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/lockdep.h>
50 #include <linux/netdevice.h>
51 #include <linux/skbuff.h> /* struct sk_buff */
53 #include <linux/security.h>
54 #include <linux/slab.h>
55 #include <linux/uaccess.h>
56 #include <linux/memcontrol.h>
57 #include <linux/res_counter.h>
59 #include <linux/filter.h>
60 #include <linux/rculist_nulls.h>
61 #include <linux/poll.h>
63 #include <linux/atomic.h>
65 #include <net/checksum.h>
70 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss);
71 void mem_cgroup_sockets_destroy(struct cgroup *cgrp);
74 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss)
79 void mem_cgroup_sockets_destroy(struct cgroup *cgrp)
84 * This structure really needs to be cleaned up.
85 * Most of it is for TCP, and not used by any of
86 * the other protocols.
89 /* Define this to get the SOCK_DBG debugging facility. */
90 #define SOCK_DEBUGGING
92 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
93 printk(KERN_DEBUG msg); } while (0)
95 /* Validate arguments and do nothing */
96 static inline __printf(2, 3)
97 void SOCK_DEBUG(struct sock *sk, const char *msg, ...)
102 /* This is the per-socket lock. The spinlock provides a synchronization
103 * between user contexts and software interrupt processing, whereas the
104 * mini-semaphore synchronizes multiple users amongst themselves.
109 wait_queue_head_t wq;
111 * We express the mutex-alike socket_lock semantics
112 * to the lock validator by explicitly managing
113 * the slock as a lock variant (in addition to
116 #ifdef CONFIG_DEBUG_LOCK_ALLOC
117 struct lockdep_map dep_map;
126 * struct sock_common - minimal network layer representation of sockets
127 * @skc_daddr: Foreign IPv4 addr
128 * @skc_rcv_saddr: Bound local IPv4 addr
129 * @skc_hash: hash value used with various protocol lookup tables
130 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
131 * @skc_family: network address family
132 * @skc_state: Connection state
133 * @skc_reuse: %SO_REUSEADDR setting
134 * @skc_bound_dev_if: bound device index if != 0
135 * @skc_bind_node: bind hash linkage for various protocol lookup tables
136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137 * @skc_prot: protocol handlers inside a network family
138 * @skc_net: reference to the network namespace of this socket
139 * @skc_node: main hash linkage for various protocol lookup tables
140 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
141 * @skc_tx_queue_mapping: tx queue number for this connection
142 * @skc_refcnt: reference count
144 * This is the minimal network layer representation of sockets, the header
145 * for struct sock and struct inet_timewait_sock.
148 /* skc_daddr and skc_rcv_saddr must be grouped :
149 * cf INET_MATCH() and INET_TW_MATCH()
152 __be32 skc_rcv_saddr;
155 unsigned int skc_hash;
156 __u16 skc_u16hashes[2];
158 unsigned short skc_family;
159 volatile unsigned char skc_state;
160 unsigned char skc_reuse;
161 int skc_bound_dev_if;
163 struct hlist_node skc_bind_node;
164 struct hlist_nulls_node skc_portaddr_node;
166 struct proto *skc_prot;
171 * fields between dontcopy_begin/dontcopy_end
172 * are not copied in sock_copy()
175 int skc_dontcopy_begin[0];
178 struct hlist_node skc_node;
179 struct hlist_nulls_node skc_nulls_node;
181 int skc_tx_queue_mapping;
184 int skc_dontcopy_end[0];
190 * struct sock - network layer representation of sockets
191 * @__sk_common: shared layout with inet_timewait_sock
192 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
193 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
194 * @sk_lock: synchronizer
195 * @sk_rcvbuf: size of receive buffer in bytes
196 * @sk_wq: sock wait queue and async head
197 * @sk_dst_cache: destination cache
198 * @sk_dst_lock: destination cache lock
199 * @sk_policy: flow policy
200 * @sk_receive_queue: incoming packets
201 * @sk_wmem_alloc: transmit queue bytes committed
202 * @sk_write_queue: Packet sending queue
203 * @sk_async_wait_queue: DMA copied packets
204 * @sk_omem_alloc: "o" is "option" or "other"
205 * @sk_wmem_queued: persistent queue size
206 * @sk_forward_alloc: space allocated forward
207 * @sk_allocation: allocation mode
208 * @sk_sndbuf: size of send buffer in bytes
209 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
210 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
211 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
212 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
213 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
214 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
215 * @sk_gso_max_size: Maximum GSO segment size to build
216 * @sk_lingertime: %SO_LINGER l_linger setting
217 * @sk_backlog: always used with the per-socket spinlock held
218 * @sk_callback_lock: used with the callbacks in the end of this struct
219 * @sk_error_queue: rarely used
220 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
221 * IPV6_ADDRFORM for instance)
222 * @sk_err: last error
223 * @sk_err_soft: errors that don't cause failure but are the cause of a
224 * persistent failure not just 'timed out'
225 * @sk_drops: raw/udp drops counter
226 * @sk_ack_backlog: current listen backlog
227 * @sk_max_ack_backlog: listen backlog set in listen()
228 * @sk_priority: %SO_PRIORITY setting
229 * @sk_type: socket type (%SOCK_STREAM, etc)
230 * @sk_protocol: which protocol this socket belongs in this network family
231 * @sk_peer_pid: &struct pid for this socket's peer
232 * @sk_peer_cred: %SO_PEERCRED setting
233 * @sk_rcvlowat: %SO_RCVLOWAT setting
234 * @sk_rcvtimeo: %SO_RCVTIMEO setting
235 * @sk_sndtimeo: %SO_SNDTIMEO setting
236 * @sk_rxhash: flow hash received from netif layer
237 * @sk_filter: socket filtering instructions
238 * @sk_protinfo: private area, net family specific, when not using slab
239 * @sk_timer: sock cleanup timer
240 * @sk_stamp: time stamp of last packet received
241 * @sk_socket: Identd and reporting IO signals
242 * @sk_user_data: RPC layer private data
243 * @sk_sndmsg_page: cached page for sendmsg
244 * @sk_sndmsg_off: cached offset for sendmsg
245 * @sk_send_head: front of stuff to transmit
246 * @sk_security: used by security modules
247 * @sk_mark: generic packet mark
248 * @sk_classid: this socket's cgroup classid
249 * @sk_cgrp: this socket's cgroup-specific proto data
250 * @sk_write_pending: a write to stream socket waits to start
251 * @sk_state_change: callback to indicate change in the state of the sock
252 * @sk_data_ready: callback to indicate there is data to be processed
253 * @sk_write_space: callback to indicate there is bf sending space available
254 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
255 * @sk_backlog_rcv: callback to process the backlog
256 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
260 * Now struct inet_timewait_sock also uses sock_common, so please just
261 * don't add nothing before this first member (__sk_common) --acme
263 struct sock_common __sk_common;
264 #define sk_node __sk_common.skc_node
265 #define sk_nulls_node __sk_common.skc_nulls_node
266 #define sk_refcnt __sk_common.skc_refcnt
267 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
269 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
270 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
271 #define sk_hash __sk_common.skc_hash
272 #define sk_family __sk_common.skc_family
273 #define sk_state __sk_common.skc_state
274 #define sk_reuse __sk_common.skc_reuse
275 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
276 #define sk_bind_node __sk_common.skc_bind_node
277 #define sk_prot __sk_common.skc_prot
278 #define sk_net __sk_common.skc_net
279 socket_lock_t sk_lock;
280 struct sk_buff_head sk_receive_queue;
282 * The backlog queue is special, it is always used with
283 * the per-socket spinlock held and requires low latency
284 * access. Therefore we special case it's implementation.
