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/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <asm/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex
);
148 static LIST_HEAD(proto_list
);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock
*sk
,
161 struct user_namespace
*user_ns
, int cap
)
163 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
164 ns_capable(user_ns
, cap
);
166 EXPORT_SYMBOL(sk_ns_capable
);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock
*sk
, int cap
)
179 return sk_ns_capable(sk
, &init_user_ns
, cap
);
181 EXPORT_SYMBOL(sk_capable
);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock
*sk
, int cap
)
194 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
196 EXPORT_SYMBOL(sk_net_capable
);
199 #ifdef CONFIG_MEMCG_KMEM
200 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
205 mutex_lock(&proto_list_mutex
);
206 list_for_each_entry(proto
, &proto_list
, node
) {
207 if (proto
->init_cgroup
) {
208 ret
= proto
->init_cgroup(memcg
, ss
);
214 mutex_unlock(&proto_list_mutex
);
217 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
218 if (proto
->destroy_cgroup
)
219 proto
->destroy_cgroup(memcg
);
220 mutex_unlock(&proto_list_mutex
);
224 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
228 mutex_lock(&proto_list_mutex
);
229 list_for_each_entry_reverse(proto
, &proto_list
, node
)
230 if (proto
->destroy_cgroup
)
231 proto
->destroy_cgroup(memcg
);
232 mutex_unlock(&proto_list_mutex
);
237 * Each address family might have different locking rules, so we have
238 * one slock key per address family:
240 static struct lock_class_key af_family_keys
[AF_MAX
];
241 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
243 #if defined(CONFIG_MEMCG_KMEM)
244 struct static_key memcg_socket_limit_enabled
;
245 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
249 * Make lock validator output more readable. (we pre-construct these
250 * strings build-time, so that runtime initialization of socket
253 static const char *const af_family_key_strings
[AF_MAX
+1] = {
254 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
255 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
256 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
257 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
258 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
259 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
260 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
261 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
262 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
263 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
264 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
265 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
266 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
267 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
269 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
270 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
271 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
272 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
273 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
274 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
275 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
276 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
277 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
278 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
279 "slock-27" , "slock-28" , "slock-AF_CAN" ,
280 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
281 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
282 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
283 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
285 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
286 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
287 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
288 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
289 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
290 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
291 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
292 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
293 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
294 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
295 "clock-27" , "clock-28" , "clock-AF_CAN" ,
296 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
297 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
298 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
299 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
303 * sk_callback_lock locking rules are per-address-family,
304 * so split the lock classes by using a per-AF key:
306 static struct lock_class_key af_callback_keys
[AF_MAX
];
308 /* Take into consideration the size of the struct sk_buff overhead in the
309 * determination of these values, since that is non-constant across
310 * platforms. This makes socket queueing behavior and performance
311 * not depend upon such differences.
313 #define _SK_MEM_PACKETS 256
314 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
315 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
316 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
318 /* Run time adjustable parameters. */
319 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
320 EXPORT_SYMBOL(sysctl_wmem_max
);
321 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
322 EXPORT_SYMBOL(sysctl_rmem_max
);
323 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
324 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
326 /* Maximal space eaten by iovec or ancillary data plus some space */
327 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
328 EXPORT_SYMBOL(sysctl_optmem_max
);
330 int sysctl_tstamp_allow_data __read_mostly
= 1;
332 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
333 EXPORT_SYMBOL_GPL(memalloc_socks
);
336 * sk_set_memalloc - sets %SOCK_MEMALLOC
337 * @sk: socket to set it on
339 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
340 * It's the responsibility of the admin to adjust min_free_kbytes
341 * to meet the requirements
343 void sk_set_memalloc(struct sock
*sk
)
345 sock_set_flag(sk
, SOCK_MEMALLOC
);
346 sk
->sk_allocation
|= __GFP_MEMALLOC
;
347 static_key_slow_inc(&memalloc_socks
);
349 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
351 void sk_clear_memalloc(struct sock
*sk
)
353 sock_reset_flag(sk
, SOCK_MEMALLOC
);
354 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
355 static_key_slow_dec(&memalloc_socks
);
358 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
359 * progress of swapping. SOCK_MEMALLOC may be cleared while
360 * it has rmem allocations due to the last swapfile being deactivated
361 * but there is a risk that the socket is unusable due to exceeding
362 * the rmem limits. Reclaim the reserves and obey rmem limits again.
366 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
368 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
371 unsigned long pflags
= current
->flags
;
373 /* these should have been dropped before queueing */
374 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
376 current
->flags
|= PF_MEMALLOC
;
377 ret
= sk
->sk_backlog_rcv(sk
, skb
);
378 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
382 EXPORT_SYMBOL(__sk_backlog_rcv
);
384 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
388 if (optlen
< sizeof(tv
))
390 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
392 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
396 static int warned __read_mostly
;
399 if (warned
< 10 && net_ratelimit()) {
401 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
402 __func__
, current
->comm
, task_pid_nr(current
));
406 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
407 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
409 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
410 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
414 static void sock_warn_obsolete_bsdism(const char *name
)
417 static char warncomm
[TASK_COMM_LEN
];
418 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
419 strcpy(warncomm
, current
->comm
);
420 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
426 static bool sock_needs_netstamp(const struct sock
*sk
)
428 switch (sk
->sk_family
) {
437 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
439 if (sk
->sk_flags
& flags
) {
440 sk
->sk_flags
&= ~flags
;
441 if (sock_needs_netstamp(sk
) &&
442 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
443 net_disable_timestamp();
448 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
452 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
454 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
455 atomic_inc(&sk
->sk_drops
);
456 trace_sock_rcvqueue_full(sk
, skb
);
460 err
