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 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <net/xfrm.h>
124 #include <linux/ipsec.h>
126 #include <linux/filter.h>
133 * Each address family might have different locking rules, so we have
134 * one slock key per address family:
136 static struct lock_class_key af_family_keys
[AF_MAX
];
137 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
141 * Make lock validator output more readable. (we pre-construct these
142 * strings build-time, so that runtime initialization of socket
145 static const char *af_family_key_strings
[AF_MAX
+1] = {
146 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
147 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
148 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
149 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
150 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
151 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
152 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
153 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
155 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
158 static const char *af_family_slock_key_strings
[AF_MAX
+1] = {
159 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
160 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
161 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
162 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
163 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
164 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
165 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
166 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
167 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
168 "slock-27" , "slock-28" , "slock-29" ,
169 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX"
174 * sk_callback_lock locking rules are per-address-family,
175 * so split the lock classes by using a per-AF key:
177 static struct lock_class_key af_callback_keys
[AF_MAX
];
179 /* Take into consideration the size of the struct sk_buff overhead in the
180 * determination of these values, since that is non-constant across
181 * platforms. This makes socket queueing behavior and performance
182 * not depend upon such differences.
184 #define _SK_MEM_PACKETS 256
185 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
186 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
187 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
189 /* Run time adjustable parameters. */
190 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
191 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
192 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
193 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
195 /* Maximal space eaten by iovec or ancilliary data plus some space */
196 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
198 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
202 if (optlen
< sizeof(tv
))
204 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
207 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
208 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
210 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
211 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
215 static void sock_warn_obsolete_bsdism(const char *name
)
218 static char warncomm
[TASK_COMM_LEN
];
219 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
220 strcpy(warncomm
, current
->comm
);
221 printk(KERN_WARNING
"process `%s' is using obsolete "
222 "%s SO_BSDCOMPAT\n", warncomm
, name
);
227 static void sock_disable_timestamp(struct sock
*sk
)
229 if (sock_flag(sk
, SOCK_TIMESTAMP
)) {
230 sock_reset_flag(sk
, SOCK_TIMESTAMP
);
231 net_disable_timestamp();
236 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
241 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
242 number of warnings when compiling with -W --ANK
244 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
245 (unsigned)sk
->sk_rcvbuf
) {
250 err
= sk_filter(sk
, skb
);
255 skb_set_owner_r(skb
, sk
);
257 /* Cache the SKB length before we tack it onto the receive
258 * queue. Once it is added it no longer belongs to us and
259 * may be freed by other threads of control pulling packets
264 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
266 if (!sock_flag(sk
, SOCK_DEAD
))
267 sk
->sk_data_ready(sk
, skb_len
);
271 EXPORT_SYMBOL(sock_queue_rcv_skb
);
273 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
275 int rc
= NET_RX_SUCCESS
;
277 if (sk_filter(sk
, skb
))
278 goto discard_and_relse
;
283 bh_lock_sock_nested(sk
);
286 if (!sock_owned_by_user(sk
)) {
288 * trylock + unlock semantics:
290 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
292 rc
= sk
->sk_backlog_rcv(sk
, skb
);
294 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
296 sk_add_backlog(sk
, skb
);
305 EXPORT_SYMBOL(sk_receive_skb
);
307 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
309 struct dst_entry
*dst
= sk
->sk_dst_cache
;
311 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
312 sk
->sk_dst_cache
= NULL
;
319 EXPORT_SYMBOL(__sk_dst_check
);
321 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
323 struct dst_entry
*dst
= sk_dst_get(sk
);
325 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
333 EXPORT_SYMBOL(sk_dst_check
);
336 * This is meant for all protocols to use and covers goings on
337 * at the socket level. Everything here is generic.
340 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
341 char __user
*optval
, int optlen
)
343 struct sock
*sk
=sock
->sk
;
344 struct sk_filter
*filter
;
351 * Options without arguments
354 #ifdef SO_DONTLINGER /* Compatibility item... */
355 if (optname
== SO_DONTLINGER
) {
357 sock_reset_flag(sk
, SOCK_LINGER
);
363 if(optlen
<sizeof(int))
366 if (get_user(val
, (int __user
*)optval
))
376 if(val
&& !capable(CAP_NET_ADMIN
))
381 sock_set_flag(sk
, SOCK_DBG
);
383 sock_reset_flag(sk
, SOCK_DBG
);
386 sk
->sk_reuse
= valbool
;
394 sock_set_flag(sk
, SOCK_LOCALROUTE
);
396 sock_reset_flag(sk
, SOCK_LOCALROUTE
);
399 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
402 /* Don't error on this BSD doesn't and if you think
403 about it this is right. Otherwise apps have to
404 play 'guess the biggest size' games. RCVBUF/SNDBUF
405 are treated in BSD as hints */
407 if (val
> sysctl_wmem_max
)
408 val
= sysctl_wmem_max
;
410 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
411 if ((val
* 2) < SOCK_MIN_SNDBUF
)
412 sk
->sk_sndbuf
= SOCK_MIN_SNDBUF
;
414 sk
->sk_sndbuf
= val
* 2;
417 * Wake up sending tasks if we
420 sk
->sk_write_space(sk
);
424 if (!capable(CAP_NET_ADMIN
)) {
431 /* Don't error on this BSD doesn't and if you think
432 about it this is right. Otherwise apps have to
433 play 'guess the biggest size' games. RCVBUF/SNDBUF
434 are treated in BSD as hints */
436 if (val
> sysctl_rmem_max
)
437 val
= sysctl_rmem_max
;
439 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
441 * We double it on the way in to account for
442 * "struct sk_buff" etc. overhead. Applications
443 * assume that the SO_RCVBUF setting they make will
444 * allow that much actual data to be received on that
447 * Applications are unaware that "struct sk_buff" and
448 * other overheads allocate from the receive buffer
449 * during socket buffer allocation.
