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.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
114 #include <asm/uaccess.h>
115 #include <asm/system.h>
117 #include <linux/netdevice.h>
118 #include <net/protocol.h>
119 #include <linux/skbuff.h>
120 #include <net/net_namespace.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <linux/net_tstamp.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
126 #include <net/cls_cgroup.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
];
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 *const 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-AF_RDS" , "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-AF_CAN" ,
157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
158 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
159 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" ,
162 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
163 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
164 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
165 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
166 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
167 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
168 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
169 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
170 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
171 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
172 "slock-27" , "slock-28" , "slock-AF_CAN" ,
173 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
174 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
175 "slock-AF_IEEE802154", "slock-AF_CAIF" ,
178 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
179 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
180 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
181 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
182 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
183 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
184 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
185 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
186 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
187 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
188 "clock-27" , "clock-28" , "clock-AF_CAN" ,
189 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
190 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
191 "clock-AF_IEEE802154", "clock-AF_CAIF" ,
196 * sk_callback_lock locking rules are per-address-family,
197 * so split the lock classes by using a per-AF key:
199 static struct lock_class_key af_callback_keys
[AF_MAX
];
201 /* Take into consideration the size of the struct sk_buff overhead in the
202 * determination of these values, since that is non-constant across
203 * platforms. This makes socket queueing behavior and performance
204 * not depend upon such differences.
206 #define _SK_MEM_PACKETS 256
207 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
208 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
209 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
211 /* Run time adjustable parameters. */
212 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
213 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
214 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
215 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
217 /* Maximal space eaten by iovec or ancilliary data plus some space */
218 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
219 EXPORT_SYMBOL(sysctl_optmem_max
);
221 #if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
222 int net_cls_subsys_id
= -1;
223 EXPORT_SYMBOL_GPL(net_cls_subsys_id
);
226 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
230 if (optlen
< sizeof(tv
))
232 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
234 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
238 static int warned __read_mostly
;
241 if (warned
< 10 && net_ratelimit()) {
243 printk(KERN_INFO
"sock_set_timeout: `%s' (pid %d) "
244 "tries to set negative timeout\n",
245 current
->comm
, task_pid_nr(current
));
249 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
250 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
252 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
253 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
257 static void sock_warn_obsolete_bsdism(const char *name
)
260 static char warncomm
[TASK_COMM_LEN
];
261 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
262 strcpy(warncomm
, current
->comm
);
263 printk(KERN_WARNING
"process `%s' is using obsolete "
264 "%s SO_BSDCOMPAT\n", warncomm
, name
);
269 static void sock_disable_timestamp(struct sock
*sk
, int flag
)
271 if (sock_flag(sk
, flag
)) {
272 sock_reset_flag(sk
, flag
);
273 if (!sock_flag(sk
, SOCK_TIMESTAMP
) &&
274 !sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
)) {
275 net_disable_timestamp();
281 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
286 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
288 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
289 number of warnings when compiling with -W --ANK
291 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
292 (unsigned)sk
->sk_rcvbuf
) {
293 atomic_inc(&sk
->sk_drops
);
297 err
= sk_filter(sk
, skb
);
301 if (!sk_rmem_schedule(sk
, skb
->truesize
)) {
302 atomic_inc(&sk
->sk_drops
);
307 skb_set_owner_r(skb
, sk
);
309 /* Cache the SKB length before we tack it onto the receive
310 * queue. Once it is added it no longer belongs to us and
311 * may be freed by other threads of control pulling packets
316 /* we escape from rcu protected region, make sure we dont leak
321 spin_lock_irqsave(&list
->lock
, flags
);
322 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
323 __skb_queue_tail(list
, skb
);
324 spin_unlock_irqrestore(&list
->lock
, flags
);
326 if (!sock_flag(sk
, SOCK_DEAD
))
327 sk
->sk_data_ready(sk
, skb_len
);
330 EXPORT_SYMBOL(sock_queue_rcv_skb
);
332 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
334 int rc
= NET_RX_SUCCESS
;
336 if (sk_filter(sk
, skb
))
337 goto discard_and_relse
;
341 if (sk_rcvqueues_full(sk
, skb
)) {
342 atomic_inc(&sk
->sk_drops
);
343 goto discard_and_relse
;
346 bh_lock_sock_nested(sk
);
349 if (!sock_owned_by_user(sk
)) {
351 * trylock + unlock semantics:
353 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
355 rc
= sk_backlog_rcv(sk
, skb
);
357 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
358 } else if (sk_add_backlog(sk
, skb
)) {
360 atomic_inc(&sk
->sk_drops
);
361 goto discard_and_relse
;
372 EXPORT_SYMBOL(sk_receive_skb
);
374 void sk_reset_txq(struct sock
*sk
)
376 sk_tx_queue_clear(sk
);
378 EXPORT_SYMBOL(sk_reset_txq
);
380 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
382 struct dst_entry
*dst
= __sk_dst_get(sk
);
384 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
385 sk_tx_queue_clear(sk
);
386 rcu_assign_pointer(sk
->sk_dst_cache
, NULL
);
393 EXPORT_SYMBOL(__sk_dst_check
);
395 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
397 struct dst_entry
*dst
= sk_dst_get(sk
);
399 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
407 EXPORT_SYMBOL(sk_dst_check
);
409 static int sock_bindtodevice(struct sock
*sk
, char __user
*optval
, int optlen
)
411 int ret
= -ENOPROTOOPT
;
412 #ifdef CONFIG_NETDEVICES
413 struct net
*net
= sock_net(sk
);
414 char devname
[IFNAMSIZ
];
419 if (!capable(CAP_NET_RAW
))
426 /* Bind this socket to a particular device like "eth0",
427 * as specified in the passed interface name. If the
428 * name is "" or the option length is zero the socket
431 if (optlen
> IFNAMSIZ
- 1)
432 optlen
= IFNAMSIZ
- 1;
433 memset(devname
, 0, sizeof(devname
));
436 if (copy_from_user(devname
, optval
, optlen
))
440 if (devname
[0] != '\0') {
441 struct net_device
*dev
;
444 dev
= dev_get_by_name_rcu(net
, devname
);
446 index
= dev
->ifindex
;
454 sk
->sk_bound_dev_if
= index
;
466 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
469 sock_set_flag(sk
, bit
);
471 sock_reset_flag(sk
, bit
);
475 * This is meant for all protocols to use and covers goings on
476 * at the socket level. Everything here is generic.
