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 $
12 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
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/config.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/major.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/sock.h>
123 #include <net/xfrm.h>
124 #include <linux/ipsec.h>
126 #include <linux/filter.h>
132 /* Take into consideration the size of the struct sk_buff overhead in the
133 * determination of these values, since that is non-constant across
134 * platforms. This makes socket queueing behavior and performance
135 * not depend upon such differences.
137 #define _SK_MEM_PACKETS 256
138 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
139 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
140 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
142 /* Run time adjustable parameters. */
143 __u32 sysctl_wmem_max
= SK_WMEM_MAX
;
144 __u32 sysctl_rmem_max
= SK_RMEM_MAX
;
145 __u32 sysctl_wmem_default
= SK_WMEM_MAX
;
146 __u32 sysctl_rmem_default
= SK_RMEM_MAX
;
148 /* Maximal space eaten by iovec or ancilliary data plus some space */
149 int sysctl_optmem_max
= sizeof(unsigned long)*(2*UIO_MAXIOV
+ 512);
151 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
155 if (optlen
< sizeof(tv
))
157 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
160 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
161 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
163 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
164 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
168 static void sock_warn_obsolete_bsdism(const char *name
)
171 static char warncomm
[TASK_COMM_LEN
];
172 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
173 strcpy(warncomm
, current
->comm
);
174 printk(KERN_WARNING
"process `%s' is using obsolete "
175 "%s SO_BSDCOMPAT\n", warncomm
, name
);
180 static void sock_disable_timestamp(struct sock
*sk
)
182 if (sock_flag(sk
, SOCK_TIMESTAMP
)) {
183 sock_reset_flag(sk
, SOCK_TIMESTAMP
);
184 net_disable_timestamp();
190 * This is meant for all protocols to use and covers goings on
191 * at the socket level. Everything here is generic.
194 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
195 char __user
*optval
, int optlen
)
197 struct sock
*sk
=sock
->sk
;
198 struct sk_filter
*filter
;
205 * Options without arguments
208 #ifdef SO_DONTLINGER /* Compatibility item... */
211 sock_reset_flag(sk
, SOCK_LINGER
);
216 if(optlen
<sizeof(int))
219 if (get_user(val
, (int __user
*)optval
))
229 if(val
&& !capable(CAP_NET_ADMIN
))
234 sock_set_flag(sk
, SOCK_DBG
);
236 sock_reset_flag(sk
, SOCK_DBG
);
239 sk
->sk_reuse
= valbool
;
247 sock_set_flag(sk
, SOCK_LOCALROUTE
);
249 sock_reset_flag(sk
, SOCK_LOCALROUTE
);
252 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
255 /* Don't error on this BSD doesn't and if you think
256 about it this is right. Otherwise apps have to
257 play 'guess the biggest size' games. RCVBUF/SNDBUF
258 are treated in BSD as hints */
260 if (val
> sysctl_wmem_max
)
261 val
= sysctl_wmem_max
;
263 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
264 if ((val
* 2) < SOCK_MIN_SNDBUF
)
265 sk
->sk_sndbuf
= SOCK_MIN_SNDBUF
;
267 sk
->sk_sndbuf
= val
* 2;
270 * Wake up sending tasks if we
273 sk
->sk_write_space(sk
);
277 /* Don't error on this BSD doesn't and if you think
278 about it this is right. Otherwise apps have to
279 play 'guess the biggest size' games. RCVBUF/SNDBUF
280 are treated in BSD as hints */
282 if (val
> sysctl_rmem_max
)
283 val
= sysctl_rmem_max
;
285 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
