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/aio.h>
61 #include <linux/sched.h>
63 #include <linux/filter.h>
64 #include <linux/rculist_nulls.h>
65 #include <linux/poll.h>
67 #include <linux/atomic.h>
69 #include <net/checksum.h>
70 #include <net/tcp_states.h>
71 #include <linux/net_tstamp.h>
76 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
77 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
80 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
85 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
90 * This structure really needs to be cleaned up.
91 * Most of it is for TCP, and not used by any of
92 * the other protocols.
95 /* Define this to get the SOCK_DBG debugging facility. */
96 #define SOCK_DEBUGGING
98 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
99 printk(KERN_DEBUG msg); } while (0)
101 /* Validate arguments and do nothing */
102 static inline __printf(2, 3)
103 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
108 /* This is the per-socket lock. The spinlock provides a synchronization
109 * between user contexts and software interrupt processing, whereas the
110 * mini-semaphore synchronizes multiple users amongst themselves.
115 wait_queue_head_t wq;
117 * We express the mutex-alike socket_lock semantics
118 * to the lock validator by explicitly managing
119 * the slock as a lock variant (in addition to
122 #ifdef CONFIG_DEBUG_LOCK_ALLOC
123 struct lockdep_map dep_map;
131 typedef __u32 __bitwise __portpair;
132 typedef __u64 __bitwise __addrpair;
135 * struct sock_common - minimal network layer representation of sockets
136 * @skc_daddr: Foreign IPv4 addr
137 * @skc_rcv_saddr: Bound local IPv4 addr
138 * @skc_hash: hash value used with various protocol lookup tables
139 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
140 * @skc_dport: placeholder for inet_dport/tw_dport
141 * @skc_num: placeholder for inet_num/tw_num
142 * @skc_family: network address family
143 * @skc_state: Connection state
144 * @skc_reuse: %SO_REUSEADDR setting
145 * @skc_reuseport: %SO_REUSEPORT setting
146 * @skc_bound_dev_if: bound device index if != 0
147 * @skc_bind_node: bind hash linkage for various protocol lookup tables
148 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
149 * @skc_prot: protocol handlers inside a network family
150 * @skc_net: reference to the network namespace of this socket
151 * @skc_node: main hash linkage for various protocol lookup tables
152 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
153 * @skc_tx_queue_mapping: tx queue number for this connection
154 * @skc_refcnt: reference count
156 * This is the minimal network layer representation of sockets, the header
157 * for struct sock and struct inet_timewait_sock.
160 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
161 * address on 64bit arches : cf INET_MATCH()
164 __addrpair skc_addrpair;
167 __be32 skc_rcv_saddr;
171 unsigned int skc_hash;
172 __u16 skc_u16hashes[2];
174 /* skc_dport && skc_num must be grouped as well */
176 __portpair skc_portpair;
183 unsigned short skc_family;
184 volatile unsigned char skc_state;
185 unsigned char skc_reuse:4;
186 unsigned char skc_reuseport:1;
187 unsigned char skc_ipv6only:1;
188 int skc_bound_dev_if;
190 struct hlist_node skc_bind_node;
191 struct hlist_nulls_node skc_portaddr_node;
193 struct proto *skc_prot;
194 possible_net_t skc_net;
196 #if IS_ENABLED(CONFIG_IPV6)
197 struct in6_addr skc_v6_daddr;
198 struct in6_addr skc_v6_rcv_saddr;
201 atomic64_t skc_cookie;
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;
217 int skc_dontcopy_end[0];
223 * struct sock - network layer representation of sockets
224 * @__sk_common: shared layout with inet_timewait_sock
225 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
226 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
227 * @sk_lock: synchronizer
228 * @sk_rcvbuf: size of receive buffer in bytes
229 * @sk_wq: sock wait queue and async head
230 * @sk_rx_dst: receive input route used by early demux
231 * @sk_dst_cache: destination cache
232 * @sk_dst_lock: destination cache lock
233 * @sk_policy: flow policy
234 * @sk_receive_queue: incoming packets
235 * @sk_wmem_alloc: transmit queue bytes committed
236 * @sk_write_queue: Packet sending queue
237 * @sk_omem_alloc: "o" is "option" or "other"
238 * @sk_wmem_queued: persistent queue size
239 * @sk_forward_alloc: space allocated forward
240 * @sk_napi_id: id of the last napi context to receive data for sk
241 * @sk_ll_usec: usecs to busypoll when there is no data
242 * @sk_allocation: allocation mode
243 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
244 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
245 * @sk_sndbuf: size of send buffer in bytes
246 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
247 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
248 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
249 * @sk_no_check_rx: allow zero checksum in RX packets
250 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
251 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
252 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
253 * @sk_gso_max_size: Maximum GSO segment size to build
254 * @sk_gso_max_segs: Maximum number of GSO segments
255 * @sk_lingertime: %SO_LINGER l_linger setting
256 * @sk_backlog: always used with the per-socket spinlock held
257 * @sk_callback_lock: used with the callbacks in the end of this struct
258 * @sk_error_queue: rarely used
259 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
260 * IPV6_ADDRFORM for instance)
261 * @sk_err: last error
262 * @sk_err_soft: errors that don't cause failure but are the cause of a
263 * persistent failure not just 'timed out'
264 * @sk_drops: raw/udp drops counter
265 * @sk_ack_backlog: current listen backlog
266 * @sk_max_ack_backlog: listen backlog set in listen()
267 * @sk_priority: %SO_PRIORITY setting
268 * @sk_cgrp_prioidx: socket group's priority map index
269 * @sk_type: socket type (%SOCK_STREAM, etc)
270 * @sk_protocol: which protocol this socket belongs in this network family
271 * @sk_peer_pid: &struct pid for this socket's peer
272 * @sk_peer_cred: %SO_PEERCRED setting
273 * @sk_rcvlowat: %SO_RCVLOWAT setting
274 * @sk_rcvtimeo: %SO_RCVTIMEO setting
275 * @sk_sndtimeo: %SO_SNDTIMEO setting
276 * @sk_rxhash: flow hash received from netif layer
277 * @sk_incoming_cpu: record cpu processing incoming packets
