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>
138 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
146 static DEFINE_MUTEX(proto_list_mutex
);
147 static LIST_HEAD(proto_list
);
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
159 bool sk_ns_capable(const struct sock
*sk
,
160 struct user_namespace
*user_ns
, int cap
)
162 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
163 ns_capable(user_ns
, cap
);
165 EXPORT_SYMBOL(sk_ns_capable
);
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
176 bool sk_capable(const struct sock
*sk
, int cap
)
178 return sk_ns_capable(sk
, &init_user_ns
, cap
);
180 EXPORT_SYMBOL(sk_capable
);
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
191 bool sk_net_capable(const struct sock
*sk
, int cap
)
193 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
195 EXPORT_SYMBOL(sk_net_capable
);
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
204 mutex_lock(&proto_list_mutex
);
205 list_for_each_entry(proto
, &proto_list
, node
) {
206 if (proto
->init_cgroup
) {
207 ret
= proto
->init_cgroup(memcg
, ss
);
213 mutex_unlock(&proto_list_mutex
);
216 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
217 if (proto
->destroy_cgroup
)
218 proto
->destroy_cgroup(memcg
);
219 mutex_unlock(&proto_list_mutex
);
223 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
227 mutex_lock(&proto_list_mutex
);
228 list_for_each_entry_reverse(proto
, &proto_list
, node
)
229 if (proto
->destroy_cgroup
)
230 proto
->destroy_cgroup(memcg
);
231 mutex_unlock(&proto_list_mutex
);
236 * Each address family might have different locking rules, so we have
237 * one slock key per address family:
239 static struct lock_class_key af_family_keys
[AF_MAX
];
240 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled
;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
248 * Make lock validator output more readable. (we pre-construct these
249 * strings build-time, so that runtime initialization of socket
252 static const char *const af_family_key_strings
[AF_MAX
+1] = {
253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
268 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
278 "slock-27" , "slock-28" , "slock-AF_CAN" ,
279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
284 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
294 "clock-27" , "clock-28" , "clock-AF_CAN" ,
295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
302 * sk_callback_lock locking rules are per-address-family,
303 * so split the lock classes by using a per-AF key:
305 static struct lock_class_key af_callback_keys
[AF_MAX
];
307 /* Take into consideration the size of the struct sk_buff overhead in the
308 * determination of these values, since that is non-constant across
309 * platforms. This makes socket queueing behavior and performance
310 * not depend upon such differences.
312 #define _SK_MEM_PACKETS 256
313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
319 EXPORT_SYMBOL(sysctl_wmem_max
);
320 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
321 EXPORT_SYMBOL(sysctl_rmem_max
);
322 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
323 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
327 EXPORT_SYMBOL(sysctl_optmem_max
);
329 int sysctl_tstamp_allow_data __read_mostly
= 1;
331 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
332 EXPORT_SYMBOL_GPL(memalloc_socks
);
335 * sk_set_memalloc - sets %SOCK_MEMALLOC
336 * @sk: socket to set it on
338 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
339 * It's the responsibility of the admin to adjust min_free_kbytes
340 * to meet the requirements
342 void sk_set_memalloc(struct sock
*sk
)
344 sock_set_flag(sk
, SOCK_MEMALLOC
);
345 sk
->sk_allocation
|= __GFP_MEMALLOC
;
346 static_key_slow_inc(&memalloc_socks
);
348 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
350 void sk_clear_memalloc(struct sock
*sk
)
352 sock_reset_flag(sk
, SOCK_MEMALLOC
);
353 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
354 static_key_slow_dec(&memalloc_socks
);
357 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
358 * progress of swapping. SOCK_MEMALLOC may be cleared while
359 * it has rmem allocations due to the last swapfile being deactivated
360 * but there is a risk that the socket is unusable due to exceeding
361 * the rmem limits. Reclaim the reserves and obey rmem limits again.
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
367 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
370 unsigned long pflags
= current
->flags
;
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
375 current
->flags
|= PF_MEMALLOC
;
376 ret
= sk
->sk_backlog_rcv(sk
, skb
);
377 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
381 EXPORT_SYMBOL(__sk_backlog_rcv
);
383 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
387 if (optlen
< sizeof(tv
))
389 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
391 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
395 static int warned __read_mostly
;
398 if (warned
< 10 && net_ratelimit()) {
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__
, current
->comm
, task_pid_nr(current
));
405 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
406 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
408 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
409 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
413 static void sock_warn_obsolete_bsdism(const char *name
)
416 static char warncomm
[TASK_COMM_LEN
];
417 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
418 strcpy(warncomm
, current
->comm
);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
425 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
427 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
429 if (sk
->sk_flags
& flags
) {
430 sk
->sk_flags
&= ~flags
;
431 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
432 net_disable_timestamp();
437 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
441 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
443 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
444 atomic_inc(&sk
->sk_drops
);
445 trace_sock_rcvqueue_full(sk
, skb
);
449 err
= sk_filter(sk
, skb
);
453 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
454 atomic_inc(&sk
->sk_drops
);
459 skb_set_owner_r(skb
, sk
);
461 /* we escape from rcu protected region, make sure we dont leak
466 spin_lock_irqsave(&list
->lock
, flags
);
467 sock_skb_set_dropcount(sk
, skb
);
468 __skb_queue_tail(list
, skb
);
469 spin_unlock_irqrestore(&list
->lock
, flags
);
471 if (!sock_flag(sk
, SOCK_DEAD
))
472 sk
->sk_data_ready(sk
);
475 EXPORT_SYMBOL(sock_queue_rcv_skb
);
477 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
479 int rc
= NET_RX_SUCCESS
;
481 if (sk_filter(sk
, skb
))
482 goto discard_and_relse
;
486 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
487 atomic_inc(&sk
->sk_drops
);
488 goto discard_and_relse
;
491 bh_lock_sock_nested(sk
);
494 if (!sock_owned_by_user(sk
)) {
496 * trylock + unlock semantics:
498 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
500 rc
= sk_backlog_rcv(sk
, skb
);
502 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
503 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
505 atomic_inc(&sk
->sk_drops
);
506 goto discard_and_relse
;
517 EXPORT_SYMBOL(sk_receive_skb
);
519 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
521 struct dst_entry
*dst
= __sk_dst_get(sk
);
523 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
524 sk_tx_queue_clear(sk
);
525 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
532 EXPORT_SYMBOL(__sk_dst_check
);
534 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
536 struct dst_entry
*dst
= sk_dst_get(sk
);
538 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
546 EXPORT_SYMBOL(sk_dst_check
);
548 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
551 int ret
= -ENOPROTOOPT
;
552 #ifdef CONFIG_NETDEVICES
553 struct net
*net
= sock_net(sk
);
554 char devname
[IFNAMSIZ
];
559 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
566 /* Bind this socket to a particular device like "eth0",
567 * as specified in the passed interface name. If the
568 * name is "" or the option length is zero the socket
571 if (optlen
> IFNAMSIZ
- 1)
572 optlen
= IFNAMSIZ
- 1;
573 memset(devname
, 0, sizeof(devname
));
576 if (copy_from_user(devname
, optval
, optlen
))
580 if (devname
[0] != '\0') {
581 struct net_device
*dev
;
584 dev
= dev_get_by_name_rcu(net
, devname
);
586 index
= dev
->ifindex
;
594 sk
->sk_bound_dev_if
= index
;
606 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
607 int __user
*optlen
, int len
)
609 int ret
= -ENOPROTOOPT
;
610 #ifdef CONFIG_NETDEVICES
611 struct net
*net
= sock_net(sk
);
612 char devname
[IFNAMSIZ
];
614 if (sk
->sk_bound_dev_if
== 0) {
623 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
627 len
= strlen(devname
) + 1;
630 if (copy_to_user(optval
, devname
, len
))
635 if (put_user(len
, optlen
))
646 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
649 sock_set_flag(sk
, bit
);
651 sock_reset_flag(sk
, bit
);
654 bool sk_mc_loop(struct sock
*sk
)
656 if (dev_recursion_level())
660 switch (sk
->sk_family
) {
662 return inet_sk(sk
)->mc_loop
;
663 #if IS_ENABLED(CONFIG_IPV6)
665 return inet6_sk(sk
)->mc_loop
;
671 EXPORT_SYMBOL(sk_mc_loop
);
674 * This is meant for all protocols to use and covers goings on
675 * at the socket level. Everything here is generic.
