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>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
117 #include <asm/uaccess.h>
118 #include <asm/system.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
132 #include <linux/filter.h>
134 #include <trace/events/sock.h>
140 static DEFINE_MUTEX(proto_list_mutex
);
141 static LIST_HEAD(proto_list
);
143 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
144 int mem_cgroup_sockets_init(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
)
149 mutex_lock(&proto_list_mutex
);
150 list_for_each_entry(proto
, &proto_list
, node
) {
151 if (proto
->init_cgroup
) {
152 ret
= proto
->init_cgroup(cgrp
, ss
);
158 mutex_unlock(&proto_list_mutex
);
161 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
162 if (proto
->destroy_cgroup
)
163 proto
->destroy_cgroup(cgrp
);
164 mutex_unlock(&proto_list_mutex
);
168 void mem_cgroup_sockets_destroy(struct cgroup
*cgrp
)
172 mutex_lock(&proto_list_mutex
);
173 list_for_each_entry_reverse(proto
, &proto_list
, node
)
174 if (proto
->destroy_cgroup
)
175 proto
->destroy_cgroup(cgrp
);
176 mutex_unlock(&proto_list_mutex
);
181 * Each address family might have different locking rules, so we have
182 * one slock key per address family:
184 static struct lock_class_key af_family_keys
[AF_MAX
];
185 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
187 struct static_key memcg_socket_limit_enabled
;
188 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
191 * Make lock validator output more readable. (we pre-construct these
192 * strings build-time, so that runtime initialization of socket
195 static const char *const af_family_key_strings
[AF_MAX
+1] = {
196 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
197 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
198 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
199 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
200 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
201 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
202 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
203 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
204 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
205 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
206 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
207 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
208 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
209 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
211 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
212 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
213 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
214 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
215 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
216 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
217 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
218 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
219 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
220 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
221 "slock-27" , "slock-28" , "slock-AF_CAN" ,
222 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
223 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
224 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
225 "slock-AF_NFC" , "slock-AF_MAX"
227 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
228 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
229 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
230 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
231 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
232 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
233 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
234 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
235 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
236 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
237 "clock-27" , "clock-28" , "clock-AF_CAN" ,
238 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
239 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
240 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
241 "clock-AF_NFC" , "clock-AF_MAX"
245 * sk_callback_lock locking rules are per-address-family,
246 * so split the lock classes by using a per-AF key:
248 static struct lock_class_key af_callback_keys
[AF_MAX
];
250 /* Take into consideration the size of the struct sk_buff overhead in the
251 * determination of these values, since that is non-constant across
252 * platforms. This makes socket queueing behavior and performance
253 * not depend upon such differences.
255 #define _SK_MEM_PACKETS 256
256 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
257 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
258 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 /* Run time adjustable parameters. */
261 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
262 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
263 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
264 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
266 /* Maximal space eaten by iovec or ancillary data plus some space */
267 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
268 EXPORT_SYMBOL(sysctl_optmem_max
);
270 #if defined(CONFIG_CGROUPS)
271 #if !defined(CONFIG_NET_CLS_CGROUP)
272 int net_cls_subsys_id
= -1;
273 EXPORT_SYMBOL_GPL(net_cls_subsys_id
);
275 #if !defined(CONFIG_NETPRIO_CGROUP)
276 int net_prio_subsys_id
= -1;
277 EXPORT_SYMBOL_GPL(net_prio_subsys_id
);
281 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
285 if (optlen
< sizeof(tv
))
287 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
289 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
293 static int warned __read_mostly
;
296 if (warned
< 10 && net_ratelimit()) {
298 printk(KERN_INFO
"sock_set_timeout: `%s' (pid %d) "
299 "tries to set negative timeout\n",
300 current
->comm
, task_pid_nr(current
));
304 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
305 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
307 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
308 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
312 static void sock_warn_obsolete_bsdism(const char *name
)
315 static char warncomm
[TASK_COMM_LEN
];
316 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
317 strcpy(warncomm
, current
->comm
);
318 printk(KERN_WARNING
"process `%s' is using obsolete "
319 "%s SO_BSDCOMPAT\n", warncomm
, name
);
324 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
326 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
328 if (sk
->sk_flags
& flags
) {
329 sk
->sk_flags
&= ~flags
;
330 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
331 net_disable_timestamp();
336 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
341 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
343 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
344 atomic_inc(&sk
->sk_drops
);
345 trace_sock_rcvqueue_full(sk
, skb
);
349 err
= sk_filter(sk
, skb
);
353 if (!sk_rmem_schedule(sk
, skb
->truesize
)) {
354 atomic_inc(&sk
->sk_drops
);
359 skb_set_owner_r(skb
, sk
);
361 /* Cache the SKB length before we tack it onto the receive
362 * queue. Once it is added it no longer belongs to us and
363 * may be freed by other threads of control pulling packets
368 /* we escape from rcu protected region, make sure we dont leak
373 spin_lock_irqsave(&list
->lock
, flags
);
374 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
375 __skb_queue_tail(list
, skb
);
376 spin_unlock_irqrestore(&list
->lock
, flags
);
378 if (!sock_flag(sk
, SOCK_DEAD
))
379 sk
->sk_data_ready(sk
, skb_len
);
382 EXPORT_SYMBOL(sock_queue_rcv_skb
);
384 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
386 int rc
= NET_RX_SUCCESS
;
388 if (sk_filter(sk
, skb
))
389 goto discard_and_relse
;
393 if (sk_rcvqueues_full(sk
, skb
)) {
394 atomic_inc(&sk
->sk_drops
);
395 goto discard_and_relse
;
398 bh_lock_sock_nested(sk
);
401 if (!sock_owned_by_user(sk
)) {
403 * trylock + unlock semantics:
405 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
407 rc
= sk_backlog_rcv(sk
, skb
);
409 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
410 } else if (sk_add_backlog(sk
, skb
)) {
412 atomic_inc(&sk
->sk_drops
);
413 goto discard_and_relse
;
424 EXPORT_SYMBOL(sk_receive_skb
);
426 void sk_reset_txq(struct sock
*sk
)
428 sk_tx_queue_clear(sk
);
430 EXPORT_SYMBOL(sk_reset_txq
);
432 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
434 struct dst_entry
*dst
= __sk_dst_get(sk
);
436 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
437 sk_tx_queue_clear(sk
);
438 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
445 EXPORT_SYMBOL(__sk_dst_check
);
447 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
449 struct dst_entry
*dst
= sk_dst_get(sk
);
451 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
459 EXPORT_SYMBOL(sk_dst_check
);
461 static int sock_bindtodevice(struct sock
*sk
, char __user
*optval
, int optlen
)
463 int ret
= -ENOPROTOOPT
;
464 #ifdef CONFIG_NETDEVICES
465 struct net
*net
= sock_net(sk
);
466 char devname
[IFNAMSIZ
];
471 if (!capable(CAP_NET_RAW
))
478 /* Bind this socket to a particular device like "eth0",
479 * as specified in the passed interface name. If the
480 * name is "" or the option length is zero the socket
483 if (optlen
> IFNAMSIZ
- 1)
484 optlen
= IFNAMSIZ
- 1;
485 memset(devname
, 0, sizeof(devname
));
488 if (copy_from_user(devname
, optval
, optlen
))
492 if (devname
[0] != '\0') {
493 struct net_device
*dev
;
496 dev
= dev_get_by_name_rcu(net
, devname
);
498 index
= dev
->ifindex
;
506 sk
->sk_bound_dev_if
= index
;
518 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
521 sock_set_flag(sk
, bit
);
523 sock_reset_flag(sk
, bit
);
527 * This is meant for all protocols to use and covers goings on
528 * at the socket level. Everything here is generic.