286 /* FIXME: is this lower bound the right one? */
287 if ((val
* 2) < SOCK_MIN_RCVBUF
)
288 sk
->sk_rcvbuf
= SOCK_MIN_RCVBUF
;
290 sk
->sk_rcvbuf
= val
* 2;
295 if (sk
->sk_protocol
== IPPROTO_TCP
)
296 tcp_set_keepalive(sk
, valbool
);
298 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
302 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
306 sk
->sk_no_check
= valbool
;
310 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
311 sk
->sk_priority
= val
;
317 if(optlen
<sizeof(ling
)) {
318 ret
= -EINVAL
; /* 1003.1g */
321 if (copy_from_user(&ling
,optval
,sizeof(ling
))) {
326 sock_reset_flag(sk
, SOCK_LINGER
);
328 #if (BITS_PER_LONG == 32)
329 if (ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
330 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
333 sk
->sk_lingertime
= ling
.l_linger
* HZ
;
334 sock_set_flag(sk
, SOCK_LINGER
);
339 sock_warn_obsolete_bsdism("setsockopt");
344 set_bit(SOCK_PASSCRED
, &sock
->flags
);
346 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
351 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
352 sock_enable_timestamp(sk
);
354 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
360 sk
->sk_rcvlowat
= val
? : 1;
364 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
368 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
371 #ifdef CONFIG_NETDEVICES
372 case SO_BINDTODEVICE
:
374 char devname
[IFNAMSIZ
];
377 if (!capable(CAP_NET_RAW
)) {
382 /* Bind this socket to a particular device like "eth0",
383 * as specified in the passed interface name. If the
384 * name is "" or the option length is zero the socket
389 sk
->sk_bound_dev_if
= 0;
391 if (optlen
> IFNAMSIZ
)
393 if (copy_from_user(devname
, optval
, optlen
)) {
398 /* Remove any cached route for this socket. */
401 if (devname
[0] == '\0') {
402 sk
->sk_bound_dev_if
= 0;
404 struct net_device
*dev
= dev_get_by_name(devname
);
409 sk
->sk_bound_dev_if
= dev
->ifindex
;
418 case SO_ATTACH_FILTER
:
420 if (optlen
== sizeof(struct sock_fprog
)) {
421 struct sock_fprog fprog
;
424 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
427 ret
= sk_attach_filter(&fprog
, sk
);
431 case SO_DETACH_FILTER
:
432 spin_lock_bh(&sk
->sk_lock
.slock
);
433 filter
= sk
->sk_filter
;
435 sk
->sk_filter
= NULL
;
436 spin_unlock_bh(&sk
->sk_lock
.slock
);
437 sk_filter_release(sk
, filter
);
440 spin_unlock_bh(&sk
->sk_lock
.slock
);
444 /* We implement the SO_SNDLOWAT etc to
445 not be settable (1003.1g 5.3) */
455 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
456 char __user
*optval
, int __user
*optlen
)
458 struct sock
*sk
= sock
->sk
;
467 unsigned int lv
= sizeof(int);
470 if(get_user(len
,optlen
))
478 v
.val
= sock_flag(sk
, SOCK_DBG
);
482 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
486 v
.val
= !!sock_flag(sk
, SOCK_BROADCAST
);
490 v
.val
= sk
->sk_sndbuf
;
494 v
.val
= sk
->sk_rcvbuf
;
498 v
.val
= sk
->sk_reuse
;
502 v
.val
= !!sock_flag(sk
, SOCK_KEEPOPEN
);
510 v
.val
= -sock_error(sk
);
512 v
.val
= xchg(&sk
->sk_err_soft
, 0);
516 v
.val
= !!sock_flag(sk
, SOCK_URGINLINE
);
520 v
.val
= sk
->sk_no_check
;
524 v
.val
= sk
->sk_priority
;
529 v
.ling
.l_onoff
= !!sock_flag(sk
, SOCK_LINGER
);
530 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
534 sock_warn_obsolete_bsdism("getsockopt");
538 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
);
542 lv
=sizeof(struct timeval
);
543 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
547 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
548 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
553 lv
=sizeof(struct timeval
);
