2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <asm/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex
);
148 static LIST_HEAD(proto_list
);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock
*sk
,
161 struct user_namespace
*user_ns
, int cap
)
163 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
164 ns_capable(user_ns
, cap
);
166 EXPORT_SYMBOL(sk_ns_capable
);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock
*sk
, int cap
)
179 return sk_ns_capable(sk
, &init_user_ns
, cap
);
181 EXPORT_SYMBOL(sk_capable
);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock
*sk
, int cap
)
194 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
196 EXPORT_SYMBOL(sk_net_capable
);
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
202 static struct lock_class_key af_family_keys
[AF_MAX
];
203 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
210 static const char *const af_family_key_strings
[AF_MAX
+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
226 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
227 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
228 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
229 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
230 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
231 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
232 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
233 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
234 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
235 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
236 "slock-27" , "slock-28" , "slock-AF_CAN" ,
237 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
238 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
239 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
240 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
242 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
243 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
244 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
245 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
246 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
247 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
248 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
249 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
250 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
251 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
252 "clock-27" , "clock-28" , "clock-AF_CAN" ,
253 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
254 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
255 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
256 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
260 * sk_callback_lock locking rules are per-address-family,
261 * so split the lock classes by using a per-AF key:
263 static struct lock_class_key af_callback_keys
[AF_MAX
];
265 /* Take into consideration the size of the struct sk_buff overhead in the
266 * determination of these values, since that is non-constant across
267 * platforms. This makes socket queueing behavior and performance
268 * not depend upon such differences.
270 #define _SK_MEM_PACKETS 256
271 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
272 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
273 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
275 /* Run time adjustable parameters. */
276 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
277 EXPORT_SYMBOL(sysctl_wmem_max
);
278 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
279 EXPORT_SYMBOL(sysctl_rmem_max
);
280 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
281 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
283 /* Maximal space eaten by iovec or ancillary data plus some space */
284 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
285 EXPORT_SYMBOL(sysctl_optmem_max
);
287 int sysctl_tstamp_allow_data __read_mostly
= 1;
289 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
290 EXPORT_SYMBOL_GPL(memalloc_socks
);
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
300 void sk_set_memalloc(struct sock
*sk
)
302 sock_set_flag(sk
, SOCK_MEMALLOC
);
303 sk
->sk_allocation
|= __GFP_MEMALLOC
;
304 static_key_slow_inc(&memalloc_socks
);
306 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
308 void sk_clear_memalloc(struct sock
*sk
)
310 sock_reset_flag(sk
, SOCK_MEMALLOC
);
311 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
312 static_key_slow_dec(&memalloc_socks
);
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
325 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
328 unsigned long pflags
= current
->flags
;
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
333 current
->flags
|= PF_MEMALLOC
;
334 ret
= sk
->sk_backlog_rcv(sk
, skb
);
335 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
339 EXPORT_SYMBOL(__sk_backlog_rcv
);
341 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
345 if (optlen
< sizeof(tv
))
347 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
349 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
353 static int warned __read_mostly
;
356 if (warned
< 10 && net_ratelimit()) {
358 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
359 __func__
, current
->comm
, task_pid_nr(current
));
363 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
364 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
366 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
367 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
371 static void sock_warn_obsolete_bsdism(const char *name
)
374 static char warncomm
[TASK_COMM_LEN
];
375 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
376 strcpy(warncomm
, current
->comm
);
377 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
383 static bool sock_needs_netstamp(const struct sock
*sk
)
385 switch (sk
->sk_family
) {
394 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
396 if (sk
->sk_flags
& flags
) {
397 sk
->sk_flags
&= ~flags
;
398 if (sock_needs_netstamp(sk
) &&
399 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
400 net_disable_timestamp();
405 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
408 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
410 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
411 atomic_inc(&sk
->sk_drops
);
412 trace_sock_rcvqueue_full(sk
, skb
);
416 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
417 atomic_inc(&sk
->sk_drops
);
422 skb_set_owner_r(skb
, sk
);
424 /* we escape from rcu protected region, make sure we dont leak
429 spin_lock_irqsave(&list
->lock
, flags
);
430 sock_skb_set_dropcount(sk
, skb
);
431 __skb_queue_tail(list
, skb
);
432 spin_unlock_irqrestore(&list
->lock
, flags
);
434 if (!sock_flag(sk
, SOCK_DEAD
))
435 sk
->sk_data_ready(sk
);
438 EXPORT_SYMBOL(__sock_queue_rcv_skb
);
440 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
444 err
= sk_filter(sk
, skb
);
448 return __sock_queue_rcv_skb(sk
, skb
);
450 EXPORT_SYMBOL(sock_queue_rcv_skb
);
452 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
454 int rc
= NET_RX_SUCCESS
;
456 if (sk_filter(sk
, skb
))
457 goto discard_and_relse
;
461 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
462 atomic_inc(&sk
->sk_drops
);
463 goto discard_and_relse
;
466 bh_lock_sock_nested(sk
);
469 if (!sock_owned_by_user(sk
)) {
471 * trylock + unlock semantics:
473 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
475 rc
= sk_backlog_rcv(sk
, skb
);
477 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
478 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
480 atomic_inc(&sk
->sk_drops
);
481 goto discard_and_relse
;
492 EXPORT_SYMBOL(sk_receive_skb
);
494 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
496 struct dst_entry
*dst
= __sk_dst_get(sk
);
498 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
499 sk_tx_queue_clear(sk
);
500 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
507 EXPORT_SYMBOL(__sk_dst_check
);
509 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
511 struct dst_entry
*dst
= sk_dst_get(sk
);
513 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
521 EXPORT_SYMBOL(sk_dst_check
);
523 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
526 int ret
= -ENOPROTOOPT
;
527 #ifdef CONFIG_NETDEVICES
528 struct net
*net
= sock_net(sk
);
529 char devname
[IFNAMSIZ
];
534 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
541 /* Bind this socket to a particular device like "eth0",
542 * as specified in the passed interface name. If the
543 * name is "" or the option length is zero the socket
546 if (optlen
> IFNAMSIZ
- 1)
547 optlen
= IFNAMSIZ
- 1;
548 memset(devname
, 0, sizeof(devname
));
551 if (copy_from_user(devname
, optval
, optlen
))
555 if (devname
[0] != '\0') {
556 struct net_device
*dev
;
559 dev
= dev_get_by_name_rcu(net
, devname
);
561 index
= dev
->ifindex
;
569 sk
->sk_bound_dev_if
= index
;
581 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
582 int __user
*optlen
, int len
)
584 int ret
= -ENOPROTOOPT
;
585 #ifdef CONFIG_NETDEVICES
586 struct net
*net
= sock_net(sk
);
587 char devname
[IFNAMSIZ
];
589 if (sk
->sk_bound_dev_if
== 0) {
598 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
602 len
= strlen(devname
) + 1;
605 if (copy_to_user(optval
, devname
, len
))
610 if (put_user(len
, optlen
))
621 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
624 sock_set_flag(sk
, bit
);
626 sock_reset_flag(sk
, bit
);
629 bool sk_mc_loop(struct sock
*sk
)
631 if (dev_recursion_level())
635 switch (sk
->sk_family
) {
637 return inet_sk(sk
)->mc_loop
;
638 #if IS_ENABLED(CONFIG_IPV6)
640 return inet6_sk(sk
)->mc_loop
;
646 EXPORT_SYMBOL(sk_mc_loop
);
649 * This is meant for all protocols to use and covers goings on
650 * at the socket level. Everything here is generic.