678 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
679 char __user
*optval
, unsigned int optlen
)
681 struct sock
*sk
= sock
->sk
;
688 * Options without arguments
691 if (optname
== SO_BINDTODEVICE
)
692 return sock_setbindtodevice(sk
, optval
, optlen
);
694 if (optlen
< sizeof(int))
697 if (get_user(val
, (int __user
*)optval
))
700 valbool
= val
? 1 : 0;
706 if (val
&& !capable(CAP_NET_ADMIN
))
709 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
712 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
715 sk
->sk_reuseport
= valbool
;
724 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
727 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
730 /* Don't error on this BSD doesn't and if you think
731 * about it this is right. Otherwise apps have to
732 * play 'guess the biggest size' games. RCVBUF/SNDBUF
733 * are treated in BSD as hints
735 val
= min_t(u32
, val
, sysctl_wmem_max
);
737 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
738 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
739 /* Wake up sending tasks if we upped the value. */
740 sk
->sk_write_space(sk
);
744 if (!capable(CAP_NET_ADMIN
)) {
751 /* Don't error on this BSD doesn't and if you think
752 * about it this is right. Otherwise apps have to
753 * play 'guess the biggest size' games. RCVBUF/SNDBUF
754 * are treated in BSD as hints
756 val
= min_t(u32
, val
, sysctl_rmem_max
);
758 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
760 * We double it on the way in to account for
761 * "struct sk_buff" etc. overhead. Applications
762 * assume that the SO_RCVBUF setting they make will
763 * allow that much actual data to be received on that
766 * Applications are unaware that "struct sk_buff" and
767 * other overheads allocate from the receive buffer
768 * during socket buffer allocation.
770 * And after considering the possible alternatives,
771 * returning the value we actually used in getsockopt
772 * is the most desirable behavior.
774 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
778 if (!capable(CAP_NET_ADMIN
)) {
786 if (sk
->sk_protocol
== IPPROTO_TCP
&&
787 sk
->sk_type
== SOCK_STREAM
)
788 tcp_set_keepalive(sk
, valbool
);
790 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
794 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
798 sk
->sk_no_check_tx
= valbool
;
802 if ((val
>= 0 && val
<= 6) ||
803 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
804 sk
->sk_priority
= val
;
810 if (optlen
< sizeof(ling
)) {
811 ret
= -EINVAL
; /* 1003.1g */
814 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
819 sock_reset_flag(sk
, SOCK_LINGER
);
821 #if (BITS_PER_LONG == 32)
822 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
823 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
826 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
827 sock_set_flag(sk
, SOCK_LINGER
);
832 sock_warn_obsolete_bsdism("setsockopt");
837 set_bit(SOCK_PASSCRED
, &sock
->flags
);
839 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
845 if (optname
== SO_TIMESTAMP
)
846 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
848 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
849 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
850 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
852 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
853 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
857 case SO_TIMESTAMPING
:
858 if (val
& ~SOF_TIMESTAMPING_MASK
) {
863 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
864 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
865 if (sk
->sk_protocol
== IPPROTO_TCP
) {
866 if (sk
->sk_state
!= TCP_ESTABLISHED
) {
870 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
875 sk
->sk_tsflags
= val
;
876 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
877 sock_enable_timestamp(sk
,
878 SOCK_TIMESTAMPING_RX_SOFTWARE
);
880 sock_disable_timestamp(sk
,
881 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
887 sk
->sk_rcvlowat
= val
? : 1;
891 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
895 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
898 case SO_ATTACH_FILTER
:
900 if (optlen
== sizeof(struct sock_fprog
)) {
901 struct sock_fprog fprog
;
904 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
907 ret
= sk_attach_filter(&fprog
, sk
);
913 if (optlen
== sizeof(u32
)) {
917 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
920 ret
= sk_attach_bpf(ufd
, sk
);
924 case SO_DETACH_FILTER
:
925 ret
= sk_detach_filter(sk
);
929 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
932 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
937 set_bit(SOCK_PASSSEC
, &sock
->flags
);
939 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
942 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
949 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
953 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
957 if (sock
->ops
->set_peek_off
)
958 ret
= sock
->ops
->set_peek_off(sk
, val
);
964 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
967 case SO_SELECT_ERR_QUEUE
:
968 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
971 #ifdef CONFIG_NET_RX_BUSY_POLL
973 /* allow unprivileged users to decrease the value */
974 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
980 sk
->sk_ll_usec
= val
;
985 case SO_MAX_PACING_RATE
:
986 sk
->sk_max_pacing_rate
= val
;
987 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
988 sk
->sk_max_pacing_rate
);
998 EXPORT_SYMBOL(sock_setsockopt
);
1001 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1002 struct ucred
*ucred
)
1004 ucred
->pid
= pid_vnr(pid
);
1005 ucred
->uid
= ucred
->gid
= -1;
1007 struct user_namespace
*current_ns
= current_user_ns();
1009 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1010 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1014 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1015 char __user
*optval
, int __user
*optlen
)
1017 struct sock
*sk
= sock
->sk
;
1025 int lv
= sizeof(int);
1028 if (get_user(len
, optlen
))
1033 memset(&v
, 0, sizeof(v
));
1037 v
.val
= sock_flag(sk
, SOCK_DBG
);
1041 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1045 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1049 v
.val
= sk
->sk_sndbuf
;
1053 v
.val
= sk
->sk_rcvbuf
;
1057 v
.val
= sk
->sk_reuse
;
1061 v
.val
= sk
->sk_reuseport
;
1065 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1069 v
.val
= sk
->sk_type
;
1073 v
.val
= sk
->sk_protocol
;
1077 v
.val
= sk
->sk_family
;
1081 v
.val
= -sock_error(sk
);
1083 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1087 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1091 v
.val
= sk
->sk_no_check_tx
;
1095 v
.val
= sk
->sk_priority
;
1099 lv
= sizeof(v
.ling
);
1100 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1101 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1105 sock_warn_obsolete_bsdism("getsockopt");
