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>
62 #include <linux/sockptr.h>
64 #include <linux/atomic.h>
65 #include <linux/refcount.h>
67 #include <net/checksum.h>
68 #include <net/tcp_states.h>
69 #include <linux/net_tstamp.h>
70 #include <net/l3mdev.h>
73 * This structure really needs to be cleaned up.
74 * Most of it is for TCP, and not used by any of
75 * the other protocols.
78 /* Define this to get the SOCK_DBG debugging facility. */
79 #define SOCK_DEBUGGING
81 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
82 printk(KERN_DEBUG msg); } while (0)
84 /* Validate arguments and do nothing */
85 static inline __printf(2, 3)
86 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
91 /* This is the per-socket lock. The spinlock provides a synchronization
92 * between user contexts and software interrupt processing, whereas the
93 * mini-semaphore synchronizes multiple users amongst themselves.
100 * We express the mutex-alike socket_lock semantics
101 * to the lock validator by explicitly managing
102 * the slock as a lock variant (in addition to
105 #ifdef CONFIG_DEBUG_LOCK_ALLOC
106 struct lockdep_map dep_map;
114 typedef __u32 __bitwise __portpair;
115 typedef __u64 __bitwise __addrpair;
118 * struct sock_common - minimal network layer representation of sockets
119 * @skc_daddr: Foreign IPv4 addr
120 * @skc_rcv_saddr: Bound local IPv4 addr
121 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
122 * @skc_hash: hash value used with various protocol lookup tables
123 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
124 * @skc_dport: placeholder for inet_dport/tw_dport
125 * @skc_num: placeholder for inet_num/tw_num
126 * @skc_portpair: __u32 union of @skc_dport & @skc_num
127 * @skc_family: network address family
128 * @skc_state: Connection state
129 * @skc_reuse: %SO_REUSEADDR setting
130 * @skc_reuseport: %SO_REUSEPORT setting
131 * @skc_ipv6only: socket is IPV6 only
132 * @skc_net_refcnt: socket is using net ref counting
133 * @skc_bound_dev_if: bound device index if != 0
134 * @skc_bind_node: bind hash linkage for various protocol lookup tables
135 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
136 * @skc_prot: protocol handlers inside a network family
137 * @skc_net: reference to the network namespace of this socket
138 * @skc_v6_daddr: IPV6 destination address
139 * @skc_v6_rcv_saddr: IPV6 source address
140 * @skc_cookie: socket's cookie value
141 * @skc_node: main hash linkage for various protocol lookup tables
142 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
143 * @skc_tx_queue_mapping: tx queue number for this connection
144 * @skc_rx_queue_mapping: rx queue number for this connection
145 * @skc_flags: place holder for sk_flags
146 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
147 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
148 * @skc_listener: connection request listener socket (aka rsk_listener)
149 * [union with @skc_flags]
150 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
151 * [union with @skc_flags]
152 * @skc_incoming_cpu: record/match cpu processing incoming packets
153 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
154 * [union with @skc_incoming_cpu]
155 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
156 * [union with @skc_incoming_cpu]
157 * @skc_refcnt: reference count
159 * This is the minimal network layer representation of sockets, the header
160 * for struct sock and struct inet_timewait_sock.
164 __addrpair skc_addrpair;
167 __be32 skc_rcv_saddr;
171 unsigned int skc_hash;
172 __u16 skc_u16hashes[2];
174 /* skc_dport && skc_num must be grouped as well */
176 __portpair skc_portpair;
183 unsigned short skc_family;
184 volatile unsigned char skc_state;
185 unsigned char skc_reuse:4;
186 unsigned char skc_reuseport:1;
187 unsigned char skc_ipv6only:1;
188 unsigned char skc_net_refcnt:1;
189 int skc_bound_dev_if;
191 struct hlist_node skc_bind_node;
192 struct hlist_node skc_portaddr_node;
194 struct proto *skc_prot;
195 possible_net_t skc_net;
197 #if IS_ENABLED(CONFIG_IPV6)
198 struct in6_addr skc_v6_daddr;
199 struct in6_addr skc_v6_rcv_saddr;
202 atomic64_t skc_cookie;
204 /* following fields are padding to force
205 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
206 * assuming IPV6 is enabled. We use this padding differently
207 * for different kind of 'sockets'
210 unsigned long skc_flags;
211 struct sock *skc_listener; /* request_sock */
212 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
215 * fields between dontcopy_begin/dontcopy_end
216 * are not copied in sock_copy()
219 int skc_dontcopy_begin[0];
222 struct hlist_node skc_node;
223 struct hlist_nulls_node skc_nulls_node;
225 unsigned short skc_tx_queue_mapping;
227 unsigned short skc_rx_queue_mapping;
230 int skc_incoming_cpu;
232 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
235 refcount_t skc_refcnt;
237 int skc_dontcopy_end[0];
240 u32 skc_window_clamp;
241 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
246 struct bpf_local_storage;
249 * struct sock - network layer representation of sockets
250 * @__sk_common: shared layout with inet_timewait_sock
251 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
252 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
253 * @sk_lock: synchronizer
254 * @sk_kern_sock: True if sock is using kernel lock classes
255 * @sk_rcvbuf: size of receive buffer in bytes
256 * @sk_wq: sock wait queue and async head
257 * @sk_rx_dst: receive input route used by early demux
258 * @sk_dst_cache: destination cache
259 * @sk_dst_pending_confirm: need to confirm neighbour
260 * @sk_policy: flow policy
261 * @sk_rx_skb_cache: cache copy of recently accessed RX skb
262 * @sk_receive_queue: incoming packets
263 * @sk_wmem_alloc: transmit queue bytes committed
264 * @sk_tsq_flags: TCP Small Queues flags
265 * @sk_write_queue: Packet sending queue
266 * @sk_omem_alloc: "o" is "option" or "other"
267 * @sk_wmem_queued: persistent queue size
268 * @sk_forward_alloc: space allocated forward
269 * @sk_napi_id: id of the last napi context to receive data for sk
270 * @sk_ll_usec: usecs to busypoll when there is no data
271 * @sk_allocation: allocation mode
272 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
273 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
274 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
275 * @sk_sndbuf: size of send buffer in bytes
276 * @__sk_flags_offset: empty field used to determine location of bitfield
277 * @sk_padding: unused element for alignment
278 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
279 * @sk_no_check_rx: allow zero checksum in RX packets
280 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
281 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
282 * @sk_route_forced_caps: static, forced route capabilities
283 * (set in tcp_init_sock())
284 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
285 * @sk_gso_max_size: Maximum GSO segment size to build
286 * @sk_gso_max_segs: Maximum number of GSO segments
287 * @sk_pacing_shift: scaling factor for TCP Small Queues
288 * @sk_lingertime: %SO_LINGER l_linger setting
289 * @sk_backlog: always used with the per-socket spinlock held
290 * @sk_callback_lock: used with the callbacks in the end of this struct
291 * @sk_error_queue: rarely used
292 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
293 * IPV6_ADDRFORM for instance)
294 * @sk_err: last error
295 * @sk_err_soft: errors that don't cause failure but are the cause of a
296 * persistent failure not just 'timed out'
297 * @sk_drops: raw/udp drops counter
298 * @sk_ack_backlog: current listen backlog
299 * @sk_max_ack_backlog: listen backlog set in listen()
300 * @sk_uid: user id of owner
301 * @sk_priority: %SO_PRIORITY setting
302 * @sk_type: socket type (%SOCK_STREAM, etc)
