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>
135 #include <linux/filter.h>
137 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex
);
146 static LIST_HEAD(proto_list
);
149 * sk_ns_capable - General socket capability test
150 * @sk: Socket to use a capability on or through
151 * @user_ns: The user namespace of the capability to use
152 * @cap: The capability to use
154 * Test to see if the opener of the socket had when the socket was
155 * created and the current process has the capability @cap in the user
156 * namespace @user_ns.
158 bool sk_ns_capable(const struct sock
*sk
,
159 struct user_namespace
*user_ns
, int cap
)
161 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
162 ns_capable(user_ns
, cap
);
164 EXPORT_SYMBOL(sk_ns_capable
);
167 * sk_capable - Socket global capability test
168 * @sk: Socket to use a capability on or through
169 * @cap: The global capability to use
171 * Test to see if the opener of the socket had when the socket was
172 * created and the current process has the capability @cap in all user
175 bool sk_capable(const struct sock
*sk
, int cap
)
177 return sk_ns_capable(sk
, &init_user_ns
, cap
);
179 EXPORT_SYMBOL(sk_capable
);
182 * sk_net_capable - Network namespace socket capability test
183 * @sk: Socket to use a capability on or through
184 * @cap: The capability to use
186 * Test to see if the opener of the socket had when the socket was created
187 * and the current process has the capability @cap over the network namespace
188 * the socket is a member of.
190 bool sk_net_capable(const struct sock
*sk
, int cap
)
192 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
194 EXPORT_SYMBOL(sk_net_capable
);
197 #ifdef CONFIG_MEMCG_KMEM
198 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
203 mutex_lock(&proto_list_mutex
);
204 list_for_each_entry(proto
, &proto_list
, node
) {
205 if (proto
->init_cgroup
) {
206 ret
= proto
->init_cgroup(memcg
, ss
);
212 mutex_unlock(&proto_list_mutex
);
215 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
216 if (proto
->destroy_cgroup
)
217 proto
->destroy_cgroup(memcg
);
218 mutex_unlock(&proto_list_mutex
);
222 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
226 mutex_lock(&proto_list_mutex
);
227 list_for_each_entry_reverse(proto
, &proto_list
, node
)
228 if (proto
->destroy_cgroup
)
229 proto
->destroy_cgroup(memcg
);
230 mutex_unlock(&proto_list_mutex
);
235 * Each address family might have different locking rules, so we have
236 * one slock key per address family:
238 static struct lock_class_key af_family_keys
[AF_MAX
];
239 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
241 #if defined(CONFIG_MEMCG_KMEM)
242 struct static_key memcg_socket_limit_enabled
;
243 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
247 * Make lock validator output more readable. (we pre-construct these
248 * strings build-time, so that runtime initialization of socket
251 static const char *const af_family_key_strings
[AF_MAX
+1] = {
252 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
253 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
254 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
255 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
256 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
257 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
258 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
259 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
260 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
261 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
262 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
263 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
264 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
265 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
267 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
268 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
269 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
270 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
271 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
272 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
273 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
274 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
275 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
276 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
277 "slock-27" , "slock-28" , "slock-AF_CAN" ,
278 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
279 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
280 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
281 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
283 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
284 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
285 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
286 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
287 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
288 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
289 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
290 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
291 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
292 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
293 "clock-27" , "clock-28" , "clock-AF_CAN" ,
294 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
295 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
296 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
297 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
301 * sk_callback_lock locking rules are per-address-family,
302 * so split the lock classes by using a per-AF key:
304 static struct lock_class_key af_callback_keys
[AF_MAX
];
306 /* Take into consideration the size of the struct sk_buff overhead in the
307 * determination of these values, since that is non-constant across
308 * platforms. This makes socket queueing behavior and performance
309 * not depend upon such differences.
311 #define _SK_MEM_PACKETS 256
312 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
313 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
314 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
316 /* Run time adjustable parameters. */
317 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
318 EXPORT_SYMBOL(sysctl_wmem_max
);
319 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
320 EXPORT_SYMBOL(sysctl_rmem_max
);
321 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
322 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
324 /* Maximal space eaten by iovec or ancillary data plus some space */
325 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
326 EXPORT_SYMBOL(sysctl_optmem_max
);
328 int sysctl_tstamp_allow_data __read_mostly
= 1;
330 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
331 EXPORT_SYMBOL_GPL(memalloc_socks
);
334 * sk_set_memalloc - sets %SOCK_MEMALLOC
335 * @sk: socket to set it on
337 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
338 * It's the responsibility of the admin to adjust min_free_kbytes
339 * to meet the requirements
341 void sk_set_memalloc(struct sock
*sk
)
343 sock_set_flag(sk
, SOCK_MEMALLOC
);
344 sk
->sk_allocation
|= __GFP_MEMALLOC
;
345 static_key_slow_inc(&memalloc_socks
);
347 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
349 void sk_clear_memalloc(struct sock
*sk
)
351 sock_reset_flag(sk
, SOCK_MEMALLOC
);
352 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
353 static_key_slow_dec(&memalloc_socks
);
356 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
357 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
358 * it has rmem allocations there is a risk that the user of the
359 * socket cannot make forward progress due to exceeding the rmem
360 * limits. By rights, sk_clear_memalloc() should only be called
361 * on sockets being torn down but warn and reset the accounting if
362 * that assumption breaks.
