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>
135 #include <linux/filter.h>
137 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex
);
146 static LIST_HEAD(proto_list
);
148 #ifdef CONFIG_MEMCG_KMEM
149 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
154 mutex_lock(&proto_list_mutex
);
155 list_for_each_entry(proto
, &proto_list
, node
) {
156 if (proto
->init_cgroup
) {
157 ret
= proto
->init_cgroup(memcg
, ss
);
163 mutex_unlock(&proto_list_mutex
);
166 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
167 if (proto
->destroy_cgroup
)
168 proto
->destroy_cgroup(memcg
);
169 mutex_unlock(&proto_list_mutex
);
173 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
177 mutex_lock(&proto_list_mutex
);
178 list_for_each_entry_reverse(proto
, &proto_list
, node
)
179 if (proto
->destroy_cgroup
)
180 proto
->destroy_cgroup(memcg
);
181 mutex_unlock(&proto_list_mutex
);
186 * Each address family might have different locking rules, so we have
187 * one slock key per address family:
189 static struct lock_class_key af_family_keys
[AF_MAX
];
190 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
192 #if defined(CONFIG_MEMCG_KMEM)
193 struct static_key memcg_socket_limit_enabled
;
194 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
198 * Make lock validator output more readable. (we pre-construct these
199 * strings build-time, so that runtime initialization of socket
202 static const char *const af_family_key_strings
[AF_MAX
+1] = {
203 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
204 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
205 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
206 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
207 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
208 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
209 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
210 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
211 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
212 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
213 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
214 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
215 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
216 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
218 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
219 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
220 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
221 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
222 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
223 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
224 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
225 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
226 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
227 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
228 "slock-27" , "slock-28" , "slock-AF_CAN" ,
229 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
230 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
231 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
232 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
234 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
235 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
236 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
237 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
238 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
239 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
240 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
241 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
242 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
243 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
244 "clock-27" , "clock-28" , "clock-AF_CAN" ,
245 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
246 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
247 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
248 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
252 * sk_callback_lock locking rules are per-address-family,
253 * so split the lock classes by using a per-AF key:
255 static struct lock_class_key af_callback_keys
[AF_MAX
];
257 /* Take into consideration the size of the struct sk_buff overhead in the
258 * determination of these values, since that is non-constant across
259 * platforms. This makes socket queueing behavior and performance
260 * not depend upon such differences.
262 #define _SK_MEM_PACKETS 256
263 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
264 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
265 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
267 /* Run time adjustable parameters. */
268 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
269 EXPORT_SYMBOL(sysctl_wmem_max
);
270 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
271 EXPORT_SYMBOL(sysctl_rmem_max
);
272 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
273 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
275 /* Maximal space eaten by iovec or ancillary data plus some space */
276 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
277 EXPORT_SYMBOL(sysctl_optmem_max
);
279 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
280 EXPORT_SYMBOL_GPL(memalloc_socks
);
283 * sk_set_memalloc - sets %SOCK_MEMALLOC
284 * @sk: socket to set it on
286 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
287 * It's the responsibility of the admin to adjust min_free_kbytes
288 * to meet the requirements
290 void sk_set_memalloc(struct sock
*sk
)
292 sock_set_flag(sk
, SOCK_MEMALLOC
);
293 sk
->sk_allocation
|= __GFP_MEMALLOC
;
294 static_key_slow_inc(&memalloc_socks
);
296 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
298 void sk_clear_memalloc(struct sock
*sk
)
300 sock_reset_flag(sk
, SOCK_MEMALLOC
);
301 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
302 static_key_slow_dec(&memalloc_socks
);
305 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
306 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
307 * it has rmem allocations there is a risk that the user of the
308 * socket cannot make forward progress due to exceeding the rmem
309 * limits. By rights, sk_clear_memalloc() should only be called
310 * on sockets being torn down but warn and reset the accounting if
311 * that assumption breaks.
313 if (WARN_ON(sk
->sk_forward_alloc
))
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
318 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
321 unsigned long pflags
= current
->flags
;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
326 current
->flags
|= PF_MEMALLOC
;
327 ret
= sk
->sk_backlog_rcv(sk
, skb
);
328 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
332 EXPORT_SYMBOL(__sk_backlog_rcv
);
334 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
338 if (optlen
< sizeof(tv
))
340 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
342 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
346 static int warned __read_mostly
;
349 if (warned
< 10 && net_ratelimit()) {
351 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
352 __func__
, current
->comm
, task_pid_nr(current
));
356 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
357 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
359 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
360 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
364 static void sock_warn_obsolete_bsdism(const char *name
)
367 static char warncomm
[TASK_COMM_LEN
];
368 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
369 strcpy(warncomm
, current
->comm
);
370 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
376 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
378 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
380 if (sk
->sk_flags
& flags
) {
381 sk
->sk_flags
&= ~flags
;
382 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
383 net_disable_timestamp();
388 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
393 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
395 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
396 atomic_inc(&sk
->sk_drops
);
397 trace_sock_rcvqueue_full(sk
, skb
);
401 err
= sk_filter(sk
, skb
);
405 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
406 atomic_inc(&sk
->sk_drops
);
411 skb_set_owner_r(skb
, sk
);
413 /* Cache the SKB length before we tack it onto the receive
414 * queue. Once it is added it no longer belongs to us and
415 * may be freed by other threads of control pulling packets
420 /* we escape from rcu protected region, make sure we dont leak
425 spin_lock_irqsave(&list
->lock
, flags
);
426 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
427 __skb_queue_tail(list
, skb
);
428 spin_unlock_irqrestore(&list
->lock
, flags
);
430 if (!sock_flag(sk
, SOCK_DEAD
))
431 sk
->sk_data_ready(sk
, skb_len
);
434 EXPORT_SYMBOL(sock_queue_rcv_skb
);
436 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
438 int rc
= NET_RX_SUCCESS
;
440 if (sk_filter(sk
, skb
))
441 goto discard_and_relse
;
445 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
)) {
446 atomic_inc(&sk
->sk_drops
);
447 goto discard_and_relse
;
450 bh_lock_sock_nested(sk
);
453 if (!sock_owned_by_user(sk
)) {
455 * trylock + unlock semantics:
457 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
459 rc
= sk_backlog_rcv(sk
, skb
);
461 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
462 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
464 atomic_inc(&sk
->sk_drops
);
465 goto discard_and_relse
;
476 EXPORT_SYMBOL(sk_receive_skb
);
478 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
480 struct dst_entry
*dst
= __sk_dst_get(sk
);
482 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
483 sk_tx_queue_clear(sk
);
484 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
491 EXPORT_SYMBOL(__sk_dst_check
);
493 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
495 struct dst_entry
*dst
= sk_dst_get(sk
);
497 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
505 EXPORT_SYMBOL(sk_dst_check
);
507 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
510 int ret
= -ENOPROTOOPT
;
511 #ifdef CONFIG_NETDEVICES
512 struct net
*net
= sock_net(sk
);
513 char devname
[IFNAMSIZ
];
518 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
525 /* Bind this socket to a particular device like "eth0",
526 * as specified in the passed interface name. If the
527 * name is "" or the option length is zero the socket
530 if (optlen
> IFNAMSIZ
- 1)
531 optlen
= IFNAMSIZ
- 1;
532 memset(devname
, 0, sizeof(devname
));
535 if (copy_from_user(devname
, optval
, optlen
))
539 if (devname
[0] != '\0') {
540 struct net_device
*dev
;
543 dev
= dev_get_by_name_rcu(net
, devname
);
545 index
= dev
->ifindex
;
553 sk
->sk_bound_dev_if
= index
;
565 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
566 int __user
*optlen
, int len
)
568 int ret
= -ENOPROTOOPT
;
569 #ifdef CONFIG_NETDEVICES
570 struct net
*net
= sock_net(sk
);
571 char devname
[IFNAMSIZ
];
573 if (sk
->sk_bound_dev_if
== 0) {
582 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
586 len
= strlen(devname
) + 1;
589 if (copy_to_user(optval
, devname
, len
))
594 if (put_user(len
, optlen
))
605 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
608 sock_set_flag(sk
, bit
);
610 sock_reset_flag(sk
, bit
);
614 * This is meant for all protocols to use and covers goings on
615 * at the socket level. Everything here is generic.
