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
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
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/config.h>
44 #include <linux/list.h>
45 #include <linux/timer.h>
46 #include <linux/cache.h>
47 #include <linux/module.h>
48 #include <linux/netdevice.h>
49 #include <linux/skbuff.h> /* struct sk_buff */
50 #include <linux/security.h>
52 #include <linux/filter.h>
53 #include <linux/vinline.h>
55 #include <asm/atomic.h>
59 * This structure really needs to be cleaned up.
60 * Most of it is for TCP, and not used by any of
61 * the other protocols.
64 /* Define this to get the sk->sk_debug debugging facility. */
65 //#define SOCK_DEBUGGING
67 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && ((sk)->sk_debug)) \
68 printk(KERN_DEBUG msg); } while (0)
70 #define SOCK_DEBUG(sk, msg...) do { } while (0)
73 /* This is the per-socket lock. The spinlock provides a synchronization
74 * between user contexts and software interrupt processing, whereas the
75 * mini-semaphore synchronizes multiple users amongst themselves.
80 struct sock_iocb *owner;
84 #define sock_lock_init(__sk) \
85 do { spin_lock_init(&((__sk)->sk_lock.slock)); \
86 (__sk)->sk_lock.owner = NULL; \
87 init_waitqueue_head(&((__sk)->sk_lock.wq)); \
93 * struct sock_common - minimal network layer representation of sockets
94 * @skc_family - network address family
95 * @skc_state - Connection state
96 * @skc_reuse - %SO_REUSEADDR setting
97 * @skc_bound_dev_if - bound device index if != 0
98 * @skc_node - main hash linkage for various protocol lookup tables
99 * @skc_bind_node - bind hash linkage for various protocol lookup tables
100 * @skc_refcnt - reference count
102 * This is the minimal network layer representation of sockets, the header
103 * for struct sock and struct tcp_tw_bucket.
106 unsigned short skc_family;
107 volatile unsigned char skc_state;
108 unsigned char skc_reuse;
109 int skc_bound_dev_if;
110 struct hlist_node skc_node;
111 struct hlist_node skc_bind_node;
114 struct vx_info *skc_vx_info;
116 struct nx_info *skc_nx_info;
120 * struct sock - network layer representation of sockets
121 * @__sk_common - shared layout with tcp_tw_bucket
122 * @sk_zapped - ax25 & ipx means !linked
123 * @sk_shutdown - mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
124 * @sk_use_write_queue - wheter to call sk->sk_write_space in sock_wfree
125 * @sk_userlocks - %SO_SNDBUF and %SO_RCVBUF settings
126 * @sk_lock - synchronizer
127 * @sk_rcvbuf - size of receive buffer in bytes
128 * @sk_sleep - sock wait queue
129 * @sk_dst_cache - destination cache
130 * @sk_dst_lock - destination cache lock
131 * @sk_policy - flow policy
132 * @sk_rmem_alloc - receive queue bytes committed
133 * @sk_receive_queue - incoming packets
134 * @sk_wmem_alloc - transmit queue bytes committed
135 * @sk_write_queue - Packet sending queue
136 * @sk_omem_alloc - "o" is "option" or "other"
137 * @sk_wmem_queued - persistent queue size
138 * @sk_forward_alloc - space allocated forward
139 * @sk_allocation - allocation mode
140 * @sk_sndbuf - size of send buffer in bytes
141 * @sk_flags - %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, %SO_OOBINLINE settings
142 * @sk_no_check - %SO_NO_CHECK setting, wether or not checkup packets
143 * @sk_debug - %SO_DEBUG setting
144 * @sk_rcvtstamp - %SO_TIMESTAMP setting
145 * @sk_no_largesend - whether to sent large segments or not
146 * @sk_route_caps - route capabilities (e.g. %NETIF_F_TSO)
