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/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
115 #include <linux/user_namespace.h>
116 #include <linux/static_key.h>
117 #include <linux/memcontrol.h>
118 #include <linux/prefetch.h>
120 #include <asm/uaccess.h>
122 #include <linux/netdevice.h>
123 #include <net/protocol.h>
124 #include <linux/skbuff.h>
125 #include <net/net_namespace.h>
126 #include <net/request_sock.h>
127 #include <net/sock.h>
128 #include <linux/net_tstamp.h>
129 #include <net/xfrm.h>
130 #include <linux/ipsec.h>
131 #include <net/cls_cgroup.h>
132 #include <net/netprio_cgroup.h>
134 #include <linux/filter.h>
136 #include <trace/events/sock.h>
142 #include <net/ll_poll.h>
144 static DEFINE_MUTEX(proto_list_mutex
);
145 static LIST_HEAD(proto_list
);
147 #ifdef CONFIG_MEMCG_KMEM
148 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
153 mutex_lock(&proto_list_mutex
);
154 list_for_each_entry(proto
, &proto_list
, node
) {
155 if (proto
->init_cgroup
) {
156 ret
= proto
->init_cgroup(memcg
, ss
);
162 mutex_unlock(&proto_list_mutex
);
165 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
166 if (proto
->destroy_cgroup
)
167 proto
->destroy_cgroup(memcg
);
168 mutex_unlock(&proto_list_mutex
);
172 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
176 mutex_lock(&proto_list_mutex
);
177 list_for_each_entry_reverse(proto
, &proto_list
, node
)
178 if (proto
->destroy_cgroup
)
179 proto
->destroy_cgroup(memcg
);
180 mutex_unlock(&proto_list_mutex
);
185 * Each address family might have different locking rules, so we have
186 * one slock key per address family:
188 static struct lock_class_key af_family_keys
[AF_MAX
];
189 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
191 #if defined(CONFIG_MEMCG_KMEM)
192 struct static_key memcg_socket_limit_enabled
;
193 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
197 * Make lock validator output more readable. (we pre-construct these
198 * strings build-time, so that runtime initialization of socket
201 static const char *const af_family_key_strings
[AF_MAX
+1] = {
202 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
203 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
204 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
205 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
206 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
207 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
208 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
209 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
210 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
211 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
212 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
213 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
214 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
215 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
217 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
218 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
219 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
220 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
221 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
222 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
223 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
224 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
225 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
226 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
227 "slock-27" , "slock-28" , "slock-AF_CAN" ,
228 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
229 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
230 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
231 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
233 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
234 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
235 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
236 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
237 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
238 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
239 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
240 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
241 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
242 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
243 "clock-27" , "clock-28" , "clock-AF_CAN" ,
244 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
245 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
246 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
247 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
251 * sk_callback_lock locking rules are per-address-family,
252 * so split the lock classes by using a per-AF key:
254 static struct lock_class_key af_callback_keys
[AF_MAX
];
256 /* Take into consideration the size of the struct sk_buff overhead in the
257 * determination of these values, since that is non-constant across
258 * platforms. This makes socket queueing behavior and performance
259 * not depend upon such differences.
261 #define _SK_MEM_PACKETS 256
262 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
263 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
264 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
266 /* Run time adjustable parameters. */
267 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
268 EXPORT_SYMBOL(sysctl_wmem_max
);
269 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
270 EXPORT_SYMBOL(sysctl_rmem_max
);
271 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
272 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
274 /* Maximal space eaten by iovec or ancillary data plus some space */
275 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
276 EXPORT_SYMBOL(sysctl_optmem_max
);
278 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
279 EXPORT_SYMBOL_GPL(memalloc_socks
);
282 * sk_set_memalloc - sets %SOCK_MEMALLOC
283 * @sk: socket to set it on
285 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
286 * It's the responsibility of the admin to adjust min_free_kbytes
287 * to meet the requirements
289 void sk_set_memalloc(struct sock
*sk
)
291 sock_set_flag(sk
, SOCK_MEMALLOC
);
292 sk
->sk_allocation
|= __GFP_MEMALLOC
;
293 static_key_slow_inc(&memalloc_socks
);
295 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
297 void sk_clear_memalloc(struct sock
*sk
)
299 sock_reset_flag(sk
, SOCK_MEMALLOC
);
300 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
301 static_key_slow_dec(&memalloc_socks
);
304 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
305 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
306 * it has rmem allocations there is a risk that the user of the
307 * socket cannot make forward progress due to exceeding the rmem
308 * limits. By rights, sk_clear_memalloc() should only be called
309 * on sockets being torn down but warn and reset the accounting if
310 * that assumption breaks.
312 if (WARN_ON(sk
->sk_forward_alloc
))
315 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
317 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
320 unsigned long pflags
= current
->flags
;
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
325 current
->flags
|= PF_MEMALLOC
;
326 ret
= sk
->sk_backlog_rcv(sk
, skb
);
327 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
331 EXPORT_SYMBOL(__sk_backlog_rcv
);
333 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
337 if (optlen
< sizeof(tv
))
339 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
341 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
345 static int warned __read_mostly
;
348 if (warned
< 10 && net_ratelimit()) {
350 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
351 __func__
, current
->comm
, task_pid_nr(current
));
355 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
356 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
358 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
359 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
363 static void sock_warn_obsolete_bsdism(const char *name
)
366 static char warncomm
[TASK_COMM_LEN
];
367 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
368 strcpy(warncomm
, current
->comm
);
369 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
375 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
377 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
379 if (sk
->sk_flags
& flags
) {
380 sk
->sk_flags
&= ~flags
;
381 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
382 net_disable_timestamp();
387 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
392 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
394 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
395 atomic_inc(&sk
->sk_drops
);
396 trace_sock_rcvqueue_full(sk
, skb
);
400 err
= sk_filter(sk
, skb
);
404 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
405 atomic_inc(&sk
->sk_drops
);
410 skb_set_owner_r(skb
