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>
146 #include <net/busy_poll.h>
148 static DEFINE_MUTEX(proto_list_mutex
);
149 static LIST_HEAD(proto_list
);
152 * sk_ns_capable - General socket capability test
153 * @sk: Socket to use a capability on or through
154 * @user_ns: The user namespace of the capability to use
155 * @cap: The capability to use
157 * Test to see if the opener of the socket had when the socket was
158 * created and the current process has the capability @cap in the user
159 * namespace @user_ns.
161 bool sk_ns_capable(const struct sock
*sk
,
162 struct user_namespace
*user_ns
, int cap
)
164 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
165 ns_capable(user_ns
, cap
);
167 EXPORT_SYMBOL(sk_ns_capable
);
170 * sk_capable - Socket global capability test
171 * @sk: Socket to use a capability on or through
172 * @cap: The global capability to use
174 * Test to see if the opener of the socket had when the socket was
175 * created and the current process has the capability @cap in all user
178 bool sk_capable(const struct sock
*sk
, int cap
)
180 return sk_ns_capable(sk
, &init_user_ns
, cap
);
182 EXPORT_SYMBOL(sk_capable
);
185 * sk_net_capable - Network namespace socket capability test
186 * @sk: Socket to use a capability on or through
187 * @cap: The capability to use
189 * Test to see if the opener of the socket had when the socket was created
190 * and the current process has the capability @cap over the network namespace
191 * the socket is a member of.
193 bool sk_net_capable(const struct sock
*sk
, int cap
)
195 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
197 EXPORT_SYMBOL(sk_net_capable
);
200 * Each address family might have different locking rules, so we have
201 * one slock key per address family and separate keys for internal and
204 static struct lock_class_key af_family_keys
[AF_MAX
];
205 static struct lock_class_key af_family_kern_keys
[AF_MAX
];
206 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
207 static struct lock_class_key af_family_kern_slock_keys
[AF_MAX
];
210 * Make lock validator output more readable. (we pre-construct these
211 * strings build-time, so that runtime initialization of socket
215 #define _sock_locks(x) \
216 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
217 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
218 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
219 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
220 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
221 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
222 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
223 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
224 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
225 x "27" , x "28" , x "AF_CAN" , \
226 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
227 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
228 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
229 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
230 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
232 static const char *const af_family_key_strings
[AF_MAX
+1] = {
233 _sock_locks("sk_lock-")
235 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
236 _sock_locks("slock-")
238 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
239 _sock_locks("clock-")
242 static const char *const af_family_kern_key_strings
[AF_MAX
+1] = {
243 _sock_locks("k-sk_lock-")
245 static const char *const af_family_kern_slock_key_strings
[AF_MAX
+1] = {
246 _sock_locks("k-slock-")
248 static const char *const af_family_kern_clock_key_strings
[AF_MAX
+1] = {
249 _sock_locks("k-clock-")
251 static const char *const af_family_rlock_key_strings
[AF_MAX
+1] = {
252 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
253 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
254 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
255 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
256 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
257 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
258 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
259 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
260 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
261 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
262 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
263 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
264 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
265 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
266 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
268 static const char *const af_family_wlock_key_strings
[AF_MAX
+1] = {
269 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
270 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
271 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
272 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
273 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
274 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
275 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
276 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
277 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
278 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
279 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
280 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
281 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
282 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
283 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
285 static const char *const af_family_elock_key_strings
[AF_MAX
+1] = {
286 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
287 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
288 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
289 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
290 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
291 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
292 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
293 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
294 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
295 "elock-27" , "elock-28" , "elock-AF_CAN" ,
296 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
297 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
298 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
299 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
300 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
304 * sk_callback_lock and sk queues locking rules are per-address-family,
305 * so split the lock classes by using a per-AF key:
307 static struct lock_class_key af_callback_keys
[AF_MAX
];
308 static struct lock_class_key af_rlock_keys
[AF_MAX
];
309 static struct lock_class_key af_wlock_keys
[AF_MAX
];
310 static struct lock_class_key af_elock_keys
[AF_MAX
];
311 static struct lock_class_key af_kern_callback_keys
[AF_MAX
];
313 /* Take into consideration the size of the struct sk_buff overhead in the
314 * determination of these values, since that is non-constant across
315 * platforms. This makes socket queueing behavior and performance
316 * not depend upon such differences.
318 #define _SK_MEM_PACKETS 256
319 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
320 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
321 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
323 /* Run time adjustable parameters. */
324 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
325 EXPORT_SYMBOL(sysctl_wmem_max
);
326 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
327 EXPORT_SYMBOL(sysctl_rmem_max
);
328 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
329 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
331 /* Maximal space eaten by iovec or ancillary data plus some space */
332 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
333 EXPORT_SYMBOL(sysctl_optmem_max
);
335 int sysctl_tstamp_allow_data __read_mostly
= 1;
337 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
338 EXPORT_SYMBOL_GPL(memalloc_socks
);
341 * sk_set_memalloc - sets %SOCK_MEMALLOC
342 * @sk: socket to set it on
344 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
345 * It's the responsibility of the admin to adjust min_free_kbytes
346 * to meet the requirements
348 void sk_set_memalloc(struct sock
*sk
)
350 sock_set_flag(sk
, SOCK_MEMALLOC
);
351 sk
->sk_allocation
|= __GFP_MEMALLOC
;
352 static_key_slow_inc(&memalloc_socks
);
354 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
356 void sk_clear_memalloc(struct sock
*sk
)
358 sock_reset_flag(sk
, SOCK_MEMALLOC
);
359 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
360 static_key_slow_dec(&memalloc_socks
);
363 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
364 * progress of swapping. SOCK_MEMALLOC may be cleared while
365 * it has rmem allocations due to the last swapfile being deactivated
366 * but there is a risk that the socket is unusable due to exceeding
367 * the rmem limits. Reclaim the reserves and obey rmem limits again.
