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 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
634 sk->sk_family == AF_INET6)
635 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
637 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
640 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
643 __sk_nulls_add_node_rcu(sk, list);
646 static inline void __sk_del_bind_node(struct sock *sk)
648 __hlist_del(&sk->sk_bind_node);
651 static inline void sk_add_bind_node(struct sock *sk,
652 struct hlist_head *list)
654 hlist_add_head(&sk->sk_bind_node, list);
657 #define sk_for_each(__sk, list) \
658 hlist_for_each_entry(__sk, list, sk_node)
659 #define sk_for_each_rcu(__sk, list) \
660 hlist_for_each_entry_rcu(__sk, list, sk_node)
661 #define sk_nulls_for_each(__sk, node, list) \
662 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
663 #define sk_nulls_for_each_rcu(__sk, node, list) \
664 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
665 #define sk_for_each_from(__sk) \
666 hlist_for_each_entry_from(__sk, sk_node)
667 #define sk_nulls_for_each_from(__sk, node) \
668 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
669 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
670 #define sk_for_each_safe(__sk, tmp, list) \
671 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
672 #define sk_for_each_bound(__sk, list) \
673 hlist_for_each_entry(__sk, list, sk_bind_node)
676 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
677 * @tpos: the type * to use as a loop cursor.
678 * @pos: the &struct hlist_node to use as a loop cursor.
679 * @head: the head for your list.
680 * @offset: offset of hlist_node within the struct.
683 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
684 for (pos = (head)->first; \
685 (!is_a_nulls(pos)) && \
686 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
689 static inline struct user_namespace *sk_user_ns(struct sock *sk)
691 /* Careful only use this in a context where these parameters
692 * can not change and must all be valid, such as recvmsg from
695 return sk->sk_socket->file->f_cred->user_ns;
709 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
710 SOCK_DBG, /* %SO_DEBUG setting */
711 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
712 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
713 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
714 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
715 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
716 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
717 SOCK_FASYNC, /* fasync() active */
719 SOCK_ZEROCOPY, /* buffers from userspace */
720 SOCK_WIFI_STATUS, /* push wifi status to userspace */
721 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
722 * Will use last 4 bytes of packet sent from
723 * user-space instead.
725 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
726 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
729 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
731 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
733 nsk->sk_flags = osk->sk_flags;
736 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
738 __set_bit(flag, &sk->sk_flags);
741 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
743 __clear_bit(flag, &sk->sk_flags);
746 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
748 return test_bit(flag, &sk->sk_flags);
752 extern struct static_key memalloc_socks;
753 static inline int sk_memalloc_socks(void)
755 return static_key_false(&memalloc_socks);
759 static inline int sk_memalloc_socks(void)
766 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
768 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
771 static inline void sk_acceptq_removed(struct sock *sk)
773 sk->sk_ack_backlog--;
776 static inline void sk_acceptq_added(struct sock *sk)
778 sk->sk_ack_backlog++;
781 static inline bool sk_acceptq_is_full(const struct sock *sk)
783 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
787 * Compute minimal free write space needed to queue new packets.
789 static inline int sk_stream_min_wspace(const struct sock *sk)
791 return sk->sk_wmem_queued >> 1;
794 static inline int sk_stream_wspace(const struct sock *sk)
796 return sk->sk_sndbuf - sk->sk_wmem_queued;
799 void sk_stream_write_space(struct sock *sk);
801 /* OOB backlog add */
802 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
804 /* dont let skb dst not refcounted, we are going to leave rcu lock */
805 skb_dst_force_safe(skb);
807 if (!sk->sk_backlog.tail)
808 sk->sk_backlog.head = skb;
810 sk->sk_backlog.tail->next = skb;
812 sk->sk_backlog.tail = skb;
817 * Take into account size of receive queue and backlog queue
818 * Do not take into account this skb truesize,
819 * to allow even a single big packet to come.