653 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
654 char __user
*optval
, unsigned int optlen
)
656 struct sock
*sk
= sock
->sk
;
663 * Options without arguments
666 if (optname
== SO_BINDTODEVICE
)
667 return sock_setbindtodevice(sk
, optval
, optlen
);
669 if (optlen
< sizeof(int))
672 if (get_user(val
, (int __user
*)optval
))
675 valbool
= val
? 1 : 0;
681 if (val
&& !capable(CAP_NET_ADMIN
))
684 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
687 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
690 sk
->sk_reuseport
= valbool
;
699 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
702 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
705 /* Don't error on this BSD doesn't and if you think
706 * about it this is right. Otherwise apps have to
707 * play 'guess the biggest size' games. RCVBUF/SNDBUF
708 * are treated in BSD as hints
710 val
= min_t(u32
, val
, sysctl_wmem_max
);
712 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
713 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
714 /* Wake up sending tasks if we upped the value. */
715 sk
->sk_write_space(sk
);
719 if (!capable(CAP_NET_ADMIN
)) {
726 /* Don't error on this BSD doesn't and if you think
727 * about it this is right. Otherwise apps have to
728 * play 'guess the biggest size' games. RCVBUF/SNDBUF
729 * are treated in BSD as hints
731 val
= min_t(u32
, val
, sysctl_rmem_max
);
733 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
735 * We double it on the way in to account for
736 * "struct sk_buff" etc. overhead. Applications
737 * assume that the SO_RCVBUF setting they make will
738 * allow that much actual data to be received on that
741 * Applications are unaware that "struct sk_buff" and
742 * other overheads allocate from the receive buffer
743 * during socket buffer allocation.
745 * And after considering the possible alternatives,
746 * returning the value we actually used in getsockopt
747 * is the most desirable behavior.
749 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
753 if (!capable(CAP_NET_ADMIN
)) {
761 if (sk
->sk_protocol
== IPPROTO_TCP
&&
762 sk
->sk_type
== SOCK_STREAM
)
763 tcp_set_keepalive(sk
, valbool
);
765 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
769 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
773 sk
->sk_no_check_tx
= valbool
;
777 if ((val
>= 0 && val
<= 6) ||
778 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
779 sk
->sk_priority
= val
;
785 if (optlen
< sizeof(ling
)) {
786 ret
= -EINVAL
; /* 1003.1g */
789 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
794 sock_reset_flag(sk
, SOCK_LINGER
);
796 #if (BITS_PER_LONG == 32)
797 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
798 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
801 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
802 sock_set_flag(sk
, SOCK_LINGER
);
807 sock_warn_obsolete_bsdism("setsockopt");
812 set_bit(SOCK_PASSCRED
, &sock
->flags
);
814 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
820 if (optname
== SO_TIMESTAMP
)
821 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
823 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
824 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
825 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
827 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
828 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
832 case SO_TIMESTAMPING
:
833 if (val
& ~SOF_TIMESTAMPING_MASK
) {
838 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
839 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
840 if (sk
->sk_protocol
== IPPROTO_TCP
&&
841 sk
->sk_type
== SOCK_STREAM
) {
842 if ((1 << sk
->sk_state
) &
843 (TCPF_CLOSE
| TCPF_LISTEN
)) {
847 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
852 sk
->sk_tsflags
= val
;
853 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
854 sock_enable_timestamp(sk
,
855 SOCK_TIMESTAMPING_RX_SOFTWARE
);
857 sock_disable_timestamp(sk
,
858 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
864 sk
->sk_rcvlowat
= val
? : 1;
868 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
872 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
875 case SO_ATTACH_FILTER
:
877 if (optlen
== sizeof(struct sock_fprog
)) {
878 struct sock_fprog fprog
;
881 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
884 ret
= sk_attach_filter(&fprog
, sk
);
890 if (optlen
== sizeof(u32
)) {
894 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
897 ret
= sk_attach_bpf(ufd
, sk
);
901 case SO_ATTACH_REUSEPORT_CBPF
:
903 if (optlen
== sizeof(struct sock_fprog
)) {
904 struct sock_fprog fprog
;
907 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
910 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
914 case SO_ATTACH_REUSEPORT_EBPF
:
916 if (optlen
== sizeof(u32
)) {
920 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
923 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
927 case SO_DETACH_FILTER
:
928 ret
= sk_detach_filter(sk
);
932 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
935 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
940 set_bit(SOCK_PASSSEC
, &sock
->flags
);
942 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
945 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
952 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
956 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
960 if (sock
->ops
->set_peek_off
)
961 ret
= sock
->ops
->set_peek_off(sk
, val
);
967 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
970 case SO_SELECT_ERR_QUEUE
:
971 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
974 #ifdef CONFIG_NET_RX_BUSY_POLL
976 /* allow unprivileged users to decrease the value */
977 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
983 sk
->sk_ll_usec
= val
;
988 case SO_MAX_PACING_RATE
:
989 sk
->sk_max_pacing_rate
= val
;
990 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
991 sk
->sk_max_pacing_rate
);
994 case SO_INCOMING_CPU
:
995 sk
->sk_incoming_cpu
= val
;
1000 dst_negative_advice(sk
);
1009 EXPORT_SYMBOL(sock_setsockopt
);
1012 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1013 struct ucred
*ucred
)
1015 ucred
->pid
= pid_vnr(pid
);
1016 ucred
->uid
= ucred
->gid
= -1;
1018 struct user_namespace
*current_ns
= current_user_ns();
1020 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1021 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1025 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1026 char __user
*optval
, int __user
*optlen
)
1028 struct sock
*sk
= sock
->sk
;
1036 int lv
= sizeof(int);
1039 if (get_user(len
, optlen
))
1044 memset(&v
, 0, sizeof(v
));
1048 v
.val
= sock_flag(sk
, SOCK_DBG
);
1052 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1056 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1060 v
.val
= sk
->sk_sndbuf
;
1064 v
.val
= sk
->sk_rcvbuf
;
1068 v
.val
= sk
->sk_reuse
;
1072 v
.val
= sk
->sk_reuseport
;
1076 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1080 v
.val
= sk
->sk_type
;
1084 v
.val
= sk
->sk_protocol
;
1088 v
.val
= sk
->sk_family
;
1092 v
.val
= -sock_error(sk
);
1094 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1098 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1102 v
.val
= sk
->sk_no_check_tx
;
1106 v
.val
= sk
->sk_priority
;
1110 lv
= sizeof(v
.ling
);
