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_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
121 * @skc_hash: hash value used with various protocol lookup tables
122 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
123 * @skc_dport: placeholder for inet_dport/tw_dport
124 * @skc_num: placeholder for inet_num/tw_num
125 * @skc_portpair: __u32 union of @skc_dport & @skc_num
126 * @skc_family: network address family
127 * @skc_state: Connection state
128 * @skc_reuse: %SO_REUSEADDR setting
129 * @skc_reuseport: %SO_REUSEPORT setting
130 * @skc_ipv6only: socket is IPV6 only
131 * @skc_net_refcnt: socket is using net ref counting
132 * @skc_bound_dev_if: bound device index if != 0
133 * @skc_bind_node: bind hash linkage for various protocol lookup tables
134 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
135 * @skc_prot: protocol handlers inside a network family
136 * @skc_net: reference to the network namespace of this socket
137 * @skc_v6_daddr: IPV6 destination address
138 * @skc_v6_rcv_saddr: IPV6 source address
139 * @skc_cookie: socket's cookie value
140 * @skc_node: main hash linkage for various protocol lookup tables
141 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
142 * @skc_tx_queue_mapping: tx queue number for this connection
143 * @skc_rx_queue_mapping: rx queue number for this connection
144 * @skc_flags: place holder for sk_flags
145 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
146 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
147 * @skc_listener: connection request listener socket (aka rsk_listener)
148 * [union with @skc_flags]
149 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
150 * [union with @skc_flags]
151 * @skc_incoming_cpu: record/match cpu processing incoming packets
152 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
153 * [union with @skc_incoming_cpu]
154 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
155 * [union with @skc_incoming_cpu]
156 * @skc_refcnt: reference count
158 * This is the minimal network layer representation of sockets, the header
159 * for struct sock and struct inet_timewait_sock.
162 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
163 * address on 64bit arches : cf INET_MATCH()
166 __addrpair skc_addrpair;
169 __be32 skc_rcv_saddr;
173 unsigned int skc_hash;
174 __u16 skc_u16hashes[2];
176 /* skc_dport && skc_num must be grouped as well */
178 __portpair skc_portpair;
185 unsigned short skc_family;
186 volatile unsigned char skc_state;
187 unsigned char skc_reuse:4;
188 unsigned char skc_reuseport:1;
189 unsigned char skc_ipv6only:1;
190 unsigned char skc_net_refcnt:1;
191 int skc_bound_dev_if;
193 struct hlist_node skc_bind_node;
194 struct hlist_node skc_portaddr_node;
196 struct proto *skc_prot;
197 possible_net_t skc_net;
199 #if IS_ENABLED(CONFIG_IPV6)
200 struct in6_addr skc_v6_daddr;
201 struct in6_addr skc_v6_rcv_saddr;
204 atomic64_t skc_cookie;
206 /* following fields are padding to force
207 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
208 * assuming IPV6 is enabled. We use this padding differently
209 * for different kind of 'sockets'
212 unsigned long skc_flags;
213 struct sock *skc_listener; /* request_sock */
214 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
217 * fields between dontcopy_begin/dontcopy_end
218 * are not copied in sock_copy()
221 int skc_dontcopy_begin[0];
224 struct hlist_node skc_node;
225 struct hlist_nulls_node skc_nulls_node;
227 unsigned short skc_tx_queue_mapping;
229 unsigned short skc_rx_queue_mapping;
232 int skc_incoming_cpu;
234 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
237 refcount_t skc_refcnt;
239 int skc_dontcopy_end[0];
242 u32 skc_window_clamp;
243 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
248 struct bpf_sk_storage;
251 * struct sock - network layer representation of sockets
252 * @__sk_common: shared layout with inet_timewait_sock
253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
255 * @sk_lock: synchronizer
256 * @sk_kern_sock: True if sock is using kernel lock classes
257 * @sk_rcvbuf: size of receive buffer in bytes
258 * @sk_wq: sock wait queue and async head
259 * @sk_rx_dst: receive input route used by early demux
260 * @sk_dst_cache: destination cache
261 * @sk_dst_pending_confirm: need to confirm neighbour
262 * @sk_policy: flow policy
263 * @sk_rx_skb_cache: cache copy of recently accessed RX skb
264 * @sk_receive_queue: incoming packets
265 * @sk_wmem_alloc: transmit queue bytes committed
266 * @sk_tsq_flags: TCP Small Queues flags
267 * @sk_write_queue: Packet sending queue
268 * @sk_omem_alloc: "o" is "option" or "other"
269 * @sk_wmem_queued: persistent queue size
270 * @sk_forward_alloc: space allocated forward
271 * @sk_napi_id: id of the last napi context to receive data for sk
272 * @sk_ll_usec: usecs to busypoll when there is no data
273 * @sk_allocation: allocation mode
274 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
275 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
276 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
277 * @sk_sndbuf: size of send buffer in bytes
278 * @__sk_flags_offset: empty field used to determine location of bitfield
279 * @sk_padding: unused element for alignment
280 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
281 * @sk_no_check_rx: allow zero checksum in RX packets
282 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
283 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
284 * @sk_route_forced_caps: static, forced route capabilities
285 * (set in tcp_init_sock())
286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
287 * @sk_gso_max_size: Maximum GSO segment size to build
288 * @sk_gso_max_segs: Maximum number of GSO segments
289 * @sk_pacing_shift: scaling factor for TCP Small Queues
290 * @sk_lingertime: %SO_LINGER l_linger setting
291 * @sk_backlog: always used with the per-socket spinlock held
292 * @sk_callback_lock: used with the callbacks in the end of this struct
293 * @sk_error_queue: rarely used
294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
295 * IPV6_ADDRFORM for instance)
296 * @sk_err: last error
297 * @sk_err_soft: errors that don't cause failure but are the cause of a
298 * persistent failure not just 'timed out'
299 * @sk_drops: raw/udp drops counter
300 * @sk_ack_backlog: current listen backlog
301 * @sk_max_ack_backlog: listen backlog set in listen()
302 * @sk_uid: user id of owner
303 * @sk_priority: %SO_PRIORITY setting
304 * @sk_type: socket type (%SOCK_STREAM, etc)
305 * @sk_protocol: which protocol this socket belongs in this network family
306 * @sk_peer_pid: &struct pid for this socket's peer
307 * @sk_peer_cred: %SO_PEERCRED setting
308 * @sk_rcvlowat: %SO_RCVLOWAT setting
309 * @sk_rcvtimeo: %SO_RCVTIMEO setting
310 * @sk_sndtimeo: %SO_SNDTIMEO setting
311 * @sk_txhash: computed flow hash for use on transmit
312 * @sk_filter: socket filtering instructions
313 * @sk_timer: sock cleanup timer
314 * @sk_stamp: time stamp of last packet received
315 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
316 * @sk_tsflags: SO_TIMESTAMPING socket options
317 * @sk_tskey: counter to disambiguate concurrent tstamp requests
318 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