= sk_filter(sk
, skb
);
464 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
465 atomic_inc(&sk
->sk_drops
);
470 skb_set_owner_r(skb
, sk
);
472 /* we escape from rcu protected region, make sure we dont leak
477 spin_lock_irqsave(&list
->lock
, flags
);
478 sock_skb_set_dropcount(sk
, skb
);
479 __skb_queue_tail(list
, skb
);
480 spin_unlock_irqrestore(&list
->lock
, flags
);
482 if (!sock_flag(sk
, SOCK_DEAD
))
483 sk
->sk_data_ready(sk
);
486 EXPORT_SYMBOL(sock_queue_rcv_skb
);
488 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
490 int rc
= NET_RX_SUCCESS
;
492 if (sk_filter(sk
, skb
))
493 goto discard_and_relse
;
497 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
498 atomic_inc(&sk
->sk_drops
);
499 goto discard_and_relse
;
502 bh_lock_sock_nested(sk
);
505 if (!sock_owned_by_user(sk
)) {
507 * trylock + unlock semantics:
509 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
511 rc
= sk_backlog_rcv(sk
, skb
);
513 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
514 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
516 atomic_inc(&sk
->sk_drops
);
517 goto discard_and_relse
;
528 EXPORT_SYMBOL(sk_receive_skb
);
530 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
532 struct dst_entry
*dst
= __sk_dst_get(sk
);
534 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
535 sk_tx_queue_clear(sk
);
536 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
543 EXPORT_SYMBOL(__sk_dst_check
);
545 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
547 struct dst_entry
*dst
= sk_dst_get(sk
);
549 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
557 EXPORT_SYMBOL(sk_dst_check
);
559 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
562 int ret
= -ENOPROTOOPT
;
563 #ifdef CONFIG_NETDEVICES
564 struct net
*net
= sock_net(sk
);
565 char devname
[IFNAMSIZ
];
570 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
577 /* Bind this socket to a particular device like "eth0",
578 * as specified in the passed interface name. If the
579 * name is "" or the option length is zero the socket
582 if (optlen
> IFNAMSIZ
- 1)
583 optlen
= IFNAMSIZ
- 1;
584 memset(devname
, 0, sizeof(devname
));
587 if (copy_from_user(devname
, optval
, optlen
))
591 if (devname
[0] != '\0') {
592 struct net_device
*dev
;
595 dev
= dev_get_by_name_rcu(net
, devname
);
597 index
= dev
->ifindex
;
605 sk
->sk_bound_dev_if
= index
;
617 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
618 int __user
*optlen
, int len
)
620 int ret
= -ENOPROTOOPT
;
621 #ifdef CONFIG_NETDEVICES
622 struct net
*net
= sock_net(sk
);
623 char devname
[IFNAMSIZ
];
625 if (sk
->sk_bound_dev_if
== 0) {
634 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
638 len
= strlen(devname
) + 1;
641 if (copy_to_user(optval
, devname
, len
))
646 if (put_user(len
, optlen
))
657 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
660 sock_set_flag(sk
, bit
);
662 sock_reset_flag(sk
, bit
);
665 bool sk_mc_loop(struct sock
*sk
)
667 if (dev_recursion_level())
671 switch (sk
->sk_family
) {
673 return inet_sk(sk
)->mc_loop
;
674 #if IS_ENABLED(CONFIG_IPV6)
676 return inet6_sk(sk
)->mc_loop
;
682 EXPORT_SYMBOL(sk_mc_loop
);
685 * This is meant for all protocols to use and covers goings on
686 * at the socket level. Everything here is generic.
689 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
690 char __user
*optval
, unsigned int optlen
)
692 struct sock
*sk
= sock
->sk
;
699 * Options without arguments
702 if (optname
== SO_BINDTODEVICE
)
703 return sock_setbindtodevice(sk
, optval
, optlen
);
705 if (optlen
< sizeof(int))
708 if (get_user(val
, (int __user
*)optval
))
711 valbool
= val
? 1 : 0;
717 if (val
&& !capable(CAP_NET_ADMIN
))
720 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
723 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
726 sk
->sk_reuseport
= valbool
;
735 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
738 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
741 /* Don't error on this BSD doesn't and if you think
742 * about it this is right. Otherwise apps have to
743 * play 'guess the biggest size' games. RCVBUF/SNDBUF
744 * are treated in BSD as hints
746 val
= min_t(u32
, val
, sysctl_wmem_max
);
748 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
749 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
750 /* Wake up sending tasks if we upped the value. */
751 sk
->sk_write_space(sk
);
755 if (!capable(CAP_NET_ADMIN
)) {
762 /* Don't error on this BSD doesn't and if you think
763 * about it this is right. Otherwise apps have to
764 * play 'guess the biggest size' games. RCVBUF/SNDBUF
765 * are treated in BSD as hints
767 val
= min_t(u32
, val
, sysctl_rmem_max
);
769 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
771 * We double it on the way in to account for
772 * "struct sk_buff" etc. overhead. Applications
773 * assume that the SO_RCVBUF setting they make will
774 * allow that much actual data to be received on that
777 * Applications are unaware that "struct sk_buff" and
778 * other overheads allocate from the receive buffer
779 * during socket buffer allocation.
781 * And after considering the possible alternatives,
782 * returning the value we actually used in getsockopt
783 * is the most desirable behavior.
785 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
789 if (!capable(CAP_NET_ADMIN
)) {
797 if (sk
->sk_protocol
== IPPROTO_TCP
&&
798 sk
->sk_type
== SOCK_STREAM
)
799 tcp_set_keepalive(sk
, valbool
);
801 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
805 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
809 sk
->sk_no_check_tx
= valbool
;
813 if ((val
>= 0 && val
<= 6) ||
814 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
815 sk
->sk_priority
= val
;
821 if (optlen
< sizeof(ling
)) {
822 ret
= -EINVAL
; /* 1003.1g */
825 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
830 sock_reset_flag(sk
, SOCK_LINGER
);
832 #if (BITS_PER_LONG == 32)
833 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
834 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
837 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
838 sock_set_flag(sk
, SOCK_LINGER
);
843 sock_warn_obsolete_bsdism("setsockopt");
848 set_bit(SOCK_PASSCRED
, &sock
->flags
);
850 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
856 if (optname
== SO_TIMESTAMP
)
857 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
859 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
860 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
861 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
863 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
864 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
868 case SO_TIMESTAMPING
:
869 if (val
& ~SOF_TIMESTAMPING_MASK
) {
874 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
875 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
876 if (sk
->sk_protocol
== IPPROTO_TCP
&&
877 sk
->sk_type
== SOCK_STREAM
) {
878 if (sk
->sk_state
!= TCP_ESTABLISHED
) {
882 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
887 sk
->sk_tsflags
= val
;
888 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
889 sock_enable_timestamp(sk
,
890 SOCK_TIMESTAMPING_RX_SOFTWARE
);
892 sock_disable_timestamp(sk
,
893 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
899 sk
->sk_rcvlowat
= val
? : 1;
903 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
907 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
910 case SO_ATTACH_FILTER
:
912 if (optlen
== sizeof(struct sock_fprog
)) {
913 struct sock_fprog fprog
;
916 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
919 ret
= sk_attach_filter(&fprog
, sk
);
925 if (optlen
== sizeof(u32
)) {
929 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
932 ret
= sk_attach_bpf(ufd
, sk
);
936 case SO_ATTACH_REUSEPORT_CBPF
:
938 if (optlen
== sizeof(struct sock_fprog
)) {
939 struct sock_fprog fprog
;
942 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
945 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
949 case SO_ATTACH_REUSEPORT_EBPF
:
951 if (optlen
== sizeof(u32
)) {
955 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
958 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
962 case SO_DETACH_FILTER
:
963 ret
= sk_detach_filter(sk
);
967 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
970 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
975 set_bit(SOCK_PASSSEC
, &sock
->flags
);
977 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
980 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
987 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
991 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
995 if (sock
->ops
->set_peek_off
)
996 ret
= sock
->ops
->set_peek_off(sk
, val
);
1002 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
1005 case SO_SELECT_ERR_QUEUE
