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>
62 #include <linux/filter.h>
63 #include <linux/rculist_nulls.h>
64 #include <linux/poll.h>
66 #include <linux/atomic.h>
68 #include <net/checksum.h>
69 #include <linux/net_tstamp.h>
74 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
75 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
78 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
83 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
88 * This structure really needs to be cleaned up.
89 * Most of it is for TCP, and not used by any of
90 * the other protocols.
93 /* Define this to get the SOCK_DBG debugging facility. */
94 #define SOCK_DEBUGGING
96 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
97 printk(KERN_DEBUG msg); } while (0)
99 /* Validate arguments and do nothing */
100 static inline __printf(2, 3)
101 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
106 /* This is the per-socket lock. The spinlock provides a synchronization
107 * between user contexts and software interrupt processing, whereas the
108 * mini-semaphore synchronizes multiple users amongst themselves.
113 wait_queue_head_t wq;
115 * We express the mutex-alike socket_lock semantics
116 * to the lock validator by explicitly managing
117 * the slock as a lock variant (in addition to
120 #ifdef CONFIG_DEBUG_LOCK_ALLOC
121 struct lockdep_map dep_map;
129 typedef __u32 __bitwise __portpair;
130 typedef __u64 __bitwise __addrpair;
133 * struct sock_common - minimal network layer representation of sockets
134 * @skc_daddr: Foreign IPv4 addr
135 * @skc_rcv_saddr: Bound local IPv4 addr
136 * @skc_hash: hash value used with various protocol lookup tables
137 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
138 * @skc_dport: placeholder for inet_dport/tw_dport
139 * @skc_num: placeholder for inet_num/tw_num
140 * @skc_family: network address family
141 * @skc_state: Connection state
142 * @skc_reuse: %SO_REUSEADDR setting
143 * @skc_reuseport: %SO_REUSEPORT setting
144 * @skc_bound_dev_if: bound device index if != 0
145 * @skc_bind_node: bind hash linkage for various protocol lookup tables
146 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
147 * @skc_prot: protocol handlers inside a network family
148 * @skc_net: reference to the network namespace of this socket
149 * @skc_node: main hash linkage for various protocol lookup tables
150 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
151 * @skc_tx_queue_mapping: tx queue number for this connection
152 * @skc_refcnt: reference count
154 * This is the minimal network layer representation of sockets, the header
155 * for struct sock and struct inet_timewait_sock.
158 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
159 * address on 64bit arches : cf INET_MATCH()
162 __addrpair skc_addrpair;
165 __be32 skc_rcv_saddr;
169 unsigned int skc_hash;
170 __u16 skc_u16hashes[2];
172 /* skc_dport && skc_num must be grouped as well */
174 __portpair skc_portpair;
181 unsigned short skc_family;
182 volatile unsigned char skc_state;
183 unsigned char skc_reuse:4;
184 unsigned char skc_reuseport:1;
185 unsigned char skc_ipv6only:1;
186 int skc_bound_dev_if;
188 struct hlist_node skc_bind_node;
189 struct hlist_nulls_node skc_portaddr_node;
191 struct proto *skc_prot;
196 #if IS_ENABLED(CONFIG_IPV6)
197 struct in6_addr skc_v6_daddr;
198 struct in6_addr skc_v6_rcv_saddr;
202 * fields between dontcopy_begin/dontcopy_end
203 * are not copied in sock_copy()
206 int skc_dontcopy_begin[0];
209 struct hlist_node skc_node;
210 struct hlist_nulls_node skc_nulls_node;
212 int skc_tx_queue_mapping;
215 int skc_dontcopy_end[0];
221 * struct sock - network layer representation of sockets
222 * @__sk_common: shared layout with inet_timewait_sock
223 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
224 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
225 * @sk_lock: synchronizer
226 * @sk_rcvbuf: size of receive buffer in bytes
227 * @sk_wq: sock wait queue and async head
228 * @sk_rx_dst: receive input route used by early demux
229 * @sk_dst_cache: destination cache
230 * @sk_dst_lock: destination cache lock
231 * @sk_policy: flow policy
232 * @sk_receive_queue: incoming packets
233 * @sk_wmem_alloc: transmit queue bytes committed
234 * @sk_write_queue: Packet sending queue
235 * @sk_omem_alloc: "o" is "option" or "other"
236 * @sk_wmem_queued: persistent queue size
237 * @sk_forward_alloc: space allocated forward
238 * @sk_napi_id: id of the last napi context to receive data for sk
239 * @sk_ll_usec: usecs to busypoll when there is no data
240 * @sk_allocation: allocation mode
241 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
242 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
243 * @sk_sndbuf: size of send buffer in bytes
244 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
245 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
246 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
247 * @sk_no_check_rx: allow zero checksum in RX packets
248 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
249 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
250 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
251 * @sk_gso_max_size: Maximum GSO segment size to build
252 * @sk_gso_max_segs: Maximum number of GSO segments
253 * @sk_lingertime: %SO_LINGER l_linger setting
254 * @sk_backlog: always used with the per-socket spinlock held
255 * @sk_callback_lock: used with the callbacks in the end of this struct
256 * @sk_error_queue: rarely used
257 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
258 * IPV6_ADDRFORM for instance)
259 * @sk_err: last error
260 * @sk_err_soft: errors that don't cause failure but are the cause of a
261 * persistent failure not just 'timed out'
262 * @sk_drops: raw/udp drops counter
263 * @sk_ack_backlog: current listen backlog
264 * @sk_max_ack_backlog: listen backlog set in listen()
265 * @sk_priority: %SO_PRIORITY setting
266 * @sk_cgrp_prioidx: socket group's priority map index
267 * @sk_type: socket type (%SOCK_STREAM, etc)
268 * @sk_protocol: which protocol this socket belongs in this network family
269 * @sk_peer_pid: &struct pid for this socket's peer
270 * @sk_peer_cred: %SO_PEERCRED setting
271 * @sk_rcvlowat: %SO_RCVLOWAT setting
272 * @sk_rcvtimeo: %SO_RCVTIMEO setting
273 * @sk_sndtimeo: %SO_SNDTIMEO setting
274 * @sk_rxhash: flow hash received from netif layer
275 * @sk_incoming_cpu: record cpu processing incoming packets
