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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
115 #include <linux/user_namespace.h>
116 #include <linux/static_key.h>
117 #include <linux/memcontrol.h>
118 #include <linux/prefetch.h>
120 #include <asm/uaccess.h>
122 #include <linux/netdevice.h>
123 #include <net/protocol.h>
124 #include <linux/skbuff.h>
125 #include <net/net_namespace.h>
126 #include <net/request_sock.h>
127 #include <net/sock.h>
128 #include <linux/net_tstamp.h>
129 #include <net/xfrm.h>
130 #include <linux/ipsec.h>
131 #include <net/cls_cgroup.h>
132 #include <net/netprio_cgroup.h>
134 #include <linux/filter.h>
136 #include <trace/events/sock.h>
142 static DEFINE_MUTEX(proto_list_mutex
);
143 static LIST_HEAD(proto_list
);
145 #ifdef CONFIG_MEMCG_KMEM
146 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
151 mutex_lock(&proto_list_mutex
);
152 list_for_each_entry(proto
, &proto_list
, node
) {
153 if (proto
->init_cgroup
) {
154 ret
= proto
->init_cgroup(memcg
, ss
);
160 mutex_unlock(&proto_list_mutex
);
163 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
164 if (proto
->destroy_cgroup
)
165 proto
->destroy_cgroup(memcg
);
166 mutex_unlock(&proto_list_mutex
);
170 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
174 mutex_lock(&proto_list_mutex
);
175 list_for_each_entry_reverse(proto
, &proto_list
, node
)
176 if (proto
->destroy_cgroup
)
177 proto
->destroy_cgroup(memcg
);
178 mutex_unlock(&proto_list_mutex
);
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
186 static struct lock_class_key af_family_keys
[AF_MAX
];
187 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
189 struct static_key memcg_socket_limit_enabled
;
190 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
197 static const char *const af_family_key_strings
[AF_MAX
+1] = {
198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
213 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
227 "slock-AF_NFC" , "slock-AF_MAX"
229 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
243 "clock-AF_NFC" , "clock-AF_MAX"
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
250 static struct lock_class_key af_callback_keys
[AF_MAX
];
252 /* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
257 #define _SK_MEM_PACKETS 256
258 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
259 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
262 /* Run time adjustable parameters. */
263 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
264 EXPORT_SYMBOL(sysctl_wmem_max
);
265 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
266 EXPORT_SYMBOL(sysctl_rmem_max
);
267 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
268 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
270 /* Maximal space eaten by iovec or ancillary data plus some space */
271 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
272 EXPORT_SYMBOL(sysctl_optmem_max
);
274 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
275 EXPORT_SYMBOL_GPL(memalloc_socks
);
278 * sk_set_memalloc - sets %SOCK_MEMALLOC
279 * @sk: socket to set it on
281 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
282 * It's the responsibility of the admin to adjust min_free_kbytes
283 * to meet the requirements
285 void sk_set_memalloc(struct sock
*sk
)
287 sock_set_flag(sk
, SOCK_MEMALLOC
);
288 sk
->sk_allocation
|= __GFP_MEMALLOC
;
289 static_key_slow_inc(&memalloc_socks
);
291 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
293 void sk_clear_memalloc(struct sock
*sk
)
295 sock_reset_flag(sk
, SOCK_MEMALLOC
);
296 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
297 static_key_slow_dec(&memalloc_socks
);
300 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
301 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
302 * it has rmem allocations there is a risk that the user of the
303 * socket cannot make forward progress due to exceeding the rmem
304 * limits. By rights, sk_clear_memalloc() should only be called
305 * on sockets being torn down but warn and reset the accounting if
306 * that assumption breaks.
308 if (WARN_ON(sk
->sk_forward_alloc
))
311 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
313 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
316 unsigned long pflags
= current
->flags
;
318 /* these should have been dropped before queueing */
319 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
321 current
->flags
|= PF_MEMALLOC
;
322 ret
= sk
->sk_backlog_rcv(sk
, skb
);
323 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
327 EXPORT_SYMBOL(__sk_backlog_rcv
);
329 #if defined(CONFIG_CGROUPS)
330 #if !defined(CONFIG_NET_CLS_CGROUP)
331 int net_cls_subsys_id
= -1;
332 EXPORT_SYMBOL_GPL(net_cls_subsys_id
);
334 #if !defined(CONFIG_NETPRIO_CGROUP)
335 int net_prio_subsys_id
= -1;
336 EXPORT_SYMBOL_GPL(net_prio_subsys_id
);
340 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
344 if (optlen
< sizeof(tv
))
346 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
348 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
352 static int warned __read_mostly
;
355 if (warned
< 10 && net_ratelimit()) {
357 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
358 __func__
, current
->comm
, task_pid_nr(current
));
362 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
363 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
365 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
366 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
370 static void sock_warn_obsolete_bsdism(const char *name
)
373 static char warncomm
[TASK_COMM_LEN
];
374 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
375 strcpy(warncomm
, current
->comm
);
376 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
382 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
384 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
386 if (sk
->sk_flags
& flags
) {
387 sk
->sk_flags
&= ~flags
;
388 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
389 net_disable_timestamp();
394 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
399 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
401 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
402 atomic_inc(&sk
->sk_drops
);
403 trace_sock_rcvqueue_full(sk
, skb
);
407 err
= sk_filter(sk
, skb
);
411 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
412 atomic_inc(&sk
->sk_drops
);
417 skb_set_owner_r(skb
, sk
);
419 /* Cache the SKB length before we tack it onto the receive
420 * queue. Once it is added it no longer belongs to us and
421 * may be freed by other threads of control pulling packets
426 /* we escape from rcu protected region, make sure we dont leak
431 spin_lock_irqsave(&list
->lock
, flags
);
432 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
433 __skb_queue_tail(list
, skb
);
434 spin_unlock_irqrestore(&list
->lock
, flags
);
436 if (!sock_flag(sk
, SOCK_DEAD
))
437 sk
->sk_data_ready(sk
, skb_len
);
440 EXPORT_SYMBOL(sock_queue_rcv_skb
);
442 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
444 int rc
= NET_RX_SUCCESS
;
446 if (sk_filter(sk
, skb
))
447 goto discard_and_relse
;
451 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
)) {
452 atomic_inc(&sk
->sk_drops
);
453 goto discard_and_relse
;
456 bh_lock_sock_nested(sk
);
459 if (!sock_owned_by_user(sk
)) {
461 * trylock + unlock semantics:
463 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
465 rc
= sk_backlog_rcv(sk
, skb
);
467 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
468 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
470 atomic_inc(&sk
->sk_drops
);
471 goto discard_and_relse
;
482 EXPORT_SYMBOL(sk_receive_skb
);
484 void sk_reset_txq(struct sock
*sk
)
486 sk_tx_queue_clear(sk
);
488 EXPORT_SYMBOL(sk_reset_txq
);
490 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
492 struct dst_entry
*dst
= __sk_dst_get(sk
);
494 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
495 sk_tx_queue_clear(sk
);
496 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
503 EXPORT_SYMBOL(__sk_dst_check
);
505 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
507 struct dst_entry
*dst
= sk_dst_get(sk
);
509 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
517 EXPORT_SYMBOL(sk_dst_check
);
519 static int sock_bindtodevice(struct sock
*sk
, char __user
*optval
, int optlen
)
521 int ret
= -ENOPROTOOPT
;
522 #ifdef CONFIG_NETDEVICES
523 struct net
*net
= sock_net(sk
);
524 char devname
[IFNAMSIZ
];
529 if (!capable(CAP_NET_RAW
))
536 /* Bind this socket to a particular device like "eth0",
537 * as specified in the passed interface name. If the
538 * name is "" or the option length is zero the socket
541 if (optlen
> IFNAMSIZ
- 1)
542 optlen
= IFNAMSIZ
- 1;
543 memset(devname
, 0, sizeof(devname
));
546 if (copy_from_user(devname
, optval
, optlen
))
550 if (devname
[0] != '\0') {
551 struct net_device
*dev
;
554 dev
= dev_get_by_name_rcu(net
, devname
);
556 index
= dev
->ifindex
;
564 sk
->sk_bound_dev_if
= index
;
576 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
579 sock_set_flag(sk
, bit
);
581 sock_reset_flag(sk
, bit
);
585 * This is meant for all protocols to use and covers goings on
586 * at the socket level. Everything here is generic.