285 * Note : rmem_alloc is in this structure to fill a hole
286 * on 64bit arches, not because its logically part of
292 struct sk_buff *head;
293 struct sk_buff *tail;
295 #define sk_rmem_alloc sk_backlog.rmem_alloc
296 int sk_forward_alloc;
303 struct sk_filter __rcu *sk_filter;
304 struct socket_wq __rcu *sk_wq;
306 #ifdef CONFIG_NET_DMA
307 struct sk_buff_head sk_async_wait_queue;
311 struct xfrm_policy *sk_policy[2];
313 unsigned long sk_flags;
314 struct dst_entry *sk_dst_cache;
315 spinlock_t sk_dst_lock;
316 atomic_t sk_wmem_alloc;
317 atomic_t sk_omem_alloc;
319 struct sk_buff_head sk_write_queue;
320 kmemcheck_bitfield_begin(flags);
321 unsigned int sk_shutdown : 2,
326 kmemcheck_bitfield_end(flags);
329 netdev_features_t sk_route_caps;
330 netdev_features_t sk_route_nocaps;
332 unsigned int sk_gso_max_size;
334 unsigned long sk_lingertime;
335 struct sk_buff_head sk_error_queue;
336 struct proto *sk_prot_creator;
337 rwlock_t sk_callback_lock;
340 unsigned short sk_ack_backlog;
341 unsigned short sk_max_ack_backlog;
343 #ifdef CONFIG_CGROUPS
344 __u32 sk_cgrp_prioidx;
346 struct pid *sk_peer_pid;
347 const struct cred *sk_peer_cred;
351 struct timer_list sk_timer;
353 struct socket *sk_socket;
355 struct page *sk_sndmsg_page;
356 struct sk_buff *sk_send_head;
358 int sk_write_pending;
359 #ifdef CONFIG_SECURITY
364 struct cg_proto *sk_cgrp;
365 void (*sk_state_change)(struct sock *sk);
366 void (*sk_data_ready)(struct sock *sk, int bytes);
367 void (*sk_write_space)(struct sock *sk);
368 void (*sk_error_report)(struct sock *sk);
369 int (*sk_backlog_rcv)(struct sock *sk,
370 struct sk_buff *skb);
371 void (*sk_destruct)(struct sock *sk);
375 * Hashed lists helper routines
377 static inline struct sock *sk_entry(const struct hlist_node *node)
379 return hlist_entry(node, struct sock, sk_node);
382 static inline struct sock *__sk_head(const struct hlist_head *head)
384 return hlist_entry(head->first, struct sock, sk_node);
387 static inline struct sock *sk_head(const struct hlist_head *head)
389 return hlist_empty(head) ? NULL : __sk_head(head);
392 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
394 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
397 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
399 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
402 static inline struct sock *sk_next(const struct sock *sk)
404 return sk->sk_node.next ?
405 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
408 static inline struct sock *sk_nulls_next(const struct sock *sk)
410 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
411 hlist_nulls_entry(sk->sk_nulls_node.next,
412 struct sock, sk_nulls_node) :
416 static inline int sk_unhashed(const struct sock *sk)
418 return hlist_unhashed(&sk->sk_node);
421 static inline int sk_hashed(const struct sock *sk)
423 return !sk_unhashed(sk);
426 static __inline__ void sk_node_init(struct hlist_node *node)
431 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
436 static __inline__ void __sk_del_node(struct sock *sk)
438 __hlist_del(&sk->sk_node);
441 /* NB: equivalent to hlist_del_init_rcu */
442 static __inline__ int __sk_del_node_init(struct sock *sk)
446 sk_node_init(&sk->sk_node);
452 /* Grab socket reference count. This operation is valid only
453 when sk is ALREADY grabbed f.e. it is found in hash table
454 or a list and the lookup is made under lock preventing hash table
458 static inline void sock_hold(struct sock *sk)
460 atomic_inc(&sk->sk_refcnt);
463 /* Ungrab socket in the context, which assumes that socket refcnt
464 cannot hit zero, f.e. it is true in context of any socketcall.
