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
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/rculist_nulls.h>
60 #include <linux/poll.h>
61 #include <linux/sockptr.h>
62 #include <linux/indirect_call_wrapper.h>
63 #include <linux/atomic.h>
64 #include <linux/refcount.h>
65 #include <linux/llist.h>
67 #include <net/checksum.h>
68 #include <net/tcp_states.h>
69 #include <linux/net_tstamp.h>
70 #include <net/l3mdev.h>
71 #include <uapi/linux/socket.h>
74 * This structure really needs to be cleaned up.
75 * Most of it is for TCP, and not used by any of
76 * the other protocols.
79 /* Define this to get the SOCK_DBG debugging facility. */
80 #define SOCK_DEBUGGING
82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
83 printk(KERN_DEBUG msg); } while (0)
85 /* Validate arguments and do nothing */
86 static inline __printf(2, 3)
87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
92 /* This is the per-socket lock. The spinlock provides a synchronization
93 * between user contexts and software interrupt processing, whereas the
94 * mini-semaphore synchronizes multiple users amongst themselves.
101 * We express the mutex-alike socket_lock semantics
102 * to the lock validator by explicitly managing
103 * the slock as a lock variant (in addition to
106 #ifdef CONFIG_DEBUG_LOCK_ALLOC
107 struct lockdep_map dep_map;
115 typedef __u32 __bitwise __portpair;
116 typedef __u64 __bitwise __addrpair;
119 * struct sock_common - minimal network layer representation of sockets
120 * @skc_daddr: Foreign IPv4 addr
121 * @skc_rcv_saddr: Bound local IPv4 addr
122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
123 * @skc_hash: hash value used with various protocol lookup tables
124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
125 * @skc_dport: placeholder for inet_dport/tw_dport
126 * @skc_num: placeholder for inet_num/tw_num
127 * @skc_portpair: __u32 union of @skc_dport & @skc_num
128 * @skc_family: network address family
129 * @skc_state: Connection state
130 * @skc_reuse: %SO_REUSEADDR setting
131 * @skc_reuseport: %SO_REUSEPORT setting
132 * @skc_ipv6only: socket is IPV6 only
133 * @skc_net_refcnt: socket is using net ref counting
134 * @skc_bound_dev_if: bound device index if != 0
135 * @skc_bind_node: bind hash linkage for various protocol lookup tables
136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137 * @skc_prot: protocol handlers inside a network family
138 * @skc_net: reference to the network namespace of this socket
139 * @skc_v6_daddr: IPV6 destination address
140 * @skc_v6_rcv_saddr: IPV6 source address
141 * @skc_cookie: socket's cookie value
142 * @skc_node: main hash linkage for various protocol lookup tables
143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
144 * @skc_tx_queue_mapping: tx queue number for this connection
145 * @skc_rx_queue_mapping: rx queue number for this connection
146 * @skc_flags: place holder for sk_flags
147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
149 * @skc_listener: connection request listener socket (aka rsk_listener)
150 * [union with @skc_flags]
151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
152 * [union with @skc_flags]
153 * @skc_incoming_cpu: record/match cpu processing incoming packets
154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
155 * [union with @skc_incoming_cpu]
156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
157 * [union with @skc_incoming_cpu]
158 * @skc_refcnt: reference count
160 * This is the minimal network layer representation of sockets, the header
161 * for struct sock and struct inet_timewait_sock.
164 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
165 * address on 64bit arches : cf INET_MATCH()
168 __addrpair skc_addrpair;
171 __be32 skc_rcv_saddr;
175 unsigned int skc_hash;
176 __u16 skc_u16hashes[2];
178 /* skc_dport && skc_num must be grouped as well */
180 __portpair skc_portpair;
187 unsigned short skc_family;
188 volatile unsigned char skc_state;
189 unsigned char skc_reuse:4;
190 unsigned char skc_reuseport:1;
191 unsigned char skc_ipv6only:1;
192 unsigned char skc_net_refcnt:1;
193 int skc_bound_dev_if;
195 struct hlist_node skc_bind_node;
196 struct hlist_node skc_portaddr_node;
198 struct proto *skc_prot;
199 possible_net_t skc_net;
201 #if IS_ENABLED(CONFIG_IPV6)
202 struct in6_addr skc_v6_daddr;
203 struct in6_addr skc_v6_rcv_saddr;
206 atomic64_t skc_cookie;
208 /* following fields are padding to force
209 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
210 * assuming IPV6 is enabled. We use this padding differently
211 * for different kind of 'sockets'
214 unsigned long skc_flags;
215 struct sock *skc_listener; /* request_sock */
216 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
219 * fields between dontcopy_begin/dontcopy_end
220 * are not copied in sock_copy()
223 int skc_dontcopy_begin[0];
226 struct hlist_node skc_node;
227 struct hlist_nulls_node skc_nulls_node;
229 unsigned short skc_tx_queue_mapping;
230 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
231 unsigned short skc_rx_queue_mapping;
234 int skc_incoming_cpu;
236 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
239 refcount_t skc_refcnt;
241 int skc_dontcopy_end[0];
244 u32 skc_window_clamp;
245 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
250 struct bpf_local_storage;
254 * struct sock - network layer representation of sockets
255 * @__sk_common: shared layout with inet_timewait_sock
256 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
257 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
258 * @sk_lock: synchronizer
259 * @sk_kern_sock: True if sock is using kernel lock classes
260 * @sk_rcvbuf: size of receive buffer in bytes
261 * @sk_wq: sock wait queue and async head
262 * @sk_rx_dst: receive input route used by early demux
263 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
264 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
265 * @sk_dst_cache: destination cache
266 * @sk_dst_pending_confirm: need to confirm neighbour
267 * @sk_policy: flow policy
268 * @sk_receive_queue: incoming packets
269 * @sk_wmem_alloc: transmit queue bytes committed
270 * @sk_tsq_flags: TCP Small Queues flags
271 * @sk_write_queue: Packet sending queue
272 * @sk_omem_alloc: "o" is "option" or "other"
273 * @sk_wmem_queued: persistent queue size
274 * @sk_forward_alloc: space allocated forward
275 * @sk_reserved_mem: space reserved and non-reclaimable for the socket
276 * @sk_napi_id: id of the last napi context to receive data for sk
277 * @sk_ll_usec: usecs to busypoll when there is no data
278 * @sk_allocation: allocation mode
279 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
280 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
281 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
282 * @sk_sndbuf: size of send buffer in bytes
283 * @__sk_flags_offset: empty field used to determine location of bitfield
284 * @sk_padding: unused element for alignment
285 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
286 * @sk_no_check_rx: allow zero checksum in RX packets
287 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
288 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
289 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
290 * @sk_gso_max_size: Maximum GSO segment size to build
291 * @sk_gso_max_segs: Maximum number of GSO segments
292 * @sk_pacing_shift: scaling factor for TCP Small Queues
293 * @sk_lingertime: %SO_LINGER l_linger setting
294 * @sk_backlog: always used with the per-socket spinlock held
295 * @defer_list: head of llist storing skbs to be freed
296 * @sk_callback_lock: used with the callbacks in the end of this struct
297 * @sk_error_queue: rarely used
298 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
299 * IPV6_ADDRFORM for instance)
300 * @sk_err: last error
301 * @sk_err_soft: errors that don't cause failure but are the cause of a
302 * persistent failure not just 'timed out'
303 * @sk_drops: raw/udp drops counter
304 * @sk_ack_backlog: current listen backlog
305 * @sk_max_ack_backlog: listen backlog set in listen()
306 * @sk_uid: user id of owner
307 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
308 * @sk_busy_poll_budget: napi processing budget when busypolling
309 * @sk_priority: %SO_PRIORITY setting
310 * @sk_type: socket type (%SOCK_STREAM, etc)
311 * @sk_protocol: which protocol this socket belongs in this network family
312 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
