2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the AF_INET socket handler.
8 * Version: @(#)sock.h 1.0.4 05/13/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche <flla@stud.uni-sb.de>
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
20 * Alan Cox : Added a zapped field for tcp to note
21 * a socket is reset and must stay shut up
22 * Alan Cox : New fields for options
23 * Pauline Middelink : identd support
24 * Alan Cox : Eliminate low level recv/recvfrom
25 * David S. Miller : New socket lookup architecture.
26 * Steve Whitehouse: Default routines for sock_ops
27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
28 * protinfo be just a void pointer, as the
29 * protocol specific parts were moved to
30 * respective headers and ipv4/v6, etc now
31 * use private slabcaches for its socks
32 * Pedro Hortas : New flags field for socket options
35 * This program is free software; you can redistribute it and/or
36 * modify it under the terms of the GNU General Public License
37 * as published by the Free Software Foundation; either version
38 * 2 of the License, or (at your option) any later version.
43 #include <linux/hardirq.h>
44 #include <linux/kernel.h>
45 #include <linux/list.h>
46 #include <linux/list_nulls.h>
47 #include <linux/timer.h>
48 #include <linux/cache.h>
49 #include <linux/bitops.h>
50 #include <linux/lockdep.h>
51 #include <linux/netdevice.h>
52 #include <linux/skbuff.h> /* struct sk_buff */
54 #include <linux/security.h>
55 #include <linux/slab.h>
56 #include <linux/uaccess.h>
57 #include <linux/page_counter.h>
58 #include <linux/memcontrol.h>
59 #include <linux/static_key.h>
60 #include <linux/sched.h>
61 #include <linux/wait.h>
62 #include <linux/cgroup-defs.h>
64 #include <linux/filter.h>
65 #include <linux/rculist_nulls.h>
66 #include <linux/poll.h>
68 #include <linux/atomic.h>
70 #include <net/checksum.h>
71 #include <net/tcp_states.h>
72 #include <linux/net_tstamp.h>
75 * This structure really needs to be cleaned up.
76 * Most of it is for TCP, and not used by any of
77 * the other protocols.
80 /* Define this to get the SOCK_DBG debugging facility. */
81 #define SOCK_DEBUGGING
83 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
84 printk(KERN_DEBUG msg); } while (0)
86 /* Validate arguments and do nothing */
87 static inline __printf(2, 3)
88 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
93 /* This is the per-socket lock. The spinlock provides a synchronization
94 * between user contexts and software interrupt processing, whereas the
95 * mini-semaphore synchronizes multiple users amongst themselves.
100 wait_queue_head_t wq;
102 * We express the mutex-alike socket_lock semantics
103 * to the lock validator by explicitly managing
104 * the slock as a lock variant (in addition to
107 #ifdef CONFIG_DEBUG_LOCK_ALLOC
108 struct lockdep_map dep_map;
116 typedef __u32 __bitwise __portpair;
117 typedef __u64 __bitwise __addrpair;
120 * struct sock_common - minimal network layer representation of sockets
121 * @skc_daddr: Foreign IPv4 addr
122 * @skc_rcv_saddr: Bound local IPv4 addr
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_family: network address family
128 * @skc_state: Connection state
129 * @skc_reuse: %SO_REUSEADDR setting
130 * @skc_reuseport: %SO_REUSEPORT setting
131 * @skc_bound_dev_if: bound device index if != 0
132 * @skc_bind_node: bind hash linkage for various protocol lookup tables
133 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
134 * @skc_prot: protocol handlers inside a network family
135 * @skc_net: reference to the network namespace of this socket
136 * @skc_node: main hash linkage for various protocol lookup tables
137 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
138 * @skc_tx_queue_mapping: tx queue number for this connection
139 * @skc_flags: place holder for sk_flags
140 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
141 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
142 * @skc_incoming_cpu: record/match cpu processing incoming packets
143 * @skc_refcnt: reference count
145 * This is the minimal network layer representation of sockets, the header
146 * for struct sock and struct inet_timewait_sock.
149 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
150 * address on 64bit arches : cf INET_MATCH()
153 __addrpair skc_addrpair;
156 __be32 skc_rcv_saddr;
160 unsigned int skc_hash;
161 __u16 skc_u16hashes[2];
163 /* skc_dport && skc_num must be grouped as well */
165 __portpair skc_portpair;
172 unsigned short skc_family;
173 volatile unsigned char skc_state;
174 unsigned char skc_reuse:4;
175 unsigned char skc_reuseport:1;
176 unsigned char skc_ipv6only:1;
177 unsigned char skc_net_refcnt:1;
178 int skc_bound_dev_if;
180 struct hlist_node skc_bind_node;
181 struct hlist_nulls_node skc_portaddr_node;
183 struct proto *skc_prot;
184 possible_net_t skc_net;
186 #if IS_ENABLED(CONFIG_IPV6)
187 struct in6_addr skc_v6_daddr;
188 struct in6_addr skc_v6_rcv_saddr;
191 atomic64_t skc_cookie;
193 /* following fields are padding to force
194 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
195 * assuming IPV6 is enabled. We use this padding differently
196 * for different kind of 'sockets'
199 unsigned long skc_flags;
200 struct sock *skc_listener; /* request_sock */
201 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
204 * fields between dontcopy_begin/dontcopy_end
205 * are not copied in sock_copy()
208 int skc_dontcopy_begin[0];
211 struct hlist_node skc_node;
212 struct hlist_nulls_node skc_nulls_node;
214 int skc_tx_queue_mapping;
216 int skc_incoming_cpu;
218 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
223 int skc_dontcopy_end[0];
226 u32 skc_window_clamp;
227 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
233 * struct sock - network layer representation of sockets
234 * @__sk_common: shared layout with inet_timewait_sock
235 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
236 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
237 * @sk_lock: synchronizer
238 * @sk_rcvbuf: size of receive buffer in bytes
239 * @sk_wq: sock wait queue and async head
240 * @sk_rx_dst: receive input route used by early demux
241 * @sk_dst_cache: destination cache
242 * @sk_policy: flow policy
243 * @sk_receive_queue: incoming packets
244 * @sk_wmem_alloc: transmit queue bytes committed
245 * @sk_write_queue: Packet sending queue
246 * @sk_omem_alloc: "o" is "option" or "other"
247 * @sk_wmem_queued: persistent queue size
248 * @sk_forward_alloc: space allocated forward
249 * @sk_napi_id: id of the last napi context to receive data for sk
250 * @sk_ll_usec: usecs to busypoll when there is no data
251 * @sk_allocation: allocation mode
252 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
253 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