618 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
619 char __user
*optval
, unsigned int optlen
)
621 struct sock
*sk
= sock
->sk
;
628 * Options without arguments
631 if (optname
== SO_BINDTODEVICE
)
632 return sock_setbindtodevice(sk
, optval
, optlen
);
634 if (optlen
< sizeof(int))
637 if (get_user(val
, (int __user
*)optval
))
640 valbool
= val
? 1 : 0;
646 if (val
&& !capable(CAP_NET_ADMIN
))
649 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
652 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
655 sk
->sk_reuseport
= valbool
;
664 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
667 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
670 /* Don't error on this BSD doesn't and if you think
671 * about it this is right. Otherwise apps have to
672 * play 'guess the biggest size' games. RCVBUF/SNDBUF
673 * are treated in BSD as hints
675 val
= min_t(u32
, val
, sysctl_wmem_max
);
677 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
678 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
679 /* Wake up sending tasks if we upped the value. */
680 sk
->sk_write_space(sk
);
684 if (!capable(CAP_NET_ADMIN
)) {
691 /* Don't error on this BSD doesn't and if you think
692 * about it this is right. Otherwise apps have to
693 * play 'guess the biggest size' games. RCVBUF/SNDBUF
694 * are treated in BSD as hints
696 val
= min_t(u32
, val
, sysctl_rmem_max
);
698 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
700 * We double it on the way in to account for
701 * "struct sk_buff" etc. overhead. Applications
702 * assume that the SO_RCVBUF setting they make will
703 * allow that much actual data to be received on that
706 * Applications are unaware that "struct sk_buff" and
707 * other overheads allocate from the receive buffer
708 * during socket buffer allocation.
710 * And after considering the possible alternatives,
711 * returning the value we actually used in getsockopt
712 * is the most desirable behavior.
714 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
718 if (!capable(CAP_NET_ADMIN
)) {
726 if (sk
->sk_protocol
== IPPROTO_TCP
&&
727 sk
->sk_type
== SOCK_STREAM
)
728 tcp_set_keepalive(sk
, valbool
);
730 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
734 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
738 sk
->sk_no_check
= valbool
;
742 if ((val
>= 0 && val
<= 6) ||
743 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
744 sk
->sk_priority
= val
;
750 if (optlen
< sizeof(ling
)) {
751 ret
= -EINVAL
; /* 1003.1g */
754 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
759 sock_reset_flag(sk
, SOCK_LINGER
);
761 #if (BITS_PER_LONG == 32)
762 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
763 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
766 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
767 sock_set_flag(sk
, SOCK_LINGER
);
772 sock_warn_obsolete_bsdism("setsockopt");
777 set_bit(SOCK_PASSCRED
, &sock
->flags
);
779 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
785 if (optname
== SO_TIMESTAMP
)
786 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
788 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
789 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
790 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
792 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
793 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
797 case SO_TIMESTAMPING
:
798 if (val
& ~SOF_TIMESTAMPING_MASK
) {
802 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
803 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
804 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
805 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
806 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
807 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
808 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
809 sock_enable_timestamp(sk
,
810 SOCK_TIMESTAMPING_RX_SOFTWARE
);
812 sock_disable_timestamp(sk
,
813 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
814 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
815 val
& SOF_TIMESTAMPING_SOFTWARE
);
816 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
817 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
818 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
819 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
825 sk
->sk_rcvlowat
= val
? : 1;
829 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
833 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
836 case SO_ATTACH_FILTER
:
838 if (optlen
== sizeof(struct sock_fprog
)) {
839 struct sock_fprog fprog
;
842 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
845 ret
= sk_attach_filter(&fprog
, sk
);
849 case SO_DETACH_FILTER
:
850 ret
= sk_detach_filter(sk
);
854 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
857 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
862 set_bit(SOCK_PASSSEC
, &sock
->flags
);
864 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
867 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
873 /* We implement the SO_SNDLOWAT etc to
874 not be settable (1003.1g 5.3) */
876 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
880 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
884 if (sock
->ops
->set_peek_off
)
885 ret
= sock
->ops
->set_peek_off(sk
, val
);
891 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
894 case SO_SELECT_ERR_QUEUE
:
895 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
898 #ifdef CONFIG_NET_RX_BUSY_POLL
900 /* allow unprivileged users to decrease the value */
901 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
907 sk
->sk_ll_usec
= val
;
912 case SO_MAX_PACING_RATE
:
913 sk
->sk_max_pacing_rate
= val
;
914 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
915 sk
->sk_max_pacing_rate
);
925 EXPORT_SYMBOL(sock_setsockopt
);
928 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
931 ucred
->pid
= pid_vnr(pid
);
932 ucred
->uid
= ucred
->gid
= -1;
934 struct user_namespace
*current_ns
= current_user_ns();
936 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
937 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
941 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
942 char __user
*optval
, int __user
*optlen
)
944 struct sock
*sk
= sock
->sk
;
952 int lv
= sizeof(int);
955 if (get_user(len
, optlen
))
960 memset(&v
, 0, sizeof(v
));
964 v
.val
= sock_flag(sk
, SOCK_DBG
);
968 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
972 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
976 v
.val
= sk
->sk_sndbuf
;
980 v
.val
= sk
->sk_rcvbuf
;
984 v
.val
= sk
->sk_reuse
;
988 v
.val
= sk
->sk_reuseport
;
992 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1000 v
.val
= sk
->sk_protocol
;
1004 v
.val
= sk
->sk_family
;
1008 v
.val
= -sock_error(sk
);
1010 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1014 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1018 v
.val
= sk
->sk_no_check
;
1022 v
.val
= sk
->sk_priority
;
1026 lv
= sizeof(v
.ling
);
1027 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1028 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1032 sock_warn_obsolete_bsdism("getsockopt");
1036 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1037 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1040 case SO_TIMESTAMPNS
:
1041 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1044 case SO_TIMESTAMPING
:
1046 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
1047 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
