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/memcontrol.h>
58 #include <linux/res_counter.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>
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:4;
185 int skc_bound_dev_if;
187 struct hlist_node skc_bind_node;
188 struct hlist_nulls_node skc_portaddr_node;
190 struct proto *skc_prot;
195 #if IS_ENABLED(CONFIG_IPV6)
196 struct in6_addr skc_v6_daddr;
197 struct in6_addr skc_v6_rcv_saddr;
201 * fields between dontcopy_begin/dontcopy_end
202 * are not copied in sock_copy()
205 int skc_dontcopy_begin[0];
208 struct hlist_node skc_node;
209 struct hlist_nulls_node skc_nulls_node;
211 int skc_tx_queue_mapping;
214 int skc_dontcopy_end[0];
220 * struct sock - network layer representation of sockets
221 * @__sk_common: shared layout with inet_timewait_sock
222 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
223 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
224 * @sk_lock: synchronizer
225 * @sk_rcvbuf: size of receive buffer in bytes
226 * @sk_wq: sock wait queue and async head
227 * @sk_rx_dst: receive input route used by early demux
228 * @sk_dst_cache: destination cache
229 * @sk_dst_lock: destination cache lock
230 * @sk_policy: flow policy
231 * @sk_receive_queue: incoming packets
232 * @sk_wmem_alloc: transmit queue bytes committed
233 * @sk_write_queue: Packet sending queue
234 * @sk_async_wait_queue: DMA copied packets
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: %SO_NO_CHECK setting, whether or not checkup packets
247 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
248 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
249 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
250 * @sk_gso_max_size: Maximum GSO segment size to build
251 * @sk_gso_max_segs: Maximum number of GSO segments
252 * @sk_lingertime: %SO_LINGER l_linger setting
253 * @sk_backlog: always used with the per-socket spinlock held
254 * @sk_callback_lock: used with the callbacks in the end of this struct
255 * @sk_error_queue: rarely used
256 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
257 * IPV6_ADDRFORM for instance)
258 * @sk_err: last error
259 * @sk_err_soft: errors that don't cause failure but are the cause of a
260 * persistent failure not just 'timed out'
261 * @sk_drops: raw/udp drops counter
262 * @sk_ack_backlog: current listen backlog
263 * @sk_max_ack_backlog: listen backlog set in listen()
264 * @sk_priority: %SO_PRIORITY setting
265 * @sk_cgrp_prioidx: socket group's priority map index
266 * @sk_type: socket type (%SOCK_STREAM, etc)
267 * @sk_protocol: which protocol this socket belongs in this network family
268 * @sk_peer_pid: &struct pid for this socket's peer
269 * @sk_peer_cred: %SO_PEERCRED setting
270 * @sk_rcvlowat: %SO_RCVLOWAT setting
271 * @sk_rcvtimeo: %SO_RCVTIMEO setting
272 * @sk_sndtimeo: %SO_SNDTIMEO setting
273 * @sk_rxhash: flow hash received from netif layer
274 * @sk_filter: socket filtering instructions
275 * @sk_protinfo: private area, net family specific, when not using slab
276 * @sk_timer: sock cleanup timer
277 * @sk_stamp: time stamp of last packet received
278 * @sk_socket: Identd and reporting IO signals
279 * @sk_user_data: RPC layer private data
280 * @sk_frag: cached page frag
281 * @sk_peek_off: current peek_offset value
282 * @sk_send_head: front of stuff to transmit
283 * @sk_security: used by security modules
284 * @sk_mark: generic packet mark
285 * @sk_classid: this socket's cgroup classid
286 * @sk_cgrp: this socket's cgroup-specific proto data
287 * @sk_write_pending: a write to stream socket waits to start
288 * @sk_state_change: callback to indicate change in the state of the sock
289 * @sk_data_ready: callback to indicate there is data to be processed
290 * @sk_write_space: callback to indicate there is bf sending space available
291 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
292 * @sk_backlog_rcv: callback to process the backlog
293 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
297 * Now struct inet_timewait_sock also uses sock_common, so please just
298 * don't add nothing before this first member (__sk_common) --acme
300 struct sock_common __sk_common;
301 #define sk_node __sk_common.skc_node
302 #define sk_nulls_node __sk_common.skc_nulls_node
303 #define sk_refcnt __sk_common.skc_refcnt
304 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
306 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
307 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
308 #define sk_hash __sk_common.skc_hash
309 #define sk_portpair __sk_common.skc_portpair
310 #define sk_num __sk_common.skc_num
311 #define sk_dport __sk_common.skc_dport
312 #define sk_addrpair __sk_common.skc_addrpair
313 #define sk_daddr __sk_common.skc_daddr
314 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
315 #define sk_family __sk_common.skc_family
316 #define sk_state __sk_common.skc_state
317 #define sk_reuse __sk_common.skc_reuse
318 #define sk_reuseport __sk_common.skc_reuseport
319 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
320 #define sk_bind_node __sk_common.skc_bind_node
321 #define sk_prot __sk_common.skc_prot
322 #define sk_net __sk_common.skc_net
323 #define sk_v6_daddr __sk_common.skc_v6_daddr
324 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
326 socket_lock_t sk_lock;
327 struct sk_buff_head sk_receive_queue;
329 * The backlog queue is special, it is always used with
330 * the per-socket spinlock held and requires low latency
331 * access. Therefore we special case it's implementation.
