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 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
333 if (optlen
< sizeof(tv
))
335 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
337 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
341 static int warned __read_mostly
;
344 if (warned
< 10 && net_ratelimit()) {
346 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
347 __func__
, current
->comm
, task_pid_nr(current
));
351 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
352 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
354 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
355 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
359 static void sock_warn_obsolete_bsdism(const char *name
)
362 static char warncomm
[TASK_COMM_LEN
];
363 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
364 strcpy(warncomm
, current
->comm
);
365 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
371 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
373 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
375 if (sk
->sk_flags
& flags
) {
376 sk
->sk_flags
&= ~flags
;
377 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
378 net_disable_timestamp();
383 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
388 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
390 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
391 atomic_inc(&sk
->sk_drops
);
392 trace_sock_rcvqueue_full(sk
, skb
);
396 err
= sk_filter(sk
, skb
);
400 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
401 atomic_inc(&sk
->sk_drops
);
406 skb_set_owner_r(skb
, sk
);
408 /* Cache the SKB length before we tack it onto the receive
409 * queue. Once it is added it no longer belongs to us and
410 * may be freed by other threads of control pulling packets
415 /* we escape from rcu protected region, make sure we dont leak
420 spin_lock_irqsave(&list
->lock
, flags
);
421 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
422 __skb_queue_tail(list
, skb
);
423 spin_unlock_irqrestore(&list
->lock
, flags
);
425 if (!sock_flag(sk
, SOCK_DEAD
))
426 sk
->sk_data_ready(sk
, skb_len
);
429 EXPORT_SYMBOL(sock_queue_rcv_skb
);
431 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
433 int rc
= NET_RX_SUCCESS
;
435 if (sk_filter(sk
, skb
))
436 goto discard_and_relse
;
440 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
)) {
441 atomic_inc(&sk
->sk_drops
);
442 goto discard_and_relse
;
445 bh_lock_sock_nested(sk
);
448 if (!sock_owned_by_user(sk
)) {
450 * trylock + unlock semantics:
452 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
454 rc
= sk_backlog_rcv(sk
, skb
);
456 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
457 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
459 atomic_inc(&sk
->sk_drops
);
460 goto discard_and_relse
;
471 EXPORT_SYMBOL(sk_receive_skb
);
473 void sk_reset_txq(struct sock
*sk
)
475 sk_tx_queue_clear(sk
);
477 EXPORT_SYMBOL(sk_reset_txq
);
479 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
481 struct dst_entry
*dst
= __sk_dst_get(sk
);
483 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
484 sk_tx_queue_clear(sk
);
485 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
492 EXPORT_SYMBOL(__sk_dst_check
);
494 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
496 struct dst_entry
*dst
= sk_dst_get(sk
);
498 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
506 EXPORT_SYMBOL(sk_dst_check
);
508 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
511 int ret
= -ENOPROTOOPT
;
512 #ifdef CONFIG_NETDEVICES
513 struct net
*net
= sock_net(sk
);
514 char devname
[IFNAMSIZ
];
519 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
526 /* Bind this socket to a particular device like "eth0",
527 * as specified in the passed interface name. If the
528 * name is "" or the option length is zero the socket
531 if (optlen
> IFNAMSIZ
- 1)
532 optlen
= IFNAMSIZ
- 1;
533 memset(devname
, 0, sizeof(devname
));
536 if (copy_from_user(devname
, optval
, optlen
))
540 if (devname
[0] != '\0') {
541 struct net_device
*dev
;
544 dev
= dev_get_by_name_rcu(net
, devname
);
546 index
= dev
->ifindex
;
554 sk
->sk_bound_dev_if
= index
;
566 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
567 int __user
*optlen
, int len
)
569 int ret
= -ENOPROTOOPT
;
570 #ifdef CONFIG_NETDEVICES
571 struct net
*net
= sock_net(sk
);
572 struct net_device
*dev
;
573 char devname
[IFNAMSIZ
];
576 if (sk
->sk_bound_dev_if
== 0) {
586 seq
= read_seqcount_begin(&devnet_rename_seq
);
588 dev
= dev_get_by_index_rcu(net
, sk
->sk_bound_dev_if
);
595 strcpy(devname
, dev
->name
);
597 if (read_seqcount_retry(&devnet_rename_seq
, seq
))
600 len
= strlen(devname
) + 1;
603 if (copy_to_user(optval
, devname
, len
))
608 if (put_user(len
, optlen
))
619 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
622 sock_set_flag(sk
, bit
);
624 sock_reset_flag(sk
, bit
);
628 * This is meant for all protocols to use and covers goings on
629 * at the socket level. Everything here is generic.
632 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
633 char __user
*optval
, unsigned int optlen
)
635 struct sock
*sk
= sock
->sk
;
642 * Options without arguments
645 if (optname
== SO_BINDTODEVICE
)
646 return sock_setbindtodevice(sk
, optval
, optlen
);
648 if (optlen
< sizeof(int))
651 if (get_user(val
, (int __user
*)optval
))
654 valbool
= val
? 1 : 0;
660 if (val
&& !capable(CAP_NET_ADMIN
))
663 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
666 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
675 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
678 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
681 /* Don't error on this BSD doesn't and if you think
682 * about it this is right. Otherwise apps have to
683 * play 'guess the biggest size' games. RCVBUF/SNDBUF
684 * are treated in BSD as hints
686 val
= min_t(u32
, val
, sysctl_wmem_max
);
688 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
689 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
690 /* Wake up sending tasks if we upped the value. */
691 sk
->sk_write_space(sk
);
695 if (!capable(CAP_NET_ADMIN
)) {
702 /* Don't error on this BSD doesn't and if you think
703 * about it this is right. Otherwise apps have to
704 * play 'guess the biggest size' games. RCVBUF/SNDBUF
705 * are treated in BSD as hints
707 val
= min_t(u32
, val
, sysctl_rmem_max
);
709 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
711 * We double it on the way in to account for
712 * "struct sk_buff" etc. overhead. Applications
713 * assume that the SO_RCVBUF setting they make will
714 * allow that much actual data to be received on that
717 * Applications are unaware that "struct sk_buff" and
718 * other overheads allocate from the receive buffer
719 * during socket buffer allocation.