:
1006 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
1009 #ifdef CONFIG_NET_RX_BUSY_POLL
1011 /* allow unprivileged users to decrease the value */
1012 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
1018 sk
->sk_ll_usec
= val
;
1023 case SO_MAX_PACING_RATE
:
1024 sk
->sk_max_pacing_rate
= val
;
1025 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
1026 sk
->sk_max_pacing_rate
);
1029 case SO_INCOMING_CPU
:
1030 sk
->sk_incoming_cpu
= val
;
1040 EXPORT_SYMBOL(sock_setsockopt
);
1043 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1044 struct ucred
*ucred
)
1046 ucred
->pid
= pid_vnr(pid
);
1047 ucred
->uid
= ucred
->gid
= -1;
1049 struct user_namespace
*current_ns
= current_user_ns();
1051 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1052 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1056 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1057 char __user
*optval
, int __user
*optlen
)
1059 struct sock
*sk
= sock
->sk
;
1067 int lv
= sizeof(int);
1070 if (get_user(len
, optlen
))
1075 memset(&v
, 0, sizeof(v
));
1079 v
.val
= sock_flag(sk
, SOCK_DBG
);
1083 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1087 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1091 v
.val
= sk
->sk_sndbuf
;
1095 v
.val
= sk
->sk_rcvbuf
;
1099 v
.val
= sk
->sk_reuse
;
1103 v
.val
= sk
->sk_reuseport
;
1107 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1111 v
.val
= sk
->sk_type
;
1115 v
.val
= sk
->sk_protocol
;
1119 v
.val
= sk
->sk_family
;
1123 v
.val
= -sock_error(sk
);
1125 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1129 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1133 v
.val
= sk
->sk_no_check_tx
;
1137 v
.val
= sk
->sk_priority
;
1141 lv
= sizeof(v
.ling
);
1142 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1143 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1147 sock_warn_obsolete_bsdism("getsockopt");
1151 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1152 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1155 case SO_TIMESTAMPNS
:
1156 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1159 case SO_TIMESTAMPING
:
1160 v
.val
= sk
->sk_tsflags
;
1164 lv
= sizeof(struct timeval
);
1165 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1169 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1170 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1175 lv
= sizeof(struct timeval
);
1176 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1180 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1181 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1186 v
.val
= sk
->sk_rcvlowat
;
1194 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1199 struct ucred peercred
;
1200 if (len
> sizeof(peercred
))
1201 len
= sizeof(peercred
);
1202 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1203 if (copy_to_user(optval
, &peercred
, len
))
1212 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1216 if (copy_to_user(optval
, address
, len
))
1221 /* Dubious BSD thing... Probably nobody even uses it, but
1222 * the UNIX standard wants it for whatever reason... -DaveM
1225 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1229 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1233 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1236 v
.val
= sk
->sk_mark
;
1240 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1243 case SO_WIFI_STATUS
:
1244 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1248 if (!sock
->ops
->set_peek_off
)
1251 v
.val
= sk
->sk_peek_off
;
1254 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1257 case SO_BINDTODEVICE
:
1258 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1261 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1267 case SO_LOCK_FILTER
:
1268 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1271 case SO_BPF_EXTENSIONS
:
1272 v
.val
= bpf_tell_extensions();
1275 case SO_SELECT_ERR_QUEUE
:
1276 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1279 #ifdef CONFIG_NET_RX_BUSY_POLL
1281 v
.val
= sk
->sk_ll_usec
;
1285 case SO_MAX_PACING_RATE
:
1286 v
.val
= sk
->sk_max_pacing_rate
;
1289 case SO_INCOMING_CPU
:
1290 v
.val
= sk
->sk_incoming_cpu
;
1294 /* We implement the SO_SNDLOWAT etc to not be settable
1297 return -ENOPROTOOPT
;
1302 if (copy_to_user(optval
, &v
, len
))
1305 if (put_user(len
, optlen
))
1311 * Initialize an sk_lock.
1313 * (We also register the sk_lock with the lock validator.)
1315 static inline void sock_lock_init(struct sock
*sk
)
1317 sock_lock_init_class_and_name(sk
,
1318 af_family_slock_key_strings
[sk
->sk_family
],
1319 af_family_slock_keys
+ sk
->sk_family
,
1320 af_family_key_strings
[sk
->sk_family
],
1321 af_family_keys
+ sk
->sk_family
);
1325 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1326 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1327 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1329 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1331 #ifdef CONFIG_SECURITY_NETWORK
1332 void *sptr
= nsk
->sk_security
;
1334 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1336 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1337 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1339 #ifdef CONFIG_SECURITY_NETWORK
1340 nsk
->sk_security
= sptr
;
1341 security_sk_clone(osk
, nsk
);
1345 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1347 unsigned long nulls1
, nulls2
;
1349 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1350 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1351 if (nulls1
> nulls2
)
1352 swap(nulls1
, nulls2
);
1355 memset((char *)sk
, 0, nulls1
);
1356 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1357 nulls2
- nulls1
- sizeof(void *));
1358 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1359 size
- nulls2
- sizeof(void *));
1361 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1363 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1367 struct kmem_cache
*slab
;
1371 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1374 if (priority
& __GFP_ZERO
) {
1376 prot
->clear_sk(sk
, prot
->obj_size
);
1378 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1381 sk
= kmalloc(prot
->obj_size
, priority
);
1384 kmemcheck_annotate_bitfield(sk
, flags
);
1386 if (security_sk_alloc(sk
, family
, priority
))
1389 if (!try_module_get(prot
->owner
))
1391 sk_tx_queue_clear(sk
);
1392 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1398 security_sk_free(sk
);
1401 kmem_cache_free(slab
, sk
);
1407 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1409 struct kmem_cache
*slab
;
1410 struct module
*owner
;
1412 owner
= prot
->owner
;
1415 cgroup_sk_free(&sk
->sk_cgrp_data
);
1416 security_sk_free(sk
);
1418 kmem_cache_free(slab
, sk
);
1425 * sk_alloc - All socket objects are allocated here
1426 * @net: the applicable net namespace
1427 * @family: protocol family
1428 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1429 * @prot: struct proto associated with this new sock instance
1430 * @kern: is this to be a kernel socket?
1432 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1433 struct proto
*prot
, int kern
)
1437 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1439 sk
->sk_family
= family
;
1441 * See comment in struct sock definition to understand
1442 * why we need sk_prot_creator -acme
1444 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1446 sk
->sk_net_refcnt
= kern
? 0 : 1;
1447 if (likely(sk
->sk_net_refcnt
))
1449 sock_net_set(sk
, net
);
1450 atomic_set(&sk
->sk_wmem_alloc
, 1);
1452 sock_update_classid(&sk
->sk_cgrp_data
);
1453 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1458 EXPORT_SYMBOL(sk_alloc
);
1460 void sk_destruct(struct sock
*sk
)
1462 struct sk_filter
*filter
;
1464 if (sk
->sk_destruct
)
1465 sk
->sk_destruct(sk
);
1467 filter
= rcu_dereference_check(sk
->sk_filter
,
1468 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1470 sk_filter_uncharge(sk
, filter
);
1471 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1473 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1474 reuseport_detach_sock(sk
);
1476 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1478 if (atomic_read(&sk
->sk_omem_alloc
))
1479 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1480 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1482 if (sk
->sk_peer_cred
)
1483 put_cred(sk
->sk_peer_cred
);
1484 put_pid(sk
->sk_peer_pid
);
1485 if (likely(sk
->sk_net_refcnt
))
1486 put_net(sock_net(sk
));
1487 sk_prot_free(sk
->sk_prot_creator
, sk
);
1490 static void __sk_free(struct sock
*sk
)
1492 if (unlikely(sock_diag_has_destroy_listeners(sk
) && sk
->sk_net_refcnt
))
1493 sock_diag_broadcast_destroy(sk
);
1498 void sk_free(struct sock
*sk
)
1501 * We subtract one from sk_wmem_alloc and can know if
1502 * some packets are still in some tx queue.