364 if (WARN_ON(sk
->sk_forward_alloc
))
367 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
369 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
372 unsigned long pflags
= current
->flags
;
374 /* these should have been dropped before queueing */
375 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
377 current
->flags
|= PF_MEMALLOC
;
378 ret
= sk
->sk_backlog_rcv(sk
, skb
);
379 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
383 EXPORT_SYMBOL(__sk_backlog_rcv
);
385 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
389 if (optlen
< sizeof(tv
))
391 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
393 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
397 static int warned __read_mostly
;
400 if (warned
< 10 && net_ratelimit()) {
402 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
403 __func__
, current
->comm
, task_pid_nr(current
));
407 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
408 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
410 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
411 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
415 static void sock_warn_obsolete_bsdism(const char *name
)
418 static char warncomm
[TASK_COMM_LEN
];
419 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
420 strcpy(warncomm
, current
->comm
);
421 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
427 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
429 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
431 if (sk
->sk_flags
& flags
) {
432 sk
->sk_flags
&= ~flags
;
433 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
434 net_disable_timestamp();
439 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 /* we escape from rcu protected region, make sure we dont leak
468 spin_lock_irqsave(&list
->lock
, flags
);
469 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
470 __skb_queue_tail(list
, skb
);
471 spin_unlock_irqrestore(&list
->lock
, flags
);
473 if (!sock_flag(sk
, SOCK_DEAD
))
474 sk
->sk_data_ready(sk
);
477 EXPORT_SYMBOL(sock_queue_rcv_skb
);
479 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
481 int rc
= NET_RX_SUCCESS
;
483 if (sk_filter(sk
, skb
))
484 goto discard_and_relse
;
488 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
489 atomic_inc(&sk
->sk_drops
);
490 goto discard_and_relse
;
493 bh_lock_sock_nested(sk
);
496 if (!sock_owned_by_user(sk
)) {
498 * trylock + unlock semantics:
500 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
502 rc
= sk_backlog_rcv(sk
, skb
);
504 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
505 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
507 atomic_inc(&sk
->sk_drops
);
508 goto discard_and_relse
;
519 EXPORT_SYMBOL(sk_receive_skb
);
521 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
523 struct dst_entry
*dst
= __sk_dst_get(sk
);
525 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
526 sk_tx_queue_clear(sk
);
527 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
534 EXPORT_SYMBOL(__sk_dst_check
);
536 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
538 struct dst_entry
*dst
= sk_dst_get(sk
);
540 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
548 EXPORT_SYMBOL(sk_dst_check
);
550 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
553 int ret
= -ENOPROTOOPT
;
554 #ifdef CONFIG_NETDEVICES
555 struct net
*net
= sock_net(sk
);
556 char devname
[IFNAMSIZ
];
561 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
568 /* Bind this socket to a particular device like "eth0",
569 * as specified in the passed interface name. If the
570 * name is "" or the option length is zero the socket
573 if (optlen
> IFNAMSIZ
- 1)
574 optlen
= IFNAMSIZ
- 1;
575 memset(devname
, 0, sizeof(devname
));
578 if (copy_from_user(devname
, optval
, optlen
))
582 if (devname
[0] != '\0') {
583 struct net_device
*dev
;
586 dev
= dev_get_by_name_rcu(net
, devname
);
588 index
= dev
->ifindex
;
596 sk
->sk_bound_dev_if
= index
;
608 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
609 int __user
*optlen
, int len
)
611 int ret
= -ENOPROTOOPT
;
612 #ifdef CONFIG_NETDEVICES
613 struct net
*net
= sock_net(sk
);
614 char devname
[IFNAMSIZ
];
616 if (sk
->sk_bound_dev_if
== 0) {
625 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
629 len
= strlen(devname
) + 1;
632 if (copy_to_user(optval
, devname
, len
))
637 if (put_user(len
, optlen
))
648 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
651 sock_set_flag(sk
, bit
);
653 sock_reset_flag(sk
, bit
);
657 * This is meant for all protocols to use and covers goings on
658 * at the socket level. Everything here is generic.
661 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
662 char __user
*optval
, unsigned int optlen
)
664 struct sock
*sk
= sock
->sk
;
671 * Options without arguments
674 if (optname
== SO_BINDTODEVICE
)
675 return sock_setbindtodevice(sk
, optval
, optlen
);
677 if (optlen
< sizeof(int))
680 if (get_user(val
, (int __user
*)optval
))
683 valbool
= val
? 1 : 0;
689 if (val
&& !capable(CAP_NET_ADMIN
))
692 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
695 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
698 sk
->sk_reuseport
= valbool
;
707 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
710 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
713 /* Don't error on this BSD doesn't and if you think
714 * about it this is right. Otherwise apps have to
715 * play 'guess the biggest size' games. RCVBUF/SNDBUF
716 * are treated in BSD as hints
718 val
= min_t(u32
, val
, sysctl_wmem_max
);
720 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
721 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
722 /* Wake up sending tasks if we upped the value. */
723 sk
->sk_write_space(sk
);
727 if (!capable(CAP_NET_ADMIN
)) {
734 /* Don't error on this BSD doesn't and if you think
735 * about it this is right. Otherwise apps have to
736 * play 'guess the biggest size' games. RCVBUF/SNDBUF
737 * are treated in BSD as hints
739 val
= min_t(u32
, val
, sysctl_rmem_max
);
741 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
743 * We double it on the way in to account for
744 * "struct sk_buff" etc. overhead. Applications
745 * assume that the SO_RCVBUF setting they make will
746 * allow that much actual data to be received on that
749 * Applications are unaware that "struct sk_buff" and
750 * other overheads allocate from the receive buffer
751 * during socket buffer allocation.
753 * And after considering the possible alternatives,
754 * returning the value we actually used in getsockopt
755 * is the most desirable behavior.
757 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
761 if (!capable(CAP_NET_ADMIN
)) {
769 if (sk
->sk_protocol
== IPPROTO_TCP
&&
770 sk
->sk_type
== SOCK_STREAM
)
771 tcp_set_keepalive(sk
, valbool
);
773 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
777 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
781 sk
->sk_no_check_tx
= valbool
;
785 if ((val
>= 0 && val
<= 6) ||
786 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
787 sk
->sk_priority
= val
;
793 if (optlen
< sizeof(ling
)) {
794 ret
= -EINVAL
; /* 1003.1g */
797 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
802 sock_reset_flag(sk
, SOCK_LINGER
);
804 #if (BITS_PER_LONG == 32)
805 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
806 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
809 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
810 sock_set_flag(sk
, SOCK_LINGER
);
815 sock_warn_obsolete_bsdism("setsockopt");
820 set_bit(SOCK_PASSCRED
, &sock
->flags
);
822 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
828 if (optname
== SO_TIMESTAMP
)
829 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
831 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
832 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
833 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
835 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
836 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
840 case SO_TIMESTAMPING
:
841 if (val
& ~SOF_TIMESTAMPING_MASK
) {
846 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
847 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
848 if (sk
->sk_protocol
== IPPROTO_TCP
) {
849 if (sk
->sk_state
!= TCP_ESTABLISHED
) {
853 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
858 sk
->sk_tsflags
= val
;
859 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
860 sock_enable_timestamp(sk
,
861 SOCK_TIMESTAMPING_RX_SOFTWARE
);
863 sock_disable_timestamp(sk
,
864 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
870 sk
->sk_rcvlowat
= val
? : 1;
874 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
878 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
881 case SO_ATTACH_FILTER
:
883 if (optlen
== sizeof(struct sock_fprog
)) {
884 struct sock_fprog fprog
;
887 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
890 ret
= sk_attach_filter(&fprog
, sk
);
896 if (optlen
== sizeof(u32
)) {
900 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