721 * And after considering the possible alternatives,
722 * returning the value we actually used in getsockopt
723 * is the most desirable behavior.
725 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
729 if (!capable(CAP_NET_ADMIN
)) {
737 if (sk
->sk_protocol
== IPPROTO_TCP
&&
738 sk
->sk_type
== SOCK_STREAM
)
739 tcp_set_keepalive(sk
, valbool
);
741 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
745 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
749 sk
->sk_no_check
= valbool
;
753 if ((val
>= 0 && val
<= 6) ||
754 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
755 sk
->sk_priority
= val
;
761 if (optlen
< sizeof(ling
)) {
762 ret
= -EINVAL
; /* 1003.1g */
765 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
770 sock_reset_flag(sk
, SOCK_LINGER
);
772 #if (BITS_PER_LONG == 32)
773 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
774 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
777 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
778 sock_set_flag(sk
, SOCK_LINGER
);
783 sock_warn_obsolete_bsdism("setsockopt");
788 set_bit(SOCK_PASSCRED
, &sock
->flags
);
790 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
796 if (optname
== SO_TIMESTAMP
)
797 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
799 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
800 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
801 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
803 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
804 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
808 case SO_TIMESTAMPING
:
809 if (val
& ~SOF_TIMESTAMPING_MASK
) {
813 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
814 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
815 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
816 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
817 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
818 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
819 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
820 sock_enable_timestamp(sk
,
821 SOCK_TIMESTAMPING_RX_SOFTWARE
);
823 sock_disable_timestamp(sk
,
824 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
825 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
826 val
& SOF_TIMESTAMPING_SOFTWARE
);
827 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
828 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
829 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
830 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
836 sk
->sk_rcvlowat
= val
? : 1;
840 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
844 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
847 case SO_ATTACH_FILTER
:
849 if (optlen
== sizeof(struct sock_fprog
)) {
850 struct sock_fprog fprog
;
853 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
856 ret
= sk_attach_filter(&fprog
, sk
);
860 case SO_DETACH_FILTER
:
861 ret
= sk_detach_filter(sk
);
866 set_bit(SOCK_PASSSEC
, &sock
->flags
);
868 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
871 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
877 /* We implement the SO_SNDLOWAT etc to
878 not be settable (1003.1g 5.3) */
880 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
884 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
888 if (sock
->ops
->set_peek_off
)
889 sock
->ops
->set_peek_off(sk
, val
);
895 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
905 EXPORT_SYMBOL(sock_setsockopt
);
908 void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
911 ucred
->pid
= pid_vnr(pid
);
912 ucred
->uid
= ucred
->gid
= -1;
914 struct user_namespace
*current_ns
= current_user_ns();
916 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
917 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
920 EXPORT_SYMBOL_GPL(cred_to_ucred
);
922 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
923 char __user
*optval
, int __user
*optlen
)
925 struct sock
*sk
= sock
->sk
;
933 int lv
= sizeof(int);
936 if (get_user(len
, optlen
))
941 memset(&v
, 0, sizeof(v
));
945 v
.val
= sock_flag(sk
, SOCK_DBG
);
949 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
953 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
957 v
.val
= sk
->sk_sndbuf
;
961 v
.val
= sk
->sk_rcvbuf
;
965 v
.val
= sk
->sk_reuse
;
969 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
977 v
.val
= sk
->sk_protocol
;
981 v
.val
= sk
->sk_family
;
985 v
.val
= -sock_error(sk
);
987 v
.val
= xchg(&sk
->sk_err_soft
, 0);
991 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
995 v
.val
= sk
->sk_no_check
;
999 v
.val
= sk
->sk_priority
;
1003 lv
= sizeof(v
.ling
);
1004 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1005 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1009 sock_warn_obsolete_bsdism("getsockopt");
1013 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1014 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1017 case SO_TIMESTAMPNS
:
1018 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1021 case SO_TIMESTAMPING
:
1023 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
1024 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
1025 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
1026 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
1027 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
1028 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
1029 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
1030 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
1031 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
1032 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
1033 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
1034 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
1035 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
1036 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
1040 lv
= sizeof(struct timeval
);
1041 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1045 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1046 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1051 lv
= sizeof(struct timeval
);
1052 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1056 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1057 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1062 v
.val
= sk
->sk_rcvlowat
;
1070 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1075 struct ucred peercred
;
1076 if (len
> sizeof(peercred
))
1077 len
= sizeof(peercred
);
1078 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1079 if (copy_to_user(optval
, &peercred
, len
))
1088 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1092 if (copy_to_user(optval
, address
, len
))
1097 /* Dubious BSD thing... Probably nobody even uses it, but
1098 * the UNIX standard wants it for whatever reason... -DaveM
1101 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1105 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1109 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1112 v
.val
= sk
->sk_mark
;
1116 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1119 case SO_WIFI_STATUS
:
1120 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1124 if (!sock
->ops
->set_peek_off
)
1127 v
.val
= sk
->sk_peek_off
;
1130 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1133 case SO_BINDTODEVICE
:
1134 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1137 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1144 return -ENOPROTOOPT
;
1149 if (copy_to_user(optval
, &v
, len
))
1152 if (put_user(len
, optlen
))
1158 * Initialize an sk_lock.
1160 * (We also register the sk_lock with the lock validator.)
1162 static inline void sock_lock_init(struct sock
*sk
)
1164 sock_lock_init_class_and_name(sk
,
1165 af_family_slock_key_strings
[sk
->sk_family
],
1166 af_family_slock_keys
+ sk
->sk_family
,
1167 af_family_key_strings
[sk
->sk_family
],
1168 af_family_keys
+ sk
->sk_family
);
1172 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1173 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1174 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1176 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1178 #ifdef CONFIG_SECURITY_NETWORK
1179 void *sptr
= nsk
->sk_security
;
1181 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1183 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1184 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1186 #ifdef CONFIG_SECURITY_NETWORK
1187 nsk
->sk_security
= sptr
;
1188 security_sk_clone(osk
, nsk
);
1193 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1194 * un-modified. Special care is taken when initializing object to zero.