479 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
480 char __user
*optval
, unsigned int optlen
)
482 struct sock
*sk
= sock
->sk
;
489 * Options without arguments
492 if (optname
== SO_BINDTODEVICE
)
493 return sock_bindtodevice(sk
, optval
, optlen
);
495 if (optlen
< sizeof(int))
498 if (get_user(val
, (int __user
*)optval
))
501 valbool
= val
? 1 : 0;
507 if (val
&& !capable(CAP_NET_ADMIN
))
510 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
513 sk
->sk_reuse
= valbool
;
522 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
525 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
528 /* Don't error on this BSD doesn't and if you think
529 about it this is right. Otherwise apps have to
530 play 'guess the biggest size' games. RCVBUF/SNDBUF
531 are treated in BSD as hints */
533 if (val
> sysctl_wmem_max
)
534 val
= sysctl_wmem_max
;
536 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
537 if ((val
* 2) < SOCK_MIN_SNDBUF
)
538 sk
->sk_sndbuf
= SOCK_MIN_SNDBUF
;
540 sk
->sk_sndbuf
= val
* 2;
543 * Wake up sending tasks if we
546 sk
->sk_write_space(sk
);
550 if (!capable(CAP_NET_ADMIN
)) {
557 /* Don't error on this BSD doesn't and if you think
558 about it this is right. Otherwise apps have to
559 play 'guess the biggest size' games. RCVBUF/SNDBUF
560 are treated in BSD as hints */
562 if (val
> sysctl_rmem_max
)
563 val
= sysctl_rmem_max
;
565 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
567 * We double it on the way in to account for
568 * "struct sk_buff" etc. overhead. Applications
569 * assume that the SO_RCVBUF setting they make will
570 * allow that much actual data to be received on that
573 * Applications are unaware that "struct sk_buff" and
574 * other overheads allocate from the receive buffer
575 * during socket buffer allocation.
577 * And after considering the possible alternatives,
578 * returning the value we actually used in getsockopt
579 * is the most desirable behavior.
581 if ((val
* 2) < SOCK_MIN_RCVBUF
)
582 sk
->sk_rcvbuf
= SOCK_MIN_RCVBUF
;
584 sk
->sk_rcvbuf
= val
* 2;
588 if (!capable(CAP_NET_ADMIN
)) {
596 if (sk
->sk_protocol
== IPPROTO_TCP
)
597 tcp_set_keepalive(sk
, valbool
);
599 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
603 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
607 sk
->sk_no_check
= valbool
;
611 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
612 sk
->sk_priority
= val
;
618 if (optlen
< sizeof(ling
)) {
619 ret
= -EINVAL
; /* 1003.1g */
622 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
627 sock_reset_flag(sk
, SOCK_LINGER
);
629 #if (BITS_PER_LONG == 32)
630 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
631 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
634 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
635 sock_set_flag(sk
, SOCK_LINGER
);
640 sock_warn_obsolete_bsdism("setsockopt");
645 set_bit(SOCK_PASSCRED
, &sock
->flags
);
647 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
653 if (optname
== SO_TIMESTAMP
)
654 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
656 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
657 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
658 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
660 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
661 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
665 case SO_TIMESTAMPING
:
666 if (val
& ~SOF_TIMESTAMPING_MASK
) {
670 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
671 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
672 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
673 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
674 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
675 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
676 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
677 sock_enable_timestamp(sk
,
678 SOCK_TIMESTAMPING_RX_SOFTWARE
);
680 sock_disable_timestamp(sk
,
681 SOCK_TIMESTAMPING_RX_SOFTWARE
);
682 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
683 val
& SOF_TIMESTAMPING_SOFTWARE
);
684 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
685 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
686 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
687 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
693 sk
->sk_rcvlowat
= val
? : 1;
697 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
701 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
704 case SO_ATTACH_FILTER
:
706 if (optlen
== sizeof(struct sock_fprog
)) {
707 struct sock_fprog fprog
;
710 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
713 ret
= sk_attach_filter(&fprog
, sk
);
717 case SO_DETACH_FILTER
:
718 ret
= sk_detach_filter(sk
);
723 set_bit(SOCK_PASSSEC
, &sock
->flags
);
725 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
728 if (!capable(CAP_NET_ADMIN
))
734 /* We implement the SO_SNDLOWAT etc to
735 not be settable (1003.1g 5.3) */
738 sock_set_flag(sk
, SOCK_RXQ_OVFL
);
740 sock_reset_flag(sk
, SOCK_RXQ_OVFL
);
749 EXPORT_SYMBOL(sock_setsockopt
);
752 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
753 char __user
*optval
, int __user
*optlen
)
755 struct sock
*sk
= sock
->sk
;
763 int lv
= sizeof(int);
766 if (get_user(len
, optlen
))
771 memset(&v
, 0, sizeof(v
));
775 v
.val
= sock_flag(sk
, SOCK_DBG
);
779 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
783 v
.val
= !!sock_flag(sk
, SOCK_BROADCAST
);
787 v
.val
= sk
->sk_sndbuf
;
791 v
.val
= sk
->sk_rcvbuf
;
795 v
.val
= sk
->sk_reuse
;
799 v
.val
= !!sock_flag(sk
, SOCK_KEEPOPEN
);
807 v
.val
= sk
->sk_protocol
;
811 v
.val
= sk
->sk_family
;
815 v
.val
= -sock_error(sk
);
817 v
.val
= xchg(&sk
->sk_err_soft
, 0);
821 v
.val
= !!sock_flag(sk
, SOCK_URGINLINE
);
825 v
.val
= sk
->sk_no_check
;
829 v
.val
= sk
->sk_priority
;
834 v
.ling
.l_onoff
= !!sock_flag(sk
, SOCK_LINGER
);
835 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
839 sock_warn_obsolete_bsdism("getsockopt");
843 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
844 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
848 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
851 case SO_TIMESTAMPING
:
853 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
854 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
855 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
856 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
857 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
858 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
859 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
860 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
861 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
862 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
863 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
864 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
865 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
866 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
870 lv
= sizeof(struct timeval
);
871 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
875 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
876 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
881 lv
= sizeof(struct timeval
);
882 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
886 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
887 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
892 v
.val
= sk
->sk_rcvlowat
;
900 v
.val
= test_bit(SOCK_PASSCRED
, &sock
->flags
) ? 1 : 0;
904 if (len
> sizeof(sk
->sk_peercred
))
905 len
= sizeof(sk
->sk_peercred
);
906 if (copy_to_user(optval
, &sk
->sk_peercred
, len
))
914 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
918 if (copy_to_user(optval
, address
, len
))
923 /* Dubious BSD thing... Probably nobody even uses it, but
924 * the UNIX standard wants it for whatever reason... -DaveM
927 v
.val
= sk
->sk_state
== TCP_LISTEN
;
931 v
.val
= test_bit(SOCK_PASSSEC
, &sock
->flags
) ? 1 : 0;
935 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
942 v
.val
= !!sock_flag(sk
, SOCK_RXQ_OVFL
);
951 if (copy_to_user(optval
, &v
, len
))
954 if (put_user(len
, optlen
))
960 * Initialize an sk_lock.