903 ret
= sk_attach_bpf(ufd
, sk
);
907 case SO_DETACH_FILTER
:
908 ret
= sk_detach_filter(sk
);
912 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
915 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
920 set_bit(SOCK_PASSSEC
, &sock
->flags
);
922 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
925 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
931 /* We implement the SO_SNDLOWAT etc to
932 not be settable (1003.1g 5.3) */
934 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
938 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
942 if (sock
->ops
->set_peek_off
)
943 ret
= sock
->ops
->set_peek_off(sk
, val
);
949 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
952 case SO_SELECT_ERR_QUEUE
:
953 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
956 #ifdef CONFIG_NET_RX_BUSY_POLL
958 /* allow unprivileged users to decrease the value */
959 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
965 sk
->sk_ll_usec
= val
;
970 case SO_MAX_PACING_RATE
:
971 sk
->sk_max_pacing_rate
= val
;
972 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
973 sk
->sk_max_pacing_rate
);
983 EXPORT_SYMBOL(sock_setsockopt
);
986 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
989 ucred
->pid
= pid_vnr(pid
);
990 ucred
->uid
= ucred
->gid
= -1;
992 struct user_namespace
*current_ns
= current_user_ns();
994 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
995 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
999 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1000 char __user
*optval
, int __user
*optlen
)
1002 struct sock
*sk
= sock
->sk
;
1010 int lv
= sizeof(int);
1013 if (get_user(len
, optlen
))
1018 memset(&v
, 0, sizeof(v
));
1022 v
.val
= sock_flag(sk
, SOCK_DBG
);
1026 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1030 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1034 v
.val
= sk
->sk_sndbuf
;
1038 v
.val
= sk
->sk_rcvbuf
;
1042 v
.val
= sk
->sk_reuse
;
1046 v
.val
= sk
->sk_reuseport
;
1050 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1054 v
.val
= sk
->sk_type
;
1058 v
.val
= sk
->sk_protocol
;
1062 v
.val
= sk
->sk_family
;
1066 v
.val
= -sock_error(sk
);
1068 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1072 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1076 v
.val
= sk
->sk_no_check_tx
;
1080 v
.val
= sk
->sk_priority
;
1084 lv
= sizeof(v
.ling
);
1085 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1086 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1090 sock_warn_obsolete_bsdism("getsockopt");
1094 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1095 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1098 case SO_TIMESTAMPNS
:
1099 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1102 case SO_TIMESTAMPING
:
1103 v
.val
= sk
->sk_tsflags
;
1107 lv
= sizeof(struct timeval
);
1108 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1112 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1113 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1118 lv
= sizeof(struct timeval
);
1119 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1123 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1124 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1129 v
.val
= sk
->sk_rcvlowat
;
1137 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1142 struct ucred peercred
;
1143 if (len
> sizeof(peercred
))
1144 len
= sizeof(peercred
);
1145 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1146 if (copy_to_user(optval
, &peercred
, len
))
1155 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1159 if (copy_to_user(optval
, address
, len
))
1164 /* Dubious BSD thing... Probably nobody even uses it, but
1165 * the UNIX standard wants it for whatever reason... -DaveM
1168 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1172 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1176 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1179 v
.val
= sk
->sk_mark
;
1183 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1186 case SO_WIFI_STATUS
:
1187 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1191 if (!sock
->ops
->set_peek_off
)
1194 v
.val
= sk
->sk_peek_off
;
1197 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1200 case SO_BINDTODEVICE
:
1201 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1204 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1210 case SO_LOCK_FILTER
:
1211 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1214 case SO_BPF_EXTENSIONS
:
1215 v
.val
= bpf_tell_extensions();
1218 case SO_SELECT_ERR_QUEUE
:
1219 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1222 #ifdef CONFIG_NET_RX_BUSY_POLL
1224 v
.val
= sk
->sk_ll_usec
;
1228 case SO_MAX_PACING_RATE
:
1229 v
.val
= sk
->sk_max_pacing_rate
;
1232 case SO_INCOMING_CPU
:
1233 v
.val
= sk
->sk_incoming_cpu
;
1237 return -ENOPROTOOPT
;
1242 if (copy_to_user(optval
, &v
, len
))
1245 if (put_user(len
, optlen
))
1251 * Initialize an sk_lock.
1253 * (We also register the sk_lock with the lock validator.)
1255 static inline void sock_lock_init(struct sock
*sk
)
1257 sock_lock_init_class_and_name(sk
,
1258 af_family_slock_key_strings
[sk
->sk_family
],
1259 af_family_slock_keys
+ sk
->sk_family
,
1260 af_family_key_strings
[sk
->sk_family
],
1261 af_family_keys
+ sk
->sk_family
);
1265 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1266 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1267 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1269 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1271 #ifdef CONFIG_SECURITY_NETWORK
1272 void *sptr
= nsk
->sk_security
;
1274 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1276 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1277 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1279 #ifdef CONFIG_SECURITY_NETWORK
1280 nsk
->sk_security
= sptr
;
1281 security_sk_clone(osk
, nsk
);
1285 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1287 unsigned long nulls1
, nulls2
;
1289 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1290 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1291 if (nulls1
> nulls2
)
1292 swap(nulls1
, nulls2
);
1295 memset((char *)sk
, 0, nulls1
);
1296 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1297 nulls2
- nulls1
- sizeof(void *));
1298 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1299 size
- nulls2
- sizeof(void *));
1301 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1303 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1307 struct kmem_cache
*slab
;
1311 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1314 if (priority
& __GFP_ZERO
) {
1316 prot
->clear_sk(sk
, prot
->obj_size
);
1318 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1321 sk
= kmalloc(prot
->obj_size
, priority
);
1324 kmemcheck_annotate_bitfield(sk
, flags
);
1326 if (security_sk_alloc(sk
, family
, priority
))
1329 if (!try_module_get(prot
->owner
))
1331 sk_tx_queue_clear(sk
);
1337 security_sk_free(sk
);
1340 kmem_cache_free(slab
, sk
);
1346 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1348 struct kmem_cache
*slab
;
1349 struct module
*owner
;
1351 owner
= prot
->owner
;
1354 security_sk_free(sk
);
1356 kmem_cache_free(slab
, sk
);
1362 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1363 void sock_update_netprioidx(struct sock
*sk
)
1368 sk
->sk_cgrp_prioidx
= task_netprioidx(current
);
1370 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1374 * sk_alloc - All socket objects are allocated here
1375 * @net: the applicable net namespace
1376 * @family: protocol family
1377 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1378 * @prot: struct proto associated with this new sock instance
1380 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1385 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1387 sk
->sk_family
= family
;
1389 * See comment in struct sock definition to understand
1390 * why we need sk_prot_creator -acme
1392 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1394 sock_net_set(sk
, get_net(net
));
1395 atomic_set(&sk
->sk_wmem_alloc
, 1);
1397 sock_update_classid(sk
);
1398 sock_update_netprioidx(sk
);
1403 EXPORT_SYMBOL(sk_alloc
);
1405 static void __sk_free(struct sock
*sk
)
1407 struct sk_filter
*filter
;
1409 if (sk
->sk_destruct
)
1410 sk
->sk_destruct(sk
);
1412 filter
= rcu_dereference_check(sk
->sk_filter
,
1413 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1415 sk_filter_uncharge(sk
, filter
);
1416 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1419 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1421 if (atomic_read(&sk
->sk_omem_alloc
))
1422 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1423 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1425 if (sk
->sk_peer_cred
)
1426 put_cred(sk
->sk_peer_cred
);
1427 put_pid(sk
->sk_peer_pid
);
1428 put_net(sock_net(sk
));
1429 sk_prot_free(sk
->sk_prot_creator
, sk
);
1432 void sk_free(struct sock
*sk
)
1435 * We subtract one from sk_wmem_alloc and can know if
1436 * some packets are still in some tx queue.