319 * @sk_socket: Identd and reporting IO signals
320 * @sk_user_data: RPC layer private data
321 * @sk_frag: cached page frag
322 * @sk_peek_off: current peek_offset value
323 * @sk_send_head: front of stuff to transmit
324 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
325 * @sk_tx_skb_cache: cache copy of recently accessed TX skb
326 * @sk_security: used by security modules
327 * @sk_mark: generic packet mark
328 * @sk_cgrp_data: cgroup data for this cgroup
329 * @sk_memcg: this socket's memory cgroup association
330 * @sk_write_pending: a write to stream socket waits to start
331 * @sk_state_change: callback to indicate change in the state of the sock
332 * @sk_data_ready: callback to indicate there is data to be processed
333 * @sk_write_space: callback to indicate there is bf sending space available
334 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
335 * @sk_backlog_rcv: callback to process the backlog
336 * @sk_validate_xmit_skb: ptr to an optional validate function
337 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
338 * @sk_reuseport_cb: reuseport group container
339 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
340 * @sk_rcu: used during RCU grace period
341 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
342 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
343 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
344 * @sk_txtime_unused: unused txtime flags
348 * Now struct inet_timewait_sock also uses sock_common, so please just
349 * don't add nothing before this first member (__sk_common) --acme
351 struct sock_common __sk_common;
352 #define sk_node __sk_common.skc_node
353 #define sk_nulls_node __sk_common.skc_nulls_node
354 #define sk_refcnt __sk_common.skc_refcnt
355 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
357 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
360 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
361 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
362 #define sk_hash __sk_common.skc_hash
363 #define sk_portpair __sk_common.skc_portpair
364 #define sk_num __sk_common.skc_num
365 #define sk_dport __sk_common.skc_dport
366 #define sk_addrpair __sk_common.skc_addrpair
367 #define sk_daddr __sk_common.skc_daddr
368 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
369 #define sk_family __sk_common.skc_family
370 #define sk_state __sk_common.skc_state
371 #define sk_reuse __sk_common.skc_reuse
372 #define sk_reuseport __sk_common.skc_reuseport
373 #define sk_ipv6only __sk_common.skc_ipv6only
374 #define sk_net_refcnt __sk_common.skc_net_refcnt
375 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
376 #define sk_bind_node __sk_common.skc_bind_node
377 #define sk_prot __sk_common.skc_prot
378 #define sk_net __sk_common.skc_net
379 #define sk_v6_daddr __sk_common.skc_v6_daddr
380 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
381 #define sk_cookie __sk_common.skc_cookie
382 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
383 #define sk_flags __sk_common.skc_flags
384 #define sk_rxhash __sk_common.skc_rxhash
386 socket_lock_t sk_lock;
389 struct sk_buff_head sk_error_queue;
390 struct sk_buff *sk_rx_skb_cache;
391 struct sk_buff_head sk_receive_queue;
393 * The backlog queue is special, it is always used with
394 * the per-socket spinlock held and requires low latency
395 * access. Therefore we special case it's implementation.
396 * Note : rmem_alloc is in this structure to fill a hole
397 * on 64bit arches, not because its logically part of
403 struct sk_buff *head;
404 struct sk_buff *tail;
406 #define sk_rmem_alloc sk_backlog.rmem_alloc
408 int sk_forward_alloc;
409 #ifdef CONFIG_NET_RX_BUSY_POLL
410 unsigned int sk_ll_usec;
411 /* ===== mostly read cache line ===== */
412 unsigned int sk_napi_id;
416 struct sk_filter __rcu *sk_filter;
418 struct socket_wq __rcu *sk_wq;
420 struct socket_wq *sk_wq_raw;
424 struct xfrm_policy __rcu *sk_policy[2];
426 struct dst_entry *sk_rx_dst;
427 struct dst_entry __rcu *sk_dst_cache;
428 atomic_t sk_omem_alloc;
431 /* ===== cache line for TX ===== */
433 refcount_t sk_wmem_alloc;
434 unsigned long sk_tsq_flags;
436 struct sk_buff *sk_send_head;
437 struct rb_root tcp_rtx_queue;
439 struct sk_buff *sk_tx_skb_cache;
440 struct sk_buff_head sk_write_queue;
442 int sk_write_pending;
443 __u32 sk_dst_pending_confirm;
444 u32 sk_pacing_status; /* see enum sk_pacing */
446 struct timer_list sk_timer;
449 unsigned long sk_pacing_rate; /* bytes per second */
450 unsigned long sk_max_pacing_rate;
451 struct page_frag sk_frag;
452 netdev_features_t sk_route_caps;
453 netdev_features_t sk_route_nocaps;
454 netdev_features_t sk_route_forced_caps;
456 unsigned int sk_gso_max_size;
461 * Because of non atomicity rules, all
462 * changes are protected by socket lock.
473 unsigned long sk_lingertime;
474 struct proto *sk_prot_creator;
475 rwlock_t sk_callback_lock;
479 u32 sk_max_ack_backlog;
481 struct pid *sk_peer_pid;
482 const struct cred *sk_peer_cred;
485 #if BITS_PER_LONG==32
486 seqlock_t sk_stamp_seq;
494 u8 sk_txtime_deadline_mode : 1,
495 sk_txtime_report_errors : 1,
496 sk_txtime_unused : 6;
498 struct socket *sk_socket;
500 #ifdef CONFIG_SECURITY
503 struct sock_cgroup_data sk_cgrp_data;
504 struct mem_cgroup *sk_memcg;
505 void (*sk_state_change)(struct sock *sk);
506 void (*sk_data_ready)(struct sock *sk);
507 void (*sk_write_space)(struct sock *sk);
508 void (*sk_error_report)(struct sock *sk);
509 int (*sk_backlog_rcv)(struct sock *sk,
510 struct sk_buff *skb);
511 #ifdef CONFIG_SOCK_VALIDATE_XMIT
512 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
513 struct net_device *dev,
514 struct sk_buff *skb);
516 void (*sk_destruct)(struct sock *sk);
517 struct sock_reuseport __rcu *sk_reuseport_cb;
518 #ifdef CONFIG_BPF_SYSCALL
519 struct bpf_sk_storage __rcu *sk_bpf_storage;
521 struct rcu_head sk_rcu;
526 SK_PACING_NEEDED = 1,
530 /* Pointer stored in sk_user_data might not be suitable for copying
531 * when cloning the socket. For instance, it can point to a reference
532 * counted object. sk_user_data bottom bit is set if pointer must not
535 #define SK_USER_DATA_NOCOPY 1UL
536 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY)
539 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
542 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
544 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
547 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
549 #define rcu_dereference_sk_user_data(sk) \
551 void *__tmp = rcu_dereference(__sk_user_data((sk))); \
552 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \
554 #define rcu_assign_sk_user_data(sk, ptr) \
556 uintptr_t __tmp = (uintptr_t)(ptr); \
557 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
558 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \
560 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \
562 uintptr_t __tmp = (uintptr_t)(ptr); \
563 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
564 rcu_assign_pointer(__sk_user_data((sk)), \
565 __tmp | SK_USER_DATA_NOCOPY); \
569 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
570 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
571 * on a socket means that the socket will reuse everybody else's port
572 * without looking at the other's sk_reuse value.
