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>
138 #include <net/af_unix.h>
144 #include <linux/xlog.h>
146 static DEFINE_MUTEX(proto_list_mutex
);
147 static LIST_HEAD(proto_list
);
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
159 bool sk_ns_capable(const struct sock
*sk
,
160 struct user_namespace
*user_ns
, int cap
)
162 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
163 ns_capable(user_ns
, cap
);
165 EXPORT_SYMBOL(sk_ns_capable
);
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capbility to use
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
176 bool sk_capable(const struct sock
*sk
, int cap
)
178 return sk_ns_capable(sk
, &init_user_ns
, cap
);
180 EXPORT_SYMBOL(sk_capable
);
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
187 * Test to see if the opener of the socket had when the socke was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
191 bool sk_net_capable(const struct sock
*sk
, int cap
)
193 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
195 EXPORT_SYMBOL(sk_net_capable
);
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
204 mutex_lock(&proto_list_mutex
);
205 list_for_each_entry(proto
, &proto_list
, node
) {
206 if (proto
->init_cgroup
) {
207 ret
= proto
->init_cgroup(memcg
, ss
);
213 mutex_unlock(&proto_list_mutex
);
216 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
217 if (proto
->destroy_cgroup
)
218 proto
->destroy_cgroup(memcg
);
219 mutex_unlock(&proto_list_mutex
);
223 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
227 mutex_lock(&proto_list_mutex
);
228 list_for_each_entry_reverse(proto
, &proto_list
, node
)
229 if (proto
->destroy_cgroup
)
230 proto
->destroy_cgroup(memcg
);
231 mutex_unlock(&proto_list_mutex
);
236 * Each address family might have different locking rules, so we have
237 * one slock key per address family:
239 static struct lock_class_key af_family_keys
[AF_MAX
];
240 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled
;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
248 * Make lock validator output more readable. (we pre-construct these
249 * strings build-time, so that runtime initialization of socket
252 static const char *const af_family_key_strings
[AF_MAX
+1] = {
253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
268 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
278 "slock-27" , "slock-28" , "slock-AF_CAN" ,
279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
284 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
294 "clock-27" , "clock-28" , "clock-AF_CAN" ,
295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
302 * sk_callback_lock locking rules are per-address-family,
303 * so split the lock classes by using a per-AF key:
305 static struct lock_class_key af_callback_keys
[AF_MAX
];
307 /* Take into consideration the size of the struct sk_buff overhead in the
308 * determination of these values, since that is non-constant across
309 * platforms. This makes socket queueing behavior and performance
310 * not depend upon such differences.
312 #define _SK_MEM_PACKETS 256
313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
319 EXPORT_SYMBOL(sysctl_wmem_max
);
320 __u32 sysctl_rmem_max __read_mostly
= (SK_RMEM_MAX
*8);
321 EXPORT_SYMBOL(sysctl_rmem_max
);
322 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
323 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
327 EXPORT_SYMBOL(sysctl_optmem_max
);
329 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
330 EXPORT_SYMBOL_GPL(memalloc_socks
);
333 * sk_set_memalloc - sets %SOCK_MEMALLOC
334 * @sk: socket to set it on
336 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
337 * It's the responsibility of the admin to adjust min_free_kbytes
338 * to meet the requirements
340 void sk_set_memalloc(struct sock
*sk
)
342 sock_set_flag(sk
, SOCK_MEMALLOC
);
343 sk
->sk_allocation
|= __GFP_MEMALLOC
;
344 static_key_slow_inc(&memalloc_socks
);
346 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
348 void sk_clear_memalloc(struct sock
*sk
)
350 sock_reset_flag(sk
, SOCK_MEMALLOC
);
351 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
352 static_key_slow_dec(&memalloc_socks
);
355 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
356 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
357 * it has rmem allocations there is a risk that the user of the
358 * socket cannot make forward progress due to exceeding the rmem
359 * limits. By rights, sk_clear_memalloc() should only be called
360 * on sockets being torn down but warn and reset the accounting if
361 * that assumption breaks.
363 if (WARN_ON(sk
->sk_forward_alloc
))
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 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
428 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
430 if (sk
->sk_flags
& flags
) {
431 sk
->sk_flags
&= ~flags
;
432 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
433 net_disable_timestamp();
438 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
443 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
445 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
446 atomic_inc(&sk
->sk_drops
);
447 trace_sock_rcvqueue_full(sk
, skb
);
451 err
= sk_filter(sk
, skb
);
455 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
456 atomic_inc(&sk
->sk_drops
);
461 skb_set_owner_r(skb
, sk
);
463 /* Cache the SKB length before we tack it onto the receive
464 * queue. Once it is added it no longer belongs to us and
465 * may be freed by other threads of control pulling packets
470 /* we escape from rcu protected region, make sure we dont leak
475 spin_lock_irqsave(&list
->lock
, flags
);
476 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
477 __skb_queue_tail(list
, skb
);
478 spin_unlock_irqrestore(&list
->lock
, flags
);
480 if (!sock_flag(sk
, SOCK_DEAD
))
481 sk
->sk_data_ready(sk
, skb_len
);
484 EXPORT_SYMBOL(sock_queue_rcv_skb
);
486 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
488 int rc
= NET_RX_SUCCESS
;
490 if (sk_filter(sk
, skb
))
491 goto discard_and_relse
;
495 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
)) {
496 atomic_inc(&sk
->sk_drops
);
497 goto discard_and_relse
;
500 bh_lock_sock_nested(sk
);
503 if (!sock_owned_by_user(sk
)) {
505 * trylock + unlock semantics:
507 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
509 rc
= sk_backlog_rcv(sk
, skb
);
511 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
512 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
514 atomic_inc(&sk
->sk_drops
);
515 goto discard_and_relse
;
526 EXPORT_SYMBOL(sk_receive_skb
);
528 void sk_reset_txq(struct sock
*sk
)
530 sk_tx_queue_clear(sk
);
532 EXPORT_SYMBOL(sk_reset_txq
);
534 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
536 struct dst_entry
*dst
= __sk_dst_get(sk
);
538 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
539 sk_tx_queue_clear(sk
);
540 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
547 EXPORT_SYMBOL(__sk_dst_check
);
549 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
551 struct dst_entry
*dst
= sk_dst_get(sk
);
553 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
561 EXPORT_SYMBOL(sk_dst_check
);
563 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
566 int ret
= -ENOPROTOOPT
;
567 #ifdef CONFIG_NETDEVICES
568 struct net
*net
= sock_net(sk
);
569 char devname
[IFNAMSIZ
];
574 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
581 /* Bind this socket to a particular device like "eth0",
582 * as specified in the passed interface name. If the
583 * name is "" or the option length is zero the socket
586 if (optlen
> IFNAMSIZ
- 1)
587 optlen
= IFNAMSIZ
- 1;
588 memset(devname
, 0, sizeof(devname
));
591 if (copy_from_user(devname
, optval
, optlen
))
595 if (devname
[0] != '\0') {
596 struct net_device
*dev
;
599 dev
= dev_get_by_name_rcu(net
, devname
);
601 index
= dev
->ifindex
;
609 sk
->sk_bound_dev_if
= index
;
621 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
622 int __user
*optlen
, int len
)
624 int ret
= -ENOPROTOOPT
;
625 #ifdef CONFIG_NETDEVICES
626 struct net
*net
= sock_net(sk
);
627 char devname
[IFNAMSIZ
];
629 if (sk
->sk_bound_dev_if
== 0) {
638 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
642 len
= strlen(devname
) + 1;
645 if (copy_to_user(optval
, devname
, len
))
650 if (put_user(len
, optlen
))
661 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
664 sock_set_flag(sk
, bit
);
666 sock_reset_flag(sk
, bit
);
670 * This is meant for all protocols to use and covers goings on
671 * at the socket level. Everything here is generic.