589 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
590 char __user
*optval
, unsigned int optlen
)
592 struct sock
*sk
= sock
->sk
;
599 * Options without arguments
602 if (optname
== SO_BINDTODEVICE
)
603 return sock_bindtodevice(sk
, optval
, optlen
);
605 if (optlen
< sizeof(int))
608 if (get_user(val
, (int __user
*)optval
))
611 valbool
= val
? 1 : 0;
617 if (val
&& !capable(CAP_NET_ADMIN
))
620 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
623 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
632 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
635 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
638 /* Don't error on this BSD doesn't and if you think
639 * about it this is right. Otherwise apps have to
640 * play 'guess the biggest size' games. RCVBUF/SNDBUF
641 * are treated in BSD as hints
643 val
= min_t(u32
, val
, sysctl_wmem_max
);
645 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
646 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
647 /* Wake up sending tasks if we upped the value. */
648 sk
->sk_write_space(sk
);
652 if (!capable(CAP_NET_ADMIN
)) {
659 /* Don't error on this BSD doesn't and if you think
660 * about it this is right. Otherwise apps have to
661 * play 'guess the biggest size' games. RCVBUF/SNDBUF
662 * are treated in BSD as hints
664 val
= min_t(u32
, val
, sysctl_rmem_max
);
666 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
668 * We double it on the way in to account for
669 * "struct sk_buff" etc. overhead. Applications
670 * assume that the SO_RCVBUF setting they make will
671 * allow that much actual data to be received on that
674 * Applications are unaware that "struct sk_buff" and
675 * other overheads allocate from the receive buffer
676 * during socket buffer allocation.
678 * And after considering the possible alternatives,
679 * returning the value we actually used in getsockopt
680 * is the most desirable behavior.
682 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
686 if (!capable(CAP_NET_ADMIN
)) {
694 if (sk
->sk_protocol
== IPPROTO_TCP
&&
695 sk
->sk_type
== SOCK_STREAM
)
696 tcp_set_keepalive(sk
, valbool
);
698 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
702 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
706 sk
->sk_no_check
= valbool
;
710 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
711 sk
->sk_priority
= val
;
717 if (optlen
< sizeof(ling
)) {
718 ret
= -EINVAL
; /* 1003.1g */
721 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
726 sock_reset_flag(sk
, SOCK_LINGER
);
728 #if (BITS_PER_LONG == 32)
729 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
730 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
733 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
734 sock_set_flag(sk
, SOCK_LINGER
);
739 sock_warn_obsolete_bsdism("setsockopt");
744 set_bit(SOCK_PASSCRED
, &sock
->flags
);
746 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
752 if (optname
== SO_TIMESTAMP
)
753 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
755 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
756 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
757 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
759 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
760 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
764 case SO_TIMESTAMPING
:
765 if (val
& ~SOF_TIMESTAMPING_MASK
) {
769 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
770 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
771 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
772 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
773 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
774 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
775 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
776 sock_enable_timestamp(sk
,
777 SOCK_TIMESTAMPING_RX_SOFTWARE
);
779 sock_disable_timestamp(sk
,
780 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
781 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
782 val
& SOF_TIMESTAMPING_SOFTWARE
);
783 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
784 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
785 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
786 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
792 sk
->sk_rcvlowat
= val
? : 1;
796 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
800 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
803 case SO_ATTACH_FILTER
:
805 if (optlen
== sizeof(struct sock_fprog
)) {
806 struct sock_fprog fprog
;
809 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
812 ret
= sk_attach_filter(&fprog
, sk
);
816 case SO_DETACH_FILTER
:
817 ret
= sk_detach_filter(sk
);
822 set_bit(SOCK_PASSSEC
, &sock
->flags
);
824 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
827 if (!capable(CAP_NET_ADMIN
))
833 /* We implement the SO_SNDLOWAT etc to
834 not be settable (1003.1g 5.3) */
836 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
840 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
844 if (sock
->ops
->set_peek_off
)
845 sock
->ops
->set_peek_off(sk
, val
);
851 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
861 EXPORT_SYMBOL(sock_setsockopt
);
864 void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
867 ucred
->pid
= pid_vnr(pid
);
868 ucred
->uid
= ucred
->gid
= -1;
870 struct user_namespace
*current_ns
= current_user_ns();
872 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
873 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
876 EXPORT_SYMBOL_GPL(cred_to_ucred
);
878 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
879 char __user
*optval
, int __user
*optlen
)
881 struct sock
*sk
= sock
->sk
;
889 int lv
= sizeof(int);
892 if (get_user(len
, optlen
))
897 memset(&v
, 0, sizeof(v
));
901 v
.val
= sock_flag(sk
, SOCK_DBG
);
905 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
909 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
913 v
.val
= sk
->sk_sndbuf
;
917 v
.val
= sk
->sk_rcvbuf
;
921 v
.val
= sk
->sk_reuse
;
925 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
933 v
.val
= sk
->sk_protocol
;
937 v
.val
= sk
->sk_family
;
941 v
.val
= -sock_error(sk
);
943 v
.val
= xchg(&sk
->sk_err_soft
, 0);
947 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
951 v
.val
= sk
->sk_no_check
;
955 v
.val
= sk
->sk_priority
;
960 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
961 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
965 sock_warn_obsolete_bsdism("getsockopt");
969 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
970 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
974 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
977 case SO_TIMESTAMPING
:
979 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
980 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
981 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
982 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
983 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
984 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
985 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
986 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
987 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
988 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
989 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
990 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
991 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
992 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
996 lv
= sizeof(struct timeval
);
997 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1001 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1002 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1007 lv
= sizeof(struct timeval
);
1008 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1012 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1013 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1018 v
.val
= sk
->sk_rcvlowat
;
1026 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1031 struct ucred peercred
;
1032 if (len
> sizeof(peercred
))
1033 len
= sizeof(peercred
);
1034 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1035 if (copy_to_user(optval
, &peercred
, len
))
1044 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1048 if (copy_to_user(optval
, address
, len
))
1053 /* Dubious BSD thing... Probably nobody even uses it, but
1054 * the UNIX standard wants it for whatever reason... -DaveM
1057 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1061 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1065 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1068 v
.val
= sk
->sk_mark
;
1072 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1075 case SO_WIFI_STATUS
:
1076 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1080 if (!sock
->ops
->set_peek_off
)
1083 v
.val
= sk
->sk_peek_off
;
1086 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1089 return -ENOPROTOOPT
;
1094 if (copy_to_user(optval
, &v
, len
))
1097 if (put_user(len
, optlen
))
1103 * Initialize an sk_lock.