, sk
);
412 /* Cache the SKB length before we tack it onto the receive
413 * queue. Once it is added it no longer belongs to us and
414 * may be freed by other threads of control pulling packets
419 /* we escape from rcu protected region, make sure we dont leak
424 spin_lock_irqsave(&list
->lock
, flags
);
425 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
426 __skb_queue_tail(list
, skb
);
427 spin_unlock_irqrestore(&list
->lock
, flags
);
429 if (!sock_flag(sk
, SOCK_DEAD
))
430 sk
->sk_data_ready(sk
, skb_len
);
433 EXPORT_SYMBOL(sock_queue_rcv_skb
);
435 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
437 int rc
= NET_RX_SUCCESS
;
439 if (sk_filter(sk
, skb
))
440 goto discard_and_relse
;
444 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
)) {
445 atomic_inc(&sk
->sk_drops
);
446 goto discard_and_relse
;
449 bh_lock_sock_nested(sk
);
452 if (!sock_owned_by_user(sk
)) {
454 * trylock + unlock semantics:
456 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
458 rc
= sk_backlog_rcv(sk
, skb
);
460 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
461 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
463 atomic_inc(&sk
->sk_drops
);
464 goto discard_and_relse
;
475 EXPORT_SYMBOL(sk_receive_skb
);
477 void sk_reset_txq(struct sock
*sk
)
479 sk_tx_queue_clear(sk
);
481 EXPORT_SYMBOL(sk_reset_txq
);
483 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
485 struct dst_entry
*dst
= __sk_dst_get(sk
);
487 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
488 sk_tx_queue_clear(sk
);
489 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
496 EXPORT_SYMBOL(__sk_dst_check
);
498 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
500 struct dst_entry
*dst
= sk_dst_get(sk
);
502 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
510 EXPORT_SYMBOL(sk_dst_check
);
512 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
515 int ret
= -ENOPROTOOPT
;
516 #ifdef CONFIG_NETDEVICES
517 struct net
*net
= sock_net(sk
);
518 char devname
[IFNAMSIZ
];
523 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
530 /* Bind this socket to a particular device like "eth0",
531 * as specified in the passed interface name. If the
532 * name is "" or the option length is zero the socket
535 if (optlen
> IFNAMSIZ
- 1)
536 optlen
= IFNAMSIZ
- 1;
537 memset(devname
, 0, sizeof(devname
));
540 if (copy_from_user(devname
, optval
, optlen
))
544 if (devname
[0] != '\0') {
545 struct net_device
*dev
;
548 dev
= dev_get_by_name_rcu(net
, devname
);
550 index
= dev
->ifindex
;
558 sk
->sk_bound_dev_if
= index
;
570 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
571 int __user
*optlen
, int len
)
573 int ret
= -ENOPROTOOPT
;
574 #ifdef CONFIG_NETDEVICES
575 struct net
*net
= sock_net(sk
);
576 struct net_device
*dev
;
577 char devname
[IFNAMSIZ
];
580 if (sk
->sk_bound_dev_if
== 0) {
590 seq
= read_seqcount_begin(&devnet_rename_seq
);
592 dev
= dev_get_by_index_rcu(net
, sk
->sk_bound_dev_if
);
599 strcpy(devname
, dev
->name
);
601 if (read_seqcount_retry(&devnet_rename_seq
, seq
))
604 len
= strlen(devname
) + 1;
607 if (copy_to_user(optval
, devname
, len
))
612 if (put_user(len
, optlen
))
623 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
626 sock_set_flag(sk
, bit
);
628 sock_reset_flag(sk
, bit
);
632 * This is meant for all protocols to use and covers goings on
633 * at the socket level. Everything here is generic.
636 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
637 char __user
*optval
, unsigned int optlen
)
639 struct sock
*sk
= sock
->sk
;
646 * Options without arguments
649 if (optname
== SO_BINDTODEVICE
)
650 return sock_setbindtodevice(sk
, optval
, optlen
);
652 if (optlen
< sizeof(int))
655 if (get_user(val
, (int __user
*)optval
))
658 valbool
= val
? 1 : 0;
664 if (val
&& !capable(CAP_NET_ADMIN
))
667 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
670 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
673 sk
->sk_reuseport
= valbool
;
682 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
685 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
688 /* Don't error on this BSD doesn't and if you think
689 * about it this is right. Otherwise apps have to
690 * play 'guess the biggest size' games. RCVBUF/SNDBUF
691 * are treated in BSD as hints
693 val
= min_t(u32
, val
, sysctl_wmem_max
);
695 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
696 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
697 /* Wake up sending tasks if we upped the value. */
698 sk
->sk_write_space(sk
);
702 if (!capable(CAP_NET_ADMIN
)) {
709 /* Don't error on this BSD doesn't and if you think
710 * about it this is right. Otherwise apps have to
711 * play 'guess the biggest size' games. RCVBUF/SNDBUF
712 * are treated in BSD as hints
714 val
= min_t(u32
, val
, sysctl_rmem_max
);
716 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
718 * We double it on the way in to account for
719 * "struct sk_buff" etc. overhead. Applications
720 * assume that the SO_RCVBUF setting they make will
721 * allow that much actual data to be received on that
724 * Applications are unaware that "struct sk_buff" and
725 * other overheads allocate from the receive buffer
726 * during socket buffer allocation.
728 * And after considering the possible alternatives,
729 * returning the value we actually used in getsockopt
730 * is the most desirable behavior.
732 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
736 if (!capable(CAP_NET_ADMIN
)) {
744 if (sk
->sk_protocol
== IPPROTO_TCP
&&
745 sk
->sk_type
== SOCK_STREAM
)
746 tcp_set_keepalive(sk
, valbool
);
748 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
752 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
756 sk
->sk_no_check
= valbool
;
760 if ((val
>= 0 && val
<= 6) ||
761 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
762 sk
->sk_priority
= val
;
768 if (optlen
< sizeof(ling
)) {
769 ret
= -EINVAL
; /* 1003.1g */
772 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
777 sock_reset_flag(sk
, SOCK_LINGER
);
779 #if (BITS_PER_LONG == 32)
780 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
781 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
784 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
785 sock_set_flag(sk
, SOCK_LINGER
);
790 sock_warn_obsolete_bsdism("setsockopt");
795 set_bit(SOCK_PASSCRED
, &sock
->flags
);
797 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
803 if (optname
== SO_TIMESTAMP
)
804 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
806 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
807 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
808 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
810 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
811 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
815 case SO_TIMESTAMPING
:
816 if (val
& ~SOF_TIMESTAMPING_MASK
) {
820 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
821 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
822 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
823 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
824 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
825 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
826 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
827 sock_enable_timestamp(sk
,
828 SOCK_TIMESTAMPING_RX_SOFTWARE
);
830 sock_disable_timestamp(sk
,
831 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
832 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
833 val
& SOF_TIMESTAMPING_SOFTWARE
);
834 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
835 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
836 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
837 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
843 sk
->sk_rcvlowat
= val
? : 1;
847 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
851 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
854 case SO_ATTACH_FILTER
:
856 if (optlen
== sizeof(struct sock_fprog
)) {
857 struct sock_fprog fprog
;
860 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
863 ret
= sk_attach_filter(&fprog
, sk
);
867 case SO_DETACH_FILTER
:
868 ret
= sk_detach_filter(sk
);
872 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
875 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
880 set_bit(SOCK_PASSSEC
, &sock
->flags
);
882 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
885 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
891 /* We implement the SO_SNDLOWAT etc to
892 not be settable (1003.1g 5.3) */
894 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
898 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
902 if (sock
->ops
->set_peek_off
)
903 sock
->ops
->set_peek_off(sk
, val
);
909 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
912 case SO_SELECT_ERR_QUEUE
:
913 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
916 #ifdef CONFIG_NET_LL_RX_POLL