303 * @sk_protocol: which protocol this socket belongs in this network family
304 * @sk_peer_pid: &struct pid for this socket's peer
305 * @sk_peer_cred: %SO_PEERCRED setting
306 * @sk_rcvlowat: %SO_RCVLOWAT setting
307 * @sk_rcvtimeo: %SO_RCVTIMEO setting
308 * @sk_sndtimeo: %SO_SNDTIMEO setting
309 * @sk_txhash: computed flow hash for use on transmit
310 * @sk_filter: socket filtering instructions
311 * @sk_timer: sock cleanup timer
312 * @sk_stamp: time stamp of last packet received
313 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
314 * @sk_tsflags: SO_TIMESTAMPING socket options
315 * @sk_tskey: counter to disambiguate concurrent tstamp requests
316 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
317 * @sk_socket: Identd and reporting IO signals
318 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
319 * @sk_frag: cached page frag
320 * @sk_peek_off: current peek_offset value
321 * @sk_send_head: front of stuff to transmit
322 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
323 * @sk_tx_skb_cache: cache copy of recently accessed TX skb
324 * @sk_security: used by security modules
325 * @sk_mark: generic packet mark
326 * @sk_cgrp_data: cgroup data for this cgroup
327 * @sk_memcg: this socket's memory cgroup association
328 * @sk_write_pending: a write to stream socket waits to start
329 * @sk_wait_pending: number of threads blocked on this socket
330 * @sk_state_change: callback to indicate change in the state of the sock
331 * @sk_data_ready: callback to indicate there is data to be processed
332 * @sk_write_space: callback to indicate there is bf sending space available
333 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
334 * @sk_backlog_rcv: callback to process the backlog
335 * @sk_validate_xmit_skb: ptr to an optional validate function
336 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
337 * @sk_reuseport_cb: reuseport group container
338 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
339 * @sk_rcu: used during RCU grace period
340 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
341 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
342 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
343 * @sk_txtime_unused: unused txtime flags
347 * Now struct inet_timewait_sock also uses sock_common, so please just
348 * don't add nothing before this first member (__sk_common) --acme
350 struct sock_common __sk_common;
351 #define sk_node __sk_common.skc_node
352 #define sk_nulls_node __sk_common.skc_nulls_node
353 #define sk_refcnt __sk_common.skc_refcnt
354 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
356 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
359 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
360 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
361 #define sk_hash __sk_common.skc_hash
362 #define sk_portpair __sk_common.skc_portpair
363 #define sk_num __sk_common.skc_num
364 #define sk_dport __sk_common.skc_dport
365 #define sk_addrpair __sk_common.skc_addrpair
366 #define sk_daddr __sk_common.skc_daddr
367 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
368 #define sk_family __sk_common.skc_family
369 #define sk_state __sk_common.skc_state
370 #define sk_reuse __sk_common.skc_reuse
371 #define sk_reuseport __sk_common.skc_reuseport
372 #define sk_ipv6only __sk_common.skc_ipv6only
373 #define sk_net_refcnt __sk_common.skc_net_refcnt
374 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
375 #define sk_bind_node __sk_common.skc_bind_node
376 #define sk_prot __sk_common.skc_prot
377 #define sk_net __sk_common.skc_net
378 #define sk_v6_daddr __sk_common.skc_v6_daddr
379 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
380 #define sk_cookie __sk_common.skc_cookie
381 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
382 #define sk_flags __sk_common.skc_flags
383 #define sk_rxhash __sk_common.skc_rxhash
385 socket_lock_t sk_lock;
388 struct sk_buff_head sk_error_queue;
389 struct sk_buff *sk_rx_skb_cache;
390 struct sk_buff_head sk_receive_queue;
392 * The backlog queue is special, it is always used with
393 * the per-socket spinlock held and requires low latency
394 * access. Therefore we special case it's implementation.
395 * Note : rmem_alloc is in this structure to fill a hole
396 * on 64bit arches, not because its logically part of
402 struct sk_buff *head;
403 struct sk_buff *tail;
405 #define sk_rmem_alloc sk_backlog.rmem_alloc
407 int sk_forward_alloc;
408 #ifdef CONFIG_NET_RX_BUSY_POLL
409 unsigned int sk_ll_usec;
410 /* ===== mostly read cache line ===== */
411 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 __rcu *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 spinlock_t sk_peer_lock;
482 struct pid *sk_peer_pid;
483 const struct cred *sk_peer_cred;
487 #if BITS_PER_LONG==32
488 seqlock_t sk_stamp_seq;
496 u8 sk_txtime_deadline_mode : 1,
497 sk_txtime_report_errors : 1,
498 sk_txtime_unused : 6;
500 struct socket *sk_socket;
502 #ifdef CONFIG_SECURITY
505 struct sock_cgroup_data sk_cgrp_data;
506 struct mem_cgroup *sk_memcg;
507 void (*sk_state_change)(struct sock *sk);
508 void (*sk_data_ready)(struct sock *sk);
509 void (*sk_write_space)(struct sock *sk);
510 void (*sk_error_report)(struct sock *sk);
511 int (*sk_backlog_rcv)(struct sock *sk,
512 struct sk_buff *skb);
513 #ifdef CONFIG_SOCK_VALIDATE_XMIT
514 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
515 struct net_device *dev,
516 struct sk_buff *skb);
518 void (*sk_destruct)(struct sock *sk);
519 struct sock_reuseport __rcu *sk_reuseport_cb;
520 #ifdef CONFIG_BPF_SYSCALL
521 struct bpf_local_storage __rcu *sk_bpf_storage;
523 struct rcu_head sk_rcu;
528 SK_PACING_NEEDED = 1,
532 /* flag bits in sk_user_data
534 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
535 * not be suitable for copying when cloning the socket. For instance,
536 * it can point to a reference counted object. sk_user_data bottom
537 * bit is set if pointer must not be copied.
539 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
540 * managed/owned by a BPF reuseport array. This bit should be set
541 * when sk_user_data's sk is added to the bpf's reuseport_array.
543 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
544 * sk_user_data points to psock type. This bit should be set
545 * when sk_user_data is assigned to a psock object.
547 #define SK_USER_DATA_NOCOPY 1UL
548 #define SK_USER_DATA_BPF 2UL
549 #define SK_USER_DATA_PSOCK 4UL
550 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
554 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
557 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
559 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
562 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
565 * __rcu_dereference_sk_user_data_with_flags - return the pointer
566 * only if argument flags all has been set in sk_user_data. Otherwise
573 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
576 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
578 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
580 if ((sk_user_data & flags) == flags)
581 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
585 #define rcu_dereference_sk_user_data(sk) \
586 __rcu_dereference_sk_user_data_with_flags(sk, 0)
587 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
589 uintptr_t __tmp1 = (uintptr_t)(ptr), \
590 __tmp2 = (uintptr_t)(flags); \
591 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
592 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
593 rcu_assign_pointer(__sk_user_data((sk)), \
596 #define rcu_assign_sk_user_data(sk, ptr) \
597 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
600 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
601 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
602 * on a socket means that the socket will reuse everybody else's port
603 * without looking at the other's sk_reuse value.