531 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
532 char __user
*optval
, unsigned int optlen
)
534 struct sock
*sk
= sock
->sk
;
541 * Options without arguments
544 if (optname
== SO_BINDTODEVICE
)
545 return sock_bindtodevice(sk
, optval
, optlen
);
547 if (optlen
< sizeof(int))
550 if (get_user(val
, (int __user
*)optval
))
553 valbool
= val
? 1 : 0;
559 if (val
&& !capable(CAP_NET_ADMIN
))
562 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
565 sk
->sk_reuse
= valbool
;
574 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
577 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
580 /* Don't error on this BSD doesn't and if you think
581 about it this is right. Otherwise apps have to
582 play 'guess the biggest size' games. RCVBUF/SNDBUF
583 are treated in BSD as hints */
585 if (val
> sysctl_wmem_max
)
586 val
= sysctl_wmem_max
;
588 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
589 if ((val
* 2) < SOCK_MIN_SNDBUF
)
590 sk
->sk_sndbuf
= SOCK_MIN_SNDBUF
;
592 sk
->sk_sndbuf
= val
* 2;
595 * Wake up sending tasks if we
598 sk
->sk_write_space(sk
);
602 if (!capable(CAP_NET_ADMIN
)) {
609 /* Don't error on this BSD doesn't and if you think
610 about it this is right. Otherwise apps have to
611 play 'guess the biggest size' games. RCVBUF/SNDBUF
612 are treated in BSD as hints */
614 if (val
> sysctl_rmem_max
)
615 val
= sysctl_rmem_max
;
617 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
619 * We double it on the way in to account for
620 * "struct sk_buff" etc. overhead. Applications
621 * assume that the SO_RCVBUF setting they make will
622 * allow that much actual data to be received on that
625 * Applications are unaware that "struct sk_buff" and
626 * other overheads allocate from the receive buffer
627 * during socket buffer allocation.
629 * And after considering the possible alternatives,
630 * returning the value we actually used in getsockopt
631 * is the most desirable behavior.
633 if ((val
* 2) < SOCK_MIN_RCVBUF
)
634 sk
->sk_rcvbuf
= SOCK_MIN_RCVBUF
;
636 sk
->sk_rcvbuf
= val
* 2;
640 if (!capable(CAP_NET_ADMIN
)) {
648 if (sk
->sk_protocol
== IPPROTO_TCP
)
649 tcp_set_keepalive(sk
, valbool
);
651 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
655 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
659 sk
->sk_no_check
= valbool
;
663 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
664 sk
->sk_priority
= val
;
670 if (optlen
< sizeof(ling
)) {
671 ret
= -EINVAL
; /* 1003.1g */
674 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
679 sock_reset_flag(sk
, SOCK_LINGER
);
681 #if (BITS_PER_LONG == 32)
682 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
683 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
686 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
687 sock_set_flag(sk
, SOCK_LINGER
);
692 sock_warn_obsolete_bsdism("setsockopt");
697 set_bit(SOCK_PASSCRED
, &sock
->flags
);
699 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
705 if (optname
== SO_TIMESTAMP
)
706 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
708 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
709 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
710 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
712 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
713 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
717 case SO_TIMESTAMPING
:
718 if (val
& ~SOF_TIMESTAMPING_MASK
) {
722 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
723 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
724 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
725 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
726 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
727 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
728 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
729 sock_enable_timestamp(sk
,
730 SOCK_TIMESTAMPING_RX_SOFTWARE
);
732 sock_disable_timestamp(sk
,
733 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
734 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
735 val
& SOF_TIMESTAMPING_SOFTWARE
);
736 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
737 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
738 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
739 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
745 sk
->sk_rcvlowat
= val
? : 1;
749 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
753 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
756 case SO_ATTACH_FILTER
:
758 if (optlen
== sizeof(struct sock_fprog
)) {
759 struct sock_fprog fprog
;
762 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
765 ret
= sk_attach_filter(&fprog
, sk
);
769 case SO_DETACH_FILTER
:
770 ret
= sk_detach_filter(sk
);
775 set_bit(SOCK_PASSSEC
, &sock
->flags
);
777 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
780 if (!capable(CAP_NET_ADMIN
))
786 /* We implement the SO_SNDLOWAT etc to
787 not be settable (1003.1g 5.3) */
789 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
793 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
797 if (sock
->ops
->set_peek_off
)
798 sock
->ops
->set_peek_off(sk
, val
);
804 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
814 EXPORT_SYMBOL(sock_setsockopt
);
817 void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
820 ucred
->pid
= pid_vnr(pid
);
821 ucred
->uid
= ucred
->gid
= -1;
823 struct user_namespace
*current_ns
= current_user_ns();
825 ucred
->uid
= user_ns_map_uid(current_ns
, cred
, cred
->euid
);
826 ucred
->gid
= user_ns_map_gid(current_ns
, cred
, cred
->egid
);
829 EXPORT_SYMBOL_GPL(cred_to_ucred
);
831 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
832 char __user
*optval
, int __user
*optlen
)
834 struct sock
*sk
= sock
->sk
;
842 int lv
= sizeof(int);
845 if (get_user(len
, optlen
))
850 memset(&v
, 0, sizeof(v
));
854 v
.val
= sock_flag(sk
, SOCK_DBG
);
858 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
862 v
.val
= !!sock_flag(sk
, SOCK_BROADCAST
);
866 v
.val
= sk
->sk_sndbuf
;
870 v
.val
= sk
->sk_rcvbuf
;
874 v
.val
= sk
->sk_reuse
;
878 v
.val
= !!sock_flag(sk
, SOCK_KEEPOPEN
);
886 v
.val
= sk
->sk_protocol
;
890 v
.val
= sk
->sk_family
;
894 v
.val
= -sock_error(sk
);
896 v
.val
= xchg(&sk
->sk_err_soft
, 0);
900 v
.val
= !!sock_flag(sk
, SOCK_URGINLINE
);
904 v
.val
= sk
->sk_no_check
;
908 v
.val
= sk
->sk_priority
;
913 v
.ling
.l_onoff
= !!sock_flag(sk
, SOCK_LINGER
);
914 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
918 sock_warn_obsolete_bsdism("getsockopt");
922 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
923 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
927 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
930 case SO_TIMESTAMPING
:
932 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
933 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
934 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