466 static inline void __sock_put(struct sock *sk)
468 atomic_dec(&sk->sk_refcnt);
471 static __inline__ int sk_del_node_init(struct sock *sk)
473 int rc = __sk_del_node_init(sk);
476 /* paranoid for a while -acme */
477 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
482 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
484 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
487 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
493 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
495 int rc = __sk_nulls_del_node_init_rcu(sk);
498 /* paranoid for a while -acme */
499 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
505 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
507 hlist_add_head(&sk->sk_node, list);
510 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
513 __sk_add_node(sk, list);
516 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
519 hlist_add_head_rcu(&sk->sk_node, list);
522 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
524 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
527 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
530 __sk_nulls_add_node_rcu(sk, list);
533 static __inline__ void __sk_del_bind_node(struct sock *sk)
535 __hlist_del(&sk->sk_bind_node);
538 static __inline__ void sk_add_bind_node(struct sock *sk,
539 struct hlist_head *list)
541 hlist_add_head(&sk->sk_bind_node, list);
544 #define sk_for_each(__sk, node, list) \
545 hlist_for_each_entry(__sk, node, list, sk_node)
546 #define sk_for_each_rcu(__sk, node, list) \
547 hlist_for_each_entry_rcu(__sk, node, list, sk_node)
548 #define sk_nulls_for_each(__sk, node, list) \
549 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
550 #define sk_nulls_for_each_rcu(__sk, node, list) \
551 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
552 #define sk_for_each_from(__sk, node) \
553 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
554 hlist_for_each_entry_from(__sk, node, sk_node)
555 #define sk_nulls_for_each_from(__sk, node) \
556 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
557 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
558 #define sk_for_each_safe(__sk, node, tmp, list) \
559 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
560 #define sk_for_each_bound(__sk, node, list) \
561 hlist_for_each_entry(__sk, node, list, sk_bind_node)
574 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
575 SOCK_DBG, /* %SO_DEBUG setting */
576 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
577 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
578 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
579 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
580 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
581 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
582 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
583 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
584 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
585 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
586 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
587 SOCK_FASYNC, /* fasync() active */
589 SOCK_ZEROCOPY, /* buffers from userspace */
590 SOCK_WIFI_STATUS, /* push wifi status to userspace */
593 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
595 nsk->sk_flags = osk->sk_flags;
598 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
600 __set_bit(flag, &sk->sk_flags);
603 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
605 __clear_bit(flag, &sk->sk_flags);
608 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
610 return test_bit(flag, &sk->sk_flags);
613 static inline void sk_acceptq_removed(struct sock *sk)
615 sk->sk_ack_backlog--;
618 static inline void sk_acceptq_added(struct sock *sk)
620 sk->sk_ack_backlog++;
623 static inline int sk_acceptq_is_full(struct sock *sk)
625 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
629 * Compute minimal free write space needed to queue new packets.
631 static inline int sk_stream_min_wspace(struct sock *sk)
633 return sk->sk_wmem_queued >> 1;
636 static inline int sk_stream_wspace(struct sock *sk)
638 return sk->sk_sndbuf - sk->sk_wmem_queued;
641 extern void sk_stream_write_space(struct sock *sk);
643 static inline int sk_stream_memory_free(struct sock *sk)
645 return sk->sk_wmem_queued < sk->sk_sndbuf;
648 /* OOB backlog add */
649 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
651 /* dont let skb dst not refcounted, we are going to leave rcu lock */
654 if (!sk->sk_backlog.tail)
655 sk->sk_backlog.head = skb;
657 sk->sk_backlog.tail->next = skb;
659 sk->sk_backlog.tail = skb;
664 * Take into account size of receive queue and backlog queue
665 * Do not take into account this skb truesize,
666 * to allow even a single big packet to come.
668 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb)
670 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
672 return qsize > sk->sk_rcvbuf;
675 /* The per-socket spinlock must be held here. */
676 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
678 if (sk_rcvqueues_full(sk, skb))
681 __sk_add_backlog(sk, skb);
682 sk->sk_backlog.len += skb->truesize;
686 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
688 return sk->sk_backlog_rcv(sk, skb);
691 static inline void sock_rps_record_flow(const struct sock *sk)
694 struct rps_sock_flow_table *sock_flow_table;
697 sock_flow_table = rcu_dereference(rps_sock_flow_table);
698 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
703 static inline void sock_rps_reset_flow(const struct sock *sk)
706 struct rps_sock_flow_table *sock_flow_table;
709 sock_flow_table = rcu_dereference(rps_sock_flow_table);
710 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
715 static inline void sock_rps_save_rxhash(struct sock *sk,
716 const struct sk_buff *skb)
719 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
720 sock_rps_reset_flow(sk);
721 sk->sk_rxhash = skb->rxhash;
726 static inline void sock_rps_reset_rxhash(struct sock *sk)
729 sock_rps_reset_flow(sk);
734 #define sk_wait_event(__sk, __timeo, __condition) \
736 release_sock(__sk); \
737 __rc = __condition; \
739 *(__timeo) = schedule_timeout(*(__timeo)); \
742 __rc = __condition; \
746 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
747 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
748 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
749 extern int sk_stream_error(struct sock *sk, int flags, int err);
750 extern void sk_stream_kill_queues(struct sock *sk);
752 extern int sk_wait_data(struct sock *sk, long *timeo);
754 struct request_sock_ops;
755 struct timewait_sock_ops;
756 struct inet_hashinfo;
760 /* Networking protocol blocks we attach to sockets.