1503 * If not null, sock_wfree() will call __sk_free(sk) later
1505 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1508 EXPORT_SYMBOL(sk_free
);
1511 * sk_clone_lock - clone a socket, and lock its clone
1512 * @sk: the socket to clone
1513 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1515 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1517 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1520 bool is_charged
= true;
1522 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1523 if (newsk
!= NULL
) {
1524 struct sk_filter
*filter
;
1526 sock_copy(newsk
, sk
);
1529 if (likely(newsk
->sk_net_refcnt
))
1530 get_net(sock_net(newsk
));
1531 sk_node_init(&newsk
->sk_node
);
1532 sock_lock_init(newsk
);
1533 bh_lock_sock(newsk
);
1534 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1535 newsk
->sk_backlog
.len
= 0;
1537 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1539 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1541 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1542 atomic_set(&newsk
->sk_omem_alloc
, 0);
1543 skb_queue_head_init(&newsk
->sk_receive_queue
);
1544 skb_queue_head_init(&newsk
->sk_write_queue
);
1546 rwlock_init(&newsk
->sk_callback_lock
);
1547 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1548 af_callback_keys
+ newsk
->sk_family
,
1549 af_family_clock_key_strings
[newsk
->sk_family
]);
1551 newsk
->sk_dst_cache
= NULL
;
1552 newsk
->sk_wmem_queued
= 0;
1553 newsk
->sk_forward_alloc
= 0;
1554 newsk
->sk_send_head
= NULL
;
1555 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1557 sock_reset_flag(newsk
, SOCK_DONE
);
1558 skb_queue_head_init(&newsk
->sk_error_queue
);
1560 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1562 /* though it's an empty new sock, the charging may fail
1563 * if sysctl_optmem_max was changed between creation of
1564 * original socket and cloning
1566 is_charged
= sk_filter_charge(newsk
, filter
);
1568 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1569 /* It is still raw copy of parent, so invalidate
1570 * destructor and make plain sk_free() */
1571 newsk
->sk_destruct
= NULL
;
1572 bh_unlock_sock(newsk
);
1579 newsk
->sk_priority
= 0;
1580 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1581 atomic64_set(&newsk
->sk_cookie
, 0);
1583 * Before updating sk_refcnt, we must commit prior changes to memory
1584 * (Documentation/RCU/rculist_nulls.txt for details)
1587 atomic_set(&newsk
->sk_refcnt
, 2);
1590 * Increment the counter in the same struct proto as the master
1591 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1592 * is the same as sk->sk_prot->socks, as this field was copied
1595 * This _changes_ the previous behaviour, where
1596 * tcp_create_openreq_child always was incrementing the
1597 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1598 * to be taken into account in all callers. -acme
1600 sk_refcnt_debug_inc(newsk
);
1601 sk_set_socket(newsk
, NULL
);
1602 newsk
->sk_wq
= NULL
;
1604 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1605 sock_update_memcg(newsk
);
1607 if (newsk
->sk_prot
->sockets_allocated
)
1608 sk_sockets_allocated_inc(newsk
);
1610 if (sock_needs_netstamp(sk
) &&
1611 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1612 net_enable_timestamp();
1617 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1619 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1623 sk_dst_set(sk
, dst
);
1624 sk
->sk_route_caps
= dst
->dev
->features
;
1625 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1626 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1627 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1628 if (sk_can_gso(sk
)) {
1629 if (dst
->header_len
) {
1630 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1632 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1633 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1634 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1637 sk
->sk_gso_max_segs
= max_segs
;
1639 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1642 * Simple resource managers for sockets.
1647 * Write buffer destructor automatically called from kfree_skb.
1649 void sock_wfree(struct sk_buff
*skb
)
1651 struct sock
*sk
= skb
->sk
;
1652 unsigned int len
= skb
->truesize
;
1654 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1656 * Keep a reference on sk_wmem_alloc, this will be released
1657 * after sk_write_space() call
1659 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1660 sk
->sk_write_space(sk
);
1664 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1665 * could not do because of in-flight packets
1667 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1670 EXPORT_SYMBOL(sock_wfree
);
1672 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1677 if (unlikely(!sk_fullsock(sk
))) {
1678 skb
->destructor
= sock_edemux
;
1683 skb
->destructor
= sock_wfree
;
1684 skb_set_hash_from_sk(skb
, sk
);
1686 * We used to take a refcount on sk, but following operation
1687 * is enough to guarantee sk_free() wont free this sock until
1688 * all in-flight packets are completed
1690 atomic_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1692 EXPORT_SYMBOL(skb_set_owner_w
);
1694 void skb_orphan_partial(struct sk_buff
*skb
)
1696 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1697 * so we do not completely orphan skb, but transfert all
1698 * accounted bytes but one, to avoid unexpected reorders.
1700 if (skb
->destructor
== sock_wfree
1702 || skb
->destructor
== tcp_wfree
1705 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1711 EXPORT_SYMBOL(skb_orphan_partial
);
1714 * Read buffer destructor automatically called from kfree_skb.
1716 void sock_rfree(struct sk_buff
*skb
)
1718 struct sock
*sk
= skb
->sk
;
1719 unsigned int len
= skb
->truesize
;
1721 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1722 sk_mem_uncharge(sk
, len
);
1724 EXPORT_SYMBOL(sock_rfree
);
1727 * Buffer destructor for skbs that are not used directly in read or write
1728 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1730 void sock_efree(struct sk_buff
*skb
)
1734 EXPORT_SYMBOL(sock_efree
);
1736 kuid_t
sock_i_uid(struct sock
*sk
)
1740 read_lock_bh(&sk
->sk_callback_lock
);
1741 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1742 read_unlock_bh(&sk
->sk_callback_lock
);
1745 EXPORT_SYMBOL(sock_i_uid
);
1747 unsigned long sock_i_ino(struct sock
*sk
)
1751 read_lock_bh(&sk
->sk_callback_lock
);
1752 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1753 read_unlock_bh(&sk
->sk_callback_lock
);
1756 EXPORT_SYMBOL(sock_i_ino
);
1759 * Allocate a skb from the socket's send buffer.
1761 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1764 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1765 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1767 skb_set_owner_w(skb
, sk
);
1773 EXPORT_SYMBOL(sock_wmalloc
);
1776 * Allocate a memory block from the socket's option memory buffer.