371 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
373 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
376 unsigned int noreclaim_flag
;
378 /* these should have been dropped before queueing */
379 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
381 noreclaim_flag
= memalloc_noreclaim_save();
382 ret
= sk
->sk_backlog_rcv(sk
, skb
);
383 memalloc_noreclaim_restore(noreclaim_flag
);
387 EXPORT_SYMBOL(__sk_backlog_rcv
);
389 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
393 if (optlen
< sizeof(tv
))
395 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
397 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
401 static int warned __read_mostly
;
404 if (warned
< 10 && net_ratelimit()) {
406 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
407 __func__
, current
->comm
, task_pid_nr(current
));
411 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
412 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
414 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
415 *timeo_p
= tv
.tv_sec
* HZ
+ DIV_ROUND_UP(tv
.tv_usec
, USEC_PER_SEC
/ HZ
);
419 static void sock_warn_obsolete_bsdism(const char *name
)
422 static char warncomm
[TASK_COMM_LEN
];
423 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
424 strcpy(warncomm
, current
->comm
);
425 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
431 static bool sock_needs_netstamp(const struct sock
*sk
)
433 switch (sk
->sk_family
) {
442 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
444 if (sk
->sk_flags
& flags
) {
445 sk
->sk_flags
&= ~flags
;
446 if (sock_needs_netstamp(sk
) &&
447 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
448 net_disable_timestamp();
453 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
456 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
458 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
459 atomic_inc(&sk
->sk_drops
);
460 trace_sock_rcvqueue_full(sk
, skb
);
464 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
465 atomic_inc(&sk
->sk_drops
);
470 skb_set_owner_r(skb
, sk
);
472 /* we escape from rcu protected region, make sure we dont leak
477 spin_lock_irqsave(&list
->lock
, flags
);
478 sock_skb_set_dropcount(sk
, skb
);
479 __skb_queue_tail(list
, skb
);
480 spin_unlock_irqrestore(&list
->lock
, flags
);
482 if (!sock_flag(sk
, SOCK_DEAD
))
483 sk
->sk_data_ready(sk
);
486 EXPORT_SYMBOL(__sock_queue_rcv_skb
);
488 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
492 err
= sk_filter(sk
, skb
);
496 return __sock_queue_rcv_skb(sk
, skb
);
498 EXPORT_SYMBOL(sock_queue_rcv_skb
);
500 int __sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
,
501 const int nested
, unsigned int trim_cap
, bool refcounted
)
503 int rc
= NET_RX_SUCCESS
;
505 if (sk_filter_trim_cap(sk
, skb
, trim_cap
))
506 goto discard_and_relse
;
510 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
511 atomic_inc(&sk
->sk_drops
);
512 goto discard_and_relse
;
515 bh_lock_sock_nested(sk
);
518 if (!sock_owned_by_user(sk
)) {
520 * trylock + unlock semantics:
522 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
524 rc
= sk_backlog_rcv(sk
, skb
);
526 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
527 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
529 atomic_inc(&sk
->sk_drops
);
530 goto discard_and_relse
;
542 EXPORT_SYMBOL(__sk_receive_skb
);
544 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
546 struct dst_entry
*dst
= __sk_dst_get(sk
);
548 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
549 sk_tx_queue_clear(sk
);
550 sk
->sk_dst_pending_confirm
= 0;
551 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
558 EXPORT_SYMBOL(__sk_dst_check
);
560 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
562 struct dst_entry
*dst
= sk_dst_get(sk
);
564 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
572 EXPORT_SYMBOL(sk_dst_check
);
574 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
577 int ret
= -ENOPROTOOPT
;
578 #ifdef CONFIG_NETDEVICES
579 struct net
*net
= sock_net(sk
);
580 char devname
[IFNAMSIZ
];
585 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
592 /* Bind this socket to a particular device like "eth0",
593 * as specified in the passed interface name. If the
594 * name is "" or the option length is zero the socket
597 if (optlen
> IFNAMSIZ
- 1)
598 optlen
= IFNAMSIZ
- 1;
599 memset(devname
, 0, sizeof(devname
));
602 if (copy_from_user(devname
, optval
, optlen
))
606 if (devname
[0] != '\0') {
607 struct net_device
*dev
;
610 dev
= dev_get_by_name_rcu(net
, devname
);
612 index
= dev
->ifindex
;
620 sk
->sk_bound_dev_if
= index
;
632 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
633 int __user
*optlen
, int len
)
635 int ret
= -ENOPROTOOPT
;
636 #ifdef CONFIG_NETDEVICES
637 struct net
*net
= sock_net(sk
);
638 char devname
[IFNAMSIZ
];
640 if (sk
->sk_bound_dev_if
== 0) {
649 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
653 len
= strlen(devname
) + 1;
656 if (copy_to_user(optval
, devname
, len
))
661 if (put_user(len
, optlen
))
672 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
675 sock_set_flag(sk
, bit
);
677 sock_reset_flag(sk
, bit
);
680 bool sk_mc_loop(struct sock
*sk
)
682 if (dev_recursion_level())
686 switch (sk
->sk_family
) {
688 return inet_sk(sk
)->mc_loop
;
689 #if IS_ENABLED(CONFIG_IPV6)
691 return inet6_sk(sk
)->mc_loop
;
697 EXPORT_SYMBOL(sk_mc_loop
);
700 * This is meant for all protocols to use and covers goings on
701 * at the socket level. Everything here is generic.
704 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
705 char __user
*optval
, unsigned int optlen
)
707 struct sock
*sk
= sock
->sk
;
714 * Options without arguments
717 if (optname
== SO_BINDTODEVICE
)
718 return sock_setbindtodevice(sk
, optval
, optlen
);
720 if (optlen
< sizeof(int))
723 if (get_user(val
, (int __user
*)optval
))
726 valbool
= val
? 1 : 0;
732 if (val
&& !capable(CAP_NET_ADMIN
))
735 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
738 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
741 sk
->sk_reuseport
= valbool
;
750 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
753 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
756 /* Don't error on this BSD doesn't and if you think
757 * about it this is right. Otherwise apps have to
758 * play 'guess the biggest size' games. RCVBUF/SNDBUF
759 * are treated in BSD as hints
761 val
= min_t(u32
, val
, sysctl_wmem_max
);
763 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
764 sk
->sk_sndbuf
= max_t(int, val
* 2, SOCK_MIN_SNDBUF
);
765 /* Wake up sending tasks if we upped the value. */
766 sk
->sk_write_space(sk
);
770 if (!capable(CAP_NET_ADMIN
)) {
777 /* Don't error on this BSD doesn't and if you think
778 * about it this is right. Otherwise apps have to
779 * play 'guess the biggest size' games. RCVBUF/SNDBUF
780 * are treated in BSD as hints
782 val
= min_t(u32
, val
, sysctl_rmem_max
);
784 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
786 * We double it on the way in to account for
787 * "struct sk_buff" etc. overhead. Applications
788 * assume that the SO_RCVBUF setting they make will
789 * allow that much actual data to be received on that
792 * Applications are unaware that "struct sk_buff" and
793 * other overheads allocate from the receive buffer
794 * during socket buffer allocation.
796 * And after considering the possible alternatives,
797 * returning the value we actually used in getsockopt
798 * is the most desirable behavior.
800 sk
->sk_rcvbuf
= max_t(int, val
* 2, SOCK_MIN_RCVBUF
);
804 if (!capable(CAP_NET_ADMIN
)) {
811 if (sk
->sk_prot
->keepalive
)
812 sk
->sk_prot
->keepalive(sk
, valbool
);
813 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
817 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
821 sk
->sk_no_check_tx
= valbool
;
825 if ((val
>= 0 && val
<= 6) ||
826 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
827 sk
->sk_priority
= val
;
833 if (optlen
< sizeof(ling
)) {
834 ret
= -EINVAL
; /* 1003.1g */
837 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
842 sock_reset_flag(sk
, SOCK_LINGER
);
844 #if (BITS_PER_LONG == 32)
845 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
846 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
849 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
850 sock_set_flag(sk
, SOCK_LINGER
);
855 sock_warn_obsolete_bsdism("setsockopt");
860 set_bit(SOCK_PASSCRED
, &sock
->flags
);
862 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
868 if (optname
== SO_TIMESTAMP
)
869 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
871 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
872 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
873 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
875 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
876 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
880 case SO_TIMESTAMPING
:
881 if (val
& ~SOF_TIMESTAMPING_MASK
) {
886 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
887 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
888 if (sk
->sk_protocol
== IPPROTO_TCP
&&
889 sk
->sk_type
== SOCK_STREAM
) {
890 if ((1 << sk
->sk_state
) &
891 (TCPF_CLOSE
| TCPF_LISTEN
)) {
895 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
901 if (val
& SOF_TIMESTAMPING_OPT_STATS
&&
902 !(val
& SOF_TIMESTAMPING_OPT_TSONLY
)) {
907 sk
->sk_tsflags
= val
;
908 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
909 sock_enable_timestamp(sk
,
910 SOCK_TIMESTAMPING_RX_SOFTWARE
);
912 sock_disable_timestamp(sk
,
913 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
919 sk
