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/sched/mm.h>
106 #include <linux/timer.h>
107 #include <linux/string.h>
108 #include <linux/sockios.h>
109 #include <linux/net.h>
110 #include <linux/mm.h>
111 #include <linux/slab.h>
112 #include <linux/interrupt.h>
113 #include <linux/poll.h>
114 #include <linux/tcp.h>
115 #include <linux/init.h>
116 #include <linux/highmem.h>
117 #include <linux/user_namespace.h>
118 #include <linux/static_key.h>
119 #include <linux/memcontrol.h>
120 #include <linux/prefetch.h>
122 #include <linux/uaccess.h>
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
140 #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 * Each address family might have different locking rules, so we have
198 * one slock key per address family and separate keys for internal and
201 static struct lock_class_key af_family_keys
[AF_MAX
];
202 static struct lock_class_key af_family_kern_keys
[AF_MAX
];
203 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
204 static struct lock_class_key af_family_kern_slock_keys
[AF_MAX
];
207 * Make lock validator output more readable. (we pre-construct these
208 * strings build-time, so that runtime initialization of socket
212 #define _sock_locks(x) \
213 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
214 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
215 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
216 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
217 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
218 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
219 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
220 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
221 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
222 x "27" , x "28" , x "AF_CAN" , \
223 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
224 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
225 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
226 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
227 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
229 static const char *const af_family_key_strings
[AF_MAX
+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings
[AF_MAX
+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings
[AF_MAX
+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings
[AF_MAX
+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings
[AF_MAX
+1] = {
249 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
250 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
251 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
252 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
253 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
254 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
255 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
256 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
257 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
258 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
259 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
260 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
261 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
262 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
263 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
265 static const char *const af_family_wlock_key_strings
[AF_MAX
+1] = {
266 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
267 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
268 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
269 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
270 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
271 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
272 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
273 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
274 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
275 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
276 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
277 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
278 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
279 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
280 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
282 static const char *const af_family_elock_key_strings
[AF_MAX
+1] = {
283 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
284 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
285 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
286 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
287 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
288 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
289 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
290 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
291 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
292 "elock-27" , "elock-28" , "elock-AF_CAN" ,
293 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
294 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
295 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
296 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
297 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
301 * sk_callback_lock and sk queues 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
];
305 static struct lock_class_key af_rlock_keys
[AF_MAX
];
306 static struct lock_class_key af_wlock_keys
[AF_MAX
];
307 static struct lock_class_key af_elock_keys
[AF_MAX
];
308 static struct lock_class_key af_kern_callback_keys
[AF_MAX
];
310 /* Run time adjustable parameters. */
311 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
312 EXPORT_SYMBOL(sysctl_wmem_max
);
313 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
314 EXPORT_SYMBOL(sysctl_rmem_max
);
315 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
316 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
318 /* Maximal space eaten by iovec or ancillary data plus some space */
319 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
320 EXPORT_SYMBOL(sysctl_optmem_max
);
322 int sysctl_tstamp_allow_data __read_mostly
= 1;
324 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
325 EXPORT_SYMBOL_GPL(memalloc_socks
);
328 * sk_set_memalloc - sets %SOCK_MEMALLOC
329 * @sk: socket to set it on
331 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
332 * It's the responsibility of the admin to adjust min_free_kbytes
333 * to meet the requirements
335 void sk_set_memalloc(struct sock
*sk
)
337 sock_set_flag(sk
, SOCK_MEMALLOC
);
338 sk
->sk_allocation
|= __GFP_MEMALLOC
;
339 static_key_slow_inc(&memalloc_socks
);
341 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
343 void sk_clear_memalloc(struct sock
*sk
)
345 sock_reset_flag(sk
, SOCK_MEMALLOC
);
346 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
347 static_key_slow_dec(&memalloc_socks
);
350 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
351 * progress of swapping. SOCK_MEMALLOC may be cleared while
352 * it has rmem allocations due to the last swapfile being deactivated
353 * but there is a risk that the socket is unusable due to exceeding
354 * the rmem limits. Reclaim the reserves and obey rmem limits again.
358 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
360 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
363 unsigned int noreclaim_flag
;
365 /* these should have been dropped before queueing */
366 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
368 noreclaim_flag
= memalloc_noreclaim_save();
369 ret
= sk
->sk_backlog_rcv(sk
, skb
);
370 memalloc_noreclaim_restore(noreclaim_flag
);
374 EXPORT_SYMBOL(__sk_backlog_rcv
);
376 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
380 if (optlen
< sizeof(tv
))
382 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
384 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
388 static int warned __read_mostly
;
391 if (warned
< 10 && net_ratelimit()) {
393 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
394 __func__
, current
->comm
, task_pid_nr(current
));
398 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
399 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
401 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
402 *timeo_p
= tv
.tv_sec
* HZ
+ DIV_ROUND_UP(tv
.tv_usec
, USEC_PER_SEC
/ HZ
);
406 static void sock_warn_obsolete_bsdism(const char *name
)
409 static char warncomm
[TASK_COMM_LEN
];
410 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
411 strcpy(warncomm
, current
->comm
);
412 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
418 static bool sock_needs_netstamp(const struct sock
*sk
)
420 switch (sk
->sk_family
) {
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 (sock_needs_netstamp(sk
) &&
434 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
435 net_disable_timestamp();
440 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 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
452 atomic_inc(&sk
->sk_drops
);
457 skb_set_owner_r(skb
, sk
);
459 /* we escape from rcu protected region, make sure we dont leak
464 spin_lock_irqsave(&list
->lock
, flags
);
465 sock_skb_set_dropcount(sk
, skb
);
466 __skb_queue_tail(list
, skb
);
467 spin_unlock_irqrestore(&list
->lock
, flags
);
469 if (!sock_flag(sk
, SOCK_DEAD
))
470 sk
->sk_data_ready(sk
);
473 EXPORT_SYMBOL(__sock_queue_rcv_skb
);
475 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
479 err
= sk_filter(sk
, skb
);
483 return __sock_queue_rcv_skb(sk
, skb
);
485 EXPORT_SYMBOL(sock_queue_rcv_skb
);
487 int __sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
,
488 const int nested
, unsigned int trim_cap
, bool refcounted
)
490 int rc
= NET_RX_SUCCESS
;
492 if (sk_filter_trim_cap(sk
, skb
, trim_cap
))
493 goto discard_and_relse
;
497 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
498 atomic_inc(&sk
->sk_drops
);
499 goto discard_and_relse
;
502 bh_lock_sock_nested(sk
);
505 if (!sock_owned_by_user(sk
)) {
507 * trylock + unlock semantics:
509 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
511 rc
= sk_backlog_rcv(sk
, skb
);
513 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
514 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
516 atomic_inc(&sk
->sk_drops
);
517 goto discard_and_relse
;
529 EXPORT_SYMBOL(__sk_receive_skb
);
531 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
533 struct dst_entry
*dst
= __sk_dst_get(sk
);
535 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
536 sk_tx_queue_clear(sk
);
537 sk
->sk_dst_pending_confirm
= 0;
538 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
545 EXPORT_SYMBOL(__sk_dst_check
);
547 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
549 struct dst_entry
*dst
= sk_dst_get(sk
);
551 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
559 EXPORT_SYMBOL(sk_dst_check
);
561 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
564 int ret
= -ENOPROTOOPT
;
565 #ifdef CONFIG_NETDEVICES
566 struct net
*net
= sock_net(sk
);
567 char devname
[IFNAMSIZ
];
572 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
579 /* Bind this socket to a particular device like "eth0",
580 * as specified in the passed interface name. If the
581 * name is "" or the option length is zero the socket
584 if (optlen
> IFNAMSIZ
- 1)
585 optlen
= IFNAMSIZ
- 1;
586 memset(devname
, 0, sizeof(devname
));
589 if (copy_from_user(devname
, optval
, optlen
))
593 if (devname
[0] != '\0') {
594 struct net_device
*dev
;
597 dev
= dev_get_by_name_rcu(net
, devname
);
599 index
= dev
->ifindex
;
607 sk
->sk_bound_dev_if
= index
;
619 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
620 int __user
*optlen
, int len
)
622 int ret
= -ENOPROTOOPT
;
623 #ifdef CONFIG_NETDEVICES
624 struct net
*net
= sock_net(sk
);
625 char devname
[IFNAMSIZ
];
627 if (sk
->sk_bound_dev_if
== 0) {
636 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
640 len
= strlen(devname
) + 1;
643 if (copy_to_user(optval
, devname
, len
))
648 if (put_user(len
, optlen
))
659 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
662 sock_set_flag(sk
, bit
);
664 sock_reset_flag(sk
, bit
);
667 bool sk_mc_loop(struct sock
*sk
)
669 if (dev_recursion_level())
673 switch (sk
->sk_family
) {
675 return inet_sk(sk
)->mc_loop
;
676 #if IS_ENABLED(CONFIG_IPV6)
678 return inet6_sk(sk
)->mc_loop
;
684 EXPORT_SYMBOL(sk_mc_loop
);
687 * This is meant for all protocols to use and covers goings on
688 * at the socket level. Everything here is generic.