254 * @sk_sndbuf: size of send buffer in bytes
255 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
256 * @sk_no_check_rx: allow zero checksum in RX packets
257 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
258 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
259 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
260 * @sk_gso_max_size: Maximum GSO segment size to build
261 * @sk_gso_max_segs: Maximum number of GSO segments
262 * @sk_lingertime: %SO_LINGER l_linger setting
263 * @sk_backlog: always used with the per-socket spinlock held
264 * @sk_callback_lock: used with the callbacks in the end of this struct
265 * @sk_error_queue: rarely used
266 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
267 * IPV6_ADDRFORM for instance)
268 * @sk_err: last error
269 * @sk_err_soft: errors that don't cause failure but are the cause of a
270 * persistent failure not just 'timed out'
271 * @sk_drops: raw/udp drops counter
272 * @sk_ack_backlog: current listen backlog
273 * @sk_max_ack_backlog: listen backlog set in listen()
274 * @sk_priority: %SO_PRIORITY setting
275 * @sk_type: socket type (%SOCK_STREAM, etc)
276 * @sk_protocol: which protocol this socket belongs in this network family
277 * @sk_peer_pid: &struct pid for this socket's peer
278 * @sk_peer_cred: %SO_PEERCRED setting
279 * @sk_rcvlowat: %SO_RCVLOWAT setting
280 * @sk_rcvtimeo: %SO_RCVTIMEO setting
281 * @sk_sndtimeo: %SO_SNDTIMEO setting
282 * @sk_txhash: computed flow hash for use on transmit
283 * @sk_filter: socket filtering instructions
284 * @sk_timer: sock cleanup timer
285 * @sk_stamp: time stamp of last packet received
286 * @sk_tsflags: SO_TIMESTAMPING socket options
287 * @sk_tskey: counter to disambiguate concurrent tstamp requests
288 * @sk_socket: Identd and reporting IO signals
289 * @sk_user_data: RPC layer private data
290 * @sk_frag: cached page frag
291 * @sk_peek_off: current peek_offset value
292 * @sk_send_head: front of stuff to transmit
293 * @sk_security: used by security modules
294 * @sk_mark: generic packet mark
295 * @sk_cgrp_data: cgroup data for this cgroup
296 * @sk_memcg: this socket's memory cgroup association
297 * @sk_write_pending: a write to stream socket waits to start
298 * @sk_state_change: callback to indicate change in the state of the sock
299 * @sk_data_ready: callback to indicate there is data to be processed
300 * @sk_write_space: callback to indicate there is bf sending space available
301 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
302 * @sk_backlog_rcv: callback to process the backlog
303 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
304 * @sk_reuseport_cb: reuseport group container
308 * Now struct inet_timewait_sock also uses sock_common, so please just
309 * don't add nothing before this first member (__sk_common) --acme
311 struct sock_common __sk_common;
312 #define sk_node __sk_common.skc_node
313 #define sk_nulls_node __sk_common.skc_nulls_node
314 #define sk_refcnt __sk_common.skc_refcnt
315 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
317 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
318 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
319 #define sk_hash __sk_common.skc_hash
320 #define sk_portpair __sk_common.skc_portpair
321 #define sk_num __sk_common.skc_num
322 #define sk_dport __sk_common.skc_dport
323 #define sk_addrpair __sk_common.skc_addrpair
324 #define sk_daddr __sk_common.skc_daddr
325 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
326 #define sk_family __sk_common.skc_family
327 #define sk_state __sk_common.skc_state
328 #define sk_reuse __sk_common.skc_reuse
329 #define sk_reuseport __sk_common.skc_reuseport
330 #define sk_ipv6only __sk_common.skc_ipv6only
331 #define sk_net_refcnt __sk_common.skc_net_refcnt
332 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
333 #define sk_bind_node __sk_common.skc_bind_node
334 #define sk_prot __sk_common.skc_prot
335 #define sk_net __sk_common.skc_net
336 #define sk_v6_daddr __sk_common.skc_v6_daddr
337 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
338 #define sk_cookie __sk_common.skc_cookie
339 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
340 #define sk_flags __sk_common.skc_flags
341 #define sk_rxhash __sk_common.skc_rxhash
343 socket_lock_t sk_lock;
344 struct sk_buff_head sk_receive_queue;
346 * The backlog queue is special, it is always used with
347 * the per-socket spinlock held and requires low latency
348 * access. Therefore we special case it's implementation.
349 * Note : rmem_alloc is in this structure to fill a hole
350 * on 64bit arches, not because its logically part of
356 struct sk_buff *head;
357 struct sk_buff *tail;
359 #define sk_rmem_alloc sk_backlog.rmem_alloc
360 int sk_forward_alloc;
363 #ifdef CONFIG_NET_RX_BUSY_POLL
364 unsigned int sk_napi_id;
365 unsigned int sk_ll_usec;
370 struct sk_filter __rcu *sk_filter;
372 struct socket_wq __rcu *sk_wq;
373 struct socket_wq *sk_wq_raw;
376 struct xfrm_policy __rcu *sk_policy[2];
378 struct dst_entry *sk_rx_dst;
379 struct dst_entry __rcu *sk_dst_cache;
380 /* Note: 32bit hole on 64bit arches */
381 atomic_t sk_wmem_alloc;
382 atomic_t sk_omem_alloc;
384 struct sk_buff_head sk_write_queue;
385 kmemcheck_bitfield_begin(flags);
386 unsigned int sk_shutdown : 2,
392 #define SK_PROTOCOL_MAX U8_MAX
393 kmemcheck_bitfield_end(flags);
396 u32 sk_pacing_rate; /* bytes per second */
397 u32 sk_max_pacing_rate;
398 netdev_features_t sk_route_caps;
399 netdev_features_t sk_route_nocaps;
401 unsigned int sk_gso_max_size;
404 unsigned long sk_lingertime;
405 struct sk_buff_head sk_error_queue;
406 struct proto *sk_prot_creator;
407 rwlock_t sk_callback_lock;
411 u32 sk_max_ack_backlog;
414 struct pid *sk_peer_pid;
415 const struct cred *sk_peer_cred;
418 struct timer_list sk_timer;
422 struct socket *sk_socket;
424 struct page_frag sk_frag;
425 struct sk_buff *sk_send_head;
427 int sk_write_pending;
428 #ifdef CONFIG_SECURITY
431 struct sock_cgroup_data sk_cgrp_data;
432 struct mem_cgroup *sk_memcg;
433 void (*sk_state_change)(struct sock *sk);
434 void (*sk_data_ready)(struct sock *sk);
435 void (*sk_write_space)(struct sock *sk);
436 void (*sk_error_report)(struct sock *sk);
437 int (*sk_backlog_rcv)(struct sock *sk,
438 struct sk_buff *skb);
439 void (*sk_destruct)(struct sock *sk);
440 struct sock_reuseport __rcu *sk_reuseport_cb;
443 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
445 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
446 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
449 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
450 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
451 * on a socket means that the socket will reuse everybody else's port
452 * without looking at the other's sk_reuse value.