1105 * (We also register the sk_lock with the lock validator.)
1107 static inline void sock_lock_init(struct sock
*sk
)
1109 sock_lock_init_class_and_name(sk
,
1110 af_family_slock_key_strings
[sk
->sk_family
],
1111 af_family_slock_keys
+ sk
->sk_family
,
1112 af_family_key_strings
[sk
->sk_family
],
1113 af_family_keys
+ sk
->sk_family
);
1117 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1118 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1119 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1121 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1123 #ifdef CONFIG_SECURITY_NETWORK
1124 void *sptr
= nsk
->sk_security
;
1126 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1128 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1129 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1131 #ifdef CONFIG_SECURITY_NETWORK
1132 nsk
->sk_security
= sptr
;
1133 security_sk_clone(osk
, nsk
);
1138 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1139 * un-modified. Special care is taken when initializing object to zero.
1141 static inline void sk_prot_clear_nulls(struct sock
*sk
, int size
)
1143 if (offsetof(struct sock
, sk_node
.next
) != 0)
1144 memset(sk
, 0, offsetof(struct sock
, sk_node
.next
));
1145 memset(&sk
->sk_node
.pprev
, 0,
1146 size
- offsetof(struct sock
, sk_node
.pprev
));
1149 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1151 unsigned long nulls1
, nulls2
;
1153 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1154 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1155 if (nulls1
> nulls2
)
1156 swap(nulls1
, nulls2
);
1159 memset((char *)sk
, 0, nulls1
);
1160 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1161 nulls2
- nulls1
- sizeof(void *));
1162 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1163 size
- nulls2
- sizeof(void *));
1165 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1167 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1171 struct kmem_cache
*slab
;
1175 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1178 if (priority
& __GFP_ZERO
) {
1180 prot
->clear_sk(sk
, prot
->obj_size
);
1182 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1185 sk
= kmalloc(prot
->obj_size
, priority
);
1188 kmemcheck_annotate_bitfield(sk
, flags
);
1190 if (security_sk_alloc(sk
, family
, priority
))
1193 if (!try_module_get(prot
->owner
))
1195 sk_tx_queue_clear(sk
);
1201 security_sk_free(sk
);
1204 kmem_cache_free(slab
, sk
);
1210 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1212 struct kmem_cache
*slab
;
1213 struct module
*owner
;
1215 owner
= prot
->owner
;
1218 security_sk_free(sk
);
1220 kmem_cache_free(slab
, sk
);
1226 #ifdef CONFIG_CGROUPS
1227 void sock_update_classid(struct sock
*sk
)
1231 rcu_read_lock(); /* doing current task, which cannot vanish. */
1232 classid
= task_cls_classid(current
);
1234 if (classid
!= sk
->sk_classid
)
1235 sk
->sk_classid
= classid
;
1237 EXPORT_SYMBOL(sock_update_classid
);
1239 void sock_update_netprioidx(struct sock
*sk
, struct task_struct
*task
)
1244 sk
->sk_cgrp_prioidx
= task_netprioidx(task
);
1246 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1250 * sk_alloc - All socket objects are allocated here
1251 * @net: the applicable net namespace
1252 * @family: protocol family
1253 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1254 * @prot: struct proto associated with this new sock instance
1256 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1261 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1263 sk
->sk_family
= family
;
1265 * See comment in struct sock definition to understand
1266 * why we need sk_prot_creator -acme
1268 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1270 sock_net_set(sk
, get_net(net
));
1271 atomic_set(&sk
->sk_wmem_alloc
, 1);
1273 sock_update_classid(sk
);
1274 sock_update_netprioidx(sk
, current
);
1279 EXPORT_SYMBOL(sk_alloc
);
1281 static void __sk_free(struct sock
*sk
)
1283 struct sk_filter
*filter
;
1285 if (sk
->sk_destruct
)
1286 sk
->sk_destruct(sk
);
1288 filter
= rcu_dereference_check(sk
->sk_filter
,
1289 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1291 sk_filter_uncharge(sk
, filter
);
1292 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1295 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1297 if (atomic_read(&sk
->sk_omem_alloc
))
1298 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1299 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1301 if (sk
->sk_peer_cred
)
1302 put_cred(sk
->sk_peer_cred
);
1303 put_pid(sk
->sk_peer_pid
);
1304 put_net(sock_net(sk
));
1305 sk_prot_free(sk
->sk_prot_creator
, sk
);
1308 void sk_free(struct sock
*sk
)
1311 * We subtract one from sk_wmem_alloc and can know if
1312 * some packets are still in some tx queue.
1313 * If not null, sock_wfree() will call __sk_free(sk) later
1315 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1318 EXPORT_SYMBOL(sk_free
);
1321 * Last sock_put should drop reference to sk->sk_net. It has already
1322 * been dropped in sk_change_net. Taking reference to stopping namespace
1324 * Take reference to a socket to remove it from hash _alive_ and after that
1325 * destroy it in the context of init_net.