1111 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1112 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1116 sock_warn_obsolete_bsdism("getsockopt");
1120 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1121 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1124 case SO_TIMESTAMPNS
:
1125 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1128 case SO_TIMESTAMPING
:
1129 v
.val
= sk
->sk_tsflags
;
1133 lv
= sizeof(struct timeval
);
1134 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1138 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1139 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1144 lv
= sizeof(struct timeval
);
1145 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1149 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1150 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1155 v
.val
= sk
->sk_rcvlowat
;
1163 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1168 struct ucred peercred
;
1169 if (len
> sizeof(peercred
))
1170 len
= sizeof(peercred
);
1171 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1172 if (copy_to_user(optval
, &peercred
, len
))
1181 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1185 if (copy_to_user(optval
, address
, len
))
1190 /* Dubious BSD thing... Probably nobody even uses it, but
1191 * the UNIX standard wants it for whatever reason... -DaveM
1194 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1198 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1202 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1205 v
.val
= sk
->sk_mark
;
1209 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1212 case SO_WIFI_STATUS
:
1213 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1217 if (!sock
->ops
->set_peek_off
)
1220 v
.val
= sk
->sk_peek_off
;
1223 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1226 case SO_BINDTODEVICE
:
1227 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1230 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1236 case SO_LOCK_FILTER
:
1237 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1240 case SO_BPF_EXTENSIONS
:
1241 v
.val
= bpf_tell_extensions();
1244 case SO_SELECT_ERR_QUEUE
:
1245 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1248 #ifdef CONFIG_NET_RX_BUSY_POLL
1250 v
.val
= sk
->sk_ll_usec
;
1254 case SO_MAX_PACING_RATE
:
1255 v
.val
= sk
->sk_max_pacing_rate
;
1258 case SO_INCOMING_CPU
:
1259 v
.val
= sk
->sk_incoming_cpu
;
1263 /* We implement the SO_SNDLOWAT etc to not be settable
1266 return -ENOPROTOOPT
;
1271 if (copy_to_user(optval
, &v
, len
))
1274 if (put_user(len
, optlen
))
1280 * Initialize an sk_lock.
1282 * (We also register the sk_lock with the lock validator.)
1284 static inline void sock_lock_init(struct sock
*sk
)
1286 sock_lock_init_class_and_name(sk
,
1287 af_family_slock_key_strings
[sk
->sk_family
],
1288 af_family_slock_keys
+ sk
->sk_family
,
1289 af_family_key_strings
[sk
->sk_family
],
1290 af_family_keys
+ sk
->sk_family
);
1294 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1295 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1296 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1298 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1300 #ifdef CONFIG_SECURITY_NETWORK
1301 void *sptr
= nsk
->sk_security
;
1303 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1305 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1306 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1308 #ifdef CONFIG_SECURITY_NETWORK
1309 nsk
->sk_security
= sptr
;
1310 security_sk_clone(osk
, nsk
);
1314 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1316 unsigned long nulls1
, nulls2
;
1318 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1319 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1320 if (nulls1
> nulls2
)
1321 swap(nulls1
, nulls2
);
1324 memset((char *)sk
, 0, nulls1
);
1325 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1326 nulls2
- nulls1
- sizeof(void *));
1327 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1328 size
- nulls2
- sizeof(void *));
1330 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1332 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1336 struct kmem_cache
*slab
;
1340 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1343 if (priority
& __GFP_ZERO
) {
1345 prot
->clear_sk(sk
, prot
->obj_size
);
1347 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1350 sk
= kmalloc(prot
->obj_size
, priority
);
1353 kmemcheck_annotate_bitfield(sk
, flags
);
1355 if (security_sk_alloc(sk
, family
, priority
))
1358 if (!try_module_get(prot
->owner
))
1360 sk_tx_queue_clear(sk
);
1361 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1367 security_sk_free(sk
);
1370 kmem_cache_free(slab
, sk
);
1376 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1378 struct kmem_cache
*slab
;
1379 struct module
*owner
;
1381 owner
= prot
->owner
;
1384 cgroup_sk_free(&sk
->sk_cgrp_data
);
1385 security_sk_free(sk
);
1387 kmem_cache_free(slab
, sk
);
1394 * sk_alloc - All socket objects are allocated here
1395 * @net: the applicable net namespace
1396 * @family: protocol family
1397 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1398 * @prot: struct proto associated with this new sock instance
1399 * @kern: is this to be a kernel socket?
1401 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1402 struct proto
*prot
, int kern
)
1406 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1408 sk
->sk_family
= family
;
1410 * See comment in struct sock definition to understand
1411 * why we need sk_prot_creator -acme
1413 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1415 sk
->sk_net_refcnt
= kern
? 0 : 1;
1416 if (likely(sk
->sk_net_refcnt
))
1418 sock_net_set(sk
, net
);
1419 atomic_set(&sk
->sk_wmem_alloc
, 1);
1421 sock_update_classid(&sk
->sk_cgrp_data
);
1422 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1427 EXPORT_SYMBOL(sk_alloc
);
1429 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1430 * grace period. This is the case for UDP sockets and TCP listeners.
1432 static void __sk_destruct(struct rcu_head
*head
)
1434 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1435 struct sk_filter
*filter
;
1437 if (sk
->sk_destruct
)
1438 sk
->sk_destruct(sk
);
1440 filter
= rcu_dereference_check(sk
->sk_filter
,
1441 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1443 sk_filter_uncharge(sk
, filter
);
1444 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1446 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1447 reuseport_detach_sock(sk
);
1449 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1451 if (atomic_read(&sk
->sk_omem_alloc
))
1452 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1453 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1455 if (sk
->sk_peer_cred
)
1456 put_cred(sk
->sk_peer_cred
);
1457 put_pid(sk
->sk_peer_pid
);
1458 if (likely(sk
->sk_net_refcnt
))
1459 put_net(sock_net(sk
));
1460 sk_prot_free(sk
->sk_prot_creator
, sk
);
1463 void sk_destruct(struct sock
*sk
)
1465 if (sock_flag(sk
, SOCK_RCU_FREE
))
1466 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1468 __sk_destruct(&sk
->sk_rcu
);
1471 static void __sk_free(struct sock
*sk
)
1473 if (unlikely(sock_diag_has_destroy_listeners(sk
) && sk
->sk_net_refcnt
))
1474 sock_diag_broadcast_destroy(sk
);
1479 void sk_free(struct sock
*sk
)
1482 * We subtract one from sk_wmem_alloc and can know if
1483 * some packets are still in some tx queue.