1437 * If not null, sock_wfree() will call __sk_free(sk) later
1439 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1442 EXPORT_SYMBOL(sk_free
);
1445 * Last sock_put should drop reference to sk->sk_net. It has already
1446 * been dropped in sk_change_net. Taking reference to stopping namespace
1448 * Take reference to a socket to remove it from hash _alive_ and after that
1449 * destroy it in the context of init_net.
1451 void sk_release_kernel(struct sock
*sk
)
1453 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1457 sock_release(sk
->sk_socket
);
1458 release_net(sock_net(sk
));
1459 sock_net_set(sk
, get_net(&init_net
));
1462 EXPORT_SYMBOL(sk_release_kernel
);
1464 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1466 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1467 sock_update_memcg(newsk
);
1471 * sk_clone_lock - clone a socket, and lock its clone
1472 * @sk: the socket to clone
1473 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1475 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1477 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1480 bool is_charged
= true;
1482 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1483 if (newsk
!= NULL
) {
1484 struct sk_filter
*filter
;
1486 sock_copy(newsk
, sk
);
1489 get_net(sock_net(newsk
));
1490 sk_node_init(&newsk
->sk_node
);
1491 sock_lock_init(newsk
);
1492 bh_lock_sock(newsk
);
1493 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1494 newsk
->sk_backlog
.len
= 0;
1496 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1498 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1500 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1501 atomic_set(&newsk
->sk_omem_alloc
, 0);
1502 skb_queue_head_init(&newsk
->sk_receive_queue
);
1503 skb_queue_head_init(&newsk
->sk_write_queue
);
1505 spin_lock_init(&newsk
->sk_dst_lock
);
1506 rwlock_init(&newsk
->sk_callback_lock
);
1507 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1508 af_callback_keys
+ newsk
->sk_family
,
1509 af_family_clock_key_strings
[newsk
->sk_family
]);
1511 newsk
->sk_dst_cache
= NULL
;
1512 newsk
->sk_wmem_queued
= 0;
1513 newsk
->sk_forward_alloc
= 0;
1514 newsk
->sk_send_head
= NULL
;
1515 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1517 sock_reset_flag(newsk
, SOCK_DONE
);
1518 skb_queue_head_init(&newsk
->sk_error_queue
);
1520 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1522 /* though it's an empty new sock, the charging may fail
1523 * if sysctl_optmem_max was changed between creation of
1524 * original socket and cloning
1526 is_charged
= sk_filter_charge(newsk
, filter
);
1528 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
))) {
1529 /* It is still raw copy of parent, so invalidate
1530 * destructor and make plain sk_free() */
1531 newsk
->sk_destruct
= NULL
;
1532 bh_unlock_sock(newsk
);
1539 newsk
->sk_priority
= 0;
1540 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1542 * Before updating sk_refcnt, we must commit prior changes to memory
1543 * (Documentation/RCU/rculist_nulls.txt for details)
1546 atomic_set(&newsk
->sk_refcnt
, 2);
1549 * Increment the counter in the same struct proto as the master
1550 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1551 * is the same as sk->sk_prot->socks, as this field was copied
1554 * This _changes_ the previous behaviour, where
1555 * tcp_create_openreq_child always was incrementing the
1556 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1557 * to be taken into account in all callers. -acme
1559 sk_refcnt_debug_inc(newsk
);
1560 sk_set_socket(newsk
, NULL
);
1561 newsk
->sk_wq
= NULL
;
1563 sk_update_clone(sk
, newsk
);
1565 if (newsk
->sk_prot
->sockets_allocated
)
1566 sk_sockets_allocated_inc(newsk
);
1568 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1569 net_enable_timestamp();
1574 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1576 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1578 __sk_dst_set(sk
, dst
);
1579 sk
->sk_route_caps
= dst
->dev
->features
;
1580 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1581 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1582 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1583 if (sk_can_gso(sk
)) {
1584 if (dst
->header_len
) {
1585 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1587 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1588 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1589 sk
->sk_gso_max_segs
= dst
->dev
->gso_max_segs
;
1593 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1596 * Simple resource managers for sockets.
1601 * Write buffer destructor automatically called from kfree_skb.
1603 void sock_wfree(struct sk_buff
*skb
)
1605 struct sock
*sk
= skb
->sk
;
1606 unsigned int len
= skb
->truesize
;
1608 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1610 * Keep a reference on sk_wmem_alloc, this will be released
1611 * after sk_write_space() call
1613 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1614 sk
->sk_write_space(sk
);
1618 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1619 * could not do because of in-flight packets
1621 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1624 EXPORT_SYMBOL(sock_wfree
);
1626 void skb_orphan_partial(struct sk_buff
*skb
)
1628 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1629 * so we do not completely orphan skb, but transfert all
1630 * accounted bytes but one, to avoid unexpected reorders.
1632 if (skb
->destructor
== sock_wfree
1634 || skb
->destructor
== tcp_wfree
1637 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1643 EXPORT_SYMBOL(skb_orphan_partial
);
1646 * Read buffer destructor automatically called from kfree_skb.
1648 void sock_rfree(struct sk_buff
*skb
)
1650 struct sock
*sk
= skb
->sk
;
1651 unsigned int len
= skb
->truesize
;
1653 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1654 sk_mem_uncharge(sk
, len
);
1656 EXPORT_SYMBOL(sock_rfree
);
1659 * Buffer destructor for skbs that are not used directly in read or write
1660 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1662 void sock_efree(struct sk_buff
*skb
)
1666 EXPORT_SYMBOL(sock_efree
);
1669 void sock_edemux(struct sk_buff
*skb
)
1671 struct sock
*sk
= skb
->sk
;
1673 if (sk
->sk_state
== TCP_TIME_WAIT
)
1674 inet_twsk_put(inet_twsk(sk
));
1678 EXPORT_SYMBOL(sock_edemux
);
1681 kuid_t
sock_i_uid(struct sock
*sk
)
1685 read_lock_bh(&sk
->sk_callback_lock
);
1686 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1687 read_unlock_bh(&sk
->sk_callback_lock
);
1690 EXPORT_SYMBOL(sock_i_uid
);
1692 unsigned long sock_i_ino(struct sock
*sk
)
1696 read_lock_bh(&sk
->sk_callback_lock
);
1697 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1698 read_unlock_bh(&sk
->sk_callback_lock
);
1701 EXPORT_SYMBOL(sock_i_ino
);
1704 * Allocate a skb from the socket's send buffer.
1706 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1709 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1710 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1712 skb_set_owner_w(skb
, sk
);
1718 EXPORT_SYMBOL(sock_wmalloc
);
1721 * Allocate a memory block from the socket's option memory buffer.
