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>
57 #include <linux/filter.h>
58 #include <linux/rculist_nulls.h>
59 #include <linux/poll.h>
61 #include <linux/atomic.h>
63 #include <net/checksum.h>
66 * This structure really needs to be cleaned up.
67 * Most of it is for TCP, and not used by any of
68 * the other protocols.
71 /* Define this to get the SOCK_DBG debugging facility. */
72 #define SOCK_DEBUGGING
74 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
75 printk(KERN_DEBUG msg); } while (0)
77 /* Validate arguments and do nothing */
78 static inline __printf(2, 3)
79 void SOCK_DEBUG(struct sock *sk, const char *msg, ...)
84 /* This is the per-socket lock. The spinlock provides a synchronization
85 * between user contexts and software interrupt processing, whereas the
86 * mini-semaphore synchronizes multiple users amongst themselves.
93 * We express the mutex-alike socket_lock semantics
94 * to the lock validator by explicitly managing
95 * the slock as a lock variant (in addition to
98 #ifdef CONFIG_DEBUG_LOCK_ALLOC
99 struct lockdep_map dep_map;
108 * struct sock_common - minimal network layer representation of sockets
109 * @skc_daddr: Foreign IPv4 addr
110 * @skc_rcv_saddr: Bound local IPv4 addr
111 * @skc_hash: hash value used with various protocol lookup tables
112 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
113 * @skc_family: network address family
114 * @skc_state: Connection state
115 * @skc_reuse: %SO_REUSEADDR setting
116 * @skc_bound_dev_if: bound device index if != 0
117 * @skc_bind_node: bind hash linkage for various protocol lookup tables
118 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
119 * @skc_prot: protocol handlers inside a network family
120 * @skc_net: reference to the network namespace of this socket
121 * @skc_node: main hash linkage for various protocol lookup tables
122 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
123 * @skc_tx_queue_mapping: tx queue number for this connection
124 * @skc_refcnt: reference count
126 * This is the minimal network layer representation of sockets, the header
127 * for struct sock and struct inet_timewait_sock.
130 /* skc_daddr and skc_rcv_saddr must be grouped :
131 * cf INET_MATCH() and INET_TW_MATCH()
134 __be32 skc_rcv_saddr;
137 unsigned int skc_hash;
138 __u16 skc_u16hashes[2];
140 unsigned short skc_family;
141 volatile unsigned char skc_state;
142 unsigned char skc_reuse;
143 int skc_bound_dev_if;
145 struct hlist_node skc_bind_node;
146 struct hlist_nulls_node skc_portaddr_node;
148 struct proto *skc_prot;
153 * fields between dontcopy_begin/dontcopy_end
154 * are not copied in sock_copy()
157 int skc_dontcopy_begin[0];
160 struct hlist_node skc_node;
161 struct hlist_nulls_node skc_nulls_node;
163 int skc_tx_queue_mapping;
166 int skc_dontcopy_end[0];
171 * struct sock - network layer representation of sockets
172 * @__sk_common: shared layout with inet_timewait_sock
173 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
174 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
175 * @sk_lock: synchronizer
176 * @sk_rcvbuf: size of receive buffer in bytes
177 * @sk_wq: sock wait queue and async head
178 * @sk_dst_cache: destination cache
179 * @sk_dst_lock: destination cache lock
180 * @sk_policy: flow policy
181 * @sk_receive_queue: incoming packets
182 * @sk_wmem_alloc: transmit queue bytes committed
183 * @sk_write_queue: Packet sending queue
184 * @sk_async_wait_queue: DMA copied packets
185 * @sk_omem_alloc: "o" is "option" or "other"
186 * @sk_wmem_queued: persistent queue size
187 * @sk_forward_alloc: space allocated forward
188 * @sk_allocation: allocation mode
189 * @sk_sndbuf: size of send buffer in bytes
190 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
191 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
192 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
193 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
194 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
195 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
196 * @sk_gso_max_size: Maximum GSO segment size to build
197 * @sk_lingertime: %SO_LINGER l_linger setting
198 * @sk_backlog: always used with the per-socket spinlock held
199 * @sk_callback_lock: used with the callbacks in the end of this struct
200 * @sk_error_queue: rarely used
201 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
202 * IPV6_ADDRFORM for instance)
203 * @sk_err: last error
204 * @sk_err_soft: errors that don't cause failure but are the cause of a
205 * persistent failure not just 'timed out'
206 * @sk_drops: raw/udp drops counter
207 * @sk_ack_backlog: current listen backlog
208 * @sk_max_ack_backlog: listen backlog set in listen()
209 * @sk_priority: %SO_PRIORITY setting
210 * @sk_type: socket type (%SOCK_STREAM, etc)
211 * @sk_protocol: which protocol this socket belongs in this network family
212 * @sk_peer_pid: &struct pid for this socket's peer
213 * @sk_peer_cred: %SO_PEERCRED setting
214 * @sk_rcvlowat: %SO_RCVLOWAT setting
215 * @sk_rcvtimeo: %SO_RCVTIMEO setting
216 * @sk_sndtimeo: %SO_SNDTIMEO setting
217 * @sk_rxhash: flow hash received from netif layer
218 * @sk_filter: socket filtering instructions
219 * @sk_protinfo: private area, net family specific, when not using slab
220 * @sk_timer: sock cleanup timer
221 * @sk_stamp: time stamp of last packet received
222 * @sk_socket: Identd and reporting IO signals