451 * And after considering the possible alternatives,
452 * returning the value we actually used in getsockopt
453 * is the most desirable behavior.
455 if ((val
* 2) < SOCK_MIN_RCVBUF
)
456 sk
->sk_rcvbuf
= SOCK_MIN_RCVBUF
;
458 sk
->sk_rcvbuf
= val
* 2;
462 if (!capable(CAP_NET_ADMIN
)) {
470 if (sk
->sk_protocol
== IPPROTO_TCP
)
471 tcp_set_keepalive(sk
, valbool
);
473 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
477 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
481 sk
->sk_no_check
= valbool
;
485 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
486 sk
->sk_priority
= val
;
492 if(optlen
<sizeof(ling
)) {
493 ret
= -EINVAL
; /* 1003.1g */
496 if (copy_from_user(&ling
,optval
,sizeof(ling
))) {
501 sock_reset_flag(sk
, SOCK_LINGER
);
503 #if (BITS_PER_LONG == 32)
504 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
505 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
508 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
509 sock_set_flag(sk
, SOCK_LINGER
);
514 sock_warn_obsolete_bsdism("setsockopt");
519 set_bit(SOCK_PASSCRED
, &sock
->flags
);
521 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
526 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
527 sock_enable_timestamp(sk
);
529 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
535 sk
->sk_rcvlowat
= val
? : 1;
539 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
543 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
546 #ifdef CONFIG_NETDEVICES
547 case SO_BINDTODEVICE
:
549 char devname
[IFNAMSIZ
];
552 if (!capable(CAP_NET_RAW
)) {
557 /* Bind this socket to a particular device like "eth0",
558 * as specified in the passed interface name. If the
559 * name is "" or the option length is zero the socket
564 sk
->sk_bound_dev_if
= 0;
566 if (optlen
> IFNAMSIZ
- 1)
567 optlen
= IFNAMSIZ
- 1;
568 memset(devname
, 0, sizeof(devname
));
569 if (copy_from_user(devname
, optval
, optlen
)) {
574 /* Remove any cached route for this socket. */
577 if (devname
[0] == '\0') {
578 sk
->sk_bound_dev_if
= 0;
580 struct net_device
*dev
= dev_get_by_name(devname
);
585 sk
->sk_bound_dev_if
= dev
->ifindex
;
594 case SO_ATTACH_FILTER
:
596 if (optlen
== sizeof(struct sock_fprog
)) {
597 struct sock_fprog fprog
;
600 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
603 ret
= sk_attach_filter(&fprog
, sk
);
607 case SO_DETACH_FILTER
:
609 filter
= rcu_dereference(sk
->sk_filter
);
611 rcu_assign_pointer(sk
->sk_filter
, NULL
);
612 sk_filter_release(sk
, filter
);
613 rcu_read_unlock_bh();
616 rcu_read_unlock_bh();
622 set_bit(SOCK_PASSSEC
, &sock
->flags
);
624 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
627 /* We implement the SO_SNDLOWAT etc to
628 not be settable (1003.1g 5.3) */
638 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
639 char __user
*optval
, int __user
*optlen
)
641 struct sock
*sk
= sock
->sk
;
650 unsigned int lv
= sizeof(int);
653 if(get_user(len
,optlen
))
661 v
.val
= sock_flag(sk
, SOCK_DBG
);
665 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
669 v
.val
= !!sock_flag(sk
, SOCK_BROADCAST
);
673 v
.val
= sk
->sk_sndbuf
;
677 v
.val
= sk
->sk_rcvbuf
;
681 v
.val
= sk
->sk_reuse
;
685 v
.val
= !!sock_flag(sk
, SOCK_KEEPOPEN
);
693 v
.val
= -sock_error(sk
);
695 v
.val
= xchg(&sk
->sk_err_soft
, 0);
699 v
.val
= !!sock_flag(sk
, SOCK_URGINLINE
);
703 v
.val
= sk
->sk_no_check
;
707 v
.val
= sk
->sk_priority
;
712 v
.ling
.l_onoff
= !!sock_flag(sk
, SOCK_LINGER
);
713 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
717 sock_warn_obsolete_bsdism("getsockopt");
721 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
);
725 lv
=sizeof(struct timeval
);
726 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
730 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
731 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
736 lv
=sizeof(struct timeval
);
737 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
741 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
742 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
747 v
.val
= sk
->sk_rcvlowat
;
755 v
.val
= test_bit(SOCK_PASSCRED
, &sock
->flags
) ? 1 : 0;
759 if (len
> sizeof(sk
->sk_peercred
))
760 len
= sizeof(sk
->sk_peercred
);
761 if (copy_to_user(optval
, &sk
->sk_peercred
, len
))
769 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
773 if (copy_to_user(optval
, address
, len
))
778 /* Dubious BSD thing... Probably nobody even uses it, but
779 * the UNIX standard wants it for whatever reason... -DaveM
782 v
.val
= sk
->sk_state
== TCP_LISTEN
;
786 v
.val
= test_bit(SOCK_PASSSEC
, &sock
->flags
) ? 1 : 0;
790 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
793 return(-ENOPROTOOPT
);
797 if (copy_to_user(optval
, &v
, len
))
800 if (put_user(len
, optlen
))
806 * Initialize an sk_lock.