1048 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
1049 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
1050 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
1051 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
1052 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
1053 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
1054 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
1055 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
1056 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
1057 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
1058 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
1059 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
1063 lv
= sizeof(struct timeval
);
1064 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1068 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1069 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1074 lv
= sizeof(struct timeval
);
1075 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1079 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1080 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1085 v
.val
= sk
->sk_rcvlowat
;
1093 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1098 struct ucred peercred
;
1099 if (len
> sizeof(peercred
))
1100 len
= sizeof(peercred
);
1101 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1102 if (copy_to_user(optval
, &peercred
, len
))
1111 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1115 if (copy_to_user(optval
, address
, len
))
1120 /* Dubious BSD thing... Probably nobody even uses it, but
1121 * the UNIX standard wants it for whatever reason... -DaveM
1124 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1128 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1132 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1135 v
.val
= sk
->sk_mark
;
1139 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1142 case SO_WIFI_STATUS
:
1143 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1147 if (!sock
->ops
->set_peek_off
)
1150 v
.val
= sk
->sk_peek_off
;
1153 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1156 case SO_BINDTODEVICE
:
1157 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1160 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1166 case SO_LOCK_FILTER
:
1167 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1170 case SO_SELECT_ERR_QUEUE
:
1171 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1174 #ifdef CONFIG_NET_RX_BUSY_POLL
1176 v
.val
= sk
->sk_ll_usec
;
1180 case SO_MAX_PACING_RATE
:
1181 v
.val
= sk
->sk_max_pacing_rate
;
1185 return -ENOPROTOOPT
;
1190 if (copy_to_user(optval
, &v
, len
))
1193 if (put_user(len
, optlen
))
1199 * Initialize an sk_lock.
1201 * (We also register the sk_lock with the lock validator.)
1203 static inline void sock_lock_init(struct sock
*sk
)
1205 sock_lock_init_class_and_name(sk
,
1206 af_family_slock_key_strings
[sk
->sk_family
],
1207 af_family_slock_keys
+ sk
->sk_family
,
1208 af_family_key_strings
[sk
->sk_family
],
1209 af_family_keys
+ sk
->sk_family
);
1213 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1214 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1215 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1217 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1219 #ifdef CONFIG_SECURITY_NETWORK
1220 void *sptr
= nsk
->sk_security
;
1222 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1224 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1225 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1227 #ifdef CONFIG_SECURITY_NETWORK
1228 nsk
->sk_security
= sptr
;
1229 security_sk_clone(osk
, nsk
);
1233 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1235 unsigned long nulls1
, nulls2
;
1237 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1238 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1239 if (nulls1
> nulls2
)
1240 swap(nulls1
, nulls2
);
1243 memset((char *)sk
, 0, nulls1
);
1244 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1245 nulls2
- nulls1
- sizeof(void *));
1246 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1247 size
- nulls2
- sizeof(void *));
1249 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1251 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1255 struct kmem_cache
*slab
;
1259 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1262 if (priority
& __GFP_ZERO
) {
1264 prot
->clear_sk(sk
, prot
->obj_size
);
1266 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1269 sk
= kmalloc(prot
->obj_size
, priority
);
1272 kmemcheck_annotate_bitfield(sk
, flags
);
1274 if (security_sk_alloc(sk
, family
, priority
))
1277 if (!try_module_get(prot
->owner
))
1279 sk_tx_queue_clear(sk
);
1285 security_sk_free(sk
);
1288 kmem_cache_free(slab
, sk
);
1294 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1296 struct kmem_cache
*slab
;
1297 struct module
*owner
;
1299 owner
= prot
->owner
;
1302 security_sk_free(sk
);
1304 kmem_cache_free(slab
, sk
);
1310 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1311 void sock_update_classid(struct sock
*sk
)
1315 classid
= task_cls_classid(current
);
1316 if (classid
!= sk
->sk_classid
)
1317 sk
->sk_classid
= classid
;
1319 EXPORT_SYMBOL(sock_update_classid
);
1322 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1323 void sock_update_netprioidx(struct sock
*sk
)
1328 sk
->sk_cgrp_prioidx
= task_netprioidx(current
);
1330 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1334 * sk_alloc - All socket objects are allocated here
1335 * @net: the applicable net namespace
1336 * @family: protocol family
1337 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1338 * @prot: struct proto associated with this new sock instance
1340 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1345 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1347 sk
->sk_family
= family
;
1349 * See comment in struct sock definition to understand
1350 * why we need sk_prot_creator -acme
1352 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1354 sock_net_set(sk
, get_net(net
));
1355 atomic_set(&sk
->sk_wmem_alloc
, 1);
1357 sock_update_classid(sk
);
1358 sock_update_netprioidx(sk
);
1363 EXPORT_SYMBOL(sk_alloc
);
1365 static void __sk_free(struct sock
*sk
)
1367 struct sk_filter
*filter
;
1369 if (sk
->sk_destruct
)
1370 sk
->sk_destruct(sk
);
1372 filter
= rcu_dereference_check(sk
->sk_filter
,
1373 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1375 sk_filter_uncharge(sk
, filter
);
1376 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1379 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1381 if (atomic_read(&sk
->sk_omem_alloc
))
1382 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1383 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1385 if (sk
->sk_peer_cred
)
1386 put_cred(sk
->sk_peer_cred
);
1387 put_pid(sk
->sk_peer_pid
);
1388 put_net(sock_net(sk
));
1389 sk_prot_free(sk
->sk_prot_creator
, sk
);
1392 void sk_free(struct sock
*sk
)
1395 * We subtract one from sk_wmem_alloc and can know if
1396 * some packets are still in some tx queue.
1397 * If not null, sock_wfree() will call __sk_free(sk) later
1399 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1402 EXPORT_SYMBOL(sk_free
);
1405 * Last sock_put should drop reference to sk->sk_net. It has already
1406 * been dropped in sk_change_net. Taking reference to stopping namespace
1408 * Take reference to a socket to remove it from hash _alive_ and after that
1409 * destroy it in the context of init_net.