1109 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1110 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1113 case SO_TIMESTAMPNS
:
1114 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1117 case SO_TIMESTAMPING
:
1118 v
.val
= sk
->sk_tsflags
;
1122 lv
= sizeof(struct timeval
);
1123 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1127 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1128 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1133 lv
= sizeof(struct timeval
);
1134 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1138 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1139 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1144 v
.val
= sk
->sk_rcvlowat
;
1152 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1157 struct ucred peercred
;
1158 if (len
> sizeof(peercred
))
1159 len
= sizeof(peercred
);
1160 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1161 if (copy_to_user(optval
, &peercred
, len
))
1170 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1174 if (copy_to_user(optval
, address
, len
))
1179 /* Dubious BSD thing... Probably nobody even uses it, but
1180 * the UNIX standard wants it for whatever reason... -DaveM
1183 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1187 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1191 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1194 v
.val
= sk
->sk_mark
;
1198 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1201 case SO_WIFI_STATUS
:
1202 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1206 if (!sock
->ops
->set_peek_off
)
1209 v
.val
= sk
->sk_peek_off
;
1212 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1215 case SO_BINDTODEVICE
:
1216 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1219 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1225 case SO_LOCK_FILTER
:
1226 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1229 case SO_BPF_EXTENSIONS
:
1230 v
.val
= bpf_tell_extensions();
1233 case SO_SELECT_ERR_QUEUE
:
1234 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1237 #ifdef CONFIG_NET_RX_BUSY_POLL
1239 v
.val
= sk
->sk_ll_usec
;
1243 case SO_MAX_PACING_RATE
:
1244 v
.val
= sk
->sk_max_pacing_rate
;
1247 case SO_INCOMING_CPU
:
1248 v
.val
= sk
->sk_incoming_cpu
;
1252 /* We implement the SO_SNDLOWAT etc to not be settable
1255 return -ENOPROTOOPT
;
1260 if (copy_to_user(optval
, &v
, len
))
1263 if (put_user(len
, optlen
))
1269 * Initialize an sk_lock.
1271 * (We also register the sk_lock with the lock validator.)
1273 static inline void sock_lock_init(struct sock
*sk
)
1275 sock_lock_init_class_and_name(sk
,
1276 af_family_slock_key_strings
[sk
->sk_family
],
1277 af_family_slock_keys
+ sk
->sk_family
,
1278 af_family_key_strings
[sk
->sk_family
],
1279 af_family_keys
+ sk
->sk_family
);
1283 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1284 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1285 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1287 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1289 #ifdef CONFIG_SECURITY_NETWORK
1290 void *sptr
= nsk
->sk_security
;
1292 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1294 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1295 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1297 #ifdef CONFIG_SECURITY_NETWORK
1298 nsk
->sk_security
= sptr
;
1299 security_sk_clone(osk
, nsk
);
1303 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1305 unsigned long nulls1
, nulls2
;
1307 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1308 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1309 if (nulls1
> nulls2
)
1310 swap(nulls1
, nulls2
);
1313 memset((char *)sk
, 0, nulls1
);
1314 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1315 nulls2
- nulls1
- sizeof(void *));
1316 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1317 size
- nulls2
- sizeof(void *));
1319 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1321 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1325 struct kmem_cache
*slab
;
1329 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1332 if (priority
& __GFP_ZERO
) {
1334 prot
->clear_sk(sk
, prot
->obj_size
);
1336 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1339 sk
= kmalloc(prot
->obj_size
, priority
);
1342 kmemcheck_annotate_bitfield(sk
, flags
);
1344 if (security_sk_alloc(sk
, family
, priority
))
1347 if (!try_module_get(prot
->owner
))
1349 sk_tx_queue_clear(sk
);
1355 security_sk_free(sk
);
1358 kmem_cache_free(slab
, sk
);
1364 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1366 struct kmem_cache
*slab
;
1367 struct module
*owner
;
1369 owner
= prot
->owner
;
1372 security_sk_free(sk
);
1374 kmem_cache_free(slab
, sk
);
1380 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1381 void sock_update_netprioidx(struct sock
*sk
)
1386 sk
->sk_cgrp_prioidx
= task_netprioidx(current
);
1388 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1392 * sk_alloc - All socket objects are allocated here
1393 * @net: the applicable net namespace
1394 * @family: protocol family
1395 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1396 * @prot: struct proto associated with this new sock instance
1397 * @kern: is this to be a kernel socket?
1399 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1400 struct proto
*prot
, int kern
)
1404 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1406 sk
->sk_family
= family
;
1408 * See comment in struct sock definition to understand
1409 * why we need sk_prot_creator -acme
1411 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1413 sk
->sk_net_refcnt
= kern
? 0 : 1;
1414 if (likely(sk
->sk_net_refcnt
))
1416 sock_net_set(sk
, net
);
1417 atomic_set(&sk
->sk_wmem_alloc
, 1);
1419 sock_update_classid(sk
);
1420 sock_update_netprioidx(sk
);
1425 EXPORT_SYMBOL(sk_alloc
);
1427 void sk_destruct(struct sock
*sk
)
1429 struct sk_filter
*filter
;
1431 if (sk
->sk_destruct
)
1432 sk
->sk_destruct(sk
);
1434 filter
= rcu_dereference_check(sk
->sk_filter
,
1435 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1437 sk_filter_uncharge(sk
, filter
);
1438 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1441 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1443 if (atomic_read(&sk
->sk_omem_alloc
))
1444 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1445 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1447 if (sk
->sk_peer_cred
)
1448 put_cred(sk
->sk_peer_cred
);
1449 put_pid(sk
->sk_peer_pid
);
1450 if (likely(sk
->sk_net_refcnt
))
1451 put_net(sock_net(sk
));
1452 sk_prot_free(sk
->sk_prot_creator
, sk
);
1455 static void __sk_free(struct sock
*sk
)
1457 if (unlikely(sock_diag_has_destroy_listeners(sk
) && sk
->sk_net_refcnt
))
1458 sock_diag_broadcast_destroy(sk
);
1463 void sk_free(struct sock
*sk
)
1466 * We subtract one from sk_wmem_alloc and can know if
1467 * some packets are still in some tx queue.