276 * @sk_txhash: computed flow hash for use on transmit
277 * @sk_filter: socket filtering instructions
278 * @sk_protinfo: private area, net family specific, when not using slab
279 * @sk_timer: sock cleanup timer
280 * @sk_stamp: time stamp of last packet received
281 * @sk_tsflags: SO_TIMESTAMPING socket options
282 * @sk_tskey: counter to disambiguate concurrent tstamp requests
283 * @sk_socket: Identd and reporting IO signals
284 * @sk_user_data: RPC layer private data
285 * @sk_frag: cached page frag
286 * @sk_peek_off: current peek_offset value
287 * @sk_send_head: front of stuff to transmit
288 * @sk_security: used by security modules
289 * @sk_mark: generic packet mark
290 * @sk_classid: this socket's cgroup classid
291 * @sk_cgrp: this socket's cgroup-specific proto data
292 * @sk_write_pending: a write to stream socket waits to start
293 * @sk_state_change: callback to indicate change in the state of the sock
294 * @sk_data_ready: callback to indicate there is data to be processed
295 * @sk_write_space: callback to indicate there is bf sending space available
296 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
297 * @sk_backlog_rcv: callback to process the backlog
298 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
302 * Now struct inet_timewait_sock also uses sock_common, so please just
303 * don't add nothing before this first member (__sk_common) --acme
305 struct sock_common __sk_common;
306 #define sk_node __sk_common.skc_node
307 #define sk_nulls_node __sk_common.skc_nulls_node
308 #define sk_refcnt __sk_common.skc_refcnt
309 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
311 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
312 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
313 #define sk_hash __sk_common.skc_hash
314 #define sk_portpair __sk_common.skc_portpair
315 #define sk_num __sk_common.skc_num
316 #define sk_dport __sk_common.skc_dport
317 #define sk_addrpair __sk_common.skc_addrpair
318 #define sk_daddr __sk_common.skc_daddr
319 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
320 #define sk_family __sk_common.skc_family
321 #define sk_state __sk_common.skc_state
322 #define sk_reuse __sk_common.skc_reuse
323 #define sk_reuseport __sk_common.skc_reuseport
324 #define sk_ipv6only __sk_common.skc_ipv6only
325 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
326 #define sk_bind_node __sk_common.skc_bind_node
327 #define sk_prot __sk_common.skc_prot
328 #define sk_net __sk_common.skc_net
329 #define sk_v6_daddr __sk_common.skc_v6_daddr
330 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
332 socket_lock_t sk_lock;
333 struct sk_buff_head sk_receive_queue;
335 * The backlog queue is special, it is always used with
336 * the per-socket spinlock held and requires low latency
337 * access. Therefore we special case it's implementation.
338 * Note : rmem_alloc is in this structure to fill a hole
339 * on 64bit arches, not because its logically part of
345 struct sk_buff *head;
346 struct sk_buff *tail;
348 #define sk_rmem_alloc sk_backlog.rmem_alloc
349 int sk_forward_alloc;
355 * Warned : sk_incoming_cpu can be set from softirq,
356 * Do not use this hole without fully understanding possible issues.
360 #ifdef CONFIG_NET_RX_BUSY_POLL
361 unsigned int sk_napi_id;
362 unsigned int sk_ll_usec;
367 struct sk_filter __rcu *sk_filter;
368 struct socket_wq __rcu *sk_wq;
371 struct xfrm_policy *sk_policy[2];
373 unsigned long sk_flags;
374 struct dst_entry *sk_rx_dst;
375 struct dst_entry __rcu *sk_dst_cache;
376 spinlock_t sk_dst_lock;
377 atomic_t sk_wmem_alloc;
378 atomic_t sk_omem_alloc;
380 struct sk_buff_head sk_write_queue;
381 kmemcheck_bitfield_begin(flags);
382 unsigned int sk_shutdown : 2,
388 kmemcheck_bitfield_end(flags);
391 u32 sk_pacing_rate; /* bytes per second */
392 u32 sk_max_pacing_rate;
393 netdev_features_t sk_route_caps;
394 netdev_features_t sk_route_nocaps;
396 unsigned int sk_gso_max_size;
399 unsigned long sk_lingertime;
400 struct sk_buff_head sk_error_queue;
401 struct proto *sk_prot_creator;
402 rwlock_t sk_callback_lock;
405 unsigned short sk_ack_backlog;
406 unsigned short sk_max_ack_backlog;
408 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
409 __u32 sk_cgrp_prioidx;
411 struct pid *sk_peer_pid;
412 const struct cred *sk_peer_cred;
416 struct timer_list sk_timer;
420 struct socket *sk_socket;
422 struct page_frag sk_frag;
423 struct sk_buff *sk_send_head;
425 int sk_write_pending;
426 #ifdef CONFIG_SECURITY
431 struct cg_proto *sk_cgrp;
432 void (*sk_state_change)(struct sock *sk);
433 void (*sk_data_ready)(struct sock *sk);
434 void (*sk_write_space)(struct sock *sk);
435 void (*sk_error_report)(struct sock *sk);
436 int (*sk_backlog_rcv)(struct sock *sk,
437 struct sk_buff *skb);
438 void (*sk_destruct)(struct sock *sk);
441 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
443 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
444 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
447 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
448 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
449 * on a socket means that the socket will reuse everybody else's port
450 * without looking at the other's sk_reuse value.
453 #define SK_NO_REUSE 0
454 #define SK_CAN_REUSE 1
455 #define SK_FORCE_REUSE 2
457 static inline int sk_peek_offset(struct sock *sk, int flags)
459 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
460 return sk->sk_peek_off;
465 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
467 if (sk->sk_peek_off >= 0) {
468 if (sk->sk_peek_off >= val)
469 sk->sk_peek_off -= val;
475 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
477 if (sk->sk_peek_off >= 0)
478 sk->sk_peek_off += val;
482 * Hashed lists helper routines
484 static inline struct sock *sk_entry(const struct hlist_node *node)
486 return hlist_entry(node, struct sock, sk_node);
489 static inline struct sock *__sk_head(const struct hlist_head *head)
491 return hlist_entry(head->first, struct sock, sk_node);
494 static inline struct sock *sk_head(const struct hlist_head *head)
496 return hlist_empty(head) ? NULL : __sk_head(head);
499 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
501 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
504 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
506 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
509 static inline struct sock *sk_next(const struct sock *sk)
511 return sk->sk_node.next ?