1723 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1725 if ((unsigned int)size
<= sysctl_optmem_max
&&
1726 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1728 /* First do the add, to avoid the race if kmalloc
1731 atomic_add(size
, &sk
->sk_omem_alloc
);
1732 mem
= kmalloc(size
, priority
);
1735 atomic_sub(size
, &sk
->sk_omem_alloc
);
1739 EXPORT_SYMBOL(sock_kmalloc
);
1741 /* Free an option memory block. Note, we actually want the inline
1742 * here as this allows gcc to detect the nullify and fold away the
1743 * condition entirely.
1745 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1748 if (WARN_ON_ONCE(!mem
))
1754 atomic_sub(size
, &sk
->sk_omem_alloc
);
1757 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1759 __sock_kfree_s(sk
, mem
, size
, false);
1761 EXPORT_SYMBOL(sock_kfree_s
);
1763 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1765 __sock_kfree_s(sk
, mem
, size
, true);
1767 EXPORT_SYMBOL(sock_kzfree_s
);
1769 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1770 I think, these locks should be removed for datagram sockets.
1772 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1776 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1780 if (signal_pending(current
))
1782 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1783 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1784 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1786 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1790 timeo
= schedule_timeout(timeo
);
1792 finish_wait(sk_sleep(sk
), &wait
);
1798 * Generic send/receive buffer handlers
1801 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1802 unsigned long data_len
, int noblock
,
1803 int *errcode
, int max_page_order
)
1805 struct sk_buff
*skb
;
1809 timeo
= sock_sndtimeo(sk
, noblock
);
1811 err
= sock_error(sk
);
1816 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1819 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1822 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1823 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1827 if (signal_pending(current
))
1829 timeo
= sock_wait_for_wmem(sk
, timeo
);
1831 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
1832 errcode
, sk
->sk_allocation
);
1834 skb_set_owner_w(skb
, sk
);
1838 err
= sock_intr_errno(timeo
);
1843 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1845 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1846 int noblock
, int *errcode
)
1848 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1850 EXPORT_SYMBOL(sock_alloc_send_skb
);
1852 /* On 32bit arches, an skb frag is limited to 2^15 */
1853 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1856 * skb_page_frag_refill - check that a page_frag contains enough room
1857 * @sz: minimum size of the fragment we want to get
1858 * @pfrag: pointer to page_frag
1859 * @gfp: priority for memory allocation
1861 * Note: While this allocator tries to use high order pages, there is
1862 * no guarantee that allocations succeed. Therefore, @sz MUST be
1863 * less or equal than PAGE_SIZE.
1865 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
1868 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1872 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1874 put_page(pfrag
->page
);
1878 if (SKB_FRAG_PAGE_ORDER
) {
1879 pfrag
->page
= alloc_pages(gfp
| __GFP_COMP
|
1880 __GFP_NOWARN
| __GFP_NORETRY
,
1881 SKB_FRAG_PAGE_ORDER
);
1882 if (likely(pfrag
->page
)) {
1883 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
1887 pfrag
->page
= alloc_page(gfp
);
1888 if (likely(pfrag
->page
)) {
1889 pfrag
->size
= PAGE_SIZE
;
1894 EXPORT_SYMBOL(skb_page_frag_refill
);
1896 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1898 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
1901 sk_enter_memory_pressure(sk
);
1902 sk_stream_moderate_sndbuf(sk
);
1905 EXPORT_SYMBOL(sk_page_frag_refill
);
1907 static void __lock_sock(struct sock
*sk
)
1908 __releases(&sk
->sk_lock
.slock
)
1909 __acquires(&sk
->sk_lock
.slock
)
1914 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1915 TASK_UNINTERRUPTIBLE
);
1916 spin_unlock_bh(&sk
->sk_lock
.slock
);
1918 spin_lock_bh(&sk
->sk_lock
.slock
);
1919 if (!sock_owned_by_user(sk
))
1922 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1925 static void __release_sock(struct sock
*sk
)
1926 __releases(&sk
->sk_lock
.slock
)
1927 __acquires(&sk
->sk_lock
.slock
)
1929 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1932 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1936 struct sk_buff
*next
= skb
->next
;
1939 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1941 sk_backlog_rcv(sk
, skb
);
1944 * We are in process context here with softirqs
1945 * disabled, use cond_resched_softirq() to preempt.
1946 * This is safe to do because we've taken the backlog
1949 cond_resched_softirq();
1952 } while (skb
!= NULL
);
1955 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1958 * Doing the zeroing here guarantee we can not loop forever
1959 * while a wild producer attempts to flood us.
1961 sk
->sk_backlog
.len
= 0;
1965 * sk_wait_data - wait for data to arrive at sk_receive_queue
1966 * @sk: sock to wait on
1967 * @timeo: for how long
1969 * Now socket state including sk->sk_err is changed only under lock,
1970 * hence we may omit checks after joining wait queue.
1971 * We check receive queue before schedule() only as optimization;
1972 * it is very likely that release_sock() added new data.
1974 int sk_wait_data(struct sock
*sk
, long *timeo
)
1979 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1980 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1981 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1982 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1983 finish_wait(sk_sleep(sk
), &wait
);
1986 EXPORT_SYMBOL(sk_wait_data
);
1989 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1991 * @size: memory size to allocate
1992 * @kind: allocation type
1994 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1995 * rmem allocation. This function assumes that protocols which have
1996 * memory_pressure use sk_wmem_queued as write buffer accounting.
1998 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2000 struct proto
*prot
= sk
->sk_prot
;
2001 int amt
= sk_mem_pages(size
);
2003 int parent_status
= UNDER_LIMIT
;
2005 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2007 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
2010 if (parent_status
== UNDER_LIMIT
&&
2011 allocated
<= sk_prot_mem_limits(sk
, 0)) {
2012 sk_leave_memory_pressure(sk
);
2016 /* Under pressure. (we or our parents) */
2017 if ((parent_status
> SOFT_LIMIT
) ||
2018 allocated
> sk_prot_mem_limits(sk
, 1))
2019 sk_enter_memory_pressure(sk
);
2021 /* Over hard limit (we or our parents) */
2022 if ((parent_status
== OVER_LIMIT
) ||
2023 (allocated
> sk_prot_mem_limits(sk
, 2)))
2024 goto suppress_allocation
;
2026 /* guarantee minimum buffer size under pressure */
2027 if (kind
== SK_MEM_RECV
) {
2028 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2031 } else { /* SK_MEM_SEND */
2032 if (sk
->sk_type
== SOCK_STREAM
) {
2033 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2035 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2036 prot
->sysctl_wmem
[0])
2040 if (sk_has_memory_pressure(sk
)) {
2043 if (!sk_under_memory_pressure(sk
))
2045 alloc
= sk_sockets_allocated_read_positive(sk
);
2046 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2047 sk_mem_pages(sk
->sk_wmem_queued
+
2048 atomic_read(&sk
->sk_rmem_alloc
) +
2049 sk
->sk_forward_alloc
))
2053 suppress_allocation
:
2055 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2056 sk_stream_moderate_sndbuf(sk
);
2058 /* Fail only if socket is _under_ its sndbuf.