575 #define SK_NO_REUSE 0
576 #define SK_CAN_REUSE 1
577 #define SK_FORCE_REUSE 2
579 int sk_set_peek_off(struct sock *sk, int val);
581 static inline int sk_peek_offset(struct sock *sk, int flags)
583 if (unlikely(flags & MSG_PEEK)) {
584 return READ_ONCE(sk->sk_peek_off);
590 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
592 s32 off = READ_ONCE(sk->sk_peek_off);
594 if (unlikely(off >= 0)) {
595 off = max_t(s32, off - val, 0);
596 WRITE_ONCE(sk->sk_peek_off, off);
600 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
602 sk_peek_offset_bwd(sk, -val);
606 * Hashed lists helper routines
608 static inline struct sock *sk_entry(const struct hlist_node *node)
610 return hlist_entry(node, struct sock, sk_node);
613 static inline struct sock *__sk_head(const struct hlist_head *head)
615 return hlist_entry(head->first, struct sock, sk_node);
618 static inline struct sock *sk_head(const struct hlist_head *head)
620 return hlist_empty(head) ? NULL : __sk_head(head);
623 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
625 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
628 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
630 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
633 static inline struct sock *sk_next(const struct sock *sk)
635 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
638 static inline struct sock *sk_nulls_next(const struct sock *sk)
640 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
641 hlist_nulls_entry(sk->sk_nulls_node.next,
642 struct sock, sk_nulls_node) :
646 static inline bool sk_unhashed(const struct sock *sk)
648 return hlist_unhashed(&sk->sk_node);
651 static inline bool sk_hashed(const struct sock *sk)
653 return !sk_unhashed(sk);
656 static inline void sk_node_init(struct hlist_node *node)
661 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
666 static inline void __sk_del_node(struct sock *sk)
668 __hlist_del(&sk->sk_node);
671 /* NB: equivalent to hlist_del_init_rcu */
672 static inline bool __sk_del_node_init(struct sock *sk)
676 sk_node_init(&sk->sk_node);
682 /* Grab socket reference count. This operation is valid only
683 when sk is ALREADY grabbed f.e. it is found in hash table
684 or a list and the lookup is made under lock preventing hash table
688 static __always_inline void sock_hold(struct sock *sk)
690 refcount_inc(&sk->sk_refcnt);
693 /* Ungrab socket in the context, which assumes that socket refcnt
694 cannot hit zero, f.e. it is true in context of any socketcall.
696 static __always_inline void __sock_put(struct sock *sk)
698 refcount_dec(&sk->sk_refcnt);
701 static inline bool sk_del_node_init(struct sock *sk)
703 bool rc = __sk_del_node_init(sk);
706 /* paranoid for a while -acme */
707 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
712 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
714 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
717 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
723 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
725 bool rc = __sk_nulls_del_node_init_rcu(sk);
728 /* paranoid for a while -acme */
729 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
735 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
737 hlist_add_head(&sk->sk_node, list);
740 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
743 __sk_add_node(sk, list);
746 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
749 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
750 sk->sk_family == AF_INET6)
751 hlist_add_tail_rcu(&sk->sk_node, list);
753 hlist_add_head_rcu(&sk->sk_node, list);
756 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
759 hlist_add_tail_rcu(&sk->sk_node, list);
762 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
764 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
767 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
769 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
772 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
775 __sk_nulls_add_node_rcu(sk, list);
778 static inline void __sk_del_bind_node(struct sock *sk)
780 __hlist_del(&sk->sk_bind_node);
783 static inline void sk_add_bind_node(struct sock *sk,
784 struct hlist_head *list)
786 hlist_add_head(&sk->sk_bind_node, list);
789 #define sk_for_each(__sk, list) \
790 hlist_for_each_entry(__sk, list, sk_node)
791 #define sk_for_each_rcu(__sk, list) \
792 hlist_for_each_entry_rcu(__sk, list, sk_node)
793 #define sk_nulls_for_each(__sk, node, list) \
794 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
795 #define sk_nulls_for_each_rcu(__sk, node, list) \
796 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
797 #define sk_for_each_from(__sk) \
798 hlist_for_each_entry_from(__sk, sk_node)
799 #define sk_nulls_for_each_from(__sk, node) \
800 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
801 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
802 #define sk_for_each_safe(__sk, tmp, list) \
803 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
804 #define sk_for_each_bound(__sk, list) \
805 hlist_for_each_entry(__sk, list, sk_bind_node)
808 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
809 * @tpos: the type * to use as a loop cursor.
810 * @pos: the &struct hlist_node to use as a loop cursor.
811 * @head: the head for your list.
812 * @offset: offset of hlist_node within the struct.
815 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
816 for (pos = rcu_dereference(hlist_first_rcu(head)); \
818 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
819 pos = rcu_dereference(hlist_next_rcu(pos)))
821 static inline struct user_namespace *sk_user_ns(struct sock *sk)
823 /* Careful only use this in a context where these parameters
824 * can not change and must all be valid, such as recvmsg from
827 return sk->sk_socket->file->f_cred->user_ns;
841 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
842 SOCK_DBG, /* %SO_DEBUG setting */
843 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
844 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
845 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
846 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
847 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
848 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
849 SOCK_FASYNC, /* fasync() active */
851 SOCK_ZEROCOPY, /* buffers from userspace */
852 SOCK_WIFI_STATUS, /* push wifi status to userspace */
853 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
854 * Will use last 4 bytes of packet sent from
855 * user-space instead.
857 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
858 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
859 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
861 SOCK_XDP, /* XDP is attached */
862 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
865 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
867 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
869 nsk->sk_flags = osk->sk_flags;
872 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
874 __set_bit(flag, &sk->sk_flags);
877 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
879 __clear_bit(flag, &sk->sk_flags);
882 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
884 return test_bit(flag, &sk->sk_flags);
888 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
889 static inline int sk_memalloc_socks(void)
891 return static_branch_unlikely(&memalloc_socks_key);
895 static inline int sk_memalloc_socks(void)
902 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
904 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
907 static inline void sk_acceptq_removed(struct sock *sk)
909 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
912 static inline void sk_acceptq_added(struct sock *sk)
914 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
917 static inline bool sk_acceptq_is_full(const struct sock *sk)
919 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
923 * Compute minimal free write space needed to queue new packets.
925 static inline int sk_stream_min_wspace(const struct sock *sk)
927 return READ_ONCE(sk->sk_wmem_queued) >> 1;
930 static inline int sk_stream_wspace(const struct sock *sk)
932 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
935 static inline void sk_wmem_queued_add(struct sock *sk, int val)
937 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
940 void sk_stream_write_space(struct sock *sk);
942 /* OOB backlog add */
943 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
945 /* dont let skb dst not refcounted, we are going to leave rcu lock */
948 if (!sk->sk_backlog.tail)
949 WRITE_ONCE(sk->sk_backlog.head, skb);
951 sk->sk_backlog.tail->next = skb;
953 WRITE_ONCE(sk->sk_backlog.tail, skb);
958 * Take into account size of receive queue and backlog queue
959 * Do not take into account this skb truesize,
960 * to allow even a single big packet to come.
962 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
964 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
966 return qsize > limit;
969 /* The per-socket spinlock must be held here. */
970 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
973 if (sk_rcvqueues_full(sk, limit))
977 * If the skb was allocated from pfmemalloc reserves, only
978 * allow SOCK_MEMALLOC sockets to use it as this socket is
979 * helping free memory
981 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
984 __sk_add_backlog(sk, skb);
985 sk->sk_backlog.len += skb->truesize;
989 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
991 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
993 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
994 return __sk_backlog_rcv(sk, skb);
996 return sk->sk_backlog_rcv(sk, skb);
999 static inline void sk_incoming_cpu_update(struct sock *sk)
1001 int cpu = raw_smp_processor_id();
1003 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1004 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1007 static inline void sock_rps_record_flow_hash(__u32 hash)
1010 struct rps_sock_flow_table *sock_flow_table;
1013 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1014 rps_record_sock_flow(sock_flow_table, hash);
1019 static inline void sock_rps_record_flow(const struct sock *sk)
1022 if (static_branch_unlikely(&rfs_needed)) {
1023 /* Reading sk->sk_rxhash might incur an expensive cache line
1026 * TCP_ESTABLISHED does cover almost all states where RFS
1027 * might be useful, and is cheaper [1] than testing :
1028 * IPv4: inet_sk(sk)->inet_daddr
1029 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1030 * OR an additional socket flag
1031 * [1] : sk_state and sk_prot are in the same cache line.