1778 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1780 if ((unsigned int)size
<= sysctl_optmem_max
&&
1781 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1783 /* First do the add, to avoid the race if kmalloc
1786 atomic_add(size
, &sk
->sk_omem_alloc
);
1787 mem
= kmalloc(size
, priority
);
1790 atomic_sub(size
, &sk
->sk_omem_alloc
);
1794 EXPORT_SYMBOL(sock_kmalloc
);
1796 /* Free an option memory block. Note, we actually want the inline
1797 * here as this allows gcc to detect the nullify and fold away the
1798 * condition entirely.
1800 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1803 if (WARN_ON_ONCE(!mem
))
1809 atomic_sub(size
, &sk
->sk_omem_alloc
);
1812 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1814 __sock_kfree_s(sk
, mem
, size
, false);
1816 EXPORT_SYMBOL(sock_kfree_s
);
1818 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1820 __sock_kfree_s(sk
, mem
, size
, true);
1822 EXPORT_SYMBOL(sock_kzfree_s
);
1824 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1825 I think, these locks should be removed for datagram sockets.
1827 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1831 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1835 if (signal_pending(current
))
1837 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1838 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1839 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1841 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1845 timeo
= schedule_timeout(timeo
);
1847 finish_wait(sk_sleep(sk
), &wait
);
1853 * Generic send/receive buffer handlers
1856 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1857 unsigned long data_len
, int noblock
,
1858 int *errcode
, int max_page_order
)
1860 struct sk_buff
*skb
;
1864 timeo
= sock_sndtimeo(sk
, noblock
);
1866 err
= sock_error(sk
);
1871 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1874 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1877 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1878 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1882 if (signal_pending(current
))
1884 timeo
= sock_wait_for_wmem(sk
, timeo
);
1886 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
1887 errcode
, sk
->sk_allocation
);
1889 skb_set_owner_w(skb
, sk
);
1893 err
= sock_intr_errno(timeo
);
1898 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1900 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1901 int noblock
, int *errcode
)
1903 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1905 EXPORT_SYMBOL(sock_alloc_send_skb
);
1907 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
1908 struct sockcm_cookie
*sockc
)
1910 struct cmsghdr
*cmsg
;
1912 for_each_cmsghdr(cmsg
, msg
) {
1913 if (!CMSG_OK(msg
, cmsg
))
1915 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
1917 switch (cmsg
->cmsg_type
) {
1919 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
1921 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1923 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
1931 EXPORT_SYMBOL(sock_cmsg_send
);
1933 /* On 32bit arches, an skb frag is limited to 2^15 */
1934 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1937 * skb_page_frag_refill - check that a page_frag contains enough room
1938 * @sz: minimum size of the fragment we want to get
1939 * @pfrag: pointer to page_frag
1940 * @gfp: priority for memory allocation
1942 * Note: While this allocator tries to use high order pages, there is
1943 * no guarantee that allocations succeed. Therefore, @sz MUST be
1944 * less or equal than PAGE_SIZE.
1946 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
1949 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1953 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1955 put_page(pfrag
->page
);
1959 if (SKB_FRAG_PAGE_ORDER
) {
1960 /* Avoid direct reclaim but allow kswapd to wake */
1961 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
1962 __GFP_COMP
| __GFP_NOWARN
|
1964 SKB_FRAG_PAGE_ORDER
);
1965 if (likely(pfrag
->page
)) {
1966 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
1970 pfrag
->page
= alloc_page(gfp
);
1971 if (likely(pfrag
->page
)) {
1972 pfrag
->size
= PAGE_SIZE
;
1977 EXPORT_SYMBOL(skb_page_frag_refill
);
1979 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1981 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
1984 sk_enter_memory_pressure(sk
);
1985 sk_stream_moderate_sndbuf(sk
);
1988 EXPORT_SYMBOL(sk_page_frag_refill
);
1990 static void __lock_sock(struct sock
*sk
)
1991 __releases(&sk
->sk_lock
.slock
)
1992 __acquires(&sk
->sk_lock
.slock
)
1997 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1998 TASK_UNINTERRUPTIBLE
);
1999 spin_unlock_bh(&sk
->sk_lock
.slock
);
2001 spin_lock_bh(&sk
->sk_lock
.slock
);
2002 if (!sock_owned_by_user(sk
))
2005 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2008 static void __release_sock(struct sock
*sk
)
2009 __releases(&sk
->sk_lock
.slock
)
2010 __acquires(&sk
->sk_lock
.slock
)
2012 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
2015 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2019 struct sk_buff
*next
= skb
->next
;
2022 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2024 sk_backlog_rcv(sk
, skb
);
2027 * We are in process context here with softirqs
2028 * disabled, use cond_resched_softirq() to preempt.
2029 * This is safe to do because we've taken the backlog
2032 cond_resched_softirq();
2035 } while (skb
!= NULL
);
2038 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
2041 * Doing the zeroing here guarantee we can not loop forever
2042 * while a wild producer attempts to flood us.
2044 sk
->sk_backlog
.len
= 0;
2048 * sk_wait_data - wait for data to arrive at sk_receive_queue
2049 * @sk: sock to wait on
2050 * @timeo: for how long
2051 * @skb: last skb seen on sk_receive_queue
2053 * Now socket state including sk->sk_err is changed only under lock,
2054 * hence we may omit checks after joining wait queue.
2055 * We check receive queue before schedule() only as optimization;
2056 * it is very likely that release_sock() added new data.
2058 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2063 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2064 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2065 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
);
2066 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2067 finish_wait(sk_sleep(sk
), &wait
);
2070 EXPORT_SYMBOL(sk_wait_data
);
2073 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2075 * @size: memory size to allocate
2076 * @kind: allocation type
2078 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2079 * rmem allocation. This function assumes that protocols which have
2080 * memory_pressure use sk_wmem_queued as write buffer accounting.
2082 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2084 struct proto
*prot
= sk
->sk_prot
;
2085 int amt
= sk_mem_pages(size
);
2087 int parent_status
= UNDER_LIMIT
;
2089 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2091 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
2094 if (parent_status
== UNDER_LIMIT
&&
2095 allocated
<= sk_prot_mem_limits(sk
, 0)) {
2096 sk_leave_memory_pressure(sk
);
2100 /* Under pressure. (we or our parents) */
2101 if ((parent_status
> SOFT_LIMIT
) ||
2102 allocated
> sk_prot_mem_limits(sk
, 1))
2103 sk_enter_memory_pressure(sk
);
2105 /* Over hard limit (we or our parents) */
2106 if ((parent_status
== OVER_LIMIT
) ||
2107 (allocated
> sk_prot_mem_limits(sk
, 2)))
2108 goto suppress_allocation
;
2110 /* guarantee minimum buffer size under pressure */
2111 if (kind
== SK_MEM_RECV
) {
2112 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2115 } else { /* SK_MEM_SEND */
2116 if (sk
->sk_type
== SOCK_STREAM
) {
2117 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2119 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2120 prot
->sysctl_wmem
[0])
2124 if (sk_has_memory_pressure(sk
)) {
2127 if (!sk_under_memory_pressure(sk
))
2129 alloc
= sk_sockets_allocated_read_positive(sk
);
2130 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2131 sk_mem_pages(sk
->sk_wmem_queued
+
2132 atomic_read(&sk
->sk_rmem_alloc
) +
2133 sk
->sk_forward_alloc
))
2137 suppress_allocation
:
2139 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2140 sk_stream_moderate_sndbuf(sk
);
2142 /* Fail only if socket is _under_ its sndbuf.
