -> sk_free
1717 * may release descendant resources allocated by the socket, but
1718 * to the time when it is called, socket is NOT referenced by any
1719 * hash tables, lists etc.
1720 * * Packets, delivered from outside (from network or from another process)
1721 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1722 * when they sit in queue. Otherwise, packets will leak to hole, when
1723 * socket is looked up by one cpu and unhasing is made by another CPU.
1724 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1725 * (leak to backlog). Packet socket does all the processing inside
1726 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1727 * use separate SMP lock, so that they are prone too.
1730 /* Ungrab socket and destroy it, if it was the last reference. */
1731 static inline void sock_put(struct sock *sk)
1733 if (refcount_dec_and_test(&sk->sk_refcnt))
1736 /* Generic version of sock_put(), dealing with all sockets
1737 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1739 void sock_gen_put(struct sock *sk);
1741 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1742 unsigned int trim_cap, bool refcounted);
1743 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1746 return __sk_receive_skb(sk, skb, nested, 1, true);
1749 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1751 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1752 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1754 sk->sk_tx_queue_mapping = tx_queue;
1757 #define NO_QUEUE_MAPPING USHRT_MAX
1759 static inline void sk_tx_queue_clear(struct sock *sk)
1761 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1764 static inline int sk_tx_queue_get(const struct sock *sk)
1766 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1767 return sk->sk_tx_queue_mapping;
1772 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1775 if (skb_rx_queue_recorded(skb)) {
1776 u16 rx_queue = skb_get_rx_queue(skb);
1778 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1781 sk->sk_rx_queue_mapping = rx_queue;
1786 static inline void sk_rx_queue_clear(struct sock *sk)
1789 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1794 static inline int sk_rx_queue_get(const struct sock *sk)
1796 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1797 return sk->sk_rx_queue_mapping;
1803 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1805 sk_tx_queue_clear(sk);
1806 sk->sk_socket = sock;
1809 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1811 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1812 return &rcu_dereference_raw(sk->sk_wq)->wait;
1814 /* Detach socket from process context.
1815 * Announce socket dead, detach it from wait queue and inode.
1816 * Note that parent inode held reference count on this struct sock,
1817 * we do not release it in this function, because protocol
1818 * probably wants some additional cleanups or even continuing
1819 * to work with this socket (TCP).
1821 static inline void sock_orphan(struct sock *sk)
1823 write_lock_bh(&sk->sk_callback_lock);
1824 sock_set_flag(sk, SOCK_DEAD);
1825 sk_set_socket(sk, NULL);
1827 write_unlock_bh(&sk->sk_callback_lock);
1830 static inline void sock_graft(struct sock *sk, struct socket *parent)
1832 WARN_ON(parent->sk);
1833 write_lock_bh(&sk->sk_callback_lock);
1834 rcu_assign_pointer(sk->sk_wq, &parent->wq);
1836 sk_set_socket(sk, parent);
1837 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1838 security_sock_graft(sk, parent);
1839 write_unlock_bh(&sk->sk_callback_lock);
1842 kuid_t sock_i_uid(struct sock *sk);
1843 unsigned long sock_i_ino(struct sock *sk);
1845 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1847 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1850 static inline u32 net_tx_rndhash(void)
1852 u32 v = prandom_u32();
1857 static inline void sk_set_txhash(struct sock *sk)
1859 sk->sk_txhash = net_tx_rndhash();
1862 static inline void sk_rethink_txhash(struct sock *sk)
1868 static inline struct dst_entry *
1869 __sk_dst_get(struct sock *sk)
1871 return rcu_dereference_check(sk->sk_dst_cache,
1872 lockdep_sock_is_held(sk));
1875 static inline struct dst_entry *
1876 sk_dst_get(struct sock *sk)
1878 struct dst_entry *dst;
1881 dst = rcu_dereference(sk->sk_dst_cache);
1882 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1888 static inline void dst_negative_advice(struct sock *sk)
1890 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1892 sk_rethink_txhash(sk);
1894 if (dst && dst->ops->negative_advice) {
1895 ndst = dst->ops->negative_advice(dst);
1898 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1899 sk_tx_queue_clear(sk);
1900 sk->sk_dst_pending_confirm = 0;
1906 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1908 struct dst_entry *old_dst;
1910 sk_tx_queue_clear(sk);
1911 sk->sk_dst_pending_confirm = 0;