512 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
515 static inline struct sock *sk_nulls_next(const struct sock *sk)
517 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
518 hlist_nulls_entry(sk->sk_nulls_node.next,
519 struct sock, sk_nulls_node) :
523 static inline bool sk_unhashed(const struct sock *sk)
525 return hlist_unhashed(&sk->sk_node);
528 static inline bool sk_hashed(const struct sock *sk)
530 return !sk_unhashed(sk);
533 static inline void sk_node_init(struct hlist_node *node)
538 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
543 static inline void __sk_del_node(struct sock *sk)
545 __hlist_del(&sk->sk_node);
548 /* NB: equivalent to hlist_del_init_rcu */
549 static inline bool __sk_del_node_init(struct sock *sk)
553 sk_node_init(&sk->sk_node);
559 /* Grab socket reference count. This operation is valid only
560 when sk is ALREADY grabbed f.e. it is found in hash table
561 or a list and the lookup is made under lock preventing hash table
565 static inline void sock_hold(struct sock *sk)
567 atomic_inc(&sk->sk_refcnt);
570 /* Ungrab socket in the context, which assumes that socket refcnt
571 cannot hit zero, f.e. it is true in context of any socketcall.
573 static inline void __sock_put(struct sock *sk)
575 atomic_dec(&sk->sk_refcnt);
578 static inline bool sk_del_node_init(struct sock *sk)
580 bool rc = __sk_del_node_init(sk);
583 /* paranoid for a while -acme */
584 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
589 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
591 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
594 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
600 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
602 bool rc = __sk_nulls_del_node_init_rcu(sk);
605 /* paranoid for a while -acme */
606 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
612 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
614 hlist_add_head(&sk->sk_node, list);
617 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
620 __sk_add_node(sk, list);
623 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
626 hlist_add_head_rcu(&sk->sk_node, list);
629 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
631 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
634 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
637 __sk_nulls_add_node_rcu(sk, list);
640 static inline void __sk_del_bind_node(struct sock *sk)
642 __hlist_del(&sk->sk_bind_node);
645 static inline void sk_add_bind_node(struct sock *sk,
646 struct hlist_head *list)
648 hlist_add_head(&sk->sk_bind_node, list);
651 #define sk_for_each(__sk, list) \
652 hlist_for_each_entry(__sk, list, sk_node)
653 #define sk_for_each_rcu(__sk, list) \
654 hlist_for_each_entry_rcu(__sk, list, sk_node)
655 #define sk_nulls_for_each(__sk, node, list) \
656 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
657 #define sk_nulls_for_each_rcu(__sk, node, list) \
658 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
659 #define sk_for_each_from(__sk) \
660 hlist_for_each_entry_from(__sk, sk_node)
661 #define sk_nulls_for_each_from(__sk, node) \
662 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
663 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
664 #define sk_for_each_safe(__sk, tmp, list) \
665 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
666 #define sk_for_each_bound(__sk, list) \
667 hlist_for_each_entry(__sk, list, sk_bind_node)
670 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
671 * @tpos: the type * to use as a loop cursor.
672 * @pos: the &struct hlist_node to use as a loop cursor.
673 * @head: the head for your list.
674 * @offset: offset of hlist_node within the struct.
677 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
678 for (pos = (head)->first; \
679 (!is_a_nulls(pos)) && \
680 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
683 static inline struct user_namespace *sk_user_ns(struct sock *sk)
685 /* Careful only use this in a context where these parameters
686 * can not change and must all be valid, such as recvmsg from
689 return sk->sk_socket->file->f_cred->user_ns;
703 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
704 SOCK_DBG, /* %SO_DEBUG setting */
705 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
706 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
707 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
708 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
709 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
710 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
711 SOCK_FASYNC, /* fasync() active */
713 SOCK_ZEROCOPY, /* buffers from userspace */
714 SOCK_WIFI_STATUS, /* push wifi status to userspace */
715 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
716 * Will use last 4 bytes of packet sent from
717 * user-space instead.
719 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
720 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
723 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
725 nsk->sk_flags = osk->sk_flags;
728 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
730 __set_bit(flag, &sk->sk_flags);
733 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
735 __clear_bit(flag, &sk->sk_flags);
738 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
740 return test_bit(flag, &sk->sk_flags);
744 extern struct static_key memalloc_socks;
745 static inline int sk_memalloc_socks(void)
747 return static_key_false(&memalloc_socks);
751 static inline int sk_memalloc_socks(void)
758 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
760 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
763 static inline void sk_acceptq_removed(struct sock *sk)
765 sk->sk_ack_backlog--;
768 static inline void sk_acceptq_added(struct sock *sk)
770 sk->sk_ack_backlog++;
773 static inline bool sk_acceptq_is_full(const struct sock *sk)
775 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
779 * Compute minimal free write space needed to queue new packets.
781 static inline int sk_stream_min_wspace(const struct sock *sk)
783 return sk->sk_wmem_queued >> 1;
786 static inline int sk_stream_wspace(const struct sock *sk)
788 return sk->sk_sndbuf - sk->sk_wmem_queued;
791 void sk_stream_write_space(struct sock *sk);
793 /* OOB backlog add */
794 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
796 /* dont let skb dst not refcounted, we are going to leave rcu lock */
799 if (!sk->sk_backlog.tail)
800 sk->sk_backlog.head = skb;
802 sk->sk_backlog.tail->next = skb;
804 sk->sk_backlog.tail = skb;
809 * Take into account size of receive queue and backlog queue
810 * Do not take into account this skb truesize,
811 * to allow even a single big packet to come.