332 * Note : rmem_alloc is in this structure to fill a hole
333 * on 64bit arches, not because its logically part of
339 struct sk_buff *head;
340 struct sk_buff *tail;
342 #define sk_rmem_alloc sk_backlog.rmem_alloc
343 int sk_forward_alloc;
347 #ifdef CONFIG_NET_RX_BUSY_POLL
348 unsigned int sk_napi_id;
349 unsigned int sk_ll_usec;
354 struct sk_filter __rcu *sk_filter;
355 struct socket_wq __rcu *sk_wq;
357 #ifdef CONFIG_NET_DMA
358 struct sk_buff_head sk_async_wait_queue;
362 struct xfrm_policy *sk_policy[2];
364 unsigned long sk_flags;
365 struct dst_entry *sk_rx_dst;
366 struct dst_entry __rcu *sk_dst_cache;
367 spinlock_t sk_dst_lock;
368 atomic_t sk_wmem_alloc;
369 atomic_t sk_omem_alloc;
371 struct sk_buff_head sk_write_queue;
372 kmemcheck_bitfield_begin(flags);
373 unsigned int sk_shutdown : 2,
378 kmemcheck_bitfield_end(flags);
381 u32 sk_pacing_rate; /* bytes per second */
382 u32 sk_max_pacing_rate;
383 netdev_features_t sk_route_caps;
384 netdev_features_t sk_route_nocaps;
386 unsigned int sk_gso_max_size;
389 unsigned long sk_lingertime;
390 struct sk_buff_head sk_error_queue;
391 struct proto *sk_prot_creator;
392 rwlock_t sk_callback_lock;
395 unsigned short sk_ack_backlog;
396 unsigned short sk_max_ack_backlog;
398 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
399 __u32 sk_cgrp_prioidx;
401 struct pid *sk_peer_pid;
402 const struct cred *sk_peer_cred;
406 struct timer_list sk_timer;
408 struct socket *sk_socket;
410 struct page_frag sk_frag;
411 struct sk_buff *sk_send_head;
413 int sk_write_pending;
414 #ifdef CONFIG_SECURITY
419 struct cg_proto *sk_cgrp;
420 void (*sk_state_change)(struct sock *sk);
421 void (*sk_data_ready)(struct sock *sk, int bytes);
422 void (*sk_write_space)(struct sock *sk);
423 void (*sk_error_report)(struct sock *sk);
424 int (*sk_backlog_rcv)(struct sock *sk,
425 struct sk_buff *skb);
426 void (*sk_destruct)(struct sock *sk);
429 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
431 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
432 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
435 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
436 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
437 * on a socket means that the socket will reuse everybody else's port
438 * without looking at the other's sk_reuse value.
441 #define SK_NO_REUSE 0
442 #define SK_CAN_REUSE 1
443 #define SK_FORCE_REUSE 2
445 static inline int sk_peek_offset(struct sock *sk, int flags)
447 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
448 return sk->sk_peek_off;
453 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
455 if (sk->sk_peek_off >= 0) {
456 if (sk->sk_peek_off >= val)
457 sk->sk_peek_off -= val;
463 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
465 if (sk->sk_peek_off >= 0)
466 sk->sk_peek_off += val;
470 * Hashed lists helper routines
472 static inline struct sock *sk_entry(const struct hlist_node *node)
474 return hlist_entry(node, struct sock, sk_node);
477 static inline struct sock *__sk_head(const struct hlist_head *head)
479 return hlist_entry(head->first, struct sock, sk_node);
482 static inline struct sock *sk_head(const struct hlist_head *head)
484 return hlist_empty(head) ? NULL : __sk_head(head);
487 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
489 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
492 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
494 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
497 static inline struct sock *sk_next(const struct sock *sk)
499 return sk->sk_node.next ?
500 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
503 static inline struct sock *sk_nulls_next(const struct sock *sk)
505 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
506 hlist_nulls_entry(sk->sk_nulls_node.next,
507 struct sock, sk_nulls_node) :
511 static inline bool sk_unhashed(const struct sock *sk)
513 return hlist_unhashed(&sk->sk_node);
516 static inline bool sk_hashed(const struct sock *sk)
518 return !sk_unhashed(sk);
521 static inline void sk_node_init(struct hlist_node *node)
526 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
531 static inline void __sk_del_node(struct sock *sk)
533 __hlist_del(&sk->sk_node);
536 /* NB: equivalent to hlist_del_init_rcu */
537 static inline bool __sk_del_node_init(struct sock *sk)
541 sk_node_init(&sk->sk_node);
547 /* Grab socket reference count. This operation is valid only
548 when sk is ALREADY grabbed f.e. it is found in hash table
549 or a list and the lookup is made under lock preventing hash table
553 static inline void sock_hold(struct sock *sk)
555 atomic_inc(&sk->sk_refcnt);
558 /* Ungrab socket in the context, which assumes that socket refcnt
559 cannot hit zero, f.e. it is true in context of any socketcall.
561 static inline void __sock_put(struct sock *sk)
563 atomic_dec(&sk->sk_refcnt);
566 static inline bool sk_del_node_init(struct sock *sk)
568 bool rc = __sk_del_node_init(sk);
571 /* paranoid for a while -acme */
572 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
577 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
579 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
582 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
588 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
590 bool rc = __sk_nulls_del_node_init_rcu(sk);
593 /* paranoid for a while -acme */
594 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
600 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
602 hlist_add_head(&sk->sk_node, list);
605 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
608 __sk_add_node(sk, list);
611 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
614 hlist_add_head_rcu(&sk->sk_node, list);
617 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
619 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
622 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
625 __sk_nulls_add_node_rcu(sk, list);
628 static inline void __sk_del_bind_node(struct sock *sk)
630 __hlist_del(&sk->sk_bind_node);
633 static inline void sk_add_bind_node(struct sock *sk,
634 struct hlist_head *list)
636 hlist_add_head(&sk->sk_bind_node, list);
639 #define sk_for_each(__sk, list) \
640 hlist_for_each_entry(__sk, list, sk_node)
641 #define sk_for_each_rcu(__sk, list) \
642 hlist_for_each_entry_rcu(__sk, list, sk_node)
643 #define sk_nulls_for_each(__sk, node, list) \
644 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
645 #define sk_nulls_for_each_rcu(__sk, node, list) \
646 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
647 #define sk_for_each_from(__sk) \
648 hlist_for_each_entry_from(__sk, sk_node)
649 #define sk_nulls_for_each_from(__sk, node) \
650 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
651 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
652 #define sk_for_each_safe(__sk, tmp, list) \
653 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
654 #define sk_for_each_bound(__sk, list) \
655 hlist_for_each_entry(__sk, list, sk_bind_node)
657 static inline struct user_namespace *sk_user_ns(struct sock *sk)
659 /* Careful only use this in a context where these parameters
660 * can not change and must all be valid, such as recvmsg from
663 return sk->sk_socket->file->f_cred->user_ns;
677 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
678 SOCK_DBG, /* %SO_DEBUG setting */
679 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
680 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
681 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
682 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
683 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
684 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
685 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
686 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
687 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
688 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
689 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
690 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
691 SOCK_FASYNC, /* fasync() active */
693 SOCK_ZEROCOPY, /* buffers from userspace */
694 SOCK_WIFI_STATUS, /* push wifi status to userspace */
695 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
696 * Will use last 4 bytes of packet sent from
697 * user-space instead.