278 * @sk_txhash: computed flow hash for use on transmit
279 * @sk_filter: socket filtering instructions
280 * @sk_protinfo: private area, net family specific, when not using slab
281 * @sk_timer: sock cleanup timer
282 * @sk_stamp: time stamp of last packet received
283 * @sk_tsflags: SO_TIMESTAMPING socket options
284 * @sk_tskey: counter to disambiguate concurrent tstamp requests
285 * @sk_socket: Identd and reporting IO signals
286 * @sk_user_data: RPC layer private data
287 * @sk_frag: cached page frag
288 * @sk_peek_off: current peek_offset value
289 * @sk_send_head: front of stuff to transmit
290 * @sk_security: used by security modules
291 * @sk_mark: generic packet mark
292 * @sk_classid: this socket's cgroup classid
293 * @sk_cgrp: this socket's cgroup-specific proto data
294 * @sk_write_pending: a write to stream socket waits to start
295 * @sk_state_change: callback to indicate change in the state of the sock
296 * @sk_data_ready: callback to indicate there is data to be processed
297 * @sk_write_space: callback to indicate there is bf sending space available
298 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
299 * @sk_backlog_rcv: callback to process the backlog
300 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
304 * Now struct inet_timewait_sock also uses sock_common, so please just
305 * don't add nothing before this first member (__sk_common) --acme
307 struct sock_common __sk_common;
308 #define sk_node __sk_common.skc_node
309 #define sk_nulls_node __sk_common.skc_nulls_node
310 #define sk_refcnt __sk_common.skc_refcnt
311 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
313 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
314 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
315 #define sk_hash __sk_common.skc_hash
316 #define sk_portpair __sk_common.skc_portpair
317 #define sk_num __sk_common.skc_num
318 #define sk_dport __sk_common.skc_dport
319 #define sk_addrpair __sk_common.skc_addrpair
320 #define sk_daddr __sk_common.skc_daddr
321 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
322 #define sk_family __sk_common.skc_family
323 #define sk_state __sk_common.skc_state
324 #define sk_reuse __sk_common.skc_reuse
325 #define sk_reuseport __sk_common.skc_reuseport
326 #define sk_ipv6only __sk_common.skc_ipv6only
327 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
328 #define sk_bind_node __sk_common.skc_bind_node
329 #define sk_prot __sk_common.skc_prot
330 #define sk_net __sk_common.skc_net
331 #define sk_v6_daddr __sk_common.skc_v6_daddr
332 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
333 #define sk_cookie __sk_common.skc_cookie
335 socket_lock_t sk_lock;
336 struct sk_buff_head sk_receive_queue;
338 * The backlog queue is special, it is always used with
339 * the per-socket spinlock held and requires low latency
340 * access. Therefore we special case it's implementation.
341 * Note : rmem_alloc is in this structure to fill a hole
342 * on 64bit arches, not because its logically part of
348 struct sk_buff *head;
349 struct sk_buff *tail;
351 #define sk_rmem_alloc sk_backlog.rmem_alloc
352 int sk_forward_alloc;
358 * Warned : sk_incoming_cpu can be set from softirq,
359 * Do not use this hole without fully understanding possible issues.
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;
371 struct socket_wq __rcu *sk_wq;
374 struct xfrm_policy *sk_policy[2];
376 unsigned long sk_flags;
377 struct dst_entry *sk_rx_dst;
378 struct dst_entry __rcu *sk_dst_cache;
379 spinlock_t sk_dst_lock;
380 atomic_t sk_wmem_alloc;
381 atomic_t sk_omem_alloc;
383 struct sk_buff_head sk_write_queue;
384 kmemcheck_bitfield_begin(flags);
385 unsigned int sk_shutdown : 2,
391 kmemcheck_bitfield_end(flags);
394 u32 sk_pacing_rate; /* bytes per second */
395 u32 sk_max_pacing_rate;
396 netdev_features_t sk_route_caps;
397 netdev_features_t sk_route_nocaps;
399 unsigned int sk_gso_max_size;
402 unsigned long sk_lingertime;
403 struct sk_buff_head sk_error_queue;
404 struct proto *sk_prot_creator;
405 rwlock_t sk_callback_lock;
409 u32 sk_max_ack_backlog;
411 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
412 __u32 sk_cgrp_prioidx;
414 struct pid *sk_peer_pid;
415 const struct cred *sk_peer_cred;
419 struct timer_list sk_timer;
423 struct socket *sk_socket;
425 struct page_frag sk_frag;
426 struct sk_buff *sk_send_head;
428 int sk_write_pending;
429 #ifdef CONFIG_SECURITY
434 struct cg_proto *sk_cgrp;
435 void (*sk_state_change)(struct sock *sk);
436 void (*sk_data_ready)(struct sock *sk);
437 void (*sk_write_space)(struct sock *sk);
438 void (*sk_error_report)(struct sock *sk);
439 int (*sk_backlog_rcv)(struct sock *sk,
440 struct sk_buff *skb);
441 void (*sk_destruct)(struct sock *sk);
444 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
446 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
447 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
450 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
451 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
452 * on a socket means that the socket will reuse everybody else's port
453 * without looking at the other's sk_reuse value.
456 #define SK_NO_REUSE 0
457 #define SK_CAN_REUSE 1
458 #define SK_FORCE_REUSE 2
460 static inline int sk_peek_offset(struct sock *sk, int flags)
462 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
463 return sk->sk_peek_off;
468 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
470 if (sk->sk_peek_off >= 0) {
471 if (sk->sk_peek_off >= val)
472 sk->sk_peek_off -= val;
478 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
480 if (sk->sk_peek_off >= 0)
481 sk->sk_peek_off += val;
485 * Hashed lists helper routines
487 static inline struct sock *sk_entry(const struct hlist_node *node)
489 return hlist_entry(node, struct sock, sk_node);
492 static inline struct sock *__sk_head(const struct hlist_head *head)
494 return hlist_entry(head->first, struct sock, sk_node);
497 static inline struct sock *sk_head(const struct hlist_head *head)
499 return hlist_empty(head) ? NULL : __sk_head(head);
502 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
504 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
507 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
509 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
512 static inline struct sock *sk_next(const struct sock *sk)
514 return sk->sk_node.next ?