761 * socket layer -> transport layer interface
762 * transport -> network interface is defined by struct inet_proto
765 void (*close)(struct sock *sk,
767 int (*connect)(struct sock *sk,
768 struct sockaddr *uaddr,
770 int (*disconnect)(struct sock *sk, int flags);
772 struct sock * (*accept) (struct sock *sk, int flags, int *err);
774 int (*ioctl)(struct sock *sk, int cmd,
776 int (*init)(struct sock *sk);
777 void (*destroy)(struct sock *sk);
778 void (*shutdown)(struct sock *sk, int how);
779 int (*setsockopt)(struct sock *sk, int level,
780 int optname, char __user *optval,
781 unsigned int optlen);
782 int (*getsockopt)(struct sock *sk, int level,
783 int optname, char __user *optval,
786 int (*compat_setsockopt)(struct sock *sk,
788 int optname, char __user *optval,
789 unsigned int optlen);
790 int (*compat_getsockopt)(struct sock *sk,
792 int optname, char __user *optval,
794 int (*compat_ioctl)(struct sock *sk,
795 unsigned int cmd, unsigned long arg);
797 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
798 struct msghdr *msg, size_t len);
799 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
801 size_t len, int noblock, int flags,
803 int (*sendpage)(struct sock *sk, struct page *page,
804 int offset, size_t size, int flags);
805 int (*bind)(struct sock *sk,
806 struct sockaddr *uaddr, int addr_len);
808 int (*backlog_rcv) (struct sock *sk,
809 struct sk_buff *skb);
811 /* Keeping track of sk's, looking them up, and port selection methods. */
812 void (*hash)(struct sock *sk);
813 void (*unhash)(struct sock *sk);
814 void (*rehash)(struct sock *sk);
815 int (*get_port)(struct sock *sk, unsigned short snum);
816 void (*clear_sk)(struct sock *sk, int size);
818 /* Keeping track of sockets in use */
819 #ifdef CONFIG_PROC_FS
820 unsigned int inuse_idx;
823 /* Memory pressure */
824 void (*enter_memory_pressure)(struct sock *sk);
825 atomic_long_t *memory_allocated; /* Current allocated memory. */
826 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
828 * Pressure flag: try to collapse.
829 * Technical note: it is used by multiple contexts non atomically.
830 * All the __sk_mem_schedule() is of this nature: accounting
831 * is strict, actions are advisory and have some latency.
833 int *memory_pressure;
840 struct kmem_cache *slab;
841 unsigned int obj_size;
844 struct percpu_counter *orphan_count;
846 struct request_sock_ops *rsk_prot;
847 struct timewait_sock_ops *twsk_prot;
850 struct inet_hashinfo *hashinfo;
851 struct udp_table *udp_table;
852 struct raw_hashinfo *raw_hash;
855 struct module *owner;
859 struct list_head node;
860 #ifdef SOCK_REFCNT_DEBUG
863 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
865 * cgroup specific init/deinit functions. Called once for all
866 * protocols that implement it, from cgroups populate function.
867 * This function has to setup any files the protocol want to
868 * appear in the kmem cgroup filesystem.
870 int (*init_cgroup)(struct cgroup *cgrp,
871 struct cgroup_subsys *ss);
872 void (*destroy_cgroup)(struct cgroup *cgrp);
873 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
878 void (*enter_memory_pressure)(struct sock *sk);
879 struct res_counter *memory_allocated; /* Current allocated memory. */
880 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
881 int *memory_pressure;
884 * memcg field is used to find which memcg we belong directly
885 * Each memcg struct can hold more than one cg_proto, so container_of
888 * The elegant solution would be having an inverse function to
889 * proto_cgroup in struct proto, but that means polluting the structure
890 * for everybody, instead of just for memcg users.
892 struct mem_cgroup *memcg;
895 extern int proto_register(struct proto *prot, int alloc_slab);
896 extern void proto_unregister(struct proto *prot);
898 #ifdef SOCK_REFCNT_DEBUG
899 static inline void sk_refcnt_debug_inc(struct sock *sk)
901 atomic_inc(&sk->sk_prot->socks);
904 static inline void sk_refcnt_debug_dec(struct sock *sk)
906 atomic_dec(&sk->sk_prot->socks);
907 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
908 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
911 inline void sk_refcnt_debug_release(const struct sock *sk)
913 if (atomic_read(&sk->sk_refcnt) != 1)
914 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
915 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
917 #else /* SOCK_REFCNT_DEBUG */
918 #define sk_refcnt_debug_inc(sk) do { } while (0)
919 #define sk_refcnt_debug_dec(sk) do { } while (0)
920 #define sk_refcnt_debug_release(sk) do { } while (0)
921 #endif /* SOCK_REFCNT_DEBUG */
923 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
924 extern struct jump_label_key memcg_socket_limit_enabled;
925 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
926 struct cg_proto *cg_proto)
928 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
930 #define mem_cgroup_sockets_enabled static_branch(&memcg_socket_limit_enabled)
932 #define mem_cgroup_sockets_enabled 0
933 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
934 struct cg_proto *cg_proto)
941 static inline bool sk_has_memory_pressure(const struct sock *sk)
943 return sk->sk_prot->memory_pressure != NULL;
946 static inline bool sk_under_memory_pressure(const struct sock *sk)
948 if (!sk->sk_prot->memory_pressure)
951 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
952 return !!*sk->sk_cgrp->memory_pressure;
954 return !!*sk->sk_prot->memory_pressure;
957 static inline void sk_leave_memory_pressure(struct sock *sk)
959 int *memory_pressure = sk->sk_prot->memory_pressure;
961 if (!memory_pressure)
964 if (*memory_pressure)
965 *memory_pressure = 0;
967 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
968 struct cg_proto *cg_proto = sk->sk_cgrp;
969 struct proto *prot = sk->sk_prot;
971 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
972 if (*cg_proto->memory_pressure)
973 *cg_proto->memory_pressure = 0;
978 static inline void sk_enter_memory_pressure(struct sock *sk)
980 if (!sk->sk_prot->enter_memory_pressure)
983 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
984 struct cg_proto *cg_proto = sk->sk_cgrp;
985 struct proto *prot = sk->sk_prot;
987 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
988 cg_proto->enter_memory_pressure(sk);
991 sk->sk_prot->enter_memory_pressure(sk);
994 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
996 long *prot = sk->sk_prot->sysctl_mem;
997 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
998 prot = sk->sk_cgrp->sysctl_mem;
1002 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1006 struct res_counter *fail;
1009 ret = res_counter_charge(prot->memory_allocated,
1010 amt << PAGE_SHIFT, &fail);
1013 *parent_status = OVER_LIMIT;
