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/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.
165 __addrpair skc_addrpair;
168 __be32 skc_rcv_saddr;
172 unsigned int skc_hash;
173 __u16 skc_u16hashes[2];
175 /* skc_dport && skc_num must be grouped as well */
177 __portpair skc_portpair;
184 unsigned short skc_family;
185 volatile unsigned char skc_state;
186 unsigned char skc_reuse:4;
187 unsigned char skc_reuseport:1;
188 unsigned char skc_ipv6only:1;
189 unsigned char skc_net_refcnt:1;
190 int skc_bound_dev_if;
192 struct hlist_node skc_bind_node;
193 struct hlist_node skc_portaddr_node;
195 struct proto *skc_prot;
196 possible_net_t skc_net;
198 #if IS_ENABLED(CONFIG_IPV6)
199 struct in6_addr skc_v6_daddr;
200 struct in6_addr skc_v6_rcv_saddr;
203 atomic64_t skc_cookie;
205 /* following fields are padding to force
206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
207 * assuming IPV6 is enabled. We use this padding differently
208 * for different kind of 'sockets'
211 unsigned long skc_flags;
212 struct sock *skc_listener; /* request_sock */
213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
216 * fields between dontcopy_begin/dontcopy_end
217 * are not copied in sock_copy()
220 int skc_dontcopy_begin[0];
223 struct hlist_node skc_node;
224 struct hlist_nulls_node skc_nulls_node;
226 unsigned short skc_tx_queue_mapping;
227 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
228 unsigned short skc_rx_queue_mapping;
231 int skc_incoming_cpu;
233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
236 refcount_t skc_refcnt;
238 int skc_dontcopy_end[0];
241 u32 skc_window_clamp;
242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
247 struct bpf_local_storage;
251 * struct sock - network layer representation of sockets
252 * @__sk_common: shared layout with inet_timewait_sock
253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
255 * @sk_lock: synchronizer
256 * @sk_kern_sock: True if sock is using kernel lock classes
257 * @sk_rcvbuf: size of receive buffer in bytes
258 * @sk_wq: sock wait queue and async head
259 * @sk_rx_dst: receive input route used by early demux
260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
262 * @sk_dst_cache: destination cache
263 * @sk_dst_pending_confirm: need to confirm neighbour
264 * @sk_policy: flow policy
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_reserved_mem: space reserved and non-reclaimable for the socket
273 * @sk_napi_id: id of the last napi context to receive data for sk
274 * @sk_ll_usec: usecs to busypoll when there is no data
275 * @sk_allocation: allocation mode
276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
279 * @sk_sndbuf: size of send buffer in bytes
280 * @__sk_flags_offset: empty field used to determine location of bitfield
281 * @sk_padding: unused element for alignment
282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
283 * @sk_no_check_rx: allow zero checksum in RX packets
284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
287 * @sk_gso_max_size: Maximum GSO segment size to build
288 * @sk_gso_max_segs: Maximum number of GSO segments
289 * @sk_pacing_shift: scaling factor for TCP Small Queues
290 * @sk_lingertime: %SO_LINGER l_linger setting
291 * @sk_backlog: always used with the per-socket spinlock held
292 * @sk_callback_lock: used with the callbacks in the end of this struct
293 * @sk_error_queue: rarely used
294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
295 * IPV6_ADDRFORM for instance)
296 * @sk_err: last error
297 * @sk_err_soft: errors that don't cause failure but are the cause of a
298 * persistent failure not just 'timed out'
299 * @sk_drops: raw/udp drops counter
300 * @sk_ack_backlog: current listen backlog
301 * @sk_max_ack_backlog: listen backlog set in listen()
302 * @sk_uid: user id of owner
303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
304 * @sk_busy_poll_budget: napi processing budget when busypolling
305 * @sk_priority: %SO_PRIORITY setting
306 * @sk_type: socket type (%SOCK_STREAM, etc)
307 * @sk_protocol: which protocol this socket belongs in this network family
308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
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_txrehash: enable TX hash rethink
316 * @sk_filter: socket filtering instructions
317 * @sk_timer: sock cleanup timer
318 * @sk_stamp: time stamp of last packet received
319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
320 * @sk_tsflags: SO_TIMESTAMPING flags
321 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
323 * @sk_tskey: counter to disambiguate concurrent tstamp requests
324 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
325 * @sk_socket: Identd and reporting IO signals
326 * @sk_user_data: RPC layer private data
327 * @sk_frag: cached page frag
328 * @sk_peek_off: current peek_offset value
329 * @sk_send_head: front of stuff to transmit
330 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
331 * @sk_security: used by security modules
332 * @sk_mark: generic packet mark
333 * @sk_cgrp_data: cgroup data for this cgroup
334 * @sk_memcg: this socket's memory cgroup association
335 * @sk_write_pending: a write to stream socket waits to start
336 * @sk_state_change: callback to indicate change in the state of the sock
337 * @sk_data_ready: callback to indicate there is data to be processed
338 * @sk_write_space: callback to indicate there is bf sending space available
339 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
340 * @sk_backlog_rcv: callback to process the backlog
341 * @sk_validate_xmit_skb: ptr to an optional validate function
342 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
343 * @sk_reuseport_cb: reuseport group container
344 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
345 * @sk_rcu: used during RCU grace period
346 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
347 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
348 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
349 * @sk_txtime_unused: unused txtime flags
350 * @ns_tracker: tracker for netns reference
354 * Now struct inet_timewait_sock also uses sock_common, so please just
355 * don't add nothing before this first member (__sk_common) --acme
357 struct sock_common __sk_common;
358 #define sk_node __sk_common.skc_node
359 #define sk_nulls_node __sk_common.skc_nulls_node
360 #define sk_refcnt __sk_common.skc_refcnt
361 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
362 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
363 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
366 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
367 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
368 #define sk_hash __sk_common.skc_hash
369 #define sk_portpair __sk_common.skc_portpair
370 #define sk_num __sk_common.skc_num
371 #define sk_dport __sk_common.skc_dport
372 #define sk_addrpair __sk_common.skc_addrpair
373 #define sk_daddr __sk_common.skc_daddr
374 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
375 #define sk_family __sk_common.skc_family
376 #define sk_state __sk_common.skc_state
377 #define sk_reuse __sk_common.skc_reuse
378 #define sk_reuseport __sk_common.skc_reuseport
379 #define sk_ipv6only __sk_common.skc_ipv6only
380 #define sk_net_refcnt __sk_common.skc_net_refcnt
381 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
382 #define sk_bind_node __sk_common.skc_bind_node
383 #define sk_prot __sk_common.skc_prot
384 #define sk_net __sk_common.skc_net
385 #define sk_v6_daddr __sk_common.skc_v6_daddr
386 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
387 #define sk_cookie __sk_common.skc_cookie
388 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
389 #define sk_flags __sk_common.skc_flags
390 #define sk_rxhash __sk_common.skc_rxhash
392 /* early demux fields */
393 struct dst_entry __rcu *sk_rx_dst;
394 int sk_rx_dst_ifindex;
395 u32 sk_rx_dst_cookie;
397 socket_lock_t sk_lock;
400 struct sk_buff_head sk_error_queue;
401 struct sk_buff_head sk_receive_queue;
403 * The backlog queue is special, it is always used with
404 * the per-socket spinlock held and requires low latency
405 * access. Therefore we special case it's implementation.