1411 void sk_release_kernel(struct sock
*sk
)
1413 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1417 sock_release(sk
->sk_socket
);
1418 release_net(sock_net(sk
));
1419 sock_net_set(sk
, get_net(&init_net
));
1422 EXPORT_SYMBOL(sk_release_kernel
);
1424 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1426 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1427 sock_update_memcg(newsk
);
1431 * sk_clone_lock - clone a socket, and lock its clone
1432 * @sk: the socket to clone
1433 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1435 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1437 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1441 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1442 if (newsk
!= NULL
) {
1443 struct sk_filter
*filter
;
1445 sock_copy(newsk
, sk
);
1448 get_net(sock_net(newsk
));
1449 sk_node_init(&newsk
->sk_node
);
1450 sock_lock_init(newsk
);
1451 bh_lock_sock(newsk
);
1452 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1453 newsk
->sk_backlog
.len
= 0;
1455 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1457 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1459 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1460 atomic_set(&newsk
->sk_omem_alloc
, 0);
1461 skb_queue_head_init(&newsk
->sk_receive_queue
);
1462 skb_queue_head_init(&newsk
->sk_write_queue
);
1463 #ifdef CONFIG_NET_DMA
1464 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1467 spin_lock_init(&newsk
->sk_dst_lock
);
1468 rwlock_init(&newsk
->sk_callback_lock
);
1469 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1470 af_callback_keys
+ newsk
->sk_family
,
1471 af_family_clock_key_strings
[newsk
->sk_family
]);
1473 newsk
->sk_dst_cache
= NULL
;
1474 newsk
->sk_wmem_queued
= 0;
1475 newsk
->sk_forward_alloc
= 0;
1476 newsk
->sk_send_head
= NULL
;
1477 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1479 sock_reset_flag(newsk
, SOCK_DONE
);
1480 skb_queue_head_init(&newsk
->sk_error_queue
);
1482 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1484 sk_filter_charge(newsk
, filter
);
1486 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1487 /* It is still raw copy of parent, so invalidate
1488 * destructor and make plain sk_free() */
1489 newsk
->sk_destruct
= NULL
;
1490 bh_unlock_sock(newsk
);
1497 newsk
->sk_priority
= 0;
1499 * Before updating sk_refcnt, we must commit prior changes to memory
1500 * (Documentation/RCU/rculist_nulls.txt for details)
1503 atomic_set(&newsk
->sk_refcnt
, 2);
1506 * Increment the counter in the same struct proto as the master
1507 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1508 * is the same as sk->sk_prot->socks, as this field was copied
1511 * This _changes_ the previous behaviour, where
1512 * tcp_create_openreq_child always was incrementing the
1513 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1514 * to be taken into account in all callers. -acme
1516 sk_refcnt_debug_inc(newsk
);
1517 sk_set_socket(newsk
, NULL
);
1518 newsk
->sk_wq
= NULL
;
1520 sk_update_clone(sk
, newsk
);
1522 if (newsk
->sk_prot
->sockets_allocated
)
1523 sk_sockets_allocated_inc(newsk
);
1525 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1526 net_enable_timestamp();
1531 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1533 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1535 __sk_dst_set(sk
, dst
);
1536 sk
->sk_route_caps
= dst
->dev
->features
;
1537 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1538 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1539 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1540 if (sk_can_gso(sk
)) {
1541 if (dst
->header_len
) {
1542 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1544 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1545 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1546 sk
->sk_gso_max_segs
= dst
->dev
->gso_max_segs
;
1550 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1553 * Simple resource managers for sockets.
1558 * Write buffer destructor automatically called from kfree_skb.
1560 void sock_wfree(struct sk_buff
*skb
)
1562 struct sock
*sk
= skb
->sk
;
1563 unsigned int len
= skb
->truesize
;
1565 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1567 * Keep a reference on sk_wmem_alloc, this will be released
1568 * after sk_write_space() call
1570 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1571 sk
->sk_write_space(sk
);
1575 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1576 * could not do because of in-flight packets
1578 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1581 EXPORT_SYMBOL(sock_wfree
);
1583 void skb_orphan_partial(struct sk_buff
*skb
)
1585 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1586 * so we do not completely orphan skb, but transfert all
1587 * accounted bytes but one, to avoid unexpected reorders.
1589 if (skb
->destructor
== sock_wfree
1591 || skb
->destructor
== tcp_wfree
1594 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1600 EXPORT_SYMBOL(skb_orphan_partial
);
1603 * Read buffer destructor automatically called from kfree_skb.
1605 void sock_rfree(struct sk_buff
*skb
)
1607 struct sock
*sk
= skb
->sk
;
1608 unsigned int len
= skb
->truesize
;
1610 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1611 sk_mem_uncharge(sk
, len
);
1613 EXPORT_SYMBOL(sock_rfree
);
1615 void sock_edemux(struct sk_buff
*skb
)
1617 struct sock
*sk
= skb
->sk
;
1620 if (sk
->sk_state
== TCP_TIME_WAIT
)
1621 inet_twsk_put(inet_twsk(sk
));
1626 EXPORT_SYMBOL(sock_edemux
);
1628 kuid_t
sock_i_uid(struct sock
*sk
)
1632 read_lock_bh(&sk
->sk_callback_lock
);
1633 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1634 read_unlock_bh(&sk
->sk_callback_lock
);
1637 EXPORT_SYMBOL(sock_i_uid
);
1639 unsigned long sock_i_ino(struct sock
*sk
)
1643 read_lock_bh(&sk
->sk_callback_lock
);
1644 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1645 read_unlock_bh(&sk
->sk_callback_lock
);
1648 EXPORT_SYMBOL(sock_i_ino
);
1651 * Allocate a skb from the socket's send buffer.
1653 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1656 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1657 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1659 skb_set_owner_w(skb
, sk
);
1665 EXPORT_SYMBOL(sock_wmalloc
);
1668 * Allocate a memory block from the socket's option memory buffer.
1670 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1672 if ((unsigned int)size
<= sysctl_optmem_max
&&
1673 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1675 /* First do the add, to avoid the race if kmalloc
1678 atomic_add(size
, &sk
->sk_omem_alloc
);
1679 mem
= kmalloc(size
, priority
);
1682 atomic_sub(size
, &sk
->sk_omem_alloc
);
1686 EXPORT_SYMBOL(sock_kmalloc
);
1689 * Free an option memory block.
1691 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1694 atomic_sub(size
, &sk
->sk_omem_alloc
);
1696 EXPORT_SYMBOL(sock_kfree_s
);
1698 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1699 I think, these locks should be removed for datagram sockets.