1468 * If not null, sock_wfree() will call __sk_free(sk) later
1470 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1473 EXPORT_SYMBOL(sk_free
);
1475 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1477 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1478 sock_update_memcg(newsk
);
1482 * sk_clone_lock - clone a socket, and lock its clone
1483 * @sk: the socket to clone
1484 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1486 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1488 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1491 bool is_charged
= true;
1493 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1494 if (newsk
!= NULL
) {
1495 struct sk_filter
*filter
;
1497 sock_copy(newsk
, sk
);
1500 get_net(sock_net(newsk
));
1501 sk_node_init(&newsk
->sk_node
);
1502 sock_lock_init(newsk
);
1503 bh_lock_sock(newsk
);
1504 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1505 newsk
->sk_backlog
.len
= 0;
1507 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1509 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1511 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1512 atomic_set(&newsk
->sk_omem_alloc
, 0);
1513 skb_queue_head_init(&newsk
->sk_receive_queue
);
1514 skb_queue_head_init(&newsk
->sk_write_queue
);
1516 spin_lock_init(&newsk
->sk_dst_lock
);
1517 rwlock_init(&newsk
->sk_callback_lock
);
1518 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1519 af_callback_keys
+ newsk
->sk_family
,
1520 af_family_clock_key_strings
[newsk
->sk_family
]);
1522 newsk
->sk_dst_cache
= NULL
;
1523 newsk
->sk_wmem_queued
= 0;
1524 newsk
->sk_forward_alloc
= 0;
1525 newsk
->sk_send_head
= NULL
;
1526 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1528 sock_reset_flag(newsk
, SOCK_DONE
);
1529 skb_queue_head_init(&newsk
->sk_error_queue
);
1531 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1533 /* though it's an empty new sock, the charging may fail
1534 * if sysctl_optmem_max was changed between creation of
1535 * original socket and cloning
1537 is_charged
= sk_filter_charge(newsk
, filter
);
1539 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
))) {
1540 /* It is still raw copy of parent, so invalidate
1541 * destructor and make plain sk_free() */
1542 newsk
->sk_destruct
= NULL
;
1543 bh_unlock_sock(newsk
);
1550 newsk
->sk_priority
= 0;
1551 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1552 atomic64_set(&newsk
->sk_cookie
, 0);
1554 * Before updating sk_refcnt, we must commit prior changes to memory
1555 * (Documentation/RCU/rculist_nulls.txt for details)
1558 atomic_set(&newsk
->sk_refcnt
, 2);
1561 * Increment the counter in the same struct proto as the master
1562 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1563 * is the same as sk->sk_prot->socks, as this field was copied
1566 * This _changes_ the previous behaviour, where
1567 * tcp_create_openreq_child always was incrementing the
1568 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1569 * to be taken into account in all callers. -acme
1571 sk_refcnt_debug_inc(newsk
);
1572 sk_set_socket(newsk
, NULL
);
1573 newsk
->sk_wq
= NULL
;
1575 sk_update_clone(sk
, newsk
);
1577 if (newsk
->sk_prot
->sockets_allocated
)
1578 sk_sockets_allocated_inc(newsk
);
1580 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1581 net_enable_timestamp();
1586 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1588 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1592 __sk_dst_set(sk
, dst
);
1593 sk
->sk_route_caps
= dst
->dev
->features
;
1594 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1595 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1596 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1597 if (sk_can_gso(sk
)) {
1598 if (dst
->header_len
) {
1599 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1601 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1602 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1603 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1606 sk
->sk_gso_max_segs
= max_segs
;
1608 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1611 * Simple resource managers for sockets.
1616 * Write buffer destructor automatically called from kfree_skb.
1618 void sock_wfree(struct sk_buff
*skb
)
1620 struct sock
*sk
= skb
->sk
;
1621 unsigned int len
= skb
->truesize
;
1623 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1625 * Keep a reference on sk_wmem_alloc, this will be released
1626 * after sk_write_space() call
1628 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1629 sk
->sk_write_space(sk
);
1633 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1634 * could not do because of in-flight packets
1636 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1639 EXPORT_SYMBOL(sock_wfree
);
1641 void skb_orphan_partial(struct sk_buff
*skb
)
1643 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1644 * so we do not completely orphan skb, but transfert all
1645 * accounted bytes but one, to avoid unexpected reorders.
1647 if (skb
->destructor
== sock_wfree
1649 || skb
->destructor
== tcp_wfree
1652 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1658 EXPORT_SYMBOL(skb_orphan_partial
);
1661 * Read buffer destructor automatically called from kfree_skb.
1663 void sock_rfree(struct sk_buff
*skb
)
1665 struct sock
*sk
= skb
->sk
;
1666 unsigned int len
= skb
->truesize
;
1668 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1669 sk_mem_uncharge(sk
, len
);
1671 EXPORT_SYMBOL(sock_rfree
);
1674 * Buffer destructor for skbs that are not used directly in read or write
1675 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1677 void sock_efree(struct sk_buff
*skb
)
1681 EXPORT_SYMBOL(sock_efree
);
1683 kuid_t
sock_i_uid(struct sock
*sk
)
1687 read_lock_bh(&sk
->sk_callback_lock
);
1688 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1689 read_unlock_bh(&sk
->sk_callback_lock
);
1692 EXPORT_SYMBOL(sock_i_uid
);
1694 unsigned long sock_i_ino(struct sock
*sk
)
1698 read_lock_bh(&sk
->sk_callback_lock
);
1699 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1700 read_unlock_bh(&sk
->sk_callback_lock
);
1703 EXPORT_SYMBOL(sock_i_ino
);
1706 * Allocate a skb from the socket's send buffer.
1708 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1711 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1712 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1714 skb_set_owner_w(skb
, sk
);
1720 EXPORT_SYMBOL(sock_wmalloc
);
1723 * Allocate a memory block from the socket's option memory buffer.
1725 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1727 if ((unsigned int)size
<= sysctl_optmem_max
&&
1728 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1730 /* First do the add, to avoid the race if kmalloc
1733 atomic_add(size
, &sk
->sk_omem_alloc
);
1734 mem
= kmalloc(size
, priority
);
1737 atomic_sub(size
, &sk
->sk_omem_alloc
);
1741 EXPORT_SYMBOL(sock_kmalloc
);
1743 /* Free an option memory block. Note, we actually want the inline
1744 * here as this allows gcc to detect the nullify and fold away the
1745 * condition entirely.