962 * (We also register the sk_lock with the lock validator.)
964 static inline void sock_lock_init(struct sock
*sk
)
966 sock_lock_init_class_and_name(sk
,
967 af_family_slock_key_strings
[sk
->sk_family
],
968 af_family_slock_keys
+ sk
->sk_family
,
969 af_family_key_strings
[sk
->sk_family
],
970 af_family_keys
+ sk
->sk_family
);
974 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
975 * even temporarly, because of RCU lookups. sk_node should also be left as is.
977 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
979 #ifdef CONFIG_SECURITY_NETWORK
980 void *sptr
= nsk
->sk_security
;
982 BUILD_BUG_ON(offsetof(struct sock
, sk_copy_start
) !=
983 sizeof(osk
->sk_node
) + sizeof(osk
->sk_refcnt
) +
984 sizeof(osk
->sk_tx_queue_mapping
));
985 memcpy(&nsk
->sk_copy_start
, &osk
->sk_copy_start
,
986 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_copy_start
));
987 #ifdef CONFIG_SECURITY_NETWORK
988 nsk
->sk_security
= sptr
;
989 security_sk_clone(osk
, nsk
);
993 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
997 struct kmem_cache
*slab
;
1001 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1004 if (priority
& __GFP_ZERO
) {
1006 * caches using SLAB_DESTROY_BY_RCU should let
1007 * sk_node.next un-modified. Special care is taken
1008 * when initializing object to zero.
1010 if (offsetof(struct sock
, sk_node
.next
) != 0)
1011 memset(sk
, 0, offsetof(struct sock
, sk_node
.next
));
1012 memset(&sk
->sk_node
.pprev
, 0,
1013 prot
->obj_size
- offsetof(struct sock
,
1018 sk
= kmalloc(prot
->obj_size
, priority
);
1021 kmemcheck_annotate_bitfield(sk
, flags
);
1023 if (security_sk_alloc(sk
, family
, priority
))
1026 if (!try_module_get(prot
->owner
))
1028 sk_tx_queue_clear(sk
);
1034 security_sk_free(sk
);
1037 kmem_cache_free(slab
, sk
);
1043 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1045 struct kmem_cache
*slab
;
1046 struct module
*owner
;
1048 owner
= prot
->owner
;
1051 security_sk_free(sk
);
1053 kmem_cache_free(slab
, sk
);
1059 #ifdef CONFIG_CGROUPS
1060 void sock_update_classid(struct sock
*sk
)
1062 u32 classid
= task_cls_classid(current
);
1064 if (classid
&& classid
!= sk
->sk_classid
)
1065 sk
->sk_classid
= classid
;
1067 EXPORT_SYMBOL(sock_update_classid
);
1071 * sk_alloc - All socket objects are allocated here
1072 * @net: the applicable net namespace
1073 * @family: protocol family
1074 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1075 * @prot: struct proto associated with this new sock instance
1077 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1082 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1084 sk
->sk_family
= family
;
1086 * See comment in struct sock definition to understand
1087 * why we need sk_prot_creator -acme
1089 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1091 sock_net_set(sk
, get_net(net
));
1092 atomic_set(&sk
->sk_wmem_alloc
, 1);
1094 sock_update_classid(sk
);
1099 EXPORT_SYMBOL(sk_alloc
);
1101 static void __sk_free(struct sock
*sk
)
1103 struct sk_filter
*filter
;
1105 if (sk
->sk_destruct
)
1106 sk
->sk_destruct(sk
);
1108 filter
= rcu_dereference_check(sk
->sk_filter
,
1109 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1111 sk_filter_uncharge(sk
, filter
);
1112 rcu_assign_pointer(sk
->sk_filter
, NULL
);
1115 sock_disable_timestamp(sk
, SOCK_TIMESTAMP
);
1116 sock_disable_timestamp(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
);
1118 if (atomic_read(&sk
->sk_omem_alloc
))
1119 printk(KERN_DEBUG
"%s: optmem leakage (%d bytes) detected.\n",
1120 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1122 put_net(sock_net(sk
));
1123 sk_prot_free(sk
->sk_prot_creator
, sk
);
1126 void sk_free(struct sock
*sk
)
1129 * We substract one from sk_wmem_alloc and can know if
1130 * some packets are still in some tx queue.
1131 * If not null, sock_wfree() will call __sk_free(sk) later
1133 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1136 EXPORT_SYMBOL(sk_free
);
1139 * Last sock_put should drop referrence to sk->sk_net. It has already
1140 * been dropped in sk_change_net. Taking referrence to stopping namespace
1142 * Take referrence to a socket to remove it from hash _alive_ and after that
1143 * destroy it in the context of init_net.
