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/request_sock.h>
123 #include <net/sock.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
127 #include <linux/filter.h>
134 * Each address family might have different locking rules, so we have
135 * one slock key per address family:
137 static struct lock_class_key af_family_keys
[AF_MAX
];
138 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
140 #ifdef CONFIG_DEBUG_LOCK_ALLOC
142 * Make lock validator output more readable. (we pre-construct these
143 * strings build-time, so that runtime initialization of socket
146 static const char *af_family_key_strings
[AF_MAX
+1] = {
147 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
148 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
149 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
150 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
151 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
152 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
153 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
154 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
155 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
156 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
159 static const char *af_family_slock_key_strings
[AF_MAX
+1] = {
160 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
161 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
162 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
163 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
164 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
165 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
166 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
167 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
168 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
169 "slock-27" , "slock-28" , "slock-29" ,
170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX"
175 * sk_callback_lock locking rules are per-address-family,
176 * so split the lock classes by using a per-AF key:
178 static struct lock_class_key af_callback_keys
[AF_MAX
];
180 /* Take into consideration the size of the struct sk_buff overhead in the
181 * determination of these values, since that is non-constant across
182 * platforms. This makes socket queueing behavior and performance
183 * not depend upon such differences.
185 #define _SK_MEM_PACKETS 256
186 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
187 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
188 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
190 /* Run time adjustable parameters. */
191 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
192 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
193 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
194 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
196 /* Maximal space eaten by iovec or ancilliary data plus some space */
197 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
199 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
203 if (optlen
< sizeof(tv
))
205 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
208 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
209 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
211 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
212 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
216 static void sock_warn_obsolete_bsdism(const char *name
)
219 static char warncomm
[TASK_COMM_LEN
];
220 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
221 strcpy(warncomm
, current
->comm
);
222 printk(KERN_WARNING
"process `%s' is using obsolete "
223 "%s SO_BSDCOMPAT\n", warncomm
, name
);
228 static void sock_disable_timestamp(struct sock
*sk
)
230 if (sock_flag(sk
, SOCK_TIMESTAMP
)) {
231 sock_reset_flag(sk
, SOCK_TIMESTAMP
);
232 net_disable_timestamp();
237 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
242 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
243 number of warnings when compiling with -W --ANK
245 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
246 (unsigned)sk
->sk_rcvbuf
) {
251 err
= sk_filter(sk
, skb
);
256 skb_set_owner_r(skb
, sk
);
258 /* Cache the SKB length before we tack it onto the receive
259 * queue. Once it is added it no longer belongs to us and
260 * may be freed by other threads of control pulling packets
265 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
267 if (!sock_flag(sk
, SOCK_DEAD
))
268 sk
->sk_data_ready(sk
, skb_len
);
272 EXPORT_SYMBOL(sock_queue_rcv_skb
);
274 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
276 int rc
= NET_RX_SUCCESS
;
278 if (sk_filter(sk
, skb
))
279 goto discard_and_relse
;
284 bh_lock_sock_nested(sk
);
287 if (!sock_owned_by_user(sk
)) {
289 * trylock + unlock semantics:
291 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
293 rc
= sk
->sk_backlog_rcv(sk
, skb
);
295 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
297 sk_add_backlog(sk
, skb
);
306 EXPORT_SYMBOL(sk_receive_skb
);
308 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
310 struct dst_entry
*dst
= sk
->sk_dst_cache
;
312 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
313 sk
->sk_dst_cache
= NULL
;
320 EXPORT_SYMBOL(__sk_dst_check
);
322 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
324 struct dst_entry
*dst
= sk_dst_get(sk
);
326 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
334 EXPORT_SYMBOL(sk_dst_check
);
337 * This is meant for all protocols to use and covers goings on
338 * at the socket level. Everything here is generic.
341 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
342 char __user
*optval
, int optlen
)
344 struct sock
*sk
=sock
->sk
;
345 struct sk_filter
*filter
;
352 * Options without arguments
355 #ifdef SO_DONTLINGER /* Compatibility item... */
356 if (optname
== SO_DONTLINGER
) {
358 sock_reset_flag(sk
, SOCK_LINGER
);
364 if (optlen
< sizeof(int))
367 if (get_user(val
, (int __user
*)optval
))
376 if (val
&& !capable(CAP_NET_ADMIN
)) {
380 sock_set_flag(sk
, SOCK_DBG
);
382 sock_reset_flag(sk
, SOCK_DBG
);
385 sk
->sk_reuse
= valbool
;
393 sock_set_flag(sk
, SOCK_LOCALROUTE
);
395 sock_reset_flag(sk
, SOCK_LOCALROUTE
);
398 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
401 /* Don't error on this BSD doesn't and if you think
402 about it this is right. Otherwise apps have to
403 play 'guess the biggest size' games. RCVBUF/SNDBUF
404 are treated in BSD as hints */
406 if (val
> sysctl_wmem_max
)
407 val
= sysctl_wmem_max
;
409 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
410 if ((val
* 2) < SOCK_MIN_SNDBUF
)
411 sk
->sk_sndbuf
= SOCK_MIN_SNDBUF
;
413 sk
->sk_sndbuf
= val
* 2;
416 * Wake up sending tasks if we
419 sk
->sk_write_space(sk
);
423 if (!capable(CAP_NET_ADMIN
)) {
430 /* Don't error on this BSD doesn't and if you think
431 about it this is right. Otherwise apps have to
432 play 'guess the biggest size' games. RCVBUF/SNDBUF
433 are treated in BSD as hints */
435 if (val
> sysctl_rmem_max
)
436 val
= sysctl_rmem_max
;
438 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
440 * We double it on the way in to account for
441 * "struct sk_buff" etc. overhead. Applications
442 * assume that the SO_RCVBUF setting they make will
443 * allow that much actual data to be received on that
446 * Applications are unaware that "struct sk_buff" and
447 * other overheads allocate from the receive buffer
448 * during socket buffer allocation.