1747 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1750 if (WARN_ON_ONCE(!mem
))
1756 atomic_sub(size
, &sk
->sk_omem_alloc
);
1759 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1761 __sock_kfree_s(sk
, mem
, size
, false);
1763 EXPORT_SYMBOL(sock_kfree_s
);
1765 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1767 __sock_kfree_s(sk
, mem
, size
, true);
1769 EXPORT_SYMBOL(sock_kzfree_s
);
1771 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1772 I think, these locks should be removed for datagram sockets.
1774 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1778 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1782 if (signal_pending(current
))
1784 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1785 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1786 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1788 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1792 timeo
= schedule_timeout(timeo
);
1794 finish_wait(sk_sleep(sk
), &wait
);
1800 * Generic send/receive buffer handlers
1803 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1804 unsigned long data_len
, int noblock
,
1805 int *errcode
, int max_page_order
)
1807 struct sk_buff
*skb
;
1811 timeo
= sock_sndtimeo(sk
, noblock
);
1813 err
= sock_error(sk
);
1818 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1821 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1824 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1825 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1829 if (signal_pending(current
))
1831 timeo
= sock_wait_for_wmem(sk
, timeo
);
1833 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
1834 errcode
, sk
->sk_allocation
);
1836 skb_set_owner_w(skb
, sk
);
1840 err
= sock_intr_errno(timeo
);
1845 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1847 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1848 int noblock
, int *errcode
)
1850 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1852 EXPORT_SYMBOL(sock_alloc_send_skb
);
1854 /* On 32bit arches, an skb frag is limited to 2^15 */
1855 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1858 * skb_page_frag_refill - check that a page_frag contains enough room
1859 * @sz: minimum size of the fragment we want to get
1860 * @pfrag: pointer to page_frag
1861 * @gfp: priority for memory allocation
1863 * Note: While this allocator tries to use high order pages, there is
1864 * no guarantee that allocations succeed. Therefore, @sz MUST be
1865 * less or equal than PAGE_SIZE.
1867 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
1870 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1874 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1876 put_page(pfrag
->page
);
1880 if (SKB_FRAG_PAGE_ORDER
) {
1881 pfrag
->page
= alloc_pages((gfp
& ~__GFP_WAIT
) | __GFP_COMP
|
1882 __GFP_NOWARN
| __GFP_NORETRY
,
1883 SKB_FRAG_PAGE_ORDER
);
1884 if (likely(pfrag
->page
)) {
1885 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
1889 pfrag
->page
= alloc_page(gfp
);
1890 if (likely(pfrag
->page
)) {
1891 pfrag
->size
= PAGE_SIZE
;
1896 EXPORT_SYMBOL(skb_page_frag_refill
);
1898 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1900 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
1903 sk_enter_memory_pressure(sk
);
1904 sk_stream_moderate_sndbuf(sk
);
1907 EXPORT_SYMBOL(sk_page_frag_refill
);
1909 static void __lock_sock(struct sock
*sk
)
1910 __releases(&sk
->sk_lock
.slock
)
1911 __acquires(&sk
->sk_lock
.slock
)
1916 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1917 TASK_UNINTERRUPTIBLE
);
1918 spin_unlock_bh(&sk
->sk_lock
.slock
);
1920 spin_lock_bh(&sk
->sk_lock
.slock
);
1921 if (!sock_owned_by_user(sk
))
1924 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1927 static void __release_sock(struct sock
*sk
)
1928 __releases(&sk
->sk_lock
.slock
)
1929 __acquires(&sk
->sk_lock
.slock
)
1931 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1934 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1938 struct sk_buff
*next
= skb
->next
;
1941 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1943 sk_backlog_rcv(sk
, skb
);
1946 * We are in process context here with softirqs
1947 * disabled, use cond_resched_softirq() to preempt.
1948 * This is safe to do because we've taken the backlog
1951 cond_resched_softirq();
1954 } while (skb
!= NULL
);
1957 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1960 * Doing the zeroing here guarantee we can not loop forever
1961 * while a wild producer attempts to flood us.
1963 sk
->sk_backlog
.len
= 0;
1967 * sk_wait_data - wait for data to arrive at sk_receive_queue
1968 * @sk: sock to wait on
1969 * @timeo: for how long
1971 * Now socket state including sk->sk_err is changed only under lock,
1972 * hence we may omit checks after joining wait queue.
1973 * We check receive queue before schedule() only as optimization;
1974 * it is very likely that release_sock() added new data.
1976 int sk_wait_data(struct sock
*sk
, long *timeo
)
1981 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1982 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1983 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1984 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1985 finish_wait(sk_sleep(sk
), &wait
);
1988 EXPORT_SYMBOL(sk_wait_data
);
1991 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1993 * @size: memory size to allocate
1994 * @kind: allocation type
1996 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1997 * rmem allocation. This function assumes that protocols which have
1998 * memory_pressure use sk_wmem_queued as write buffer accounting.
2000 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2002 struct proto
*prot
= sk
->sk_prot
;
2003 int amt
= sk_mem_pages(size
);
2005 int parent_status
= UNDER_LIMIT
;
2007 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2009 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
2012 if (parent_status
== UNDER_LIMIT
&&
2013 allocated
<= sk_prot_mem_limits(sk
, 0)) {
2014 sk_leave_memory_pressure(sk
);
2018 /* Under pressure. (we or our parents) */
2019 if ((parent_status
> SOFT_LIMIT
) ||
2020 allocated
> sk_prot_mem_limits(sk
, 1))
2021 sk_enter_memory_pressure(sk
);
2023 /* Over hard limit (we or our parents) */
2024 if ((parent_status
== OVER_LIMIT
) ||
2025 (allocated
> sk_prot_mem_limits(sk
, 2)))
2026 goto suppress_allocation
;
2028 /* guarantee minimum buffer size under pressure */
2029 if (kind
== SK_MEM_RECV
) {
2030 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2033 } else { /* SK_MEM_SEND */
2034 if (sk
->sk_type
== SOCK_STREAM
) {
2035 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2037 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2038 prot
->sysctl_wmem
[0])
2042 if (sk_has_memory_pressure(sk
)) {
2045 if (!sk_under_memory_pressure(sk
))
2047 alloc
= sk_sockets_allocated_read_positive(sk
);
2048 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2049 sk_mem_pages(sk
->sk_wmem_queued
+
2050 atomic_read(&sk
->sk_rmem_alloc
) +
2051 sk
->sk_forward_alloc
))
2055 suppress_allocation
:
2057 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2058 sk_stream_moderate_sndbuf(sk
);
2060 /* Fail only if socket is _under_ its sndbuf.
2061 * In this case we cannot block, so that we have to fail.