406 * Note : rmem_alloc is in this structure to fill a hole
407 * on 64bit arches, not because its logically part of
413 struct sk_buff *head;
414 struct sk_buff *tail;
417 #define sk_rmem_alloc sk_backlog.rmem_alloc
419 int sk_forward_alloc;
421 #ifdef CONFIG_NET_RX_BUSY_POLL
422 unsigned int sk_ll_usec;
423 /* ===== mostly read cache line ===== */
424 unsigned int sk_napi_id;
428 struct sk_filter __rcu *sk_filter;
430 struct socket_wq __rcu *sk_wq;
432 struct socket_wq *sk_wq_raw;
436 struct xfrm_policy __rcu *sk_policy[2];
439 struct dst_entry __rcu *sk_dst_cache;
440 atomic_t sk_omem_alloc;
443 /* ===== cache line for TX ===== */
445 refcount_t sk_wmem_alloc;
446 unsigned long sk_tsq_flags;
448 struct sk_buff *sk_send_head;
449 struct rb_root tcp_rtx_queue;
451 struct sk_buff_head sk_write_queue;
453 int sk_write_pending;
454 __u32 sk_dst_pending_confirm;
455 u32 sk_pacing_status; /* see enum sk_pacing */
457 struct timer_list sk_timer;
460 unsigned long sk_pacing_rate; /* bytes per second */
461 unsigned long sk_max_pacing_rate;
462 struct page_frag sk_frag;
463 netdev_features_t sk_route_caps;
465 unsigned int sk_gso_max_size;
470 * Because of non atomicity rules, all
471 * changes are protected by socket lock.
473 u8 sk_gso_disabled : 1,
482 unsigned long sk_lingertime;
483 struct proto *sk_prot_creator;
484 rwlock_t sk_callback_lock;
488 u32 sk_max_ack_backlog;
491 #ifdef CONFIG_NET_RX_BUSY_POLL
492 u8 sk_prefer_busy_poll;
493 u16 sk_busy_poll_budget;
495 spinlock_t sk_peer_lock;
497 struct pid *sk_peer_pid;
498 const struct cred *sk_peer_cred;
502 #if BITS_PER_LONG==32
503 seqlock_t sk_stamp_seq;
511 u8 sk_txtime_deadline_mode : 1,
512 sk_txtime_report_errors : 1,
513 sk_txtime_unused : 6;
515 struct socket *sk_socket;
517 #ifdef CONFIG_SECURITY
520 struct sock_cgroup_data sk_cgrp_data;
521 struct mem_cgroup *sk_memcg;
522 void (*sk_state_change)(struct sock *sk);
523 void (*sk_data_ready)(struct sock *sk);
524 void (*sk_write_space)(struct sock *sk);
525 void (*sk_error_report)(struct sock *sk);
526 int (*sk_backlog_rcv)(struct sock *sk,
527 struct sk_buff *skb);
528 #ifdef CONFIG_SOCK_VALIDATE_XMIT
529 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
530 struct net_device *dev,
531 struct sk_buff *skb);
533 void (*sk_destruct)(struct sock *sk);
534 struct sock_reuseport __rcu *sk_reuseport_cb;
535 #ifdef CONFIG_BPF_SYSCALL
536 struct bpf_local_storage __rcu *sk_bpf_storage;
538 struct rcu_head sk_rcu;
539 netns_tracker ns_tracker;
544 SK_PACING_NEEDED = 1,
548 /* Pointer stored in sk_user_data might not be suitable for copying
549 * when cloning the socket. For instance, it can point to a reference
550 * counted object. sk_user_data bottom bit is set if pointer must not
553 #define SK_USER_DATA_NOCOPY 1UL
554 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */
555 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF)
558 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
561 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
563 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
566 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
568 #define rcu_dereference_sk_user_data(sk) \
570 void *__tmp = rcu_dereference(__sk_user_data((sk))); \
571 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \
573 #define rcu_assign_sk_user_data(sk, ptr) \
575 uintptr_t __tmp = (uintptr_t)(ptr); \
576 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
577 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \
579 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \
581 uintptr_t __tmp = (uintptr_t)(ptr); \
582 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
583 rcu_assign_pointer(__sk_user_data((sk)), \
584 __tmp | SK_USER_DATA_NOCOPY); \
588 struct net *sock_net(const struct sock *sk)
590 return read_pnet(&sk->sk_net);
594 void sock_net_set(struct sock *sk, struct net *net)
596 write_pnet(&sk->sk_net, net);
600 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
601 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
602 * on a socket means that the socket will reuse everybody else's port
603 * without looking at the other's sk_reuse value.
606 #define SK_NO_REUSE 0
607 #define SK_CAN_REUSE 1
608 #define SK_FORCE_REUSE 2
610 int sk_set_peek_off(struct sock *sk, int val);
612 static inline int sk_peek_offset(const struct sock *sk, int flags)
614 if (unlikely(flags & MSG_PEEK)) {
615 return READ_ONCE(sk->sk_peek_off);
621 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
623 s32 off = READ_ONCE(sk->sk_peek_off);
625 if (unlikely(off >= 0)) {
626 off = max_t(s32, off - val, 0);
627 WRITE_ONCE(sk->sk_peek_off, off);
631 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
633 sk_peek_offset_bwd(sk, -val);
637 * Hashed lists helper routines
639 static inline struct sock *sk_entry(const struct hlist_node *node)
641 return hlist_entry(node, struct sock, sk_node);
644 static inline struct sock *__sk_head(const struct hlist_head *head)
646 return hlist_entry(head->first, struct sock, sk_node);
649 static inline struct sock *sk_head(const struct hlist_head *head)
651 return hlist_empty(head) ? NULL : __sk_head(head);
654 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
656 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
659 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
661 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
664 static inline struct sock *sk_next(const struct sock *sk)
666 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
669 static inline struct sock *sk_nulls_next(const struct sock *sk)
671 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
672 hlist_nulls_entry(sk->sk_nulls_node.next,
673 struct sock, sk_nulls_node) :
677 static inline bool sk_unhashed(const struct sock *sk)
679 return hlist_unhashed(&sk->sk_node);
682 static inline bool sk_hashed(const struct sock *sk)
684 return !sk_unhashed(sk);
687 static inline void sk_node_init(struct hlist_node *node)
692 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
697 static inline void __sk_del_node(struct sock *sk)
699 __hlist_del(&sk->sk_node);
702 /* NB: equivalent to hlist_del_init_rcu */
703 static inline bool __sk_del_node_init(struct sock *sk)
707 sk_node_init(&sk->sk_node);
713 /* Grab socket reference count. This operation is valid only
714 when sk is ALREADY grabbed f.e. it is found in hash table
715 or a list and the lookup is made under lock preventing hash table
719 static __always_inline void sock_hold(struct sock *sk)
721 refcount_inc(&sk->sk_refcnt);
724 /* Ungrab socket in the context, which assumes that socket refcnt
725 cannot hit zero, f.e. it is true in context of any socketcall.