808 * (We also register the sk_lock with the lock validator.)
810 static void inline sock_lock_init(struct sock
*sk
)
812 spin_lock_init(&sk
->sk_lock
.slock
);
813 sk
->sk_lock
.owner
= NULL
;
814 init_waitqueue_head(&sk
->sk_lock
.wq
);
816 * Make sure we are not reinitializing a held lock:
818 debug_check_no_locks_freed((void *)&sk
->sk_lock
, sizeof(sk
->sk_lock
));
821 * Mark both the sk_lock and the sk_lock.slock as a
822 * per-address-family lock class:
824 lockdep_set_class_and_name(&sk
->sk_lock
.slock
,
825 af_family_slock_keys
+ sk
->sk_family
,
826 af_family_slock_key_strings
[sk
->sk_family
]);
827 lockdep_init_map(&sk
->sk_lock
.dep_map
,
828 af_family_key_strings
[sk
->sk_family
],
829 af_family_keys
+ sk
->sk_family
, 0);
833 * sk_alloc - All socket objects are allocated here
834 * @family: protocol family
835 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
836 * @prot: struct proto associated with this new sock instance
837 * @zero_it: if we should zero the newly allocated sock
839 struct sock
*sk_alloc(int family
, gfp_t priority
,
840 struct proto
*prot
, int zero_it
)
842 struct sock
*sk
= NULL
;
843 kmem_cache_t
*slab
= prot
->slab
;
846 sk
= kmem_cache_alloc(slab
, priority
);
848 sk
= kmalloc(prot
->obj_size
, priority
);
852 memset(sk
, 0, prot
->obj_size
);
853 sk
->sk_family
= family
;
855 * See comment in struct sock definition to understand
856 * why we need sk_prot_creator -acme
858 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
862 if (security_sk_alloc(sk
, family
, priority
))
865 if (!try_module_get(prot
->owner
))
872 kmem_cache_free(slab
, sk
);
878 void sk_free(struct sock
*sk
)
880 struct sk_filter
*filter
;
881 struct module
*owner
= sk
->sk_prot_creator
->owner
;
886 filter
= rcu_dereference(sk
->sk_filter
);
888 sk_filter_release(sk
, filter
);
889 rcu_assign_pointer(sk
->sk_filter
, NULL
);
892 sock_disable_timestamp(sk
);
894 if (atomic_read(&sk
->sk_omem_alloc
))
895 printk(KERN_DEBUG
"%s: optmem leakage (%d bytes) detected.\n",
896 __FUNCTION__
, atomic_read(&sk
->sk_omem_alloc
));
898 security_sk_free(sk
);
899 if (sk
->sk_prot_creator
->slab
!= NULL
)
900 kmem_cache_free(sk
->sk_prot_creator
->slab
, sk
);
906 struct sock
*sk_clone(const struct sock
*sk
, const gfp_t priority
)
908 struct sock
*newsk
= sk_alloc(sk
->sk_family
, priority
, sk
->sk_prot
, 0);
911 struct sk_filter
*filter
;
913 sock_copy(newsk
, sk
);
916 sk_node_init(&newsk
->sk_node
);
917 sock_lock_init(newsk
);
920 atomic_set(&newsk
->sk_rmem_alloc
, 0);
921 atomic_set(&newsk
->sk_wmem_alloc
, 0);
922 atomic_set(&newsk
->sk_omem_alloc
, 0);
923 skb_queue_head_init(&newsk
->sk_receive_queue
);
924 skb_queue_head_init(&newsk
->sk_write_queue
);
925 #ifdef CONFIG_NET_DMA
926 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
929 rwlock_init(&newsk
->sk_dst_lock
);
930 rwlock_init(&newsk
->sk_callback_lock
);
931 lockdep_set_class(&newsk
->sk_callback_lock
,
932 af_callback_keys
+ newsk
->sk_family
);
934 newsk
->sk_dst_cache
= NULL
;
935 newsk
->sk_wmem_queued
= 0;
936 newsk
->sk_forward_alloc
= 0;
937 newsk
->sk_send_head
= NULL
;
938 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
939 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
941 sock_reset_flag(newsk
, SOCK_DONE
);
942 skb_queue_head_init(&newsk
->sk_error_queue
);
944 filter
= newsk
->sk_filter
;
946 sk_filter_charge(newsk
, filter
);
948 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
949 /* It is still raw copy of parent, so invalidate
950 * destructor and make plain sk_free() */
951 newsk
->sk_destruct
= NULL
;
958 newsk
->sk_priority
= 0;
959 atomic_set(&newsk
->sk_refcnt
, 2);
962 * Increment the counter in the same struct proto as the master
963 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
964 * is the same as sk->sk_prot->socks, as this field was copied
967 * This _changes_ the previous behaviour, where
968 * tcp_create_openreq_child always was incrementing the
969 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
970 * to be taken into account in all callers. -acme
972 sk_refcnt_debug_inc(newsk
);
973 newsk
->sk_socket
= NULL
;
974 newsk
->sk_sleep
= NULL
;
976 if (newsk
->sk_prot
->sockets_allocated
)
977 atomic_inc(newsk
->sk_prot
->sockets_allocated
);
983 EXPORT_SYMBOL_GPL(sk_clone
);
985 void __init
sk_init(void)
987 if (num_physpages
<= 4096) {
988 sysctl_wmem_max
= 32767;
989 sysctl_rmem_max
= 32767;
990 sysctl_wmem_default
= 32767;
991 sysctl_rmem_default
= 32767;
992 } else if (num_physpages
>= 131072) {
993 sysctl_wmem_max
= 131071;
994 sysctl_rmem_max
= 131071;
999 * Simple resource managers for sockets.