1145 void sk_release_kernel(struct sock
*sk
)
1147 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1151 sock_release(sk
->sk_socket
);
1152 release_net(sock_net(sk
));
1153 sock_net_set(sk
, get_net(&init_net
));
1156 EXPORT_SYMBOL(sk_release_kernel
);
1158 struct sock
*sk_clone(const struct sock
*sk
, const gfp_t priority
)
1162 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1163 if (newsk
!= NULL
) {
1164 struct sk_filter
*filter
;
1166 sock_copy(newsk
, sk
);
1169 get_net(sock_net(newsk
));
1170 sk_node_init(&newsk
->sk_node
);
1171 sock_lock_init(newsk
);
1172 bh_lock_sock(newsk
);
1173 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1174 newsk
->sk_backlog
.len
= 0;
1176 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1178 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1180 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1181 atomic_set(&newsk
->sk_omem_alloc
, 0);
1182 skb_queue_head_init(&newsk
->sk_receive_queue
);
1183 skb_queue_head_init(&newsk
->sk_write_queue
);
1184 #ifdef CONFIG_NET_DMA
1185 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1188 spin_lock_init(&newsk
->sk_dst_lock
);
1189 rwlock_init(&newsk
->sk_callback_lock
);
1190 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1191 af_callback_keys
+ newsk
->sk_family
,
1192 af_family_clock_key_strings
[newsk
->sk_family
]);
1194 newsk
->sk_dst_cache
= NULL
;
1195 newsk
->sk_wmem_queued
= 0;
1196 newsk
->sk_forward_alloc
= 0;
1197 newsk
->sk_send_head
= NULL
;
1198 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1200 sock_reset_flag(newsk
, SOCK_DONE
);
1201 skb_queue_head_init(&newsk
->sk_error_queue
);
1203 filter
= newsk
->sk_filter
;
1205 sk_filter_charge(newsk
, filter
);
1207 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1208 /* It is still raw copy of parent, so invalidate
1209 * destructor and make plain sk_free() */
1210 newsk
->sk_destruct
= NULL
;
1217 newsk
->sk_priority
= 0;
1219 * Before updating sk_refcnt, we must commit prior changes to memory
1220 * (Documentation/RCU/rculist_nulls.txt for details)
1223 atomic_set(&newsk
->sk_refcnt
, 2);
1226 * Increment the counter in the same struct proto as the master
1227 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1228 * is the same as sk->sk_prot->socks, as this field was copied
1231 * This _changes_ the previous behaviour, where
1232 * tcp_create_openreq_child always was incrementing the
1233 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1234 * to be taken into account in all callers. -acme
1236 sk_refcnt_debug_inc(newsk
);
1237 sk_set_socket(newsk
, NULL
);
1238 newsk
->sk_wq
= NULL
;
1240 if (newsk
->sk_prot
->sockets_allocated
)
1241 percpu_counter_inc(newsk
->sk_prot
->sockets_allocated
);
1243 if (sock_flag(newsk
, SOCK_TIMESTAMP
) ||
1244 sock_flag(newsk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
1245 net_enable_timestamp();
1250 EXPORT_SYMBOL_GPL(sk_clone
);
1252 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1254 __sk_dst_set(sk
, dst
);
1255 sk
->sk_route_caps
= dst
->dev
->features
;
1256 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1257 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1258 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1259 if (sk_can_gso(sk
)) {
1260 if (dst
->header_len
) {
1261 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1263 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1264 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1268 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1270 void __init
sk_init(void)
1272 if (totalram_pages
<= 4096) {
1273 sysctl_wmem_max
= 32767;
1274 sysctl_rmem_max
= 32767;
1275 sysctl_wmem_default
= 32767;
1276 sysctl_rmem_default
= 32767;
1277 } else if (totalram_pages
>= 131072) {
1278 sysctl_wmem_max
= 131071;
1279 sysctl_rmem_max
= 131071;
1284 * Simple resource managers for sockets.
1289 * Write buffer destructor automatically called from kfree_skb.
1291 void sock_wfree(struct sk_buff
*skb
)
1293 struct sock
*sk
= skb
->sk
;
1294 unsigned int len
= skb
->truesize
;
1296 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1298 * Keep a reference on sk_wmem_alloc, this will be released
1299 * after sk_write_space() call
1301 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1302 sk
->sk_write_space(sk
);
1306 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1307 * could not do because of in-flight packets
1309 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1312 EXPORT_SYMBOL(sock_wfree
);
1315 * Read buffer destructor automatically called from kfree_skb.
1317 void sock_rfree(struct sk_buff
*skb
)
1319 struct sock
*sk
= skb
->sk
;
1321 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1322 sk_mem_uncharge(skb
->sk
, skb
->truesize
);
1324 EXPORT_SYMBOL(sock_rfree
);
1327 int sock_i_uid(struct sock
*sk
)
1331 read_lock(&sk
->sk_callback_lock
);
1332 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: 0;
1333 read_unlock(&sk
->sk_callback_lock
);
1336 EXPORT_SYMBOL(sock_i_uid
);
1338 unsigned long sock_i_ino(struct sock
*sk
)
1342 read_lock(&sk
->sk_callback_lock
);
1343 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1344 read_unlock(&sk
->sk_callback_lock
);
1347 EXPORT_SYMBOL(sock_i_ino
);
1350 * Allocate a skb from the socket's send buffer.
1352 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1355 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1356 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1358 skb_set_owner_w(skb
, sk
);
1364 EXPORT_SYMBOL(sock_wmalloc
);
1367 * Allocate a skb from the socket's receive buffer.
1369 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1372 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1373 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1375 skb_set_owner_r(skb
, sk
);
1383 * Allocate a memory block from the socket's option memory buffer.
1385 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1387 if ((unsigned)size
<= sysctl_optmem_max
&&
1388 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1390 /* First do the add, to avoid the race if kmalloc
1393 atomic_add(size
, &sk
->sk_omem_alloc
);
1394 mem
= kmalloc(size
, priority
);
1397 atomic_sub(size
, &sk
->sk_omem_alloc
);
1401 EXPORT_SYMBOL(sock_kmalloc
);
1404 * Free an option memory block.
1406 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1409 atomic_sub(size
, &sk
->sk_omem_alloc
);
1411 EXPORT_SYMBOL(sock_kfree_s
);
1413 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1414 I think, these locks should be removed for datagram sockets.