1484 * If not null, sock_wfree() will call __sk_free(sk) later
1486 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1489 EXPORT_SYMBOL(sk_free
);
1492 * sk_clone_lock - clone a socket, and lock its clone
1493 * @sk: the socket to clone
1494 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1496 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1498 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1501 bool is_charged
= true;
1503 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1504 if (newsk
!= NULL
) {
1505 struct sk_filter
*filter
;
1507 sock_copy(newsk
, sk
);
1510 if (likely(newsk
->sk_net_refcnt
))
1511 get_net(sock_net(newsk
));
1512 sk_node_init(&newsk
->sk_node
);
1513 sock_lock_init(newsk
);
1514 bh_lock_sock(newsk
);
1515 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1516 newsk
->sk_backlog
.len
= 0;
1518 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1520 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1522 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1523 atomic_set(&newsk
->sk_omem_alloc
, 0);
1524 skb_queue_head_init(&newsk
->sk_receive_queue
);
1525 skb_queue_head_init(&newsk
->sk_write_queue
);
1527 rwlock_init(&newsk
->sk_callback_lock
);
1528 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1529 af_callback_keys
+ newsk
->sk_family
,
1530 af_family_clock_key_strings
[newsk
->sk_family
]);
1532 newsk
->sk_dst_cache
= NULL
;
1533 newsk
->sk_wmem_queued
= 0;
1534 newsk
->sk_forward_alloc
= 0;
1535 atomic_set(&newsk
->sk_drops
, 0);
1536 newsk
->sk_send_head
= NULL
;
1537 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1539 sock_reset_flag(newsk
, SOCK_DONE
);
1540 skb_queue_head_init(&newsk
->sk_error_queue
);
1542 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1544 /* though it's an empty new sock, the charging may fail
1545 * if sysctl_optmem_max was changed between creation of
1546 * original socket and cloning
1548 is_charged
= sk_filter_charge(newsk
, filter
);
1550 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1551 /* It is still raw copy of parent, so invalidate
1552 * destructor and make plain sk_free() */
1553 newsk
->sk_destruct
= NULL
;
1554 bh_unlock_sock(newsk
);
1559 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1562 newsk
->sk_priority
= 0;
1563 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1564 atomic64_set(&newsk
->sk_cookie
, 0);
1566 * Before updating sk_refcnt, we must commit prior changes to memory
1567 * (Documentation/RCU/rculist_nulls.txt for details)
1570 atomic_set(&newsk
->sk_refcnt
, 2);
1573 * Increment the counter in the same struct proto as the master
1574 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1575 * is the same as sk->sk_prot->socks, as this field was copied
1578 * This _changes_ the previous behaviour, where
1579 * tcp_create_openreq_child always was incrementing the
1580 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1581 * to be taken into account in all callers. -acme
1583 sk_refcnt_debug_inc(newsk
);
1584 sk_set_socket(newsk
, NULL
);
1585 newsk
->sk_wq
= NULL
;
1587 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
1588 sock_update_memcg(newsk
);
1590 if (newsk
->sk_prot
->sockets_allocated
)
1591 sk_sockets_allocated_inc(newsk
);
1593 if (sock_needs_netstamp(sk
) &&
1594 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1595 net_enable_timestamp();
1600 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1602 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1606 sk_dst_set(sk
, dst
);
1607 sk
->sk_route_caps
= dst
->dev
->features
;
1608 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1609 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1610 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1611 if (sk_can_gso(sk
)) {
1612 if (dst
->header_len
) {
1613 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1615 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1616 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1617 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1620 sk
->sk_gso_max_segs
= max_segs
;
1622 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1625 * Simple resource managers for sockets.
1630 * Write buffer destructor automatically called from kfree_skb.
1632 void sock_wfree(struct sk_buff
*skb
)
1634 struct sock
*sk
= skb
->sk
;
1635 unsigned int len
= skb
->truesize
;
1637 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1639 * Keep a reference on sk_wmem_alloc, this will be released
1640 * after sk_write_space() call
1642 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1643 sk
->sk_write_space(sk
);
1647 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1648 * could not do because of in-flight packets
1650 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1653 EXPORT_SYMBOL(sock_wfree
);
1655 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1660 if (unlikely(!sk_fullsock(sk
))) {
1661 skb
->destructor
= sock_edemux
;
1666 skb
->destructor
= sock_wfree
;
1667 skb_set_hash_from_sk(skb
, sk
);
1669 * We used to take a refcount on sk, but following operation
1670 * is enough to guarantee sk_free() wont free this sock until
1671 * all in-flight packets are completed
1673 atomic_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1675 EXPORT_SYMBOL(skb_set_owner_w
);
1677 void skb_orphan_partial(struct sk_buff
*skb
)
1679 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1680 * so we do not completely orphan skb, but transfert all
1681 * accounted bytes but one, to avoid unexpected reorders.
1683 if (skb
->destructor
== sock_wfree
1685 || skb
->destructor
== tcp_wfree
1688 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1694 EXPORT_SYMBOL(skb_orphan_partial
);
1697 * Read buffer destructor automatically called from kfree_skb.
1699 void sock_rfree(struct sk_buff
*skb
)
1701 struct sock
*sk
= skb
->sk
;
1702 unsigned int len
= skb
->truesize
;
1704 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1705 sk_mem_uncharge(sk
, len
);
1707 EXPORT_SYMBOL(sock_rfree
);
1710 * Buffer destructor for skbs that are not used directly in read or write
1711 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1713 void sock_efree(struct sk_buff
*skb
)
1717 EXPORT_SYMBOL(sock_efree
);
1719 kuid_t
sock_i_uid(struct sock
*sk
)
1723 read_lock_bh(&sk
->sk_callback_lock
);
1724 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1725 read_unlock_bh(&sk
->sk_callback_lock
);
1728 EXPORT_SYMBOL(sock_i_uid
);
1730 unsigned long sock_i_ino(struct sock
*sk
)
1734 read_lock_bh(&sk
->sk_callback_lock
);
1735 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1736 read_unlock_bh(&sk
->sk_callback_lock
);
1739 EXPORT_SYMBOL(sock_i_ino
);
1742 * Allocate a skb from the socket's send buffer.
1744 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1747 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1748 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1750 skb_set_owner_w(skb
, sk
);
1756 EXPORT_SYMBOL(sock_wmalloc
);
1759 * Allocate a memory block from the socket's option memory buffer.
1761 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1763 if ((unsigned int)size
<= sysctl_optmem_max
&&
1764 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1766 /* First do the add, to avoid the race if kmalloc
1769 atomic_add(size
, &sk
->sk_omem_alloc
);
1770 mem
= kmalloc(size
, priority
);
1773 atomic_sub(size
, &sk
->sk_omem_alloc
);
1777 EXPORT_SYMBOL(sock_kmalloc
);
1779 /* Free an option memory block. Note, we actually want the inline
1780 * here as this allows gcc to detect the nullify and fold away the
1781 * condition entirely.