606 #define SK_NO_REUSE 0
607 #define SK_CAN_REUSE 1
608 #define SK_FORCE_REUSE 2
610 int sk_set_peek_off(struct sock *sk, int val);
612 static inline int sk_peek_offset(struct sock *sk, int flags)
614 if (unlikely(flags & MSG_PEEK)) {
615 return READ_ONCE(sk->sk_peek_off);
621 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
623 s32 off = READ_ONCE(sk->sk_peek_off);
625 if (unlikely(off >= 0)) {
626 off = max_t(s32, off - val, 0);
627 WRITE_ONCE(sk->sk_peek_off, off);
631 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
633 sk_peek_offset_bwd(sk, -val);
637 * Hashed lists helper routines
639 static inline struct sock *sk_entry(const struct hlist_node *node)
641 return hlist_entry(node, struct sock, sk_node);
644 static inline struct sock *__sk_head(const struct hlist_head *head)
646 return hlist_entry(head->first, struct sock, sk_node);
649 static inline struct sock *sk_head(const struct hlist_head *head)
651 return hlist_empty(head) ? NULL : __sk_head(head);
654 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
656 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
659 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
661 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
664 static inline struct sock *sk_next(const struct sock *sk)
666 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
669 static inline struct sock *sk_nulls_next(const struct sock *sk)
671 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
672 hlist_nulls_entry(sk->sk_nulls_node.next,
673 struct sock, sk_nulls_node) :
677 static inline bool sk_unhashed(const struct sock *sk)
679 return hlist_unhashed(&sk->sk_node);
682 static inline bool sk_hashed(const struct sock *sk)
684 return !sk_unhashed(sk);
687 static inline void sk_node_init(struct hlist_node *node)
692 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
697 static inline void __sk_del_node(struct sock *sk)
699 __hlist_del(&sk->sk_node);
702 /* NB: equivalent to hlist_del_init_rcu */
703 static inline bool __sk_del_node_init(struct sock *sk)
707 sk_node_init(&sk->sk_node);
713 /* Grab socket reference count. This operation is valid only
714 when sk is ALREADY grabbed f.e. it is found in hash table
715 or a list and the lookup is made under lock preventing hash table
719 static __always_inline void sock_hold(struct sock *sk)
721 refcount_inc(&sk->sk_refcnt);
724 /* Ungrab socket in the context, which assumes that socket refcnt
725 cannot hit zero, f.e. it is true in context of any socketcall.
727 static __always_inline void __sock_put(struct sock *sk)
729 refcount_dec(&sk->sk_refcnt);
732 static inline bool sk_del_node_init(struct sock *sk)
734 bool rc = __sk_del_node_init(sk);
737 /* paranoid for a while -acme */
738 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
743 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
745 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
748 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
754 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
756 bool rc = __sk_nulls_del_node_init_rcu(sk);
759 /* paranoid for a while -acme */
760 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
766 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
768 hlist_add_head(&sk->sk_node, list);
771 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
774 __sk_add_node(sk, list);
777 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
780 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
781 sk->sk_family == AF_INET6)
782 hlist_add_tail_rcu(&sk->sk_node, list);
784 hlist_add_head_rcu(&sk->sk_node, list);
787 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
790 hlist_add_tail_rcu(&sk->sk_node, list);
793 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
795 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
798 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
800 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
803 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
806 __sk_nulls_add_node_rcu(sk, list);
809 static inline void __sk_del_bind_node(struct sock *sk)
811 __hlist_del(&sk->sk_bind_node);
814 static inline void sk_add_bind_node(struct sock *sk,
815 struct hlist_head *list)
817 hlist_add_head(&sk->sk_bind_node, list);
820 #define sk_for_each(__sk, list) \
821 hlist_for_each_entry(__sk, list, sk_node)
822 #define sk_for_each_rcu(__sk, list) \
823 hlist_for_each_entry_rcu(__sk, list, sk_node)
824 #define sk_nulls_for_each(__sk, node, list) \
825 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
826 #define sk_nulls_for_each_rcu(__sk, node, list) \
827 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
828 #define sk_for_each_from(__sk) \
829 hlist_for_each_entry_from(__sk, sk_node)
830 #define sk_nulls_for_each_from(__sk, node) \
831 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
832 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
833 #define sk_for_each_safe(__sk, tmp, list) \
834 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
835 #define sk_for_each_bound(__sk, list) \
836 hlist_for_each_entry(__sk, list, sk_bind_node)
839 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
840 * @tpos: the type * to use as a loop cursor.
841 * @pos: the &struct hlist_node to use as a loop cursor.
842 * @head: the head for your list.
843 * @offset: offset of hlist_node within the struct.
846 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
847 for (pos = rcu_dereference(hlist_first_rcu(head)); \
849 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
850 pos = rcu_dereference(hlist_next_rcu(pos)))
852 static inline struct user_namespace *sk_user_ns(struct sock *sk)
854 /* Careful only use this in a context where these parameters
855 * can not change and must all be valid, such as recvmsg from
858 return sk->sk_socket->file->f_cred->user_ns;
872 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
873 SOCK_DBG, /* %SO_DEBUG setting */
874 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
875 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
876 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
877 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
878 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
879 SOCK_FASYNC, /* fasync() active */
881 SOCK_ZEROCOPY, /* buffers from userspace */
882 SOCK_WIFI_STATUS, /* push wifi status to userspace */
883 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
884 * Will use last 4 bytes of packet sent from
885 * user-space instead.
887 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
888 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
889 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
891 SOCK_XDP, /* XDP is attached */
892 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
895 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
897 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
899 nsk->sk_flags = osk->sk_flags;
902 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
904 __set_bit(flag, &sk->sk_flags);
907 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
909 __clear_bit(flag, &sk->sk_flags);
912 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
916 sock_set_flag(sk, bit);
918 sock_reset_flag(sk, bit);
921 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
923 return test_bit(flag, &sk->sk_flags);
927 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
928 static inline int sk_memalloc_socks(void)
930 return static_branch_unlikely(&memalloc_socks_key);
933 void __receive_sock(struct file *file);
936 static inline int sk_memalloc_socks(void)
941 static inline void __receive_sock(struct file *file)
945 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
947 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
950 static inline void sk_acceptq_removed(struct sock *sk)
952 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
955 static inline void sk_acceptq_added(struct sock *sk)
957 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
960 static inline bool sk_acceptq_is_full(const struct sock *sk)
962 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
966 * Compute minimal free write space needed to queue new packets.
968 static inline int sk_stream_min_wspace(const struct sock *sk)
970 return READ_ONCE(sk->sk_wmem_queued) >> 1;
973 static inline int sk_stream_wspace(const struct sock *sk)
975 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
978 static inline void sk_wmem_queued_add(struct sock *sk, int val)
980 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
983 void sk_stream_write_space(struct sock *sk);
985 /* OOB backlog add */
986 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
988 /* dont let skb dst not refcounted, we are going to leave rcu lock */
991 if (!sk->sk_backlog.tail)
992 WRITE_ONCE(sk->sk_backlog.head, skb);
994 sk->sk_backlog.tail->next = skb;
996 WRITE_ONCE(sk->sk_backlog.tail, skb);
1001 * Take into account size of receive queue and backlog queue
1002 * Do not take into account this skb truesize,
1003 * to allow even a single big packet to come.
1005 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1007 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1009 return qsize > limit;
1012 /* The per-socket spinlock must be held here. */
1013 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1016 if (sk_rcvqueues_full(sk, limit))
1020 * If the skb was allocated from pfmemalloc reserves, only
1021 * allow SOCK_MEMALLOC sockets to use it as this socket is
1022 * helping free memory
1024 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1027 __sk_add_backlog(sk, skb);
1028 sk->sk_backlog.len += skb->truesize;
1032 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1034 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1036 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1037 return __sk_backlog_rcv(sk, skb);
1039 return sk->sk_backlog_rcv(sk, skb);
1042 static inline void sk_incoming_cpu_update(struct sock *sk)
1044 int cpu = raw_smp_processor_id();
1046 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1047 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1050 static inline void sock_rps_record_flow_hash(__u32 hash)
1053 struct rps_sock_flow_table *sock_flow_table;
1056 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1057 rps_record_sock_flow(sock_flow_table, hash);
1062 static inline void sock_rps_record_flow(const struct sock *sk)
1065 if (static_branch_unlikely(&rfs_needed)) {
1066 /* Reading sk->sk_rxhash might incur an expensive cache line
1069 * TCP_ESTABLISHED does cover almost all states where RFS
1070 * might be useful, and is cheaper [1] than testing :
1071 * IPv4: inet_sk(sk)->inet_daddr
1072 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1073 * OR an additional socket flag
1074 * [1] : sk_state and sk_prot are in the same cache line.