515 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
518 static inline struct sock *sk_nulls_next(const struct sock *sk)
520 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
521 hlist_nulls_entry(sk->sk_nulls_node.next,
522 struct sock, sk_nulls_node) :
526 static inline bool sk_unhashed(const struct sock *sk)
528 return hlist_unhashed(&sk->sk_node);
531 static inline bool sk_hashed(const struct sock *sk)
533 return !sk_unhashed(sk);
536 static inline void sk_node_init(struct hlist_node *node)
541 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
546 static inline void __sk_del_node(struct sock *sk)
548 __hlist_del(&sk->sk_node);
551 /* NB: equivalent to hlist_del_init_rcu */
552 static inline bool __sk_del_node_init(struct sock *sk)
556 sk_node_init(&sk->sk_node);
562 /* Grab socket reference count. This operation is valid only
563 when sk is ALREADY grabbed f.e. it is found in hash table
564 or a list and the lookup is made under lock preventing hash table
568 static inline void sock_hold(struct sock *sk)
570 atomic_inc(&sk->sk_refcnt);
573 /* Ungrab socket in the context, which assumes that socket refcnt
574 cannot hit zero, f.e. it is true in context of any socketcall.
576 static inline void __sock_put(struct sock *sk)
578 atomic_dec(&sk->sk_refcnt);
581 static inline bool sk_del_node_init(struct sock *sk)
583 bool rc = __sk_del_node_init(sk);
586 /* paranoid for a while -acme */
587 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
592 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
594 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
597 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
603 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
605 bool rc = __sk_nulls_del_node_init_rcu(sk);
608 /* paranoid for a while -acme */
609 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
615 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
617 hlist_add_head(&sk->sk_node, list);
620 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
623 __sk_add_node(sk, list);
626 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
629 hlist_add_head_rcu(&sk->sk_node, list);
632 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
634 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
637 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
640 __sk_nulls_add_node_rcu(sk, list);
643 static inline void __sk_del_bind_node(struct sock *sk)
645 __hlist_del(&sk->sk_bind_node);
648 static inline void sk_add_bind_node(struct sock *sk,
649 struct hlist_head *list)
651 hlist_add_head(&sk->sk_bind_node, list);
654 #define sk_for_each(__sk, list) \
655 hlist_for_each_entry(__sk, list, sk_node)
656 #define sk_for_each_rcu(__sk, list) \
657 hlist_for_each_entry_rcu(__sk, list, sk_node)
658 #define sk_nulls_for_each(__sk, node, list) \
659 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
660 #define sk_nulls_for_each_rcu(__sk, node, list) \
661 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
662 #define sk_for_each_from(__sk) \
663 hlist_for_each_entry_from(__sk, sk_node)
664 #define sk_nulls_for_each_from(__sk, node) \
665 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
666 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
667 #define sk_for_each_safe(__sk, tmp, list) \
668 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
669 #define sk_for_each_bound(__sk, list) \
670 hlist_for_each_entry(__sk, list, sk_bind_node)
673 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
674 * @tpos: the type * to use as a loop cursor.
675 * @pos: the &struct hlist_node to use as a loop cursor.
676 * @head: the head for your list.
677 * @offset: offset of hlist_node within the struct.
680 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
681 for (pos = (head)->first; \
682 (!is_a_nulls(pos)) && \
683 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
686 static inline struct user_namespace *sk_user_ns(struct sock *sk)
688 /* Careful only use this in a context where these parameters
689 * can not change and must all be valid, such as recvmsg from
692 return sk->sk_socket->file->f_cred->user_ns;
706 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
707 SOCK_DBG, /* %SO_DEBUG setting */
708 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
709 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
710 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
711 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
712 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
713 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
714 SOCK_FASYNC, /* fasync() active */
716 SOCK_ZEROCOPY, /* buffers from userspace */
717 SOCK_WIFI_STATUS, /* push wifi status to userspace */
718 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
719 * Will use last 4 bytes of packet sent from
720 * user-space instead.
722 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
723 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
726 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
728 nsk->sk_flags = osk->sk_flags;
731 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
733 __set_bit(flag, &sk->sk_flags);
736 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
738 __clear_bit(flag, &sk->sk_flags);
741 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
743 return test_bit(flag, &sk->sk_flags);
747 extern struct static_key memalloc_socks;
748 static inline int sk_memalloc_socks(void)
750 return static_key_false(&memalloc_socks);
754 static inline int sk_memalloc_socks(void)
761 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
763 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
766 static inline void sk_acceptq_removed(struct sock *sk)
768 sk->sk_ack_backlog--;
771 static inline void sk_acceptq_added(struct sock *sk)
773 sk->sk_ack_backlog++;
776 static inline bool sk_acceptq_is_full(const struct sock *sk)
778 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
782 * Compute minimal free write space needed to queue new packets.
784 static inline int sk_stream_min_wspace(const struct sock *sk)
786 return sk->sk_wmem_queued >> 1;
789 static inline int sk_stream_wspace(const struct sock *sk)
791 return sk->sk_sndbuf - sk->sk_wmem_queued;
794 void sk_stream_write_space(struct sock *sk);
796 /* OOB backlog add */
797 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
799 /* dont let skb dst not refcounted, we are going to leave rcu lock */
802 if (!sk->sk_backlog.tail)
803 sk->sk_backlog.head = skb;
805 sk->sk_backlog.tail->next = skb;
807 sk->sk_backlog.tail = skb;
812 * Take into account size of receive queue and backlog queue
813 * Do not take into account this skb truesize,
814 * to allow even a single big packet to come.
816 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
818 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
820 return qsize > limit;
823 /* The per-socket spinlock must be held here. */
824 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
827 if (sk_rcvqueues_full(sk, limit))
830 __sk_add_backlog(sk, skb);
831 sk->sk_backlog.len += skb->truesize;
835 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
837 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
839 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
840 return __sk_backlog_rcv(sk, skb);
842 return sk->sk_backlog_rcv(sk, skb);
845 static inline void sk_incoming_cpu_update(struct sock *sk)
847 sk->sk_incoming_cpu = raw_smp_processor_id();
850 static inline void sock_rps_record_flow_hash(__u32 hash)
853 struct rps_sock_flow_table *sock_flow_table;
856 sock_flow_table = rcu_dereference(rps_sock_flow_table);
857 rps_record_sock_flow(sock_flow_table, hash);
862 static inline void sock_rps_record_flow(const struct sock *sk)
865 sock_rps_record_flow_hash(sk->sk_rxhash);
869 static inline void sock_rps_save_rxhash(struct sock *sk,
870 const struct sk_buff *skb)
873 if (unlikely(sk->sk_rxhash != skb->hash))
874 sk->sk_rxhash = skb->hash;
878 static inline void sock_rps_reset_rxhash(struct sock *sk)
885 #define sk_wait_event(__sk, __timeo, __condition) \
887 release_sock(__sk); \
888 __rc = __condition; \
890 *(__timeo) = schedule_timeout(*(__timeo)); \
892 sched_annotate_sleep(); \
894 __rc = __condition; \
898 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
899 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
900 void sk_stream_wait_close(struct sock *sk, long timeo_p);
901 int sk_stream_error(struct sock *sk, int flags, int err);
902 void sk_stream_kill_queues(struct sock *sk);
903 void sk_set_memalloc(struct sock *sk);
904 void sk_clear_memalloc(struct sock *sk);
906 int sk_wait_data(struct sock *sk, long *timeo);
908 struct request_sock_ops;
909 struct timewait_sock_ops;
910 struct inet_hashinfo;
915 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
916 * un-modified. Special care is taken when initializing object to zero.
