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>
63 #include <linux/indirect_call_wrapper.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.
163 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
164 * address on 64bit arches : cf INET_MATCH()
167 __addrpair skc_addrpair;
170 __be32 skc_rcv_saddr;
174 unsigned int skc_hash;
175 __u16 skc_u16hashes[2];
177 /* skc_dport && skc_num must be grouped as well */
179 __portpair skc_portpair;
186 unsigned short skc_family;
187 volatile unsigned char skc_state;
188 unsigned char skc_reuse:4;
189 unsigned char skc_reuseport:1;
190 unsigned char skc_ipv6only:1;
191 unsigned char skc_net_refcnt:1;
192 int skc_bound_dev_if;
194 struct hlist_node skc_bind_node;
195 struct hlist_node skc_portaddr_node;
197 struct proto *skc_prot;
198 possible_net_t skc_net;
200 #if IS_ENABLED(CONFIG_IPV6)
201 struct in6_addr skc_v6_daddr;
202 struct in6_addr skc_v6_rcv_saddr;
205 atomic64_t skc_cookie;
207 /* following fields are padding to force
208 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
209 * assuming IPV6 is enabled. We use this padding differently
210 * for different kind of 'sockets'
213 unsigned long skc_flags;
214 struct sock *skc_listener; /* request_sock */
215 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
218 * fields between dontcopy_begin/dontcopy_end
219 * are not copied in sock_copy()
222 int skc_dontcopy_begin[0];
225 struct hlist_node skc_node;
226 struct hlist_nulls_node skc_nulls_node;
228 unsigned short skc_tx_queue_mapping;
230 unsigned short skc_rx_queue_mapping;
233 int skc_incoming_cpu;
235 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
238 refcount_t skc_refcnt;
240 int skc_dontcopy_end[0];
243 u32 skc_window_clamp;
244 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
249 struct bpf_local_storage;
252 * struct sock - network layer representation of sockets
253 * @__sk_common: shared layout with inet_timewait_sock
254 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
255 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
256 * @sk_lock: synchronizer
257 * @sk_kern_sock: True if sock is using kernel lock classes
258 * @sk_rcvbuf: size of receive buffer in bytes
259 * @sk_wq: sock wait queue and async head
260 * @sk_rx_dst: receive input route used by early demux
261 * @sk_dst_cache: destination cache
262 * @sk_dst_pending_confirm: need to confirm neighbour
263 * @sk_policy: flow policy
264 * @sk_rx_skb_cache: cache copy of recently accessed RX skb
265 * @sk_receive_queue: incoming packets
266 * @sk_wmem_alloc: transmit queue bytes committed
267 * @sk_tsq_flags: TCP Small Queues flags
268 * @sk_write_queue: Packet sending queue
269 * @sk_omem_alloc: "o" is "option" or "other"
270 * @sk_wmem_queued: persistent queue size
271 * @sk_forward_alloc: space allocated forward
272 * @sk_napi_id: id of the last napi context to receive data for sk
273 * @sk_ll_usec: usecs to busypoll when there is no data
274 * @sk_allocation: allocation mode
275 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
276 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
277 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
278 * @sk_sndbuf: size of send buffer in bytes
279 * @__sk_flags_offset: empty field used to determine location of bitfield
280 * @sk_padding: unused element for alignment
281 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
282 * @sk_no_check_rx: allow zero checksum in RX packets
283 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
284 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
285 * @sk_route_forced_caps: static, forced route capabilities
286 * (set in tcp_init_sock())
287 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
288 * @sk_gso_max_size: Maximum GSO segment size to build
289 * @sk_gso_max_segs: Maximum number of GSO segments
290 * @sk_pacing_shift: scaling factor for TCP Small Queues
291 * @sk_lingertime: %SO_LINGER l_linger setting
292 * @sk_backlog: always used with the per-socket spinlock held
293 * @sk_callback_lock: used with the callbacks in the end of this struct
294 * @sk_error_queue: rarely used
295 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
296 * IPV6_ADDRFORM for instance)
297 * @sk_err: last error
298 * @sk_err_soft: errors that don't cause failure but are the cause of a
299 * persistent failure not just 'timed out'
300 * @sk_drops: raw/udp drops counter
301 * @sk_ack_backlog: current listen backlog
302 * @sk_max_ack_backlog: listen backlog set in listen()
303 * @sk_uid: user id of owner
304 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
305 * @sk_busy_poll_budget: napi processing budget when busypolling
306 * @sk_priority: %SO_PRIORITY setting
307 * @sk_type: socket type (%SOCK_STREAM, etc)
308 * @sk_protocol: which protocol this socket belongs in this network family
309 * @sk_peer_pid: &struct pid for this socket's peer
310 * @sk_peer_cred: %SO_PEERCRED setting
311 * @sk_rcvlowat: %SO_RCVLOWAT setting
312 * @sk_rcvtimeo: %SO_RCVTIMEO setting
313 * @sk_sndtimeo: %SO_SNDTIMEO setting
314 * @sk_txhash: computed flow hash for use on transmit
315 * @sk_filter: socket filtering instructions
316 * @sk_timer: sock cleanup timer
317 * @sk_stamp: time stamp of last packet received
318 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
319 * @sk_tsflags: SO_TIMESTAMPING socket options
320 * @sk_tskey: counter to disambiguate concurrent tstamp requests
321 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
322 * @sk_socket: Identd and reporting IO signals
323 * @sk_user_data: RPC layer private data
324 * @sk_frag: cached page frag
325 * @sk_peek_off: current peek_offset value
326 * @sk_send_head: front of stuff to transmit
327 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
328 * @sk_tx_skb_cache: cache copy of recently accessed TX skb
329 * @sk_security: used by security modules
330 * @sk_mark: generic packet mark
331 * @sk_cgrp_data: cgroup data for this cgroup
332 * @sk_memcg: this socket's memory cgroup association
333 * @sk_write_pending: a write to stream socket waits to start
334 * @sk_state_change: callback to indicate change in the state of the sock
335 * @sk_data_ready: callback to indicate there is data to be processed
336 * @sk_write_space: callback to indicate there is bf sending space available
337 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
338 * @sk_backlog_rcv: callback to process the backlog
339 * @sk_validate_xmit_skb: ptr to an optional validate function
340 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
341 * @sk_reuseport_cb: reuseport group container
342 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
343 * @sk_rcu: used during RCU grace period
344 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
345 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
346 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
347 * @sk_txtime_unused: unused txtime flags
351 * Now struct inet_timewait_sock also uses sock_common, so please just
352 * don't add nothing before this first member (__sk_common) --acme
354 struct sock_common __sk_common;
355 #define sk_node __sk_common.skc_node
356 #define sk_nulls_node __sk_common.skc_nulls_node
357 #define sk_refcnt __sk_common.skc_refcnt
358 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
360 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
363 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
364 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
365 #define sk_hash __sk_common.skc_hash
366 #define sk_portpair __sk_common.skc_portpair
367 #define sk_num __sk_common.skc_num
368 #define sk_dport __sk_common.skc_dport
369 #define sk_addrpair __sk_common.skc_addrpair
370 #define sk_daddr __sk_common.skc_daddr
371 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
372 #define sk_family __sk_common.skc_family
373 #define sk_state __sk_common.skc_state
374 #define sk_reuse __sk_common.skc_reuse
375 #define sk_reuseport __sk_common.skc_reuseport
376 #define sk_ipv6only __sk_common.skc_ipv6only
377 #define sk_net_refcnt __sk_common.skc_net_refcnt
378 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
379 #define sk_bind_node __sk_common.skc_bind_node
380 #define sk_prot __sk_common.skc_prot
381 #define sk_net __sk_common.skc_net
382 #define sk_v6_daddr __sk_common.skc_v6_daddr
383 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
384 #define sk_cookie __sk_common.skc_cookie
385 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
386 #define sk_flags __sk_common.skc_flags
387 #define sk_rxhash __sk_common.skc_rxhash
389 socket_lock_t sk_lock;
392 struct sk_buff_head sk_error_queue;
393 struct sk_buff *sk_rx_skb_cache;
394 struct sk_buff_head sk_receive_queue;
396 * The backlog queue is special, it is always used with
397 * the per-socket spinlock held and requires low latency
398 * access. Therefore we special case it's implementation.