554 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
558 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
559 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
564 v
.val
= sk
->sk_rcvlowat
;
572 v
.val
= test_bit(SOCK_PASSCRED
, &sock
->flags
) ? 1 : 0;
576 if (len
> sizeof(sk
->sk_peercred
))
577 len
= sizeof(sk
->sk_peercred
);
578 if (copy_to_user(optval
, &sk
->sk_peercred
, len
))
586 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
590 if (copy_to_user(optval
, address
, len
))
595 /* Dubious BSD thing... Probably nobody even uses it, but
596 * the UNIX standard wants it for whatever reason... -DaveM
599 v
.val
= sk
->sk_state
== TCP_LISTEN
;
603 return security_socket_getpeersec(sock
, optval
, optlen
, len
);
606 return(-ENOPROTOOPT
);
610 if (copy_to_user(optval
, &v
, len
))
613 if (put_user(len
, optlen
))
619 * sk_alloc - All socket objects are allocated here
620 * @family - protocol family
621 * @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
622 * @prot - struct proto associated with this new sock instance
623 * @zero_it - if we should zero the newly allocated sock
625 struct sock
*sk_alloc(int family
, int priority
, struct proto
*prot
, int zero_it
)
627 struct sock
*sk
= NULL
;
628 kmem_cache_t
*slab
= prot
->slab
;
631 sk
= kmem_cache_alloc(slab
, priority
);
633 sk
= kmalloc(prot
->obj_size
, priority
);
637 memset(sk
, 0, prot
->obj_size
);
638 sk
->sk_family
= family
;
643 if (security_sk_alloc(sk
, family
, priority
)) {
645 kmem_cache_free(slab
, sk
);
650 __module_get(prot
->owner
);
655 void sk_free(struct sock
*sk
)
657 struct sk_filter
*filter
;
658 struct module
*owner
= sk
->sk_prot
->owner
;
663 filter
= sk
->sk_filter
;
665 sk_filter_release(sk
, filter
);
666 sk
->sk_filter
= NULL
;
669 sock_disable_timestamp(sk
);
671 if (atomic_read(&sk
->sk_omem_alloc
))
672 printk(KERN_DEBUG
"%s: optmem leakage (%d bytes) detected.\n",
673 __FUNCTION__
, atomic_read(&sk
->sk_omem_alloc
));
675 security_sk_free(sk
);
676 if (sk
->sk_prot
->slab
!= NULL
)
677 kmem_cache_free(sk
->sk_prot
->slab
, sk
);
683 void __init
sk_init(void)
685 if (num_physpages
<= 4096) {
686 sysctl_wmem_max
= 32767;
687 sysctl_rmem_max
= 32767;
688 sysctl_wmem_default
= 32767;
689 sysctl_rmem_default
= 32767;
690 } else if (num_physpages
>= 131072) {
691 sysctl_wmem_max
= 131071;
692 sysctl_rmem_max
= 131071;
697 * Simple resource managers for sockets.
702 * Write buffer destructor automatically called from kfree_skb.
704 void sock_wfree(struct sk_buff
*skb
)
706 struct sock
*sk
= skb
->sk
;
708 /* In case it might be waiting for more memory. */
709 atomic_sub(skb
->truesize
, &sk
->sk_wmem_alloc
);
710 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
))
711 sk
->sk_write_space(sk
);
716 * Read buffer destructor automatically called from kfree_skb.
718 void sock_rfree(struct sk_buff
*skb
)
720 struct sock
*sk
= skb
->sk
;
722 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
726 int sock_i_uid(struct sock
*sk
)
730 read_lock(&sk
->sk_callback_lock
);
731 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: 0;
732 read_unlock(&sk
->sk_callback_lock
);
736 unsigned long sock_i_ino(struct sock
*sk
)
740 read_lock(&sk
->sk_callback_lock
);
741 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
742 read_unlock(&sk
->sk_callback_lock
);
747 * Allocate a skb from the socket's send buffer.
749 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
, int priority
)
751 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
752 struct sk_buff
* skb
= alloc_skb(size
, priority
);
754 skb_set_owner_w(skb
, sk
);
762 * Allocate a skb from the socket's receive buffer.