->sk_rcvlowat
= val
? : 1;
923 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
927 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
930 case SO_ATTACH_FILTER
:
932 if (optlen
== sizeof(struct sock_fprog
)) {
933 struct sock_fprog fprog
;
936 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
939 ret
= sk_attach_filter(&fprog
, sk
);
945 if (optlen
== sizeof(u32
)) {
949 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
952 ret
= sk_attach_bpf(ufd
, sk
);
956 case SO_ATTACH_REUSEPORT_CBPF
:
958 if (optlen
== sizeof(struct sock_fprog
)) {
959 struct sock_fprog fprog
;
962 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
965 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
969 case SO_ATTACH_REUSEPORT_EBPF
:
971 if (optlen
== sizeof(u32
)) {
975 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
978 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
982 case SO_DETACH_FILTER
:
983 ret
= sk_detach_filter(sk
);
987 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
990 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
995 set_bit(SOCK_PASSSEC
, &sock
->flags
);
997 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
1000 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
1007 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
1010 case SO_WIFI_STATUS
:
1011 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
1015 if (sock
->ops
->set_peek_off
)
1016 ret
= sock
->ops
->set_peek_off(sk
, val
);
1022 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
1025 case SO_SELECT_ERR_QUEUE
:
1026 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
1029 #ifdef CONFIG_NET_RX_BUSY_POLL
1031 /* allow unprivileged users to decrease the value */
1032 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
1038 sk
->sk_ll_usec
= val
;
1043 case SO_MAX_PACING_RATE
:
1044 sk
->sk_max_pacing_rate
= val
;
1045 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
1046 sk
->sk_max_pacing_rate
);
1049 case SO_INCOMING_CPU
:
1050 sk
->sk_incoming_cpu
= val
;
1055 dst_negative_advice(sk
);
1064 EXPORT_SYMBOL(sock_setsockopt
);
1067 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1068 struct ucred
*ucred
)
1070 ucred
->pid
= pid_vnr(pid
);
1071 ucred
->uid
= ucred
->gid
= -1;
1073 struct user_namespace
*current_ns
= current_user_ns();
1075 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1076 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1080 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1081 char __user
*optval
, int __user
*optlen
)
1083 struct sock
*sk
= sock
->sk
;
1092 int lv
= sizeof(int);
1095 if (get_user(len
, optlen
))
1100 memset(&v
, 0, sizeof(v
));
1104 v
.val
= sock_flag(sk
, SOCK_DBG
);
1108 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1112 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1116 v
.val
= sk
->sk_sndbuf
;
1120 v
.val
= sk
->sk_rcvbuf
;
1124 v
.val
= sk
->sk_reuse
;
1128 v
.val
= sk
->sk_reuseport
;
1132 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1136 v
.val
= sk
->sk_type
;
1140 v
.val
= sk
->sk_protocol
;
1144 v
.val
= sk
->sk_family
;
1148 v
.val
= -sock_error(sk
);
1150 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1154 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1158 v
.val
= sk
->sk_no_check_tx
;
1162 v
.val
= sk
->sk_priority
;
1166 lv
= sizeof(v
.ling
);
1167 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1168 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1172 sock_warn_obsolete_bsdism("getsockopt");
1176 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1177 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1180 case SO_TIMESTAMPNS
:
1181 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1184 case SO_TIMESTAMPING
:
1185 v
.val
= sk
->sk_tsflags
;
1189 lv
= sizeof(struct timeval
);
1190 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1194 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1195 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * USEC_PER_SEC
) / HZ
;
1200 lv
= sizeof(struct timeval
);
1201 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1205 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1206 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * USEC_PER_SEC
) / HZ
;
1211 v
.val
= sk
->sk_rcvlowat
;
1219 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1224 struct ucred peercred
;
1225 if (len
> sizeof(peercred
))
1226 len
= sizeof(peercred
);
1227 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1228 if (copy_to_user(optval
, &peercred
, len
))
1237 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1241 if (copy_to_user(optval
, address
, len
))
1246 /* Dubious BSD thing... Probably nobody even uses it, but
1247 * the UNIX standard wants it for whatever reason... -DaveM
1250 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1254 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1258 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1261 v
.val
= sk
->sk_mark
;
1265 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1268 case SO_WIFI_STATUS
:
1269 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1273 if (!sock
->ops
->set_peek_off
)
1276 v
.val
= sk
->sk_peek_off
;
1279 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1282 case SO_BINDTODEVICE
:
1283 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1286 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1292 case SO_LOCK_FILTER
:
1293 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1296 case SO_BPF_EXTENSIONS
:
1297 v
.val
= bpf_tell_extensions();
1300 case SO_SELECT_ERR_QUEUE
:
1301 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1304 #ifdef CONFIG_NET_RX_BUSY_POLL
1306 v
.val
= sk
->sk_ll_usec
;
1310 case SO_MAX_PACING_RATE
:
1311 v
.val
= sk
->sk_max_pacing_rate
;
1314 case SO_INCOMING_CPU
:
1315 v
.val
= sk
->sk_incoming_cpu
;
1320 u32 meminfo
[SK_MEMINFO_VARS
];
1322 if (get_user(len
, optlen
))
1325 sk_get_meminfo(sk
, meminfo
);
1327 len
= min_t(unsigned int, len
, sizeof(meminfo
));
1328 if (copy_to_user(optval
, &meminfo
, len
))
1334 #ifdef CONFIG_NET_RX_BUSY_POLL
1335 case SO_INCOMING_NAPI_ID
:
1336 v
.val
= READ_ONCE(sk
->sk_napi_id
);
1338 /* aggregate non-NAPI IDs down to 0 */
1339 if (v
.val
< MIN_NAPI_ID
)
1349 v
.val64
= sock_gen_cookie(sk
);
1353 /* We implement the SO_SNDLOWAT etc to not be settable
1356 return -ENOPROTOOPT
;
1361 if (copy_to_user(optval
, &v
, len
))
1364 if (put_user(len
, optlen
))
1370 * Initialize an sk_lock.
1372 * (We also register the sk_lock with the lock validator.)
1374 static inline void sock_lock_init(struct sock
*sk
)
1376 if (sk
->sk_kern_sock
)
1377 sock_lock_init_class_and_name(
1379 af_family_kern_slock_key_strings
[sk
->sk_family
],
1380 af_family_kern_slock_keys
+ sk
->sk_family
,
1381 af_family_kern_key_strings
[sk
->sk_family
],
1382 af_family_kern_keys
+ sk
->sk_family
);
1384 sock_lock_init_class_and_name(
1386 af_family_slock_key_strings
[sk
->sk_family
],
1387 af_family_slock_keys
+ sk
->sk_family
,
1388 af_family_key_strings
[sk
->sk_family
],
1389 af_family_keys
+ sk
->sk_family
);
1393 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1394 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1395 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1397 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1399 #ifdef CONFIG_SECURITY_NETWORK
1400 void *sptr
= nsk
->sk_security
;
1402 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1404 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1405 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1407 #ifdef CONFIG_SECURITY_NETWORK
1408 nsk
->sk_security
= sptr
;
1409 security_sk_clone(osk
, nsk
);
1413 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1417 struct kmem_cache
*slab
;
1421 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1424 if (priority
& __GFP_ZERO
)
1425 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1427 sk
= kmalloc(prot
->obj_size
, priority
);
1430 kmemcheck_annotate_bitfield(sk
, flags
);
1432 if (security_sk_alloc(sk
, family
, priority
))
1435 if (!try_module_get(prot
->owner
))
1437 sk_tx_queue_clear(sk
);
1443 security_sk_free(sk
);
1446 kmem_cache_free(slab
, sk
);
1452 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1454 struct kmem_cache
*slab
;
1455 struct module
*owner
;
1457 owner
= prot
->owner
;
1460 cgroup_sk_free(&sk
->sk_cgrp_data
);
1461 mem_cgroup_sk_free(sk
);
1462 security_sk_free(sk
);
1464 kmem_cache_free(slab
, sk
);
1471 * sk_alloc - All socket objects are allocated here
1472 * @net: the applicable net namespace
1473 * @family: protocol family
1474 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1475 * @prot: struct proto associated with this new sock instance
1476 * @kern: is this to be a kernel socket?
1478 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1479 struct proto
*prot
, int kern
)
1483 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1485 sk
->sk_family
= family
;
1487 * See comment in struct sock definition to understand
1488 * why we need sk_prot_creator -acme
1490 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1491 sk
->sk_kern_sock
= kern
;
1493 sk
->sk_net_refcnt
= kern
? 0 : 1;
1494 if (likely(sk
->sk_net_refcnt
))
1496 sock_net_set(sk
, net
);
1497 atomic_set(&sk
->sk_wmem_alloc
, 1);
1499 mem_cgroup_sk_alloc(sk
);
1500 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1501 sock_update_classid(&sk
->sk_cgrp_data
);
1502 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1507 EXPORT_SYMBOL(sk_alloc
);
1509 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1510 * grace period. This is the case for UDP sockets and TCP listeners.