821 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
823 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
825 return qsize > limit;
828 /* The per-socket spinlock must be held here. */
829 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
832 if (sk_rcvqueues_full(sk, limit))
836 * If the skb was allocated from pfmemalloc reserves, only
837 * allow SOCK_MEMALLOC sockets to use it as this socket is
838 * helping free memory
840 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
843 __sk_add_backlog(sk, skb);
844 sk->sk_backlog.len += skb->truesize;
848 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
850 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
852 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
853 return __sk_backlog_rcv(sk, skb);
855 return sk->sk_backlog_rcv(sk, skb);
858 static inline void sk_incoming_cpu_update(struct sock *sk)
860 sk->sk_incoming_cpu = raw_smp_processor_id();
863 static inline void sock_rps_record_flow_hash(__u32 hash)
866 struct rps_sock_flow_table *sock_flow_table;
869 sock_flow_table = rcu_dereference(rps_sock_flow_table);
870 rps_record_sock_flow(sock_flow_table, hash);
875 static inline void sock_rps_record_flow(const struct sock *sk)
878 sock_rps_record_flow_hash(sk->sk_rxhash);
882 static inline void sock_rps_save_rxhash(struct sock *sk,
883 const struct sk_buff *skb)
886 if (unlikely(sk->sk_rxhash != skb->hash))
887 sk->sk_rxhash = skb->hash;
891 static inline void sock_rps_reset_rxhash(struct sock *sk)
898 #define sk_wait_event(__sk, __timeo, __condition) \
900 release_sock(__sk); \
901 __rc = __condition; \
903 *(__timeo) = schedule_timeout(*(__timeo)); \
905 sched_annotate_sleep(); \
907 __rc = __condition; \
911 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
912 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
913 void sk_stream_wait_close(struct sock *sk, long timeo_p);
914 int sk_stream_error(struct sock *sk, int flags, int err);
915 void sk_stream_kill_queues(struct sock *sk);
916 void sk_set_memalloc(struct sock *sk);
917 void sk_clear_memalloc(struct sock *sk);
919 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
921 struct request_sock_ops;
922 struct timewait_sock_ops;
923 struct inet_hashinfo;
928 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
929 * un-modified. Special care is taken when initializing object to zero.
931 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
933 if (offsetof(struct sock, sk_node.next) != 0)
934 memset(sk, 0, offsetof(struct sock, sk_node.next));
935 memset(&sk->sk_node.pprev, 0,
936 size - offsetof(struct sock, sk_node.pprev));
939 /* Networking protocol blocks we attach to sockets.
940 * socket layer -> transport layer interface
943 void (*close)(struct sock *sk,
945 int (*connect)(struct sock *sk,
946 struct sockaddr *uaddr,
948 int (*disconnect)(struct sock *sk, int flags);
950 struct sock * (*accept)(struct sock *sk, int flags, int *err);
952 int (*ioctl)(struct sock *sk, int cmd,
954 int (*init)(struct sock *sk);
955 void (*destroy)(struct sock *sk);
956 void (*shutdown)(struct sock *sk, int how);
957 int (*setsockopt)(struct sock *sk, int level,
958 int optname, char __user *optval,
959 unsigned int optlen);
960 int (*getsockopt)(struct sock *sk, int level,
961 int optname, char __user *optval,
964 int (*compat_setsockopt)(struct sock *sk,
966 int optname, char __user *optval,
967 unsigned int optlen);
968 int (*compat_getsockopt)(struct sock *sk,
970 int optname, char __user *optval,
972 int (*compat_ioctl)(struct sock *sk,
973 unsigned int cmd, unsigned long arg);
975 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
977 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
978 size_t len, int noblock, int flags,
980 int (*sendpage)(struct sock *sk, struct page *page,
981 int offset, size_t size, int flags);
982 int (*bind)(struct sock *sk,
983 struct sockaddr *uaddr, int addr_len);
985 int (*backlog_rcv) (struct sock *sk,
986 struct sk_buff *skb);
988 void (*release_cb)(struct sock *sk);
990 /* Keeping track of sk's, looking them up, and port selection methods. */
991 int (*hash)(struct sock *sk);
992 void (*unhash)(struct sock *sk);
993 void (*rehash)(struct sock *sk);
994 int (*get_port)(struct sock *sk, unsigned short snum);
995 void (*clear_sk)(struct sock *sk, int size);
997 /* Keeping track of sockets in use */
998 #ifdef CONFIG_PROC_FS
999 unsigned int inuse_idx;
1002 bool (*stream_memory_free)(const struct sock *sk);
1003 /* Memory pressure */
1004 void (*enter_memory_pressure)(struct sock *sk);
1005 atomic_long_t *memory_allocated; /* Current allocated memory. */
1006 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1008 * Pressure flag: try to collapse.
