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 static DEFINE_MUTEX(proto_list_mutex
);
143 static LIST_HEAD(proto_list
);
145 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
146 int mem_cgroup_sockets_init(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
151 mutex_lock(&proto_list_mutex
);
152 list_for_each_entry(proto
, &proto_list
, node
) {
153 if (proto
->init_cgroup
) {
154 ret
= proto
->init_cgroup(memcg
, ss
);
160 mutex_unlock(&proto_list_mutex
);
163 list_for_each_entry_continue_reverse(proto
, &proto_list
, node
)
164 if (proto
->destroy_cgroup
)
165 proto
->destroy_cgroup(memcg
);
166 mutex_unlock(&proto_list_mutex
);
170 void mem_cgroup_sockets_destroy(struct mem_cgroup
*memcg
)
174 mutex_lock(&proto_list_mutex
);
175 list_for_each_entry_reverse(proto
, &proto_list
, node
)
176 if (proto
->destroy_cgroup
)
177 proto
->destroy_cgroup(memcg
);
178 mutex_unlock(&proto_list_mutex
);
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
186 static struct lock_class_key af_family_keys
[AF_MAX
];
187 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
189 struct static_key memcg_socket_limit_enabled
;
190 EXPORT_SYMBOL(memcg_socket_limit_enabled
);
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
197 static const char *const af_family_key_strings
[AF_MAX
+1] = {
198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
213 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
227 "slock-AF_NFC" , "slock-AF_MAX"
229 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
243 "clock-AF_NFC" , "clock-AF_MAX"
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
250 static struct lock_class_key af_callback_keys
[AF_MAX
];
252 /* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
257 #define _SK_MEM_PACKETS 256
258 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
259 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
262 /* Run time adjustable parameters. */
263 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
264 EXPORT_SYMBOL(sysctl_wmem_max
);
265 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
266 EXPORT_SYMBOL(sysctl_rmem_max
);
267 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
268 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
270 /* Maximal space eaten by iovec or ancillary data plus some space */
271 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
272 EXPORT_SYMBOL(sysctl_optmem_max
);
274 #if defined(CONFIG_CGROUPS)
275 #if !defined(CONFIG_NET_CLS_CGROUP)
276 int net_cls_subsys_id
= -1;
277 EXPORT_SYMBOL_GPL(net_cls_subsys_id
);
279 #if !defined(CONFIG_NETPRIO_CGROUP)
280 int net_prio_subsys_id
= -1;
281 EXPORT_SYMBOL_GPL(net_prio_subsys_id
);
285 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
289 if (optlen
< sizeof(tv
))
291 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
293 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
297 static int warned __read_mostly
;
300 if (warned
< 10 && net_ratelimit()) {
302 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
303 __func__
, current
->comm
, task_pid_nr(current
));
307 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
308 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
310 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
311 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
315 static void sock_warn_obsolete_bsdism(const char *name
)
318 static char warncomm
[TASK_COMM_LEN
];
319 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
320 strcpy(warncomm
, current
->comm
);
321 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
327 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
329 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
331 if (sk
->sk_flags
& flags
) {
332 sk
->sk_flags
&= ~flags
;
333 if (!(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
334 net_disable_timestamp();
339 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
344 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
346 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
347 atomic_inc(&sk
->sk_drops
);
348 trace_sock_rcvqueue_full(sk
, skb
);
352 err
= sk_filter(sk
, skb
);
356 if (!sk_rmem_schedule(sk
, skb
->truesize
)) {
357 atomic_inc(&sk
->sk_drops
);
362 skb_set_owner_r(skb
, sk
);
364 /* Cache the SKB length before we tack it onto the receive
365 * queue. Once it is added it no longer belongs to us and
366 * may be freed by other threads of control pulling packets
371 /* we escape from rcu protected region, make sure we dont leak
376 spin_lock_irqsave(&list
->lock
, flags
);
377 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
378 __skb_queue_tail(list
, skb
);
379 spin_unlock_irqrestore(&list
->lock
, flags
);
381 if (!sock_flag(sk
, SOCK_DEAD
))
382 sk
->sk_data_ready(sk
, skb_len
);
385 EXPORT_SYMBOL(sock_queue_rcv_skb
);
387 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
389 int rc
= NET_RX_SUCCESS
;
391 if (sk_filter(sk
, skb
))
392 goto discard_and_relse
;
396 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
)) {
397 atomic_inc(&sk
->sk_drops
);
398 goto discard_and_relse
;
401 bh_lock_sock_nested(sk
);
404 if (!sock_owned_by_user(sk
)) {
406 * trylock + unlock semantics:
408 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
410 rc
= sk_backlog_rcv(sk
, skb
);
412 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
