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>
114 #include <linux/highmem.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/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
128 #include <linux/filter.h>
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
138 static struct lock_class_key af_family_keys
[AF_MAX
];
139 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
147 static const char *af_family_key_strings
[AF_MAX
+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
161 static const char *af_family_slock_key_strings
[AF_MAX
+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
171 "slock-27" , "slock-28" , "slock-29" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
175 static const char *af_family_clock_key_strings
[AF_MAX
+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-29" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_MAX"
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
195 static struct lock_class_key af_callback_keys
[AF_MAX
];
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
209 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
210 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
211 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
216 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
220 if (optlen
< sizeof(tv
))
222 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
224 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
228 static int warned __read_mostly
;
231 if (warned
< 10 && net_ratelimit())
233 printk(KERN_INFO
"sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
235 current
->comm
, task_pid_nr(current
));
238 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
239 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
241 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
242 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
246 static void sock_warn_obsolete_bsdism(const char *name
)
249 static char warncomm
[TASK_COMM_LEN
];
250 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
251 strcpy(warncomm
, current
->comm
);
252 printk(KERN_WARNING
"process `%s' is using obsolete "
253 "%s SO_BSDCOMPAT\n", warncomm
, name
);
258 static void sock_disable_timestamp(struct sock
*sk
)
260 if (sock_flag(sk
, SOCK_TIMESTAMP
)) {
261 sock_reset_flag(sk
, SOCK_TIMESTAMP
);
262 net_disable_timestamp();
267 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
272 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
273 number of warnings when compiling with -W --ANK
275 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
276 (unsigned)sk
->sk_rcvbuf
) {
281 err
= sk_filter(sk
, skb
);
286 skb_set_owner_r(skb
, sk
);
288 /* Cache the SKB length before we tack it onto the receive
289 * queue. Once it is added it no longer belongs to us and
290 * may be freed by other threads of control pulling packets
295 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
297 if (!sock_flag(sk
, SOCK_DEAD
))
298 sk
->sk_data_ready(sk
, skb_len
);
302 EXPORT_SYMBOL(sock_queue_rcv_skb
);
304 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
306 int rc
= NET_RX_SUCCESS
;
308 if (sk_filter(sk
, skb
))
309 goto discard_and_relse
;
314 bh_lock_sock_nested(sk
);
317 if (!sock_owned_by_user(sk
)) {
319 * trylock + unlock semantics:
321 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
323 rc
= sk
->sk_backlog_rcv(sk
, skb
);
325 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
327 sk_add_backlog(sk
, skb
);
336 EXPORT_SYMBOL(sk_receive_skb
);
338 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
340 struct dst_entry
*dst
= sk
->sk_dst_cache
;
342 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
343 sk
->sk_dst_cache
= NULL
;
350 EXPORT_SYMBOL(__sk_dst_check
);
352 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
354 struct dst_entry
*dst
= sk_dst_get(sk
);
356 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
364 EXPORT_SYMBOL(sk_dst_check
);
366 static int sock_bindtodevice(struct sock
*sk
, char __user
*optval
, int optlen
)
368 int ret
= -ENOPROTOOPT
;
369 #ifdef CONFIG_NETDEVICES
370 struct net
*net
= sk
->sk_net
;
371 char devname
[IFNAMSIZ
];
376 if (!capable(CAP_NET_RAW
))
383 /* Bind this socket to a particular device like "eth0",
384 * as specified in the passed interface name. If the
385 * name is "" or the option length is zero the socket
388 if (optlen
> IFNAMSIZ
- 1)
389 optlen
= IFNAMSIZ
- 1;
390 memset(devname
, 0, sizeof(devname
));
393 if (copy_from_user(devname
, optval
, optlen
))
396 if (devname
[0] == '\0') {
399 struct net_device
*dev
= dev_get_by_name(net
, devname
);
405 index
= dev
->ifindex
;
410 sk
->sk_bound_dev_if
= index
;
423 * This is meant for all protocols to use and covers goings on
424 * at the socket level. Everything here is generic.
427 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
428 char __user
*optval
, int optlen
)
430 struct sock
*sk
=sock
->sk
;
437 * Options without arguments
440 #ifdef SO_DONTLINGER /* Compatibility item... */
441 if (optname
== SO_DONTLINGER
) {
443 sock_reset_flag(sk
, SOCK_LINGER
);
449 if (optname
== SO_BINDTODEVICE
)
450 return sock_bindtodevice(sk
, optval
, optlen
);
452 if (optlen
< sizeof(int))
455 if (get_user(val
, (int __user
*)optval
))
464 if (val
&& !capable(CAP_NET_ADMIN
)) {
468 sock_set_flag(sk
, SOCK_DBG
);
470 sock_reset_flag(sk
, SOCK_DBG
);
473 sk
->sk_reuse
= valbool
;
481 sock_set_flag(sk
, SOCK_LOCALROUTE
);
483 sock_reset_flag(sk
, SOCK_LOCALROUTE
);
486 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
489 /* Don't error on this BSD doesn't and if you think
490 about it this is right. Otherwise apps have to
491 play 'guess the biggest size' games. RCVBUF/SNDBUF
492 are treated in BSD as hints */
494 if (val
> sysctl_wmem_max
)
495 val
= sysctl_wmem_max
;
497 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
498 if ((val
* 2) < SOCK_MIN_SNDBUF
)
499 sk
->sk_sndbuf
= SOCK_MIN_SNDBUF
;
501 sk
->sk_sndbuf
= val
* 2;
504 * Wake up sending tasks if we
507 sk
->sk_write_space(sk
);
511 if (!capable(CAP_NET_ADMIN
)) {
518 /* Don't error on this BSD doesn't and if you think
519 about it this is right. Otherwise apps have to
520 play 'guess the biggest size' games. RCVBUF/SNDBUF
521 are treated in BSD as hints */
523 if (val
> sysctl_rmem_max
)
524 val
= sysctl_rmem_max
;
526 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
528 * We double it on the way in to account for
529 * "struct sk_buff" etc. overhead. Applications
530 * assume that the SO_RCVBUF setting they make will
531 * allow that much actual data to be received on that
534 * Applications are unaware that "struct sk_buff" and
535 * other overheads allocate from the receive buffer
536 * during socket buffer allocation.