935 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
936 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
937 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
938 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
939 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
940 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
941 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
942 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
943 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
944 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
945 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
949 lv
= sizeof(struct timeval
);
950 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
954 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
955 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
960 lv
= sizeof(struct timeval
);
961 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
965 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
966 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
971 v
.val
= sk
->sk_rcvlowat
;
979 v
.val
= test_bit(SOCK_PASSCRED
, &sock
->flags
) ? 1 : 0;
984 struct ucred peercred
;
985 if (len
> sizeof(peercred
))
986 len
= sizeof(peercred
);
987 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
988 if (copy_to_user(optval
, &peercred
, len
))
997 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1001 if (copy_to_user(optval
, address
, len
))
1006 /* Dubious BSD thing... Probably nobody even uses it, but
1007 * the UNIX standard wants it for whatever reason... -DaveM
1010 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1014 v
.val
= test_bit(SOCK_PASSSEC
, &sock
->flags
) ? 1 : 0;
1018 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1021 v
.val
= sk
->sk_mark
;
1025 v
.val
= !!sock_flag(sk
, SOCK_RXQ_OVFL
);
1028 case SO_WIFI_STATUS
:
1029 v
.val
= !!sock_flag(sk
, SOCK_WIFI_STATUS
);
1033 if (!sock
->ops
->set_peek_off
)
1036 v
.val
= sk
->sk_peek_off
;
1039 v
.val
= !!sock_flag(sk
, SOCK_NOFCS
);
1042 return -ENOPROTOOPT
;
1047 if (copy_to_user(optval
, &v
, len
))
1050 if (put_user(len
, optlen
))
1056 * Initialize an sk_lock.
1058 * (We also register the sk_lock with the lock validator.)
1060 static inline void sock_lock_init(struct sock
*sk
)
1062 sock_lock_init_class_and_name(sk
,
1063 af_family_slock_key_strings
[sk
->sk_family
],
1064 af_family_slock_keys
+ sk
->sk_family
,
1065 af_family_key_strings
[sk
->sk_family
],
1066 af_family_keys
+ sk
->sk_family
);
1070 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1071 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1072 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1074 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1076 #ifdef CONFIG_SECURITY_NETWORK
1077 void *sptr
= nsk
->sk_security
;
1079 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1081 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1082 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1084 #ifdef CONFIG_SECURITY_NETWORK
1085 nsk
->sk_security
= sptr
;
1086 security_sk_clone(osk
, nsk
);
1091 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1092 * un-modified. Special care is taken when initializing object to zero.
1094 static inline void sk_prot_clear_nulls(struct sock
*sk
, int size
)
1096 if (offsetof(struct sock
, sk_node
.next
) != 0)
1097 memset(sk
, 0, offsetof(struct sock
, sk_node
.next
));
1098 memset(&sk
->sk_node
.pprev
, 0,
1099 size
- offsetof(struct sock
, sk_node
.pprev
));
1102 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1104 unsigned long nulls1
, nulls2
;
1106 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1107 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1108 if (nulls1
> nulls2
)
1109 swap(nulls1
, nulls2
);
1112 memset((char *)sk
, 0, nulls1
);
1113 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1114 nulls2
- nulls1
- sizeof(void *));
1115 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1116 size
- nulls2
- sizeof(void *));
1118 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1120 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1124 struct kmem_cache
*slab
;
1128 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1131 if (priority
& __GFP_ZERO
) {
1133 prot
->clear_sk(sk
, prot
->obj_size
);
1135 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1138 sk
= kmalloc(prot
->obj_size
, priority
);
1141 kmemcheck_annotate_bitfield(sk
, flags
);
1143 if (security_sk_alloc(sk
, family
, priority
))
1146 if (!try_module_get(prot
->owner
))
1148 sk_tx_queue_clear(sk
);
1154 security_sk_free(sk
);
1157 kmem_cache_free(slab
, sk
);
1163 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1165 struct kmem_cache
*slab
;
1166 struct module
*owner
;
1168 owner
= prot
->owner
;
1171 security_sk_free(sk
);
1173 kmem_cache_free(slab
, sk
);
1179 #ifdef CONFIG_CGROUPS
1180 void sock_update_classid(struct sock
*sk
)
1184 rcu_read_lock(); /* doing current task, which cannot vanish. */
1185 classid
= task_cls_classid(current
);
1187 if (classid
&& classid
!= sk
->sk_classid
)
1188 sk
->sk_classid
= classid
;
1190 EXPORT_SYMBOL(sock_update_classid
);
1192 void sock_update_netprioidx(struct sock
*sk
)
1197 sk
->sk_cgrp_prioidx
= task_netprioidx(current
);
1199 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1203 * sk_alloc - All socket objects are allocated here
1204 * @net: the applicable net namespace
1205 * @family: protocol family
1206 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1207 * @prot: struct proto associated with this new sock instance
1209 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1214 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1216 sk
->sk_family
= family
;
1218 * See comment in struct sock definition to understand
1219 * why we need sk_prot_creator -acme
1221 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1223 sock_net_set(sk
, get_net(net
));
1224 atomic_set(&sk
->sk_wmem_alloc
, 1);
1226 sock_update_classid(sk
);
1227 sock_update_netprioidx(sk
);
1232 EXPORT_SYMBOL(sk_alloc
);
1234 static void __sk_free(struct sock
*sk
)
1236 struct sk_filter
*filter
;
1238 if (sk
->sk_destruct
)
1239 sk
->sk_destruct(sk
);
1241 filter
= rcu_dereference_check(sk
->sk_filter
,
1242 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1244 sk_filter_uncharge(sk
, filter
);
1245 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1248 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1250 if (atomic_read(&sk
->sk_omem_alloc
))
1251 printk(KERN_DEBUG
"%s: optmem leakage (%d bytes) detected.\n",
1252 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1254 if (sk
->sk_peer_cred
)
1255 put_cred(sk
->sk_peer_cred
);
1256 put_pid(sk
->sk_peer_pid
);
1257 put_net(sock_net(sk
));
1258 sk_prot_free(sk
->sk_prot_creator
, sk
);
1261 void sk_free(struct sock
*sk
)
1264 * We subtract one from sk_wmem_alloc and can know if
1265 * some packets are still in some tx queue.