455 #define SK_NO_REUSE 0
456 #define SK_CAN_REUSE 1
457 #define SK_FORCE_REUSE 2
459 static inline int sk_peek_offset(struct sock *sk, int flags)
461 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
462 return sk->sk_peek_off;
467 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
469 if (sk->sk_peek_off >= 0) {
470 if (sk->sk_peek_off >= val)
471 sk->sk_peek_off -= val;
477 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
479 if (sk->sk_peek_off >= 0)
480 sk->sk_peek_off += val;
484 * Hashed lists helper routines
486 static inline struct sock *sk_entry(const struct hlist_node *node)
488 return hlist_entry(node, struct sock, sk_node);
491 static inline struct sock *__sk_head(const struct hlist_head *head)
493 return hlist_entry(head->first, struct sock, sk_node);
496 static inline struct sock *sk_head(const struct hlist_head *head)
498 return hlist_empty(head) ? NULL : __sk_head(head);
501 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
503 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
506 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
508 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
511 static inline struct sock *sk_next(const struct sock *sk)
513 return sk->sk_node.next ?
514 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
517 static inline struct sock *sk_nulls_next(const struct sock *sk)
519 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
520 hlist_nulls_entry(sk->sk_nulls_node.next,
521 struct sock, sk_nulls_node) :
525 static inline bool sk_unhashed(const struct sock *sk)
527 return hlist_unhashed(&sk->sk_node);
530 static inline bool sk_hashed(const struct sock *sk)
532 return !sk_unhashed(sk);
535 static inline void sk_node_init(struct hlist_node *node)
540 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
545 static inline void __sk_del_node(struct sock *sk)
547 __hlist_del(&sk->sk_node);
550 /* NB: equivalent to hlist_del_init_rcu */
551 static inline bool __sk_del_node_init(struct sock *sk)
555 sk_node_init(&sk->sk_node);
561 /* Grab socket reference count. This operation is valid only
562 when sk is ALREADY grabbed f.e. it is found in hash table
563 or a list and the lookup is made under lock preventing hash table
567 static inline void sock_hold(struct sock *sk)
569 atomic_inc(&sk->sk_refcnt);
572 /* Ungrab socket in the context, which assumes that socket refcnt
573 cannot hit zero, f.e. it is true in context of any socketcall.
575 static inline void __sock_put(struct sock *sk)
577 atomic_dec(&sk->sk_refcnt);
580 static inline bool sk_del_node_init(struct sock *sk)
582 bool rc = __sk_del_node_init(sk);
585 /* paranoid for a while -acme */
586 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
591 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
593 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
596 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
602 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
604 bool rc = __sk_nulls_del_node_init_rcu(sk);
607 /* paranoid for a while -acme */
608 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
614 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
616 hlist_add_head(&sk->sk_node, list);
619 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
622 __sk_add_node(sk, list);
625 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
628 hlist_add_head_rcu(&sk->sk_node, list);
631 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
633 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
636 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
639 __sk_nulls_add_node_rcu(sk, list);
642 static inline void __sk_del_bind_node(struct sock *sk)
644 __hlist_del(&sk->sk_bind_node);
647 static inline void sk_add_bind_node(struct sock *sk,
648 struct hlist_head *list)
650 hlist_add_head(&sk->sk_bind_node, list);
653 #define sk_for_each(__sk, list) \
654 hlist_for_each_entry(__sk, list, sk_node)
655 #define sk_for_each_rcu(__sk, list) \
656 hlist_for_each_entry_rcu(__sk, list, sk_node)
657 #define sk_nulls_for_each(__sk, node, list) \
658 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
659 #define sk_nulls_for_each_rcu(__sk, node, list) \
660 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
661 #define sk_for_each_from(__sk) \
662 hlist_for_each_entry_from(__sk, sk_node)
663 #define sk_nulls_for_each_from(__sk, node) \
664 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
665 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
666 #define sk_for_each_safe(__sk, tmp, list) \
667 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
668 #define sk_for_each_bound(__sk, list) \
669 hlist_for_each_entry(__sk, list, sk_bind_node)
672 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
673 * @tpos: the type * to use as a loop cursor.
674 * @pos: the &struct hlist_node to use as a loop cursor.
675 * @head: the head for your list.
676 * @offset: offset of hlist_node within the struct.
679 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
680 for (pos = (head)->first; \
681 (!is_a_nulls(pos)) && \
682 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
685 static inline struct user_namespace *sk_user_ns(struct sock *sk)
687 /* Careful only use this in a context where these parameters
688 * can not change and must all be valid, such as recvmsg from
691 return sk->sk_socket->file->f_cred->user_ns;
705 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
706 SOCK_DBG, /* %SO_DEBUG setting */
707 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
708 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
709 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
710 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
711 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
712 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
713 SOCK_FASYNC, /* fasync() active */
715 SOCK_ZEROCOPY, /* buffers from userspace */
716 SOCK_WIFI_STATUS, /* push wifi status to userspace */
717 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
718 * Will use last 4 bytes of packet sent from
719 * user-space instead.
721 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
722 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
725 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
727 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
729 nsk->sk_flags = osk->sk_flags;
732 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
734 __set_bit(flag, &sk->sk_flags);
737 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
739 __clear_bit(flag, &sk->sk_flags);
742 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
744 return test_bit(flag, &sk->sk_flags);
748 extern struct static_key memalloc_socks;
749 static inline int sk_memalloc_socks(void)
751 return static_key_false(&memalloc_socks);
755 static inline int sk_memalloc_socks(void)
762 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
764 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
767 static inline void sk_acceptq_removed(struct sock *sk)
769 sk->sk_ack_backlog--;
772 static inline void sk_acceptq_added(struct sock *sk)
774 sk->sk_ack_backlog++;
777 static inline bool sk_acceptq_is_full(const struct sock *sk)
779 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
783 * Compute minimal free write space needed to queue new packets.