1327 void sk_release_kernel(struct sock
*sk
)
1329 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1333 sock_release(sk
->sk_socket
);
1334 release_net(sock_net(sk
));
1335 sock_net_set(sk
, get_net(&init_net
));
1338 EXPORT_SYMBOL(sk_release_kernel
);
1340 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1342 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1343 sock_update_memcg(newsk
);
1347 * sk_clone_lock - clone a socket, and lock its clone
1348 * @sk: the socket to clone
1349 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1351 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1353 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1357 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1358 if (newsk
!= NULL
) {
1359 struct sk_filter
*filter
;
1361 sock_copy(newsk
, sk
);
1364 get_net(sock_net(newsk
));
1365 sk_node_init(&newsk
->sk_node
);
1366 sock_lock_init(newsk
);
1367 bh_lock_sock(newsk
);
1368 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1369 newsk
->sk_backlog
.len
= 0;
1371 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1373 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1375 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1376 atomic_set(&newsk
->sk_omem_alloc
, 0);
1377 skb_queue_head_init(&newsk
->sk_receive_queue
);
1378 skb_queue_head_init(&newsk
->sk_write_queue
);
1379 #ifdef CONFIG_NET_DMA
1380 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1383 spin_lock_init(&newsk
->sk_dst_lock
);
1384 rwlock_init(&newsk
->sk_callback_lock
);
1385 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1386 af_callback_keys
+ newsk
->sk_family
,
1387 af_family_clock_key_strings
[newsk
->sk_family
]);
1389 newsk
->sk_dst_cache
= NULL
;
1390 newsk
->sk_wmem_queued
= 0;
1391 newsk
->sk_forward_alloc
= 0;
1392 newsk
->sk_send_head
= NULL
;
1393 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1395 sock_reset_flag(newsk
, SOCK_DONE
);
1396 skb_queue_head_init(&newsk
->sk_error_queue
);
1398 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1400 sk_filter_charge(newsk
, filter
);
1402 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1403 /* It is still raw copy of parent, so invalidate
1404 * destructor and make plain sk_free() */
1405 newsk
->sk_destruct
= NULL
;
1406 bh_unlock_sock(newsk
);
1413 newsk
->sk_priority
= 0;
1415 * Before updating sk_refcnt, we must commit prior changes to memory
1416 * (Documentation/RCU/rculist_nulls.txt for details)
1419 atomic_set(&newsk
->sk_refcnt
, 2);
1422 * Increment the counter in the same struct proto as the master
1423 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1424 * is the same as sk->sk_prot->socks, as this field was copied
1427 * This _changes_ the previous behaviour, where
1428 * tcp_create_openreq_child always was incrementing the
1429 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1430 * to be taken into account in all callers. -acme
1432 sk_refcnt_debug_inc(newsk
);
1433 sk_set_socket(newsk
, NULL
);
1434 newsk
->sk_wq
= NULL
;
1436 sk_update_clone(sk
, newsk
);
1438 if (newsk
->sk_prot
->sockets_allocated
)
1439 sk_sockets_allocated_inc(newsk
);
1441 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1442 net_enable_timestamp();
1447 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1449 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1451 __sk_dst_set(sk
, dst
);
1452 sk
->sk_route_caps
= dst
->dev
->features
;
1453 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1454 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1455 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1456 if (sk_can_gso(sk
)) {
1457 if (dst
->header_len
) {
1458 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1460 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1461 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1462 sk
->sk_gso_max_segs
= dst
->dev
->gso_max_segs
;
1466 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1469 * Simple resource managers for sockets.
1474 * Write buffer destructor automatically called from kfree_skb.
1476 void sock_wfree(struct sk_buff
*skb
)
1478 struct sock
*sk
= skb
->sk
;
1479 unsigned int len
= skb
->truesize
;
1481 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1483 * Keep a reference on sk_wmem_alloc, this will be released
1484 * after sk_write_space() call
1486 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1487 sk
->sk_write_space(sk
);
1491 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1492 * could not do because of in-flight packets
1494 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1497 EXPORT_SYMBOL(sock_wfree
);
1500 * Read buffer destructor automatically called from kfree_skb.
1502 void sock_rfree(struct sk_buff
*skb
)
1504 struct sock
*sk
= skb
->sk
;
1505 unsigned int len
= skb
->truesize
;
1507 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1508 sk_mem_uncharge(sk
, len
);
1510 EXPORT_SYMBOL(sock_rfree
);
1512 void sock_edemux(struct sk_buff
*skb
)
1514 struct sock
*sk
= skb
->sk
;
1517 if (sk
->sk_state
== TCP_TIME_WAIT
)
1518 inet_twsk_put(inet_twsk(sk
));
1523 EXPORT_SYMBOL(sock_edemux
);
1525 kuid_t
sock_i_uid(struct sock
*sk
)
1529 read_lock_bh(&sk
->sk_callback_lock
);
1530 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1531 read_unlock_bh(&sk
->sk_callback_lock
);
1534 EXPORT_SYMBOL(sock_i_uid
);
1536 unsigned long sock_i_ino(struct sock
*sk
)
1540 read_lock_bh(&sk
->sk_callback_lock
);
1541 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1542 read_unlock_bh(&sk
->sk_callback_lock
);
1545 EXPORT_SYMBOL(sock_i_ino
);
1548 * Allocate a skb from the socket's send buffer.
1550 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1553 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1554 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1556 skb_set_owner_w(skb
, sk
);
1562 EXPORT_SYMBOL(sock_wmalloc
);
1565 * Allocate a skb from the socket's receive buffer.
1567 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1570 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1571 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1573 skb_set_owner_r(skb
, sk
);
1581 * Allocate a memory block from the socket's option memory buffer.
1583 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1585 if ((unsigned int)size
<= sysctl_optmem_max
&&
1586 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1588 /* First do the add, to avoid the race if kmalloc
1591 atomic_add(size
, &sk
->sk_omem_alloc
);
1592 mem
= kmalloc(size
, priority
);
1595 atomic_sub(size
, &sk
->sk_omem_alloc
);
1599 EXPORT_SYMBOL(sock_kmalloc
);
1602 * Free an option memory block.
1604 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1607 atomic_sub(size
, &sk
->sk_omem_alloc
);
1609 EXPORT_SYMBOL(sock_kfree_s
);
1611 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1612 I think, these locks should be removed for datagram sockets.