1512 static void __sk_destruct(struct rcu_head
*head
)
1514 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1515 struct sk_filter
*filter
;
1517 if (sk
->sk_destruct
)
1518 sk
->sk_destruct(sk
);
1520 filter
= rcu_dereference_check(sk
->sk_filter
,
1521 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1523 sk_filter_uncharge(sk
, filter
);
1524 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1526 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1527 reuseport_detach_sock(sk
);
1529 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1531 if (atomic_read(&sk
->sk_omem_alloc
))
1532 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1533 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1535 if (sk
->sk_frag
.page
) {
1536 put_page(sk
->sk_frag
.page
);
1537 sk
->sk_frag
.page
= NULL
;
1540 if (sk
->sk_peer_cred
)
1541 put_cred(sk
->sk_peer_cred
);
1542 put_pid(sk
->sk_peer_pid
);
1543 if (likely(sk
->sk_net_refcnt
))
1544 put_net(sock_net(sk
));
1545 sk_prot_free(sk
->sk_prot_creator
, sk
);
1548 void sk_destruct(struct sock
*sk
)
1550 if (sock_flag(sk
, SOCK_RCU_FREE
))
1551 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1553 __sk_destruct(&sk
->sk_rcu
);
1556 static void __sk_free(struct sock
*sk
)
1558 if (unlikely(sock_diag_has_destroy_listeners(sk
) && sk
->sk_net_refcnt
))
1559 sock_diag_broadcast_destroy(sk
);
1564 void sk_free(struct sock
*sk
)
1567 * We subtract one from sk_wmem_alloc and can know if
1568 * some packets are still in some tx queue.
1569 * If not null, sock_wfree() will call __sk_free(sk) later
1571 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1574 EXPORT_SYMBOL(sk_free
);
1576 static void sk_init_common(struct sock
*sk
)
1578 skb_queue_head_init(&sk
->sk_receive_queue
);
1579 skb_queue_head_init(&sk
->sk_write_queue
);
1580 skb_queue_head_init(&sk
->sk_error_queue
);
1582 rwlock_init(&sk
->sk_callback_lock
);
1583 lockdep_set_class_and_name(&sk
->sk_receive_queue
.lock
,
1584 af_rlock_keys
+ sk
->sk_family
,
1585 af_family_rlock_key_strings
[sk
->sk_family
]);
1586 lockdep_set_class_and_name(&sk
->sk_write_queue
.lock
,
1587 af_wlock_keys
+ sk
->sk_family
,
1588 af_family_wlock_key_strings
[sk
->sk_family
]);
1589 lockdep_set_class_and_name(&sk
->sk_error_queue
.lock
,
1590 af_elock_keys
+ sk
->sk_family
,
1591 af_family_elock_key_strings
[sk
->sk_family
]);
1592 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
1593 af_callback_keys
+ sk
->sk_family
,
1594 af_family_clock_key_strings
[sk
->sk_family
]);
1598 * sk_clone_lock - clone a socket, and lock its clone
1599 * @sk: the socket to clone
1600 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1602 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1604 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1607 bool is_charged
= true;
1609 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1610 if (newsk
!= NULL
) {
1611 struct sk_filter
*filter
;
1613 sock_copy(newsk
, sk
);
1616 if (likely(newsk
->sk_net_refcnt
))
1617 get_net(sock_net(newsk
));
1618 sk_node_init(&newsk
->sk_node
);
1619 sock_lock_init(newsk
);
1620 bh_lock_sock(newsk
);
1621 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1622 newsk
->sk_backlog
.len
= 0;
1624 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1626 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1628 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1629 atomic_set(&newsk
->sk_omem_alloc
, 0);
1630 sk_init_common(newsk
);
1632 newsk
->sk_dst_cache
= NULL
;
1633 newsk
->sk_dst_pending_confirm
= 0;
1634 newsk
->sk_wmem_queued
= 0;
1635 newsk
->sk_forward_alloc
= 0;
1636 atomic_set(&newsk
->sk_drops
, 0);
1637 newsk
->sk_send_head
= NULL
;
1638 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1640 sock_reset_flag(newsk
, SOCK_DONE
);
1642 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1644 /* though it's an empty new sock, the charging may fail
1645 * if sysctl_optmem_max was changed between creation of
1646 * original socket and cloning
1648 is_charged
= sk_filter_charge(newsk
, filter
);
1650 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1651 /* We need to make sure that we don't uncharge the new
1652 * socket if we couldn't charge it in the first place
1653 * as otherwise we uncharge the parent's filter.
1656 RCU_INIT_POINTER(newsk
->sk_filter
, NULL
);
1657 sk_free_unlock_clone(newsk
);
1661 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1664 newsk
->sk_err_soft
= 0;
1665 newsk
->sk_priority
= 0;
1666 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1667 atomic64_set(&newsk
->sk_cookie
, 0);
1669 mem_cgroup_sk_alloc(newsk
);
1670 cgroup_sk_alloc(&newsk
->sk_cgrp_data
);
1673 * Before updating sk_refcnt, we must commit prior changes to memory
1674 * (Documentation/RCU/rculist_nulls.txt for details)
1677 atomic_set(&newsk
->sk_refcnt
, 2);
1680 * Increment the counter in the same struct proto as the master
1681 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1682 * is the same as sk->sk_prot->socks, as this field was copied
1685 * This _changes_ the previous behaviour, where
1686 * tcp_create_openreq_child always was incrementing the
1687 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1688 * to be taken into account in all callers. -acme
1690 sk_refcnt_debug_inc(newsk
);
1691 sk_set_socket(newsk
, NULL
);
1692 newsk
->sk_wq
= NULL
;
1694 if (newsk
->sk_prot
->sockets_allocated
)
1695 sk_sockets_allocated_inc(newsk
);
1697 if (sock_needs_netstamp(sk
) &&
1698 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1699 net_enable_timestamp();
1704 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1706 void sk_free_unlock_clone(struct sock
*sk
)
1708 /* It is still raw copy of parent, so invalidate
1709 * destructor and make plain sk_free() */
1710 sk
->sk_destruct
= NULL
;
1714 EXPORT_SYMBOL_GPL(sk_free_unlock_clone
);
1716 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1720 sk_dst_set(sk
, dst
);
1721 sk
->sk_route_caps
= dst
->dev
->features
;
1722 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1723 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1724 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1725 if (sk_can_gso(sk
)) {
1726 if (dst
->header_len
) {
1727 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1729 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1730 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1731 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1734 sk
->sk_gso_max_segs
= max_segs
;
1736 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1739 * Simple resource managers for sockets.
1744 * Write buffer destructor automatically called from kfree_skb.
1746 void sock_wfree(struct sk_buff
*skb
)
1748 struct sock
*sk
= skb
->sk
;
1749 unsigned int len
= skb
->truesize
;
1751 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1753 * Keep a reference on sk_wmem_alloc, this will be released
1754 * after sk_write_space() call
1756 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1757 sk
->sk_write_space(sk
);
1761 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1762 * could not do because of in-flight packets
1764 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1767 EXPORT_SYMBOL(sock_wfree
);
1769 /* This variant of sock_wfree() is used by TCP,
1770 * since it sets SOCK_USE_WRITE_QUEUE.
1772 void __sock_wfree(struct sk_buff
*skb
)
1774 struct sock
*sk
= skb
->sk
;
1776 if (atomic_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
))
1780 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1785 if (unlikely(!sk_fullsock(sk
))) {
1786 skb
->destructor
= sock_edemux
;
1791 skb
->destructor
= sock_wfree
;
1792 skb_set_hash_from_sk(skb
, sk
);
1794 * We used to take a refcount on sk, but following operation
1795 * is enough to guarantee sk_free() wont free this sock until
1796 * all in-flight packets are completed
1798 atomic_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1800 EXPORT_SYMBOL(skb_set_owner_w
);
1802 /* This helper is used by netem, as it can hold packets in its
1803 * delay queue. We want to allow the owner socket to send more
1804 * packets, as if they were already TX completed by a typical driver.
1805 * But we also want to keep skb->sk set because some packet schedulers
1806 * rely on it (sch_fq for example). So we set skb->truesize to a small
1807 * amount (1) and decrease sk_wmem_alloc accordingly.
1809 void skb_orphan_partial(struct sk_buff
*skb
)
1811 /* If this skb is a TCP pure ACK or already went here,
1812 * we have nothing to do. 2 is already a very small truesize.
1814 if (skb
->truesize
<= 2)
1817 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1818 * so we do not completely orphan skb, but transfert all
1819 * accounted bytes but one, to avoid unexpected reorders.
1821 if (skb
->destructor
== sock_wfree
1823 || skb
->destructor
== tcp_wfree
1826 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1832 EXPORT_SYMBOL(skb_orphan_partial
);
1835 * Read buffer destructor automatically called from kfree_skb.