147 * @sk_lingertime - %SO_LINGER l_linger setting
148 * @sk_hashent - hash entry in several tables (e.g. tcp_ehash)
149 * @sk_pair - socket pair (e.g. AF_UNIX/unix_peer)
150 * @sk_backlog - always used with the per-socket spinlock held
151 * @sk_callback_lock - used with the callbacks in the end of this struct
152 * @sk_error_queue - rarely used
153 * @sk_prot - protocol handlers inside a network family
154 * @sk_err - last error
155 * @sk_err_soft - errors that don't cause failure but are the cause of a persistent failure not just 'timed out'
156 * @sk_ack_backlog - current listen backlog
157 * @sk_max_ack_backlog - listen backlog set in listen()
158 * @sk_priority - %SO_PRIORITY setting
159 * @sk_type - socket type (%SOCK_STREAM, etc)
160 * @sk_localroute - route locally only, %SO_DONTROUTE setting
161 * @sk_protocol - which protocol this socket belongs in this network family
162 * @sk_peercred - %SO_PEERCRED setting
163 * @sk_rcvlowat - %SO_RCVLOWAT setting
164 * @sk_rcvtimeo - %SO_RCVTIMEO setting
165 * @sk_sndtimeo - %SO_SNDTIMEO setting
166 * @sk_filter - socket filtering instructions
167 * @sk_protinfo - private area, net family specific, when not using slab
168 * @sk_slab - the slabcache this instance was allocated from
169 * @sk_timer - sock cleanup timer
170 * @sk_stamp - time stamp of last packet received
171 * @sk_socket - Identd and reporting IO signals
172 * @sk_user_data - RPC and Tux layer private data
173 * @sk_owner - module that owns this socket
174 * @sk_state_change - callback to indicate change in the state of the sock
175 * @sk_data_ready - callback to indicate there is data to be processed
176 * @sk_write_space - callback to indicate there is bf sending space available
177 * @sk_error_report - callback to indicate errors (e.g. %MSG_ERRQUEUE)
178 * @sk_create_child - callback to get new socket events
179 * @sk_backlog_rcv - callback to process the backlog
180 * @sk_destruct - called at sock freeing time, i.e. when all refcnt == 0
184 * Now struct tcp_tw_bucket also uses sock_common, so please just
185 * don't add nothing before this first member (__sk_common) --acme
187 struct sock_common __sk_common;
188 #define sk_family __sk_common.skc_family
189 #define sk_state __sk_common.skc_state
190 #define sk_reuse __sk_common.skc_reuse
191 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
192 #define sk_node __sk_common.skc_node
193 #define sk_bind_node __sk_common.skc_bind_node
194 #define sk_refcnt __sk_common.skc_refcnt
195 #define sk_xid __sk_common.skc_xid
196 #define sk_vx_info __sk_common.skc_vx_info
197 #define sk_nid __sk_common.skc_nid
198 #define sk_nx_info __sk_common.skc_nx_info
199 volatile unsigned char sk_zapped;
200 unsigned char sk_shutdown;
201 unsigned char sk_use_write_queue;
202 unsigned char sk_userlocks;
203 socket_lock_t sk_lock;
205 wait_queue_head_t *sk_sleep;
206 struct dst_entry *sk_dst_cache;
207 rwlock_t sk_dst_lock;
208 struct xfrm_policy *sk_policy[2];
209 atomic_t sk_rmem_alloc;
210 struct sk_buff_head sk_receive_queue;
211 atomic_t sk_wmem_alloc;
212 struct sk_buff_head sk_write_queue;
213 atomic_t sk_omem_alloc;
215 int sk_forward_alloc;
216 unsigned int sk_allocation;
218 unsigned long sk_flags;
220 unsigned char sk_debug;
221 unsigned char sk_rcvtstamp;
222 unsigned char sk_no_largesend;
224 unsigned long sk_lingertime;
226 struct sock *sk_pair;
228 * The backlog queue is special, it is always used with
229 * the per-socket spinlock held and requires low latency
230 * access. Therefore we special case it's implementation.