413 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
415 atomic_inc(&sk
->sk_drops
);
416 goto discard_and_relse
;
427 EXPORT_SYMBOL(sk_receive_skb
);
429 void sk_reset_txq(struct sock
*sk
)
431 sk_tx_queue_clear(sk
);
433 EXPORT_SYMBOL(sk_reset_txq
);
435 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
437 struct dst_entry
*dst
= __sk_dst_get(sk
);
439 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
440 sk_tx_queue_clear(sk
);
441 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
448 EXPORT_SYMBOL(__sk_dst_check
);
450 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
452 struct dst_entry
*dst
= sk_dst_get(sk
);
454 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
462 EXPORT_SYMBOL(sk_dst_check
);
464 static int sock_bindtodevice(struct sock
*sk
, char __user
*optval
, int optlen
)
466 int ret
= -ENOPROTOOPT
;
467 #ifdef CONFIG_NETDEVICES
468 struct net
*net
= sock_net(sk
);
469 char devname
[IFNAMSIZ
];
474 if (!capable(CAP_NET_RAW
))
481 /* Bind this socket to a particular device like "eth0",
482 * as specified in the passed interface name. If the
483 * name is "" or the option length is zero the socket
486 if (optlen
> IFNAMSIZ
- 1)
487 optlen
= IFNAMSIZ
- 1;
488 memset(devname
, 0, sizeof(devname
));
491 if (copy_from_user(devname
, optval
, optlen
))
495 if (devname
[0] != '\0') {
496 struct net_device
*dev
;
499 dev
= dev_get_by_name_rcu(net
, devname
);
501 index
= dev
->ifindex
;
509 sk
->sk_bound_dev_if
= index
;
521 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
524 sock_set_flag(sk
, bit
);
526 sock_reset_flag(sk
, bit
);
530 * This is meant for all protocols to use and covers goings on
531 * at the socket level. Everything here is generic.
534 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
535 char __user
*optval
, unsigned int optlen
)
537 struct sock
*sk
= sock
->sk
;
544 * Options without arguments
547 if (optname
== SO_BINDTODEVICE
)
548 return sock_bindtodevice(sk
, optval
, optlen
);
550 if (optlen
< sizeof(int))
553 if (get_user(val
, (int __user
*)optval
))
556 valbool
= val
? 1 : 0;
562 if (val
&& !capable(CAP_NET_ADMIN
))
565 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
568 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
577 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
580 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
583 /* Don't error on this BSD doesn't and if you think
584 * about it this is right. Otherwise apps have to
585 * play 'guess the biggest size' games. RCVBUF/SNDBUF
586 * are treated in BSD as hints
588 val
= min_t(u32
, val
, sysctl_wmem_max
);
590 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
591 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
592 /* Wake up sending tasks if we upped the value. */
593 sk
->sk_write_space(sk
);
597 if (!capable(CAP_NET_ADMIN
)) {
604 /* Don't error on this BSD doesn't and if you think
605 * about it this is right. Otherwise apps have to
606 * play 'guess the biggest size' games. RCVBUF/SNDBUF
607 * are treated in BSD as hints
609 val
= min_t(u32
, val
, sysctl_rmem_max
);
611 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
613 * We double it on the way in to account for
614 * "struct sk_buff" etc. overhead. Applications
615 * assume that the SO_RCVBUF setting they make will
616 * allow that much actual data to be received on that
619 * Applications are unaware that "struct sk_buff" and
620 * other overheads allocate from the receive buffer
621 * during socket buffer allocation.
623 * And after considering the possible alternatives,
624 * returning the value we actually used in getsockopt
625 * is the most desirable behavior.
627 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
631 if (!capable(CAP_NET_ADMIN
)) {
639 if (sk
->sk_protocol
== IPPROTO_TCP
)
640 tcp_set_keepalive(sk
, valbool
);
642 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
646 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
650 sk
->sk_no_check
= valbool
;
654 if ((val
>= 0 && val
<= 6) || capable(CAP_NET_ADMIN
))
655 sk
->sk_priority
= val
;
661 if (optlen
< sizeof(ling
)) {
662 ret
= -EINVAL
; /* 1003.1g */
665 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
670 sock_reset_flag(sk
, SOCK_LINGER
);
672 #if (BITS_PER_LONG == 32)
673 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
674 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
677 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
678 sock_set_flag(sk
, SOCK_LINGER
);
683 sock_warn_obsolete_bsdism("setsockopt");
688 set_bit(SOCK_PASSCRED
, &sock
->flags
);
690 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
696 if (optname
== SO_TIMESTAMP
)
697 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
699 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
700 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
701 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
703 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
704 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
708 case SO_TIMESTAMPING
:
709 if (val
& ~SOF_TIMESTAMPING_MASK
) {
713 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
,
714 val
& SOF_TIMESTAMPING_TX_HARDWARE
);
715 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
,
716 val
& SOF_TIMESTAMPING_TX_SOFTWARE
);
717 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
,
718 val