223 * @sk_user_data: RPC layer private data
224 * @sk_sndmsg_page: cached page for sendmsg
225 * @sk_sndmsg_off: cached offset for sendmsg
226 * @sk_send_head: front of stuff to transmit
227 * @sk_security: used by security modules
228 * @sk_mark: generic packet mark
229 * @sk_classid: this socket's cgroup classid
230 * @sk_write_pending: a write to stream socket waits to start
231 * @sk_state_change: callback to indicate change in the state of the sock
232 * @sk_data_ready: callback to indicate there is data to be processed
233 * @sk_write_space: callback to indicate there is bf sending space available
234 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
235 * @sk_backlog_rcv: callback to process the backlog
236 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
240 * Now struct inet_timewait_sock also uses sock_common, so please just
241 * don't add nothing before this first member (__sk_common) --acme
243 struct sock_common __sk_common;
244 #define sk_node __sk_common.skc_node
245 #define sk_nulls_node __sk_common.skc_nulls_node
246 #define sk_refcnt __sk_common.skc_refcnt
247 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
249 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
250 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
251 #define sk_hash __sk_common.skc_hash
252 #define sk_family __sk_common.skc_family
253 #define sk_state __sk_common.skc_state
254 #define sk_reuse __sk_common.skc_reuse
255 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
256 #define sk_bind_node __sk_common.skc_bind_node
257 #define sk_prot __sk_common.skc_prot
258 #define sk_net __sk_common.skc_net
259 socket_lock_t sk_lock;
260 struct sk_buff_head sk_receive_queue;
262 * The backlog queue is special, it is always used with
263 * the per-socket spinlock held and requires low latency
264 * access. Therefore we special case it's implementation.
265 * Note : rmem_alloc is in this structure to fill a hole
266 * on 64bit arches, not because its logically part of
272 struct sk_buff *head;
273 struct sk_buff *tail;
275 #define sk_rmem_alloc sk_backlog.rmem_alloc
276 int sk_forward_alloc;
283 struct sk_filter __rcu *sk_filter;
284 struct socket_wq __rcu *sk_wq;
286 #ifdef CONFIG_NET_DMA
287 struct sk_buff_head sk_async_wait_queue;
291 struct xfrm_policy *sk_policy[2];
293 unsigned long sk_flags;
294 struct dst_entry *sk_dst_cache;
295 spinlock_t sk_dst_lock;
296 atomic_t sk_wmem_alloc;
297 atomic_t sk_omem_alloc;
299 struct sk_buff_head sk_write_queue;
300 kmemcheck_bitfield_begin(flags);
301 unsigned int sk_shutdown : 2,
306 #define SK_PROTOCOL_MAX ((u8)~0U)
307 kmemcheck_bitfield_end(flags);
313 unsigned int sk_gso_max_size;
315 unsigned long sk_lingertime;
316 struct sk_buff_head sk_error_queue;
317 struct proto *sk_prot_creator;
318 rwlock_t sk_callback_lock;
321 unsigned short sk_ack_backlog;
322 unsigned short sk_max_ack_backlog;
324 struct pid *sk_peer_pid;
325 const struct cred *sk_peer_cred;
329 struct timer_list sk_timer;
331 struct socket *sk_socket;
333 struct page *sk_sndmsg_page;
334 struct sk_buff *sk_send_head;
336 int sk_write_pending;
337 #ifdef CONFIG_SECURITY
342 void (*sk_state_change)(struct sock *sk);
343 void (*sk_data_ready)(struct sock *sk, int bytes);
344 void (*sk_write_space)(struct sock *sk);
345 void (*sk_error_report)(struct sock *sk);
346 int (*sk_backlog_rcv)(struct sock *sk,
347 struct sk_buff *skb);
348 void (*sk_destruct)(struct sock *sk);
352 * Hashed lists helper routines
354 static inline struct sock *sk_entry(const struct hlist_node *node)
356 return hlist_entry(node, struct sock, sk_node);
359 static inline struct sock *__sk_head(const struct hlist_head *head)
361 return hlist_entry(head->first, struct sock, sk_node);
364 static inline struct sock *sk_head(const struct hlist_head *head)
366 return hlist_empty(head) ? NULL : __sk_head(head);
369 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
371 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
374 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
376 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
379 static inline struct sock *sk_next(const struct sock *sk)
381 return sk->sk_node.next ?
382 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
385 static inline struct sock *sk_nulls_next(const struct sock *sk)
387 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
388 hlist_nulls_entry(sk->sk_nulls_node.next,
389 struct sock, sk_nulls_node) :
393 static inline int sk_unhashed(const struct sock *sk)
395 return hlist_unhashed(&sk->sk_node);
398 static inline int sk_hashed(const struct sock *sk)
400 return !sk_unhashed(sk);
403 static __inline__ void sk_node_init(struct hlist_node *node)
408 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
413 static __inline__ void __sk_del_node(struct sock *sk)
415 __hlist_del(&sk->sk_node);
418 /* NB: equivalent to hlist_del_init_rcu */
419 static __inline__ int __sk_del_node_init(struct sock *sk)
423 sk_node_init(&sk->sk_node);
429 /* Grab socket reference count. This operation is valid only
430 when sk is ALREADY grabbed f.e. it is found in hash table
431 or a list and the lookup is made under lock preventing hash table
435 static inline void sock_hold(struct sock *sk)
437 atomic_inc(&sk->sk_refcnt);
440 /* Ungrab socket in the context, which assumes that socket refcnt
441 cannot hit zero, f.e. it is true in context of any socketcall.