1783 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1786 if (WARN_ON_ONCE(!mem
))
1792 atomic_sub(size
, &sk
->sk_omem_alloc
);
1795 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1797 __sock_kfree_s(sk
, mem
, size
, false);
1799 EXPORT_SYMBOL(sock_kfree_s
);
1801 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1803 __sock_kfree_s(sk
, mem
, size
, true);
1805 EXPORT_SYMBOL(sock_kzfree_s
);
1807 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1808 I think, these locks should be removed for datagram sockets.
1810 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1814 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1818 if (signal_pending(current
))
1820 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1821 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1822 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1824 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1828 timeo
= schedule_timeout(timeo
);
1830 finish_wait(sk_sleep(sk
), &wait
);
1836 * Generic send/receive buffer handlers
1839 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1840 unsigned long data_len
, int noblock
,
1841 int *errcode
, int max_page_order
)
1843 struct sk_buff
*skb
;
1847 timeo
= sock_sndtimeo(sk
, noblock
);
1849 err
= sock_error(sk
);
1854 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1857 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1860 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1861 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1865 if (signal_pending(current
))
1867 timeo
= sock_wait_for_wmem(sk
, timeo
);
1869 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
1870 errcode
, sk
->sk_allocation
);
1872 skb_set_owner_w(skb
, sk
);
1876 err
= sock_intr_errno(timeo
);
1881 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1883 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1884 int noblock
, int *errcode
)
1886 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1888 EXPORT_SYMBOL(sock_alloc_send_skb
);
1890 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
1891 struct sockcm_cookie
*sockc
)
1895 switch (cmsg
->cmsg_type
) {
1897 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
1899 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1901 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
1903 case SO_TIMESTAMPING
:
1904 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1907 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
1908 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
1911 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
1912 sockc
->tsflags
|= tsflags
;
1919 EXPORT_SYMBOL(__sock_cmsg_send
);
1921 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
1922 struct sockcm_cookie
*sockc
)
1924 struct cmsghdr
*cmsg
;
1927 for_each_cmsghdr(cmsg
, msg
) {
1928 if (!CMSG_OK(msg
, cmsg
))
1930 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
1932 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
1938 EXPORT_SYMBOL(sock_cmsg_send
);
1940 /* On 32bit arches, an skb frag is limited to 2^15 */
1941 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1944 * skb_page_frag_refill - check that a page_frag contains enough room
1945 * @sz: minimum size of the fragment we want to get
1946 * @pfrag: pointer to page_frag
1947 * @gfp: priority for memory allocation
1949 * Note: While this allocator tries to use high order pages, there is
1950 * no guarantee that allocations succeed. Therefore, @sz MUST be
1951 * less or equal than PAGE_SIZE.
1953 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
1956 if (page_ref_count(pfrag
->page
) == 1) {
1960 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1962 put_page(pfrag
->page
);
1966 if (SKB_FRAG_PAGE_ORDER
) {
1967 /* Avoid direct reclaim but allow kswapd to wake */
1968 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
1969 __GFP_COMP
| __GFP_NOWARN
|
1971 SKB_FRAG_PAGE_ORDER
);
1972 if (likely(pfrag
->page
)) {
1973 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
1977 pfrag
->page
= alloc_page(gfp
);
1978 if (likely(pfrag
->page
)) {
1979 pfrag
->size
= PAGE_SIZE
;
1984 EXPORT_SYMBOL(skb_page_frag_refill
);
1986 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1988 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
1991 sk_enter_memory_pressure(sk
);
1992 sk_stream_moderate_sndbuf(sk
);
1995 EXPORT_SYMBOL(sk_page_frag_refill
);
1997 static void __lock_sock(struct sock
*sk
)
1998 __releases(&sk
->sk_lock
.slock
)
1999 __acquires(&sk
->sk_lock
.slock
)
2004 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2005 TASK_UNINTERRUPTIBLE
);
2006 spin_unlock_bh(&sk
->sk_lock
.slock
);
2008 spin_lock_bh(&sk
->sk_lock
.slock
);
2009 if (!sock_owned_by_user(sk
))
2012 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2015 static void __release_sock(struct sock
*sk
)
2016 __releases(&sk
->sk_lock
.slock
)
2017 __acquires(&sk
->sk_lock
.slock
)
2019 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
2022 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2026 struct sk_buff
*next
= skb
->next
;
2029 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2031 sk_backlog_rcv(sk
, skb
);
2034 * We are in process context here with softirqs
2035 * disabled, use cond_resched_softirq() to preempt.
2036 * This is safe to do because we've taken the backlog
2039 cond_resched_softirq();
2042 } while (skb
!= NULL
);
2045 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
2048 * Doing the zeroing here guarantee we can not loop forever
2049 * while a wild producer attempts to flood us.
2051 sk
->sk_backlog
.len
= 0;
2055 * sk_wait_data - wait for data to arrive at sk_receive_queue
2056 * @sk: sock to wait on
2057 * @timeo: for how long
2058 * @skb: last skb seen on sk_receive_queue
2060 * Now socket state including sk->sk_err is changed only under lock,
2061 * hence we may omit checks after joining wait queue.
2062 * We check receive queue before schedule() only as optimization;
2063 * it is very likely that release_sock() added new data.
2065 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2070 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2071 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2072 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
);
2073 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2074 finish_wait(sk_sleep(sk
), &wait
);
2077 EXPORT_SYMBOL(sk_wait_data
);
2080 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2082 * @size: memory size to allocate
2083 * @kind: allocation type
2085 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2086 * rmem allocation. This function assumes that protocols which have
2087 * memory_pressure use sk_wmem_queued as write buffer accounting.
2089 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2091 struct proto
*prot
= sk
->sk_prot
;
2092 int amt
= sk_mem_pages(size
);
2095 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2097 allocated
= sk_memory_allocated_add(sk
, amt
);
2099 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2100 !mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
))
2101 goto suppress_allocation
;
2104 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2105 sk_leave_memory_pressure(sk
);
2109 /* Under pressure. */
2110 if (allocated
> sk_prot_mem_limits(sk
, 1))
2111 sk_enter_memory_pressure(sk
);
2113 /* Over hard limit. */
2114 if (allocated
> sk_prot_mem_limits(sk
, 2))
2115 goto suppress_allocation
;
2117 /* guarantee minimum buffer size under pressure */
2118 if (kind
== SK_MEM_RECV
) {
2119 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2122 } else { /* SK_MEM_SEND */
2123 if (sk
->sk_type
== SOCK_STREAM
) {
2124 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2126 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2127 prot
->sysctl_wmem
[0])
2131 if (sk_has_memory_pressure(sk
)) {
2134 if (!sk_under_memory_pressure(sk
))
2136 alloc
= sk_sockets_allocated_read_positive(sk
);
2137 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2138 sk_mem_pages(sk
->sk_wmem_queued
+
2139 atomic_read(&sk
->sk_rmem_alloc
) +
2140 sk
->sk_forward_alloc
))
2144 suppress_allocation
:
2146 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2147 sk_stream_moderate_sndbuf(sk
);
2149 /* Fail only if socket is _under_ its sndbuf.