1416 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1420 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1424 if (signal_pending(current
))
1426 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1427 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1428 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1430 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1434 timeo
= schedule_timeout(timeo
);
1436 finish_wait(sk_sleep(sk
), &wait
);
1442 * Generic send/receive buffer handlers
1445 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1446 unsigned long data_len
, int noblock
,
1449 struct sk_buff
*skb
;
1454 gfp_mask
= sk
->sk_allocation
;
1455 if (gfp_mask
& __GFP_WAIT
)
1456 gfp_mask
|= __GFP_REPEAT
;
1458 timeo
= sock_sndtimeo(sk
, noblock
);
1460 err
= sock_error(sk
);
1465 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1468 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1469 skb
= alloc_skb(header_len
, gfp_mask
);
1474 /* No pages, we're done... */
1478 npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1479 skb
->truesize
+= data_len
;
1480 skb_shinfo(skb
)->nr_frags
= npages
;
1481 for (i
= 0; i
< npages
; i
++) {
1485 page
= alloc_pages(sk
->sk_allocation
, 0);
1488 skb_shinfo(skb
)->nr_frags
= i
;
1493 frag
= &skb_shinfo(skb
)->frags
[i
];
1495 frag
->page_offset
= 0;
1496 frag
->size
= (data_len
>= PAGE_SIZE
?
1499 data_len
-= PAGE_SIZE
;
1502 /* Full success... */
1508 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1509 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1513 if (signal_pending(current
))
1515 timeo
= sock_wait_for_wmem(sk
, timeo
);
1518 skb_set_owner_w(skb
, sk
);
1522 err
= sock_intr_errno(timeo
);
1527 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1529 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1530 int noblock
, int *errcode
)
1532 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1534 EXPORT_SYMBOL(sock_alloc_send_skb
);
1536 static void __lock_sock(struct sock
*sk
)
1541 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1542 TASK_UNINTERRUPTIBLE
);
1543 spin_unlock_bh(&sk
->sk_lock
.slock
);
1545 spin_lock_bh(&sk
->sk_lock
.slock
);
1546 if (!sock_owned_by_user(sk
))
1549 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1552 static void __release_sock(struct sock
*sk
)
1554 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1557 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1561 struct sk_buff
*next
= skb
->next
;
1563 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1565 sk_backlog_rcv(sk
, skb
);
1568 * We are in process context here with softirqs
1569 * disabled, use cond_resched_softirq() to preempt.
1570 * This is safe to do because we've taken the backlog
1573 cond_resched_softirq();
1576 } while (skb
!= NULL
);
1579 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1582 * Doing the zeroing here guarantee we can not loop forever
1583 * while a wild producer attempts to flood us.
1585 sk
->sk_backlog
.len
= 0;
1589 * sk_wait_data - wait for data to arrive at sk_receive_queue
1590 * @sk: sock to wait on
1591 * @timeo: for how long
1593 * Now socket state including sk->sk_err is changed only under lock,
1594 * hence we may omit checks after joining wait queue.
1595 * We check receive queue before schedule() only as optimization;
1596 * it is very likely that release_sock() added new data.
1598 int sk_wait_data(struct sock
*sk
, long *timeo
)
1603 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1604 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1605 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1606 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1607 finish_wait(sk_sleep(sk
), &wait
);
1610 EXPORT_SYMBOL(sk_wait_data
);
1613 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1615 * @size: memory size to allocate
1616 * @kind: allocation type
1618 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1619 * rmem allocation. This function assumes that protocols which have
1620 * memory_pressure use sk_wmem_queued as write buffer accounting.
1622 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
1624 struct proto
*prot
= sk
->sk_prot
;
1625 int amt
= sk_mem_pages(size
);
1628 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
1629 allocated
= atomic_add_return(amt
, prot
->memory_allocated
);
1632 if (allocated
<= prot
->sysctl_mem
[0]) {
1633 if (prot
->memory_pressure
&& *prot
->memory_pressure
)
1634 *prot
->memory_pressure
= 0;
1638 /* Under pressure. */
1639 if (allocated
> prot
->sysctl_mem
[1])
1640 if (prot
->enter_memory_pressure
)
1641 prot
->enter_memory_pressure(sk
);
1643 /* Over hard limit. */
1644 if (allocated
> prot
->sysctl_mem
[2])
1645 goto suppress_allocation
;
1647 /* guarantee minimum buffer size under pressure */
1648 if (kind
== SK_MEM_RECV
) {
1649 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
1651 } else { /* SK_MEM_SEND */
1652 if (sk
->sk_type
== SOCK_STREAM
) {
1653 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
1655 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
1656 prot
->sysctl_wmem
[0])
1660 if (prot
->memory_pressure
) {
1663 if (!*prot
->memory_pressure
)
1665 alloc
= percpu_counter_read_positive(prot
->sockets_allocated
);
1666 if (prot
->sysctl_mem
[2] > alloc
*
1667 sk_mem_pages(sk
->sk_wmem_queued
+
1668 atomic_read(&sk
->sk_rmem_alloc
) +
1669 sk
->sk_forward_alloc
))
1673 suppress_allocation
:
1675 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
1676 sk_stream_moderate_sndbuf(sk
);
1678 /* Fail only if socket is _under_ its sndbuf.
1679 * In this case we cannot block, so that we have to fail.
1681 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
1685 /* Alas. Undo changes. */
1686 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
1687 atomic_sub(amt
, prot
->memory_allocated
);
1690 EXPORT_SYMBOL(__sk_mem_schedule
);
1693 * __sk_reclaim - reclaim memory_allocated
1696 void __sk_mem_reclaim(struct sock
*sk
)
1698 struct proto
*prot
= sk
->sk_prot
;
1700 atomic_sub(sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
,
1701 prot
->memory_allocated
);
1702 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
1704 if (prot
->memory_pressure
&& *prot
->memory_pressure
&&
1705 (atomic_read(prot
->memory_allocated
) < prot
->sysctl_mem
[0]))
1706 *prot
->memory_pressure
= 0;
1708 EXPORT_SYMBOL(__sk_mem_reclaim
);
1712 * Set of default routines for initialising struct proto_ops when
1713 * the protocol does not support a particular function. In certain
1714 * cases where it makes no sense for a protocol to have a "do nothing"
1715 * function, some default processing is provided.