443 static inline void __sock_put(struct sock *sk)
445 atomic_dec(&sk->sk_refcnt);
448 static __inline__ int sk_del_node_init(struct sock *sk)
450 int rc = __sk_del_node_init(sk);
453 /* paranoid for a while -acme */
454 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
459 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
461 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
464 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
470 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
472 int rc = __sk_nulls_del_node_init_rcu(sk);
475 /* paranoid for a while -acme */
476 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
482 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
484 hlist_add_head(&sk->sk_node, list);
487 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
490 __sk_add_node(sk, list);
493 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
496 hlist_add_head_rcu(&sk->sk_node, list);
499 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
501 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
504 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
507 __sk_nulls_add_node_rcu(sk, list);
510 static __inline__ void __sk_del_bind_node(struct sock *sk)
512 __hlist_del(&sk->sk_bind_node);
515 static __inline__ void sk_add_bind_node(struct sock *sk,
516 struct hlist_head *list)
518 hlist_add_head(&sk->sk_bind_node, list);
521 #define sk_for_each(__sk, node, list) \
522 hlist_for_each_entry(__sk, node, list, sk_node)
523 #define sk_for_each_rcu(__sk, node, list) \
524 hlist_for_each_entry_rcu(__sk, node, list, sk_node)
525 #define sk_nulls_for_each(__sk, node, list) \
526 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
527 #define sk_nulls_for_each_rcu(__sk, node, list) \
528 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
529 #define sk_for_each_from(__sk, node) \
530 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
531 hlist_for_each_entry_from(__sk, node, sk_node)
532 #define sk_nulls_for_each_from(__sk, node) \
533 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
534 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
535 #define sk_for_each_safe(__sk, node, tmp, list) \
536 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
537 #define sk_for_each_bound(__sk, node, list) \
538 hlist_for_each_entry(__sk, node, list, sk_bind_node)
551 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
552 SOCK_DBG, /* %SO_DEBUG setting */
553 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
554 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
555 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
556 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
557 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
558 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
559 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
560 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
561 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
562 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
563 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
564 SOCK_FASYNC, /* fasync() active */
566 SOCK_ZEROCOPY, /* buffers from userspace */
569 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
571 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
573 nsk->sk_flags = osk->sk_flags;
576 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
578 __set_bit(flag, &sk->sk_flags);
581 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
583 __clear_bit(flag, &sk->sk_flags);
586 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
588 return test_bit(flag, &sk->sk_flags);
591 static inline void sk_acceptq_removed(struct sock *sk)
593 sk->sk_ack_backlog--;
596 static inline void sk_acceptq_added(struct sock *sk)
598 sk->sk_ack_backlog++;
601 static inline int sk_acceptq_is_full(struct sock *sk)
603 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
607 * Compute minimal free write space needed to queue new packets.
609 static inline int sk_stream_min_wspace(struct sock *sk)
611 return sk->sk_wmem_queued >> 1;
614 static inline int sk_stream_wspace(struct sock *sk)
616 return sk->sk_sndbuf - sk->sk_wmem_queued;
619 extern void sk_stream_write_space(struct sock *sk);
621 static inline int sk_stream_memory_free(struct sock *sk)
623 return sk->sk_wmem_queued < sk->sk_sndbuf;
626 /* OOB backlog add */
627 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
629 /* dont let skb dst not refcounted, we are going to leave rcu lock */
632 if (!sk->sk_backlog.tail)
633 sk->sk_backlog.head = skb;
635 sk->sk_backlog.tail->next = skb;
637 sk->sk_backlog.tail = skb;
642 * Take into account size of receive queue and backlog queue
643 * Do not take into account this skb truesize,
644 * to allow even a single big packet to come.
646 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb)
648 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
650 return qsize > sk->sk_rcvbuf;
653 /* The per-socket spinlock must be held here. */
654 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
656 if (sk_rcvqueues_full(sk, skb))
659 __sk_add_backlog(sk, skb);
660 sk->sk_backlog.len += skb->truesize;
664 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
666 return sk->sk_backlog_rcv(sk, skb);
669 static inline void sock_rps_record_flow(const struct sock *sk)
672 struct rps_sock_flow_table *sock_flow_table;
675 sock_flow_table = rcu_dereference(rps_sock_flow_table);
676 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
681 static inline void sock_rps_reset_flow(const struct sock *sk)
684 struct rps_sock_flow_table *sock_flow_table;
687 sock_flow_table = rcu_dereference(rps_sock_flow_table);
688 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
693 static inline void sock_rps_save_rxhash(struct sock *sk,
694 const struct sk_buff *skb)
697 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
698 sock_rps_reset_flow(sk);
699 sk->sk_rxhash = skb->rxhash;
704 static inline void sock_rps_reset_rxhash(struct sock *sk)
707 sock_rps_reset_flow(sk);
712 #define sk_wait_event(__sk, __timeo, __condition) \
714 release_sock(__sk); \
715 __rc = __condition; \
717 *(__timeo) = schedule_timeout(*(__timeo)); \
720 __rc = __condition; \
724 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
725 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
726 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
727 extern int sk_stream_error(struct sock *sk, int flags, int err);
728 extern void sk_stream_kill_queues(struct sock *sk);
730 extern int sk_wait_data(struct sock *sk, long *timeo);
732 struct request_sock_ops;
733 struct timewait_sock_ops;
734 struct inet_hashinfo;
739 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
740 * un-modified. Special care is taken when initializing object to zero.
