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/kernel.h>
44 #include <linux/list.h>
45 #include <linux/timer.h>
46 #include <linux/cache.h>
47 #include <linux/module.h>
48 #include <linux/lockdep.h>
49 #include <linux/netdevice.h>
50 #include <linux/pcounter.h>
51 #include <linux/skbuff.h> /* struct sk_buff */
53 #include <linux/security.h>
55 #include <linux/filter.h>
57 #include <asm/atomic.h>
59 #include <net/checksum.h>
62 * This structure really needs to be cleaned up.
63 * Most of it is for TCP, and not used by any of
64 * the other protocols.
67 /* Define this to get the SOCK_DBG debugging facility. */
68 #define SOCK_DEBUGGING
70 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
71 printk(KERN_DEBUG msg); } while (0)
73 /* Validate arguments and do nothing */
74 static void inline int __attribute__ ((format (printf, 2, 3)))
75 SOCK_DEBUG(struct sock *sk, const char *msg, ...)
80 /* This is the per-socket lock. The spinlock provides a synchronization
81 * between user contexts and software interrupt processing, whereas the
82 * mini-semaphore synchronizes multiple users amongst themselves.
89 * We express the mutex-alike socket_lock semantics
90 * to the lock validator by explicitly managing
91 * the slock as a lock variant (in addition to
94 #ifdef CONFIG_DEBUG_LOCK_ALLOC
95 struct lockdep_map dep_map;
104 * struct sock_common - minimal network layer representation of sockets
105 * @skc_family: network address family
106 * @skc_state: Connection state
107 * @skc_reuse: %SO_REUSEADDR setting
108 * @skc_bound_dev_if: bound device index if != 0
109 * @skc_node: main hash linkage for various protocol lookup tables
110 * @skc_bind_node: bind hash linkage for various protocol lookup tables
111 * @skc_refcnt: reference count
112 * @skc_hash: hash value used with various protocol lookup tables
113 * @skc_prot: protocol handlers inside a network family
114 * @skc_net: reference to the network namespace of this socket
116 * This is the minimal network layer representation of sockets, the header
117 * for struct sock and struct inet_timewait_sock.
120 unsigned short skc_family;
121 volatile unsigned char skc_state;
122 unsigned char skc_reuse;
123 int skc_bound_dev_if;
124 struct hlist_node skc_node;
125 struct hlist_node skc_bind_node;
127 unsigned int skc_hash;
128 struct proto *skc_prot;
135 * struct sock - network layer representation of sockets
136 * @__sk_common: shared layout with inet_timewait_sock
137 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
138 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
139 * @sk_lock: synchronizer
140 * @sk_rcvbuf: size of receive buffer in bytes
141 * @sk_sleep: sock wait queue
142 * @sk_dst_cache: destination cache
143 * @sk_dst_lock: destination cache lock
144 * @sk_policy: flow policy
145 * @sk_rmem_alloc: receive queue bytes committed
146 * @sk_receive_queue: incoming packets
147 * @sk_wmem_alloc: transmit queue bytes committed
148 * @sk_write_queue: Packet sending queue
149 * @sk_async_wait_queue: DMA copied packets
150 * @sk_omem_alloc: "o" is "option" or "other"
151 * @sk_wmem_queued: persistent queue size
152 * @sk_forward_alloc: space allocated forward
153 * @sk_allocation: allocation mode
154 * @sk_sndbuf: size of send buffer in bytes
155 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
156 * %SO_OOBINLINE settings
157 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
158 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
159 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
160 * @sk_gso_max_size: Maximum GSO segment size to build
161 * @sk_lingertime: %SO_LINGER l_linger setting
162 * @sk_backlog: always used with the per-socket spinlock held
163 * @sk_callback_lock: used with the callbacks in the end of this struct
164 * @sk_error_queue: rarely used
165 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
166 * IPV6_ADDRFORM for instance)
167 * @sk_err: last error
168 * @sk_err_soft: errors that don't cause failure but are the cause of a
169 * persistent failure not just 'timed out'