399 * Note : rmem_alloc is in this structure to fill a hole
400 * on 64bit arches, not because its logically part of
406 struct sk_buff *head;
407 struct sk_buff *tail;
409 #define sk_rmem_alloc sk_backlog.rmem_alloc
411 int sk_forward_alloc;
412 #ifdef CONFIG_NET_RX_BUSY_POLL
413 unsigned int sk_ll_usec;
414 /* ===== mostly read cache line ===== */
415 unsigned int sk_napi_id;
419 struct sk_filter __rcu *sk_filter;
421 struct socket_wq __rcu *sk_wq;
423 struct socket_wq *sk_wq_raw;
427 struct xfrm_policy __rcu *sk_policy[2];
429 struct dst_entry *sk_rx_dst;
430 struct dst_entry __rcu *sk_dst_cache;
431 atomic_t sk_omem_alloc;
434 /* ===== cache line for TX ===== */
436 refcount_t sk_wmem_alloc;
437 unsigned long sk_tsq_flags;
439 struct sk_buff *sk_send_head;
440 struct rb_root tcp_rtx_queue;
442 struct sk_buff *sk_tx_skb_cache;
443 struct sk_buff_head sk_write_queue;
445 int sk_write_pending;
446 __u32 sk_dst_pending_confirm;
447 u32 sk_pacing_status; /* see enum sk_pacing */
449 struct timer_list sk_timer;
452 unsigned long sk_pacing_rate; /* bytes per second */
453 unsigned long sk_max_pacing_rate;
454 struct page_frag sk_frag;
455 netdev_features_t sk_route_caps;
456 netdev_features_t sk_route_nocaps;
457 netdev_features_t sk_route_forced_caps;
459 unsigned int sk_gso_max_size;
464 * Because of non atomicity rules, all
465 * changes are protected by socket lock.
476 unsigned long sk_lingertime;
477 struct proto *sk_prot_creator;
478 rwlock_t sk_callback_lock;
482 u32 sk_max_ack_backlog;
484 #ifdef CONFIG_NET_RX_BUSY_POLL
485 u8 sk_prefer_busy_poll;
486 u16 sk_busy_poll_budget;
488 struct pid *sk_peer_pid;
489 const struct cred *sk_peer_cred;
492 #if BITS_PER_LONG==32
493 seqlock_t sk_stamp_seq;
501 u8 sk_txtime_deadline_mode : 1,
502 sk_txtime_report_errors : 1,
503 sk_txtime_unused : 6;
505 struct socket *sk_socket;
507 #ifdef CONFIG_SECURITY
510 struct sock_cgroup_data sk_cgrp_data;
511 struct mem_cgroup *sk_memcg;
512 void (*sk_state_change)(struct sock *sk);
513 void (*sk_data_ready)(struct sock *sk);
514 void (*sk_write_space)(struct sock *sk);
515 void (*sk_error_report)(struct sock *sk);
516 int (*sk_backlog_rcv)(struct sock *sk,
517 struct sk_buff *skb);
518 #ifdef CONFIG_SOCK_VALIDATE_XMIT
519 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
520 struct net_device *dev,
521 struct sk_buff *skb);
523 void (*sk_destruct)(struct sock *sk);
524 struct sock_reuseport __rcu *sk_reuseport_cb;
525 #ifdef CONFIG_BPF_SYSCALL
526 struct bpf_local_storage __rcu *sk_bpf_storage;
528 struct rcu_head sk_rcu;
533 SK_PACING_NEEDED = 1,
537 /* Pointer stored in sk_user_data might not be suitable for copying
538 * when cloning the socket. For instance, it can point to a reference
539 * counted object. sk_user_data bottom bit is set if pointer must not
542 #define SK_USER_DATA_NOCOPY 1UL
543 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */
544 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF)
547 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
550 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
552 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
555 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
557 #define rcu_dereference_sk_user_data(sk) \
559 void *__tmp = rcu_dereference(__sk_user_data((sk))); \
560 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \
562 #define rcu_assign_sk_user_data(sk, ptr) \
564 uintptr_t __tmp = (uintptr_t)(ptr); \
565 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
566 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \
568 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \
570 uintptr_t __tmp = (uintptr_t)(ptr); \
571 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
572 rcu_assign_pointer(__sk_user_data((sk)), \
573 __tmp | SK_USER_DATA_NOCOPY); \
577 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
578 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
579 * on a socket means that the socket will reuse everybody else's port
580 * without looking at the other's sk_reuse value.
583 #define SK_NO_REUSE 0
584 #define SK_CAN_REUSE 1
585 #define SK_FORCE_REUSE 2
587 int sk_set_peek_off(struct sock *sk, int val);
589 static inline int sk_peek_offset(struct sock *sk, int flags)
591 if (unlikely(flags & MSG_PEEK)) {
592 return READ_ONCE(sk->sk_peek_off);
598 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
600 s32 off = READ_ONCE(sk->sk_peek_off);
602 if (unlikely(off >= 0)) {
603 off = max_t(s32, off - val, 0);
604 WRITE_ONCE(sk->sk_peek_off, off);
608 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
610 sk_peek_offset_bwd(sk, -val);
614 * Hashed lists helper routines
616 static inline struct sock *sk_entry(const struct hlist_node *node)
618 return hlist_entry(node, struct sock, sk_node);
621 static inline struct sock *__sk_head(const struct hlist_head *head)
623 return hlist_entry(head->first, struct sock, sk_node);
626 static inline struct sock *sk_head(const struct hlist_head *head)
628 return hlist_empty(head) ? NULL : __sk_head(head);
631 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
633 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
636 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
638 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
641 static inline struct sock *sk_next(const struct sock *sk)
643 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
646 static inline struct sock *sk_nulls_next(const struct sock *sk)
648 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
649 hlist_nulls_entry(sk->sk_nulls_node.next,
650 struct sock, sk_nulls_node) :
654 static inline bool sk_unhashed(const struct sock *sk)
656 return hlist_unhashed(&sk->sk_node);
659 static inline bool sk_hashed(const struct sock *sk)
661 return !sk_unhashed(sk);
664 static inline void sk_node_init(struct hlist_node *node)
669 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
674 static inline void __sk_del_node(struct sock *sk)
676 __hlist_del(&sk->sk_node);
679 /* NB: equivalent to hlist_del_init_rcu */
680 static inline bool __sk_del_node_init(struct sock *sk)
684 sk_node_init(&sk->sk_node);
690 /* Grab socket reference count. This operation is valid only
691 when sk is ALREADY grabbed f.e. it is found in hash table
692 or a list and the lookup is made under lock preventing hash table
696 static __always_inline void sock_hold(struct sock *sk)
698 refcount_inc(&sk->sk_refcnt);
701 /* Ungrab socket in the context, which assumes that socket refcnt
702 cannot hit zero, f.e. it is true in context of any socketcall.