742 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
744 if (offsetof(struct sock, sk_node.next) != 0)
745 memset(sk, 0, offsetof(struct sock, sk_node.next));
746 memset(&sk->sk_node.pprev, 0,
747 size - offsetof(struct sock, sk_node.pprev));
750 /* Networking protocol blocks we attach to sockets.
751 * socket layer -> transport layer interface
752 * transport -> network interface is defined by struct inet_proto
755 void (*close)(struct sock *sk,
757 int (*connect)(struct sock *sk,
758 struct sockaddr *uaddr,
760 int (*disconnect)(struct sock *sk, int flags);
762 struct sock * (*accept) (struct sock *sk, int flags, int *err);
764 int (*ioctl)(struct sock *sk, int cmd,
766 int (*init)(struct sock *sk);
767 void (*destroy)(struct sock *sk);
768 void (*shutdown)(struct sock *sk, int how);
769 int (*setsockopt)(struct sock *sk, int level,
770 int optname, char __user *optval,
771 unsigned int optlen);
772 int (*getsockopt)(struct sock *sk, int level,
773 int optname, char __user *optval,
776 int (*compat_setsockopt)(struct sock *sk,
778 int optname, char __user *optval,
779 unsigned int optlen);
780 int (*compat_getsockopt)(struct sock *sk,
782 int optname, char __user *optval,
784 int (*compat_ioctl)(struct sock *sk,
785 unsigned int cmd, unsigned long arg);
787 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
788 struct msghdr *msg, size_t len);
789 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
791 size_t len, int noblock, int flags,
793 int (*sendpage)(struct sock *sk, struct page *page,
794 int offset, size_t size, int flags);
795 int (*bind)(struct sock *sk,
796 struct sockaddr *uaddr, int addr_len);
798 int (*backlog_rcv) (struct sock *sk,
799 struct sk_buff *skb);
801 /* Keeping track of sk's, looking them up, and port selection methods. */
802 void (*hash)(struct sock *sk);
803 void (*unhash)(struct sock *sk);
804 void (*rehash)(struct sock *sk);
805 int (*get_port)(struct sock *sk, unsigned short snum);
806 void (*clear_sk)(struct sock *sk, int size);
808 /* Keeping track of sockets in use */
809 #ifdef CONFIG_PROC_FS
810 unsigned int inuse_idx;
813 /* Memory pressure */
814 void (*enter_memory_pressure)(struct sock *sk);
815 atomic_long_t *memory_allocated; /* Current allocated memory. */
816 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
818 * Pressure flag: try to collapse.
819 * Technical note: it is used by multiple contexts non atomically.
820 * All the __sk_mem_schedule() is of this nature: accounting
821 * is strict, actions are advisory and have some latency.
823 int *memory_pressure;
830 struct kmem_cache *slab;
831 unsigned int obj_size;
834 struct percpu_counter *orphan_count;
836 struct request_sock_ops *rsk_prot;
837 struct timewait_sock_ops *twsk_prot;
840 struct inet_hashinfo *hashinfo;
841 struct udp_table *udp_table;
842 struct raw_hashinfo *raw_hash;
845 struct module *owner;
849 struct list_head node;
850 #ifdef SOCK_REFCNT_DEBUG
855 extern int proto_register(struct proto *prot, int alloc_slab);
856 extern void proto_unregister(struct proto *prot);
858 #ifdef SOCK_REFCNT_DEBUG
859 static inline void sk_refcnt_debug_inc(struct sock *sk)
861 atomic_inc(&sk->sk_prot->socks);
864 static inline void sk_refcnt_debug_dec(struct sock *sk)
866 atomic_dec(&sk->sk_prot->socks);
867 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
868 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
871 static inline void sk_refcnt_debug_release(const struct sock *sk)
873 if (atomic_read(&sk->sk_refcnt) != 1)
874 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
875 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
877 #else /* SOCK_REFCNT_DEBUG */
878 #define sk_refcnt_debug_inc(sk) do { } while (0)
879 #define sk_refcnt_debug_dec(sk) do { } while (0)
880 #define sk_refcnt_debug_release(sk) do { } while (0)
881 #endif /* SOCK_REFCNT_DEBUG */
884 #ifdef CONFIG_PROC_FS
885 /* Called with local bh disabled */
886 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
887 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
889 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
896 /* With per-bucket locks this operation is not-atomic, so that
897 * this version is not worse.