1266 * If not null, sock_wfree() will call __sk_free(sk) later
1268 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1271 EXPORT_SYMBOL(sk_free
);
1274 * Last sock_put should drop reference to sk->sk_net. It has already
1275 * been dropped in sk_change_net. Taking reference to stopping namespace
1277 * Take reference to a socket to remove it from hash _alive_ and after that
1278 * destroy it in the context of init_net.
1280 void sk_release_kernel(struct sock
*sk
)
1282 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1286 sock_release(sk
->sk_socket
);
1287 release_net(sock_net(sk
));
1288 sock_net_set(sk
, get_net(&init_net
));
1291 EXPORT_SYMBOL(sk_release_kernel
);
1293 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1295 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1296 sock_update_memcg(newsk
);
1300 * sk_clone_lock - clone a socket, and lock its clone
1301 * @sk: the socket to clone
1302 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1304 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1306 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1310 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1311 if (newsk
!= NULL
) {
1312 struct sk_filter
*filter
;
1314 sock_copy(newsk
, sk
);
1317 get_net(sock_net(newsk
));
1318 sk_node_init(&newsk
->sk_node
);
1319 sock_lock_init(newsk
);
1320 bh_lock_sock(newsk
);
1321 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1322 newsk
->sk_backlog
.len
= 0;
1324 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1326 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1328 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1329 atomic_set(&newsk
->sk_omem_alloc
, 0);
1330 skb_queue_head_init(&newsk
->sk_receive_queue
);
1331 skb_queue_head_init(&newsk
->sk_write_queue
);
1332 #ifdef CONFIG_NET_DMA
1333 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1336 spin_lock_init(&newsk
->sk_dst_lock
);
1337 rwlock_init(&newsk
->sk_callback_lock
);
1338 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1339 af_callback_keys
+ newsk
->sk_family
,
1340 af_family_clock_key_strings
[newsk
->sk_family
]);
1342 newsk
->sk_dst_cache
= NULL
;
1343 newsk
->sk_wmem_queued
= 0;
1344 newsk
->sk_forward_alloc
= 0;
1345 newsk
->sk_send_head
= NULL
;
1346 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1348 sock_reset_flag(newsk
, SOCK_DONE
);
1349 skb_queue_head_init(&newsk
->sk_error_queue
);
1351 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1353 sk_filter_charge(newsk
, filter
);
1355 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1356 /* It is still raw copy of parent, so invalidate
1357 * destructor and make plain sk_free() */
1358 newsk
->sk_destruct
= NULL
;
1359 bh_unlock_sock(newsk
);
1366 newsk
->sk_priority
= 0;
1368 * Before updating sk_refcnt, we must commit prior changes to memory
1369 * (Documentation/RCU/rculist_nulls.txt for details)
1372 atomic_set(&newsk
->sk_refcnt
, 2);
1375 * Increment the counter in the same struct proto as the master
1376 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1377 * is the same as sk->sk_prot->socks, as this field was copied
1380 * This _changes_ the previous behaviour, where
1381 * tcp_create_openreq_child always was incrementing the
1382 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1383 * to be taken into account in all callers. -acme
1385 sk_refcnt_debug_inc(newsk
);
1386 sk_set_socket(newsk
, NULL
);
1387 newsk
->sk_wq
= NULL
;
1389 sk_update_clone(sk
, newsk
);
1391 if (newsk
->sk_prot
->sockets_allocated
)
1392 sk_sockets_allocated_inc(newsk
);
1394 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1395 net_enable_timestamp();
1400 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1402 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1404 __sk_dst_set(sk
, dst
);
1405 sk
->sk_route_caps
= dst
->dev
->features
;
1406 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1407 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1408 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1409 if (sk_can_gso(sk
)) {
1410 if (dst
->header_len
) {
1411 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1413 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1414 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1418 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1420 void __init
sk_init(void)
1422 if (totalram_pages
<= 4096) {
1423 sysctl_wmem_max
= 32767;
1424 sysctl_rmem_max
= 32767;
1425 sysctl_wmem_default
= 32767;
1426 sysctl_rmem_default
= 32767;
1427 } else if (totalram_pages
>= 131072) {
1428 sysctl_wmem_max
= 131071;
1429 sysctl_rmem_max
= 131071;
1434 * Simple resource managers for sockets.
1439 * Write buffer destructor automatically called from kfree_skb.
1441 void sock_wfree(struct sk_buff
*skb
)
1443 struct sock
*sk
= skb
->sk
;
1444 unsigned int len
= skb
->truesize
;
1446 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1448 * Keep a reference on sk_wmem_alloc, this will be released
1449 * after sk_write_space() call
1451 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1452 sk
->sk_write_space(sk
);
1456 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1457 * could not do because of in-flight packets
1459 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1462 EXPORT_SYMBOL(sock_wfree
);
1465 * Read buffer destructor automatically called from kfree_skb.
1467 void sock_rfree(struct sk_buff
*skb
)
1469 struct sock
*sk
= skb
->sk
;
1470 unsigned int len
= skb
->truesize
;
1472 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1473 sk_mem_uncharge(sk
, len
);
1475 EXPORT_SYMBOL(sock_rfree
);
1478 int sock_i_uid(struct sock
*sk
)
1482 read_lock_bh(&sk
->sk_callback_lock
);
1483 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: 0;
1484 read_unlock_bh(&sk
->sk_callback_lock
);
1487 EXPORT_SYMBOL(sock_i_uid
);
1489 unsigned long sock_i_ino(struct sock
*sk
)
1493 read_lock_bh(&sk
->sk_callback_lock
);
1494 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1495 read_unlock_bh(&sk
->sk_callback_lock
);
1498 EXPORT_SYMBOL(sock_i_ino
);
1501 * Allocate a skb from the socket's send buffer.
1503 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1506 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1507 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1509 skb_set_owner_w(skb
, sk
);
1515 EXPORT_SYMBOL(sock_wmalloc
);
1518 * Allocate a skb from the socket's receive buffer.