& SOF_TIMESTAMPING_RX_HARDWARE
);
719 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
720 sock_enable_timestamp(sk
,
721 SOCK_TIMESTAMPING_RX_SOFTWARE
);
723 sock_disable_timestamp(sk
,
724 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
725 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
,
726 val
& SOF_TIMESTAMPING_SOFTWARE
);
727 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
,
728 val
& SOF_TIMESTAMPING_SYS_HARDWARE
);
729 sock_valbool_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
,
730 val
& SOF_TIMESTAMPING_RAW_HARDWARE
);
736 sk
->sk_rcvlowat
= val
? : 1;
740 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
744 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
747 case SO_ATTACH_FILTER
:
749 if (optlen
== sizeof(struct sock_fprog
)) {
750 struct sock_fprog fprog
;
753 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
756 ret
= sk_attach_filter(&fprog
, sk
);
760 case SO_DETACH_FILTER
:
761 ret
= sk_detach_filter(sk
);
766 set_bit(SOCK_PASSSEC
, &sock
->flags
);
768 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
771 if (!capable(CAP_NET_ADMIN
))
777 /* We implement the SO_SNDLOWAT etc to
778 not be settable (1003.1g 5.3) */
780 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
784 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
788 if (sock
->ops
->set_peek_off
)
789 sock
->ops
->set_peek_off(sk
, val
);
795 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
805 EXPORT_SYMBOL(sock_setsockopt
);
808 void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
811 ucred
->pid
= pid_vnr(pid
);
812 ucred
->uid
= ucred
->gid
= -1;
814 struct user_namespace
*current_ns
= current_user_ns();
816 ucred
->uid
= from_kuid(current_ns
, cred
->euid
);
817 ucred
->gid
= from_kgid(current_ns
, cred
->egid
);
820 EXPORT_SYMBOL_GPL(cred_to_ucred
);
822 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
823 char __user
*optval
, int __user
*optlen
)
825 struct sock
*sk
= sock
->sk
;
833 int lv
= sizeof(int);
836 if (get_user(len
, optlen
))
841 memset(&v
, 0, sizeof(v
));
845 v
.val
= sock_flag(sk
, SOCK_DBG
);
849 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
853 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
857 v
.val
= sk
->sk_sndbuf
;
861 v
.val
= sk
->sk_rcvbuf
;
865 v
.val
= sk
->sk_reuse
;
869 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
877 v
.val
= sk
->sk_protocol
;
881 v
.val
= sk
->sk_family
;
885 v
.val
= -sock_error(sk
);
887 v
.val
= xchg(&sk
->sk_err_soft
, 0);
891 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
895 v
.val
= sk
->sk_no_check
;
899 v
.val
= sk
->sk_priority
;
904 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
905 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
909 sock_warn_obsolete_bsdism("getsockopt");
913 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
914 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
918 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
921 case SO_TIMESTAMPING
:
923 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_HARDWARE
))
924 v
.val
|= SOF_TIMESTAMPING_TX_HARDWARE
;
925 if (sock_flag(sk
, SOCK_TIMESTAMPING_TX_SOFTWARE
))
926 v
.val
|= SOF_TIMESTAMPING_TX_SOFTWARE
;
927 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_HARDWARE
))
928 v
.val
|= SOF_TIMESTAMPING_RX_HARDWARE
;
929 if (sock_flag(sk
, SOCK_TIMESTAMPING_RX_SOFTWARE
))
930 v
.val
|= SOF_TIMESTAMPING_RX_SOFTWARE
;
931 if (sock_flag(sk
, SOCK_TIMESTAMPING_SOFTWARE
))
932 v
.val
|= SOF_TIMESTAMPING_SOFTWARE
;
933 if (sock_flag(sk
, SOCK_TIMESTAMPING_SYS_HARDWARE
))
934 v
.val
|= SOF_TIMESTAMPING_SYS_HARDWARE
;
935 if (sock_flag(sk
, SOCK_TIMESTAMPING_RAW_HARDWARE
))
936 v
.val
|= SOF_TIMESTAMPING_RAW_HARDWARE
;
940 lv
= sizeof(struct timeval
);
941 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
945 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
946 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
951 lv
= sizeof(struct timeval
);
952 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
956 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
957 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
962 v
.val
= sk
->sk_rcvlowat
;
970 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
975 struct ucred peercred
;
976 if (len
> sizeof(peercred
))
977 len
= sizeof(peercred
);
978 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
979 if (copy_to_user(optval
, &peercred
, len
))
988 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
992 if (copy_to_user(optval
, address
, len
))
997 /* Dubious BSD thing... Probably nobody even uses it, but
998 * the UNIX standard wants it for whatever reason... -DaveM
1001 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1005 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1009 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1012 v
.val
= sk
->sk_mark
;
1016 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1019 case SO_WIFI_STATUS
:
1020 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1024 if (!sock
->ops
->set_peek_off
)
1027 v
.val
= sk
->sk_peek_off
;
1030 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1033 return -ENOPROTOOPT
;
1038 if (copy_to_user(optval
, &v
, len
))
1041 if (put_user(len
, optlen
))
1047 * Initialize an sk_lock.
1049 * (We also register the sk_lock with the lock validator.)