1033 if (sk->sk_state == TCP_ESTABLISHED)
1034 sock_rps_record_flow_hash(sk->sk_rxhash);
1039 static inline void sock_rps_save_rxhash(struct sock *sk,
1040 const struct sk_buff *skb)
1043 if (unlikely(sk->sk_rxhash != skb->hash))
1044 sk->sk_rxhash = skb->hash;
1048 static inline void sock_rps_reset_rxhash(struct sock *sk)
1055 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1057 release_sock(__sk); \
1058 __rc = __condition; \
1060 *(__timeo) = wait_woken(__wait, \
1061 TASK_INTERRUPTIBLE, \
1064 sched_annotate_sleep(); \
1066 __rc = __condition; \
1070 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1071 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1072 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1073 int sk_stream_error(struct sock *sk, int flags, int err);
1074 void sk_stream_kill_queues(struct sock *sk);
1075 void sk_set_memalloc(struct sock *sk);
1076 void sk_clear_memalloc(struct sock *sk);
1078 void __sk_flush_backlog(struct sock *sk);
1080 static inline bool sk_flush_backlog(struct sock *sk)
1082 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1083 __sk_flush_backlog(sk);
1089 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1091 struct request_sock_ops;
1092 struct timewait_sock_ops;
1093 struct inet_hashinfo;
1094 struct raw_hashinfo;
1095 struct smc_hashinfo;
1099 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1100 * un-modified. Special care is taken when initializing object to zero.
1102 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1104 if (offsetof(struct sock, sk_node.next) != 0)
1105 memset(sk, 0, offsetof(struct sock, sk_node.next));
1106 memset(&sk->sk_node.pprev, 0,
1107 size - offsetof(struct sock, sk_node.pprev));
1110 /* Networking protocol blocks we attach to sockets.
1111 * socket layer -> transport layer interface
1114 void (*close)(struct sock *sk,
1116 int (*pre_connect)(struct sock *sk,
1117 struct sockaddr *uaddr,
1119 int (*connect)(struct sock *sk,
1120 struct sockaddr *uaddr,
1122 int (*disconnect)(struct sock *sk, int flags);
1124 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1127 int (*ioctl)(struct sock *sk, int cmd,
1129 int (*init)(struct sock *sk);
1130 void (*destroy)(struct sock *sk);
1131 void (*shutdown)(struct sock *sk, int how);
1132 int (*setsockopt)(struct sock *sk, int level,
1133 int optname, char __user *optval,
1134 unsigned int optlen);
1135 int (*getsockopt)(struct sock *sk, int level,
1136 int optname, char __user *optval,
1137 int __user *option);
1138 void (*keepalive)(struct sock *sk, int valbool);
1139 #ifdef CONFIG_COMPAT
1140 int (*compat_setsockopt)(struct sock *sk,
1142 int optname, char __user *optval,
1143 unsigned int optlen);
1144 int (*compat_getsockopt)(struct sock *sk,
1146 int optname, char __user *optval,
1147 int __user *option);
1148 int (*compat_ioctl)(struct sock *sk,
1149 unsigned int cmd, unsigned long arg);
1151 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1153 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1154 size_t len, int noblock, int flags,
1156 int (*sendpage)(struct sock *sk, struct page *page,
1157 int offset, size_t size, int flags);
1158 int (*bind)(struct sock *sk,
1159 struct sockaddr *uaddr, int addr_len);
1161 int (*backlog_rcv) (struct sock *sk,
1162 struct sk_buff *skb);
1164 void (*release_cb)(struct sock *sk);
1166 /* Keeping track of sk's, looking them up, and port selection methods. */
1167 int (*hash)(struct sock *sk);
1168 void (*unhash)(struct sock *sk);
1169 void (*rehash)(struct sock *sk);
1170 int (*get_port)(struct sock *sk, unsigned short snum);
1172 /* Keeping track of sockets in use */
1173 #ifdef CONFIG_PROC_FS
1174 unsigned int inuse_idx;
1177 bool (*stream_memory_free)(const struct sock *sk, int wake);
1178 bool (*stream_memory_read)(const struct sock *sk);
1179 /* Memory pressure */
1180 void (*enter_memory_pressure)(struct sock *sk);
1181 void (*leave_memory_pressure)(struct sock *sk);
1182 atomic_long_t *memory_allocated; /* Current allocated memory. */
1183 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1185 * Pressure flag: try to collapse.
1186 * Technical note: it is used by multiple contexts non atomically.
1187 * All the __sk_mem_schedule() is of this nature: accounting
1188 * is strict, actions are advisory and have some latency.
1190 unsigned long *memory_pressure;
1195 u32 sysctl_wmem_offset;
1196 u32 sysctl_rmem_offset;
1201 struct kmem_cache *slab;
1202 unsigned int obj_size;
1203 slab_flags_t slab_flags;
1204 unsigned int useroffset; /* Usercopy region offset */
1205 unsigned int usersize; /* Usercopy region size */
1207 struct percpu_counter *orphan_count;
1209 struct request_sock_ops *rsk_prot;
1210 struct timewait_sock_ops *twsk_prot;
1213 struct inet_hashinfo *hashinfo;
1214 struct udp_table *udp_table;
1215 struct raw_hashinfo *raw_hash;
1216 struct smc_hashinfo *smc_hash;
1219 struct module *owner;
1223 struct list_head node;
1224 #ifdef SOCK_REFCNT_DEBUG
1227 int (*diag_destroy)(struct sock *sk, int err);
1228 } __randomize_layout;
1230 int proto_register(struct proto *prot, int alloc_slab);
1231 void proto_unregister(struct proto *prot);
1232 int sock_load_diag_module(int family, int protocol);
1234 #ifdef SOCK_REFCNT_DEBUG
1235 static inline void sk_refcnt_debug_inc(struct sock *sk)
1237 atomic_inc(&sk->sk_prot->socks);
1240 static inline void sk_refcnt_debug_dec(struct sock *sk)
1242 atomic_dec(&sk->sk_prot->socks);
1243 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1244 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1247 static inline void sk_refcnt_debug_release(const struct sock *sk)
1249 if (refcount_read(&sk->sk_refcnt) != 1)
1250 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1251 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1253 #else /* SOCK_REFCNT_DEBUG */
1254 #define sk_refcnt_debug_inc(sk) do { } while (0)
1255 #define sk_refcnt_debug_dec(sk) do { } while (0)
1256 #define sk_refcnt_debug_release(sk) do { } while (0)
1257 #endif /* SOCK_REFCNT_DEBUG */
1259 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1261 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1264 return sk->sk_prot->stream_memory_free ?