1520 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1523 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1524 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1526 skb_set_owner_r(skb
, sk
);
1534 * Allocate a memory block from the socket's option memory buffer.
1536 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1538 if ((unsigned)size
<= sysctl_optmem_max
&&
1539 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1541 /* First do the add, to avoid the race if kmalloc
1544 atomic_add(size
, &sk
->sk_omem_alloc
);
1545 mem
= kmalloc(size
, priority
);
1548 atomic_sub(size
, &sk
->sk_omem_alloc
);
1552 EXPORT_SYMBOL(sock_kmalloc
);
1555 * Free an option memory block.
1557 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1560 atomic_sub(size
, &sk
->sk_omem_alloc
);
1562 EXPORT_SYMBOL(sock_kfree_s
);
1564 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1565 I think, these locks should be removed for datagram sockets.
1567 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1571 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1575 if (signal_pending(current
))
1577 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1578 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1579 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1581 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1585 timeo
= schedule_timeout(timeo
);
1587 finish_wait(sk_sleep(sk
), &wait
);
1593 * Generic send/receive buffer handlers
1596 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1597 unsigned long data_len
, int noblock
,
1600 struct sk_buff
*skb
;
1605 gfp_mask
= sk
->sk_allocation
;
1606 if (gfp_mask
& __GFP_WAIT
)
1607 gfp_mask
|= __GFP_REPEAT
;
1609 timeo
= sock_sndtimeo(sk
, noblock
);
1611 err
= sock_error(sk
);
1616 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1619 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1620 skb
= alloc_skb(header_len
, gfp_mask
);
1625 /* No pages, we're done... */
1629 npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1630 skb
->truesize
+= data_len
;
1631 skb_shinfo(skb
)->nr_frags
= npages
;
1632 for (i
= 0; i
< npages
; i
++) {
1635 page
= alloc_pages(sk
->sk_allocation
, 0);
1638 skb_shinfo(skb
)->nr_frags
= i
;
1643 __skb_fill_page_desc(skb
, i
,
1645 (data_len
>= PAGE_SIZE
?
1648 data_len
-= PAGE_SIZE
;
1651 /* Full success... */
1657 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1658 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1662 if (signal_pending(current
))
1664 timeo
= sock_wait_for_wmem(sk
, timeo
);
1667 skb_set_owner_w(skb
, sk
);
1671 err
= sock_intr_errno(timeo
);
1676 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1678 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1679 int noblock
, int *errcode
)
1681 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1683 EXPORT_SYMBOL(sock_alloc_send_skb
);
1685 static void __lock_sock(struct sock
*sk
)
1686 __releases(&sk
->sk_lock
.slock
)
1687 __acquires(&sk
->sk_lock
.slock
)
1692 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1693 TASK_UNINTERRUPTIBLE
);
1694 spin_unlock_bh(&sk
->sk_lock
.slock
);
1696 spin_lock_bh(&sk
->sk_lock
.slock
);
1697 if (!sock_owned_by_user(sk
))
1700 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1703 static void __release_sock(struct sock
*sk
)
1704 __releases(&sk
->sk_lock
.slock
)
1705 __acquires(&sk
->sk_lock
.slock
)
1707 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1710 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1714 struct sk_buff
*next
= skb
->next
;
1716 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1718 sk_backlog_rcv(sk
, skb
);
1721 * We are in process context here with softirqs
1722 * disabled, use cond_resched_softirq() to preempt.
1723 * This is safe to do because we've taken the backlog
1726 cond_resched_softirq();
1729 } while (skb
!= NULL
);
1732 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1735 * Doing the zeroing here guarantee we can not loop forever
1736 * while a wild producer attempts to flood us.
1738 sk
->sk_backlog
.len
= 0;
1742 * sk_wait_data - wait for data to arrive at sk_receive_queue
1743 * @sk: sock to wait on
1744 * @timeo: for how long
1746 * Now socket state including sk->sk_err is changed only under lock,
1747 * hence we may omit checks after joining wait queue.
1748 * We check receive queue before schedule() only as optimization;
1749 * it is very likely that release_sock() added new data.
1751 int sk_wait_data(struct sock
*sk
, long *timeo
)
1756 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1757 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1758 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1759 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1760 finish_wait(sk_sleep(sk
), &wait
);
1763 EXPORT_SYMBOL(sk_wait_data
);
1766 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1768 * @size: memory size to allocate
1769 * @kind: allocation type
1771 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1772 * rmem allocation. This function assumes that protocols which have
1773 * memory_pressure use sk_wmem_queued as write buffer accounting.
1775 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
1777 struct proto
*prot
= sk
->sk_prot
;
1778 int amt
= sk_mem_pages(size
);
1780 int parent_status
= UNDER_LIMIT
;
1782 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
1784 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
1787 if (parent_status
== UNDER_LIMIT
&&
1788 allocated
<= sk_prot_mem_limits(sk
, 0)) {
1789 sk_leave_memory_pressure(sk
);
1793 /* Under pressure. (we or our parents) */
1794 if ((parent_status
> SOFT_LIMIT
) ||
1795 allocated
> sk_prot_mem_limits(sk
, 1))
1796 sk_enter_memory_pressure(sk
);
1798 /* Over hard limit (we or our parents) */
1799 if ((parent_status
== OVER_LIMIT
) ||
1800 (allocated
> sk_prot_mem_limits(sk
, 2)))
1801 goto suppress_allocation
;
1803 /* guarantee minimum buffer size under pressure */
1804 if (kind
== SK_MEM_RECV
) {
1805 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
1808 } else { /* SK_MEM_SEND */
1809 if (sk
->sk_type
== SOCK_STREAM
) {
1810 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
1812 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
1813 prot
->sysctl_wmem
[0])
1817 if (sk_has_memory_pressure(sk
)) {
1820 if (!sk_under_memory_pressure(sk
))
1822 alloc
= sk_sockets_allocated_read_positive(sk
);
1823 if (sk_prot_mem_limits(sk
, 2) > alloc
*
1824 sk_mem_pages(sk
->sk_wmem_queued
+
1825 atomic_read(&sk
->sk_rmem_alloc
) +
1826 sk
->sk_forward_alloc
))
1830 suppress_allocation
:
1832 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
1833 sk_stream_moderate_sndbuf(sk
);
1835 /* Fail only if socket is _under_ its sndbuf.
1836 * In this case we cannot block, so that we have to fail.