1614 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1618 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1622 if (signal_pending(current
))
1624 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1625 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1626 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1628 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1632 timeo
= schedule_timeout(timeo
);
1634 finish_wait(sk_sleep(sk
), &wait
);
1640 * Generic send/receive buffer handlers
1643 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1644 unsigned long data_len
, int noblock
,
1647 struct sk_buff
*skb
;
1651 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1654 if (npages
> MAX_SKB_FRAGS
)
1657 gfp_mask
= sk
->sk_allocation
;
1658 if (gfp_mask
& __GFP_WAIT
)
1659 gfp_mask
|= __GFP_REPEAT
;
1661 timeo
= sock_sndtimeo(sk
, noblock
);
1663 err
= sock_error(sk
);
1668 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1671 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1672 skb
= alloc_skb(header_len
, gfp_mask
);
1676 /* No pages, we're done... */
1680 skb
->truesize
+= data_len
;
1681 skb_shinfo(skb
)->nr_frags
= npages
;
1682 for (i
= 0; i
< npages
; i
++) {
1685 page
= alloc_pages(sk
->sk_allocation
, 0);
1688 skb_shinfo(skb
)->nr_frags
= i
;
1693 __skb_fill_page_desc(skb
, i
,
1695 (data_len
>= PAGE_SIZE
?
1698 data_len
-= PAGE_SIZE
;
1701 /* Full success... */
1707 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1708 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1712 if (signal_pending(current
))
1714 timeo
= sock_wait_for_wmem(sk
, timeo
);
1717 skb_set_owner_w(skb
, sk
);
1721 err
= sock_intr_errno(timeo
);
1726 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1728 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1729 int noblock
, int *errcode
)
1731 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1733 EXPORT_SYMBOL(sock_alloc_send_skb
);
1735 /* On 32bit arches, an skb frag is limited to 2^15 */
1736 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1738 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1743 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1747 if (pfrag
->offset
< pfrag
->size
)
1749 put_page(pfrag
->page
);
1752 /* We restrict high order allocations to users that can afford to wait */
1753 order
= (sk
->sk_allocation
& __GFP_WAIT
) ? SKB_FRAG_PAGE_ORDER
: 0;
1756 gfp_t gfp
= sk
->sk_allocation
;
1759 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
1760 pfrag
->page
= alloc_pages(gfp
, order
);
1761 if (likely(pfrag
->page
)) {
1763 pfrag
->size
= PAGE_SIZE
<< order
;
1766 } while (--order
>= 0);
1768 sk_enter_memory_pressure(sk
);
1769 sk_stream_moderate_sndbuf(sk
);
1772 EXPORT_SYMBOL(sk_page_frag_refill
);
1774 static void __lock_sock(struct sock
*sk
)
1775 __releases(&sk
->sk_lock
.slock
)
1776 __acquires(&sk
->sk_lock
.slock
)
1781 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1782 TASK_UNINTERRUPTIBLE
);
1783 spin_unlock_bh(&sk
->sk_lock
.slock
);
1785 spin_lock_bh(&sk
->sk_lock
.slock
);
1786 if (!sock_owned_by_user(sk
))
1789 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1792 static void __release_sock(struct sock
*sk
)
1793 __releases(&sk
->sk_lock
.slock
)
1794 __acquires(&sk
->sk_lock
.slock
)
1796 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1799 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1803 struct sk_buff
*next
= skb
->next
;
1806 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1808 sk_backlog_rcv(sk
, skb
);
1811 * We are in process context here with softirqs
1812 * disabled, use cond_resched_softirq() to preempt.
1813 * This is safe to do because we've taken the backlog
1816 cond_resched_softirq();
1819 } while (skb
!= NULL
);
1822 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1825 * Doing the zeroing here guarantee we can not loop forever
1826 * while a wild producer attempts to flood us.
1828 sk
->sk_backlog
.len
= 0;
1832 * sk_wait_data - wait for data to arrive at sk_receive_queue
1833 * @sk: sock to wait on
1834 * @timeo: for how long
1836 * Now socket state including sk->sk_err is changed only under lock,
1837 * hence we may omit checks after joining wait queue.
1838 * We check receive queue before schedule() only as optimization;
1839 * it is very likely that release_sock() added new data.
1841 int sk_wait_data(struct sock
*sk
, long *timeo
)
1846 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1847 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1848 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1849 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1850 finish_wait(sk_sleep(sk
), &wait
);
1853 EXPORT_SYMBOL(sk_wait_data
);
1856 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1858 * @size: memory size to allocate
1859 * @kind: allocation type
1861 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1862 * rmem allocation. This function assumes that protocols which have
1863 * memory_pressure use sk_wmem_queued as write buffer accounting.
1865 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
1867 struct proto
*prot
= sk
->sk_prot
;
1868 int amt
= sk_mem_pages(size
);
1870 int parent_status
= UNDER_LIMIT
;
1872 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
1874 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
1877 if (parent_status
== UNDER_LIMIT
&&
1878 allocated
<= sk_prot_mem_limits(sk
, 0)) {
1879 sk_leave_memory_pressure(sk
);
1883 /* Under pressure. (we or our parents) */
1884 if ((parent_status
> SOFT_LIMIT
) ||
1885 allocated
> sk_prot_mem_limits(sk
, 1))
1886 sk_enter_memory_pressure(sk
);
1888 /* Over hard limit (we or our parents) */
1889 if ((parent_status
== OVER_LIMIT
) ||
1890 (allocated
> sk_prot_mem_limits(sk
, 2)))
1891 goto suppress_allocation
;
1893 /* guarantee minimum buffer size under pressure */
1894 if (kind
== SK_MEM_RECV
) {
1895 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
1898 } else { /* SK_MEM_SEND */
1899 if (sk
->sk_type
== SOCK_STREAM
) {
1900 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
1902 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
1903 prot
->sysctl_wmem
[0])
1907 if (sk_has_memory_pressure(sk
)) {
1910 if (!sk_under_memory_pressure(sk
))
1912 alloc
= sk_sockets_allocated_read_positive(sk
);
1913 if (sk_prot_mem_limits(sk
, 2) > alloc
*
1914 sk_mem_pages(sk
->sk_wmem_queued
+
1915 atomic_read(&sk
->sk_rmem_alloc
) +
1916 sk
->sk_forward_alloc
))
1920 suppress_allocation
:
1922 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
1923 sk_stream_moderate_sndbuf(sk
);
1925 /* Fail only if socket is _under_ its sndbuf.
1926 * In this case we cannot block, so that we have to fail.