170 * @sk_drops: raw drops counter
171 * @sk_ack_backlog: current listen backlog
172 * @sk_max_ack_backlog: listen backlog set in listen()
173 * @sk_priority: %SO_PRIORITY setting
174 * @sk_type: socket type (%SOCK_STREAM, etc)
175 * @sk_protocol: which protocol this socket belongs in this network family
176 * @sk_peercred: %SO_PEERCRED setting
177 * @sk_rcvlowat: %SO_RCVLOWAT setting
178 * @sk_rcvtimeo: %SO_RCVTIMEO setting
179 * @sk_sndtimeo: %SO_SNDTIMEO setting
180 * @sk_filter: socket filtering instructions
181 * @sk_protinfo: private area, net family specific, when not using slab
182 * @sk_timer: sock cleanup timer
183 * @sk_stamp: time stamp of last packet received
184 * @sk_socket: Identd and reporting IO signals
185 * @sk_user_data: RPC layer private data
186 * @sk_sndmsg_page: cached page for sendmsg
187 * @sk_sndmsg_off: cached offset for sendmsg
188 * @sk_send_head: front of stuff to transmit
189 * @sk_security: used by security modules
190 * @sk_mark: generic packet mark
191 * @sk_write_pending: a write to stream socket waits to start
192 * @sk_state_change: callback to indicate change in the state of the sock
193 * @sk_data_ready: callback to indicate there is data to be processed
194 * @sk_write_space: callback to indicate there is bf sending space available
195 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
196 * @sk_backlog_rcv: callback to process the backlog
197 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
201 * Now struct inet_timewait_sock also uses sock_common, so please just
202 * don't add nothing before this first member (__sk_common) --acme
204 struct sock_common __sk_common;
205 #define sk_family __sk_common.skc_family
206 #define sk_state __sk_common.skc_state
207 #define sk_reuse __sk_common.skc_reuse
208 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
209 #define sk_node __sk_common.skc_node
210 #define sk_bind_node __sk_common.skc_bind_node
211 #define sk_refcnt __sk_common.skc_refcnt
212 #define sk_hash __sk_common.skc_hash
213 #define sk_prot __sk_common.skc_prot
214 #define sk_net __sk_common.skc_net
215 unsigned char sk_shutdown : 2,
218 unsigned char sk_protocol;
219 unsigned short sk_type;
221 socket_lock_t sk_lock;
223 * The backlog queue is special, it is always used with
224 * the per-socket spinlock held and requires low latency
225 * access. Therefore we special case it's implementation.
228 struct sk_buff *head;
229 struct sk_buff *tail;
231 wait_queue_head_t *sk_sleep;
232 struct dst_entry *sk_dst_cache;
233 struct xfrm_policy *sk_policy[2];
234 rwlock_t sk_dst_lock;
235 atomic_t sk_rmem_alloc;
236 atomic_t sk_wmem_alloc;
237 atomic_t sk_omem_alloc;
239 struct sk_buff_head sk_receive_queue;
240 struct sk_buff_head sk_write_queue;
241 struct sk_buff_head sk_async_wait_queue;
243 int sk_forward_alloc;
247 unsigned int sk_gso_max_size;
249 unsigned long sk_flags;
250 unsigned long sk_lingertime;
251 struct sk_buff_head sk_error_queue;
252 struct proto *sk_prot_creator;
253 rwlock_t sk_callback_lock;
257 unsigned short sk_ack_backlog;
258 unsigned short sk_max_ack_backlog;
260 struct ucred sk_peercred;
263 struct sk_filter *sk_filter;
265 struct timer_list sk_timer;
267 struct socket *sk_socket;
269 struct page *sk_sndmsg_page;
270 struct sk_buff *sk_send_head;
272 int sk_write_pending;
275 /* XXX 4 bytes hole on 64 bit */
276 void (*sk_state_change)(struct sock *sk);
277 void (*sk_data_ready)(struct sock *sk, int bytes);
278 void (*sk_write_space)(struct sock *sk);
279 void (*sk_error_report)(struct sock *sk);
280 int (*sk_backlog_rcv)(struct sock *sk,
281 struct sk_buff *skb);
282 void (*sk_destruct)(struct sock *sk);
286 * Hashed lists helper routines
288 static inline struct sock *__sk_head(const struct hlist_head *head)
290 return hlist_entry(head->first, struct sock, sk_node);
293 static inline struct sock *sk_head(const struct hlist_head *head)
295 return hlist_empty(head) ? NULL : __sk_head(head);
298 static inline struct sock *sk_next(const struct sock *sk)
300 return sk->sk_node.next ?