313 * @sk_peer_pid: &struct pid for this socket's peer
314 * @sk_peer_cred: %SO_PEERCRED setting
315 * @sk_rcvlowat: %SO_RCVLOWAT setting
316 * @sk_rcvtimeo: %SO_RCVTIMEO setting
317 * @sk_sndtimeo: %SO_SNDTIMEO setting
318 * @sk_txhash: computed flow hash for use on transmit
319 * @sk_txrehash: enable TX hash rethink
320 * @sk_filter: socket filtering instructions
321 * @sk_timer: sock cleanup timer
322 * @sk_stamp: time stamp of last packet received
323 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
324 * @sk_tsflags: SO_TIMESTAMPING flags
325 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
327 * @sk_tskey: counter to disambiguate concurrent tstamp requests
328 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
329 * @sk_socket: Identd and reporting IO signals
330 * @sk_user_data: RPC layer private data
331 * @sk_frag: cached page frag
332 * @sk_peek_off: current peek_offset value
333 * @sk_send_head: front of stuff to transmit
334 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
335 * @sk_security: used by security modules
336 * @sk_mark: generic packet mark
337 * @sk_cgrp_data: cgroup data for this cgroup
338 * @sk_memcg: this socket's memory cgroup association
339 * @sk_write_pending: a write to stream socket waits to start
340 * @sk_state_change: callback to indicate change in the state of the sock
341 * @sk_data_ready: callback to indicate there is data to be processed
342 * @sk_write_space: callback to indicate there is bf sending space available
343 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
344 * @sk_backlog_rcv: callback to process the backlog
345 * @sk_validate_xmit_skb: ptr to an optional validate function
346 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
347 * @sk_reuseport_cb: reuseport group container
348 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
349 * @sk_rcu: used during RCU grace period
350 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
351 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
352 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
353 * @sk_txtime_unused: unused txtime flags
354 * @ns_tracker: tracker for netns reference
358 * Now struct inet_timewait_sock also uses sock_common, so please just
359 * don't add nothing before this first member (__sk_common) --acme
361 struct sock_common __sk_common;
362 #define sk_node __sk_common.skc_node
363 #define sk_nulls_node __sk_common.skc_nulls_node
364 #define sk_refcnt __sk_common.skc_refcnt
365 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
366 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
367 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
370 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
371 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
372 #define sk_hash __sk_common.skc_hash
373 #define sk_portpair __sk_common.skc_portpair
374 #define sk_num __sk_common.skc_num
375 #define sk_dport __sk_common.skc_dport
376 #define sk_addrpair __sk_common.skc_addrpair
377 #define sk_daddr __sk_common.skc_daddr
378 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
379 #define sk_family __sk_common.skc_family
380 #define sk_state __sk_common.skc_state
381 #define sk_reuse __sk_common.skc_reuse
382 #define sk_reuseport __sk_common.skc_reuseport
383 #define sk_ipv6only __sk_common.skc_ipv6only
384 #define sk_net_refcnt __sk_common.skc_net_refcnt
385 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
386 #define sk_bind_node __sk_common.skc_bind_node
387 #define sk_prot __sk_common.skc_prot
388 #define sk_net __sk_common.skc_net
389 #define sk_v6_daddr __sk_common.skc_v6_daddr
390 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
391 #define sk_cookie __sk_common.skc_cookie
392 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
393 #define sk_flags __sk_common.skc_flags
394 #define sk_rxhash __sk_common.skc_rxhash
396 /* early demux fields */
397 struct dst_entry __rcu *sk_rx_dst;
398 int sk_rx_dst_ifindex;
399 u32 sk_rx_dst_cookie;
401 socket_lock_t sk_lock;
404 struct sk_buff_head sk_error_queue;
405 struct sk_buff_head sk_receive_queue;
407 * The backlog queue is special, it is always used with
408 * the per-socket spinlock held and requires low latency
409 * access. Therefore we special case it's implementation.
410 * Note : rmem_alloc is in this structure to fill a hole
411 * on 64bit arches, not because its logically part of
417 struct sk_buff *head;
418 struct sk_buff *tail;
420 struct llist_head defer_list;
422 #define sk_rmem_alloc sk_backlog.rmem_alloc
424 int sk_forward_alloc;
426 #ifdef CONFIG_NET_RX_BUSY_POLL
427 unsigned int sk_ll_usec;
428 /* ===== mostly read cache line ===== */
429 unsigned int sk_napi_id;
433 struct sk_filter __rcu *sk_filter;
435 struct socket_wq __rcu *sk_wq;
437 struct socket_wq *sk_wq_raw;
441 struct xfrm_policy __rcu *sk_policy[2];
444 struct dst_entry __rcu *sk_dst_cache;
445 atomic_t sk_omem_alloc;
448 /* ===== cache line for TX ===== */
450 refcount_t sk_wmem_alloc;
451 unsigned long sk_tsq_flags;
453 struct sk_buff *sk_send_head;
454 struct rb_root tcp_rtx_queue;
456 struct sk_buff_head sk_write_queue;
458 int sk_write_pending;
459 __u32 sk_dst_pending_confirm;
460 u32 sk_pacing_status; /* see enum sk_pacing */
462 struct timer_list sk_timer;
465 unsigned long sk_pacing_rate; /* bytes per second */
466 unsigned long sk_max_pacing_rate;
467 struct page_frag sk_frag;
468 netdev_features_t sk_route_caps;
470 unsigned int sk_gso_max_size;
475 * Because of non atomicity rules, all
476 * changes are protected by socket lock.
478 u8 sk_gso_disabled : 1,
487 unsigned long sk_lingertime;
488 struct proto *sk_prot_creator;
489 rwlock_t sk_callback_lock;
493 u32 sk_max_ack_backlog;
496 #ifdef CONFIG_NET_RX_BUSY_POLL
497 u8 sk_prefer_busy_poll;
498 u16 sk_busy_poll_budget;
500 spinlock_t sk_peer_lock;
502 struct pid *sk_peer_pid;
503 const struct cred *sk_peer_cred;
507 #if BITS_PER_LONG==32
508 seqlock_t sk_stamp_seq;
516 u8 sk_txtime_deadline_mode : 1,
517 sk_txtime_report_errors : 1,
518 sk_txtime_unused : 6;
520 struct socket *sk_socket;
522 #ifdef CONFIG_SECURITY
525 struct sock_cgroup_data sk_cgrp_data;
526 struct mem_cgroup *sk_memcg;
527 void (*sk_state_change)(struct sock *sk);
528 void (*sk_data_ready)(struct sock *sk);
529 void (*sk_write_space)(struct sock *sk);
530 void (*sk_error_report)(struct sock *sk);
531 int (*sk_backlog_rcv)(struct sock *sk,
532 struct sk_buff *skb);
533 #ifdef CONFIG_SOCK_VALIDATE_XMIT
534 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
535 struct net_device *dev,
536 struct sk_buff *skb);
538 void (*sk_destruct)(struct sock *sk);
539 struct sock_reuseport __rcu *sk_reuseport_cb;
540 #ifdef CONFIG_BPF_SYSCALL
541 struct bpf_local_storage __rcu *sk_bpf_storage;
543 struct rcu_head sk_rcu;
544 netns_tracker ns_tracker;
549 SK_PACING_NEEDED = 1,
553 /* Pointer stored in sk_user_data might not be suitable for copying
554 * when cloning the socket. For instance, it can point to a reference
555 * counted object. sk_user_data bottom bit is set if pointer must not
558 #define SK_USER_DATA_NOCOPY 1UL
559 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */
560 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF)
563 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
566 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
568 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
571 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
573 #define rcu_dereference_sk_user_data(sk) \
575 void *__tmp = rcu_dereference(__sk_user_data((sk))); \
576 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \
578 #define rcu_assign_sk_user_data(sk, ptr) \
580 uintptr_t __tmp = (uintptr_t)(ptr); \
581 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
582 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \
584 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \
586 uintptr_t __tmp = (uintptr_t)(ptr); \
587 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
588 rcu_assign_pointer(__sk_user_data((sk)), \
589 __tmp | SK_USER_DATA_NOCOPY); \
593 struct net *sock_net(const struct sock *sk)
595 return read_pnet(&sk->sk_net);
599 void sock_net_set(struct sock *sk, struct net *net)
601 write_pnet(&sk->sk_net, net);
605 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
606 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
607 * on a socket means that the socket will reuse everybody else's port
608 * without looking at the other's sk_reuse value.