674 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
675 char __user
*optval
, unsigned int optlen
)
677 struct sock
*sk
= sock
->sk
;
684 * Options without arguments
687 if (optname
== SO_BINDTODEVICE
)
688 return sock_setbindtodevice(sk
, optval
, optlen
);
690 if (optlen
< sizeof(int))
693 if (get_user(val
, (int __user
*)optval
))
696 valbool
= val
? 1 : 0;
702 if (val
&& !capable(CAP_NET_ADMIN
))
705 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
708 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
711 sk
->sk_reuseport
= valbool
;
720 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
723 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
726 /* Don't error on this BSD doesn't and if you think
727 * about it this is right. Otherwise apps have to
728 * play 'guess the biggest size' games. RCVBUF/SNDBUF
729 * are treated in BSD as hints
731 val
= min_t(u32
, val
, sysctl_wmem_max
);
733 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
734 sk
->sk_sndbuf
= max_t(int, val
* 2, SOCK_MIN_SNDBUF
);
735 /* Wake up sending tasks if we upped the value. */
736 sk
->sk_write_space(sk
);
740 if (!capable(CAP_NET_ADMIN
)) {
747 /* Don't error on this BSD doesn't and if you think
748 * about it this is right. Otherwise apps have to
749 * play 'guess the biggest size' games. RCVBUF/SNDBUF
750 * are treated in BSD as hints
752 val
= min_t(u32
, val
, sysctl_rmem_max
);
754 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
756 * We double it on the way in to account for
757 * "struct sk_buff" etc. overhead. Applications
758 * assume that the SO_RCVBUF setting they make will
759 * allow that much actual data to be received on that
762 * Applications are unaware that "struct sk_buff" and
763 * other overheads allocate from the receive buffer
764 * during socket buffer allocation.
766 * And after considering the possible alternatives,
767 * returning the value we actually used in getsockopt
768 * is the most desirable behavior.
770 sk
->sk_rcvbuf
= max_t(int, val
* 2, SOCK_MIN_RCVBUF
);
774 if (!capable(CAP_NET_ADMIN
)) {
782 if (sk
->sk_protocol
== IPPROTO_TCP
&&
783 sk
->sk_type
== SOCK_STREAM
)
784 tcp_set_keepalive(sk
, valbool
);
786 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
790 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
794 sk
->sk_no_check
= valbool
;
798 if ((val
>= 0 && val
<= 6) ||
799 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
800 sk
->sk_priority
= val
;
806 if (optlen
< sizeof(ling
)) {
807 ret
= -EINVAL
; /* 1003.1g */
810 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
815 sock_reset_flag(sk
, SOCK_LINGER
);
817 #if (BITS_PER_LONG == 32)
818 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
819 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
822 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
823 sock_set_flag(sk
, SOCK_LINGER
);
828 sock_warn_obsolete_bsdism("setsockopt");
833 set_bit(SOCK_PASSCRED
, &sock
->flags
);
835 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
841 if (optname
== SO_TIMESTAMP
)
842 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
844 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
845 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
846 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
848 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
849 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
853 case SO_TIMESTAMPING
:
854 if (val
& ~SOF_TIMESTAMPING_MASK
) {
858 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
859 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
860 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
861 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
862 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
863 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
864 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
865 sock_enable_timestamp(sk
,
866 SOCK_TIMESTAMPING_RX_SOFTWARE
);
868 sock_disable_timestamp(sk
,
869 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
870 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
871 val
& SOF_TIMESTAMPING_SOFTWARE
);
872 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
873 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
874 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
875 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
881 sk
->sk_rcvlowat
= val
? : 1;
885 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
889 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
892 case SO_ATTACH_FILTER
:
894 if (optlen
== sizeof(struct sock_fprog
)) {
895 struct sock_fprog fprog
;
898 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
901 ret
= sk_attach_filter(&fprog
, sk
);
905 case SO_DETACH_FILTER
:
906 ret
= sk_detach_filter(sk
);
910 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
913 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
918 set_bit(SOCK_PASSSEC
, &sock
->flags
);
920 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
923 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
929 /* We implement the SO_SNDLOWAT etc to
930 not be settable (1003.1g 5.3) */
932 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
936 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
940 if (sock
->ops
->set_peek_off
)
941 ret
= sock
->ops
->set_peek_off(sk
, val
);
947 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
950 case SO_SELECT_ERR_QUEUE
:
951 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
961 EXPORT_SYMBOL(sock_setsockopt
);
964 void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
967 ucred
->pid
= pid_vnr(pid
);
968 ucred
->uid
= ucred
->gid
= -1;
970 struct user_namespace
*current_ns
= current_user_ns();
972 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
973 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
976 EXPORT_SYMBOL_GPL(cred_to_ucred
);
978 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
979 char __user
*optval
, int __user
*optlen
)
981 struct sock
*sk
= sock
->sk
;
989 int lv
= sizeof(int);
992 if (get_user(len
, optlen
))
997 memset(&v
, 0, sizeof(v
));
1001 v
.val
= sock_flag(sk
, SOCK_DBG
);
1005 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1009 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1013 v
.val
= sk
->sk_sndbuf
;
1017 v
.val
= sk
->sk_rcvbuf
;
1021 v
.val
= sk
->sk_reuse
;
1025 v
.val
= sk
->sk_reuseport
;
1029 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1033 v
.val
= sk
->sk_type
;
1037 v
.val
= sk
->sk_protocol
;
1041 v
.val
= sk
->sk_family
;
1045 v
.val
= -sock_error(sk
);
1047 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1051 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1055 v
.val
= sk
->sk_no_check
;
1059 v
.val
= sk
->sk_priority
;
1063 lv
= sizeof(v
.ling
);
1064 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1065 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1069 sock_warn_obsolete_bsdism("getsockopt");
1073 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1074 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1077 case SO_TIMESTAMPNS
:
1078 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1081 case SO_TIMESTAMPING
:
1083 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
1084 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
1085 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
1086 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
1087 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
1088 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
1089 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
1090 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
1091 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
1092 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
1093 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
1094 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
1095 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
1096 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
1100 lv
= sizeof(struct timeval
);
1101 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1105 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1106 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1111 lv
= sizeof(struct timeval
);
1112 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1116 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1117 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1122 v
.val
= sk
->sk_rcvlowat
;
1130 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1135 struct ucred peercred
;
1136 if (len
> sizeof(peercred
))
1137 len
= sizeof(peercred
);
1138 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1139 if (copy_to_user(optval
, &peercred
, len
))
1148 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1152 if (copy_to_user(optval
, address
, len
))
1157 /* Dubious BSD thing... Probably nobody even uses it, but
1158 * the UNIX standard wants it for whatever reason... -DaveM
1161 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1165 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1169 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1172 v
.val
= sk
->sk_mark
;
1176 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1179 case SO_WIFI_STATUS
:
1180 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1184 if (!sock
->ops
->set_peek_off
)
1187 v
.val
= sk
->sk_peek_off
;
1190 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1193 case SO_BINDTODEVICE
:
1194 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1197 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1203 case SO_LOCK_FILTER
:
1204 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1207 case SO_SELECT_ERR_QUEUE
:
1208 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1212 return -ENOPROTOOPT
;
1217 if (copy_to_user(optval
, &v
, len
))
1220 if (put_user(len
, optlen
))
1226 * Initialize an sk_lock.