1196 static inline void sk_prot_clear_nulls(struct sock
*sk
, int size
)
1198 if (offsetof(struct sock
, sk_node
.next
) != 0)
1199 memset(sk
, 0, offsetof(struct sock
, sk_node
.next
));
1200 memset(&sk
->sk_node
.pprev
, 0,
1201 size
- offsetof(struct sock
, sk_node
.pprev
));
1204 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1206 unsigned long nulls1
, nulls2
;
1208 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1209 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1210 if (nulls1
> nulls2
)
1211 swap(nulls1
, nulls2
);
1214 memset((char *)sk
, 0, nulls1
);
1215 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1216 nulls2
- nulls1
- sizeof(void *));
1217 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1218 size
- nulls2
- sizeof(void *));
1220 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1222 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1226 struct kmem_cache
*slab
;
1230 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1233 if (priority
& __GFP_ZERO
) {
1235 prot
->clear_sk(sk
, prot
->obj_size
);
1237 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1240 sk
= kmalloc(prot
->obj_size
, priority
);
1243 kmemcheck_annotate_bitfield(sk
, flags
);
1245 if (security_sk_alloc(sk
, family
, priority
))
1248 if (!try_module_get(prot
->owner
))
1250 sk_tx_queue_clear(sk
);
1256 security_sk_free(sk
);
1259 kmem_cache_free(slab
, sk
);
1265 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1267 struct kmem_cache
*slab
;
1268 struct module
*owner
;
1270 owner
= prot
->owner
;
1273 security_sk_free(sk
);
1275 kmem_cache_free(slab
, sk
);
1281 #ifdef CONFIG_CGROUPS
1282 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1283 void sock_update_classid(struct sock
*sk
, struct task_struct
*task
)
1287 classid
= task_cls_classid(task
);
1288 if (classid
!= sk
->sk_classid
)
1289 sk
->sk_classid
= classid
;
1291 EXPORT_SYMBOL(sock_update_classid
);
1294 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1295 void sock_update_netprioidx(struct sock
*sk
, struct task_struct
*task
)
1300 sk
->sk_cgrp_prioidx
= task_netprioidx(task
);
1302 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1307 * sk_alloc - All socket objects are allocated here
1308 * @net: the applicable net namespace
1309 * @family: protocol family
1310 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1311 * @prot: struct proto associated with this new sock instance
1313 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1318 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1320 sk
->sk_family
= family
;
1322 * See comment in struct sock definition to understand
1323 * why we need sk_prot_creator -acme
1325 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1327 sock_net_set(sk
, get_net(net
));
1328 atomic_set(&sk
->sk_wmem_alloc
, 1);
1330 sock_update_classid(sk
, current
);
1331 sock_update_netprioidx(sk
, current
);
1336 EXPORT_SYMBOL(sk_alloc
);
1338 static void __sk_free(struct sock
*sk
)
1340 struct sk_filter
*filter
;
1342 if (sk
->sk_destruct
)
1343 sk
->sk_destruct(sk
);
1345 filter
= rcu_dereference_check(sk
->sk_filter
,
1346 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1348 sk_filter_uncharge(sk
, filter
);
1349 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1352 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1354 if (atomic_read(&sk
->sk_omem_alloc
))
1355 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1356 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1358 if (sk
->sk_peer_cred
)
1359 put_cred(sk
->sk_peer_cred
);
1360 put_pid(sk
->sk_peer_pid
);
1361 put_net(sock_net(sk
));
1362 sk_prot_free(sk
->sk_prot_creator
, sk
);
1365 void sk_free(struct sock
*sk
)
1368 * We subtract one from sk_wmem_alloc and can know if
1369 * some packets are still in some tx queue.
1370 * If not null, sock_wfree() will call __sk_free(sk) later
1372 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1375 EXPORT_SYMBOL(sk_free
);
1378 * Last sock_put should drop reference to sk->sk_net. It has already
1379 * been dropped in sk_change_net. Taking reference to stopping namespace
1381 * Take reference to a socket to remove it from hash _alive_ and after that
1382 * destroy it in the context of init_net.
1384 void sk_release_kernel(struct sock
*sk
)
1386 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1390 sock_release(sk
->sk_socket
);
1391 release_net(sock_net(sk
));
1392 sock_net_set(sk
, get_net(&init_net
));
1395 EXPORT_SYMBOL(sk_release_kernel
);
1397 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1399 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1400 sock_update_memcg(newsk
);
1404 * sk_clone_lock - clone a socket, and lock its clone
1405 * @sk: the socket to clone
1406 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1408 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1410 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1414 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1415 if (newsk
!= NULL
) {
1416 struct sk_filter
*filter
;
1418 sock_copy(newsk
, sk
);
1421 get_net(sock_net(newsk
));
1422 sk_node_init(&newsk
->sk_node
);
1423 sock_lock_init(newsk
);
1424 bh_lock_sock(newsk
);
1425 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1426 newsk
->sk_backlog
.len
= 0;
1428 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1430 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1432 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1433 atomic_set(&newsk
->sk_omem_alloc
, 0);
1434 skb_queue_head_init(&newsk
->sk_receive_queue
);
1435 skb_queue_head_init(&newsk
->sk_write_queue
);
1436 #ifdef CONFIG_NET_DMA
1437 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1440 spin_lock_init(&newsk
->sk_dst_lock
);
1441 rwlock_init(&newsk
->sk_callback_lock
);
1442 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1443 af_callback_keys
+ newsk
->sk_family
,
1444 af_family_clock_key_strings
[newsk
->sk_family
]);
1446 newsk
->sk_dst_cache
= NULL
;
1447 newsk
->sk_wmem_queued
= 0;
1448 newsk
->sk_forward_alloc
= 0;
1449 newsk
->sk_send_head
= NULL
;
1450 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1452 sock_reset_flag(newsk
, SOCK_DONE
);
1453 skb_queue_head_init(&newsk
->sk_error_queue
);
1455 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1457 sk_filter_charge(newsk
, filter
);
1459 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1460 /* It is still raw copy of parent, so invalidate
1461 * destructor and make plain sk_free() */
1462 newsk
->sk_destruct
= NULL
;
1463 bh_unlock_sock(newsk
);
1470 newsk
->sk_priority
= 0;
1472 * Before updating sk_refcnt, we must commit prior changes to memory
1473 * (Documentation/RCU/rculist_nulls.txt for details)
1476 atomic_set(&newsk
->sk_refcnt
, 2);
1479 * Increment the counter in the same struct proto as the master
1480 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1481 * is the same as sk->sk_prot->socks, as this field was copied
1484 * This _changes_ the previous behaviour, where
1485 * tcp_create_openreq_child always was incrementing the
1486 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1487 * to be taken into account in all callers. -acme
1489 sk_refcnt_debug_inc(newsk
);
1490 sk_set_socket(newsk
, NULL
);
1491 newsk
->sk_wq
= NULL
;
1493 sk_update_clone(sk
, newsk
);
1495 if (newsk
->sk_prot
->sockets_allocated
)
1496 sk_sockets_allocated_inc(newsk
);
1498 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1499 net_enable_timestamp();
1504 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1506 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1508 __sk_dst_set(sk
, dst
);
1509 sk
->sk_route_caps
= dst
->dev
->features
;
1510 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1511 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1512 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1513 if (sk_can_gso(sk
)) {
1514 if (dst
->header_len
) {
1515 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1517 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1518 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1519 sk
->sk_gso_max_segs
= dst
->dev
->gso_max_segs
;
1523 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1526 * Simple resource managers for sockets.