704 static __always_inline void __sock_put(struct sock *sk)
706 refcount_dec(&sk->sk_refcnt);
709 static inline bool sk_del_node_init(struct sock *sk)
711 bool rc = __sk_del_node_init(sk);
714 /* paranoid for a while -acme */
715 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
720 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
722 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
725 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
731 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
733 bool rc = __sk_nulls_del_node_init_rcu(sk);
736 /* paranoid for a while -acme */
737 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
743 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
745 hlist_add_head(&sk->sk_node, list);
748 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
751 __sk_add_node(sk, list);
754 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
757 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
758 sk->sk_family == AF_INET6)
759 hlist_add_tail_rcu(&sk->sk_node, list);
761 hlist_add_head_rcu(&sk->sk_node, list);
764 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
767 hlist_add_tail_rcu(&sk->sk_node, list);
770 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
772 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
775 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
777 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
780 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
783 __sk_nulls_add_node_rcu(sk, list);
786 static inline void __sk_del_bind_node(struct sock *sk)
788 __hlist_del(&sk->sk_bind_node);
791 static inline void sk_add_bind_node(struct sock *sk,
792 struct hlist_head *list)
794 hlist_add_head(&sk->sk_bind_node, list);
797 #define sk_for_each(__sk, list) \
798 hlist_for_each_entry(__sk, list, sk_node)
799 #define sk_for_each_rcu(__sk, list) \
800 hlist_for_each_entry_rcu(__sk, list, sk_node)
801 #define sk_nulls_for_each(__sk, node, list) \
802 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
803 #define sk_nulls_for_each_rcu(__sk, node, list) \
804 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
805 #define sk_for_each_from(__sk) \
806 hlist_for_each_entry_from(__sk, sk_node)
807 #define sk_nulls_for_each_from(__sk, node) \
808 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
809 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
810 #define sk_for_each_safe(__sk, tmp, list) \
811 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
812 #define sk_for_each_bound(__sk, list) \
813 hlist_for_each_entry(__sk, list, sk_bind_node)
816 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
817 * @tpos: the type * to use as a loop cursor.
818 * @pos: the &struct hlist_node to use as a loop cursor.
819 * @head: the head for your list.
820 * @offset: offset of hlist_node within the struct.
823 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
824 for (pos = rcu_dereference(hlist_first_rcu(head)); \
826 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
827 pos = rcu_dereference(hlist_next_rcu(pos)))
829 static inline struct user_namespace *sk_user_ns(struct sock *sk)
831 /* Careful only use this in a context where these parameters
832 * can not change and must all be valid, such as recvmsg from
835 return sk->sk_socket->file->f_cred->user_ns;
849 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
850 SOCK_DBG, /* %SO_DEBUG setting */
851 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
852 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
853 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
854 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
855 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
856 SOCK_FASYNC, /* fasync() active */
858 SOCK_ZEROCOPY, /* buffers from userspace */
859 SOCK_WIFI_STATUS, /* push wifi status to userspace */
860 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
861 * Will use last 4 bytes of packet sent from
862 * user-space instead.
864 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
865 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
866 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
868 SOCK_XDP, /* XDP is attached */
869 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
872 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
874 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
876 nsk->sk_flags = osk->sk_flags;
879 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
881 __set_bit(flag, &sk->sk_flags);
884 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
886 __clear_bit(flag, &sk->sk_flags);
889 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
893 sock_set_flag(sk, bit);
895 sock_reset_flag(sk, bit);
898 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
900 return test_bit(flag, &sk->sk_flags);
904 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
905 static inline int sk_memalloc_socks(void)
907 return static_branch_unlikely(&memalloc_socks_key);
910 void __receive_sock(struct file *file);
913 static inline int sk_memalloc_socks(void)
918 static inline void __receive_sock(struct file *file)
922 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
924 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
927 static inline void sk_acceptq_removed(struct sock *sk)
929 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
932 static inline void sk_acceptq_added(struct sock *sk)
934 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
937 static inline bool sk_acceptq_is_full(const struct sock *sk)
939 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
943 * Compute minimal free write space needed to queue new packets.
945 static inline int sk_stream_min_wspace(const struct sock *sk)
947 return READ_ONCE(sk->sk_wmem_queued) >> 1;
950 static inline int sk_stream_wspace(const struct sock *sk)
952 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
955 static inline void sk_wmem_queued_add(struct sock *sk, int val)
957 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
960 void sk_stream_write_space(struct sock *sk);
962 /* OOB backlog add */
963 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
965 /* dont let skb dst not refcounted, we are going to leave rcu lock */
968 if (!sk->sk_backlog.tail)
969 WRITE_ONCE(sk->sk_backlog.head, skb);
971 sk->sk_backlog.tail->next = skb;
973 WRITE_ONCE(sk->sk_backlog.tail, skb);
978 * Take into account size of receive queue and backlog queue
979 * Do not take into account this skb truesize,
980 * to allow even a single big packet to come.
982 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
984 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
986 return qsize > limit;
989 /* The per-socket spinlock must be held here. */
990 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
993 if (sk_rcvqueues_full(sk, limit))
997 * If the skb was allocated from pfmemalloc reserves, only
998 * allow SOCK_MEMALLOC sockets to use it as this socket is
999 * helping free memory
1001 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1004 __sk_add_backlog(sk, skb);
1005 sk->sk_backlog.len += skb->truesize;
1009 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1011 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1013 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1014 return __sk_backlog_rcv(sk, skb);
1016 return sk->sk_backlog_rcv(sk, skb);
1019 static inline void sk_incoming_cpu_update(struct sock *sk)
1021 int cpu = raw_smp_processor_id();
1023 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1024 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1027 static inline void sock_rps_record_flow_hash(__u32 hash)
1030 struct rps_sock_flow_table *sock_flow_table;
1033 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1034 rps_record_sock_flow(sock_flow_table, hash);
1039 static inline void sock_rps_record_flow(const struct sock *sk)
1042 if (static_branch_unlikely(&rfs_needed)) {
1043 /* Reading sk->sk_rxhash might incur an expensive cache line
1046 * TCP_ESTABLISHED does cover almost all states where RFS
1047 * might be useful, and is cheaper [1] than testing :
1048 * IPv4: inet_sk(sk)->inet_daddr
1049 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1050 * OR an additional socket flag
1051 * [1] : sk_state and sk_prot are in the same cache line.