1837 void sock_rfree(struct sk_buff
*skb
)
1839 struct sock
*sk
= skb
->sk
;
1840 unsigned int len
= skb
->truesize
;
1842 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1843 sk_mem_uncharge(sk
, len
);
1845 EXPORT_SYMBOL(sock_rfree
);
1848 * Buffer destructor for skbs that are not used directly in read or write
1849 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1851 void sock_efree(struct sk_buff
*skb
)
1855 EXPORT_SYMBOL(sock_efree
);
1857 kuid_t
sock_i_uid(struct sock
*sk
)
1861 read_lock_bh(&sk
->sk_callback_lock
);
1862 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1863 read_unlock_bh(&sk
->sk_callback_lock
);
1866 EXPORT_SYMBOL(sock_i_uid
);
1868 unsigned long sock_i_ino(struct sock
*sk
)
1872 read_lock_bh(&sk
->sk_callback_lock
);
1873 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1874 read_unlock_bh(&sk
->sk_callback_lock
);
1877 EXPORT_SYMBOL(sock_i_ino
);
1880 * Allocate a skb from the socket's send buffer.
1882 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1885 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1886 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1888 skb_set_owner_w(skb
, sk
);
1894 EXPORT_SYMBOL(sock_wmalloc
);
1897 * Allocate a memory block from the socket's option memory buffer.
1899 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1901 if ((unsigned int)size
<= sysctl_optmem_max
&&
1902 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1904 /* First do the add, to avoid the race if kmalloc
1907 atomic_add(size
, &sk
->sk_omem_alloc
);
1908 mem
= kmalloc(size
, priority
);
1911 atomic_sub(size
, &sk
->sk_omem_alloc
);
1915 EXPORT_SYMBOL(sock_kmalloc
);
1917 /* Free an option memory block. Note, we actually want the inline
1918 * here as this allows gcc to detect the nullify and fold away the
1919 * condition entirely.
1921 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1924 if (WARN_ON_ONCE(!mem
))
1930 atomic_sub(size
, &sk
->sk_omem_alloc
);
1933 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1935 __sock_kfree_s(sk
, mem
, size
, false);
1937 EXPORT_SYMBOL(sock_kfree_s
);
1939 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1941 __sock_kfree_s(sk
, mem
, size
, true);
1943 EXPORT_SYMBOL(sock_kzfree_s
);
1945 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1946 I think, these locks should be removed for datagram sockets.
1948 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1952 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1956 if (signal_pending(current
))
1958 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1959 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1960 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1962 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1966 timeo
= schedule_timeout(timeo
);
1968 finish_wait(sk_sleep(sk
), &wait
);
1974 * Generic send/receive buffer handlers
1977 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1978 unsigned long data_len
, int noblock
,
1979 int *errcode
, int max_page_order
)
1981 struct sk_buff
*skb
;
1985 timeo
= sock_sndtimeo(sk
, noblock
);
1987 err
= sock_error(sk
);
1992 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1995 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1998 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1999 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2003 if (signal_pending(current
))
2005 timeo
= sock_wait_for_wmem(sk
, timeo
);
2007 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
2008 errcode
, sk
->sk_allocation
);
2010 skb_set_owner_w(skb
, sk
);
2014 err
= sock_intr_errno(timeo
);
2019 EXPORT_SYMBOL(sock_alloc_send_pskb
);
2021 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
2022 int noblock
, int *errcode
)
2024 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
2026 EXPORT_SYMBOL(sock_alloc_send_skb
);
2028 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
2029 struct sockcm_cookie
*sockc
)
2033 switch (cmsg
->cmsg_type
) {
2035 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
2037 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2039 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
2041 case SO_TIMESTAMPING
:
2042 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2045 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
2046 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
2049 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
2050 sockc
->tsflags
|= tsflags
;
2052 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2054 case SCM_CREDENTIALS
:
2061 EXPORT_SYMBOL(__sock_cmsg_send
);
2063 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
2064 struct sockcm_cookie
*sockc
)
2066 struct cmsghdr
*cmsg
;
2069 for_each_cmsghdr(cmsg
, msg
) {
2070 if (!CMSG_OK(msg
, cmsg
))
2072 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
2074 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
2080 EXPORT_SYMBOL(sock_cmsg_send
);
2082 /* On 32bit arches, an skb frag is limited to 2^15 */
2083 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2086 * skb_page_frag_refill - check that a page_frag contains enough room
2087 * @sz: minimum size of the fragment we want to get
2088 * @pfrag: pointer to page_frag
2089 * @gfp: priority for memory allocation
2091 * Note: While this allocator tries to use high order pages, there is
2092 * no guarantee that allocations succeed. Therefore, @sz MUST be
2093 * less or equal than PAGE_SIZE.
2095 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
2098 if (page_ref_count(pfrag
->page
) == 1) {
2102 if (pfrag
->offset
+ sz
<= pfrag
->size
)
2104 put_page(pfrag
->page
);
2108 if (SKB_FRAG_PAGE_ORDER
) {
2109 /* Avoid direct reclaim but allow kswapd to wake */
2110 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
2111 __GFP_COMP
| __GFP_NOWARN
|
2113 SKB_FRAG_PAGE_ORDER
);
2114 if (likely(pfrag
->page
)) {
2115 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
2119 pfrag
->page
= alloc_page(gfp
);
2120 if (likely(pfrag
->page
)) {
2121 pfrag
->size
= PAGE_SIZE
;
2126 EXPORT_SYMBOL(skb_page_frag_refill
);
2128 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
2130 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
2133 sk_enter_memory_pressure(sk
);
2134 sk_stream_moderate_sndbuf(sk
);
2137 EXPORT_SYMBOL(sk_page_frag_refill
);
2139 static void __lock_sock(struct sock
*sk
)
2140 __releases(&sk
->sk_lock
.slock
)
2141 __acquires(&sk
->sk_lock
.slock
)
2146 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2147 TASK_UNINTERRUPTIBLE
);
2148 spin_unlock_bh(&sk
->sk_lock
.slock
);
2150 spin_lock_bh(&sk
->sk_lock
.slock
);
2151 if (!sock_owned_by_user(sk
))
2154 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2157 static void __release_sock(struct sock
*sk
)
2158 __releases(&sk
->sk_lock
.slock
)
2159 __acquires(&sk
->sk_lock
.slock
)
2161 struct sk_buff
*skb
, *next
;
2163 while ((skb
= sk
->sk_backlog
.head
) != NULL
) {
2164 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2166 spin_unlock_bh(&sk
->sk_lock
.slock
);
2171 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2173 sk_backlog_rcv(sk
, skb
);
2178 } while (skb
!= NULL
);
2180 spin_lock_bh(&sk
->sk_lock
.slock
);
2184 * Doing the zeroing here guarantee we can not loop forever
2185 * while a wild producer attempts to flood us.
2187 sk
->sk_backlog
.len
= 0;
2190 void __sk_flush_backlog(struct sock
*sk
)
2192 spin_lock_bh(&sk
->sk_lock
.slock
);
2194 spin_unlock_bh(&sk
->sk_lock
.slock
);
2198 * sk_wait_data - wait for data to arrive at sk_receive_queue
2199 * @sk: sock to wait on
2200 * @timeo: for how long
2201 * @skb: last skb seen on sk_receive_queue
2203 * Now socket state including sk->sk_err is changed only under lock,
2204 * hence we may omit checks after joining wait queue.
2205 * We check receive queue before schedule() only as optimization;
2206 * it is very likely that release_sock() added new data.