1928 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
1932 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
1934 /* Alas. Undo changes. */
1935 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
1937 sk_memory_allocated_sub(sk
, amt
);
1941 EXPORT_SYMBOL(__sk_mem_schedule
);
1944 * __sk_reclaim - reclaim memory_allocated
1947 void __sk_mem_reclaim(struct sock
*sk
)
1949 sk_memory_allocated_sub(sk
,
1950 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
1951 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
1953 if (sk_under_memory_pressure(sk
) &&
1954 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
1955 sk_leave_memory_pressure(sk
);
1957 EXPORT_SYMBOL(__sk_mem_reclaim
);
1961 * Set of default routines for initialising struct proto_ops when
1962 * the protocol does not support a particular function. In certain
1963 * cases where it makes no sense for a protocol to have a "do nothing"
1964 * function, some default processing is provided.
1967 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1971 EXPORT_SYMBOL(sock_no_bind
);
1973 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1978 EXPORT_SYMBOL(sock_no_connect
);
1980 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1984 EXPORT_SYMBOL(sock_no_socketpair
);
1986 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1990 EXPORT_SYMBOL(sock_no_accept
);
1992 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1997 EXPORT_SYMBOL(sock_no_getname
);
1999 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2003 EXPORT_SYMBOL(sock_no_poll
);
2005 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2009 EXPORT_SYMBOL(sock_no_ioctl
);
2011 int sock_no_listen(struct socket
*sock
, int backlog
)
2015 EXPORT_SYMBOL(sock_no_listen
);
2017 int sock_no_shutdown(struct socket
*sock
, int how
)
2021 EXPORT_SYMBOL(sock_no_shutdown
);
2023 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2024 char __user
*optval
, unsigned int optlen
)
2028 EXPORT_SYMBOL(sock_no_setsockopt
);
2030 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2031 char __user
*optval
, int __user
*optlen
)
2035 EXPORT_SYMBOL(sock_no_getsockopt
);
2037 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2042 EXPORT_SYMBOL(sock_no_sendmsg
);
2044 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2045 size_t len
, int flags
)
2049 EXPORT_SYMBOL(sock_no_recvmsg
);
2051 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2053 /* Mirror missing mmap method error code */
2056 EXPORT_SYMBOL(sock_no_mmap
);
2058 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2061 struct msghdr msg
= {.msg_flags
= flags
};
2063 char *kaddr
= kmap(page
);
2064 iov
.iov_base
= kaddr
+ offset
;
2066 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2070 EXPORT_SYMBOL(sock_no_sendpage
);
2073 * Default Socket Callbacks
2076 static void sock_def_wakeup(struct sock
*sk
)
2078 struct socket_wq
*wq
;
2081 wq
= rcu_dereference(sk
->sk_wq
);
2082 if (wq_has_sleeper(wq
))
2083 wake_up_interruptible_all(&wq
->wait
);
2087 static void sock_def_error_report(struct sock
*sk
)
2089 struct socket_wq
*wq
;
2092 wq
= rcu_dereference(sk
->sk_wq
);
2093 if (wq_has_sleeper(wq
))
2094 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2095 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2099 static void sock_def_readable(struct sock
*sk
, int len
)
2101 struct socket_wq
*wq
;
2104 wq
= rcu_dereference(sk
->sk_wq
);
2105 if (wq_has_sleeper(wq
))
2106 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2107 POLLRDNORM
| POLLRDBAND
);
2108 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2112 static void sock_def_write_space(struct sock
*sk
)
2114 struct socket_wq
*wq
;
2118 /* Do not wake up a writer until he can make "significant"
2121 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2122 wq
= rcu_dereference(sk
->sk_wq
);
2123 if (wq_has_sleeper(wq
))
2124 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2125 POLLWRNORM
| POLLWRBAND
);
2127 /* Should agree with poll, otherwise some programs break */
2128 if (sock_writeable(sk
))
2129 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2135 static void sock_def_destruct(struct sock
*sk
)
2137 kfree(sk
->sk_protinfo
);
2140 void sk_send_sigurg(struct sock
*sk
)
2142 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2143 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2144 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2146 EXPORT_SYMBOL(sk_send_sigurg
);
2148 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2149 unsigned long expires
)
2151 if (!mod_timer(timer
, expires
))
2154 EXPORT_SYMBOL(sk_reset_timer
);
2156 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2158 if (timer_pending(timer
) && del_timer(timer
))
2161 EXPORT_SYMBOL(sk_stop_timer
);
2163 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2165 skb_queue_head_init(&sk
->sk_receive_queue
);
2166 skb_queue_head_init(&sk
->sk_write_queue
);
2167 skb_queue_head_init(&sk
->sk_error_queue
);
2168 #ifdef CONFIG_NET_DMA
2169 skb_queue_head_init(&sk
->sk_async_wait_queue
);
2172 sk
->sk_send_head
= NULL
;
2174 init_timer(&sk
->sk_timer
);
2176 sk
->sk_allocation
= GFP_KERNEL
;
2177 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2178 sk
->sk_sndbuf
= sysctl_wmem_default
;
2179 sk
->sk_state
= TCP_CLOSE
;
2180 sk_set_socket(sk
, sock
);
2182 sock_set_flag(sk
, SOCK_ZAPPED
);
2185 sk
->sk_type
= sock
->type
;
2186 sk
->sk_wq
= sock
->wq
;
2191 spin_lock_init(&sk
->sk_dst_lock
);
2192 rwlock_init(&sk
->sk_callback_lock
);
2193 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2194 af_callback_keys
+ sk
->sk_family
,
2195 af_family_clock_key_strings
[sk
->sk_family
]);
2197 sk
->sk_state_change
= sock_def_wakeup
;
2198 sk
->sk_data_ready
= sock_def_readable
;
2199 sk
->sk_write_space
= sock_def_write_space
;
2200 sk
->sk_error_report
= sock_def_error_report
;
2201 sk
->sk_destruct
= sock_def_destruct
;
2203 sk
->sk_frag
.page
= NULL
;
2204 sk
->sk_frag
.offset
= 0;
2205 sk
->sk_peek_off
= -1;
2207 sk
->sk_peer_pid
= NULL
;
2208 sk
->sk_peer_cred
= NULL
;
2209 sk
->sk_write_pending
= 0;
2210 sk