1053 if (sk->sk_state == TCP_ESTABLISHED)
1054 sock_rps_record_flow_hash(sk->sk_rxhash);
1059 static inline void sock_rps_save_rxhash(struct sock *sk,
1060 const struct sk_buff *skb)
1063 if (unlikely(sk->sk_rxhash != skb->hash))
1064 sk->sk_rxhash = skb->hash;
1068 static inline void sock_rps_reset_rxhash(struct sock *sk)
1075 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1077 release_sock(__sk); \
1078 __rc = __condition; \
1080 *(__timeo) = wait_woken(__wait, \
1081 TASK_INTERRUPTIBLE, \
1084 sched_annotate_sleep(); \
1086 __rc = __condition; \
1090 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1091 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1092 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1093 int sk_stream_error(struct sock *sk, int flags, int err);
1094 void sk_stream_kill_queues(struct sock *sk);
1095 void sk_set_memalloc(struct sock *sk);
1096 void sk_clear_memalloc(struct sock *sk);
1098 void __sk_flush_backlog(struct sock *sk);
1100 static inline bool sk_flush_backlog(struct sock *sk)
1102 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1103 __sk_flush_backlog(sk);
1109 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1111 struct request_sock_ops;
1112 struct timewait_sock_ops;
1113 struct inet_hashinfo;
1114 struct raw_hashinfo;
1115 struct smc_hashinfo;
1119 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1120 * un-modified. Special care is taken when initializing object to zero.
1122 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1124 if (offsetof(struct sock, sk_node.next) != 0)
1125 memset(sk, 0, offsetof(struct sock, sk_node.next));
1126 memset(&sk->sk_node.pprev, 0,
1127 size - offsetof(struct sock, sk_node.pprev));
1130 /* Networking protocol blocks we attach to sockets.
1131 * socket layer -> transport layer interface
1134 void (*close)(struct sock *sk,
1136 int (*pre_connect)(struct sock *sk,
1137 struct sockaddr *uaddr,
1139 int (*connect)(struct sock *sk,
1140 struct sockaddr *uaddr,
1142 int (*disconnect)(struct sock *sk, int flags);
1144 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1147 int (*ioctl)(struct sock *sk, int cmd,
1149 int (*init)(struct sock *sk);
1150 void (*destroy)(struct sock *sk);
1151 void (*shutdown)(struct sock *sk, int how);
1152 int (*setsockopt)(struct sock *sk, int level,
1153 int optname, sockptr_t optval,
1154 unsigned int optlen);
1155 int (*getsockopt)(struct sock *sk, int level,
1156 int optname, char __user *optval,
1157 int __user *option);
1158 void (*keepalive)(struct sock *sk, int valbool);
1159 #ifdef CONFIG_COMPAT
1160 int (*compat_ioctl)(struct sock *sk,
1161 unsigned int cmd, unsigned long arg);
1163 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1165 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1166 size_t len, int noblock, int flags,
1168 int (*sendpage)(struct sock *sk, struct page *page,
1169 int offset, size_t size, int flags);
1170 int (*bind)(struct sock *sk,
1171 struct sockaddr *addr, int addr_len);
1172 int (*bind_add)(struct sock *sk,
1173 struct sockaddr *addr, int addr_len);
1175 int (*backlog_rcv) (struct sock *sk,
1176 struct sk_buff *skb);
1178 void (*release_cb)(struct sock *sk);
1180 /* Keeping track of sk's, looking them up, and port selection methods. */
1181 int (*hash)(struct sock *sk);
1182 void (*unhash)(struct sock *sk);
1183 void (*rehash)(struct sock *sk);
1184 int (*get_port)(struct sock *sk, unsigned short snum);
1186 /* Keeping track of sockets in use */
1187 #ifdef CONFIG_PROC_FS
1188 unsigned int inuse_idx;
1191 bool (*stream_memory_free)(const struct sock *sk, int wake);
1192 bool (*stream_memory_read)(const struct sock *sk);
1193 /* Memory pressure */
1194 void (*enter_memory_pressure)(struct sock *sk);
1195 void (*leave_memory_pressure)(struct sock *sk);
1196 atomic_long_t *memory_allocated; /* Current allocated memory. */
1197 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1199 * Pressure flag: try to collapse.
1200 * Technical note: it is used by multiple contexts non atomically.
1201 * All the __sk_mem_schedule() is of this nature: accounting
1202 * is strict, actions are advisory and have some latency.
1204 unsigned long *memory_pressure;
1209 u32 sysctl_wmem_offset;
1210 u32 sysctl_rmem_offset;
1215 struct kmem_cache *slab;
1216 unsigned int obj_size;
1217 slab_flags_t slab_flags;
1218 unsigned int useroffset; /* Usercopy region offset */
1219 unsigned int usersize; /* Usercopy region size */
1221 struct percpu_counter *orphan_count;
1223 struct request_sock_ops *rsk_prot;
1224 struct timewait_sock_ops *twsk_prot;
1227 struct inet_hashinfo *hashinfo;
1228 struct udp_table *udp_table;
1229 struct raw_hashinfo *raw_hash;
1230 struct smc_hashinfo *smc_hash;
1233 struct module *owner;
1237 struct list_head node;
1238 #ifdef SOCK_REFCNT_DEBUG
1241 int (*diag_destroy)(struct sock *sk, int err);
1242 } __randomize_layout;
1244 int proto_register(struct proto *prot, int alloc_slab);
1245 void proto_unregister(struct proto *prot);
1246 int sock_load_diag_module(int family, int protocol);
1248 #ifdef SOCK_REFCNT_DEBUG
1249 static inline void sk_refcnt_debug_inc(struct sock *sk)
1251 atomic_inc(&sk->sk_prot->socks);
1254 static inline void sk_refcnt_debug_dec(struct sock *sk)
1256 atomic_dec(&sk->sk_prot->socks);
1257 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1258 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1261 static inline void sk_refcnt_debug_release(const struct sock *sk)
1263 if (refcount_read(&sk->sk_refcnt) != 1)
1264 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1265 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1267 #else /* SOCK_REFCNT_DEBUG */
1268 #define sk_refcnt_debug_inc(sk) do { } while (0)
1269 #define sk_refcnt_debug_dec(sk) do { } while (0)
1270 #define sk_refcnt_debug_release(sk) do { } while (0)
1271 #endif /* SOCK_REFCNT_DEBUG */
1273 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1275 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1277 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1281 return sk->sk_prot->stream_memory_free ?
1282 INDIRECT_CALL_1(sk->sk_prot->stream_memory_free,
1283 tcp_stream_memory_free,
1286 return sk->sk_prot->stream_memory_free ?