691 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
692 char __user
*optval
, unsigned int optlen
)
694 struct sock
*sk
= sock
->sk
;
701 * Options without arguments
704 if (optname
== SO_BINDTODEVICE
)
705 return sock_setbindtodevice(sk
, optval
, optlen
);
707 if (optlen
< sizeof(int))
710 if (get_user(val
, (int __user
*)optval
))
713 valbool
= val
? 1 : 0;
719 if (val
&& !capable(CAP_NET_ADMIN
))
722 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
725 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
728 sk
->sk_reuseport
= valbool
;
737 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
741 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
744 /* Don't error on this BSD doesn't and if you think
745 * about it this is right. Otherwise apps have to
746 * play 'guess the biggest size' games. RCVBUF/SNDBUF
747 * are treated in BSD as hints
749 val
= min_t(u32
, val
, sysctl_wmem_max
);
751 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
752 sk
->sk_sndbuf
= max_t(int, val
* 2, SOCK_MIN_SNDBUF
);
753 /* Wake up sending tasks if we upped the value. */
754 sk
->sk_write_space(sk
);
758 if (!capable(CAP_NET_ADMIN
)) {
765 /* Don't error on this BSD doesn't and if you think
766 * about it this is right. Otherwise apps have to
767 * play 'guess the biggest size' games. RCVBUF/SNDBUF
768 * are treated in BSD as hints
770 val
= min_t(u32
, val
, sysctl_rmem_max
);
772 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
774 * We double it on the way in to account for
775 * "struct sk_buff" etc. overhead. Applications
776 * assume that the SO_RCVBUF setting they make will
777 * allow that much actual data to be received on that
780 * Applications are unaware that "struct sk_buff" and
781 * other overheads allocate from the receive buffer
782 * during socket buffer allocation.
784 * And after considering the possible alternatives,
785 * returning the value we actually used in getsockopt
786 * is the most desirable behavior.
788 sk
->sk_rcvbuf
= max_t(int, val
* 2, SOCK_MIN_RCVBUF
);
792 if (!capable(CAP_NET_ADMIN
)) {
799 if (sk
->sk_prot
->keepalive
)
800 sk
->sk_prot
->keepalive(sk
, valbool
);
801 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
805 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
809 sk
->sk_no_check_tx
= valbool
;
813 if ((val
>= 0 && val
<= 6) ||
814 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
815 sk
->sk_priority
= val
;
821 if (optlen
< sizeof(ling
)) {
822 ret
= -EINVAL
; /* 1003.1g */
825 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
830 sock_reset_flag(sk
, SOCK_LINGER
);
832 #if (BITS_PER_LONG == 32)
833 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
834 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
837 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
838 sock_set_flag(sk
, SOCK_LINGER
);
843 sock_warn_obsolete_bsdism("setsockopt");
848 set_bit(SOCK_PASSCRED
, &sock
->flags
);
850 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
856 if (optname
== SO_TIMESTAMP
)
857 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
859 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
860 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
861 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
863 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
864 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
868 case SO_TIMESTAMPING
:
869 if (val
& ~SOF_TIMESTAMPING_MASK
) {
874 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
875 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
876 if (sk
->sk_protocol
== IPPROTO_TCP
&&
877 sk
->sk_type
== SOCK_STREAM
) {
878 if ((1 << sk
->sk_state
) &
879 (TCPF_CLOSE
| TCPF_LISTEN
)) {
883 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
889 if (val
& SOF_TIMESTAMPING_OPT_STATS
&&
890 !(val
& SOF_TIMESTAMPING_OPT_TSONLY
)) {
895 sk
->sk_tsflags
= val
;
896 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
897 sock_enable_timestamp(sk
,
898 SOCK_TIMESTAMPING_RX_SOFTWARE
);
900 sock_disable_timestamp(sk
,
901 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
907 sk
->sk_rcvlowat
= val
? : 1;
911 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
915 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
918 case SO_ATTACH_FILTER
:
920 if (optlen
== sizeof(struct sock_fprog
)) {
921 struct sock_fprog fprog
;
924 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
927 ret
= sk_attach_filter(&fprog
, sk
);
933 if (optlen
== sizeof(u32
)) {
937 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
940 ret
= sk_attach_bpf(ufd
, sk
);
944 case SO_ATTACH_REUSEPORT_CBPF
:
946 if (optlen
== sizeof(struct sock_fprog
)) {
947 struct sock_fprog fprog
;
950 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
953 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
957 case SO_ATTACH_REUSEPORT_EBPF
:
959 if (optlen
== sizeof(u32
)) {
963 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
966 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
970 case SO_DETACH_FILTER
:
971 ret
= sk_detach_filter(sk
);
975 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
978 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
983 set_bit(SOCK_PASSSEC
, &sock
->flags
);
985 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
988 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
995 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
999 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
1003 if (sock
->ops
->set_peek_off
)
1004 ret
= sock
->ops
->set_peek_off(sk
, val
);
1010 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
1013 case SO_SELECT_ERR_QUEUE
:
1014 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
1017 #ifdef CONFIG_NET_RX_BUSY_POLL
1019 /* allow unprivileged users to decrease the value */
1020 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
1026 sk
->sk_ll_usec
= val
;
1031 case SO_MAX_PACING_RATE
:
1033 cmpxchg(&sk
->sk_pacing_status
,
1036 sk
->sk_max_pacing_rate
= val
;
1037 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
1038 sk
->sk_max_pacing_rate
);
1041 case SO_INCOMING_CPU
:
1042 sk
->sk_incoming_cpu
= val
;
1047 dst_negative_advice(sk
);
1051 if (sk
->sk_family
!= PF_INET
&& sk
->sk_family
!= PF_INET6
)
1053 else if (sk
->sk_protocol
!= IPPROTO_TCP
)
1055 else if (sk
->sk_state
!= TCP_CLOSE
)
1057 else if (val
< 0 || val
> 1)
1060 sock_valbool_flag(sk
, SOCK_ZEROCOPY
, valbool
);
1070 EXPORT_SYMBOL(sock_setsockopt
);
1073 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1074 struct ucred
*ucred
)
1076 ucred
->pid
= pid_vnr(pid
);
1077 ucred
->uid
= ucred
->gid
= -1;
1079 struct user_namespace
*current_ns
= current_user_ns();
1081 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1082 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1086 static int groups_to_user(gid_t __user
*dst
, const struct group_info
*src
)
1088 struct user_namespace
*user_ns
= current_user_ns();
1091 for (i
= 0; i
< src
->ngroups
; i
++)
1092 if (put_user(from_kgid_munged(user_ns
, src
->gid
[i
]), dst
+ i
))
1098 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1099 char __user
*optval
, int __user
*optlen
)
1101 struct sock
*sk
= sock
->sk
;
1110 int lv
= sizeof(int);
1113 if (get_user(len
, optlen
))
1118 memset(&v
, 0, sizeof(v
));
1122 v
.val
= sock_flag(sk
, SOCK_DBG
);
1126 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1130 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1134 v
.val
= sk
->sk_sndbuf
;
1138 v
.val
= sk
->sk_rcvbuf
;
1142 v
.val
= sk
->sk_reuse
;
1146 v
.val
= sk
->sk_reuseport
;
1150 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1154 v
.val
= sk
->sk_type
;
1158 v
.val
= sk
->sk_protocol
;
1162 v
.val
= sk
->sk_family
;
1166 v
.val
= -sock_error(sk
);
1168 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1172 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1176 v
.val
= sk
->sk_no_check_tx
;
1180 v
.val
= sk
->sk_priority
;
1184 lv
= sizeof(v
.ling
);
1185 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1186 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1190 sock_warn_obsolete_bsdism("getsockopt");