1016 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1019 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1022 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1025 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1026 return ret >> PAGE_SHIFT;
1030 sk_memory_allocated(const struct sock *sk)
1032 struct proto *prot = sk->sk_prot;
1033 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1034 return memcg_memory_allocated_read(sk->sk_cgrp);
1036 return atomic_long_read(prot->memory_allocated);
1040 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1042 struct proto *prot = sk->sk_prot;
1044 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1045 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1046 /* update the root cgroup regardless */
1047 atomic_long_add_return(amt, prot->memory_allocated);
1048 return memcg_memory_allocated_read(sk->sk_cgrp);
1051 return atomic_long_add_return(amt, prot->memory_allocated);
1055 sk_memory_allocated_sub(struct sock *sk, int amt, int parent_status)
1057 struct proto *prot = sk->sk_prot;
1059 if (mem_cgroup_sockets_enabled && sk->sk_cgrp &&
1060 parent_status != OVER_LIMIT) /* Otherwise was uncharged already */
1061 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1063 atomic_long_sub(amt, prot->memory_allocated);
1066 static inline void sk_sockets_allocated_dec(struct sock *sk)
1068 struct proto *prot = sk->sk_prot;
1070 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1071 struct cg_proto *cg_proto = sk->sk_cgrp;
1073 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1074 percpu_counter_dec(cg_proto->sockets_allocated);
1077 percpu_counter_dec(prot->sockets_allocated);
1080 static inline void sk_sockets_allocated_inc(struct sock *sk)
1082 struct proto *prot = sk->sk_prot;
1084 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1085 struct cg_proto *cg_proto = sk->sk_cgrp;
1087 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1088 percpu_counter_inc(cg_proto->sockets_allocated);
1091 percpu_counter_inc(prot->sockets_allocated);
1095 sk_sockets_allocated_read_positive(struct sock *sk)
1097 struct proto *prot = sk->sk_prot;
1099 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1100 return percpu_counter_sum_positive(sk->sk_cgrp->sockets_allocated);
1102 return percpu_counter_sum_positive(prot->sockets_allocated);
1106 proto_sockets_allocated_sum_positive(struct proto *prot)
1108 return percpu_counter_sum_positive(prot->sockets_allocated);
1112 proto_memory_allocated(struct proto *prot)
1114 return atomic_long_read(prot->memory_allocated);
1118 proto_memory_pressure(struct proto *prot)
1120 if (!prot->memory_pressure)
1122 return !!*prot->memory_pressure;
1126 #ifdef CONFIG_PROC_FS
1127 /* Called with local bh disabled */
1128 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1129 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1131 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
1138 /* With per-bucket locks this operation is not-atomic, so that
1139 * this version is not worse.
1141 static inline void __sk_prot_rehash(struct sock *sk)
1143 sk->sk_prot->unhash(sk);
1144 sk->sk_prot->hash(sk);
1147 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1149 /* About 10 seconds */
1150 #define SOCK_DESTROY_TIME (10*HZ)
1152 /* Sockets 0-1023 can't be bound to unless you are superuser */
1153 #define PROT_SOCK 1024
1155 #define SHUTDOWN_MASK 3
1156 #define RCV_SHUTDOWN 1
1157 #define SEND_SHUTDOWN 2
1159 #define SOCK_SNDBUF_LOCK 1
1160 #define SOCK_RCVBUF_LOCK 2
1161 #define SOCK_BINDADDR_LOCK 4
1162 #define SOCK_BINDPORT_LOCK 8
1164 /* sock_iocb: used to kick off async processing of socket ios */
1166 struct list_head list;
1170 struct socket *sock;
1172 struct scm_cookie *scm;
1173 struct msghdr *msg, async_msg;
1174 struct kiocb *kiocb;
1177 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1179 return (struct sock_iocb *)iocb->private;
1182 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1187 struct socket_alloc {
1188 struct socket socket;
1189 struct inode vfs_inode;
1192 static inline struct socket *SOCKET_I(struct inode *inode)
1194 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1197 static inline struct inode *SOCK_INODE(struct socket *socket)
1199 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1203 * Functions for memory accounting
1205 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1206 extern void __sk_mem_reclaim(struct sock *sk);
1208 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1209 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1210 #define SK_MEM_SEND 0
1211 #define SK_MEM_RECV 1
1213 static inline int sk_mem_pages(int amt)
1215 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1218 static inline int sk_has_account(struct sock *sk)
1220 /* return true if protocol supports memory accounting */
1221 return !!sk->sk_prot->memory_allocated;
1224 static inline int sk_wmem_schedule(struct sock *sk, int size)
1226 if (!sk_has_account(sk))
1228 return size <= sk->sk_forward_alloc ||
1229 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1232 static inline int sk_rmem_schedule(struct sock *sk, int size)
1234 if (!sk_has_account(sk))
1236 return size <= sk->sk_forward_alloc ||
1237 __sk_mem_schedule(sk, size, SK_MEM_RECV);
1240 static inline void sk_mem_reclaim(struct sock *sk)
1242 if (!sk_has_account(sk))
1244 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1245 __sk_mem_reclaim(sk);
1248 static inline void sk_mem_reclaim_partial(struct sock *sk)
1250 if (!sk_has_account(sk))
1252 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1253 __sk_mem_reclaim(sk);
1256 static inline void sk_mem_charge(struct sock *sk, int size)
1258 if (!sk_has_account(sk))
1260 sk->sk_forward_alloc -= size;
1263 static inline void sk_mem_uncharge(struct sock *sk, int size)
1265 if (!sk_has_account(sk))
1267 sk->sk_forward_alloc += size;
1270 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1272 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1273 sk->sk_wmem_queued -= skb->truesize;
1274 sk_mem_uncharge(sk, skb->truesize);
1278 /* Used by processes to "lock" a socket state, so that
1279 * interrupts and bottom half handlers won't change it
1280 * from under us. It essentially blocks any incoming
1281 * packets, so that we won't get any new data or any
1282 * packets that change the state of the socket.
1284 * While locked, BH processing will add new packets to
1285 * the backlog queue. This queue is processed by the
1286 * owner of the socket lock right before it is released.
1288 * Since ~2.3.5 it is also exclusive sleep lock serializing
1289 * accesses from user process context.
1291 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1294 * Macro so as to not evaluate some arguments when
1295 * lockdep is not enabled.
1297 * Mark both the sk_lock and the sk_lock.slock as a
1298 * per-address-family lock class.