301 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
304 static inline int sk_unhashed(const struct sock *sk)
306 return hlist_unhashed(&sk->sk_node);
309 static inline int sk_hashed(const struct sock *sk)
311 return !sk_unhashed(sk);
314 static __inline__ void sk_node_init(struct hlist_node *node)
319 static __inline__ void __sk_del_node(struct sock *sk)
321 __hlist_del(&sk->sk_node);
324 static __inline__ int __sk_del_node_init(struct sock *sk)
328 sk_node_init(&sk->sk_node);
334 /* Grab socket reference count. This operation is valid only
335 when sk is ALREADY grabbed f.e. it is found in hash table
336 or a list and the lookup is made under lock preventing hash table
340 static inline void sock_hold(struct sock *sk)
342 atomic_inc(&sk->sk_refcnt);
345 /* Ungrab socket in the context, which assumes that socket refcnt
346 cannot hit zero, f.e. it is true in context of any socketcall.
348 static inline void __sock_put(struct sock *sk)
350 atomic_dec(&sk->sk_refcnt);
353 static __inline__ int sk_del_node_init(struct sock *sk)
355 int rc = __sk_del_node_init(sk);
358 /* paranoid for a while -acme */
359 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
365 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
367 hlist_add_head(&sk->sk_node, list);
370 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
373 __sk_add_node(sk, list);
376 static __inline__ void __sk_del_bind_node(struct sock *sk)
378 __hlist_del(&sk->sk_bind_node);
381 static __inline__ void sk_add_bind_node(struct sock *sk,
382 struct hlist_head *list)
384 hlist_add_head(&sk->sk_bind_node, list);
387 #define sk_for_each(__sk, node, list) \
388 hlist_for_each_entry(__sk, node, list, sk_node)
389 #define sk_for_each_from(__sk, node) \
390 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
391 hlist_for_each_entry_from(__sk, node, sk_node)
392 #define sk_for_each_continue(__sk, node) \
393 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
394 hlist_for_each_entry_continue(__sk, node, sk_node)
395 #define sk_for_each_safe(__sk, node, tmp, list) \
396 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
397 #define sk_for_each_bound(__sk, node, list) \
398 hlist_for_each_entry(__sk, node, list, sk_bind_node)
411 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
412 SOCK_DBG, /* %SO_DEBUG setting */
413 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
414 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
415 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
416 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
419 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
421 nsk->sk_flags = osk->sk_flags;
424 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
426 __set_bit(flag, &sk->sk_flags);
429 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
431 __clear_bit(flag, &sk->sk_flags);
434 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
436 return test_bit(flag, &sk->sk_flags);
439 static inline void sk_acceptq_removed(struct sock *sk)
441 sk->sk_ack_backlog--;
444 static inline void sk_acceptq_added(struct sock *sk)
446 sk->sk_ack_backlog++;
449 static inline int sk_acceptq_is_full(struct sock *sk)
451 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
455 * Compute minimal free write space needed to queue new packets.
457 static inline int sk_stream_min_wspace(struct sock *sk)
459 return sk->sk_wmem_queued >> 1;
462 static inline int sk_stream_wspace(struct sock *sk)
464 return sk->sk_sndbuf - sk->sk_wmem_queued;
467 extern void sk_stream_write_space(struct sock *sk);
469 static inline int sk_stream_memory_free(struct sock *sk)
471 return sk->sk_wmem_queued < sk->sk_sndbuf;
474 /* The per-socket spinlock must be held here. */
475 static inline void sk_add_backlog(struct sock *sk, struct sk_buff *skb)
477 if (!sk->sk_backlog.tail) {
478 sk->sk_backlog.head = sk->sk_backlog.tail = skb;
480 sk->sk_backlog.tail->next = skb;
481 sk->sk_backlog.tail = skb;
486 #define sk_wait_event(__sk, __timeo, __condition) \
488 release_sock(__sk); \
489 __rc = __condition; \
491 *(__timeo) = schedule_timeout(*(__timeo)); \
494 __rc = __condition; \
498 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
499 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
500 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
501 extern int sk_stream_error(struct sock *sk, int flags, int err);
502 extern void sk_stream_kill_queues(struct sock *sk);
504 extern int sk_wait_data(struct sock *sk, long *timeo);
506 struct request_sock_ops;
507 struct timewait_sock_ops;
508 struct inet_hashinfo;
511 /* Networking protocol blocks we attach to sockets.