918 /* allow unprivileged users to decrease the value */
919 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
925 sk
->sk_ll_usec
= val
;
936 EXPORT_SYMBOL(sock_setsockopt
);
939 void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
942 ucred
->pid
= pid_vnr(pid
);
943 ucred
->uid
= ucred
->gid
= -1;
945 struct user_namespace
*current_ns
= current_user_ns();
947 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
948 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
951 EXPORT_SYMBOL_GPL(cred_to_ucred
);
953 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
954 char __user
*optval
, int __user
*optlen
)
956 struct sock
*sk
= sock
->sk
;
964 int lv
= sizeof(int);
967 if (get_user(len
, optlen
))
972 memset(&v
, 0, sizeof(v
));
976 v
.val
= sock_flag(sk
, SOCK_DBG
);
980 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
984 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
988 v
.val
= sk
->sk_sndbuf
;
992 v
.val
= sk
->sk_rcvbuf
;
996 v
.val
= sk
->sk_reuse
;
1000 v
.val
= sk
->sk_reuseport
;
1004 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1008 v
.val
= sk
->sk_type
;
1012 v
.val
= sk
->sk_protocol
;
1016 v
.val
= sk
->sk_family
;
1020 v
.val
= -sock_error(sk
);
1022 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1026 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1030 v
.val
= sk
->sk_no_check
;
1034 v
.val
= sk
->sk_priority
;
1038 lv
= sizeof(v
.ling
);
1039 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1040 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1044 sock_warn_obsolete_bsdism("getsockopt");
1048 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1049 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1052 case SO_TIMESTAMPNS
:
1053 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1056 case SO_TIMESTAMPING
:
1058 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
1059 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
1060 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
1061 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
1062 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
1063 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
1064 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
1065 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
1066 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
1067 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
1068 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
1069 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
1070 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
1071 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
1075 lv
= sizeof(struct timeval
);
1076 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1080 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1081 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1086 lv
= sizeof(struct timeval
);
1087 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1091 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1092 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1097 v
.val
= sk
->sk_rcvlowat
;
1105 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1110 struct ucred peercred
;
1111 if (len
> sizeof(peercred
))
1112 len
= sizeof(peercred
);
1113 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1114 if (copy_to_user(optval
, &peercred
, len
))
1123 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1127 if (copy_to_user(optval
, address
, len
))
1132 /* Dubious BSD thing... Probably nobody even uses it, but
1133 * the UNIX standard wants it for whatever reason... -DaveM
1136 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1140 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1144 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1147 v
.val
= sk
->sk_mark
;
1151 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1154 case SO_WIFI_STATUS
:
1155 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1159 if (!sock
->ops
->set_peek_off
)
1162 v
.val
= sk
->sk_peek_off
;
1165 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1168 case SO_BINDTODEVICE
:
1169 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1172 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1178 case SO_LOCK_FILTER
:
1179 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1182 case SO_SELECT_ERR_QUEUE
:
1183 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1186 #ifdef CONFIG_NET_LL_RX_POLL
1188 v
.val
= sk
->sk_ll_usec
;
1193 return -ENOPROTOOPT
;
1198 if (copy_to_user(optval
, &v
, len
))
1201 if (put_user(len
, optlen
))
1207 * Initialize an sk_lock.
1209 * (We also register the sk_lock with the lock validator.)
1211 static inline void sock_lock_init(struct sock
*sk
)
1213 sock_lock_init_class_and_name(sk
,
1214 af_family_slock_key_strings
[sk
->sk_family
],
1215 af_family_slock_keys
+ sk
->sk_family
,
1216 af_family_key_strings
[sk
->sk_family
],
1217 af_family_keys
+ sk
->sk_family
);
1221 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1222 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1223 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1225 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1227 #ifdef CONFIG_SECURITY_NETWORK
1228 void *sptr
= nsk
->sk_security
;
1230 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1232 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1233 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1235 #ifdef CONFIG_SECURITY_NETWORK
1236 nsk
->sk_security
= sptr
;
1237 security_sk_clone(osk
, nsk
);
1241 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1243 unsigned long nulls1
, nulls2
;
1245 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1246 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1247 if (nulls1
> nulls2
)
1248 swap(nulls1
, nulls2
);
1251 memset((char *)sk
, 0, nulls1
);
1252 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1253 nulls2
- nulls1
- sizeof(void *));
1254 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1255 size
- nulls2
- sizeof(void *));
1257 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1259 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1263 struct kmem_cache
*slab
;
1267 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1270 if (priority
& __GFP_ZERO
) {
1272 prot
->clear_sk(sk
, prot
->obj_size
);
1274 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1277 sk
= kmalloc(prot
->obj_size
, priority
);
1280 kmemcheck_annotate_bitfield(sk
, flags
);
1282 if (security_sk_alloc(sk
, family
, priority
))
1285 if (!try_module_get(prot
->owner
))
1287 sk_tx_queue_clear(sk
);
1293 security_sk_free(sk
);
1296 kmem_cache_free(slab
, sk
);
1302 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1304 struct kmem_cache
*slab
;
1305 struct module
*owner
;
1307 owner
= prot
->owner
;
1310 security_sk_free(sk
);
1312 kmem_cache_free(slab
, sk
);
1318 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1319 void sock_update_classid(struct sock
*sk
)
1323 classid
= task_cls_classid(current
);
1324 if (classid
!= sk
->sk_classid
)
1325 sk
->sk_classid
= classid
;
1327 EXPORT_SYMBOL(sock_update_classid
);
1330 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1331 void sock_update_netprioidx(struct sock
*sk
)
1336 sk
->sk_cgrp_prioidx
= task_netprioidx(current
);
1338 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1342 * sk_alloc - All socket objects are allocated here
1343 * @net: the applicable net namespace
1344 * @family: protocol family
1345 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1346 * @prot: struct proto associated with this new sock instance
1348 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1353 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1355 sk
->sk_family
= family
;
1357 * See comment in struct sock definition to understand
1358 * why we need sk_prot_creator -acme
1360 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1362 sock_net_set(sk
, get_net(net
));
1363 atomic_set(&sk
->sk_wmem_alloc
, 1);
1365 sock_update_classid(sk
);
1366 sock_update_netprioidx(sk
);
1371 EXPORT_SYMBOL(sk_alloc
);
1373 static void __sk_free(struct sock
*sk
)
1375 struct sk_filter
*filter
;
1377 if (sk
->sk_destruct
)
1378 sk
->sk_destruct(sk
);
1380 filter
= rcu_dereference_check(sk
->sk_filter
,
1381 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1383 sk_filter_uncharge(sk
, filter
);
1384 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1387 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1389 if (atomic_read(&sk
->sk_omem_alloc
))
1390 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1391 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1393 if (sk
->sk_peer_cred
)
1394 put_cred(sk
->sk_peer_cred
);
1395 put_pid(sk
->sk_peer_pid
);
1396 put_net(sock_net(sk
));
1397 sk_prot_free(sk
->sk_prot_creator
, sk
);
1400 void sk_free(struct sock
*sk
)
1403 * We subtract one from sk_wmem_alloc and can know if
1404 * some packets are still in some tx queue.