1004 * Write buffer destructor automatically called from kfree_skb.
1006 void sock_wfree(struct sk_buff
*skb
)
1008 struct sock
*sk
= skb
->sk
;
1010 /* In case it might be waiting for more memory. */
1011 atomic_sub(skb
->truesize
, &sk
->sk_wmem_alloc
);
1012 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
))
1013 sk
->sk_write_space(sk
);
1018 * Read buffer destructor automatically called from kfree_skb.
1020 void sock_rfree(struct sk_buff
*skb
)
1022 struct sock
*sk
= skb
->sk
;
1024 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1028 int sock_i_uid(struct sock
*sk
)
1032 read_lock(&sk
->sk_callback_lock
);
1033 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: 0;
1034 read_unlock(&sk
->sk_callback_lock
);
1038 unsigned long sock_i_ino(struct sock
*sk
)
1042 read_lock(&sk
->sk_callback_lock
);
1043 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1044 read_unlock(&sk
->sk_callback_lock
);
1049 * Allocate a skb from the socket's send buffer.
1051 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1054 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1055 struct sk_buff
* skb
= alloc_skb(size
, priority
);
1057 skb_set_owner_w(skb
, sk
);
1065 * Allocate a skb from the socket's receive buffer.
1067 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1070 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1071 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1073 skb_set_owner_r(skb
, sk
);
1081 * Allocate a memory block from the socket's option memory buffer.
1083 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1085 if ((unsigned)size
<= sysctl_optmem_max
&&
1086 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1088 /* First do the add, to avoid the race if kmalloc
1091 atomic_add(size
, &sk
->sk_omem_alloc
);
1092 mem
= kmalloc(size
, priority
);
1095 atomic_sub(size
, &sk
->sk_omem_alloc
);
1101 * Free an option memory block.
1103 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1106 atomic_sub(size
, &sk
->sk_omem_alloc
);
1109 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1110 I think, these locks should be removed for datagram sockets.
1112 static long sock_wait_for_wmem(struct sock
* sk
, long timeo
)
1116 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1120 if (signal_pending(current
))
1122 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1123 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
1124 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1126 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1130 timeo
= schedule_timeout(timeo
);
1132 finish_wait(sk
->sk_sleep
, &wait
);
1138 * Generic send/receive buffer handlers
1141 static struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
,
1142 unsigned long header_len
,
1143 unsigned long data_len
,
1144 int noblock
, int *errcode
)
1146 struct sk_buff
*skb
;
1151 gfp_mask
= sk
->sk_allocation
;
1152 if (gfp_mask
& __GFP_WAIT
)
1153 gfp_mask
|= __GFP_REPEAT
;
1155 timeo
= sock_sndtimeo(sk
, noblock
);
1157 err
= sock_error(sk
);
1162 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1165 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1166 skb
= alloc_skb(header_len
, gfp_mask
);
1171 /* No pages, we're done... */
1175 npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1176 skb
->truesize
+= data_len
;
1177 skb_shinfo(skb
)->nr_frags
= npages
;
1178 for (i
= 0; i
< npages
; i
++) {
1182 page
= alloc_pages(sk
->sk_allocation
, 0);
1185 skb_shinfo(skb
)->nr_frags
= i
;
1190 frag
= &skb_shinfo(skb
)->frags
[i
];
1192 frag
->page_offset
= 0;
1193 frag
->size
= (data_len
>= PAGE_SIZE
?