512 * socket layer -> transport layer interface
513 * transport -> network interface is defined by struct inet_proto
516 void (*close)(struct sock *sk,
518 int (*connect)(struct sock *sk,
519 struct sockaddr *uaddr,
521 int (*disconnect)(struct sock *sk, int flags);
523 struct sock * (*accept) (struct sock *sk, int flags, int *err);
525 int (*ioctl)(struct sock *sk, int cmd,
527 int (*init)(struct sock *sk);
528 int (*destroy)(struct sock *sk);
529 void (*shutdown)(struct sock *sk, int how);
530 int (*setsockopt)(struct sock *sk, int level,
531 int optname, char __user *optval,
533 int (*getsockopt)(struct sock *sk, int level,
534 int optname, char __user *optval,
536 int (*compat_setsockopt)(struct sock *sk,
538 int optname, char __user *optval,
540 int (*compat_getsockopt)(struct sock *sk,
542 int optname, char __user *optval,
544 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
545 struct msghdr *msg, size_t len);
546 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
548 size_t len, int noblock, int flags,
550 int (*sendpage)(struct sock *sk, struct page *page,
551 int offset, size_t size, int flags);
552 int (*bind)(struct sock *sk,
553 struct sockaddr *uaddr, int addr_len);
555 int (*backlog_rcv) (struct sock *sk,
556 struct sk_buff *skb);
558 /* Keeping track of sk's, looking them up, and port selection methods. */
559 void (*hash)(struct sock *sk);
560 void (*unhash)(struct sock *sk);
561 int (*get_port)(struct sock *sk, unsigned short snum);
563 /* Keeping track of sockets in use */
564 #ifdef CONFIG_PROC_FS
565 struct pcounter inuse;
568 /* Memory pressure */
569 void (*enter_memory_pressure)(void);
570 atomic_t *memory_allocated; /* Current allocated memory. */
571 atomic_t *sockets_allocated; /* Current number of sockets. */
573 * Pressure flag: try to collapse.
574 * Technical note: it is used by multiple contexts non atomically.
575 * All the __sk_mem_schedule() is of this nature: accounting
576 * is strict, actions are advisory and have some latency.
578 int *memory_pressure;
584 struct kmem_cache *slab;
585 unsigned int obj_size;
587 atomic_t *orphan_count;
589 struct request_sock_ops *rsk_prot;
590 struct timewait_sock_ops *twsk_prot;
593 struct inet_hashinfo *hashinfo;
594 struct hlist_head *udp_hash;
595 struct raw_hashinfo *raw_hash;
598 struct module *owner;
602 struct list_head node;
603 #ifdef SOCK_REFCNT_DEBUG
608 extern int proto_register(struct proto *prot, int alloc_slab);
609 extern void proto_unregister(struct proto *prot);
611 #ifdef SOCK_REFCNT_DEBUG
612 static inline void sk_refcnt_debug_inc(struct sock *sk)
614 atomic_inc(&sk->sk_prot->socks);
617 static inline void sk_refcnt_debug_dec(struct sock *sk)
619 atomic_dec(&sk->sk_prot->socks);
620 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
621 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
624 static inline void sk_refcnt_debug_release(const struct sock *sk)
626 if (atomic_read(&sk->sk_refcnt) != 1)
627 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
628 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
630 #else /* SOCK_REFCNT_DEBUG */
631 #define sk_refcnt_debug_inc(sk) do { } while (0)
632 #define sk_refcnt_debug_dec(sk) do { } while (0)
633 #define sk_refcnt_debug_release(sk) do { } while (0)
634 #endif /* SOCK_REFCNT_DEBUG */
637 #ifdef CONFIG_PROC_FS
638 # define DEFINE_PROTO_INUSE(NAME) DEFINE_PCOUNTER(NAME)
639 # define REF_PROTO_INUSE(NAME) PCOUNTER_MEMBER_INITIALIZER(NAME, .inuse)
640 /* Called with local bh disabled */
641 static inline void sock_prot_inuse_add(struct proto *prot, int inc)
643 pcounter_add(&prot->inuse, inc);
645 static inline int sock_prot_inuse_init(struct proto *proto)
647 return pcounter_alloc(&proto->inuse);
649 static inline int sock_prot_inuse_get(struct proto *proto)
651 return pcounter_getval(&proto->inuse);
653 static inline void sock_prot_inuse_free(struct proto *proto)
655 pcounter_free(&proto->inuse);
658 # define DEFINE_PROTO_INUSE(NAME)
659 # define REF_PROTO_INUSE(NAME)
660 static void inline sock_prot_inuse_add(struct proto *prot, int inc)
663 static int inline sock_prot_inuse_init(struct proto *proto)
667 static void inline sock_prot_inuse_free(struct proto *proto)
673 /* With per-bucket locks this operation is not-atomic, so that
674 * this version is not worse.