233 struct sk_buff *head;
234 struct sk_buff *tail;
236 rwlock_t sk_callback_lock;
237 struct sk_buff_head sk_error_queue;
238 struct proto *sk_prot;
241 unsigned short sk_ack_backlog;
242 unsigned short sk_max_ack_backlog;
244 unsigned short sk_type;
245 unsigned char sk_localroute;
246 unsigned char sk_protocol;
247 struct ucred sk_peercred;
251 struct sk_filter *sk_filter;
253 kmem_cache_t *sk_slab;
254 struct timer_list sk_timer;
255 struct timeval sk_stamp;
256 struct socket *sk_socket;
258 void *sk_ns; // For use by CKRM
259 struct module *sk_owner;
261 void (*sk_state_change)(struct sock *sk);
262 void (*sk_data_ready)(struct sock *sk, int bytes);
263 void (*sk_write_space)(struct sock *sk);
264 void (*sk_error_report)(struct sock *sk);
265 int (*sk_backlog_rcv)(struct sock *sk,
266 struct sk_buff *skb);
267 void (*sk_create_child)(struct sock *sk, struct sock *newsk);
268 void (*sk_destruct)(struct sock *sk);
272 * Hashed lists helper routines
274 static inline struct sock *__sk_head(struct hlist_head *head)
276 return hlist_entry(head->first, struct sock, sk_node);
279 static inline struct sock *sk_head(struct hlist_head *head)
281 return hlist_empty(head) ? NULL : __sk_head(head);
284 static inline struct sock *sk_next(struct sock *sk)
286 return sk->sk_node.next ?
287 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
290 static inline int sk_unhashed(struct sock *sk)
292 return hlist_unhashed(&sk->sk_node);
295 static inline int sk_hashed(struct sock *sk)
297 return sk->sk_node.pprev != NULL;
300 static __inline__ void sk_node_init(struct hlist_node *node)
305 static __inline__ void __sk_del_node(struct sock *sk)
307 __hlist_del(&sk->sk_node);
310 static __inline__ int __sk_del_node_init(struct sock *sk)
314 sk_node_init(&sk->sk_node);
320 /* Grab socket reference count. This operation is valid only
321 when sk is ALREADY grabbed f.e. it is found in hash table
322 or a list and the lookup is made under lock preventing hash table
326 static inline void sock_hold(struct sock *sk)
328 atomic_inc(&sk->sk_refcnt);
331 /* Ungrab socket in the context, which assumes that socket refcnt
332 cannot hit zero, f.e. it is true in context of any socketcall.
334 static inline void __sock_put(struct sock *sk)
336 atomic_dec(&sk->sk_refcnt);
339 static __inline__ int sk_del_node_init(struct sock *sk)
341 int rc = __sk_del_node_init(sk);
344 /* paranoid for a while -acme */
345 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
351 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
353 hlist_add_head(&sk->sk_node, list);
356 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
359 __sk_add_node(sk, list);
362 static __inline__ void __sk_del_bind_node(struct sock *sk)
364 __hlist_del(&sk->sk_bind_node);
367 static __inline__ void sk_add_bind_node(struct sock *sk,
368 struct hlist_head *list)
370 hlist_add_head(&sk->sk_bind_node, list);
373 #define sk_for_each(__sk, node, list) \
374 hlist_for_each_entry(__sk, node, list, sk_node)
375 #define sk_for_each_from(__sk, node) \
376 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
377 hlist_for_each_entry_from(__sk, node, sk_node)
378 #define sk_for_each_continue(__sk, node) \
379 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
380 hlist_for_each_entry_continue(__sk, node, sk_node)
381 #define sk_for_each_safe(__sk, node, tmp, list) \
382 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
383 #define sk_for_each_bound(__sk, node, list) \
384 hlist_for_each_entry(__sk, node, list, sk_bind_node)
398 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
400 __set_bit(flag, &sk->sk_flags);
403 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
405 __clear_bit(flag, &sk->sk_flags);
408 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
410 return test_bit(flag, &sk->sk_flags);
413 /* The per-socket spinlock must be held here. */
414 #define sk_add_backlog(__sk, __skb) \
415 do { if (!(__sk)->sk_backlog.tail) { \
416 (__sk)->sk_backlog.head = \
417 (__sk)->sk_backlog.tail = (__skb); \
419 ((__sk)->sk_backlog.tail)->next = (__skb); \
420 (__sk)->sk_backlog.tail = (__skb); \
422 (__skb)->next = NULL; \
425 /* IP protocol blocks we attach to sockets.