2208 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2210 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
2213 add_wait_queue(sk_sleep(sk
), &wait
);
2214 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2215 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
, &wait
);
2216 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2217 remove_wait_queue(sk_sleep(sk
), &wait
);
2220 EXPORT_SYMBOL(sk_wait_data
);
2223 * __sk_mem_raise_allocated - increase memory_allocated
2225 * @size: memory size to allocate
2226 * @amt: pages to allocate
2227 * @kind: allocation type
2229 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2231 int __sk_mem_raise_allocated(struct sock
*sk
, int size
, int amt
, int kind
)
2233 struct proto
*prot
= sk
->sk_prot
;
2234 long allocated
= sk_memory_allocated_add(sk
, amt
);
2236 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2237 !mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
))
2238 goto suppress_allocation
;
2241 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2242 sk_leave_memory_pressure(sk
);
2246 /* Under pressure. */
2247 if (allocated
> sk_prot_mem_limits(sk
, 1))
2248 sk_enter_memory_pressure(sk
);
2250 /* Over hard limit. */
2251 if (allocated
> sk_prot_mem_limits(sk
, 2))
2252 goto suppress_allocation
;
2254 /* guarantee minimum buffer size under pressure */
2255 if (kind
== SK_MEM_RECV
) {
2256 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2259 } else { /* SK_MEM_SEND */
2260 if (sk
->sk_type
== SOCK_STREAM
) {
2261 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2263 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2264 prot
->sysctl_wmem
[0])
2268 if (sk_has_memory_pressure(sk
)) {
2271 if (!sk_under_memory_pressure(sk
))
2273 alloc
= sk_sockets_allocated_read_positive(sk
);
2274 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2275 sk_mem_pages(sk
->sk_wmem_queued
+
2276 atomic_read(&sk
->sk_rmem_alloc
) +
2277 sk
->sk_forward_alloc
))
2281 suppress_allocation
:
2283 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2284 sk_stream_moderate_sndbuf(sk
);
2286 /* Fail only if socket is _under_ its sndbuf.
2287 * In this case we cannot block, so that we have to fail.
2289 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2293 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2295 sk_memory_allocated_sub(sk
, amt
);
2297 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2298 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2302 EXPORT_SYMBOL(__sk_mem_raise_allocated
);
2305 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2307 * @size: memory size to allocate
2308 * @kind: allocation type
2310 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2311 * rmem allocation. This function assumes that protocols which have
2312 * memory_pressure use sk_wmem_queued as write buffer accounting.
2314 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2316 int ret
, amt
= sk_mem_pages(size
);
2318 sk
->sk_forward_alloc
+= amt
<< SK_MEM_QUANTUM_SHIFT
;
2319 ret
= __sk_mem_raise_allocated(sk
, size
, amt
, kind
);
2321 sk
->sk_forward_alloc
-= amt
<< SK_MEM_QUANTUM_SHIFT
;
2324 EXPORT_SYMBOL(__sk_mem_schedule
);
2327 * __sk_mem_reduce_allocated - reclaim memory_allocated
2329 * @amount: number of quanta
2331 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2333 void __sk_mem_reduce_allocated(struct sock
*sk
, int amount
)
2335 sk_memory_allocated_sub(sk
, amount
);
2337 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2338 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2340 if (sk_under_memory_pressure(sk
) &&
2341 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2342 sk_leave_memory_pressure(sk
);
2344 EXPORT_SYMBOL(__sk_mem_reduce_allocated
);
2347 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2349 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2351 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2353 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2354 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2355 __sk_mem_reduce_allocated(sk
, amount
);
2357 EXPORT_SYMBOL(__sk_mem_reclaim
);
2359 int sk_set_peek_off(struct sock
*sk
, int val
)
2364 sk
->sk_peek_off
= val
;
2367 EXPORT_SYMBOL_GPL(sk_set_peek_off
);
2370 * Set of default routines for initialising struct proto_ops when
2371 * the protocol does not support a particular function. In certain
2372 * cases where it makes no sense for a protocol to have a "do nothing"
2373 * function, some default processing is provided.
2376 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2380 EXPORT_SYMBOL(sock_no_bind
);
2382 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2387 EXPORT_SYMBOL(sock_no_connect
);
2389 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2393 EXPORT_SYMBOL(sock_no_socketpair
);
2395 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
,
2400 EXPORT_SYMBOL(sock_no_accept
);
2402 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2407 EXPORT_SYMBOL(sock_no_getname
);
2409 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2413 EXPORT_SYMBOL(sock_no_poll
);
2415 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2419 EXPORT_SYMBOL(sock_no_ioctl
);
2421 int sock_no_listen(struct socket
*sock
, int backlog
)
2425 EXPORT_SYMBOL(sock_no_listen
);
2427 int sock_no_shutdown(struct socket
*sock
, int how
)
2431 EXPORT_SYMBOL(sock_no_shutdown
);
2433 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2434 char __user
*optval
, unsigned int optlen
)
2438 EXPORT_SYMBOL(sock_no_setsockopt
);
2440 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2441 char __user
*optval
, int __user
*optlen
)
2445 EXPORT_SYMBOL(sock_no_getsockopt
);
2447 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2451 EXPORT_SYMBOL(sock_no_sendmsg
);
2453 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2458 EXPORT_SYMBOL(sock_no_recvmsg
);
2460 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2462 /* Mirror missing mmap method error code */
2465 EXPORT_SYMBOL(sock_no_mmap
);
2467 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2470 struct msghdr msg
= {.msg_flags
= flags
};
2472 char *kaddr
= kmap(page
);
2473 iov
.iov_base
= kaddr
+ offset
;
2475 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2479 EXPORT_SYMBOL(sock_no_sendpage
);
2482 * Default Socket Callbacks
2485 static void sock_def_wakeup(struct sock
*sk
)
2487 struct socket_wq
*wq
;
2490 wq
= rcu_dereference(sk
->sk_wq
);
2491 if (skwq_has_sleeper(wq
))
2492 wake_up_interruptible_all(&wq
->wait
);
2496 static void sock_def_error_report(struct sock
*sk
)
2498 struct socket_wq
*wq
;
2501 wq
= rcu_dereference(sk
->sk_wq
);
2502 if (skwq_has_sleeper(wq
))
2503 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2504 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2508 static void sock_def_readable(struct sock
*sk
)
2510 struct socket_wq
*wq
;
2513 wq
= rcu_dereference(sk
->sk_wq
);
2514 if (skwq_has_sleeper(wq
))
2515 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2516 POLLRDNORM
| POLLRDBAND
);
2517 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2521 static void sock_def_write_space(struct sock
*sk
)
2523 struct socket_wq
*wq
;
2527 /* Do not wake up a writer until he can make "significant"
2530 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2531 wq
= rcu_dereference(sk
->sk_wq
);
2532 if (skwq_has_sleeper(wq
))
2533 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2534 POLLWRNORM
| POLLWRBAND
);
2536 /* Should agree with poll, otherwise some programs break */
2537 if (sock_writeable(sk
))
2538 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2544 static void sock_def_destruct(struct sock
*sk
)
2548 void sk_send_sigurg(struct sock
*sk
)
2550 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2551 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2552 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2554 EXPORT_SYMBOL(sk_send_sigurg
);
2556 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2557 unsigned long expires
)
2559 if (!mod_timer(timer
, expires
))
2562 EXPORT_SYMBOL(sk_reset_timer
);
2564 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2566 if (del_timer(timer
))
2569 EXPORT_SYMBOL(sk_stop_timer
);
2571 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2574 sk
->sk_send_head
= NULL
;
2576 init_timer(&sk
->sk_timer
);
2578 sk
->sk_allocation
= GFP_KERNEL
;
2579 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2580 sk
->sk_sndbuf
= sysctl_wmem_default
;
2581 sk
->sk_state
= TCP_CLOSE
;
2582 sk_set_socket(sk
, sock
);
2584 sock_set_flag(sk
, SOCK_ZAPPED
);
2587 sk
->sk_type
= sock
->type
;
2588 sk
->sk_wq
= sock
->wq
;
2590 sk
->sk_uid
= SOCK_INODE(sock
)->i_uid
;
2593 sk
->sk_uid
= make_kuid(sock_net(sk
)->user_ns
, 0);
2596 rwlock_init(&sk
->sk_callback_lock
);
2597 if (sk
->sk_kern_sock
)
2598 lockdep_set_class_and_name(
2599 &sk
->sk_callback_lock
,
2600 af_kern_callback_keys
+ sk
->sk_family
,
2601 af_family_kern_clock_key_strings
[sk
->sk_family
]);
2603 lockdep_set_class_and_name(
2604 &sk
->sk_callback_lock
,
2605 af_callback_keys
+ sk
->sk_family
,
2606 af_family_clock_key_strings
[sk
->sk_family
]);
2608 sk
->sk_state_change
= sock_def_wakeup
;
2609 sk
->sk_data_ready
= sock_def_readable
;
2610 sk
->sk_write_space
= sock_def_write_space
;
2611 sk
->sk_error_report
= sock_def_error_report
;
2612 sk
->sk_destruct
= sock_def_destruct
;
2614 sk
->sk_frag
.page
= NULL
;
2615 sk
->sk_frag
.offset
= 0;
2616 sk
->sk_peek_off
= -1;
2618 sk
->sk_peer_pid
= NULL
;
2619 sk
->sk_peer_cred
= NULL
;
2620 sk
->sk_write_pending
= 0;
2621 sk
->sk_rcvlowat
= 1;
2622 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2623 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2625 sk
->sk_stamp
= SK_DEFAULT_STAMP
;
2627 #ifdef CONFIG_NET_RX_BUSY_POLL
2629 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2632 sk
->sk_max_pacing_rate
= ~0U;
2633 sk
->sk_pacing_rate
= ~0U;
2634 sk
->sk_incoming_cpu
= -1;
2636 * Before updating sk_refcnt, we must commit prior changes to memory
2637 * (Documentation/RCU/rculist_nulls.txt for details)
2640 atomic_set(&sk
->sk_refcnt
, 1);
2641 atomic_set(&sk
->sk_drops
, 0);
2643 EXPORT_SYMBOL(sock_init_data
);
2645 void lock_sock_nested(struct sock
*sk
, int subclass
)
2648 spin_lock_bh(&sk
->sk_lock
.slock
);
2649 if (sk
->sk_lock
.owned
)
2651 sk
->sk_lock
.owned
= 1;
2652 spin_unlock(&sk
->sk_lock
.slock
);
2654 * The sk_lock has mutex_lock() semantics here:
2656 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2659 EXPORT_SYMBOL(lock_sock_nested
);
2661 void release_sock(struct sock
*sk
)
2663 spin_lock_bh(&sk
->sk_lock
.slock
);
2664 if (sk
->sk_backlog
.tail
)
2667 /* Warning : release_cb() might need to release sk ownership,
2668 * ie call sock_release_ownership(sk) before us.