1912 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1913 lockdep_sock_is_held(sk));
1914 rcu_assign_pointer(sk->sk_dst_cache, dst);
1915 dst_release(old_dst);
1919 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1921 struct dst_entry *old_dst;
1923 sk_tx_queue_clear(sk);
1924 sk->sk_dst_pending_confirm = 0;
1925 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1926 dst_release(old_dst);
1930 __sk_dst_reset(struct sock *sk)
1932 __sk_dst_set(sk, NULL);
1936 sk_dst_reset(struct sock *sk)
1938 sk_dst_set(sk, NULL);
1941 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1943 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1945 static inline void sk_dst_confirm(struct sock *sk)
1947 if (!READ_ONCE(sk->sk_dst_pending_confirm))
1948 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
1951 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1953 if (skb_get_dst_pending_confirm(skb)) {
1954 struct sock *sk = skb->sk;
1955 unsigned long now = jiffies;
1957 /* avoid dirtying neighbour */
1958 if (READ_ONCE(n->confirmed) != now)
1959 WRITE_ONCE(n->confirmed, now);
1960 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
1961 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
1965 bool sk_mc_loop(struct sock *sk);
1967 static inline bool sk_can_gso(const struct sock *sk)
1969 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1972 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1974 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1976 sk->sk_route_nocaps |= flags;
1977 sk->sk_route_caps &= ~flags;
1980 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1981 struct iov_iter *from, char *to,
1982 int copy, int offset)
1984 if (skb->ip_summed == CHECKSUM_NONE) {
1986 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1988 skb->csum = csum_block_add(skb->csum, csum, offset);
1989 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1990 if (!copy_from_iter_full_nocache(to, copy, from))
1992 } else if (!copy_from_iter_full(to, copy, from))
1998 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1999 struct iov_iter *from, int copy)
2001 int err, offset = skb->len;
2003 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2006 __skb_trim(skb, offset);
2011 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2012 struct sk_buff *skb,
2018 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2024 skb->data_len += copy;
2025 skb->truesize += copy;
2026 sk_wmem_queued_add(sk, copy);
2027 sk_mem_charge(sk, copy);
2032 * sk_wmem_alloc_get - returns write allocations
2035 * Returns sk_wmem_alloc minus initial offset of one
2037 static inline int sk_wmem_alloc_get(const struct sock *sk)
2039 return refcount_read(&sk->sk_wmem_alloc) - 1;
2043 * sk_rmem_alloc_get - returns read allocations
2046 * Returns sk_rmem_alloc
2048 static inline int sk_rmem_alloc_get(const struct sock *sk)
2050 return atomic_read(&sk->sk_rmem_alloc);
2054 * sk_has_allocations - check if allocations are outstanding
2057 * Returns true if socket has write or read allocations
2059 static inline bool sk_has_allocations(const struct sock *sk)
2061 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2065 * skwq_has_sleeper - check if there are any waiting processes
2066 * @wq: struct socket_wq
2068 * Returns true if socket_wq has waiting processes
2070 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2071 * barrier call. They were added due to the race found within the tcp code.
2073 * Consider following tcp code paths::
2076 * sys_select receive packet
2078 * __add_wait_queue update tp->rcv_nxt
2080 * tp->rcv_nxt check sock_def_readable
2082 * schedule rcu_read_lock();
2083 * wq = rcu_dereference(sk->sk_wq);
2084 * if (wq && waitqueue_active(&wq->wait))
2085 * wake_up_interruptible(&wq->wait)
2089 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2090 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2091 * could then endup calling schedule and sleep forever if there are no more
2092 * data on the socket.
2095 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2097 return wq && wq_has_sleeper(&wq->wait);
2101 * sock_poll_wait - place memory barrier behind the poll_wait call.
2103 * @sock: socket to wait on
2106 * See the comments in the wq_has_sleeper function.
2108 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2111 if (!poll_does_not_wait(p)) {
2112 poll_wait(filp, &sock->wq.wait, p);
2113 /* We need to be sure we are in sync with the
2114 * socket flags modification.
2116 * This memory barrier is paired in the wq_has_sleeper.
2122 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2124 if (sk->sk_txhash) {
2126 skb->hash = sk->sk_txhash;
2130 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2133 * Queue a received datagram if it will fit. Stream and sequenced
2134 * protocols can't normally use this as they need to fit buffers in
2135 * and play with them.