699 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
700 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
703 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
705 nsk->sk_flags = osk->sk_flags;
708 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
710 __set_bit(flag, &sk->sk_flags);
713 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
715 __clear_bit(flag, &sk->sk_flags);
718 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
720 return test_bit(flag, &sk->sk_flags);
724 extern struct static_key memalloc_socks;
725 static inline int sk_memalloc_socks(void)
727 return static_key_false(&memalloc_socks);
731 static inline int sk_memalloc_socks(void)
738 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
740 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
743 static inline void sk_acceptq_removed(struct sock *sk)
745 sk->sk_ack_backlog--;
748 static inline void sk_acceptq_added(struct sock *sk)
750 sk->sk_ack_backlog++;
753 static inline bool sk_acceptq_is_full(const struct sock *sk)
755 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
759 * Compute minimal free write space needed to queue new packets.
761 static inline int sk_stream_min_wspace(const struct sock *sk)
763 return sk->sk_wmem_queued >> 1;
766 static inline int sk_stream_wspace(const struct sock *sk)
768 return sk->sk_sndbuf - sk->sk_wmem_queued;
771 void sk_stream_write_space(struct sock *sk);
773 /* OOB backlog add */
774 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
776 /* dont let skb dst not refcounted, we are going to leave rcu lock */
779 if (!sk->sk_backlog.tail)
780 sk->sk_backlog.head = skb;
782 sk->sk_backlog.tail->next = skb;
784 sk->sk_backlog.tail = skb;
789 * Take into account size of receive queue and backlog queue
790 * Do not take into account this skb truesize,
791 * to allow even a single big packet to come.
793 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
796 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
798 return qsize > limit;
801 /* The per-socket spinlock must be held here. */
802 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
805 if (sk_rcvqueues_full(sk, skb, limit))
808 __sk_add_backlog(sk, skb);
809 sk->sk_backlog.len += skb->truesize;
813 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
815 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
817 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
818 return __sk_backlog_rcv(sk, skb);
820 return sk->sk_backlog_rcv(sk, skb);
823 static inline void sock_rps_record_flow_hash(__u32 hash)
826 struct rps_sock_flow_table *sock_flow_table;
829 sock_flow_table = rcu_dereference(rps_sock_flow_table);
830 rps_record_sock_flow(sock_flow_table, hash);
835 static inline void sock_rps_reset_flow_hash(__u32 hash)
838 struct rps_sock_flow_table *sock_flow_table;
841 sock_flow_table = rcu_dereference(rps_sock_flow_table);
842 rps_reset_sock_flow(sock_flow_table, hash);
847 static inline void sock_rps_record_flow(const struct sock *sk)
850 sock_rps_record_flow_hash(sk->sk_rxhash);
854 static inline void sock_rps_reset_flow(const struct sock *sk)
857 sock_rps_reset_flow_hash(sk->sk_rxhash);
861 static inline void sock_rps_save_rxhash(struct sock *sk,
862 const struct sk_buff *skb)
865 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
866 sock_rps_reset_flow(sk);
867 sk->sk_rxhash = skb->rxhash;
872 static inline void sock_rps_reset_rxhash(struct sock *sk)
875 sock_rps_reset_flow(sk);
880 #define sk_wait_event(__sk, __timeo, __condition) \
882 release_sock(__sk); \
883 __rc = __condition; \
885 *(__timeo) = schedule_timeout(*(__timeo)); \
888 __rc = __condition; \
892 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
893 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
894 void sk_stream_wait_close(struct sock *sk, long timeo_p);
895 int sk_stream_error(struct sock *sk, int flags, int err);
896 void sk_stream_kill_queues(struct sock *sk);
897 void sk_set_memalloc(struct sock *sk);
898 void sk_clear_memalloc(struct sock *sk);
900 int sk_wait_data(struct sock *sk, long *timeo);
902 struct request_sock_ops;
903 struct timewait_sock_ops;
904 struct inet_hashinfo;
909 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
910 * un-modified. Special care is taken when initializing object to zero.
912 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
914 if (offsetof(struct sock, sk_node.next) != 0)
915 memset(sk, 0, offsetof(struct sock, sk_node.next));
916 memset(&sk->sk_node.pprev, 0,
917 size - offsetof(struct sock, sk_node.pprev));
920 /* Networking protocol blocks we attach to sockets.
921 * socket layer -> transport layer interface
922 * transport -> network interface is defined by struct inet_proto
925 void (*close)(struct sock *sk,
927 int (*connect)(struct sock *sk,
928 struct sockaddr *uaddr,
930 int (*disconnect)(struct sock *sk, int flags);
932 struct sock * (*accept)(struct sock *sk, int flags, int *err);
934 int (*ioctl)(struct sock *sk, int cmd,
936 int (*init)(struct sock *sk);
937 void (*destroy)(struct sock *sk);
938 void (*shutdown)(struct sock *sk, int how);
939 int (*setsockopt)(struct sock *sk, int level,
940 int optname, char __user *optval,
941 unsigned int optlen);
942 int (*getsockopt)(struct sock *sk, int level,
943 int optname, char __user *optval,
946 int (*compat_setsockopt)(struct sock *sk,
948 int optname, char __user *optval,
949 unsigned int optlen);
950 int (*compat_getsockopt)(struct sock *sk,
952 int optname, char __user *optval,
954 int (*compat_ioctl)(struct sock *sk,
955 unsigned int cmd, unsigned long arg);
957 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
958 struct msghdr *msg, size_t len);
959 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
961 size_t len, int noblock, int flags,
963 int (*sendpage)(struct sock *sk, struct page *page,
964 int offset, size_t size, int flags);
965 int (*bind)(struct sock *sk,
966 struct sockaddr *uaddr, int addr_len);
968 int (*backlog_rcv) (struct sock *sk,
969 struct sk_buff *skb);
971 void (*release_cb)(struct sock *sk);
972 void (*mtu_reduced)(struct sock *sk);
974 /* Keeping track of sk's, looking them up, and port selection methods. */
975 void (*hash)(struct sock *sk);
976 void (*unhash)(struct sock *sk);
977 void (*rehash)(struct sock *sk);
978 int (*get_port)(struct sock *sk, unsigned short snum);
979 void (*clear_sk)(struct sock *sk, int size);
981 /* Keeping track of sockets in use */
982 #ifdef CONFIG_PROC_FS
983 unsigned int inuse_idx;
986 bool (*stream_memory_free)(const struct sock *sk);
987 /* Memory pressure */
988 void (*enter_memory_pressure)(struct sock *sk);
989 atomic_long_t *memory_allocated; /* Current allocated memory. */
990 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
992 * Pressure flag: try to collapse.