1009 * Technical note: it is used by multiple contexts non atomically.
1010 * All the __sk_mem_schedule() is of this nature: accounting
1011 * is strict, actions are advisory and have some latency.
1013 int *memory_pressure;
1020 struct kmem_cache *slab;
1021 unsigned int obj_size;
1024 struct percpu_counter *orphan_count;
1026 struct request_sock_ops *rsk_prot;
1027 struct timewait_sock_ops *twsk_prot;
1030 struct inet_hashinfo *hashinfo;
1031 struct udp_table *udp_table;
1032 struct raw_hashinfo *raw_hash;
1035 struct module *owner;
1039 struct list_head node;
1040 #ifdef SOCK_REFCNT_DEBUG
1043 int (*diag_destroy)(struct sock *sk, int err);
1046 int proto_register(struct proto *prot, int alloc_slab);
1047 void proto_unregister(struct proto *prot);
1049 #ifdef SOCK_REFCNT_DEBUG
1050 static inline void sk_refcnt_debug_inc(struct sock *sk)
1052 atomic_inc(&sk->sk_prot->socks);
1055 static inline void sk_refcnt_debug_dec(struct sock *sk)
1057 atomic_dec(&sk->sk_prot->socks);
1058 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1059 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1062 static inline void sk_refcnt_debug_release(const struct sock *sk)
1064 if (atomic_read(&sk->sk_refcnt) != 1)
1065 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1066 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1068 #else /* SOCK_REFCNT_DEBUG */
1069 #define sk_refcnt_debug_inc(sk) do { } while (0)
1070 #define sk_refcnt_debug_dec(sk) do { } while (0)
1071 #define sk_refcnt_debug_release(sk) do { } while (0)
1072 #endif /* SOCK_REFCNT_DEBUG */
1074 static inline bool sk_stream_memory_free(const struct sock *sk)
1076 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1079 return sk->sk_prot->stream_memory_free ?
1080 sk->sk_prot->stream_memory_free(sk) : true;
1083 static inline bool sk_stream_is_writeable(const struct sock *sk)
1085 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1086 sk_stream_memory_free(sk);
1090 static inline bool sk_has_memory_pressure(const struct sock *sk)
1092 return sk->sk_prot->memory_pressure != NULL;
1095 static inline bool sk_under_memory_pressure(const struct sock *sk)
1097 if (!sk->sk_prot->memory_pressure)
1100 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1101 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1104 return !!*sk->sk_prot->memory_pressure;
1107 static inline void sk_leave_memory_pressure(struct sock *sk)
1109 int *memory_pressure = sk->sk_prot->memory_pressure;
1111 if (!memory_pressure)
1114 if (*memory_pressure)
1115 *memory_pressure = 0;
1118 static inline void sk_enter_memory_pressure(struct sock *sk)
1120 if (!sk->sk_prot->enter_memory_pressure)
1123 sk->sk_prot->enter_memory_pressure(sk);
1126 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1128 return sk->sk_prot->sysctl_mem[index];
1132 sk_memory_allocated(const struct sock *sk)
1134 return atomic_long_read(sk->sk_prot->memory_allocated);
1138 sk_memory_allocated_add(struct sock *sk, int amt)
1140 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1144 sk_memory_allocated_sub(struct sock *sk, int amt)
1146 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1149 static inline void sk_sockets_allocated_dec(struct sock *sk)
1151 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1154 static inline void sk_sockets_allocated_inc(struct sock *sk)
1156 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1160 sk_sockets_allocated_read_positive(struct sock *sk)
1162 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1166 proto_sockets_allocated_sum_positive(struct proto *prot)
1168 return percpu_counter_sum_positive(prot->sockets_allocated);
1172 proto_memory_allocated(struct proto *prot)
1174 return atomic_long_read(prot->memory_allocated);
1178 proto_memory_pressure(struct proto *prot)
1180 if (!prot->memory_pressure)
1182 return !!*prot->memory_pressure;
1186 #ifdef CONFIG_PROC_FS
1187 /* Called with local bh disabled */
1188 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1189 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1191 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1198 /* With per-bucket locks this operation is not-atomic, so that
1199 * this version is not worse.