2150 * In this case we cannot block, so that we have to fail.
2152 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2156 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2158 /* Alas. Undo changes. */
2159 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2161 sk_memory_allocated_sub(sk
, amt
);
2163 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2164 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2168 EXPORT_SYMBOL(__sk_mem_schedule
);
2171 * __sk_mem_reclaim - reclaim memory_allocated
2173 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2175 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2177 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2178 sk_memory_allocated_sub(sk
, amount
);
2179 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2181 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2182 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2184 if (sk_under_memory_pressure(sk
) &&
2185 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2186 sk_leave_memory_pressure(sk
);
2188 EXPORT_SYMBOL(__sk_mem_reclaim
);
2192 * Set of default routines for initialising struct proto_ops when
2193 * the protocol does not support a particular function. In certain
2194 * cases where it makes no sense for a protocol to have a "do nothing"
2195 * function, some default processing is provided.
2198 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2202 EXPORT_SYMBOL(sock_no_bind
);
2204 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2209 EXPORT_SYMBOL(sock_no_connect
);
2211 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2215 EXPORT_SYMBOL(sock_no_socketpair
);
2217 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2221 EXPORT_SYMBOL(sock_no_accept
);
2223 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2228 EXPORT_SYMBOL(sock_no_getname
);
2230 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2234 EXPORT_SYMBOL(sock_no_poll
);
2236 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2240 EXPORT_SYMBOL(sock_no_ioctl
);
2242 int sock_no_listen(struct socket
*sock
, int backlog
)
2246 EXPORT_SYMBOL(sock_no_listen
);
2248 int sock_no_shutdown(struct socket
*sock
, int how
)
2252 EXPORT_SYMBOL(sock_no_shutdown
);
2254 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2255 char __user
*optval
, unsigned int optlen
)
2259 EXPORT_SYMBOL(sock_no_setsockopt
);
2261 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2262 char __user
*optval
, int __user
*optlen
)
2266 EXPORT_SYMBOL(sock_no_getsockopt
);
2268 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2272 EXPORT_SYMBOL(sock_no_sendmsg
);
2274 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2279 EXPORT_SYMBOL(sock_no_recvmsg
);
2281 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2283 /* Mirror missing mmap method error code */
2286 EXPORT_SYMBOL(sock_no_mmap
);
2288 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2291 struct msghdr msg
= {.msg_flags
= flags
};
2293 char *kaddr
= kmap(page
);
2294 iov
.iov_base
= kaddr
+ offset
;
2296 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2300 EXPORT_SYMBOL(sock_no_sendpage
);
2303 * Default Socket Callbacks
2306 static void sock_def_wakeup(struct sock
*sk
)
2308 struct socket_wq
*wq
;
2311 wq
= rcu_dereference(sk
->sk_wq
);
2312 if (skwq_has_sleeper(wq
))
2313 wake_up_interruptible_all(&wq
->wait
);
2317 static void sock_def_error_report(struct sock
*sk
)
2319 struct socket_wq
*wq
;
2322 wq
= rcu_dereference(sk
->sk_wq
);
2323 if (skwq_has_sleeper(wq
))
2324 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2325 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2329 static void sock_def_readable(struct sock
*sk
)
2331 struct socket_wq
*wq
;
2334 wq
= rcu_dereference(sk
->sk_wq
);
2335 if (skwq_has_sleeper(wq
))
2336 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2337 POLLRDNORM
| POLLRDBAND
);
2338 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2342 static void sock_def_write_space(struct sock
*sk
)
2344 struct socket_wq
*wq
;
2348 /* Do not wake up a writer until he can make "significant"
2351 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2352 wq
= rcu_dereference(sk
->sk_wq
);
2353 if (skwq_has_sleeper(wq
))
2354 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2355 POLLWRNORM
| POLLWRBAND
);
2357 /* Should agree with poll, otherwise some programs break */
2358 if (sock_writeable(sk
))
2359 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2365 static void sock_def_destruct(struct sock
*sk
)
2369 void sk_send_sigurg(struct sock
*sk
)
2371 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2372 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2373 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2375 EXPORT_SYMBOL(sk_send_sigurg
);
2377 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2378 unsigned long expires
)
2380 if (!mod_timer(timer
, expires
))
2383 EXPORT_SYMBOL(sk_reset_timer
);
2385 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2387 if (del_timer(timer
))
2390 EXPORT_SYMBOL(sk_stop_timer
);
2392 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2394 skb_queue_head_init(&sk
->sk_receive_queue
);
2395 skb_queue_head_init(&sk
->sk_write_queue
);
2396 skb_queue_head_init(&sk
->sk_error_queue
);
2398 sk
->sk_send_head
= NULL
;
2400 init_timer(&sk
->sk_timer
);
2402 sk
->sk_allocation
= GFP_KERNEL
;
2403 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2404 sk
->sk_sndbuf
= sysctl_wmem_default
;
2405 sk
->sk_state
= TCP_CLOSE
;
2406 sk_set_socket(sk
, sock
);
2408 sock_set_flag(sk
, SOCK_ZAPPED
);
2411 sk
->sk_type
= sock
->type
;
2412 sk
->sk_wq
= sock
->wq
;
2417 rwlock_init(&sk
->sk_callback_lock
);
2418 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2419 af_callback_keys
+ sk
->sk_family
,
2420 af_family_clock_key_strings
[sk
->sk_family
]);
2422 sk
->sk_state_change
= sock_def_wakeup
;
2423 sk
->sk_data_ready
= sock_def_readable
;
2424 sk
->sk_write_space
= sock_def_write_space
;
2425 sk
->sk_error_report
= sock_def_error_report
;
2426 sk
->sk_destruct
= sock_def_destruct
;
2428 sk
->sk_frag
.page
= NULL
;
2429 sk
->sk_frag
.offset
= 0;
2430 sk
->sk_peek_off
= -1;
2432 sk
->sk_peer_pid
= NULL
;
2433 sk
->sk_peer_cred
= NULL
;
2434 sk
->sk_write_pending
= 0;
2435 sk
->sk_rcvlowat
= 1;
2436 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2437 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2439 sk
->sk_stamp
= ktime_set(-1L, 0);
2441 #ifdef CONFIG_NET_RX_BUSY_POLL
2443 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2446 sk
->sk_max_pacing_rate
= ~0U;
2447 sk
->sk_pacing_rate
= ~0U;
2448 sk
->sk_incoming_cpu
= -1;
2450 * Before updating sk_refcnt, we must commit prior changes to memory
2451 * (Documentation/RCU/rculist_nulls.txt for details)
2454 atomic_set(&sk
->sk_refcnt
, 1);
2455 atomic_set(&sk
->sk_drops
, 0);
2457 EXPORT_SYMBOL(sock_init_data
);
2459 void lock_sock_nested(struct sock
*sk
, int subclass
)
2462 spin_lock_bh(&sk
->sk_lock
.slock
);
2463 if (sk
->sk_lock
.owned
)
2465 sk
->sk_lock
.owned
= 1;
2466 spin_unlock(&sk
->sk_lock
.slock
);
2468 * The sk_lock has mutex_lock() semantics here:
2470 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2473 EXPORT_SYMBOL(lock_sock_nested
);
2475 void release_sock(struct sock
*sk
)
2478 * The sk_lock has mutex_unlock() semantics:
2480 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2482 spin_lock_bh(&sk
->sk_lock
.slock
);
2483 if (sk
->sk_backlog
.tail
)
2486 /* Warning : release_cb() might need to release sk ownership,
2487 * ie call sock_release_ownership(sk) before us.