813 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
815 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
817 return qsize > limit;
820 /* The per-socket spinlock must be held here. */
821 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
824 if (sk_rcvqueues_full(sk, limit))
827 __sk_add_backlog(sk, skb);
828 sk->sk_backlog.len += skb->truesize;
832 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
834 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
836 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
837 return __sk_backlog_rcv(sk, skb);
839 return sk->sk_backlog_rcv(sk, skb);
842 static inline void sk_incoming_cpu_update(struct sock *sk)
844 sk->sk_incoming_cpu = raw_smp_processor_id();
847 static inline void sock_rps_record_flow_hash(__u32 hash)
850 struct rps_sock_flow_table *sock_flow_table;
853 sock_flow_table = rcu_dereference(rps_sock_flow_table);
854 rps_record_sock_flow(sock_flow_table, hash);
859 static inline void sock_rps_record_flow(const struct sock *sk)
862 sock_rps_record_flow_hash(sk->sk_rxhash);
866 static inline void sock_rps_save_rxhash(struct sock *sk,
867 const struct sk_buff *skb)
870 if (unlikely(sk->sk_rxhash != skb->hash))
871 sk->sk_rxhash = skb->hash;
875 static inline void sock_rps_reset_rxhash(struct sock *sk)
882 #define sk_wait_event(__sk, __timeo, __condition) \
884 release_sock(__sk); \
885 __rc = __condition; \
887 *(__timeo) = schedule_timeout(*(__timeo)); \
889 sched_annotate_sleep(); \
891 __rc = __condition; \
895 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
896 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
897 void sk_stream_wait_close(struct sock *sk, long timeo_p);
898 int sk_stream_error(struct sock *sk, int flags, int err);
899 void sk_stream_kill_queues(struct sock *sk);
900 void sk_set_memalloc(struct sock *sk);
901 void sk_clear_memalloc(struct sock *sk);
903 int sk_wait_data(struct sock *sk, long *timeo);
905 struct request_sock_ops;
906 struct timewait_sock_ops;
907 struct inet_hashinfo;
912 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
913 * un-modified. Special care is taken when initializing object to zero.
915 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
917 if (offsetof(struct sock, sk_node.next) != 0)
918 memset(sk, 0, offsetof(struct sock, sk_node.next));
919 memset(&sk->sk_node.pprev, 0,
920 size - offsetof(struct sock, sk_node.pprev));
923 /* Networking protocol blocks we attach to sockets.
924 * socket layer -> transport layer interface
925 * transport -> network interface is defined by struct inet_proto
928 void (*close)(struct sock *sk,
930 int (*connect)(struct sock *sk,
931 struct sockaddr *uaddr,
933 int (*disconnect)(struct sock *sk, int flags);
935 struct sock * (*accept)(struct sock *sk, int flags, int *err);
937 int (*ioctl)(struct sock *sk, int cmd,
939 int (*init)(struct sock *sk);
940 void (*destroy)(struct sock *sk);
941 void (*shutdown)(struct sock *sk, int how);
942 int (*setsockopt)(struct sock *sk, int level,
943 int optname, char __user *optval,
944 unsigned int optlen);
945 int (*getsockopt)(struct sock *sk, int level,
946 int optname, char __user *optval,
949 int (*compat_setsockopt)(struct sock *sk,
951 int optname, char __user *optval,
952 unsigned int optlen);
953 int (*compat_getsockopt)(struct sock *sk,
955 int optname, char __user *optval,
957 int (*compat_ioctl)(struct sock *sk,
958 unsigned int cmd, unsigned long arg);
960 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
961 struct msghdr *msg, size_t len);
962 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
964 size_t len, int noblock, int flags,
966 int (*sendpage)(struct sock *sk, struct page *page,
967 int offset, size_t size, int flags);
968 int (*bind)(struct sock *sk,
969 struct sockaddr *uaddr, int addr_len);
971 int (*backlog_rcv) (struct sock *sk,
972 struct sk_buff *skb);
974 void (*release_cb)(struct sock *sk);
976 /* Keeping track of sk's, looking them up, and port selection methods. */
977 void (*hash)(struct sock *sk);
978 void (*unhash)(struct sock *sk);
979 void (*rehash)(struct sock *sk);
980 int (*get_port)(struct sock *sk, unsigned short snum);
981 void (*clear_sk)(struct sock *sk, int size);
983 /* Keeping track of sockets in use */
984 #ifdef CONFIG_PROC_FS
985 unsigned int inuse_idx;
988 bool (*stream_memory_free)(const struct sock *sk);
989 /* Memory pressure */
990 void (*enter_memory_pressure)(struct sock *sk);
991 atomic_long_t *memory_allocated; /* Current allocated memory. */
992 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
994 * Pressure flag: try to collapse.
995 * Technical note: it is used by multiple contexts non atomically.
996 * All the __sk_mem_schedule() is of this nature: accounting
997 * is strict, actions are advisory and have some latency.
999 int *memory_pressure;
1006 struct kmem_cache *slab;
1007 unsigned int obj_size;
1010 struct percpu_counter *orphan_count;
1012 struct request_sock_ops *rsk_prot;
1013 struct timewait_sock_ops *twsk_prot;
1016 struct inet_hashinfo *hashinfo;
1017 struct udp_table *udp_table;
1018 struct raw_hashinfo *raw_hash;
1021 struct module *owner;
1025 struct list_head node;
1026 #ifdef SOCK_REFCNT_DEBUG
1029 #ifdef CONFIG_MEMCG_KMEM
1031 * cgroup specific init/deinit functions. Called once for all
1032 * protocols that implement it, from cgroups populate function.
1033 * This function has to setup any files the protocol want to
1034 * appear in the kmem cgroup filesystem.
1036 int (*init_cgroup)(struct mem_cgroup *memcg,
1037 struct cgroup_subsys *ss);
1038 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1039 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1044 * Bits in struct cg_proto.flags
1046 enum cg_proto_flags {
1047 /* Currently active and new sockets should be assigned to cgroups */
1049 /* It was ever activated; we must disarm static keys on destruction */
1050 MEMCG_SOCK_ACTIVATED,
1054 struct page_counter memory_allocated; /* Current allocated memory. */
1055 struct percpu_counter sockets_allocated; /* Current number of sockets. */
1056 int memory_pressure;
1058 unsigned long flags;
1060 * memcg field is used to find which memcg we belong directly
1061 * Each memcg struct can hold more than one cg_proto, so container_of
1064 * The elegant solution would be having an inverse function to
1065 * proto_cgroup in struct proto, but that means polluting the structure
1066 * for everybody, instead of just for memcg users.