1194 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1195 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1198 case SO_TIMESTAMPNS
:
1199 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1202 case SO_TIMESTAMPING
:
1203 v
.val
= sk
->sk_tsflags
;
1207 lv
= sizeof(struct timeval
);
1208 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1212 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1213 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * USEC_PER_SEC
) / HZ
;
1218 lv
= sizeof(struct timeval
);
1219 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1223 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1224 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * USEC_PER_SEC
) / HZ
;
1229 v
.val
= sk
->sk_rcvlowat
;
1237 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1242 struct ucred peercred
;
1243 if (len
> sizeof(peercred
))
1244 len
= sizeof(peercred
);
1245 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1246 if (copy_to_user(optval
, &peercred
, len
))
1255 if (!sk
->sk_peer_cred
)
1258 n
= sk
->sk_peer_cred
->group_info
->ngroups
;
1259 if (len
< n
* sizeof(gid_t
)) {
1260 len
= n
* sizeof(gid_t
);
1261 return put_user(len
, optlen
) ? -EFAULT
: -ERANGE
;
1263 len
= n
* sizeof(gid_t
);
1265 ret
= groups_to_user((gid_t __user
*)optval
,
1266 sk
->sk_peer_cred
->group_info
);
1276 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1280 if (copy_to_user(optval
, address
, len
))
1285 /* Dubious BSD thing... Probably nobody even uses it, but
1286 * the UNIX standard wants it for whatever reason... -DaveM
1289 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1293 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1297 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1300 v
.val
= sk
->sk_mark
;
1304 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1307 case SO_WIFI_STATUS
:
1308 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1312 if (!sock
->ops
->set_peek_off
)
1315 v
.val
= sk
->sk_peek_off
;
1318 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1321 case SO_BINDTODEVICE
:
1322 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1325 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1331 case SO_LOCK_FILTER
:
1332 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1335 case SO_BPF_EXTENSIONS
:
1336 v
.val
= bpf_tell_extensions();
1339 case SO_SELECT_ERR_QUEUE
:
1340 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1343 #ifdef CONFIG_NET_RX_BUSY_POLL
1345 v
.val
= sk
->sk_ll_usec
;
1349 case SO_MAX_PACING_RATE
:
1350 v
.val
= sk
->sk_max_pacing_rate
;
1353 case SO_INCOMING_CPU
:
1354 v
.val
= sk
->sk_incoming_cpu
;
1359 u32 meminfo
[SK_MEMINFO_VARS
];
1361 if (get_user(len
, optlen
))
1364 sk_get_meminfo(sk
, meminfo
);
1366 len
= min_t(unsigned int, len
, sizeof(meminfo
));
1367 if (copy_to_user(optval
, &meminfo
, len
))
1373 #ifdef CONFIG_NET_RX_BUSY_POLL
1374 case SO_INCOMING_NAPI_ID
:
1375 v
.val
= READ_ONCE(sk
->sk_napi_id
);
1377 /* aggregate non-NAPI IDs down to 0 */
1378 if (v
.val
< MIN_NAPI_ID
)
1388 v
.val64
= sock_gen_cookie(sk
);
1392 v
.val
= sock_flag(sk
, SOCK_ZEROCOPY
);
1396 /* We implement the SO_SNDLOWAT etc to not be settable
1399 return -ENOPROTOOPT
;
1404 if (copy_to_user(optval
, &v
, len
))
1407 if (put_user(len
, optlen
))
1413 * Initialize an sk_lock.
1415 * (We also register the sk_lock with the lock validator.)
1417 static inline void sock_lock_init(struct sock
*sk
)
1419 if (sk
->sk_kern_sock
)
1420 sock_lock_init_class_and_name(
1422 af_family_kern_slock_key_strings
[sk
->sk_family
],
1423 af_family_kern_slock_keys
+ sk
->sk_family
,
1424 af_family_kern_key_strings
[sk
->sk_family
],
1425 af_family_kern_keys
+ sk
->sk_family
);
1427 sock_lock_init_class_and_name(
1429 af_family_slock_key_strings
[sk
->sk_family
],
1430 af_family_slock_keys
+ sk
->sk_family
,
1431 af_family_key_strings
[sk
->sk_family
],
1432 af_family_keys
+ sk
->sk_family
);
1436 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1437 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1438 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1440 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1442 #ifdef CONFIG_SECURITY_NETWORK
1443 void *sptr
= nsk
->sk_security
;
1445 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1447 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1448 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1450 #ifdef CONFIG_SECURITY_NETWORK
1451 nsk
->sk_security
= sptr
;
1452 security_sk_clone(osk
, nsk
);
1456 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1460 struct kmem_cache
*slab
;
1464 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1467 if (priority
& __GFP_ZERO
)
1468 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1470 sk
= kmalloc(prot
->obj_size
, priority
);
1473 if (security_sk_alloc(sk
, family
, priority
))
1476 if (!try_module_get(prot
->owner
))
1478 sk_tx_queue_clear(sk
);
1484 security_sk_free(sk
);
1487 kmem_cache_free(slab
, sk
);
1493 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1495 struct kmem_cache
*slab
;
1496 struct module
*owner
;
1498 owner
= prot
->owner
;
1501 cgroup_sk_free(&sk
->sk_cgrp_data
);
1502 mem_cgroup_sk_free(sk
);
1503 security_sk_free(sk
);
1505 kmem_cache_free(slab
, sk
);
1512 * sk_alloc - All socket objects are allocated here
1513 * @net: the applicable net namespace
1514 * @family: protocol family
1515 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1516 * @prot: struct proto associated with this new sock instance
1517 * @kern: is this to be a kernel socket?
1519 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1520 struct proto
*prot
, int kern
)
1524 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1526 sk
->sk_family
= family
;
1528 * See comment in struct sock definition to understand
1529 * why we need sk_prot_creator -acme
1531 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1532 sk
->sk_kern_sock
= kern
;
1534 sk
->sk_net_refcnt
= kern
? 0 : 1;
1535 if (likely(sk
->sk_net_refcnt
))
1537 sock_net_set(sk
, net
);
1538 refcount_set(&sk
->sk_wmem_alloc
, 1);
1540 mem_cgroup_sk_alloc(sk
);
1541 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1542 sock_update_classid(&sk
->sk_cgrp_data
);
1543 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1548 EXPORT_SYMBOL(sk_alloc
);
1550 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1551 * grace period. This is the case for UDP sockets and TCP listeners.
1553 static void __sk_destruct(struct rcu_head
*head
)
1555 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1556 struct sk_filter
*filter
;
1558 if (sk
->sk_destruct
)
1559 sk
->sk_destruct(sk
);
1561 filter
= rcu_dereference_check(sk
->sk_filter
,
1562 refcount_read(&sk
->sk_wmem_alloc
) == 0);
1564 sk_filter_uncharge(sk
, filter
);
1565 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1567 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1568 reuseport_detach_sock(sk
);
1570 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1572 if (atomic_read(&sk
->sk_omem_alloc
))
1573 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1574 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1576 if (sk
->sk_frag
.page
) {
1577 put_page(sk
->sk_frag
.page
);
1578 sk
->sk_frag
.page
= NULL
;
1581 if (sk
->sk_peer_cred
)
1582 put_cred(sk
->sk_peer_cred
);
1583 put_pid(sk
->sk_peer_pid
);
1584 if (likely(sk
->sk_net_refcnt
))
1585 put_net(sock_net(sk
));
1586 sk_prot_free(sk
->sk_prot_creator
, sk
);
1589 void sk_destruct(struct sock
*sk
)
1591 if (sock_flag(sk
, SOCK_RCU_FREE
))
1592 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1594 __sk_destruct(&sk
->sk_rcu
);
1597 static void __sk_free(struct sock
*sk
)
1599 if (unlikely(sk
->sk_net_refcnt
&& sock_diag_has_destroy_listeners(sk
)))
1600 sock_diag_broadcast_destroy(sk
);
1605 void sk_free(struct sock
*sk
)
1608 * We subtract one from sk_wmem_alloc and can know if
1609 * some packets are still in some tx queue.