2670 if (sk
->sk_prot
->release_cb
)
2671 sk
->sk_prot
->release_cb(sk
);
2673 sock_release_ownership(sk
);
2674 if (waitqueue_active(&sk
->sk_lock
.wq
))
2675 wake_up(&sk
->sk_lock
.wq
);
2676 spin_unlock_bh(&sk
->sk_lock
.slock
);
2678 EXPORT_SYMBOL(release_sock
);
2681 * lock_sock_fast - fast version of lock_sock
2684 * This version should be used for very small section, where process wont block
2685 * return false if fast path is taken
2686 * sk_lock.slock locked, owned = 0, BH disabled
2687 * return true if slow path is taken
2688 * sk_lock.slock unlocked, owned = 1, BH enabled
2690 bool lock_sock_fast(struct sock
*sk
)
2693 spin_lock_bh(&sk
->sk_lock
.slock
);
2695 if (!sk
->sk_lock
.owned
)
2697 * Note : We must disable BH
2702 sk
->sk_lock
.owned
= 1;
2703 spin_unlock(&sk
->sk_lock
.slock
);
2705 * The sk_lock has mutex_lock() semantics here:
2707 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2711 EXPORT_SYMBOL(lock_sock_fast
);
2713 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2716 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2717 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2718 tv
= ktime_to_timeval(sk
->sk_stamp
);
2719 if (tv
.tv_sec
== -1)
2721 if (tv
.tv_sec
== 0) {
2722 sk
->sk_stamp
= ktime_get_real();
2723 tv
= ktime_to_timeval(sk
->sk_stamp
);
2725 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2727 EXPORT_SYMBOL(sock_get_timestamp
);
2729 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2732 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2733 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2734 ts
= ktime_to_timespec(sk
->sk_stamp
);
2735 if (ts
.tv_sec
== -1)
2737 if (ts
.tv_sec
== 0) {
2738 sk
->sk_stamp
= ktime_get_real();
2739 ts
= ktime_to_timespec(sk
->sk_stamp
);
2741 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2743 EXPORT_SYMBOL(sock_get_timestampns
);
2745 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2747 if (!sock_flag(sk
, flag
)) {
2748 unsigned long previous_flags
= sk
->sk_flags
;
2750 sock_set_flag(sk
, flag
);
2752 * we just set one of the two flags which require net
2753 * time stamping, but time stamping might have been on
2754 * already because of the other one
2756 if (sock_needs_netstamp(sk
) &&
2757 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
2758 net_enable_timestamp();
2762 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2763 int level
, int type
)
2765 struct sock_exterr_skb
*serr
;
2766 struct sk_buff
*skb
;
2770 skb
= sock_dequeue_err_skb(sk
);
2776 msg
->msg_flags
|= MSG_TRUNC
;
2779 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2783 sock_recv_timestamp(msg
, sk
, skb
);
2785 serr
= SKB_EXT_ERR(skb
);
2786 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2788 msg
->msg_flags
|= MSG_ERRQUEUE
;
2796 EXPORT_SYMBOL(sock_recv_errqueue
);
2799 * Get a socket option on an socket.
2801 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2802 * asynchronous errors should be reported by getsockopt. We assume
2803 * this means if you specify SO_ERROR (otherwise whats the point of it).
2805 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2806 char __user
*optval
, int __user
*optlen
)
2808 struct sock
*sk
= sock
->sk
;
2810 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2812 EXPORT_SYMBOL(sock_common_getsockopt
);
2814 #ifdef CONFIG_COMPAT
2815 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2816 char __user
*optval
, int __user
*optlen
)
2818 struct sock
*sk
= sock
->sk
;
2820 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2821 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2823 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2825 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2828 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2831 struct sock
*sk
= sock
->sk
;
2835 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2836 flags
& ~MSG_DONTWAIT
, &addr_len
);
2838 msg
->msg_namelen
= addr_len
;
2841 EXPORT_SYMBOL(sock_common_recvmsg
);
2844 * Set socket options on an inet socket.
2846 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2847 char __user
*optval
, unsigned int optlen
)
2849 struct sock
*sk
= sock
->sk
;
2851 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2853 EXPORT_SYMBOL(sock_common_setsockopt
);
2855 #ifdef CONFIG_COMPAT
2856 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2857 char __user
*optval
, unsigned int optlen
)
2859 struct sock
*sk
= sock
->sk
;
2861 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2862 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2864 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2866 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2869 void sk_common_release(struct sock
*sk
)
2871 if (sk
->sk_prot
->destroy
)
2872 sk
->sk_prot
->destroy(sk
);
2875 * Observation: when sock_common_release is called, processes have
2876 * no access to socket. But net still has.
2877 * Step one, detach it from networking:
2879 * A. Remove from hash tables.
2882 sk
->sk_prot
->unhash(sk
);
2885 * In this point socket cannot receive new packets, but it is possible
2886 * that some packets are in flight because some CPU runs receiver and
2887 * did hash table lookup before we unhashed socket. They will achieve
2888 * receive queue and will be purged by socket destructor.
2890 * Also we still have packets pending on receive queue and probably,
2891 * our own packets waiting in device queues. sock_destroy will drain
2892 * receive queue, but transmitted packets will delay socket destruction
2893 * until the last reference will be released.