1718 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1722 EXPORT_SYMBOL(sock_no_bind
);
1724 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1729 EXPORT_SYMBOL(sock_no_connect
);
1731 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1735 EXPORT_SYMBOL(sock_no_socketpair
);
1737 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1741 EXPORT_SYMBOL(sock_no_accept
);
1743 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1748 EXPORT_SYMBOL(sock_no_getname
);
1750 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
1754 EXPORT_SYMBOL(sock_no_poll
);
1756 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1760 EXPORT_SYMBOL(sock_no_ioctl
);
1762 int sock_no_listen(struct socket
*sock
, int backlog
)
1766 EXPORT_SYMBOL(sock_no_listen
);
1768 int sock_no_shutdown(struct socket
*sock
, int how
)
1772 EXPORT_SYMBOL(sock_no_shutdown
);
1774 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
1775 char __user
*optval
, unsigned int optlen
)
1779 EXPORT_SYMBOL(sock_no_setsockopt
);
1781 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
1782 char __user
*optval
, int __user
*optlen
)
1786 EXPORT_SYMBOL(sock_no_getsockopt
);
1788 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1793 EXPORT_SYMBOL(sock_no_sendmsg
);
1795 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1796 size_t len
, int flags
)
1800 EXPORT_SYMBOL(sock_no_recvmsg
);
1802 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
1804 /* Mirror missing mmap method error code */
1807 EXPORT_SYMBOL(sock_no_mmap
);
1809 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
1812 struct msghdr msg
= {.msg_flags
= flags
};
1814 char *kaddr
= kmap(page
);
1815 iov
.iov_base
= kaddr
+ offset
;
1817 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1821 EXPORT_SYMBOL(sock_no_sendpage
);
1824 * Default Socket Callbacks
1827 static void sock_def_wakeup(struct sock
*sk
)
1829 struct socket_wq
*wq
;
1832 wq
= rcu_dereference(sk
->sk_wq
);
1833 if (wq_has_sleeper(wq
))
1834 wake_up_interruptible_all(&wq
->wait
);
1838 static void sock_def_error_report(struct sock
*sk
)
1840 struct socket_wq
*wq
;
1843 wq
= rcu_dereference(sk
->sk_wq
);
1844 if (wq_has_sleeper(wq
))
1845 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
1846 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
1850 static void sock_def_readable(struct sock
*sk
, int len
)
1852 struct socket_wq
*wq
;
1855 wq
= rcu_dereference(sk
->sk_wq
);
1856 if (wq_has_sleeper(wq
))
1857 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
|
1858 POLLRDNORM
| POLLRDBAND
);
1859 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
1863 static void sock_def_write_space(struct sock
*sk
)
1865 struct socket_wq
*wq
;
1869 /* Do not wake up a writer until he can make "significant"
1872 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
1873 wq
= rcu_dereference(sk
->sk_wq
);
1874 if (wq_has_sleeper(wq
))
1875 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
1876 POLLWRNORM
| POLLWRBAND
);
1878 /* Should agree with poll, otherwise some programs break */
1879 if (sock_writeable(sk
))
1880 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
1886 static void sock_def_destruct(struct sock
*sk
)
1888 kfree(sk
->sk_protinfo
);
1891 void sk_send_sigurg(struct sock
*sk
)
1893 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
1894 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
1895 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
1897 EXPORT_SYMBOL(sk_send_sigurg
);
1899 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
1900 unsigned long expires
)
1902 if (!mod_timer(timer
, expires
))
1905 EXPORT_SYMBOL(sk_reset_timer
);
1907 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
1909 if (timer_pending(timer
) && del_timer(timer
))
1912 EXPORT_SYMBOL(sk_stop_timer
);
1914 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
1916 skb_queue_head_init(&sk
->sk_receive_queue
);
1917 skb_queue_head_init(&sk
->sk_write_queue
);
1918 skb_queue_head_init(&sk
->sk_error_queue
);
1919 #ifdef CONFIG_NET_DMA
1920 skb_queue_head_init(&sk
->sk_async_wait_queue
);
1923 sk
->sk_send_head
= NULL
;
1925 init_timer(&sk
->sk_timer
);
1927 sk
->sk_allocation
= GFP_KERNEL
;
1928 sk
->sk_rcvbuf
= sysctl_rmem_default
;
1929 sk
->sk_sndbuf
= sysctl_wmem_default
;
1930 sk
->sk_state
= TCP_CLOSE
;
1931 sk_set_socket(sk
, sock
);
1933 sock_set_flag(sk
, SOCK_ZAPPED
);
1936 sk
->sk_type
= sock
->type
;
1937 sk
->sk_wq
= sock
->wq
;
1942 spin_lock_init(&sk
->sk_dst_lock
);
1943 rwlock_init(&sk
->sk_callback_lock
);
1944 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
1945 af_callback_keys
+ sk
->sk_family
,
1946 af_family_clock_key_strings
[sk
->sk_family
]);
1948 sk
->sk_state_change
= sock_def_wakeup
;
1949 sk
->sk_data_ready
= sock_def_readable
;
1950 sk
->sk_write_space
= sock_def_write_space
;
1951 sk
->sk_error_report
= sock_def_error_report
;
1952 sk
->sk_destruct
= sock_def_destruct
;
1954 sk
->sk_sndmsg_page
= NULL
;
1955 sk
->sk_sndmsg_off
= 0;
1957 sk
->sk_peercred
.pid
= 0;
1958 sk
->sk_peercred
.uid
= -1;
1959 sk
->sk_peercred
.gid
= -1;
1960 sk
->sk_write_pending
= 0;
1961 sk
->sk_rcvlowat
= 1;
1962 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
1963 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
1965 sk
->sk_stamp
= ktime_set(-1L, 0);
1968 * Before updating sk_refcnt, we must commit prior changes to memory
1969 * (Documentation/RCU/rculist_nulls.txt for details)
1972 atomic_set(&sk
->sk_refcnt
, 1);
1973 atomic_set(&sk
->sk_drops
, 0);
1975 EXPORT_SYMBOL(sock_init_data
);