1201 static inline int __sk_prot_rehash(struct sock *sk)
1203 sk->sk_prot->unhash(sk);
1204 return sk->sk_prot->hash(sk);
1207 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1209 /* About 10 seconds */
1210 #define SOCK_DESTROY_TIME (10*HZ)
1212 /* Sockets 0-1023 can't be bound to unless you are superuser */
1213 #define PROT_SOCK 1024
1215 #define SHUTDOWN_MASK 3
1216 #define RCV_SHUTDOWN 1
1217 #define SEND_SHUTDOWN 2
1219 #define SOCK_SNDBUF_LOCK 1
1220 #define SOCK_RCVBUF_LOCK 2
1221 #define SOCK_BINDADDR_LOCK 4
1222 #define SOCK_BINDPORT_LOCK 8
1224 struct socket_alloc {
1225 struct socket socket;
1226 struct inode vfs_inode;
1229 static inline struct socket *SOCKET_I(struct inode *inode)
1231 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1234 static inline struct inode *SOCK_INODE(struct socket *socket)
1236 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1240 * Functions for memory accounting
1242 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1243 void __sk_mem_reclaim(struct sock *sk, int amount);
1245 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1246 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1247 #define SK_MEM_SEND 0
1248 #define SK_MEM_RECV 1
1250 static inline int sk_mem_pages(int amt)
1252 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1255 static inline bool sk_has_account(struct sock *sk)
1257 /* return true if protocol supports memory accounting */
1258 return !!sk->sk_prot->memory_allocated;
1261 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1263 if (!sk_has_account(sk))
1265 return size <= sk->sk_forward_alloc ||
1266 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1270 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1272 if (!sk_has_account(sk))
1274 return size<= sk->sk_forward_alloc ||
1275 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1276 skb_pfmemalloc(skb);
1279 static inline void sk_mem_reclaim(struct sock *sk)
1281 if (!sk_has_account(sk))
1283 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1284 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1287 static inline void sk_mem_reclaim_partial(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 - 1);
1295 static inline void sk_mem_charge(struct sock *sk, int size)
1297 if (!sk_has_account(sk))
1299 sk->sk_forward_alloc -= size;
1302 static inline void sk_mem_uncharge(struct sock *sk, int size)
1304 if (!sk_has_account(sk))
1306 sk->sk_forward_alloc += size;
1309 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1311 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1312 sk->sk_wmem_queued -= skb->truesize;
1313 sk_mem_uncharge(sk, skb->truesize);
1317 /* Used by processes to "lock" a socket state, so that
1318 * interrupts and bottom half handlers won't change it
1319 * from under us. It essentially blocks any incoming
1320 * packets, so that we won't get any new data or any
1321 * packets that change the state of the socket.
1323 * While locked, BH processing will add new packets to
1324 * the backlog queue. This queue is processed by the
1325 * owner of the socket lock right before it is released.
1327 * Since ~2.3.5 it is also exclusive sleep lock serializing
1328 * accesses from user process context.
1330 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1332 static inline void sock_release_ownership(struct sock *sk)
1334 sk->sk_lock.owned = 0;
1338 * Macro so as to not evaluate some arguments when
1339 * lockdep is not enabled.
1341 * Mark both the sk_lock and the sk_lock.slock as a
1342 * per-address-family lock class.
1344 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1346 sk->sk_lock.owned = 0; \
1347 init_waitqueue_head(&sk->sk_lock.wq); \
1348 spin_lock_init(&(sk)->sk_lock.slock); \
1349 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1350 sizeof((sk)->sk_lock)); \
1351 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1353 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1356 void lock_sock_nested(struct sock *sk, int subclass);
1358 static inline void lock_sock(struct sock *sk)
1360 lock_sock_nested(sk, 0);
1363 void release_sock(struct sock *sk);
1365 /* BH context may only use the following locking interface. */
1366 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1367 #define bh_lock_sock_nested(__sk) \
1368 spin_lock_nested(&((__sk)->sk_lock.slock), \
1369 SINGLE_DEPTH_NESTING)
1370 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1372 bool lock_sock_fast(struct sock *sk);
1374 * unlock_sock_fast - complement of lock_sock_fast
1378 * fast unlock socket for user context.