611 #define SK_NO_REUSE 0
612 #define SK_CAN_REUSE 1
613 #define SK_FORCE_REUSE 2
615 int sk_set_peek_off(struct sock *sk, int val);
617 static inline int sk_peek_offset(struct sock *sk, int flags)
619 if (unlikely(flags & MSG_PEEK)) {
620 return READ_ONCE(sk->sk_peek_off);
626 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
628 s32 off = READ_ONCE(sk->sk_peek_off);
630 if (unlikely(off >= 0)) {
631 off = max_t(s32, off - val, 0);
632 WRITE_ONCE(sk->sk_peek_off, off);
636 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
638 sk_peek_offset_bwd(sk, -val);
642 * Hashed lists helper routines
644 static inline struct sock *sk_entry(const struct hlist_node *node)
646 return hlist_entry(node, struct sock, sk_node);
649 static inline struct sock *__sk_head(const struct hlist_head *head)
651 return hlist_entry(head->first, struct sock, sk_node);
654 static inline struct sock *sk_head(const struct hlist_head *head)
656 return hlist_empty(head) ? NULL : __sk_head(head);
659 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
661 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
664 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
666 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
669 static inline struct sock *sk_next(const struct sock *sk)
671 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
674 static inline struct sock *sk_nulls_next(const struct sock *sk)
676 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
677 hlist_nulls_entry(sk->sk_nulls_node.next,
678 struct sock, sk_nulls_node) :
682 static inline bool sk_unhashed(const struct sock *sk)
684 return hlist_unhashed(&sk->sk_node);
687 static inline bool sk_hashed(const struct sock *sk)
689 return !sk_unhashed(sk);
692 static inline void sk_node_init(struct hlist_node *node)
697 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
702 static inline void __sk_del_node(struct sock *sk)
704 __hlist_del(&sk->sk_node);
707 /* NB: equivalent to hlist_del_init_rcu */
708 static inline bool __sk_del_node_init(struct sock *sk)
712 sk_node_init(&sk->sk_node);
718 /* Grab socket reference count. This operation is valid only
719 when sk is ALREADY grabbed f.e. it is found in hash table
720 or a list and the lookup is made under lock preventing hash table
724 static __always_inline void sock_hold(struct sock *sk)
726 refcount_inc(&sk->sk_refcnt);
729 /* Ungrab socket in the context, which assumes that socket refcnt
730 cannot hit zero, f.e. it is true in context of any socketcall.
732 static __always_inline void __sock_put(struct sock *sk)
734 refcount_dec(&sk->sk_refcnt);
737 static inline bool sk_del_node_init(struct sock *sk)
739 bool rc = __sk_del_node_init(sk);
742 /* paranoid for a while -acme */
743 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
748 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
750 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
753 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
759 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
761 bool rc = __sk_nulls_del_node_init_rcu(sk);
764 /* paranoid for a while -acme */
765 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
771 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
773 hlist_add_head(&sk->sk_node, list);
776 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
779 __sk_add_node(sk, list);
782 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
785 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
786 sk->sk_family == AF_INET6)
787 hlist_add_tail_rcu(&sk->sk_node, list);
789 hlist_add_head_rcu(&sk->sk_node, list);
792 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
795 hlist_add_tail_rcu(&sk->sk_node, list);
798 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
800 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
803 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
805 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
808 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
811 __sk_nulls_add_node_rcu(sk, list);
814 static inline void __sk_del_bind_node(struct sock *sk)
816 __hlist_del(&sk->sk_bind_node);
819 static inline void sk_add_bind_node(struct sock *sk,
820 struct hlist_head *list)
822 hlist_add_head(&sk->sk_bind_node, list);
825 #define sk_for_each(__sk, list) \
826 hlist_for_each_entry(__sk, list, sk_node)
827 #define sk_for_each_rcu(__sk, list) \
828 hlist_for_each_entry_rcu(__sk, list, sk_node)
829 #define sk_nulls_for_each(__sk, node, list) \
830 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
831 #define sk_nulls_for_each_rcu(__sk, node, list) \
832 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
833 #define sk_for_each_from(__sk) \
834 hlist_for_each_entry_from(__sk, sk_node)
835 #define sk_nulls_for_each_from(__sk, node) \
836 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
837 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
838 #define sk_for_each_safe(__sk, tmp, list) \
839 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
840 #define sk_for_each_bound(__sk, list) \
841 hlist_for_each_entry(__sk, list, sk_bind_node)
844 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
845 * @tpos: the type * to use as a loop cursor.
846 * @pos: the &struct hlist_node to use as a loop cursor.
847 * @head: the head for your list.
848 * @offset: offset of hlist_node within the struct.
851 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
852 for (pos = rcu_dereference(hlist_first_rcu(head)); \
854 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
855 pos = rcu_dereference(hlist_next_rcu(pos)))
857 static inline struct user_namespace *sk_user_ns(struct sock *sk)
859 /* Careful only use this in a context where these parameters
860 * can not change and must all be valid, such as recvmsg from
863 return sk->sk_socket->file->f_cred->user_ns;
877 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
878 SOCK_DBG, /* %SO_DEBUG setting */
879 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
880 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
881 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
882 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
883 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
884 SOCK_FASYNC, /* fasync() active */
886 SOCK_ZEROCOPY, /* buffers from userspace */
887 SOCK_WIFI_STATUS, /* push wifi status to userspace */
888 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
889 * Will use last 4 bytes of packet sent from
890 * user-space instead.
892 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
893 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
894 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
896 SOCK_XDP, /* XDP is attached */
897 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
900 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
902 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
904 nsk->sk_flags = osk->sk_flags;
907 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
909 __set_bit(flag, &sk->sk_flags);
912 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
914 __clear_bit(flag, &sk->sk_flags);
917 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
921 sock_set_flag(sk, bit);
923 sock_reset_flag(sk, bit);
926 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
928 return test_bit(flag, &sk->sk_flags);
932 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
933 static inline int sk_memalloc_socks(void)
935 return static_branch_unlikely(&memalloc_socks_key);
938 void __receive_sock(struct file *file);
941 static inline int sk_memalloc_socks(void)
946 static inline void __receive_sock(struct file *file)
950 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
952 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
955 static inline void sk_acceptq_removed(struct sock *sk)
957 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
960 static inline void sk_acceptq_added(struct sock *sk)
962 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
965 /* Note: If you think the test should be:
966 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
967 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
969 static inline bool sk_acceptq_is_full(const struct sock *sk)
971 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
975 * Compute minimal free write space needed to queue new packets.
977 static inline int sk_stream_min_wspace(const struct sock *sk)
979 return READ_ONCE(sk->sk_wmem_queued) >> 1;
982 static inline int sk_stream_wspace(const struct sock *sk)
984 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
987 static inline void sk_wmem_queued_add(struct sock *sk, int val)
989 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
992 void sk_stream_write_space(struct sock *sk);
994 /* OOB backlog add */
995 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
997 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1000 if (!sk->sk_backlog.tail)
1001 WRITE_ONCE(sk->sk_backlog.head, skb);
1003 sk->sk_backlog.tail->next = skb;
1005 WRITE_ONCE(sk->sk_backlog.tail, skb);
1010 * Take into account size of receive queue and backlog queue
1011 * Do not take into account this skb truesize,
1012 * to allow even a single big packet to come.