1196 data_len
-= PAGE_SIZE
;
1199 /* Full success... */
1205 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1206 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1210 if (signal_pending(current
))
1212 timeo
= sock_wait_for_wmem(sk
, timeo
);
1215 skb_set_owner_w(skb
, sk
);
1219 err
= sock_intr_errno(timeo
);
1225 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1226 int noblock
, int *errcode
)
1228 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1231 static void __lock_sock(struct sock
*sk
)
1236 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1237 TASK_UNINTERRUPTIBLE
);
1238 spin_unlock_bh(&sk
->sk_lock
.slock
);
1240 spin_lock_bh(&sk
->sk_lock
.slock
);
1241 if(!sock_owned_by_user(sk
))
1244 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1247 static void __release_sock(struct sock
*sk
)
1249 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1252 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1256 struct sk_buff
*next
= skb
->next
;
1259 sk
->sk_backlog_rcv(sk
, skb
);
1262 * We are in process context here with softirqs
1263 * disabled, use cond_resched_softirq() to preempt.
1264 * This is safe to do because we've taken the backlog
1267 cond_resched_softirq();
1270 } while (skb
!= NULL
);
1273 } while((skb
= sk
->sk_backlog
.head
) != NULL
);
1277 * sk_wait_data - wait for data to arrive at sk_receive_queue
1278 * @sk: sock to wait on
1279 * @timeo: for how long
1281 * Now socket state including sk->sk_err is changed only under lock,
1282 * hence we may omit checks after joining wait queue.
1283 * We check receive queue before schedule() only as optimization;
1284 * it is very likely that release_sock() added new data.
1286 int sk_wait_data(struct sock
*sk
, long *timeo
)
1291 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
1292 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1293 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1294 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1295 finish_wait(sk
->sk_sleep
, &wait
);
1299 EXPORT_SYMBOL(sk_wait_data
);
1302 * Set of default routines for initialising struct proto_ops when
1303 * the protocol does not support a particular function. In certain
1304 * cases where it makes no sense for a protocol to have a "do nothing"
1305 * function, some default processing is provided.
1308 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1313 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1319 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1324 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1329 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1335 unsigned int sock_no_poll(struct file
* file
, struct socket
*sock
, poll_table
*pt
)
1340 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1345 int sock_no_listen(struct socket
*sock
, int backlog
)
1350 int sock_no_shutdown(struct socket
*sock
, int how
)
1355 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
1356 char __user
*optval
, int optlen
)
1361 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
1362 char __user
*optval
, int __user
*optlen
)
1367 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1373 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1374 size_t len
, int flags
)
1379 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
1381 /* Mirror missing mmap method error code */
1385 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
1388 struct msghdr msg
= {.msg_flags
= flags
};
1390 char *kaddr
= kmap(page
);
1391 iov
.iov_base
= kaddr
+ offset
;
1393 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1399 * Default Socket Callbacks
1402 static void sock_def_wakeup(struct sock
*sk
)
1404 read_lock(&sk
->sk_callback_lock
);
1405 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1406 wake_up_interruptible_all(sk
->sk_sleep
);
1407 read_unlock(&sk
->sk_callback_lock
);
1410 static void sock_def_error_report(struct sock
*sk
)
1412 read_lock(&sk
->sk_callback_lock
);
1413 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1414 wake_up_interruptible(sk
->sk_sleep
);
1415 sk_wake_async(sk
,0,POLL_ERR
);
1416 read_unlock(&sk
->sk_callback_lock
);
1419 static void sock_def_readable(struct sock
*sk
, int len
)
1421 read_lock(&sk
->sk_callback_lock
);
1422 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1423 wake_up_interruptible(sk
->sk_sleep
);
1424 sk_wake_async(sk
,1,POLL_IN
);
1425 read_unlock(&sk
->sk_callback_lock
);
1428 static void sock_def_write_space(struct sock
*sk
)
1430 read_lock(&sk
->sk_callback_lock
);
1432 /* Do not wake up a writer until he can make "significant"
1435 if((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
1436 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1437 wake_up_interruptible(sk
->sk_sleep
);
1439 /* Should agree with poll, otherwise some programs break */
1440 if (sock_writeable(sk
))
1441 sk_wake_async(sk
, 2, POLL_OUT
);
1444 read_unlock(&sk
->sk_callback_lock
);
1447 static void sock_def_destruct(struct sock
*sk
)
1449 kfree(sk
->sk_protinfo
);
1452 void sk_send_sigurg(struct sock
*sk
)
1454 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
1455 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
1456 sk_wake_async(sk
, 3, POLL_PRI
);
1459 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
1460 unsigned long expires