993 * Technical note: it is used by multiple contexts non atomically.
994 * All the __sk_mem_schedule() is of this nature: accounting
995 * is strict, actions are advisory and have some latency.
997 int *memory_pressure;
1004 struct kmem_cache *slab;
1005 unsigned int obj_size;
1008 struct percpu_counter *orphan_count;
1010 struct request_sock_ops *rsk_prot;
1011 struct timewait_sock_ops *twsk_prot;
1014 struct inet_hashinfo *hashinfo;
1015 struct udp_table *udp_table;
1016 struct raw_hashinfo *raw_hash;
1019 struct module *owner;
1023 struct list_head node;
1024 #ifdef SOCK_REFCNT_DEBUG
1027 #ifdef CONFIG_MEMCG_KMEM
1029 * cgroup specific init/deinit functions. Called once for all
1030 * protocols that implement it, from cgroups populate function.
1031 * This function has to setup any files the protocol want to
1032 * appear in the kmem cgroup filesystem.
1034 int (*init_cgroup)(struct mem_cgroup *memcg,
1035 struct cgroup_subsys *ss);
1036 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1037 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1042 * Bits in struct cg_proto.flags
1044 enum cg_proto_flags {
1045 /* Currently active and new sockets should be assigned to cgroups */
1047 /* It was ever activated; we must disarm static keys on destruction */
1048 MEMCG_SOCK_ACTIVATED,
1052 struct res_counter memory_allocated; /* Current allocated memory. */
1053 struct percpu_counter sockets_allocated; /* Current number of sockets. */
1054 int memory_pressure;
1056 unsigned long flags;
1058 * memcg field is used to find which memcg we belong directly
1059 * Each memcg struct can hold more than one cg_proto, so container_of
1062 * The elegant solution would be having an inverse function to
1063 * proto_cgroup in struct proto, but that means polluting the structure
1064 * for everybody, instead of just for memcg users.
1066 struct mem_cgroup *memcg;
1069 int proto_register(struct proto *prot, int alloc_slab);
1070 void proto_unregister(struct proto *prot);
1072 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1074 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1077 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1079 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1082 #ifdef SOCK_REFCNT_DEBUG
1083 static inline void sk_refcnt_debug_inc(struct sock *sk)
1085 atomic_inc(&sk->sk_prot->socks);
1088 static inline void sk_refcnt_debug_dec(struct sock *sk)
1090 atomic_dec(&sk->sk_prot->socks);
1091 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1092 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1095 static inline void sk_refcnt_debug_release(const struct sock *sk)
1097 if (atomic_read(&sk->sk_refcnt) != 1)
1098 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1099 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1101 #else /* SOCK_REFCNT_DEBUG */
1102 #define sk_refcnt_debug_inc(sk) do { } while (0)
1103 #define sk_refcnt_debug_dec(sk) do { } while (0)
1104 #define sk_refcnt_debug_release(sk) do { } while (0)
1105 #endif /* SOCK_REFCNT_DEBUG */
1107 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1108 extern struct static_key memcg_socket_limit_enabled;
1109 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1110 struct cg_proto *cg_proto)
1112 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1114 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1116 #define mem_cgroup_sockets_enabled 0
1117 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1118 struct cg_proto *cg_proto)
1124 static inline bool sk_stream_memory_free(const struct sock *sk)
1126 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1129 return sk->sk_prot->stream_memory_free ?