2063 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2067 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2069 /* Alas. Undo changes. */
2070 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2072 sk_memory_allocated_sub(sk
, amt
);
2076 EXPORT_SYMBOL(__sk_mem_schedule
);
2079 * __sk_reclaim - reclaim memory_allocated
2081 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2083 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2085 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2086 sk_memory_allocated_sub(sk
, amount
);
2087 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2089 if (sk_under_memory_pressure(sk
) &&
2090 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2091 sk_leave_memory_pressure(sk
);
2093 EXPORT_SYMBOL(__sk_mem_reclaim
);
2097 * Set of default routines for initialising struct proto_ops when
2098 * the protocol does not support a particular function. In certain
2099 * cases where it makes no sense for a protocol to have a "do nothing"
2100 * function, some default processing is provided.
2103 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2107 EXPORT_SYMBOL(sock_no_bind
);
2109 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2114 EXPORT_SYMBOL(sock_no_connect
);
2116 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2120 EXPORT_SYMBOL(sock_no_socketpair
);
2122 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2126 EXPORT_SYMBOL(sock_no_accept
);
2128 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2133 EXPORT_SYMBOL(sock_no_getname
);
2135 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2139 EXPORT_SYMBOL(sock_no_poll
);
2141 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2145 EXPORT_SYMBOL(sock_no_ioctl
);
2147 int sock_no_listen(struct socket
*sock
, int backlog
)
2151 EXPORT_SYMBOL(sock_no_listen
);
2153 int sock_no_shutdown(struct socket
*sock
, int how
)
2157 EXPORT_SYMBOL(sock_no_shutdown
);
2159 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2160 char __user
*optval
, unsigned int optlen
)
2164 EXPORT_SYMBOL(sock_no_setsockopt
);
2166 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2167 char __user
*optval
, int __user
*optlen
)
2171 EXPORT_SYMBOL(sock_no_getsockopt
);
2173 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2177 EXPORT_SYMBOL(sock_no_sendmsg
);
2179 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2184 EXPORT_SYMBOL(sock_no_recvmsg
);
2186 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2188 /* Mirror missing mmap method error code */
2191 EXPORT_SYMBOL(sock_no_mmap
);
2193 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2196 struct msghdr msg
= {.msg_flags
= flags
};
2198 char *kaddr
= kmap(page
);
2199 iov
.iov_base
= kaddr
+ offset
;
2201 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2205 EXPORT_SYMBOL(sock_no_sendpage
);
2208 * Default Socket Callbacks
2211 static void sock_def_wakeup(struct sock
*sk
)
2213 struct socket_wq
*wq
;
2216 wq
= rcu_dereference(sk
->sk_wq
);
2217 if (wq_has_sleeper(wq
))
2218 wake_up_interruptible_all(&wq
->wait
);
2222 static void sock_def_error_report(struct sock
*sk
)
2224 struct socket_wq
*wq
;
2227 wq
= rcu_dereference(sk
->sk_wq
);
2228 if (wq_has_sleeper(wq
))
2229 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2230 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2234 static void sock_def_readable(struct sock
*sk
)
2236 struct socket_wq
*wq
;
2239 wq
= rcu_dereference(sk
->sk_wq
);
2240 if (wq_has_sleeper(wq
))
2241 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2242 POLLRDNORM
| POLLRDBAND
);
2243 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2247 static void sock_def_write_space(struct sock
*sk
)
2249 struct socket_wq
*wq
;
2253 /* Do not wake up a writer until he can make "significant"
2256 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2257 wq
= rcu_dereference(sk
->sk_wq
);
2258 if (wq_has_sleeper(wq
))
2259 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2260 POLLWRNORM
| POLLWRBAND
);
2262 /* Should agree with poll, otherwise some programs break */
2263 if (sock_writeable(sk
))
2264 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2270 static void sock_def_destruct(struct sock
*sk
)
2274 void sk_send_sigurg(struct sock
*sk
)
2276 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2277 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2278 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2280 EXPORT_SYMBOL(sk_send_sigurg
);
2282 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2283 unsigned long expires
)
2285 if (!mod_timer(timer
, expires
))
2288 EXPORT_SYMBOL(sk_reset_timer
);
2290 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2292 if (del_timer(timer
))
2295 EXPORT_SYMBOL(sk_stop_timer
);
2297 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2299 skb_queue_head_init(&sk
->sk_receive_queue
);
2300 skb_queue_head_init(&sk
->sk_write_queue
);
2301 skb_queue_head_init(&sk
->sk_error_queue
);
2303 sk
->sk_send_head
= NULL
;
2305 init_timer(&sk
->sk_timer
);
2307 sk
->sk_allocation
= GFP_KERNEL
;
2308 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2309 sk
->sk_sndbuf
= sysctl_wmem_default
;
2310 sk
->sk_state
= TCP_CLOSE
;
2311 sk_set_socket(sk
, sock
);
2313 sock_set_flag(sk
, SOCK_ZAPPED
);
2316 sk
->sk_type
= sock
->type
;
2317 sk
->sk_wq
= sock
->wq
;
2322 spin_lock_init(&sk
->sk_dst_lock
);
2323 rwlock_init(&sk
->sk_callback_lock
);
2324 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2325 af_callback_keys
+ sk
->sk_family
,
2326 af_family_clock_key_strings
[sk
->sk_family
]);
2328 sk
->sk_state_change
= sock_def_wakeup
;
2329 sk
->sk_data_ready
= sock_def_readable
;
2330 sk
->sk_write_space
= sock_def_write_space
;
2331 sk
->sk_error_report
= sock_def_error_report
;
2332 sk
->sk_destruct
= sock_def_destruct
;
2334 sk
->sk_frag
.page
= NULL
;
2335 sk
->sk_frag
.offset
= 0;
2336 sk
->sk_peek_off
= -1;
2338 sk
->sk_peer_pid
= NULL
;
2339 sk
->sk_peer_cred
= NULL
;
2340 sk
->sk_write_pending
= 0;
2341 sk
->sk_rcvlowat
= 1;
2342 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2343 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2345 sk
->sk_stamp
= ktime_set(-1L, 0);
2347 #ifdef CONFIG_NET_RX_BUSY_POLL
2349 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2352 sk
->sk_max_pacing_rate
= ~0U;
2353 sk
->sk_pacing_rate
= ~0U;
2355 * Before updating sk_refcnt, we must commit prior changes to memory
2356 * (Documentation/RCU/rculist_nulls.txt for details)
2359 atomic_set(&sk
->sk_refcnt
, 1);
2360 atomic_set(&sk
->sk_drops
, 0);
2362 EXPORT_SYMBOL(sock_init_data
);
2364 void lock_sock_nested(struct sock
*sk
, int subclass
)
2367 spin_lock_bh(&sk
->sk_lock
.slock
);
2368 if (sk
->sk_lock
.owned
)
2370 sk
->sk_lock
.owned
= 1;
2371 spin_unlock(&sk
->sk_lock
.slock
);
2373 * The sk_lock has mutex_lock() semantics here:
2375 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2378 EXPORT_SYMBOL(lock_sock_nested
);
2380 void release_sock(struct sock
*sk
)
2383 * The sk_lock has mutex_unlock() semantics:
2385 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2387 spin_lock_bh(&sk
->sk_lock
.slock
);
2388 if (sk
->sk_backlog
.tail
)
2391 /* Warning : release_cb() might need to release sk ownership,
2392 * ie call sock_release_ownership(sk) before us.