727 static __always_inline void __sock_put(struct sock *sk)
729 refcount_dec(&sk->sk_refcnt);
732 static inline bool sk_del_node_init(struct sock *sk)
734 bool rc = __sk_del_node_init(sk);
737 /* paranoid for a while -acme */
738 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
743 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
745 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
748 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
754 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
756 bool rc = __sk_nulls_del_node_init_rcu(sk);
759 /* paranoid for a while -acme */
760 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
766 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
768 hlist_add_head(&sk->sk_node, list);
771 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
774 __sk_add_node(sk, list);
777 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
780 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
781 sk->sk_family == AF_INET6)
782 hlist_add_tail_rcu(&sk->sk_node, list);
784 hlist_add_head_rcu(&sk->sk_node, list);
787 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
790 hlist_add_tail_rcu(&sk->sk_node, list);
793 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
795 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
798 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
800 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
803 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
806 __sk_nulls_add_node_rcu(sk, list);
809 static inline void __sk_del_bind_node(struct sock *sk)
811 __hlist_del(&sk->sk_bind_node);
814 static inline void sk_add_bind_node(struct sock *sk,
815 struct hlist_head *list)
817 hlist_add_head(&sk->sk_bind_node, list);
820 #define sk_for_each(__sk, list) \
821 hlist_for_each_entry(__sk, list, sk_node)
822 #define sk_for_each_rcu(__sk, list) \
823 hlist_for_each_entry_rcu(__sk, list, sk_node)
824 #define sk_nulls_for_each(__sk, node, list) \
825 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
826 #define sk_nulls_for_each_rcu(__sk, node, list) \
827 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
828 #define sk_for_each_from(__sk) \
829 hlist_for_each_entry_from(__sk, sk_node)
830 #define sk_nulls_for_each_from(__sk, node) \
831 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
832 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
833 #define sk_for_each_safe(__sk, tmp, list) \
834 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
835 #define sk_for_each_bound(__sk, list) \
836 hlist_for_each_entry(__sk, list, sk_bind_node)
839 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
840 * @tpos: the type * to use as a loop cursor.
841 * @pos: the &struct hlist_node to use as a loop cursor.
842 * @head: the head for your list.
843 * @offset: offset of hlist_node within the struct.
846 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
847 for (pos = rcu_dereference(hlist_first_rcu(head)); \
849 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
850 pos = rcu_dereference(hlist_next_rcu(pos)))
852 static inline struct user_namespace *sk_user_ns(const struct sock *sk)
854 /* Careful only use this in a context where these parameters
855 * can not change and must all be valid, such as recvmsg from
858 return sk->sk_socket->file->f_cred->user_ns;
872 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
873 SOCK_DBG, /* %SO_DEBUG setting */
874 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
875 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
876 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
877 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
878 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
879 SOCK_FASYNC, /* fasync() active */
881 SOCK_ZEROCOPY, /* buffers from userspace */
882 SOCK_WIFI_STATUS, /* push wifi status to userspace */
883 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
884 * Will use last 4 bytes of packet sent from
885 * user-space instead.
887 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
888 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
889 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
891 SOCK_XDP, /* XDP is attached */
892 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
893 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */
896 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
898 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
900 nsk->sk_flags = osk->sk_flags;
903 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
905 __set_bit(flag, &sk->sk_flags);
908 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
910 __clear_bit(flag, &sk->sk_flags);
913 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
917 sock_set_flag(sk, bit);
919 sock_reset_flag(sk, bit);
922 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
924 return test_bit(flag, &sk->sk_flags);
928 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
929 static inline int sk_memalloc_socks(void)
931 return static_branch_unlikely(&memalloc_socks_key);
934 void __receive_sock(struct file *file);
937 static inline int sk_memalloc_socks(void)
942 static inline void __receive_sock(struct file *file)
946 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
948 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
951 static inline void sk_acceptq_removed(struct sock *sk)
953 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
956 static inline void sk_acceptq_added(struct sock *sk)
958 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
961 /* Note: If you think the test should be:
962 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
963 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
965 static inline bool sk_acceptq_is_full(const struct sock *sk)
967 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
971 * Compute minimal free write space needed to queue new packets.
973 static inline int sk_stream_min_wspace(const struct sock *sk)
975 return READ_ONCE(sk->sk_wmem_queued) >> 1;
978 static inline int sk_stream_wspace(const struct sock *sk)
980 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
983 static inline void sk_wmem_queued_add(struct sock *sk, int val)
985 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
988 void sk_stream_write_space(struct sock *sk);
990 /* OOB backlog add */
991 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
993 /* dont let skb dst not refcounted, we are going to leave rcu lock */
996 if (!sk->sk_backlog.tail)
997 WRITE_ONCE(sk->sk_backlog.head, skb);
999 sk->sk_backlog.tail->next = skb;
1001 WRITE_ONCE(sk->sk_backlog.tail, skb);
1006 * Take into account size of receive queue and backlog queue
1007 * Do not take into account this skb truesize,
1008 * to allow even a single big packet to come.
1010 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1012 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1014 return qsize > limit;
1017 /* The per-socket spinlock must be held here. */
1018 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1021 if (sk_rcvqueues_full(sk, limit))
1025 * If the skb was allocated from pfmemalloc reserves, only
1026 * allow SOCK_MEMALLOC sockets to use it as this socket is
1027 * helping free memory
1029 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1032 __sk_add_backlog(sk, skb);
1033 sk->sk_backlog.len += skb->truesize;
1037 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1039 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1040 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1042 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1044 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1045 return __sk_backlog_rcv(sk, skb);
1047 return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1053 static inline void sk_incoming_cpu_update(struct sock *sk)
1055 int cpu = raw_smp_processor_id();
1057 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1058 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1061 static inline void sock_rps_record_flow_hash(__u32 hash)
1064 struct rps_sock_flow_table *sock_flow_table;
1067 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1068 rps_record_sock_flow(sock_flow_table, hash);
1073 static inline void sock_rps_record_flow(const struct sock *sk)
1076 if (static_branch_unlikely(&rfs_needed)) {
1077 /* Reading sk->sk_rxhash might incur an expensive cache line
1080 * TCP_ESTABLISHED does cover almost all states where RFS
1081 * might be useful, and is cheaper [1] than testing :
1082 * IPv4: inet_sk(sk)->inet_daddr
1083 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1084 * OR an additional socket flag
1085 * [1] : sk_state and sk_prot are in the same cache line.
1087 if (sk->sk_state == TCP_ESTABLISHED)
1088 sock_rps_record_flow_hash(sk->sk_rxhash);
1093 static inline void sock_rps_save_rxhash(struct sock *sk,
1094 const struct sk_buff *skb)
1097 if (unlikely(sk->sk_rxhash != skb->hash))
1098 sk->sk_rxhash = skb->hash;
1102 static inline void sock_rps_reset_rxhash(struct sock *sk)
1109 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1111 release_sock(__sk); \
1112 __rc = __condition; \
1114 *(__timeo) = wait_woken(__wait, \
1115 TASK_INTERRUPTIBLE, \
1118 sched_annotate_sleep(); \
1120 __rc = __condition; \
1124 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1125 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1126 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1127 int sk_stream_error(struct sock *sk, int flags, int err);
1128 void sk_stream_kill_queues(struct sock *sk);
1129 void sk_set_memalloc(struct sock *sk);
1130 void sk_clear_memalloc(struct sock *sk);
1132 void __sk_flush_backlog(struct sock *sk);
1134 static inline bool sk_flush_backlog(struct sock *sk)
1136 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1137 __sk_flush_backlog(sk);
1143 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1145 struct request_sock_ops;
1146 struct timewait_sock_ops;
1147 struct inet_hashinfo;
1148 struct raw_hashinfo;
1149 struct smc_hashinfo;
1154 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1155 * un-modified. Special care is taken when initializing object to zero.
1157 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1159 if (offsetof(struct sock, sk_node.next) != 0)
1160 memset(sk, 0, offsetof(struct sock, sk_node.next));
1161 memset(&sk->sk_node.pprev, 0,
1162 size - offsetof(struct sock, sk_node.pprev));
1165 /* Networking protocol blocks we attach to sockets.