1701 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1705 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1709 if (signal_pending(current
))
1711 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1712 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1713 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1715 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1719 timeo
= schedule_timeout(timeo
);
1721 finish_wait(sk_sleep(sk
), &wait
);
1727 * Generic send/receive buffer handlers
1730 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1731 unsigned long data_len
, int noblock
,
1732 int *errcode
, int max_page_order
)
1734 struct sk_buff
*skb
= NULL
;
1735 unsigned long chunk
;
1739 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1744 if (npages
> MAX_SKB_FRAGS
)
1747 timeo
= sock_sndtimeo(sk
, noblock
);
1749 err
= sock_error(sk
);
1754 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1757 if (atomic_read(&sk
->sk_wmem_alloc
) >= sk
->sk_sndbuf
) {
1758 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1759 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1763 if (signal_pending(current
))
1765 timeo
= sock_wait_for_wmem(sk
, timeo
);
1770 gfp_mask
= sk
->sk_allocation
;
1771 if (gfp_mask
& __GFP_WAIT
)
1772 gfp_mask
|= __GFP_REPEAT
;
1774 skb
= alloc_skb(header_len
, gfp_mask
);
1778 skb
->truesize
+= data_len
;
1780 for (i
= 0; npages
> 0; i
++) {
1781 int order
= max_page_order
;
1784 if (npages
>= 1 << order
) {
1785 page
= alloc_pages(sk
->sk_allocation
|
1786 __GFP_COMP
| __GFP_NOWARN
,
1793 page
= alloc_page(sk
->sk_allocation
);
1797 chunk
= min_t(unsigned long, data_len
,
1798 PAGE_SIZE
<< order
);
1799 skb_fill_page_desc(skb
, i
, page
, 0, chunk
);
1801 npages
-= 1 << order
;
1805 skb_set_owner_w(skb
, sk
);
1809 err
= sock_intr_errno(timeo
);
1815 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1817 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1818 int noblock
, int *errcode
)
1820 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1822 EXPORT_SYMBOL(sock_alloc_send_skb
);
1824 /* On 32bit arches, an skb frag is limited to 2^15 */
1825 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1828 * skb_page_frag_refill - check that a page_frag contains enough room
1829 * @sz: minimum size of the fragment we want to get
1830 * @pfrag: pointer to page_frag
1831 * @prio: priority for memory allocation
1833 * Note: While this allocator tries to use high order pages, there is
1834 * no guarantee that allocations succeed. Therefore, @sz MUST be
1835 * less or equal than PAGE_SIZE.
1837 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t prio
)
1842 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1846 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1848 put_page(pfrag
->page
);
1851 /* We restrict high order allocations to users that can afford to wait */
1852 order
= (prio
& __GFP_WAIT
) ? SKB_FRAG_PAGE_ORDER
: 0;
1858 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
1859 pfrag
->page
= alloc_pages(gfp
, order
);
1860 if (likely(pfrag
->page
)) {
1862 pfrag
->size
= PAGE_SIZE
<< order
;
1865 } while (--order
>= 0);
1869 EXPORT_SYMBOL(skb_page_frag_refill
);
1871 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1873 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
1876 sk_enter_memory_pressure(sk
);
1877 sk_stream_moderate_sndbuf(sk
);
1880 EXPORT_SYMBOL(sk_page_frag_refill
);
1882 static void __lock_sock(struct sock
*sk
)
1883 __releases(&sk
->sk_lock
.slock
)
1884 __acquires(&sk
->sk_lock
.slock
)
1889 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1890 TASK_UNINTERRUPTIBLE
);
1891 spin_unlock_bh(&sk
->sk_lock
.slock
);
1893 spin_lock_bh(&sk
->sk_lock
.slock
);
1894 if (!sock_owned_by_user(sk
))
1897 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1900 static void __release_sock(struct sock
*sk
)
1901 __releases(&sk
->sk_lock
.slock
)
1902 __acquires(&sk
->sk_lock
.slock
)
1904 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1907 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1911 struct sk_buff
*next
= skb
->next
;
1914 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1916 sk_backlog_rcv(sk
, skb
);
1919 * We are in process context here with softirqs
1920 * disabled, use cond_resched_softirq() to preempt.
1921 * This is safe to do because we've taken the backlog
1924 cond_resched_softirq();
1927 } while (skb
!= NULL
);
1930 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1933 * Doing the zeroing here guarantee we can not loop forever
1934 * while a wild producer attempts to flood us.
1936 sk
->sk_backlog
.len
= 0;
1940 * sk_wait_data - wait for data to arrive at sk_receive_queue
1941 * @sk: sock to wait on
1942 * @timeo: for how long
1944 * Now socket state including sk->sk_err is changed only under lock,
1945 * hence we may omit checks after joining wait queue.
1946 * We check receive queue before schedule() only as optimization;
1947 * it is very likely that release_sock() added new data.
1949 int sk_wait_data(struct sock
*sk
, long *timeo
)
1954 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1955 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1956 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1957 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1958 finish_wait(sk_sleep(sk
), &wait
);
1961 EXPORT_SYMBOL(sk_wait_data
);
1964 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1966 * @size: memory size to allocate
1967 * @kind: allocation type
1969 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1970 * rmem allocation. This function assumes that protocols which have
1971 * memory_pressure use sk_wmem_queued as write buffer accounting.
1973 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
1975 struct proto
*prot
= sk
->sk_prot
;
1976 int amt
= sk_mem_pages(size
);
1978 int parent_status
= UNDER_LIMIT
;
1980 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
1982 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
1985 if (parent_status
== UNDER_LIMIT
&&
1986 allocated
<= sk_prot_mem_limits(sk
, 0)) {
1987 sk_leave_memory_pressure(sk
);
1991 /* Under pressure. (we or our parents) */
1992 if ((parent_status
> SOFT_LIMIT
) ||
1993 allocated
> sk_prot_mem_limits(sk
, 1))
1994 sk_enter_memory_pressure(sk
);
1996 /* Over hard limit (we or our parents) */
1997 if ((parent_status
== OVER_LIMIT
) ||
1998 (allocated
> sk_prot_mem_limits(sk
, 2)))
1999 goto suppress_allocation
;
2001 /* guarantee minimum buffer size under pressure */
2002 if (kind
== SK_MEM_RECV
) {
2003 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2006 } else { /* SK_MEM_SEND */
2007 if (sk
->sk_type
== SOCK_STREAM
) {
2008 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2010 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2011 prot
->sysctl_wmem
[0])
2015 if (sk_has_memory_pressure(sk
)) {
2018 if (!sk_under_memory_pressure(sk
))
2020 alloc
= sk_sockets_allocated_read_positive(sk
);
2021 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2022 sk_mem_pages(sk
->sk_wmem_queued
+
2023 atomic_read(&sk
->sk_rmem_alloc
) +
2024 sk
->sk_forward_alloc
))
2028 suppress_allocation
:
2030 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2031 sk_stream_moderate_sndbuf(sk
);
2033 /* Fail only if socket is _under_ its sndbuf.
2034 * In this case we cannot block, so that we have to fail.
2036 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2040 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2042 /* Alas. Undo changes. */
2043 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2045 sk_memory_allocated_sub(sk
, amt
);
2049 EXPORT_SYMBOL(__sk_mem_schedule
);
2052 * __sk_reclaim - reclaim memory_allocated
2055 void __sk_mem_reclaim(struct sock
*sk
)
2057 sk_memory_allocated_sub(sk
,
2058 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
2059 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
2061 if (sk_under_memory_pressure(sk
) &&
2062 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2063 sk_leave_memory_pressure(sk
);
2065 EXPORT_SYMBOL(__sk_mem_reclaim
);
2069 * Set of default routines for initialising struct proto_ops when
2070 * the protocol does not support a particular function. In certain
2071 * cases where it makes no sense for a protocol to have a "do nothing"
2072 * function, some default processing is provided.