1014 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1016 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1018 return qsize > limit;
1021 /* The per-socket spinlock must be held here. */
1022 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1025 if (sk_rcvqueues_full(sk, limit))
1029 * If the skb was allocated from pfmemalloc reserves, only
1030 * allow SOCK_MEMALLOC sockets to use it as this socket is
1031 * helping free memory
1033 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1036 __sk_add_backlog(sk, skb);
1037 sk->sk_backlog.len += skb->truesize;
1041 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1043 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1044 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1046 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1048 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1049 return __sk_backlog_rcv(sk, skb);
1051 return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1057 static inline void sk_incoming_cpu_update(struct sock *sk)
1059 int cpu = raw_smp_processor_id();
1061 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1062 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1065 static inline void sock_rps_record_flow_hash(__u32 hash)
1068 struct rps_sock_flow_table *sock_flow_table;
1071 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1072 rps_record_sock_flow(sock_flow_table, hash);
1077 static inline void sock_rps_record_flow(const struct sock *sk)
1080 if (static_branch_unlikely(&rfs_needed)) {
1081 /* Reading sk->sk_rxhash might incur an expensive cache line
1084 * TCP_ESTABLISHED does cover almost all states where RFS
1085 * might be useful, and is cheaper [1] than testing :
1086 * IPv4: inet_sk(sk)->inet_daddr
1087 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1088 * OR an additional socket flag
1089 * [1] : sk_state and sk_prot are in the same cache line.
1091 if (sk->sk_state == TCP_ESTABLISHED)
1092 sock_rps_record_flow_hash(sk->sk_rxhash);
1097 static inline void sock_rps_save_rxhash(struct sock *sk,
1098 const struct sk_buff *skb)
1101 if (unlikely(sk->sk_rxhash != skb->hash))
1102 sk->sk_rxhash = skb->hash;
1106 static inline void sock_rps_reset_rxhash(struct sock *sk)
1113 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1115 release_sock(__sk); \
1116 __rc = __condition; \
1118 *(__timeo) = wait_woken(__wait, \
1119 TASK_INTERRUPTIBLE, \
1122 sched_annotate_sleep(); \
1124 __rc = __condition; \
1128 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1129 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1130 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1131 int sk_stream_error(struct sock *sk, int flags, int err);
1132 void sk_stream_kill_queues(struct sock *sk);
1133 void sk_set_memalloc(struct sock *sk);
1134 void sk_clear_memalloc(struct sock *sk);
1136 void __sk_flush_backlog(struct sock *sk);
1138 static inline bool sk_flush_backlog(struct sock *sk)
1140 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1141 __sk_flush_backlog(sk);
1147 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1149 struct request_sock_ops;
1150 struct timewait_sock_ops;
1151 struct inet_hashinfo;
1152 struct raw_hashinfo;
1153 struct smc_hashinfo;
1158 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1159 * un-modified. Special care is taken when initializing object to zero.
1161 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1163 if (offsetof(struct sock, sk_node.next) != 0)
1164 memset(sk, 0, offsetof(struct sock, sk_node.next));
1165 memset(&sk->sk_node.pprev, 0,
1166 size - offsetof(struct sock, sk_node.pprev));
1169 /* Networking protocol blocks we attach to sockets.
1170 * socket layer -> transport layer interface
1173 void (*close)(struct sock *sk,
1175 int (*pre_connect)(struct sock *sk,
1176 struct sockaddr *uaddr,
1178 int (*connect)(struct sock *sk,
1179 struct sockaddr *uaddr,
1181 int (*disconnect)(struct sock *sk, int flags);
1183 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1186 int (*ioctl)(struct sock *sk, int cmd,
1188 int (*init)(struct sock *sk);
1189 void (*destroy)(struct sock *sk);
1190 void (*shutdown)(struct sock *sk, int how);
1191 int (*setsockopt)(struct sock *sk, int level,
1192 int optname, sockptr_t optval,
1193 unsigned int optlen);
1194 int (*getsockopt)(struct sock *sk, int level,
1195 int optname, char __user *optval,
1196 int __user *option);
1197 void (*keepalive)(struct sock *sk, int valbool);
1198 #ifdef CONFIG_COMPAT
1199 int (*compat_ioctl)(struct sock *sk,
1200 unsigned int cmd, unsigned long arg);
1202 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1204 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1205 size_t len, int noblock, int flags,
1207 int (*sendpage)(struct sock *sk, struct page *page,
1208 int offset, size_t size, int flags);
1209 int (*bind)(struct sock *sk,
1210 struct sockaddr *addr, int addr_len);
1211 int (*bind_add)(struct sock *sk,
1212 struct sockaddr *addr, int addr_len);
1214 int (*backlog_rcv) (struct sock *sk,
1215 struct sk_buff *skb);
1216 bool (*bpf_bypass_getsockopt)(int level,
1219 void (*release_cb)(struct sock *sk);
1221 /* Keeping track of sk's, looking them up, and port selection methods. */
1222 int (*hash)(struct sock *sk);
1223 void (*unhash)(struct sock *sk);
1224 void (*rehash)(struct sock *sk);
1225 int (*get_port)(struct sock *sk, unsigned short snum);
1226 void (*put_port)(struct sock *sk);
1227 #ifdef CONFIG_BPF_SYSCALL
1228 int (*psock_update_sk_prot)(struct sock *sk,
1229 struct sk_psock *psock,
1233 /* Keeping track of sockets in use */
1234 #ifdef CONFIG_PROC_FS
1235 unsigned int inuse_idx;
1238 #if IS_ENABLED(CONFIG_MPTCP)
1239 int (*forward_alloc_get)(const struct sock *sk);
1242 bool (*stream_memory_free)(const struct sock *sk, int wake);
1243 bool (*sock_is_readable)(struct sock *sk);
1244 /* Memory pressure */
1245 void (*enter_memory_pressure)(struct sock *sk);
1246 void (*leave_memory_pressure)(struct sock *sk);
1247 atomic_long_t *memory_allocated; /* Current allocated memory. */
1248 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1251 * Pressure flag: try to collapse.
1252 * Technical note: it is used by multiple contexts non atomically.
1253 * All the __sk_mem_schedule() is of this nature: accounting
1254 * is strict, actions are advisory and have some latency.
1256 unsigned long *memory_pressure;
1261 u32 sysctl_wmem_offset;
1262 u32 sysctl_rmem_offset;
1267 struct kmem_cache *slab;
1268 unsigned int obj_size;
1269 slab_flags_t slab_flags;
1270 unsigned int useroffset; /* Usercopy region offset */
1271 unsigned int usersize; /* Usercopy region size */
1273 unsigned int __percpu *orphan_count;
1275 struct request_sock_ops *rsk_prot;
1276 struct timewait_sock_ops *twsk_prot;
1279 struct inet_hashinfo *hashinfo;
1280 struct udp_table *udp_table;
1281 struct raw_hashinfo *raw_hash;
1282 struct smc_hashinfo *smc_hash;
1285 struct module *owner;
1289 struct list_head node;
1290 #ifdef SOCK_REFCNT_DEBUG
1293 int (*diag_destroy)(struct sock *sk, int err);
1294 } __randomize_layout;
1296 int proto_register(struct proto *prot, int alloc_slab);
1297 void proto_unregister(struct proto *prot);
1298 int sock_load_diag_module(int family, int protocol);
1300 #ifdef SOCK_REFCNT_DEBUG
1301 static inline void sk_refcnt_debug_inc(struct sock *sk)
1303 atomic_inc(&sk->sk_prot->socks);
1306 static inline void sk_refcnt_debug_dec(struct sock *sk)
1308 atomic_dec(&sk->sk_prot->socks);
1309 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1310 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1313 static inline void sk_refcnt_debug_release(const struct sock *sk)
1315 if (refcount_read(&sk->sk_refcnt) != 1)
1316 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1317 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1319 #else /* SOCK_REFCNT_DEBUG */
1320 #define sk_refcnt_debug_inc(sk) do { } while (0)
1321 #define sk_refcnt_debug_dec(sk) do { } while (0)
1322 #define sk_refcnt_debug_release(sk) do { } while (0)
1323 #endif /* SOCK_REFCNT_DEBUG */
1325 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1327 static inline int sk_forward_alloc_get(const struct sock *sk)
1329 #if IS_ENABLED(CONFIG_MPTCP)
1330 if (sk->sk_prot->forward_alloc_get)
1331 return sk->sk_prot->forward_alloc_get(sk);
1333 return sk->sk_forward_alloc;
1336 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1338 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1341 return sk->sk_prot->stream_memory_free ?