1130 sk->sk_prot->stream_memory_free(sk) : true;
1133 static inline bool sk_stream_is_writeable(const struct sock *sk)
1135 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1136 sk_stream_memory_free(sk);
1140 static inline bool sk_has_memory_pressure(const struct sock *sk)
1142 return sk->sk_prot->memory_pressure != NULL;
1145 static inline bool sk_under_memory_pressure(const struct sock *sk)
1147 if (!sk->sk_prot->memory_pressure)
1150 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1151 return !!sk->sk_cgrp->memory_pressure;
1153 return !!*sk->sk_prot->memory_pressure;
1156 static inline void sk_leave_memory_pressure(struct sock *sk)
1158 int *memory_pressure = sk->sk_prot->memory_pressure;
1160 if (!memory_pressure)
1163 if (*memory_pressure)
1164 *memory_pressure = 0;
1166 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1167 struct cg_proto *cg_proto = sk->sk_cgrp;
1168 struct proto *prot = sk->sk_prot;
1170 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1171 cg_proto->memory_pressure = 0;
1176 static inline void sk_enter_memory_pressure(struct sock *sk)
1178 if (!sk->sk_prot->enter_memory_pressure)
1181 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1182 struct cg_proto *cg_proto = sk->sk_cgrp;
1183 struct proto *prot = sk->sk_prot;
1185 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1186 cg_proto->memory_pressure = 1;
1189 sk->sk_prot->enter_memory_pressure(sk);
1192 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1194 long *prot = sk->sk_prot->sysctl_mem;
1195 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1196 prot = sk->sk_cgrp->sysctl_mem;
1200 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1204 struct res_counter *fail;
1207 ret = res_counter_charge_nofail(&prot->memory_allocated,
1208 amt << PAGE_SHIFT, &fail);
1210 *parent_status = OVER_LIMIT;
1213 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1216 res_counter_uncharge(&prot->memory_allocated, amt << PAGE_SHIFT);
1219 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1222 ret = res_counter_read_u64(&prot->memory_allocated, RES_USAGE);
1223 return ret >> PAGE_SHIFT;
1227 sk_memory_allocated(const struct sock *sk)
1229 struct proto *prot = sk->sk_prot;
1230 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1231 return memcg_memory_allocated_read(sk->sk_cgrp);
1233 return atomic_long_read(prot->memory_allocated);
1237 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1239 struct proto *prot = sk->sk_prot;
1241 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1242 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1243 /* update the root cgroup regardless */
1244 atomic_long_add_return(amt, prot->memory_allocated);
1245 return memcg_memory_allocated_read(sk->sk_cgrp);
1248 return atomic_long_add_return(amt, prot->memory_allocated);
1252 sk_memory_allocated_sub(struct sock *sk, int amt)
1254 struct proto *prot = sk->sk_prot;
1256 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1257 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1259 atomic_long_sub(amt, prot->memory_allocated);
1262 static inline void sk_sockets_allocated_dec(struct sock *sk)
1264 struct proto *prot = sk->sk_prot;
1266 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1267 struct cg_proto *cg_proto = sk->sk_cgrp;
1269 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1270 percpu_counter_dec(&cg_proto->sockets_allocated);
1273 percpu_counter_dec(prot->sockets_allocated);
1276 static inline void sk_sockets_allocated_inc(struct sock *sk)
1278 struct proto *prot = sk->sk_prot;
1280 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1281 struct cg_proto *cg_proto = sk->sk_cgrp;
1283 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1284 percpu_counter_inc(&cg_proto->sockets_allocated);
1287 percpu_counter_inc(prot->sockets_allocated);
1291 sk_sockets_allocated_read_positive(struct sock *sk)
1293 struct proto *prot = sk->sk_prot;
1295 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1296 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1298 return percpu_counter_read_positive(prot->sockets_allocated);
1302 proto_sockets_allocated_sum_positive(struct proto *prot)
1304 return percpu_counter_sum_positive(prot->sockets_allocated);
1308 proto_memory_allocated(struct proto *prot)
1310 return atomic_long_read(prot->memory_allocated);
1314 proto_memory_pressure(struct proto *prot)
1316 if (!prot->memory_pressure)
1318 return !!*prot->memory_pressure;
1322 #ifdef CONFIG_PROC_FS
1323 /* Called with local bh disabled */
1324 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1325 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1327 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1334 /* With per-bucket locks this operation is not-atomic, so that
1335 * this version is not worse.
1337 static inline void __sk_prot_rehash(struct sock *sk)
1339 sk->sk_prot->unhash(sk);
1340 sk->sk_prot->hash(sk);
1343 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1345 /* About 10 seconds */
1346 #define SOCK_DESTROY_TIME (10*HZ)
1348 /* Sockets 0-1023 can't be bound to unless you are superuser */
1349 #define PROT_SOCK 1024
1351 #define SHUTDOWN_MASK 3
1352 #define RCV_SHUTDOWN 1
1353 #define SEND_SHUTDOWN 2
1355 #define SOCK_SNDBUF_LOCK 1
1356 #define SOCK_RCVBUF_LOCK 2
1357 #define SOCK_BINDADDR_LOCK 4
1358 #define SOCK_BINDPORT_LOCK 8
1360 /* sock_iocb: used to kick off async processing of socket ios */
1362 struct list_head list;
1366 struct socket *sock;
1368 struct scm_cookie *scm;
1369 struct msghdr *msg, async_msg;
1370 struct kiocb *kiocb;
1373 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1375 return (struct sock_iocb *)iocb->private;
1378 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1383 struct socket_alloc {
1384 struct socket socket;
1385 struct inode vfs_inode;
1388 static inline struct socket *SOCKET_I(struct inode *inode)
1390 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1393 static inline struct inode *SOCK_INODE(struct socket *socket)
1395 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1399 * Functions for memory accounting
1401 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1402 void __sk_mem_reclaim(struct sock *sk);
1404 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1405 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1406 #define SK_MEM_SEND 0
1407 #define SK_MEM_RECV 1
1409 static inline int sk_mem_pages(int amt)
1411 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1414 static inline bool sk_has_account(struct sock *sk)
1416 /* return true if protocol supports memory accounting */
1417 return !!sk->sk_prot->memory_allocated;
1420 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1422 if (!sk_has_account(sk))
1424 return size <= sk->sk_forward_alloc ||
1425 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1429 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1431 if (!sk_has_account(sk))
1433 return size<= sk->sk_forward_alloc ||
1434 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1435 skb_pfmemalloc(skb);
1438 static inline void sk_mem_reclaim(struct sock *sk)
1440 if (!sk_has_account(sk))
1442 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1443 __sk_mem_reclaim(sk);
1446 static inline void sk_mem_reclaim_partial(struct sock *sk)
1448 if (!sk_has_account(sk))
1450 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1451 __sk_mem_reclaim(sk);
1454 static inline void sk_mem_charge(struct sock *sk, int size)
1456 if (!sk_has_account(sk))
1458 sk->sk_forward_alloc -= size;
1461 static inline void sk_mem_uncharge(struct sock *sk, int size)
1463 if (!sk_has_account(sk))
1465 sk->sk_forward_alloc += size;
1468 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1470 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1471 sk->sk_wmem_queued -= skb->truesize;
1472 sk_mem_uncharge(sk, skb->truesize);
1476 /* Used by processes to "lock" a socket state, so that
1477 * interrupts and bottom half handlers won't change it
1478 * from under us. It essentially blocks any incoming
1479 * packets, so that we won't get any new data or any
1480 * packets that change the state of the socket.
1482 * While locked, BH processing will add new packets to
1483 * the backlog queue. This queue is processed by the
1484 * owner of the socket lock right before it is released.