918 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
920 if (offsetof(struct sock, sk_node.next) != 0)
921 memset(sk, 0, offsetof(struct sock, sk_node.next));
922 memset(&sk->sk_node.pprev, 0,
923 size - offsetof(struct sock, sk_node.pprev));
926 /* Networking protocol blocks we attach to sockets.
927 * socket layer -> transport layer interface
928 * transport -> network interface is defined by struct inet_proto
931 void (*close)(struct sock *sk,
933 int (*connect)(struct sock *sk,
934 struct sockaddr *uaddr,
936 int (*disconnect)(struct sock *sk, int flags);
938 struct sock * (*accept)(struct sock *sk, int flags, int *err);
940 int (*ioctl)(struct sock *sk, int cmd,
942 int (*init)(struct sock *sk);
943 void (*destroy)(struct sock *sk);
944 void (*shutdown)(struct sock *sk, int how);
945 int (*setsockopt)(struct sock *sk, int level,
946 int optname, char __user *optval,
947 unsigned int optlen);
948 int (*getsockopt)(struct sock *sk, int level,
949 int optname, char __user *optval,
952 int (*compat_setsockopt)(struct sock *sk,
954 int optname, char __user *optval,
955 unsigned int optlen);
956 int (*compat_getsockopt)(struct sock *sk,
958 int optname, char __user *optval,
960 int (*compat_ioctl)(struct sock *sk,
961 unsigned int cmd, unsigned long arg);
963 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
965 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
966 size_t len, int noblock, int flags,
968 int (*sendpage)(struct sock *sk, struct page *page,
969 int offset, size_t size, int flags);
970 int (*bind)(struct sock *sk,
971 struct sockaddr *uaddr, int addr_len);
973 int (*backlog_rcv) (struct sock *sk,
974 struct sk_buff *skb);
976 void (*release_cb)(struct sock *sk);
978 /* Keeping track of sk's, looking them up, and port selection methods. */
979 void (*hash)(struct sock *sk);
980 void (*unhash)(struct sock *sk);
981 void (*rehash)(struct sock *sk);
982 int (*get_port)(struct sock *sk, unsigned short snum);
983 void (*clear_sk)(struct sock *sk, int size);
985 /* Keeping track of sockets in use */
986 #ifdef CONFIG_PROC_FS
987 unsigned int inuse_idx;
990 bool (*stream_memory_free)(const struct sock *sk);
991 /* Memory pressure */
992 void (*enter_memory_pressure)(struct sock *sk);
993 atomic_long_t *memory_allocated; /* Current allocated memory. */
994 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
996 * Pressure flag: try to collapse.
997 * Technical note: it is used by multiple contexts non atomically.
998 * All the __sk_mem_schedule() is of this nature: accounting
999 * is strict, actions are advisory and have some latency.
1001 int *memory_pressure;
1008 struct kmem_cache *slab;
1009 unsigned int obj_size;
1012 struct percpu_counter *orphan_count;
1014 struct request_sock_ops *rsk_prot;
1015 struct timewait_sock_ops *twsk_prot;
1018 struct inet_hashinfo *hashinfo;
1019 struct udp_table *udp_table;
1020 struct raw_hashinfo *raw_hash;
1023 struct module *owner;
1027 struct list_head node;
1028 #ifdef SOCK_REFCNT_DEBUG
1031 #ifdef CONFIG_MEMCG_KMEM
1033 * cgroup specific init/deinit functions. Called once for all
1034 * protocols that implement it, from cgroups populate function.
1035 * This function has to setup any files the protocol want to
1036 * appear in the kmem cgroup filesystem.
1038 int (*init_cgroup)(struct mem_cgroup *memcg,
1039 struct cgroup_subsys *ss);
1040 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1041 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1046 * Bits in struct cg_proto.flags
1048 enum cg_proto_flags {
1049 /* Currently active and new sockets should be assigned to cgroups */
1051 /* It was ever activated; we must disarm static keys on destruction */
1052 MEMCG_SOCK_ACTIVATED,
1056 struct page_counter memory_allocated; /* Current allocated memory. */
1057 struct percpu_counter sockets_allocated; /* Current number of sockets. */
1058 int memory_pressure;
1060 unsigned long flags;
1062 * memcg field is used to find which memcg we belong directly
1063 * Each memcg struct can hold more than one cg_proto, so container_of
1066 * The elegant solution would be having an inverse function to
1067 * proto_cgroup in struct proto, but that means polluting the structure
1068 * for everybody, instead of just for memcg users.
1070 struct mem_cgroup *memcg;
1073 int proto_register(struct proto *prot, int alloc_slab);
1074 void proto_unregister(struct proto *prot);
1076 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1078 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1081 #ifdef SOCK_REFCNT_DEBUG
1082 static inline void sk_refcnt_debug_inc(struct sock *sk)
1084 atomic_inc(&sk->sk_prot->socks);
1087 static inline void sk_refcnt_debug_dec(struct sock *sk)
1089 atomic_dec(&sk->sk_prot->socks);
1090 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1091 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1094 static inline void sk_refcnt_debug_release(const struct sock *sk)
1096 if (atomic_read(&sk->sk_refcnt) != 1)
1097 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1098 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1100 #else /* SOCK_REFCNT_DEBUG */
1101 #define sk_refcnt_debug_inc(sk) do { } while (0)
1102 #define sk_refcnt_debug_dec(sk) do { } while (0)
1103 #define sk_refcnt_debug_release(sk) do { } while (0)
1104 #endif /* SOCK_REFCNT_DEBUG */
1106 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1107 extern struct static_key memcg_socket_limit_enabled;
1108 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1109 struct cg_proto *cg_proto)
1111 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1113 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1115 #define mem_cgroup_sockets_enabled 0
1116 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1117 struct cg_proto *cg_proto)
1123 static inline bool sk_stream_memory_free(const struct sock *sk)
1125 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1128 return sk->sk_prot->stream_memory_free ?