)
1462 if (!mod_timer(timer
, expires
))
1466 EXPORT_SYMBOL(sk_reset_timer
);
1468 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
1470 if (timer_pending(timer
) && del_timer(timer
))
1474 EXPORT_SYMBOL(sk_stop_timer
);
1476 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
1478 skb_queue_head_init(&sk
->sk_receive_queue
);
1479 skb_queue_head_init(&sk
->sk_write_queue
);
1480 skb_queue_head_init(&sk
->sk_error_queue
);
1481 #ifdef CONFIG_NET_DMA
1482 skb_queue_head_init(&sk
->sk_async_wait_queue
);
1485 sk
->sk_send_head
= NULL
;
1487 init_timer(&sk
->sk_timer
);
1489 sk
->sk_allocation
= GFP_KERNEL
;
1490 sk
->sk_rcvbuf
= sysctl_rmem_default
;
1491 sk
->sk_sndbuf
= sysctl_wmem_default
;
1492 sk
->sk_state
= TCP_CLOSE
;
1493 sk
->sk_socket
= sock
;
1495 sock_set_flag(sk
, SOCK_ZAPPED
);
1499 sk
->sk_type
= sock
->type
;
1500 sk
->sk_sleep
= &sock
->wait
;
1503 sk
->sk_sleep
= NULL
;
1505 rwlock_init(&sk
->sk_dst_lock
);
1506 rwlock_init(&sk
->sk_callback_lock
);
1507 lockdep_set_class(&sk
->sk_callback_lock
,
1508 af_callback_keys
+ sk
->sk_family
);
1510 sk
->sk_state_change
= sock_def_wakeup
;
1511 sk
->sk_data_ready
= sock_def_readable
;
1512 sk
->sk_write_space
= sock_def_write_space
;
1513 sk
->sk_error_report
= sock_def_error_report
;
1514 sk
->sk_destruct
= sock_def_destruct
;
1516 sk
->sk_sndmsg_page
= NULL
;
1517 sk
->sk_sndmsg_off
= 0;
1519 sk
->sk_peercred
.pid
= 0;
1520 sk
->sk_peercred
.uid
= -1;
1521 sk
->sk_peercred
.gid
= -1;
1522 sk
->sk_write_pending
= 0;
1523 sk
->sk_rcvlowat
= 1;
1524 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
1525 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
1527 sk
->sk_stamp
.tv_sec
= -1L;
1528 sk
->sk_stamp
.tv_usec
= -1L;
1530 atomic_set(&sk
->sk_refcnt
, 1);
1533 void fastcall
lock_sock_nested(struct sock
*sk
, int subclass
)
1536 spin_lock_bh(&sk
->sk_lock
.slock
);
1537 if (sk
->sk_lock
.owner
)
1539 sk
->sk_lock
.owner
= (void *)1;
1540 spin_unlock(&sk
->sk_lock
.slock
);
1542 * The sk_lock has mutex_lock() semantics here:
1544 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
1548 EXPORT_SYMBOL(lock_sock_nested
);
1550 void fastcall
release_sock(struct sock
*sk
)
1553 * The sk_lock has mutex_unlock() semantics:
1555 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
1557 spin_lock_bh(&sk
->sk_lock
.slock
);
1558 if (sk
->sk_backlog
.tail
)
1560 sk
->sk_lock
.owner
= NULL
;
1561 if (waitqueue_active(&sk
->sk_lock
.wq
))
1562 wake_up(&sk
->sk_lock
.wq
);
1563 spin_unlock_bh(&sk
->sk_lock
.slock
);
1565 EXPORT_SYMBOL(release_sock
);
1567 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
1569 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
1570 sock_enable_timestamp(sk
);
1571 if (sk
->sk_stamp
.tv_sec
== -1)
1573 if (sk
->sk_stamp
.tv_sec
== 0)
1574 do_gettimeofday(&sk
->sk_stamp
);
1575 return copy_to_user(userstamp
, &sk
->sk_stamp
, sizeof(struct timeval
)) ?
1578 EXPORT_SYMBOL(sock_get_timestamp
);
1580 void sock_enable_timestamp(struct sock
*sk
)
1582 if (!sock_flag(sk
, SOCK_TIMESTAMP
)) {
1583 sock_set_flag(sk
, SOCK_TIMESTAMP
);
1584 net_enable_timestamp();
1587 EXPORT_SYMBOL(sock_enable_timestamp
);
1590 * Get a socket option on an socket.
1592 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1593 * asynchronous errors should be reported by getsockopt. We assume
1594 * this means if you specify SO_ERROR (otherwise whats the point of it).
1596 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1597 char __user
*optval
, int __user
*optlen
)
1599 struct sock
*sk
= sock
->sk
;
1601 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
1604 EXPORT_SYMBOL(sock_common_getsockopt
);
1606 #ifdef CONFIG_COMPAT
1607 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1608 char __user
*optval
, int __user
*optlen
)
1610 struct sock
*sk
= sock
->sk
;
1612 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
1613 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
1615 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
1617 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
1620 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1621 struct msghdr
*msg
, size_t size
, int flags
)
1623 struct sock
*sk
= sock
->sk
;
1627 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
1628 flags
& ~MSG_DONTWAIT
, &addr_len
);
1630 msg
->msg_namelen
= addr_len
;
1634 EXPORT_SYMBOL(sock_common_recvmsg
);
1637 * Set socket options on an inet socket.
1639 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1640 char __user
*optval
, int optlen
)
1642 struct sock
*sk
= sock
->sk
;
1644 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
1647 EXPORT_SYMBOL(sock_common_setsockopt
);
1649 #ifdef CONFIG_COMPAT
1650 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1651 char __user
*optval
, int optlen
)
1653 struct sock
*sk
= sock
->sk
;
1655 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
1656 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
1658 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
1660 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
1663 void sk_common_release(struct sock
*sk
)
1665 if (sk
->sk_prot
->destroy
)
1666 sk
->sk_prot
->destroy(sk
);
1669 * Observation: when sock_common_release is called, processes have
1670 * no access to socket. But net still has.