1287 sk->sk_prot->stream_memory_free(sk, wake) : true;
1291 static inline bool sk_stream_memory_free(const struct sock *sk)
1293 return __sk_stream_memory_free(sk, 0);
1296 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1298 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1299 __sk_stream_memory_free(sk, wake);
1302 static inline bool sk_stream_is_writeable(const struct sock *sk)
1304 return __sk_stream_is_writeable(sk, 0);
1307 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1308 struct cgroup *ancestor)
1310 #ifdef CONFIG_SOCK_CGROUP_DATA
1311 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1318 static inline bool sk_has_memory_pressure(const struct sock *sk)
1320 return sk->sk_prot->memory_pressure != NULL;
1323 static inline bool sk_under_memory_pressure(const struct sock *sk)
1325 if (!sk->sk_prot->memory_pressure)
1328 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1329 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1332 return !!*sk->sk_prot->memory_pressure;
1336 sk_memory_allocated(const struct sock *sk)
1338 return atomic_long_read(sk->sk_prot->memory_allocated);
1342 sk_memory_allocated_add(struct sock *sk, int amt)
1344 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1348 sk_memory_allocated_sub(struct sock *sk, int amt)
1350 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1353 static inline void sk_sockets_allocated_dec(struct sock *sk)
1355 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1358 static inline void sk_sockets_allocated_inc(struct sock *sk)
1360 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1364 sk_sockets_allocated_read_positive(struct sock *sk)
1366 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1370 proto_sockets_allocated_sum_positive(struct proto *prot)
1372 return percpu_counter_sum_positive(prot->sockets_allocated);
1376 proto_memory_allocated(struct proto *prot)
1378 return atomic_long_read(prot->memory_allocated);
1382 proto_memory_pressure(struct proto *prot)
1384 if (!prot->memory_pressure)
1386 return !!*prot->memory_pressure;
1390 #ifdef CONFIG_PROC_FS
1391 /* Called with local bh disabled */
1392 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1393 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1394 int sock_inuse_get(struct net *net);
1396 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1403 /* With per-bucket locks this operation is not-atomic, so that
1404 * this version is not worse.
1406 static inline int __sk_prot_rehash(struct sock *sk)
1408 sk->sk_prot->unhash(sk);
1409 return sk->sk_prot->hash(sk);
1412 /* About 10 seconds */
1413 #define SOCK_DESTROY_TIME (10*HZ)
1415 /* Sockets 0-1023 can't be bound to unless you are superuser */
1416 #define PROT_SOCK 1024
1418 #define SHUTDOWN_MASK 3
1419 #define RCV_SHUTDOWN 1
1420 #define SEND_SHUTDOWN 2
1422 #define SOCK_SNDBUF_LOCK 1
1423 #define SOCK_RCVBUF_LOCK 2
1424 #define SOCK_BINDADDR_LOCK 4
1425 #define SOCK_BINDPORT_LOCK 8
1427 struct socket_alloc {
1428 struct socket socket;
1429 struct inode vfs_inode;
1432 static inline struct socket *SOCKET_I(struct inode *inode)
1434 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1437 static inline struct inode *SOCK_INODE(struct socket *socket)
1439 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1443 * Functions for memory accounting
1445 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1446 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1447 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1448 void __sk_mem_reclaim(struct sock *sk, int amount);
1450 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1451 * do not necessarily have 16x time more memory than 4KB ones.
1453 #define SK_MEM_QUANTUM 4096
1454 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1455 #define SK_MEM_SEND 0
1456 #define SK_MEM_RECV 1
1458 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1459 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1461 long val = sk->sk_prot->sysctl_mem[index];
1463 #if PAGE_SIZE > SK_MEM_QUANTUM
1464 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1465 #elif PAGE_SIZE < SK_MEM_QUANTUM
1466 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1471 static inline int sk_mem_pages(int amt)
1473 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1476 static inline bool sk_has_account(struct sock *sk)
1478 /* return true if protocol supports memory accounting */
1479 return !!sk->sk_prot->memory_allocated;
1482 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1484 if (!sk_has_account(sk))
1486 return size <= sk->sk_forward_alloc ||
1487 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1491 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1493 if (!sk_has_account(sk))
1495 return size <= sk->sk_forward_alloc ||
1496 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1497 skb_pfmemalloc(skb);
1500 static inline void sk_mem_reclaim(struct sock *sk)
1502 if (!sk_has_account(sk))
1504 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1505 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1508 static inline void sk_mem_reclaim_partial(struct sock *sk)
1510 if (!sk_has_account(sk))
1512 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1513 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1516 static inline void sk_mem_charge(struct sock *sk, int size)
1518 if (!sk_has_account(sk))
1520 sk->sk_forward_alloc -= size;
1523 static inline void sk_mem_uncharge(struct sock *sk, int size)
1525 if (!sk_has_account(sk))
1527 sk->sk_forward_alloc += size;
1529 /* Avoid a possible overflow.
1530 * TCP send queues can make this happen, if sk_mem_reclaim()
1531 * is not called and more than 2 GBytes are released at once.
1533 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1534 * no need to hold that much forward allocation anyway.
1536 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1537 __sk_mem_reclaim(sk, 1 << 20);
1540 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
1541 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1543 sk_wmem_queued_add(sk, -skb->truesize);
1544 sk_mem_uncharge(sk, skb->truesize);
1545 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1546 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1548 skb_zcopy_clear(skb, true);
1549 sk->sk_tx_skb_cache = skb;
1555 static inline void sock_release_ownership(struct sock *sk)
1557 if (sk->sk_lock.owned) {
1558 sk->sk_lock.owned = 0;
1560 /* The sk_lock has mutex_unlock() semantics: */
1561 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1566 * Macro so as to not evaluate some arguments when
1567 * lockdep is not enabled.
1569 * Mark both the sk_lock and the sk_lock.slock as a
1570 * per-address-family lock class.
1572 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1574 sk->sk_lock.owned = 0; \
1575 init_waitqueue_head(&sk->sk_lock.wq); \
1576 spin_lock_init(&(sk)->sk_lock.slock); \
1577 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1578 sizeof((sk)->sk_lock)); \
1579 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1581 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1584 #ifdef CONFIG_LOCKDEP
1585 static inline bool lockdep_sock_is_held(const struct sock *sk)
1587 return lockdep_is_held(&sk->sk_lock) ||
1588 lockdep_is_held(&sk->sk_lock.slock);
1592 void lock_sock_nested(struct sock *sk, int subclass);
1594 static inline void lock_sock(struct sock *sk)
1596 lock_sock_nested(sk, 0);
1599 void __lock_sock(struct sock *sk);
1600 void __release_sock(struct sock *sk);
1601 void release_sock(struct sock *sk);
1603 /* BH context may only use the following locking interface. */
1604 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1605 #define bh_lock_sock_nested(__sk) \
1606 spin_lock_nested(&((__sk)->sk_lock.slock), \
1607 SINGLE_DEPTH_NESTING)
1608 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1610 bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1613 * unlock_sock_fast - complement of lock_sock_fast
1617 * fast unlock socket for user context.
1618 * If slow mode is on, we call regular release_sock()
1620 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1621 __releases(&sk->sk_lock.slock)
1625 __release(&sk->sk_lock.slock);
1627 spin_unlock_bh(&sk->sk_lock.slock);
1631 /* Used by processes to "lock" a socket state, so that
1632 * interrupts and bottom half handlers won't change it
1633 * from under us. It essentially blocks any incoming
1634 * packets, so that we won't get any new data or any
1635 * packets that change the state of the socket.
1637 * While locked, BH processing will add new packets to
1638 * the backlog queue. This queue is processed by the
1639 * owner of the socket lock right before it is released.
1641 * Since ~2.3.5 it is also exclusive sleep lock serializing
1642 * accesses from user process context.