1265 sk->sk_prot->stream_memory_free(sk, wake) : true;
1268 static inline bool sk_stream_memory_free(const struct sock *sk)
1270 return __sk_stream_memory_free(sk, 0);
1273 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1275 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1276 __sk_stream_memory_free(sk, wake);
1279 static inline bool sk_stream_is_writeable(const struct sock *sk)
1281 return __sk_stream_is_writeable(sk, 0);
1284 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1285 struct cgroup *ancestor)
1287 #ifdef CONFIG_SOCK_CGROUP_DATA
1288 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1295 static inline bool sk_has_memory_pressure(const struct sock *sk)
1297 return sk->sk_prot->memory_pressure != NULL;
1300 static inline bool sk_under_memory_pressure(const struct sock *sk)
1302 if (!sk->sk_prot->memory_pressure)
1305 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1306 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1309 return !!*sk->sk_prot->memory_pressure;
1313 sk_memory_allocated(const struct sock *sk)
1315 return atomic_long_read(sk->sk_prot->memory_allocated);
1319 sk_memory_allocated_add(struct sock *sk, int amt)
1321 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1325 sk_memory_allocated_sub(struct sock *sk, int amt)
1327 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1330 static inline void sk_sockets_allocated_dec(struct sock *sk)
1332 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1335 static inline void sk_sockets_allocated_inc(struct sock *sk)
1337 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1341 sk_sockets_allocated_read_positive(struct sock *sk)
1343 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1347 proto_sockets_allocated_sum_positive(struct proto *prot)
1349 return percpu_counter_sum_positive(prot->sockets_allocated);
1353 proto_memory_allocated(struct proto *prot)
1355 return atomic_long_read(prot->memory_allocated);
1359 proto_memory_pressure(struct proto *prot)
1361 if (!prot->memory_pressure)
1363 return !!*prot->memory_pressure;
1367 #ifdef CONFIG_PROC_FS
1368 /* Called with local bh disabled */
1369 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1370 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1371 int sock_inuse_get(struct net *net);
1373 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1380 /* With per-bucket locks this operation is not-atomic, so that
1381 * this version is not worse.
1383 static inline int __sk_prot_rehash(struct sock *sk)
1385 sk->sk_prot->unhash(sk);
1386 return sk->sk_prot->hash(sk);
1389 /* About 10 seconds */
1390 #define SOCK_DESTROY_TIME (10*HZ)
1392 /* Sockets 0-1023 can't be bound to unless you are superuser */
1393 #define PROT_SOCK 1024
1395 #define SHUTDOWN_MASK 3
1396 #define RCV_SHUTDOWN 1
1397 #define SEND_SHUTDOWN 2
1399 #define SOCK_SNDBUF_LOCK 1
1400 #define SOCK_RCVBUF_LOCK 2
1401 #define SOCK_BINDADDR_LOCK 4
1402 #define SOCK_BINDPORT_LOCK 8
1404 struct socket_alloc {
1405 struct socket socket;
1406 struct inode vfs_inode;
1409 static inline struct socket *SOCKET_I(struct inode *inode)
1411 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1414 static inline struct inode *SOCK_INODE(struct socket *socket)
1416 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1420 * Functions for memory accounting
1422 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1423 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1424 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1425 void __sk_mem_reclaim(struct sock *sk, int amount);
1427 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1428 * do not necessarily have 16x time more memory than 4KB ones.
1430 #define SK_MEM_QUANTUM 4096
1431 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1432 #define SK_MEM_SEND 0
1433 #define SK_MEM_RECV 1
1435 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1436 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1438 long val = sk->sk_prot->sysctl_mem[index];
1440 #if PAGE_SIZE > SK_MEM_QUANTUM
1441 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1442 #elif PAGE_SIZE < SK_MEM_QUANTUM
1443 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1448 static inline int sk_mem_pages(int amt)
1450 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1453 static inline bool sk_has_account(struct sock *sk)
1455 /* return true if protocol supports memory accounting */
1456 return !!sk->sk_prot->memory_allocated;
1459 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1461 if (!sk_has_account(sk))
1463 return size <= sk->sk_forward_alloc ||
1464 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1468 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1470 if (!sk_has_account(sk))
1472 return size<= sk->sk_forward_alloc ||
1473 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1474 skb_pfmemalloc(skb);
1477 static inline void sk_mem_reclaim(struct sock *sk)
1479 if (!sk_has_account(sk))
1481 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1482 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1485 static inline void sk_mem_reclaim_partial(struct sock *sk)
1487 if (!sk_has_account(sk))
1489 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1490 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1493 static inline void sk_mem_charge(struct sock *sk, int size)
1495 if (!sk_has_account(sk))
1497 sk->sk_forward_alloc -= size;
1500 static inline void sk_mem_uncharge(struct sock *sk, int size)
1502 if (!sk_has_account(sk))
1504 sk->sk_forward_alloc += size;
1506 /* Avoid a possible overflow.
1507 * TCP send queues can make this happen, if sk_mem_reclaim()
1508 * is not called and more than 2 GBytes are released at once.
1510 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1511 * no need to hold that much forward allocation anyway.
1513 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1514 __sk_mem_reclaim(sk, 1 << 20);
1517 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
1518 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1520 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1521 sk_wmem_queued_add(sk, -skb->truesize);
1522 sk_mem_uncharge(sk, skb->truesize);
1523 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1524 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1526 skb_zcopy_clear(skb, true);
1527 sk->sk_tx_skb_cache = skb;
1533 static inline void sock_release_ownership(struct sock *sk)
1535 if (sk->sk_lock.owned) {
1536 sk->sk_lock.owned = 0;
1538 /* The sk_lock has mutex_unlock() semantics: */
1539 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1544 * Macro so as to not evaluate some arguments when
1545 * lockdep is not enabled.
1547 * Mark both the sk_lock and the sk_lock.slock as a
1548 * per-address-family lock class.
1550 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1552 sk->sk_lock.owned = 0; \
1553 init_waitqueue_head(&sk->sk_lock.wq); \
1554 spin_lock_init(&(sk)->sk_lock.slock); \
1555 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1556 sizeof((sk)->sk_lock)); \
1557 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1559 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1562 #ifdef CONFIG_LOCKDEP
1563 static inline bool lockdep_sock_is_held(const struct sock *sk)
1565 return lockdep_is_held(&sk->sk_lock) ||
1566 lockdep_is_held(&sk->sk_lock.slock);
1570 void lock_sock_nested(struct sock *sk, int subclass);
1572 static inline void lock_sock(struct sock *sk)
1574 lock_sock_nested(sk, 0);
1577 void __release_sock(struct sock *sk);
1578 void release_sock(struct sock *sk);
1580 /* BH context may only use the following locking interface. */
1581 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1582 #define bh_lock_sock_nested(__sk) \
1583 spin_lock_nested(&((__sk)->sk_lock.slock), \
1584 SINGLE_DEPTH_NESTING)
1585 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1587 bool lock_sock_fast(struct sock *sk);
1589 * unlock_sock_fast - complement of lock_sock_fast
1593 * fast unlock socket for user context.
1594 * If slow mode is on, we call regular release_sock()
1596 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1601 spin_unlock_bh(&sk->sk_lock.slock);
1604 /* Used by processes to "lock" a socket state, so that
1605 * interrupts and bottom half handlers won't change it
1606 * from under us. It essentially blocks any incoming
1607 * packets, so that we won't get any new data or any
1608 * packets that change the state of the socket.
1610 * While locked, BH processing will add new packets to
1611 * the backlog queue. This queue is processed by the
1612 * owner of the socket lock right before it is released.