1166 * socket layer -> transport layer interface
1169 void (*close)(struct sock *sk,
1171 int (*pre_connect)(struct sock *sk,
1172 struct sockaddr *uaddr,
1174 int (*connect)(struct sock *sk,
1175 struct sockaddr *uaddr,
1177 int (*disconnect)(struct sock *sk, int flags);
1179 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1182 int (*ioctl)(struct sock *sk, int cmd,
1184 int (*init)(struct sock *sk);
1185 void (*destroy)(struct sock *sk);
1186 void (*shutdown)(struct sock *sk, int how);
1187 int (*setsockopt)(struct sock *sk, int level,
1188 int optname, sockptr_t optval,
1189 unsigned int optlen);
1190 int (*getsockopt)(struct sock *sk, int level,
1191 int optname, char __user *optval,
1192 int __user *option);
1193 void (*keepalive)(struct sock *sk, int valbool);
1194 #ifdef CONFIG_COMPAT
1195 int (*compat_ioctl)(struct sock *sk,
1196 unsigned int cmd, unsigned long arg);
1198 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1200 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1201 size_t len, int flags, int *addr_len);
1202 int (*sendpage)(struct sock *sk, struct page *page,
1203 int offset, size_t size, int flags);
1204 int (*bind)(struct sock *sk,
1205 struct sockaddr *addr, int addr_len);
1206 int (*bind_add)(struct sock *sk,
1207 struct sockaddr *addr, int addr_len);
1209 int (*backlog_rcv) (struct sock *sk,
1210 struct sk_buff *skb);
1211 bool (*bpf_bypass_getsockopt)(int level,
1214 void (*release_cb)(struct sock *sk);
1216 /* Keeping track of sk's, looking them up, and port selection methods. */
1217 int (*hash)(struct sock *sk);
1218 void (*unhash)(struct sock *sk);
1219 void (*rehash)(struct sock *sk);
1220 int (*get_port)(struct sock *sk, unsigned short snum);
1221 void (*put_port)(struct sock *sk);
1222 #ifdef CONFIG_BPF_SYSCALL
1223 int (*psock_update_sk_prot)(struct sock *sk,
1224 struct sk_psock *psock,
1228 /* Keeping track of sockets in use */
1229 #ifdef CONFIG_PROC_FS
1230 unsigned int inuse_idx;
1233 #if IS_ENABLED(CONFIG_MPTCP)
1234 int (*forward_alloc_get)(const struct sock *sk);
1237 bool (*stream_memory_free)(const struct sock *sk, int wake);
1238 bool (*sock_is_readable)(struct sock *sk);
1239 /* Memory pressure */
1240 void (*enter_memory_pressure)(struct sock *sk);
1241 void (*leave_memory_pressure)(struct sock *sk);
1242 atomic_long_t *memory_allocated; /* Current allocated memory. */
1243 int __percpu *per_cpu_fw_alloc;
1244 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1247 * Pressure flag: try to collapse.
1248 * Technical note: it is used by multiple contexts non atomically.
1249 * All the __sk_mem_schedule() is of this nature: accounting
1250 * is strict, actions are advisory and have some latency.
1252 unsigned long *memory_pressure;
1257 u32 sysctl_wmem_offset;
1258 u32 sysctl_rmem_offset;
1263 struct kmem_cache *slab;
1264 unsigned int obj_size;
1265 slab_flags_t slab_flags;
1266 unsigned int useroffset; /* Usercopy region offset */
1267 unsigned int usersize; /* Usercopy region size */
1269 unsigned int __percpu *orphan_count;
1271 struct request_sock_ops *rsk_prot;
1272 struct timewait_sock_ops *twsk_prot;
1275 struct inet_hashinfo *hashinfo;
1276 struct udp_table *udp_table;
1277 struct raw_hashinfo *raw_hash;
1278 struct smc_hashinfo *smc_hash;
1281 struct module *owner;
1285 struct list_head node;
1286 #ifdef SOCK_REFCNT_DEBUG
1289 int (*diag_destroy)(struct sock *sk, int err);
1290 } __randomize_layout;
1292 int proto_register(struct proto *prot, int alloc_slab);
1293 void proto_unregister(struct proto *prot);
1294 int sock_load_diag_module(int family, int protocol);
1296 #ifdef SOCK_REFCNT_DEBUG
1297 static inline void sk_refcnt_debug_inc(struct sock *sk)
1299 atomic_inc(&sk->sk_prot->socks);
1302 static inline void sk_refcnt_debug_dec(struct sock *sk)
1304 atomic_dec(&sk->sk_prot->socks);
1305 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1306 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1309 static inline void sk_refcnt_debug_release(const struct sock *sk)
1311 if (refcount_read(&sk->sk_refcnt) != 1)
1312 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1313 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1315 #else /* SOCK_REFCNT_DEBUG */
1316 #define sk_refcnt_debug_inc(sk) do { } while (0)
1317 #define sk_refcnt_debug_dec(sk) do { } while (0)
1318 #define sk_refcnt_debug_release(sk) do { } while (0)
1319 #endif /* SOCK_REFCNT_DEBUG */
1321 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1323 static inline int sk_forward_alloc_get(const struct sock *sk)
1325 #if IS_ENABLED(CONFIG_MPTCP)
1326 if (sk->sk_prot->forward_alloc_get)
1327 return sk->sk_prot->forward_alloc_get(sk);
1329 return sk->sk_forward_alloc;
1332 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1334 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1337 return sk->sk_prot->stream_memory_free ?
1338 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1339 tcp_stream_memory_free, sk, wake) : true;
1342 static inline bool sk_stream_memory_free(const struct sock *sk)
1344 return __sk_stream_memory_free(sk, 0);
1347 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1349 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1350 __sk_stream_memory_free(sk, wake);
1353 static inline bool sk_stream_is_writeable(const struct sock *sk)
1355 return __sk_stream_is_writeable(sk, 0);
1358 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1359 struct cgroup *ancestor)
1361 #ifdef CONFIG_SOCK_CGROUP_DATA
1362 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1369 static inline bool sk_has_memory_pressure(const struct sock *sk)
1371 return sk->sk_prot->memory_pressure != NULL;
1374 static inline bool sk_under_memory_pressure(const struct sock *sk)
1376 if (!sk->sk_prot->memory_pressure)
1379 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1380 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1383 return !!*sk->sk_prot->memory_pressure;
1387 proto_memory_allocated(const struct proto *prot)
1389 return max(0L, atomic_long_read(prot->memory_allocated));
1393 sk_memory_allocated(const struct sock *sk)
1395 return proto_memory_allocated(sk->sk_prot);
1398 /* 1 MB per cpu, in page units */
1399 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT))
1402 sk_memory_allocated_add(struct sock *sk, int amt)
1407 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1408 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) {
1409 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1410 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1416 sk_memory_allocated_sub(struct sock *sk, int amt)
1421 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1422 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) {
1423 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1424 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1429 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1431 static inline void sk_sockets_allocated_dec(struct sock *sk)
1433 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1434 SK_ALLOC_PERCPU_COUNTER_BATCH);
1437 static inline void sk_sockets_allocated_inc(struct sock *sk)
1439 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1440 SK_ALLOC_PERCPU_COUNTER_BATCH);
1444 sk_sockets_allocated_read_positive(struct sock *sk)
1446 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1450 proto_sockets_allocated_sum_positive(struct proto *prot)
1452 return percpu_counter_sum_positive(prot->sockets_allocated);
1456 proto_memory_pressure(struct proto *prot)
1458 if (!prot->memory_pressure)
1460 return !!*prot->memory_pressure;
1464 #ifdef CONFIG_PROC_FS
1465 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1468 int val[PROTO_INUSE_NR];
1471 static inline void sock_prot_inuse_add(const struct net *net,
1472 const struct proto *prot, int val)
1474 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1477 static inline void sock_inuse_add(const struct net *net, int val)
1479 this_cpu_add(net->core.prot_inuse->all, val);
1482 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1483 int sock_inuse_get(struct net *net);
1485 static inline void sock_prot_inuse_add(const struct net *net,
1486 const struct proto *prot, int val)
1490 static inline void sock_inuse_add(const struct net *net, int val)
1496 /* With per-bucket locks this operation is not-atomic, so that
1497 * this version is not worse.