1531 * Write buffer destructor automatically called from kfree_skb.
1533 void sock_wfree(struct sk_buff
*skb
)
1535 struct sock
*sk
= skb
->sk
;
1536 unsigned int len
= skb
->truesize
;
1538 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1540 * Keep a reference on sk_wmem_alloc, this will be released
1541 * after sk_write_space() call
1543 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1544 sk
->sk_write_space(sk
);
1548 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1549 * could not do because of in-flight packets
1551 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1554 EXPORT_SYMBOL(sock_wfree
);
1557 * Read buffer destructor automatically called from kfree_skb.
1559 void sock_rfree(struct sk_buff
*skb
)
1561 struct sock
*sk
= skb
->sk
;
1562 unsigned int len
= skb
->truesize
;
1564 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1565 sk_mem_uncharge(sk
, len
);
1567 EXPORT_SYMBOL(sock_rfree
);
1569 void sock_edemux(struct sk_buff
*skb
)
1571 struct sock
*sk
= skb
->sk
;
1574 if (sk
->sk_state
== TCP_TIME_WAIT
)
1575 inet_twsk_put(inet_twsk(sk
));
1580 EXPORT_SYMBOL(sock_edemux
);
1582 kuid_t
sock_i_uid(struct sock
*sk
)
1586 read_lock_bh(&sk
->sk_callback_lock
);
1587 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1588 read_unlock_bh(&sk
->sk_callback_lock
);
1591 EXPORT_SYMBOL(sock_i_uid
);
1593 unsigned long sock_i_ino(struct sock
*sk
)
1597 read_lock_bh(&sk
->sk_callback_lock
);
1598 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1599 read_unlock_bh(&sk
->sk_callback_lock
);
1602 EXPORT_SYMBOL(sock_i_ino
);
1605 * Allocate a skb from the socket's send buffer.
1607 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1610 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1611 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1613 skb_set_owner_w(skb
, sk
);
1619 EXPORT_SYMBOL(sock_wmalloc
);
1622 * Allocate a skb from the socket's receive buffer.
1624 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1627 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1628 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1630 skb_set_owner_r(skb
, sk
);
1638 * Allocate a memory block from the socket's option memory buffer.
1640 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1642 if ((unsigned int)size
<= sysctl_optmem_max
&&
1643 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1645 /* First do the add, to avoid the race if kmalloc
1648 atomic_add(size
, &sk
->sk_omem_alloc
);
1649 mem
= kmalloc(size
, priority
);
1652 atomic_sub(size
, &sk
->sk_omem_alloc
);
1656 EXPORT_SYMBOL(sock_kmalloc
);
1659 * Free an option memory block.
1661 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1664 atomic_sub(size
, &sk
->sk_omem_alloc
);
1666 EXPORT_SYMBOL(sock_kfree_s
);
1668 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1669 I think, these locks should be removed for datagram sockets.
1671 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1675 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1679 if (signal_pending(current
))
1681 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1682 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1683 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1685 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1689 timeo
= schedule_timeout(timeo
);
1691 finish_wait(sk_sleep(sk
), &wait
);
1697 * Generic send/receive buffer handlers
1700 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1701 unsigned long data_len
, int noblock
,
1704 struct sk_buff
*skb
;
1708 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1711 if (npages
> MAX_SKB_FRAGS
)
1714 gfp_mask
= sk
->sk_allocation
;
1715 if (gfp_mask
& __GFP_WAIT
)
1716 gfp_mask
|= __GFP_REPEAT
;
1718 timeo
= sock_sndtimeo(sk
, noblock
);
1720 err
= sock_error(sk
);
1725 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1728 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1729 skb
= alloc_skb(header_len
, gfp_mask
);
1733 /* No pages, we're done... */
1737 skb
->truesize
+= data_len
;
1738 skb_shinfo(skb
)->nr_frags
= npages
;
1739 for (i
= 0; i
< npages
; i
++) {
1742 page
= alloc_pages(sk
->sk_allocation
, 0);
1745 skb_shinfo(skb
)->nr_frags
= i
;
1750 __skb_fill_page_desc(skb
, i
,
1752 (data_len
>= PAGE_SIZE
?