1068 struct mem_cgroup *memcg;
1071 int proto_register(struct proto *prot, int alloc_slab);
1072 void proto_unregister(struct proto *prot);
1074 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1076 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1079 #ifdef SOCK_REFCNT_DEBUG
1080 static inline void sk_refcnt_debug_inc(struct sock *sk)
1082 atomic_inc(&sk->sk_prot->socks);
1085 static inline void sk_refcnt_debug_dec(struct sock *sk)
1087 atomic_dec(&sk->sk_prot->socks);
1088 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1089 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1092 static inline void sk_refcnt_debug_release(const struct sock *sk)
1094 if (atomic_read(&sk->sk_refcnt) != 1)
1095 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1096 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1098 #else /* SOCK_REFCNT_DEBUG */
1099 #define sk_refcnt_debug_inc(sk) do { } while (0)
1100 #define sk_refcnt_debug_dec(sk) do { } while (0)
1101 #define sk_refcnt_debug_release(sk) do { } while (0)
1102 #endif /* SOCK_REFCNT_DEBUG */
1104 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1105 extern struct static_key memcg_socket_limit_enabled;
1106 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1107 struct cg_proto *cg_proto)
1109 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1111 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1113 #define mem_cgroup_sockets_enabled 0
1114 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1115 struct cg_proto *cg_proto)
1121 static inline bool sk_stream_memory_free(const struct sock *sk)
1123 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1126 return sk->sk_prot->stream_memory_free ?
1127 sk->sk_prot->stream_memory_free(sk) : true;
1130 static inline bool sk_stream_is_writeable(const struct sock *sk)
1132 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1133 sk_stream_memory_free(sk);
1137 static inline bool sk_has_memory_pressure(const struct sock *sk)
1139 return sk->sk_prot->memory_pressure != NULL;
1142 static inline bool sk_under_memory_pressure(const struct sock *sk)
1144 if (!sk->sk_prot->memory_pressure)
1147 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1148 return !!sk->sk_cgrp->memory_pressure;
1150 return !!*sk->sk_prot->memory_pressure;
1153 static inline void sk_leave_memory_pressure(struct sock *sk)
1155 int *memory_pressure = sk->sk_prot->memory_pressure;
1157 if (!memory_pressure)
1160 if (*memory_pressure)
1161 *memory_pressure = 0;
1163 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1164 struct cg_proto *cg_proto = sk->sk_cgrp;
1165 struct proto *prot = sk->sk_prot;
1167 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1168 cg_proto->memory_pressure = 0;
1173 static inline void sk_enter_memory_pressure(struct sock *sk)
1175 if (!sk->sk_prot->enter_memory_pressure)
1178 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1179 struct cg_proto *cg_proto = sk->sk_cgrp;
1180 struct proto *prot = sk->sk_prot;
1182 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1183 cg_proto->memory_pressure = 1;
1186 sk->sk_prot->enter_memory_pressure(sk);
1189 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1191 long *prot = sk->sk_prot->sysctl_mem;
1192 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1193 prot = sk->sk_cgrp->sysctl_mem;
1197 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1201 page_counter_charge(&prot->memory_allocated, amt);
1203 if (page_counter_read(&prot->memory_allocated) >
1204 prot->memory_allocated.limit)
1205 *parent_status = OVER_LIMIT;
1208 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1211 page_counter_uncharge(&prot->memory_allocated, amt);
1215 sk_memory_allocated(const struct sock *sk)
1217 struct proto *prot = sk->sk_prot;
1219 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1220 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1222 return atomic_long_read(prot->memory_allocated);
1226 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1228 struct proto *prot = sk->sk_prot;
1230 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1231 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1232 /* update the root cgroup regardless */
1233 atomic_long_add_return(amt, prot->memory_allocated);
1234 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1237 return atomic_long_add_return(amt, prot->memory_allocated);
1241 sk_memory_allocated_sub(struct sock *sk, int amt)
1243 struct proto *prot = sk->sk_prot;
1245 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1246 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1248 atomic_long_sub(amt, prot->memory_allocated);
1251 static inline void sk_sockets_allocated_dec(struct sock *sk)
1253 struct proto *prot = sk->sk_prot;
1255 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1256 struct cg_proto *cg_proto = sk->sk_cgrp;
1258 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1259 percpu_counter_dec(&cg_proto->sockets_allocated);
1262 percpu_counter_dec(prot->sockets_allocated);
1265 static inline void sk_sockets_allocated_inc(struct sock *sk)
1267 struct proto *prot = sk->sk_prot;
1269 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1270 struct cg_proto *cg_proto = sk->sk_cgrp;
1272 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1273 percpu_counter_inc(&cg_proto->sockets_allocated);
1276 percpu_counter_inc(prot->sockets_allocated);
1280 sk_sockets_allocated_read_positive(struct sock *sk)
1282 struct proto *prot = sk->sk_prot;
1284 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1285 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1287 return percpu_counter_read_positive(prot->sockets_allocated);
1291 proto_sockets_allocated_sum_positive(struct proto *prot)
1293 return percpu_counter_sum_positive(prot->sockets_allocated);
1297 proto_memory_allocated(struct proto *prot)
1299 return atomic_long_read(prot->memory_allocated);
1303 proto_memory_pressure(struct proto *prot)
1305 if (!prot->memory_pressure)
1307 return !!*prot->memory_pressure;
1311 #ifdef CONFIG_PROC_FS
1312 /* Called with local bh disabled */
1313 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1314 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1316 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1323 /* With per-bucket locks this operation is not-atomic, so that
1324 * this version is not worse.