1228 * (We also register the sk_lock with the lock validator.)
1230 static inline void sock_lock_init(struct sock
*sk
)
1232 sock_lock_init_class_and_name(sk
,
1233 af_family_slock_key_strings
[sk
->sk_family
],
1234 af_family_slock_keys
+ sk
->sk_family
,
1235 af_family_key_strings
[sk
->sk_family
],
1236 af_family_keys
+ sk
->sk_family
);
1240 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1241 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1242 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1244 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1246 #ifdef CONFIG_SECURITY_NETWORK
1247 void *sptr
= nsk
->sk_security
;
1249 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1251 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1252 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1254 #ifdef CONFIG_SECURITY_NETWORK
1255 nsk
->sk_security
= sptr
;
1256 security_sk_clone(osk
, nsk
);
1260 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1262 unsigned long nulls1
, nulls2
;
1264 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1265 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1266 if (nulls1
> nulls2
)
1267 swap(nulls1
, nulls2
);
1270 memset((char *)sk
, 0, nulls1
);
1271 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1272 nulls2
- nulls1
- sizeof(void *));
1273 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1274 size
- nulls2
- sizeof(void *));
1276 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1278 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1282 struct kmem_cache
*slab
;
1286 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1289 if (priority
& __GFP_ZERO
) {
1291 prot
->clear_sk(sk
, prot
->obj_size
);
1293 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1296 sk
= kmalloc(prot
->obj_size
, priority
);
1299 kmemcheck_annotate_bitfield(sk
, flags
);
1301 if (security_sk_alloc(sk
, family
, priority
))
1304 if (!try_module_get(prot
->owner
))
1306 sk_tx_queue_clear(sk
);
1312 security_sk_free(sk
);
1315 kmem_cache_free(slab
, sk
);
1321 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1323 struct kmem_cache
*slab
;
1324 struct module
*owner
;
1326 owner
= prot
->owner
;
1329 security_sk_free(sk
);
1331 kmem_cache_free(slab
, sk
);
1337 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1338 void sock_update_classid(struct sock
*sk
)
1342 classid
= task_cls_classid(current
);
1343 if (classid
!= sk
->sk_classid
)
1344 sk
->sk_classid
= classid
;
1346 EXPORT_SYMBOL(sock_update_classid
);
1349 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1350 void sock_update_netprioidx(struct sock
*sk
)
1355 sk
->sk_cgrp_prioidx
= task_netprioidx(current
);
1357 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1361 * sk_alloc - All socket objects are allocated here
1362 * @net: the applicable net namespace
1363 * @family: protocol family
1364 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1365 * @prot: struct proto associated with this new sock instance
1367 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1372 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1374 sk
->sk_family
= family
;
1376 * See comment in struct sock definition to understand
1377 * why we need sk_prot_creator -acme
1379 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1381 sock_net_set(sk
, get_net(net
));
1382 atomic_set(&sk
->sk_wmem_alloc
, 1);
1384 sock_update_classid(sk
);
1385 sock_update_netprioidx(sk
);
1390 EXPORT_SYMBOL(sk_alloc
);
1392 static void __sk_free(struct sock
*sk
)
1394 struct sk_filter
*filter
;
1396 if (sk
->sk_destruct
)
1397 sk
->sk_destruct(sk
);
1399 filter
= rcu_dereference_check(sk
->sk_filter
,
1400 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1402 sk_filter_uncharge(sk
, filter
);
1403 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1406 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1408 if (atomic_read(&sk
->sk_omem_alloc
))
1409 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1410 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1412 if (sk
->sk_peer_cred
)
1413 put_cred(sk
->sk_peer_cred
);
1414 put_pid(sk
->sk_peer_pid
);
1415 put_net(sock_net(sk
));
1416 sk_prot_free(sk
->sk_prot_creator
, sk
);
1419 void sk_free(struct sock
*sk
)
1422 * We subtract one from sk_wmem_alloc and can know if
1423 * some packets are still in some tx queue.
1424 * If not null, sock_wfree() will call __sk_free(sk) later
1426 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1429 EXPORT_SYMBOL(sk_free
);
1432 * Last sock_put should drop reference to sk->sk_net. It has already
1433 * been dropped in sk_change_net. Taking reference to stopping namespace
1435 * Take reference to a socket to remove it from hash _alive_ and after that
1436 * destroy it in the context of init_net.
1438 void sk_release_kernel(struct sock
*sk
)
1440 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1444 sock_release(sk
->sk_socket
);
1445 release_net(sock_net(sk
));
1446 sock_net_set(sk
, get_net(&init_net
));
1449 EXPORT_SYMBOL(sk_release_kernel
);
1451 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1453 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1454 sock_update_memcg(newsk
);
1458 * sk_clone_lock - clone a socket, and lock its clone
1459 * @sk: the socket to clone
1460 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1462 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1464 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1468 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1469 if (newsk
!= NULL
) {
1470 struct sk_filter
*filter
;
1472 sock_copy(newsk
, sk
);
1475 get_net(sock_net(newsk
));
1476 sk_node_init(&newsk
->sk_node
);
1477 sock_lock_init(newsk
);
1478 bh_lock_sock(newsk
);
1479 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1480 newsk
->sk_backlog
.len
= 0;
1482 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1484 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1486 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1487 atomic_set(&newsk
->sk_omem_alloc
, 0);
1488 skb_queue_head_init(&newsk
->sk_receive_queue
);
1489 skb_queue_head_init(&newsk
->sk_write_queue
);
1490 #ifdef CONFIG_NET_DMA
1491 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1494 spin_lock_init(&newsk
->sk_dst_lock
);
1495 rwlock_init(&newsk
->sk_callback_lock
);
1496 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1497 af_callback_keys
+ newsk
->sk_family
,
1498 af_family_clock_key_strings
[newsk
->sk_family
]);
1500 newsk
->sk_dst_cache
= NULL
;
1501 newsk
->sk_wmem_queued
= 0;
1502 newsk
->sk_forward_alloc
= 0;
1503 newsk
->sk_send_head
= NULL
;
1504 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1506 sock_reset_flag(newsk
, SOCK_DONE
);
1507 skb_queue_head_init(&newsk
->sk_error_queue
);
1509 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1511 sk_filter_charge(newsk
, filter
);
1513 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1514 /* It is still raw copy of parent, so invalidate
1515 * destructor and make plain sk_free() */
1516 newsk
->sk_destruct
= NULL
;
1517 bh_unlock_sock(newsk
);
1524 newsk
->sk_priority
= 0;
1526 * Before updating sk_refcnt, we must commit prior changes to memory
1527 * (Documentation/RCU/rculist_nulls.txt for details)
1530 atomic_set(&newsk
->sk_refcnt
, 2);
1533 * Increment the counter in the same struct proto as the master
1534 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1535 * is the same as sk->sk_prot->socks, as this field was copied
1538 * This _changes_ the previous behaviour, where
1539 * tcp_create_openreq_child always was incrementing the
1540 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1541 * to be taken into account in all callers. -acme
1543 sk_refcnt_debug_inc(newsk
);
1544 sk_set_socket(newsk
, NULL
);
1545 newsk
->sk_wq
= NULL
;
1547 sk_update_clone(sk
, newsk
);
1549 if (newsk
->sk_prot
->sockets_allocated
)
1550 sk_sockets_allocated_inc(newsk
);
1552 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1553 net_enable_timestamp();
1558 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1560 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1562 __sk_dst_set(sk
, dst
);
1563 sk
->sk_route_caps
= dst
->dev
->features
;
1564 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1565 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1566 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1567 if (sk_can_gso(sk
)) {
1568 if (dst
->header_len
) {
1569 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1571 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1572 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1573 sk
->sk_gso_max_segs
= dst
->dev
->gso_max_segs
;
1577 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1580 * Simple resource managers for sockets.