1342 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1343 tcp_stream_memory_free, sk, wake) : true;
1346 static inline bool sk_stream_memory_free(const struct sock *sk)
1348 return __sk_stream_memory_free(sk, 0);
1351 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1353 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1354 __sk_stream_memory_free(sk, wake);
1357 static inline bool sk_stream_is_writeable(const struct sock *sk)
1359 return __sk_stream_is_writeable(sk, 0);
1362 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1363 struct cgroup *ancestor)
1365 #ifdef CONFIG_SOCK_CGROUP_DATA
1366 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1373 static inline bool sk_has_memory_pressure(const struct sock *sk)
1375 return sk->sk_prot->memory_pressure != NULL;
1378 static inline bool sk_under_memory_pressure(const struct sock *sk)
1380 if (!sk->sk_prot->memory_pressure)
1383 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1384 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1387 return !!*sk->sk_prot->memory_pressure;
1391 sk_memory_allocated(const struct sock *sk)
1393 return atomic_long_read(sk->sk_prot->memory_allocated);
1397 sk_memory_allocated_add(struct sock *sk, int amt)
1399 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1403 sk_memory_allocated_sub(struct sock *sk, int amt)
1405 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1408 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1410 static inline void sk_sockets_allocated_dec(struct sock *sk)
1412 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1413 SK_ALLOC_PERCPU_COUNTER_BATCH);
1416 static inline void sk_sockets_allocated_inc(struct sock *sk)
1418 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1419 SK_ALLOC_PERCPU_COUNTER_BATCH);
1423 sk_sockets_allocated_read_positive(struct sock *sk)
1425 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1429 proto_sockets_allocated_sum_positive(struct proto *prot)
1431 return percpu_counter_sum_positive(prot->sockets_allocated);
1435 proto_memory_allocated(struct proto *prot)
1437 return atomic_long_read(prot->memory_allocated);
1441 proto_memory_pressure(struct proto *prot)
1443 if (!prot->memory_pressure)
1445 return !!*prot->memory_pressure;
1449 #ifdef CONFIG_PROC_FS
1450 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1453 int val[PROTO_INUSE_NR];
1456 static inline void sock_prot_inuse_add(const struct net *net,
1457 const struct proto *prot, int val)
1459 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1462 static inline void sock_inuse_add(const struct net *net, int val)
1464 this_cpu_add(net->core.prot_inuse->all, val);
1467 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1468 int sock_inuse_get(struct net *net);
1470 static inline void sock_prot_inuse_add(const struct net *net,
1471 const struct proto *prot, int val)
1475 static inline void sock_inuse_add(const struct net *net, int val)
1481 /* With per-bucket locks this operation is not-atomic, so that
1482 * this version is not worse.
1484 static inline int __sk_prot_rehash(struct sock *sk)
1486 sk->sk_prot->unhash(sk);
1487 return sk->sk_prot->hash(sk);
1490 /* About 10 seconds */
1491 #define SOCK_DESTROY_TIME (10*HZ)
1493 /* Sockets 0-1023 can't be bound to unless you are superuser */
1494 #define PROT_SOCK 1024
1496 #define SHUTDOWN_MASK 3
1497 #define RCV_SHUTDOWN 1
1498 #define SEND_SHUTDOWN 2
1500 #define SOCK_BINDADDR_LOCK 4
1501 #define SOCK_BINDPORT_LOCK 8
1503 struct socket_alloc {
1504 struct socket socket;
1505 struct inode vfs_inode;
1508 static inline struct socket *SOCKET_I(struct inode *inode)
1510 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1513 static inline struct inode *SOCK_INODE(struct socket *socket)
1515 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1519 * Functions for memory accounting
1521 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1522 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1523 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1524 void __sk_mem_reclaim(struct sock *sk, int amount);
1526 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1527 * do not necessarily have 16x time more memory than 4KB ones.
1529 #define SK_MEM_QUANTUM 4096
1530 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1531 #define SK_MEM_SEND 0
1532 #define SK_MEM_RECV 1
1534 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1535 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1537 long val = sk->sk_prot->sysctl_mem[index];
1539 #if PAGE_SIZE > SK_MEM_QUANTUM
1540 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1541 #elif PAGE_SIZE < SK_MEM_QUANTUM
1542 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1547 static inline int sk_mem_pages(int amt)
1549 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1552 static inline bool sk_has_account(struct sock *sk)
1554 /* return true if protocol supports memory accounting */
1555 return !!sk->sk_prot->memory_allocated;
1558 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1560 if (!sk_has_account(sk))
1562 return size <= sk->sk_forward_alloc ||
1563 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1567 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1569 if (!sk_has_account(sk))
1571 return size <= sk->sk_forward_alloc ||
1572 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1573 skb_pfmemalloc(skb);
1576 static inline int sk_unused_reserved_mem(const struct sock *sk)
1580 if (likely(!sk->sk_reserved_mem))
1583 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1584 atomic_read(&sk->sk_rmem_alloc);
1586 return unused_mem > 0 ? unused_mem : 0;
1589 static inline void sk_mem_reclaim(struct sock *sk)
1593 if (!sk_has_account(sk))
1596 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1598 if (reclaimable >= SK_MEM_QUANTUM)
1599 __sk_mem_reclaim(sk, reclaimable);
1602 static inline void sk_mem_reclaim_final(struct sock *sk)
1604 sk->sk_reserved_mem = 0;
1608 static inline void sk_mem_reclaim_partial(struct sock *sk)
1612 if (!sk_has_account(sk))
1615 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1617 if (reclaimable > SK_MEM_QUANTUM)
1618 __sk_mem_reclaim(sk, reclaimable - 1);
1621 static inline void sk_mem_charge(struct sock *sk, int size)
1623 if (!sk_has_account(sk))
1625 sk->sk_forward_alloc -= size;
1628 /* the following macros control memory reclaiming in sk_mem_uncharge()
1630 #define SK_RECLAIM_THRESHOLD (1 << 21)
1631 #define SK_RECLAIM_CHUNK (1 << 20)
1633 static inline void sk_mem_uncharge(struct sock *sk, int size)
1637 if (!sk_has_account(sk))
1639 sk->sk_forward_alloc += size;
1640 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1642 /* Avoid a possible overflow.
1643 * TCP send queues can make this happen, if sk_mem_reclaim()
1644 * is not called and more than 2 GBytes are released at once.
1646 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1647 * no need to hold that much forward allocation anyway.
1649 if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD))
1650 __sk_mem_reclaim(sk, SK_RECLAIM_CHUNK);
1654 * Macro so as to not evaluate some arguments when
1655 * lockdep is not enabled.
1657 * Mark both the sk_lock and the sk_lock.slock as a
1658 * per-address-family lock class.
1660 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1662 sk->sk_lock.owned = 0; \
1663 init_waitqueue_head(&sk->sk_lock.wq); \
1664 spin_lock_init(&(sk)->sk_lock.slock); \
1665 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1666 sizeof((sk)->sk_lock)); \
1667 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1669 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1672 static inline bool lockdep_sock_is_held(const struct sock *sk)
1674 return lockdep_is_held(&sk->sk_lock) ||
1675 lockdep_is_held(&sk->sk_lock.slock);
1678 void lock_sock_nested(struct sock *sk, int subclass);
1680 static inline void lock_sock(struct sock *sk)
1682 lock_sock_nested(sk, 0);
1685 void __lock_sock(struct sock *sk);
1686 void __release_sock(struct sock *sk);
1687 void release_sock(struct sock *sk);
1689 /* BH context may only use the following locking interface. */
1690 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1691 #define bh_lock_sock_nested(__sk) \
1692 spin_lock_nested(&((__sk)->sk_lock.slock), \
1693 SINGLE_DEPTH_NESTING)
1694 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1696 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1699 * lock_sock_fast - fast version of lock_sock
1702 * This version should be used for very small section, where process wont block
1703 * return false if fast path is taken:
1705 * sk_lock.slock locked, owned = 0, BH disabled
1707 * return true if slow path is taken:
1709 * sk_lock.slock unlocked, owned = 1, BH enabled
1711 static inline bool lock_sock_fast(struct sock *sk)
1713 /* The sk_lock has mutex_lock() semantics here. */
1714 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1716 return __lock_sock_fast(sk);
1719 /* fast socket lock variant for caller already holding a [different] socket lock */
1720 static inline bool lock_sock_fast_nested(struct sock *sk)
1722 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1724 return __lock_sock_fast(sk);
1728 * unlock_sock_fast - complement of lock_sock_fast
1732 * fast unlock socket for user context.