1405 * If not null, sock_wfree() will call __sk_free(sk) later
1407 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1410 EXPORT_SYMBOL(sk_free
);
1413 * Last sock_put should drop reference to sk->sk_net. It has already
1414 * been dropped in sk_change_net. Taking reference to stopping namespace
1416 * Take reference to a socket to remove it from hash _alive_ and after that
1417 * destroy it in the context of init_net.
1419 void sk_release_kernel(struct sock
*sk
)
1421 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1425 sock_release(sk
->sk_socket
);
1426 release_net(sock_net(sk
));
1427 sock_net_set(sk
, get_net(&init_net
));
1430 EXPORT_SYMBOL(sk_release_kernel
);
1432 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1434 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1435 sock_update_memcg(newsk
);
1439 * sk_clone_lock - clone a socket, and lock its clone
1440 * @sk: the socket to clone
1441 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1443 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1445 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1449 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1450 if (newsk
!= NULL
) {
1451 struct sk_filter
*filter
;
1453 sock_copy(newsk
, sk
);
1456 get_net(sock_net(newsk
));
1457 sk_node_init(&newsk
->sk_node
);
1458 sock_lock_init(newsk
);
1459 bh_lock_sock(newsk
);
1460 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1461 newsk
->sk_backlog
.len
= 0;
1463 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1465 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1467 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1468 atomic_set(&newsk
->sk_omem_alloc
, 0);
1469 skb_queue_head_init(&newsk
->sk_receive_queue
);
1470 skb_queue_head_init(&newsk
->sk_write_queue
);
1471 #ifdef CONFIG_NET_DMA
1472 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1475 spin_lock_init(&newsk
->sk_dst_lock
);
1476 rwlock_init(&newsk
->sk_callback_lock
);
1477 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1478 af_callback_keys
+ newsk
->sk_family
,
1479 af_family_clock_key_strings
[newsk
->sk_family
]);
1481 newsk
->sk_dst_cache
= NULL
;
1482 newsk
->sk_wmem_queued
= 0;
1483 newsk
->sk_forward_alloc
= 0;
1484 newsk
->sk_send_head
= NULL
;
1485 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1487 sock_reset_flag(newsk
, SOCK_DONE
);
1488 skb_queue_head_init(&newsk
->sk_error_queue
);
1490 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1492 sk_filter_charge(newsk
, filter
);
1494 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1495 /* It is still raw copy of parent, so invalidate
1496 * destructor and make plain sk_free() */
1497 newsk
->sk_destruct
= NULL
;
1498 bh_unlock_sock(newsk
);
1505 newsk
->sk_priority
= 0;
1507 * Before updating sk_refcnt, we must commit prior changes to memory
1508 * (Documentation/RCU/rculist_nulls.txt for details)
1511 atomic_set(&newsk
->sk_refcnt
, 2);
1514 * Increment the counter in the same struct proto as the master
1515 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1516 * is the same as sk->sk_prot->socks, as this field was copied
1519 * This _changes_ the previous behaviour, where
1520 * tcp_create_openreq_child always was incrementing the
1521 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1522 * to be taken into account in all callers. -acme
1524 sk_refcnt_debug_inc(newsk
);
1525 sk_set_socket(newsk
, NULL
);
1526 newsk
->sk_wq
= NULL
;
1528 sk_update_clone(sk
, newsk
);
1530 if (newsk
->sk_prot
->sockets_allocated
)
1531 sk_sockets_allocated_inc(newsk
);
1533 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1534 net_enable_timestamp();
1539 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1541 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1543 __sk_dst_set(sk
, dst
);
1544 sk
->sk_route_caps
= dst
->dev
->features
;
1545 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1546 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1547 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1548 if (sk_can_gso(sk
)) {
1549 if (dst
->header_len
) {
1550 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1552 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1553 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1554 sk
->sk_gso_max_segs
= dst
->dev
->gso_max_segs
;
1558 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1561 * Simple resource managers for sockets.
1566 * Write buffer destructor automatically called from kfree_skb.
1568 void sock_wfree(struct sk_buff
*skb
)
1570 struct sock
*sk
= skb
->sk
;
1571 unsigned int len
= skb
->truesize
;
1573 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1575 * Keep a reference on sk_wmem_alloc, this will be released
1576 * after sk_write_space() call
1578 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1579 sk
->sk_write_space(sk
);
1583 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1584 * could not do because of in-flight packets
1586 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1589 EXPORT_SYMBOL(sock_wfree
);
1592 * Read buffer destructor automatically called from kfree_skb.
1594 void sock_rfree(struct sk_buff
*skb
)
1596 struct sock
*sk
= skb
->sk
;
1597 unsigned int len
= skb
->truesize
;
1599 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1600 sk_mem_uncharge(sk
, len
);
1602 EXPORT_SYMBOL(sock_rfree
);
1604 void sock_edemux(struct sk_buff
*skb
)
1606 struct sock
*sk
= skb
->sk
;
1609 if (sk
->sk_state
== TCP_TIME_WAIT
)
1610 inet_twsk_put(inet_twsk(sk
));
1615 EXPORT_SYMBOL(sock_edemux
);
1617 kuid_t
sock_i_uid(struct sock
*sk
)
1621 read_lock_bh(&sk
->sk_callback_lock
);
1622 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1623 read_unlock_bh(&sk
->sk_callback_lock
);
1626 EXPORT_SYMBOL(sock_i_uid
);
1628 unsigned long sock_i_ino(struct sock
*sk
)
1632 read_lock_bh(&sk
->sk_callback_lock
);
1633 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1634 read_unlock_bh(&sk
->sk_callback_lock
);
1637 EXPORT_SYMBOL(sock_i_ino
);
1640 * Allocate a skb from the socket's send buffer.
1642 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1645 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1646 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1648 skb_set_owner_w(skb
, sk
);
1654 EXPORT_SYMBOL(sock_wmalloc
);
1657 * Allocate a skb from the socket's receive buffer.
1659 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1662 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1663 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1665 skb_set_owner_r(skb
, sk
);
1673 * Allocate a memory block from the socket's option memory buffer.