1585 * Write buffer destructor automatically called from kfree_skb.
1587 void sock_wfree(struct sk_buff
*skb
)
1589 struct sock
*sk
= skb
->sk
;
1590 unsigned int len
= skb
->truesize
;
1592 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1594 * Keep a reference on sk_wmem_alloc, this will be released
1595 * after sk_write_space() call
1597 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1598 sk
->sk_write_space(sk
);
1602 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1603 * could not do because of in-flight packets
1605 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1608 EXPORT_SYMBOL(sock_wfree
);
1611 * Read buffer destructor automatically called from kfree_skb.
1613 void sock_rfree(struct sk_buff
*skb
)
1615 struct sock
*sk
= skb
->sk
;
1616 unsigned int len
= skb
->truesize
;
1618 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1619 sk_mem_uncharge(sk
, len
);
1621 EXPORT_SYMBOL(sock_rfree
);
1623 void sock_edemux(struct sk_buff
*skb
)
1625 struct sock
*sk
= skb
->sk
;
1628 if (sk
->sk_state
== TCP_TIME_WAIT
)
1629 inet_twsk_put(inet_twsk(sk
));
1634 EXPORT_SYMBOL(sock_edemux
);
1636 kuid_t
sock_i_uid(struct sock
*sk
)
1640 /*mtk_net: fix kernel bug*/
1642 pr_info("sk == NULL for sock_i_uid\n");
1643 return GLOBAL_ROOT_UID
;
1646 read_lock_bh(&sk
->sk_callback_lock
);
1647 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1648 read_unlock_bh(&sk
->sk_callback_lock
);
1651 EXPORT_SYMBOL(sock_i_uid
);
1653 unsigned long sock_i_ino(struct sock
*sk
)
1657 read_lock_bh(&sk
->sk_callback_lock
);
1658 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1659 read_unlock_bh(&sk
->sk_callback_lock
);
1662 EXPORT_SYMBOL(sock_i_ino
);
1665 * Allocate a skb from the socket's send buffer.
1667 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1670 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1671 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1673 skb_set_owner_w(skb
, sk
);
1679 EXPORT_SYMBOL(sock_wmalloc
);
1682 * Allocate a skb from the socket's receive buffer.
1684 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1687 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1688 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1690 skb_set_owner_r(skb
, sk
);
1698 * Allocate a memory block from the socket's option memory buffer.
1700 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1702 if ((unsigned int)size
<= sysctl_optmem_max
&&
1703 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1705 /* First do the add, to avoid the race if kmalloc
1708 atomic_add(size
, &sk
->sk_omem_alloc
);
1709 mem
= kmalloc(size
, priority
);
1712 atomic_sub(size
, &sk
->sk_omem_alloc
);
1716 EXPORT_SYMBOL(sock_kmalloc
);
1719 * Free an option memory block.
1721 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1724 atomic_sub(size
, &sk
->sk_omem_alloc
);
1726 EXPORT_SYMBOL(sock_kfree_s
);
1728 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1729 I think, these locks should be removed for datagram sockets.
1731 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1735 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1739 if (signal_pending(current
))
1741 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1742 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1743 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1745 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1749 timeo
= schedule_timeout(timeo
);
1751 finish_wait(sk_sleep(sk
), &wait
);
1758 static int sock_dump_info(struct sock
*sk
)
1760 //dump receiver queue 128 bytes
1761 //struct sk_buff *skb;
1763 //dump receiver queue 128 bytes end
1765 if(sk
->sk_family
== AF_UNIX
)
1767 struct unix_sock
*u
= unix_sk(sk
);
1768 struct sock
*other
= NULL
;
1769 if( (u
->path
.dentry
!=NULL
)&&(u
->path
.dentry
->d_iname
!=NULL
))
1770 //if( (u->dentry !=NULL)&&(u->dentry->d_iname!=NULL))
1772 #ifdef CONFIG_MTK_NET_LOGGING
1773 printk(KERN_INFO
"[mtk_net][sock]sockdbg: socket-Name:%s \n",u
->path
.dentry
->d_iname
);
1778 #ifdef CONFIG_MTK_NET_LOGGING
1779 printk(KERN_INFO
"[mtk_net][sock]sockdbg:socket Name (NULL)\n" );
1783 if(sk
->sk_socket
&& SOCK_INODE(sk
->sk_socket
))
1785 #ifdef CONFIG_MTK_NET_LOGGING
1786 printk(KERN_INFO
"[mtk_net][sock]sockdbg:socket Inode[%lu]\n" ,SOCK_INODE(sk
->sk_socket
)->i_ino
);
1790 other
= unix_sk(sk
)->peer
;
1793 #ifdef CONFIG_MTK_NET_LOGGING
1794 printk(KERN_INFO
"[mtk_net][sock]sockdbg:peer is (NULL) \n");
1798 if ((((struct unix_sock
*)other
)->path
.dentry
!= NULL
)&&(((struct unix_sock
*)other
)->path
.dentry
->d_iname
!= NULL
))
1799 //if ((((struct unix_sock *)other)->dentry != NULL)&&(((struct unix_sock *)other)->dentry->d_iname != NULL))
1801 #ifdef CONFIG_MTK_NET_LOGGING
1802 printk(KERN_INFO
"[mtk_net][sock]sockdbg: Peer Name:%s \n",((struct unix_sock
*)other
)->path
.dentry
->d_iname
);
1807 #ifdef CONFIG_MTK_NET_LOGGING
1808 printk(KERN_INFO
"[mtk_net][sock]sockdbg: Peer Name (NULL) \n");
1812 if(other
->sk_socket
&& SOCK_INODE(other
->sk_socket
))
1814 #ifdef CONFIG_MTK_NET_LOGGING
1815 printk(KERN_INFO
"[mtk_net][sock]sockdbg: Peer Inode [%lu] \n", SOCK_INODE(other
->sk_socket
)->i_ino
);
1818 #ifdef CONFIG_MTK_NET_LOGGING
1819 printk(KERN_INFO
"[mtk_net][sock]sockdbg: Peer Recieve Queue len:%d \n",other
->sk_receive_queue
.qlen
);
1821 //dump receiver queue 128 bytes
1822 /* if ((skb = skb_peek_tail(&other->sk_receive_queue)) == NULL) {