1076 if (sk->sk_state == TCP_ESTABLISHED) {
1077 /* This READ_ONCE() is paired with the WRITE_ONCE()
1078 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
1080 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1086 static inline void sock_rps_save_rxhash(struct sock *sk,
1087 const struct sk_buff *skb)
1090 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1091 * here, and another one in sock_rps_record_flow().
1093 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1094 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1098 static inline void sock_rps_reset_rxhash(struct sock *sk)
1101 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1102 WRITE_ONCE(sk->sk_rxhash, 0);
1106 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1108 __sk->sk_wait_pending++; \
1109 release_sock(__sk); \
1110 __rc = __condition; \
1112 *(__timeo) = wait_woken(__wait, \
1113 TASK_INTERRUPTIBLE, \
1116 sched_annotate_sleep(); \
1118 __sk->sk_wait_pending--; \
1119 __rc = __condition; \
1123 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1124 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1125 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1126 int sk_stream_error(struct sock *sk, int flags, int err);
1127 void sk_stream_kill_queues(struct sock *sk);
1128 void sk_set_memalloc(struct sock *sk);
1129 void sk_clear_memalloc(struct sock *sk);
1131 void __sk_flush_backlog(struct sock *sk);
1133 static inline bool sk_flush_backlog(struct sock *sk)
1135 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1136 __sk_flush_backlog(sk);
1142 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1144 struct request_sock_ops;
1145 struct timewait_sock_ops;
1146 struct inet_hashinfo;
1147 struct raw_hashinfo;
1148 struct smc_hashinfo;
1152 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1153 * un-modified. Special care is taken when initializing object to zero.
1155 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1157 if (offsetof(struct sock, sk_node.next) != 0)
1158 memset(sk, 0, offsetof(struct sock, sk_node.next));
1159 memset(&sk->sk_node.pprev, 0,
1160 size - offsetof(struct sock, sk_node.pprev));
1163 /* Networking protocol blocks we attach to sockets.
1164 * socket layer -> transport layer interface
1167 void (*close)(struct sock *sk,
1169 int (*pre_connect)(struct sock *sk,
1170 struct sockaddr *uaddr,
1172 int (*connect)(struct sock *sk,
1173 struct sockaddr *uaddr,
1175 int (*disconnect)(struct sock *sk, int flags);
1177 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1180 int (*ioctl)(struct sock *sk, int cmd,
1182 int (*init)(struct sock *sk);
1183 void (*destroy)(struct sock *sk);
1184 void (*shutdown)(struct sock *sk, int how);
1185 int (*setsockopt)(struct sock *sk, int level,
1186 int optname, sockptr_t optval,
1187 unsigned int optlen);
1188 int (*getsockopt)(struct sock *sk, int level,
1189 int optname, char __user *optval,
1190 int __user *option);
1191 void (*keepalive)(struct sock *sk, int valbool);
1192 #ifdef CONFIG_COMPAT
1193 int (*compat_ioctl)(struct sock *sk,
1194 unsigned int cmd, unsigned long arg);
1196 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1198 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1199 size_t len, int noblock, int flags,
1201 int (*sendpage)(struct sock *sk, struct page *page,
1202 int offset, size_t size, int flags);
1203 int (*bind)(struct sock *sk,
1204 struct sockaddr *addr, int addr_len);
1205 int (*bind_add)(struct sock *sk,
1206 struct sockaddr *addr, int addr_len);
1208 int (*backlog_rcv) (struct sock *sk,
1209 struct sk_buff *skb);
1210 bool (*bpf_bypass_getsockopt)(int level,
1213 void (*release_cb)(struct sock *sk);
1215 /* Keeping track of sk's, looking them up, and port selection methods. */
1216 int (*hash)(struct sock *sk);
1217 void (*unhash)(struct sock *sk);
1218 void (*rehash)(struct sock *sk);
1219 int (*get_port)(struct sock *sk, unsigned short snum);
1221 /* Keeping track of sockets in use */
1222 #ifdef CONFIG_PROC_FS
1223 unsigned int inuse_idx;
1226 bool (*stream_memory_free)(const struct sock *sk, int wake);
1227 bool (*stream_memory_read)(const struct sock *sk);
1228 /* Memory pressure */
1229 void (*enter_memory_pressure)(struct sock *sk);
1230 void (*leave_memory_pressure)(struct sock *sk);
1231 atomic_long_t *memory_allocated; /* Current allocated memory. */
1232 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1234 * Pressure flag: try to collapse.
1235 * Technical note: it is used by multiple contexts non atomically.
1236 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1237 * All the __sk_mem_schedule() is of this nature: accounting
1238 * is strict, actions are advisory and have some latency.
1240 unsigned long *memory_pressure;
1245 u32 sysctl_wmem_offset;
1246 u32 sysctl_rmem_offset;
1251 struct kmem_cache *slab;
1252 unsigned int obj_size;
1253 slab_flags_t slab_flags;
1254 unsigned int useroffset; /* Usercopy region offset */
1255 unsigned int usersize; /* Usercopy region size */
1257 unsigned int __percpu *orphan_count;
1259 struct request_sock_ops *rsk_prot;
1260 struct timewait_sock_ops *twsk_prot;
1263 struct inet_hashinfo *hashinfo;
1264 struct udp_table *udp_table;
1265 struct raw_hashinfo *raw_hash;
1266 struct smc_hashinfo *smc_hash;
1269 struct module *owner;
1273 struct list_head node;
1274 #ifdef SOCK_REFCNT_DEBUG
1277 int (*diag_destroy)(struct sock *sk, int err);
1278 } __randomize_layout;
1280 int proto_register(struct proto *prot, int alloc_slab);
1281 void proto_unregister(struct proto *prot);
1282 int sock_load_diag_module(int family, int protocol);
1284 #ifdef SOCK_REFCNT_DEBUG
1285 static inline void sk_refcnt_debug_inc(struct sock *sk)
1287 atomic_inc(&sk->sk_prot->socks);
1290 static inline void sk_refcnt_debug_dec(struct sock *sk)
1292 atomic_dec(&sk->sk_prot->socks);
1293 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1294 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1297 static inline void sk_refcnt_debug_release(const struct sock *sk)
1299 if (refcount_read(&sk->sk_refcnt) != 1)
1300 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1301 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1303 #else /* SOCK_REFCNT_DEBUG */
1304 #define sk_refcnt_debug_inc(sk) do { } while (0)
1305 #define sk_refcnt_debug_dec(sk) do { } while (0)
1306 #define sk_refcnt_debug_release(sk) do { } while (0)
1307 #endif /* SOCK_REFCNT_DEBUG */
1309 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1311 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1314 return sk->sk_prot->stream_memory_free ?