1379 * If slow mode is on, we call regular release_sock()
1381 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1386 spin_unlock_bh(&sk->sk_lock.slock);
1390 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1391 struct proto *prot, int kern);
1392 void sk_free(struct sock *sk);
1393 void sk_destruct(struct sock *sk);
1394 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1396 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1398 void sock_wfree(struct sk_buff *skb);
1399 void skb_orphan_partial(struct sk_buff *skb);
1400 void sock_rfree(struct sk_buff *skb);
1401 void sock_efree(struct sk_buff *skb);
1403 void sock_edemux(struct sk_buff *skb);
1405 #define sock_edemux(skb) sock_efree(skb)
1408 int sock_setsockopt(struct socket *sock, int level, int op,
1409 char __user *optval, unsigned int optlen);
1411 int sock_getsockopt(struct socket *sock, int level, int op,
1412 char __user *optval, int __user *optlen);
1413 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1414 int noblock, int *errcode);
1415 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1416 unsigned long data_len, int noblock,
1417 int *errcode, int max_page_order);
1418 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1419 void sock_kfree_s(struct sock *sk, void *mem, int size);
1420 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1421 void sk_send_sigurg(struct sock *sk);
1423 struct sockcm_cookie {
1427 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1428 struct sockcm_cookie *sockc);
1431 * Functions to fill in entries in struct proto_ops when a protocol
1432 * does not implement a particular function.
1434 int sock_no_bind(struct socket *, struct sockaddr *, int);
1435 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1436 int sock_no_socketpair(struct socket *, struct socket *);
1437 int sock_no_accept(struct socket *, struct socket *, int);
1438 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1439 unsigned int sock_no_poll(struct file *, struct socket *,
1440 struct poll_table_struct *);
1441 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1442 int sock_no_listen(struct socket *, int);
1443 int sock_no_shutdown(struct socket *, int);
1444 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1445 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1446 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1447 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1448 int sock_no_mmap(struct file *file, struct socket *sock,
1449 struct vm_area_struct *vma);
1450 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1451 size_t size, int flags);
1454 * Functions to fill in entries in struct proto_ops when a protocol
1455 * uses the inet style.
1457 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1458 char __user *optval, int __user *optlen);
1459 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1461 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1462 char __user *optval, unsigned int optlen);
1463 int compat_sock_common_getsockopt(struct socket *sock, int level,
1464 int optname, char __user *optval, int __user *optlen);
1465 int compat_sock_common_setsockopt(struct socket *sock, int level,
1466 int optname, char __user *optval, unsigned int optlen);
1468 void sk_common_release(struct sock *sk);
1471 * Default socket callbacks and setup code
1474 /* Initialise core socket variables */
1475 void sock_init_data(struct socket *sock, struct sock *sk);
1478 * Socket reference counting postulates.
1480 * * Each user of socket SHOULD hold a reference count.
1481 * * Each access point to socket (an hash table bucket, reference from a list,
1482 * running timer, skb in flight MUST hold a reference count.
1483 * * When reference count hits 0, it means it will never increase back.
1484 * * When reference count hits 0, it means that no references from
1485 * outside exist to this socket and current process on current CPU
1486 * is last user and may/should destroy this socket.
1487 * * sk_free is called from any context: process, BH, IRQ. When
1488 * it is called, socket has no references from outside -> sk_free
1489 * may release descendant resources allocated by the socket, but
1490 * to the time when it is called, socket is NOT referenced by any
1491 * hash tables, lists etc.
1492 * * Packets, delivered from outside (from network or from another process)
1493 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1494 * when they sit in queue. Otherwise, packets will leak to hole, when
1495 * socket is looked up by one cpu and unhasing is made by another CPU.
1496 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1497 * (leak to backlog). Packet socket does all the processing inside
1498 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1499 * use separate SMP lock, so that they are prone too.
1502 /* Ungrab socket and destroy it, if it was the last reference. */
1503 static inline void sock_put(struct sock *sk)
1505 if (atomic_dec_and_test(&sk->sk_refcnt))
1508 /* Generic version of sock_put(), dealing with all sockets
1509 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1511 void sock_gen_put(struct sock *sk);
1513 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1515 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1517 sk->sk_tx_queue_mapping = tx_queue;
1520 static inline void sk_tx_queue_clear(struct sock *sk)
1522 sk->sk_tx_queue_mapping = -1;
1525 static inline int sk_tx_queue_get(const struct sock *sk)
1527 return sk ? sk->sk_tx_queue_mapping : -1;
1530 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1532 sk_tx_queue_clear(sk);
1533 sk->sk_socket = sock;
1536 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1538 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1539 return &rcu_dereference_raw(sk->sk_wq)->wait;
1541 /* Detach socket from process context.