1499 static inline int __sk_prot_rehash(struct sock *sk)
1501 sk->sk_prot->unhash(sk);
1502 return sk->sk_prot->hash(sk);
1505 /* About 10 seconds */
1506 #define SOCK_DESTROY_TIME (10*HZ)
1508 /* Sockets 0-1023 can't be bound to unless you are superuser */
1509 #define PROT_SOCK 1024
1511 #define SHUTDOWN_MASK 3
1512 #define RCV_SHUTDOWN 1
1513 #define SEND_SHUTDOWN 2
1515 #define SOCK_BINDADDR_LOCK 4
1516 #define SOCK_BINDPORT_LOCK 8
1518 struct socket_alloc {
1519 struct socket socket;
1520 struct inode vfs_inode;
1523 static inline struct socket *SOCKET_I(struct inode *inode)
1525 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1528 static inline struct inode *SOCK_INODE(struct socket *socket)
1530 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1534 * Functions for memory accounting
1536 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1537 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1538 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1539 void __sk_mem_reclaim(struct sock *sk, int amount);
1541 #define SK_MEM_SEND 0
1542 #define SK_MEM_RECV 1
1544 /* sysctl_mem values are in pages */
1545 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1547 return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1550 static inline int sk_mem_pages(int amt)
1552 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1555 static inline bool sk_has_account(struct sock *sk)
1557 /* return true if protocol supports memory accounting */
1558 return !!sk->sk_prot->memory_allocated;
1561 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1565 if (!sk_has_account(sk))
1567 delta = size - sk->sk_forward_alloc;
1568 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1572 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1576 if (!sk_has_account(sk))
1578 delta = size - sk->sk_forward_alloc;
1579 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1580 skb_pfmemalloc(skb);
1583 static inline int sk_unused_reserved_mem(const struct sock *sk)
1587 if (likely(!sk->sk_reserved_mem))
1590 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1591 atomic_read(&sk->sk_rmem_alloc);
1593 return unused_mem > 0 ? unused_mem : 0;
1596 static inline void sk_mem_reclaim(struct sock *sk)
1600 if (!sk_has_account(sk))
1603 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1605 if (reclaimable >= (int)PAGE_SIZE)
1606 __sk_mem_reclaim(sk, reclaimable);
1609 static inline void sk_mem_reclaim_final(struct sock *sk)
1611 sk->sk_reserved_mem = 0;
1615 static inline void sk_mem_charge(struct sock *sk, int size)
1617 if (!sk_has_account(sk))
1619 sk->sk_forward_alloc -= size;
1622 static inline void sk_mem_uncharge(struct sock *sk, int size)
1624 if (!sk_has_account(sk))
1626 sk->sk_forward_alloc += size;
1631 * Macro so as to not evaluate some arguments when
1632 * lockdep is not enabled.
1634 * Mark both the sk_lock and the sk_lock.slock as a
1635 * per-address-family lock class.
1637 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1639 sk->sk_lock.owned = 0; \
1640 init_waitqueue_head(&sk->sk_lock.wq); \
1641 spin_lock_init(&(sk)->sk_lock.slock); \
1642 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1643 sizeof((sk)->sk_lock)); \
1644 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1646 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1649 static inline bool lockdep_sock_is_held(const struct sock *sk)
1651 return lockdep_is_held(&sk->sk_lock) ||
1652 lockdep_is_held(&sk->sk_lock.slock);
1655 void lock_sock_nested(struct sock *sk, int subclass);
1657 static inline void lock_sock(struct sock *sk)
1659 lock_sock_nested(sk, 0);
1662 void __lock_sock(struct sock *sk);
1663 void __release_sock(struct sock *sk);
1664 void release_sock(struct sock *sk);
1666 /* BH context may only use the following locking interface. */
1667 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1668 #define bh_lock_sock_nested(__sk) \
1669 spin_lock_nested(&((__sk)->sk_lock.slock), \
1670 SINGLE_DEPTH_NESTING)
1671 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1673 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1676 * lock_sock_fast - fast version of lock_sock
1679 * This version should be used for very small section, where process wont block
1680 * return false if fast path is taken:
1682 * sk_lock.slock locked, owned = 0, BH disabled
1684 * return true if slow path is taken:
1686 * sk_lock.slock unlocked, owned = 1, BH enabled
1688 static inline bool lock_sock_fast(struct sock *sk)
1690 /* The sk_lock has mutex_lock() semantics here. */
1691 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1693 return __lock_sock_fast(sk);
1696 /* fast socket lock variant for caller already holding a [different] socket lock */
1697 static inline bool lock_sock_fast_nested(struct sock *sk)
1699 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1701 return __lock_sock_fast(sk);
1705 * unlock_sock_fast - complement of lock_sock_fast
1709 * fast unlock socket for user context.
1710 * If slow mode is on, we call regular release_sock()
1712 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1713 __releases(&sk->sk_lock.slock)
1717 __release(&sk->sk_lock.slock);
1719 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1720 spin_unlock_bh(&sk->sk_lock.slock);
1724 /* Used by processes to "lock" a socket state, so that
1725 * interrupts and bottom half handlers won't change it
1726 * from under us. It essentially blocks any incoming
1727 * packets, so that we won't get any new data or any
1728 * packets that change the state of the socket.
1730 * While locked, BH processing will add new packets to
1731 * the backlog queue. This queue is processed by the
1732 * owner of the socket lock right before it is released.
1734 * Since ~2.3.5 it is also exclusive sleep lock serializing
1735 * accesses from user process context.