1315 sk->sk_prot->stream_memory_free(sk, wake) : true;
1318 static inline bool sk_stream_memory_free(const struct sock *sk)
1320 return __sk_stream_memory_free(sk, 0);
1323 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1325 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1326 __sk_stream_memory_free(sk, wake);
1329 static inline bool sk_stream_is_writeable(const struct sock *sk)
1331 return __sk_stream_is_writeable(sk, 0);
1334 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1335 struct cgroup *ancestor)
1337 #ifdef CONFIG_SOCK_CGROUP_DATA
1338 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1345 static inline bool sk_has_memory_pressure(const struct sock *sk)
1347 return sk->sk_prot->memory_pressure != NULL;
1350 static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1352 return sk->sk_prot->memory_pressure &&
1353 !!READ_ONCE(*sk->sk_prot->memory_pressure);
1356 static inline bool sk_under_memory_pressure(const struct sock *sk)
1358 if (!sk->sk_prot->memory_pressure)
1361 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1362 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1365 return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1369 sk_memory_allocated(const struct sock *sk)
1371 return atomic_long_read(sk->sk_prot->memory_allocated);
1375 sk_memory_allocated_add(struct sock *sk, int amt)
1377 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1381 sk_memory_allocated_sub(struct sock *sk, int amt)
1383 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1386 static inline void sk_sockets_allocated_dec(struct sock *sk)
1388 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1391 static inline void sk_sockets_allocated_inc(struct sock *sk)
1393 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1397 sk_sockets_allocated_read_positive(struct sock *sk)
1399 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1403 proto_sockets_allocated_sum_positive(struct proto *prot)
1405 return percpu_counter_sum_positive(prot->sockets_allocated);
1409 proto_memory_allocated(struct proto *prot)
1411 return atomic_long_read(prot->memory_allocated);
1415 proto_memory_pressure(struct proto *prot)
1417 if (!prot->memory_pressure)
1419 return !!READ_ONCE(*prot->memory_pressure);
1423 #ifdef CONFIG_PROC_FS
1424 /* Called with local bh disabled */
1425 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1426 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1427 int sock_inuse_get(struct net *net);
1429 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1436 /* With per-bucket locks this operation is not-atomic, so that
1437 * this version is not worse.
1439 static inline int __sk_prot_rehash(struct sock *sk)
1441 sk->sk_prot->unhash(sk);
1442 return sk->sk_prot->hash(sk);
1445 /* About 10 seconds */
1446 #define SOCK_DESTROY_TIME (10*HZ)
1448 /* Sockets 0-1023 can't be bound to unless you are superuser */
1449 #define PROT_SOCK 1024
1451 #define SHUTDOWN_MASK 3
1452 #define RCV_SHUTDOWN 1
1453 #define SEND_SHUTDOWN 2
1455 #define SOCK_SNDBUF_LOCK 1
1456 #define SOCK_RCVBUF_LOCK 2
1457 #define SOCK_BINDADDR_LOCK 4
1458 #define SOCK_BINDPORT_LOCK 8
1460 struct socket_alloc {
1461 struct socket socket;
1462 struct inode vfs_inode;
1465 static inline struct socket *SOCKET_I(struct inode *inode)
1467 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1470 static inline struct inode *SOCK_INODE(struct socket *socket)
1472 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1476 * Functions for memory accounting
1478 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1479 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1480 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1481 void __sk_mem_reclaim(struct sock *sk, int amount);
1483 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1484 * do not necessarily have 16x time more memory than 4KB ones.
1486 #define SK_MEM_QUANTUM 4096
1487 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1488 #define SK_MEM_SEND 0
1489 #define SK_MEM_RECV 1
1491 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1492 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1494 long val = READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1496 #if PAGE_SIZE > SK_MEM_QUANTUM
1497 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1498 #elif PAGE_SIZE < SK_MEM_QUANTUM
1499 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1504 static inline int sk_mem_pages(int amt)
1506 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1509 static inline bool sk_has_account(struct sock *sk)
1511 /* return true if protocol supports memory accounting */
1512 return !!sk->sk_prot->memory_allocated;
1515 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1519 if (!sk_has_account(sk))
1521 delta = size - sk->sk_forward_alloc;
1522 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1526 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1530 if (!sk_has_account(sk))
1532 delta = size - sk->sk_forward_alloc;
1533 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1534 skb_pfmemalloc(skb);
1537 static inline void sk_mem_reclaim(struct sock *sk)
1539 if (!sk_has_account(sk))
1541 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1542 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1545 static inline void sk_mem_reclaim_partial(struct sock *sk)
1547 if (!sk_has_account(sk))
1549 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1550 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1553 static inline void sk_mem_charge(struct sock *sk, int size)
1555 if (!sk_has_account(sk))
1557 sk->sk_forward_alloc -= size;
1560 static inline void sk_mem_uncharge(struct sock *sk, int size)
1562 if (!sk_has_account(sk))
1564 sk->sk_forward_alloc += size;
1566 /* Avoid a possible overflow.
1567 * TCP send queues can make this happen, if sk_mem_reclaim()
1568 * is not called and more than 2 GBytes are released at once.
1570 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1571 * no need to hold that much forward allocation anyway.
1573 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1574 __sk_mem_reclaim(sk, 1 << 20);
1577 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
1578 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1580 sk_wmem_queued_add(sk, -skb->truesize);
1581 sk_mem_uncharge(sk, skb->truesize);
1582 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1583 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1585 skb_zcopy_clear(skb, true);
1586 sk->sk_tx_skb_cache = skb;
1592 static inline void sock_release_ownership(struct sock *sk)
1594 if (sk->sk_lock.owned) {
1595 sk->sk_lock.owned = 0;
1597 /* The sk_lock has mutex_unlock() semantics: */
1598 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1603 * Macro so as to not evaluate some arguments when
1604 * lockdep is not enabled.
1606 * Mark both the sk_lock and the sk_lock.slock as a
1607 * per-address-family lock class.
1609 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1611 sk->sk_lock.owned = 0; \
1612 init_waitqueue_head(&sk->sk_lock.wq); \
1613 spin_lock_init(&(sk)->sk_lock.slock); \
1614 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1615 sizeof((sk)->sk_lock)); \
1616 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1618 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1621 #ifdef CONFIG_LOCKDEP
1622 static inline bool lockdep_sock_is_held(const struct sock *sk)
1624 return lockdep_is_held(&sk->sk_lock) ||
1625 lockdep_is_held(&sk->sk_lock.slock);
1629 void lock_sock_nested(struct sock *sk, int subclass);
1631 static inline void lock_sock(struct sock *sk)
1633 lock_sock_nested(sk, 0);
1636 void __release_sock(struct sock *sk);
1637 void release_sock(struct sock *sk);
1639 /* BH context may only use the following locking interface. */
1640 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1641 #define bh_lock_sock_nested(__sk) \
1642 spin_lock_nested(&((__sk)->sk_lock.slock), \
1643 SINGLE_DEPTH_NESTING)
1644 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1646 bool lock_sock_fast(struct sock *sk);
1648 * unlock_sock_fast - complement of lock_sock_fast
1652 * fast unlock socket for user context.
1653 * If slow mode is on, we call regular release_sock()
1655 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1660 spin_unlock_bh(&sk->sk_lock.slock);
1663 /* Used by processes to "lock" a socket state, so that
1664 * interrupts and bottom half handlers won't change it
1665 * from under us. It essentially blocks any incoming
1666 * packets, so that we won't get any new data or any
1667 * packets that change the state of the socket.
1669 * While locked, BH processing will add new packets to
1670 * the backlog queue. This queue is processed by the
1671 * owner of the socket lock right before it is released.
1673 * Since ~2.3.5 it is also exclusive sleep lock serializing
1674 * accesses from user process context.