2075 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2079 EXPORT_SYMBOL(sock_no_bind
);
2081 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2086 EXPORT_SYMBOL(sock_no_connect
);
2088 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2092 EXPORT_SYMBOL(sock_no_socketpair
);
2094 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2098 EXPORT_SYMBOL(sock_no_accept
);
2100 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2105 EXPORT_SYMBOL(sock_no_getname
);
2107 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2111 EXPORT_SYMBOL(sock_no_poll
);
2113 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2117 EXPORT_SYMBOL(sock_no_ioctl
);
2119 int sock_no_listen(struct socket
*sock
, int backlog
)
2123 EXPORT_SYMBOL(sock_no_listen
);
2125 int sock_no_shutdown(struct socket
*sock
, int how
)
2129 EXPORT_SYMBOL(sock_no_shutdown
);
2131 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2132 char __user
*optval
, unsigned int optlen
)
2136 EXPORT_SYMBOL(sock_no_setsockopt
);
2138 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2139 char __user
*optval
, int __user
*optlen
)
2143 EXPORT_SYMBOL(sock_no_getsockopt
);
2145 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2150 EXPORT_SYMBOL(sock_no_sendmsg
);
2152 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
2153 size_t len
, int flags
)
2157 EXPORT_SYMBOL(sock_no_recvmsg
);
2159 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2161 /* Mirror missing mmap method error code */
2164 EXPORT_SYMBOL(sock_no_mmap
);
2166 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2169 struct msghdr msg
= {.msg_flags
= flags
};
2171 char *kaddr
= kmap(page
);
2172 iov
.iov_base
= kaddr
+ offset
;
2174 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2178 EXPORT_SYMBOL(sock_no_sendpage
);
2181 * Default Socket Callbacks
2184 static void sock_def_wakeup(struct sock
*sk
)
2186 struct socket_wq
*wq
;
2189 wq
= rcu_dereference(sk
->sk_wq
);
2190 if (wq_has_sleeper(wq
))
2191 wake_up_interruptible_all(&wq
->wait
);
2195 static void sock_def_error_report(struct sock
*sk
)
2197 struct socket_wq
*wq
;
2200 wq
= rcu_dereference(sk
->sk_wq
);
2201 if (wq_has_sleeper(wq
))
2202 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2203 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2207 static void sock_def_readable(struct sock
*sk
, int len
)
2209 struct socket_wq
*wq
;
2212 wq
= rcu_dereference(sk
->sk_wq
);
2213 if (wq_has_sleeper(wq
))
2214 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2215 POLLRDNORM
| POLLRDBAND
);
2216 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2220 static void sock_def_write_space(struct sock
*sk
)
2222 struct socket_wq
*wq
;
2226 /* Do not wake up a writer until he can make "significant"
2229 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2230 wq
= rcu_dereference(sk
->sk_wq
);
2231 if (wq_has_sleeper(wq
))
2232 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2233 POLLWRNORM
| POLLWRBAND
);
2235 /* Should agree with poll, otherwise some programs break */
2236 if (sock_writeable(sk
))
2237 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2243 static void sock_def_destruct(struct sock
*sk
)
2245 kfree(sk
->sk_protinfo
);
2248 void sk_send_sigurg(struct sock
*sk
)
2250 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2251 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2252 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2254 EXPORT_SYMBOL(sk_send_sigurg
);
2256 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2257 unsigned long expires
)
2259 if (!mod_timer(timer
, expires
))
2262 EXPORT_SYMBOL(sk_reset_timer
);
2264 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2266 if (del_timer(timer
))
2269 EXPORT_SYMBOL(sk_stop_timer
);
2271 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2273 skb_queue_head_init(&sk
->sk_receive_queue
);
2274 skb_queue_head_init(&sk
->sk_write_queue
);
2275 skb_queue_head_init(&sk
->sk_error_queue
);
2276 #ifdef CONFIG_NET_DMA
2277 skb_queue_head_init(&sk
->sk_async_wait_queue
);
2280 sk
->sk_send_head
= NULL
;
2282 init_timer(&sk
->sk_timer
);
2284 sk
->sk_allocation
= GFP_KERNEL
;
2285 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2286 sk
->sk_sndbuf
= sysctl_wmem_default
;
2287 sk
->sk_state
= TCP_CLOSE
;
2288 sk_set_socket(sk
, sock
);
2290 sock_set_flag(sk
, SOCK_ZAPPED
);
2293 sk
->sk_type
= sock
->type
;
2294 sk
->sk_wq
= sock
->wq
;
2299 spin_lock_init(&sk
->sk_dst_lock
);
2300 rwlock_init(&sk
->sk_callback_lock
);
2301 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2302 af_callback_keys
+ sk
->sk_family
,
2303 af_family_clock_key_strings
[sk
->sk_family
]);
2305 sk
->sk_state_change
= sock_def_wakeup
;
2306 sk
->sk_data_ready
= sock_def_readable
;
2307 sk
->sk_write_space
= sock_def_write_space
;
2308 sk
->sk_error_report
= sock_def_error_report
;
2309 sk
->sk_destruct
= sock_def_destruct
;
2311 sk
->sk_frag
.page
= NULL
;
2312 sk
->sk_frag
.offset
= 0;
2313 sk
->sk_peek_off
= -1;
2315 sk
->sk_peer_pid
= NULL
;
2316 sk
->sk_peer_cred
= NULL
;
2317 sk
->sk_write_pending
= 0;
2318 sk
->sk_rcvlowat
= 1;
2319 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2320 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2322 sk
->sk_stamp
= ktime_set(-1L, 0);
2324 #ifdef CONFIG_NET_RX_BUSY_POLL
2326 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2329 sk
->sk_max_pacing_rate
= ~0U;
2330 sk
->sk_pacing_rate
= ~0U;
2332 * Before updating sk_refcnt, we must commit prior changes to memory
2333 * (Documentation/RCU/rculist_nulls.txt for details)
2336 atomic_set(&sk
->sk_refcnt
, 1);
2337 atomic_set(&sk
->sk_drops
, 0);
2339 EXPORT_SYMBOL(sock_init_data
);
2341 void lock_sock_nested(struct sock
*sk
, int subclass
)
2344 spin_lock_bh(&sk
->sk_lock
.slock
);
2345 if (sk
->sk_lock
.owned
)
2347 sk
->sk_lock
.owned
= 1;
2348 spin_unlock(&sk
->sk_lock
.slock
);
2350 * The sk_lock has mutex_lock() semantics here:
2352 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2355 EXPORT_SYMBOL(lock_sock_nested
);
2357 void release_sock(struct sock
*sk
)
2360 * The sk_lock has mutex_unlock() semantics:
2362 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2364 spin_lock_bh(&sk
->sk_lock
.slock
);
2365 if (sk
->sk_backlog
.tail
)
2368 if (sk
->sk_prot
->release_cb
)
2369 sk
->sk_prot
->release_cb(sk
);
2371 sk
->sk_lock
.owned
= 0;
2372 if (waitqueue_active(&sk
->sk_lock
.wq
))
2373 wake_up(&sk
->sk_lock
.wq
);
2374 spin_unlock_bh(&sk
->sk_lock
.slock
);
2376 EXPORT_SYMBOL(release_sock
);
2379 * lock_sock_fast - fast version of lock_sock
2382 * This version should be used for very small section, where process wont block
2383 * return false if fast path is taken
2384 * sk_lock.slock locked, owned = 0, BH disabled
2385 * return true if slow path is taken
2386 * sk_lock.slock unlocked, owned = 1, BH enabled
2388 bool lock_sock_fast(struct sock
*sk
)
2391 spin_lock_bh(&sk
->sk_lock
.slock
);
2393 if (!sk
->sk_lock
.owned
)
2395 * Note : We must disable BH
2400 sk
->sk_lock
.owned
= 1;
2401 spin_unlock(&sk
->sk_lock
.slock
);
2403 * The sk_lock has mutex_lock() semantics here:
2405 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2409 EXPORT_SYMBOL(lock_sock_fast
);
2411 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2414 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2415 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2416 tv
= ktime_to_timeval(sk
->sk_stamp
);
2417 if (tv
.tv_sec
== -1)
2419 if (tv
.tv_sec
== 0) {
2420 sk
->sk_stamp
= ktime_get_real();
2421 tv
= ktime_to_timeval(sk
->sk_stamp
);
2423 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2425 EXPORT_SYMBOL(sock_get_timestamp
);
2427 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2430 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2431 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2432 ts
= ktime_to_timespec(sk
->sk_stamp
);
2433 if (ts
.tv_sec
== -1)
2435 if (ts
.tv_sec
== 0) {
2436 sk
->sk_stamp
= ktime_get_real();
2437 ts
= ktime_to_timespec(sk
->sk_stamp
);
2439 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2441 EXPORT_SYMBOL(sock_get_timestampns
);
2443 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2445 if (!sock_flag(sk
, flag
)) {
2446 unsigned long previous_flags
= sk
->sk_flags
;
2448 sock_set_flag(sk
, flag
);
2450 * we just set one of the two flags which require net
2451 * time stamping, but time stamping might have been on
2452 * already because of the other one
2454 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2455 net_enable_timestamp();
2459 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2460 int level
, int type
)
2462 struct sock_exterr_skb
*serr
;
2463 struct sk_buff
*skb
, *skb2
;
2467 skb
= skb_dequeue(&sk
->sk_error_queue
);
2473 msg
->msg_flags
|= MSG_TRUNC
;
2476 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2480 sock_recv_timestamp(msg
, sk
, skb
);
2482 serr
= SKB_EXT_ERR(skb
);
2483 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2485 msg
->msg_flags
|= MSG_ERRQUEUE
;
2488 /* Reset and regenerate socket error */
2489 spin_lock_bh(&sk
->sk_error_queue
.lock
);
2491 if ((skb2
= skb_peek(&sk
->sk_error_queue
)) != NULL
) {
2492 sk
->sk_err
= SKB_EXT_ERR(skb2
)->ee
.ee_errno
;
2493 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2494 sk
->sk_error_report(sk
);
2496 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2503 EXPORT_SYMBOL(sock_recv_errqueue
);
2506 * Get a socket option on an socket.
2508 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2509 * asynchronous errors should be reported by getsockopt. We assume
2510 * this means if you specify SO_ERROR (otherwise whats the point of it).
2512 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2513 char __user
*optval
, int __user
*optlen
)
2515 struct sock
*sk
= sock
->sk
;
2517 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2519 EXPORT_SYMBOL(sock_common_getsockopt
);
2521 #ifdef CONFIG_COMPAT
2522 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2523 char __user
*optval
, int __user
*optlen
)
2525 struct sock
*sk
= sock
->sk
;
2527 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2528 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2530 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2532 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2535 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2536 struct msghdr
*msg
, size_t size
, int flags
)
2538 struct sock
*sk
= sock
->sk
;
2542 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2543 flags
& ~MSG_DONTWAIT
, &addr_len
);
2545 msg
->msg_namelen
= addr_len
;
2548 EXPORT_SYMBOL(sock_common_recvmsg
);
2551 * Set socket options on an inet socket.
2553 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2554 char __user
*optval
, unsigned int optlen
)
2556 struct sock
*sk
= sock
->sk
;
2558 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2560 EXPORT_SYMBOL(sock_common_setsockopt
);
2562 #ifdef CONFIG_COMPAT
2563 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2564 char __user
*optval
, unsigned int optlen
)
2566 struct sock
*sk
= sock
->sk
;
2568 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2569 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2571 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2573 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2576 void sk_common_release(struct sock
*sk
)
2578 if (sk
->sk_prot
->destroy
)
2579 sk
->sk_prot
->destroy(sk
);
2582 * Observation: when sock_common_release is called, processes have
2583 * no access to socket. But net still has.
2584 * Step one, detach it from networking:
2586 * A. Remove from hash tables.
2589 sk
->sk_prot
->unhash(sk
);
2592 * In this point socket cannot receive new packets, but it is possible
2593 * that some packets are in flight because some CPU runs receiver and
2594 * did hash table lookup before we unhashed socket. They will achieve
2595 * receive queue and will be purged by socket destructor.
2597 * Also we still have packets pending on receive queue and probably,
2598 * our own packets waiting in device queues. sock_destroy will drain
2599 * receive queue, but transmitted packets will delay socket destruction
2600 * until the last reference will be released.