764 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
, int priority
)
766 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
767 struct sk_buff
*skb
= alloc_skb(size
, priority
);
769 skb_set_owner_r(skb
, sk
);
777 * Allocate a memory block from the socket's option memory buffer.
779 void *sock_kmalloc(struct sock
*sk
, int size
, int priority
)
781 if ((unsigned)size
<= sysctl_optmem_max
&&
782 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
784 /* First do the add, to avoid the race if kmalloc
787 atomic_add(size
, &sk
->sk_omem_alloc
);
788 mem
= kmalloc(size
, priority
);
791 atomic_sub(size
, &sk
->sk_omem_alloc
);
797 * Free an option memory block.
799 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
802 atomic_sub(size
, &sk
->sk_omem_alloc
);
805 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
806 I think, these locks should be removed for datagram sockets.
808 static long sock_wait_for_wmem(struct sock
* sk
, long timeo
)
812 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
816 if (signal_pending(current
))
818 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
819 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
820 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
822 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
826 timeo
= schedule_timeout(timeo
);
828 finish_wait(sk
->sk_sleep
, &wait
);
834 * Generic send/receive buffer handlers
837 static struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
,
838 unsigned long header_len
,
839 unsigned long data_len
,
840 int noblock
, int *errcode
)
843 unsigned int gfp_mask
;
847 gfp_mask
= sk
->sk_allocation
;
848 if (gfp_mask
& __GFP_WAIT
)
849 gfp_mask
|= __GFP_REPEAT
;
851 timeo
= sock_sndtimeo(sk
, noblock
);
853 err
= sock_error(sk
);
858 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
861 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
862 skb
= alloc_skb(header_len
, sk
->sk_allocation
);
867 /* No pages, we're done... */
871 npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
872 skb
->truesize
+= data_len
;
873 skb_shinfo(skb
)->nr_frags
= npages
;
874 for (i
= 0; i
< npages
; i
++) {
878 page
= alloc_pages(sk
->sk_allocation
, 0);
881 skb_shinfo(skb
)->nr_frags
= i
;
886 frag
= &skb_shinfo(skb
)->frags
[i
];
888 frag
->page_offset
= 0;
889 frag
->size
= (data_len
>= PAGE_SIZE
?
892 data_len
-= PAGE_SIZE
;
895 /* Full success... */
901 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
902 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
906 if (signal_pending(current
))
908 timeo
= sock_wait_for_wmem(sk
, timeo
);
911 skb_set_owner_w(skb
, sk
);
915 err
= sock_intr_errno(timeo
);
921 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
922 int noblock
, int *errcode
)
924 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
927 static void __lock_sock(struct sock
*sk
)
932 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
933 TASK_UNINTERRUPTIBLE
);
934 spin_unlock_bh(&sk
->sk_lock
.slock
);
936 spin_lock_bh(&sk
->sk_lock
.slock
);
937 if(!sock_owned_by_user(sk
))
940 finish_wait(&sk
->sk_lock
.wq
, &wait
);
943 static void __release_sock(struct sock
*sk
)
945 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
948 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
952 struct sk_buff
*next
= skb
->next
;
955 sk
->sk_backlog_rcv(sk
, skb
);
958 * We are in process context here with softirqs
959 * disabled, use cond_resched_softirq() to preempt.
960 * This is safe to do because we've taken the backlog
963 cond_resched_softirq();
966 } while (skb
!= NULL
);
969 } while((skb
= sk
->sk_backlog
.head
) != NULL
);
973 * sk_wait_data - wait for data to arrive at sk_receive_queue
974 * sk - sock to wait on
975 * timeo - for how long
977 * Now socket state including sk->sk_err is changed only under lock,
978 * hence we may omit checks after joining wait queue.
979 * We check receive queue before schedule() only as optimization;
980 * it is very likely that release_sock() added new data.
982 int sk_wait_data(struct sock
*sk
, long *timeo
)
987 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
988 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
989 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
990 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
991 finish_wait(sk
->sk_sleep
, &wait
);
995 EXPORT_SYMBOL(sk_wait_data
);
998 * Set of default routines for initialising struct proto_ops when
999 * the protocol does not support a particular function. In certain
1000 * cases where it makes no sense for a protocol to have a "do nothing"
1001 * function, some default processing is provided.