1733 * If slow mode is on, we call regular release_sock()
1735 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1736 __releases(&sk->sk_lock.slock)
1740 __release(&sk->sk_lock.slock);
1742 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1743 spin_unlock_bh(&sk->sk_lock.slock);
1747 /* Used by processes to "lock" a socket state, so that
1748 * interrupts and bottom half handlers won't change it
1749 * from under us. It essentially blocks any incoming
1750 * packets, so that we won't get any new data or any
1751 * packets that change the state of the socket.
1753 * While locked, BH processing will add new packets to
1754 * the backlog queue. This queue is processed by the
1755 * owner of the socket lock right before it is released.
1757 * Since ~2.3.5 it is also exclusive sleep lock serializing
1758 * accesses from user process context.
1761 static inline void sock_owned_by_me(const struct sock *sk)
1763 #ifdef CONFIG_LOCKDEP
1764 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1768 static inline bool sock_owned_by_user(const struct sock *sk)
1770 sock_owned_by_me(sk);
1771 return sk->sk_lock.owned;
1774 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1776 return sk->sk_lock.owned;
1779 static inline void sock_release_ownership(struct sock *sk)
1781 if (sock_owned_by_user_nocheck(sk)) {
1782 sk->sk_lock.owned = 0;
1784 /* The sk_lock has mutex_unlock() semantics: */
1785 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1789 /* no reclassification while locks are held */
1790 static inline bool sock_allow_reclassification(const struct sock *csk)
1792 struct sock *sk = (struct sock *)csk;
1794 return !sock_owned_by_user_nocheck(sk) &&
1795 !spin_is_locked(&sk->sk_lock.slock);
1798 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1799 struct proto *prot, int kern);
1800 void sk_free(struct sock *sk);
1801 void sk_destruct(struct sock *sk);
1802 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1803 void sk_free_unlock_clone(struct sock *sk);
1805 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1807 void __sock_wfree(struct sk_buff *skb);
1808 void sock_wfree(struct sk_buff *skb);
1809 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1811 void skb_orphan_partial(struct sk_buff *skb);
1812 void sock_rfree(struct sk_buff *skb);
1813 void sock_efree(struct sk_buff *skb);
1815 void sock_edemux(struct sk_buff *skb);
1816 void sock_pfree(struct sk_buff *skb);
1818 #define sock_edemux sock_efree
1821 int sock_setsockopt(struct socket *sock, int level, int op,
1822 sockptr_t optval, unsigned int optlen);
1824 int sock_getsockopt(struct socket *sock, int level, int op,
1825 char __user *optval, int __user *optlen);
1826 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1827 bool timeval, bool time32);
1828 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1829 int noblock, int *errcode);
1830 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1831 unsigned long data_len, int noblock,
1832 int *errcode, int max_page_order);
1833 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1834 void sock_kfree_s(struct sock *sk, void *mem, int size);
1835 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1836 void sk_send_sigurg(struct sock *sk);
1838 struct sockcm_cookie {
1844 static inline void sockcm_init(struct sockcm_cookie *sockc,
1845 const struct sock *sk)
1847 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1850 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1851 struct sockcm_cookie *sockc);
1852 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1853 struct sockcm_cookie *sockc);
1856 * Functions to fill in entries in struct proto_ops when a protocol
1857 * does not implement a particular function.
1859 int sock_no_bind(struct socket *, struct sockaddr *, int);
1860 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1861 int sock_no_socketpair(struct socket *, struct socket *);
1862 int sock_no_accept(struct socket *, struct socket *, int, bool);
1863 int sock_no_getname(struct socket *, struct sockaddr *, int);
1864 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1865 int sock_no_listen(struct socket *, int);
1866 int sock_no_shutdown(struct socket *, int);
1867 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1868 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1869 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1870 int sock_no_mmap(struct file *file, struct socket *sock,
1871 struct vm_area_struct *vma);
1872 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1873 size_t size, int flags);
1874 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1875 int offset, size_t size, int flags);
1878 * Functions to fill in entries in struct proto_ops when a protocol
1879 * uses the inet style.
1881 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1882 char __user *optval, int __user *optlen);
1883 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1885 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1886 sockptr_t optval, unsigned int optlen);
1888 void sk_common_release(struct sock *sk);
1891 * Default socket callbacks and setup code
1894 /* Initialise core socket variables */
1895 void sock_init_data(struct socket *sock, struct sock *sk);
1898 * Socket reference counting postulates.
1900 * * Each user of socket SHOULD hold a reference count.
1901 * * Each access point to socket (an hash table bucket, reference from a list,
1902 * running timer, skb in flight MUST hold a reference count.
1903 * * When reference count hits 0, it means it will never increase back.
1904 * * When reference count hits 0, it means that no references from
1905 * outside exist to this socket and current process on current CPU
1906 * is last user and may/should destroy this socket.
1907 * * sk_free is called from any context: process, BH, IRQ. When
1908 * it is called, socket has no references from outside -> sk_free
1909 * may release descendant resources allocated by the socket, but
1910 * to the time when it is called, socket is NOT referenced by any
1911 * hash tables, lists etc.
1912 * * Packets, delivered from outside (from network or from another process)
1913 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1914 * when they sit in queue. Otherwise, packets will leak to hole, when
1915 * socket is looked up by one cpu and unhasing is made by another CPU.
1916 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1917 * (leak to backlog). Packet socket does all the processing inside
1918 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1919 * use separate SMP lock, so that they are prone too.
1922 /* Ungrab socket and destroy it, if it was the last reference. */
1923 static inline void sock_put(struct sock *sk)
1925 if (refcount_dec_and_test(&sk->sk_refcnt))
1928 /* Generic version of sock_put(), dealing with all sockets
1929 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1931 void sock_gen_put(struct sock *sk);
1933 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1934 unsigned int trim_cap, bool refcounted);
1935 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1938 return __sk_receive_skb(sk, skb, nested, 1, true);
1941 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1943 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1944 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1946 sk->sk_tx_queue_mapping = tx_queue;
1949 #define NO_QUEUE_MAPPING USHRT_MAX
1951 static inline void sk_tx_queue_clear(struct sock *sk)
1953 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1956 static inline int sk_tx_queue_get(const struct sock *sk)
1958 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1959 return sk->sk_tx_queue_mapping;
1964 static inline void __sk_rx_queue_set(struct sock *sk,
1965 const struct sk_buff *skb,
1968 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1969 if (skb_rx_queue_recorded(skb)) {
1970 u16 rx_queue = skb_get_rx_queue(skb);
1973 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
1974 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
1979 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1981 __sk_rx_queue_set(sk, skb, true);
1984 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
1986 __sk_rx_queue_set(sk, skb, false);
1989 static inline void sk_rx_queue_clear(struct sock *sk)
1991 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1992 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
1996 static inline int sk_rx_queue_get(const struct sock *sk)
1998 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2000 int res = READ_ONCE(sk->sk_rx_queue_mapping);
2002 if (res != NO_QUEUE_MAPPING)
2010 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2012 sk->sk_socket = sock;
2015 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2017 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2018 return &rcu_dereference_raw(sk->sk_wq)->wait;
2020 /* Detach socket from process context.
2021 * Announce socket dead, detach it from wait queue and inode.
2022 * Note that parent inode held reference count on this struct sock,
2023 * we do not release it in this function, because protocol
2024 * probably wants some additional cleanups or even continuing
2025 * to work with this socket (TCP).