1051 static inline void sock_lock_init(struct sock
*sk
)
1053 sock_lock_init_class_and_name(sk
,
1054 af_family_slock_key_strings
[sk
->sk_family
],
1055 af_family_slock_keys
+ sk
->sk_family
,
1056 af_family_key_strings
[sk
->sk_family
],
1057 af_family_keys
+ sk
->sk_family
);
1061 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1062 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1063 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1065 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1067 #ifdef CONFIG_SECURITY_NETWORK
1068 void *sptr
= nsk
->sk_security
;
1070 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1072 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1073 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1075 #ifdef CONFIG_SECURITY_NETWORK
1076 nsk
->sk_security
= sptr
;
1077 security_sk_clone(osk
, nsk
);
1082 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1083 * un-modified. Special care is taken when initializing object to zero.
1085 static inline void sk_prot_clear_nulls(struct sock
*sk
, int size
)
1087 if (offsetof(struct sock
, sk_node
.next
) != 0)
1088 memset(sk
, 0, offsetof(struct sock
, sk_node
.next
));
1089 memset(&sk
->sk_node
.pprev
, 0,
1090 size
- offsetof(struct sock
, sk_node
.pprev
));
1093 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1095 unsigned long nulls1
, nulls2
;
1097 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1098 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1099 if (nulls1
> nulls2
)
1100 swap(nulls1
, nulls2
);
1103 memset((char *)sk
, 0, nulls1
);
1104 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1105 nulls2
- nulls1
- sizeof(void *));
1106 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1107 size
- nulls2
- sizeof(void *));
1109 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1111 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1115 struct kmem_cache
*slab
;
1119 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1122 if (priority
& __GFP_ZERO
) {
1124 prot
->clear_sk(sk
, prot
->obj_size
);
1126 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1129 sk
= kmalloc(prot
->obj_size
, priority
);
1132 kmemcheck_annotate_bitfield(sk
, flags
);
1134 if (security_sk_alloc(sk
, family
, priority
))
1137 if (!try_module_get(prot
->owner
))
1139 sk_tx_queue_clear(sk
);
1145 security_sk_free(sk
);
1148 kmem_cache_free(slab
, sk
);
1154 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1156 struct kmem_cache
*slab
;
1157 struct module
*owner
;
1159 owner
= prot
->owner
;
1162 security_sk_free(sk
);
1164 kmem_cache_free(slab
, sk
);
1170 #ifdef CONFIG_CGROUPS
1171 void sock_update_classid(struct sock
*sk
)
1175 rcu_read_lock(); /* doing current task, which cannot vanish. */
1176 classid
= task_cls_classid(current
);
1178 if (classid
&& classid
!= sk
->sk_classid
)
1179 sk
->sk_classid
= classid
;
1181 EXPORT_SYMBOL(sock_update_classid
);
1183 void sock_update_netprioidx(struct sock
*sk
, struct task_struct
*task
)
1188 sk
->sk_cgrp_prioidx
= task_netprioidx(task
);
1190 EXPORT_SYMBOL_GPL(sock_update_netprioidx
);
1194 * sk_alloc - All socket objects are allocated here
1195 * @net: the applicable net namespace
1196 * @family: protocol family
1197 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1198 * @prot: struct proto associated with this new sock instance
1200 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1205 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1207 sk
->sk_family
= family
;
1209 * See comment in struct sock definition to understand
1210 * why we need sk_prot_creator -acme
1212 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1214 sock_net_set(sk
, get_net(net
));
1215 atomic_set(&sk
->sk_wmem_alloc
, 1);
1217 sock_update_classid(sk
);
1218 sock_update_netprioidx(sk
, current
);
1223 EXPORT_SYMBOL(sk_alloc
);
1225 static void __sk_free(struct sock
*sk
)
1227 struct sk_filter
*filter
;
1229 if (sk
->sk_destruct
)
1230 sk
->sk_destruct(sk
);
1232 filter
= rcu_dereference_check(sk
->sk_filter
,
1233 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1235 sk_filter_uncharge(sk
, filter
);
1236 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1239 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1241 if (atomic_read(&sk
->sk_omem_alloc
))
1242 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1243 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1245 if (sk
->sk_peer_cred
)
1246 put_cred(sk
->sk_peer_cred
);
1247 put_pid(sk
->sk_peer_pid
);
1248 put_net(sock_net(sk
));
1249 sk_prot_free(sk
->sk_prot_creator
, sk
);
1252 void sk_free(struct sock
*sk
)
1255 * We subtract one from sk_wmem_alloc and can know if
1256 * some packets are still in some tx queue.
1257 * If not null, sock_wfree() will call __sk_free(sk) later
1259 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1262 EXPORT_SYMBOL(sk_free
);
1265 * Last sock_put should drop reference to sk->sk_net. It has already
1266 * been dropped in sk_change_net. Taking reference to stopping namespace
1268 * Take reference to a socket to remove it from hash _alive_ and after that
1269 * destroy it in the context of init_net.