785 static inline int sk_stream_min_wspace(const struct sock *sk)
787 return sk->sk_wmem_queued >> 1;
790 static inline int sk_stream_wspace(const struct sock *sk)
792 return sk->sk_sndbuf - sk->sk_wmem_queued;
795 void sk_stream_write_space(struct sock *sk);
797 /* OOB backlog add */
798 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
800 /* dont let skb dst not refcounted, we are going to leave rcu lock */
801 skb_dst_force_safe(skb);
803 if (!sk->sk_backlog.tail)
804 sk->sk_backlog.head = skb;
806 sk->sk_backlog.tail->next = skb;
808 sk->sk_backlog.tail = skb;
813 * Take into account size of receive queue and backlog queue
814 * Do not take into account this skb truesize,
815 * to allow even a single big packet to come.
817 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
819 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
821 return qsize > limit;
824 /* The per-socket spinlock must be held here. */
825 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
828 if (sk_rcvqueues_full(sk, limit))
832 * If the skb was allocated from pfmemalloc reserves, only
833 * allow SOCK_MEMALLOC sockets to use it as this socket is
834 * helping free memory
836 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
839 __sk_add_backlog(sk, skb);
840 sk->sk_backlog.len += skb->truesize;
844 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
846 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
848 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
849 return __sk_backlog_rcv(sk, skb);
851 return sk->sk_backlog_rcv(sk, skb);
854 static inline void sk_incoming_cpu_update(struct sock *sk)
856 sk->sk_incoming_cpu = raw_smp_processor_id();
859 static inline void sock_rps_record_flow_hash(__u32 hash)
862 struct rps_sock_flow_table *sock_flow_table;
865 sock_flow_table = rcu_dereference(rps_sock_flow_table);
866 rps_record_sock_flow(sock_flow_table, hash);
871 static inline void sock_rps_record_flow(const struct sock *sk)
874 sock_rps_record_flow_hash(sk->sk_rxhash);
878 static inline void sock_rps_save_rxhash(struct sock *sk,
879 const struct sk_buff *skb)
882 if (unlikely(sk->sk_rxhash != skb->hash))
883 sk->sk_rxhash = skb->hash;
887 static inline void sock_rps_reset_rxhash(struct sock *sk)
894 #define sk_wait_event(__sk, __timeo, __condition) \
896 release_sock(__sk); \
897 __rc = __condition; \
899 *(__timeo) = schedule_timeout(*(__timeo)); \
901 sched_annotate_sleep(); \
903 __rc = __condition; \
907 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
908 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
909 void sk_stream_wait_close(struct sock *sk, long timeo_p);
910 int sk_stream_error(struct sock *sk, int flags, int err);
911 void sk_stream_kill_queues(struct sock *sk);
912 void sk_set_memalloc(struct sock *sk);
913 void sk_clear_memalloc(struct sock *sk);
915 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
917 struct request_sock_ops;
918 struct timewait_sock_ops;
919 struct inet_hashinfo;
924 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
925 * un-modified. Special care is taken when initializing object to zero.
927 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
929 if (offsetof(struct sock, sk_node.next) != 0)
930 memset(sk, 0, offsetof(struct sock, sk_node.next));
931 memset(&sk->sk_node.pprev, 0,
932 size - offsetof(struct sock, sk_node.pprev));
935 /* Networking protocol blocks we attach to sockets.
936 * socket layer -> transport layer interface
939 void (*close)(struct sock *sk,
941 int (*connect)(struct sock *sk,
942 struct sockaddr *uaddr,
944 int (*disconnect)(struct sock *sk, int flags);
946 struct sock * (*accept)(struct sock *sk, int flags, int *err);
948 int (*ioctl)(struct sock *sk, int cmd,
950 int (*init)(struct sock *sk);
951 void (*destroy)(struct sock *sk);
952 void (*shutdown)(struct sock *sk, int how);
953 int (*setsockopt)(struct sock *sk, int level,
954 int optname, char __user *optval,
955 unsigned int optlen);
956 int (*getsockopt)(struct sock *sk, int level,
957 int optname, char __user *optval,
960 int (*compat_setsockopt)(struct sock *sk,
962 int optname, char __user *optval,
963 unsigned int optlen);
964 int (*compat_getsockopt)(struct sock *sk,
966 int optname, char __user *optval,
968 int (*compat_ioctl)(struct sock *sk,
969 unsigned int cmd, unsigned long arg);
971 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
973 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
974 size_t len, int noblock, int flags,
976 int (*sendpage)(struct sock *sk, struct page *page,
977 int offset, size_t size, int flags);
978 int (*bind)(struct sock *sk,
979 struct sockaddr *uaddr, int addr_len);
981 int (*backlog_rcv) (struct sock *sk,
982 struct sk_buff *skb);
984 void (*release_cb)(struct sock *sk);
986 /* Keeping track of sk's, looking them up, and port selection methods. */
987 void (*hash)(struct sock *sk);
988 void (*unhash)(struct sock *sk);
989 void (*rehash)(struct sock *sk);
990 int (*get_port)(struct sock *sk, unsigned short snum);
991 void (*clear_sk)(struct sock *sk, int size);
993 /* Keeping track of sockets in use */
994 #ifdef CONFIG_PROC_FS
995 unsigned int inuse_idx;
998 bool (*stream_memory_free)(const struct sock *sk);
999 /* Memory pressure */
1000 void (*enter_memory_pressure)(struct sock *sk);
1001 atomic_long_t *memory_allocated; /* Current allocated memory. */
1002 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1004 * Pressure flag: try to collapse.
1005 * Technical note: it is used by multiple contexts non atomically.
1006 * All the __sk_mem_schedule() is of this nature: accounting
1007 * is strict, actions are advisory and have some latency.
1009 int *memory_pressure;
1016 struct kmem_cache *slab;
1017 unsigned int obj_size;
1020 struct percpu_counter *orphan_count;
1022 struct request_sock_ops *rsk_prot;
1023 struct timewait_sock_ops *twsk_prot;
1026 struct inet_hashinfo *hashinfo;
1027 struct udp_table *udp_table;
1028 struct raw_hashinfo *raw_hash;
1031 struct module *owner;
1035 struct list_head node;
1036 #ifdef SOCK_REFCNT_DEBUG
1039 #ifdef CONFIG_MEMCG_KMEM
1041 * cgroup specific init/deinit functions. Called once for all
1042 * protocols that implement it, from cgroups populate function.