1004 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1009 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1015 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1020 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1025 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1031 unsigned int sock_no_poll(struct file
* file
, struct socket
*sock
, poll_table
*pt
)
1036 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1041 int sock_no_listen(struct socket
*sock
, int backlog
)
1046 int sock_no_shutdown(struct socket
*sock
, int how
)
1051 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
1052 char __user
*optval
, int optlen
)
1057 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
1058 char __user
*optval
, int __user
*optlen
)
1063 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1069 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1070 size_t len
, int flags
)
1075 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
1077 /* Mirror missing mmap method error code */
1081 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
1084 struct msghdr msg
= {.msg_flags
= flags
};
1086 char *kaddr
= kmap(page
);
1087 iov
.iov_base
= kaddr
+ offset
;
1089 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1095 * Default Socket Callbacks
1098 static void sock_def_wakeup(struct sock
*sk
)
1100 read_lock(&sk
->sk_callback_lock
);
1101 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1102 wake_up_interruptible_all(sk
->sk_sleep
);
1103 read_unlock(&sk
->sk_callback_lock
);
1106 static void sock_def_error_report(struct sock
*sk
)
1108 read_lock(&sk
->sk_callback_lock
);
1109 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1110 wake_up_interruptible(sk
->sk_sleep
);
1111 sk_wake_async(sk
,0,POLL_ERR
);
1112 read_unlock(&sk
->sk_callback_lock
);
1115 static void sock_def_readable(struct sock
*sk
, int len
)
1117 read_lock(&sk
->sk_callback_lock
);
1118 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1119 wake_up_interruptible(sk
->sk_sleep
);
1120 sk_wake_async(sk
,1,POLL_IN
);
1121 read_unlock(&sk
->sk_callback_lock
);
1124 static void sock_def_write_space(struct sock
*sk
)
1126 read_lock(&sk
->sk_callback_lock
);
1128 /* Do not wake up a writer until he can make "significant"
1131 if((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
1132 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1133 wake_up_interruptible(sk
->sk_sleep
);
1135 /* Should agree with poll, otherwise some programs break */
1136 if (sock_writeable(sk
))
1137 sk_wake_async(sk
, 2, POLL_OUT
);
1140 read_unlock(&sk
->sk_callback_lock
);
1143 static void sock_def_destruct(struct sock
*sk
)
1145 if (sk
->sk_protinfo
)
1146 kfree(sk
->sk_protinfo
);
1149 void sk_send_sigurg(struct sock
*sk
)
1151 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
1152 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
1153 sk_wake_async(sk
, 3, POLL_PRI
);
1156 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
1157 unsigned long expires
)
1159 if (!mod_timer(timer
, expires
))
1163 EXPORT_SYMBOL(sk_reset_timer
);
1165 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
1167 if (timer_pending(timer
) && del_timer(timer
))
1171 EXPORT_SYMBOL(sk_stop_timer
);
1173 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
1175 skb_queue_head_init(&sk
->sk_receive_queue
);
1176 skb_queue_head_init(&sk
->sk_write_queue
);
1177 skb_queue_head_init(&sk
->sk_error_queue
);
1179 sk
->sk_send_head
= NULL
;
1181 init_timer(&sk
->sk_timer
);
1183 sk
->sk_allocation
= GFP_KERNEL
;
1184 sk