899 static inline void __sk_prot_rehash(struct sock *sk)
901 sk->sk_prot->unhash(sk);
902 sk->sk_prot->hash(sk);
905 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
907 /* About 10 seconds */
908 #define SOCK_DESTROY_TIME (10*HZ)
910 /* Sockets 0-1023 can't be bound to unless you are superuser */
911 #define PROT_SOCK 1024
913 #define SHUTDOWN_MASK 3
914 #define RCV_SHUTDOWN 1
915 #define SEND_SHUTDOWN 2
917 #define SOCK_SNDBUF_LOCK 1
918 #define SOCK_RCVBUF_LOCK 2
919 #define SOCK_BINDADDR_LOCK 4
920 #define SOCK_BINDPORT_LOCK 8
922 /* sock_iocb: used to kick off async processing of socket ios */
924 struct list_head list;
930 struct scm_cookie *scm;
931 struct msghdr *msg, async_msg;
935 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
937 return (struct sock_iocb *)iocb->private;
940 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
945 struct socket_alloc {
946 struct socket socket;
947 struct inode vfs_inode;
950 static inline struct socket *SOCKET_I(struct inode *inode)
952 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
955 static inline struct inode *SOCK_INODE(struct socket *socket)
957 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
961 * Functions for memory accounting
963 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
964 void __sk_mem_reclaim(struct sock *sk, int amount);
966 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
967 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
968 #define SK_MEM_SEND 0
969 #define SK_MEM_RECV 1
971 static inline int sk_mem_pages(int amt)
973 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
976 static inline int sk_has_account(struct sock *sk)
978 /* return true if protocol supports memory accounting */
979 return !!sk->sk_prot->memory_allocated;
982 static inline int sk_wmem_schedule(struct sock *sk, int size)
984 if (!sk_has_account(sk))
986 return size <= sk->sk_forward_alloc ||
987 __sk_mem_schedule(sk, size, SK_MEM_SEND);
990 static inline int sk_rmem_schedule(struct sock *sk, int size)
992 if (!sk_has_account(sk))
994 return size <= sk->sk_forward_alloc ||
995 __sk_mem_schedule(sk, size, SK_MEM_RECV);
998 static inline void sk_mem_reclaim(struct sock *sk)
1000 if (!sk_has_account(sk))
1002 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1003 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1006 static inline void sk_mem_reclaim_partial(struct sock *sk)
1008 if (!sk_has_account(sk))
1010 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1011 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1014 static inline void sk_mem_charge(struct sock *sk, int size)
1016 if (!sk_has_account(sk))
1018 sk->sk_forward_alloc -= size;
1021 static inline void sk_mem_uncharge(struct sock *sk, int size)
1023 if (!sk_has_account(sk))
1025 sk->sk_forward_alloc += size;
1027 /* Avoid a possible overflow.
1028 * TCP send queues can make this happen, if sk_mem_reclaim()
1029 * is not called and more than 2 GBytes are released at once.
1031 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1032 * no need to hold that much forward allocation anyway.
1034 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1035 __sk_mem_reclaim(sk, 1 << 20);
1038 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1040 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1041 sk->sk_wmem_queued -= skb->truesize;
1042 sk_mem_uncharge(sk, skb->truesize);
1046 /* Used by processes to "lock" a socket state, so that
1047 * interrupts and bottom half handlers won't change it
1048 * from under us. It essentially blocks any incoming
1049 * packets, so that we won't get any new data or any
1050 * packets that change the state of the socket.
1052 * While locked, BH processing will add new packets to
1053 * the backlog queue. This queue is processed by the
1054 * owner of the socket lock right before it is released.
1056 * Since ~2.3.5 it is also exclusive sleep lock serializing
1057 * accesses from user process context.
1059 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1062 * Macro so as to not evaluate some arguments when
1063 * lockdep is not enabled.
1065 * Mark both the sk_lock and the sk_lock.slock as a
1066 * per-address-family lock class.
1068 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1070 sk->sk_lock.owned = 0; \
1071 init_waitqueue_head(&sk->sk_lock.wq); \
1072 spin_lock_init(&(sk)->sk_lock.slock); \
1073 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1074 sizeof((sk)->sk_lock)); \
1075 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1077 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1080 extern void lock_sock_nested(struct sock *sk, int subclass);
1082 static inline void lock_sock(struct sock *sk)
1084 lock_sock_nested(sk, 0);
1087 extern void release_sock(struct sock *sk);
1089 /* BH context may only use the following locking interface. */
1090 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1091 #define bh_lock_sock_nested(__sk) \
1092 spin_lock_nested(&((__sk)->sk_lock.slock), \
1093 SINGLE_DEPTH_NESTING)
1094 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1096 extern bool lock_sock_fast(struct sock *sk);
1098 * unlock_sock_fast - complement of lock_sock_fast
1102 * fast unlock socket for user context.
1103 * If slow mode is on, we call regular release_sock()
1105 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1110 spin_unlock_bh(&sk->sk_lock.slock);
1114 extern struct sock *sk_alloc(struct net *net, int family,
1116 struct proto *prot);
1117 extern void sk_free(struct sock *sk);
1118 extern void sk_release_kernel(struct sock *sk);
1119 extern struct sock *sk_clone(const struct sock *sk,
1120 const gfp_t priority);
1122 extern struct sk_buff *sock_wmalloc(struct sock *sk,
1123 unsigned long size, int force,
1125 extern struct sk_buff *sock_rmalloc(struct sock *sk,
1126 unsigned long size, int force,
1128 extern void sock_wfree(struct sk_buff *skb);
1129 extern void sock_rfree(struct sk_buff *skb);
1130 void sock_efree(struct sk_buff *skb);
1132 extern int sock_setsockopt(struct socket *sock, int level,
1133 int op, char __user *optval,
1134 unsigned int optlen);
1136 extern int sock_getsockopt(struct socket *sock, int level,
1137 int op, char __user *optval,
1138 int __user *optlen);
1139 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1143 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1144 unsigned long header_len,
1145 unsigned long data_len,
1148 extern void *sock_kmalloc(struct sock *sk, int size,
1150 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1151 extern void sk_send_sigurg(struct sock *sk);
1153 #ifdef CONFIG_CGROUPS
1154 extern void sock_update_classid(struct sock *sk);
1156 static inline void sock_update_classid(struct sock *sk)
1162 * Functions to fill in entries in struct proto_ops when a protocol
1163 * does not implement a particular function.