1326 static inline void __sk_prot_rehash(struct sock *sk)
1328 sk->sk_prot->unhash(sk);
1329 sk->sk_prot->hash(sk);
1332 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1334 /* About 10 seconds */
1335 #define SOCK_DESTROY_TIME (10*HZ)
1337 /* Sockets 0-1023 can't be bound to unless you are superuser */
1338 #define PROT_SOCK 1024
1340 #define SHUTDOWN_MASK 3
1341 #define RCV_SHUTDOWN 1
1342 #define SEND_SHUTDOWN 2
1344 #define SOCK_SNDBUF_LOCK 1
1345 #define SOCK_RCVBUF_LOCK 2
1346 #define SOCK_BINDADDR_LOCK 4
1347 #define SOCK_BINDPORT_LOCK 8
1349 struct socket_alloc {
1350 struct socket socket;
1351 struct inode vfs_inode;
1354 static inline struct socket *SOCKET_I(struct inode *inode)
1356 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1359 static inline struct inode *SOCK_INODE(struct socket *socket)
1361 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1365 * Functions for memory accounting
1367 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1368 void __sk_mem_reclaim(struct sock *sk);
1370 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1371 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1372 #define SK_MEM_SEND 0
1373 #define SK_MEM_RECV 1
1375 static inline int sk_mem_pages(int amt)
1377 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1380 static inline bool sk_has_account(struct sock *sk)
1382 /* return true if protocol supports memory accounting */
1383 return !!sk->sk_prot->memory_allocated;
1386 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1388 if (!sk_has_account(sk))
1390 return size <= sk->sk_forward_alloc ||
1391 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1395 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1397 if (!sk_has_account(sk))
1399 return size<= sk->sk_forward_alloc ||
1400 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1401 skb_pfmemalloc(skb);
1404 static inline void sk_mem_reclaim(struct sock *sk)
1406 if (!sk_has_account(sk))
1408 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1409 __sk_mem_reclaim(sk);
1412 static inline void sk_mem_reclaim_partial(struct sock *sk)
1414 if (!sk_has_account(sk))
1416 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1417 __sk_mem_reclaim(sk);
1420 static inline void sk_mem_charge(struct sock *sk, int size)
1422 if (!sk_has_account(sk))
1424 sk->sk_forward_alloc -= size;
1427 static inline void sk_mem_uncharge(struct sock *sk, int size)
1429 if (!sk_has_account(sk))
1431 sk->sk_forward_alloc += size;
1434 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1436 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1437 sk->sk_wmem_queued -= skb->truesize;
1438 sk_mem_uncharge(sk, skb->truesize);
1442 /* Used by processes to "lock" a socket state, so that
1443 * interrupts and bottom half handlers won't change it
1444 * from under us. It essentially blocks any incoming
1445 * packets, so that we won't get any new data or any
1446 * packets that change the state of the socket.
1448 * While locked, BH processing will add new packets to
1449 * the backlog queue. This queue is processed by the
1450 * owner of the socket lock right before it is released.
1452 * Since ~2.3.5 it is also exclusive sleep lock serializing
1453 * accesses from user process context.
1455 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1457 static inline void sock_release_ownership(struct sock *sk)
1459 sk->sk_lock.owned = 0;
1463 * Macro so as to not evaluate some arguments when
1464 * lockdep is not enabled.
1466 * Mark both the sk_lock and the sk_lock.slock as a
1467 * per-address-family lock class.
1469 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1471 sk->sk_lock.owned = 0; \
1472 init_waitqueue_head(&sk->sk_lock.wq); \
1473 spin_lock_init(&(sk)->sk_lock.slock); \
1474 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1475 sizeof((sk)->sk_lock)); \
1476 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1478 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1481 void lock_sock_nested(struct sock *sk, int subclass);
1483 static inline void lock_sock(struct sock *sk)
1485 lock_sock_nested(sk, 0);
1488 void release_sock(struct sock *sk);
1490 /* BH context may only use the following locking interface. */
1491 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1492 #define bh_lock_sock_nested(__sk) \
1493 spin_lock_nested(&((__sk)->sk_lock.slock), \
1494 SINGLE_DEPTH_NESTING)
1495 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1497 bool lock_sock_fast(struct sock *sk);
1499 * unlock_sock_fast - complement of lock_sock_fast
1503 * fast unlock socket for user context.
1504 * If slow mode is on, we call regular release_sock()
1506 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1511 spin_unlock_bh(&sk->sk_lock.slock);
1515 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1516 struct proto *prot);
1517 void sk_free(struct sock *sk);
1518 void sk_release_kernel(struct sock *sk);
1519 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1521 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1523 void sock_wfree(struct sk_buff *skb);
1524 void skb_orphan_partial(struct sk_buff *skb);
1525 void sock_rfree(struct sk_buff *skb);
1526 void sock_efree(struct sk_buff *skb);
1528 void sock_edemux(struct sk_buff *skb);
1530 #define sock_edemux(skb) sock_efree(skb)
1533 int sock_setsockopt(struct socket *sock, int level, int op,
1534 char __user *optval, unsigned int optlen);
1536 int sock_getsockopt(struct socket *sock, int level, int op,
1537 char __user *optval, int __user *optlen);
1538 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1539 int noblock, int *errcode);
1540 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1541 unsigned long data_len, int noblock,
1542 int *errcode, int max_page_order);
1543 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1544 void sock_kfree_s(struct sock *sk, void *mem, int size);
1545 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1546 void sk_send_sigurg(struct sock *sk);
1549 * Functions to fill in entries in struct proto_ops when a protocol
1550 * does not implement a particular function.
1552 int sock_no_bind(struct socket *, struct sockaddr *, int);
1553 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1554 int sock_no_socketpair(struct socket *, struct socket *);
1555 int sock_no_accept(struct socket *, struct socket *, int);
1556 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1557 unsigned int sock_no_poll(struct file *, struct socket *,
1558 struct poll_table_struct *);
1559 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1560 int sock_no_listen(struct socket *, int);
1561 int sock_no_shutdown(struct socket *, int);
1562 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1563 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1564 int sock_no_sendmsg(struct kiocb *, struct socket *, struct msghdr *, size_t);
1565 int sock_no_recvmsg(struct kiocb *, struct socket *, struct msghdr *, size_t,
1567 int sock_no_mmap(struct file *file, struct socket *sock,
1568 struct vm_area_struct *vma);
1569 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1570 size_t size, int flags);
1573 * Functions to fill in entries in struct proto_ops when a protocol
1574 * uses the inet style.
1576 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1577 char __user *optval, int __user *optlen);
1578 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1579 struct msghdr *msg, size_t size, int flags);
1580 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1581 char __user *optval, unsigned int optlen);
1582 int compat_sock_common_getsockopt(struct socket *sock, int level,
1583 int optname, char __user *optval, int __user *optlen);
1584 int compat_sock_common_setsockopt(struct socket *sock, int level,
1585 int optname, char __user *optval, unsigned int optlen);
1587 void sk_common_release(struct sock *sk);
1590 * Default socket callbacks and setup code
1593 /* Initialise core socket variables */
1594 void sock_init_data(struct socket *sock, struct sock *sk);
1597 * Socket reference counting postulates.
1599 * * Each user of socket SHOULD hold a reference count.
1600 * * Each access point to socket (an hash table bucket, reference from a list,
1601 * running timer, skb in flight MUST hold a reference count.
1602 * * When reference count hits 0, it means it will never increase back.
1603 * * When reference count hits 0, it means that no references from
1604 * outside exist to this socket and current process on current CPU
1605 * is last user and may/should destroy this socket.
1606 * * sk_free is called from any context: process, BH, IRQ. When
1607 * it is called, socket has no references from outside -> sk_free
1608 * may release descendant resources allocated by the socket, but
1609 * to the time when it is called, socket is NOT referenced by any
1610 * hash tables, lists etc.