1043 * This function has to setup any files the protocol want to
1044 * appear in the kmem cgroup filesystem.
1046 int (*init_cgroup)(struct mem_cgroup *memcg,
1047 struct cgroup_subsys *ss);
1048 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1049 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1051 int (*diag_destroy)(struct sock *sk, int err);
1054 int proto_register(struct proto *prot, int alloc_slab);
1055 void proto_unregister(struct proto *prot);
1057 #ifdef SOCK_REFCNT_DEBUG
1058 static inline void sk_refcnt_debug_inc(struct sock *sk)
1060 atomic_inc(&sk->sk_prot->socks);
1063 static inline void sk_refcnt_debug_dec(struct sock *sk)
1065 atomic_dec(&sk->sk_prot->socks);
1066 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1067 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1070 static inline void sk_refcnt_debug_release(const struct sock *sk)
1072 if (atomic_read(&sk->sk_refcnt) != 1)
1073 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1074 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1076 #else /* SOCK_REFCNT_DEBUG */
1077 #define sk_refcnt_debug_inc(sk) do { } while (0)
1078 #define sk_refcnt_debug_dec(sk) do { } while (0)
1079 #define sk_refcnt_debug_release(sk) do { } while (0)
1080 #endif /* SOCK_REFCNT_DEBUG */
1082 static inline bool sk_stream_memory_free(const struct sock *sk)
1084 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1087 return sk->sk_prot->stream_memory_free ?
1088 sk->sk_prot->stream_memory_free(sk) : true;
1091 static inline bool sk_stream_is_writeable(const struct sock *sk)
1093 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1094 sk_stream_memory_free(sk);
1098 static inline bool sk_has_memory_pressure(const struct sock *sk)
1100 return sk->sk_prot->memory_pressure != NULL;
1103 static inline bool sk_under_memory_pressure(const struct sock *sk)
1105 if (!sk->sk_prot->memory_pressure)
1108 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1109 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1112 return !!*sk->sk_prot->memory_pressure;
1115 static inline void sk_leave_memory_pressure(struct sock *sk)
1117 int *memory_pressure = sk->sk_prot->memory_pressure;
1119 if (!memory_pressure)
1122 if (*memory_pressure)
1123 *memory_pressure = 0;
1126 static inline void sk_enter_memory_pressure(struct sock *sk)
1128 if (!sk->sk_prot->enter_memory_pressure)
1131 sk->sk_prot->enter_memory_pressure(sk);
1134 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1136 return sk->sk_prot->sysctl_mem[index];
1140 sk_memory_allocated(const struct sock *sk)
1142 return atomic_long_read(sk->sk_prot->memory_allocated);
1146 sk_memory_allocated_add(struct sock *sk, int amt)
1148 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1152 sk_memory_allocated_sub(struct sock *sk, int amt)
1154 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1157 static inline void sk_sockets_allocated_dec(struct sock *sk)
1159 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1162 static inline void sk_sockets_allocated_inc(struct sock *sk)
1164 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1168 sk_sockets_allocated_read_positive(struct sock *sk)
1170 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1174 proto_sockets_allocated_sum_positive(struct proto *prot)
1176 return percpu_counter_sum_positive(prot->sockets_allocated);
1180 proto_memory_allocated(struct proto *prot)
1182 return atomic_long_read(prot->memory_allocated);
1186 proto_memory_pressure(struct proto *prot)
1188 if (!prot->memory_pressure)
1190 return !!*prot->memory_pressure;
1194 #ifdef CONFIG_PROC_FS
1195 /* Called with local bh disabled */
1196 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1197 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1199 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1206 /* With per-bucket locks this operation is not-atomic, so that
1207 * this version is not worse.
1209 static inline void __sk_prot_rehash(struct sock *sk)
1211 sk->sk_prot->unhash(sk);
1212 sk->sk_prot->hash(sk);
1215 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1217 /* About 10 seconds */
1218 #define SOCK_DESTROY_TIME (10*HZ)
1220 /* Sockets 0-1023 can't be bound to unless you are superuser */
1221 #define PROT_SOCK 1024
1223 #define SHUTDOWN_MASK 3
1224 #define RCV_SHUTDOWN 1
1225 #define SEND_SHUTDOWN 2
1227 #define SOCK_SNDBUF_LOCK 1
1228 #define SOCK_RCVBUF_LOCK 2
1229 #define SOCK_BINDADDR_LOCK 4
1230 #define SOCK_BINDPORT_LOCK 8
1232 struct socket_alloc {
1233 struct socket socket;
1234 struct inode vfs_inode;
1237 static inline struct socket *SOCKET_I(struct inode *inode)
1239 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1242 static inline struct inode *SOCK_INODE(struct socket *socket)
1244 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1248 * Functions for memory accounting
1250 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1251 void __sk_mem_reclaim(struct sock *sk, int amount);
1253 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1254 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1255 #define SK_MEM_SEND 0
1256 #define SK_MEM_RECV 1
1258 static inline int sk_mem_pages(int amt)
1260 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1263 static inline bool sk_has_account(struct sock *sk)
1265 /* return true if protocol supports memory accounting */
1266 return !!sk->sk_prot->memory_allocated;
1269 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1271 if (!sk_has_account(sk))
1273 return size <= sk->sk_forward_alloc ||
1274 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1278 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1280 if (!sk_has_account(sk))
1282 return size<= sk->sk_forward_alloc ||
1283 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1284 skb_pfmemalloc(skb);
1287 static inline void sk_mem_reclaim(struct sock *sk)
1289 if (!sk_has_account(sk))
1291 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1292 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1295 static inline void sk_mem_reclaim_partial(struct sock *sk)
1297 if (!sk_has_account(sk))
1299 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1300 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1303 static inline void sk_mem_charge(struct sock *sk, int size)
1305 if (!sk_has_account(sk))
1307 sk->sk_forward_alloc -= size;
1310 static inline void sk_mem_uncharge(struct sock *sk, int size)
1312 if (!sk_has_account(sk))
1314 sk->sk_forward_alloc += size;
1317 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1319 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1320 sk->sk_wmem_queued -= skb->truesize;
1321 sk_mem_uncharge(sk, skb->truesize);
1325 /* Used by processes to "lock" a socket state, so that
1326 * interrupts and bottom half handlers won't change it
1327 * from under us. It essentially blocks any incoming
1328 * packets, so that we won't get any new data or any
1329 * packets that change the state of the socket.