->sk_rcvlowat
= 1;
2211 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2212 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2214 sk
->sk_stamp
= ktime_set(-1L, 0);
2217 * Before updating sk_refcnt, we must commit prior changes to memory
2218 * (Documentation/RCU/rculist_nulls.txt for details)
2221 atomic_set(&sk
->sk_refcnt
, 1);
2222 atomic_set(&sk
->sk_drops
, 0);
2224 EXPORT_SYMBOL(sock_init_data
);
2226 void lock_sock_nested(struct sock
*sk
, int subclass
)
2229 spin_lock_bh(&sk
->sk_lock
.slock
);
2230 if (sk
->sk_lock
.owned
)
2232 sk
->sk_lock
.owned
= 1;
2233 spin_unlock(&sk
->sk_lock
.slock
);
2235 * The sk_lock has mutex_lock() semantics here:
2237 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2240 EXPORT_SYMBOL(lock_sock_nested
);
2242 void release_sock(struct sock
*sk
)
2245 * The sk_lock has mutex_unlock() semantics:
2247 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2249 spin_lock_bh(&sk
->sk_lock
.slock
);
2250 if (sk
->sk_backlog
.tail
)
2253 if (sk
->sk_prot
->release_cb
)
2254 sk
->sk_prot
->release_cb(sk
);
2256 sk
->sk_lock
.owned
= 0;
2257 if (waitqueue_active(&sk
->sk_lock
.wq
))
2258 wake_up(&sk
->sk_lock
.wq
);
2259 spin_unlock_bh(&sk
->sk_lock
.slock
);
2261 EXPORT_SYMBOL(release_sock
);
2264 * lock_sock_fast - fast version of lock_sock
2267 * This version should be used for very small section, where process wont block
2268 * return false if fast path is taken
2269 * sk_lock.slock locked, owned = 0, BH disabled
2270 * return true if slow path is taken
2271 * sk_lock.slock unlocked, owned = 1, BH enabled
2273 bool lock_sock_fast(struct sock
*sk
)
2276 spin_lock_bh(&sk
->sk_lock
.slock
);
2278 if (!sk
->sk_lock
.owned
)
2280 * Note : We must disable BH
2285 sk
->sk_lock
.owned
= 1;
2286 spin_unlock(&sk
->sk_lock
.slock
);
2288 * The sk_lock has mutex_lock() semantics here:
2290 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2294 EXPORT_SYMBOL(lock_sock_fast
);
2296 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2299 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2300 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2301 tv
= ktime_to_timeval(sk
->sk_stamp
);
2302 if (tv
.tv_sec
== -1)
2304 if (tv
.tv_sec
== 0) {
2305 sk
->sk_stamp
= ktime_get_real();
2306 tv
= ktime_to_timeval(sk
->sk_stamp
);
2308 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2310 EXPORT_SYMBOL(sock_get_timestamp
);
2312 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2315 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2316 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2317 ts
= ktime_to_timespec(sk
->sk_stamp
);
2318 if (ts
.tv_sec
== -1)
2320 if (ts
.tv_sec
== 0) {
2321 sk
->sk_stamp
= ktime_get_real();
2322 ts
= ktime_to_timespec(sk
->sk_stamp
);
2324 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2326 EXPORT_SYMBOL(sock_get_timestampns
);
2328 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2330 if (!sock_flag(sk
, flag
)) {
2331 unsigned long previous_flags
= sk
->sk_flags
;
2333 sock_set_flag(sk
, flag
);
2335 * we just set one of the two flags which require net
2336 * time stamping, but time stamping might have been on
2337 * already because of the other one
2339 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2340 net_enable_timestamp();
2345 * Get a socket option on an socket.
2347 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2348 * asynchronous errors should be reported by getsockopt. We assume
2349 * this means if you specify SO_ERROR (otherwise whats the point of it).
2351 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2352 char __user
*optval
, int __user
*optlen
)
2354 struct sock
*sk
= sock
->sk
;
2356 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2358 EXPORT_SYMBOL(sock_common_getsockopt
);
2360 #ifdef CONFIG_COMPAT
2361 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2362 char __user
*optval
, int __user
*optlen
)
2364 struct sock
*sk
= sock
->sk
;
2366 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2367 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2369 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2371 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2374 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2375 struct msghdr
*msg
, size_t size
, int flags
)
2377 struct sock
*sk
= sock
->sk
;
2381 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2382 flags
& ~MSG_DONTWAIT
, &addr_len
);
2384 msg
->msg_namelen
= addr_len
;
2387 EXPORT_SYMBOL(sock_common_recvmsg
);
2390 * Set socket options on an inet socket.
2392 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2393 char __user
*optval
, unsigned int optlen
)
2395 struct sock
*sk
= sock
->sk
;
2397 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2399 EXPORT_SYMBOL(sock_common_setsockopt
);
2401 #ifdef CONFIG_COMPAT
2402 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2403 char __user
*optval
, unsigned int optlen
)
2405 struct sock
*sk
= sock
->sk
;
2407 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2408 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2410 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2412 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2415 void sk_common_release(struct sock
*sk
)
2417 if (sk
->sk_prot
->destroy
)
2418 sk
->sk_prot
->destroy(sk
);
2421 * Observation: when sock_common_release is called, processes have
2422 * no access to socket. But net still has.
2423 * Step one, detach it from networking:
2425 * A. Remove from hash tables.
2428 sk
->sk_prot
->unhash(sk
);
2431 * In this point socket cannot receive new packets, but it is possible
2432 * that some packets are in flight because some CPU runs receiver and
2433 * did hash table lookup before we unhashed socket. They will achieve
2434 * receive queue and will be purged by socket destructor.
2436 * Also we still have packets pending on receive queue and probably,
2437 * our own packets waiting in device queues. sock_destroy will drain
2438 * receive queue, but transmitted packets will delay socket destruction
2439 * until the last reference will be released.