1977 void lock_sock_nested(struct sock
*sk
, int subclass
)
1980 spin_lock_bh(&sk
->sk_lock
.slock
);
1981 if (sk
->sk_lock
.owned
)
1983 sk
->sk_lock
.owned
= 1;
1984 spin_unlock(&sk
->sk_lock
.slock
);
1986 * The sk_lock has mutex_lock() semantics here:
1988 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
1991 EXPORT_SYMBOL(lock_sock_nested
);
1993 void release_sock(struct sock
*sk
)
1996 * The sk_lock has mutex_unlock() semantics:
1998 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2000 spin_lock_bh(&sk
->sk_lock
.slock
);
2001 if (sk
->sk_backlog
.tail
)
2003 sk
->sk_lock
.owned
= 0;
2004 if (waitqueue_active(&sk
->sk_lock
.wq
))
2005 wake_up(&sk
->sk_lock
.wq
);
2006 spin_unlock_bh(&sk
->sk_lock
.slock
);
2008 EXPORT_SYMBOL(release_sock
);
2011 * lock_sock_fast - fast version of lock_sock
2014 * This version should be used for very small section, where process wont block
2015 * return false if fast path is taken
2016 * sk_lock.slock locked, owned = 0, BH disabled
2017 * return true if slow path is taken
2018 * sk_lock.slock unlocked, owned = 1, BH enabled
2020 bool lock_sock_fast(struct sock
*sk
)
2023 spin_lock_bh(&sk
->sk_lock
.slock
);
2025 if (!sk
->sk_lock
.owned
)
2027 * Note : We must disable BH
2032 sk
->sk_lock
.owned
= 1;
2033 spin_unlock(&sk
->sk_lock
.slock
);
2035 * The sk_lock has mutex_lock() semantics here:
2037 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2041 EXPORT_SYMBOL(lock_sock_fast
);
2043 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2046 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2047 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2048 tv
= ktime_to_timeval(sk
->sk_stamp
);
2049 if (tv
.tv_sec
== -1)
2051 if (tv
.tv_sec
== 0) {
2052 sk
->sk_stamp
= ktime_get_real();
2053 tv
= ktime_to_timeval(sk
->sk_stamp
);
2055 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2057 EXPORT_SYMBOL(sock_get_timestamp
);
2059 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2062 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2063 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2064 ts
= ktime_to_timespec(sk
->sk_stamp
);
2065 if (ts
.tv_sec
== -1)
2067 if (ts
.tv_sec
== 0) {
2068 sk
->sk_stamp
= ktime_get_real();
2069 ts
= ktime_to_timespec(sk
->sk_stamp
);
2071 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2073 EXPORT_SYMBOL(sock_get_timestampns
);
2075 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2077 if (!sock_flag(sk
, flag
)) {
2078 sock_set_flag(sk
, flag
);
2080 * we just set one of the two flags which require net
2081 * time stamping, but time stamping might have been on
2082 * already because of the other one
2085 flag
== SOCK_TIMESTAMP
?
2086 SOCK_TIMESTAMPING_RX_SOFTWARE
:
2088 net_enable_timestamp();
2093 * Get a socket option on an socket.
2095 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2096 * asynchronous errors should be reported by getsockopt. We assume
2097 * this means if you specify SO_ERROR (otherwise whats the point of it).
2099 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2100 char __user
*optval
, int __user
*optlen
)
2102 struct sock
*sk
= sock
->sk
;
2104 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2106 EXPORT_SYMBOL(sock_common_getsockopt
);
2108 #ifdef CONFIG_COMPAT
2109 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2110 char __user
*optval
, int __user
*optlen
)
2112 struct sock
*sk
= sock
->sk
;
2114 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2115 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2117 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2119 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2122 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2123 struct msghdr
*msg
, size_t size
, int flags
)
2125 struct sock
*sk
= sock
->sk
;
2129 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2130 flags
& ~MSG_DONTWAIT
, &addr_len
);
2132 msg
->msg_namelen
= addr_len
;
2135 EXPORT_SYMBOL(sock_common_recvmsg
);
2138 * Set socket options on an inet socket.
2140 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2141 char __user
*optval
, unsigned int optlen
)
2143 struct sock
*sk
= sock
->sk
;
2145 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2147 EXPORT_SYMBOL(sock_common_setsockopt
);
2149 #ifdef CONFIG_COMPAT
2150 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2151 char __user
*optval
, unsigned int optlen
)
2153 struct sock
*sk
= sock
->sk
;
2155 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2156 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2158 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2160 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2163 void sk_common_release(struct sock
*sk
)
2165 if (sk
->sk_prot
->destroy
)
2166 sk
->sk_prot
->destroy(sk
);
2169 * Observation: when sock_common_release is called, processes have
2170 * no access to socket. But net still has.
2171 * Step one, detach it from networking:
2173 * A. Remove from hash tables.
2176 sk
->sk_prot
->unhash(sk
);
2179 * In this point socket cannot receive new packets, but it is possible
2180 * that some packets are in flight because some CPU runs receiver and
2181 * did hash table lookup before we unhashed socket. They will achieve
2182 * receive queue and will be purged by socket destructor.
2184 * Also we still have packets pending on receive queue and probably,
2185 * our own packets waiting in device queues. sock_destroy will drain
2186 * receive queue, but transmitted packets will delay socket destruction
2187 * until the last reference will be released.