538 * And after considering the possible alternatives,
539 * returning the value we actually used in getsockopt
540 * is the most desirable behavior.
542 if ((val
* 2) < SOCK_MIN_RCVBUF
)
543 sk
->sk_rcvbuf
= SOCK_MIN_RCVBUF
;
545 sk
->sk_rcvbuf
= val
* 2;
549 if (!capable(CAP_NET_ADMIN
)) {
557 if (sk
->sk_protocol
== IPPROTO_TCP
)
558 tcp_set_keepalive(sk
, valbool
);
560 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
564 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
568 sk
->sk_no_check
= valbool
;
572 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
573 sk
->sk_priority
= val
;
579 if (optlen
< sizeof(ling
)) {
580 ret
= -EINVAL
; /* 1003.1g */
583 if (copy_from_user(&ling
,optval
,sizeof(ling
))) {
588 sock_reset_flag(sk
, SOCK_LINGER
);
590 #if (BITS_PER_LONG == 32)
591 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
592 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
595 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
596 sock_set_flag(sk
, SOCK_LINGER
);
601 sock_warn_obsolete_bsdism("setsockopt");
606 set_bit(SOCK_PASSCRED
, &sock
->flags
);
608 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
614 if (optname
== SO_TIMESTAMP
)
615 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
617 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
618 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
619 sock_enable_timestamp(sk
);
621 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
622 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
629 sk
->sk_rcvlowat
= val
? : 1;
633 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
637 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
640 case SO_ATTACH_FILTER
:
642 if (optlen
== sizeof(struct sock_fprog
)) {
643 struct sock_fprog fprog
;
646 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
649 ret
= sk_attach_filter(&fprog
, sk
);
653 case SO_DETACH_FILTER
:
654 ret
= sk_detach_filter(sk
);
659 set_bit(SOCK_PASSSEC
, &sock
->flags
);
661 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
664 /* We implement the SO_SNDLOWAT etc to
665 not be settable (1003.1g 5.3) */
675 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
676 char __user
*optval
, int __user
*optlen
)
678 struct sock
*sk
= sock
->sk
;
686 unsigned int lv
= sizeof(int);
689 if (get_user(len
, optlen
))
696 v
.val
= sock_flag(sk
, SOCK_DBG
);
700 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
704 v
.val
= !!sock_flag(sk
, SOCK_BROADCAST
);
708 v
.val
= sk
->sk_sndbuf
;
712 v
.val
= sk
->sk_rcvbuf
;
716 v
.val
= sk
->sk_reuse
;
720 v
.val
= !!sock_flag(sk
, SOCK_KEEPOPEN
);
728 v
.val
= -sock_error(sk
);
730 v
.val
= xchg(&sk
->sk_err_soft
, 0);
734 v
.val
= !!sock_flag(sk
, SOCK_URGINLINE
);
738 v
.val
= sk
->sk_no_check
;
742 v
.val
= sk
->sk_priority
;
747 v
.ling
.l_onoff
= !!sock_flag(sk
, SOCK_LINGER
);
748 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
752 sock_warn_obsolete_bsdism("getsockopt");
756 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
757 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
761 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
765 lv
=sizeof(struct timeval
);
766 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
770 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
771 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
776 lv
=sizeof(struct timeval
);
777 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
781 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
782 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
787 v
.val
= sk
->sk_rcvlowat
;
795 v
.val
= test_bit(SOCK_PASSCRED
, &sock
->flags
) ? 1 : 0;
799 if (len
> sizeof(sk
->sk_peercred
))
800 len
= sizeof(sk
->sk_peercred
);
801 if (copy_to_user(optval
, &sk
->sk_peercred
, len
))
809 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
813 if (copy_to_user(optval
, address
, len
))
818 /* Dubious BSD thing... Probably nobody even uses it, but
819 * the UNIX standard wants it for whatever reason... -DaveM
822 v
.val
= sk
->sk_state
== TCP_LISTEN
;
826 v
.val
= test_bit(SOCK_PASSSEC
, &sock
->flags
) ? 1 : 0;
830 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
838 if (copy_to_user(optval
, &v
, len
))
841 if (put_user(len
, optlen
))
847 * Initialize an sk_lock.