1610 * If not null, sock_wfree() will call __sk_free(sk) later
1612 if (refcount_dec_and_test(&sk
->sk_wmem_alloc
))
1615 EXPORT_SYMBOL(sk_free
);
1617 static void sk_init_common(struct sock
*sk
)
1619 skb_queue_head_init(&sk
->sk_receive_queue
);
1620 skb_queue_head_init(&sk
->sk_write_queue
);
1621 skb_queue_head_init(&sk
->sk_error_queue
);
1623 rwlock_init(&sk
->sk_callback_lock
);
1624 lockdep_set_class_and_name(&sk
->sk_receive_queue
.lock
,
1625 af_rlock_keys
+ sk
->sk_family
,
1626 af_family_rlock_key_strings
[sk
->sk_family
]);
1627 lockdep_set_class_and_name(&sk
->sk_write_queue
.lock
,
1628 af_wlock_keys
+ sk
->sk_family
,
1629 af_family_wlock_key_strings
[sk
->sk_family
]);
1630 lockdep_set_class_and_name(&sk
->sk_error_queue
.lock
,
1631 af_elock_keys
+ sk
->sk_family
,
1632 af_family_elock_key_strings
[sk
->sk_family
]);
1633 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
1634 af_callback_keys
+ sk
->sk_family
,
1635 af_family_clock_key_strings
[sk
->sk_family
]);
1639 * sk_clone_lock - clone a socket, and lock its clone
1640 * @sk: the socket to clone
1641 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1643 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1645 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1648 bool is_charged
= true;
1650 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1651 if (newsk
!= NULL
) {
1652 struct sk_filter
*filter
;
1654 sock_copy(newsk
, sk
);
1656 newsk
->sk_prot_creator
= sk
->sk_prot
;
1659 if (likely(newsk
->sk_net_refcnt
))
1660 get_net(sock_net(newsk
));
1661 sk_node_init(&newsk
->sk_node
);
1662 sock_lock_init(newsk
);
1663 bh_lock_sock(newsk
);
1664 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1665 newsk
->sk_backlog
.len
= 0;
1667 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1669 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1671 refcount_set(&newsk
->sk_wmem_alloc
, 1);
1672 atomic_set(&newsk
->sk_omem_alloc
, 0);
1673 sk_init_common(newsk
);
1675 newsk
->sk_dst_cache
= NULL
;
1676 newsk
->sk_dst_pending_confirm
= 0;
1677 newsk
->sk_wmem_queued
= 0;
1678 newsk
->sk_forward_alloc
= 0;
1679 atomic_set(&newsk
->sk_drops
, 0);
1680 newsk
->sk_send_head
= NULL
;
1681 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1682 atomic_set(&newsk
->sk_zckey
, 0);
1684 sock_reset_flag(newsk
, SOCK_DONE
);
1685 mem_cgroup_sk_alloc(newsk
);
1686 cgroup_sk_alloc(&newsk
->sk_cgrp_data
);
1689 filter
= rcu_dereference(sk
->sk_filter
);
1691 /* though it's an empty new sock, the charging may fail
1692 * if sysctl_optmem_max was changed between creation of
1693 * original socket and cloning
1695 is_charged
= sk_filter_charge(newsk
, filter
);
1696 RCU_INIT_POINTER(newsk
->sk_filter
, filter
);
1699 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1700 /* We need to make sure that we don't uncharge the new
1701 * socket if we couldn't charge it in the first place
1702 * as otherwise we uncharge the parent's filter.
1705 RCU_INIT_POINTER(newsk
->sk_filter
, NULL
);
1706 sk_free_unlock_clone(newsk
);
1710 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1713 newsk
->sk_err_soft
= 0;
1714 newsk
->sk_priority
= 0;
1715 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1716 atomic64_set(&newsk
->sk_cookie
, 0);
1719 * Before updating sk_refcnt, we must commit prior changes to memory
1720 * (Documentation/RCU/rculist_nulls.txt for details)
1723 refcount_set(&newsk
->sk_refcnt
, 2);
1726 * Increment the counter in the same struct proto as the master
1727 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1728 * is the same as sk->sk_prot->socks, as this field was copied
1731 * This _changes_ the previous behaviour, where
1732 * tcp_create_openreq_child always was incrementing the
1733 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1734 * to be taken into account in all callers. -acme
1736 sk_refcnt_debug_inc(newsk
);
1737 sk_set_socket(newsk
, NULL
);
1738 newsk
->sk_wq
= NULL
;
1740 if (newsk
->sk_prot
->sockets_allocated
)
1741 sk_sockets_allocated_inc(newsk
);
1743 if (sock_needs_netstamp(sk
) &&
1744 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1745 net_enable_timestamp();
1750 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1752 void sk_free_unlock_clone(struct sock
*sk
)
1754 /* It is still raw copy of parent, so invalidate
1755 * destructor and make plain sk_free() */
1756 sk
->sk_destruct
= NULL
;
1760 EXPORT_SYMBOL_GPL(sk_free_unlock_clone
);
1762 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1766 sk_dst_set(sk
, dst
);
1767 sk
->sk_route_caps
= dst
->dev
->features
;
1768 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1769 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1770 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1771 if (sk_can_gso(sk
)) {
1772 if (dst
->header_len
&& !xfrm_dst_offload_ok(dst
)) {
1773 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1775 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1776 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1777 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1780 sk
->sk_gso_max_segs
= max_segs
;
1782 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1785 * Simple resource managers for sockets.
1790 * Write buffer destructor automatically called from kfree_skb.
1792 void sock_wfree(struct sk_buff
*skb
)
1794 struct sock
*sk
= skb
->sk
;
1795 unsigned int len
= skb
->truesize
;
1797 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1799 * Keep a reference on sk_wmem_alloc, this will be released
1800 * after sk_write_space() call
1802 WARN_ON(refcount_sub_and_test(len
- 1, &sk
->sk_wmem_alloc
));
1803 sk
->sk_write_space(sk
);
1807 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1808 * could not do because of in-flight packets
1810 if (refcount_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1813 EXPORT_SYMBOL(sock_wfree
);
1815 /* This variant of sock_wfree() is used by TCP,
1816 * since it sets SOCK_USE_WRITE_QUEUE.
1818 void __sock_wfree(struct sk_buff
*skb
)
1820 struct sock
*sk
= skb
->sk
;
1822 if (refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
))
1826 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1831 if (unlikely(!sk_fullsock(sk
))) {
1832 skb
->destructor
= sock_edemux
;
1837 skb
->destructor
= sock_wfree
;
1838 skb_set_hash_from_sk(skb
, sk
);
1840 * We used to take a refcount on sk, but following operation
1841 * is enough to guarantee sk_free() wont free this sock until
1842 * all in-flight packets are completed
1844 refcount_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1846 EXPORT_SYMBOL(skb_set_owner_w
);
1848 /* This helper is used by netem, as it can hold packets in its
1849 * delay queue. We want to allow the owner socket to send more
1850 * packets, as if they were already TX completed by a typical driver.
1851 * But we also want to keep skb->sk set because some packet schedulers
1852 * rely on it (sch_fq for example).
1854 void skb_orphan_partial(struct sk_buff
*skb
)
1856 if (skb_is_tcp_pure_ack(skb
))
1859 if (skb
->destructor
== sock_wfree
1861 || skb
->destructor
== tcp_wfree
1864 struct sock
*sk
= skb
->sk
;
1866 if (refcount_inc_not_zero(&sk
->sk_refcnt
)) {
1867 WARN_ON(refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
));
1868 skb
->destructor
= sock_efree
;
1874 EXPORT_SYMBOL(skb_orphan_partial
);
1877 * Read buffer destructor automatically called from kfree_skb.
1879 void sock_rfree(struct sk_buff
*skb
)
1881 struct sock
*sk
= skb
->sk
;
1882 unsigned int len
= skb
->truesize
;
1884 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1885 sk_mem_uncharge(sk
, len
);
1887 EXPORT_SYMBOL(sock_rfree
);
1890 * Buffer destructor for skbs that are not used directly in read or write
1891 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1893 void sock_efree(struct sk_buff
*skb
)
1897 EXPORT_SYMBOL(sock_efree
);
1899 kuid_t
sock_i_uid(struct sock
*sk
)
1903 read_lock_bh(&sk
->sk_callback_lock
);
1904 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1905 read_unlock_bh(&sk
->sk_callback_lock
);
1908 EXPORT_SYMBOL(sock_i_uid
);
1910 unsigned long sock_i_ino(struct sock
*sk
)
1914 read_lock_bh(&sk
->sk_callback_lock
);
1915 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1916 read_unlock_bh(&sk
->sk_callback_lock
);
1919 EXPORT_SYMBOL(sock_i_ino
);
1922 * Allocate a skb from the socket's send buffer.