2143 * In this case we cannot block, so that we have to fail.
2145 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2149 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2151 /* Alas. Undo changes. */
2152 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2154 sk_memory_allocated_sub(sk
, amt
);
2158 EXPORT_SYMBOL(__sk_mem_schedule
);
2161 * __sk_mem_reclaim - reclaim memory_allocated
2163 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2165 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2167 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2168 sk_memory_allocated_sub(sk
, amount
);
2169 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2171 if (sk_under_memory_pressure(sk
) &&
2172 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2173 sk_leave_memory_pressure(sk
);
2175 EXPORT_SYMBOL(__sk_mem_reclaim
);
2179 * Set of default routines for initialising struct proto_ops when
2180 * the protocol does not support a particular function. In certain
2181 * cases where it makes no sense for a protocol to have a "do nothing"
2182 * function, some default processing is provided.
2185 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2189 EXPORT_SYMBOL(sock_no_bind
);
2191 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2196 EXPORT_SYMBOL(sock_no_connect
);
2198 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2202 EXPORT_SYMBOL(sock_no_socketpair
);
2204 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2208 EXPORT_SYMBOL(sock_no_accept
);
2210 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2215 EXPORT_SYMBOL(sock_no_getname
);
2217 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2221 EXPORT_SYMBOL(sock_no_poll
);
2223 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2227 EXPORT_SYMBOL(sock_no_ioctl
);
2229 int sock_no_listen(struct socket
*sock
, int backlog
)
2233 EXPORT_SYMBOL(sock_no_listen
);
2235 int sock_no_shutdown(struct socket
*sock
, int how
)
2239 EXPORT_SYMBOL(sock_no_shutdown
);
2241 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2242 char __user
*optval
, unsigned int optlen
)
2246 EXPORT_SYMBOL(sock_no_setsockopt
);
2248 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2249 char __user
*optval
, int __user
*optlen
)
2253 EXPORT_SYMBOL(sock_no_getsockopt
);
2255 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2259 EXPORT_SYMBOL(sock_no_sendmsg
);
2261 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2266 EXPORT_SYMBOL(sock_no_recvmsg
);
2268 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2270 /* Mirror missing mmap method error code */
2273 EXPORT_SYMBOL(sock_no_mmap
);
2275 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2278 struct msghdr msg
= {.msg_flags
= flags
};
2280 char *kaddr
= kmap(page
);
2281 iov
.iov_base
= kaddr
+ offset
;
2283 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2287 EXPORT_SYMBOL(sock_no_sendpage
);
2290 * Default Socket Callbacks
2293 static void sock_def_wakeup(struct sock
*sk
)
2295 struct socket_wq
*wq
;
2298 wq
= rcu_dereference(sk
->sk_wq
);
2299 if (skwq_has_sleeper(wq
))
2300 wake_up_interruptible_all(&wq
->wait
);
2304 static void sock_def_error_report(struct sock
*sk
)
2306 struct socket_wq
*wq
;
2309 wq
= rcu_dereference(sk
->sk_wq
);
2310 if (skwq_has_sleeper(wq
))
2311 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2312 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2316 static void sock_def_readable(struct sock
*sk
)
2318 struct socket_wq
*wq
;
2321 wq
= rcu_dereference(sk
->sk_wq
);
2322 if (skwq_has_sleeper(wq
))
2323 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2324 POLLRDNORM
| POLLRDBAND
);
2325 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2329 static void sock_def_write_space(struct sock
*sk
)
2331 struct socket_wq
*wq
;
2335 /* Do not wake up a writer until he can make "significant"
2338 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2339 wq
= rcu_dereference(sk
->sk_wq
);
2340 if (skwq_has_sleeper(wq
))
2341 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2342 POLLWRNORM
| POLLWRBAND
);
2344 /* Should agree with poll, otherwise some programs break */
2345 if (sock_writeable(sk
))
2346 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2352 static void sock_def_destruct(struct sock
*sk
)
2356 void sk_send_sigurg(struct sock
*sk
)
2358 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2359 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2360 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2362 EXPORT_SYMBOL(sk_send_sigurg
);
2364 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2365 unsigned long expires
)
2367 if (!mod_timer(timer
, expires
))
2370 EXPORT_SYMBOL(sk_reset_timer
);
2372 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2374 if (del_timer(timer
))
2377 EXPORT_SYMBOL(sk_stop_timer
);
2379 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2381 skb_queue_head_init(&sk
->sk_receive_queue
);
2382 skb_queue_head_init(&sk
->sk_write_queue
);
2383 skb_queue_head_init(&sk
->sk_error_queue
);
2385 sk
->sk_send_head
= NULL
;
2387 init_timer(&sk
->sk_timer
);
2389 sk
->sk_allocation
= GFP_KERNEL
;
2390 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2391 sk
->sk_sndbuf
= sysctl_wmem_default
;
2392 sk
->sk_state
= TCP_CLOSE
;
2393 sk_set_socket(sk
, sock
);
2395 sock_set_flag(sk
, SOCK_ZAPPED
);
2398 sk
->sk_type
= sock
->type
;
2399 sk
->sk_wq
= sock
->wq
;
2404 rwlock_init(&sk
->sk_callback_lock
);
2405 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2406 af_callback_keys
+ sk
->sk_family
,
2407 af_family_clock_key_strings
[sk
->sk_family
]);
2409 sk
->sk_state_change
= sock_def_wakeup
;
2410 sk
->sk_data_ready
= sock_def_readable
;
2411 sk
->sk_write_space
= sock_def_write_space
;
2412 sk
->sk_error_report
= sock_def_error_report
;
2413 sk
->sk_destruct
= sock_def_destruct
;
2415 sk
->sk_frag
.page
= NULL
;
2416 sk
->sk_frag
.offset
= 0;
2417 sk
->sk_peek_off
= -1;
2419 sk
->sk_peer_pid
= NULL
;
2420 sk
->sk_peer_cred
= NULL
;
2421 sk
->sk_write_pending
= 0;
2422 sk
->sk_rcvlowat
= 1;
2423 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2424 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2426 sk
->sk_stamp
= ktime_set(-1L, 0);
2428 #ifdef CONFIG_NET_RX_BUSY_POLL
2430 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2433 sk
->sk_max_pacing_rate
= ~0U;
2434 sk
->sk_pacing_rate
= ~0U;
2435 sk
->sk_incoming_cpu
= -1;
2437 * Before updating sk_refcnt, we must commit prior changes to memory
2438 * (Documentation/RCU/rculist_nulls.txt for details)
2441 atomic_set(&sk
->sk_refcnt
, 1);
2442 atomic_set(&sk
->sk_drops
, 0);
2444 EXPORT_SYMBOL(sock_init_data
);
2446 void lock_sock_nested(struct sock
*sk
, int subclass
)
2449 spin_lock_bh(&sk
->sk_lock
.slock
);
2450 if (sk
->sk_lock
.owned
)
2452 sk
->sk_lock
.owned
= 1;
2453 spin_unlock(&sk
->sk_lock
.slock
);
2455 * The sk_lock has mutex_lock() semantics here:
2457 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2460 EXPORT_SYMBOL(lock_sock_nested
);
2462 void release_sock(struct sock
*sk
)
2465 * The sk_lock has mutex_unlock() semantics:
2467 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2469 spin_lock_bh(&sk
->sk_lock
.slock
);
2470 if (sk
->sk_backlog
.tail
)
2473 /* Warning : release_cb() might need to release sk ownership,
2474 * ie call sock_release_ownership(sk) before us.