1129 sk->sk_prot->stream_memory_free(sk) : true;
1132 static inline bool sk_stream_is_writeable(const struct sock *sk)
1134 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1135 sk_stream_memory_free(sk);
1139 static inline bool sk_has_memory_pressure(const struct sock *sk)
1141 return sk->sk_prot->memory_pressure != NULL;
1144 static inline bool sk_under_memory_pressure(const struct sock *sk)
1146 if (!sk->sk_prot->memory_pressure)
1149 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1150 return !!sk->sk_cgrp->memory_pressure;
1152 return !!*sk->sk_prot->memory_pressure;
1155 static inline void sk_leave_memory_pressure(struct sock *sk)
1157 int *memory_pressure = sk->sk_prot->memory_pressure;
1159 if (!memory_pressure)
1162 if (*memory_pressure)
1163 *memory_pressure = 0;
1165 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1166 struct cg_proto *cg_proto = sk->sk_cgrp;
1167 struct proto *prot = sk->sk_prot;
1169 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1170 cg_proto->memory_pressure = 0;
1175 static inline void sk_enter_memory_pressure(struct sock *sk)
1177 if (!sk->sk_prot->enter_memory_pressure)
1180 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1181 struct cg_proto *cg_proto = sk->sk_cgrp;
1182 struct proto *prot = sk->sk_prot;
1184 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1185 cg_proto->memory_pressure = 1;
1188 sk->sk_prot->enter_memory_pressure(sk);
1191 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1193 long *prot = sk->sk_prot->sysctl_mem;
1194 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1195 prot = sk->sk_cgrp->sysctl_mem;
1199 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1203 page_counter_charge(&prot->memory_allocated, amt);
1205 if (page_counter_read(&prot->memory_allocated) >
1206 prot->memory_allocated.limit)
1207 *parent_status = OVER_LIMIT;
1210 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1213 page_counter_uncharge(&prot->memory_allocated, amt);
1217 sk_memory_allocated(const struct sock *sk)
1219 struct proto *prot = sk->sk_prot;
1221 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1222 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1224 return atomic_long_read(prot->memory_allocated);
1228 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1230 struct proto *prot = sk->sk_prot;
1232 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1233 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1234 /* update the root cgroup regardless */
1235 atomic_long_add_return(amt, prot->memory_allocated);
1236 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1239 return atomic_long_add_return(amt, prot->memory_allocated);
1243 sk_memory_allocated_sub(struct sock *sk, int amt)
1245 struct proto *prot = sk->sk_prot;
1247 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1248 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1250 atomic_long_sub(amt, prot->memory_allocated);
1253 static inline void sk_sockets_allocated_dec(struct sock *sk)
1255 struct proto *prot = sk->sk_prot;
1257 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1258 struct cg_proto *cg_proto = sk->sk_cgrp;
1260 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1261 percpu_counter_dec(&cg_proto->sockets_allocated);
1264 percpu_counter_dec(prot->sockets_allocated);
1267 static inline void sk_sockets_allocated_inc(struct sock *sk)
1269 struct proto *prot = sk->sk_prot;
1271 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1272 struct cg_proto *cg_proto = sk->sk_cgrp;
1274 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1275 percpu_counter_inc(&cg_proto->sockets_allocated);
1278 percpu_counter_inc(prot->sockets_allocated);
1282 sk_sockets_allocated_read_positive(struct sock *sk)
1284 struct proto *prot = sk->sk_prot;
1286 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1287 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1289 return percpu_counter_read_positive(prot->sockets_allocated);
1293 proto_sockets_allocated_sum_positive(struct proto *prot)
1295 return percpu_counter_sum_positive(prot->sockets_allocated);
1299 proto_memory_allocated(struct proto *prot)
1301 return atomic_long_read(prot->memory_allocated);
1305 proto_memory_pressure(struct proto *prot)
1307 if (!prot->memory_pressure)
1309 return !!*prot->memory_pressure;
1313 #ifdef CONFIG_PROC_FS
1314 /* Called with local bh disabled */
1315 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1316 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1318 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1325 /* With per-bucket locks this operation is not-atomic, so that
1326 * this version is not worse.
1328 static inline void __sk_prot_rehash(struct sock *sk)
1330 sk->sk_prot->unhash(sk);
1331 sk->sk_prot->hash(sk);
1334 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1336 /* About 10 seconds */
1337 #define SOCK_DESTROY_TIME (10*HZ)
1339 /* Sockets 0-1023 can't be bound to unless you are superuser */
1340 #define PROT_SOCK 1024
1342 #define SHUTDOWN_MASK 3
1343 #define RCV_SHUTDOWN 1
1344 #define SEND_SHUTDOWN 2
1346 #define SOCK_SNDBUF_LOCK 1
1347 #define SOCK_RCVBUF_LOCK 2
1348 #define SOCK_BINDADDR_LOCK 4
1349 #define SOCK_BINDPORT_LOCK 8
1351 struct socket_alloc {
1352 struct socket socket;
1353 struct inode vfs_inode;
1356 static inline struct socket *SOCKET_I(struct inode *inode)
1358 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1361 static inline struct inode *SOCK_INODE(struct socket *socket)
1363 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1367 * Functions for memory accounting
1369 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1370 void __sk_mem_reclaim(struct sock *sk);
1372 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1373 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1374 #define SK_MEM_SEND 0
1375 #define SK_MEM_RECV 1
1377 static inline int sk_mem_pages(int amt)
1379 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1382 static inline bool sk_has_account(struct sock *sk)
1384 /* return true if protocol supports memory accounting */
1385 return !!sk->sk_prot->memory_allocated;
1388 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1390 if (!sk_has_account(sk))
1392 return size <= sk->sk_forward_alloc ||
1393 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1397 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1399 if (!sk_has_account(sk))
1401 return size<= sk->sk_forward_alloc ||
1402 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1403 skb_pfmemalloc(skb);
1406 static inline void sk_mem_reclaim(struct sock *sk)
1408 if (!sk_has_account(sk))
1410 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1411 __sk_mem_reclaim(sk);
1414 static inline void sk_mem_reclaim_partial(struct sock *sk)
1416 if (!sk_has_account(sk))
1418 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1419 __sk_mem_reclaim(sk);
1422 static inline void sk_mem_charge(struct sock *sk, int size)
1424 if (!sk_has_account(sk))
1426 sk->sk_forward_alloc -= size;
1429 static inline void sk_mem_uncharge(struct sock *sk, int size)
1431 if (!sk_has_account(sk))
1433 sk->sk_forward_alloc += size;
1436 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1438 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1439 sk->sk_wmem_queued -= skb->truesize;
1440 sk_mem_uncharge(sk, skb->truesize);
1444 /* Used by processes to "lock" a socket state, so that
1445 * interrupts and bottom half handlers won't change it
1446 * from under us. It essentially blocks any incoming
1447 * packets, so that we won't get any new data or any
1448 * packets that change the state of the socket.