1738 static inline void sock_owned_by_me(const struct sock *sk)
1740 #ifdef CONFIG_LOCKDEP
1741 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1745 static inline bool sock_owned_by_user(const struct sock *sk)
1747 sock_owned_by_me(sk);
1748 return sk->sk_lock.owned;
1751 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1753 return sk->sk_lock.owned;
1756 static inline void sock_release_ownership(struct sock *sk)
1758 if (sock_owned_by_user_nocheck(sk)) {
1759 sk->sk_lock.owned = 0;
1761 /* The sk_lock has mutex_unlock() semantics: */
1762 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1766 /* no reclassification while locks are held */
1767 static inline bool sock_allow_reclassification(const struct sock *csk)
1769 struct sock *sk = (struct sock *)csk;
1771 return !sock_owned_by_user_nocheck(sk) &&
1772 !spin_is_locked(&sk->sk_lock.slock);
1775 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1776 struct proto *prot, int kern);
1777 void sk_free(struct sock *sk);
1778 void sk_destruct(struct sock *sk);
1779 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1780 void sk_free_unlock_clone(struct sock *sk);
1782 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1784 void __sock_wfree(struct sk_buff *skb);
1785 void sock_wfree(struct sk_buff *skb);
1786 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1788 void skb_orphan_partial(struct sk_buff *skb);
1789 void sock_rfree(struct sk_buff *skb);
1790 void sock_efree(struct sk_buff *skb);
1792 void sock_edemux(struct sk_buff *skb);
1793 void sock_pfree(struct sk_buff *skb);
1795 #define sock_edemux sock_efree
1798 int sock_setsockopt(struct socket *sock, int level, int op,
1799 sockptr_t optval, unsigned int optlen);
1801 int sock_getsockopt(struct socket *sock, int level, int op,
1802 char __user *optval, int __user *optlen);
1803 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1804 bool timeval, bool time32);
1805 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1806 unsigned long data_len, int noblock,
1807 int *errcode, int max_page_order);
1809 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1811 int noblock, int *errcode)
1813 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1816 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1817 void sock_kfree_s(struct sock *sk, void *mem, int size);
1818 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1819 void sk_send_sigurg(struct sock *sk);
1821 struct sockcm_cookie {
1827 static inline void sockcm_init(struct sockcm_cookie *sockc,
1828 const struct sock *sk)
1830 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1833 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1834 struct sockcm_cookie *sockc);
1835 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1836 struct sockcm_cookie *sockc);
1839 * Functions to fill in entries in struct proto_ops when a protocol
1840 * does not implement a particular function.
1842 int sock_no_bind(struct socket *, struct sockaddr *, int);
1843 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1844 int sock_no_socketpair(struct socket *, struct socket *);
1845 int sock_no_accept(struct socket *, struct socket *, int, bool);
1846 int sock_no_getname(struct socket *, struct sockaddr *, int);
1847 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1848 int sock_no_listen(struct socket *, int);
1849 int sock_no_shutdown(struct socket *, int);
1850 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1851 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1852 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1853 int sock_no_mmap(struct file *file, struct socket *sock,
1854 struct vm_area_struct *vma);
1855 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1856 size_t size, int flags);
1857 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1858 int offset, size_t size, int flags);
1861 * Functions to fill in entries in struct proto_ops when a protocol
1862 * uses the inet style.
1864 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1865 char __user *optval, int __user *optlen);
1866 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1868 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1869 sockptr_t optval, unsigned int optlen);
1871 void sk_common_release(struct sock *sk);
1874 * Default socket callbacks and setup code
1877 /* Initialise core socket variables */
1878 void sock_init_data(struct socket *sock, struct sock *sk);
1881 * Socket reference counting postulates.
1883 * * Each user of socket SHOULD hold a reference count.
1884 * * Each access point to socket (an hash table bucket, reference from a list,
1885 * running timer, skb in flight MUST hold a reference count.
1886 * * When reference count hits 0, it means it will never increase back.
1887 * * When reference count hits 0, it means that no references from
1888 * outside exist to this socket and current process on current CPU
1889 * is last user and may/should destroy this socket.
1890 * * sk_free is called from any context: process, BH, IRQ. When
1891 * it is called, socket has no references from outside -> sk_free
1892 * may release descendant resources allocated by the socket, but
1893 * to the time when it is called, socket is NOT referenced by any
1894 * hash tables, lists etc.
1895 * * Packets, delivered from outside (from network or from another process)
1896 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1897 * when they sit in queue. Otherwise, packets will leak to hole, when
1898 * socket is looked up by one cpu and unhasing is made by another CPU.
1899 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1900 * (leak to backlog). Packet socket does all the processing inside
1901 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1902 * use separate SMP lock, so that they are prone too.
1905 /* Ungrab socket and destroy it, if it was the last reference. */
1906 static inline void sock_put(struct sock *sk)
1908 if (refcount_dec_and_test(&sk->sk_refcnt))
1911 /* Generic version of sock_put(), dealing with all sockets
1912 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1914 void sock_gen_put(struct sock *sk);
1916 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1917 unsigned int trim_cap, bool refcounted);
1918 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1921 return __sk_receive_skb(sk, skb, nested, 1, true);
1924 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1926 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1927 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1929 sk->sk_tx_queue_mapping = tx_queue;
1932 #define NO_QUEUE_MAPPING USHRT_MAX
1934 static inline void sk_tx_queue_clear(struct sock *sk)
1936 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1939 static inline int sk_tx_queue_get(const struct sock *sk)
1941 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1942 return sk->sk_tx_queue_mapping;
1947 static inline void __sk_rx_queue_set(struct sock *sk,
1948 const struct sk_buff *skb,
1951 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1952 if (skb_rx_queue_recorded(skb)) {
1953 u16 rx_queue = skb_get_rx_queue(skb);
1956 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
1957 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
1962 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1964 __sk_rx_queue_set(sk, skb, true);
1967 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
1969 __sk_rx_queue_set(sk, skb, false);
1972 static inline void sk_rx_queue_clear(struct sock *sk)
1974 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1975 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
1979 static inline int sk_rx_queue_get(const struct sock *sk)
1981 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1983 int res = READ_ONCE(sk->sk_rx_queue_mapping);
1985 if (res != NO_QUEUE_MAPPING)
1993 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1995 sk->sk_socket = sock;
1998 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2000 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2001 return &rcu_dereference_raw(sk->sk_wq)->wait;
2003 /* Detach socket from process context.
2004 * Announce socket dead, detach it from wait queue and inode.
2005 * Note that parent inode held reference count on this struct sock,
2006 * we do not release it in this function, because protocol
2007 * probably wants some additional cleanups or even continuing
2008 * to work with this socket (TCP).