849 * (We also register the sk_lock with the lock validator.)
851 static inline void sock_lock_init(struct sock
*sk
)
853 sock_lock_init_class_and_name(sk
,
854 af_family_slock_key_strings
[sk
->sk_family
],
855 af_family_slock_keys
+ sk
->sk_family
,
856 af_family_key_strings
[sk
->sk_family
],
857 af_family_keys
+ sk
->sk_family
);
860 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
862 #ifdef CONFIG_SECURITY_NETWORK
863 void *sptr
= nsk
->sk_security
;
866 memcpy(nsk
, osk
, osk
->sk_prot
->obj_size
);
867 #ifdef CONFIG_SECURITY_NETWORK
868 nsk
->sk_security
= sptr
;
869 security_sk_clone(osk
, nsk
);
873 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
877 struct kmem_cache
*slab
;
881 sk
= kmem_cache_alloc(slab
, priority
);
883 sk
= kmalloc(prot
->obj_size
, priority
);
886 if (security_sk_alloc(sk
, family
, priority
))
889 if (!try_module_get(prot
->owner
))
896 security_sk_free(sk
);
899 kmem_cache_free(slab
, sk
);
905 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
907 struct kmem_cache
*slab
;
908 struct module
*owner
;
913 security_sk_free(sk
);
915 kmem_cache_free(slab
, sk
);
922 * sk_alloc - All socket objects are allocated here
923 * @net: the applicable net namespace
924 * @family: protocol family
925 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
926 * @prot: struct proto associated with this new sock instance
927 * @zero_it: if we should zero the newly allocated sock
929 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
934 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
936 sk
->sk_family
= family
;
938 * See comment in struct sock definition to understand
939 * why we need sk_prot_creator -acme
941 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
943 sk
->sk_net
= get_net(net
);
949 void sk_free(struct sock
*sk
)
951 struct sk_filter
*filter
;
956 filter
= rcu_dereference(sk
->sk_filter
);
958 sk_filter_uncharge(sk
, filter
);
959 rcu_assign_pointer(sk
->sk_filter
, NULL
);
962 sock_disable_timestamp(sk
);
964 if (atomic_read(&sk
->sk_omem_alloc
))
965 printk(KERN_DEBUG
"%s: optmem leakage (%d bytes) detected.\n",
966 __FUNCTION__
, atomic_read(&sk
->sk_omem_alloc
));
969 sk_prot_free(sk
->sk_prot_creator
, sk
);
972 struct sock
*sk_clone(const struct sock
*sk
, const gfp_t priority
)
976 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
978 struct sk_filter
*filter
;
980 sock_copy(newsk
, sk
);
983 get_net(newsk
->sk_net
);
984 sk_node_init(&newsk
->sk_node
);
985 sock_lock_init(newsk
);
987 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
989 atomic_set(&newsk
->sk_rmem_alloc
, 0);
990 atomic_set(&newsk
->sk_wmem_alloc
, 0);
991 atomic_set(&newsk
->sk_omem_alloc
, 0);
992 skb_queue_head_init(&newsk
->sk_receive_queue
);
993 skb_queue_head_init(&newsk
->sk_write_queue
);
994 #ifdef CONFIG_NET_DMA
995 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
998 rwlock_init(&newsk
->sk_dst_lock
);
999 rwlock_init(&newsk
->sk_callback_lock
);
1000 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1001 af_callback_keys
+ newsk
->sk_family
,
1002 af_family_clock_key_strings
[newsk
->sk_family
]);
1004 newsk
->sk_dst_cache
= NULL
;
1005 newsk
->sk_wmem_queued
= 0;
1006 newsk
->sk_forward_alloc
= 0;
1007 newsk
->sk_send_head
= NULL
;
1008 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1010 sock_reset_flag(newsk
, SOCK_DONE
);
1011 skb_queue_head_init(&newsk
->sk_error_queue
);
1013 filter
= newsk
->sk_filter
;
1015 sk_filter_charge(newsk
, filter
);
1017 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1018 /* It is still raw copy of parent, so invalidate
1019 * destructor and make plain sk_free() */
1020 newsk
->sk_destruct
= NULL
;
1027 newsk
->sk_priority
= 0;
1028 atomic_set(&newsk
->sk_refcnt
, 2);
1031 * Increment the counter in the same struct proto as the master
1032 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1033 * is the same as sk->sk_prot->socks, as this field was copied
1036 * This _changes_ the previous behaviour, where
1037 * tcp_create_openreq_child always was incrementing the
1038 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1039 * to be taken into account in all callers. -acme
1041 sk_refcnt_debug_inc(newsk
);
1042 newsk
->sk_socket
= NULL
;
1043 newsk
->sk_sleep
= NULL
;
1045 if (newsk
->sk_prot
->sockets_allocated
)
1046 atomic_inc(newsk
->sk_prot
->sockets_allocated
);
1052 EXPORT_SYMBOL_GPL(sk_clone
);
1054 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1056 __sk_dst_set(sk
, dst
);
1057 sk
->sk_route_caps
= dst
->dev
->features
;
1058 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1059 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1060 if (sk_can_gso(sk
)) {
1061 if (dst
->header_len
)
1062 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1064 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1067 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1069 void __init
sk_init(void)
1071 if (num_physpages
<= 4096) {
1072 sysctl_wmem_max
= 32767;
1073 sysctl_rmem_max
= 32767;
1074 sysctl_wmem_default
= 32767;
1075 sysctl_rmem_default
= 32767;
1076 } else if (num_physpages
>= 131072) {
1077 sysctl_wmem_max
= 131071;
1078 sysctl_rmem_max
= 131071;
1083 * Simple resource managers for sockets.