2059 * In this case we cannot block, so that we have to fail.
2061 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2065 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2067 /* Alas. Undo changes. */
2068 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2070 sk_memory_allocated_sub(sk
, amt
);
2074 EXPORT_SYMBOL(__sk_mem_schedule
);
2077 * __sk_reclaim - reclaim memory_allocated
2080 void __sk_mem_reclaim(struct sock
*sk
)
2082 sk_memory_allocated_sub(sk
,
2083 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
2084 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
2086 if (sk_under_memory_pressure(sk
) &&
2087 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2088 sk_leave_memory_pressure(sk
);
2090 EXPORT_SYMBOL(__sk_mem_reclaim
);
2094 * Set of default routines for initialising struct proto_ops when
2095 * the protocol does not support a particular function. In certain
2096 * cases where it makes no sense for a protocol to have a "do nothing"
2097 * function, some default processing is provided.
2100 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2104 EXPORT_SYMBOL(sock_no_bind
);
2106 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2111 EXPORT_SYMBOL(sock_no_connect
);
2113 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2117 EXPORT_SYMBOL(sock_no_socketpair
);
2119 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2123 EXPORT_SYMBOL(sock_no_accept
);
2125 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2130 EXPORT_SYMBOL(sock_no_getname
);
2132 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2136 EXPORT_SYMBOL(sock_no_poll
);
2138 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2142 EXPORT_SYMBOL(sock_no_ioctl
);
2144 int sock_no_listen(struct socket
*sock
, int backlog
)
2148 EXPORT_SYMBOL(sock_no_listen
);
2150 int sock_no_shutdown(struct socket
*sock
, int how
)
2154 EXPORT_SYMBOL(sock_no_shutdown
);
2156 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2157 char __user
*optval
, unsigned int optlen
)
2161 EXPORT_SYMBOL(sock_no_setsockopt
);
2163 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2164 char __user
*optval
, int __user
*optlen
)
2168 EXPORT_SYMBOL(sock_no_getsockopt
);
2170 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2175 EXPORT_SYMBOL(sock_no_sendmsg
);
2177 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2178 size_t len
, int flags
)
2182 EXPORT_SYMBOL(sock_no_recvmsg
);
2184 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2186 /* Mirror missing mmap method error code */
2189 EXPORT_SYMBOL(sock_no_mmap
);
2191 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2194 struct msghdr msg
= {.msg_flags
= flags
};
2196 char *kaddr
= kmap(page
);
2197 iov
.iov_base
= kaddr
+ offset
;
2199 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2203 EXPORT_SYMBOL(sock_no_sendpage
);
2206 * Default Socket Callbacks
2209 static void sock_def_wakeup(struct sock
*sk
)
2211 struct socket_wq
*wq
;
2214 wq
= rcu_dereference(sk
->sk_wq
);
2215 if (wq_has_sleeper(wq
))
2216 wake_up_interruptible_all(&wq
->wait
);
2220 static void sock_def_error_report(struct sock
*sk
)
2222 struct socket_wq
*wq
;
2225 wq
= rcu_dereference(sk
->sk_wq
);
2226 if (wq_has_sleeper(wq
))
2227 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2228 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2232 static void sock_def_readable(struct sock
*sk
)
2234 struct socket_wq
*wq
;
2237 wq
= rcu_dereference(sk
->sk_wq
);
2238 if (wq_has_sleeper(wq
))
2239 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2240 POLLRDNORM
| POLLRDBAND
);
2241 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2245 static void sock_def_write_space(struct sock
*sk
)
2247 struct socket_wq
*wq
;
2251 /* Do not wake up a writer until he can make "significant"
2254 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2255 wq
= rcu_dereference(sk
->sk_wq
);
2256 if (wq_has_sleeper(wq
))
2257 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2258 POLLWRNORM
| POLLWRBAND
);
2260 /* Should agree with poll, otherwise some programs break */
2261 if (sock_writeable(sk
))
2262 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2268 static void sock_def_destruct(struct sock
*sk
)
2270 kfree(sk
->sk_protinfo
);
2273 void sk_send_sigurg(struct sock
*sk
)
2275 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2276 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2277 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2279 EXPORT_SYMBOL(sk_send_sigurg
);
2281 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2282 unsigned long expires
)
2284 if (!mod_timer(timer
, expires
))
2287 EXPORT_SYMBOL(sk_reset_timer
);
2289 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2291 if (del_timer(timer
))
2294 EXPORT_SYMBOL(sk_stop_timer
);
2296 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2298 skb_queue_head_init(&sk
->sk_receive_queue
);
2299 skb_queue_head_init(&sk
->sk_write_queue
);
2300 skb_queue_head_init(&sk
->sk_error_queue
);
2302 sk
->sk_send_head
= NULL
;
2304 init_timer(&sk
->sk_timer
);
2306 sk
->sk_allocation
= GFP_KERNEL
;
2307 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2308 sk
->sk_sndbuf
= sysctl_wmem_default
;
2309 sk
->sk_state
= TCP_CLOSE
;
2310 sk_set_socket(sk
, sock
);
2312 sock_set_flag(sk
, SOCK_ZAPPED
);
2315 sk
->sk_type
= sock
->type
;
2316 sk
->sk_wq
= sock
->wq
;
2321 spin_lock_init(&sk
->sk_dst_lock
);
2322 rwlock_init(&sk
->sk_callback_lock
);
2323 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2324 af_callback_keys
+ sk
->sk_family
,
2325 af_family_clock_key_strings
[sk
->sk_family
]);
2327 sk
->sk_state_change
= sock_def_wakeup
;
2328 sk
->sk_data_ready
= sock_def_readable
;
2329 sk
->sk_write_space
= sock_def_write_space
;
2330 sk
->sk_error_report
= sock_def_error_report
;
2331 sk
->sk_destruct
= sock_def_destruct
;
2333 sk
->sk_frag
.page
= NULL
;
2334 sk
->sk_frag
.offset
= 0;
2335 sk
->sk_peek_off
= -1;
2337 sk
->sk_peer_pid
= NULL
;
2338 sk
->sk_peer_cred
= NULL
;
2339 sk
->sk_write_pending
= 0;
2340 sk
->sk_rcvlowat
= 1;
2341 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2342 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2344 sk
->sk_stamp
= ktime_set(-1L, 0);
2346 #ifdef CONFIG_NET_RX_BUSY_POLL
2348 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2351 sk
->sk_max_pacing_rate
= ~0U;
2352 sk
->sk_pacing_rate
= ~0U;
2354 * Before updating sk_refcnt, we must commit prior changes to memory
2355 * (Documentation/RCU/rculist_nulls.txt for details)
2358 atomic_set(&sk
->sk_refcnt
, 1);
2359 atomic_set(&sk
->sk_drops
, 0);
2361 EXPORT_SYMBOL(sock_init_data
);
2363 void lock_sock_nested(struct sock
*sk
, int subclass
)
2366 spin_lock_bh(&sk
->sk_lock
.slock
);
2367 if (sk
->sk_lock
.owned
)
2369 sk
->sk_lock
.owned
= 1;
2370 spin_unlock(&sk
->sk_lock
.slock
);
2372 * The sk_lock has mutex_lock() semantics here:
2374 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2377 EXPORT_SYMBOL(lock_sock_nested
);
2379 void release_sock(struct sock
*sk
)
2382 * The sk_lock has mutex_unlock() semantics:
2384 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2386 spin_lock_bh(&sk
->sk_lock
.slock
);
2387 if (sk
->sk_backlog
.tail
)
2390 /* Warning : release_cb() might need to release sk ownership,
2391 * ie call sock_release_ownership(sk) before us.