1675 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1677 if ((unsigned int)size
<= sysctl_optmem_max
&&
1678 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1680 /* First do the add, to avoid the race if kmalloc
1683 atomic_add(size
, &sk
->sk_omem_alloc
);
1684 mem
= kmalloc(size
, priority
);
1687 atomic_sub(size
, &sk
->sk_omem_alloc
);
1691 EXPORT_SYMBOL(sock_kmalloc
);
1694 * Free an option memory block.
1696 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1699 atomic_sub(size
, &sk
->sk_omem_alloc
);
1701 EXPORT_SYMBOL(sock_kfree_s
);
1703 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1704 I think, these locks should be removed for datagram sockets.
1706 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1710 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1714 if (signal_pending(current
))
1716 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1717 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1718 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1720 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1724 timeo
= schedule_timeout(timeo
);
1726 finish_wait(sk_sleep(sk
), &wait
);
1732 * Generic send/receive buffer handlers
1735 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1736 unsigned long data_len
, int noblock
,
1739 struct sk_buff
*skb
;
1743 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1746 if (npages
> MAX_SKB_FRAGS
)
1749 gfp_mask
= sk
->sk_allocation
;
1750 if (gfp_mask
& __GFP_WAIT
)
1751 gfp_mask
|= __GFP_REPEAT
;
1753 timeo
= sock_sndtimeo(sk
, noblock
);
1755 err
= sock_error(sk
);
1760 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1763 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1764 skb
= alloc_skb(header_len
, gfp_mask
);
1768 /* No pages, we're done... */
1772 skb
->truesize
+= data_len
;
1773 skb_shinfo(skb
)->nr_frags
= npages
;
1774 for (i
= 0; i
< npages
; i
++) {
1777 page
= alloc_pages(sk
->sk_allocation
, 0);
1780 skb_shinfo(skb
)->nr_frags
= i
;
1785 __skb_fill_page_desc(skb
, i
,
1787 (data_len
>= PAGE_SIZE
?
1790 data_len
-= PAGE_SIZE
;
1793 /* Full success... */
1799 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1800 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1804 if (signal_pending(current
))
1806 timeo
= sock_wait_for_wmem(sk
, timeo
);
1809 skb_set_owner_w(skb
, sk
);
1813 err
= sock_intr_errno(timeo
);
1818 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1820 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1821 int noblock
, int *errcode
)
1823 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1825 EXPORT_SYMBOL(sock_alloc_send_skb
);
1827 /* On 32bit arches, an skb frag is limited to 2^15 */
1828 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1830 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1835 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1839 if (pfrag
->offset
< pfrag
->size
)
1841 put_page(pfrag
->page
);
1844 /* We restrict high order allocations to users that can afford to wait */
1845 order
= (sk
->sk_allocation
& __GFP_WAIT
) ? SKB_FRAG_PAGE_ORDER
: 0;
1848 gfp_t gfp
= sk
->sk_allocation
;
1851 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
1852 pfrag
->page
= alloc_pages(gfp
, order
);
1853 if (likely(pfrag
->page
)) {
1855 pfrag
->size
= PAGE_SIZE
<< order
;
1858 } while (--order
>= 0);
1860 sk_enter_memory_pressure(sk
);
1861 sk_stream_moderate_sndbuf(sk
);
1864 EXPORT_SYMBOL(sk_page_frag_refill
);
1866 static void __lock_sock(struct sock
*sk
)
1867 __releases(&sk
->sk_lock
.slock
)
1868 __acquires(&sk
->sk_lock
.slock
)
1873 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1874 TASK_UNINTERRUPTIBLE
);
1875 spin_unlock_bh(&sk
->sk_lock
.slock
);
1877 spin_lock_bh(&sk
->sk_lock
.slock
);
1878 if (!sock_owned_by_user(sk
))
1881 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1884 static void __release_sock(struct sock
*sk
)
1885 __releases(&sk
->sk_lock
.slock
)
1886 __acquires(&sk
->sk_lock
.slock
)
1888 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1891 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1895 struct sk_buff
*next
= skb
->next
;
1898 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1900 sk_backlog_rcv(sk
, skb
);
1903 * We are in process context here with softirqs
1904 * disabled, use cond_resched_softirq() to preempt.
1905 * This is safe to do because we've taken the backlog
1908 cond_resched_softirq();
1911 } while (skb
!= NULL
);
1914 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1917 * Doing the zeroing here guarantee we can not loop forever
1918 * while a wild producer attempts to flood us.
1920 sk
->sk_backlog
.len
= 0;
1924 * sk_wait_data - wait for data to arrive at sk_receive_queue
1925 * @sk: sock to wait on
1926 * @timeo: for how long
1928 * Now socket state including sk->sk_err is changed only under lock,
1929 * hence we may omit checks after joining wait queue.
1930 * We check receive queue before schedule() only as optimization;
1931 * it is very likely that release_sock() added new data.
1933 int sk_wait_data(struct sock
*sk
, long *timeo
)
1938 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1939 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1940 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1941 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1942 finish_wait(sk_sleep(sk
), &wait
);
1945 EXPORT_SYMBOL(sk_wait_data
);
1948 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1950 * @size: memory size to allocate
1951 * @kind: allocation type
1953 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1954 * rmem allocation. This function assumes that protocols which have
1955 * memory_pressure use sk_wmem_queued as write buffer accounting.
1957 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
1959 struct proto
*prot
= sk
->sk_prot
;
1960 int amt
= sk_mem_pages(size
);
1962 int parent_status
= UNDER_LIMIT
;
1964 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
1966 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
1969 if (parent_status
== UNDER_LIMIT
&&
1970 allocated
<= sk_prot_mem_limits(sk
, 0)) {
1971 sk_leave_memory_pressure(sk
);
1975 /* Under pressure. (we or our parents) */
1976 if ((parent_status
> SOFT_LIMIT
) ||
1977 allocated
> sk_prot_mem_limits(sk
, 1))
1978 sk_enter_memory_pressure(sk
);
1980 /* Over hard limit (we or our parents) */
1981 if ((parent_status
== OVER_LIMIT
) ||
1982 (allocated
> sk_prot_mem_limits(sk
, 2)))
1983 goto suppress_allocation
;
1985 /* guarantee minimum buffer size under pressure */
1986 if (kind
== SK_MEM_RECV
) {
1987 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
1990 } else { /* SK_MEM_SEND */
1991 if (sk
->sk_type
== SOCK_STREAM
) {
1992 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
1994 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
1995 prot
->sysctl_wmem
[0])
1999 if (sk_has_memory_pressure(sk
)) {
2002 if (!sk_under_memory_pressure(sk
))
2004 alloc
= sk_sockets_allocated_read_positive(sk
);
2005 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2006 sk_mem_pages(sk
->sk_wmem_queued
+
2007 atomic_read(&sk
->sk_rmem_alloc
) +
2008 sk
->sk_forward_alloc
))
2012 suppress_allocation
:
2014 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2015 sk_stream_moderate_sndbuf(sk
);
2017 /* Fail only if socket is _under_ its sndbuf.