1755 data_len
-= PAGE_SIZE
;
1758 /* Full success... */
1764 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1765 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1769 if (signal_pending(current
))
1771 timeo
= sock_wait_for_wmem(sk
, timeo
);
1774 skb_set_owner_w(skb
, sk
);
1778 err
= sock_intr_errno(timeo
);
1783 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1785 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1786 int noblock
, int *errcode
)
1788 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1790 EXPORT_SYMBOL(sock_alloc_send_skb
);
1792 /* On 32bit arches, an skb frag is limited to 2^15 */
1793 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1795 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1800 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1804 if (pfrag
->offset
< pfrag
->size
)
1806 put_page(pfrag
->page
);
1809 /* We restrict high order allocations to users that can afford to wait */
1810 order
= (sk
->sk_allocation
& __GFP_WAIT
) ? SKB_FRAG_PAGE_ORDER
: 0;
1813 gfp_t gfp
= sk
->sk_allocation
;
1816 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
1817 pfrag
->page
= alloc_pages(gfp
, order
);
1818 if (likely(pfrag
->page
)) {
1820 pfrag
->size
= PAGE_SIZE
<< order
;
1823 } while (--order
>= 0);
1825 sk_enter_memory_pressure(sk
);
1826 sk_stream_moderate_sndbuf(sk
);
1829 EXPORT_SYMBOL(sk_page_frag_refill
);
1831 static void __lock_sock(struct sock
*sk
)
1832 __releases(&sk
->sk_lock
.slock
)
1833 __acquires(&sk
->sk_lock
.slock
)
1838 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1839 TASK_UNINTERRUPTIBLE
);
1840 spin_unlock_bh(&sk
->sk_lock
.slock
);
1842 spin_lock_bh(&sk
->sk_lock
.slock
);
1843 if (!sock_owned_by_user(sk
))
1846 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1849 static void __release_sock(struct sock
*sk
)
1850 __releases(&sk
->sk_lock
.slock
)
1851 __acquires(&sk
->sk_lock
.slock
)
1853 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1856 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1860 struct sk_buff
*next
= skb
->next
;
1863 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1865 sk_backlog_rcv(sk
, skb
);
1868 * We are in process context here with softirqs
1869 * disabled, use cond_resched_softirq() to preempt.
1870 * This is safe to do because we've taken the backlog
1873 cond_resched_softirq();
1876 } while (skb
!= NULL
);
1879 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1882 * Doing the zeroing here guarantee we can not loop forever
1883 * while a wild producer attempts to flood us.
1885 sk
->sk_backlog
.len
= 0;
1889 * sk_wait_data - wait for data to arrive at sk_receive_queue
1890 * @sk: sock to wait on
1891 * @timeo: for how long
1893 * Now socket state including sk->sk_err is changed only under lock,
1894 * hence we may omit checks after joining wait queue.
1895 * We check receive queue before schedule() only as optimization;
1896 * it is very likely that release_sock() added new data.
1898 int sk_wait_data(struct sock
*sk
, long *timeo
)
1903 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1904 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1905 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1906 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1907 finish_wait(sk_sleep(sk
), &wait
);
1910 EXPORT_SYMBOL(sk_wait_data
);
1913 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1915 * @size: memory size to allocate
1916 * @kind: allocation type
1918 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1919 * rmem allocation. This function assumes that protocols which have
1920 * memory_pressure use sk_wmem_queued as write buffer accounting.
1922 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
1924 struct proto
*prot
= sk
->sk_prot
;
1925 int amt
= sk_mem_pages(size
);
1927 int parent_status
= UNDER_LIMIT
;
1929 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
1931 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
1934 if (parent_status
== UNDER_LIMIT
&&
1935 allocated
<= sk_prot_mem_limits(sk
, 0)) {
1936 sk_leave_memory_pressure(sk
);
1940 /* Under pressure. (we or our parents) */
1941 if ((parent_status
> SOFT_LIMIT
) ||
1942 allocated
> sk_prot_mem_limits(sk
, 1))
1943 sk_enter_memory_pressure(sk
);
1945 /* Over hard limit (we or our parents) */
1946 if ((parent_status
== OVER_LIMIT
) ||
1947 (allocated
> sk_prot_mem_limits(sk
, 2)))
1948 goto suppress_allocation
;
1950 /* guarantee minimum buffer size under pressure */
1951 if (kind
== SK_MEM_RECV
) {
1952 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
1955 } else { /* SK_MEM_SEND */
1956 if (sk
->sk_type
== SOCK_STREAM
) {
1957 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
1959 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
1960 prot
->sysctl_wmem
[0])
1964 if (sk_has_memory_pressure(sk
)) {
1967 if (!sk_under_memory_pressure(sk
))
1969 alloc
= sk_sockets_allocated_read_positive(sk
);
1970 if (sk_prot_mem_limits(sk
, 2) > alloc
*
1971 sk_mem_pages(sk
->sk_wmem_queued
+
1972 atomic_read(&sk
->sk_rmem_alloc
) +
1973 sk
->sk_forward_alloc
))
1977 suppress_allocation
:
1979 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
1980 sk_stream_moderate_sndbuf(sk
);
1982 /* Fail only if socket is _under_ its sndbuf.
1983 * In this case we cannot block, so that we have to fail.
1985 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
1989 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
1991 /* Alas. Undo changes. */
1992 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
1994 sk_memory_allocated_sub(sk
, amt
);
1998 EXPORT_SYMBOL(__sk_mem_schedule
);
2001 * __sk_reclaim - reclaim memory_allocated
2004 void __sk_mem_reclaim(struct sock
*sk
)
2006 sk_memory_allocated_sub(sk
,
2007 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
2008 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
2010 if (sk_under_memory_pressure(sk
) &&
2011 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2012 sk_leave_memory_pressure(sk
);
2014 EXPORT_SYMBOL(__sk_mem_reclaim
);
2018 * Set of default routines for initialising struct proto_ops when
2019 * the protocol does not support a particular function. In certain
2020 * cases where it makes no sense for a protocol to have a "do nothing"
2021 * function, some default processing is provided.