1088 * Write buffer destructor automatically called from kfree_skb.
1090 void sock_wfree(struct sk_buff
*skb
)
1092 struct sock
*sk
= skb
->sk
;
1094 /* In case it might be waiting for more memory. */
1095 atomic_sub(skb
->truesize
, &sk
->sk_wmem_alloc
);
1096 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
))
1097 sk
->sk_write_space(sk
);
1102 * Read buffer destructor automatically called from kfree_skb.
1104 void sock_rfree(struct sk_buff
*skb
)
1106 struct sock
*sk
= skb
->sk
;
1108 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1112 int sock_i_uid(struct sock
*sk
)
1116 read_lock(&sk
->sk_callback_lock
);
1117 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: 0;
1118 read_unlock(&sk
->sk_callback_lock
);
1122 unsigned long sock_i_ino(struct sock
*sk
)
1126 read_lock(&sk
->sk_callback_lock
);
1127 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1128 read_unlock(&sk
->sk_callback_lock
);
1133 * Allocate a skb from the socket's send buffer.
1135 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1138 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1139 struct sk_buff
* skb
= alloc_skb(size
, priority
);
1141 skb_set_owner_w(skb
, sk
);
1149 * Allocate a skb from the socket's receive buffer.
1151 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1154 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1155 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1157 skb_set_owner_r(skb
, sk
);
1165 * Allocate a memory block from the socket's option memory buffer.
1167 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1169 if ((unsigned)size
<= sysctl_optmem_max
&&
1170 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1172 /* First do the add, to avoid the race if kmalloc
1175 atomic_add(size
, &sk
->sk_omem_alloc
);
1176 mem
= kmalloc(size
, priority
);
1179 atomic_sub(size
, &sk
->sk_omem_alloc
);
1185 * Free an option memory block.
1187 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1190 atomic_sub(size
, &sk
->sk_omem_alloc
);
1193 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1194 I think, these locks should be removed for datagram sockets.
1196 static long sock_wait_for_wmem(struct sock
* sk
, long timeo
)
1200 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1204 if (signal_pending(current
))
1206 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1207 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
1208 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1210 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1214 timeo
= schedule_timeout(timeo
);
1216 finish_wait(sk
->sk_sleep
, &wait
);
1222 * Generic send/receive buffer handlers
1225 static struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
,
1226 unsigned long header_len
,
1227 unsigned long data_len
,
1228 int noblock
, int *errcode
)
1230 struct sk_buff
*skb
;
1235 gfp_mask
= sk
->sk_allocation
;
1236 if (gfp_mask
& __GFP_WAIT
)
1237 gfp_mask
|= __GFP_REPEAT
;
1239 timeo
= sock_sndtimeo(sk
, noblock
);
1241 err
= sock_error(sk
);
1246 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1249 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1250 skb
= alloc_skb(header_len
, gfp_mask
);
1255 /* No pages, we're done... */
1259 npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1260 skb
->truesize
+= data_len
;
1261 skb_shinfo(skb
)->nr_frags
= npages
;
1262 for (i
= 0; i
< npages
; i
++) {
1266 page
= alloc_pages(sk
->sk_allocation
, 0);
1269 skb_shinfo(skb
)->nr_frags
= i
;
1274 frag
= &skb_shinfo(skb
)->frags
[i
];
1276 frag
->page_offset
= 0;
1277 frag
->size
= (data_len
>= PAGE_SIZE
?
1280 data_len
-= PAGE_SIZE
;
1283 /* Full success... */
1289 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1290 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1294 if (signal_pending(current
))
1296 timeo
= sock_wait_for_wmem(sk
, timeo
);
1299 skb_set_owner_w(skb
, sk
);
1303 err
= sock_intr_errno(timeo
);
1309 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1310 int noblock
, int *errcode
)
1312 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1315 static void __lock_sock(struct sock
*sk
)
1320 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1321 TASK_UNINTERRUPTIBLE
);
1322 spin_unlock_bh(&sk
->sk_lock
.slock
);
1324 spin_lock_bh(&sk
->sk_lock
.slock
);
1325 if (!sock_owned_by_user(sk
))
1328 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1331 static void __release_sock(struct sock
*sk
)
1333 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1336 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1340 struct sk_buff
*next
= skb
->next
;
1343 sk
->sk_backlog_rcv(sk
, skb
);
1346 * We are in process context here with softirqs
1347 * disabled, use cond_resched_softirq() to preempt.
1348 * This is safe to do because we've taken the backlog
1351 cond_resched_softirq();
1354 } while (skb
!= NULL
);
1357 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1361 * sk_wait_data - wait for data to arrive at sk_receive_queue
1362 * @sk: sock to wait on
1363 * @timeo: for how long
1365 * Now socket state including sk->sk_err is changed only under lock,
1366 * hence we may omit checks after joining wait queue.