2018 * In this case we cannot block, so that we have to fail.
2020 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2024 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2026 /* Alas. Undo changes. */
2027 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2029 sk_memory_allocated_sub(sk
, amt
);
2033 EXPORT_SYMBOL(__sk_mem_schedule
);
2036 * __sk_reclaim - reclaim memory_allocated
2039 void __sk_mem_reclaim(struct sock
*sk
)
2041 sk_memory_allocated_sub(sk
,
2042 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
2043 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
2045 if (sk_under_memory_pressure(sk
) &&
2046 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2047 sk_leave_memory_pressure(sk
);
2049 EXPORT_SYMBOL(__sk_mem_reclaim
);
2053 * Set of default routines for initialising struct proto_ops when
2054 * the protocol does not support a particular function. In certain
2055 * cases where it makes no sense for a protocol to have a "do nothing"
2056 * function, some default processing is provided.
2059 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2063 EXPORT_SYMBOL(sock_no_bind
);
2065 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2070 EXPORT_SYMBOL(sock_no_connect
);
2072 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2076 EXPORT_SYMBOL(sock_no_socketpair
);
2078 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2082 EXPORT_SYMBOL(sock_no_accept
);
2084 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2089 EXPORT_SYMBOL(sock_no_getname
);
2091 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2095 EXPORT_SYMBOL(sock_no_poll
);
2097 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2101 EXPORT_SYMBOL(sock_no_ioctl
);
2103 int sock_no_listen(struct socket
*sock
, int backlog
)
2107 EXPORT_SYMBOL(sock_no_listen
);
2109 int sock_no_shutdown(struct socket
*sock
, int how
)
2113 EXPORT_SYMBOL(sock_no_shutdown
);
2115 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2116 char __user
*optval
, unsigned int optlen
)
2120 EXPORT_SYMBOL(sock_no_setsockopt
);
2122 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2123 char __user
*optval
, int __user
*optlen
)
2127 EXPORT_SYMBOL(sock_no_getsockopt
);
2129 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2134 EXPORT_SYMBOL(sock_no_sendmsg
);
2136 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2137 size_t len
, int flags
)
2141 EXPORT_SYMBOL(sock_no_recvmsg
);
2143 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2145 /* Mirror missing mmap method error code */
2148 EXPORT_SYMBOL(sock_no_mmap
);
2150 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2153 struct msghdr msg
= {.msg_flags
= flags
};
2155 char *kaddr
= kmap(page
);
2156 iov
.iov_base
= kaddr
+ offset
;
2158 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2162 EXPORT_SYMBOL(sock_no_sendpage
);
2165 * Default Socket Callbacks
2168 static void sock_def_wakeup(struct sock
*sk
)
2170 struct socket_wq
*wq
;
2173 wq
= rcu_dereference(sk
->sk_wq
);
2174 if (wq_has_sleeper(wq
))
2175 wake_up_interruptible_all(&wq
->wait
);
2179 static void sock_def_error_report(struct sock
*sk
)
2181 struct socket_wq
*wq
;
2184 wq
= rcu_dereference(sk
->sk_wq
);
2185 if (wq_has_sleeper(wq
))
2186 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2187 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2191 static void sock_def_readable(struct sock
*sk
, int len
)
2193 struct socket_wq
*wq
;
2196 wq
= rcu_dereference(sk
->sk_wq
);
2197 if (wq_has_sleeper(wq
))
2198 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2199 POLLRDNORM
| POLLRDBAND
);
2200 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2204 static void sock_def_write_space(struct sock
*sk
)
2206 struct socket_wq
*wq
;
2210 /* Do not wake up a writer until he can make "significant"
2213 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2214 wq
= rcu_dereference(sk
->sk_wq
);
2215 if (wq_has_sleeper(wq
))
2216 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2217 POLLWRNORM
| POLLWRBAND
);
2219 /* Should agree with poll, otherwise some programs break */
2220 if (sock_writeable(sk
))
2221 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2227 static void sock_def_destruct(struct sock
*sk
)
2229 kfree(sk
->sk_protinfo
);
2232 void sk_send_sigurg(struct sock
*sk
)
2234 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2235 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2236 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2238 EXPORT_SYMBOL(sk_send_sigurg
);
2240 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2241 unsigned long expires
)
2243 if (!mod_timer(timer
, expires
))
2246 EXPORT_SYMBOL(sk_reset_timer
);
2248 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2250 if (del_timer(timer
))
2253 EXPORT_SYMBOL(sk_stop_timer
);
2255 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2257 skb_queue_head_init(&sk
->sk_receive_queue
);
2258 skb_queue_head_init(&sk
->sk_write_queue
);
2259 skb_queue_head_init(&sk
->sk_error_queue
);
2260 #ifdef CONFIG_NET_DMA
2261 skb_queue_head_init(&sk
->sk_async_wait_queue
);
2264 sk
->sk_send_head
= NULL
;
2266 init_timer(&sk
->sk_timer
);
2268 sk
->sk_allocation
= GFP_KERNEL
;
2269 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2270 sk
->sk_sndbuf
= sysctl_wmem_default
;
2271 sk
->sk_state
= TCP_CLOSE
;
2272 sk_set_socket(sk
, sock
);
2274 sock_set_flag(sk
, SOCK_ZAPPED
);
2277 sk
->sk_type
= sock
->type
;
2278 sk
->sk_wq
= sock
->wq
;
2283 spin_lock_init(&sk
->sk_dst_lock
);
2284 rwlock_init(&sk
->sk_callback_lock
);
2285 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2286 af_callback_keys
+ sk
->sk_family
,
2287 af_family_clock_key_strings
[sk
->sk_family
]);
2289 sk
->sk_state_change
= sock_def_wakeup
;
2290 sk
->sk_data_ready
= sock_def_readable
;
2291 sk
->sk_write_space
= sock_def_write_space
;
2292 sk
->sk_error_report
= sock_def_error_report
;
2293 sk
->sk_destruct
= sock_def_destruct
;
2295 sk
->sk_frag
.page
= NULL
;
2296 sk
->sk_frag
.offset
= 0;
2297 sk
->sk_peek_off
= -1;
2299 sk
->sk_peer_pid
= NULL
;
2300 sk
->sk_peer_cred
= NULL
;
2301 sk
->sk_write_pending
= 0;
2302 sk
->sk_rcvlowat
= 1;