1300 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1302 sk->sk_lock.owned = 0; \
1303 init_waitqueue_head(&sk->sk_lock.wq); \
1304 spin_lock_init(&(sk)->sk_lock.slock); \
1305 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1306 sizeof((sk)->sk_lock)); \
1307 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1309 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1312 extern void lock_sock_nested(struct sock *sk, int subclass);
1314 static inline void lock_sock(struct sock *sk)
1316 lock_sock_nested(sk, 0);
1319 extern void release_sock(struct sock *sk);
1321 /* BH context may only use the following locking interface. */
1322 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1323 #define bh_lock_sock_nested(__sk) \
1324 spin_lock_nested(&((__sk)->sk_lock.slock), \
1325 SINGLE_DEPTH_NESTING)
1326 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1328 extern bool lock_sock_fast(struct sock *sk);
1330 * unlock_sock_fast - complement of lock_sock_fast
1334 * fast unlock socket for user context.
1335 * If slow mode is on, we call regular release_sock()
1337 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1342 spin_unlock_bh(&sk->sk_lock.slock);
1346 extern struct sock *sk_alloc(struct net *net, int family,
1348 struct proto *prot);
1349 extern void sk_free(struct sock *sk);
1350 extern void sk_release_kernel(struct sock *sk);
1351 extern struct sock *sk_clone_lock(const struct sock *sk,
1352 const gfp_t priority);
1354 extern struct sk_buff *sock_wmalloc(struct sock *sk,
1355 unsigned long size, int force,
1357 extern struct sk_buff *sock_rmalloc(struct sock *sk,
1358 unsigned long size, int force,
1360 extern void sock_wfree(struct sk_buff *skb);
1361 extern void sock_rfree(struct sk_buff *skb);
1363 extern int sock_setsockopt(struct socket *sock, int level,
1364 int op, char __user *optval,
1365 unsigned int optlen);
1367 extern int sock_getsockopt(struct socket *sock, int level,
1368 int op, char __user *optval,
1369 int __user *optlen);
1370 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1374 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1375 unsigned long header_len,
1376 unsigned long data_len,
1379 extern void *sock_kmalloc(struct sock *sk, int size,
1381 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1382 extern void sk_send_sigurg(struct sock *sk);
1384 #ifdef CONFIG_CGROUPS
1385 extern void sock_update_classid(struct sock *sk);
1387 static inline void sock_update_classid(struct sock *sk)
1393 * Functions to fill in entries in struct proto_ops when a protocol
1394 * does not implement a particular function.
1396 extern int sock_no_bind(struct socket *,
1397 struct sockaddr *, int);
1398 extern int sock_no_connect(struct socket *,
1399 struct sockaddr *, int, int);
1400 extern int sock_no_socketpair(struct socket *,
1402 extern int sock_no_accept(struct socket *,
1403 struct socket *, int);
1404 extern int sock_no_getname(struct socket *,
1405 struct sockaddr *, int *, int);
1406 extern unsigned int sock_no_poll(struct file *, struct socket *,
1407 struct poll_table_struct *);
1408 extern int sock_no_ioctl(struct socket *, unsigned int,
1410 extern int sock_no_listen(struct socket *, int);
1411 extern int sock_no_shutdown(struct socket *, int);
1412 extern int sock_no_getsockopt(struct socket *, int , int,
1413 char __user *, int __user *);
1414 extern int sock_no_setsockopt(struct socket *, int, int,
1415 char __user *, unsigned int);
1416 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1417 struct msghdr *, size_t);
1418 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1419 struct msghdr *, size_t, int);
1420 extern int sock_no_mmap(struct file *file,
1421 struct socket *sock,
1422 struct vm_area_struct *vma);
1423 extern ssize_t sock_no_sendpage(struct socket *sock,
1425 int offset, size_t size,
1429 * Functions to fill in entries in struct proto_ops when a protocol
1430 * uses the inet style.
1432 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1433 char __user *optval, int __user *optlen);
1434 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1435 struct msghdr *msg, size_t size, int flags);
1436 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1437 char __user *optval, unsigned int optlen);
1438 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1439 int optname, char __user *optval, int __user *optlen);
1440 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1441 int optname, char __user *optval, unsigned int optlen);
1443 extern void sk_common_release(struct sock *sk);
1446 * Default socket callbacks and setup code
1449 /* Initialise core socket variables */
1450 extern void sock_init_data(struct socket *sock, struct sock *sk);
1452 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1455 * sk_filter_release - release a socket filter
1456 * @fp: filter to remove
1458 * Remove a filter from a socket and release its resources.
1461 static inline void sk_filter_release(struct sk_filter *fp)
1463 if (atomic_dec_and_test(&fp->refcnt))
1464 call_rcu(&fp->rcu, sk_filter_release_rcu);
1467 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1469 unsigned int size = sk_filter_len(fp);
1471 atomic_sub(size, &sk->sk_omem_alloc);
1472 sk_filter_release(fp);
1475 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1477 atomic_inc(&fp->refcnt);
1478 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1482 * Socket reference counting postulates.
1484 * * Each user of socket SHOULD hold a reference count.
1485 * * Each access point to socket (an hash table bucket, reference from a list,
1486 * running timer, skb in flight MUST hold a reference count.
1487 * * When reference count hits 0, it means it will never increase back.
1488 * * When reference count hits 0, it means that no references from
1489 * outside exist to this socket and current process on current CPU
1490 * is last user and may/should destroy this socket.
1491 * * sk_free is called from any context: process, BH, IRQ. When
1492 * it is called, socket has no references from outside -> sk_free
1493 * may release descendant resources allocated by the socket, but
1494 * to the time when it is called, socket is NOT referenced by any
1495 * hash tables, lists etc.
1496 * * Packets, delivered from outside (from network or from another process)
1497 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1498 * when they sit in queue. Otherwise, packets will leak to hole, when
1499 * socket is looked up by one cpu and unhasing is made by another CPU.
1500 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1501 * (leak to backlog). Packet socket does all the processing inside
1502 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1503 * use separate SMP lock, so that they are prone too.