2024 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2028 EXPORT_SYMBOL(sock_no_bind
);
2030 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2035 EXPORT_SYMBOL(sock_no_connect
);
2037 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2041 EXPORT_SYMBOL(sock_no_socketpair
);
2043 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2047 EXPORT_SYMBOL(sock_no_accept
);
2049 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2054 EXPORT_SYMBOL(sock_no_getname
);
2056 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2060 EXPORT_SYMBOL(sock_no_poll
);
2062 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2066 EXPORT_SYMBOL(sock_no_ioctl
);
2068 int sock_no_listen(struct socket
*sock
, int backlog
)
2072 EXPORT_SYMBOL(sock_no_listen
);
2074 int sock_no_shutdown(struct socket
*sock
, int how
)
2078 EXPORT_SYMBOL(sock_no_shutdown
);
2080 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2081 char __user
*optval
, unsigned int optlen
)
2085 EXPORT_SYMBOL(sock_no_setsockopt
);
2087 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2088 char __user
*optval
, int __user
*optlen
)
2092 EXPORT_SYMBOL(sock_no_getsockopt
);
2094 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2099 EXPORT_SYMBOL(sock_no_sendmsg
);
2101 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2102 size_t len
, int flags
)
2106 EXPORT_SYMBOL(sock_no_recvmsg
);
2108 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2110 /* Mirror missing mmap method error code */
2113 EXPORT_SYMBOL(sock_no_mmap
);
2115 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2118 struct msghdr msg
= {.msg_flags
= flags
};
2120 char *kaddr
= kmap(page
);
2121 iov
.iov_base
= kaddr
+ offset
;
2123 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2127 EXPORT_SYMBOL(sock_no_sendpage
);
2130 * Default Socket Callbacks
2133 static void sock_def_wakeup(struct sock
*sk
)
2135 struct socket_wq
*wq
;
2138 wq
= rcu_dereference(sk
->sk_wq
);
2139 if (wq_has_sleeper(wq
))
2140 wake_up_interruptible_all(&wq
->wait
);
2144 static void sock_def_error_report(struct sock
*sk
)
2146 struct socket_wq
*wq
;
2149 wq
= rcu_dereference(sk
->sk_wq
);
2150 if (wq_has_sleeper(wq
))
2151 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2152 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2156 static void sock_def_readable(struct sock
*sk
, int len
)
2158 struct socket_wq
*wq
;
2161 wq
= rcu_dereference(sk
->sk_wq
);
2162 if (wq_has_sleeper(wq
))
2163 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2164 POLLRDNORM
| POLLRDBAND
);
2165 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2169 static void sock_def_write_space(struct sock
*sk
)
2171 struct socket_wq
*wq
;
2175 /* Do not wake up a writer until he can make "significant"
2178 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2179 wq
= rcu_dereference(sk
->sk_wq
);
2180 if (wq_has_sleeper(wq
))
2181 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2182 POLLWRNORM
| POLLWRBAND
);
2184 /* Should agree with poll, otherwise some programs break */
2185 if (sock_writeable(sk
))
2186 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2192 static void sock_def_destruct(struct sock
*sk
)
2194 kfree(sk
->sk_protinfo
);
2197 void sk_send_sigurg(struct sock
*sk
)
2199 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2200 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2201 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2203 EXPORT_SYMBOL(sk_send_sigurg
);
2205 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2206 unsigned long expires
)
2208 if (!mod_timer(timer
, expires
))
2211 EXPORT_SYMBOL(sk_reset_timer
);
2213 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2215 if (timer_pending(timer
) && del_timer(timer
))
2218 EXPORT_SYMBOL(sk_stop_timer
);
2220 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2222 skb_queue_head_init(&sk
->sk_receive_queue
);
2223 skb_queue_head_init(&sk
->sk_write_queue
);
2224 skb_queue_head_init(&sk
->sk_error_queue
);
2225 #ifdef CONFIG_NET_DMA
2226 skb_queue_head_init(&sk
->sk_async_wait_queue
);
2229 sk
->sk_send_head
= NULL
;
2231 init_timer(&sk
->sk_timer
);
2233 sk
->sk_allocation
= GFP_KERNEL
;
2234 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2235 sk
->sk_sndbuf
= sysctl_wmem_default
;
2236 sk
->sk_state
= TCP_CLOSE
;
2237 sk_set_socket(sk
, sock
);
2239 sock_set_flag(sk
, SOCK_ZAPPED
);
2242 sk
->sk_type
= sock
->type
;
2243 sk
->sk_wq
= sock
->wq
;
2248 spin_lock_init(&sk
->sk_dst_lock
);
2249 rwlock_init(&sk
->sk_callback_lock
);
2250 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2251 af_callback_keys
+ sk
->sk_family
,
2252 af_family_clock_key_strings
[sk
->sk_family
]);
2254 sk
->sk_state_change
= sock_def_wakeup
;
2255 sk
->sk_data_ready
= sock_def_readable
;
2256 sk
->sk_write_space
= sock_def_write_space
;
2257 sk
->sk_error_report
= sock_def_error_report
;
2258 sk
->sk_destruct
= sock_def_destruct
;
2260 sk
->sk_frag
.page
= NULL
;
2261 sk
->sk_frag
.offset
= 0;
2262 sk
->sk_peek_off
= -1;
2264 sk
->sk_peer_pid
= NULL
;
2265 sk
->sk_peer_cred
= NULL
;
2266 sk
->sk_write_pending
= 0;
2267 sk
->sk_rcvlowat
= 1;
2268 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2269 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2271 sk
->sk_stamp
= ktime_set(-1L, 0);
2274 * Before updating sk_refcnt, we must commit prior changes to memory
2275 * (Documentation/RCU/rculist_nulls.txt for details)
2278 atomic_set(&sk
->sk_refcnt
, 1);
2279 atomic_set(&sk
->sk_drops
, 0);
2281 EXPORT_SYMBOL(sock_init_data
);
2283 void lock_sock_nested(struct sock
*sk
, int subclass
)
2286 spin_lock_bh(&sk