1486 * Since ~2.3.5 it is also exclusive sleep lock serializing
1487 * accesses from user process context.
1489 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1491 static inline void sock_release_ownership(struct sock *sk)
1493 sk->sk_lock.owned = 0;
1497 * Macro so as to not evaluate some arguments when
1498 * lockdep is not enabled.
1500 * Mark both the sk_lock and the sk_lock.slock as a
1501 * per-address-family lock class.
1503 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1505 sk->sk_lock.owned = 0; \
1506 init_waitqueue_head(&sk->sk_lock.wq); \
1507 spin_lock_init(&(sk)->sk_lock.slock); \
1508 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1509 sizeof((sk)->sk_lock)); \
1510 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1512 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1515 void lock_sock_nested(struct sock *sk, int subclass);
1517 static inline void lock_sock(struct sock *sk)
1519 lock_sock_nested(sk, 0);
1522 void release_sock(struct sock *sk);
1524 /* BH context may only use the following locking interface. */
1525 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1526 #define bh_lock_sock_nested(__sk) \
1527 spin_lock_nested(&((__sk)->sk_lock.slock), \
1528 SINGLE_DEPTH_NESTING)
1529 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1531 bool lock_sock_fast(struct sock *sk);
1533 * unlock_sock_fast - complement of lock_sock_fast
1537 * fast unlock socket for user context.
1538 * If slow mode is on, we call regular release_sock()
1540 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1545 spin_unlock_bh(&sk->sk_lock.slock);
1549 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1550 struct proto *prot);
1551 void sk_free(struct sock *sk);
1552 void sk_release_kernel(struct sock *sk);
1553 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1555 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1557 void sock_wfree(struct sk_buff *skb);
1558 void skb_orphan_partial(struct sk_buff *skb);
1559 void sock_rfree(struct sk_buff *skb);
1560 void sock_edemux(struct sk_buff *skb);
1562 int sock_setsockopt(struct socket *sock, int level, int op,
1563 char __user *optval, unsigned int optlen);
1565 int sock_getsockopt(struct socket *sock, int level, int op,
1566 char __user *optval, int __user *optlen);
1567 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1568 int noblock, int *errcode);
1569 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1570 unsigned long data_len, int noblock,
1571 int *errcode, int max_page_order);
1572 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1573 void sock_kfree_s(struct sock *sk, void *mem, int size);
1574 void sk_send_sigurg(struct sock *sk);
1577 * Functions to fill in entries in struct proto_ops when a protocol
1578 * does not implement a particular function.
1580 int sock_no_bind(struct socket *, struct sockaddr *, int);
1581 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1582 int sock_no_socketpair(struct socket *, struct socket *);
1583 int sock_no_accept(struct socket *, struct socket *, int);
1584 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1585 unsigned int sock_no_poll(struct file *, struct socket *,
1586 struct poll_table_struct *);
1587 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1588 int sock_no_listen(struct socket *, int);
1589 int sock_no_shutdown(struct socket *, int);
1590 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1591 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1592 int sock_no_sendmsg(struct kiocb *, struct socket *, struct msghdr *, size_t);
1593 int sock_no_recvmsg(struct kiocb *, struct socket *, struct msghdr *, size_t,
1595 int sock_no_mmap(struct file *file, struct socket *sock,
1596 struct vm_area_struct *vma);
1597 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1598 size_t size, int flags);
1601 * Functions to fill in entries in struct proto_ops when a protocol
1602 * uses the inet style.
1604 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1605 char __user *optval, int __user *optlen);
1606 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1607 struct msghdr *msg, size_t size, int flags);
1608 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1609 char __user *optval, unsigned int optlen);
1610 int compat_sock_common_getsockopt(struct socket *sock, int level,
1611 int optname, char __user *optval, int __user *optlen);
1612 int compat_sock_common_setsockopt(struct socket *sock, int level,
1613 int optname, char __user *optval, unsigned int optlen);
1615 void sk_common_release(struct sock *sk);
1618 * Default socket callbacks and setup code
1621 /* Initialise core socket variables */
1622 void sock_init_data(struct socket *sock, struct sock *sk);
1624 void sk_filter_release_rcu(struct rcu_head *rcu);
1627 * sk_filter_release - release a socket filter
1628 * @fp: filter to remove
1630 * Remove a filter from a socket and release its resources.
1633 static inline void sk_filter_release(struct sk_filter *fp)
1635 if (atomic_dec_and_test(&fp->refcnt))
1636 call_rcu(&fp->rcu, sk_filter_release_rcu);
1639 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1641 atomic_sub(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1642 sk_filter_release(fp);
1645 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1647 atomic_inc(&fp->refcnt);
1648 atomic_add(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1652 * Socket reference counting postulates.
1654 * * Each user of socket SHOULD hold a reference count.
1655 * * Each access point to socket (an hash table bucket, reference from a list,
1656 * running timer, skb in flight MUST hold a reference count.
1657 * * When reference count hits 0, it means it will never increase back.
1658 * * When reference count hits 0, it means that no references from
1659 * outside exist to this socket and current process on current CPU
1660 * is last user and may/should destroy this socket.
1661 * * sk_free is called from any context: process, BH, IRQ. When
1662 * it is called, socket has no references from outside -> sk_free
1663 * may release descendant resources allocated by the socket, but
1664 * to the time when it is called, socket is NOT referenced by any
1665 * hash tables, lists etc.
1666 * * Packets, delivered from outside (from network or from another process)
1667 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1668 * when they sit in queue. Otherwise, packets will leak to hole, when
1669 * socket is looked up by one cpu and unhasing is made by another CPU.
1670 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1671 * (leak to backlog). Packet socket does all the processing inside
1672 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1673 * use separate SMP lock, so that they are prone too.
1676 /* Ungrab socket and destroy it, if it was the last reference. */
1677 static inline void sock_put(struct sock *sk)
1679 if (atomic_dec_and_test(&sk->sk_refcnt))
1682 /* Generic version of sock_put(), dealing with all sockets
1683 * (TCP_TIMEWAIT, ESTABLISHED...)