2444 xfrm_sk_free_policy(sk
);
2446 sk_refcnt_debug_release(sk
);
2448 if (sk
->sk_frag
.page
) {
2449 put_page(sk
->sk_frag
.page
);
2450 sk
->sk_frag
.page
= NULL
;
2455 EXPORT_SYMBOL(sk_common_release
);
2457 #ifdef CONFIG_PROC_FS
2458 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2460 int val
[PROTO_INUSE_NR
];
2463 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2465 #ifdef CONFIG_NET_NS
2466 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2468 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2470 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2472 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2474 int cpu
, idx
= prot
->inuse_idx
;
2477 for_each_possible_cpu(cpu
)
2478 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2480 return res
>= 0 ? res
: 0;
2482 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2484 static int __net_init
sock_inuse_init_net(struct net
*net
)
2486 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2487 return net
->core
.inuse
? 0 : -ENOMEM
;
2490 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2492 free_percpu(net
->core
.inuse
);
2495 static struct pernet_operations net_inuse_ops
= {
2496 .init
= sock_inuse_init_net
,
2497 .exit
= sock_inuse_exit_net
,
2500 static __init
int net_inuse_init(void)
2502 if (register_pernet_subsys(&net_inuse_ops
))
2503 panic("Cannot initialize net inuse counters");
2508 core_initcall(net_inuse_init
);
2510 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2512 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2514 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2516 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2518 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2520 int cpu
, idx
= prot
->inuse_idx
;
2523 for_each_possible_cpu(cpu
)
2524 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2526 return res
>= 0 ? res
: 0;
2528 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2531 static void assign_proto_idx(struct proto
*prot
)
2533 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2535 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2536 pr_err("PROTO_INUSE_NR exhausted\n");
2540 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2543 static void release_proto_idx(struct proto
*prot
)
2545 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2546 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2549 static inline void assign_proto_idx(struct proto
*prot
)
2553 static inline void release_proto_idx(struct proto
*prot
)
2558 int proto_register(struct proto
*prot
, int alloc_slab
)
2561 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2562 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2565 if (prot
->slab
== NULL
) {
2566 pr_crit("%s: Can't create sock SLAB cache!\n",
2571 if (prot
->rsk_prot
!= NULL
) {
2572 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2573 if (prot
->rsk_prot
->slab_name
== NULL
)
2574 goto out_free_sock_slab
;
2576 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2577 prot
->rsk_prot
->obj_size
, 0,
2578 SLAB_HWCACHE_ALIGN
, NULL
);
2580 if (prot
->rsk_prot
->slab
== NULL
) {
2581 pr_crit("%s: Can't create request sock SLAB cache!\n",
2583 goto out_free_request_sock_slab_name
;
2587 if (prot
->twsk_prot
!= NULL
) {
2588 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2590 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2591 goto out_free_request_sock_slab
;
2593 prot
->twsk_prot
->twsk_slab
=
2594 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2595 prot
->twsk_prot
->twsk_obj_size
,
2597 SLAB_HWCACHE_ALIGN
|
2600 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2601 goto out_free_timewait_sock_slab_name
;
2605 mutex_lock(&proto_list_mutex
);
2606 list_add(&prot
->node
, &proto_list
);
2607 assign_proto_idx(prot
);
2608 mutex_unlock(&proto_list_mutex
);
2611 out_free_timewait_sock_slab_name
:
2612 kfree(prot
->twsk_prot
->twsk_slab_name
);
2613 out_free_request_sock_slab
:
2614 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2615 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2616 prot
->rsk_prot
->slab
= NULL
;
2618 out_free_request_sock_slab_name
:
2620 kfree(prot
->rsk_prot
->slab_name
);
2622 kmem_cache_destroy(prot
->slab
);
2627 EXPORT_SYMBOL(proto_register
);
2629 void proto_unregister(struct proto
*prot
)
2631 mutex_lock(&proto_list_mutex
);
2632 release_proto_idx(prot
);
2633 list_del(&prot
->node
);
2634 mutex_unlock(&proto_list_mutex
);
2636 if (prot
->slab
!= NULL
) {
2637 kmem_cache_destroy(prot
->slab
);
2641 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2642 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2643 kfree(prot
->rsk_prot
->slab_name
);
2644 prot
->rsk_prot
->slab
= NULL
;
2647 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2648 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2649 kfree(prot
->twsk_prot
->twsk_slab_name
);
2650 prot
->twsk_prot
->twsk_slab
= NULL
;
2653 EXPORT_SYMBOL(proto_unregister
);
2655 #ifdef CONFIG_PROC_FS
2656 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2657 __acquires(proto_list_mutex
)
2659 mutex_lock(&proto_list_mutex
);
2660 return seq_list_start_head(&proto_list
, *pos
);
2663 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2665 return seq_list_next(v
, &proto_list
, pos
);
2668 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2669 __releases(proto_list_mutex
)
2671 mutex_unlock(&proto_list_mutex
);
2674 static char proto_method_implemented(const void *method
)
2676 return method
== NULL
? 'n' : 'y';
2678 static long sock_prot_memory_allocated(struct proto
*proto
)
2680 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2683 static char *sock_prot_memory_pressure(struct proto
*proto
)
2685 return proto
->memory_pressure
!= NULL
?
2686 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2689 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2692 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2693 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2696 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2697 sock_prot_memory_allocated(proto
),
2698 sock_prot_memory_pressure(proto
),
2700 proto
->slab
== NULL
? "no" : "yes",
2701 module_name(proto
->owner
),
2702 proto_method_implemented(proto
->close
),
2703 proto_method_implemented(proto
->connect
),
2704 proto_method_implemented(proto
->disconnect
),
2705 proto_method_implemented(proto
->accept
),
2706 proto_method_implemented(proto
->ioctl
),
2707 proto_method_implemented(proto
->init
),
2708 proto_method_implemented(proto
->destroy
),
2709 proto_method_implemented(proto
->shutdown
),
2710 proto_method_implemented(proto
->setsockopt
),
2711 proto_method_implemented(proto
->getsockopt
),
2712 proto_method_implemented(proto
->sendmsg
),
2713 proto_method_implemented(proto
->recvmsg
),
2714 proto_method_implemented(proto
->sendpage
),
2715 proto_method_implemented(proto
->bind
),
2716 proto_method_implemented(proto
->backlog_rcv
),
2717 proto_method_implemented(proto
->hash
),
2718 proto_method_implemented(proto
->unhash
),
2719 proto_method_implemented(proto
->get_port
),
2720 proto_method_implemented(proto
->enter_memory_pressure
));
2723 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2725 if (v
== &proto_list
)
2726 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2735 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2737 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2741 static const struct seq_operations proto_seq_ops
= {
2742 .start
= proto_seq_start
,
2743 .next
= proto_seq_next
,
2744 .stop
= proto_seq_stop
,
2745 .show
= proto_seq_show
,
2748 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2750 return seq_open_net(inode
, file
, &proto_seq_ops
,
2751 sizeof(struct seq_net_private
));
2754 static const struct file_operations proto_seq_fops
= {
2755 .owner
= THIS_MODULE
,
2756 .open
= proto_seq_open
,
2758 .llseek
= seq_lseek
,
2759 .release
= seq_release_net
,
2762 static __net_init
int proto_init_net(struct net
*net
)
2764 if (!proc_net_fops_create(net
, "protocols", S_IRUGO
, &proto_seq_fops
))
2770 static __net_exit
void proto_exit_net(struct net
*net
)
2772 proc_net_remove(net
, "protocols");
2776 static __net_initdata
struct pernet_operations proto_net_ops
= {
2777 .init
= proto_init_net
,
2778 .exit
= proto_exit_net
,
2781 static int __init
proto_init(void)
2783 return register_pernet_subsys(&proto_net_ops
);
2786 subsys_initcall(proto_init
);
2788 #endif /* PROC_FS */