676 static inline void __sk_prot_rehash(struct sock *sk)
678 sk->sk_prot->unhash(sk);
679 sk->sk_prot->hash(sk);
682 /* About 10 seconds */
683 #define SOCK_DESTROY_TIME (10*HZ)
685 /* Sockets 0-1023 can't be bound to unless you are superuser */
686 #define PROT_SOCK 1024
688 #define SHUTDOWN_MASK 3
689 #define RCV_SHUTDOWN 1
690 #define SEND_SHUTDOWN 2
692 #define SOCK_SNDBUF_LOCK 1
693 #define SOCK_RCVBUF_LOCK 2
694 #define SOCK_BINDADDR_LOCK 4
695 #define SOCK_BINDPORT_LOCK 8
697 /* sock_iocb: used to kick off async processing of socket ios */
699 struct list_head list;
705 struct scm_cookie *scm;
706 struct msghdr *msg, async_msg;
710 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
712 return (struct sock_iocb *)iocb->private;
715 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
720 struct socket_alloc {
721 struct socket socket;
722 struct inode vfs_inode;
725 static inline struct socket *SOCKET_I(struct inode *inode)
727 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
730 static inline struct inode *SOCK_INODE(struct socket *socket)
732 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
736 * Functions for memory accounting
738 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
739 extern void __sk_mem_reclaim(struct sock *sk);
741 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
742 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
743 #define SK_MEM_SEND 0
744 #define SK_MEM_RECV 1
746 static inline int sk_mem_pages(int amt)
748 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
751 static inline int sk_has_account(struct sock *sk)
753 /* return true if protocol supports memory accounting */
754 return !!sk->sk_prot->memory_allocated;
757 static inline int sk_wmem_schedule(struct sock *sk, int size)
759 if (!sk_has_account(sk))
761 return size <= sk->sk_forward_alloc ||
762 __sk_mem_schedule(sk, size, SK_MEM_SEND);
765 static inline int sk_rmem_schedule(struct sock *sk, int size)
767 if (!sk_has_account(sk))
769 return size <= sk->sk_forward_alloc ||
770 __sk_mem_schedule(sk, size, SK_MEM_RECV);
773 static inline void sk_mem_reclaim(struct sock *sk)
775 if (!sk_has_account(sk))
777 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
778 __sk_mem_reclaim(sk);
781 static inline void sk_mem_reclaim_partial(struct sock *sk)
783 if (!sk_has_account(sk))
785 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
786 __sk_mem_reclaim(sk);
789 static inline void sk_mem_charge(struct sock *sk, int size)
791 if (!sk_has_account(sk))
793 sk->sk_forward_alloc -= size;
796 static inline void sk_mem_uncharge(struct sock *sk, int size)
798 if (!sk_has_account(sk))
800 sk->sk_forward_alloc += size;
803 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
805 skb_truesize_check(skb);
806 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
807 sk->sk_wmem_queued -= skb->truesize;
808 sk_mem_uncharge(sk, skb->truesize);
812 /* Used by processes to "lock" a socket state, so that
813 * interrupts and bottom half handlers won't change it
814 * from under us. It essentially blocks any incoming
815 * packets, so that we won't get any new data or any
816 * packets that change the state of the socket.
818 * While locked, BH processing will add new packets to
819 * the backlog queue. This queue is processed by the
820 * owner of the socket lock right before it is released.
822 * Since ~2.3.5 it is also exclusive sleep lock serializing
823 * accesses from user process context.
825 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
828 * Macro so as to not evaluate some arguments when
829 * lockdep is not enabled.
831 * Mark both the sk_lock and the sk_lock.slock as a
832 * per-address-family lock class.