450 * And after considering the possible alternatives,
451 * returning the value we actually used in getsockopt
452 * is the most desirable behavior.
454 if ((val
* 2) < SOCK_MIN_RCVBUF
)
455 sk
->sk_rcvbuf
= SOCK_MIN_RCVBUF
;
457 sk
->sk_rcvbuf
= val
* 2;
461 if (!capable(CAP_NET_ADMIN
)) {
469 if (sk
->sk_protocol
== IPPROTO_TCP
)
470 tcp_set_keepalive(sk
, valbool
);
472 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
476 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
480 sk
->sk_no_check
= valbool
;
484 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
485 sk
->sk_priority
= val
;
491 if (optlen
< sizeof(ling
)) {
492 ret
= -EINVAL
; /* 1003.1g */
495 if (copy_from_user(&ling
,optval
,sizeof(ling
))) {
500 sock_reset_flag(sk
, SOCK_LINGER
);
502 #if (BITS_PER_LONG == 32)
503 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
504 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
507 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
508 sock_set_flag(sk
, SOCK_LINGER
);
513 sock_warn_obsolete_bsdism("setsockopt");
518 set_bit(SOCK_PASSCRED
, &sock
->flags
);
520 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
526 if (optname
== SO_TIMESTAMP
)
527 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
529 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
530 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
531 sock_enable_timestamp(sk
);
533 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
534 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
541 sk
->sk_rcvlowat
= val
? : 1;
545 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
549 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
552 #ifdef CONFIG_NETDEVICES
553 case SO_BINDTODEVICE
:
555 char devname
[IFNAMSIZ
];
558 if (!capable(CAP_NET_RAW
)) {
563 /* Bind this socket to a particular device like "eth0",
564 * as specified in the passed interface name. If the
565 * name is "" or the option length is zero the socket
570 sk
->sk_bound_dev_if
= 0;
572 if (optlen
> IFNAMSIZ
- 1)
573 optlen
= IFNAMSIZ
- 1;
574 memset(devname
, 0, sizeof(devname
));
575 if (copy_from_user(devname
, optval
, optlen
)) {
580 /* Remove any cached route for this socket. */
583 if (devname
[0] == '\0') {
584 sk
->sk_bound_dev_if
= 0;
586 struct net_device
*dev
= dev_get_by_name(devname
);
591 sk
->sk_bound_dev_if
= dev
->ifindex
;
600 case SO_ATTACH_FILTER
:
602 if (optlen
== sizeof(struct sock_fprog
)) {
603 struct sock_fprog fprog
;
606 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
609 ret
= sk_attach_filter(&fprog
, sk
);
613 case SO_DETACH_FILTER
:
615 filter
= rcu_dereference(sk
->sk_filter
);
617 rcu_assign_pointer(sk
->sk_filter
, NULL
);
618 sk_filter_release(sk
, filter
);
619 rcu_read_unlock_bh();
622 rcu_read_unlock_bh();
628 set_bit(SOCK_PASSSEC
, &sock
->flags
);
630 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
633 /* We implement the SO_SNDLOWAT etc to
634 not be settable (1003.1g 5.3) */
644 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
645 char __user
*optval
, int __user
*optlen
)
647 struct sock
*sk
= sock
->sk
;
655 unsigned int lv
= sizeof(int);
658 if (get_user(len
, optlen
))
665 v
.val
= sock_flag(sk
, SOCK_DBG
);
669 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
673 v
.val
= !!sock_flag(sk
, SOCK_BROADCAST
);
677 v
.val
= sk
->sk_sndbuf
;
681 v
.val
= sk
->sk_rcvbuf
;
685 v
.val
= sk
->sk_reuse
;
689 v
.val
= !!sock_flag(sk
, SOCK_KEEPOPEN
);
697 v
.val
= -sock_error(sk
);
699 v
.val
= xchg(&sk
->sk_err_soft
, 0);
703 v
.val
= !!sock_flag(sk
, SOCK_URGINLINE
);
707 v
.val
= sk
->sk_no_check
;
711 v
.val
= sk
->sk_priority
;
716 v
.ling
.l_onoff
= !!sock_flag(sk
, SOCK_LINGER
);
717 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
721 sock_warn_obsolete_bsdism("getsockopt");
725 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
726 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
730 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
734 lv
=sizeof(struct timeval
);
735 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
739 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
740 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
745 lv
=sizeof(struct timeval
);
746 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
750 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
751 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
756 v
.val
= sk
->sk_rcvlowat
;
764 v
.val
= test_bit(SOCK_PASSCRED
, &sock
->flags
) ? 1 : 0;
768 if (len
> sizeof(sk
->sk_peercred
))
769 len
= sizeof(sk
->sk_peercred
);
770 if (copy_to_user(optval
, &sk
->sk_peercred
, len
))
778 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
782 if (copy_to_user(optval
, address
, len
))
787 /* Dubious BSD thing... Probably nobody even uses it, but
788 * the UNIX standard wants it for whatever reason... -DaveM
791 v
.val
= sk
->sk_state
== TCP_LISTEN
;
795 v
.val
= test_bit(SOCK_PASSSEC
, &sock
->flags
) ? 1 : 0;
799 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
807 if (copy_to_user(optval
, &v
, len
))
810 if (put_user(len
, optlen
))
816 * Initialize an sk_lock.