2010 static inline void sock_orphan(struct sock *sk)
2012 write_lock_bh(&sk->sk_callback_lock);
2013 sock_set_flag(sk, SOCK_DEAD);
2014 sk_set_socket(sk, NULL);
2016 write_unlock_bh(&sk->sk_callback_lock);
2019 static inline void sock_graft(struct sock *sk, struct socket *parent)
2021 WARN_ON(parent->sk);
2022 write_lock_bh(&sk->sk_callback_lock);
2023 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2025 sk_set_socket(sk, parent);
2026 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2027 security_sock_graft(sk, parent);
2028 write_unlock_bh(&sk->sk_callback_lock);
2031 kuid_t sock_i_uid(struct sock *sk);
2032 unsigned long sock_i_ino(struct sock *sk);
2034 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2036 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2039 static inline u32 net_tx_rndhash(void)
2041 u32 v = prandom_u32();
2046 static inline void sk_set_txhash(struct sock *sk)
2048 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2049 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2052 static inline bool sk_rethink_txhash(struct sock *sk)
2054 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2061 static inline struct dst_entry *
2062 __sk_dst_get(struct sock *sk)
2064 return rcu_dereference_check(sk->sk_dst_cache,
2065 lockdep_sock_is_held(sk));
2068 static inline struct dst_entry *
2069 sk_dst_get(struct sock *sk)
2071 struct dst_entry *dst;
2074 dst = rcu_dereference(sk->sk_dst_cache);
2075 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
2081 static inline void __dst_negative_advice(struct sock *sk)
2083 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2085 if (dst && dst->ops->negative_advice) {
2086 ndst = dst->ops->negative_advice(dst);
2089 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2090 sk_tx_queue_clear(sk);
2091 sk->sk_dst_pending_confirm = 0;
2096 static inline void dst_negative_advice(struct sock *sk)
2098 sk_rethink_txhash(sk);
2099 __dst_negative_advice(sk);
2103 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2105 struct dst_entry *old_dst;
2107 sk_tx_queue_clear(sk);
2108 sk->sk_dst_pending_confirm = 0;
2109 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2110 lockdep_sock_is_held(sk));
2111 rcu_assign_pointer(sk->sk_dst_cache, dst);
2112 dst_release(old_dst);
2116 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2118 struct dst_entry *old_dst;
2120 sk_tx_queue_clear(sk);
2121 sk->sk_dst_pending_confirm = 0;
2122 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2123 dst_release(old_dst);
2127 __sk_dst_reset(struct sock *sk)
2129 __sk_dst_set(sk, NULL);
2133 sk_dst_reset(struct sock *sk)
2135 sk_dst_set(sk, NULL);
2138 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2140 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2142 static inline void sk_dst_confirm(struct sock *sk)
2144 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2145 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2148 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2150 if (skb_get_dst_pending_confirm(skb)) {
2151 struct sock *sk = skb->sk;
2153 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2154 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2159 bool sk_mc_loop(struct sock *sk);
2161 static inline bool sk_can_gso(const struct sock *sk)
2163 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2166 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2168 static inline void sk_gso_disable(struct sock *sk)
2170 sk->sk_gso_disabled = 1;
2171 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2174 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2175 struct iov_iter *from, char *to,
2176 int copy, int offset)
2178 if (skb->ip_summed == CHECKSUM_NONE) {
2180 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2182 skb->csum = csum_block_add(skb->csum, csum, offset);
2183 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2184 if (!copy_from_iter_full_nocache(to, copy, from))
2186 } else if (!copy_from_iter_full(to, copy, from))
2192 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2193 struct iov_iter *from, int copy)
2195 int err, offset = skb->len;
2197 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2200 __skb_trim(skb, offset);
2205 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2206 struct sk_buff *skb,
2212 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2217 skb_len_add(skb, copy);
2218 sk_wmem_queued_add(sk, copy);
2219 sk_mem_charge(sk, copy);
2224 * sk_wmem_alloc_get - returns write allocations
2227 * Return: sk_wmem_alloc minus initial offset of one
2229 static inline int sk_wmem_alloc_get(const struct sock *sk)
2231 return refcount_read(&sk->sk_wmem_alloc) - 1;
2235 * sk_rmem_alloc_get - returns read allocations
2238 * Return: sk_rmem_alloc
2240 static inline int sk_rmem_alloc_get(const struct sock *sk)
2242 return atomic_read(&sk->sk_rmem_alloc);
2246 * sk_has_allocations - check if allocations are outstanding
2249 * Return: true if socket has write or read allocations
2251 static inline bool sk_has_allocations(const struct sock *sk)
2253 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2257 * skwq_has_sleeper - check if there are any waiting processes
2258 * @wq: struct socket_wq
2260 * Return: true if socket_wq has waiting processes
2262 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2263 * barrier call. They were added due to the race found within the tcp code.
2265 * Consider following tcp code paths::
2268 * sys_select receive packet
2270 * __add_wait_queue update tp->rcv_nxt
2272 * tp->rcv_nxt check sock_def_readable
2274 * schedule rcu_read_lock();
2275 * wq = rcu_dereference(sk->sk_wq);
2276 * if (wq && waitqueue_active(&wq->wait))
2277 * wake_up_interruptible(&wq->wait)
2281 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2282 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2283 * could then endup calling schedule and sleep forever if there are no more
2284 * data on the socket.
2287 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2289 return wq && wq_has_sleeper(&wq->wait);
2293 * sock_poll_wait - place memory barrier behind the poll_wait call.
2295 * @sock: socket to wait on
2298 * See the comments in the wq_has_sleeper function.
2300 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2303 if (!poll_does_not_wait(p)) {
2304 poll_wait(filp, &sock->wq.wait, p);
2305 /* We need to be sure we are in sync with the
2306 * socket flags modification.
2308 * This memory barrier is paired in the wq_has_sleeper.
2314 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2316 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2317 u32 txhash = READ_ONCE(sk->sk_txhash);
2325 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2328 * Queue a received datagram if it will fit. Stream and sequenced
2329 * protocols can't normally use this as they need to fit buffers in
2330 * and play with them.
2332 * Inlined as it's very short and called for pretty much every
2333 * packet ever received.
2335 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2339 skb->destructor = sock_rfree;
2340 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2341 sk_mem_charge(sk, skb->truesize);
2344 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2346 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2348 skb->destructor = sock_efree;
2355 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2357 if (skb->destructor != sock_wfree) {
2364 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2365 unsigned long expires);
2367 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2369 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2371 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2372 struct sk_buff *skb, unsigned int flags,
2373 void (*destructor)(struct sock *sk,
2374 struct sk_buff *skb));
2375 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2377 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2378 enum skb_drop_reason *reason);
2380 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2382 return sock_queue_rcv_skb_reason(sk, skb, NULL);
2385 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2386 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2389 * Recover an error report and clear atomically
2392 static inline int sock_error(struct sock *sk)
2396 /* Avoid an atomic operation for the common case.
2397 * This is racy since another cpu/thread can change sk_err under us.
2399 if (likely(data_race(!sk->sk_err)))
2402 err = xchg(&sk->sk_err, 0);
2406 void sk_error_report(struct sock *sk);
2408 static inline unsigned long sock_wspace(struct sock *sk)
2412 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2413 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2421 * We use sk->sk_wq_raw, from contexts knowing this
2422 * pointer is not NULL and cannot disappear/change.
2424 static inline void sk_set_bit(int nr, struct sock *sk)
2426 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2427 !sock_flag(sk, SOCK_FASYNC))
2430 set_bit(nr, &sk->sk_wq_raw->flags);
2433 static inline void sk_clear_bit(int nr, struct sock *sk)
2435 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2436 !sock_flag(sk, SOCK_FASYNC))
2439 clear_bit(nr, &sk->sk_wq_raw->flags);
2442 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2444 if (sock_flag(sk, SOCK_FASYNC)) {
2446 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2451 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2452 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2453 * Note: for send buffers, TCP works better if we can build two skbs at
2456 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2458 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2459 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2461 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2465 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2468 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2469 val = max_t(u32, val, sk_unused_reserved_mem(sk));
2471 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2475 * sk_page_frag - return an appropriate page_frag
2478 * Use the per task page_frag instead of the per socket one for
2479 * optimization when we know that we're in process context and own
2480 * everything that's associated with %current.
2482 * Both direct reclaim and page faults can nest inside other
2483 * socket operations and end up recursing into sk_page_frag()
2484 * while it's already in use: explicitly avoid task page_frag
2485 * usage if the caller is potentially doing any of them.
2486 * This assumes that page fault handlers use the GFP_NOFS flags.
2488 * Return: a per task page_frag if context allows that,
2489 * otherwise a per socket one.
2491 static inline struct page_frag *sk_page_frag(struct sock *sk)
2493 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) ==
2494 (__GFP_DIRECT_RECLAIM | __GFP_FS))
2495 return ¤t->task_frag;
2497 return &sk->sk_frag;
2500 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2503 * Default write policy as shown to user space via poll/select/SIGIO
2505 static inline bool sock_writeable(const struct sock *sk)
2507 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2510 static inline gfp_t gfp_any(void)
2512 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2515 static inline gfp_t gfp_memcg_charge(void)
2517 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL;
2520 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2522 return noblock ? 0 : sk->sk_rcvtimeo;
2525 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2527 return noblock ? 0 : sk->sk_sndtimeo;
2530 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2532 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2537 /* Alas, with timeout socket operations are not restartable.