1506 /* Ungrab socket and destroy it, if it was the last reference. */
1507 static inline void sock_put(struct sock *sk)
1509 if (atomic_dec_and_test(&sk->sk_refcnt))
1513 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1516 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1518 sk->sk_tx_queue_mapping = tx_queue;
1521 static inline void sk_tx_queue_clear(struct sock *sk)
1523 sk->sk_tx_queue_mapping = -1;
1526 static inline int sk_tx_queue_get(const struct sock *sk)
1528 return sk ? sk->sk_tx_queue_mapping : -1;
1531 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1533 sk_tx_queue_clear(sk);
1534 sk->sk_socket = sock;
1537 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1539 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1540 return &rcu_dereference_raw(sk->sk_wq)->wait;
1542 /* Detach socket from process context.
1543 * Announce socket dead, detach it from wait queue and inode.
1544 * Note that parent inode held reference count on this struct sock,
1545 * we do not release it in this function, because protocol
1546 * probably wants some additional cleanups or even continuing
1547 * to work with this socket (TCP).
1549 static inline void sock_orphan(struct sock *sk)
1551 write_lock_bh(&sk->sk_callback_lock);
1552 sock_set_flag(sk, SOCK_DEAD);
1553 sk_set_socket(sk, NULL);
1555 write_unlock_bh(&sk->sk_callback_lock);
1558 static inline void sock_graft(struct sock *sk, struct socket *parent)
1560 write_lock_bh(&sk->sk_callback_lock);
1561 sk->sk_wq = parent->wq;
1563 sk_set_socket(sk, parent);
1564 security_sock_graft(sk, parent);
1565 write_unlock_bh(&sk->sk_callback_lock);
1568 extern int sock_i_uid(struct sock *sk);
1569 extern unsigned long sock_i_ino(struct sock *sk);
1571 static inline struct dst_entry *
1572 __sk_dst_get(struct sock *sk)
1574 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1575 lockdep_is_held(&sk->sk_lock.slock));
1578 static inline struct dst_entry *
1579 sk_dst_get(struct sock *sk)
1581 struct dst_entry *dst;
1584 dst = rcu_dereference(sk->sk_dst_cache);
1591 extern void sk_reset_txq(struct sock *sk);
1593 static inline void dst_negative_advice(struct sock *sk)
1595 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1597 if (dst && dst->ops->negative_advice) {
1598 ndst = dst->ops->negative_advice(dst);
1601 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1608 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1610 struct dst_entry *old_dst;
1612 sk_tx_queue_clear(sk);
1614 * This can be called while sk is owned by the caller only,
1615 * with no state that can be checked in a rcu_dereference_check() cond
1617 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1618 rcu_assign_pointer(sk->sk_dst_cache, dst);
1619 dst_release(old_dst);
1623 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1625 spin_lock(&sk->sk_dst_lock);
1626 __sk_dst_set(sk, dst);
1627 spin_unlock(&sk->sk_dst_lock);
1631 __sk_dst_reset(struct sock *sk)
1633 __sk_dst_set(sk, NULL);
1637 sk_dst_reset(struct sock *sk)
1639 spin_lock(&sk->sk_dst_lock);
1641 spin_unlock(&sk->sk_dst_lock);
1644 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1646 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1648 static inline int sk_can_gso(const struct sock *sk)
1650 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1653 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1655 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1657 sk->sk_route_nocaps |= flags;
1658 sk->sk_route_caps &= ~flags;
1661 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1662 char __user *from, char *to,
1663 int copy, int offset)
1665 if (skb->ip_summed == CHECKSUM_NONE) {
1667 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1670 skb->csum = csum_block_add(skb->csum, csum, offset);
1671 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1672 if (!access_ok(VERIFY_READ, from, copy) ||
1673 __copy_from_user_nocache(to, from, copy))
1675 } else if (copy_from_user(to, from, copy))
1681 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1682 char __user *from, int copy)
1684 int err, offset = skb->len;
1686 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1689 __skb_trim(skb, offset);
1694 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1695 struct sk_buff *skb,
1701 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1707 skb->data_len += copy;
1708 skb->truesize += copy;
1709 sk->sk_wmem_queued += copy;
1710 sk_mem_charge(sk, copy);
1714 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1715 struct sk_buff *skb, struct page *page,
1718 if (skb->ip_summed == CHECKSUM_NONE) {
1720 __wsum csum = csum_and_copy_from_user(from,
1721 page_address(page) + off,
1725 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1726 } else if (copy_from_user(page_address(page) + off, from, copy))
1730 skb->data_len += copy;
1731 skb->truesize += copy;
1732 sk->sk_wmem_queued += copy;
1733 sk_mem_charge(sk, copy);
1738 * sk_wmem_alloc_get - returns write allocations
1741 * Returns sk_wmem_alloc minus initial offset of one
1743 static inline int sk_wmem_alloc_get(const struct sock *sk)
1745 return atomic_read(&sk->sk_wmem_alloc) - 1;
1749 * sk_rmem_alloc_get - returns read allocations
1752 * Returns sk_rmem_alloc
1754 static inline int sk_rmem_alloc_get(const struct sock *sk)
1756 return atomic_read(&sk->sk_rmem_alloc);
1760 * sk_has_allocations - check if allocations are outstanding
1763 * Returns true if socket has write or read allocations
1765 static inline int sk_has_allocations(const struct sock *sk)
1767 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1771 * wq_has_sleeper - check if there are any waiting processes
1772 * @wq: struct socket_wq
1774 * Returns true if socket_wq has waiting processes
1776 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1777 * barrier call. They were added due to the race found within the tcp code.
1779 * Consider following tcp code paths:
1783 * sys_select receive packet
1785 * __add_wait_queue update tp->rcv_nxt
1787 * tp->rcv_nxt check sock_def_readable
1789 * schedule rcu_read_lock();
1790 * wq = rcu_dereference(sk->sk_wq);
1791 * if (wq && waitqueue_active(&wq->wait))
1792 * wake_up_interruptible(&wq->wait)
1796 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1797 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1798 * could then endup calling schedule and sleep forever if there are no more
1799 * data on the socket.