818 * (We also register the sk_lock with the lock validator.)
820 static inline void sock_lock_init(struct sock
*sk
)
822 sock_lock_init_class_and_name(sk
,
823 af_family_slock_key_strings
[sk
->sk_family
],
824 af_family_slock_keys
+ sk
->sk_family
,
825 af_family_key_strings
[sk
->sk_family
],
826 af_family_keys
+ sk
->sk_family
);
830 * sk_alloc - All socket objects are allocated here
831 * @family: protocol family
832 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
833 * @prot: struct proto associated with this new sock instance
834 * @zero_it: if we should zero the newly allocated sock
836 struct sock
*sk_alloc(int family
, gfp_t priority
,
837 struct proto
*prot
, int zero_it
)
839 struct sock
*sk
= NULL
;
840 struct kmem_cache
*slab
= prot
->slab
;
843 sk
= kmem_cache_alloc(slab
, priority
);
845 sk
= kmalloc(prot
->obj_size
, priority
);
849 memset(sk
, 0, prot
->obj_size
);
850 sk
->sk_family
= family
;
852 * See comment in struct sock definition to understand
853 * why we need sk_prot_creator -acme
855 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
859 if (security_sk_alloc(sk
, family
, priority
))
862 if (!try_module_get(prot
->owner
))
869 kmem_cache_free(slab
, sk
);
875 void sk_free(struct sock
*sk
)
877 struct sk_filter
*filter
;
878 struct module
*owner
= sk
->sk_prot_creator
->owner
;
883 filter
= rcu_dereference(sk
->sk_filter
);
885 sk_filter_release(sk
, filter
);
886 rcu_assign_pointer(sk
->sk_filter
, NULL
);
889 sock_disable_timestamp(sk
);
891 if (atomic_read(&sk
->sk_omem_alloc
))
892 printk(KERN_DEBUG
"%s: optmem leakage (%d bytes) detected.\n",
893 __FUNCTION__
, atomic_read(&sk
->sk_omem_alloc
));
895 security_sk_free(sk
);
896 if (sk
->sk_prot_creator
->slab
!= NULL
)
897 kmem_cache_free(sk
->sk_prot_creator
->slab
, sk
);
903 struct sock
*sk_clone(const struct sock
*sk
, const gfp_t priority
)
905 struct sock
*newsk
= sk_alloc(sk
->sk_family
, priority
, sk
->sk_prot
, 0);
908 struct sk_filter
*filter
;
910 sock_copy(newsk
, sk
);
913 sk_node_init(&newsk
->sk_node
);
914 sock_lock_init(newsk
);
916 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
918 atomic_set(&newsk
->sk_rmem_alloc
, 0);
919 atomic_set(&newsk
->sk_wmem_alloc
, 0);
920 atomic_set(&newsk
->sk_omem_alloc
, 0);
921 skb_queue_head_init(&newsk
->sk_receive_queue
);
922 skb_queue_head_init(&newsk
->sk_write_queue
);
923 #ifdef CONFIG_NET_DMA
924 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
927 rwlock_init(&newsk
->sk_dst_lock
);
928 rwlock_init(&newsk
->sk_callback_lock
);
929 lockdep_set_class(&newsk
->sk_callback_lock
,
930 af_callback_keys
+ newsk
->sk_family
);
932 newsk
->sk_dst_cache
= NULL
;
933 newsk
->sk_wmem_queued
= 0;
934 newsk
->sk_forward_alloc
= 0;
935 newsk
->sk_send_head
= NULL
;
936 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
938 sock_reset_flag(newsk
, SOCK_DONE
);
939 skb_queue_head_init(&newsk
->sk_error_queue
);
941 filter
= newsk
->sk_filter
;
943 sk_filter_charge(newsk
, filter
);
945 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
946 /* It is still raw copy of parent, so invalidate
947 * destructor and make plain sk_free() */
948 newsk
->sk_destruct
= NULL
;
955 newsk
->sk_priority
= 0;
956 atomic_set(&newsk
->sk_refcnt
, 2);
959 * Increment the counter in the same struct proto as the master
960 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
961 * is the same as sk->sk_prot->socks, as this field was copied
964 * This _changes_ the previous behaviour, where
965 * tcp_create_openreq_child always was incrementing the
966 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
967 * to be taken into account in all callers. -acme
969 sk_refcnt_debug_inc(newsk
);
970 newsk
->sk_socket
= NULL
;
971 newsk
->sk_sleep
= NULL
;
973 if (newsk
->sk_prot
->sockets_allocated
)
974 atomic_inc(newsk
->sk_prot
->sockets_allocated
);
980 EXPORT_SYMBOL_GPL(sk_clone
);
982 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
984 __sk_dst_set(sk
, dst
);
985 sk
->sk_route_caps
= dst
->dev
->features
;
986 if (sk
->sk_route_caps
& NETIF_F_GSO
)
987 sk
->sk_route_caps
|= NETIF_F_GSO_MASK
;
988 if (sk_can_gso(sk
)) {
990 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
992 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
995 EXPORT_SYMBOL_GPL(sk_setup_caps
);
997 void __init
sk_init(void)
999 if (num_physpages
<= 4096) {
1000 sysctl_wmem_max
= 32767;
1001 sysctl_rmem_max
= 32767;
1002 sysctl_wmem_default
= 32767;
1003 sysctl_rmem_default
= 32767;
1004 } else if (num_physpages
>= 131072) {
1005 sysctl_wmem_max
= 131071;
1006 sysctl_rmem_max
= 131071;
1011 * Simple resource managers for sockets.