1611 * * Packets, delivered from outside (from network or from another process)
1612 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1613 * when they sit in queue. Otherwise, packets will leak to hole, when
1614 * socket is looked up by one cpu and unhasing is made by another CPU.
1615 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1616 * (leak to backlog). Packet socket does all the processing inside
1617 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1618 * use separate SMP lock, so that they are prone too.
1621 /* Ungrab socket and destroy it, if it was the last reference. */
1622 static inline void sock_put(struct sock *sk)
1624 if (atomic_dec_and_test(&sk->sk_refcnt))
1627 /* Generic version of sock_put(), dealing with all sockets
1628 * (TCP_TIMEWAIT, ESTABLISHED...)
1630 void sock_gen_put(struct sock *sk);
1632 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1634 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1636 sk->sk_tx_queue_mapping = tx_queue;
1639 static inline void sk_tx_queue_clear(struct sock *sk)
1641 sk->sk_tx_queue_mapping = -1;
1644 static inline int sk_tx_queue_get(const struct sock *sk)
1646 return sk ? sk->sk_tx_queue_mapping : -1;
1649 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1651 sk_tx_queue_clear(sk);
1652 sk->sk_socket = sock;
1655 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1657 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1658 return &rcu_dereference_raw(sk->sk_wq)->wait;
1660 /* Detach socket from process context.
1661 * Announce socket dead, detach it from wait queue and inode.
1662 * Note that parent inode held reference count on this struct sock,
1663 * we do not release it in this function, because protocol
1664 * probably wants some additional cleanups or even continuing
1665 * to work with this socket (TCP).
1667 static inline void sock_orphan(struct sock *sk)
1669 write_lock_bh(&sk->sk_callback_lock);
1670 sock_set_flag(sk, SOCK_DEAD);
1671 sk_set_socket(sk, NULL);
1673 write_unlock_bh(&sk->sk_callback_lock);
1676 static inline void sock_graft(struct sock *sk, struct socket *parent)
1678 write_lock_bh(&sk->sk_callback_lock);
1679 sk->sk_wq = parent->wq;
1681 sk_set_socket(sk, parent);
1682 security_sock_graft(sk, parent);
1683 write_unlock_bh(&sk->sk_callback_lock);
1686 kuid_t sock_i_uid(struct sock *sk);
1687 unsigned long sock_i_ino(struct sock *sk);
1689 static inline struct dst_entry *
1690 __sk_dst_get(struct sock *sk)
1692 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1693 lockdep_is_held(&sk->sk_lock.slock));
1696 static inline struct dst_entry *
1697 sk_dst_get(struct sock *sk)
1699 struct dst_entry *dst;
1702 dst = rcu_dereference(sk->sk_dst_cache);
1703 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1709 static inline void dst_negative_advice(struct sock *sk)
1711 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1713 if (dst && dst->ops->negative_advice) {
1714 ndst = dst->ops->negative_advice(dst);
1717 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1718 sk_tx_queue_clear(sk);
1724 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1726 struct dst_entry *old_dst;
1728 sk_tx_queue_clear(sk);
1730 * This can be called while sk is owned by the caller only,
1731 * with no state that can be checked in a rcu_dereference_check() cond
1733 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1734 rcu_assign_pointer(sk->sk_dst_cache, dst);
1735 dst_release(old_dst);
1739 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1741 struct dst_entry *old_dst;
1743 sk_tx_queue_clear(sk);
1744 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1745 dst_release(old_dst);
1749 __sk_dst_reset(struct sock *sk)
1751 __sk_dst_set(sk, NULL);
1755 sk_dst_reset(struct sock *sk)
1757 sk_dst_set(sk, NULL);
1760 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1762 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1764 bool sk_mc_loop(struct sock *sk);
1766 static inline bool sk_can_gso(const struct sock *sk)
1768 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1771 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1773 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1775 sk->sk_route_nocaps |= flags;
1776 sk->sk_route_caps &= ~flags;
1779 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1780 struct iov_iter *from, char *to,
1781 int copy, int offset)
1783 if (skb->ip_summed == CHECKSUM_NONE) {
1785 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1787 skb->csum = csum_block_add(skb->csum, csum, offset);
1788 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1789 if (copy_from_iter_nocache(to, copy, from) != copy)
1791 } else if (copy_from_iter(to, copy, from) != copy)
1797 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1798 struct iov_iter *from, int copy)
1800 int err, offset = skb->len;
1802 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1805 __skb_trim(skb, offset);
1810 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1811 struct sk_buff *skb,
1817 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1823 skb->data_len += copy;
1824 skb->truesize += copy;
1825 sk->sk_wmem_queued += copy;
1826 sk_mem_charge(sk, copy);
1831 * sk_wmem_alloc_get - returns write allocations
1834 * Returns sk_wmem_alloc minus initial offset of one
1836 static inline int sk_wmem_alloc_get(const struct sock *sk)
1838 return atomic_read(&sk->sk_wmem_alloc) - 1;
1842 * sk_rmem_alloc_get - returns read allocations
1845 * Returns sk_rmem_alloc
1847 static inline int sk_rmem_alloc_get(const struct sock *sk)
1849 return atomic_read(&sk->sk_rmem_alloc);
1853 * sk_has_allocations - check if allocations are outstanding
1856 * Returns true if socket has write or read allocations
1858 static inline bool sk_has_allocations(const struct sock *sk)
1860 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1864 * wq_has_sleeper - check if there are any waiting processes
1865 * @wq: struct socket_wq
1867 * Returns true if socket_wq has waiting processes
1869 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1870 * barrier call. They were added due to the race found within the tcp code.
1872 * Consider following tcp code paths:
1876 * sys_select receive packet
1878 * __add_wait_queue update tp->rcv_nxt
1880 * tp->rcv_nxt check sock_def_readable
1882 * schedule rcu_read_lock();
1883 * wq = rcu_dereference(sk->sk_wq);
1884 * if (wq && waitqueue_active(&wq->wait))
1885 * wake_up_interruptible(&wq->wait)
1889 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1890 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1891 * could then endup calling schedule and sleep forever if there are no more
1892 * data on the socket.
1895 static inline bool wq_has_sleeper(struct socket_wq *wq)
1897 /* We need to be sure we are in sync with the
1898 * add_wait_queue modifications to the wait queue.