1671 * Step one, detach it from networking:
1673 * A. Remove from hash tables.
1676 sk
->sk_prot
->unhash(sk
);
1679 * In this point socket cannot receive new packets, but it is possible
1680 * that some packets are in flight because some CPU runs receiver and
1681 * did hash table lookup before we unhashed socket. They will achieve
1682 * receive queue and will be purged by socket destructor.
1684 * Also we still have packets pending on receive queue and probably,
1685 * our own packets waiting in device queues. sock_destroy will drain
1686 * receive queue, but transmitted packets will delay socket destruction
1687 * until the last reference will be released.
1692 xfrm_sk_free_policy(sk
);
1694 sk_refcnt_debug_release(sk
);
1698 EXPORT_SYMBOL(sk_common_release
);
1700 static DEFINE_RWLOCK(proto_list_lock
);
1701 static LIST_HEAD(proto_list
);
1703 int proto_register(struct proto
*prot
, int alloc_slab
)
1705 char *request_sock_slab_name
= NULL
;
1706 char *timewait_sock_slab_name
;
1710 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
1711 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1713 if (prot
->slab
== NULL
) {
1714 printk(KERN_CRIT
"%s: Can't create sock SLAB cache!\n",
1719 if (prot
->rsk_prot
!= NULL
) {
1720 static const char mask
[] = "request_sock_%s";
1722 request_sock_slab_name
= kmalloc(strlen(prot
->name
) + sizeof(mask
) - 1, GFP_KERNEL
);
1723 if (request_sock_slab_name
== NULL
)
1724 goto out_free_sock_slab
;
1726 sprintf(request_sock_slab_name
, mask
, prot
->name
);
1727 prot
->rsk_prot
->slab
= kmem_cache_create(request_sock_slab_name
,
1728 prot
->rsk_prot
->obj_size
, 0,
1729 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1731 if (prot
->rsk_prot
->slab
== NULL
) {
1732 printk(KERN_CRIT
"%s: Can't create request sock SLAB cache!\n",
1734 goto out_free_request_sock_slab_name
;
1738 if (prot
->twsk_prot
!= NULL
) {
1739 static const char mask
[] = "tw_sock_%s";
1741 timewait_sock_slab_name
= kmalloc(strlen(prot
->name
) + sizeof(mask
) - 1, GFP_KERNEL
);
1743 if (timewait_sock_slab_name
== NULL
)
1744 goto out_free_request_sock_slab
;
1746 sprintf(timewait_sock_slab_name
, mask
, prot
->name
);
1747 prot
->twsk_prot
->twsk_slab
=
1748 kmem_cache_create(timewait_sock_slab_name
,
1749 prot
->twsk_prot
->twsk_obj_size
,
1750 0, SLAB_HWCACHE_ALIGN
,
1752 if (prot
->twsk_prot
->twsk_slab
== NULL
)
1753 goto out_free_timewait_sock_slab_name
;
1757 write_lock(&proto_list_lock
);
1758 list_add(&prot
->node
, &proto_list
);
1759 write_unlock(&proto_list_lock
);
1763 out_free_timewait_sock_slab_name
:
1764 kfree(timewait_sock_slab_name
);
1765 out_free_request_sock_slab
:
1766 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
1767 kmem_cache_destroy(prot
->rsk_prot
->slab
);
1768 prot
->rsk_prot
->slab
= NULL
;
1770 out_free_request_sock_slab_name
:
1771 kfree(request_sock_slab_name
);
1773 kmem_cache_destroy(prot
->slab
);
1778 EXPORT_SYMBOL(proto_register
);
1780 void proto_unregister(struct proto
*prot
)
1782 write_lock(&proto_list_lock
);
1783 list_del(&prot
->node
);
1784 write_unlock(&proto_list_lock
);
1786 if (prot
->slab
!= NULL
) {
1787 kmem_cache_destroy(prot
->slab
);
1791 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
1792 const char *name
= kmem_cache_name(prot
->rsk_prot
->slab
);
1794 kmem_cache_destroy(prot
->rsk_prot
->slab
);
1796 prot
->rsk_prot
->slab
= NULL
;
1799 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
1800 const char *name
= kmem_cache_name(prot
->twsk_prot
->twsk_slab
);
1802 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
1804 prot
->twsk_prot
->twsk_slab
= NULL
;
1808 EXPORT_SYMBOL(proto_unregister
);
1810 #ifdef CONFIG_PROC_FS
1811 static inline struct proto
*__proto_head(void)
1813 return list_entry(proto_list
.next
, struct proto
, node
);
1816 static inline struct proto
*proto_head(void)
1818 return list_empty(&proto_list
) ? NULL
: __proto_head();
1821 static inline struct proto
*proto_next(struct proto
*proto
)
1823 return proto
->node
.next
== &proto_list
? NULL
:
1824 list_entry(proto
->node
.next
, struct proto
, node
);
1827 static inline struct proto
*proto_get_idx(loff_t pos