->sk_lock
.slock
);
2287 if (sk
->sk_lock
.owned
)
2289 sk
->sk_lock
.owned
= 1;
2290 spin_unlock(&sk
->sk_lock
.slock
);
2292 * The sk_lock has mutex_lock() semantics here:
2294 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2297 EXPORT_SYMBOL(lock_sock_nested
);
2299 void release_sock(struct sock
*sk
)
2302 * The sk_lock has mutex_unlock() semantics:
2304 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2306 spin_lock_bh(&sk
->sk_lock
.slock
);
2307 if (sk
->sk_backlog
.tail
)
2310 if (sk
->sk_prot
->release_cb
)
2311 sk
->sk_prot
->release_cb(sk
);
2313 sk
->sk_lock
.owned
= 0;
2314 if (waitqueue_active(&sk
->sk_lock
.wq
))
2315 wake_up(&sk
->sk_lock
.wq
);
2316 spin_unlock_bh(&sk
->sk_lock
.slock
);
2318 EXPORT_SYMBOL(release_sock
);
2321 * lock_sock_fast - fast version of lock_sock
2324 * This version should be used for very small section, where process wont block
2325 * return false if fast path is taken
2326 * sk_lock.slock locked, owned = 0, BH disabled
2327 * return true if slow path is taken
2328 * sk_lock.slock unlocked, owned = 1, BH enabled
2330 bool lock_sock_fast(struct sock
*sk
)
2333 spin_lock_bh(&sk
->sk_lock
.slock
);
2335 if (!sk
->sk_lock
.owned
)
2337 * Note : We must disable BH
2342 sk
->sk_lock
.owned
= 1;
2343 spin_unlock(&sk
->sk_lock
.slock
);
2345 * The sk_lock has mutex_lock() semantics here:
2347 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2351 EXPORT_SYMBOL(lock_sock_fast
);
2353 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2356 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2357 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2358 tv
= ktime_to_timeval(sk
->sk_stamp
);
2359 if (tv
.tv_sec
== -1)
2361 if (tv
.tv_sec
== 0) {
2362 sk
->sk_stamp
= ktime_get_real();
2363 tv
= ktime_to_timeval(sk
->sk_stamp
);
2365 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2367 EXPORT_SYMBOL(sock_get_timestamp
);
2369 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2372 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2373 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2374 ts
= ktime_to_timespec(sk
->sk_stamp
);
2375 if (ts
.tv_sec
== -1)
2377 if (ts
.tv_sec
== 0) {
2378 sk
->sk_stamp
= ktime_get_real();
2379 ts
= ktime_to_timespec(sk
->sk_stamp
);
2381 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2383 EXPORT_SYMBOL(sock_get_timestampns
);
2385 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2387 if (!sock_flag(sk
, flag
)) {
2388 unsigned long previous_flags
= sk
->sk_flags
;
2390 sock_set_flag(sk
, flag
);
2392 * we just set one of the two flags which require net
2393 * time stamping, but time stamping might have been on
2394 * already because of the other one
2396 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2397 net_enable_timestamp();
2402 * Get a socket option on an socket.
2404 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2405 * asynchronous errors should be reported by getsockopt. We assume
2406 * this means if you specify SO_ERROR (otherwise whats the point of it).
2408 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2409 char __user
*optval
, int __user
*optlen
)
2411 struct sock
*sk
= sock
->sk
;
2413 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2415 EXPORT_SYMBOL(sock_common_getsockopt
);
2417 #ifdef CONFIG_COMPAT
2418 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2419 char __user
*optval
, int __user
*optlen
)
2421 struct sock
*sk
= sock
->sk
;
2423 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2424 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2426 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2428 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2431 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2432 struct msghdr
*msg
, size_t size
, int flags
)
2434 struct sock
*sk
= sock
->sk
;
2438 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2439 flags
& ~MSG_DONTWAIT
, &addr_len
);
2441 msg
->msg_namelen
= addr_len
;
2444 EXPORT_SYMBOL(sock_common_recvmsg
);
2447 * Set socket options on an inet socket.
2449 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2450 char __user
*optval
, unsigned int optlen
)
2452 struct sock
*sk
= sock
->sk
;
2454 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2456 EXPORT_SYMBOL(sock_common_setsockopt
);
2458 #ifdef CONFIG_COMPAT
2459 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2460 char __user
*optval
, unsigned int optlen
)
2462 struct sock
*sk
= sock
->sk
;
2464 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2465 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2467 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2469 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2472 void sk_common_release(struct sock
*sk
)
2474 if (sk
->sk_prot
->destroy
)
2475 sk
->sk_prot
->destroy(sk
);
2478 * Observation: when sock_common_release is called, processes have
2479 * no access to socket. But net still has.
2480 * Step one, detach it from networking:
2482 * A. Remove from hash tables.
2485 sk
->sk_prot
->unhash(sk
);
2488 * In this point socket cannot receive new packets, but it is possible
2489 * that some packets are in flight because some CPU runs receiver and
2490 * did hash table lookup before we unhashed socket. They will achieve
2491 * receive queue and will be purged by socket destructor.
2493 * Also we still have packets pending on receive queue and probably,
2494 * our own packets waiting in device queues. sock_destroy will drain
2495 * receive queue, but transmitted packets will delay socket destruction
2496 * until the last reference will be released.