1542 * Announce socket dead, detach it from wait queue and inode.
1543 * Note that parent inode held reference count on this struct sock,
1544 * we do not release it in this function, because protocol
1545 * probably wants some additional cleanups or even continuing
1546 * to work with this socket (TCP).
1548 static inline void sock_orphan(struct sock *sk)
1550 write_lock_bh(&sk->sk_callback_lock);
1551 sock_set_flag(sk, SOCK_DEAD);
1552 sk_set_socket(sk, NULL);
1554 write_unlock_bh(&sk->sk_callback_lock);
1557 static inline void sock_graft(struct sock *sk, struct socket *parent)
1559 write_lock_bh(&sk->sk_callback_lock);
1560 sk->sk_wq = parent->wq;
1562 sk_set_socket(sk, parent);
1563 security_sock_graft(sk, parent);
1564 write_unlock_bh(&sk->sk_callback_lock);
1567 kuid_t sock_i_uid(struct sock *sk);
1568 unsigned long sock_i_ino(struct sock *sk);
1570 static inline u32 net_tx_rndhash(void)
1572 u32 v = prandom_u32();
1577 static inline void sk_set_txhash(struct sock *sk)
1579 sk->sk_txhash = net_tx_rndhash();
1582 static inline void sk_rethink_txhash(struct sock *sk)
1588 static inline struct dst_entry *
1589 __sk_dst_get(struct sock *sk)
1591 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1592 lockdep_is_held(&sk->sk_lock.slock));
1595 static inline struct dst_entry *
1596 sk_dst_get(struct sock *sk)
1598 struct dst_entry *dst;
1601 dst = rcu_dereference(sk->sk_dst_cache);
1602 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1608 static inline void dst_negative_advice(struct sock *sk)
1610 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1612 sk_rethink_txhash(sk);
1614 if (dst && dst->ops->negative_advice) {
1615 ndst = dst->ops->negative_advice(dst);
1618 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1619 sk_tx_queue_clear(sk);
1625 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1627 struct dst_entry *old_dst;
1629 sk_tx_queue_clear(sk);
1631 * This can be called while sk is owned by the caller only,
1632 * with no state that can be checked in a rcu_dereference_check() cond
1634 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1635 rcu_assign_pointer(sk->sk_dst_cache, dst);
1636 dst_release(old_dst);
1640 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1642 struct dst_entry *old_dst;
1644 sk_tx_queue_clear(sk);
1645 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1646 dst_release(old_dst);
1650 __sk_dst_reset(struct sock *sk)
1652 __sk_dst_set(sk, NULL);
1656 sk_dst_reset(struct sock *sk)
1658 sk_dst_set(sk, NULL);
1661 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1663 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1665 bool sk_mc_loop(struct sock *sk);
1667 static inline bool sk_can_gso(const struct sock *sk)
1669 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1672 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1674 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1676 sk->sk_route_nocaps |= flags;
1677 sk->sk_route_caps &= ~flags;
1680 static inline bool sk_check_csum_caps(struct sock *sk)
1682 return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1683 (sk->sk_family == PF_INET &&
1684 (sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1685 (sk->sk_family == PF_INET6 &&
1686 (sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1689 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1690 struct iov_iter *from, char *to,
1691 int copy, int offset)
1693 if (skb->ip_summed == CHECKSUM_NONE) {
1695 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1697 skb->csum = csum_block_add(skb->csum, csum, offset);
1698 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1699 if (copy_from_iter_nocache(to, copy, from) != copy)
1701 } else if (copy_from_iter(to, copy, from) != copy)
1707 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1708 struct iov_iter *from, int copy)
1710 int err, offset = skb->len;
1712 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1715 __skb_trim(skb, offset);
1720 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1721 struct sk_buff *skb,
1727 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1733 skb->data_len += copy;
1734 skb->truesize += copy;
1735 sk->sk_wmem_queued += copy;
1736 sk_mem_charge(sk, copy);
1741 * sk_wmem_alloc_get - returns write allocations
1744 * Returns sk_wmem_alloc minus initial offset of one
1746 static inline int sk_wmem_alloc_get(const struct sock *sk)
1748 return atomic_read(&sk->sk_wmem_alloc) - 1;
1752 * sk_rmem_alloc_get - returns read allocations
1755 * Returns sk_rmem_alloc
1757 static inline int sk_rmem_alloc_get(const struct sock *sk)
1759 return atomic_read(&sk->sk_rmem_alloc);
1763 * sk_has_allocations - check if allocations are outstanding
1766 * Returns true if socket has write or read allocations
1768 static inline bool sk_has_allocations(const struct sock *sk)
1770 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1774 * skwq_has_sleeper - check if there are any waiting processes
1775 * @wq: struct socket_wq
1777 * Returns true if socket_wq has waiting processes
1779 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1780 * barrier call. They were added due to the race found within the tcp code.