1331 * While locked, BH processing will add new packets to
1332 * the backlog queue. This queue is processed by the
1333 * owner of the socket lock right before it is released.
1335 * Since ~2.3.5 it is also exclusive sleep lock serializing
1336 * accesses from user process context.
1338 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1340 static inline void sock_release_ownership(struct sock *sk)
1342 sk->sk_lock.owned = 0;
1346 * Macro so as to not evaluate some arguments when
1347 * lockdep is not enabled.
1349 * Mark both the sk_lock and the sk_lock.slock as a
1350 * per-address-family lock class.
1352 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1354 sk->sk_lock.owned = 0; \
1355 init_waitqueue_head(&sk->sk_lock.wq); \
1356 spin_lock_init(&(sk)->sk_lock.slock); \
1357 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1358 sizeof((sk)->sk_lock)); \
1359 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1361 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1364 void lock_sock_nested(struct sock *sk, int subclass);
1366 static inline void lock_sock(struct sock *sk)
1368 lock_sock_nested(sk, 0);
1371 void release_sock(struct sock *sk);
1373 /* BH context may only use the following locking interface. */
1374 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1375 #define bh_lock_sock_nested(__sk) \
1376 spin_lock_nested(&((__sk)->sk_lock.slock), \
1377 SINGLE_DEPTH_NESTING)
1378 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1380 bool lock_sock_fast(struct sock *sk);
1382 * unlock_sock_fast - complement of lock_sock_fast
1386 * fast unlock socket for user context.
1387 * If slow mode is on, we call regular release_sock()
1389 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1394 spin_unlock_bh(&sk->sk_lock.slock);
1398 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1399 struct proto *prot, int kern);
1400 void sk_free(struct sock *sk);
1401 void sk_destruct(struct sock *sk);
1402 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1404 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1406 void sock_wfree(struct sk_buff *skb);
1407 void skb_orphan_partial(struct sk_buff *skb);
1408 void sock_rfree(struct sk_buff *skb);
1409 void sock_efree(struct sk_buff *skb);
1411 void sock_edemux(struct sk_buff *skb);
1413 #define sock_edemux(skb) sock_efree(skb)
1416 int sock_setsockopt(struct socket *sock, int level, int op,
1417 char __user *optval, unsigned int optlen);
1419 int sock_getsockopt(struct socket *sock, int level, int op,
1420 char __user *optval, int __user *optlen);
1421 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1422 int noblock, int *errcode);
1423 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1424 unsigned long data_len, int noblock,
1425 int *errcode, int max_page_order);
1426 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1427 void sock_kfree_s(struct sock *sk, void *mem, int size);
1428 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1429 void sk_send_sigurg(struct sock *sk);
1431 struct sockcm_cookie {
1435 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1436 struct sockcm_cookie *sockc);
1439 * Functions to fill in entries in struct proto_ops when a protocol
1440 * does not implement a particular function.
1442 int sock_no_bind(struct socket *, struct sockaddr *, int);
1443 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1444 int sock_no_socketpair(struct socket *, struct socket *);
1445 int sock_no_accept(struct socket *, struct socket *, int);
1446 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1447 unsigned int sock_no_poll(struct file *, struct socket *,
1448 struct poll_table_struct *);
1449 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1450 int sock_no_listen(struct socket *, int);
1451 int sock_no_shutdown(struct socket *, int);
1452 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1453 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1454 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1455 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1456 int sock_no_mmap(struct file *file, struct socket *sock,
1457 struct vm_area_struct *vma);
1458 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1459 size_t size, int flags);
1462 * Functions to fill in entries in struct proto_ops when a protocol
1463 * uses the inet style.
1465 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1466 char __user *optval, int __user *optlen);
1467 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1469 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1470 char __user *optval, unsigned int optlen);
1471 int compat_sock_common_getsockopt(struct socket *sock, int level,
1472 int optname, char __user *optval, int __user *optlen);
1473 int compat_sock_common_setsockopt(struct socket *sock, int level,
1474 int optname, char __user *optval, unsigned int optlen);
1476 void sk_common_release(struct sock *sk);
1479 * Default socket callbacks and setup code
1482 /* Initialise core socket variables */
1483 void sock_init_data(struct socket *sock, struct sock *sk);
1486 * Socket reference counting postulates.
1488 * * Each user of socket SHOULD hold a reference count.
1489 * * Each access point to socket (an hash table bucket, reference from a list,
1490 * running timer, skb in flight MUST hold a reference count.
1491 * * When reference count hits 0, it means it will never increase back.
1492 * * When reference count hits 0, it means that no references from
1493 * outside exist to this socket and current process on current CPU
1494 * is last user and may/should destroy this socket.
1495 * * sk_free is called from any context: process, BH, IRQ. When
1496 * it is called, socket has no references from outside -> sk_free
1497 * may release descendant resources allocated by the socket, but
1498 * to the time when it is called, socket is NOT referenced by any
1499 * hash tables, lists etc.
1500 * * Packets, delivered from outside (from network or from another process)
1501 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1502 * when they sit in queue. Otherwise, packets will leak to hole, when
1503 * socket is looked up by one cpu and unhasing is made by another CPU.
1504 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1505 * (leak to backlog). Packet socket does all the processing inside
1506 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1507 * use separate SMP lock, so that they are prone too.
1510 /* Ungrab socket and destroy it, if it was the last reference. */
1511 static inline void sock_put(struct sock *sk)
1513 if (atomic_dec_and_test(&sk->sk_refcnt))
1516 /* Generic version of sock_put(), dealing with all sockets
1517 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1519 void sock_gen_put(struct sock *sk);
1521 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1523 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1525 sk->sk_tx_queue_mapping = tx_queue;
1528 static inline void sk_tx_queue_clear(struct sock *sk)
1530 sk->sk_tx_queue_mapping = -1;
1533 static inline int sk_tx_queue_get(const struct sock *sk)
1535 return sk ? sk->sk_tx_queue_mapping : -1;
1538 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1540 sk_tx_queue_clear(sk);
1541 sk->sk_socket = sock;
1544 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1546 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1547 return &rcu_dereference_raw(sk->sk_wq)->wait;
1549 /* Detach socket from process context.