1450 * While locked, BH processing will add new packets to
1451 * the backlog queue. This queue is processed by the
1452 * owner of the socket lock right before it is released.
1454 * Since ~2.3.5 it is also exclusive sleep lock serializing
1455 * accesses from user process context.
1457 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1459 static inline void sock_release_ownership(struct sock *sk)
1461 sk->sk_lock.owned = 0;
1465 * Macro so as to not evaluate some arguments when
1466 * lockdep is not enabled.
1468 * Mark both the sk_lock and the sk_lock.slock as a
1469 * per-address-family lock class.
1471 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1473 sk->sk_lock.owned = 0; \
1474 init_waitqueue_head(&sk->sk_lock.wq); \
1475 spin_lock_init(&(sk)->sk_lock.slock); \
1476 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1477 sizeof((sk)->sk_lock)); \
1478 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1480 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1483 void lock_sock_nested(struct sock *sk, int subclass);
1485 static inline void lock_sock(struct sock *sk)
1487 lock_sock_nested(sk, 0);
1490 void release_sock(struct sock *sk);
1492 /* BH context may only use the following locking interface. */
1493 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1494 #define bh_lock_sock_nested(__sk) \
1495 spin_lock_nested(&((__sk)->sk_lock.slock), \
1496 SINGLE_DEPTH_NESTING)
1497 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1499 bool lock_sock_fast(struct sock *sk);
1501 * unlock_sock_fast - complement of lock_sock_fast
1505 * fast unlock socket for user context.
1506 * If slow mode is on, we call regular release_sock()
1508 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1513 spin_unlock_bh(&sk->sk_lock.slock);
1517 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1518 struct proto *prot);
1519 void sk_free(struct sock *sk);
1520 void sk_release_kernel(struct sock *sk);
1521 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1523 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1525 void sock_wfree(struct sk_buff *skb);
1526 void skb_orphan_partial(struct sk_buff *skb);
1527 void sock_rfree(struct sk_buff *skb);
1528 void sock_efree(struct sk_buff *skb);
1530 void sock_edemux(struct sk_buff *skb);
1532 #define sock_edemux(skb) sock_efree(skb)
1535 int sock_setsockopt(struct socket *sock, int level, int op,
1536 char __user *optval, unsigned int optlen);
1538 int sock_getsockopt(struct socket *sock, int level, int op,
1539 char __user *optval, int __user *optlen);
1540 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1541 int noblock, int *errcode);
1542 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1543 unsigned long data_len, int noblock,
1544 int *errcode, int max_page_order);
1545 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1546 void sock_kfree_s(struct sock *sk, void *mem, int size);
1547 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1548 void sk_send_sigurg(struct sock *sk);
1551 * Functions to fill in entries in struct proto_ops when a protocol
1552 * does not implement a particular function.
1554 int sock_no_bind(struct socket *, struct sockaddr *, int);
1555 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1556 int sock_no_socketpair(struct socket *, struct socket *);
1557 int sock_no_accept(struct socket *, struct socket *, int);
1558 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1559 unsigned int sock_no_poll(struct file *, struct socket *,
1560 struct poll_table_struct *);
1561 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1562 int sock_no_listen(struct socket *, int);
1563 int sock_no_shutdown(struct socket *, int);
1564 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1565 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1566 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1567 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1568 int sock_no_mmap(struct file *file, struct socket *sock,
1569 struct vm_area_struct *vma);
1570 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1571 size_t size, int flags);
1574 * Functions to fill in entries in struct proto_ops when a protocol
1575 * uses the inet style.
1577 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1578 char __user *optval, int __user *optlen);
1579 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1581 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1582 char __user *optval, unsigned int optlen);
1583 int compat_sock_common_getsockopt(struct socket *sock, int level,
1584 int optname, char __user *optval, int __user *optlen);
1585 int compat_sock_common_setsockopt(struct socket *sock, int level,
1586 int optname, char __user *optval, unsigned int optlen);
1588 void sk_common_release(struct sock *sk);
1591 * Default socket callbacks and setup code
1594 /* Initialise core socket variables */
1595 void sock_init_data(struct socket *sock, struct sock *sk);
1598 * Socket reference counting postulates.
1600 * * Each user of socket SHOULD hold a reference count.
1601 * * Each access point to socket (an hash table bucket, reference from a list,
1602 * running timer, skb in flight MUST hold a reference count.
1603 * * When reference count hits 0, it means it will never increase back.
1604 * * When reference count hits 0, it means that no references from
1605 * outside exist to this socket and current process on current CPU
1606 * is last user and may/should destroy this socket.
1607 * * sk_free is called from any context: process, BH, IRQ. When
1608 * it is called, socket has no references from outside -> sk_free
1609 * may release descendant resources allocated by the socket, but
1610 * to the time when it is called, socket is NOT referenced by any
1611 * hash tables, lists etc.
1612 * * Packets, delivered from outside (from network or from another process)
1613 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1614 * when they sit in queue. Otherwise, packets will leak to hole, when
1615 * socket is looked up by one cpu and unhasing is made by another CPU.
1616 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1617 * (leak to backlog). Packet socket does all the processing inside
1618 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1619 * use separate SMP lock, so that they are prone too.