1924 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1927 if (force
|| refcount_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1928 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1930 skb_set_owner_w(skb
, sk
);
1936 EXPORT_SYMBOL(sock_wmalloc
);
1938 static void sock_ofree(struct sk_buff
*skb
)
1940 struct sock
*sk
= skb
->sk
;
1942 atomic_sub(skb
->truesize
, &sk
->sk_omem_alloc
);
1945 struct sk_buff
*sock_omalloc(struct sock
*sk
, unsigned long size
,
1948 struct sk_buff
*skb
;
1950 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1951 if (atomic_read(&sk
->sk_omem_alloc
) + SKB_TRUESIZE(size
) >
1955 skb
= alloc_skb(size
, priority
);
1959 atomic_add(skb
->truesize
, &sk
->sk_omem_alloc
);
1961 skb
->destructor
= sock_ofree
;
1966 * Allocate a memory block from the socket's option memory buffer.
1968 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1970 if ((unsigned int)size
<= sysctl_optmem_max
&&
1971 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1973 /* First do the add, to avoid the race if kmalloc
1976 atomic_add(size
, &sk
->sk_omem_alloc
);
1977 mem
= kmalloc(size
, priority
);
1980 atomic_sub(size
, &sk
->sk_omem_alloc
);
1984 EXPORT_SYMBOL(sock_kmalloc
);
1986 /* Free an option memory block. Note, we actually want the inline
1987 * here as this allows gcc to detect the nullify and fold away the
1988 * condition entirely.
1990 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1993 if (WARN_ON_ONCE(!mem
))
1999 atomic_sub(size
, &sk
->sk_omem_alloc
);
2002 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
2004 __sock_kfree_s(sk
, mem
, size
, false);
2006 EXPORT_SYMBOL(sock_kfree_s
);
2008 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
2010 __sock_kfree_s(sk
, mem
, size
, true);
2012 EXPORT_SYMBOL(sock_kzfree_s
);
2014 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2015 I think, these locks should be removed for datagram sockets.
2017 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
2021 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2025 if (signal_pending(current
))
2027 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2028 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2029 if (refcount_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
2031 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2035 timeo
= schedule_timeout(timeo
);
2037 finish_wait(sk_sleep(sk
), &wait
);
2043 * Generic send/receive buffer handlers
2046 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
2047 unsigned long data_len
, int noblock
,
2048 int *errcode
, int max_page_order
)
2050 struct sk_buff
*skb
;
2054 timeo
= sock_sndtimeo(sk
, noblock
);
2056 err
= sock_error(sk
);
2061 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2064 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
2067 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2068 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2072 if (signal_pending(current
))
2074 timeo
= sock_wait_for_wmem(sk
, timeo
);
2076 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
2077 errcode
, sk
->sk_allocation
);
2079 skb_set_owner_w(skb
, sk
);
2083 err
= sock_intr_errno(timeo
);
2088 EXPORT_SYMBOL(sock_alloc_send_pskb
);
2090 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
2091 int noblock
, int *errcode
)
2093 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
2095 EXPORT_SYMBOL(sock_alloc_send_skb
);
2097 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
2098 struct sockcm_cookie
*sockc
)
2102 switch (cmsg
->cmsg_type
) {
2104 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
2106 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2108 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
2110 case SO_TIMESTAMPING
:
2111 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2114 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
2115 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
2118 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
2119 sockc
->tsflags
|= tsflags
;
2121 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2123 case SCM_CREDENTIALS
:
2130 EXPORT_SYMBOL(__sock_cmsg_send
);
2132 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
2133 struct sockcm_cookie
*sockc
)
2135 struct cmsghdr
*cmsg
;
2138 for_each_cmsghdr(cmsg
, msg
) {
2139 if (!CMSG_OK(msg
, cmsg
))
2141 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
2143 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
2149 EXPORT_SYMBOL(sock_cmsg_send
);
2151 static void sk_enter_memory_pressure(struct sock
*sk
)
2153 if (!sk
->sk_prot
->enter_memory_pressure
)
2156 sk
->sk_prot
->enter_memory_pressure(sk
);
2159 static void sk_leave_memory_pressure(struct sock
*sk
)
2161 if (sk
->sk_prot
->leave_memory_pressure
) {
2162 sk
->sk_prot
->leave_memory_pressure(sk
);
2164 unsigned long *memory_pressure
= sk
->sk_prot
->memory_pressure
;
2166 if (memory_pressure
&& *memory_pressure
)
2167 *memory_pressure
= 0;
2171 /* On 32bit arches, an skb frag is limited to 2^15 */
2172 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2175 * skb_page_frag_refill - check that a page_frag contains enough room
2176 * @sz: minimum size of the fragment we want to get
2177 * @pfrag: pointer to page_frag
2178 * @gfp: priority for memory allocation
2180 * Note: While this allocator tries to use high order pages, there is
2181 * no guarantee that allocations succeed. Therefore, @sz MUST be
2182 * less or equal than PAGE_SIZE.
2184 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
2187 if (page_ref_count(pfrag
->page
) == 1) {
2191 if (pfrag
->offset
+ sz
<= pfrag
->size
)
2193 put_page(pfrag
->page
);
2197 if (SKB_FRAG_PAGE_ORDER
) {
2198 /* Avoid direct reclaim but allow kswapd to wake */
2199 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
2200 __GFP_COMP
| __GFP_NOWARN
|
2202 SKB_FRAG_PAGE_ORDER
);
2203 if (likely(pfrag
->page
)) {
2204 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
2208 pfrag
->page
= alloc_page(gfp
);
2209 if (likely(pfrag
->page
)) {
2210 pfrag
->size
= PAGE_SIZE
;
2215 EXPORT_SYMBOL(skb_page_frag_refill
);
2217 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
2219 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
2222 sk_enter_memory_pressure(sk
);
2223 sk_stream_moderate_sndbuf(sk
);
2226 EXPORT_SYMBOL(sk_page_frag_refill
);
2228 static void __lock_sock(struct sock
*sk
)
2229 __releases(&sk
->sk_lock
.slock
)
2230 __acquires(&sk
->sk_lock
.slock
)
2235 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2236 TASK_UNINTERRUPTIBLE
);
2237 spin_unlock_bh(&sk
->sk_lock
.slock
);
2239 spin_lock_bh(&sk
->sk_lock
.slock
);
2240 if (!sock_owned_by_user(sk
))
2243 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2246 void __release_sock(struct sock
*sk
)
2247 __releases(&sk
->sk_lock
.slock
)
2248 __acquires(&sk
->sk_lock
.slock
)
2250 struct sk_buff
*skb
, *next
;
2252 while ((skb
= sk
->sk_backlog
.head
) != NULL
) {
2253 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2255 spin_unlock_bh(&sk
->sk_lock
.slock
);
2260 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2262 sk_backlog_rcv(sk
, skb
);
2267 } while (skb
!= NULL
);
2269 spin_lock_bh(&sk
->sk_lock
.slock
);
2273 * Doing the zeroing here guarantee we can not loop forever
2274 * while a wild producer attempts to flood us.
2276 sk
->sk_backlog
.len
= 0;
2279 void __sk_flush_backlog(struct sock
*sk
)
2281 spin_lock_bh(&sk
->sk_lock
.slock
);
2283 spin_unlock_bh(&sk
->sk_lock
.slock
);
2287 * sk_wait_data - wait for data to arrive at sk_receive_queue
2288 * @sk: sock to wait on
2289 * @timeo: for how long
2290 * @skb: last skb seen on sk_receive_queue
2292 * Now socket state including sk->sk_err is changed only under lock,
2293 * hence we may omit checks after joining wait queue.
2294 * We check receive queue before schedule() only as optimization;
2295 * it is very likely that release_sock() added new data.