1677 static inline void sock_owned_by_me(const struct sock *sk)
1679 #ifdef CONFIG_LOCKDEP
1680 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1684 static inline void sock_not_owned_by_me(const struct sock *sk)
1686 #ifdef CONFIG_LOCKDEP
1687 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1691 static inline bool sock_owned_by_user(const struct sock *sk)
1693 sock_owned_by_me(sk);
1694 return sk->sk_lock.owned;
1697 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1699 return sk->sk_lock.owned;
1702 /* no reclassification while locks are held */
1703 static inline bool sock_allow_reclassification(const struct sock *csk)
1705 struct sock *sk = (struct sock *)csk;
1707 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1710 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1711 struct proto *prot, int kern);
1712 void sk_free(struct sock *sk);
1713 void sk_destruct(struct sock *sk);
1714 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1715 void sk_free_unlock_clone(struct sock *sk);
1717 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1719 void __sock_wfree(struct sk_buff *skb);
1720 void sock_wfree(struct sk_buff *skb);
1721 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1723 void skb_orphan_partial(struct sk_buff *skb);
1724 void sock_rfree(struct sk_buff *skb);
1725 void sock_efree(struct sk_buff *skb);
1727 void sock_edemux(struct sk_buff *skb);
1728 void sock_pfree(struct sk_buff *skb);
1730 #define sock_edemux sock_efree
1733 int sock_setsockopt(struct socket *sock, int level, int op,
1734 sockptr_t optval, unsigned int optlen);
1736 int sock_getsockopt(struct socket *sock, int level, int op,
1737 char __user *optval, int __user *optlen);
1738 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1739 bool timeval, bool time32);
1740 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1741 int noblock, int *errcode);
1742 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1743 unsigned long data_len, int noblock,
1744 int *errcode, int max_page_order);
1745 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1746 void sock_kfree_s(struct sock *sk, void *mem, int size);
1747 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1748 void sk_send_sigurg(struct sock *sk);
1750 struct sockcm_cookie {
1756 static inline void sockcm_init(struct sockcm_cookie *sockc,
1757 const struct sock *sk)
1759 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1762 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1763 struct sockcm_cookie *sockc);
1764 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1765 struct sockcm_cookie *sockc);
1768 * Functions to fill in entries in struct proto_ops when a protocol
1769 * does not implement a particular function.
1771 int sock_no_bind(struct socket *, struct sockaddr *, int);
1772 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1773 int sock_no_socketpair(struct socket *, struct socket *);
1774 int sock_no_accept(struct socket *, struct socket *, int, bool);
1775 int sock_no_getname(struct socket *, struct sockaddr *, int);
1776 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1777 int sock_no_listen(struct socket *, int);
1778 int sock_no_shutdown(struct socket *, int);
1779 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1780 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1781 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1782 int sock_no_mmap(struct file *file, struct socket *sock,
1783 struct vm_area_struct *vma);
1784 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1785 size_t size, int flags);
1786 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1787 int offset, size_t size, int flags);
1790 * Functions to fill in entries in struct proto_ops when a protocol
1791 * uses the inet style.
1793 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1794 char __user *optval, int __user *optlen);
1795 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1797 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1798 sockptr_t optval, unsigned int optlen);
1800 void sk_common_release(struct sock *sk);
1803 * Default socket callbacks and setup code
1806 /* Initialise core socket variables using an explicit uid. */
1807 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1809 /* Initialise core socket variables.
1810 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1812 void sock_init_data(struct socket *sock, struct sock *sk);
1815 * Socket reference counting postulates.
1817 * * Each user of socket SHOULD hold a reference count.
1818 * * Each access point to socket (an hash table bucket, reference from a list,
1819 * running timer, skb in flight MUST hold a reference count.
1820 * * When reference count hits 0, it means it will never increase back.
1821 * * When reference count hits 0, it means that no references from
1822 * outside exist to this socket and current process on current CPU
1823 * is last user and may/should destroy this socket.
1824 * * sk_free is called from any context: process, BH, IRQ. When
1825 * it is called, socket has no references from outside -> sk_free
1826 * may release descendant resources allocated by the socket, but
1827 * to the time when it is called, socket is NOT referenced by any
1828 * hash tables, lists etc.
1829 * * Packets, delivered from outside (from network or from another process)
1830 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1831 * when they sit in queue. Otherwise, packets will leak to hole, when
1832 * socket is looked up by one cpu and unhasing is made by another CPU.
1833 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1834 * (leak to backlog). Packet socket does all the processing inside
1835 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1836 * use separate SMP lock, so that they are prone too.
1839 /* Ungrab socket and destroy it, if it was the last reference. */
1840 static inline void sock_put(struct sock *sk)
1842 if (refcount_dec_and_test(&sk->sk_refcnt))
1845 /* Generic version of sock_put(), dealing with all sockets
1846 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1848 void sock_gen_put(struct sock *sk);
1850 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1851 unsigned int trim_cap, bool refcounted);
1852 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1855 return __sk_receive_skb(sk, skb, nested, 1, true);
1858 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1860 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1861 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1863 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1864 * other WRITE_ONCE() because socket lock might be not held.
1866 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1869 #define NO_QUEUE_MAPPING USHRT_MAX
1871 static inline void sk_tx_queue_clear(struct sock *sk)
1873 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1874 * other WRITE_ONCE() because socket lock might be not held.
1876 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1879 static inline int sk_tx_queue_get(const struct sock *sk)
1882 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
1883 * and sk_tx_queue_set().
1885 int val = READ_ONCE(sk->sk_tx_queue_mapping);
1887 if (val != NO_QUEUE_MAPPING)
1893 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1896 if (skb_rx_queue_recorded(skb)) {
1897 u16 rx_queue = skb_get_rx_queue(skb);
1899 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1902 sk->sk_rx_queue_mapping = rx_queue;
1907 static inline void sk_rx_queue_clear(struct sock *sk)
1910 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1915 static inline int sk_rx_queue_get(const struct sock *sk)
1917 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1918 return sk->sk_rx_queue_mapping;
1924 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1926 sk->sk_socket = sock;
1929 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1931 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1932 return &rcu_dereference_raw(sk->sk_wq)->wait;
1934 /* Detach socket from process context.
1935 * Announce socket dead, detach it from wait queue and inode.
1936 * Note that parent inode held reference count on this struct sock,
1937 * we do not release it in this function, because protocol
1938 * probably wants some additional cleanups or even continuing
1939 * to work with this socket (TCP).