1271 void sk_release_kernel(struct sock
*sk
)
1273 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
1277 sock_release(sk
->sk_socket
);
1278 release_net(sock_net(sk
));
1279 sock_net_set(sk
, get_net(&init_net
));
1282 EXPORT_SYMBOL(sk_release_kernel
);
1284 static void sk_update_clone(const struct sock
*sk
, struct sock
*newsk
)
1286 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
)
1287 sock_update_memcg(newsk
);
1291 * sk_clone_lock - clone a socket, and lock its clone
1292 * @sk: the socket to clone
1293 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1295 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1297 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1301 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1302 if (newsk
!= NULL
) {
1303 struct sk_filter
*filter
;
1305 sock_copy(newsk
, sk
);
1308 get_net(sock_net(newsk
));
1309 sk_node_init(&newsk
->sk_node
);
1310 sock_lock_init(newsk
);
1311 bh_lock_sock(newsk
);
1312 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1313 newsk
->sk_backlog
.len
= 0;
1315 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1317 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1319 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1320 atomic_set(&newsk
->sk_omem_alloc
, 0);
1321 skb_queue_head_init(&newsk
->sk_receive_queue
);
1322 skb_queue_head_init(&newsk
->sk_write_queue
);
1323 #ifdef CONFIG_NET_DMA
1324 skb_queue_head_init(&newsk
->sk_async_wait_queue
);
1327 spin_lock_init(&newsk
->sk_dst_lock
);
1328 rwlock_init(&newsk
->sk_callback_lock
);
1329 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1330 af_callback_keys
+ newsk
->sk_family
,
1331 af_family_clock_key_strings
[newsk
->sk_family
]);
1333 newsk
->sk_dst_cache
= NULL
;
1334 newsk
->sk_wmem_queued
= 0;
1335 newsk
->sk_forward_alloc
= 0;
1336 newsk
->sk_send_head
= NULL
;
1337 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1339 sock_reset_flag(newsk
, SOCK_DONE
);
1340 skb_queue_head_init(&newsk
->sk_error_queue
);
1342 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1344 sk_filter_charge(newsk
, filter
);
1346 if (unlikely(xfrm_sk_clone_policy(newsk
))) {
1347 /* It is still raw copy of parent, so invalidate
1348 * destructor and make plain sk_free() */
1349 newsk
->sk_destruct
= NULL
;
1350 bh_unlock_sock(newsk
);
1357 newsk
->sk_priority
= 0;
1359 * Before updating sk_refcnt, we must commit prior changes to memory
1360 * (Documentation/RCU/rculist_nulls.txt for details)
1363 atomic_set(&newsk
->sk_refcnt
, 2);
1366 * Increment the counter in the same struct proto as the master
1367 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1368 * is the same as sk->sk_prot->socks, as this field was copied
1371 * This _changes_ the previous behaviour, where
1372 * tcp_create_openreq_child always was incrementing the
1373 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1374 * to be taken into account in all callers. -acme
1376 sk_refcnt_debug_inc(newsk
);
1377 sk_set_socket(newsk
, NULL
);
1378 newsk
->sk_wq
= NULL
;
1380 sk_update_clone(sk
, newsk
);
1382 if (newsk
->sk_prot
->sockets_allocated
)
1383 sk_sockets_allocated_inc(newsk
);
1385 if (newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1386 net_enable_timestamp();
1391 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1393 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1395 __sk_dst_set(sk
, dst
);
1396 sk
->sk_route_caps
= dst
->dev
->features
;
1397 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1398 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1399 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1400 if (sk_can_gso(sk
)) {
1401 if (dst
->header_len
) {
1402 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1404 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1405 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1409 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1411 void __init
sk_init(void)
1413 if (totalram_pages
<= 4096) {
1414 sysctl_wmem_max
= 32767;
1415 sysctl_rmem_max
= 32767;
1416 sysctl_wmem_default
= 32767;
1417 sysctl_rmem_default
= 32767;
1418 } else if (totalram_pages
>= 131072) {
1419 sysctl_wmem_max
= 131071;
1420 sysctl_rmem_max
= 131071;
1425 * Simple resource managers for sockets.
1430 * Write buffer destructor automatically called from kfree_skb.
1432 void sock_wfree(struct sk_buff
*skb
)
1434 struct sock
*sk
= skb
->sk
;
1435 unsigned int len
= skb
->truesize
;
1437 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1439 * Keep a reference on sk_wmem_alloc, this will be released
1440 * after sk_write_space() call
1442 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1443 sk
->sk_write_space(sk
);
1447 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1448 * could not do because of in-flight packets
1450 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1453 EXPORT_SYMBOL(sock_wfree
);
1456 * Read buffer destructor automatically called from kfree_skb.