1645 static inline void sock_owned_by_me(const struct sock *sk)
1647 #ifdef CONFIG_LOCKDEP
1648 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1652 static inline bool sock_owned_by_user(const struct sock *sk)
1654 sock_owned_by_me(sk);
1655 return sk->sk_lock.owned;
1658 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1660 return sk->sk_lock.owned;
1663 /* no reclassification while locks are held */
1664 static inline bool sock_allow_reclassification(const struct sock *csk)
1666 struct sock *sk = (struct sock *)csk;
1668 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1671 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1672 struct proto *prot, int kern);
1673 void sk_free(struct sock *sk);
1674 void sk_destruct(struct sock *sk);
1675 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1676 void sk_free_unlock_clone(struct sock *sk);
1678 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1680 void __sock_wfree(struct sk_buff *skb);
1681 void sock_wfree(struct sk_buff *skb);
1682 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1684 void skb_orphan_partial(struct sk_buff *skb);
1685 void sock_rfree(struct sk_buff *skb);
1686 void sock_efree(struct sk_buff *skb);
1688 void sock_edemux(struct sk_buff *skb);
1689 void sock_pfree(struct sk_buff *skb);
1691 #define sock_edemux sock_efree
1694 int sock_setsockopt(struct socket *sock, int level, int op,
1695 sockptr_t optval, unsigned int optlen);
1697 int sock_getsockopt(struct socket *sock, int level, int op,
1698 char __user *optval, int __user *optlen);
1699 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1700 bool timeval, bool time32);
1701 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1702 int noblock, int *errcode);
1703 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1704 unsigned long data_len, int noblock,
1705 int *errcode, int max_page_order);
1706 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1707 void sock_kfree_s(struct sock *sk, void *mem, int size);
1708 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1709 void sk_send_sigurg(struct sock *sk);
1711 struct sockcm_cookie {
1717 static inline void sockcm_init(struct sockcm_cookie *sockc,
1718 const struct sock *sk)
1720 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1723 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1724 struct sockcm_cookie *sockc);
1725 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1726 struct sockcm_cookie *sockc);
1729 * Functions to fill in entries in struct proto_ops when a protocol
1730 * does not implement a particular function.
1732 int sock_no_bind(struct socket *, struct sockaddr *, int);
1733 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1734 int sock_no_socketpair(struct socket *, struct socket *);
1735 int sock_no_accept(struct socket *, struct socket *, int, bool);
1736 int sock_no_getname(struct socket *, struct sockaddr *, int);
1737 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1738 int sock_no_listen(struct socket *, int);
1739 int sock_no_shutdown(struct socket *, int);
1740 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1741 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1742 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1743 int sock_no_mmap(struct file *file, struct socket *sock,
1744 struct vm_area_struct *vma);
1745 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1746 size_t size, int flags);
1747 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1748 int offset, size_t size, int flags);
1751 * Functions to fill in entries in struct proto_ops when a protocol
1752 * uses the inet style.
1754 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1755 char __user *optval, int __user *optlen);
1756 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1758 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1759 sockptr_t optval, unsigned int optlen);
1761 void sk_common_release(struct sock *sk);
1764 * Default socket callbacks and setup code
1767 /* Initialise core socket variables */
1768 void sock_init_data(struct socket *sock, struct sock *sk);
1771 * Socket reference counting postulates.
1773 * * Each user of socket SHOULD hold a reference count.
1774 * * Each access point to socket (an hash table bucket, reference from a list,
1775 * running timer, skb in flight MUST hold a reference count.
1776 * * When reference count hits 0, it means it will never increase back.
1777 * * When reference count hits 0, it means that no references from
1778 * outside exist to this socket and current process on current CPU
1779 * is last user and may/should destroy this socket.
1780 * * sk_free is called from any context: process, BH, IRQ. When
1781 * it is called, socket has no references from outside -> sk_free
1782 * may release descendant resources allocated by the socket, but
1783 * to the time when it is called, socket is NOT referenced by any
1784 * hash tables, lists etc.
1785 * * Packets, delivered from outside (from network or from another process)
1786 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1787 * when they sit in queue. Otherwise, packets will leak to hole, when
1788 * socket is looked up by one cpu and unhasing is made by another CPU.
1789 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1790 * (leak to backlog). Packet socket does all the processing inside
1791 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1792 * use separate SMP lock, so that they are prone too.
1795 /* Ungrab socket and destroy it, if it was the last reference. */
1796 static inline void sock_put(struct sock *sk)
1798 if (refcount_dec_and_test(&sk->sk_refcnt))
1801 /* Generic version of sock_put(), dealing with all sockets
1802 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1804 void sock_gen_put(struct sock *sk);
1806 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1807 unsigned int trim_cap, bool refcounted);
1808 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1811 return __sk_receive_skb(sk, skb, nested, 1, true);
1814 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1816 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1817 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1819 sk->sk_tx_queue_mapping = tx_queue;
1822 #define NO_QUEUE_MAPPING USHRT_MAX
1824 static inline void sk_tx_queue_clear(struct sock *sk)
1826 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1829 static inline int sk_tx_queue_get(const struct sock *sk)
1831 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1832 return sk->sk_tx_queue_mapping;
1837 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1840 if (skb_rx_queue_recorded(skb)) {
1841 u16 rx_queue = skb_get_rx_queue(skb);
1843 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1846 sk->sk_rx_queue_mapping = rx_queue;
1851 static inline void sk_rx_queue_clear(struct sock *sk)
1854 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1859 static inline int sk_rx_queue_get(const struct sock *sk)
1861 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1862 return sk->sk_rx_queue_mapping;
1868 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1870 sk->sk_socket = sock;
1873 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1875 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1876 return &rcu_dereference_raw(sk->sk_wq)->wait;
1878 /* Detach socket from process context.
1879 * Announce socket dead, detach it from wait queue and inode.
1880 * Note that parent inode held reference count on this struct sock,
1881 * we do not release it in this function, because protocol
1882 * probably wants some additional cleanups or even continuing
1883 * to work with this socket (TCP).