2394 if (sk
->sk_prot
->release_cb
)
2395 sk
->sk_prot
->release_cb(sk
);
2397 sock_release_ownership(sk
);
2398 if (waitqueue_active(&sk
->sk_lock
.wq
))
2399 wake_up(&sk
->sk_lock
.wq
);
2400 spin_unlock_bh(&sk
->sk_lock
.slock
);
2402 EXPORT_SYMBOL(release_sock
);
2405 * lock_sock_fast - fast version of lock_sock
2408 * This version should be used for very small section, where process wont block
2409 * return false if fast path is taken
2410 * sk_lock.slock locked, owned = 0, BH disabled
2411 * return true if slow path is taken
2412 * sk_lock.slock unlocked, owned = 1, BH enabled
2414 bool lock_sock_fast(struct sock
*sk
)
2417 spin_lock_bh(&sk
->sk_lock
.slock
);
2419 if (!sk
->sk_lock
.owned
)
2421 * Note : We must disable BH
2426 sk
->sk_lock
.owned
= 1;
2427 spin_unlock(&sk
->sk_lock
.slock
);
2429 * The sk_lock has mutex_lock() semantics here:
2431 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2435 EXPORT_SYMBOL(lock_sock_fast
);
2437 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2440 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2441 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2442 tv
= ktime_to_timeval(sk
->sk_stamp
);
2443 if (tv
.tv_sec
== -1)
2445 if (tv
.tv_sec
== 0) {
2446 sk
->sk_stamp
= ktime_get_real();
2447 tv
= ktime_to_timeval(sk
->sk_stamp
);
2449 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2451 EXPORT_SYMBOL(sock_get_timestamp
);
2453 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2456 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2457 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2458 ts
= ktime_to_timespec(sk
->sk_stamp
);
2459 if (ts
.tv_sec
== -1)
2461 if (ts
.tv_sec
== 0) {
2462 sk
->sk_stamp
= ktime_get_real();
2463 ts
= ktime_to_timespec(sk
->sk_stamp
);
2465 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2467 EXPORT_SYMBOL(sock_get_timestampns
);
2469 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2471 if (!sock_flag(sk
, flag
)) {
2472 unsigned long previous_flags
= sk
->sk_flags
;
2474 sock_set_flag(sk
, flag
);
2476 * we just set one of the two flags which require net
2477 * time stamping, but time stamping might have been on
2478 * already because of the other one
2480 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2481 net_enable_timestamp();
2485 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2486 int level
, int type
)
2488 struct sock_exterr_skb
*serr
;
2489 struct sk_buff
*skb
;
2493 skb
= sock_dequeue_err_skb(sk
);
2499 msg
->msg_flags
|= MSG_TRUNC
;
2502 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2506 sock_recv_timestamp(msg
, sk
, skb
);
2508 serr
= SKB_EXT_ERR(skb
);
2509 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2511 msg
->msg_flags
|= MSG_ERRQUEUE
;
2519 EXPORT_SYMBOL(sock_recv_errqueue
);
2522 * Get a socket option on an socket.
2524 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2525 * asynchronous errors should be reported by getsockopt. We assume
2526 * this means if you specify SO_ERROR (otherwise whats the point of it).
2528 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2529 char __user
*optval
, int __user
*optlen
)
2531 struct sock
*sk
= sock
->sk
;
2533 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2535 EXPORT_SYMBOL(sock_common_getsockopt
);
2537 #ifdef CONFIG_COMPAT
2538 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2539 char __user
*optval
, int __user
*optlen
)
2541 struct sock
*sk
= sock
->sk
;
2543 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2544 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2546 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2548 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2551 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2554 struct sock
*sk
= sock
->sk
;
2558 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2559 flags
& ~MSG_DONTWAIT
, &addr_len
);
2561 msg
->msg_namelen
= addr_len
;
2564 EXPORT_SYMBOL(sock_common_recvmsg
);
2567 * Set socket options on an inet socket.
2569 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2570 char __user
*optval
, unsigned int optlen
)
2572 struct sock
*sk
= sock
->sk
;
2574 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2576 EXPORT_SYMBOL(sock_common_setsockopt
);
2578 #ifdef CONFIG_COMPAT
2579 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2580 char __user
*optval
, unsigned int optlen
)
2582 struct sock
*sk
= sock
->sk
;
2584 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2585 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2587 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2589 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2592 void sk_common_release(struct sock
*sk
)
2594 if (sk
->sk_prot
->destroy
)
2595 sk
->sk_prot
->destroy(sk
);
2598 * Observation: when sock_common_release is called, processes have
2599 * no access to socket. But net still has.
2600 * Step one, detach it from networking:
2602 * A. Remove from hash tables.
2605 sk
->sk_prot
->unhash(sk
);
2608 * In this point socket cannot receive new packets, but it is possible
2609 * that some packets are in flight because some CPU runs receiver and
2610 * did hash table lookup before we unhashed socket. They will achieve
2611 * receive queue and will be purged by socket destructor.
2613 * Also we still have packets pending on receive queue and probably,
2614 * our own packets waiting in device queues. sock_destroy will drain
2615 * receive queue, but transmitted packets will delay socket destruction
2616 * until the last reference will be released.