1824 printk(KERN_INFO "sockdbg: Peer Recieve Queue is null (warning) \n");
1827 if((skb->len !=0) && (skb->data != NULL)){
1829 if(skb->len >= 127){
1836 sprintf(skbmsg+i, "%x", skb->data[i]);
1840 printk(KERN_INFO "sockdbg: Peer Recieve Queue dump(%d bytes):%s\n", len, skbmsg);
1844 printk(KERN_INFO "sockdbg: Peer Recieve skb error \n");
1846 //dump receiver queue 128 bytes end
1849 //dump receiver queue 128 bytes end
1862 * Generic send/receive buffer handlers
1865 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1866 unsigned long data_len
, int noblock
,
1869 struct sk_buff
*skb
;
1873 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1876 if (npages
> MAX_SKB_FRAGS
)
1879 gfp_mask
= sk
->sk_allocation
;
1880 if (gfp_mask
& __GFP_WAIT
)
1881 gfp_mask
|= __GFP_REPEAT
;
1883 timeo
= sock_sndtimeo(sk
, noblock
);
1885 err
= sock_error(sk
);
1890 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1893 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1894 skb
= alloc_skb(header_len
, gfp_mask
);
1898 /* No pages, we're done... */
1902 skb
->truesize
+= data_len
;
1903 skb_shinfo(skb
)->nr_frags
= npages
;
1904 for (i
= 0; i
< npages
; i
++) {
1907 page
= alloc_pages(sk
->sk_allocation
, 0);
1910 skb_shinfo(skb
)->nr_frags
= i
;
1915 __skb_fill_page_desc(skb
, i
,
1917 (data_len
>= PAGE_SIZE
?
1920 data_len
-= PAGE_SIZE
;
1923 /* Full success... */
1929 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1930 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1934 if (signal_pending(current
))
1938 #ifdef CONFIG_MTK_NET_LOGGING
1939 printk(KERN_INFO
"[mtk_net][sock]sockdbg: wait_for_wmem, timeo =%ld, wmem =%d, snd buf =%d \n",
1940 timeo
, atomic_read(&sk
->sk_wmem_alloc
), sk
->sk_sndbuf
);
1942 timeo
= sock_wait_for_wmem(sk
, timeo
);
1943 #ifdef CONFIG_MTK_NET_LOGGING
1944 printk(KERN_INFO
"[mtk_net][sock]sockdbg: wait_for_wmem done, header_len=0x%lx, data_len=0x%lx,timeo =%ld \n",
1945 header_len
, data_len
,timeo
);
1949 skb_set_owner_w(skb
, sk
);
1953 err
= sock_intr_errno(timeo
);
1958 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1960 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1961 int noblock
, int *errcode
)
1963 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1965 EXPORT_SYMBOL(sock_alloc_send_skb
);
1967 /* On 32bit arches, an skb frag is limited to 2^15 */
1968 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1970 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1975 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1979 if (pfrag
->offset
< pfrag
->size
)
1981 put_page(pfrag
->page
);
1984 /* We restrict high order allocations to users that can afford to wait */
1985 order
= (sk
->sk_allocation
& __GFP_WAIT
) ? SKB_FRAG_PAGE_ORDER
: 0;
1988 gfp_t gfp
= sk
->sk_allocation
;
1991 gfp
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
;
1992 pfrag
->page
= alloc_pages(gfp
, order
);
1993 if (likely(pfrag
->page
)) {
1995 pfrag
->size
= PAGE_SIZE
<< order
;
1998 } while (--order
>= 0);
2000 sk_enter_memory_pressure(sk
);
2001 sk_stream_moderate_sndbuf(sk
);
2004 EXPORT_SYMBOL(sk_page_frag_refill
);
2006 static void __lock_sock(struct sock
*sk
)
2007 __releases(&sk
->sk_lock
.slock
)
2008 __acquires(&sk
->sk_lock
.slock
)
2013 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2014 TASK_UNINTERRUPTIBLE
);
2015 spin_unlock_bh(&sk
->sk_lock
.slock
);
2017 spin_lock_bh(&sk
->sk_lock
.slock
);
2018 if (!sock_owned_by_user(sk
))
2021 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2024 static void __release_sock(struct sock
*sk
)
2025 __releases(&sk
->sk_lock
.slock
)
2026 __acquires(&sk
->sk_lock
.slock
)
2028 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
2031 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2035 struct sk_buff
*next
= skb
->next
;
2038 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2040 sk_backlog_rcv(sk
, skb
);
2043 * We are in process context here with softirqs
2044 * disabled, use cond_resched_softirq() to preempt.
2045 * This is safe to do because we've taken the backlog
2048 cond_resched_softirq();
2051 } while (skb
!= NULL
);
2054 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
2057 * Doing the zeroing here guarantee we can not loop forever
2058 * while a wild producer attempts to flood us.
2060 sk
->sk_backlog
.len
= 0;
2064 * sk_wait_data - wait for data to arrive at sk_receive_queue
2065 * @sk: sock to wait on
2066 * @timeo: for how long
2068 * Now socket state including sk->sk_err is changed only under lock,
2069 * hence we may omit checks after joining wait queue.
2070 * We check receive queue before schedule() only as optimization;
2071 * it is very likely that release_sock() added new data.
2073 int sk_wait_data(struct sock
*sk
, long *timeo
)
2078 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2079 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
2080 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
2081 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
2082 finish_wait(sk_sleep(sk
), &wait
);
2085 EXPORT_SYMBOL(sk_wait_data
);
2088 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2090 * @size: memory size to allocate
2091 * @kind: allocation type
2093 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2094 * rmem allocation. This function assumes that protocols which have
2095 * memory_pressure use sk_wmem_queued as write buffer accounting.