1685 void sock_gen_put(struct sock *sk);
1687 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1689 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1691 sk->sk_tx_queue_mapping = tx_queue;
1694 static inline void sk_tx_queue_clear(struct sock *sk)
1696 sk->sk_tx_queue_mapping = -1;
1699 static inline int sk_tx_queue_get(const struct sock *sk)
1701 return sk ? sk->sk_tx_queue_mapping : -1;
1704 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1706 sk_tx_queue_clear(sk);
1707 sk->sk_socket = sock;
1710 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1712 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1713 return &rcu_dereference_raw(sk->sk_wq)->wait;
1715 /* Detach socket from process context.
1716 * Announce socket dead, detach it from wait queue and inode.
1717 * Note that parent inode held reference count on this struct sock,
1718 * we do not release it in this function, because protocol
1719 * probably wants some additional cleanups or even continuing
1720 * to work with this socket (TCP).
1722 static inline void sock_orphan(struct sock *sk)
1724 write_lock_bh(&sk->sk_callback_lock);
1725 sock_set_flag(sk, SOCK_DEAD);
1726 sk_set_socket(sk, NULL);
1728 write_unlock_bh(&sk->sk_callback_lock);
1731 static inline void sock_graft(struct sock *sk, struct socket *parent)
1733 write_lock_bh(&sk->sk_callback_lock);
1734 sk->sk_wq = parent->wq;
1736 sk_set_socket(sk, parent);
1737 security_sock_graft(sk, parent);
1738 write_unlock_bh(&sk->sk_callback_lock);
1741 kuid_t sock_i_uid(struct sock *sk);
1742 unsigned long sock_i_ino(struct sock *sk);
1744 static inline struct dst_entry *
1745 __sk_dst_get(struct sock *sk)
1747 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1748 lockdep_is_held(&sk->sk_lock.slock));
1751 static inline struct dst_entry *
1752 sk_dst_get(struct sock *sk)
1754 struct dst_entry *dst;
1757 dst = rcu_dereference(sk->sk_dst_cache);
1764 static inline void dst_negative_advice(struct sock *sk)
1766 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1768 if (dst && dst->ops->negative_advice) {
1769 ndst = dst->ops->negative_advice(dst);
1772 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1773 sk_tx_queue_clear(sk);
1779 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1781 struct dst_entry *old_dst;
1783 sk_tx_queue_clear(sk);
1785 * This can be called while sk is owned by the caller only,
1786 * with no state that can be checked in a rcu_dereference_check() cond
1788 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1789 rcu_assign_pointer(sk->sk_dst_cache, dst);
1790 dst_release(old_dst);
1794 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1796 spin_lock(&sk->sk_dst_lock);
1797 __sk_dst_set(sk, dst);
1798 spin_unlock(&sk->sk_dst_lock);
1802 __sk_dst_reset(struct sock *sk)
1804 __sk_dst_set(sk, NULL);
1808 sk_dst_reset(struct sock *sk)
1810 spin_lock(&sk->sk_dst_lock);
1812 spin_unlock(&sk->sk_dst_lock);
1815 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1817 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1819 static inline bool sk_can_gso(const struct sock *sk)
1821 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1824 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1826 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1828 sk->sk_route_nocaps |= flags;
1829 sk->sk_route_caps &= ~flags;
1832 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1833 char __user *from, char *to,
1834 int copy, int offset)
1836 if (skb->ip_summed == CHECKSUM_NONE) {
1838 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1841 skb->csum = csum_block_add(skb->csum, csum, offset);
1842 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1843 if (!access_ok(VERIFY_READ, from, copy) ||
1844 __copy_from_user_nocache(to, from, copy))
1846 } else if (copy_from_user(to, from, copy))
1852 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1853 char __user *from, int copy)
1855 int err, offset = skb->len;
1857 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1860 __skb_trim(skb, offset);
1865 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1866 struct sk_buff *skb,
1872 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1878 skb->data_len += copy;
1879 skb->truesize += copy;
1880 sk->sk_wmem_queued += copy;
1881 sk_mem_charge(sk, copy);
1885 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1886 struct sk_buff *skb, struct page *page,
1889 if (skb->ip_summed == CHECKSUM_NONE) {
1891 __wsum csum = csum_and_copy_from_user(from,
1892 page_address(page) + off,
1896 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1897 } else if (copy_from_user(page_address(page) + off, from, copy))
1901 skb->data_len += copy;
1902 skb->truesize += copy;
1903 sk->sk_wmem_queued += copy;
1904 sk_mem_charge(sk, copy);
1909 * sk_wmem_alloc_get - returns write allocations
1912 * Returns sk_wmem_alloc minus initial offset of one
1914 static inline int sk_wmem_alloc_get(const struct sock *sk)
1916 return atomic_read(&sk->sk_wmem_alloc) - 1;
1920 * sk_rmem_alloc_get - returns read allocations
1923 * Returns sk_rmem_alloc
1925 static inline int sk_rmem_alloc_get(const struct sock *sk)
1927 return atomic_read(&sk->sk_rmem_alloc);
1931 * sk_has_allocations - check if allocations are outstanding
1934 * Returns true if socket has write or read allocations
1936 static inline bool sk_has_allocations(const struct sock *sk)
1938 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1942 * wq_has_sleeper - check if there are any waiting processes
1943 * @wq: struct socket_wq
1945 * Returns true if socket_wq has waiting processes
1947 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1948 * barrier call. They were added due to the race found within the tcp code.
1950 * Consider following tcp code paths:
1954 * sys_select receive packet
1956 * __add_wait_queue update tp->rcv_nxt
1958 * tp->rcv_nxt check sock_def_readable
1960 * schedule rcu_read_lock();
1961 * wq = rcu_dereference(sk->sk_wq);
1962 * if (wq && waitqueue_active(&wq->wait))
1963 * wake_up_interruptible(&wq->wait)
1967 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1968 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1969 * could then endup calling schedule and sleep forever if there are no more
1970 * data on the socket.
1973 static inline bool wq_has_sleeper(struct socket_wq *wq)
1975 /* We need to be sure we are in sync with the
1976 * add_wait_queue modifications to the wait queue.