2192 xfrm_sk_free_policy(sk
);
2194 sk_refcnt_debug_release(sk
);
2197 EXPORT_SYMBOL(sk_common_release
);
2199 static DEFINE_RWLOCK(proto_list_lock
);
2200 static LIST_HEAD(proto_list
);
2202 #ifdef CONFIG_PROC_FS
2203 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2205 int val
[PROTO_INUSE_NR
];
2208 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2210 #ifdef CONFIG_NET_NS
2211 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2213 int cpu
= smp_processor_id();
2214 per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[prot
->inuse_idx
] += val
;
2216 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2218 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2220 int cpu
, idx
= prot
->inuse_idx
;
2223 for_each_possible_cpu(cpu
)
2224 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2226 return res
>= 0 ? res
: 0;
2228 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2230 static int __net_init
sock_inuse_init_net(struct net
*net
)
2232 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2233 return net
->core
.inuse
? 0 : -ENOMEM
;
2236 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2238 free_percpu(net
->core
.inuse
);
2241 static struct pernet_operations net_inuse_ops
= {
2242 .init
= sock_inuse_init_net
,
2243 .exit
= sock_inuse_exit_net
,
2246 static __init
int net_inuse_init(void)
2248 if (register_pernet_subsys(&net_inuse_ops
))
2249 panic("Cannot initialize net inuse counters");
2254 core_initcall(net_inuse_init
);
2256 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2258 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2260 __get_cpu_var(prot_inuse
).val
[prot
->inuse_idx
] += val
;
2262 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2264 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2266 int cpu
, idx
= prot
->inuse_idx
;
2269 for_each_possible_cpu(cpu
)
2270 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2272 return res
>= 0 ? res
: 0;
2274 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2277 static void assign_proto_idx(struct proto
*prot
)
2279 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2281 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2282 printk(KERN_ERR
"PROTO_INUSE_NR exhausted\n");
2286 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2289 static void release_proto_idx(struct proto
*prot
)
2291 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2292 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2295 static inline void assign_proto_idx(struct proto
*prot
)
2299 static inline void release_proto_idx(struct proto
*prot
)
2304 int proto_register(struct proto
*prot
, int alloc_slab
)
2307 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2308 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2311 if (prot
->slab
== NULL
) {
2312 printk(KERN_CRIT
"%s: Can't create sock SLAB cache!\n",
2317 if (prot
->rsk_prot
!= NULL
) {
2318 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2319 if (prot
->rsk_prot
->slab_name
== NULL
)
2320 goto out_free_sock_slab
;
2322 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2323 prot
->rsk_prot
->obj_size
, 0,
2324 SLAB_HWCACHE_ALIGN
, NULL
);
2326 if (prot
->rsk_prot
->slab
== NULL
) {
2327 printk(KERN_CRIT
"%s: Can't create request sock SLAB cache!\n",
2329 goto out_free_request_sock_slab_name
;
2333 if (prot
->twsk_prot
!= NULL
) {
2334 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2336 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2337 goto out_free_request_sock_slab
;
2339 prot
->twsk_prot
->twsk_slab
=
2340 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2341 prot
->twsk_prot
->twsk_obj_size
,
2343 SLAB_HWCACHE_ALIGN
|
2346 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2347 goto out_free_timewait_sock_slab_name
;
2351 write_lock(&proto_list_lock
);
2352 list_add(&prot
->node
, &proto_list
);
2353 assign_proto_idx(prot
);
2354 write_unlock(&proto_list_lock
);
2357 out_free_timewait_sock_slab_name
:
2358 kfree(prot
->twsk_prot
->twsk_slab_name
);
2359 out_free_request_sock_slab
:
2360 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2361 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2362 prot
->rsk_prot
->slab
= NULL
;
2364 out_free_request_sock_slab_name
:
2366 kfree(prot
->rsk_prot
->slab_name
);
2368 kmem_cache_destroy(prot
->slab
);
2373 EXPORT_SYMBOL(proto_register
);
2375 void proto_unregister(struct proto
*prot
)
2377 write_lock(&proto_list_lock
);
2378 release_proto_idx(prot
);
2379 list_del(&prot
->node
);
2380 write_unlock(&proto_list_lock
);
2382 if (prot
->slab
!= NULL
) {
2383 kmem_cache_destroy(prot
->slab
);
2387 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2388 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2389 kfree(prot
->rsk_prot
->slab_name
);
2390 prot
->rsk_prot
->slab
= NULL
;
2393 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2394 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2395 kfree(prot
->twsk_prot
->twsk_slab_name
);
2396 prot
->twsk_prot
->twsk_slab
= NULL
;
2399 EXPORT_SYMBOL(proto_unregister
);
2401 #ifdef CONFIG_PROC_FS
2402 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2403 __acquires(proto_list_lock
)
2405 read_lock(&proto_list_lock
);
2406 return seq_list_start_head(&proto_list
, *pos
);
2409 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2411 return seq_list_next(v
, &proto_list
, pos
);
2414 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2415 __releases(proto_list_lock
)
2417 read_unlock(&proto_list_lock
);
2420 static char proto_method_implemented(const void *method
)
2422 return method
== NULL
? 'n' : 'y';
2425 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2427 seq_printf(seq
, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2428 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2431 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2432 proto
->memory_allocated
!= NULL
? atomic_read(proto
->memory_allocated
) : -1,
2433 proto
->memory_pressure
!= NULL
? *proto
->memory_pressure
? "yes" : "no" : "NI",
2435 proto
->slab
== NULL
? "no" : "yes",
2436 module_name(proto
->owner
),
2437 proto_method_implemented(proto
->close
),
2438 proto_method_implemented(proto
->connect
),
2439 proto_method_implemented(proto
->disconnect
),
2440 proto_method_implemented(proto
->accept
),
2441 proto_method_implemented(proto
->ioctl
),
2442 proto_method_implemented(proto
->init
),
2443 proto_method_implemented(proto
->destroy
),
2444 proto_method_implemented(proto
->shutdown
),
2445 proto_method_implemented(proto
->setsockopt
),
2446 proto_method_implemented(proto
->getsockopt
),
2447 proto_method_implemented(proto
->sendmsg
),
2448 proto_method_implemented(proto
->recvmsg
),
2449 proto_method_implemented(proto
->sendpage
),
2450 proto_method_implemented(proto
->bind
),
2451 proto_method_implemented(proto
->backlog_rcv
),
2452 proto_method_implemented(proto
->hash
),
2453 proto_method_implemented(proto
->unhash
),
2454 proto_method_implemented(proto
->get_port
),
2455 proto_method_implemented(proto
->enter_memory_pressure
));
2458 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2460 if (v
== &proto_list
)
2461 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2470 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2472 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2476 static const struct seq_operations proto_seq_ops
= {
2477 .start
= proto_seq_start
,
2478 .next
= proto_seq_next
,
2479 .stop
= proto_seq_stop
,
2480 .show
= proto_seq_show
,
2483 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2485 return seq_open_net(inode
, file
, &proto_seq_ops
,
2486 sizeof(struct seq_net_private
));
2489 static const struct file_operations proto_seq_fops
= {
2490 .owner
= THIS_MODULE
,
2491 .open
= proto_seq_open
,
2493 .llseek
= seq_lseek
,
2494 .release
= seq_release_net
,
2497 static __net_init
int proto_init_net(struct net
*net
)
2499 if (!proc_net_fops_create(net
, "protocols", S_IRUGO
, &proto_seq_fops
))
2505 static __net_exit
void proto_exit_net(struct net
*net
)
2507 proc_net_remove(net
, "protocols");
2511 static __net_initdata
struct pernet_operations proto_net_ops
= {
2512 .init
= proto_init_net
,
2513 .exit
= proto_exit_net
,
2516 static int __init
proto_init(void)
2518 return register_pernet_subsys(&proto_net_ops
);
2521 subsys_initcall(proto_init
);
2523 #endif /* PROC_FS */