426 * socket layer -> transport layer interface
427 * transport -> network interface is defined by struct inet_proto
430 void (*close)(struct sock *sk,
432 int (*connect)(struct sock *sk,
433 struct sockaddr *uaddr,
435 int (*disconnect)(struct sock *sk, int flags);
437 struct sock * (*accept) (struct sock *sk, int flags, int *err);
439 int (*ioctl)(struct sock *sk, int cmd,
441 int (*init)(struct sock *sk);
442 int (*destroy)(struct sock *sk);
443 void (*shutdown)(struct sock *sk, int how);
444 int (*setsockopt)(struct sock *sk, int level,
445 int optname, char __user *optval,
447 int (*getsockopt)(struct sock *sk, int level,
448 int optname, char __user *optval,
450 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
451 struct msghdr *msg, size_t len);
452 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
454 size_t len, int noblock, int flags,
456 int (*sendpage)(struct sock *sk, struct page *page,
457 int offset, size_t size, int flags);
458 int (*bind)(struct sock *sk,
459 struct sockaddr *uaddr, int addr_len);
461 int (*backlog_rcv) (struct sock *sk,
462 struct sk_buff *skb);
464 /* Keeping track of sk's, looking them up, and port selection methods. */
465 void (*hash)(struct sock *sk);
466 void (*unhash)(struct sock *sk);
467 int (*get_port)(struct sock *sk, unsigned short snum);
473 u8 __pad[SMP_CACHE_BYTES - sizeof(int)];
477 static __inline__ void sk_set_owner(struct sock *sk, struct module *owner)
480 * One should use sk_set_owner just once, after struct sock creation,
481 * be it shortly after sk_alloc or after a function that returns a new
482 * struct sock (and that down the call chain called sk_alloc), e.g. the
483 * IPv4 and IPv6 modules share tcp_create_openreq_child, so if
484 * tcp_create_openreq_child called sk_set_owner IPv6 would have to
485 * change the ownership of this struct sock, with one not needed
486 * transient sk_set_owner call.
488 BUG_ON(sk->sk_owner != NULL);
490 sk->sk_owner = owner;
494 /* Called with local bh disabled */
495 static __inline__ void sock_prot_inc_use(struct proto *prot)
497 prot->stats[smp_processor_id()].inuse++;
500 static __inline__ void sock_prot_dec_use(struct proto *prot)
502 prot->stats[smp_processor_id()].inuse--;
505 /* About 10 seconds */
506 #define SOCK_DESTROY_TIME (10*HZ)
508 /* Sockets 0-1023 can't be bound to unless you are superuser */
509 #define PROT_SOCK 1024
511 #define SHUTDOWN_MASK 3
512 #define RCV_SHUTDOWN 1
513 #define SEND_SHUTDOWN 2
515 #define SOCK_SNDBUF_LOCK 1
516 #define SOCK_RCVBUF_LOCK 2
517 #define SOCK_BINDADDR_LOCK 4
518 #define SOCK_BINDPORT_LOCK 8
520 /* sock_iocb: used to kick off async processing of socket ios */
522 struct list_head list;
528 struct scm_cookie *scm;
529 struct msghdr *msg, async_msg;
530 struct iovec async_iov;
533 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
535 BUG_ON(sizeof(struct sock_iocb) > KIOCB_PRIVATE_SIZE);
536 return (struct sock_iocb *)iocb->private;
539 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
541 return container_of((void *)si, struct kiocb, private);
544 struct socket_alloc {
545 struct socket socket;
546 struct inode vfs_inode;
549 static inline struct socket *SOCKET_I(struct inode *inode)
551 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
554 static inline struct inode *SOCK_INODE(struct socket *socket)
556 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
559 /* Used by processes to "lock" a socket state, so that
560 * interrupts and bottom half handlers won't change it
561 * from under us. It essentially blocks any incoming
562 * packets, so that we won't get any new data or any
563 * packets that change the state of the socket.
565 * While locked, BH processing will add new packets to
566 * the backlog queue. This queue is processed by the
567 * owner of the socket lock right before it is released.
569 * Since ~2.3.5 it is also exclusive sleep lock serializing
570 * accesses from user process context.