2393 if (sk
->sk_prot
->release_cb
)
2394 sk
->sk_prot
->release_cb(sk
);
2396 sock_release_ownership(sk
);
2397 if (waitqueue_active(&sk
->sk_lock
.wq
))
2398 wake_up(&sk
->sk_lock
.wq
);
2399 spin_unlock_bh(&sk
->sk_lock
.slock
);
2401 EXPORT_SYMBOL(release_sock
);
2404 * lock_sock_fast - fast version of lock_sock
2407 * This version should be used for very small section, where process wont block
2408 * return false if fast path is taken
2409 * sk_lock.slock locked, owned = 0, BH disabled
2410 * return true if slow path is taken
2411 * sk_lock.slock unlocked, owned = 1, BH enabled
2413 bool lock_sock_fast(struct sock
*sk
)
2416 spin_lock_bh(&sk
->sk_lock
.slock
);
2418 if (!sk
->sk_lock
.owned
)
2420 * Note : We must disable BH
2425 sk
->sk_lock
.owned
= 1;
2426 spin_unlock(&sk
->sk_lock
.slock
);
2428 * The sk_lock has mutex_lock() semantics here:
2430 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2434 EXPORT_SYMBOL(lock_sock_fast
);
2436 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2439 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2440 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2441 tv
= ktime_to_timeval(sk
->sk_stamp
);
2442 if (tv
.tv_sec
== -1)
2444 if (tv
.tv_sec
== 0) {
2445 sk
->sk_stamp
= ktime_get_real();
2446 tv
= ktime_to_timeval(sk
->sk_stamp
);
2448 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2450 EXPORT_SYMBOL(sock_get_timestamp
);
2452 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2455 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2456 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2457 ts
= ktime_to_timespec(sk
->sk_stamp
);
2458 if (ts
.tv_sec
== -1)
2460 if (ts
.tv_sec
== 0) {
2461 sk
->sk_stamp
= ktime_get_real();
2462 ts
= ktime_to_timespec(sk
->sk_stamp
);
2464 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2466 EXPORT_SYMBOL(sock_get_timestampns
);
2468 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2470 if (!sock_flag(sk
, flag
)) {
2471 unsigned long previous_flags
= sk
->sk_flags
;
2473 sock_set_flag(sk
, flag
);
2475 * we just set one of the two flags which require net
2476 * time stamping, but time stamping might have been on
2477 * already because of the other one
2479 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2480 net_enable_timestamp();
2484 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2485 int level
, int type
)
2487 struct sock_exterr_skb
*serr
;
2488 struct sk_buff
*skb
;
2492 skb
= sock_dequeue_err_skb(sk
);
2498 msg
->msg_flags
|= MSG_TRUNC
;
2501 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2505 sock_recv_timestamp(msg
, sk
, skb
);
2507 serr
= SKB_EXT_ERR(skb
);
2508 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2510 msg
->msg_flags
|= MSG_ERRQUEUE
;
2518 EXPORT_SYMBOL(sock_recv_errqueue
);
2521 * Get a socket option on an socket.
2523 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2524 * asynchronous errors should be reported by getsockopt. We assume
2525 * this means if you specify SO_ERROR (otherwise whats the point of it).
2527 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2528 char __user
*optval
, int __user
*optlen
)
2530 struct sock
*sk
= sock
->sk
;
2532 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2534 EXPORT_SYMBOL(sock_common_getsockopt
);
2536 #ifdef CONFIG_COMPAT
2537 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2538 char __user
*optval
, int __user
*optlen
)
2540 struct sock
*sk
= sock
->sk
;
2542 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2543 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2545 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2547 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2550 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2551 struct msghdr
*msg
, size_t size
, int flags
)
2553 struct sock
*sk
= sock
->sk
;
2557 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2558 flags
& ~MSG_DONTWAIT
, &addr_len
);
2560 msg
->msg_namelen
= addr_len
;
2563 EXPORT_SYMBOL(sock_common_recvmsg
);
2566 * Set socket options on an inet socket.
2568 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2569 char __user
*optval
, unsigned int optlen
)
2571 struct sock
*sk
= sock
->sk
;
2573 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2575 EXPORT_SYMBOL(sock_common_setsockopt
);
2577 #ifdef CONFIG_COMPAT
2578 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2579 char __user
*optval
, unsigned int optlen
)
2581 struct sock
*sk
= sock
->sk
;
2583 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2584 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2586 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2588 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2591 void sk_common_release(struct sock
*sk
)
2593 if (sk
->sk_prot
->destroy
)
2594 sk
->sk_prot
->destroy(sk
);
2597 * Observation: when sock_common_release is called, processes have
2598 * no access to socket. But net still has.
2599 * Step one, detach it from networking:
2601 * A. Remove from hash tables.
2604 sk
->sk_prot
->unhash(sk
);
2607 * In this point socket cannot receive new packets, but it is possible
2608 * that some packets are in flight because some CPU runs receiver and
2609 * did hash table lookup before we unhashed socket. They will achieve
2610 * receive queue and will be purged by socket destructor.
2612 * Also we still have packets pending on receive queue and probably,
2613 * our own packets waiting in device queues. sock_destroy will drain
2614 * receive queue, but transmitted packets will delay socket destruction
2615 * until the last reference will be released.