2097 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2099 struct proto
*prot
= sk
->sk_prot
;
2100 int amt
= sk_mem_pages(size
);
2102 int parent_status
= UNDER_LIMIT
;
2104 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2106 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
2109 if (parent_status
== UNDER_LIMIT
&&
2110 allocated
<= sk_prot_mem_limits(sk
, 0)) {
2111 sk_leave_memory_pressure(sk
);
2115 /* Under pressure. (we or our parents) */
2116 if ((parent_status
> SOFT_LIMIT
) ||
2117 allocated
> sk_prot_mem_limits(sk
, 1))
2118 sk_enter_memory_pressure(sk
);
2120 /* Over hard limit (we or our parents) */
2121 if ((parent_status
== OVER_LIMIT
) ||
2122 (allocated
> sk_prot_mem_limits(sk
, 2)))
2123 goto suppress_allocation
;
2125 /* guarantee minimum buffer size under pressure */
2126 if (kind
== SK_MEM_RECV
) {
2127 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2130 } else { /* SK_MEM_SEND */
2131 if (sk
->sk_type
== SOCK_STREAM
) {
2132 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2134 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2135 prot
->sysctl_wmem
[0])
2139 if (sk_has_memory_pressure(sk
)) {
2142 if (!sk_under_memory_pressure(sk
))
2144 alloc
= sk_sockets_allocated_read_positive(sk
);
2145 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2146 sk_mem_pages(sk
->sk_wmem_queued
+
2147 atomic_read(&sk
->sk_rmem_alloc
) +
2148 sk
->sk_forward_alloc
))
2152 suppress_allocation
:
2154 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2155 sk_stream_moderate_sndbuf(sk
);
2157 /* Fail only if socket is _under_ its sndbuf.
2158 * In this case we cannot block, so that we have to fail.
2160 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2164 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2166 /* Alas. Undo changes. */
2167 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2169 sk_memory_allocated_sub(sk
, amt
);
2173 EXPORT_SYMBOL(__sk_mem_schedule
);
2176 * __sk_reclaim - reclaim memory_allocated
2179 void __sk_mem_reclaim(struct sock
*sk
)
2181 sk_memory_allocated_sub(sk
,
2182 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
2183 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
2185 if (sk_under_memory_pressure(sk
) &&
2186 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2187 sk_leave_memory_pressure(sk
);
2189 EXPORT_SYMBOL(__sk_mem_reclaim
);
2193 * Set of default routines for initialising struct proto_ops when
2194 * the protocol does not support a particular function. In certain
2195 * cases where it makes no sense for a protocol to have a "do nothing"
2196 * function, some default processing is provided.
2199 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2203 EXPORT_SYMBOL(sock_no_bind
);
2205 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2210 EXPORT_SYMBOL(sock_no_connect
);
2212 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2216 EXPORT_SYMBOL(sock_no_socketpair
);
2218 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2222 EXPORT_SYMBOL(sock_no_accept
);
2224 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2229 EXPORT_SYMBOL(sock_no_getname
);
2231 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2235 EXPORT_SYMBOL(sock_no_poll
);
2237 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2241 EXPORT_SYMBOL(sock_no_ioctl
);
2243 int sock_no_listen(struct socket
*sock
, int backlog
)
2247 EXPORT_SYMBOL(sock_no_listen
);
2249 int sock_no_shutdown(struct socket
*sock
, int how
)
2253 EXPORT_SYMBOL(sock_no_shutdown
);
2255 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2256 char __user
*optval
, unsigned int optlen
)
2260 EXPORT_SYMBOL(sock_no_setsockopt
);
2262 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2263 char __user
*optval
, int __user
*optlen
)
2267 EXPORT_SYMBOL(sock_no_getsockopt
);
2269 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2274 EXPORT_SYMBOL(sock_no_sendmsg
);
2276 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2277 size_t len
, int flags
)
2281 EXPORT_SYMBOL(sock_no_recvmsg
);
2283 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2285 /* Mirror missing mmap method error code */
2288 EXPORT_SYMBOL(sock_no_mmap
);
2290 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2293 struct msghdr msg
= {.msg_flags
= flags
};
2295 char *kaddr
= kmap(page
);
2296 iov
.iov_base
= kaddr
+ offset
;
2298 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2302 EXPORT_SYMBOL(sock_no_sendpage
);
2305 * Default Socket Callbacks
2308 static void sock_def_wakeup(struct sock
*sk
)
2310 struct socket_wq
*wq
;
2313 wq
= rcu_dereference(sk
->sk_wq
);
2314 if (wq_has_sleeper(wq
))
2315 wake_up_interruptible_all(&wq
->wait
);
2319 static void sock_def_error_report(struct sock
*sk
)
2321 struct socket_wq
*wq
;
2324 wq
= rcu_dereference(sk
->sk_wq
);
2325 if (wq_has_sleeper(wq
))
2326 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2327 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2331 static void sock_def_readable(struct sock
*sk
, int len
)
2333 struct socket_wq
*wq
;
2336 wq
= rcu_dereference(sk
->sk_wq
);
2337 if (wq_has_sleeper(wq
))
2338 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2339 POLLRDNORM
| POLLRDBAND
);
2340 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2344 static void sock_def_write_space(struct sock
*sk
)
2346 struct socket_wq
*wq
;
2350 /* Do not wake up a writer until he can make "significant"
2353 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2354 wq
= rcu_dereference(sk
->sk_wq
);
2355 if (wq_has_sleeper(wq
))
2356 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2357 POLLWRNORM
| POLLWRBAND
);
2359 /* Should agree with poll, otherwise some programs break */
2360 if (sock_writeable(sk
))
2361 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2367 static void sock_def_destruct(struct sock
*sk
)
2369 kfree(sk
->sk_protinfo
);
2372 void sk_send_sigurg(struct sock
*sk
)
2374 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2375 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2376 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2378 EXPORT_SYMBOL(sk_send_sigurg
);
2380 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2381 unsigned long expires
)
2383 if (!mod_timer(timer
, expires
))
2386 EXPORT_SYMBOL(sk_reset_timer
);
2388 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2390 if (del_timer(timer
))
2393 EXPORT_SYMBOL(sk_stop_timer
);
2395 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2397 skb_queue_head_init(&sk
->sk_receive_queue
);
2398 skb_queue_head_init(&sk
->sk_write_queue
);
2399 skb_queue_head_init(&sk
->sk_error_queue
);
2400 #ifdef CONFIG_NET_DMA
2401 skb_queue_head_init(&sk
->sk_async_wait_queue
);
2404 sk
->sk_send_head
= NULL
;
2406 init_timer(&sk
->sk_timer
);
2408 sk
->sk_allocation
= GFP_KERNEL
;
2409 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2410 sk
->sk_sndbuf
= sysctl_wmem_default
;
2411 sk
->sk_state
= TCP_CLOSE
;
2412 sk_set_socket(sk
, sock
);
2414 sock_set_flag(sk
, SOCK_ZAPPED
);
2417 sk
->sk_type
= sock
->type
;
2418 sk
->sk_wq
= sock
->wq
;
2423 spin_lock_init(&sk
->sk_dst_lock
);
2424 rwlock_init(&sk
->sk_callback_lock
);
2425 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2426 af_callback_keys
+ sk
->sk_family
,
2427 af_family_clock_key_strings
[sk
->sk_family
]);
2429 sk
->sk_state_change
= sock_def_wakeup
;
2430 sk
->sk_data_ready
= sock_def_readable
;
2431 sk
->sk_write_space
= sock_def_write_space
;
2432 sk
->sk_error_report
= sock_def_error_report
;
2433 sk
->sk_destruct
= sock_def_destruct
;
2435 sk
->sk_frag
.page
= NULL
;
2436 sk
->sk_frag
.offset
= 0;
2437 sk
->sk_peek_off
= -1;
2439 sk
->sk_peer_pid
= NULL
;
2440 sk
->sk_peer_cred
= NULL
;
2441 sk
->sk_write_pending
= 0;
2442 sk
->sk_rcvlowat
= 1;
2443 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2444 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2446 sk
->sk_stamp
= ktime_set(-1L, 0);
2448 sk
->sk_pacing_rate
= ~0U;
2450 * Before updating sk_refcnt, we must commit prior changes to memory
2451 * (Documentation/RCU/rculist_nulls.txt for details)
2454 atomic_set(&sk
->sk_refcnt
, 1);
2455 atomic_set(&sk
->sk_drops
, 0);
2457 EXPORT_SYMBOL(sock_init_data
);
2459 void lock_sock_nested(struct sock
*sk
, int subclass
)
2462 spin_lock_bh(&sk
->sk_lock
.slock
);
2463 if (sk
->sk_lock
.owned
)
2465 sk
->sk_lock
.owned
= 1;
2466 spin_unlock(&sk
->sk_lock
.slock
);
2468 * The sk_lock has mutex_lock() semantics here:
2470 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2473 EXPORT_SYMBOL(lock_sock_nested
);
2475 void release_sock(struct sock
*sk
)
2478 * The sk_lock has mutex_unlock() semantics:
2480 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2482 spin_lock_bh(&sk
->sk_lock
.slock
);
2483 if (sk
->sk_backlog
.tail
)
2486 /* Warning : release_cb() might need to release sk ownership,
2487 * ie call sock_release_ownership(sk) before us.