2303 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2304 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2306 sk
->sk_stamp
= ktime_set(-1L, 0);
2308 #ifdef CONFIG_NET_LL_RX_POLL
2310 sk
->sk_ll_usec
= sysctl_net_ll_read
;
2314 * Before updating sk_refcnt, we must commit prior changes to memory
2315 * (Documentation/RCU/rculist_nulls.txt for details)
2318 atomic_set(&sk
->sk_refcnt
, 1);
2319 atomic_set(&sk
->sk_drops
, 0);
2321 EXPORT_SYMBOL(sock_init_data
);
2323 void lock_sock_nested(struct sock
*sk
, int subclass
)
2326 spin_lock_bh(&sk
->sk_lock
.slock
);
2327 if (sk
->sk_lock
.owned
)
2329 sk
->sk_lock
.owned
= 1;
2330 spin_unlock(&sk
->sk_lock
.slock
);
2332 * The sk_lock has mutex_lock() semantics here:
2334 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2337 EXPORT_SYMBOL(lock_sock_nested
);
2339 void release_sock(struct sock
*sk
)
2342 * The sk_lock has mutex_unlock() semantics:
2344 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2346 spin_lock_bh(&sk
->sk_lock
.slock
);
2347 if (sk
->sk_backlog
.tail
)
2350 if (sk
->sk_prot
->release_cb
)
2351 sk
->sk_prot
->release_cb(sk
);
2353 sk
->sk_lock
.owned
= 0;
2354 if (waitqueue_active(&sk
->sk_lock
.wq
))
2355 wake_up(&sk
->sk_lock
.wq
);
2356 spin_unlock_bh(&sk
->sk_lock
.slock
);
2358 EXPORT_SYMBOL(release_sock
);
2361 * lock_sock_fast - fast version of lock_sock
2364 * This version should be used for very small section, where process wont block
2365 * return false if fast path is taken
2366 * sk_lock.slock locked, owned = 0, BH disabled
2367 * return true if slow path is taken
2368 * sk_lock.slock unlocked, owned = 1, BH enabled
2370 bool lock_sock_fast(struct sock
*sk
)
2373 spin_lock_bh(&sk
->sk_lock
.slock
);
2375 if (!sk
->sk_lock
.owned
)
2377 * Note : We must disable BH
2382 sk
->sk_lock
.owned
= 1;
2383 spin_unlock(&sk
->sk_lock
.slock
);
2385 * The sk_lock has mutex_lock() semantics here:
2387 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2391 EXPORT_SYMBOL(lock_sock_fast
);
2393 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2396 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2397 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2398 tv
= ktime_to_timeval(sk
->sk_stamp
);
2399 if (tv
.tv_sec
== -1)
2401 if (tv
.tv_sec
== 0) {
2402 sk
->sk_stamp
= ktime_get_real();
2403 tv
= ktime_to_timeval(sk
->sk_stamp
);
2405 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2407 EXPORT_SYMBOL(sock_get_timestamp
);
2409 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2412 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2413 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2414 ts
= ktime_to_timespec(sk
->sk_stamp
);
2415 if (ts
.tv_sec
== -1)
2417 if (ts
.tv_sec
== 0) {
2418 sk
->sk_stamp
= ktime_get_real();
2419 ts
= ktime_to_timespec(sk
->sk_stamp
);
2421 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2423 EXPORT_SYMBOL(sock_get_timestampns
);
2425 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2427 if (!sock_flag(sk
, flag
)) {
2428 unsigned long previous_flags
= sk
->sk_flags
;
2430 sock_set_flag(sk
, flag
);
2432 * we just set one of the two flags which require net
2433 * time stamping, but time stamping might have been on
2434 * already because of the other one
2436 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2437 net_enable_timestamp();
2442 * Get a socket option on an socket.
2444 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2445 * asynchronous errors should be reported by getsockopt. We assume
2446 * this means if you specify SO_ERROR (otherwise whats the point of it).
2448 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2449 char __user
*optval
, int __user
*optlen
)
2451 struct sock
*sk
= sock
->sk
;
2453 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2455 EXPORT_SYMBOL(sock_common_getsockopt
);
2457 #ifdef CONFIG_COMPAT
2458 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2459 char __user
*optval
, int __user
*optlen
)
2461 struct sock
*sk
= sock
->sk
;
2463 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2464 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2466 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2468 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2471 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2472 struct msghdr
*msg
, size_t size
, int flags
)
2474 struct sock
*sk
= sock
->sk
;
2478 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2479 flags
& ~MSG_DONTWAIT
, &addr_len
);
2481 msg
->msg_namelen
= addr_len
;
2484 EXPORT_SYMBOL(sock_common_recvmsg
);
2487 * Set socket options on an inet socket.
2489 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2490 char __user
*optval
, unsigned int optlen
)
2492 struct sock
*sk
= sock
->sk
;
2494 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2496 EXPORT_SYMBOL(sock_common_setsockopt
);
2498 #ifdef CONFIG_COMPAT
2499 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2500 char __user
*optval
, unsigned int optlen
)
2502 struct sock
*sk
= sock
->sk
;
2504 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2505 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2507 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2509 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2512 void sk_common_release(struct sock
*sk
)
2514 if (sk
->sk_prot
->destroy
)
2515 sk
->sk_prot
->destroy(sk
);
2518 * Observation: when sock_common_release is called, processes have
2519 * no access to socket. But net still has.
2520 * Step one, detach it from networking:
2522 * A. Remove from hash tables.
2525 sk
->sk_prot
->unhash(sk
);
2528 * In this point socket cannot receive new packets, but it is possible
2529 * that some packets are in flight because some CPU runs receiver and
2530 * did hash table lookup before we unhashed socket. They will achieve
2531 * receive queue and will be purged by socket destructor.
2533 * Also we still have packets pending on receive queue and probably,
2534 * our own packets waiting in device queues. sock_destroy will drain
2535 * receive queue, but transmitted packets will delay socket destruction
2536 * until the last reference will be released.