1838 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
1842 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
1844 /* Alas. Undo changes. */
1845 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
1847 sk_memory_allocated_sub(sk
, amt
);
1851 EXPORT_SYMBOL(__sk_mem_schedule
);
1854 * __sk_reclaim - reclaim memory_allocated
1857 void __sk_mem_reclaim(struct sock
*sk
)
1859 sk_memory_allocated_sub(sk
,
1860 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
1861 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
1863 if (sk_under_memory_pressure(sk
) &&
1864 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
1865 sk_leave_memory_pressure(sk
);
1867 EXPORT_SYMBOL(__sk_mem_reclaim
);
1871 * Set of default routines for initialising struct proto_ops when
1872 * the protocol does not support a particular function. In certain
1873 * cases where it makes no sense for a protocol to have a "do nothing"
1874 * function, some default processing is provided.
1877 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1881 EXPORT_SYMBOL(sock_no_bind
);
1883 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1888 EXPORT_SYMBOL(sock_no_connect
);
1890 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1894 EXPORT_SYMBOL(sock_no_socketpair
);
1896 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1900 EXPORT_SYMBOL(sock_no_accept
);
1902 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1907 EXPORT_SYMBOL(sock_no_getname
);
1909 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
1913 EXPORT_SYMBOL(sock_no_poll
);
1915 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1919 EXPORT_SYMBOL(sock_no_ioctl
);
1921 int sock_no_listen(struct socket
*sock
, int backlog
)
1925 EXPORT_SYMBOL(sock_no_listen
);
1927 int sock_no_shutdown(struct socket
*sock
, int how
)
1931 EXPORT_SYMBOL(sock_no_shutdown
);
1933 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
1934 char __user
*optval
, unsigned int optlen
)
1938 EXPORT_SYMBOL(sock_no_setsockopt
);
1940 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
1941 char __user
*optval
, int __user
*optlen
)
1945 EXPORT_SYMBOL(sock_no_getsockopt
);
1947 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1952 EXPORT_SYMBOL(sock_no_sendmsg
);
1954 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1955 size_t len
, int flags
)
1959 EXPORT_SYMBOL(sock_no_recvmsg
);
1961 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
1963 /* Mirror missing mmap method error code */
1966 EXPORT_SYMBOL(sock_no_mmap
);
1968 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
1971 struct msghdr msg
= {.msg_flags
= flags
};
1973 char *kaddr
= kmap(page
);
1974 iov
.iov_base
= kaddr
+ offset
;
1976 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1980 EXPORT_SYMBOL(sock_no_sendpage
);
1983 * Default Socket Callbacks
1986 static void sock_def_wakeup(struct sock
*sk
)
1988 struct socket_wq
*wq
;
1991 wq
= rcu_dereference(sk
->sk_wq
);
1992 if (wq_has_sleeper(wq
))
1993 wake_up_interruptible_all(&wq
->wait
);
1997 static void sock_def_error_report(struct sock
*sk
)
1999 struct socket_wq
*wq
;
2002 wq
= rcu_dereference(sk
->sk_wq
);
2003 if (wq_has_sleeper(wq
))
2004 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2005 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2009 static void sock_def_readable(struct sock
*sk
, int len
)
2011 struct socket_wq
*wq
;
2014 wq
= rcu_dereference(sk
->sk_wq
);
2015 if (wq_has_sleeper(wq
))
2016 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2017 POLLRDNORM
| POLLRDBAND
);
2018 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2022 static void sock_def_write_space(struct sock
*sk
)
2024 struct socket_wq
*wq
;
2028 /* Do not wake up a writer until he can make "significant"
2031 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2032 wq
= rcu_dereference(sk
->sk_wq
);
2033 if (wq_has_sleeper(wq
))
2034 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2035 POLLWRNORM
| POLLWRBAND
);
2037 /* Should agree with poll, otherwise some programs break */
2038 if (sock_writeable(sk
))
2039 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2045 static void sock_def_destruct(struct sock
*sk
)
2047 kfree(sk
->sk_protinfo
);
2050 void sk_send_sigurg(struct sock
*sk
)
2052 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2053 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2054 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2056 EXPORT_SYMBOL(sk_send_sigurg
);
2058 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2059 unsigned long expires
)
2061 if (!mod_timer(timer
, expires
))
2064 EXPORT_SYMBOL(sk_reset_timer
);
2066 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2068 if (timer_pending(timer
) && del_timer(timer
))
2071 EXPORT_SYMBOL(sk_stop_timer
);
2073 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2075 skb_queue_head_init(&sk
->sk_receive_queue
);
2076 skb_queue_head_init(&sk
->sk_write_queue
);
2077 skb_queue_head_init(&sk
->sk_error_queue
);
2078 #ifdef CONFIG_NET_DMA
2079 skb_queue_head_init(&sk
->sk_async_wait_queue
);
2082 sk
->sk_send_head
= NULL
;
2084 init_timer(&sk
->sk_timer
);
2086 sk
->sk_allocation
= GFP_KERNEL
;
2087 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2088 sk
->sk_sndbuf
= sysctl_wmem_default
;
2089 sk
->sk_state
= TCP_CLOSE
;
2090 sk_set_socket(sk
, sock
);
2092 sock_set_flag(sk
, SOCK_ZAPPED
);
2095 sk
->sk_type
= sock
->type
;
2096 sk
->sk_wq
= sock
->wq
;
2101 spin_lock_init(&sk
->sk_dst_lock
);
2102 rwlock_init(&sk
->sk_callback_lock
);
2103 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2104 af_callback_keys
+ sk
->sk_family
,
2105 af_family_clock_key_strings
[sk
->sk_family
]);
2107 sk
->sk_state_change
= sock_def_wakeup
;
2108 sk
->sk_data_ready
= sock_def_readable
;
2109 sk
->sk_write_space
= sock_def_write_space
;
2110 sk
->sk_error_report
= sock_def_error_report
;
2111 sk
->sk_destruct
= sock_def_destruct
;
2113 sk
->sk_sndmsg_page
= NULL
;
2114 sk
->sk_sndmsg_off
= 0;
2115 sk
->sk_peek_off
= -1;
2117 sk
->sk_peer_pid
= NULL
;
2118 sk
->sk_peer_cred
= NULL
;
2119 sk