1016 * Write buffer destructor automatically called from kfree_skb.
1018 void sock_wfree(struct sk_buff
*skb
)
1020 struct sock
*sk
= skb
->sk
;
1022 /* In case it might be waiting for more memory. */
1023 atomic_sub(skb
->truesize
, &sk
->sk_wmem_alloc
);
1024 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
))
1025 sk
->sk_write_space(sk
);
1030 * Read buffer destructor automatically called from kfree_skb.
1032 void sock_rfree(struct sk_buff
*skb
)
1034 struct sock
*sk
= skb
->sk
;
1036 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1040 int sock_i_uid(struct sock
*sk
)
1044 read_lock(&sk
->sk_callback_lock
);
1045 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: 0;
1046 read_unlock(&sk
->sk_callback_lock
);
1050 unsigned long sock_i_ino(struct sock
*sk
)
1054 read_lock(&sk
->sk_callback_lock
);
1055 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1056 read_unlock(&sk
->sk_callback_lock
);
1061 * Allocate a skb from the socket's send buffer.
1063 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1066 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1067 struct sk_buff
* skb
= alloc_skb(size
, priority
);
1069 skb_set_owner_w(skb
, sk
);
1077 * Allocate a skb from the socket's receive buffer.
1079 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1082 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1083 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1085 skb_set_owner_r(skb
, sk
);
1093 * Allocate a memory block from the socket's option memory buffer.
1095 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1097 if ((unsigned)size
<= sysctl_optmem_max
&&
1098 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1100 /* First do the add, to avoid the race if kmalloc
1103 atomic_add(size
, &sk
->sk_omem_alloc
);
1104 mem
= kmalloc(size
, priority
);
1107 atomic_sub(size
, &sk
->sk_omem_alloc
);
1113 * Free an option memory block.
1115 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1118 atomic_sub(size
, &sk
->sk_omem_alloc
);
1121 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1122 I think, these locks should be removed for datagram sockets.
1124 static long sock_wait_for_wmem(struct sock
* sk
, long timeo
)
1128 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1132 if (signal_pending(current
))
1134 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1135 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
1136 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1138 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1142 timeo
= schedule_timeout(timeo
);
1144 finish_wait(sk
->sk_sleep
, &wait
);
1150 * Generic send/receive buffer handlers
1153 static struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
,
1154 unsigned long header_len
,
1155 unsigned long data_len
,
1156 int noblock
, int *errcode
)
1158 struct sk_buff
*skb
;
1163 gfp_mask
= sk
->sk_allocation
;
1164 if (gfp_mask
& __GFP_WAIT
)
1165 gfp_mask
|= __GFP_REPEAT
;
1167 timeo
= sock_sndtimeo(sk
, noblock
);
1169 err
= sock_error(sk
);
1174 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1177 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1178 skb
= alloc_skb(header_len
, gfp_mask
);
1183 /* No pages, we're done... */
1187 npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1188 skb
->truesize
+= data_len
;
1189 skb_shinfo(skb
)->nr_frags
= npages
;
1190 for (i
= 0; i
< npages
; i
++) {
1194 page
= alloc_pages(sk
->sk_allocation
, 0);
1197 skb_shinfo(skb
)->nr_frags
= i
;
1202 frag
= &skb_shinfo(skb
)->frags
[i
];
1204 frag
->page_offset
= 0;
1205 frag
->size
= (data_len
>= PAGE_SIZE
?