572 extern void __lock_sock(struct sock *sk);
573 extern void __release_sock(struct sock *sk);
574 #define sock_owned_by_user(sk) ((sk)->sk_lock.owner)
576 extern void FASTCALL(lock_sock(struct sock *sk));
577 extern void FASTCALL(release_sock(struct sock *sk));
579 /* BH context may only use the following locking interface. */
580 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
581 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
583 extern struct sock * sk_alloc(int family, int priority, int zero_it,
585 extern void sk_free(struct sock *sk);
587 extern struct sk_buff *sock_wmalloc(struct sock *sk,
588 unsigned long size, int force,
590 extern struct sk_buff *sock_rmalloc(struct sock *sk,
591 unsigned long size, int force,
593 extern void sock_wfree(struct sk_buff *skb);
594 extern void sock_rfree(struct sk_buff *skb);
596 extern int sock_setsockopt(struct socket *sock, int level,
597 int op, char __user *optval,
600 extern int sock_getsockopt(struct socket *sock, int level,
601 int op, char __user *optval,
603 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
607 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
608 unsigned long header_len,
609 unsigned long data_len,
612 extern void *sock_kmalloc(struct sock *sk, int size, int priority);
613 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
614 extern void sk_send_sigurg(struct sock *sk);
617 * Functions to fill in entries in struct proto_ops when a protocol
618 * does not implement a particular function.
620 extern int sock_no_release(struct socket *);
621 extern int sock_no_bind(struct socket *,
622 struct sockaddr *, int);
623 extern int sock_no_connect(struct socket *,
624 struct sockaddr *, int, int);
625 extern int sock_no_socketpair(struct socket *,
627 extern int sock_no_accept(struct socket *,
628 struct socket *, int);
629 extern int sock_no_getname(struct socket *,
630 struct sockaddr *, int *, int);
631 extern unsigned int sock_no_poll(struct file *, struct socket *,
632 struct poll_table_struct *);
633 extern int sock_no_ioctl(struct socket *, unsigned int,
635 extern int sock_no_listen(struct socket *, int);
636 extern int sock_no_shutdown(struct socket *, int);
637 extern int sock_no_getsockopt(struct socket *, int , int,
638 char __user *, int __user *);
639 extern int sock_no_setsockopt(struct socket *, int, int,
641 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
642 struct msghdr *, size_t);
643 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
644 struct msghdr *, size_t, int);
645 extern int sock_no_mmap(struct file *file,
647 struct vm_area_struct *vma);
648 extern ssize_t sock_no_sendpage(struct socket *sock,
650 int offset, size_t size,
654 * Default socket callbacks and setup code
657 extern void sock_def_destruct(struct sock *);
659 /* Initialise core socket variables */
660 extern void sock_init_data(struct socket *sock, struct sock *sk);
663 * sk_filter - run a packet through a socket filter
664 * @sk: sock associated with &sk_buff
665 * @skb: buffer to filter
666 * @needlock: set to 1 if the sock is not locked by caller.
668 * Run the filter code and then cut skb->data to correct size returned by
669 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
670 * than pkt_len we keep whole skb->data. This is the socket level
671 * wrapper to sk_run_filter. It returns 0 if the packet should
672 * be accepted or -EPERM if the packet should be tossed.
676 static inline int sk_filter(struct sock *sk, struct sk_buff *skb, int needlock)
680 err = security_sock_rcv_skb(sk, skb);
685 struct sk_filter *filter;
690 filter = sk->sk_filter;
692 int pkt_len = sk_run_filter(skb, filter->insns,
697 skb_trim(skb, pkt_len);
707 * sk_filter_release: Release a socket filter
709 * @fp: filter to remove
711 * Remove a filter from a socket and release its resources.
714 static inline void sk_filter_release(struct sock *sk, struct sk_filter *fp)
716 unsigned int size = sk_filter_len(fp);
718 atomic_sub(size, &sk->sk_omem_alloc);
720 if (atomic_dec_and_test(&fp->refcnt))
724 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
726 atomic_inc(&fp->refcnt);
727 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
731 * Socket reference counting postulates.
733 * * Each user of socket SHOULD hold a reference count.
734 * * Each access point to socket (an hash table bucket, reference from a list,
735 * running timer, skb in flight MUST hold a reference count.
736 * * When reference count hits 0, it means it will never increase back.
737 * * When reference count hits 0, it means that no references from
738 * outside exist to this socket and current process on current CPU
739 * is last user and may/should destroy this socket.
740 * * sk_free is called from any context: process, BH, IRQ. When
741 * it is called, socket has no references from outside -> sk_free
742 * may release descendant resources allocated by the socket, but
743 * to the time when it is called, socket is NOT referenced by any
744 * hash tables, lists etc.