2027 static inline void sock_orphan(struct sock *sk)
2029 write_lock_bh(&sk->sk_callback_lock);
2030 sock_set_flag(sk, SOCK_DEAD);
2031 sk_set_socket(sk, NULL);
2033 write_unlock_bh(&sk->sk_callback_lock);
2036 static inline void sock_graft(struct sock *sk, struct socket *parent)
2038 WARN_ON(parent->sk);
2039 write_lock_bh(&sk->sk_callback_lock);
2040 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2042 sk_set_socket(sk, parent);
2043 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2044 security_sock_graft(sk, parent);
2045 write_unlock_bh(&sk->sk_callback_lock);
2048 kuid_t sock_i_uid(struct sock *sk);
2049 unsigned long sock_i_ino(struct sock *sk);
2051 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2053 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2056 static inline u32 net_tx_rndhash(void)
2058 u32 v = prandom_u32();
2063 static inline void sk_set_txhash(struct sock *sk)
2065 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2066 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2069 static inline bool sk_rethink_txhash(struct sock *sk)
2071 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2078 static inline struct dst_entry *
2079 __sk_dst_get(struct sock *sk)
2081 return rcu_dereference_check(sk->sk_dst_cache,
2082 lockdep_sock_is_held(sk));
2085 static inline struct dst_entry *
2086 sk_dst_get(struct sock *sk)
2088 struct dst_entry *dst;
2091 dst = rcu_dereference(sk->sk_dst_cache);
2092 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
2098 static inline void __dst_negative_advice(struct sock *sk)
2100 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2102 if (dst && dst->ops->negative_advice) {
2103 ndst = dst->ops->negative_advice(dst);
2106 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2107 sk_tx_queue_clear(sk);
2108 sk->sk_dst_pending_confirm = 0;
2113 static inline void dst_negative_advice(struct sock *sk)
2115 sk_rethink_txhash(sk);
2116 __dst_negative_advice(sk);
2120 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2122 struct dst_entry *old_dst;
2124 sk_tx_queue_clear(sk);
2125 sk->sk_dst_pending_confirm = 0;
2126 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2127 lockdep_sock_is_held(sk));
2128 rcu_assign_pointer(sk->sk_dst_cache, dst);
2129 dst_release(old_dst);
2133 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2135 struct dst_entry *old_dst;
2137 sk_tx_queue_clear(sk);
2138 sk->sk_dst_pending_confirm = 0;
2139 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2140 dst_release(old_dst);
2144 __sk_dst_reset(struct sock *sk)
2146 __sk_dst_set(sk, NULL);
2150 sk_dst_reset(struct sock *sk)
2152 sk_dst_set(sk, NULL);
2155 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2157 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2159 static inline void sk_dst_confirm(struct sock *sk)
2161 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2162 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2165 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2167 if (skb_get_dst_pending_confirm(skb)) {
2168 struct sock *sk = skb->sk;
2170 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2171 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2176 bool sk_mc_loop(struct sock *sk);
2178 static inline bool sk_can_gso(const struct sock *sk)
2180 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2183 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2185 static inline void sk_gso_disable(struct sock *sk)
2187 sk->sk_gso_disabled = 1;
2188 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2191 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2192 struct iov_iter *from, char *to,
2193 int copy, int offset)
2195 if (skb->ip_summed == CHECKSUM_NONE) {
2197 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2199 skb->csum = csum_block_add(skb->csum, csum, offset);
2200 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2201 if (!copy_from_iter_full_nocache(to, copy, from))
2203 } else if (!copy_from_iter_full(to, copy, from))
2209 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2210 struct iov_iter *from, int copy)
2212 int err, offset = skb->len;
2214 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2217 __skb_trim(skb, offset);
2222 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2223 struct sk_buff *skb,
2229 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2235 skb->data_len += copy;
2236 skb->truesize += copy;
2237 sk_wmem_queued_add(sk, copy);
2238 sk_mem_charge(sk, copy);
2243 * sk_wmem_alloc_get - returns write allocations
2246 * Return: sk_wmem_alloc minus initial offset of one
2248 static inline int sk_wmem_alloc_get(const struct sock *sk)
2250 return refcount_read(&sk->sk_wmem_alloc) - 1;
2254 * sk_rmem_alloc_get - returns read allocations
2257 * Return: sk_rmem_alloc
2259 static inline int sk_rmem_alloc_get(const struct sock *sk)
2261 return atomic_read(&sk->sk_rmem_alloc);
2265 * sk_has_allocations - check if allocations are outstanding
2268 * Return: true if socket has write or read allocations
2270 static inline bool sk_has_allocations(const struct sock *sk)
2272 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2276 * skwq_has_sleeper - check if there are any waiting processes
2277 * @wq: struct socket_wq
2279 * Return: true if socket_wq has waiting processes
2281 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2282 * barrier call. They were added due to the race found within the tcp code.
2284 * Consider following tcp code paths::
2287 * sys_select receive packet
2289 * __add_wait_queue update tp->rcv_nxt
2291 * tp->rcv_nxt check sock_def_readable
2293 * schedule rcu_read_lock();
2294 * wq = rcu_dereference(sk->sk_wq);
2295 * if (wq && waitqueue_active(&wq->wait))
2296 * wake_up_interruptible(&wq->wait)
2300 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2301 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2302 * could then endup calling schedule and sleep forever if there are no more
2303 * data on the socket.
2306 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2308 return wq && wq_has_sleeper(&wq->wait);
2312 * sock_poll_wait - place memory barrier behind the poll_wait call.
2314 * @sock: socket to wait on
2317 * See the comments in the wq_has_sleeper function.
2319 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2322 if (!poll_does_not_wait(p)) {
2323 poll_wait(filp, &sock->wq.wait, p);
2324 /* We need to be sure we are in sync with the
2325 * socket flags modification.
2327 * This memory barrier is paired in the wq_has_sleeper.
2333 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2335 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2336 u32 txhash = READ_ONCE(sk->sk_txhash);
2344 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2347 * Queue a received datagram if it will fit. Stream and sequenced
2348 * protocols can't normally use this as they need to fit buffers in
2349 * and play with them.
2351 * Inlined as it's very short and called for pretty much every
2352 * packet ever received.
2354 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2358 skb->destructor = sock_rfree;
2359 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2360 sk_mem_charge(sk, skb->truesize);
2363 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2365 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2367 skb->destructor = sock_efree;
2374 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2376 if (skb->destructor != sock_wfree) {
2383 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2384 unsigned long expires);
2386 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2388 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2390 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2391 struct sk_buff *skb, unsigned int flags,
2392 void (*destructor)(struct sock *sk,
2393 struct sk_buff *skb));
2394 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2395 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2397 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2398 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2401 * Recover an error report and clear atomically
2404 static inline int sock_error(struct sock *sk)
2408 /* Avoid an atomic operation for the common case.
2409 * This is racy since another cpu/thread can change sk_err under us.
2411 if (likely(data_race(!sk->sk_err)))
2414 err = xchg(&sk->sk_err, 0);
2418 void sk_error_report(struct sock *sk);
2420 static inline unsigned long sock_wspace(struct sock *sk)
2424 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2425 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2433 * We use sk->sk_wq_raw, from contexts knowing this
2434 * pointer is not NULL and cannot disappear/change.
2436 static inline void sk_set_bit(int nr, struct sock *sk)
2438 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2439 !sock_flag(sk, SOCK_FASYNC))
2442 set_bit(nr, &sk->sk_wq_raw->flags);
2445 static inline void sk_clear_bit(int nr, struct sock *sk)
2447 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2448 !sock_flag(sk, SOCK_FASYNC))
2451 clear_bit(nr, &sk->sk_wq_raw->flags);
2454 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2456 if (sock_flag(sk, SOCK_FASYNC)) {
2458 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2463 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2464 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2465 * Note: for send buffers, TCP works better if we can build two skbs at
2468 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2470 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2471 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2473 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2477 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2480 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2481 val = max_t(u32, val, sk_unused_reserved_mem(sk));
2483 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2487 * sk_page_frag - return an appropriate page_frag
2490 * Use the per task page_frag instead of the per socket one for
2491 * optimization when we know that we're in process context and own
2492 * everything that's associated with %current.
2494 * Both direct reclaim and page faults can nest inside other
2495 * socket operations and end up recursing into sk_page_frag()
2496 * while it's already in use: explicitly avoid task page_frag
2497 * usage if the caller is potentially doing any of them.
2498 * This assumes that page fault handlers use the GFP_NOFS flags.
2500 * Return: a per task page_frag if context allows that,
2501 * otherwise a per socket one.