834 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
836 sk->sk_lock.owned = 0; \
837 init_waitqueue_head(&sk->sk_lock.wq); \
838 spin_lock_init(&(sk)->sk_lock.slock); \
839 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
840 sizeof((sk)->sk_lock)); \
841 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
843 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
846 extern void lock_sock_nested(struct sock *sk, int subclass);
848 static inline void lock_sock(struct sock *sk)
850 lock_sock_nested(sk, 0);
853 extern void release_sock(struct sock *sk);
855 /* BH context may only use the following locking interface. */
856 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
857 #define bh_lock_sock_nested(__sk) \
858 spin_lock_nested(&((__sk)->sk_lock.slock), \
859 SINGLE_DEPTH_NESTING)
860 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
862 extern struct sock *sk_alloc(struct net *net, int family,
865 extern void sk_free(struct sock *sk);
866 extern void sk_release_kernel(struct sock *sk);
867 extern struct sock *sk_clone(const struct sock *sk,
868 const gfp_t priority);
870 extern struct sk_buff *sock_wmalloc(struct sock *sk,
871 unsigned long size, int force,
873 extern struct sk_buff *sock_rmalloc(struct sock *sk,
874 unsigned long size, int force,
876 extern void sock_wfree(struct sk_buff *skb);
877 extern void sock_rfree(struct sk_buff *skb);
879 extern int sock_setsockopt(struct socket *sock, int level,
880 int op, char __user *optval,
883 extern int sock_getsockopt(struct socket *sock, int level,
884 int op, char __user *optval,
886 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
890 extern void *sock_kmalloc(struct sock *sk, int size,
892 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
893 extern void sk_send_sigurg(struct sock *sk);
896 * Functions to fill in entries in struct proto_ops when a protocol
897 * does not implement a particular function.
899 extern int sock_no_bind(struct socket *,
900 struct sockaddr *, int);
901 extern int sock_no_connect(struct socket *,
902 struct sockaddr *, int, int);
903 extern int sock_no_socketpair(struct socket *,
905 extern int sock_no_accept(struct socket *,
906 struct socket *, int);
907 extern int sock_no_getname(struct socket *,
908 struct sockaddr *, int *, int);
909 extern unsigned int sock_no_poll(struct file *, struct socket *,
910 struct poll_table_struct *);
911 extern int sock_no_ioctl(struct socket *, unsigned int,
913 extern int sock_no_listen(struct socket *, int);
914 extern int sock_no_shutdown(struct socket *, int);
915 extern int sock_no_getsockopt(struct socket *, int , int,
916 char __user *, int __user *);
917 extern int sock_no_setsockopt(struct socket *, int, int,
919 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
920 struct msghdr *, size_t);
921 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
922 struct msghdr *, size_t, int);
923 extern int sock_no_mmap(struct file *file,
925 struct vm_area_struct *vma);
926 extern ssize_t sock_no_sendpage(struct socket *sock,
928 int offset, size_t size,
932 * Functions to fill in entries in struct proto_ops when a protocol
933 * uses the inet style.
935 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
936 char __user *optval, int __user *optlen);
937 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
938 struct msghdr *msg, size_t size, int flags);
939 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
940 char __user *optval, int optlen);
941 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
942 int optname, char __user *optval, int __user *optlen);
943 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
944 int optname, char __user *optval, int optlen);
946 extern void sk_common_release(struct sock *sk);
949 * Default socket callbacks and setup code
952 /* Initialise core socket variables */
953 extern void sock_init_data(struct socket *sock, struct sock *sk);
956 * sk_filter - run a packet through a socket filter
957 * @sk: sock associated with &sk_buff
958 * @skb: buffer to filter
959 * @needlock: set to 1 if the sock is not locked by caller.
961 * Run the filter code and then cut skb->data to correct size returned by
962 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
963 * than pkt_len we keep whole skb->data. This is the socket level
964 * wrapper to sk_run_filter. It returns 0 if the packet should
965 * be accepted or -EPERM if the packet should be tossed.
969 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
972 struct sk_filter *filter;
974 err = security_sock_rcv_skb(sk, skb);
979 filter = rcu_dereference(sk->sk_filter);
981 unsigned int pkt_len = sk_run_filter(skb, filter->insns,
983 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
985 rcu_read_unlock_bh();
991 * sk_filter_release: Release a socket filter
993 * @fp: filter to remove
995 * Remove a filter from a socket and release its resources.
998 static inline void sk_filter_release(struct sk_filter *fp)
1000 if (atomic_dec_and_test(&fp->refcnt))
1004 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1006 unsigned int size = sk_filter_len(fp);
1008 atomic_sub(size, &sk->sk_omem_alloc);
1009 sk_filter_release(fp);
1012 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1014 atomic_inc(&fp->refcnt);
1015 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1019 * Socket reference counting postulates.
1021 * * Each user of socket SHOULD hold a reference count.
1022 * * Each access point to socket (an hash table bucket, reference from a list,
1023 * running timer, skb in flight MUST hold a reference count.
1024 * * When reference count hits 0, it means it will never increase back.
1025 * * When reference count hits 0, it means that no references from
1026 * outside exist to this socket and current process on current CPU
1027 * is last user and may/should destroy this socket.