1622 /* Ungrab socket and destroy it, if it was the last reference. */
1623 static inline void sock_put(struct sock *sk)
1625 if (atomic_dec_and_test(&sk->sk_refcnt))
1628 /* Generic version of sock_put(), dealing with all sockets
1629 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1631 void sock_gen_put(struct sock *sk);
1633 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1635 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1637 sk->sk_tx_queue_mapping = tx_queue;
1640 static inline void sk_tx_queue_clear(struct sock *sk)
1642 sk->sk_tx_queue_mapping = -1;
1645 static inline int sk_tx_queue_get(const struct sock *sk)
1647 return sk ? sk->sk_tx_queue_mapping : -1;
1650 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1652 sk_tx_queue_clear(sk);
1653 sk->sk_socket = sock;
1656 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1658 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1659 return &rcu_dereference_raw(sk->sk_wq)->wait;
1661 /* Detach socket from process context.
1662 * Announce socket dead, detach it from wait queue and inode.
1663 * Note that parent inode held reference count on this struct sock,
1664 * we do not release it in this function, because protocol
1665 * probably wants some additional cleanups or even continuing
1666 * to work with this socket (TCP).
1668 static inline void sock_orphan(struct sock *sk)
1670 write_lock_bh(&sk->sk_callback_lock);
1671 sock_set_flag(sk, SOCK_DEAD);
1672 sk_set_socket(sk, NULL);
1674 write_unlock_bh(&sk->sk_callback_lock);
1677 static inline void sock_graft(struct sock *sk, struct socket *parent)
1679 write_lock_bh(&sk->sk_callback_lock);
1680 sk->sk_wq = parent->wq;
1682 sk_set_socket(sk, parent);
1683 security_sock_graft(sk, parent);
1684 write_unlock_bh(&sk->sk_callback_lock);
1687 kuid_t sock_i_uid(struct sock *sk);
1688 unsigned long sock_i_ino(struct sock *sk);
1690 static inline struct dst_entry *
1691 __sk_dst_get(struct sock *sk)
1693 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1694 lockdep_is_held(&sk->sk_lock.slock));
1697 static inline struct dst_entry *
1698 sk_dst_get(struct sock *sk)
1700 struct dst_entry *dst;
1703 dst = rcu_dereference(sk->sk_dst_cache);
1704 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1710 static inline void dst_negative_advice(struct sock *sk)
1712 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1714 if (dst && dst->ops->negative_advice) {
1715 ndst = dst->ops->negative_advice(dst);
1718 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1719 sk_tx_queue_clear(sk);
1725 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1727 struct dst_entry *old_dst;
1729 sk_tx_queue_clear(sk);
1731 * This can be called while sk is owned by the caller only,
1732 * with no state that can be checked in a rcu_dereference_check() cond
1734 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1735 rcu_assign_pointer(sk->sk_dst_cache, dst);
1736 dst_release(old_dst);
1740 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1742 struct dst_entry *old_dst;
1744 sk_tx_queue_clear(sk);
1745 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1746 dst_release(old_dst);
1750 __sk_dst_reset(struct sock *sk)
1752 __sk_dst_set(sk, NULL);
1756 sk_dst_reset(struct sock *sk)
1758 sk_dst_set(sk, NULL);
1761 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1763 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1765 static inline bool sk_can_gso(const struct sock *sk)
1767 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1770 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1772 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1774 sk->sk_route_nocaps |= flags;
1775 sk->sk_route_caps &= ~flags;
1778 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1779 struct iov_iter *from, char *to,
1780 int copy, int offset)
1782 if (skb->ip_summed == CHECKSUM_NONE) {
1784 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1786 skb->csum = csum_block_add(skb->csum, csum, offset);
1787 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1788 if (copy_from_iter_nocache(to, copy, from) != copy)
1790 } else if (copy_from_iter(to, copy, from) != copy)
1796 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1797 struct iov_iter *from, int copy)
1799 int err, offset = skb->len;
1801 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1804 __skb_trim(skb, offset);
1809 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1810 struct sk_buff *skb,
1816 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1822 skb->data_len += copy;
1823 skb->truesize += copy;
1824 sk->sk_wmem_queued += copy;
1825 sk_mem_charge(sk, copy);
1830 * sk_wmem_alloc_get - returns write allocations
1833 * Returns sk_wmem_alloc minus initial offset of one
1835 static inline int sk_wmem_alloc_get(const struct sock *sk)
1837 return atomic_read(&sk->sk_wmem_alloc) - 1;
1841 * sk_rmem_alloc_get - returns read allocations
1844 * Returns sk_rmem_alloc
1846 static inline int sk_rmem_alloc_get(const struct sock *sk)
1848 return atomic_read(&sk->sk_rmem_alloc);
1852 * sk_has_allocations - check if allocations are outstanding
1855 * Returns true if socket has write or read allocations
1857 static inline bool sk_has_allocations(const struct sock *sk)
1859 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1863 * wq_has_sleeper - check if there are any waiting processes
1864 * @wq: struct socket_wq
1866 * Returns true if socket_wq has waiting processes
1868 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1869 * barrier call. They were added due to the race found within the tcp code.
1871 * Consider following tcp code paths:
1875 * sys_select receive packet
1877 * __add_wait_queue update tp->rcv_nxt
1879 * tp->rcv_nxt check sock_def_readable
1881 * schedule rcu_read_lock();
1882 * wq = rcu_dereference(sk->sk_wq);
1883 * if (wq && waitqueue_active(&wq->wait))
1884 * wake_up_interruptible(&wq->wait)
1888 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1889 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1890 * could then endup calling schedule and sleep forever if there are no more
1891 * data on the socket.
1894 static inline bool wq_has_sleeper(struct socket_wq *wq)
1896 /* We need to be sure we are in sync with the
1897 * add_wait_queue modifications to the wait queue.
1899 * This memory barrier is paired in the sock_poll_wait.
1902 return wq && waitqueue_active(&wq->wait);
1906 * sock_poll_wait - place memory barrier behind the poll_wait call.
1908 * @wait_address: socket wait queue
1911 * See the comments in the wq_has_sleeper function.
1913 static inline void sock_poll_wait(struct file *filp,
1914 wait_queue_head_t *wait_address, poll_table *p)
1916 if (!poll_does_not_wait(p) && wait_address) {
1917 poll_wait(filp, wait_address, p);
1918 /* We need to be sure we are in sync with the
1919 * socket flags modification.