2489 if (sk
->sk_prot
->release_cb
)
2490 sk
->sk_prot
->release_cb(sk
);
2492 sock_release_ownership(sk
);
2493 if (waitqueue_active(&sk
->sk_lock
.wq
))
2494 wake_up(&sk
->sk_lock
.wq
);
2495 spin_unlock_bh(&sk
->sk_lock
.slock
);
2497 EXPORT_SYMBOL(release_sock
);
2500 * lock_sock_fast - fast version of lock_sock
2503 * This version should be used for very small section, where process wont block
2504 * return false if fast path is taken
2505 * sk_lock.slock locked, owned = 0, BH disabled
2506 * return true if slow path is taken
2507 * sk_lock.slock unlocked, owned = 1, BH enabled
2509 bool lock_sock_fast(struct sock
*sk
)
2512 spin_lock_bh(&sk
->sk_lock
.slock
);
2514 if (!sk
->sk_lock
.owned
)
2516 * Note : We must disable BH
2521 sk
->sk_lock
.owned
= 1;
2522 spin_unlock(&sk
->sk_lock
.slock
);
2524 * The sk_lock has mutex_lock() semantics here:
2526 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2530 EXPORT_SYMBOL(lock_sock_fast
);
2532 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2535 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2536 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2537 tv
= ktime_to_timeval(sk
->sk_stamp
);
2538 if (tv
.tv_sec
== -1)
2540 if (tv
.tv_sec
== 0) {
2541 sk
->sk_stamp
= ktime_get_real();
2542 tv
= ktime_to_timeval(sk
->sk_stamp
);
2544 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2546 EXPORT_SYMBOL(sock_get_timestamp
);
2548 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2551 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2552 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2553 ts
= ktime_to_timespec(sk
->sk_stamp
);
2554 if (ts
.tv_sec
== -1)
2556 if (ts
.tv_sec
== 0) {
2557 sk
->sk_stamp
= ktime_get_real();
2558 ts
= ktime_to_timespec(sk
->sk_stamp
);
2560 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2562 EXPORT_SYMBOL(sock_get_timestampns
);
2564 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2566 if (!sock_flag(sk
, flag
)) {
2567 unsigned long previous_flags
= sk
->sk_flags
;
2569 sock_set_flag(sk
, flag
);
2571 * we just set one of the two flags which require net
2572 * time stamping, but time stamping might have been on
2573 * already because of the other one
2575 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2576 net_enable_timestamp();
2581 * Get a socket option on an socket.
2583 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2584 * asynchronous errors should be reported by getsockopt. We assume
2585 * this means if you specify SO_ERROR (otherwise whats the point of it).
2587 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2588 char __user
*optval
, int __user
*optlen
)
2590 struct sock
*sk
= sock
->sk
;
2592 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2594 EXPORT_SYMBOL(sock_common_getsockopt
);
2596 #ifdef CONFIG_COMPAT
2597 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2598 char __user
*optval
, int __user
*optlen
)
2600 struct sock
*sk
= sock
->sk
;
2602 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2603 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2605 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2607 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2610 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2611 struct msghdr
*msg
, size_t size
, int flags
)
2613 struct sock
*sk
= sock
->sk
;
2617 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2618 flags
& ~MSG_DONTWAIT
, &addr_len
);
2620 msg
->msg_namelen
= addr_len
;
2623 EXPORT_SYMBOL(sock_common_recvmsg
);
2626 * Set socket options on an inet socket.
2628 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2629 char __user
*optval
, unsigned int optlen
)
2631 struct sock
*sk
= sock
->sk
;
2633 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2635 EXPORT_SYMBOL(sock_common_setsockopt
);
2637 #ifdef CONFIG_COMPAT
2638 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2639 char __user
*optval
, unsigned int optlen
)
2641 struct sock
*sk
= sock
->sk
;
2643 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2644 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2646 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2648 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2651 void sk_common_release(struct sock
*sk
)
2653 if (sk
->sk_prot
->destroy
)
2654 sk
->sk_prot
->destroy(sk
);
2657 * Observation: when sock_common_release is called, processes have
2658 * no access to socket. But net still has.
2659 * Step one, detach it from networking:
2661 * A. Remove from hash tables.
2664 sk
->sk_prot
->unhash(sk
);
2667 * In this point socket cannot receive new packets, but it is possible
2668 * that some packets are in flight because some CPU runs receiver and
2669 * did hash table lookup before we unhashed socket. They will achieve
2670 * receive queue and will be purged by socket destructor.
2672 * Also we still have packets pending on receive queue and probably,
2673 * our own packets waiting in device queues. sock_destroy will drain
2674 * receive queue, but transmitted packets will delay socket destruction
2675 * until the last reference will be released.