2297 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2299 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
2302 add_wait_queue(sk_sleep(sk
), &wait
);
2303 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2304 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
, &wait
);
2305 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2306 remove_wait_queue(sk_sleep(sk
), &wait
);
2309 EXPORT_SYMBOL(sk_wait_data
);
2312 * __sk_mem_raise_allocated - increase memory_allocated
2314 * @size: memory size to allocate
2315 * @amt: pages to allocate
2316 * @kind: allocation type
2318 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2320 int __sk_mem_raise_allocated(struct sock
*sk
, int size
, int amt
, int kind
)
2322 struct proto
*prot
= sk
->sk_prot
;
2323 long allocated
= sk_memory_allocated_add(sk
, amt
);
2325 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2326 !mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
))
2327 goto suppress_allocation
;
2330 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2331 sk_leave_memory_pressure(sk
);
2335 /* Under pressure. */
2336 if (allocated
> sk_prot_mem_limits(sk
, 1))
2337 sk_enter_memory_pressure(sk
);
2339 /* Over hard limit. */
2340 if (allocated
> sk_prot_mem_limits(sk
, 2))
2341 goto suppress_allocation
;
2343 /* guarantee minimum buffer size under pressure */
2344 if (kind
== SK_MEM_RECV
) {
2345 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2348 } else { /* SK_MEM_SEND */
2349 if (sk
->sk_type
== SOCK_STREAM
) {
2350 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2352 } else if (refcount_read(&sk
->sk_wmem_alloc
) <
2353 prot
->sysctl_wmem
[0])
2357 if (sk_has_memory_pressure(sk
)) {
2360 if (!sk_under_memory_pressure(sk
))
2362 alloc
= sk_sockets_allocated_read_positive(sk
);
2363 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2364 sk_mem_pages(sk
->sk_wmem_queued
+
2365 atomic_read(&sk
->sk_rmem_alloc
) +
2366 sk
->sk_forward_alloc
))
2370 suppress_allocation
:
2372 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2373 sk_stream_moderate_sndbuf(sk
);
2375 /* Fail only if socket is _under_ its sndbuf.
2376 * In this case we cannot block, so that we have to fail.
2378 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2382 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2384 sk_memory_allocated_sub(sk
, amt
);
2386 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2387 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2391 EXPORT_SYMBOL(__sk_mem_raise_allocated
);
2394 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2396 * @size: memory size to allocate
2397 * @kind: allocation type
2399 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2400 * rmem allocation. This function assumes that protocols which have
2401 * memory_pressure use sk_wmem_queued as write buffer accounting.
2403 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2405 int ret
, amt
= sk_mem_pages(size
);
2407 sk
->sk_forward_alloc
+= amt
<< SK_MEM_QUANTUM_SHIFT
;
2408 ret
= __sk_mem_raise_allocated(sk
, size
, amt
, kind
);
2410 sk
->sk_forward_alloc
-= amt
<< SK_MEM_QUANTUM_SHIFT
;
2413 EXPORT_SYMBOL(__sk_mem_schedule
);
2416 * __sk_mem_reduce_allocated - reclaim memory_allocated
2418 * @amount: number of quanta
2420 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2422 void __sk_mem_reduce_allocated(struct sock
*sk
, int amount
)
2424 sk_memory_allocated_sub(sk
, amount
);
2426 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2427 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2429 if (sk_under_memory_pressure(sk
) &&
2430 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2431 sk_leave_memory_pressure(sk
);
2433 EXPORT_SYMBOL(__sk_mem_reduce_allocated
);
2436 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2438 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2440 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2442 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2443 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2444 __sk_mem_reduce_allocated(sk
, amount
);
2446 EXPORT_SYMBOL(__sk_mem_reclaim
);
2448 int sk_set_peek_off(struct sock
*sk
, int val
)
2450 sk
->sk_peek_off
= val
;
2453 EXPORT_SYMBOL_GPL(sk_set_peek_off
);
2456 * Set of default routines for initialising struct proto_ops when
2457 * the protocol does not support a particular function. In certain
2458 * cases where it makes no sense for a protocol to have a "do nothing"
2459 * function, some default processing is provided.
2462 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2466 EXPORT_SYMBOL(sock_no_bind
);
2468 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2473 EXPORT_SYMBOL(sock_no_connect
);
2475 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2479 EXPORT_SYMBOL(sock_no_socketpair
);
2481 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
,
2486 EXPORT_SYMBOL(sock_no_accept
);
2488 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2493 EXPORT_SYMBOL(sock_no_getname
);
2495 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2499 EXPORT_SYMBOL(sock_no_poll
);
2501 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2505 EXPORT_SYMBOL(sock_no_ioctl
);
2507 int sock_no_listen(struct socket
*sock
, int backlog
)
2511 EXPORT_SYMBOL(sock_no_listen
);
2513 int sock_no_shutdown(struct socket
*sock
, int how
)
2517 EXPORT_SYMBOL(sock_no_shutdown
);
2519 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2520 char __user
*optval
, unsigned int optlen
)
2524 EXPORT_SYMBOL(sock_no_setsockopt
);
2526 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2527 char __user
*optval
, int __user
*optlen
)
2531 EXPORT_SYMBOL(sock_no_getsockopt
);
2533 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2537 EXPORT_SYMBOL(sock_no_sendmsg
);
2539 int sock_no_sendmsg_locked(struct sock
*sk
, struct msghdr
*m
, size_t len
)
2543 EXPORT_SYMBOL(sock_no_sendmsg_locked
);
2545 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2550 EXPORT_SYMBOL(sock_no_recvmsg
);
2552 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2554 /* Mirror missing mmap method error code */
2557 EXPORT_SYMBOL(sock_no_mmap
);
2559 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2562 struct msghdr msg
= {.msg_flags
= flags
};
2564 char *kaddr
= kmap(page
);
2565 iov
.iov_base
= kaddr
+ offset
;
2567 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2571 EXPORT_SYMBOL(sock_no_sendpage
);
2573 ssize_t
sock_no_sendpage_locked(struct sock
*sk
, struct page
*page
,
2574 int offset
, size_t size
, int flags
)
2577 struct msghdr msg
= {.msg_flags
= flags
};
2579 char *kaddr
= kmap(page
);
2581 iov
.iov_base
= kaddr
+ offset
;
2583 res
= kernel_sendmsg_locked(sk
, &msg
, &iov
, 1, size
);
2587 EXPORT_SYMBOL(sock_no_sendpage_locked
);
2590 * Default Socket Callbacks
2593 static void sock_def_wakeup(struct sock
*sk
)
2595 struct socket_wq
*wq
;
2598 wq
= rcu_dereference(sk
->sk_wq
);
2599 if (skwq_has_sleeper(wq
))
2600 wake_up_interruptible_all(&wq
->wait
);
2604 static void sock_def_error_report(struct sock
*sk
)
2606 struct socket_wq
*wq
;
2609 wq
= rcu_dereference(sk
->sk_wq
);
2610 if (skwq_has_sleeper(wq
))
2611 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2612 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2616 static void sock_def_readable(struct sock
*sk
)
2618 struct socket_wq
*wq
;
2621 wq
= rcu_dereference(sk
->sk_wq
);
2622 if (skwq_has_sleeper(wq
))
2623 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2624 POLLRDNORM
| POLLRDBAND
);
2625 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2629 static void sock_def_write_space(struct sock
*sk
)
2631 struct socket_wq
*wq
;
2635 /* Do not wake up a writer until he can make "significant"
2638 if ((refcount_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2639 wq
= rcu_dereference(sk
->sk_wq
);
2640 if (skwq_has_sleeper(wq
))
2641 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2642 POLLWRNORM
| POLLWRBAND
);
2644 /* Should agree with poll, otherwise some programs break */
2645 if (sock_writeable(sk
))
2646 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2652 static void sock_def_destruct(struct sock
*sk
)
2656 void sk_send_sigurg(struct sock
*sk
)
2658 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2659 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2660 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2662 EXPORT_SYMBOL(sk_send_sigurg
);
2664 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2665 unsigned long expires
)
2667 if (!mod_timer(timer
, expires
))
2670 EXPORT_SYMBOL(sk_reset_timer
);
2672 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2674 if (del_timer(timer
))
2677 EXPORT_SYMBOL(sk_stop_timer
);
2679 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2682 sk
->sk_send_head
= NULL
;
2684 init_timer(&sk
->sk_timer
);
2686 sk
->sk_allocation
= GFP_KERNEL
;
2687 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2688 sk
->sk_sndbuf
= sysctl_wmem_default
;
2689 sk
->sk_state
= TCP_CLOSE
;
2690 sk_set_socket(sk
, sock
);
2692 sock_set_flag(sk
, SOCK_ZAPPED
);
2695 sk
->sk_type
= sock
->type
;
2696 sk
->sk_wq
= sock
->wq
;
2698 sk
->sk_uid
= SOCK_INODE(sock
)->i_uid
;
2701 sk
->sk_uid
= make_kuid(sock_net(sk
)->user_ns
, 0);
2704 rwlock_init(&sk
->sk_callback_lock
);
2705 if (sk
->sk_kern_sock
)
2706 lockdep_set_class_and_name(
2707 &sk
->sk_callback_lock
,
2708 af_kern_callback_keys
+ sk
->sk_family
,
2709 af_family_kern_clock_key_strings
[sk
->sk_family
]);
2711 lockdep_set_class_and_name(
2712 &sk
->sk_callback_lock
,
2713 af_callback_keys
+ sk
->sk_family
,
2714 af_family_clock_key_strings
[sk
->sk_family
]);
2716 sk
->sk_state_change
= sock_def_wakeup
;
2717 sk
->sk_data_ready
= sock_def_readable
;
2718 sk
->sk_write_space
= sock_def_write_space
;
2719 sk
->sk_error_report
= sock_def_error_report
;
2720 sk
->sk_destruct
= sock_def_destruct
;
2722 sk
->sk_frag
.page
= NULL
;
2723 sk
->sk_frag
.offset
= 0;
2724 sk
->sk_peek_off
= -1;
2726 sk
->sk_peer_pid
= NULL
;
2727 sk
->sk_peer_cred
= NULL
;
2728 sk
->sk_write_pending
= 0;
2729 sk
->sk_rcvlowat
= 1;
2730 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2731 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2733 sk
->sk_stamp
= SK_DEFAULT_STAMP
;
2734 #if BITS_PER_LONG==32
2735 seqlock_init(&sk
->sk_stamp_seq
);
2737 atomic_set(&sk
->sk_zckey
, 0);
2739 #ifdef CONFIG_NET_RX_BUSY_POLL
2741 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2744 sk
->sk_max_pacing_rate
= ~0U;
2745 sk
->sk_pacing_rate
= ~0U;
2746 sk
->sk_incoming_cpu
= -1;
2748 * Before updating sk_refcnt, we must commit prior changes to memory
2749 * (Documentation/RCU/rculist_nulls.txt for details)
2752 refcount_set(&sk
->sk_refcnt
, 1);
2753 atomic_set(&sk
->sk_drops
, 0);
2755 EXPORT_SYMBOL(sock_init_data
);
2757 void lock_sock_nested(struct sock
*sk
, int subclass
)
2760 spin_lock_bh(&sk
->sk_lock
.slock
);
2761 if (sk
->sk_lock
.owned
)
2763 sk
->sk_lock
.owned
= 1;
2764 spin_unlock(&sk
->sk_lock
.slock
);
2766 * The sk_lock has mutex_lock() semantics here:
2768 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2771 EXPORT_SYMBOL(lock_sock_nested
);
2773 void release_sock(struct sock
*sk
)
2775 spin_lock_bh(&sk
->sk_lock
.slock
);
2776 if (sk
->sk_backlog
.tail
)
2779 /* Warning : release_cb() might need to release sk ownership,
2780 * ie call sock_release_ownership(sk) before us.