1550 * Announce socket dead, detach it from wait queue and inode.
1551 * Note that parent inode held reference count on this struct sock,
1552 * we do not release it in this function, because protocol
1553 * probably wants some additional cleanups or even continuing
1554 * to work with this socket (TCP).
1556 static inline void sock_orphan(struct sock *sk)
1558 write_lock_bh(&sk->sk_callback_lock);
1559 sock_set_flag(sk, SOCK_DEAD);
1560 sk_set_socket(sk, NULL);
1562 write_unlock_bh(&sk->sk_callback_lock);
1565 static inline void sock_graft(struct sock *sk, struct socket *parent)
1567 write_lock_bh(&sk->sk_callback_lock);
1568 sk->sk_wq = parent->wq;
1570 sk_set_socket(sk, parent);
1571 security_sock_graft(sk, parent);
1572 write_unlock_bh(&sk->sk_callback_lock);
1575 kuid_t sock_i_uid(struct sock *sk);
1576 unsigned long sock_i_ino(struct sock *sk);
1578 static inline u32 net_tx_rndhash(void)
1580 u32 v = prandom_u32();
1585 static inline void sk_set_txhash(struct sock *sk)
1587 sk->sk_txhash = net_tx_rndhash();
1590 static inline void sk_rethink_txhash(struct sock *sk)
1596 static inline struct dst_entry *
1597 __sk_dst_get(struct sock *sk)
1599 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1600 lockdep_is_held(&sk->sk_lock.slock));
1603 static inline struct dst_entry *
1604 sk_dst_get(struct sock *sk)
1606 struct dst_entry *dst;
1609 dst = rcu_dereference(sk->sk_dst_cache);
1610 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1616 static inline void dst_negative_advice(struct sock *sk)
1618 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1620 sk_rethink_txhash(sk);
1622 if (dst && dst->ops->negative_advice) {
1623 ndst = dst->ops->negative_advice(dst);
1626 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1627 sk_tx_queue_clear(sk);
1633 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1635 struct dst_entry *old_dst;
1637 sk_tx_queue_clear(sk);
1639 * This can be called while sk is owned by the caller only,
1640 * with no state that can be checked in a rcu_dereference_check() cond
1642 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1643 rcu_assign_pointer(sk->sk_dst_cache, dst);
1644 dst_release(old_dst);
1648 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1650 struct dst_entry *old_dst;
1652 sk_tx_queue_clear(sk);
1653 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1654 dst_release(old_dst);
1658 __sk_dst_reset(struct sock *sk)
1660 __sk_dst_set(sk, NULL);
1664 sk_dst_reset(struct sock *sk)
1666 sk_dst_set(sk, NULL);
1669 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1671 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1673 bool sk_mc_loop(struct sock *sk);
1675 static inline bool sk_can_gso(const struct sock *sk)
1677 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1680 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1682 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1684 sk->sk_route_nocaps |= flags;
1685 sk->sk_route_caps &= ~flags;
1688 static inline bool sk_check_csum_caps(struct sock *sk)
1690 return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1691 (sk->sk_family == PF_INET &&
1692 (sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1693 (sk->sk_family == PF_INET6 &&
1694 (sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1697 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1698 struct iov_iter *from, char *to,
1699 int copy, int offset)
1701 if (skb->ip_summed == CHECKSUM_NONE) {
1703 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1705 skb->csum = csum_block_add(skb->csum, csum, offset);
1706 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1707 if (copy_from_iter_nocache(to, copy, from) != copy)
1709 } else if (copy_from_iter(to, copy, from) != copy)
1715 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1716 struct iov_iter *from, int copy)
1718 int err, offset = skb->len;
1720 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1723 __skb_trim(skb, offset);
1728 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1729 struct sk_buff *skb,
1735 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1741 skb->data_len += copy;
1742 skb->truesize += copy;
1743 sk->sk_wmem_queued += copy;
1744 sk_mem_charge(sk, copy);
1749 * sk_wmem_alloc_get - returns write allocations
1752 * Returns sk_wmem_alloc minus initial offset of one
1754 static inline int sk_wmem_alloc_get(const struct sock *sk)
1756 return atomic_read(&sk->sk_wmem_alloc) - 1;
1760 * sk_rmem_alloc_get - returns read allocations
1763 * Returns sk_rmem_alloc
1765 static inline int sk_rmem_alloc_get(const struct sock *sk)
1767 return atomic_read(&sk->sk_rmem_alloc);
1771 * sk_has_allocations - check if allocations are outstanding
1774 * Returns true if socket has write or read allocations
1776 static inline bool sk_has_allocations(const struct sock *sk)
1778 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1782 * skwq_has_sleeper - check if there are any waiting processes
1783 * @wq: struct socket_wq
1785 * Returns true if socket_wq has waiting processes
1787 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1788 * barrier call. They were added due to the race found within the tcp code.
1790 * Consider following tcp code paths:
1794 * sys_select receive packet
1796 * __add_wait_queue update tp->rcv_nxt
1798 * tp->rcv_nxt check sock_def_readable
1800 * schedule rcu_read_lock();
1801 * wq = rcu_dereference(sk->sk_wq);
1802 * if (wq && waitqueue_active(&wq->wait))
1803 * wake_up_interruptible(&wq->wait)
1807 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1808 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1809 * could then endup calling schedule and sleep forever if there are no more
1810 * data on the socket.
1813 static inline bool skwq_has_sleeper(struct socket_wq *wq)
1815 return wq && wq_has_sleeper(&wq->wait);
1819 * sock_poll_wait - place memory barrier behind the poll_wait call.
1821 * @wait_address: socket wait queue
1824 * See the comments in the wq_has_sleeper function.
1826 static inline void sock_poll_wait(struct file *filp,
1827 wait_queue_head_t *wait_address, poll_table *p)
1829 if (!poll_does_not_wait(p) && wait_address) {
1830 poll_wait(filp, wait_address, p);
1831 /* We need to be sure we are in sync with the
1832 * socket flags modification.
1834 * This memory barrier is paired in the wq_has_sleeper.
1840 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1842 if (sk->sk_txhash) {
1844 skb->hash = sk->sk_txhash;
1848 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
1851 * Queue a received datagram if it will fit. Stream and sequenced
1852 * protocols can't normally use this as they need to fit buffers in
1853 * and play with them.