2501 xfrm_sk_free_policy(sk
);
2503 sk_refcnt_debug_release(sk
);
2505 if (sk
->sk_frag
.page
) {
2506 put_page(sk
->sk_frag
.page
);
2507 sk
->sk_frag
.page
= NULL
;
2512 EXPORT_SYMBOL(sk_common_release
);
2514 #ifdef CONFIG_PROC_FS
2515 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2517 int val
[PROTO_INUSE_NR
];
2520 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2522 #ifdef CONFIG_NET_NS
2523 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2525 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2527 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2529 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2531 int cpu
, idx
= prot
->inuse_idx
;
2534 for_each_possible_cpu(cpu
)
2535 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2537 return res
>= 0 ? res
: 0;
2539 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2541 static int __net_init
sock_inuse_init_net(struct net
*net
)
2543 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2544 return net
->core
.inuse
? 0 : -ENOMEM
;
2547 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2549 free_percpu(net
->core
.inuse
);
2552 static struct pernet_operations net_inuse_ops
= {
2553 .init
= sock_inuse_init_net
,
2554 .exit
= sock_inuse_exit_net
,
2557 static __init
int net_inuse_init(void)
2559 if (register_pernet_subsys(&net_inuse_ops
))
2560 panic("Cannot initialize net inuse counters");
2565 core_initcall(net_inuse_init
);
2567 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2569 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2571 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2573 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2575 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2577 int cpu
, idx
= prot
->inuse_idx
;
2580 for_each_possible_cpu(cpu
)
2581 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2583 return res
>= 0 ? res
: 0;
2585 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2588 static void assign_proto_idx(struct proto
*prot
)
2590 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2592 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2593 pr_err("PROTO_INUSE_NR exhausted\n");
2597 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2600 static void release_proto_idx(struct proto
*prot
)
2602 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2603 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2606 static inline void assign_proto_idx(struct proto
*prot
)
2610 static inline void release_proto_idx(struct proto
*prot
)
2615 int proto_register(struct proto
*prot
, int alloc_slab
)
2618 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2619 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2622 if (prot
->slab
== NULL
) {
2623 pr_crit("%s: Can't create sock SLAB cache!\n",
2628 if (prot
->rsk_prot
!= NULL
) {
2629 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2630 if (prot
->rsk_prot
->slab_name
== NULL
)
2631 goto out_free_sock_slab
;
2633 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2634 prot
->rsk_prot
->obj_size
, 0,
2635 SLAB_HWCACHE_ALIGN
, NULL
);
2637 if (prot
->rsk_prot
->slab
== NULL
) {
2638 pr_crit("%s: Can't create request sock SLAB cache!\n",
2640 goto out_free_request_sock_slab_name
;
2644 if (prot
->twsk_prot
!= NULL
) {
2645 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2647 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2648 goto out_free_request_sock_slab
;
2650 prot
->twsk_prot
->twsk_slab
=
2651 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2652 prot
->twsk_prot
->twsk_obj_size
,
2654 SLAB_HWCACHE_ALIGN
|
2657 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2658 goto out_free_timewait_sock_slab_name
;
2662 mutex_lock(&proto_list_mutex
);
2663 list_add(&prot
->node
, &proto_list
);
2664 assign_proto_idx(prot
);
2665 mutex_unlock(&proto_list_mutex
);
2668 out_free_timewait_sock_slab_name
:
2669 kfree(prot
->twsk_prot
->twsk_slab_name
);
2670 out_free_request_sock_slab
:
2671 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2672 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2673 prot
->rsk_prot
->slab
= NULL
;
2675 out_free_request_sock_slab_name
:
2677 kfree(prot
->rsk_prot
->slab_name
);
2679 kmem_cache_destroy(prot
->slab
);
2684 EXPORT_SYMBOL(proto_register
);
2686 void proto_unregister(struct proto
*prot
)
2688 mutex_lock(&proto_list_mutex
);
2689 release_proto_idx(prot
);
2690 list_del(&prot
->node
);
2691 mutex_unlock(&proto_list_mutex
);
2693 if (prot
->slab
!= NULL
) {
2694 kmem_cache_destroy(prot
->slab
);
2698 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2699 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2700 kfree(prot
->rsk_prot
->slab_name
);
2701 prot
->rsk_prot
->slab
= NULL
;
2704 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2705 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2706 kfree(prot
->twsk_prot
->twsk_slab_name
);
2707 prot
->twsk_prot
->twsk_slab
= NULL
;
2710 EXPORT_SYMBOL(proto_unregister
);
2712 #ifdef CONFIG_PROC_FS
2713 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2714 __acquires(proto_list_mutex
)
2716 mutex_lock(&proto_list_mutex
);
2717 return seq_list_start_head(&proto_list
, *pos
);
2720 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2722 return seq_list_next(v
, &proto_list
, pos
);
2725 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2726 __releases(proto_list_mutex
)
2728 mutex_unlock(&proto_list_mutex
);
2731 static char proto_method_implemented(const void *method
)
2733 return method
== NULL
? 'n' : 'y';
2735 static long sock_prot_memory_allocated(struct proto
*proto
)
2737 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2740 static char *sock_prot_memory_pressure(struct proto
*proto
)
2742 return proto
->memory_pressure
!= NULL
?
2743 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2746 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2749 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2750 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2753 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2754 sock_prot_memory_allocated(proto
),
2755 sock_prot_memory_pressure(proto
),
2757 proto
->slab
== NULL
? "no" : "yes",
2758 module_name(proto
->owner
),
2759 proto_method_implemented(proto
->close
),
2760 proto_method_implemented(proto
->connect
),
2761 proto_method_implemented(proto
->disconnect
),
2762 proto_method_implemented(proto
->accept
),
2763 proto_method_implemented(proto
->ioctl
),
2764 proto_method_implemented(proto
->init
),
2765 proto_method_implemented(proto
->destroy
),
2766 proto_method_implemented(proto
->shutdown
),
2767 proto_method_implemented(proto
->setsockopt
),
2768 proto_method_implemented(proto
->getsockopt
),
2769 proto_method_implemented(proto
->sendmsg
),
2770 proto_method_implemented(proto
->recvmsg
),
2771 proto_method_implemented(proto
->sendpage
),
2772 proto_method_implemented(proto
->bind
),
2773 proto_method_implemented(proto
->backlog_rcv
),
2774 proto_method_implemented(proto
->hash
),
2775 proto_method_implemented(proto
->unhash
),
2776 proto_method_implemented(proto
->get_port
),
2777 proto_method_implemented(proto
->enter_memory_pressure
));
2780 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2782 if (v
== &proto_list
)
2783 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2792 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2794 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2798 static const struct seq_operations proto_seq_ops
= {
2799 .start
= proto_seq_start
,
2800 .next
= proto_seq_next
,
2801 .stop
= proto_seq_stop
,
2802 .show
= proto_seq_show
,
2805 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2807 return seq_open_net(inode
, file
, &proto_seq_ops
,
2808 sizeof(struct seq_net_private
));
2811 static const struct file_operations proto_seq_fops
= {
2812 .owner
= THIS_MODULE
,
2813 .open
= proto_seq_open
,
2815 .llseek
= seq_lseek
,
2816 .release
= seq_release_net
,
2819 static __net_init
int proto_init_net(struct net
*net
)
2821 if (!proc_net_fops_create(net
, "protocols", S_IRUGO
, &proto_seq_fops
))
2827 static __net_exit
void proto_exit_net(struct net
*net
)
2829 proc_net_remove(net
, "protocols");
2833 static __net_initdata
struct pernet_operations proto_net_ops
= {
2834 .init
= proto_init_net
,
2835 .exit
= proto_exit_net
,
2838 static int __init
proto_init(void)
2840 return register_pernet_subsys(&proto_net_ops
);
2843 subsys_initcall(proto_init
);
2845 #endif /* PROC_FS */