1208 data_len
-= PAGE_SIZE
;
1211 /* Full success... */
1217 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1218 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1222 if (signal_pending(current
))
1224 timeo
= sock_wait_for_wmem(sk
, timeo
);
1227 skb_set_owner_w(skb
, sk
);
1231 err
= sock_intr_errno(timeo
);
1237 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1238 int noblock
, int *errcode
)
1240 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1243 static void __lock_sock(struct sock
*sk
)
1248 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1249 TASK_UNINTERRUPTIBLE
);
1250 spin_unlock_bh(&sk
->sk_lock
.slock
);
1252 spin_lock_bh(&sk
->sk_lock
.slock
);
1253 if (!sock_owned_by_user(sk
))
1256 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1259 static void __release_sock(struct sock
*sk
)
1261 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1264 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1268 struct sk_buff
*next
= skb
->next
;
1271 sk
->sk_backlog_rcv(sk
, skb
);
1274 * We are in process context here with softirqs
1275 * disabled, use cond_resched_softirq() to preempt.
1276 * This is safe to do because we've taken the backlog
1279 cond_resched_softirq();
1282 } while (skb
!= NULL
);
1285 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1289 * sk_wait_data - wait for data to arrive at sk_receive_queue
1290 * @sk: sock to wait on
1291 * @timeo: for how long
1293 * Now socket state including sk->sk_err is changed only under lock,
1294 * hence we may omit checks after joining wait queue.
1295 * We check receive queue before schedule() only as optimization;
1296 * it is very likely that release_sock() added new data.
1298 int sk_wait_data(struct sock
*sk
, long *timeo
)
1303 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
1304 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1305 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1306 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1307 finish_wait(sk
->sk_sleep
, &wait
);
1311 EXPORT_SYMBOL(sk_wait_data
);
1314 * Set of default routines for initialising struct proto_ops when
1315 * the protocol does not support a particular function. In certain
1316 * cases where it makes no sense for a protocol to have a "do nothing"
1317 * function, some default processing is provided.
1320 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1325 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1331 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1336 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1341 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1347 unsigned int sock_no_poll(struct file
* file
, struct socket
*sock
, poll_table
*pt
)
1352 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1357 int sock_no_listen(struct socket
*sock
, int backlog
)
1362 int sock_no_shutdown(struct socket
*sock
, int how
)
1367 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
1368 char __user
*optval
, int optlen
)
1373 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
1374 char __user
*optval
, int __user
*optlen
)
1379 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1385 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1386 size_t len
, int flags
)
1391 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
1393 /* Mirror missing mmap method error code */
1397 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
1400 struct msghdr msg
= {.msg_flags
= flags
};
1402 char *kaddr
= kmap(page
);
1403 iov
.iov_base
= kaddr
+ offset
;
1405 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1411 * Default Socket Callbacks
1414 static void sock_def_wakeup(struct sock
*sk
)
1416 read_lock(&sk
->sk_callback_lock
);
1417 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1418 wake_up_interruptible_all(sk
->sk_sleep
);
1419 read_unlock(&sk
->sk_callback_lock
);
1422 static void sock_def_error_report(struct sock
*sk
)
1424 read_lock(&sk
->sk_callback_lock
);
1425 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1426 wake_up_interruptible(sk
->sk_sleep
);
1427 sk_wake_async(sk
,0,POLL_ERR
);
1428 read_unlock(&sk
->sk_callback_lock
);
1431 static void sock_def_readable(struct sock
*sk
, int len
)
1433 read_lock(&sk
->sk_callback_lock
);
1434 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1435 wake_up_interruptible(sk
->sk_sleep
);
1436 sk_wake_async(sk
,1,POLL_IN
);
1437 read_unlock(&sk
->sk_callback_lock
);
1440 static void sock_def_write_space(struct sock
*sk
)
1442 read_lock(&sk
->sk_callback_lock
);
1444 /* Do not wake up a writer until he can make "significant"
1447 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
1448 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1449 wake_up_interruptible(sk
->sk_sleep
);
1451 /* Should agree with poll, otherwise some programs break */
1452 if (sock_writeable(sk
))
1453 sk_wake_async(sk
, 2, POLL_OUT
);
1456 read_unlock(&sk
->sk_callback_lock
);
1459 static void sock_def_destruct(struct sock
*sk
)
1461 kfree(sk
->sk_protinfo
);
1464 void sk_send_sigurg(struct sock
*sk
)
1466 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
1467 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
1468 sk_wake_async(sk
, 3, POLL_PRI
);
1471 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
1472 unsigned long expires
)
1474 if (!mod_timer(timer
, expires
))
1478 EXPORT_SYMBOL(sk_reset_timer
);
1480 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
1482 if (timer_pending(timer
) && del_timer(timer
))
1486 EXPORT_SYMBOL(sk_stop_timer
);