2680 xfrm_sk_free_policy(sk
);
2682 sk_refcnt_debug_release(sk
);
2684 if (sk
->sk_frag
.page
) {
2685 put_page(sk
->sk_frag
.page
);
2686 sk
->sk_frag
.page
= NULL
;
2691 EXPORT_SYMBOL(sk_common_release
);
2693 #ifdef CONFIG_PROC_FS
2694 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2696 int val
[PROTO_INUSE_NR
];
2699 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2701 #ifdef CONFIG_NET_NS
2702 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2704 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2706 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2708 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2710 int cpu
, idx
= prot
->inuse_idx
;
2713 for_each_possible_cpu(cpu
)
2714 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2716 return res
>= 0 ? res
: 0;
2718 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2720 static int __net_init
sock_inuse_init_net(struct net
*net
)
2722 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2723 return net
->core
.inuse
? 0 : -ENOMEM
;
2726 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2728 free_percpu(net
->core
.inuse
);
2731 static struct pernet_operations net_inuse_ops
= {
2732 .init
= sock_inuse_init_net
,
2733 .exit
= sock_inuse_exit_net
,
2736 static __init
int net_inuse_init(void)
2738 if (register_pernet_subsys(&net_inuse_ops
))
2739 panic("Cannot initialize net inuse counters");
2744 core_initcall(net_inuse_init
);
2746 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2748 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2750 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2752 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2754 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2756 int cpu
, idx
= prot
->inuse_idx
;
2759 for_each_possible_cpu(cpu
)
2760 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2762 return res
>= 0 ? res
: 0;
2764 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2767 static void assign_proto_idx(struct proto
*prot
)
2769 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2771 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2772 pr_err("PROTO_INUSE_NR exhausted\n");
2776 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2779 static void release_proto_idx(struct proto
*prot
)
2781 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2782 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2785 static inline void assign_proto_idx(struct proto
*prot
)
2789 static inline void release_proto_idx(struct proto
*prot
)
2794 int proto_register(struct proto
*prot
, int alloc_slab
)
2797 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2798 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2801 if (prot
->slab
== NULL
) {
2802 pr_crit("%s: Can't create sock SLAB cache!\n",
2807 if (prot
->rsk_prot
!= NULL
) {
2808 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2809 if (prot
->rsk_prot
->slab_name
== NULL
)
2810 goto out_free_sock_slab
;
2812 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2813 prot
->rsk_prot
->obj_size
, 0,
2814 SLAB_HWCACHE_ALIGN
, NULL
);
2816 if (prot
->rsk_prot
->slab
== NULL
) {
2817 pr_crit("%s: Can't create request sock SLAB cache!\n",
2819 goto out_free_request_sock_slab_name
;
2823 if (prot
->twsk_prot
!= NULL
) {
2824 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2826 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2827 goto out_free_request_sock_slab
;
2829 prot
->twsk_prot
->twsk_slab
=
2830 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2831 prot
->twsk_prot
->twsk_obj_size
,
2833 SLAB_HWCACHE_ALIGN
|
2836 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2837 goto out_free_timewait_sock_slab_name
;
2841 mutex_lock(&proto_list_mutex
);
2842 list_add(&prot
->node
, &proto_list
);
2843 assign_proto_idx(prot
);
2844 mutex_unlock(&proto_list_mutex
);
2847 out_free_timewait_sock_slab_name
:
2848 kfree(prot
->twsk_prot
->twsk_slab_name
);
2849 out_free_request_sock_slab
:
2850 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2851 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2852 prot
->rsk_prot
->slab
= NULL
;
2854 out_free_request_sock_slab_name
:
2856 kfree(prot
->rsk_prot
->slab_name
);
2858 kmem_cache_destroy(prot
->slab
);
2863 EXPORT_SYMBOL(proto_register
);
2865 void proto_unregister(struct proto
*prot
)
2867 mutex_lock(&proto_list_mutex
);
2868 release_proto_idx(prot
);
2869 list_del(&prot
->node
);
2870 mutex_unlock(&proto_list_mutex
);
2872 if (prot
->slab
!= NULL
) {
2873 kmem_cache_destroy(prot
->slab
);
2877 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2878 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2879 kfree(prot
->rsk_prot
->slab_name
);
2880 prot
->rsk_prot
->slab
= NULL
;
2883 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2884 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2885 kfree(prot
->twsk_prot
->twsk_slab_name
);
2886 prot
->twsk_prot
->twsk_slab
= NULL
;
2889 EXPORT_SYMBOL(proto_unregister
);
2891 #ifdef CONFIG_PROC_FS
2892 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2893 __acquires(proto_list_mutex
)
2895 mutex_lock(&proto_list_mutex
);
2896 return seq_list_start_head(&proto_list
, *pos
);
2899 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2901 return seq_list_next(v
, &proto_list
, pos
);
2904 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2905 __releases(proto_list_mutex
)
2907 mutex_unlock(&proto_list_mutex
);
2910 static char proto_method_implemented(const void *method
)
2912 return method
== NULL
? 'n' : 'y';
2914 static long sock_prot_memory_allocated(struct proto
*proto
)
2916 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2919 static char *sock_prot_memory_pressure(struct proto
*proto
)
2921 return proto
->memory_pressure
!= NULL
?
2922 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2925 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2928 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2929 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2932 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2933 sock_prot_memory_allocated(proto
),
2934 sock_prot_memory_pressure(proto
),
2936 proto
->slab
== NULL
? "no" : "yes",
2937 module_name(proto
->owner
),
2938 proto_method_implemented(proto
->close
),
2939 proto_method_implemented(proto
->connect
),
2940 proto_method_implemented(proto
->disconnect
),
2941 proto_method_implemented(proto
->accept
),
2942 proto_method_implemented(proto
->ioctl
),
2943 proto_method_implemented(proto
->init
),
2944 proto_method_implemented(proto
->destroy
),
2945 proto_method_implemented(proto
->shutdown
),
2946 proto_method_implemented(proto
->setsockopt
),
2947 proto_method_implemented(proto
->getsockopt
),
2948 proto_method_implemented(proto
->sendmsg
),
2949 proto_method_implemented(proto
->recvmsg
),
2950 proto_method_implemented(proto
->sendpage
),
2951 proto_method_implemented(proto
->bind
),
2952 proto_method_implemented(proto
->backlog_rcv
),
2953 proto_method_implemented(proto
->hash
),
2954 proto_method_implemented(proto
->unhash
),
2955 proto_method_implemented(proto
->get_port
),
2956 proto_method_implemented(proto
->enter_memory_pressure
));
2959 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2961 if (v
== &proto_list
)
2962 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2971 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2973 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2977 static const struct seq_operations proto_seq_ops
= {
2978 .start
= proto_seq_start
,
2979 .next
= proto_seq_next
,
2980 .stop
= proto_seq_stop
,
2981 .show
= proto_seq_show
,
2984 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2986 return seq_open_net(inode
, file
, &proto_seq_ops
,
2987 sizeof(struct seq_net_private
));
2990 static const struct file_operations proto_seq_fops
= {
2991 .owner
= THIS_MODULE
,
2992 .open
= proto_seq_open
,
2994 .llseek
= seq_lseek
,
2995 .release
= seq_release_net
,
2998 static __net_init
int proto_init_net(struct net
*net
)
3000 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3006 static __net_exit
void proto_exit_net(struct net
*net
)
3008 remove_proc_entry("protocols", net
->proc_net
);
3012 static __net_initdata
struct pernet_operations proto_net_ops
= {
3013 .init
= proto_init_net
,
3014 .exit
= proto_exit_net
,
3017 static int __init
proto_init(void)
3019 return register_pernet_subsys(&proto_net_ops
);
3022 subsys_initcall(proto_init
);
3024 #endif /* PROC_FS */