2898 xfrm_sk_free_policy(sk
);
2900 sk_refcnt_debug_release(sk
);
2904 EXPORT_SYMBOL(sk_common_release
);
2906 void sk_get_meminfo(const struct sock
*sk
, u32
*mem
)
2908 memset(mem
, 0, sizeof(*mem
) * SK_MEMINFO_VARS
);
2910 mem
[SK_MEMINFO_RMEM_ALLOC
] = sk_rmem_alloc_get(sk
);
2911 mem
[SK_MEMINFO_RCVBUF
] = sk
->sk_rcvbuf
;
2912 mem
[SK_MEMINFO_WMEM_ALLOC
] = sk_wmem_alloc_get(sk
);
2913 mem
[SK_MEMINFO_SNDBUF
] = sk
->sk_sndbuf
;
2914 mem
[SK_MEMINFO_FWD_ALLOC
] = sk
->sk_forward_alloc
;
2915 mem
[SK_MEMINFO_WMEM_QUEUED
] = sk
->sk_wmem_queued
;
2916 mem
[SK_MEMINFO_OPTMEM
] = atomic_read(&sk
->sk_omem_alloc
);
2917 mem
[SK_MEMINFO_BACKLOG
] = sk
->sk_backlog
.len
;
2918 mem
[SK_MEMINFO_DROPS
] = atomic_read(&sk
->sk_drops
);
2921 #ifdef CONFIG_PROC_FS
2922 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2924 int val
[PROTO_INUSE_NR
];
2927 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2929 #ifdef CONFIG_NET_NS
2930 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2932 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2934 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2936 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2938 int cpu
, idx
= prot
->inuse_idx
;
2941 for_each_possible_cpu(cpu
)
2942 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2944 return res
>= 0 ? res
: 0;
2946 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2948 static int __net_init
sock_inuse_init_net(struct net
*net
)
2950 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2951 return net
->core
.inuse
? 0 : -ENOMEM
;
2954 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2956 free_percpu(net
->core
.inuse
);
2959 static struct pernet_operations net_inuse_ops
= {
2960 .init
= sock_inuse_init_net
,
2961 .exit
= sock_inuse_exit_net
,
2964 static __init
int net_inuse_init(void)
2966 if (register_pernet_subsys(&net_inuse_ops
))
2967 panic("Cannot initialize net inuse counters");
2972 core_initcall(net_inuse_init
);
2974 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2976 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2978 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2980 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2982 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2984 int cpu
, idx
= prot
->inuse_idx
;
2987 for_each_possible_cpu(cpu
)
2988 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2990 return res
>= 0 ? res
: 0;
2992 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2995 static void assign_proto_idx(struct proto
*prot
)
2997 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2999 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
3000 pr_err("PROTO_INUSE_NR exhausted\n");
3004 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
3007 static void release_proto_idx(struct proto
*prot
)
3009 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
3010 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
3013 static inline void assign_proto_idx(struct proto
*prot
)
3017 static inline void release_proto_idx(struct proto
*prot
)
3022 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
3026 kfree(rsk_prot
->slab_name
);
3027 rsk_prot
->slab_name
= NULL
;
3028 kmem_cache_destroy(rsk_prot
->slab
);
3029 rsk_prot
->slab
= NULL
;
3032 static int req_prot_init(const struct proto
*prot
)
3034 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
3039 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
3041 if (!rsk_prot
->slab_name
)
3044 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
3045 rsk_prot
->obj_size
, 0,
3046 prot
->slab_flags
, NULL
);
3048 if (!rsk_prot
->slab
) {
3049 pr_crit("%s: Can't create request sock SLAB cache!\n",
3056 int proto_register(struct proto
*prot
, int alloc_slab
)
3059 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
3060 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
3063 if (prot
->slab
== NULL
) {
3064 pr_crit("%s: Can't create sock SLAB cache!\n",
3069 if (req_prot_init(prot
))
3070 goto out_free_request_sock_slab
;
3072 if (prot
->twsk_prot
!= NULL
) {
3073 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
3075 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
3076 goto out_free_request_sock_slab
;
3078 prot
->twsk_prot
->twsk_slab
=
3079 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
3080 prot
->twsk_prot
->twsk_obj_size
,
3084 if (prot
->twsk_prot
->twsk_slab
== NULL
)
3085 goto out_free_timewait_sock_slab_name
;
3089 mutex_lock(&proto_list_mutex
);
3090 list_add(&prot
->node
, &proto_list
);
3091 assign_proto_idx(prot
);
3092 mutex_unlock(&proto_list_mutex
);
3095 out_free_timewait_sock_slab_name
:
3096 kfree(prot
->twsk_prot
->twsk_slab_name
);
3097 out_free_request_sock_slab
:
3098 req_prot_cleanup(prot
->rsk_prot
);
3100 kmem_cache_destroy(prot
->slab
);
3105 EXPORT_SYMBOL(proto_register
);
3107 void proto_unregister(struct proto
*prot
)
3109 mutex_lock(&proto_list_mutex
);
3110 release_proto_idx(prot
);
3111 list_del(&prot
->node
);
3112 mutex_unlock(&proto_list_mutex
);
3114 kmem_cache_destroy(prot
->slab
);
3117 req_prot_cleanup(prot
->rsk_prot
);
3119 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
3120 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
3121 kfree(prot
->twsk_prot
->twsk_slab_name
);
3122 prot
->twsk_prot
->twsk_slab
= NULL
;
3125 EXPORT_SYMBOL(proto_unregister
);
3127 #ifdef CONFIG_PROC_FS
3128 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3129 __acquires(proto_list_mutex
)
3131 mutex_lock(&proto_list_mutex
);
3132 return seq_list_start_head(&proto_list
, *pos
);
3135 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3137 return seq_list_next(v
, &proto_list
, pos
);
3140 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
3141 __releases(proto_list_mutex
)
3143 mutex_unlock(&proto_list_mutex
);
3146 static char proto_method_implemented(const void *method
)
3148 return method
== NULL
? 'n' : 'y';
3150 static long sock_prot_memory_allocated(struct proto
*proto
)
3152 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
3155 static char *sock_prot_memory_pressure(struct proto
*proto
)
3157 return proto
->memory_pressure
!= NULL
?
3158 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
3161 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
3164 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3165 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3168 sock_prot_inuse_get(seq_file_net(seq
), proto
),
3169 sock_prot_memory_allocated(proto
),
3170 sock_prot_memory_pressure(proto
),
3172 proto
->slab
== NULL
? "no" : "yes",
3173 module_name(proto
->owner
),
3174 proto_method_implemented(proto
->close
),
3175 proto_method_implemented(proto
->connect
),
3176 proto_method_implemented(proto
->disconnect
),
3177 proto_method_implemented(proto
->accept
),
3178 proto_method_implemented(proto
->ioctl
),
3179 proto_method_implemented(proto
->init
),
3180 proto_method_implemented(proto
->destroy
),
3181 proto_method_implemented(proto
->shutdown
),
3182 proto_method_implemented(proto
->setsockopt
),
3183 proto_method_implemented(proto
->getsockopt
),
3184 proto_method_implemented(proto
->sendmsg
),
3185 proto_method_implemented(proto
->recvmsg
),
3186 proto_method_implemented(proto
->sendpage
),
3187 proto_method_implemented(proto
->bind
),
3188 proto_method_implemented(proto
->backlog_rcv
),
3189 proto_method_implemented(proto
->hash
),
3190 proto_method_implemented(proto
->unhash
),
3191 proto_method_implemented(proto
->get_port
),
3192 proto_method_implemented(proto
->enter_memory_pressure
));
3195 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3197 if (v
== &proto_list
)
3198 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3207 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3209 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3213 static const struct seq_operations proto_seq_ops
= {
3214 .start
= proto_seq_start
,
3215 .next
= proto_seq_next
,
3216 .stop
= proto_seq_stop
,
3217 .show
= proto_seq_show
,
3220 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
3222 return seq_open_net(inode
, file
, &proto_seq_ops
,
3223 sizeof(struct seq_net_private
));
3226 static const struct file_operations proto_seq_fops
= {
3227 .owner
= THIS_MODULE
,
3228 .open
= proto_seq_open
,
3230 .llseek
= seq_lseek
,
3231 .release
= seq_release_net
,
3234 static __net_init
int proto_init_net(struct net
*net
)
3236 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3242 static __net_exit
void proto_exit_net(struct net
*net
)
3244 remove_proc_entry("protocols", net
->proc_net
);
3248 static __net_initdata
struct pernet_operations proto_net_ops
= {
3249 .init
= proto_init_net
,
3250 .exit
= proto_exit_net
,
3253 static int __init
proto_init(void)
3255 return register_pernet_subsys(&proto_net_ops
);
3258 subsys_initcall(proto_init
);
3260 #endif /* PROC_FS */
3262 #ifdef CONFIG_NET_RX_BUSY_POLL
3263 bool sk_busy_loop_end(void *p
, unsigned long start_time
)
3265 struct sock
*sk
= p
;
3267 return !skb_queue_empty(&sk
->sk_receive_queue
) ||
3268 sk_busy_loop_timeout(sk
, start_time
);
3270 EXPORT_SYMBOL(sk_busy_loop_end
);
3271 #endif /* CONFIG_NET_RX_BUSY_POLL */