->sk_rcvbuf
= sysctl_rmem_default
;
1185 sk
->sk_sndbuf
= sysctl_wmem_default
;
1186 sk
->sk_state
= TCP_CLOSE
;
1187 sk
->sk_socket
= sock
;
1189 sock_set_flag(sk
, SOCK_ZAPPED
);
1193 sk
->sk_type
= sock
->type
;
1194 sk
->sk_sleep
= &sock
->wait
;
1197 sk
->sk_sleep
= NULL
;
1199 rwlock_init(&sk
->sk_dst_lock
);
1200 rwlock_init(&sk
->sk_callback_lock
);
1202 sk
->sk_state_change
= sock_def_wakeup
;
1203 sk
->sk_data_ready
= sock_def_readable
;
1204 sk
->sk_write_space
= sock_def_write_space
;
1205 sk
->sk_error_report
= sock_def_error_report
;
1206 sk
->sk_destruct
= sock_def_destruct
;
1208 sk
->sk_sndmsg_page
= NULL
;
1209 sk
->sk_sndmsg_off
= 0;
1211 sk
->sk_peercred
.pid
= 0;
1212 sk
->sk_peercred
.uid
= -1;
1213 sk
->sk_peercred
.gid
= -1;
1214 sk
->sk_write_pending
= 0;
1215 sk
->sk_rcvlowat
= 1;
1216 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
1217 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
1219 sk
->sk_stamp
.tv_sec
= -1L;
1220 sk
->sk_stamp
.tv_usec
= -1L;
1222 atomic_set(&sk
->sk_refcnt
, 1);
1225 void fastcall
lock_sock(struct sock
*sk
)
1228 spin_lock_bh(&(sk
->sk_lock
.slock
));
1229 if (sk
->sk_lock
.owner
)
1231 sk
->sk_lock
.owner
= (void *)1;
1232 spin_unlock_bh(&(sk
->sk_lock
.slock
));
1235 EXPORT_SYMBOL(lock_sock
);
1237 void fastcall
release_sock(struct sock
*sk
)
1239 spin_lock_bh(&(sk
->sk_lock
.slock
));
1240 if (sk
->sk_backlog
.tail
)
1242 sk
->sk_lock
.owner
= NULL
;
1243 if (waitqueue_active(&(sk
->sk_lock
.wq
)))
1244 wake_up(&(sk
->sk_lock
.wq
));
1245 spin_unlock_bh(&(sk
->sk_lock
.slock
));
1247 EXPORT_SYMBOL(release_sock
);
1249 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
1251 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
1252 sock_enable_timestamp(sk
);
1253 if (sk
->sk_stamp
.tv_sec
== -1)
1255 if (sk
->sk_stamp
.tv_sec
== 0)
1256 do_gettimeofday(&sk
->sk_stamp
);
1257 return copy_to_user(userstamp
, &sk
->sk_stamp
, sizeof(struct timeval
)) ?
1260 EXPORT_SYMBOL(sock_get_timestamp
);
1262 void sock_enable_timestamp(struct sock
*sk
)
1264 if (!sock_flag(sk
, SOCK_TIMESTAMP
)) {
1265 sock_set_flag(sk
, SOCK_TIMESTAMP
);
1266 net_enable_timestamp();
1269 EXPORT_SYMBOL(sock_enable_timestamp
);
1272 * Get a socket option on an socket.
1274 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1275 * asynchronous errors should be reported by getsockopt. We assume
1276 * this means if you specify SO_ERROR (otherwise whats the point of it).
1278 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1279 char __user
*optval
, int __user
*optlen
)
1281 struct sock
*sk
= sock
->sk
;
1283 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
1286 EXPORT_SYMBOL(sock_common_getsockopt
);
1288 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1289 struct msghdr
*msg
, size_t size
, int flags
)
1291 struct sock
*sk
= sock
->sk
;
1295 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
1296 flags
& ~MSG_DONTWAIT
, &addr_len
);
1298 msg
->msg_namelen
= addr_len
;
1302 EXPORT_SYMBOL(sock_common_recvmsg
);
1305 * Set socket options on an inet socket.
1307 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1308 char __user
*optval
, int optlen
)
1310 struct sock
*sk
= sock
->sk
;
1312 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
1315 EXPORT_SYMBOL(sock_common_setsockopt
);
1317 void sk_common_release(struct sock
*sk
)
1319 if (sk
->sk_prot
->destroy
)
1320 sk
->sk_prot
->destroy(sk
);
1323 * Observation: when sock_common_release is called, processes have
1324 * no access to socket. But net still has.