1367 * We check receive queue before schedule() only as optimization;
1368 * it is very likely that release_sock() added new data.
1370 int sk_wait_data(struct sock
*sk
, long *timeo
)
1375 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
1376 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1377 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1378 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1379 finish_wait(sk
->sk_sleep
, &wait
);
1383 EXPORT_SYMBOL(sk_wait_data
);
1386 * Set of default routines for initialising struct proto_ops when
1387 * the protocol does not support a particular function. In certain
1388 * cases where it makes no sense for a protocol to have a "do nothing"
1389 * function, some default processing is provided.
1392 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1397 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1403 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1408 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1413 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1419 unsigned int sock_no_poll(struct file
* file
, struct socket
*sock
, poll_table
*pt
)
1424 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1429 int sock_no_listen(struct socket
*sock
, int backlog
)
1434 int sock_no_shutdown(struct socket
*sock
, int how
)
1439 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
1440 char __user
*optval
, int optlen
)
1445 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
1446 char __user
*optval
, int __user
*optlen
)
1451 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1457 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1458 size_t len
, int flags
)
1463 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
1465 /* Mirror missing mmap method error code */
1469 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
1472 struct msghdr msg
= {.msg_flags
= flags
};
1474 char *kaddr
= kmap(page
);
1475 iov
.iov_base
= kaddr
+ offset
;
1477 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1483 * Default Socket Callbacks
1486 static void sock_def_wakeup(struct sock
*sk
)
1488 read_lock(&sk
->sk_callback_lock
);
1489 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1490 wake_up_interruptible_all(sk
->sk_sleep
);
1491 read_unlock(&sk
->sk_callback_lock
);
1494 static void sock_def_error_report(struct sock
*sk
)
1496 read_lock(&sk
->sk_callback_lock
);
1497 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1498 wake_up_interruptible(sk
->sk_sleep
);
1499 sk_wake_async(sk
,0,POLL_ERR
);
1500 read_unlock(&sk
->sk_callback_lock
);
1503 static void sock_def_readable(struct sock
*sk
, int len
)
1505 read_lock(&sk
->sk_callback_lock
);
1506 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1507 wake_up_interruptible(sk
->sk_sleep
);
1508 sk_wake_async(sk
,1,POLL_IN
);
1509 read_unlock(&sk
->sk_callback_lock
);
1512 static void sock_def_write_space(struct sock
*sk
)
1514 read_lock(&sk
->sk_callback_lock
);
1516 /* Do not wake up a writer until he can make "significant"
1519 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
1520 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1521 wake_up_interruptible(sk
->sk_sleep
);
1523 /* Should agree with poll, otherwise some programs break */
1524 if (sock_writeable(sk
))
1525 sk_wake_async(sk
, 2, POLL_OUT
);
1528 read_unlock(&sk
->sk_callback_lock
);
1531 static void sock_def_destruct(struct sock
*sk
)
1533 kfree(sk
->sk_protinfo
);
1536 void sk_send_sigurg(struct sock
*sk
)
1538 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
1539 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
1540 sk_wake_async(sk
, 3, POLL_PRI
);
1543 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
1544 unsigned long expires
)
1546 if (!mod_timer(timer
, expires
))
1550 EXPORT_SYMBOL(sk_reset_timer
);
1552 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
1554 if (timer_pending(timer
) && del_timer(timer
))
1558 EXPORT_SYMBOL(sk_stop_timer
);
1560 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
1562 skb_queue_head_init(&sk
->sk_receive_queue
);
1563 skb_queue_head_init(&sk
->sk_write_queue
);
1564 skb_queue_head_init(&sk
->sk_error_queue
);
1565 #ifdef CONFIG_NET_DMA
1566 skb_queue_head_init(&sk
->sk_async_wait_queue
);
1569 sk
->sk_send_head
= NULL
;
1571 init_timer(&sk
->sk_timer
);
1573 sk
->sk_allocation
= GFP_KERNEL
;
1574 sk
->sk_rcvbuf
= sysctl_rmem_default
;
1575 sk
->sk_sndbuf
= sysctl_wmem_default
;
1576 sk
->sk_state
= TCP_CLOSE
;
1577 sk
->sk_socket
= sock
;
1579 sock_set_flag(sk
, SOCK_ZAPPED
);
1582 sk
->sk_type
= sock
->type
;
1583 sk
->sk_sleep
= &sock
->wait
;
1586 sk
->sk_sleep
= NULL
;
1588 rwlock_init(&sk
->sk_dst_lock
);
1589 rwlock_init(&sk
->sk_callback_lock
);
1590 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
1591 af_callback_keys
+ sk
->sk_family
,
1592 af_family_clock_key_strings
[sk
->sk_family
]);
1594 sk
->sk_state_change
= sock_def_wakeup
;
1595 sk
->sk_data_ready
= sock_def_readable
;
1596 sk
->sk_write_space
= sock_def_write_space
;
1597 sk
->sk_error_report
= sock_def_error_report
;
1598 sk