1488 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
1490 skb_queue_head_init(&sk
->sk_receive_queue
);
1491 skb_queue_head_init(&sk
->sk_write_queue
);
1492 skb_queue_head_init(&sk
->sk_error_queue
);
1493 #ifdef CONFIG_NET_DMA
1494 skb_queue_head_init(&sk
->sk_async_wait_queue
);
1497 sk
->sk_send_head
= NULL
;
1499 init_timer(&sk
->sk_timer
);
1501 sk
->sk_allocation
= GFP_KERNEL
;
1502 sk
->sk_rcvbuf
= sysctl_rmem_default
;
1503 sk
->sk_sndbuf
= sysctl_wmem_default
;
1504 sk
->sk_state
= TCP_CLOSE
;
1505 sk
->sk_socket
= sock
;
1507 sock_set_flag(sk
, SOCK_ZAPPED
);
1510 sk
->sk_type
= sock
->type
;
1511 sk
->sk_sleep
= &sock
->wait
;
1514 sk
->sk_sleep
= NULL
;
1516 rwlock_init(&sk
->sk_dst_lock
);
1517 rwlock_init(&sk
->sk_callback_lock
);
1518 lockdep_set_class(&sk
->sk_callback_lock
,
1519 af_callback_keys
+ sk
->sk_family
);
1521 sk
->sk_state_change
= sock_def_wakeup
;
1522 sk
->sk_data_ready
= sock_def_readable
;
1523 sk
->sk_write_space
= sock_def_write_space
;
1524 sk
->sk_error_report
= sock_def_error_report
;
1525 sk
->sk_destruct
= sock_def_destruct
;
1527 sk
->sk_sndmsg_page
= NULL
;
1528 sk
->sk_sndmsg_off
= 0;
1530 sk
->sk_peercred
.pid
= 0;
1531 sk
->sk_peercred
.uid
= -1;
1532 sk
->sk_peercred
.gid
= -1;
1533 sk
->sk_write_pending
= 0;
1534 sk
->sk_rcvlowat
= 1;
1535 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
1536 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
1538 sk
->sk_stamp
= ktime_set(-1L, -1L);
1540 atomic_set(&sk
->sk_refcnt
, 1);
1543 void fastcall
lock_sock_nested(struct sock
*sk
, int subclass
)
1546 spin_lock_bh(&sk
->sk_lock
.slock
);
1547 if (sk
->sk_lock
.owner
)
1549 sk
->sk_lock
.owner
= (void *)1;
1550 spin_unlock(&sk
->sk_lock
.slock
);
1552 * The sk_lock has mutex_lock() semantics here:
1554 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
1558 EXPORT_SYMBOL(lock_sock_nested
);
1560 void fastcall
release_sock(struct sock
*sk
)
1563 * The sk_lock has mutex_unlock() semantics:
1565 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
1567 spin_lock_bh(&sk
->sk_lock
.slock
);
1568 if (sk
->sk_backlog
.tail
)
1570 sk
->sk_lock
.owner
= NULL
;
1571 if (waitqueue_active(&sk
->sk_lock
.wq
))
1572 wake_up(&sk
->sk_lock
.wq
);
1573 spin_unlock_bh(&sk
->sk_lock
.slock
);
1575 EXPORT_SYMBOL(release_sock
);
1577 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
1580 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
1581 sock_enable_timestamp(sk
);
1582 tv
= ktime_to_timeval(sk
->sk_stamp
);
1583 if (tv
.tv_sec
== -1)
1585 if (tv
.tv_sec
== 0) {
1586 sk
->sk_stamp
= ktime_get_real();
1587 tv
= ktime_to_timeval(sk
->sk_stamp
);
1589 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
1591 EXPORT_SYMBOL(sock_get_timestamp
);
1593 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
1596 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
1597 sock_enable_timestamp(sk
);
1598 ts
= ktime_to_timespec(sk
->sk_stamp
);
1599 if (ts
.tv_sec
== -1)
1601 if (ts
.tv_sec
== 0) {
1602 sk
->sk_stamp
= ktime_get_real();
1603 ts
= ktime_to_timespec(sk
->sk_stamp
);
1605 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
1607 EXPORT_SYMBOL(sock_get_timestampns
);
1609 void sock_enable_timestamp(struct sock
*sk
)
1611 if (!sock_flag(sk
, SOCK_TIMESTAMP
)) {
1612 sock_set_flag(sk
, SOCK_TIMESTAMP
);
1613 net_enable_timestamp();
1616 EXPORT_SYMBOL(sock_enable_timestamp
);
1619 * Get a socket option on an socket.
1621 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1622 * asynchronous errors should be reported by getsockopt. We assume
1623 * this means if you specify SO_ERROR (otherwise whats the point of it).
1625 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1626 char __user
*optval
, int __user
*optlen
)
1628 struct sock
*sk
= sock
->sk
;
1630 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
1633 EXPORT_SYMBOL(sock_common_getsockopt
);
1635 #ifdef CONFIG_COMPAT
1636 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
1637 char __user
*optval
, int __user
*optlen
)
1639 struct sock
*sk
= sock
->sk
;
1641 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
1642 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
1644 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
1646 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
1649 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1650 struct msghdr
*msg
, size_t size
, int flags
)
1652 struct sock
*sk
= sock
->sk
;
1656 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
1657 flags
& ~MSG_DONTWAIT
, &addr_len
);
1659 msg
->msg_namelen
= addr_len
;
1663 EXPORT_SYMBOL(sock_common_recvmsg
);
1666 * Set socket options on an inet socket.
1668 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1669 char __user
*optval
, int optlen
)
1671 struct sock
*sk
= sock
->sk
;
1673 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
1676 EXPORT_SYMBOL(sock_common_setsockopt
);
1678 #ifdef CONFIG_COMPAT
1679 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
1680 char __user
*optval
, int optlen
)
1682 struct sock
*sk
= sock
->sk
;
1684 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
1685 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
1687 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
1689 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
1692 void sk_common_release(struct sock
*sk
)
1694 if (sk
->sk_prot
->destroy
)
1695 sk
->sk_prot
->destroy(sk
);
1698 * Observation: when sock_common_release is called, processes have
1699 * no access to socket. But net still has.