2476 if (sk
->sk_prot
->release_cb
)
2477 sk
->sk_prot
->release_cb(sk
);
2479 sock_release_ownership(sk
);
2480 if (waitqueue_active(&sk
->sk_lock
.wq
))
2481 wake_up(&sk
->sk_lock
.wq
);
2482 spin_unlock_bh(&sk
->sk_lock
.slock
);
2484 EXPORT_SYMBOL(release_sock
);
2487 * lock_sock_fast - fast version of lock_sock
2490 * This version should be used for very small section, where process wont block
2491 * return false if fast path is taken
2492 * sk_lock.slock locked, owned = 0, BH disabled
2493 * return true if slow path is taken
2494 * sk_lock.slock unlocked, owned = 1, BH enabled
2496 bool lock_sock_fast(struct sock
*sk
)
2499 spin_lock_bh(&sk
->sk_lock
.slock
);
2501 if (!sk
->sk_lock
.owned
)
2503 * Note : We must disable BH
2508 sk
->sk_lock
.owned
= 1;
2509 spin_unlock(&sk
->sk_lock
.slock
);
2511 * The sk_lock has mutex_lock() semantics here:
2513 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2517 EXPORT_SYMBOL(lock_sock_fast
);
2519 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2522 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2523 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2524 tv
= ktime_to_timeval(sk
->sk_stamp
);
2525 if (tv
.tv_sec
== -1)
2527 if (tv
.tv_sec
== 0) {
2528 sk
->sk_stamp
= ktime_get_real();
2529 tv
= ktime_to_timeval(sk
->sk_stamp
);
2531 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2533 EXPORT_SYMBOL(sock_get_timestamp
);
2535 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2538 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2539 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2540 ts
= ktime_to_timespec(sk
->sk_stamp
);
2541 if (ts
.tv_sec
== -1)
2543 if (ts
.tv_sec
== 0) {
2544 sk
->sk_stamp
= ktime_get_real();
2545 ts
= ktime_to_timespec(sk
->sk_stamp
);
2547 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2549 EXPORT_SYMBOL(sock_get_timestampns
);
2551 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2553 if (!sock_flag(sk
, flag
)) {
2554 unsigned long previous_flags
= sk
->sk_flags
;
2556 sock_set_flag(sk
, flag
);
2558 * we just set one of the two flags which require net
2559 * time stamping, but time stamping might have been on
2560 * already because of the other one
2562 if (sock_needs_netstamp(sk
) &&
2563 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
2564 net_enable_timestamp();
2568 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2569 int level
, int type
)
2571 struct sock_exterr_skb
*serr
;
2572 struct sk_buff
*skb
;
2576 skb
= sock_dequeue_err_skb(sk
);
2582 msg
->msg_flags
|= MSG_TRUNC
;
2585 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2589 sock_recv_timestamp(msg
, sk
, skb
);
2591 serr
= SKB_EXT_ERR(skb
);
2592 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2594 msg
->msg_flags
|= MSG_ERRQUEUE
;
2602 EXPORT_SYMBOL(sock_recv_errqueue
);
2605 * Get a socket option on an socket.
2607 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2608 * asynchronous errors should be reported by getsockopt. We assume
2609 * this means if you specify SO_ERROR (otherwise whats the point of it).
2611 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2612 char __user
*optval
, int __user
*optlen
)
2614 struct sock
*sk
= sock
->sk
;
2616 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2618 EXPORT_SYMBOL(sock_common_getsockopt
);
2620 #ifdef CONFIG_COMPAT
2621 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2622 char __user
*optval
, int __user
*optlen
)
2624 struct sock
*sk
= sock
->sk
;
2626 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2627 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2629 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2631 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2634 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2637 struct sock
*sk
= sock
->sk
;
2641 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2642 flags
& ~MSG_DONTWAIT
, &addr_len
);
2644 msg
->msg_namelen
= addr_len
;
2647 EXPORT_SYMBOL(sock_common_recvmsg
);
2650 * Set socket options on an inet socket.
2652 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2653 char __user
*optval
, unsigned int optlen
)
2655 struct sock
*sk
= sock
->sk
;
2657 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2659 EXPORT_SYMBOL(sock_common_setsockopt
);
2661 #ifdef CONFIG_COMPAT
2662 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2663 char __user
*optval
, unsigned int optlen
)
2665 struct sock
*sk
= sock
->sk
;
2667 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2668 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2670 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2672 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2675 void sk_common_release(struct sock
*sk
)
2677 if (sk
->sk_prot
->destroy
)
2678 sk
->sk_prot
->destroy(sk
);
2681 * Observation: when sock_common_release is called, processes have
2682 * no access to socket. But net still has.
2683 * Step one, detach it from networking:
2685 * A. Remove from hash tables.
2688 sk
->sk_prot
->unhash(sk
);
2691 * In this point socket cannot receive new packets, but it is possible
2692 * that some packets are in flight because some CPU runs receiver and
2693 * did hash table lookup before we unhashed socket. They will achieve
2694 * receive queue and will be purged by socket destructor.
2696 * Also we still have packets pending on receive queue and probably,
2697 * our own packets waiting in device queues. sock_destroy will drain
2698 * receive queue, but transmitted packets will delay socket destruction
2699 * until the last reference will be released.