->sk_destruct
= sock_def_destruct
;
1600 sk
->sk_sndmsg_page
= NULL
;
1601 sk
->sk_sndmsg_off
= 0;
1603 sk
->sk_peercred
.pid
= 0;
1604 sk
->sk_peercred
.uid
= -1;
1605 sk
->sk_peercred
.gid
= -1;
1606 sk
->sk_write_pending
= 0;
1607 sk
->sk_rcvlowat
= 1;
1608 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
1609 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
1611 sk
->sk_stamp
= ktime_set(-1L, -1L);
1613 atomic_set(&sk
->sk_refcnt
, 1);
1616 void fastcall
lock_sock_nested(struct sock
*sk
, int subclass
)
1619 spin_lock_bh(&sk
->sk_lock
.slock
);
1620 if (sk
->sk_lock
.owned
)
1622 sk
->sk_lock
.owned
= 1;
1623 spin_unlock(&sk
->sk_lock
.slock
);
1625 * The sk_lock has mutex_lock() semantics here:
1627 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
1631 EXPORT_SYMBOL(lock_sock_nested
);
1633 void fastcall
release_sock(struct sock
*sk
)
1636 * The sk_lock has mutex_unlock() semantics:
1638 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
1640 spin_lock_bh(&sk
->sk_lock
.slock
);
1641 if (sk
->sk_backlog
.tail
)
1643 sk
->sk_lock
.owned
= 0;
1644 if (waitqueue_active(&sk
->sk_lock
.wq
))
1645 wake_up(&sk
->sk_lock
.wq
);
1646 spin_unlock_bh(&sk
->sk_lock
.slock
);
1648 EXPORT_SYMBOL(release_sock
);
1650 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
1653 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
1654 sock_enable_timestamp(sk
);
1655 tv
= ktime_to_timeval(sk
->sk_stamp
);
1656 if (tv
.tv_sec
== -1)
1658 if (tv
.tv_sec
== 0) {
1659 sk
->sk_stamp
= ktime_get_real();
1660 tv
= ktime_to_timeval(sk
->sk_stamp
);
1662 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
1664 EXPORT_SYMBOL(sock_get_timestamp
);
1666 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
1669 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
1670 sock_enable_timestamp(sk
);
1671 ts
= ktime_to_timespec(sk
->sk_stamp
);
1672 if (ts
.tv_sec
== -1)
1674 if (ts
.tv_sec
== 0) {
1675 sk
->sk_stamp
= ktime_get_real();
1676 ts
= ktime_to_timespec(sk
->sk_stamp
);
1678 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
1680 EXPORT_SYMBOL(sock_get_timestampns
);
1682 void sock_enable_timestamp(struct sock
*sk
)
1684 if (!sock_flag(sk
, SOCK_TIMESTAMP
)) {
1685 sock_set_flag(sk
, SOCK_TIMESTAMP
);
1686 net_enable_timestamp();
1691 * Get a socket option on an socket.
1693 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1694 * asynchronous errors should be reported by getsockopt. We assume
1695 * this means if you specify SO_ERROR (otherwise whats the point of it).
1697 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1698 char __user
*optval
, int __user
*optlen
)
1700 struct sock
*sk
= sock
->sk
;
1702 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
1705 EXPORT_SYMBOL(sock_common_getsockopt
);
1707 #ifdef CONFIG_COMPAT
1708 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1709 char __user
*optval
, int __user
*optlen
)
1711 struct sock
*sk
= sock
->sk
;
1713 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
1714 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
1716 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
1718 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
1721 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1722 struct msghdr
*msg
, size_t size
, int flags
)
1724 struct sock
*sk
= sock
->sk
;
1728 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
1729 flags
& ~MSG_DONTWAIT
, &addr_len
);
1731 msg
->msg_namelen
= addr_len
;
1735 EXPORT_SYMBOL(sock_common_recvmsg
);
1738 * Set socket options on an inet socket.
1740 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1741 char __user
*optval
, int optlen
)
1743 struct sock
*sk
= sock
->sk
;
1745 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
1748 EXPORT_SYMBOL(sock_common_setsockopt
);
1750 #ifdef CONFIG_COMPAT
1751 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1752 char __user
*optval
, int optlen
)
1754 struct sock
*sk
= sock
->sk
;
1756 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
1757 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
1759 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
1761 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
1764 void sk_common_release(struct sock
*sk
)
1766 if (sk
->sk_prot
->destroy
)
1767 sk
->sk_prot
->destroy(sk
);
1770 * Observation: when sock_common_release is called, processes have
1771 * no access to socket. But net still has.
1772 * Step one, detach it from networking:
1774 * A. Remove from hash tables.
1777 sk
->sk_prot
->unhash(sk
);
1780 * In this point socket cannot receive new packets, but it is possible
1781 * that some packets are in flight because some CPU runs receiver and
1782 * did hash table lookup before we unhashed socket. They will achieve
1783 * receive queue and will be purged by socket destructor.
1785 * Also we still have packets pending on receive queue and probably,
1786 * our own packets waiting in device queues. sock_destroy will drain
1787 * receive queue, but transmitted packets will delay socket destruction
1788 * until the last reference will be released.