1614 * Since ~2.3.5 it is also exclusive sleep lock serializing
1615 * accesses from user process context.
1618 static inline void sock_owned_by_me(const struct sock *sk)
1620 #ifdef CONFIG_LOCKDEP
1621 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1625 static inline bool sock_owned_by_user(const struct sock *sk)
1627 sock_owned_by_me(sk);
1628 return sk->sk_lock.owned;
1631 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1633 return sk->sk_lock.owned;
1636 /* no reclassification while locks are held */
1637 static inline bool sock_allow_reclassification(const struct sock *csk)
1639 struct sock *sk = (struct sock *)csk;
1641 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1644 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1645 struct proto *prot, int kern);
1646 void sk_free(struct sock *sk);
1647 void sk_destruct(struct sock *sk);
1648 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1649 void sk_free_unlock_clone(struct sock *sk);
1651 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1653 void __sock_wfree(struct sk_buff *skb);
1654 void sock_wfree(struct sk_buff *skb);
1655 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1657 void skb_orphan_partial(struct sk_buff *skb);
1658 void sock_rfree(struct sk_buff *skb);
1659 void sock_efree(struct sk_buff *skb);
1661 void sock_edemux(struct sk_buff *skb);
1662 void sock_pfree(struct sk_buff *skb);
1664 #define sock_edemux sock_efree
1667 int sock_setsockopt(struct socket *sock, int level, int op,
1668 char __user *optval, unsigned int optlen);
1670 int sock_getsockopt(struct socket *sock, int level, int op,
1671 char __user *optval, int __user *optlen);
1672 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1673 bool timeval, bool time32);
1674 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1675 int noblock, int *errcode);
1676 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1677 unsigned long data_len, int noblock,
1678 int *errcode, int max_page_order);
1679 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1680 void sock_kfree_s(struct sock *sk, void *mem, int size);
1681 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1682 void sk_send_sigurg(struct sock *sk);
1684 struct sockcm_cookie {
1690 static inline void sockcm_init(struct sockcm_cookie *sockc,
1691 const struct sock *sk)
1693 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1696 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1697 struct sockcm_cookie *sockc);
1698 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1699 struct sockcm_cookie *sockc);
1702 * Functions to fill in entries in struct proto_ops when a protocol
1703 * does not implement a particular function.
1705 int sock_no_bind(struct socket *, struct sockaddr *, int);
1706 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1707 int sock_no_socketpair(struct socket *, struct socket *);
1708 int sock_no_accept(struct socket *, struct socket *, int, bool);
1709 int sock_no_getname(struct socket *, struct sockaddr *, int);
1710 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1711 int sock_no_listen(struct socket *, int);
1712 int sock_no_shutdown(struct socket *, int);
1713 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1714 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1715 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1716 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1717 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1718 int sock_no_mmap(struct file *file, struct socket *sock,
1719 struct vm_area_struct *vma);
1720 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1721 size_t size, int flags);
1722 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1723 int offset, size_t size, int flags);
1726 * Functions to fill in entries in struct proto_ops when a protocol
1727 * uses the inet style.
1729 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1730 char __user *optval, int __user *optlen);
1731 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1733 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1734 char __user *optval, unsigned int optlen);
1735 int compat_sock_common_getsockopt(struct socket *sock, int level,
1736 int optname, char __user *optval, int __user *optlen);
1737 int compat_sock_common_setsockopt(struct socket *sock, int level,
1738 int optname, char __user *optval, unsigned int optlen);
1740 void sk_common_release(struct sock *sk);
1743 * Default socket callbacks and setup code
1746 /* Initialise core socket variables */
1747 void sock_init_data(struct socket *sock, struct sock *sk);
1750 * Socket reference counting postulates.
1752 * * Each user of socket SHOULD hold a reference count.
1753 * * Each access point to socket (an hash table bucket, reference from a list,
1754 * running timer, skb in flight MUST hold a reference count.
1755 * * When reference count hits 0, it means it will never increase back.
1756 * * When reference count hits 0, it means that no references from
1757 * outside exist to this socket and current process on current CPU
1758 * is last user and may/should destroy this socket.
1759 * * sk_free is called from any context: process, BH, IRQ. When
1760 * it is called, socket has no references from outside -> sk_free
1761 * may release descendant resources allocated by the socket, but
1762 * to the time when it is called, socket is NOT referenced by any
1763 * hash tables, lists etc.
1764 * * Packets, delivered from outside (from network or from another process)
1765 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1766 * when they sit in queue. Otherwise, packets will leak to hole, when
1767 * socket is looked up by one cpu and unhasing is made by another CPU.
1768 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1769 * (leak to backlog). Packet socket does all the processing inside
1770 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1771 * use separate SMP lock, so that they are prone too.
1774 /* Ungrab socket and destroy it, if it was the last reference. */
1775 static inline void sock_put(struct sock *sk)
1777 if (refcount_dec_and_test(&sk->sk_refcnt))
1780 /* Generic version of sock_put(), dealing with all sockets
1781 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1783 void sock_gen_put(struct sock *sk);
1785 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1786 unsigned int trim_cap, bool refcounted);
1787 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1790 return __sk_receive_skb(sk, skb, nested, 1, true);
1793 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1795 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1796 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1798 sk->sk_tx_queue_mapping = tx_queue;
1801 #define NO_QUEUE_MAPPING USHRT_MAX
1803 static inline void sk_tx_queue_clear(struct sock *sk)
1805 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1808 static inline int sk_tx_queue_get(const struct sock *sk)
1810 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1811 return sk->sk_tx_queue_mapping;
1816 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1819 if (skb_rx_queue_recorded(skb)) {
1820 u16 rx_queue = skb_get_rx_queue(skb);
1822 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1825 sk->sk_rx_queue_mapping = rx_queue;
1830 static inline void sk_rx_queue_clear(struct sock *sk)
1833 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1838 static inline int sk_rx_queue_get(const struct sock *sk)
1840 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1841 return sk->sk_rx_queue_mapping;
1847 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1849 sk_tx_queue_clear(sk);
1850 sk->sk_socket = sock;
1853 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1855 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1856 return &rcu_dereference_raw(sk->sk_wq)->wait;
1858 /* Detach socket from process context.
1859 * Announce socket dead, detach it from wait queue and inode.
1860 * Note that parent inode held reference count on this struct sock,
1861 * we do not release it in this function, because protocol
1862 * probably wants some additional cleanups or even continuing
1863 * to work with this socket (TCP).