1700 * Step one, detach it from networking:
1702 * A. Remove from hash tables.
1705 sk
->sk_prot
->unhash(sk
);
1708 * In this point socket cannot receive new packets, but it is possible
1709 * that some packets are in flight because some CPU runs receiver and
1710 * did hash table lookup before we unhashed socket. They will achieve
1711 * receive queue and will be purged by socket destructor.
1713 * Also we still have packets pending on receive queue and probably,
1714 * our own packets waiting in device queues. sock_destroy will drain
1715 * receive queue, but transmitted packets will delay socket destruction
1716 * until the last reference will be released.
1721 xfrm_sk_free_policy(sk
);
1723 sk_refcnt_debug_release(sk
);
1727 EXPORT_SYMBOL(sk_common_release
);
1729 static DEFINE_RWLOCK(proto_list_lock
);
1730 static LIST_HEAD(proto_list
);
1732 int proto_register(struct proto
*prot
, int alloc_slab
)
1734 char *request_sock_slab_name
= NULL
;
1735 char *timewait_sock_slab_name
;
1739 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
1740 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1742 if (prot
->slab
== NULL
) {
1743 printk(KERN_CRIT
"%s: Can't create sock SLAB cache!\n",
1748 if (prot
->rsk_prot
!= NULL
) {
1749 static const char mask
[] = "request_sock_%s";
1751 request_sock_slab_name
= kmalloc(strlen(prot
->name
) + sizeof(mask
) - 1, GFP_KERNEL
);
1752 if (request_sock_slab_name
== NULL
)
1753 goto out_free_sock_slab
;
1755 sprintf(request_sock_slab_name
, mask
, prot
->name
);
1756 prot
->rsk_prot
->slab
= kmem_cache_create(request_sock_slab_name
,
1757 prot
->rsk_prot
->obj_size
, 0,
1758 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1760 if (prot
->rsk_prot
->slab
== NULL
) {
1761 printk(KERN_CRIT
"%s: Can't create request sock SLAB cache!\n",
1763 goto out_free_request_sock_slab_name
;
1767 if (prot
->twsk_prot
!= NULL
) {
1768 static const char mask
[] = "tw_sock_%s";
1770 timewait_sock_slab_name
= kmalloc(strlen(prot
->name
) + sizeof(mask
) - 1, GFP_KERNEL
);
1772 if (timewait_sock_slab_name
== NULL
)
1773 goto out_free_request_sock_slab
;
1775 sprintf(timewait_sock_slab_name
, mask
, prot
->name
);
1776 prot
->twsk_prot
->twsk_slab
=
1777 kmem_cache_create(timewait_sock_slab_name
,
1778 prot
->twsk_prot
->twsk_obj_size
,
1779 0, SLAB_HWCACHE_ALIGN
,
1781 if (prot
->twsk_prot
->twsk_slab
== NULL
)
1782 goto out_free_timewait_sock_slab_name
;
1786 write_lock(&proto_list_lock
);
1787 list_add(&prot
->node
, &proto_list
);
1788 write_unlock(&proto_list_lock
);
1792 out_free_timewait_sock_slab_name
:
1793 kfree(timewait_sock_slab_name
);
1794 out_free_request_sock_slab
:
1795 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
1796 kmem_cache_destroy(prot
->rsk_prot
->slab
);
1797 prot
->rsk_prot
->slab
= NULL
;
1799 out_free_request_sock_slab_name
:
1800 kfree(request_sock_slab_name
);
1802 kmem_cache_destroy(prot
->slab
);
1807 EXPORT_SYMBOL(proto_register
);
1809 void proto_unregister(struct proto
*prot
)
1811 write_lock(&proto_list_lock
);
1812 list_del(&prot
->node
);
1813 write_unlock(&proto_list_lock
);
1815 if (prot
->slab
!= NULL
) {
1816 kmem_cache_destroy(prot
->slab
);
1820 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
1821 const char *name
= kmem_cache_name(prot
->rsk_prot
->slab
);
1823 kmem_cache_destroy(prot
->rsk_prot
->slab
);
1825 prot
->rsk_prot
->slab
= NULL
;
1828 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
1829 const char *name
= kmem_cache_name(prot
->twsk_prot
->twsk_slab
);
1831 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
1833 prot
->twsk_prot
->twsk_slab
= NULL
;
1837 EXPORT_SYMBOL(proto_unregister
);
1839 #ifdef CONFIG_PROC_FS
1840 static inline struct proto
*__proto_head(void)
1842 return list_entry(proto_list
.next
, struct proto
, node
);
1845 static inline struct proto
*proto_head(void)
1847 return list_empty(&proto_list
) ? NULL
: __proto_head();
1850 static inline struct proto
*proto_next(struct proto
*proto
)
1852 return proto
->node
.next
== &proto_list
? NULL
:
1853 list_entry(proto
->node
.next
, struct proto
, node
);
1856 static inline struct proto
*proto_get_idx(loff_t pos
)