1793 xfrm_sk_free_policy(sk
);
1795 sk_refcnt_debug_release(sk
);
1799 EXPORT_SYMBOL(sk_common_release
);
1801 static DEFINE_RWLOCK(proto_list_lock
);
1802 static LIST_HEAD(proto_list
);
1806 * Define default functions to keep track of inuse sockets per protocol
1807 * Note that often used protocols use dedicated functions to get a speed increase.
1808 * (see DEFINE_PROTO_INUSE/REF_PROTO_INUSE)
1810 static void inuse_add(struct proto
*prot
, int inc
)
1812 per_cpu_ptr(prot
->inuse_ptr
, smp_processor_id())[0] += inc
;
1815 static int inuse_get(const struct proto
*prot
)
1818 for_each_possible_cpu(cpu
)
1819 res
+= per_cpu_ptr(prot
->inuse_ptr
, cpu
)[0];
1824 int proto_register(struct proto
*prot
, int alloc_slab
)
1826 char *request_sock_slab_name
= NULL
;
1827 char *timewait_sock_slab_name
;
1831 if (!prot
->inuse_getval
|| !prot
->inuse_add
) {
1832 prot
->inuse_ptr
= alloc_percpu(int);
1833 if (prot
->inuse_ptr
== NULL
)
1835 prot
->inuse_getval
= inuse_get
;
1836 prot
->inuse_add
= inuse_add
;
1840 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
1841 SLAB_HWCACHE_ALIGN
, NULL
);
1843 if (prot
->slab
== NULL
) {
1844 printk(KERN_CRIT
"%s: Can't create sock SLAB cache!\n",
1846 goto out_free_inuse
;
1849 if (prot
->rsk_prot
!= NULL
) {
1850 static const char mask
[] = "request_sock_%s";
1852 request_sock_slab_name
= kmalloc(strlen(prot
->name
) + sizeof(mask
) - 1, GFP_KERNEL
);
1853 if (request_sock_slab_name
== NULL
)
1854 goto out_free_sock_slab
;
1856 sprintf(request_sock_slab_name
, mask
, prot
->name
);
1857 prot
->rsk_prot
->slab
= kmem_cache_create(request_sock_slab_name
,
1858 prot
->rsk_prot
->obj_size
, 0,
1859 SLAB_HWCACHE_ALIGN
, NULL
);
1861 if (prot
->rsk_prot
->slab
== NULL
) {
1862 printk(KERN_CRIT
"%s: Can't create request sock SLAB cache!\n",
1864 goto out_free_request_sock_slab_name
;
1868 if (prot
->twsk_prot
!= NULL
) {
1869 static const char mask
[] = "tw_sock_%s";
1871 timewait_sock_slab_name
= kmalloc(strlen(prot
->name
) + sizeof(mask
) - 1, GFP_KERNEL
);
1873 if (timewait_sock_slab_name
== NULL
)
1874 goto out_free_request_sock_slab
;
1876 sprintf(timewait_sock_slab_name
, mask
, prot
->name
);
1877 prot
->twsk_prot
->twsk_slab
=
1878 kmem_cache_create(timewait_sock_slab_name
,
1879 prot
->twsk_prot
->twsk_obj_size
,
1880 0, SLAB_HWCACHE_ALIGN
,
1882 if (prot
->twsk_prot
->twsk_slab
== NULL
)
1883 goto out_free_timewait_sock_slab_name
;
1887 write_lock(&proto_list_lock
);
1888 list_add(&prot
->node
, &proto_list
);
1889 write_unlock(&proto_list_lock
);
1893 out_free_timewait_sock_slab_name
:
1894 kfree(timewait_sock_slab_name
);
1895 out_free_request_sock_slab
:
1896 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
1897 kmem_cache_destroy(prot
->rsk_prot
->slab
);
1898 prot
->rsk_prot
->slab
= NULL
;
1900 out_free_request_sock_slab_name
:
1901 kfree(request_sock_slab_name
);
1903 kmem_cache_destroy(prot
->slab
);
1907 if (prot
->inuse_ptr
!= NULL
) {
1908 free_percpu(prot
->inuse_ptr
);
1909 prot
->inuse_ptr
= NULL
;
1910 prot
->inuse_getval
= NULL
;
1911 prot
->inuse_add
= NULL
;
1917 EXPORT_SYMBOL(proto_register
);
1919 void proto_unregister(struct proto
*prot
)
1921 write_lock(&proto_list_lock
);
1922 list_del(&prot
->node
);
1923 write_unlock(&proto_list_lock
);
1926 if (prot
->inuse_ptr
!= NULL
) {
1927 free_percpu(prot
->inuse_ptr
);
1928 prot
->inuse_ptr
= NULL
;
1929 prot
->inuse_getval
= NULL
;
1930 prot
->inuse_add
= NULL
;
1933 if (prot
->slab
!= NULL
) {
1934 kmem_cache_destroy(prot
->slab
);
1938 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
1939 const char *name
= kmem_cache_name(prot
->rsk_prot
->slab
);
1941 kmem_cache_destroy(prot
->rsk_prot
->slab
);
1943 prot
->rsk_prot
->slab
= NULL
;
1946 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
1947 const char *name
= kmem_cache_name(prot
->twsk_prot
->twsk_slab
);
1949 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
1951 prot
->twsk_prot
->twsk_slab
= NULL
;
1955 EXPORT_SYMBOL(proto_unregister
);
1957 #ifdef CONFIG_PROC_FS