2704 xfrm_sk_free_policy(sk
);
2706 sk_refcnt_debug_release(sk
);
2708 if (sk
->sk_frag
.page
) {
2709 put_page(sk
->sk_frag
.page
);
2710 sk
->sk_frag
.page
= NULL
;
2715 EXPORT_SYMBOL(sk_common_release
);
2717 #ifdef CONFIG_PROC_FS
2718 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2720 int val
[PROTO_INUSE_NR
];
2723 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2725 #ifdef CONFIG_NET_NS
2726 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2728 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2730 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2732 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2734 int cpu
, idx
= prot
->inuse_idx
;
2737 for_each_possible_cpu(cpu
)
2738 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2740 return res
>= 0 ? res
: 0;
2742 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2744 static int __net_init
sock_inuse_init_net(struct net
*net
)
2746 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2747 return net
->core
.inuse
? 0 : -ENOMEM
;
2750 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2752 free_percpu(net
->core
.inuse
);
2755 static struct pernet_operations net_inuse_ops
= {
2756 .init
= sock_inuse_init_net
,
2757 .exit
= sock_inuse_exit_net
,
2760 static __init
int net_inuse_init(void)
2762 if (register_pernet_subsys(&net_inuse_ops
))
2763 panic("Cannot initialize net inuse counters");
2768 core_initcall(net_inuse_init
);
2770 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2772 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2774 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2776 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2778 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2780 int cpu
, idx
= prot
->inuse_idx
;
2783 for_each_possible_cpu(cpu
)
2784 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2786 return res
>= 0 ? res
: 0;
2788 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2791 static void assign_proto_idx(struct proto
*prot
)
2793 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2795 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2796 pr_err("PROTO_INUSE_NR exhausted\n");
2800 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2803 static void release_proto_idx(struct proto
*prot
)
2805 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2806 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2809 static inline void assign_proto_idx(struct proto
*prot
)
2813 static inline void release_proto_idx(struct proto
*prot
)
2818 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
2822 kfree(rsk_prot
->slab_name
);
2823 rsk_prot
->slab_name
= NULL
;
2824 kmem_cache_destroy(rsk_prot
->slab
);
2825 rsk_prot
->slab
= NULL
;
2828 static int req_prot_init(const struct proto
*prot
)
2830 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
2835 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
2837 if (!rsk_prot
->slab_name
)
2840 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
2841 rsk_prot
->obj_size
, 0,
2842 prot
->slab_flags
, NULL
);
2844 if (!rsk_prot
->slab
) {
2845 pr_crit("%s: Can't create request sock SLAB cache!\n",
2852 int proto_register(struct proto
*prot
, int alloc_slab
)
2855 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2856 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2859 if (prot
->slab
== NULL
) {
2860 pr_crit("%s: Can't create sock SLAB cache!\n",
2865 if (req_prot_init(prot
))
2866 goto out_free_request_sock_slab
;
2868 if (prot
->twsk_prot
!= NULL
) {
2869 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2871 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2872 goto out_free_request_sock_slab
;
2874 prot
->twsk_prot
->twsk_slab
=
2875 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2876 prot
->twsk_prot
->twsk_obj_size
,
2880 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2881 goto out_free_timewait_sock_slab_name
;
2885 mutex_lock(&proto_list_mutex
);
2886 list_add(&prot
->node
, &proto_list
);
2887 assign_proto_idx(prot
);
2888 mutex_unlock(&proto_list_mutex
);
2891 out_free_timewait_sock_slab_name
:
2892 kfree(prot
->twsk_prot
->twsk_slab_name
);
2893 out_free_request_sock_slab
:
2894 req_prot_cleanup(prot
->rsk_prot
);
2896 kmem_cache_destroy(prot
->slab
);
2901 EXPORT_SYMBOL(proto_register
);
2903 void proto_unregister(struct proto
*prot
)
2905 mutex_lock(&proto_list_mutex
);
2906 release_proto_idx(prot
);
2907 list_del(&prot
->node
);
2908 mutex_unlock(&proto_list_mutex
);
2910 kmem_cache_destroy(prot
->slab
);
2913 req_prot_cleanup(prot
->rsk_prot
);
2915 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2916 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2917 kfree(prot
->twsk_prot
->twsk_slab_name
);
2918 prot
->twsk_prot
->twsk_slab
= NULL
;
2921 EXPORT_SYMBOL(proto_unregister
);
2923 #ifdef CONFIG_PROC_FS
2924 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2925 __acquires(proto_list_mutex
)
2927 mutex_lock(&proto_list_mutex
);
2928 return seq_list_start_head(&proto_list
, *pos
);
2931 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2933 return seq_list_next(v
, &proto_list
, pos
);
2936 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2937 __releases(proto_list_mutex
)
2939 mutex_unlock(&proto_list_mutex
);
2942 static char proto_method_implemented(const void *method
)
2944 return method
== NULL
? 'n' : 'y';
2946 static long sock_prot_memory_allocated(struct proto
*proto
)
2948 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2951 static char *sock_prot_memory_pressure(struct proto
*proto
)
2953 return proto
->memory_pressure
!= NULL
?
2954 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2957 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2960 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2961 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2964 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2965 sock_prot_memory_allocated(proto
),
2966 sock_prot_memory_pressure(proto
),
2968 proto
->slab
== NULL
? "no" : "yes",
2969 module_name(proto
->owner
),
2970 proto_method_implemented(proto
->close
),
2971 proto_method_implemented(proto
->connect
),
2972 proto_method_implemented(proto
->disconnect
),
2973 proto_method_implemented(proto
->accept
),
2974 proto_method_implemented(proto
->ioctl
),
2975 proto_method_implemented(proto
->init
),
2976 proto_method_implemented(proto
->destroy
),
2977 proto_method_implemented(proto
->shutdown
),
2978 proto_method_implemented(proto
->setsockopt
),
2979 proto_method_implemented(proto
->getsockopt
),
2980 proto_method_implemented(proto
->sendmsg
),
2981 proto_method_implemented(proto
->recvmsg
),
2982 proto_method_implemented(proto
->sendpage
),
2983 proto_method_implemented(proto
->bind
),
2984 proto_method_implemented(proto
->backlog_rcv
),
2985 proto_method_implemented(proto
->hash
),
2986 proto_method_implemented(proto
->unhash
),
2987 proto_method_implemented(proto
->get_port
),
2988 proto_method_implemented(proto
->enter_memory_pressure
));
2991 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2993 if (v
== &proto_list
)
2994 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3003 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3005 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3009 static const struct seq_operations proto_seq_ops
= {
3010 .start
= proto_seq_start
,
3011 .next
= proto_seq_next
,
3012 .stop
= proto_seq_stop
,
3013 .show
= proto_seq_show
,
3016 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
3018 return seq_open_net(inode
, file
, &proto_seq_ops
,
3019 sizeof(struct seq_net_private
));
3022 static const struct file_operations proto_seq_fops
= {
3023 .owner
= THIS_MODULE
,
3024 .open
= proto_seq_open
,
3026 .llseek
= seq_lseek
,
3027 .release
= seq_release_net
,
3030 static __net_init
int proto_init_net(struct net
*net
)
3032 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3038 static __net_exit
void proto_exit_net(struct net
*net
)
3040 remove_proc_entry("protocols", net
->proc_net
);
3044 static __net_initdata
struct pernet_operations proto_net_ops
= {
3045 .init
= proto_init_net
,
3046 .exit
= proto_exit_net
,
3049 static int __init
proto_init(void)
3051 return register_pernet_subsys(&proto_net_ops
);
3054 subsys_initcall(proto_init
);
3056 #endif /* PROC_FS */