)
1829 struct proto
*proto
;
1832 list_for_each_entry(proto
, &proto_list
, node
)
1841 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1843 read_lock(&proto_list_lock
);
1844 return *pos
? proto_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1847 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1850 return v
== SEQ_START_TOKEN
? proto_head() : proto_next(v
);
1853 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
1855 read_unlock(&proto_list_lock
);
1858 static char proto_method_implemented(const void *method
)
1860 return method
== NULL
? 'n' : 'y';
1863 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
1865 seq_printf(seq
, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1866 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1869 proto
->sockets_allocated
!= NULL
? atomic_read(proto
->sockets_allocated
) : -1,
1870 proto
->memory_allocated
!= NULL
? atomic_read(proto
->memory_allocated
) : -1,
1871 proto
->memory_pressure
!= NULL
? *proto
->memory_pressure
? "yes" : "no" : "NI",
1873 proto
->slab
== NULL
? "no" : "yes",
1874 module_name(proto
->owner
),
1875 proto_method_implemented(proto
->close
),
1876 proto_method_implemented(proto
->connect
),
1877 proto_method_implemented(proto
->disconnect
),
1878 proto_method_implemented(proto
->accept
),
1879 proto_method_implemented(proto
->ioctl
),
1880 proto_method_implemented(proto
->init
),
1881 proto_method_implemented(proto
->destroy
),
1882 proto_method_implemented(proto
->shutdown
),
1883 proto_method_implemented(proto
->setsockopt
),
1884 proto_method_implemented(proto
->getsockopt
),
1885 proto_method_implemented(proto
->sendmsg
),
1886 proto_method_implemented(proto
->recvmsg
),
1887 proto_method_implemented(proto
->sendpage
),
1888 proto_method_implemented(proto
->bind
),
1889 proto_method_implemented(proto
->backlog_rcv
),
1890 proto_method_implemented(proto
->hash
),
1891 proto_method_implemented(proto
->unhash
),
1892 proto_method_implemented(proto
->get_port
),
1893 proto_method_implemented(proto
->enter_memory_pressure
));
1896 static int proto_seq_show(struct seq_file
*seq
, void *v
)
1898 if (v
== SEQ_START_TOKEN
)
1899 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1908 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1910 proto_seq_printf(seq
, v
);
1914 static struct seq_operations proto_seq_ops
= {
1915 .start
= proto_seq_start
,
1916 .next
= proto_seq_next
,
1917 .stop
= proto_seq_stop
,
1918 .show
= proto_seq_show
,
1921 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
1923 return seq_open(file
, &proto_seq_ops
);
1926 static struct file_operations proto_seq_fops
= {
1927 .owner
= THIS_MODULE
,
1928 .open
= proto_seq_open
,
1930 .llseek
= seq_lseek
,
1931 .release
= seq_release
,
1934 static int __init
proto_init(void)
1936 /* register /proc/net/protocols */
1937 return proc_net_fops_create("protocols", S_IRUGO
, &proto_seq_fops
) == NULL
? -ENOBUFS
: 0;
1940 subsys_initcall(proto_init
);
1942 #endif /* PROC_FS */
1944 EXPORT_SYMBOL(sk_alloc
);
1945 EXPORT_SYMBOL(sk_free
);
1946 EXPORT_SYMBOL(sk_send_sigurg
);
1947 EXPORT_SYMBOL(sock_alloc_send_skb
);
1948 EXPORT_SYMBOL(sock_init_data
);
1949 EXPORT_SYMBOL(sock_kfree_s
);
1950 EXPORT_SYMBOL(sock_kmalloc
);
1951 EXPORT_SYMBOL(sock_no_accept
);
1952 EXPORT_SYMBOL(sock_no_bind
);
1953 EXPORT_SYMBOL(sock_no_connect
);
1954 EXPORT_SYMBOL(sock_no_getname
);
1955 EXPORT_SYMBOL(sock_no_getsockopt
);
1956 EXPORT_SYMBOL(sock_no_ioctl
);
1957 EXPORT_SYMBOL(sock_no_listen
);
1958 EXPORT_SYMBOL(sock_no_mmap
);
1959 EXPORT_SYMBOL(sock_no_poll
);
1960 EXPORT_SYMBOL(sock_no_recvmsg
);
1961 EXPORT_SYMBOL(sock_no_sendmsg
);
1962 EXPORT_SYMBOL(sock_no_sendpage
);
1963 EXPORT_SYMBOL(sock_no_setsockopt
);
1964 EXPORT_SYMBOL(sock_no_shutdown
);
1965 EXPORT_SYMBOL(sock_no_socketpair
);
1966 EXPORT_SYMBOL(sock_rfree
);
1967 EXPORT_SYMBOL(sock_setsockopt
);
1968 EXPORT_SYMBOL(sock_wfree
);
1969 EXPORT_SYMBOL(sock_wmalloc
);
1970 EXPORT_SYMBOL(sock_i_uid
);
1971 EXPORT_SYMBOL(sock_i_ino
);
1972 EXPORT_SYMBOL(sysctl_optmem_max
);
1973 #ifdef CONFIG_SYSCTL
1974 EXPORT_SYMBOL(sysctl_rmem_max
);
1975 EXPORT_SYMBOL(sysctl_wmem_max
);