2489 if (sk
->sk_prot
->release_cb
)
2490 sk
->sk_prot
->release_cb(sk
);
2492 sock_release_ownership(sk
);
2493 if (waitqueue_active(&sk
->sk_lock
.wq
))
2494 wake_up(&sk
->sk_lock
.wq
);
2495 spin_unlock_bh(&sk
->sk_lock
.slock
);
2497 EXPORT_SYMBOL(release_sock
);
2500 * lock_sock_fast - fast version of lock_sock
2503 * This version should be used for very small section, where process wont block
2504 * return false if fast path is taken
2505 * sk_lock.slock locked, owned = 0, BH disabled
2506 * return true if slow path is taken
2507 * sk_lock.slock unlocked, owned = 1, BH enabled
2509 bool lock_sock_fast(struct sock
*sk
)
2512 spin_lock_bh(&sk
->sk_lock
.slock
);
2514 if (!sk
->sk_lock
.owned
)
2516 * Note : We must disable BH
2521 sk
->sk_lock
.owned
= 1;
2522 spin_unlock(&sk
->sk_lock
.slock
);
2524 * The sk_lock has mutex_lock() semantics here:
2526 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2530 EXPORT_SYMBOL(lock_sock_fast
);
2532 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2535 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2536 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2537 tv
= ktime_to_timeval(sk
->sk_stamp
);
2538 if (tv
.tv_sec
== -1)
2540 if (tv
.tv_sec
== 0) {
2541 sk
->sk_stamp
= ktime_get_real();
2542 tv
= ktime_to_timeval(sk
->sk_stamp
);
2544 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2546 EXPORT_SYMBOL(sock_get_timestamp
);
2548 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2551 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2552 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2553 ts
= ktime_to_timespec(sk
->sk_stamp
);
2554 if (ts
.tv_sec
== -1)
2556 if (ts
.tv_sec
== 0) {
2557 sk
->sk_stamp
= ktime_get_real();
2558 ts
= ktime_to_timespec(sk
->sk_stamp
);
2560 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2562 EXPORT_SYMBOL(sock_get_timestampns
);
2564 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2566 if (!sock_flag(sk
, flag
)) {
2567 unsigned long previous_flags
= sk
->sk_flags
;
2569 sock_set_flag(sk
, flag
);
2571 * we just set one of the two flags which require net
2572 * time stamping, but time stamping might have been on
2573 * already because of the other one
2575 if (sock_needs_netstamp(sk
) &&
2576 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
2577 net_enable_timestamp();
2581 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2582 int level
, int type
)
2584 struct sock_exterr_skb
*serr
;
2585 struct sk_buff
*skb
;
2589 skb
= sock_dequeue_err_skb(sk
);
2595 msg
->msg_flags
|= MSG_TRUNC
;
2598 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2602 sock_recv_timestamp(msg
, sk
, skb
);
2604 serr
= SKB_EXT_ERR(skb
);
2605 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2607 msg
->msg_flags
|= MSG_ERRQUEUE
;
2615 EXPORT_SYMBOL(sock_recv_errqueue
);
2618 * Get a socket option on an socket.
2620 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2621 * asynchronous errors should be reported by getsockopt. We assume
2622 * this means if you specify SO_ERROR (otherwise whats the point of it).
2624 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2625 char __user
*optval
, int __user
*optlen
)
2627 struct sock
*sk
= sock
->sk
;
2629 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2631 EXPORT_SYMBOL(sock_common_getsockopt
);
2633 #ifdef CONFIG_COMPAT
2634 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2635 char __user
*optval
, int __user
*optlen
)
2637 struct sock
*sk
= sock
->sk
;
2639 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2640 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2642 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2644 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2647 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2650 struct sock
*sk
= sock
->sk
;
2654 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2655 flags
& ~MSG_DONTWAIT
, &addr_len
);
2657 msg
->msg_namelen
= addr_len
;
2660 EXPORT_SYMBOL(sock_common_recvmsg
);
2663 * Set socket options on an inet socket.
2665 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2666 char __user
*optval
, unsigned int optlen
)
2668 struct sock
*sk
= sock
->sk
;
2670 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2672 EXPORT_SYMBOL(sock_common_setsockopt
);
2674 #ifdef CONFIG_COMPAT
2675 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2676 char __user
*optval
, unsigned int optlen
)
2678 struct sock
*sk
= sock
->sk
;
2680 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2681 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2683 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2685 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2688 void sk_common_release(struct sock
*sk
)
2690 if (sk
->sk_prot
->destroy
)
2691 sk
->sk_prot
->destroy(sk
);
2694 * Observation: when sock_common_release is called, processes have
2695 * no access to socket. But net still has.
2696 * Step one, detach it from networking:
2698 * A. Remove from hash tables.
2701 sk
->sk_prot
->unhash(sk
);
2704 * In this point socket cannot receive new packets, but it is possible
2705 * that some packets are in flight because some CPU runs receiver and
2706 * did hash table lookup before we unhashed socket. They will achieve
2707 * receive queue and will be purged by socket destructor.
2709 * Also we still have packets pending on receive queue and probably,
2710 * our own packets waiting in device queues. sock_destroy will drain
2711 * receive queue, but transmitted packets will delay socket destruction
2712 * until the last reference will be released.