1941 static inline void sock_orphan(struct sock *sk)
1943 write_lock_bh(&sk->sk_callback_lock);
1944 sock_set_flag(sk, SOCK_DEAD);
1945 sk_set_socket(sk, NULL);
1947 write_unlock_bh(&sk->sk_callback_lock);
1950 static inline void sock_graft(struct sock *sk, struct socket *parent)
1952 WARN_ON(parent->sk);
1953 write_lock_bh(&sk->sk_callback_lock);
1954 rcu_assign_pointer(sk->sk_wq, &parent->wq);
1956 sk_set_socket(sk, parent);
1957 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1958 security_sock_graft(sk, parent);
1959 write_unlock_bh(&sk->sk_callback_lock);
1962 kuid_t sock_i_uid(struct sock *sk);
1963 unsigned long __sock_i_ino(struct sock *sk);
1964 unsigned long sock_i_ino(struct sock *sk);
1966 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1968 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1971 static inline u32 net_tx_rndhash(void)
1973 u32 v = prandom_u32();
1978 static inline void sk_set_txhash(struct sock *sk)
1980 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
1981 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
1984 static inline bool sk_rethink_txhash(struct sock *sk)
1986 if (sk->sk_txhash) {
1993 static inline struct dst_entry *
1994 __sk_dst_get(struct sock *sk)
1996 return rcu_dereference_check(sk->sk_dst_cache,
1997 lockdep_sock_is_held(sk));
2000 static inline struct dst_entry *
2001 sk_dst_get(struct sock *sk)
2003 struct dst_entry *dst;
2006 dst = rcu_dereference(sk->sk_dst_cache);
2007 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
2013 static inline void __dst_negative_advice(struct sock *sk)
2015 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2017 if (dst && dst->ops->negative_advice) {
2018 ndst = dst->ops->negative_advice(dst);
2021 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2022 sk_tx_queue_clear(sk);
2023 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2028 static inline void dst_negative_advice(struct sock *sk)
2030 sk_rethink_txhash(sk);
2031 __dst_negative_advice(sk);
2035 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2037 struct dst_entry *old_dst;
2039 sk_tx_queue_clear(sk);
2040 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2041 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2042 lockdep_sock_is_held(sk));
2043 rcu_assign_pointer(sk->sk_dst_cache, dst);
2044 dst_release(old_dst);
2048 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2050 struct dst_entry *old_dst;
2052 sk_tx_queue_clear(sk);
2053 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2054 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2055 dst_release(old_dst);
2059 __sk_dst_reset(struct sock *sk)
2061 __sk_dst_set(sk, NULL);
2065 sk_dst_reset(struct sock *sk)
2067 sk_dst_set(sk, NULL);
2070 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2072 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2074 static inline void sk_dst_confirm(struct sock *sk)
2076 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2077 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2080 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2082 if (skb_get_dst_pending_confirm(skb)) {
2083 struct sock *sk = skb->sk;
2084 unsigned long now = jiffies;
2086 /* avoid dirtying neighbour */
2087 if (READ_ONCE(n->confirmed) != now)
2088 WRITE_ONCE(n->confirmed, now);
2089 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2090 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2094 bool sk_mc_loop(struct sock *sk);
2096 static inline bool sk_can_gso(const struct sock *sk)
2098 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2101 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2103 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2105 sk->sk_route_nocaps |= flags;
2106 sk->sk_route_caps &= ~flags;
2109 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2110 struct iov_iter *from, char *to,
2111 int copy, int offset)
2113 if (skb->ip_summed == CHECKSUM_NONE) {
2115 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2117 skb->csum = csum_block_add(skb->csum, csum, offset);
2118 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2119 if (!copy_from_iter_full_nocache(to, copy, from))
2121 } else if (!copy_from_iter_full(to, copy, from))
2127 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2128 struct iov_iter *from, int copy)
2130 int err, offset = skb->len;
2132 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2135 __skb_trim(skb, offset);
2140 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2141 struct sk_buff *skb,
2147 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2153 skb->data_len += copy;
2154 skb->truesize += copy;
2155 sk_wmem_queued_add(sk, copy);
2156 sk_mem_charge(sk, copy);
2161 * sk_wmem_alloc_get - returns write allocations
2164 * Return: sk_wmem_alloc minus initial offset of one
2166 static inline int sk_wmem_alloc_get(const struct sock *sk)
2168 return refcount_read(&sk->sk_wmem_alloc) - 1;
2172 * sk_rmem_alloc_get - returns read allocations
2175 * Return: sk_rmem_alloc
2177 static inline int sk_rmem_alloc_get(const struct sock *sk)
2179 return atomic_read(&sk->sk_rmem_alloc);
2183 * sk_has_allocations - check if allocations are outstanding
2186 * Return: true if socket has write or read allocations
2188 static inline bool sk_has_allocations(const struct sock *sk)
2190 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2194 * skwq_has_sleeper - check if there are any waiting processes
2195 * @wq: struct socket_wq
2197 * Return: true if socket_wq has waiting processes
2199 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2200 * barrier call. They were added due to the race found within the tcp code.
2202 * Consider following tcp code paths::
2205 * sys_select receive packet
2207 * __add_wait_queue update tp->rcv_nxt
2209 * tp->rcv_nxt check sock_def_readable
2211 * schedule rcu_read_lock();
2212 * wq = rcu_dereference(sk->sk_wq);
2213 * if (wq && waitqueue_active(&wq->wait))
2214 * wake_up_interruptible(&wq->wait)
2218 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2219 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2220 * could then endup calling schedule and sleep forever if there are no more
2221 * data on the socket.
2224 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2226 return wq && wq_has_sleeper(&wq->wait);
2230 * sock_poll_wait - place memory barrier behind the poll_wait call.
2232 * @sock: socket to wait on
2235 * See the comments in the wq_has_sleeper function.
2237 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2240 if (!poll_does_not_wait(p)) {
2241 poll_wait(filp, &sock->wq.wait, p);
2242 /* We need to be sure we are in sync with the
2243 * socket flags modification.
2245 * This memory barrier is paired in the wq_has_sleeper.
2251 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2253 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2254 u32 txhash = READ_ONCE(sk->sk_txhash);
2262 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2265 * Queue a received datagram if it will fit. Stream and sequenced
2266 * protocols can't normally use this as they need to fit buffers in
2267 * and play with them.
2269 * Inlined as it's very short and called for pretty much every
2270 * packet ever received.
2272 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2276 skb->destructor = sock_rfree;
2277 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2278 sk_mem_charge(sk, skb->truesize);
2281 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2283 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2285 skb->destructor = sock_efree;
2292 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2294 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2296 if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2297 skb_set_owner_r(skb, sk);
2305 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2306 unsigned long expires);
2308 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2310 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2312 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2313 struct sk_buff *skb, unsigned int flags,
2314 void (*destructor)(struct sock *sk,
2315 struct sk_buff *skb));
2316 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2317 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2319 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2320 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2323 * Recover an error report and clear atomically
2326 static inline int sock_error(struct sock *sk)
2330 /* Avoid an atomic operation for the common case.
2331 * This is racy since another cpu/thread can change sk_err under us.
2333 if (likely(data_race(!sk->sk_err)))
2336 err = xchg(&sk->sk_err, 0);
2340 static inline unsigned long sock_wspace(struct sock *sk)
2344 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2345 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2353 * We use sk->sk_wq_raw, from contexts knowing this
2354 * pointer is not NULL and cannot disappear/change.
2356 static inline void sk_set_bit(int nr, struct sock *sk)
2358 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2359 !sock_flag(sk, SOCK_FASYNC))
2362 set_bit(nr, &sk->sk_wq_raw->flags);
2365 static inline void sk_clear_bit(int nr, struct sock *sk)
2367 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2368 !sock_flag(sk, SOCK_FASYNC))
2371 clear_bit(nr, &sk->sk_wq_raw->flags);
2374 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2376 if (sock_flag(sk, SOCK_FASYNC)) {
2378 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2383 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2384 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2385 * Note: for send buffers, TCP works better if we can build two skbs at
2388 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2390 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2391 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2393 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2397 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2400 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2402 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2405 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2406 bool force_schedule);
2409 * sk_page_frag - return an appropriate page_frag
2412 * Use the per task page_frag instead of the per socket one for
2413 * optimization when we know that we're in process context and own
2414 * everything that's associated with %current.
2416 * Both direct reclaim and page faults can nest inside other
2417 * socket operations and end up recursing into sk_page_frag()
2418 * while it's already in use: explicitly avoid task page_frag
2419 * usage if the caller is potentially doing any of them.
2420 * This assumes that page fault handlers use the GFP_NOFS flags.
2422 * Return: a per task page_frag if context allows that,
2423 * otherwise a per socket one.