2782 if (sk
->sk_prot
->release_cb
)
2783 sk
->sk_prot
->release_cb(sk
);
2785 sock_release_ownership(sk
);
2786 if (waitqueue_active(&sk
->sk_lock
.wq
))
2787 wake_up(&sk
->sk_lock
.wq
);
2788 spin_unlock_bh(&sk
->sk_lock
.slock
);
2790 EXPORT_SYMBOL(release_sock
);
2793 * lock_sock_fast - fast version of lock_sock
2796 * This version should be used for very small section, where process wont block
2797 * return false if fast path is taken:
2799 * sk_lock.slock locked, owned = 0, BH disabled
2801 * return true if slow path is taken:
2803 * sk_lock.slock unlocked, owned = 1, BH enabled
2805 bool lock_sock_fast(struct sock
*sk
)
2808 spin_lock_bh(&sk
->sk_lock
.slock
);
2810 if (!sk
->sk_lock
.owned
)
2812 * Note : We must disable BH
2817 sk
->sk_lock
.owned
= 1;
2818 spin_unlock(&sk
->sk_lock
.slock
);
2820 * The sk_lock has mutex_lock() semantics here:
2822 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2826 EXPORT_SYMBOL(lock_sock_fast
);
2828 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2831 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2832 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2833 tv
= ktime_to_timeval(sk
->sk_stamp
);
2834 if (tv
.tv_sec
== -1)
2836 if (tv
.tv_sec
== 0) {
2837 sk
->sk_stamp
= ktime_get_real();
2838 tv
= ktime_to_timeval(sk
->sk_stamp
);
2840 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2842 EXPORT_SYMBOL(sock_get_timestamp
);
2844 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2847 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2848 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2849 ts
= ktime_to_timespec(sk
->sk_stamp
);
2850 if (ts
.tv_sec
== -1)
2852 if (ts
.tv_sec
== 0) {
2853 sk
->sk_stamp
= ktime_get_real();
2854 ts
= ktime_to_timespec(sk
->sk_stamp
);
2856 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2858 EXPORT_SYMBOL(sock_get_timestampns
);
2860 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2862 if (!sock_flag(sk
, flag
)) {
2863 unsigned long previous_flags
= sk
->sk_flags
;
2865 sock_set_flag(sk
, flag
);
2867 * we just set one of the two flags which require net
2868 * time stamping, but time stamping might have been on
2869 * already because of the other one
2871 if (sock_needs_netstamp(sk
) &&
2872 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
2873 net_enable_timestamp();
2877 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2878 int level
, int type
)
2880 struct sock_exterr_skb
*serr
;
2881 struct sk_buff
*skb
;
2885 skb
= sock_dequeue_err_skb(sk
);
2891 msg
->msg_flags
|= MSG_TRUNC
;
2894 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2898 sock_recv_timestamp(msg
, sk
, skb
);
2900 serr
= SKB_EXT_ERR(skb
);
2901 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2903 msg
->msg_flags
|= MSG_ERRQUEUE
;
2911 EXPORT_SYMBOL(sock_recv_errqueue
);
2914 * Get a socket option on an socket.
2916 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2917 * asynchronous errors should be reported by getsockopt. We assume
2918 * this means if you specify SO_ERROR (otherwise whats the point of it).
2920 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2921 char __user
*optval
, int __user
*optlen
)
2923 struct sock
*sk
= sock
->sk
;
2925 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2927 EXPORT_SYMBOL(sock_common_getsockopt
);
2929 #ifdef CONFIG_COMPAT
2930 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2931 char __user
*optval
, int __user
*optlen
)
2933 struct sock
*sk
= sock
->sk
;
2935 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2936 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2938 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2940 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2943 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2946 struct sock
*sk
= sock
->sk
;
2950 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2951 flags
& ~MSG_DONTWAIT
, &addr_len
);
2953 msg
->msg_namelen
= addr_len
;
2956 EXPORT_SYMBOL(sock_common_recvmsg
);
2959 * Set socket options on an inet socket.
2961 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2962 char __user
*optval
, unsigned int optlen
)
2964 struct sock
*sk
= sock
->sk
;
2966 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2968 EXPORT_SYMBOL(sock_common_setsockopt
);
2970 #ifdef CONFIG_COMPAT
2971 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2972 char __user
*optval
, unsigned int optlen
)
2974 struct sock
*sk
= sock
->sk
;
2976 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2977 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2979 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2981 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2984 void sk_common_release(struct sock
*sk
)
2986 if (sk
->sk_prot
->destroy
)
2987 sk
->sk_prot
->destroy(sk
);
2990 * Observation: when sock_common_release is called, processes have
2991 * no access to socket. But net still has.
2992 * Step one, detach it from networking:
2994 * A. Remove from hash tables.
2997 sk
->sk_prot
->unhash(sk
);
3000 * In this point socket cannot receive new packets, but it is possible
3001 * that some packets are in flight because some CPU runs receiver and
3002 * did hash table lookup before we unhashed socket. They will achieve
3003 * receive queue and will be purged by socket destructor.
3005 * Also we still have packets pending on receive queue and probably,
3006 * our own packets waiting in device queues. sock_destroy will drain
3007 * receive queue, but transmitted packets will delay socket destruction
3008 * until the last reference will be released.