1900 * This memory barrier is paired in the sock_poll_wait.
1903 return wq && waitqueue_active(&wq->wait);
1907 * sock_poll_wait - place memory barrier behind the poll_wait call.
1909 * @wait_address: socket wait queue
1912 * See the comments in the wq_has_sleeper function.
1914 static inline void sock_poll_wait(struct file *filp,
1915 wait_queue_head_t *wait_address, poll_table *p)
1917 if (!poll_does_not_wait(p) && wait_address) {
1918 poll_wait(filp, wait_address, p);
1919 /* We need to be sure we are in sync with the
1920 * socket flags modification.
1922 * This memory barrier is paired in the wq_has_sleeper.
1928 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1930 if (sk->sk_txhash) {
1932 skb->hash = sk->sk_txhash;
1937 * Queue a received datagram if it will fit. Stream and sequenced
1938 * protocols can't normally use this as they need to fit buffers in
1939 * and play with them.
1941 * Inlined as it's very short and called for pretty much every
1942 * packet ever received.
1945 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1949 skb->destructor = sock_wfree;
1950 skb_set_hash_from_sk(skb, sk);
1952 * We used to take a refcount on sk, but following operation
1953 * is enough to guarantee sk_free() wont free this sock until
1954 * all in-flight packets are completed
1956 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1959 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1963 skb->destructor = sock_rfree;
1964 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1965 sk_mem_charge(sk, skb->truesize);
1968 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1969 unsigned long expires);
1971 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1973 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1975 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1976 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1979 * Recover an error report and clear atomically
1982 static inline int sock_error(struct sock *sk)
1985 if (likely(!sk->sk_err))
1987 err = xchg(&sk->sk_err, 0);
1991 static inline unsigned long sock_wspace(struct sock *sk)
1995 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1996 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2003 static inline void sk_wake_async(struct sock *sk, int how, int band)
2005 if (sock_flag(sk, SOCK_FASYNC))
2006 sock_wake_async(sk->sk_socket, how, band);
2009 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2010 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2011 * Note: for send buffers, TCP works better if we can build two skbs at
2014 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2016 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2017 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2019 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2021 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2022 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2023 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2027 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2030 * sk_page_frag - return an appropriate page_frag
2033 * If socket allocation mode allows current thread to sleep, it means its
2034 * safe to use the per task page_frag instead of the per socket one.
2036 static inline struct page_frag *sk_page_frag(struct sock *sk)
2038 if (sk->sk_allocation & __GFP_WAIT)
2039 return ¤t->task_frag;
2041 return &sk->sk_frag;
2044 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2047 * Default write policy as shown to user space via poll/select/SIGIO
2049 static inline bool sock_writeable(const struct sock *sk)
2051 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2054 static inline gfp_t gfp_any(void)
2056 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2059 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2061 return noblock ? 0 : sk->sk_rcvtimeo;
2064 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2066 return noblock ? 0 : sk->sk_sndtimeo;
2069 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2071 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2074 /* Alas, with timeout socket operations are not restartable.
2075 * Compare this to poll().
2077 static inline int sock_intr_errno(long timeo)
2079 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2082 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2083 struct sk_buff *skb);
2084 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2085 struct sk_buff *skb);
2088 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2090 ktime_t kt = skb->tstamp;
2091 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2094 * generate control messages if
2095 * - receive time stamping in software requested
2096 * - software time stamp available and wanted
2097 * - hardware time stamps available and wanted
2099 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2100 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2101 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2102 (hwtstamps->hwtstamp.tv64 &&
2103 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2104 __sock_recv_timestamp(msg, sk, skb);
2108 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2109 __sock_recv_wifi_status(msg, sk, skb);
2112 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2113 struct sk_buff *skb);
2115 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2116 struct sk_buff *skb)
2118 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2119 (1UL << SOCK_RCVTSTAMP))
2120 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2121 SOF_TIMESTAMPING_RAW_HARDWARE)
2123 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2124 __sock_recv_ts_and_drops(msg, sk, skb);
2126 sk->sk_stamp = skb->tstamp;
2129 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2132 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2133 * @sk: socket sending this packet
2134 * @tx_flags: completed with instructions for time stamping
2136 * Note : callers should take care of initial *tx_flags value (usually 0)
2138 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2140 if (unlikely(sk->sk_tsflags))
2141 __sock_tx_timestamp(sk, tx_flags);
2142 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2143 *tx_flags |= SKBTX_WIFI_STATUS;
2147 * sk_eat_skb - Release a skb if it is no longer needed
2148 * @sk: socket to eat this skb from
2149 * @skb: socket buffer to eat
2151 * This routine must be called with interrupts disabled or with the socket
2152 * locked so that the sk_buff queue operation is ok.
2154 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2156 __skb_unlink(skb, &sk->sk_receive_queue);
2161 struct net *sock_net(const struct sock *sk)
2163 return read_pnet(&sk->sk_net);
2167 void sock_net_set(struct sock *sk, struct net *net)
2169 write_pnet(&sk->sk_net, net);
2173 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2174 * They should not hold a reference to a namespace in order to allow
2176 * Sockets after sk_change_net should be released using sk_release_kernel
2178 static inline void sk_change_net(struct sock *sk, struct net *net)
2180 struct net *current_net = sock_net(sk);
2182 if (!net_eq(current_net, net)) {
2183 put_net(current_net);
2184 sock_net_set(sk, hold_net(net));
2188 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2191 struct sock *sk = skb->sk;
2193 skb->destructor = NULL;
2200 void sock_enable_timestamp(struct sock *sk, int flag);
2201 int sock_get_timestamp(struct sock *, struct timeval __user *);
2202 int sock_get_timestampns(struct sock *, struct timespec __user *);
2203 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2206 bool sk_ns_capable(const struct sock *sk,
2207 struct user_namespace *user_ns, int cap);
2208 bool sk_capable(const struct sock *sk, int cap);
2209 bool sk_net_capable(const struct sock *sk, int cap);
2211 extern __u32 sysctl_wmem_max;
2212 extern __u32 sysctl_rmem_max;
2214 extern int sysctl_tstamp_allow_data;
2215 extern int sysctl_optmem_max;
2217 extern __u32 sysctl_wmem_default;
2218 extern __u32 sysctl_rmem_default;
2220 #endif /* _SOCK_H */