1958 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1960 read_lock(&proto_list_lock
);
1961 return seq_list_start_head(&proto_list
, *pos
);
1964 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1966 return seq_list_next(v
, &proto_list
, pos
);
1969 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
1971 read_unlock(&proto_list_lock
);
1974 static char proto_method_implemented(const void *method
)
1976 return method
== NULL
? 'n' : 'y';
1979 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
1981 seq_printf(seq
, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1982 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1985 proto
->sockets_allocated
!= NULL
? atomic_read(proto
->sockets_allocated
) : -1,
1986 proto
->memory_allocated
!= NULL
? atomic_read(proto
->memory_allocated
) : -1,
1987 proto
->memory_pressure
!= NULL
? *proto
->memory_pressure
? "yes" : "no" : "NI",
1989 proto
->slab
== NULL
? "no" : "yes",
1990 module_name(proto
->owner
),
1991 proto_method_implemented(proto
->close
),
1992 proto_method_implemented(proto
->connect
),
1993 proto_method_implemented(proto
->disconnect
),
1994 proto_method_implemented(proto
->accept
),
1995 proto_method_implemented(proto
->ioctl
),
1996 proto_method_implemented(proto
->init
),
1997 proto_method_implemented(proto
->destroy
),
1998 proto_method_implemented(proto
->shutdown
),
1999 proto_method_implemented(proto
->setsockopt
),
2000 proto_method_implemented(proto
->getsockopt
),
2001 proto_method_implemented(proto
->sendmsg
),
2002 proto_method_implemented(proto
->recvmsg
),
2003 proto_method_implemented(proto
->sendpage
),
2004 proto_method_implemented(proto
->bind
),
2005 proto_method_implemented(proto
->backlog_rcv
),
2006 proto_method_implemented(proto
->hash
),
2007 proto_method_implemented(proto
->unhash
),
2008 proto_method_implemented(proto
->get_port
),
2009 proto_method_implemented(proto
->enter_memory_pressure
));
2012 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2014 if (v
== &proto_list
)
2015 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2024 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2026 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2030 static const struct seq_operations proto_seq_ops
= {
2031 .start
= proto_seq_start
,
2032 .next
= proto_seq_next
,
2033 .stop
= proto_seq_stop
,
2034 .show
= proto_seq_show
,
2037 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2039 return seq_open(file
, &proto_seq_ops
);
2042 static const struct file_operations proto_seq_fops
= {
2043 .owner
= THIS_MODULE
,
2044 .open
= proto_seq_open
,
2046 .llseek
= seq_lseek
,
2047 .release
= seq_release
,
2050 static int __init
proto_init(void)
2052 /* register /proc/net/protocols */
2053 return proc_net_fops_create(&init_net
, "protocols", S_IRUGO
, &proto_seq_fops
) == NULL
? -ENOBUFS
: 0;
2056 subsys_initcall(proto_init
);
2058 #endif /* PROC_FS */
2060 EXPORT_SYMBOL(sk_alloc
);
2061 EXPORT_SYMBOL(sk_free
);
2062 EXPORT_SYMBOL(sk_send_sigurg
);
2063 EXPORT_SYMBOL(sock_alloc_send_skb
);
2064 EXPORT_SYMBOL(sock_init_data
);
2065 EXPORT_SYMBOL(sock_kfree_s
);
2066 EXPORT_SYMBOL(sock_kmalloc
);
2067 EXPORT_SYMBOL(sock_no_accept
);
2068 EXPORT_SYMBOL(sock_no_bind
);
2069 EXPORT_SYMBOL(sock_no_connect
);
2070 EXPORT_SYMBOL(sock_no_getname
);
2071 EXPORT_SYMBOL(sock_no_getsockopt
);
2072 EXPORT_SYMBOL(sock_no_ioctl
);
2073 EXPORT_SYMBOL(sock_no_listen
);
2074 EXPORT_SYMBOL(sock_no_mmap
);
2075 EXPORT_SYMBOL(sock_no_poll
);
2076 EXPORT_SYMBOL(sock_no_recvmsg
);
2077 EXPORT_SYMBOL(sock_no_sendmsg
);
2078 EXPORT_SYMBOL(sock_no_sendpage
);
2079 EXPORT_SYMBOL(sock_no_setsockopt
);
2080 EXPORT_SYMBOL(sock_no_shutdown
);
2081 EXPORT_SYMBOL(sock_no_socketpair
);
2082 EXPORT_SYMBOL(sock_rfree
);
2083 EXPORT_SYMBOL(sock_setsockopt
);
2084 EXPORT_SYMBOL(sock_wfree
);
2085 EXPORT_SYMBOL(sock_wmalloc
);
2086 EXPORT_SYMBOL(sock_i_uid
);
2087 EXPORT_SYMBOL(sock_i_ino
);
2088 EXPORT_SYMBOL(sysctl_optmem_max
);
2089 #ifdef CONFIG_SYSCTL
2090 EXPORT_SYMBOL(sysctl_rmem_max
);
2091 EXPORT_SYMBOL(sysctl_wmem_max
);