1865 static inline void sock_orphan(struct sock *sk)
1867 write_lock_bh(&sk->sk_callback_lock);
1868 sock_set_flag(sk, SOCK_DEAD);
1869 sk_set_socket(sk, NULL);
1871 write_unlock_bh(&sk->sk_callback_lock);
1874 static inline void sock_graft(struct sock *sk, struct socket *parent)
1876 WARN_ON(parent->sk);
1877 write_lock_bh(&sk->sk_callback_lock);
1878 rcu_assign_pointer(sk->sk_wq, &parent->wq);
1880 sk_set_socket(sk, parent);
1881 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1882 security_sock_graft(sk, parent);
1883 write_unlock_bh(&sk->sk_callback_lock);
1886 kuid_t sock_i_uid(struct sock *sk);
1887 unsigned long sock_i_ino(struct sock *sk);
1889 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1891 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1894 static inline u32 net_tx_rndhash(void)
1896 u32 v = prandom_u32();
1901 static inline void sk_set_txhash(struct sock *sk)
1903 sk->sk_txhash = net_tx_rndhash();
1906 static inline void sk_rethink_txhash(struct sock *sk)
1912 static inline struct dst_entry *
1913 __sk_dst_get(struct sock *sk)
1915 return rcu_dereference_check(sk->sk_dst_cache,
1916 lockdep_sock_is_held(sk));
1919 static inline struct dst_entry *
1920 sk_dst_get(struct sock *sk)
1922 struct dst_entry *dst;
1925 dst = rcu_dereference(sk->sk_dst_cache);
1926 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1932 static inline void dst_negative_advice(struct sock *sk)
1934 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1936 sk_rethink_txhash(sk);
1938 if (dst && dst->ops->negative_advice) {
1939 ndst = dst->ops->negative_advice(dst);
1942 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1943 sk_tx_queue_clear(sk);
1944 sk->sk_dst_pending_confirm = 0;
1950 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1952 struct dst_entry *old_dst;
1954 sk_tx_queue_clear(sk);
1955 sk->sk_dst_pending_confirm = 0;
1956 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1957 lockdep_sock_is_held(sk));
1958 rcu_assign_pointer(sk->sk_dst_cache, dst);
1959 dst_release(old_dst);
1963 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1965 struct dst_entry *old_dst;
1967 sk_tx_queue_clear(sk);
1968 sk->sk_dst_pending_confirm = 0;
1969 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1970 dst_release(old_dst);
1974 __sk_dst_reset(struct sock *sk)
1976 __sk_dst_set(sk, NULL);
1980 sk_dst_reset(struct sock *sk)
1982 sk_dst_set(sk, NULL);
1985 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1987 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1989 static inline void sk_dst_confirm(struct sock *sk)
1991 if (!READ_ONCE(sk->sk_dst_pending_confirm))
1992 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
1995 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1997 if (skb_get_dst_pending_confirm(skb)) {
1998 struct sock *sk = skb->sk;
1999 unsigned long now = jiffies;
2001 /* avoid dirtying neighbour */
2002 if (READ_ONCE(n->confirmed) != now)
2003 WRITE_ONCE(n->confirmed, now);
2004 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2005 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2009 bool sk_mc_loop(struct sock *sk);
2011 static inline bool sk_can_gso(const struct sock *sk)
2013 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2016 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2018 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2020 sk->sk_route_nocaps |= flags;
2021 sk->sk_route_caps &= ~flags;
2024 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2025 struct iov_iter *from, char *to,
2026 int copy, int offset)
2028 if (skb->ip_summed == CHECKSUM_NONE) {
2030 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2032 skb->csum = csum_block_add(skb->csum, csum, offset);
2033 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2034 if (!copy_from_iter_full_nocache(to, copy, from))
2036 } else if (!copy_from_iter_full(to, copy, from))
2042 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2043 struct iov_iter *from, int copy)
2045 int err, offset = skb->len;
2047 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2050 __skb_trim(skb, offset);
2055 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2056 struct sk_buff *skb,
2062 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2068 skb->data_len += copy;
2069 skb->truesize += copy;
2070 sk_wmem_queued_add(sk, copy);
2071 sk_mem_charge(sk, copy);
2076 * sk_wmem_alloc_get - returns write allocations
2079 * Return: sk_wmem_alloc minus initial offset of one
2081 static inline int sk_wmem_alloc_get(const struct sock *sk)
2083 return refcount_read(&sk->sk_wmem_alloc) - 1;
2087 * sk_rmem_alloc_get - returns read allocations
2090 * Return: sk_rmem_alloc
2092 static inline int sk_rmem_alloc_get(const struct sock *sk)
2094 return atomic_read(&sk->sk_rmem_alloc);
2098 * sk_has_allocations - check if allocations are outstanding
2101 * Return: true if socket has write or read allocations
2103 static inline bool sk_has_allocations(const struct sock *sk)
2105 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2109 * skwq_has_sleeper - check if there are any waiting processes
2110 * @wq: struct socket_wq
2112 * Return: true if socket_wq has waiting processes
2114 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2115 * barrier call. They were added due to the race found within the tcp code.
2117 * Consider following tcp code paths::
2120 * sys_select receive packet
2122 * __add_wait_queue update tp->rcv_nxt
2124 * tp->rcv_nxt check sock_def_readable
2126 * schedule rcu_read_lock();
2127 * wq = rcu_dereference(sk->sk_wq);
2128 * if (wq && waitqueue_active(&wq->wait))
2129 * wake_up_interruptible(&wq->wait)
2133 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2134 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2135 * could then endup calling schedule and sleep forever if there are no more
2136 * data on the socket.
2139 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2141 return wq && wq_has_sleeper(&wq->wait);
2145 * sock_poll_wait - place memory barrier behind the poll_wait call.
2147 * @sock: socket to wait on
2150 * See the comments in the wq_has_sleeper function.
2152 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2155 if (!poll_does_not_wait(p)) {
2156 poll_wait(filp, &sock->wq.wait, p);
2157 /* We need to be sure we are in sync with the
2158 * socket flags modification.
2160 * This memory barrier is paired in the wq_has_sleeper.
2166 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2168 if (sk->sk_txhash) {
2170 skb->hash = sk->sk_txhash;
2174 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2177 * Queue a received datagram if it will fit. Stream and sequenced
2178 * protocols can't normally use this as they need to fit buffers in
2179 * and play with them.
2181 * Inlined as it's very short and called for pretty much every
2182 * packet ever received.
2184 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2188 skb->destructor = sock_rfree;
2189 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2190 sk_mem_charge(sk, skb->truesize);
2193 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2194 unsigned long expires);
2196 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2198 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2199 struct sk_buff *skb, unsigned int flags,
2200 void (*destructor)(struct sock *sk,
2201 struct sk_buff *skb));
2202 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2203 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2205 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2206 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2209 * Recover an error report and clear atomically
2212 static inline int sock_error(struct sock *sk)
2215 if (likely(!sk->sk_err))
2217 err = xchg(&sk->sk_err, 0);
2221 static inline unsigned long sock_wspace(struct sock *sk)
2225 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2226 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2234 * We use sk->sk_wq_raw, from contexts knowing this
2235 * pointer is not NULL and cannot disappear/change.
2237 static inline void sk_set_bit(int nr, struct sock *sk)
2239 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2240 !sock_flag(sk, SOCK_FASYNC))
2243 set_bit(nr, &sk->sk_wq_raw->flags);
2246 static inline void sk_clear_bit(int nr, struct sock *sk)
2248 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2249 !sock_flag(sk, SOCK_FASYNC))
2252 clear_bit(nr, &sk->sk_wq_raw->flags);
2255 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2257 if (sock_flag(sk, SOCK_FASYNC)) {
2259 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2264 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2265 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2266 * Note: for send buffers, TCP works better if we can build two skbs at
2269 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2271 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2272 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2274 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2278 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2281 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2283 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2286 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2287 bool force_schedule);
2290 * sk_page_frag - return an appropriate page_frag
2293 * Use the per task page_frag instead of the per socket one for
2294 * optimization when we know that we're in the normal context and owns
2295 * everything that's associated with %current.
2297 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2298 * inside other socket operations and end up recursing into sk_page_frag()
2299 * while it's already in use.