1855 * Inlined as it's very short and called for pretty much every
1856 * packet ever received.
1858 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1862 skb->destructor = sock_rfree;
1863 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1864 sk_mem_charge(sk, skb->truesize);
1867 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1868 unsigned long expires);
1870 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1872 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1874 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1875 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1878 * Recover an error report and clear atomically
1881 static inline int sock_error(struct sock *sk)
1884 if (likely(!sk->sk_err))
1886 err = xchg(&sk->sk_err, 0);
1890 static inline unsigned long sock_wspace(struct sock *sk)
1894 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1895 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1903 * We use sk->sk_wq_raw, from contexts knowing this
1904 * pointer is not NULL and cannot disappear/change.
1906 static inline void sk_set_bit(int nr, struct sock *sk)
1908 set_bit(nr, &sk->sk_wq_raw->flags);
1911 static inline void sk_clear_bit(int nr, struct sock *sk)
1913 clear_bit(nr, &sk->sk_wq_raw->flags);
1916 static inline void sk_wake_async(const struct sock *sk, int how, int band)
1918 if (sock_flag(sk, SOCK_FASYNC)) {
1920 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
1925 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
1926 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
1927 * Note: for send buffers, TCP works better if we can build two skbs at
1930 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
1932 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
1933 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
1935 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1937 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1938 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1939 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1943 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
1944 bool force_schedule);
1947 * sk_page_frag - return an appropriate page_frag
1950 * If socket allocation mode allows current thread to sleep, it means its
1951 * safe to use the per task page_frag instead of the per socket one.
1953 static inline struct page_frag *sk_page_frag(struct sock *sk)
1955 if (gfpflags_allow_blocking(sk->sk_allocation))
1956 return ¤t->task_frag;
1958 return &sk->sk_frag;
1961 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
1964 * Default write policy as shown to user space via poll/select/SIGIO
1966 static inline bool sock_writeable(const struct sock *sk)
1968 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1971 static inline gfp_t gfp_any(void)
1973 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1976 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
1978 return noblock ? 0 : sk->sk_rcvtimeo;
1981 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
1983 return noblock ? 0 : sk->sk_sndtimeo;
1986 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1988 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1991 /* Alas, with timeout socket operations are not restartable.
1992 * Compare this to poll().
1994 static inline int sock_intr_errno(long timeo)
1996 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1999 struct sock_skb_cb {
2003 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2004 * using skb->cb[] would keep using it directly and utilize its
2005 * alignement guarantee.
2007 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2008 sizeof(struct sock_skb_cb)))
2010 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2011 SOCK_SKB_CB_OFFSET))
2013 #define sock_skb_cb_check_size(size) \
2014 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2017 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2019 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2022 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2023 struct sk_buff *skb);
2024 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2025 struct sk_buff *skb);
2028 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2030 ktime_t kt = skb->tstamp;
2031 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2034 * generate control messages if
2035 * - receive time stamping in software requested
2036 * - software time stamp available and wanted
2037 * - hardware time stamps available and wanted
2039 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2040 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2041 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2042 (hwtstamps->hwtstamp.tv64 &&
2043 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2044 __sock_recv_timestamp(msg, sk, skb);
2048 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2049 __sock_recv_wifi_status(msg, sk, skb);
2052 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2053 struct sk_buff *skb);
2055 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2056 struct sk_buff *skb)
2058 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2059 (1UL << SOCK_RCVTSTAMP))
2060 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2061 SOF_TIMESTAMPING_RAW_HARDWARE)
2063 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2064 __sock_recv_ts_and_drops(msg, sk, skb);
2066 sk->sk_stamp = skb->tstamp;
2069 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2072 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2073 * @sk: socket sending this packet
2074 * @tx_flags: completed with instructions for time stamping
2076 * Note : callers should take care of initial *tx_flags value (usually 0)
2078 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2080 if (unlikely(sk->sk_tsflags))
2081 __sock_tx_timestamp(sk, tx_flags);
2082 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2083 *tx_flags |= SKBTX_WIFI_STATUS;
2087 * sk_eat_skb - Release a skb if it is no longer needed
2088 * @sk: socket to eat this skb from
2089 * @skb: socket buffer to eat
2091 * This routine must be called with interrupts disabled or with the socket
2092 * locked so that the sk_buff queue operation is ok.
2094 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2096 __skb_unlink(skb, &sk->sk_receive_queue);
2101 struct net *sock_net(const struct sock *sk)
2103 return read_pnet(&sk->sk_net);
2107 void sock_net_set(struct sock *sk, struct net *net)
2109 write_pnet(&sk->sk_net, net);
2112 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2115 struct sock *sk = skb->sk;
2117 skb->destructor = NULL;
2124 /* This helper checks if a socket is a full socket,
2125 * ie _not_ a timewait or request socket.
2127 static inline bool sk_fullsock(const struct sock *sk)
2129 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2132 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2133 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2135 static inline bool sk_listener(const struct sock *sk)
2137 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2141 * sk_state_load - read sk->sk_state for lockless contexts
2142 * @sk: socket pointer
2144 * Paired with sk_state_store(). Used in places we do not hold socket lock :
2145 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2147 static inline int sk_state_load(const struct sock *sk)
2149 return smp_load_acquire(&sk->sk_state);
2153 * sk_state_store - update sk->sk_state
2154 * @sk: socket pointer
2155 * @newstate: new state
2157 * Paired with sk_state_load(). Should be used in contexts where
2158 * state change might impact lockless readers.
2160 static inline void sk_state_store(struct sock *sk, int newstate)
2162 smp_store_release(&sk->sk_state, newstate);
2165 void sock_enable_timestamp(struct sock *sk, int flag);
2166 int sock_get_timestamp(struct sock *, struct timeval __user *);
2167 int sock_get_timestampns(struct sock *, struct timespec __user *);
2168 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2171 bool sk_ns_capable(const struct sock *sk,
2172 struct user_namespace *user_ns, int cap);
2173 bool sk_capable(const struct sock *sk, int cap);
2174 bool sk_net_capable(const struct sock *sk, int cap);
2176 extern __u32 sysctl_wmem_max;
2177 extern __u32 sysctl_rmem_max;
2179 extern int sysctl_tstamp_allow_data;
2180 extern int sysctl_optmem_max;
2182 extern __u32 sysctl_wmem_default;
2183 extern __u32 sysctl_rmem_default;
2185 #endif /* _SOCK_H */