2137 * Inlined as it's very short and called for pretty much every
2138 * packet ever received.
2140 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2144 skb->destructor = sock_rfree;
2145 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2146 sk_mem_charge(sk, skb->truesize);
2149 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2150 unsigned long expires);
2152 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2154 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2155 struct sk_buff *skb, unsigned int flags,
2156 void (*destructor)(struct sock *sk,
2157 struct sk_buff *skb));
2158 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2159 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2161 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2162 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2165 * Recover an error report and clear atomically
2168 static inline int sock_error(struct sock *sk)
2171 if (likely(!sk->sk_err))
2173 err = xchg(&sk->sk_err, 0);
2177 static inline unsigned long sock_wspace(struct sock *sk)
2181 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2182 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2190 * We use sk->sk_wq_raw, from contexts knowing this
2191 * pointer is not NULL and cannot disappear/change.
2193 static inline void sk_set_bit(int nr, struct sock *sk)
2195 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2196 !sock_flag(sk, SOCK_FASYNC))
2199 set_bit(nr, &sk->sk_wq_raw->flags);
2202 static inline void sk_clear_bit(int nr, struct sock *sk)
2204 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2205 !sock_flag(sk, SOCK_FASYNC))
2208 clear_bit(nr, &sk->sk_wq_raw->flags);
2211 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2213 if (sock_flag(sk, SOCK_FASYNC)) {
2215 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2220 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2221 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2222 * Note: for send buffers, TCP works better if we can build two skbs at
2225 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2227 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2228 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2230 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2234 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2237 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2239 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2242 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2243 bool force_schedule);
2246 * sk_page_frag - return an appropriate page_frag
2249 * Use the per task page_frag instead of the per socket one for
2250 * optimization when we know that we're in the normal context and owns
2251 * everything that's associated with %current.
2253 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2254 * inside other socket operations and end up recursing into sk_page_frag()
2255 * while it's already in use.
2257 static inline struct page_frag *sk_page_frag(struct sock *sk)
2259 if (gfpflags_normal_context(sk->sk_allocation))
2260 return ¤t->task_frag;
2262 return &sk->sk_frag;
2265 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2268 * Default write policy as shown to user space via poll/select/SIGIO
2270 static inline bool sock_writeable(const struct sock *sk)
2272 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2275 static inline gfp_t gfp_any(void)
2277 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2280 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2282 return noblock ? 0 : sk->sk_rcvtimeo;
2285 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2287 return noblock ? 0 : sk->sk_sndtimeo;
2290 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2292 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2297 /* Alas, with timeout socket operations are not restartable.
2298 * Compare this to poll().
2300 static inline int sock_intr_errno(long timeo)
2302 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2305 struct sock_skb_cb {
2309 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2310 * using skb->cb[] would keep using it directly and utilize its
2311 * alignement guarantee.
2313 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2314 sizeof(struct sock_skb_cb)))
2316 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2317 SOCK_SKB_CB_OFFSET))
2319 #define sock_skb_cb_check_size(size) \
2320 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2323 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2325 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2326 atomic_read(&sk->sk_drops) : 0;
2329 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2331 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2333 atomic_add(segs, &sk->sk_drops);
2336 static inline ktime_t sock_read_timestamp(struct sock *sk)
2338 #if BITS_PER_LONG==32
2343 seq = read_seqbegin(&sk->sk_stamp_seq);
2345 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2349 return READ_ONCE(sk->sk_stamp);
2353 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2355 #if BITS_PER_LONG==32
2356 write_seqlock(&sk->sk_stamp_seq);
2358 write_sequnlock(&sk->sk_stamp_seq);
2360 WRITE_ONCE(sk->sk_stamp, kt);
2364 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2365 struct sk_buff *skb);
2366 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2367 struct sk_buff *skb);
2370 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2372 ktime_t kt = skb->tstamp;
2373 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2376 * generate control messages if
2377 * - receive time stamping in software requested
2378 * - software time stamp available and wanted
2379 * - hardware time stamps available and wanted
2381 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2382 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2383 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2384 (hwtstamps->hwtstamp &&
2385 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2386 __sock_recv_timestamp(msg, sk, skb);
2388 sock_write_timestamp(sk, kt);
2390 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2391 __sock_recv_wifi_status(msg, sk, skb);