1028 * * sk_free is called from any context: process, BH, IRQ. When
1029 * it is called, socket has no references from outside -> sk_free
1030 * may release descendant resources allocated by the socket, but
1031 * to the time when it is called, socket is NOT referenced by any
1032 * hash tables, lists etc.
1033 * * Packets, delivered from outside (from network or from another process)
1034 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1035 * when they sit in queue. Otherwise, packets will leak to hole, when
1036 * socket is looked up by one cpu and unhasing is made by another CPU.
1037 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1038 * (leak to backlog). Packet socket does all the processing inside
1039 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1040 * use separate SMP lock, so that they are prone too.
1043 /* Ungrab socket and destroy it, if it was the last reference. */
1044 static inline void sock_put(struct sock *sk)
1046 if (atomic_dec_and_test(&sk->sk_refcnt))
1050 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1053 /* Detach socket from process context.
1054 * Announce socket dead, detach it from wait queue and inode.
1055 * Note that parent inode held reference count on this struct sock,
1056 * we do not release it in this function, because protocol
1057 * probably wants some additional cleanups or even continuing
1058 * to work with this socket (TCP).
1060 static inline void sock_orphan(struct sock *sk)
1062 write_lock_bh(&sk->sk_callback_lock);
1063 sock_set_flag(sk, SOCK_DEAD);
1064 sk->sk_socket = NULL;
1065 sk->sk_sleep = NULL;
1066 write_unlock_bh(&sk->sk_callback_lock);
1069 static inline void sock_graft(struct sock *sk, struct socket *parent)
1071 write_lock_bh(&sk->sk_callback_lock);
1072 sk->sk_sleep = &parent->wait;
1074 sk->sk_socket = parent;
1075 security_sock_graft(sk, parent);
1076 write_unlock_bh(&sk->sk_callback_lock);
1079 extern int sock_i_uid(struct sock *sk);
1080 extern unsigned long sock_i_ino(struct sock *sk);
1082 static inline struct dst_entry *
1083 __sk_dst_get(struct sock *sk)
1085 return sk->sk_dst_cache;
1088 static inline struct dst_entry *
1089 sk_dst_get(struct sock *sk)
1091 struct dst_entry *dst;
1093 read_lock(&sk->sk_dst_lock);
1094 dst = sk->sk_dst_cache;
1097 read_unlock(&sk->sk_dst_lock);
1102 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1104 struct dst_entry *old_dst;
1106 old_dst = sk->sk_dst_cache;
1107 sk->sk_dst_cache = dst;
1108 dst_release(old_dst);
1112 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1114 write_lock(&sk->sk_dst_lock);
1115 __sk_dst_set(sk, dst);
1116 write_unlock(&sk->sk_dst_lock);
1120 __sk_dst_reset(struct sock *sk)
1122 struct dst_entry *old_dst;
1124 old_dst = sk->sk_dst_cache;
1125 sk->sk_dst_cache = NULL;
1126 dst_release(old_dst);
1130 sk_dst_reset(struct sock *sk)
1132 write_lock(&sk->sk_dst_lock);
1134 write_unlock(&sk->sk_dst_lock);
1137 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1139 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1141 static inline int sk_can_gso(const struct sock *sk)
1143 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1146 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1148 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1149 struct sk_buff *skb, struct page *page,
1152 if (skb->ip_summed == CHECKSUM_NONE) {
1154 __wsum csum = csum_and_copy_from_user(from,
1155 page_address(page) + off,
1159 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1160 } else if (copy_from_user(page_address(page) + off, from, copy))
1164 skb->data_len += copy;
1165 skb->truesize += copy;
1166 sk->sk_wmem_queued += copy;
1167 sk_mem_charge(sk, copy);
1172 * Queue a received datagram if it will fit. Stream and sequenced
1173 * protocols can't normally use this as they need to fit buffers in
1174 * and play with them.
1176 * Inlined as it's very short and called for pretty much every
1177 * packet ever received.