2425 static inline struct page_frag *sk_page_frag(struct sock *sk)
2427 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) ==
2428 (__GFP_DIRECT_RECLAIM | __GFP_FS))
2429 return ¤t->task_frag;
2431 return &sk->sk_frag;
2434 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2437 * Default write policy as shown to user space via poll/select/SIGIO
2439 static inline bool sock_writeable(const struct sock *sk)
2441 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2444 static inline gfp_t gfp_any(void)
2446 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2449 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2451 return noblock ? 0 : sk->sk_rcvtimeo;
2454 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2456 return noblock ? 0 : sk->sk_sndtimeo;
2459 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2461 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2466 /* Alas, with timeout socket operations are not restartable.
2467 * Compare this to poll().
2469 static inline int sock_intr_errno(long timeo)
2471 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2474 struct sock_skb_cb {
2478 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2479 * using skb->cb[] would keep using it directly and utilize its
2480 * alignement guarantee.
2482 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2483 sizeof(struct sock_skb_cb)))
2485 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2486 SOCK_SKB_CB_OFFSET))
2488 #define sock_skb_cb_check_size(size) \
2489 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2492 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2494 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2495 atomic_read(&sk->sk_drops) : 0;
2498 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2500 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2502 atomic_add(segs, &sk->sk_drops);
2505 static inline ktime_t sock_read_timestamp(struct sock *sk)
2507 #if BITS_PER_LONG==32
2512 seq = read_seqbegin(&sk->sk_stamp_seq);
2514 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2518 return READ_ONCE(sk->sk_stamp);
2522 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2524 #if BITS_PER_LONG==32
2525 write_seqlock(&sk->sk_stamp_seq);
2527 write_sequnlock(&sk->sk_stamp_seq);
2529 WRITE_ONCE(sk->sk_stamp, kt);
2533 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2534 struct sk_buff *skb);
2535 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2536 struct sk_buff *skb);
2539 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2541 ktime_t kt = skb->tstamp;
2542 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2545 * generate control messages if
2546 * - receive time stamping in software requested
2547 * - software time stamp available and wanted
2548 * - hardware time stamps available and wanted
2550 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2551 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2552 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2553 (hwtstamps->hwtstamp &&
2554 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2555 __sock_recv_timestamp(msg, sk, skb);
2557 sock_write_timestamp(sk, kt);
2559 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2560 __sock_recv_wifi_status(msg, sk, skb);
2563 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2564 struct sk_buff *skb);
2566 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2567 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2568 struct sk_buff *skb)
2570 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2571 (1UL << SOCK_RCVTSTAMP))
2572 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2573 SOF_TIMESTAMPING_RAW_HARDWARE)
2575 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2576 __sock_recv_ts_and_drops(msg, sk, skb);
2577 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2578 sock_write_timestamp(sk, skb->tstamp);
2579 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2580 sock_write_timestamp(sk, 0);
2583 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2586 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2587 * @sk: socket sending this packet
2588 * @tsflags: timestamping flags to use
2589 * @tx_flags: completed with instructions for time stamping
2590 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2592 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2594 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2595 __u8 *tx_flags, __u32 *tskey)
2597 if (unlikely(tsflags)) {
2598 __sock_tx_timestamp(tsflags, tx_flags);
2599 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2600 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2601 *tskey = sk->sk_tskey++;
2603 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2604 *tx_flags |= SKBTX_WIFI_STATUS;
2607 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2610 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2613 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2615 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2616 &skb_shinfo(skb)->tskey);
2619 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2621 * sk_eat_skb - Release a skb if it is no longer needed
2622 * @sk: socket to eat this skb from
2623 * @skb: socket buffer to eat
2625 * This routine must be called with interrupts disabled or with the socket
2626 * locked so that the sk_buff queue operation is ok.
2628 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2630 __skb_unlink(skb, &sk->sk_receive_queue);
2631 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2632 !sk->sk_rx_skb_cache) {
2633 sk->sk_rx_skb_cache = skb;
2641 struct net *sock_net(const struct sock *sk)
2643 return read_pnet(&sk->sk_net);
2647 void sock_net_set(struct sock *sk, struct net *net)
2649 write_pnet(&sk->sk_net, net);
2653 skb_sk_is_prefetched(struct sk_buff *skb)
2656 return skb->destructor == sock_pfree;
2659 #endif /* CONFIG_INET */
2662 /* This helper checks if a socket is a full socket,
2663 * ie _not_ a timewait or request socket.
2665 static inline bool sk_fullsock(const struct sock *sk)
2667 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2671 sk_is_refcounted(struct sock *sk)
2673 /* Only full sockets have sk->sk_flags. */
2674 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2678 * skb_steal_sock - steal a socket from an sk_buff
2679 * @skb: sk_buff to steal the socket from
2680 * @refcounted: is set to true if the socket is reference-counted
2682 static inline struct sock *
2683 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2686 struct sock *sk = skb->sk;
2689 if (skb_sk_is_prefetched(skb))
2690 *refcounted = sk_is_refcounted(sk);
2691 skb->destructor = NULL;
2695 *refcounted = false;
2699 /* Checks if this SKB belongs to an HW offloaded socket
2700 * and whether any SW fallbacks are required based on dev.
2701 * Check decrypted mark in case skb_orphan() cleared socket.
2703 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2704 struct net_device *dev)
2706 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2707 struct sock *sk = skb->sk;
2709 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2710 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2711 #ifdef CONFIG_TLS_DEVICE
2712 } else if (unlikely(skb->decrypted)) {
2713 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2723 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2724 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2726 static inline bool sk_listener(const struct sock *sk)
2728 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2731 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2732 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2735 bool sk_ns_capable(const struct sock *sk,
2736 struct user_namespace *user_ns, int cap);
2737 bool sk_capable(const struct sock *sk, int cap);
2738 bool sk_net_capable(const struct sock *sk, int cap);
2740 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2742 /* Take into consideration the size of the struct sk_buff overhead in the
2743 * determination of these values, since that is non-constant across
2744 * platforms. This makes socket queueing behavior and performance
2745 * not depend upon such differences.
2747 #define _SK_MEM_PACKETS 256
2748 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2749 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2750 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2752 extern __u32 sysctl_wmem_max;
2753 extern __u32 sysctl_rmem_max;
2755 extern int sysctl_tstamp_allow_data;
2756 extern int sysctl_optmem_max;
2758 extern __u32 sysctl_wmem_default;
2759 extern __u32 sysctl_rmem_default;
2761 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2762 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2764 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2766 /* Does this proto have per netns sysctl_wmem ? */
2767 if (proto->sysctl_wmem_offset)
2768 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2770 return READ_ONCE(*proto->sysctl_wmem);
2773 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2775 /* Does this proto have per netns sysctl_rmem ? */
2776 if (proto->sysctl_rmem_offset)
2777 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2779 return READ_ONCE(*proto->sysctl_rmem);
2782 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2783 * Some wifi drivers need to tweak it to get more chunks.
2784 * They can use this helper from their ndo_start_xmit()
2786 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2788 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2790 WRITE_ONCE(sk->sk_pacing_shift, val);
2793 /* if a socket is bound to a device, check that the given device
2794 * index is either the same or that the socket is bound to an L3
2795 * master device and the given device index is also enslaved to
2798 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2802 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2805 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2806 if (mdif && mdif == sk->sk_bound_dev_if)
2812 void sock_def_readable(struct sock *sk);
2814 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2815 void sock_enable_timestamps(struct sock *sk);
2816 void sock_no_linger(struct sock *sk);
2817 void sock_set_keepalive(struct sock *sk);
2818 void sock_set_priority(struct sock *sk, u32 priority);
2819 void sock_set_rcvbuf(struct sock *sk, int val);
2820 void sock_set_mark(struct sock *sk, u32 val);
2821 void sock_set_reuseaddr(struct sock *sk);
2822 void sock_set_reuseport(struct sock *sk);
2823 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2825 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2827 #endif /* _SOCK_H */