1458 void sock_rfree(struct sk_buff
*skb
)
1460 struct sock
*sk
= skb
->sk
;
1461 unsigned int len
= skb
->truesize
;
1463 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1464 sk_mem_uncharge(sk
, len
);
1466 EXPORT_SYMBOL(sock_rfree
);
1468 void sock_edemux(struct sk_buff
*skb
)
1472 EXPORT_SYMBOL(sock_edemux
);
1474 int sock_i_uid(struct sock
*sk
)
1478 read_lock_bh(&sk
->sk_callback_lock
);
1479 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: 0;
1480 read_unlock_bh(&sk
->sk_callback_lock
);
1483 EXPORT_SYMBOL(sock_i_uid
);
1485 unsigned long sock_i_ino(struct sock
*sk
)
1489 read_lock_bh(&sk
->sk_callback_lock
);
1490 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1491 read_unlock_bh(&sk
->sk_callback_lock
);
1494 EXPORT_SYMBOL(sock_i_ino
);
1497 * Allocate a skb from the socket's send buffer.
1499 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1502 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1503 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1505 skb_set_owner_w(skb
, sk
);
1511 EXPORT_SYMBOL(sock_wmalloc
);
1514 * Allocate a skb from the socket's receive buffer.
1516 struct sk_buff
*sock_rmalloc(struct sock
*sk
, unsigned long size
, int force
,
1519 if (force
|| atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1520 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1522 skb_set_owner_r(skb
, sk
);
1530 * Allocate a memory block from the socket's option memory buffer.
1532 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1534 if ((unsigned int)size
<= sysctl_optmem_max
&&
1535 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1537 /* First do the add, to avoid the race if kmalloc
1540 atomic_add(size
, &sk
->sk_omem_alloc
);
1541 mem
= kmalloc(size
, priority
);
1544 atomic_sub(size
, &sk
->sk_omem_alloc
);
1548 EXPORT_SYMBOL(sock_kmalloc
);
1551 * Free an option memory block.
1553 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1556 atomic_sub(size
, &sk
->sk_omem_alloc
);
1558 EXPORT_SYMBOL(sock_kfree_s
);
1560 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1561 I think, these locks should be removed for datagram sockets.
1563 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1567 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1571 if (signal_pending(current
))
1573 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1574 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1575 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1577 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1581 timeo
= schedule_timeout(timeo
);
1583 finish_wait(sk_sleep(sk
), &wait
);
1589 * Generic send/receive buffer handlers
1592 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1593 unsigned long data_len
, int noblock
,
1596 struct sk_buff
*skb
;
1600 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
1603 if (npages
> MAX_SKB_FRAGS
)
1606 gfp_mask
= sk
->sk_allocation
;
1607 if (gfp_mask
& __GFP_WAIT
)
1608 gfp_mask
|= __GFP_REPEAT
;
1610 timeo
= sock_sndtimeo(sk
, noblock
);
1612 err
= sock_error(sk
);
1617 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1620 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1621 skb
= alloc_skb(header_len
, gfp_mask
);
1625 /* No pages, we're done... */
1629 skb
->truesize
+= data_len
;
1630 skb_shinfo(skb
)->nr_frags
= npages
;
1631 for (i
= 0; i
< npages
; i
++) {
1634 page
= alloc_pages(sk
->sk_allocation
, 0);
1637 skb_shinfo(skb
)->nr_frags
= i
;
1642 __skb_fill_page_desc(skb
, i
,
1644 (data_len
>= PAGE_SIZE
?
1647 data_len
-= PAGE_SIZE
;
1650 /* Full success... */
1656 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
1657 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1661 if (signal_pending(current
))
1663 timeo
= sock_wait_for_wmem(sk
, timeo
);
1666 skb_set_owner_w(skb
, sk
);
1670 err
= sock_intr_errno(timeo
);
1675 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1677 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1678 int noblock
, int *errcode
)
1680 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
);
1682 EXPORT_SYMBOL(sock_alloc_send_skb
);
1684 static void __lock_sock(struct sock
*sk
)
1685 __releases(&sk
->sk_lock
.slock
)
1686 __acquires(&sk
->sk_lock
.slock
)
1691 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1692 TASK_UNINTERRUPTIBLE
);
1693 spin_unlock_bh(&sk
->sk_lock
.slock
);
1695 spin_lock_bh(&sk
->sk_lock
.slock
);
1696 if (!sock_owned_by_user(sk
))
1699 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1702 static void __release_sock(struct sock
*sk
)
1703 __releases(&sk
->sk_lock
.slock
)
1704 __acquires(&sk
->sk_lock
.slock
)
1706 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1709 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1713 struct sk_buff
*next
= skb
->next
;
1716 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1718 sk_backlog_rcv(sk
, skb
);
1721 * We are in process context here with softirqs
1722 * disabled, use cond_resched_softirq() to preempt.
1723 * This is safe to do because we've taken the backlog
1726 cond_resched_softirq();
1729 } while (skb
!= NULL
);
1732 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1735 * Doing the zeroing here guarantee we can not loop forever
1736 * while a wild producer attempts to flood us.
1738 sk
->sk_backlog
.len
= 0;
1742 * sk_wait_data - wait for data to arrive at sk_receive_queue
1743 * @sk: sock to wait on
1744 * @timeo: for how long
1746 * Now socket state including sk->sk_err is changed only under lock,
1747 * hence we may omit checks after joining wait queue.