745 * * Packets, delivered from outside (from network or from another process)
746 * and enqueued on receive/error queues SHOULD NOT grab reference count,
747 * when they sit in queue. Otherwise, packets will leak to hole, when
748 * socket is looked up by one cpu and unhasing is made by another CPU.
749 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
750 * (leak to backlog). Packet socket does all the processing inside
751 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
752 * use separate SMP lock, so that they are prone too.
755 /* Ungrab socket and destroy it, if it was the last reference. */
756 static inline void sock_put(struct sock *sk)
758 if (atomic_dec_and_test(&sk->sk_refcnt))
762 /* Detach socket from process context.
763 * Announce socket dead, detach it from wait queue and inode.
764 * Note that parent inode held reference count on this struct sock,
765 * we do not release it in this function, because protocol
766 * probably wants some additional cleanups or even continuing
767 * to work with this socket (TCP).
769 static inline void sock_orphan(struct sock *sk)
771 write_lock_bh(&sk->sk_callback_lock);
772 sock_set_flag(sk, SOCK_DEAD);
773 sk->sk_socket = NULL;
775 write_unlock_bh(&sk->sk_callback_lock);
778 static inline void sock_graft(struct sock *sk, struct socket *parent)
780 write_lock_bh(&sk->sk_callback_lock);
781 sk->sk_sleep = &parent->wait;
783 sk->sk_socket = parent;
784 write_unlock_bh(&sk->sk_callback_lock);
787 static inline int sock_i_uid(struct sock *sk)
791 read_lock(&sk->sk_callback_lock);
792 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
793 read_unlock(&sk->sk_callback_lock);
797 static inline unsigned long sock_i_ino(struct sock *sk)
801 read_lock(&sk->sk_callback_lock);
802 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
803 read_unlock(&sk->sk_callback_lock);
807 static inline struct dst_entry *
808 __sk_dst_get(struct sock *sk)
810 return sk->sk_dst_cache;
813 static inline struct dst_entry *
814 sk_dst_get(struct sock *sk)
816 struct dst_entry *dst;
818 read_lock(&sk->sk_dst_lock);
819 dst = sk->sk_dst_cache;
822 read_unlock(&sk->sk_dst_lock);
827 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
829 struct dst_entry *old_dst;
831 old_dst = sk->sk_dst_cache;
832 sk->sk_dst_cache = dst;
833 dst_release(old_dst);
837 sk_dst_set(struct sock *sk, struct dst_entry *dst)
839 write_lock(&sk->sk_dst_lock);
840 __sk_dst_set(sk, dst);
841 write_unlock(&sk->sk_dst_lock);
845 __sk_dst_reset(struct sock *sk)
847 struct dst_entry *old_dst;
849 old_dst = sk->sk_dst_cache;
850 sk->sk_dst_cache = NULL;
851 dst_release(old_dst);
855 sk_dst_reset(struct sock *sk)
857 write_lock(&sk->sk_dst_lock);
859 write_unlock(&sk->sk_dst_lock);
862 static inline struct dst_entry *
863 __sk_dst_check(struct sock *sk, u32 cookie)
865 struct dst_entry *dst = sk->sk_dst_cache;
867 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
868 sk->sk_dst_cache = NULL;
875 static inline struct dst_entry *
876 sk_dst_check(struct sock *sk, u32 cookie)
878 struct dst_entry *dst = sk_dst_get(sk);
880 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
890 * Queue a received datagram if it will fit. Stream and sequenced
891 * protocols can't normally use this as they need to fit buffers in
892 * and play with them.
894 * Inlined as it's very short and called for pretty much every
895 * packet ever received.
898 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
902 skb->destructor = sock_wfree;
903 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
906 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
909 skb->destructor = sock_rfree;
910 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
913 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
918 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
919 number of warnings when compiling with -W --ANK
921 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
922 (unsigned)sk->sk_rcvbuf) {
927 /* It would be deadlock, if sock_queue_rcv_skb is used
928 with socket lock! We assume that users of this
929 function are lock free.