1885 static inline void sock_orphan(struct sock *sk)
1887 write_lock_bh(&sk->sk_callback_lock);
1888 sock_set_flag(sk, SOCK_DEAD);
1889 sk_set_socket(sk, NULL);
1891 write_unlock_bh(&sk->sk_callback_lock);
1894 static inline void sock_graft(struct sock *sk, struct socket *parent)
1896 WARN_ON(parent->sk);
1897 write_lock_bh(&sk->sk_callback_lock);
1898 rcu_assign_pointer(sk->sk_wq, &parent->wq);
1900 sk_set_socket(sk, parent);
1901 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1902 security_sock_graft(sk, parent);
1903 write_unlock_bh(&sk->sk_callback_lock);
1906 kuid_t sock_i_uid(struct sock *sk);
1907 unsigned long sock_i_ino(struct sock *sk);
1909 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1911 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1914 static inline u32 net_tx_rndhash(void)
1916 u32 v = prandom_u32();
1921 static inline void sk_set_txhash(struct sock *sk)
1923 sk->sk_txhash = net_tx_rndhash();
1926 static inline void sk_rethink_txhash(struct sock *sk)
1932 static inline struct dst_entry *
1933 __sk_dst_get(struct sock *sk)
1935 return rcu_dereference_check(sk->sk_dst_cache,
1936 lockdep_sock_is_held(sk));
1939 static inline struct dst_entry *
1940 sk_dst_get(struct sock *sk)
1942 struct dst_entry *dst;
1945 dst = rcu_dereference(sk->sk_dst_cache);
1946 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1952 static inline void dst_negative_advice(struct sock *sk)
1954 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1956 sk_rethink_txhash(sk);
1958 if (dst && dst->ops->negative_advice) {
1959 ndst = dst->ops->negative_advice(dst);
1962 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1963 sk_tx_queue_clear(sk);
1964 sk->sk_dst_pending_confirm = 0;
1970 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1972 struct dst_entry *old_dst;
1974 sk_tx_queue_clear(sk);
1975 sk->sk_dst_pending_confirm = 0;
1976 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1977 lockdep_sock_is_held(sk));
1978 rcu_assign_pointer(sk->sk_dst_cache, dst);
1979 dst_release(old_dst);
1983 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1985 struct dst_entry *old_dst;
1987 sk_tx_queue_clear(sk);
1988 sk->sk_dst_pending_confirm = 0;
1989 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1990 dst_release(old_dst);
1994 __sk_dst_reset(struct sock *sk)
1996 __sk_dst_set(sk, NULL);
2000 sk_dst_reset(struct sock *sk)
2002 sk_dst_set(sk, NULL);
2005 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2007 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2009 static inline void sk_dst_confirm(struct sock *sk)
2011 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2012 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2015 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2017 if (skb_get_dst_pending_confirm(skb)) {
2018 struct sock *sk = skb->sk;
2019 unsigned long now = jiffies;
2021 /* avoid dirtying neighbour */
2022 if (READ_ONCE(n->confirmed) != now)
2023 WRITE_ONCE(n->confirmed, now);
2024 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2025 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2029 bool sk_mc_loop(struct sock *sk);
2031 static inline bool sk_can_gso(const struct sock *sk)
2033 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2036 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2038 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2040 sk->sk_route_nocaps |= flags;
2041 sk->sk_route_caps &= ~flags;
2044 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2045 struct iov_iter *from, char *to,
2046 int copy, int offset)
2048 if (skb->ip_summed == CHECKSUM_NONE) {
2050 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2052 skb->csum = csum_block_add(skb->csum, csum, offset);
2053 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2054 if (!copy_from_iter_full_nocache(to, copy, from))
2056 } else if (!copy_from_iter_full(to, copy, from))
2062 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2063 struct iov_iter *from, int copy)
2065 int err, offset = skb->len;
2067 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2070 __skb_trim(skb, offset);
2075 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2076 struct sk_buff *skb,
2082 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2088 skb->data_len += copy;
2089 skb->truesize += copy;
2090 sk_wmem_queued_add(sk, copy);
2091 sk_mem_charge(sk, copy);
2096 * sk_wmem_alloc_get - returns write allocations
2099 * Return: sk_wmem_alloc minus initial offset of one
2101 static inline int sk_wmem_alloc_get(const struct sock *sk)
2103 return refcount_read(&sk->sk_wmem_alloc) - 1;
2107 * sk_rmem_alloc_get - returns read allocations
2110 * Return: sk_rmem_alloc
2112 static inline int sk_rmem_alloc_get(const struct sock *sk)
2114 return atomic_read(&sk->sk_rmem_alloc);
2118 * sk_has_allocations - check if allocations are outstanding
2121 * Return: true if socket has write or read allocations
2123 static inline bool sk_has_allocations(const struct sock *sk)
2125 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2129 * skwq_has_sleeper - check if there are any waiting processes
2130 * @wq: struct socket_wq
2132 * Return: true if socket_wq has waiting processes
2134 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2135 * barrier call. They were added due to the race found within the tcp code.
2137 * Consider following tcp code paths::
2140 * sys_select receive packet
2142 * __add_wait_queue update tp->rcv_nxt
2144 * tp->rcv_nxt check sock_def_readable
2146 * schedule rcu_read_lock();
2147 * wq = rcu_dereference(sk->sk_wq);
2148 * if (wq && waitqueue_active(&wq->wait))
2149 * wake_up_interruptible(&wq->wait)
2153 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2154 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2155 * could then endup calling schedule and sleep forever if there are no more
2156 * data on the socket.
2159 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2161 return wq && wq_has_sleeper(&wq->wait);
2165 * sock_poll_wait - place memory barrier behind the poll_wait call.
2167 * @sock: socket to wait on
2170 * See the comments in the wq_has_sleeper function.
2172 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2175 if (!poll_does_not_wait(p)) {
2176 poll_wait(filp, &sock->wq.wait, p);
2177 /* We need to be sure we are in sync with the
2178 * socket flags modification.
2180 * This memory barrier is paired in the wq_has_sleeper.
2186 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2188 if (sk->sk_txhash) {
2190 skb->hash = sk->sk_txhash;
2194 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2197 * Queue a received datagram if it will fit. Stream and sequenced
2198 * protocols can't normally use this as they need to fit buffers in
2199 * and play with them.
2201 * Inlined as it's very short and called for pretty much every
2202 * packet ever received.
2204 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2208 skb->destructor = sock_rfree;
2209 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2210 sk_mem_charge(sk, skb->truesize);
2213 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2214 unsigned long expires);
2216 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2218 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2220 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2221 struct sk_buff *skb, unsigned int flags,
2222 void (*destructor)(struct sock *sk,
2223 struct sk_buff *skb));
2224 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2225 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2227 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2228 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2231 * Recover an error report and clear atomically
2234 static inline int sock_error(struct sock *sk)
2237 if (likely(!sk->sk_err))
2239 err = xchg(&sk->sk_err, 0);
2243 static inline unsigned long sock_wspace(struct sock *sk)
2247 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2248 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2256 * We use sk->sk_wq_raw, from contexts knowing this
2257 * pointer is not NULL and cannot disappear/change.
2259 static inline void sk_set_bit(int nr, struct sock *sk)
2261 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2262 !sock_flag(sk, SOCK_FASYNC))
2265 set_bit(nr, &sk->sk_wq_raw->flags);
2268 static inline void sk_clear_bit(int nr, struct sock *sk)
2270 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2271 !sock_flag(sk, SOCK_FASYNC))
2274 clear_bit(nr, &sk->sk_wq_raw->flags);
2277 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2279 if (sock_flag(sk, SOCK_FASYNC)) {
2281 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2286 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2287 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2288 * Note: for send buffers, TCP works better if we can build two skbs at
2291 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2293 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2294 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2296 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2300 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2303 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2305 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2308 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2309 bool force_schedule);
2312 * sk_page_frag - return an appropriate page_frag
2315 * Use the per task page_frag instead of the per socket one for
2316 * optimization when we know that we're in the normal context and owns
2317 * everything that's associated with %current.
2319 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2320 * inside other socket operations and end up recursing into sk_page_frag()
2321 * while it's already in use.
2323 * Return: a per task page_frag if context allows that,
2324 * otherwise a per socket one.
2326 static inline struct page_frag *sk_page_frag(struct sock *sk)
2328 if (gfpflags_normal_context(sk->sk_allocation))
2329 return ¤t->task_frag;
2331 return &sk->sk_frag;
2334 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2337 * Default write policy as shown to user space via poll/select/SIGIO
2339 static inline bool sock_writeable(const struct sock *sk)
2341 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2344 static inline gfp_t gfp_any(void)
2346 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2349 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2351 return noblock ? 0 : sk->sk_rcvtimeo;
2354 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2356 return noblock ? 0 : sk->sk_sndtimeo;
2359 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2361 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2366 /* Alas, with timeout socket operations are not restartable.