->sk_write_pending
= 0;
2120 sk
->sk_rcvlowat
= 1;
2121 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2122 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2124 sk
->sk_stamp
= ktime_set(-1L, 0);
2127 * Before updating sk_refcnt, we must commit prior changes to memory
2128 * (Documentation/RCU/rculist_nulls.txt for details)
2131 atomic_set(&sk
->sk_refcnt
, 1);
2132 atomic_set(&sk
->sk_drops
, 0);
2134 EXPORT_SYMBOL(sock_init_data
);
2136 void lock_sock_nested(struct sock
*sk
, int subclass
)
2139 spin_lock_bh(&sk
->sk_lock
.slock
);
2140 if (sk
->sk_lock
.owned
)
2142 sk
->sk_lock
.owned
= 1;
2143 spin_unlock(&sk
->sk_lock
.slock
);
2145 * The sk_lock has mutex_lock() semantics here:
2147 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2150 EXPORT_SYMBOL(lock_sock_nested
);
2152 void release_sock(struct sock
*sk
)
2155 * The sk_lock has mutex_unlock() semantics:
2157 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2159 spin_lock_bh(&sk
->sk_lock
.slock
);
2160 if (sk
->sk_backlog
.tail
)
2162 sk
->sk_lock
.owned
= 0;
2163 if (waitqueue_active(&sk
->sk_lock
.wq
))
2164 wake_up(&sk
->sk_lock
.wq
);
2165 spin_unlock_bh(&sk
->sk_lock
.slock
);
2167 EXPORT_SYMBOL(release_sock
);
2170 * lock_sock_fast - fast version of lock_sock
2173 * This version should be used for very small section, where process wont block
2174 * return false if fast path is taken
2175 * sk_lock.slock locked, owned = 0, BH disabled
2176 * return true if slow path is taken
2177 * sk_lock.slock unlocked, owned = 1, BH enabled
2179 bool lock_sock_fast(struct sock
*sk
)
2182 spin_lock_bh(&sk
->sk_lock
.slock
);
2184 if (!sk
->sk_lock
.owned
)
2186 * Note : We must disable BH
2191 sk
->sk_lock
.owned
= 1;
2192 spin_unlock(&sk
->sk_lock
.slock
);
2194 * The sk_lock has mutex_lock() semantics here:
2196 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2200 EXPORT_SYMBOL(lock_sock_fast
);
2202 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2205 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2206 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2207 tv
= ktime_to_timeval(sk
->sk_stamp
);
2208 if (tv
.tv_sec
== -1)
2210 if (tv
.tv_sec
== 0) {
2211 sk
->sk_stamp
= ktime_get_real();
2212 tv
= ktime_to_timeval(sk
->sk_stamp
);
2214 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2216 EXPORT_SYMBOL(sock_get_timestamp
);
2218 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2221 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2222 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2223 ts
= ktime_to_timespec(sk
->sk_stamp
);
2224 if (ts
.tv_sec
== -1)
2226 if (ts
.tv_sec
== 0) {
2227 sk
->sk_stamp
= ktime_get_real();
2228 ts
= ktime_to_timespec(sk
->sk_stamp
);
2230 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2232 EXPORT_SYMBOL(sock_get_timestampns
);
2234 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2236 if (!sock_flag(sk
, flag
)) {
2237 unsigned long previous_flags
= sk
->sk_flags
;
2239 sock_set_flag(sk
, flag
);
2241 * we just set one of the two flags which require net
2242 * time stamping, but time stamping might have been on
2243 * already because of the other one
2245 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2246 net_enable_timestamp();
2251 * Get a socket option on an socket.
2253 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2254 * asynchronous errors should be reported by getsockopt. We assume
2255 * this means if you specify SO_ERROR (otherwise whats the point of it).
2257 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2258 char __user
*optval
, int __user
*optlen
)
2260 struct sock
*sk
= sock
->sk
;
2262 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2264 EXPORT_SYMBOL(sock_common_getsockopt
);
2266 #ifdef CONFIG_COMPAT
2267 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2268 char __user
*optval
, int __user
*optlen
)
2270 struct sock
*sk
= sock
->sk
;
2272 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2273 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2275 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2277 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2280 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2281 struct msghdr
*msg
, size_t size
, int flags
)
2283 struct sock
*sk
= sock
->sk
;
2287 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2288 flags
& ~MSG_DONTWAIT
, &addr_len
);
2290 msg
->msg_namelen
= addr_len
;
2293 EXPORT_SYMBOL(sock_common_recvmsg
);
2296 * Set socket options on an inet socket.
2298 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2299 char __user
*optval
, unsigned int optlen
)
2301 struct sock
*sk
= sock
->sk
;
2303 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2305 EXPORT_SYMBOL(sock_common_setsockopt
);
2307 #ifdef CONFIG_COMPAT
2308 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2309 char __user
*optval
, unsigned int optlen
)
2311 struct sock
*sk
= sock
->sk
;
2313 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2314 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2316 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2318 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2321 void sk_common_release(struct sock
*sk
)
2323 if (sk
->sk_prot
->destroy
)
2324 sk
->sk_prot
->destroy(sk
);
2327 * Observation: when sock_common_release is called, processes have
2328 * no access to socket. But net still has.
2329 * Step one, detach it from networking:
2331 * A. Remove from hash tables.
2334 sk
->sk_prot
->unhash(sk
);
2337 * In this point socket cannot receive new packets, but it is possible
2338 * that some packets are in flight because some CPU runs receiver and
2339 * did hash table lookup before we unhashed socket. They will achieve
2340 * receive queue and will be purged by socket destructor.
2342 * Also we still have packets pending on receive queue and probably,
2343 * our own packets waiting in device queues. sock_destroy will drain
2344 * receive queue, but transmitted packets will delay socket destruction
2345 * until the last reference will be released.