2620 xfrm_sk_free_policy(sk
);
2622 sk_refcnt_debug_release(sk
);
2624 if (sk
->sk_frag
.page
) {
2625 put_page(sk
->sk_frag
.page
);
2626 sk
->sk_frag
.page
= NULL
;
2631 EXPORT_SYMBOL(sk_common_release
);
2633 #ifdef CONFIG_PROC_FS
2634 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2636 int val
[PROTO_INUSE_NR
];
2639 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2641 #ifdef CONFIG_NET_NS
2642 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2644 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2646 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2648 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2650 int cpu
, idx
= prot
->inuse_idx
;
2653 for_each_possible_cpu(cpu
)
2654 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2656 return res
>= 0 ? res
: 0;
2658 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2660 static int __net_init
sock_inuse_init_net(struct net
*net
)
2662 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2663 return net
->core
.inuse
? 0 : -ENOMEM
;
2666 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2668 free_percpu(net
->core
.inuse
);
2671 static struct pernet_operations net_inuse_ops
= {
2672 .init
= sock_inuse_init_net
,
2673 .exit
= sock_inuse_exit_net
,
2676 static __init
int net_inuse_init(void)
2678 if (register_pernet_subsys(&net_inuse_ops
))
2679 panic("Cannot initialize net inuse counters");
2684 core_initcall(net_inuse_init
);
2686 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2688 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2690 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2692 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2694 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2696 int cpu
, idx
= prot
->inuse_idx
;
2699 for_each_possible_cpu(cpu
)
2700 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2702 return res
>= 0 ? res
: 0;
2704 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2707 static void assign_proto_idx(struct proto
*prot
)
2709 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2711 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2712 pr_err("PROTO_INUSE_NR exhausted\n");
2716 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2719 static void release_proto_idx(struct proto
*prot
)
2721 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2722 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2725 static inline void assign_proto_idx(struct proto
*prot
)
2729 static inline void release_proto_idx(struct proto
*prot
)
2734 int proto_register(struct proto
*prot
, int alloc_slab
)
2737 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2738 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2741 if (prot
->slab
== NULL
) {
2742 pr_crit("%s: Can't create sock SLAB cache!\n",
2747 if (prot
->rsk_prot
!= NULL
) {
2748 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2749 if (prot
->rsk_prot
->slab_name
== NULL
)
2750 goto out_free_sock_slab
;
2752 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2753 prot
->rsk_prot
->obj_size
, 0,
2754 SLAB_HWCACHE_ALIGN
, NULL
);
2756 if (prot
->rsk_prot
->slab
== NULL
) {
2757 pr_crit("%s: Can't create request sock SLAB cache!\n",
2759 goto out_free_request_sock_slab_name
;
2763 if (prot
->twsk_prot
!= NULL
) {
2764 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2766 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2767 goto out_free_request_sock_slab
;
2769 prot
->twsk_prot
->twsk_slab
=
2770 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2771 prot
->twsk_prot
->twsk_obj_size
,
2773 SLAB_HWCACHE_ALIGN
|
2776 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2777 goto out_free_timewait_sock_slab_name
;
2781 mutex_lock(&proto_list_mutex
);
2782 list_add(&prot
->node
, &proto_list
);
2783 assign_proto_idx(prot
);
2784 mutex_unlock(&proto_list_mutex
);
2787 out_free_timewait_sock_slab_name
:
2788 kfree(prot
->twsk_prot
->twsk_slab_name
);
2789 out_free_request_sock_slab
:
2790 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2791 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2792 prot
->rsk_prot
->slab
= NULL
;
2794 out_free_request_sock_slab_name
:
2796 kfree(prot
->rsk_prot
->slab_name
);
2798 kmem_cache_destroy(prot
->slab
);
2803 EXPORT_SYMBOL(proto_register
);
2805 void proto_unregister(struct proto
*prot
)
2807 mutex_lock(&proto_list_mutex
);
2808 release_proto_idx(prot
);
2809 list_del(&prot
->node
);
2810 mutex_unlock(&proto_list_mutex
);
2812 if (prot
->slab
!= NULL
) {
2813 kmem_cache_destroy(prot
->slab
);
2817 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2818 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2819 kfree(prot
->rsk_prot
->slab_name
);
2820 prot
->rsk_prot
->slab
= NULL
;
2823 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2824 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2825 kfree(prot
->twsk_prot
->twsk_slab_name
);
2826 prot
->twsk_prot
->twsk_slab
= NULL
;
2829 EXPORT_SYMBOL(proto_unregister
);
2831 #ifdef CONFIG_PROC_FS
2832 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2833 __acquires(proto_list_mutex
)
2835 mutex_lock(&proto_list_mutex
);
2836 return seq_list_start_head(&proto_list
, *pos
);
2839 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2841 return seq_list_next(v
, &proto_list
, pos
);
2844 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2845 __releases(proto_list_mutex
)
2847 mutex_unlock(&proto_list_mutex
);
2850 static char proto_method_implemented(const void *method
)
2852 return method
== NULL
? 'n' : 'y';
2854 static long sock_prot_memory_allocated(struct proto
*proto
)
2856 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2859 static char *sock_prot_memory_pressure(struct proto
*proto
)
2861 return proto
->memory_pressure
!= NULL
?
2862 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2865 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2868 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2869 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2872 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2873 sock_prot_memory_allocated(proto
),
2874 sock_prot_memory_pressure(proto
),
2876 proto
->slab
== NULL
? "no" : "yes",
2877 module_name(proto
->owner
),
2878 proto_method_implemented(proto
->close
),
2879 proto_method_implemented(proto
->connect
),
2880 proto_method_implemented(proto
->disconnect
),
2881 proto_method_implemented(proto
->accept
),
2882 proto_method_implemented(proto
->ioctl
),
2883 proto_method_implemented(proto
->init
),
2884 proto_method_implemented(proto
->destroy
),
2885 proto_method_implemented(proto
->shutdown
),
2886 proto_method_implemented(proto
->setsockopt
),
2887 proto_method_implemented(proto
->getsockopt
),
2888 proto_method_implemented(proto
->sendmsg
),
2889 proto_method_implemented(proto
->recvmsg
),
2890 proto_method_implemented(proto
->sendpage
),
2891 proto_method_implemented(proto
->bind
),
2892 proto_method_implemented(proto
->backlog_rcv
),
2893 proto_method_implemented(proto
->hash
),
2894 proto_method_implemented(proto
->unhash
),
2895 proto_method_implemented(proto
->get_port
),
2896 proto_method_implemented(proto
->enter_memory_pressure
));
2899 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2901 if (v
== &proto_list
)
2902 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2911 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2913 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2917 static const struct seq_operations proto_seq_ops
= {
2918 .start
= proto_seq_start
,
2919 .next
= proto_seq_next
,
2920 .stop
= proto_seq_stop
,
2921 .show
= proto_seq_show
,
2924 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2926 return seq_open_net(inode
, file
, &proto_seq_ops
,
2927 sizeof(struct seq_net_private
));
2930 static const struct file_operations proto_seq_fops
= {
2931 .owner
= THIS_MODULE
,
2932 .open
= proto_seq_open
,
2934 .llseek
= seq_lseek
,
2935 .release
= seq_release_net
,
2938 static __net_init
int proto_init_net(struct net
*net
)
2940 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
2946 static __net_exit
void proto_exit_net(struct net
*net
)
2948 remove_proc_entry("protocols", net
->proc_net
);
2952 static __net_initdata
struct pernet_operations proto_net_ops
= {
2953 .init
= proto_init_net
,
2954 .exit
= proto_exit_net
,
2957 static int __init
proto_init(void)
2959 return register_pernet_subsys(&proto_net_ops
);
2962 subsys_initcall(proto_init
);
2964 #endif /* PROC_FS */