1165 extern int sock_no_bind(struct socket *,
1166 struct sockaddr *, int);
1167 extern int sock_no_connect(struct socket *,
1168 struct sockaddr *, int, int);
1169 extern int sock_no_socketpair(struct socket *,
1171 extern int sock_no_accept(struct socket *,
1172 struct socket *, int);
1173 extern int sock_no_getname(struct socket *,
1174 struct sockaddr *, int *, int);
1175 extern unsigned int sock_no_poll(struct file *, struct socket *,
1176 struct poll_table_struct *);
1177 extern int sock_no_ioctl(struct socket *, unsigned int,
1179 extern int sock_no_listen(struct socket *, int);
1180 extern int sock_no_shutdown(struct socket *, int);
1181 extern int sock_no_getsockopt(struct socket *, int , int,
1182 char __user *, int __user *);
1183 extern int sock_no_setsockopt(struct socket *, int, int,
1184 char __user *, unsigned int);
1185 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1186 struct msghdr *, size_t);
1187 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1188 struct msghdr *, size_t, int);
1189 extern int sock_no_mmap(struct file *file,
1190 struct socket *sock,
1191 struct vm_area_struct *vma);
1192 extern ssize_t sock_no_sendpage(struct socket *sock,
1194 int offset, size_t size,
1198 * Functions to fill in entries in struct proto_ops when a protocol
1199 * uses the inet style.
1201 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1202 char __user *optval, int __user *optlen);
1203 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1204 struct msghdr *msg, size_t size, int flags);
1205 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1206 char __user *optval, unsigned int optlen);
1207 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1208 int optname, char __user *optval, int __user *optlen);
1209 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1210 int optname, char __user *optval, unsigned int optlen);
1212 extern void sk_common_release(struct sock *sk);
1215 * Default socket callbacks and setup code
1218 /* Initialise core socket variables */
1219 extern void sock_init_data(struct socket *sock, struct sock *sk);
1221 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1224 * sk_filter_release - release a socket filter
1225 * @fp: filter to remove
1227 * Remove a filter from a socket and release its resources.
1230 static inline void sk_filter_release(struct sk_filter *fp)
1232 if (atomic_dec_and_test(&fp->refcnt))
1233 call_rcu(&fp->rcu, sk_filter_release_rcu);
1236 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1238 unsigned int size = sk_filter_len(fp);
1240 atomic_sub(size, &sk->sk_omem_alloc);
1241 sk_filter_release(fp);
1244 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1246 atomic_inc(&fp->refcnt);
1247 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1251 * Socket reference counting postulates.
1253 * * Each user of socket SHOULD hold a reference count.
1254 * * Each access point to socket (an hash table bucket, reference from a list,
1255 * running timer, skb in flight MUST hold a reference count.
1256 * * When reference count hits 0, it means it will never increase back.
1257 * * When reference count hits 0, it means that no references from
1258 * outside exist to this socket and current process on current CPU
1259 * is last user and may/should destroy this socket.
1260 * * sk_free is called from any context: process, BH, IRQ. When
1261 * it is called, socket has no references from outside -> sk_free
1262 * may release descendant resources allocated by the socket, but
1263 * to the time when it is called, socket is NOT referenced by any
1264 * hash tables, lists etc.
1265 * * Packets, delivered from outside (from network or from another process)
1266 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1267 * when they sit in queue. Otherwise, packets will leak to hole, when
1268 * socket is looked up by one cpu and unhasing is made by another CPU.
1269 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1270 * (leak to backlog). Packet socket does all the processing inside
1271 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1272 * use separate SMP lock, so that they are prone too.
1275 /* Ungrab socket and destroy it, if it was the last reference. */
1276 static inline void sock_put(struct sock *sk)
1278 if (atomic_dec_and_test(&sk->sk_refcnt))
1282 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1283 unsigned int trim_cap);
1284 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1287 return __sk_receive_skb(sk, skb, nested, 1);
1290 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1292 sk->sk_tx_queue_mapping = tx_queue;
1295 static inline void sk_tx_queue_clear(struct sock *sk)
1297 sk->sk_tx_queue_mapping = -1;
1300 static inline int sk_tx_queue_get(const struct sock *sk)
1302 return sk ? sk->sk_tx_queue_mapping : -1;
1305 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1307 sk_tx_queue_clear(sk);
1308 sk->sk_socket = sock;
1311 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1313 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1314 return &rcu_dereference_raw(sk->sk_wq)->wait;
1316 /* Detach socket from process context.
1317 * Announce socket dead, detach it from wait queue and inode.
1318 * Note that parent inode held reference count on this struct sock,
1319 * we do not release it in this function, because protocol
1320 * probably wants some additional cleanups or even continuing
1321 * to work with this socket (TCP).