2367 * Compare this to poll().
2369 static inline int sock_intr_errno(long timeo)
2371 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2374 struct sock_skb_cb {
2378 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2379 * using skb->cb[] would keep using it directly and utilize its
2380 * alignement guarantee.
2382 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2383 sizeof(struct sock_skb_cb)))
2385 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2386 SOCK_SKB_CB_OFFSET))
2388 #define sock_skb_cb_check_size(size) \
2389 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2392 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2394 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2395 atomic_read(&sk->sk_drops) : 0;
2398 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2400 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2402 atomic_add(segs, &sk->sk_drops);
2405 static inline ktime_t sock_read_timestamp(struct sock *sk)
2407 #if BITS_PER_LONG==32
2412 seq = read_seqbegin(&sk->sk_stamp_seq);
2414 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2418 return READ_ONCE(sk->sk_stamp);
2422 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2424 #if BITS_PER_LONG==32
2425 write_seqlock(&sk->sk_stamp_seq);
2427 write_sequnlock(&sk->sk_stamp_seq);
2429 WRITE_ONCE(sk->sk_stamp, kt);
2433 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2434 struct sk_buff *skb);
2435 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2436 struct sk_buff *skb);
2439 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2441 ktime_t kt = skb->tstamp;
2442 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2445 * generate control messages if
2446 * - receive time stamping in software requested
2447 * - software time stamp available and wanted
2448 * - hardware time stamps available and wanted
2450 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2451 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2452 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2453 (hwtstamps->hwtstamp &&
2454 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2455 __sock_recv_timestamp(msg, sk, skb);
2457 sock_write_timestamp(sk, kt);
2459 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2460 __sock_recv_wifi_status(msg, sk, skb);
2463 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2464 struct sk_buff *skb);
2466 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2467 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2468 struct sk_buff *skb)
2470 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2471 (1UL << SOCK_RCVTSTAMP))
2472 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2473 SOF_TIMESTAMPING_RAW_HARDWARE)
2475 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2476 __sock_recv_ts_and_drops(msg, sk, skb);
2477 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2478 sock_write_timestamp(sk, skb->tstamp);
2479 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2480 sock_write_timestamp(sk, 0);
2483 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2486 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2487 * @sk: socket sending this packet
2488 * @tsflags: timestamping flags to use
2489 * @tx_flags: completed with instructions for time stamping
2490 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2492 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2494 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2495 __u8 *tx_flags, __u32 *tskey)
2497 if (unlikely(tsflags)) {
2498 __sock_tx_timestamp(tsflags, tx_flags);
2499 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2500 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2501 *tskey = sk->sk_tskey++;
2503 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2504 *tx_flags |= SKBTX_WIFI_STATUS;
2507 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2510 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2513 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2515 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2516 &skb_shinfo(skb)->tskey);
2519 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2521 * sk_eat_skb - Release a skb if it is no longer needed
2522 * @sk: socket to eat this skb from
2523 * @skb: socket buffer to eat
2525 * This routine must be called with interrupts disabled or with the socket
2526 * locked so that the sk_buff queue operation is ok.
2528 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2530 __skb_unlink(skb, &sk->sk_receive_queue);
2531 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2532 !sk->sk_rx_skb_cache) {
2533 sk->sk_rx_skb_cache = skb;
2541 struct net *sock_net(const struct sock *sk)
2543 return read_pnet(&sk->sk_net);
2547 void sock_net_set(struct sock *sk, struct net *net)
2549 write_pnet(&sk->sk_net, net);
2553 skb_sk_is_prefetched(struct sk_buff *skb)
2556 return skb->destructor == sock_pfree;
2559 #endif /* CONFIG_INET */
2562 /* This helper checks if a socket is a full socket,
2563 * ie _not_ a timewait or request socket.
2565 static inline bool sk_fullsock(const struct sock *sk)
2567 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2571 sk_is_refcounted(struct sock *sk)
2573 /* Only full sockets have sk->sk_flags. */
2574 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2578 * skb_steal_sock - steal a socket from an sk_buff
2579 * @skb: sk_buff to steal the socket from
2580 * @refcounted: is set to true if the socket is reference-counted
2582 static inline struct sock *
2583 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2586 struct sock *sk = skb->sk;
2589 if (skb_sk_is_prefetched(skb))
2590 *refcounted = sk_is_refcounted(sk);
2591 skb->destructor = NULL;
2595 *refcounted = false;
2599 /* Checks if this SKB belongs to an HW offloaded socket
2600 * and whether any SW fallbacks are required based on dev.
2601 * Check decrypted mark in case skb_orphan() cleared socket.
2603 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2604 struct net_device *dev)
2606 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2607 struct sock *sk = skb->sk;
2609 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2610 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2611 #ifdef CONFIG_TLS_DEVICE
2612 } else if (unlikely(skb->decrypted)) {
2613 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2623 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2624 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2626 static inline bool sk_listener(const struct sock *sk)
2628 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2631 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2632 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2635 bool sk_ns_capable(const struct sock *sk,
2636 struct user_namespace *user_ns, int cap);
2637 bool sk_capable(const struct sock *sk, int cap);
2638 bool sk_net_capable(const struct sock *sk, int cap);
2640 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2642 /* Take into consideration the size of the struct sk_buff overhead in the
2643 * determination of these values, since that is non-constant across
2644 * platforms. This makes socket queueing behavior and performance
2645 * not depend upon such differences.
2647 #define _SK_MEM_PACKETS 256
2648 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2649 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2650 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2652 extern __u32 sysctl_wmem_max;
2653 extern __u32 sysctl_rmem_max;
2655 extern int sysctl_tstamp_allow_data;
2656 extern int sysctl_optmem_max;
2658 extern __u32 sysctl_wmem_default;
2659 extern __u32 sysctl_rmem_default;
2661 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2663 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2665 /* Does this proto have per netns sysctl_wmem ? */
2666 if (proto->sysctl_wmem_offset)
2667 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2669 return *proto->sysctl_wmem;
2672 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2674 /* Does this proto have per netns sysctl_rmem ? */
2675 if (proto->sysctl_rmem_offset)
2676 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2678 return *proto->sysctl_rmem;
2681 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2682 * Some wifi drivers need to tweak it to get more chunks.
2683 * They can use this helper from their ndo_start_xmit()
2685 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2687 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2689 WRITE_ONCE(sk->sk_pacing_shift, val);
2692 /* if a socket is bound to a device, check that the given device
2693 * index is either the same or that the socket is bound to an L3
2694 * master device and the given device index is also enslaved to
2697 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2701 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2704 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2705 if (mdif && mdif == sk->sk_bound_dev_if)
2711 void sock_def_readable(struct sock *sk);
2713 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2714 void sock_enable_timestamps(struct sock *sk);
2715 void sock_no_linger(struct sock *sk);
2716 void sock_set_keepalive(struct sock *sk);
2717 void sock_set_priority(struct sock *sk, u32 priority);
2718 void sock_set_rcvbuf(struct sock *sk, int val);
2719 void sock_set_mark(struct sock *sk, u32 val);
2720 void sock_set_reuseaddr(struct sock *sk);
2721 void sock_set_reuseport(struct sock *sk);
2722 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2724 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2726 #endif /* _SOCK_H */