1782 * Consider following tcp code paths:
1786 * sys_select receive packet
1788 * __add_wait_queue update tp->rcv_nxt
1790 * tp->rcv_nxt check sock_def_readable
1792 * schedule rcu_read_lock();
1793 * wq = rcu_dereference(sk->sk_wq);
1794 * if (wq && waitqueue_active(&wq->wait))
1795 * wake_up_interruptible(&wq->wait)
1799 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1800 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1801 * could then endup calling schedule and sleep forever if there are no more
1802 * data on the socket.
1805 static inline bool skwq_has_sleeper(struct socket_wq *wq)
1807 return wq && wq_has_sleeper(&wq->wait);
1811 * sock_poll_wait - place memory barrier behind the poll_wait call.
1813 * @wait_address: socket wait queue
1816 * See the comments in the wq_has_sleeper function.
1818 static inline void sock_poll_wait(struct file *filp,
1819 wait_queue_head_t *wait_address, poll_table *p)
1821 if (!poll_does_not_wait(p) && wait_address) {
1822 poll_wait(filp, wait_address, p);
1823 /* We need to be sure we are in sync with the
1824 * socket flags modification.
1826 * This memory barrier is paired in the wq_has_sleeper.
1832 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1834 if (sk->sk_txhash) {
1836 skb->hash = sk->sk_txhash;
1840 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
1843 * Queue a received datagram if it will fit. Stream and sequenced
1844 * protocols can't normally use this as they need to fit buffers in
1845 * and play with them.
1847 * Inlined as it's very short and called for pretty much every
1848 * packet ever received.
1850 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1854 skb->destructor = sock_rfree;
1855 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1856 sk_mem_charge(sk, skb->truesize);
1859 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1860 unsigned long expires);
1862 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1864 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1866 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1867 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1870 * Recover an error report and clear atomically
1873 static inline int sock_error(struct sock *sk)
1876 if (likely(!sk->sk_err))
1878 err = xchg(&sk->sk_err, 0);
1882 static inline unsigned long sock_wspace(struct sock *sk)
1886 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1887 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1895 * We use sk->sk_wq_raw, from contexts knowing this
1896 * pointer is not NULL and cannot disappear/change.
1898 static inline void sk_set_bit(int nr, struct sock *sk)
1900 set_bit(nr, &sk->sk_wq_raw->flags);
1903 static inline void sk_clear_bit(int nr, struct sock *sk)
1905 clear_bit(nr, &sk->sk_wq_raw->flags);
1908 static inline void sk_wake_async(const struct sock *sk, int how, int band)
1910 if (sock_flag(sk, SOCK_FASYNC)) {
1912 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
1917 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
1918 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
1919 * Note: for send buffers, TCP works better if we can build two skbs at
1922 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
1924 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
1925 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
1927 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1929 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1930 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1931 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1935 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
1936 bool force_schedule);
1939 * sk_page_frag - return an appropriate page_frag
1942 * If socket allocation mode allows current thread to sleep, it means its
1943 * safe to use the per task page_frag instead of the per socket one.
1945 static inline struct page_frag *sk_page_frag(struct sock *sk)
1947 if (gfpflags_allow_blocking(sk->sk_allocation))
1948 return ¤t->task_frag;
1950 return &sk->sk_frag;
1953 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
1956 * Default write policy as shown to user space via poll/select/SIGIO
1958 static inline bool sock_writeable(const struct sock *sk)
1960 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1963 static inline gfp_t gfp_any(void)
1965 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1968 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
1970 return noblock ? 0 : sk->sk_rcvtimeo;
1973 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
1975 return noblock ? 0 : sk->sk_sndtimeo;
1978 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1980 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1983 /* Alas, with timeout socket operations are not restartable.