2541 xfrm_sk_free_policy(sk
);
2543 sk_refcnt_debug_release(sk
);
2545 if (sk
->sk_frag
.page
) {
2546 put_page(sk
->sk_frag
.page
);
2547 sk
->sk_frag
.page
= NULL
;
2552 EXPORT_SYMBOL(sk_common_release
);
2554 #ifdef CONFIG_PROC_FS
2555 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2557 int val
[PROTO_INUSE_NR
];
2560 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2562 #ifdef CONFIG_NET_NS
2563 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2565 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2567 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2569 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2571 int cpu
, idx
= prot
->inuse_idx
;
2574 for_each_possible_cpu(cpu
)
2575 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2577 return res
>= 0 ? res
: 0;
2579 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2581 static int __net_init
sock_inuse_init_net(struct net
*net
)
2583 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2584 return net
->core
.inuse
? 0 : -ENOMEM
;
2587 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2589 free_percpu(net
->core
.inuse
);
2592 static struct pernet_operations net_inuse_ops
= {
2593 .init
= sock_inuse_init_net
,
2594 .exit
= sock_inuse_exit_net
,
2597 static __init
int net_inuse_init(void)
2599 if (register_pernet_subsys(&net_inuse_ops
))
2600 panic("Cannot initialize net inuse counters");
2605 core_initcall(net_inuse_init
);
2607 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2609 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2611 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2613 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2615 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2617 int cpu
, idx
= prot
->inuse_idx
;
2620 for_each_possible_cpu(cpu
)
2621 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2623 return res
>= 0 ? res
: 0;
2625 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2628 static void assign_proto_idx(struct proto
*prot
)
2630 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2632 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2633 pr_err("PROTO_INUSE_NR exhausted\n");
2637 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2640 static void release_proto_idx(struct proto
*prot
)
2642 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2643 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2646 static inline void assign_proto_idx(struct proto
*prot
)
2650 static inline void release_proto_idx(struct proto
*prot
)
2655 int proto_register(struct proto
*prot
, int alloc_slab
)
2658 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2659 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2662 if (prot
->slab
== NULL
) {
2663 pr_crit("%s: Can't create sock SLAB cache!\n",
2668 if (prot
->rsk_prot
!= NULL
) {
2669 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2670 if (prot
->rsk_prot
->slab_name
== NULL
)
2671 goto out_free_sock_slab
;
2673 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2674 prot
->rsk_prot
->obj_size
, 0,
2675 SLAB_HWCACHE_ALIGN
, NULL
);
2677 if (prot
->rsk_prot
->slab
== NULL
) {
2678 pr_crit("%s: Can't create request sock SLAB cache!\n",
2680 goto out_free_request_sock_slab_name
;
2684 if (prot
->twsk_prot
!= NULL
) {
2685 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2687 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2688 goto out_free_request_sock_slab
;
2690 prot
->twsk_prot
->twsk_slab
=
2691 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2692 prot
->twsk_prot
->twsk_obj_size
,
2694 SLAB_HWCACHE_ALIGN
|
2697 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2698 goto out_free_timewait_sock_slab_name
;
2702 mutex_lock(&proto_list_mutex
);
2703 list_add(&prot
->node
, &proto_list
);
2704 assign_proto_idx(prot
);
2705 mutex_unlock(&proto_list_mutex
);
2708 out_free_timewait_sock_slab_name
:
2709 kfree(prot
->twsk_prot
->twsk_slab_name
);
2710 out_free_request_sock_slab
:
2711 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2712 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2713 prot
->rsk_prot
->slab
= NULL
;
2715 out_free_request_sock_slab_name
:
2717 kfree(prot
->rsk_prot
->slab_name
);
2719 kmem_cache_destroy(prot
->slab
);
2724 EXPORT_SYMBOL(proto_register
);
2726 void proto_unregister(struct proto
*prot
)
2728 mutex_lock(&proto_list_mutex
);
2729 release_proto_idx(prot
);
2730 list_del(&prot
->node
);
2731 mutex_unlock(&proto_list_mutex
);
2733 if (prot
->slab
!= NULL
) {
2734 kmem_cache_destroy(prot
->slab
);
2738 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2739 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2740 kfree(prot
->rsk_prot
->slab_name
);
2741 prot
->rsk_prot
->slab
= NULL
;
2744 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2745 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2746 kfree(prot
->twsk_prot
->twsk_slab_name
);
2747 prot
->twsk_prot
->twsk_slab
= NULL
;
2750 EXPORT_SYMBOL(proto_unregister
);
2752 #ifdef CONFIG_PROC_FS
2753 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2754 __acquires(proto_list_mutex
)
2756 mutex_lock(&proto_list_mutex
);
2757 return seq_list_start_head(&proto_list
, *pos
);
2760 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2762 return seq_list_next(v
, &proto_list
, pos
);
2765 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2766 __releases(proto_list_mutex
)
2768 mutex_unlock(&proto_list_mutex
);
2771 static char proto_method_implemented(const void *method
)
2773 return method
== NULL
? 'n' : 'y';
2775 static long sock_prot_memory_allocated(struct proto
*proto
)
2777 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2780 static char *sock_prot_memory_pressure(struct proto
*proto
)
2782 return proto
->memory_pressure
!= NULL
?
2783 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2786 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2789 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2790 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2793 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2794 sock_prot_memory_allocated(proto
),
2795 sock_prot_memory_pressure(proto
),
2797 proto
->slab
== NULL
? "no" : "yes",
2798 module_name(proto
->owner
),
2799 proto_method_implemented(proto
->close
),
2800 proto_method_implemented(proto
->connect
),
2801 proto_method_implemented(proto
->disconnect
),
2802 proto_method_implemented(proto
->accept
),
2803 proto_method_implemented(proto
->ioctl
),
2804 proto_method_implemented(proto
->init
),
2805 proto_method_implemented(proto
->destroy
),
2806 proto_method_implemented(proto
->shutdown
),
2807 proto_method_implemented(proto
->setsockopt
),
2808 proto_method_implemented(proto
->getsockopt
),
2809 proto_method_implemented(proto
->sendmsg
),
2810 proto_method_implemented(proto
->recvmsg
),
2811 proto_method_implemented(proto
->sendpage
),
2812 proto_method_implemented(proto
->bind
),
2813 proto_method_implemented(proto
->backlog_rcv
),
2814 proto_method_implemented(proto
->hash
),
2815 proto_method_implemented(proto
->unhash
),
2816 proto_method_implemented(proto
->get_port
),
2817 proto_method_implemented(proto
->enter_memory_pressure
));
2820 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2822 if (v
== &proto_list
)
2823 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2832 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2834 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2838 static const struct seq_operations proto_seq_ops
= {
2839 .start
= proto_seq_start
,
2840 .next
= proto_seq_next
,
2841 .stop
= proto_seq_stop
,
2842 .show
= proto_seq_show
,
2845 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2847 return seq_open_net(inode
, file
, &proto_seq_ops
,
2848 sizeof(struct seq_net_private
));
2851 static const struct file_operations proto_seq_fops
= {
2852 .owner
= THIS_MODULE
,
2853 .open
= proto_seq_open
,
2855 .llseek
= seq_lseek
,
2856 .release
= seq_release_net
,
2859 static __net_init
int proto_init_net(struct net
*net
)
2861 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
2867 static __net_exit
void proto_exit_net(struct net
*net
)
2869 remove_proc_entry("protocols", net
->proc_net
);
2873 static __net_initdata
struct pernet_operations proto_net_ops
= {
2874 .init
= proto_init_net
,
2875 .exit
= proto_exit_net
,
2878 static int __init
proto_init(void)
2880 return register_pernet_subsys(&proto_net_ops
);
2883 subsys_initcall(proto_init
);
2885 #endif /* PROC_FS */