2538 * Compare this to poll().
2540 static inline int sock_intr_errno(long timeo)
2542 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2545 struct sock_skb_cb {
2549 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2550 * using skb->cb[] would keep using it directly and utilize its
2551 * alignement guarantee.
2553 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2554 sizeof(struct sock_skb_cb)))
2556 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2557 SOCK_SKB_CB_OFFSET))
2559 #define sock_skb_cb_check_size(size) \
2560 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2563 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2565 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2566 atomic_read(&sk->sk_drops) : 0;
2569 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2571 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2573 atomic_add(segs, &sk->sk_drops);
2576 static inline ktime_t sock_read_timestamp(struct sock *sk)
2578 #if BITS_PER_LONG==32
2583 seq = read_seqbegin(&sk->sk_stamp_seq);
2585 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2589 return READ_ONCE(sk->sk_stamp);
2593 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2595 #if BITS_PER_LONG==32
2596 write_seqlock(&sk->sk_stamp_seq);
2598 write_sequnlock(&sk->sk_stamp_seq);
2600 WRITE_ONCE(sk->sk_stamp, kt);
2604 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2605 struct sk_buff *skb);
2606 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2607 struct sk_buff *skb);
2610 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2612 ktime_t kt = skb->tstamp;
2613 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2616 * generate control messages if
2617 * - receive time stamping in software requested
2618 * - software time stamp available and wanted
2619 * - hardware time stamps available and wanted
2621 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2622 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2623 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2624 (hwtstamps->hwtstamp &&
2625 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2626 __sock_recv_timestamp(msg, sk, skb);
2628 sock_write_timestamp(sk, kt);
2630 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2631 __sock_recv_wifi_status(msg, sk, skb);
2634 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2635 struct sk_buff *skb);
2637 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2638 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2639 struct sk_buff *skb)
2641 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \
2642 (1UL << SOCK_RCVTSTAMP) | \
2643 (1UL << SOCK_RCVMARK))
2644 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2645 SOF_TIMESTAMPING_RAW_HARDWARE)
2647 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY)
2648 __sock_recv_cmsgs(msg, sk, skb);
2649 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2650 sock_write_timestamp(sk, skb->tstamp);
2651 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2652 sock_write_timestamp(sk, 0);
2655 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2658 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2659 * @sk: socket sending this packet
2660 * @tsflags: timestamping flags to use
2661 * @tx_flags: completed with instructions for time stamping
2662 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2664 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2666 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2667 __u8 *tx_flags, __u32 *tskey)
2669 if (unlikely(tsflags)) {
2670 __sock_tx_timestamp(tsflags, tx_flags);
2671 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2672 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2673 *tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2675 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2676 *tx_flags |= SKBTX_WIFI_STATUS;
2679 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2682 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2685 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2687 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2688 &skb_shinfo(skb)->tskey);
2691 static inline bool sk_is_tcp(const struct sock *sk)
2693 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP;
2697 * sk_eat_skb - Release a skb if it is no longer needed
2698 * @sk: socket to eat this skb from
2699 * @skb: socket buffer to eat
2701 * This routine must be called with interrupts disabled or with the socket
2702 * locked so that the sk_buff queue operation is ok.
2704 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2706 __skb_unlink(skb, &sk->sk_receive_queue);
2711 skb_sk_is_prefetched(struct sk_buff *skb)
2714 return skb->destructor == sock_pfree;
2717 #endif /* CONFIG_INET */
2720 /* This helper checks if a socket is a full socket,
2721 * ie _not_ a timewait or request socket.
2723 static inline bool sk_fullsock(const struct sock *sk)
2725 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2729 sk_is_refcounted(struct sock *sk)
2731 /* Only full sockets have sk->sk_flags. */
2732 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2736 * skb_steal_sock - steal a socket from an sk_buff
2737 * @skb: sk_buff to steal the socket from
2738 * @refcounted: is set to true if the socket is reference-counted
2740 static inline struct sock *
2741 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2744 struct sock *sk = skb->sk;
2747 if (skb_sk_is_prefetched(skb))
2748 *refcounted = sk_is_refcounted(sk);
2749 skb->destructor = NULL;
2753 *refcounted = false;
2757 /* Checks if this SKB belongs to an HW offloaded socket
2758 * and whether any SW fallbacks are required based on dev.
2759 * Check decrypted mark in case skb_orphan() cleared socket.
2761 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2762 struct net_device *dev)
2764 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2765 struct sock *sk = skb->sk;
2767 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2768 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2769 #ifdef CONFIG_TLS_DEVICE
2770 } else if (unlikely(skb->decrypted)) {
2771 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2781 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2782 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2784 static inline bool sk_listener(const struct sock *sk)
2786 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2789 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2790 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2793 bool sk_ns_capable(const struct sock *sk,
2794 struct user_namespace *user_ns, int cap);
2795 bool sk_capable(const struct sock *sk, int cap);
2796 bool sk_net_capable(const struct sock *sk, int cap);
2798 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2800 /* Take into consideration the size of the struct sk_buff overhead in the
2801 * determination of these values, since that is non-constant across
2802 * platforms. This makes socket queueing behavior and performance
2803 * not depend upon such differences.
2805 #define _SK_MEM_PACKETS 256
2806 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2807 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2808 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2810 extern __u32 sysctl_wmem_max;
2811 extern __u32 sysctl_rmem_max;
2813 extern int sysctl_tstamp_allow_data;
2814 extern int sysctl_optmem_max;
2816 extern __u32 sysctl_wmem_default;
2817 extern __u32 sysctl_rmem_default;
2819 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2820 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2822 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2824 /* Does this proto have per netns sysctl_wmem ? */
2825 if (proto->sysctl_wmem_offset)
2826 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2828 return READ_ONCE(*proto->sysctl_wmem);
2831 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2833 /* Does this proto have per netns sysctl_rmem ? */
2834 if (proto->sysctl_rmem_offset)
2835 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2837 return READ_ONCE(*proto->sysctl_rmem);
2840 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2841 * Some wifi drivers need to tweak it to get more chunks.
2842 * They can use this helper from their ndo_start_xmit()
2844 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2846 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2848 WRITE_ONCE(sk->sk_pacing_shift, val);
2851 /* if a socket is bound to a device, check that the given device
2852 * index is either the same or that the socket is bound to an L3
2853 * master device and the given device index is also enslaved to
2856 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2858 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2861 if (!bound_dev_if || bound_dev_if == dif)
2864 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2865 if (mdif && mdif == bound_dev_if)
2871 void sock_def_readable(struct sock *sk);
2873 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2874 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2875 int sock_set_timestamping(struct sock *sk, int optname,
2876 struct so_timestamping timestamping);
2878 void sock_enable_timestamps(struct sock *sk);
2879 void sock_no_linger(struct sock *sk);
2880 void sock_set_keepalive(struct sock *sk);
2881 void sock_set_priority(struct sock *sk, u32 priority);
2882 void sock_set_rcvbuf(struct sock *sk, int val);
2883 void sock_set_mark(struct sock *sk, u32 val);
2884 void sock_set_reuseaddr(struct sock *sk);
2885 void sock_set_reuseport(struct sock *sk);
2886 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2888 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2890 int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2891 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2892 sockptr_t optval, int optlen, bool old_timeval);
2894 static inline bool sk_is_readable(struct sock *sk)
2896 if (sk->sk_prot->sock_is_readable)
2897 return sk->sk_prot->sock_is_readable(sk);
2900 #endif /* _SOCK_H */