1325 * Step one, detach it from networking:
1327 * A. Remove from hash tables.
1330 sk
->sk_prot
->unhash(sk
);
1333 * In this point socket cannot receive new packets, but it is possible
1334 * that some packets are in flight because some CPU runs receiver and
1335 * did hash table lookup before we unhashed socket. They will achieve
1336 * receive queue and will be purged by socket destructor.
1338 * Also we still have packets pending on receive queue and probably,
1339 * our own packets waiting in device queues. sock_destroy will drain
1340 * receive queue, but transmitted packets will delay socket destruction
1341 * until the last reference will be released.
1346 xfrm_sk_free_policy(sk
);
1348 #ifdef INET_REFCNT_DEBUG
1349 if (atomic_read(&sk
->sk_refcnt
) != 1)
1350 printk(KERN_DEBUG
"Destruction of the socket %p delayed, c=%d\n",
1351 sk
, atomic_read(&sk
->sk_refcnt
));
1356 EXPORT_SYMBOL(sk_common_release
);
1358 static DEFINE_RWLOCK(proto_list_lock
);
1359 static LIST_HEAD(proto_list
);
1361 int proto_register(struct proto
*prot
, int alloc_slab
)
1366 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
1367 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1369 if (prot
->slab
== NULL
) {
1370 printk(KERN_CRIT
"%s: Can't create sock SLAB cache!\n",
1376 write_lock(&proto_list_lock
);
1377 list_add(&prot
->node
, &proto_list
);
1378 write_unlock(&proto_list_lock
);
1384 EXPORT_SYMBOL(proto_register
);
1386 void proto_unregister(struct proto
*prot
)
1388 write_lock(&proto_list_lock
);
1390 if (prot
->slab
!= NULL
) {
1391 kmem_cache_destroy(prot
->slab
);
1395 list_del(&prot
->node
);
1396 write_unlock(&proto_list_lock
);
1399 EXPORT_SYMBOL(proto_unregister
);
1401 #ifdef CONFIG_PROC_FS
1402 static inline struct proto
*__proto_head(void)
1404 return list_entry(proto_list
.next
, struct proto
, node
);
1407 static inline struct proto
*proto_head(void)
1409 return list_empty(&proto_list
) ? NULL
: __proto_head();
1412 static inline struct proto
*proto_next(struct proto
*proto
)
1414 return proto
->node
.next
== &proto_list
? NULL
:
1415 list_entry(proto
->node
.next
, struct proto
, node
);
1418 static inline struct proto
*proto_get_idx(loff_t pos
)
1420 struct proto
*proto
;
1423 list_for_each_entry(proto
, &proto_list
, node
)
1432 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1434 read_lock(&proto_list_lock
);
1435 return *pos
? proto_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1438 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1441 return v
== SEQ_START_TOKEN
? proto_head() : proto_next(v
);
1444 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
1446 read_unlock(&proto_list_lock
);
1449 static char proto_method_implemented(const void *method
)
1451 return method
== NULL
? 'n' : 'y';
1454 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
1456 seq_printf(seq
, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1457 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1460 proto
->sockets_allocated
!= NULL
? atomic_read(proto
->sockets_allocated
) : -1,
1461 proto
->memory_allocated
!= NULL
? atomic_read(proto
->memory_allocated
) : -1,
1462 proto
->memory_pressure
!= NULL