2301 * Return: a per task page_frag if context allows that,
2302 * otherwise a per socket one.
2304 static inline struct page_frag *sk_page_frag(struct sock *sk)
2306 if (gfpflags_normal_context(sk->sk_allocation))
2307 return ¤t->task_frag;
2309 return &sk->sk_frag;
2312 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2315 * Default write policy as shown to user space via poll/select/SIGIO
2317 static inline bool sock_writeable(const struct sock *sk)
2319 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2322 static inline gfp_t gfp_any(void)
2324 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2327 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2329 return noblock ? 0 : sk->sk_rcvtimeo;
2332 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2334 return noblock ? 0 : sk->sk_sndtimeo;
2337 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2339 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2344 /* Alas, with timeout socket operations are not restartable.
2345 * Compare this to poll().
2347 static inline int sock_intr_errno(long timeo)
2349 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2352 struct sock_skb_cb {
2356 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2357 * using skb->cb[] would keep using it directly and utilize its
2358 * alignement guarantee.
2360 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2361 sizeof(struct sock_skb_cb)))
2363 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2364 SOCK_SKB_CB_OFFSET))
2366 #define sock_skb_cb_check_size(size) \
2367 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2370 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2372 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2373 atomic_read(&sk->sk_drops) : 0;
2376 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2378 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2380 atomic_add(segs, &sk->sk_drops);
2383 static inline ktime_t sock_read_timestamp(struct sock *sk)
2385 #if BITS_PER_LONG==32
2390 seq = read_seqbegin(&sk->sk_stamp_seq);
2392 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2396 return READ_ONCE(sk->sk_stamp);
2400 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2402 #if BITS_PER_LONG==32
2403 write_seqlock(&sk->sk_stamp_seq);
2405 write_sequnlock(&sk->sk_stamp_seq);
2407 WRITE_ONCE(sk->sk_stamp, kt);
2411 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2412 struct sk_buff *skb);
2413 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2414 struct sk_buff *skb);
2417 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2419 ktime_t kt = skb->tstamp;
2420 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2423 * generate control messages if
2424 * - receive time stamping in software requested
2425 * - software time stamp available and wanted
2426 * - hardware time stamps available and wanted
2428 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2429 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2430 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2431 (hwtstamps->hwtstamp &&
2432 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2433 __sock_recv_timestamp(msg, sk, skb);
2435 sock_write_timestamp(sk, kt);
2437 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2438 __sock_recv_wifi_status(msg, sk, skb);
2441 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2442 struct sk_buff *skb);
2444 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2445 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2446 struct sk_buff *skb)
2448 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2449 (1UL << SOCK_RCVTSTAMP))
2450 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2451 SOF_TIMESTAMPING_RAW_HARDWARE)
2453 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2454 __sock_recv_ts_and_drops(msg, sk, skb);
2455 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2456 sock_write_timestamp(sk, skb->tstamp);
2457 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2458 sock_write_timestamp(sk, 0);
2461 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2464 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2465 * @sk: socket sending this packet
2466 * @tsflags: timestamping flags to use
2467 * @tx_flags: completed with instructions for time stamping
2468 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2470 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2472 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2473 __u8 *tx_flags, __u32 *tskey)
2475 if (unlikely(tsflags)) {
2476 __sock_tx_timestamp(tsflags, tx_flags);
2477 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2478 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2479 *tskey = sk->sk_tskey++;
2481 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2482 *tx_flags |= SKBTX_WIFI_STATUS;
2485 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2488 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2491 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2493 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2494 &skb_shinfo(skb)->tskey);
2497 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2499 * sk_eat_skb - Release a skb if it is no longer needed
2500 * @sk: socket to eat this skb from
2501 * @skb: socket buffer to eat
2503 * This routine must be called with interrupts disabled or with the socket
2504 * locked so that the sk_buff queue operation is ok.
2506 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2508 __skb_unlink(skb, &sk->sk_receive_queue);
2509 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2510 !sk->sk_rx_skb_cache) {
2511 sk->sk_rx_skb_cache = skb;
2519 struct net *sock_net(const struct sock *sk)
2521 return read_pnet(&sk->sk_net);
2525 void sock_net_set(struct sock *sk, struct net *net)
2527 write_pnet(&sk->sk_net, net);
2531 skb_sk_is_prefetched(struct sk_buff *skb)
2534 return skb->destructor == sock_pfree;
2537 #endif /* CONFIG_INET */
2540 /* This helper checks if a socket is a full socket,
2541 * ie _not_ a timewait or request socket.
2543 static inline bool sk_fullsock(const struct sock *sk)
2545 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2549 sk_is_refcounted(struct sock *sk)
2551 /* Only full sockets have sk->sk_flags. */
2552 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2557 * @skb to steal the socket from
2558 * @refcounted is set to true if the socket is reference-counted
2560 static inline struct sock *
2561 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2564 struct sock *sk = skb->sk;
2567 if (skb_sk_is_prefetched(skb))
2568 *refcounted = sk_is_refcounted(sk);
2569 skb->destructor = NULL;
2573 *refcounted = false;
2577 /* Checks if this SKB belongs to an HW offloaded socket
2578 * and whether any SW fallbacks are required based on dev.
2579 * Check decrypted mark in case skb_orphan() cleared socket.
2581 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2582 struct net_device *dev)
2584 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2585 struct sock *sk = skb->sk;
2587 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2588 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2589 #ifdef CONFIG_TLS_DEVICE
2590 } else if (unlikely(skb->decrypted)) {
2591 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2601 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2602 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2604 static inline bool sk_listener(const struct sock *sk)
2606 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2609 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2610 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2613 bool sk_ns_capable(const struct sock *sk,
2614 struct user_namespace *user_ns, int cap);
2615 bool sk_capable(const struct sock *sk, int cap);
2616 bool sk_net_capable(const struct sock *sk, int cap);
2618 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2620 /* Take into consideration the size of the struct sk_buff overhead in the
2621 * determination of these values, since that is non-constant across
2622 * platforms. This makes socket queueing behavior and performance
2623 * not depend upon such differences.
2625 #define _SK_MEM_PACKETS 256
2626 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2627 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2628 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2630 extern __u32 sysctl_wmem_max;
2631 extern __u32 sysctl_rmem_max;
2633 extern int sysctl_tstamp_allow_data;
2634 extern int sysctl_optmem_max;
2636 extern __u32 sysctl_wmem_default;
2637 extern __u32 sysctl_rmem_default;
2639 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2641 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2643 /* Does this proto have per netns sysctl_wmem ? */
2644 if (proto->sysctl_wmem_offset)
2645 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2647 return *proto->sysctl_wmem;
2650 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2652 /* Does this proto have per netns sysctl_rmem ? */
2653 if (proto->sysctl_rmem_offset)
2654 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2656 return *proto->sysctl_rmem;
2659 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2660 * Some wifi drivers need to tweak it to get more chunks.
2661 * They can use this helper from their ndo_start_xmit()
2663 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2665 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2667 WRITE_ONCE(sk->sk_pacing_shift, val);
2670 /* if a socket is bound to a device, check that the given device
2671 * index is either the same or that the socket is bound to an L3
2672 * master device and the given device index is also enslaved to
2675 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2679 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2682 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2683 if (mdif && mdif == sk->sk_bound_dev_if)
2689 void sock_def_readable(struct sock *sk);
2691 #endif /* _SOCK_H */