2621 xfrm_sk_free_policy(sk
);
2623 sk_refcnt_debug_release(sk
);
2625 if (sk
->sk_frag
.page
) {
2626 put_page(sk
->sk_frag
.page
);
2627 sk
->sk_frag
.page
= NULL
;
2632 EXPORT_SYMBOL(sk_common_release
);
2634 #ifdef CONFIG_PROC_FS
2635 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2637 int val
[PROTO_INUSE_NR
];
2640 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2642 #ifdef CONFIG_NET_NS
2643 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2645 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2647 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2649 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2651 int cpu
, idx
= prot
->inuse_idx
;
2654 for_each_possible_cpu(cpu
)
2655 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2657 return res
>= 0 ? res
: 0;
2659 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2661 static int __net_init
sock_inuse_init_net(struct net
*net
)
2663 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2664 return net
->core
.inuse
? 0 : -ENOMEM
;
2667 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2669 free_percpu(net
->core
.inuse
);
2672 static struct pernet_operations net_inuse_ops
= {
2673 .init
= sock_inuse_init_net
,
2674 .exit
= sock_inuse_exit_net
,
2677 static __init
int net_inuse_init(void)
2679 if (register_pernet_subsys(&net_inuse_ops
))
2680 panic("Cannot initialize net inuse counters");
2685 core_initcall(net_inuse_init
);
2687 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2689 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2691 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2693 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2695 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2697 int cpu
, idx
= prot
->inuse_idx
;
2700 for_each_possible_cpu(cpu
)
2701 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2703 return res
>= 0 ? res
: 0;
2705 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2708 static void assign_proto_idx(struct proto
*prot
)
2710 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2712 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2713 pr_err("PROTO_INUSE_NR exhausted\n");
2717 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2720 static void release_proto_idx(struct proto
*prot
)
2722 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2723 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2726 static inline void assign_proto_idx(struct proto
*prot
)
2730 static inline void release_proto_idx(struct proto
*prot
)
2735 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
2739 kfree(rsk_prot
->slab_name
);
2740 rsk_prot
->slab_name
= NULL
;
2741 if (rsk_prot
->slab
) {
2742 kmem_cache_destroy(rsk_prot
->slab
);
2743 rsk_prot
->slab
= NULL
;
2747 static int req_prot_init(const struct proto
*prot
)
2749 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
2754 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
2756 if (!rsk_prot
->slab_name
)
2759 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
2760 rsk_prot
->obj_size
, 0,
2763 if (!rsk_prot
->slab
) {
2764 pr_crit("%s: Can't create request sock SLAB cache!\n",
2771 int proto_register(struct proto
*prot
, int alloc_slab
)
2774 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2775 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2778 if (prot
->slab
== NULL
) {
2779 pr_crit("%s: Can't create sock SLAB cache!\n",
2784 if (req_prot_init(prot
))
2785 goto out_free_request_sock_slab
;
2787 if (prot
->twsk_prot
!= NULL
) {
2788 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2790 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2791 goto out_free_request_sock_slab
;
2793 prot
->twsk_prot
->twsk_slab
=
2794 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2795 prot
->twsk_prot
->twsk_obj_size
,
2799 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2800 goto out_free_timewait_sock_slab_name
;
2804 mutex_lock(&proto_list_mutex
);
2805 list_add(&prot
->node
, &proto_list
);
2806 assign_proto_idx(prot
);
2807 mutex_unlock(&proto_list_mutex
);
2810 out_free_timewait_sock_slab_name
:
2811 kfree(prot
->twsk_prot
->twsk_slab_name
);
2812 out_free_request_sock_slab
:
2813 req_prot_cleanup(prot
->rsk_prot
);
2815 kmem_cache_destroy(prot
->slab
);
2820 EXPORT_SYMBOL(proto_register
);
2822 void proto_unregister(struct proto
*prot
)
2824 mutex_lock(&proto_list_mutex
);
2825 release_proto_idx(prot
);
2826 list_del(&prot
->node
);
2827 mutex_unlock(&proto_list_mutex
);
2829 if (prot
->slab
!= NULL
) {
2830 kmem_cache_destroy(prot
->slab
);
2834 req_prot_cleanup(prot
->rsk_prot
);
2836 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2837 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2838 kfree(prot
->twsk_prot
->twsk_slab_name
);
2839 prot
->twsk_prot
->twsk_slab
= NULL
;
2842 EXPORT_SYMBOL(proto_unregister
);
2844 #ifdef CONFIG_PROC_FS
2845 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2846 __acquires(proto_list_mutex
)
2848 mutex_lock(&proto_list_mutex
);
2849 return seq_list_start_head(&proto_list
, *pos
);
2852 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2854 return seq_list_next(v
, &proto_list
, pos
);
2857 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2858 __releases(proto_list_mutex
)
2860 mutex_unlock(&proto_list_mutex
);
2863 static char proto_method_implemented(const void *method
)
2865 return method
== NULL
? 'n' : 'y';
2867 static long sock_prot_memory_allocated(struct proto
*proto
)
2869 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2872 static char *sock_prot_memory_pressure(struct proto
*proto
)
2874 return proto
->memory_pressure
!= NULL
?
2875 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2878 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2881 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2882 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2885 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2886 sock_prot_memory_allocated(proto
),
2887 sock_prot_memory_pressure(proto
),
2889 proto
->slab
== NULL
? "no" : "yes",
2890 module_name(proto
->owner
),
2891 proto_method_implemented(proto
->close
),
2892 proto_method_implemented(proto
->connect
),
2893 proto_method_implemented(proto
->disconnect
),
2894 proto_method_implemented(proto
->accept
),
2895 proto_method_implemented(proto
->ioctl
),
2896 proto_method_implemented(proto
->init
),
2897 proto_method_implemented(proto
->destroy
),
2898 proto_method_implemented(proto
->shutdown
),
2899 proto_method_implemented(proto
->setsockopt
),
2900 proto_method_implemented(proto
->getsockopt
),
2901 proto_method_implemented(proto
->sendmsg
),
2902 proto_method_implemented(proto
->recvmsg
),
2903 proto_method_implemented(proto
->sendpage
),
2904 proto_method_implemented(proto
->bind
),
2905 proto_method_implemented(proto
->backlog_rcv
),
2906 proto_method_implemented(proto
->hash
),
2907 proto_method_implemented(proto
->unhash
),
2908 proto_method_implemented(proto
->get_port
),
2909 proto_method_implemented(proto
->enter_memory_pressure
));
2912 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2914 if (v
== &proto_list
)
2915 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2924 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2926 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2930 static const struct seq_operations proto_seq_ops
= {
2931 .start
= proto_seq_start
,
2932 .next
= proto_seq_next
,
2933 .stop
= proto_seq_stop
,
2934 .show
= proto_seq_show
,
2937 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2939 return seq_open_net(inode
, file
, &proto_seq_ops
,
2940 sizeof(struct seq_net_private
));
2943 static const struct file_operations proto_seq_fops
= {
2944 .owner
= THIS_MODULE
,
2945 .open
= proto_seq_open
,
2947 .llseek
= seq_lseek
,
2948 .release
= seq_release_net
,
2951 static __net_init
int proto_init_net(struct net
*net
)
2953 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
2959 static __net_exit
void proto_exit_net(struct net
*net
)
2961 remove_proc_entry("protocols", net
->proc_net
);
2965 static __net_initdata
struct pernet_operations proto_net_ops
= {
2966 .init
= proto_init_net
,
2967 .exit
= proto_exit_net
,
2970 static int __init
proto_init(void)
2972 return register_pernet_subsys(&proto_net_ops
);
2975 subsys_initcall(proto_init
);
2977 #endif /* PROC_FS */