1323 static inline void sock_orphan(struct sock *sk)
1325 write_lock_bh(&sk->sk_callback_lock);
1326 sock_set_flag(sk, SOCK_DEAD);
1327 sk_set_socket(sk, NULL);
1329 write_unlock_bh(&sk->sk_callback_lock);
1332 static inline void sock_graft(struct sock *sk, struct socket *parent)
1334 write_lock_bh(&sk->sk_callback_lock);
1335 sk->sk_wq = parent->wq;
1337 sk_set_socket(sk, parent);
1338 security_sock_graft(sk, parent);
1339 write_unlock_bh(&sk->sk_callback_lock);
1342 extern int sock_i_uid(struct sock *sk);
1343 extern unsigned long sock_i_ino(struct sock *sk);
1345 static inline struct dst_entry *
1346 __sk_dst_get(struct sock *sk)
1348 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1349 lockdep_is_held(&sk->sk_lock.slock));
1352 static inline struct dst_entry *
1353 sk_dst_get(struct sock *sk)
1355 struct dst_entry *dst;
1358 dst = rcu_dereference(sk->sk_dst_cache);
1365 extern void sk_reset_txq(struct sock *sk);
1367 static inline void dst_negative_advice(struct sock *sk)
1369 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1371 if (dst && dst->ops->negative_advice) {
1372 ndst = dst->ops->negative_advice(dst);
1375 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1382 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1384 struct dst_entry *old_dst;
1386 sk_tx_queue_clear(sk);
1388 * This can be called while sk is owned by the caller only,
1389 * with no state that can be checked in a rcu_dereference_check() cond
1391 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1392 rcu_assign_pointer(sk->sk_dst_cache, dst);
1393 dst_release(old_dst);
1397 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1399 spin_lock(&sk->sk_dst_lock);
1400 __sk_dst_set(sk, dst);
1401 spin_unlock(&sk->sk_dst_lock);
1405 __sk_dst_reset(struct sock *sk)
1407 __sk_dst_set(sk, NULL);
1411 sk_dst_reset(struct sock *sk)
1413 spin_lock(&sk->sk_dst_lock);
1415 spin_unlock(&sk->sk_dst_lock);
1418 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1420 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1422 static inline int sk_can_gso(const struct sock *sk)
1424 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1427 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1429 static inline void sk_nocaps_add(struct sock *sk, int flags)
1431 sk->sk_route_nocaps |= flags;
1432 sk->sk_route_caps &= ~flags;
1435 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1436 char __user *from, char *to,
1437 int copy, int offset)
1439 if (skb->ip_summed == CHECKSUM_NONE) {
1441 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1444 skb->csum = csum_block_add(skb->csum, csum, offset);
1445 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1446 if (!access_ok(VERIFY_READ, from, copy) ||
1447 __copy_from_user_nocache(to, from, copy))
1449 } else if (copy_from_user(to, from, copy))
1455 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1456 char __user *from, int copy)
1458 int err, offset = skb->len;
1460 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1463 __skb_trim(skb, offset);
1468 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1469 struct sk_buff *skb,
1475 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1481 skb->data_len += copy;
1482 skb->truesize += copy;
1483 sk->sk_wmem_queued += copy;
1484 sk_mem_charge(sk, copy);
1488 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1489 struct sk_buff *skb, struct page *page,
1492 if (skb->ip_summed == CHECKSUM_NONE) {
1494 __wsum csum = csum_and_copy_from_user(from,
1495 page_address(page) + off,
1499 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1500 } else if (copy_from_user(page_address(page) + off, from, copy))
1504 skb->data_len += copy;
1505 skb->truesize += copy;
1506 sk->sk_wmem_queued += copy;
1507 sk_mem_charge(sk, copy);
1512 * sk_wmem_alloc_get - returns write allocations
1515 * Returns sk_wmem_alloc minus initial offset of one
1517 static inline int sk_wmem_alloc_get(const struct sock *sk)
1519 return atomic_read(&sk->sk_wmem_alloc) - 1;
1523 * sk_rmem_alloc_get - returns read allocations
1526 * Returns sk_rmem_alloc
1528 static inline int sk_rmem_alloc_get(const struct sock *sk)
1530 return atomic_read(&sk->sk_rmem_alloc);
1534 * sk_has_allocations - check if allocations are outstanding
1537 * Returns true if socket has write or read allocations
1539 static inline int sk_has_allocations(const struct sock *sk)
1541 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1545 * wq_has_sleeper - check if there are any waiting processes
1546 * @wq: struct socket_wq
1548 * Returns true if socket_wq has waiting processes
1550 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1551 * barrier call. They were added due to the race found within the tcp code.
1553 * Consider following tcp code paths:
1557 * sys_select receive packet
1559 * __add_wait_queue update tp->rcv_nxt
1561 * tp->rcv_nxt check sock_def_readable
1563 * schedule rcu_read_lock();
1564 * wq = rcu_dereference(sk->sk_wq);
1565 * if (wq && waitqueue_active(&wq->wait))
1566 * wake_up_interruptible(&wq->wait)
1570 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1571 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1572 * could then endup calling schedule and sleep forever if there are no more
1573 * data on the socket.
1576 static inline bool wq_has_sleeper(struct socket_wq *wq)
1580 * We need to be sure we are in sync with the
1581 * add_wait_queue modifications to the wait queue.
1583 * This memory barrier is paired in the sock_poll_wait.
1586 return wq && waitqueue_active(&wq->wait);
1590 * sock_poll_wait - place memory barrier behind the poll_wait call.
1592 * @wait_address: socket wait queue
1595 * See the comments in the wq_has_sleeper function.
1597 static inline void sock_poll_wait(struct file *filp,
1598 wait_queue_head_t *wait_address, poll_table *p)
1600 if (p && wait_address) {
1601 poll_wait(filp, wait_address, p);
1603 * We need to be sure we are in sync with the
1604 * socket flags modification.