1180 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1184 skb->destructor = sock_wfree;
1185 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1188 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1191 skb->destructor = sock_rfree;
1192 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1193 sk_mem_charge(sk, skb->truesize);
1196 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1197 unsigned long expires);
1199 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1201 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1203 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
1205 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
1206 number of warnings when compiling with -W --ANK
1208 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1209 (unsigned)sk->sk_rcvbuf)
1211 skb_set_owner_r(skb, sk);
1212 skb_queue_tail(&sk->sk_error_queue, skb);
1213 if (!sock_flag(sk, SOCK_DEAD))
1214 sk->sk_data_ready(sk, skb->len);
1219 * Recover an error report and clear atomically
1222 static inline int sock_error(struct sock *sk)
1225 if (likely(!sk->sk_err))
1227 err = xchg(&sk->sk_err, 0);
1231 static inline unsigned long sock_wspace(struct sock *sk)
1235 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1236 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1243 static inline void sk_wake_async(struct sock *sk, int how, int band)
1245 if (sk->sk_socket && sk->sk_socket->fasync_list)
1246 sock_wake_async(sk->sk_socket, how, band);
1249 #define SOCK_MIN_SNDBUF 2048
1250 #define SOCK_MIN_RCVBUF 256
1252 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1254 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1255 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1256 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1260 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1262 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1264 struct page *page = NULL;
1266 page = alloc_pages(sk->sk_allocation, 0);
1268 sk->sk_prot->enter_memory_pressure();
1269 sk_stream_moderate_sndbuf(sk);
1275 * Default write policy as shown to user space via poll/select/SIGIO
1277 static inline int sock_writeable(const struct sock *sk)
1279 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1282 static inline gfp_t gfp_any(void)
1284 return in_atomic() ? GFP_ATOMIC : GFP_KERNEL;
1287 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1289 return noblock ? 0 : sk->sk_rcvtimeo;
1292 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1294 return noblock ? 0 : sk->sk_sndtimeo;
1297 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1299 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1302 /* Alas, with timeout socket operations are not restartable.
1303 * Compare this to poll().
1305 static inline int sock_intr_errno(long timeo)
1307 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1310 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1311 struct sk_buff *skb);
1313 static __inline__ void
1314 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1316 ktime_t kt = skb->tstamp;
1318 if (sock_flag(sk, SOCK_RCVTSTAMP))
1319 __sock_recv_timestamp(msg, sk, skb);
1325 * sk_eat_skb - Release a skb if it is no longer needed
1326 * @sk: socket to eat this skb from
1327 * @skb: socket buffer to eat
1328 * @copied_early: flag indicating whether DMA operations copied this data early
1330 * This routine must be called with interrupts disabled or with the socket
1331 * locked so that the sk_buff queue operation is ok.
1333 #ifdef CONFIG_NET_DMA
1334 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1336 __skb_unlink(skb, &sk->sk_receive_queue);
1340 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
1343 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1345 __skb_unlink(skb, &sk->sk_receive_queue);
1351 struct net *sock_net(const struct sock *sk)
1353 #ifdef CONFIG_NET_NS
1361 void sock_net_set(struct sock *sk, const struct net *net)
1363 #ifdef CONFIG_NET_NS
1369 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
1370 * They should not hold a referrence to a namespace in order to allow
1372 * Sockets after sk_change_net should be released using sk_release_kernel
1374 static inline void sk_change_net(struct sock *sk, struct net *net)
1376 put_net(sock_net(sk));
1377 sock_net_set(sk, net);
1380 extern void sock_enable_timestamp(struct sock *sk);
1381 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1382 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1385 * Enable debug/info messages
1387 extern int net_msg_warn;
1388 #define NETDEBUG(fmt, args...) \
1389 do { if (net_msg_warn) printk(fmt,##args); } while (0)
1391 #define LIMIT_NETDEBUG(fmt, args...) \
1392 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
1395 * Macros for sleeping on a socket. Use them like this:
1397 * SOCK_SLEEP_PRE(sk)
1400 * SOCK_SLEEP_POST(sk)
1402 * N.B. These are now obsolete and were, afaik, only ever used in DECnet
1403 * and when the last use of them in DECnet has gone, I'm intending to
1407 #define SOCK_SLEEP_PRE(sk) { struct task_struct *tsk = current; \
1408 DECLARE_WAITQUEUE(wait, tsk); \
1409 tsk->state = TASK_INTERRUPTIBLE; \
1410 add_wait_queue((sk)->sk_sleep, &wait); \
1413 #define SOCK_SLEEP_POST(sk) tsk->state = TASK_RUNNING; \
1414 remove_wait_queue((sk)->sk_sleep, &wait); \
1418 extern __u32 sysctl_wmem_max;
1419 extern __u32 sysctl_rmem_max;
1421 extern void sk_init(void);
1423 extern int sysctl_optmem_max;
1425 extern __u32 sysctl_wmem_default;
1426 extern __u32 sysctl_rmem_default;
1428 #endif /* _SOCK_H */