3013 xfrm_sk_free_policy(sk
);
3015 sk_refcnt_debug_release(sk
);
3019 EXPORT_SYMBOL(sk_common_release
);
3021 void sk_get_meminfo(const struct sock
*sk
, u32
*mem
)
3023 memset(mem
, 0, sizeof(*mem
) * SK_MEMINFO_VARS
);
3025 mem
[SK_MEMINFO_RMEM_ALLOC
] = sk_rmem_alloc_get(sk
);
3026 mem
[SK_MEMINFO_RCVBUF
] = sk
->sk_rcvbuf
;
3027 mem
[SK_MEMINFO_WMEM_ALLOC
] = sk_wmem_alloc_get(sk
);
3028 mem
[SK_MEMINFO_SNDBUF
] = sk
->sk_sndbuf
;
3029 mem
[SK_MEMINFO_FWD_ALLOC
] = sk
->sk_forward_alloc
;
3030 mem
[SK_MEMINFO_WMEM_QUEUED
] = sk
->sk_wmem_queued
;
3031 mem
[SK_MEMINFO_OPTMEM
] = atomic_read(&sk
->sk_omem_alloc
);
3032 mem
[SK_MEMINFO_BACKLOG
] = sk
->sk_backlog
.len
;
3033 mem
[SK_MEMINFO_DROPS
] = atomic_read(&sk
->sk_drops
);
3036 #ifdef CONFIG_PROC_FS
3037 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3039 int val
[PROTO_INUSE_NR
];
3042 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
3044 #ifdef CONFIG_NET_NS
3045 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
3047 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
3049 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
3051 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
3053 int cpu
, idx
= prot
->inuse_idx
;
3056 for_each_possible_cpu(cpu
)
3057 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
3059 return res
>= 0 ? res
: 0;
3061 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
3063 static int __net_init
sock_inuse_init_net(struct net
*net
)
3065 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
3066 return net
->core
.inuse
? 0 : -ENOMEM
;
3069 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
3071 free_percpu(net
->core
.inuse
);
3074 static struct pernet_operations net_inuse_ops
= {
3075 .init
= sock_inuse_init_net
,
3076 .exit
= sock_inuse_exit_net
,
3079 static __init
int net_inuse_init(void)
3081 if (register_pernet_subsys(&net_inuse_ops
))
3082 panic("Cannot initialize net inuse counters");
3087 core_initcall(net_inuse_init
);
3089 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
3091 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
3093 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
3095 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
3097 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
3099 int cpu
, idx
= prot
->inuse_idx
;
3102 for_each_possible_cpu(cpu
)
3103 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
3105 return res
>= 0 ? res
: 0;
3107 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
3110 static void assign_proto_idx(struct proto
*prot
)
3112 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
3114 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
3115 pr_err("PROTO_INUSE_NR exhausted\n");
3119 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
3122 static void release_proto_idx(struct proto
*prot
)
3124 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
3125 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
3128 static inline void assign_proto_idx(struct proto
*prot
)
3132 static inline void release_proto_idx(struct proto
*prot
)
3137 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
3141 kfree(rsk_prot
->slab_name
);
3142 rsk_prot
->slab_name
= NULL
;
3143 kmem_cache_destroy(rsk_prot
->slab
);
3144 rsk_prot
->slab
= NULL
;
3147 static int req_prot_init(const struct proto
*prot
)
3149 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
3154 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
3156 if (!rsk_prot
->slab_name
)
3159 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
3160 rsk_prot
->obj_size
, 0,
3161 prot
->slab_flags
, NULL
);
3163 if (!rsk_prot
->slab
) {
3164 pr_crit("%s: Can't create request sock SLAB cache!\n",
3171 int proto_register(struct proto
*prot
, int alloc_slab
)
3174 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
3175 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
3178 if (prot
->slab
== NULL
) {
3179 pr_crit("%s: Can't create sock SLAB cache!\n",
3184 if (req_prot_init(prot
))
3185 goto out_free_request_sock_slab
;
3187 if (prot
->twsk_prot
!= NULL
) {
3188 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
3190 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
3191 goto out_free_request_sock_slab
;
3193 prot
->twsk_prot
->twsk_slab
=
3194 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
3195 prot
->twsk_prot
->twsk_obj_size
,
3199 if (prot
->twsk_prot
->twsk_slab
== NULL
)
3200 goto out_free_timewait_sock_slab_name
;
3204 mutex_lock(&proto_list_mutex
);
3205 list_add(&prot
->node
, &proto_list
);
3206 assign_proto_idx(prot
);
3207 mutex_unlock(&proto_list_mutex
);
3210 out_free_timewait_sock_slab_name
:
3211 kfree(prot
->twsk_prot
->twsk_slab_name
);
3212 out_free_request_sock_slab
:
3213 req_prot_cleanup(prot
->rsk_prot
);
3215 kmem_cache_destroy(prot
->slab
);
3220 EXPORT_SYMBOL(proto_register
);
3222 void proto_unregister(struct proto
*prot
)
3224 mutex_lock(&proto_list_mutex
);
3225 release_proto_idx(prot
);
3226 list_del(&prot
->node
);
3227 mutex_unlock(&proto_list_mutex
);
3229 kmem_cache_destroy(prot
->slab
);
3232 req_prot_cleanup(prot
->rsk_prot
);
3234 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
3235 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
3236 kfree(prot
->twsk_prot
->twsk_slab_name
);
3237 prot
->twsk_prot
->twsk_slab
= NULL
;
3240 EXPORT_SYMBOL(proto_unregister
);
3242 #ifdef CONFIG_PROC_FS
3243 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3244 __acquires(proto_list_mutex
)
3246 mutex_lock(&proto_list_mutex
);
3247 return seq_list_start_head(&proto_list
, *pos
);
3250 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3252 return seq_list_next(v
, &proto_list
, pos
);
3255 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
3256 __releases(proto_list_mutex
)
3258 mutex_unlock(&proto_list_mutex
);
3261 static char proto_method_implemented(const void *method
)
3263 return method
== NULL
? 'n' : 'y';
3265 static long sock_prot_memory_allocated(struct proto
*proto
)
3267 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
3270 static char *sock_prot_memory_pressure(struct proto
*proto
)
3272 return proto
->memory_pressure
!= NULL
?
3273 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
3276 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
3279 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3280 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3283 sock_prot_inuse_get(seq_file_net(seq
), proto
),
3284 sock_prot_memory_allocated(proto
),
3285 sock_prot_memory_pressure(proto
),
3287 proto
->slab
== NULL
? "no" : "yes",
3288 module_name(proto
->owner
),
3289 proto_method_implemented(proto
->close
),
3290 proto_method_implemented(proto
->connect
),
3291 proto_method_implemented(proto
->disconnect
),
3292 proto_method_implemented(proto
->accept
),
3293 proto_method_implemented(proto
->ioctl
),
3294 proto_method_implemented(proto
->init
),
3295 proto_method_implemented(proto
->destroy
),
3296 proto_method_implemented(proto
->shutdown
),
3297 proto_method_implemented(proto
->setsockopt
),
3298 proto_method_implemented(proto
->getsockopt
),
3299 proto_method_implemented(proto
->sendmsg
),
3300 proto_method_implemented(proto
->recvmsg
),
3301 proto_method_implemented(proto
->sendpage
),
3302 proto_method_implemented(proto
->bind
),
3303 proto_method_implemented(proto
->backlog_rcv
),
3304 proto_method_implemented(proto
->hash
),
3305 proto_method_implemented(proto
->unhash
),
3306 proto_method_implemented(proto
->get_port
),
3307 proto_method_implemented(proto
->enter_memory_pressure
));
3310 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3312 if (v
== &proto_list
)
3313 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3322 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3324 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3328 static const struct seq_operations proto_seq_ops
= {
3329 .start
= proto_seq_start
,
3330 .next
= proto_seq_next
,
3331 .stop
= proto_seq_stop
,
3332 .show
= proto_seq_show
,
3335 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
3337 return seq_open_net(inode
, file
, &proto_seq_ops
,
3338 sizeof(struct seq_net_private
));
3341 static const struct file_operations proto_seq_fops
= {
3342 .owner
= THIS_MODULE
,
3343 .open
= proto_seq_open
,
3345 .llseek
= seq_lseek
,
3346 .release
= seq_release_net
,
3349 static __net_init
int proto_init_net(struct net
*net
)
3351 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3357 static __net_exit
void proto_exit_net(struct net
*net
)
3359 remove_proc_entry("protocols", net
->proc_net
);
3363 static __net_initdata
struct pernet_operations proto_net_ops
= {
3364 .init
= proto_init_net
,
3365 .exit
= proto_exit_net
,
3368 static int __init
proto_init(void)
3370 return register_pernet_subsys(&proto_net_ops
);
3373 subsys_initcall(proto_init
);
3375 #endif /* PROC_FS */
3377 #ifdef CONFIG_NET_RX_BUSY_POLL
3378 bool sk_busy_loop_end(void *p
, unsigned long start_time
)
3380 struct sock
*sk
= p
;
3382 return !skb_queue_empty(&sk
->sk_receive_queue
) ||
3383 sk_busy_loop_timeout(sk
, start_time
);
3385 EXPORT_SYMBOL(sk_busy_loop_end
);
3386 #endif /* CONFIG_NET_RX_BUSY_POLL */