1921 * This memory barrier is paired in the wq_has_sleeper.
1927 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1929 if (sk->sk_txhash) {
1931 skb->hash = sk->sk_txhash;
1936 * Queue a received datagram if it will fit. Stream and sequenced
1937 * protocols can't normally use this as they need to fit buffers in
1938 * and play with them.
1940 * Inlined as it's very short and called for pretty much every
1941 * packet ever received.
1944 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1948 skb->destructor = sock_wfree;
1949 skb_set_hash_from_sk(skb, sk);
1951 * We used to take a refcount on sk, but following operation
1952 * is enough to guarantee sk_free() wont free this sock until
1953 * all in-flight packets are completed
1955 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1958 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1962 skb->destructor = sock_rfree;
1963 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1964 sk_mem_charge(sk, skb->truesize);
1967 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1968 unsigned long expires);
1970 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1972 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1974 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1975 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1978 * Recover an error report and clear atomically
1981 static inline int sock_error(struct sock *sk)
1984 if (likely(!sk->sk_err))
1986 err = xchg(&sk->sk_err, 0);
1990 static inline unsigned long sock_wspace(struct sock *sk)
1994 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1995 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2002 static inline void sk_wake_async(struct sock *sk, int how, int band)
2004 if (sock_flag(sk, SOCK_FASYNC))
2005 sock_wake_async(sk->sk_socket, how, band);
2008 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2009 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2010 * Note: for send buffers, TCP works better if we can build two skbs at
2013 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2015 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2016 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2018 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2020 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2021 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2022 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2026 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2029 * sk_page_frag - return an appropriate page_frag
2032 * If socket allocation mode allows current thread to sleep, it means its
2033 * safe to use the per task page_frag instead of the per socket one.
2035 static inline struct page_frag *sk_page_frag(struct sock *sk)
2037 if (sk->sk_allocation & __GFP_WAIT)
2038 return ¤t->task_frag;
2040 return &sk->sk_frag;
2043 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2046 * Default write policy as shown to user space via poll/select/SIGIO
2048 static inline bool sock_writeable(const struct sock *sk)
2050 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2053 static inline gfp_t gfp_any(void)
2055 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2058 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2060 return noblock ? 0 : sk->sk_rcvtimeo;
2063 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2065 return noblock ? 0 : sk->sk_sndtimeo;
2068 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2070 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2073 /* Alas, with timeout socket operations are not restartable.
2074 * Compare this to poll().
2076 static inline int sock_intr_errno(long timeo)
2078 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2081 struct sock_skb_cb {
2085 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2086 * using skb->cb[] would keep using it directly and utilize its
2087 * alignement guarantee.
2089 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2090 sizeof(struct sock_skb_cb)))
2092 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2093 SOCK_SKB_CB_OFFSET))
2095 #define sock_skb_cb_check_size(size) \
2096 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2099 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2101 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2104 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2105 struct sk_buff *skb);
2106 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2107 struct sk_buff *skb);
2110 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2112 ktime_t kt = skb->tstamp;
2113 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2116 * generate control messages if
2117 * - receive time stamping in software requested
2118 * - software time stamp available and wanted
2119 * - hardware time stamps available and wanted
2121 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2122 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2123 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2124 (hwtstamps->hwtstamp.tv64 &&
2125 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2126 __sock_recv_timestamp(msg, sk, skb);
2130 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2131 __sock_recv_wifi_status(msg, sk, skb);
2134 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2135 struct sk_buff *skb);
2137 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2138 struct sk_buff *skb)
2140 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2141 (1UL << SOCK_RCVTSTAMP))
2142 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2143 SOF_TIMESTAMPING_RAW_HARDWARE)
2145 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2146 __sock_recv_ts_and_drops(msg, sk, skb);
2148 sk->sk_stamp = skb->tstamp;
2151 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2154 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2155 * @sk: socket sending this packet
2156 * @tx_flags: completed with instructions for time stamping
2158 * Note : callers should take care of initial *tx_flags value (usually 0)
2160 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2162 if (unlikely(sk->sk_tsflags))
2163 __sock_tx_timestamp(sk, tx_flags);
2164 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2165 *tx_flags |= SKBTX_WIFI_STATUS;
2169 * sk_eat_skb - Release a skb if it is no longer needed
2170 * @sk: socket to eat this skb from
2171 * @skb: socket buffer to eat
2173 * This routine must be called with interrupts disabled or with the socket
2174 * locked so that the sk_buff queue operation is ok.
2176 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2178 __skb_unlink(skb, &sk->sk_receive_queue);
2183 struct net *sock_net(const struct sock *sk)
2185 return read_pnet(&sk->sk_net);
2189 void sock_net_set(struct sock *sk, struct net *net)
2191 write_pnet(&sk->sk_net, net);
2195 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2196 * They should not hold a reference to a namespace in order to allow
2198 * Sockets after sk_change_net should be released using sk_release_kernel
2200 static inline void sk_change_net(struct sock *sk, struct net *net)
2202 struct net *current_net = sock_net(sk);
2204 if (!net_eq(current_net, net)) {
2205 put_net(current_net);
2206 sock_net_set(sk, net);
2210 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2213 struct sock *sk = skb->sk;
2215 skb->destructor = NULL;
2222 /* This helper checks if a socket is a full socket,
2223 * ie _not_ a timewait or request socket.
2225 static inline bool sk_fullsock(const struct sock *sk)
2227 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2230 void sock_enable_timestamp(struct sock *sk, int flag);
2231 int sock_get_timestamp(struct sock *, struct timeval __user *);
2232 int sock_get_timestampns(struct sock *, struct timespec __user *);
2233 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2236 bool sk_ns_capable(const struct sock *sk,
2237 struct user_namespace *user_ns, int cap);
2238 bool sk_capable(const struct sock *sk, int cap);
2239 bool sk_net_capable(const struct sock *sk, int cap);
2241 extern __u32 sysctl_wmem_max;
2242 extern __u32 sysctl_rmem_max;
2244 extern int sysctl_tstamp_allow_data;
2245 extern int sysctl_optmem_max;
2247 extern __u32 sysctl_wmem_default;
2248 extern __u32 sysctl_rmem_default;
2250 #endif /* _SOCK_H */