1858 struct proto
*proto
;
1861 list_for_each_entry(proto
, &proto_list
, node
)
1870 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1872 read_lock(&proto_list_lock
);
1873 return *pos
? proto_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1876 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1879 return v
== SEQ_START_TOKEN
? proto_head() : proto_next(v
);
1882 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
1884 read_unlock(&proto_list_lock
);
1887 static char proto_method_implemented(const void *method
)
1889 return method
== NULL
? 'n' : 'y';
1892 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
1894 seq_printf(seq
, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1895 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1898 proto
->sockets_allocated
!= NULL
? atomic_read(proto
->sockets_allocated
) : -1,
1899 proto
->memory_allocated
!= NULL
? atomic_read(proto
->memory_allocated
) : -1,
1900 proto
->memory_pressure
!= NULL
? *proto
->memory_pressure
? "yes" : "no" : "NI",
1902 proto
->slab
== NULL
? "no" : "yes",
1903 module_name(proto
->owner
),
1904 proto_method_implemented(proto
->close
),
1905 proto_method_implemented(proto
->connect
),
1906 proto_method_implemented(proto
->disconnect
),
1907 proto_method_implemented(proto
->accept
),
1908 proto_method_implemented(proto
->ioctl
),
1909 proto_method_implemented(proto
->init
),
1910 proto_method_implemented(proto
->destroy
),
1911 proto_method_implemented(proto
->shutdown
),
1912 proto_method_implemented(proto
->setsockopt
),
1913 proto_method_implemented(proto
->getsockopt
),
1914 proto_method_implemented(proto
->sendmsg
),
1915 proto_method_implemented(proto
->recvmsg
),
1916 proto_method_implemented(proto
->sendpage
),
1917 proto_method_implemented(proto
->bind
),
1918 proto_method_implemented(proto
->backlog_rcv
),
1919 proto_method_implemented(proto
->hash
),
1920 proto_method_implemented(proto
->unhash
),
1921 proto_method_implemented(proto
->get_port
),
1922 proto_method_implemented(proto
->enter_memory_pressure
));
1925 static int proto_seq_show(struct seq_file
*seq
, void *v
)
1927 if (v
== SEQ_START_TOKEN
)
1928 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1937 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1939 proto_seq_printf(seq
, v
);
1943 static const struct seq_operations proto_seq_ops
= {
1944 .start
= proto_seq_start
,
1945 .next
= proto_seq_next
,
1946 .stop
= proto_seq_stop
,
1947 .show
= proto_seq_show
,
1950 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
1952 return seq_open(file
, &proto_seq_ops
);
1955 static const struct file_operations proto_seq_fops
= {
1956 .owner
= THIS_MODULE
,
1957 .open
= proto_seq_open
,
1959 .llseek
= seq_lseek
,
1960 .release
= seq_release
,
1963 static int __init
proto_init(void)
1965 /* register /proc/net/protocols */
1966 return proc_net_fops_create("protocols", S_IRUGO
, &proto_seq_fops
) == NULL
? -ENOBUFS
: 0;
1969 subsys_initcall(proto_init
);
1971 #endif /* PROC_FS */
1973 EXPORT_SYMBOL(sk_alloc
);
1974 EXPORT_SYMBOL(sk_free
);
1975 EXPORT_SYMBOL(sk_send_sigurg
);
1976 EXPORT_SYMBOL(sock_alloc_send_skb
);
1977 EXPORT_SYMBOL(sock_init_data
);
1978 EXPORT_SYMBOL(sock_kfree_s
);
1979 EXPORT_SYMBOL(sock_kmalloc
);
1980 EXPORT_SYMBOL(sock_no_accept
);
1981 EXPORT_SYMBOL(sock_no_bind
);
1982 EXPORT_SYMBOL(sock_no_connect
);
1983 EXPORT_SYMBOL(sock_no_getname
);
1984 EXPORT_SYMBOL(sock_no_getsockopt
);
1985 EXPORT_SYMBOL(sock_no_ioctl
);
1986 EXPORT_SYMBOL(sock_no_listen
);
1987 EXPORT_SYMBOL(sock_no_mmap
);
1988 EXPORT_SYMBOL(sock_no_poll
);
1989 EXPORT_SYMBOL(sock_no_recvmsg
);
1990 EXPORT_SYMBOL(sock_no_sendmsg
);
1991 EXPORT_SYMBOL(sock_no_sendpage
);
1992 EXPORT_SYMBOL(sock_no_setsockopt
);
1993 EXPORT_SYMBOL(sock_no_shutdown
);
1994 EXPORT_SYMBOL(sock_no_socketpair
);
1995 EXPORT_SYMBOL(sock_rfree
);
1996 EXPORT_SYMBOL(sock_setsockopt
);
1997 EXPORT_SYMBOL(sock_wfree
);
1998 EXPORT_SYMBOL(sock_wmalloc
);
1999 EXPORT_SYMBOL(sock_i_uid
);
2000 EXPORT_SYMBOL(sock_i_ino
);
2001 EXPORT_SYMBOL(sysctl_optmem_max
);
2002 #ifdef CONFIG_SYSCTL
2003 EXPORT_SYMBOL(sysctl_rmem_max
);
2004 EXPORT_SYMBOL(sysctl_wmem_max
);