1984 * Compare this to poll().
1986 static inline int sock_intr_errno(long timeo)
1988 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1991 struct sock_skb_cb {
1995 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
1996 * using skb->cb[] would keep using it directly and utilize its
1997 * alignement guarantee.
1999 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2000 sizeof(struct sock_skb_cb)))
2002 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2003 SOCK_SKB_CB_OFFSET))
2005 #define sock_skb_cb_check_size(size) \
2006 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2009 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2011 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2014 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2015 struct sk_buff *skb);
2016 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2017 struct sk_buff *skb);
2020 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2022 ktime_t kt = skb->tstamp;
2023 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2026 * generate control messages if
2027 * - receive time stamping in software requested
2028 * - software time stamp available and wanted
2029 * - hardware time stamps available and wanted
2031 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2032 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2033 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2034 (hwtstamps->hwtstamp.tv64 &&
2035 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2036 __sock_recv_timestamp(msg, sk, skb);
2040 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2041 __sock_recv_wifi_status(msg, sk, skb);
2044 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2045 struct sk_buff *skb);
2047 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2048 struct sk_buff *skb)
2050 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2051 (1UL << SOCK_RCVTSTAMP))
2052 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2053 SOF_TIMESTAMPING_RAW_HARDWARE)
2055 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2056 __sock_recv_ts_and_drops(msg, sk, skb);
2058 sk->sk_stamp = skb->tstamp;
2061 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2064 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2065 * @sk: socket sending this packet
2066 * @tx_flags: completed with instructions for time stamping
2068 * Note : callers should take care of initial *tx_flags value (usually 0)
2070 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2072 if (unlikely(sk->sk_tsflags))
2073 __sock_tx_timestamp(sk, tx_flags);
2074 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2075 *tx_flags |= SKBTX_WIFI_STATUS;
2079 * sk_eat_skb - Release a skb if it is no longer needed
2080 * @sk: socket to eat this skb from
2081 * @skb: socket buffer to eat
2083 * This routine must be called with interrupts disabled or with the socket
2084 * locked so that the sk_buff queue operation is ok.
2086 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2088 __skb_unlink(skb, &sk->sk_receive_queue);
2093 struct net *sock_net(const struct sock *sk)
2095 return read_pnet(&sk->sk_net);
2099 void sock_net_set(struct sock *sk, struct net *net)
2101 write_pnet(&sk->sk_net, net);
2104 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2107 struct sock *sk = skb->sk;
2109 skb->destructor = NULL;
2116 /* This helper checks if a socket is a full socket,
2117 * ie _not_ a timewait or request socket.
2119 static inline bool sk_fullsock(const struct sock *sk)
2121 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2124 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2125 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2127 static inline bool sk_listener(const struct sock *sk)
2129 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2133 * sk_state_load - read sk->sk_state for lockless contexts
2134 * @sk: socket pointer
2136 * Paired with sk_state_store(). Used in places we do not hold socket lock :
2137 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2139 static inline int sk_state_load(const struct sock *sk)
2141 return smp_load_acquire(&sk->sk_state);
2145 * sk_state_store - update sk->sk_state
2146 * @sk: socket pointer
2147 * @newstate: new state
2149 * Paired with sk_state_load(). Should be used in contexts where
2150 * state change might impact lockless readers.
2152 static inline void sk_state_store(struct sock *sk, int newstate)
2154 smp_store_release(&sk->sk_state, newstate);
2157 void sock_enable_timestamp(struct sock *sk, int flag);
2158 int sock_get_timestamp(struct sock *, struct timeval __user *);
2159 int sock_get_timestampns(struct sock *, struct timespec __user *);
2160 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2163 bool sk_ns_capable(const struct sock *sk,
2164 struct user_namespace *user_ns, int cap);
2165 bool sk_capable(const struct sock *sk, int cap);
2166 bool sk_net_capable(const struct sock *sk, int cap);
2168 extern __u32 sysctl_wmem_max;
2169 extern __u32 sysctl_rmem_max;
2171 extern int sysctl_tstamp_allow_data;
2172 extern int sysctl_optmem_max;
2174 extern __u32 sysctl_wmem_default;
2175 extern __u32 sysctl_rmem_default;
2177 #endif /* _SOCK_H */