2394 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2395 struct sk_buff *skb);
2397 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2398 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2399 struct sk_buff *skb)
2401 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2402 (1UL << SOCK_RCVTSTAMP))
2403 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2404 SOF_TIMESTAMPING_RAW_HARDWARE)
2406 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2407 __sock_recv_ts_and_drops(msg, sk, skb);
2408 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2409 sock_write_timestamp(sk, skb->tstamp);
2410 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2411 sock_write_timestamp(sk, 0);
2414 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2417 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2418 * @sk: socket sending this packet
2419 * @tsflags: timestamping flags to use
2420 * @tx_flags: completed with instructions for time stamping
2421 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2423 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2425 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2426 __u8 *tx_flags, __u32 *tskey)
2428 if (unlikely(tsflags)) {
2429 __sock_tx_timestamp(tsflags, tx_flags);
2430 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2431 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2432 *tskey = sk->sk_tskey++;
2434 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2435 *tx_flags |= SKBTX_WIFI_STATUS;
2438 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2441 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2444 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2446 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2447 &skb_shinfo(skb)->tskey);
2451 * sk_eat_skb - Release a skb if it is no longer needed
2452 * @sk: socket to eat this skb from
2453 * @skb: socket buffer to eat
2455 * This routine must be called with interrupts disabled or with the socket
2456 * locked so that the sk_buff queue operation is ok.
2458 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2459 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2461 __skb_unlink(skb, &sk->sk_receive_queue);
2462 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2463 !sk->sk_rx_skb_cache) {
2464 sk->sk_rx_skb_cache = skb;
2472 struct net *sock_net(const struct sock *sk)
2474 return read_pnet(&sk->sk_net);
2478 void sock_net_set(struct sock *sk, struct net *net)
2480 write_pnet(&sk->sk_net, net);
2483 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2486 struct sock *sk = skb->sk;
2488 skb->destructor = NULL;
2495 /* This helper checks if a socket is a full socket,
2496 * ie _not_ a timewait or request socket.
2498 static inline bool sk_fullsock(const struct sock *sk)
2500 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2503 /* Checks if this SKB belongs to an HW offloaded socket
2504 * and whether any SW fallbacks are required based on dev.
2505 * Check decrypted mark in case skb_orphan() cleared socket.
2507 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2508 struct net_device *dev)
2510 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2511 struct sock *sk = skb->sk;
2513 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2514 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2515 #ifdef CONFIG_TLS_DEVICE
2516 } else if (unlikely(skb->decrypted)) {
2517 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2527 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2528 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2530 static inline bool sk_listener(const struct sock *sk)
2532 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2535 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2536 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2539 bool sk_ns_capable(const struct sock *sk,
2540 struct user_namespace *user_ns, int cap);
2541 bool sk_capable(const struct sock *sk, int cap);
2542 bool sk_net_capable(const struct sock *sk, int cap);
2544 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2546 /* Take into consideration the size of the struct sk_buff overhead in the
2547 * determination of these values, since that is non-constant across
2548 * platforms. This makes socket queueing behavior and performance
2549 * not depend upon such differences.
2551 #define _SK_MEM_PACKETS 256
2552 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2553 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2554 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2556 extern __u32 sysctl_wmem_max;
2557 extern __u32 sysctl_rmem_max;
2559 extern int sysctl_tstamp_allow_data;
2560 extern int sysctl_optmem_max;
2562 extern __u32 sysctl_wmem_default;
2563 extern __u32 sysctl_rmem_default;
2565 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2567 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2569 /* Does this proto have per netns sysctl_wmem ? */
2570 if (proto->sysctl_wmem_offset)
2571 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2573 return *proto->sysctl_wmem;
2576 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2578 /* Does this proto have per netns sysctl_rmem ? */
2579 if (proto->sysctl_rmem_offset)
2580 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2582 return *proto->sysctl_rmem;
2585 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2586 * Some wifi drivers need to tweak it to get more chunks.
2587 * They can use this helper from their ndo_start_xmit()
2589 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2591 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2593 WRITE_ONCE(sk->sk_pacing_shift, val);
2596 /* if a socket is bound to a device, check that the given device
2597 * index is either the same or that the socket is bound to an L3
2598 * master device and the given device index is also enslaved to
2601 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2605 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2608 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2609 if (mdif && mdif == sk->sk_bound_dev_if)
2615 #endif /* _SOCK_H */