? *proto
->memory_pressure
? "yes" : "no" : "NI",
1464 proto
->slab
== NULL
? "no" : "yes",
1465 module_name(proto
->owner
),
1466 proto_method_implemented(proto
->close
),
1467 proto_method_implemented(proto
->connect
),
1468 proto_method_implemented(proto
->disconnect
),
1469 proto_method_implemented(proto
->accept
),
1470 proto_method_implemented(proto
->ioctl
),
1471 proto_method_implemented(proto
->init
),
1472 proto_method_implemented(proto
->destroy
),
1473 proto_method_implemented(proto
->shutdown
),
1474 proto_method_implemented(proto
->setsockopt
),
1475 proto_method_implemented(proto
->getsockopt
),
1476 proto_method_implemented(proto
->sendmsg
),
1477 proto_method_implemented(proto
->recvmsg
),
1478 proto_method_implemented(proto
->sendpage
),
1479 proto_method_implemented(proto
->bind
),
1480 proto_method_implemented(proto
->backlog_rcv
),
1481 proto_method_implemented(proto
->hash
),
1482 proto_method_implemented(proto
->unhash
),
1483 proto_method_implemented(proto
->get_port
),
1484 proto_method_implemented(proto
->enter_memory_pressure
));
1487 static int proto_seq_show(struct seq_file
*seq
, void *v
)
1489 if (v
== SEQ_START_TOKEN
)
1490 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1499 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1501 proto_seq_printf(seq
, v
);
1505 static struct seq_operations proto_seq_ops
= {
1506 .start
= proto_seq_start
,
1507 .next
= proto_seq_next
,
1508 .stop
= proto_seq_stop
,
1509 .show
= proto_seq_show
,
1512 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
1514 return seq_open(file
, &proto_seq_ops
);
1517 static struct file_operations proto_seq_fops
= {
1518 .owner
= THIS_MODULE
,
1519 .open
= proto_seq_open
,
1521 .llseek
= seq_lseek
,
1522 .release
= seq_release
,
1525 static int __init
proto_init(void)
1527 /* register /proc/net/protocols */
1528 return proc_net_fops_create("protocols", S_IRUGO
, &proto_seq_fops
) == NULL
? -ENOBUFS
: 0;
1531 subsys_initcall(proto_init
);
1533 #endif /* PROC_FS */
1535 EXPORT_SYMBOL(sk_alloc
);
1536 EXPORT_SYMBOL(sk_free
);
1537 EXPORT_SYMBOL(sk_send_sigurg
);
1538 EXPORT_SYMBOL(sock_alloc_send_skb
);
1539 EXPORT_SYMBOL(sock_init_data
);
1540 EXPORT_SYMBOL(sock_kfree_s
);
1541 EXPORT_SYMBOL(sock_kmalloc
);
1542 EXPORT_SYMBOL(sock_no_accept
);
1543 EXPORT_SYMBOL(sock_no_bind
);
1544 EXPORT_SYMBOL(sock_no_connect
);
1545 EXPORT_SYMBOL(sock_no_getname
);
1546 EXPORT_SYMBOL(sock_no_getsockopt
);
1547 EXPORT_SYMBOL(sock_no_ioctl
);
1548 EXPORT_SYMBOL(sock_no_listen
);
1549 EXPORT_SYMBOL(sock_no_mmap
);
1550 EXPORT_SYMBOL(sock_no_poll
);
1551 EXPORT_SYMBOL(sock_no_recvmsg
);
1552 EXPORT_SYMBOL(sock_no_sendmsg
);
1553 EXPORT_SYMBOL(sock_no_sendpage
);
1554 EXPORT_SYMBOL(sock_no_setsockopt
);
1555 EXPORT_SYMBOL(sock_no_shutdown
);
1556 EXPORT_SYMBOL(sock_no_socketpair
);
1557 EXPORT_SYMBOL(sock_rfree
);
1558 EXPORT_SYMBOL(sock_setsockopt
);
1559 EXPORT_SYMBOL(sock_wfree
);
1560 EXPORT_SYMBOL(sock_wmalloc
);
1561 EXPORT_SYMBOL(sock_i_uid
);
1562 EXPORT_SYMBOL(sock_i_ino
);
1563 #ifdef CONFIG_SYSCTL
1564 EXPORT_SYMBOL(sysctl_optmem_max
);
1565 EXPORT_SYMBOL(sysctl_rmem_max
);
1566 EXPORT_SYMBOL(sysctl_wmem_max
);