2350 xfrm_sk_free_policy(sk
);
2352 sk_refcnt_debug_release(sk
);
2355 EXPORT_SYMBOL(sk_common_release
);
2357 #ifdef CONFIG_PROC_FS
2358 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2360 int val
[PROTO_INUSE_NR
];
2363 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2365 #ifdef CONFIG_NET_NS
2366 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2368 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2370 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2372 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2374 int cpu
, idx
= prot
->inuse_idx
;
2377 for_each_possible_cpu(cpu
)
2378 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2380 return res
>= 0 ? res
: 0;
2382 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2384 static int __net_init
sock_inuse_init_net(struct net
*net
)
2386 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2387 return net
->core
.inuse
? 0 : -ENOMEM
;
2390 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2392 free_percpu(net
->core
.inuse
);
2395 static struct pernet_operations net_inuse_ops
= {
2396 .init
= sock_inuse_init_net
,
2397 .exit
= sock_inuse_exit_net
,
2400 static __init
int net_inuse_init(void)
2402 if (register_pernet_subsys(&net_inuse_ops
))
2403 panic("Cannot initialize net inuse counters");
2408 core_initcall(net_inuse_init
);
2410 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2412 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2414 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2416 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2418 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2420 int cpu
, idx
= prot
->inuse_idx
;
2423 for_each_possible_cpu(cpu
)
2424 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2426 return res
>= 0 ? res
: 0;
2428 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2431 static void assign_proto_idx(struct proto
*prot
)
2433 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2435 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2436 printk(KERN_ERR
"PROTO_INUSE_NR exhausted\n");
2440 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2443 static void release_proto_idx(struct proto
*prot
)
2445 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2446 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2449 static inline void assign_proto_idx(struct proto
*prot
)
2453 static inline void release_proto_idx(struct proto
*prot
)
2458 int proto_register(struct proto
*prot
, int alloc_slab
)
2461 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2462 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2465 if (prot
->slab
== NULL
) {
2466 printk(KERN_CRIT
"%s: Can't create sock SLAB cache!\n",
2471 if (prot
->rsk_prot
!= NULL
) {
2472 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2473 if (prot
->rsk_prot
->slab_name
== NULL
)
2474 goto out_free_sock_slab
;
2476 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2477 prot
->rsk_prot
->obj_size
, 0,
2478 SLAB_HWCACHE_ALIGN
, NULL
);
2480 if (prot
->rsk_prot
->slab
== NULL
) {
2481 printk(KERN_CRIT
"%s: Can't create request sock SLAB cache!\n",
2483 goto out_free_request_sock_slab_name
;
2487 if (prot
->twsk_prot
!= NULL
) {
2488 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2490 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2491 goto out_free_request_sock_slab
;
2493 prot
->twsk_prot
->twsk_slab
=
2494 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2495 prot
->twsk_prot
->twsk_obj_size
,
2497 SLAB_HWCACHE_ALIGN
|
2500 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2501 goto out_free_timewait_sock_slab_name
;
2505 mutex_lock(&proto_list_mutex
);
2506 list_add(&prot
->node
, &proto_list
);
2507 assign_proto_idx(prot
);
2508 mutex_unlock(&proto_list_mutex
);
2511 out_free_timewait_sock_slab_name
:
2512 kfree(prot
->twsk_prot
->twsk_slab_name
);
2513 out_free_request_sock_slab
:
2514 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2515 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2516 prot
->rsk_prot
->slab
= NULL
;
2518 out_free_request_sock_slab_name
:
2520 kfree(prot
->rsk_prot
->slab_name
);
2522 kmem_cache_destroy(prot
->slab
);
2527 EXPORT_SYMBOL(proto_register
);
2529 void proto_unregister(struct proto
*prot
)
2531 mutex_lock(&proto_list_mutex
);
2532 release_proto_idx(prot
);
2533 list_del(&prot
->node
);
2534 mutex_unlock(&proto_list_mutex
);
2536 if (prot
->slab
!= NULL
) {
2537 kmem_cache_destroy(prot
->slab
);
2541 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2542 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2543 kfree(prot
->rsk_prot
->slab_name
);
2544 prot
->rsk_prot
->slab
= NULL
;
2547 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2548 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2549 kfree(prot
->twsk_prot
->twsk_slab_name
);
2550 prot
->twsk_prot
->twsk_slab
= NULL
;
2553 EXPORT_SYMBOL(proto_unregister
);
2555 #ifdef CONFIG_PROC_FS
2556 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2557 __acquires(proto_list_mutex
)
2559 mutex_lock(&proto_list_mutex
);
2560 return seq_list_start_head(&proto_list
, *pos
);
2563 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2565 return seq_list_next(v
, &proto_list
, pos
);
2568 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2569 __releases(proto_list_mutex
)
2571 mutex_unlock(&proto_list_mutex
);
2574 static char proto_method_implemented(const void *method
)
2576 return method
== NULL
? 'n' : 'y';
2578 static long sock_prot_memory_allocated(struct proto
*proto
)
2580 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
): -1L;
2583 static char *sock_prot_memory_pressure(struct proto
*proto
)
2585 return proto
->memory_pressure
!= NULL
?
2586 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2589 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2592 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2593 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2596 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2597 sock_prot_memory_allocated(proto
),
2598 sock_prot_memory_pressure(proto
),
2600 proto
->slab
== NULL
? "no" : "yes",
2601 module_name(proto
->owner
),
2602 proto_method_implemented(proto
->close
),
2603 proto_method_implemented(proto
->connect
),
2604 proto_method_implemented(proto
->disconnect
),
2605 proto_method_implemented(proto
->accept
),
2606 proto_method_implemented(proto
->ioctl
),
2607 proto_method_implemented(proto
->init
),
2608 proto_method_implemented(proto
->destroy
),
2609 proto_method_implemented(proto
->shutdown
),
2610 proto_method_implemented(proto
->setsockopt
),
2611 proto_method_implemented(proto
->getsockopt
),
2612 proto_method_implemented(proto
->sendmsg
),
2613 proto_method_implemented(proto
->recvmsg
),
2614 proto_method_implemented(proto
->sendpage
),
2615 proto_method_implemented(proto
->bind
),
2616 proto_method_implemented(proto
->backlog_rcv
),
2617 proto_method_implemented(proto
->hash
),
2618 proto_method_implemented(proto
->unhash
),
2619 proto_method_implemented(proto
->get_port
),
2620 proto_method_implemented(proto
->enter_memory_pressure
));
2623 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2625 if (v
== &proto_list
)
2626 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2635 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2637 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2641 static const struct seq_operations proto_seq_ops
= {
2642 .start
= proto_seq_start
,
2643 .next
= proto_seq_next
,
2644 .stop
= proto_seq_stop
,
2645 .show
= proto_seq_show
,
2648 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2650 return seq_open_net(inode
, file
, &proto_seq_ops
,
2651 sizeof(struct seq_net_private
));
2654 static const struct file_operations proto_seq_fops
= {
2655 .owner
= THIS_MODULE
,
2656 .open
= proto_seq_open
,
2658 .llseek
= seq_lseek
,
2659 .release
= seq_release_net
,
2662 static __net_init
int proto_init_net(struct net
*net
)
2664 if (!proc_net_fops_create(net
, "protocols", S_IRUGO
, &proto_seq_fops
))
2670 static __net_exit
void proto_exit_net(struct net
*net
)
2672 proc_net_remove(net
, "protocols");
2676 static __net_initdata
struct pernet_operations proto_net_ops
= {
2677 .init
= proto_init_net
,
2678 .exit
= proto_exit_net
,
2681 static int __init
proto_init(void)
2683 return register_pernet_subsys(&proto_net_ops
);
2686 subsys_initcall(proto_init
);
2688 #endif /* PROC_FS */