1748 * We check receive queue before schedule() only as optimization;
1749 * it is very likely that release_sock() added new data.
1751 int sk_wait_data(struct sock
*sk
, long *timeo
)
1756 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1757 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1758 rc
= sk_wait_event(sk
, timeo
, !skb_queue_empty(&sk
->sk_receive_queue
));
1759 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1760 finish_wait(sk_sleep(sk
), &wait
);
1763 EXPORT_SYMBOL(sk_wait_data
);
1766 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1768 * @size: memory size to allocate
1769 * @kind: allocation type
1771 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1772 * rmem allocation. This function assumes that protocols which have
1773 * memory_pressure use sk_wmem_queued as write buffer accounting.
1775 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
1777 struct proto
*prot
= sk
->sk_prot
;
1778 int amt
= sk_mem_pages(size
);
1780 int parent_status
= UNDER_LIMIT
;
1782 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
1784 allocated
= sk_memory_allocated_add(sk
, amt
, &parent_status
);
1787 if (parent_status
== UNDER_LIMIT
&&
1788 allocated
<= sk_prot_mem_limits(sk
, 0)) {
1789 sk_leave_memory_pressure(sk
);
1793 /* Under pressure. (we or our parents) */
1794 if ((parent_status
> SOFT_LIMIT
) ||
1795 allocated
> sk_prot_mem_limits(sk
, 1))
1796 sk_enter_memory_pressure(sk
);
1798 /* Over hard limit (we or our parents) */
1799 if ((parent_status
== OVER_LIMIT
) ||
1800 (allocated
> sk_prot_mem_limits(sk
, 2)))
1801 goto suppress_allocation
;
1803 /* guarantee minimum buffer size under pressure */
1804 if (kind
== SK_MEM_RECV
) {
1805 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
1808 } else { /* SK_MEM_SEND */
1809 if (sk
->sk_type
== SOCK_STREAM
) {
1810 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
1812 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
1813 prot
->sysctl_wmem
[0])
1817 if (sk_has_memory_pressure(sk
)) {
1820 if (!sk_under_memory_pressure(sk
))
1822 alloc
= sk_sockets_allocated_read_positive(sk
);
1823 if (sk_prot_mem_limits(sk
, 2) > alloc
*
1824 sk_mem_pages(sk
->sk_wmem_queued
+
1825 atomic_read(&sk
->sk_rmem_alloc
) +
1826 sk
->sk_forward_alloc
))
1830 suppress_allocation
:
1832 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
1833 sk_stream_moderate_sndbuf(sk
);
1835 /* Fail only if socket is _under_ its sndbuf.
1836 * In this case we cannot block, so that we have to fail.
1838 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
1842 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
1844 /* Alas. Undo changes. */
1845 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
1847 sk_memory_allocated_sub(sk
, amt
);
1851 EXPORT_SYMBOL(__sk_mem_schedule
);
1854 * __sk_reclaim - reclaim memory_allocated
1857 void __sk_mem_reclaim(struct sock
*sk
)
1859 sk_memory_allocated_sub(sk
,
1860 sk
->sk_forward_alloc
>> SK_MEM_QUANTUM_SHIFT
);
1861 sk
->sk_forward_alloc
&= SK_MEM_QUANTUM
- 1;
1863 if (sk_under_memory_pressure(sk
) &&
1864 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
1865 sk_leave_memory_pressure(sk
);
1867 EXPORT_SYMBOL(__sk_mem_reclaim
);
1871 * Set of default routines for initialising struct proto_ops when
1872 * the protocol does not support a particular function. In certain
1873 * cases where it makes no sense for a protocol to have a "do nothing"
1874 * function, some default processing is provided.