2717 xfrm_sk_free_policy(sk
);
2719 sk_refcnt_debug_release(sk
);
2721 if (sk
->sk_frag
.page
) {
2722 put_page(sk
->sk_frag
.page
);
2723 sk
->sk_frag
.page
= NULL
;
2728 EXPORT_SYMBOL(sk_common_release
);
2730 #ifdef CONFIG_PROC_FS
2731 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2733 int val
[PROTO_INUSE_NR
];
2736 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2738 #ifdef CONFIG_NET_NS
2739 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2741 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2743 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2745 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2747 int cpu
, idx
= prot
->inuse_idx
;
2750 for_each_possible_cpu(cpu
)
2751 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2753 return res
>= 0 ? res
: 0;
2755 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2757 static int __net_init
sock_inuse_init_net(struct net
*net
)
2759 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2760 return net
->core
.inuse
? 0 : -ENOMEM
;
2763 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2765 free_percpu(net
->core
.inuse
);
2768 static struct pernet_operations net_inuse_ops
= {
2769 .init
= sock_inuse_init_net
,
2770 .exit
= sock_inuse_exit_net
,
2773 static __init
int net_inuse_init(void)
2775 if (register_pernet_subsys(&net_inuse_ops
))
2776 panic("Cannot initialize net inuse counters");
2781 core_initcall(net_inuse_init
);
2783 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2785 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2787 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2789 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2791 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2793 int cpu
, idx
= prot
->inuse_idx
;
2796 for_each_possible_cpu(cpu
)
2797 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2799 return res
>= 0 ? res
: 0;
2801 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2804 static void assign_proto_idx(struct proto
*prot
)
2806 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2808 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2809 pr_err("PROTO_INUSE_NR exhausted\n");
2813 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2816 static void release_proto_idx(struct proto
*prot
)
2818 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2819 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2822 static inline void assign_proto_idx(struct proto
*prot
)
2826 static inline void release_proto_idx(struct proto
*prot
)
2831 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
2835 kfree(rsk_prot
->slab_name
);
2836 rsk_prot
->slab_name
= NULL
;
2837 kmem_cache_destroy(rsk_prot
->slab
);
2838 rsk_prot
->slab
= NULL
;
2841 static int req_prot_init(const struct proto
*prot
)
2843 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
2848 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
2850 if (!rsk_prot
->slab_name
)
2853 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
2854 rsk_prot
->obj_size
, 0,
2855 prot
->slab_flags
, NULL
);
2857 if (!rsk_prot
->slab
) {
2858 pr_crit("%s: Can't create request sock SLAB cache!\n",
2865 int proto_register(struct proto
*prot
, int alloc_slab
)
2868 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2869 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2872 if (prot
->slab
== NULL
) {
2873 pr_crit("%s: Can't create sock SLAB cache!\n",
2878 if (req_prot_init(prot
))
2879 goto out_free_request_sock_slab
;
2881 if (prot
->twsk_prot
!= NULL
) {
2882 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2884 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2885 goto out_free_request_sock_slab
;
2887 prot
->twsk_prot
->twsk_slab
=
2888 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2889 prot
->twsk_prot
->twsk_obj_size
,
2893 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2894 goto out_free_timewait_sock_slab_name
;
2898 mutex_lock(&proto_list_mutex
);
2899 list_add(&prot
->node
, &proto_list
);
2900 assign_proto_idx(prot
);
2901 mutex_unlock(&proto_list_mutex
);
2904 out_free_timewait_sock_slab_name
:
2905 kfree(prot
->twsk_prot
->twsk_slab_name
);
2906 out_free_request_sock_slab
:
2907 req_prot_cleanup(prot
->rsk_prot
);
2909 kmem_cache_destroy(prot
->slab
);
2914 EXPORT_SYMBOL(proto_register
);
2916 void proto_unregister(struct proto
*prot
)
2918 mutex_lock(&proto_list_mutex
);
2919 release_proto_idx(prot
);
2920 list_del(&prot
->node
);
2921 mutex_unlock(&proto_list_mutex
);
2923 kmem_cache_destroy(prot
->slab
);
2926 req_prot_cleanup(prot
->rsk_prot
);
2928 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2929 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2930 kfree(prot
->twsk_prot
->twsk_slab_name
);
2931 prot
->twsk_prot
->twsk_slab
= NULL
;
2934 EXPORT_SYMBOL(proto_unregister
);
2936 #ifdef CONFIG_PROC_FS
2937 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2938 __acquires(proto_list_mutex
)
2940 mutex_lock(&proto_list_mutex
);
2941 return seq_list_start_head(&proto_list
, *pos
);
2944 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2946 return seq_list_next(v
, &proto_list
, pos
);
2949 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2950 __releases(proto_list_mutex
)
2952 mutex_unlock(&proto_list_mutex
);
2955 static char proto_method_implemented(const void *method
)
2957 return method
== NULL
? 'n' : 'y';
2959 static long sock_prot_memory_allocated(struct proto
*proto
)
2961 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2964 static char *sock_prot_memory_pressure(struct proto
*proto
)
2966 return proto
->memory_pressure
!= NULL
?
2967 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2970 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2973 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2974 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2977 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2978 sock_prot_memory_allocated(proto
),
2979 sock_prot_memory_pressure(proto
),
2981 proto
->slab
== NULL
? "no" : "yes",
2982 module_name(proto
->owner
),
2983 proto_method_implemented(proto
->close
),
2984 proto_method_implemented(proto
->connect
),
2985 proto_method_implemented(proto
->disconnect
),
2986 proto_method_implemented(proto
->accept
),
2987 proto_method_implemented(proto
->ioctl
),
2988 proto_method_implemented(proto
->init
),
2989 proto_method_implemented(proto
->destroy
),
2990 proto_method_implemented(proto
->shutdown
),
2991 proto_method_implemented(proto
->setsockopt
),
2992 proto_method_implemented(proto
->getsockopt
),
2993 proto_method_implemented(proto
->sendmsg
),
2994 proto_method_implemented(proto
->recvmsg
),
2995 proto_method_implemented(proto
->sendpage
),
2996 proto_method_implemented(proto
->bind
),
2997 proto_method_implemented(proto
->backlog_rcv
),
2998 proto_method_implemented(proto
->hash
),
2999 proto_method_implemented(proto
->unhash
),
3000 proto_method_implemented(proto
->get_port
),
3001 proto_method_implemented(proto
->enter_memory_pressure
));
3004 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3006 if (v
== &proto_list
)
3007 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3016 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3018 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3022 static const struct seq_operations proto_seq_ops
= {
3023 .start
= proto_seq_start
,
3024 .next
= proto_seq_next
,
3025 .stop
= proto_seq_stop
,
3026 .show
= proto_seq_show
,
3029 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
3031 return seq_open_net(inode
, file
, &proto_seq_ops
,
3032 sizeof(struct seq_net_private
));
3035 static const struct file_operations proto_seq_fops
= {
3036 .owner
= THIS_MODULE
,
3037 .open
= proto_seq_open
,
3039 .llseek
= seq_lseek
,
3040 .release
= seq_release_net
,
3043 static __net_init
int proto_init_net(struct net
*net
)
3045 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3051 static __net_exit
void proto_exit_net(struct net
*net
)
3053 remove_proc_entry("protocols", net
->proc_net
);
3057 static __net_initdata
struct pernet_operations proto_net_ops
= {
3058 .init
= proto_init_net
,
3059 .exit
= proto_exit_net
,
3062 static int __init
proto_init(void)
3064 return register_pernet_subsys(&proto_net_ops
);
3067 subsys_initcall(proto_init
);
3069 #endif /* PROC_FS */