2503 static inline struct page_frag *sk_page_frag(struct sock *sk)
2505 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) ==
2506 (__GFP_DIRECT_RECLAIM | __GFP_FS))
2507 return ¤t->task_frag;
2509 return &sk->sk_frag;
2512 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2515 * Default write policy as shown to user space via poll/select/SIGIO
2517 static inline bool sock_writeable(const struct sock *sk)
2519 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2522 static inline gfp_t gfp_any(void)
2524 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2527 static inline gfp_t gfp_memcg_charge(void)
2529 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL;
2532 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2534 return noblock ? 0 : sk->sk_rcvtimeo;
2537 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2539 return noblock ? 0 : sk->sk_sndtimeo;
2542 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2544 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2549 /* Alas, with timeout socket operations are not restartable.
2550 * Compare this to poll().
2552 static inline int sock_intr_errno(long timeo)
2554 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2557 struct sock_skb_cb {
2561 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2562 * using skb->cb[] would keep using it directly and utilize its
2563 * alignement guarantee.
2565 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2566 sizeof(struct sock_skb_cb)))
2568 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2569 SOCK_SKB_CB_OFFSET))
2571 #define sock_skb_cb_check_size(size) \
2572 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2575 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2577 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2578 atomic_read(&sk->sk_drops) : 0;
2581 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2583 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2585 atomic_add(segs, &sk->sk_drops);
2588 static inline ktime_t sock_read_timestamp(struct sock *sk)
2590 #if BITS_PER_LONG==32
2595 seq = read_seqbegin(&sk->sk_stamp_seq);
2597 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2601 return READ_ONCE(sk->sk_stamp);
2605 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2607 #if BITS_PER_LONG==32
2608 write_seqlock(&sk->sk_stamp_seq);
2610 write_sequnlock(&sk->sk_stamp_seq);
2612 WRITE_ONCE(sk->sk_stamp, kt);
2616 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2617 struct sk_buff *skb);
2618 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2619 struct sk_buff *skb);
2622 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2624 ktime_t kt = skb->tstamp;
2625 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2628 * generate control messages if
2629 * - receive time stamping in software requested
2630 * - software time stamp available and wanted
2631 * - hardware time stamps available and wanted
2633 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2634 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2635 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2636 (hwtstamps->hwtstamp &&
2637 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2638 __sock_recv_timestamp(msg, sk, skb);
2640 sock_write_timestamp(sk, kt);
2642 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2643 __sock_recv_wifi_status(msg, sk, skb);
2646 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2647 struct sk_buff *skb);
2649 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2650 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2651 struct sk_buff *skb)
2653 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2654 (1UL << SOCK_RCVTSTAMP))
2655 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2656 SOF_TIMESTAMPING_RAW_HARDWARE)
2658 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2659 __sock_recv_ts_and_drops(msg, sk, skb);
2660 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2661 sock_write_timestamp(sk, skb->tstamp);
2662 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2663 sock_write_timestamp(sk, 0);
2666 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2669 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2670 * @sk: socket sending this packet
2671 * @tsflags: timestamping flags to use
2672 * @tx_flags: completed with instructions for time stamping
2673 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2675 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2677 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2678 __u8 *tx_flags, __u32 *tskey)
2680 if (unlikely(tsflags)) {
2681 __sock_tx_timestamp(tsflags, tx_flags);
2682 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2683 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2684 *tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2686 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2687 *tx_flags |= SKBTX_WIFI_STATUS;
2690 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2693 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2696 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2698 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2699 &skb_shinfo(skb)->tskey);
2702 static inline bool sk_is_tcp(const struct sock *sk)
2704 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP;
2708 * sk_eat_skb - Release a skb if it is no longer needed
2709 * @sk: socket to eat this skb from
2710 * @skb: socket buffer to eat
2712 * This routine must be called with interrupts disabled or with the socket
2713 * locked so that the sk_buff queue operation is ok.
2715 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2717 __skb_unlink(skb, &sk->sk_receive_queue);
2722 skb_sk_is_prefetched(struct sk_buff *skb)
2725 return skb->destructor == sock_pfree;
2728 #endif /* CONFIG_INET */
2731 /* This helper checks if a socket is a full socket,
2732 * ie _not_ a timewait or request socket.
2734 static inline bool sk_fullsock(const struct sock *sk)
2736 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2740 sk_is_refcounted(struct sock *sk)
2742 /* Only full sockets have sk->sk_flags. */
2743 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2747 * skb_steal_sock - steal a socket from an sk_buff
2748 * @skb: sk_buff to steal the socket from
2749 * @refcounted: is set to true if the socket is reference-counted
2751 static inline struct sock *
2752 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2755 struct sock *sk = skb->sk;
2758 if (skb_sk_is_prefetched(skb))
2759 *refcounted = sk_is_refcounted(sk);
2760 skb->destructor = NULL;
2764 *refcounted = false;
2768 /* Checks if this SKB belongs to an HW offloaded socket
2769 * and whether any SW fallbacks are required based on dev.
2770 * Check decrypted mark in case skb_orphan() cleared socket.
2772 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2773 struct net_device *dev)
2775 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2776 struct sock *sk = skb->sk;
2778 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2779 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2780 #ifdef CONFIG_TLS_DEVICE
2781 } else if (unlikely(skb->decrypted)) {
2782 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2792 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2793 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2795 static inline bool sk_listener(const struct sock *sk)
2797 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2800 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2801 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2804 bool sk_ns_capable(const struct sock *sk,
2805 struct user_namespace *user_ns, int cap);
2806 bool sk_capable(const struct sock *sk, int cap);
2807 bool sk_net_capable(const struct sock *sk, int cap);
2809 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2811 /* Take into consideration the size of the struct sk_buff overhead in the
2812 * determination of these values, since that is non-constant across
2813 * platforms. This makes socket queueing behavior and performance
2814 * not depend upon such differences.
2816 #define _SK_MEM_PACKETS 256
2817 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2818 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2819 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2821 extern __u32 sysctl_wmem_max;
2822 extern __u32 sysctl_rmem_max;
2824 extern int sysctl_tstamp_allow_data;
2825 extern int sysctl_optmem_max;
2827 extern __u32 sysctl_wmem_default;
2828 extern __u32 sysctl_rmem_default;
2830 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2831 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2833 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2835 /* Does this proto have per netns sysctl_wmem ? */
2836 if (proto->sysctl_wmem_offset)
2837 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2839 return *proto->sysctl_wmem;
2842 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2844 /* Does this proto have per netns sysctl_rmem ? */
2845 if (proto->sysctl_rmem_offset)
2846 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2848 return *proto->sysctl_rmem;
2851 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2852 * Some wifi drivers need to tweak it to get more chunks.
2853 * They can use this helper from their ndo_start_xmit()
2855 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2857 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2859 WRITE_ONCE(sk->sk_pacing_shift, val);
2862 /* if a socket is bound to a device, check that the given device
2863 * index is either the same or that the socket is bound to an L3
2864 * master device and the given device index is also enslaved to
2867 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2871 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2874 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2875 if (mdif && mdif == sk->sk_bound_dev_if)
2881 void sock_def_readable(struct sock *sk);
2883 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2884 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2885 int sock_set_timestamping(struct sock *sk, int optname,
2886 struct so_timestamping timestamping);
2888 void sock_enable_timestamps(struct sock *sk);
2889 void sock_no_linger(struct sock *sk);
2890 void sock_set_keepalive(struct sock *sk);
2891 void sock_set_priority(struct sock *sk, u32 priority);
2892 void sock_set_rcvbuf(struct sock *sk, int val);
2893 void sock_set_mark(struct sock *sk, u32 val);
2894 void sock_set_reuseaddr(struct sock *sk);
2895 void sock_set_reuseport(struct sock *sk);
2896 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2898 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2900 int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2901 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2902 sockptr_t optval, int optlen, bool old_timeval);
2904 static inline bool sk_is_readable(struct sock *sk)
2906 if (sk->sk_prot->sock_is_readable)
2907 return sk->sk_prot->sock_is_readable(sk);
2910 #endif /* _SOCK_H */