931 err = sk_filter(sk, skb, 1);
936 skb_set_owner_r(skb, sk);
938 /* Cache the SKB length before we tack it onto the receive
939 * queue. Once it is added it no longer belongs to us and
940 * may be freed by other threads of control pulling packets
945 skb_queue_tail(&sk->sk_receive_queue, skb);
947 if (!sock_flag(sk, SOCK_DEAD))
948 sk->sk_data_ready(sk, skb_len);
953 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
955 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
956 number of warnings when compiling with -W --ANK
958 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
959 (unsigned)sk->sk_rcvbuf)
961 skb_set_owner_r(skb, sk);
962 skb_queue_tail(&sk->sk_error_queue, skb);
963 if (!sock_flag(sk, SOCK_DEAD))
964 sk->sk_data_ready(sk, skb->len);
969 * Recover an error report and clear atomically
972 static inline int sock_error(struct sock *sk)
974 int err = xchg(&sk->sk_err, 0);
978 static inline unsigned long sock_wspace(struct sock *sk)
982 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
983 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
990 static inline void sk_wake_async(struct sock *sk, int how, int band)
992 if (sk->sk_socket && sk->sk_socket->fasync_list)
993 sock_wake_async(sk->sk_socket, how, band);
996 #define SOCK_MIN_SNDBUF 2048
997 #define SOCK_MIN_RCVBUF 256
1000 * Default write policy as shown to user space via poll/select/SIGIO
1002 static inline int sock_writeable(struct sock *sk)
1004 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf / 2);
1007 static inline int gfp_any(void)
1009 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1012 static inline long sock_rcvtimeo(struct sock *sk, int noblock)
1014 return noblock ? 0 : sk->sk_rcvtimeo;
1017 static inline long sock_sndtimeo(struct sock *sk, int noblock)
1019 return noblock ? 0 : sk->sk_sndtimeo;
1022 static inline int sock_rcvlowat(struct sock *sk, int waitall, int len)
1024 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1027 /* Alas, with timeout socket operations are not restartable.
1028 * Compare this to poll().
1030 static inline int sock_intr_errno(long timeo)
1032 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1035 static __inline__ void
1036 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1038 struct timeval *stamp = &skb->stamp;
1039 if (sk->sk_rcvtstamp) {
1040 /* Race occurred between timestamp enabling and packet
1041 receiving. Fill in the current time for now. */
1042 if (stamp->tv_sec == 0)
1043 do_gettimeofday(stamp);
1044 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP, sizeof(struct timeval),
1047 sk->sk_stamp = *stamp;
1050 extern atomic_t netstamp_needed;
1051 extern void sock_enable_timestamp(struct sock *sk);
1052 extern void sock_disable_timestamp(struct sock *sk);
1054 static inline void net_timestamp(struct timeval *stamp)
1056 if (atomic_read(&netstamp_needed))
1057 do_gettimeofday(stamp);
1064 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1067 * Enable debug/info messages
1071 #define NETDEBUG(x) do { } while (0)
1072 #define LIMIT_NETDEBUG(x) do {} while(0)
1074 #define NETDEBUG(x) do { x; } while (0)
1075 #define LIMIT_NETDEBUG(x) do { if (net_ratelimit()) { x; } } while(0)
1079 * Macros for sleeping on a socket. Use them like this:
1081 * SOCK_SLEEP_PRE(sk)
1084 * SOCK_SLEEP_POST(sk)
1086 * N.B. These are now obsolete and were, afaik, only ever used in DECnet
1087 * and when the last use of them in DECnet has gone, I'm intending to
1091 #define SOCK_SLEEP_PRE(sk) { struct task_struct *tsk = current; \
1092 DECLARE_WAITQUEUE(wait, tsk); \
1093 tsk->state = TASK_INTERRUPTIBLE; \
1094 add_wait_queue((sk)->sk_sleep, &wait); \
1097 #define SOCK_SLEEP_POST(sk) tsk->state = TASK_RUNNING; \
1098 remove_wait_queue((sk)->sk_sleep, &wait); \
1102 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
1105 sock_set_flag(sk, bit);
1107 sock_reset_flag(sk, bit);
1110 extern __u32 sysctl_wmem_max;
1111 extern __u32 sysctl_rmem_max;
1113 int siocdevprivate_ioctl(unsigned int fd, unsigned int cmd, unsigned long arg);
1115 #endif /* _SOCK_H */