1978 * This memory barrier is paired in the sock_poll_wait.
1981 return wq && waitqueue_active(&wq->wait);
1985 * sock_poll_wait - place memory barrier behind the poll_wait call.
1987 * @wait_address: socket wait queue
1990 * See the comments in the wq_has_sleeper function.
1992 static inline void sock_poll_wait(struct file *filp,
1993 wait_queue_head_t *wait_address, poll_table *p)
1995 if (!poll_does_not_wait(p) && wait_address) {
1996 poll_wait(filp, wait_address, p);
1997 /* We need to be sure we are in sync with the
1998 * socket flags modification.
2000 * This memory barrier is paired in the wq_has_sleeper.
2007 * Queue a received datagram if it will fit. Stream and sequenced
2008 * protocols can't normally use this as they need to fit buffers in
2009 * and play with them.
2011 * Inlined as it's very short and called for pretty much every
2012 * packet ever received.
2015 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2019 skb->destructor = sock_wfree;
2021 * We used to take a refcount on sk, but following operation
2022 * is enough to guarantee sk_free() wont free this sock until
2023 * all in-flight packets are completed
2025 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
2028 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2032 skb->destructor = sock_rfree;
2033 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2034 sk_mem_charge(sk, skb->truesize);
2037 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2038 unsigned long expires);
2040 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2042 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2044 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2047 * Recover an error report and clear atomically
2050 static inline int sock_error(struct sock *sk)
2053 if (likely(!sk->sk_err))
2055 err = xchg(&sk->sk_err, 0);
2059 static inline unsigned long sock_wspace(struct sock *sk)
2063 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2064 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2071 static inline void sk_wake_async(struct sock *sk, int how, int band)
2073 if (sock_flag(sk, SOCK_FASYNC))
2074 sock_wake_async(sk->sk_socket, how, band);
2077 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2078 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2079 * Note: for send buffers, TCP works better if we can build two skbs at
2082 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2084 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2085 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2087 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2089 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2090 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2091 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2095 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2098 * sk_page_frag - return an appropriate page_frag
2101 * If socket allocation mode allows current thread to sleep, it means its
2102 * safe to use the per task page_frag instead of the per socket one.
2104 static inline struct page_frag *sk_page_frag(struct sock *sk)
2106 if (sk->sk_allocation & __GFP_WAIT)
2107 return ¤t->task_frag;
2109 return &sk->sk_frag;
2112 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2115 * Default write policy as shown to user space via poll/select/SIGIO
2117 static inline bool sock_writeable(const struct sock *sk)
2119 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2122 static inline gfp_t gfp_any(void)
2124 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2127 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2129 return noblock ? 0 : sk->sk_rcvtimeo;
2132 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2134 return noblock ? 0 : sk->sk_sndtimeo;
2137 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2139 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2142 /* Alas, with timeout socket operations are not restartable.
2143 * Compare this to poll().
2145 static inline int sock_intr_errno(long timeo)
2147 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2150 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2151 struct sk_buff *skb);
2152 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2153 struct sk_buff *skb);
2156 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2158 ktime_t kt = skb->tstamp;
2159 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2162 * generate control messages if
2163 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2164 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2165 * - software time stamp available and wanted
2166 * (SOCK_TIMESTAMPING_SOFTWARE)
2167 * - hardware time stamps available and wanted
2168 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2169 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2171 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2172 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2173 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2174 (hwtstamps->hwtstamp.tv64 &&
2175 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2176 (hwtstamps->syststamp.tv64 &&
2177 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2178 __sock_recv_timestamp(msg, sk, skb);
2182 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2183 __sock_recv_wifi_status(msg, sk, skb);
2186 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2187 struct sk_buff *skb);
2189 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2190 struct sk_buff *skb)
2192 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2193 (1UL << SOCK_RCVTSTAMP) | \
2194 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2195 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2196 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2198 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2199 __sock_recv_ts_and_drops(msg, sk, skb);
2201 sk->sk_stamp = skb->tstamp;
2205 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2206 * @sk: socket sending this packet
2207 * @tx_flags: filled with instructions for time stamping
2209 * Currently only depends on SOCK_TIMESTAMPING* flags.
2211 void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2214 * sk_eat_skb - Release a skb if it is no longer needed
2215 * @sk: socket to eat this skb from
2216 * @skb: socket buffer to eat
2217 * @copied_early: flag indicating whether DMA operations copied this data early
2219 * This routine must be called with interrupts disabled or with the socket
2220 * locked so that the sk_buff queue operation is ok.
2222 #ifdef CONFIG_NET_DMA
2223 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2225 __skb_unlink(skb, &sk->sk_receive_queue);
2229 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2232 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2234 __skb_unlink(skb, &sk->sk_receive_queue);
2240 struct net *sock_net(const struct sock *sk)
2242 return read_pnet(&sk->sk_net);
2246 void sock_net_set(struct sock *sk, struct net *net)
2248 write_pnet(&sk->sk_net, net);
2252 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2253 * They should not hold a reference to a namespace in order to allow
2255 * Sockets after sk_change_net should be released using sk_release_kernel
2257 static inline void sk_change_net(struct sock *sk, struct net *net)
2259 put_net(sock_net(sk));
2260 sock_net_set(sk, hold_net(net));
2263 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2266 struct sock *sk = skb->sk;
2268 skb->destructor = NULL;
2275 void sock_enable_timestamp(struct sock *sk, int flag);
2276 int sock_get_timestamp(struct sock *, struct timeval __user *);
2277 int sock_get_timestampns(struct sock *, struct timespec __user *);
2278 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2282 * Enable debug/info messages
2284 extern int net_msg_warn;
2285 #define NETDEBUG(fmt, args...) \
2286 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2288 #define LIMIT_NETDEBUG(fmt, args...) \
2289 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2291 extern __u32 sysctl_wmem_max;
2292 extern __u32 sysctl_rmem_max;
2294 extern int sysctl_optmem_max;
2296 extern __u32 sysctl_wmem_default;
2297 extern __u32 sysctl_rmem_default;
2299 #endif /* _SOCK_H */