2605 xfrm_sk_free_policy(sk
);
2607 sk_refcnt_debug_release(sk
);
2609 if (sk
->sk_frag
.page
) {
2610 put_page(sk
->sk_frag
.page
);
2611 sk
->sk_frag
.page
= NULL
;
2616 EXPORT_SYMBOL(sk_common_release
);
2618 #ifdef CONFIG_PROC_FS
2619 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2621 int val
[PROTO_INUSE_NR
];
2624 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2626 #ifdef CONFIG_NET_NS
2627 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2629 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2631 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2633 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2635 int cpu
, idx
= prot
->inuse_idx
;
2638 for_each_possible_cpu(cpu
)
2639 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2641 return res
>= 0 ? res
: 0;
2643 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2645 static int __net_init
sock_inuse_init_net(struct net
*net
)
2647 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2648 return net
->core
.inuse
? 0 : -ENOMEM
;
2651 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2653 free_percpu(net
->core
.inuse
);
2656 static struct pernet_operations net_inuse_ops
= {
2657 .init
= sock_inuse_init_net
,
2658 .exit
= sock_inuse_exit_net
,
2661 static __init
int net_inuse_init(void)
2663 if (register_pernet_subsys(&net_inuse_ops
))
2664 panic("Cannot initialize net inuse counters");
2669 core_initcall(net_inuse_init
);
2671 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2673 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2675 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2677 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2679 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2681 int cpu
, idx
= prot
->inuse_idx
;
2684 for_each_possible_cpu(cpu
)
2685 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2687 return res
>= 0 ? res
: 0;
2689 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2692 static void assign_proto_idx(struct proto
*prot
)
2694 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2696 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2697 pr_err("PROTO_INUSE_NR exhausted\n");
2701 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2704 static void release_proto_idx(struct proto
*prot
)
2706 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2707 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2710 static inline void assign_proto_idx(struct proto
*prot
)
2714 static inline void release_proto_idx(struct proto
*prot
)
2719 int proto_register(struct proto
*prot
, int alloc_slab
)
2722 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2723 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2726 if (prot
->slab
== NULL
) {
2727 pr_crit("%s: Can't create sock SLAB cache!\n",
2732 if (prot
->rsk_prot
!= NULL
) {
2733 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2734 if (prot
->rsk_prot
->slab_name
== NULL
)
2735 goto out_free_sock_slab
;
2737 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2738 prot
->rsk_prot
->obj_size
, 0,
2739 SLAB_HWCACHE_ALIGN
, NULL
);
2741 if (prot
->rsk_prot
->slab
== NULL
) {
2742 pr_crit("%s: Can't create request sock SLAB cache!\n",
2744 goto out_free_request_sock_slab_name
;
2748 if (prot
->twsk_prot
!= NULL
) {
2749 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2751 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2752 goto out_free_request_sock_slab
;
2754 prot
->twsk_prot
->twsk_slab
=
2755 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2756 prot
->twsk_prot
->twsk_obj_size
,
2758 SLAB_HWCACHE_ALIGN
|
2761 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2762 goto out_free_timewait_sock_slab_name
;
2766 mutex_lock(&proto_list_mutex
);
2767 list_add(&prot
->node
, &proto_list
);
2768 assign_proto_idx(prot
);
2769 mutex_unlock(&proto_list_mutex
);
2772 out_free_timewait_sock_slab_name
:
2773 kfree(prot
->twsk_prot
->twsk_slab_name
);
2774 out_free_request_sock_slab
:
2775 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2776 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2777 prot
->rsk_prot
->slab
= NULL
;
2779 out_free_request_sock_slab_name
:
2781 kfree(prot
->rsk_prot
->slab_name
);
2783 kmem_cache_destroy(prot
->slab
);
2788 EXPORT_SYMBOL(proto_register
);
2790 void proto_unregister(struct proto
*prot
)
2792 mutex_lock(&proto_list_mutex
);
2793 release_proto_idx(prot
);
2794 list_del(&prot
->node
);
2795 mutex_unlock(&proto_list_mutex
);
2797 if (prot
->slab
!= NULL
) {
2798 kmem_cache_destroy(prot
->slab
);
2802 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2803 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2804 kfree(prot
->rsk_prot
->slab_name
);
2805 prot
->rsk_prot
->slab
= NULL
;
2808 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2809 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2810 kfree(prot
->twsk_prot
->twsk_slab_name
);
2811 prot
->twsk_prot
->twsk_slab
= NULL
;
2814 EXPORT_SYMBOL(proto_unregister
);
2816 #ifdef CONFIG_PROC_FS
2817 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2818 __acquires(proto_list_mutex
)
2820 mutex_lock(&proto_list_mutex
);
2821 return seq_list_start_head(&proto_list
, *pos
);
2824 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2826 return seq_list_next(v
, &proto_list
, pos
);
2829 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2830 __releases(proto_list_mutex
)
2832 mutex_unlock(&proto_list_mutex
);
2835 static char proto_method_implemented(const void *method
)
2837 return method
== NULL
? 'n' : 'y';
2839 static long sock_prot_memory_allocated(struct proto
*proto
)
2841 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2844 static char *sock_prot_memory_pressure(struct proto
*proto
)
2846 return proto
->memory_pressure
!= NULL
?
2847 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2850 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2853 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2854 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2857 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2858 sock_prot_memory_allocated(proto
),
2859 sock_prot_memory_pressure(proto
),
2861 proto
->slab
== NULL
? "no" : "yes",
2862 module_name(proto
->owner
),
2863 proto_method_implemented(proto
->close
),
2864 proto_method_implemented(proto
->connect
),
2865 proto_method_implemented(proto
->disconnect
),
2866 proto_method_implemented(proto
->accept
),
2867 proto_method_implemented(proto
->ioctl
),
2868 proto_method_implemented(proto
->init
),
2869 proto_method_implemented(proto
->destroy
),
2870 proto_method_implemented(proto
->shutdown
),
2871 proto_method_implemented(proto
->setsockopt
),
2872 proto_method_implemented(proto
->getsockopt
),
2873 proto_method_implemented(proto
->sendmsg
),
2874 proto_method_implemented(proto
->recvmsg
),
2875 proto_method_implemented(proto
->sendpage
),
2876 proto_method_implemented(proto
->bind
),
2877 proto_method_implemented(proto
->backlog_rcv
),
2878 proto_method_implemented(proto
->hash
),
2879 proto_method_implemented(proto
->unhash
),
2880 proto_method_implemented(proto
->get_port
),
2881 proto_method_implemented(proto
->enter_memory_pressure
));
2884 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2886 if (v
== &proto_list
)
2887 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2896 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2898 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2902 static const struct seq_operations proto_seq_ops
= {
2903 .start
= proto_seq_start
,
2904 .next
= proto_seq_next
,
2905 .stop
= proto_seq_stop
,
2906 .show
= proto_seq_show
,
2909 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2911 return seq_open_net(inode
, file
, &proto_seq_ops
,
2912 sizeof(struct seq_net_private
));
2915 static const struct file_operations proto_seq_fops
= {
2916 .owner
= THIS_MODULE
,
2917 .open
= proto_seq_open
,
2919 .llseek
= seq_lseek
,
2920 .release
= seq_release_net
,
2923 static __net_init
int proto_init_net(struct net
*net
)
2925 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
2931 static __net_exit
void proto_exit_net(struct net
*net
)
2933 remove_proc_entry("protocols", net
->proc_net
);
2937 static __net_initdata
struct pernet_operations proto_net_ops
= {
2938 .init
= proto_init_net
,
2939 .exit
= proto_exit_net
,
2942 static int __init
proto_init(void)
2944 return register_pernet_subsys(&proto_net_ops
);
2947 subsys_initcall(proto_init
);
2949 #endif /* PROC_FS */