1802 static inline bool wq_has_sleeper(struct socket_wq *wq)
1806 * We need to be sure we are in sync with the
1807 * add_wait_queue modifications to the wait queue.
1809 * This memory barrier is paired in the sock_poll_wait.
1812 return wq && waitqueue_active(&wq->wait);
1816 * sock_poll_wait - place memory barrier behind the poll_wait call.
1818 * @wait_address: socket wait queue
1821 * See the comments in the wq_has_sleeper function.
1823 static inline void sock_poll_wait(struct file *filp,
1824 wait_queue_head_t *wait_address, poll_table *p)
1826 if (p && wait_address) {
1827 poll_wait(filp, wait_address, p);
1829 * We need to be sure we are in sync with the
1830 * socket flags modification.
1832 * This memory barrier is paired in the wq_has_sleeper.
1839 * Queue a received datagram if it will fit. Stream and sequenced
1840 * protocols can't normally use this as they need to fit buffers in
1841 * and play with them.
1843 * Inlined as it's very short and called for pretty much every
1844 * packet ever received.
1847 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1851 skb->destructor = sock_wfree;
1853 * We used to take a refcount on sk, but following operation
1854 * is enough to guarantee sk_free() wont free this sock until
1855 * all in-flight packets are completed
1857 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1860 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1864 skb->destructor = sock_rfree;
1865 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1866 sk_mem_charge(sk, skb->truesize);
1869 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1870 unsigned long expires);
1872 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1874 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1876 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1879 * Recover an error report and clear atomically
1882 static inline int sock_error(struct sock *sk)
1885 if (likely(!sk->sk_err))
1887 err = xchg(&sk->sk_err, 0);
1891 static inline unsigned long sock_wspace(struct sock *sk)
1895 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1896 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1903 static inline void sk_wake_async(struct sock *sk, int how, int band)
1905 if (sock_flag(sk, SOCK_FASYNC))
1906 sock_wake_async(sk->sk_socket, how, band);
1909 #define SOCK_MIN_SNDBUF 2048
1911 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1912 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1914 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1916 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1918 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1919 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1920 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1924 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1926 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1928 struct page *page = NULL;
1930 page = alloc_pages(sk->sk_allocation, 0);
1932 sk_enter_memory_pressure(sk);
1933 sk_stream_moderate_sndbuf(sk);
1939 * Default write policy as shown to user space via poll/select/SIGIO
1941 static inline int sock_writeable(const struct sock *sk)
1943 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1946 static inline gfp_t gfp_any(void)
1948 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1951 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1953 return noblock ? 0 : sk->sk_rcvtimeo;
1956 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1958 return noblock ? 0 : sk->sk_sndtimeo;
1961 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1963 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1966 /* Alas, with timeout socket operations are not restartable.
1967 * Compare this to poll().
1969 static inline int sock_intr_errno(long timeo)
1971 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1974 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1975 struct sk_buff *skb);
1976 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
1977 struct sk_buff *skb);
1979 static __inline__ void
1980 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1982 ktime_t kt = skb->tstamp;
1983 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1986 * generate control messages if
1987 * - receive time stamping in software requested (SOCK_RCVTSTAMP
1988 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
1989 * - software time stamp available and wanted
1990 * (SOCK_TIMESTAMPING_SOFTWARE)
1991 * - hardware time stamps available and wanted
1992 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
1993 * SOCK_TIMESTAMPING_RAW_HARDWARE)
1995 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
1996 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
1997 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
1998 (hwtstamps->hwtstamp.tv64 &&
1999 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2000 (hwtstamps->syststamp.tv64 &&
2001 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2002 __sock_recv_timestamp(msg, sk, skb);
2006 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2007 __sock_recv_wifi_status(msg, sk, skb);
2010 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2011 struct sk_buff *skb);
2013 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2014 struct sk_buff *skb)
2016 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2017 (1UL << SOCK_RCVTSTAMP) | \
2018 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
2019 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2020 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2021 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2023 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2024 __sock_recv_ts_and_drops(msg, sk, skb);
2026 sk->sk_stamp = skb->tstamp;
2030 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2031 * @sk: socket sending this packet
2032 * @tx_flags: filled with instructions for time stamping
2034 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2035 * parameters are invalid.
2037 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2040 * sk_eat_skb - Release a skb if it is no longer needed
2041 * @sk: socket to eat this skb from
2042 * @skb: socket buffer to eat
2043 * @copied_early: flag indicating whether DMA operations copied this data early
2045 * This routine must be called with interrupts disabled or with the socket
2046 * locked so that the sk_buff queue operation is ok.
2048 #ifdef CONFIG_NET_DMA
2049 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
2051 __skb_unlink(skb, &sk->sk_receive_queue);
2055 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2058 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
2060 __skb_unlink(skb, &sk->sk_receive_queue);
2066 struct net *sock_net(const struct sock *sk)
2068 return read_pnet(&sk->sk_net);
2072 void sock_net_set(struct sock *sk, struct net *net)
2074 write_pnet(&sk->sk_net, net);
2078 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2079 * They should not hold a reference to a namespace in order to allow
2081 * Sockets after sk_change_net should be released using sk_release_kernel
2083 static inline void sk_change_net(struct sock *sk, struct net *net)
2085 put_net(sock_net(sk));
2086 sock_net_set(sk, hold_net(net));
2089 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2091 if (unlikely(skb->sk)) {
2092 struct sock *sk = skb->sk;
2094 skb->destructor = NULL;
2101 extern void sock_enable_timestamp(struct sock *sk, int flag);
2102 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2103 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2106 * Enable debug/info messages
2108 extern int net_msg_warn;
2109 #define NETDEBUG(fmt, args...) \
2110 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2112 #define LIMIT_NETDEBUG(fmt, args...) \
2113 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2115 extern __u32 sysctl_wmem_max;
2116 extern __u32 sysctl_rmem_max;
2118 extern void sk_init(void);
2120 extern int sysctl_optmem_max;
2122 extern __u32 sysctl_wmem_default;
2123 extern __u32 sysctl_rmem_default;
2125 #endif /* _SOCK_H */