1606 * This memory barrier is paired in the wq_has_sleeper.
1613 * Queue a received datagram if it will fit. Stream and sequenced
1614 * protocols can't normally use this as they need to fit buffers in
1615 * and play with them.
1617 * Inlined as it's very short and called for pretty much every
1618 * packet ever received.
1621 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1625 skb->destructor = sock_wfree;
1627 * We used to take a refcount on sk, but following operation
1628 * is enough to guarantee sk_free() wont free this sock until
1629 * all in-flight packets are completed
1631 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1634 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1638 skb->destructor = sock_rfree;
1639 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1640 sk_mem_charge(sk, skb->truesize);
1643 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1644 unsigned long expires);
1646 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1648 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1649 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1651 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1654 * Recover an error report and clear atomically
1657 static inline int sock_error(struct sock *sk)
1660 if (likely(!sk->sk_err))
1662 err = xchg(&sk->sk_err, 0);
1666 static inline unsigned long sock_wspace(struct sock *sk)
1670 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1671 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1678 static inline void sk_wake_async(struct sock *sk, int how, int band)
1680 if (sock_flag(sk, SOCK_FASYNC))
1681 sock_wake_async(sk->sk_socket, how, band);
1684 #define SOCK_MIN_SNDBUF 2048
1686 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1687 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1689 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1691 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1693 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1694 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1695 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1699 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1701 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1703 struct page *page = NULL;
1705 page = alloc_pages(sk->sk_allocation, 0);
1707 sk->sk_prot->enter_memory_pressure(sk);
1708 sk_stream_moderate_sndbuf(sk);
1714 * Default write policy as shown to user space via poll/select/SIGIO
1716 static inline int sock_writeable(const struct sock *sk)
1718 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1721 static inline gfp_t gfp_any(void)
1723 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1726 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1728 return noblock ? 0 : sk->sk_rcvtimeo;
1731 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1733 return noblock ? 0 : sk->sk_sndtimeo;
1736 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1738 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1741 /* Alas, with timeout socket operations are not restartable.
1742 * Compare this to poll().
1744 static inline int sock_intr_errno(long timeo)
1746 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1749 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1750 struct sk_buff *skb);
1752 static __inline__ void
1753 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1755 ktime_t kt = skb->tstamp;
1756 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1759 * generate control messages if
1760 * - receive time stamping in software requested (SOCK_RCVTSTAMP
1761 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
1762 * - software time stamp available and wanted
1763 * (SOCK_TIMESTAMPING_SOFTWARE)
1764 * - hardware time stamps available and wanted
1765 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
1766 * SOCK_TIMESTAMPING_RAW_HARDWARE)
1768 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
1769 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
1770 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
1771 (hwtstamps->hwtstamp.tv64 &&
1772 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
1773 (hwtstamps->syststamp.tv64 &&
1774 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
1775 __sock_recv_timestamp(msg, sk, skb);
1780 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
1781 struct sk_buff *skb);
1783 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
1784 struct sk_buff *skb)
1786 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
1787 (1UL << SOCK_RCVTSTAMP) | \
1788 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
1789 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
1790 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
1791 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
1793 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
1794 __sock_recv_ts_and_drops(msg, sk, skb);
1796 sk->sk_stamp = skb->tstamp;
1800 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
1801 * @sk: socket sending this packet
1802 * @tx_flags: filled with instructions for time stamping
1804 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
1805 * parameters are invalid.
1807 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
1810 * sk_eat_skb - Release a skb if it is no longer needed
1811 * @sk: socket to eat this skb from
1812 * @skb: socket buffer to eat
1813 * @copied_early: flag indicating whether DMA operations copied this data early
1815 * This routine must be called with interrupts disabled or with the socket
1816 * locked so that the sk_buff queue operation is ok.
1818 #ifdef CONFIG_NET_DMA
1819 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1821 __skb_unlink(skb, &sk->sk_receive_queue);
1825 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
1828 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1830 __skb_unlink(skb, &sk->sk_receive_queue);
1836 struct net *sock_net(const struct sock *sk)
1838 return read_pnet(&sk->sk_net);
1842 void sock_net_set(struct sock *sk, struct net *net)
1844 write_pnet(&sk->sk_net, net);
1848 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
1849 * They should not hold a reference to a namespace in order to allow
1851 * Sockets after sk_change_net should be released using sk_release_kernel
1853 static inline void sk_change_net(struct sock *sk, struct net *net)
1855 put_net(sock_net(sk));
1856 sock_net_set(sk, hold_net(net));
1859 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
1861 if (unlikely(skb->sk)) {
1862 struct sock *sk = skb->sk;
1864 skb->destructor = NULL;
1871 extern void sock_enable_timestamp(struct sock *sk, int flag);
1872 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1873 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1876 * Enable debug/info messages
1878 extern int net_msg_warn;
1879 #define NETDEBUG(fmt, args...) \
1880 do { if (net_msg_warn) printk(fmt,##args); } while (0)
1882 #define LIMIT_NETDEBUG(fmt, args...) \
1883 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
1885 extern __u32 sysctl_wmem_max;
1886 extern __u32 sysctl_rmem_max;
1888 extern void sk_init(void);
1890 extern int sysctl_optmem_max;
1892 extern __u32 sysctl_wmem_default;
1893 extern __u32 sysctl_rmem_default;
1895 #endif /* _SOCK_H */