1877 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
1881 EXPORT_SYMBOL(sock_no_bind
);
1883 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
1888 EXPORT_SYMBOL(sock_no_connect
);
1890 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
1894 EXPORT_SYMBOL(sock_no_socketpair
);
1896 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1900 EXPORT_SYMBOL(sock_no_accept
);
1902 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
1907 EXPORT_SYMBOL(sock_no_getname
);
1909 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
1913 EXPORT_SYMBOL(sock_no_poll
);
1915 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1919 EXPORT_SYMBOL(sock_no_ioctl
);
1921 int sock_no_listen(struct socket
*sock
, int backlog
)
1925 EXPORT_SYMBOL(sock_no_listen
);
1927 int sock_no_shutdown(struct socket
*sock
, int how
)
1931 EXPORT_SYMBOL(sock_no_shutdown
);
1933 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
1934 char __user
*optval
, unsigned int optlen
)
1938 EXPORT_SYMBOL(sock_no_setsockopt
);
1940 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
1941 char __user
*optval
, int __user
*optlen
)
1945 EXPORT_SYMBOL(sock_no_getsockopt
);
1947 int sock_no_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1952 EXPORT_SYMBOL(sock_no_sendmsg
);
1954 int sock_no_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*m
,
1955 size_t len
, int flags
)
1959 EXPORT_SYMBOL(sock_no_recvmsg
);
1961 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
1963 /* Mirror missing mmap method error code */
1966 EXPORT_SYMBOL(sock_no_mmap
);
1968 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
1971 struct msghdr msg
= {.msg_flags
= flags
};
1973 char *kaddr
= kmap(page
);
1974 iov
.iov_base
= kaddr
+ offset
;
1976 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1980 EXPORT_SYMBOL(sock_no_sendpage
);
1983 * Default Socket Callbacks
1986 static void sock_def_wakeup(struct sock
*sk
)
1988 struct socket_wq
*wq
;
1991 wq
= rcu_dereference(sk
->sk_wq
);
1992 if (wq_has_sleeper(wq
))
1993 wake_up_interruptible_all(&wq
->wait
);
1997 static void sock_def_error_report(struct sock
*sk
)
1999 struct socket_wq
*wq
;
2002 wq
= rcu_dereference(sk
->sk_wq
);
2003 if (wq_has_sleeper(wq
))
2004 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2005 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2009 static void sock_def_readable(struct sock
*sk
, int len
)
2011 struct socket_wq
*wq
;
2014 wq
= rcu_dereference(sk
->sk_wq
);
2015 if (wq_has_sleeper(wq
))
2016 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2017 POLLRDNORM
| POLLRDBAND
);
2018 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2022 static void sock_def_write_space(struct sock
*sk
)
2024 struct socket_wq
*wq
;
2028 /* Do not wake up a writer until he can make "significant"
2031 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2032 wq
= rcu_dereference(sk
->sk_wq
);
2033 if (wq_has_sleeper(wq
))
2034 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2035 POLLWRNORM
| POLLWRBAND
);
2037 /* Should agree with poll, otherwise some programs break */
2038 if (sock_writeable(sk
))
2039 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2045 static void sock_def_destruct(struct sock
*sk
)
2047 kfree(sk
->sk_protinfo
);
2050 void sk_send_sigurg(struct sock
*sk
)
2052 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2053 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2054 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2056 EXPORT_SYMBOL(sk_send_sigurg
);
2058 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2059 unsigned long expires
)
2061 if (!mod_timer(timer
, expires
))
2064 EXPORT_SYMBOL(sk_reset_timer
);
2066 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2068 if (timer_pending(timer
) && del_timer(timer
))
2071 EXPORT_SYMBOL(sk_stop_timer
);
2073 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2075 skb_queue_head_init(&sk
->sk_receive_queue
);
2076 skb_queue_head_init(&sk
->sk_write_queue
);
2077 skb_queue_head_init(&sk
->sk_error_queue
);
2078 #ifdef CONFIG_NET_DMA
2079 skb_queue_head_init(&sk
->sk_async_wait_queue
);
2082 sk
->sk_send_head
= NULL
;
2084 init_timer(&sk
->sk_timer
);
2086 sk
->sk_allocation
= GFP_KERNEL
;
2087 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2088 sk
->sk_sndbuf
= sysctl_wmem_default
;
2089 sk
->sk_state
= TCP_CLOSE
;
2090 sk_set_socket(sk
, sock
);
2092 sock_set_flag(sk
, SOCK_ZAPPED
);
2095 sk
->sk_type
= sock
->type
;
2096 sk
->sk_wq
= sock
->wq
;
2101 spin_lock_init(&sk
->sk_dst_lock
);
2102 rwlock_init(&sk
->sk_callback_lock
);
2103 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2104 af_callback_keys
+ sk
->sk_family
,
2105 af_family_clock_key_strings
[sk
->sk_family
]);
2107 sk
->sk_state_change
= sock_def_wakeup
;
2108 sk
->sk_data_ready
= sock_def_readable
;
2109 sk
->sk_write_space
= sock_def_write_space
;
2110 sk
->sk_error_report
= sock_def_error_report
;
2111 sk
->sk_destruct
= sock_def_destruct
;
2113 sk
->sk_sndmsg_page
= NULL
;
2114 sk
->sk_sndmsg_off
= 0;
2115 sk
->sk_peek_off
= -1;
2117 sk
->sk_peer_pid
= NULL
;
2118 sk
->sk_peer_cred
= NULL
;
2119 sk
->sk_write_pending
= 0;
2120 sk
->sk_rcvlowat
= 1;
2121 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2122 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2124 sk
->sk_stamp
= ktime_set(-1L, 0);
2127 * Before updating sk_refcnt, we must commit prior changes to memory
2128 * (Documentation/RCU/rculist_nulls.txt for details)
2131 atomic_set(&sk
->sk_refcnt
, 1);
2132 atomic_set(&sk
->sk_drops
, 0);
2134 EXPORT_SYMBOL(sock_init_data
);
2136 void lock_sock_nested(struct sock
*sk
, int subclass
)
2139 spin_lock_bh(&sk
->sk_lock
.slock
);
2140 if (sk
->sk_lock
.owned
)
2142 sk
->sk_lock
.owned
= 1;
2143 spin_unlock(&sk
->sk_lock
.slock
);
2145 * The sk_lock has mutex_lock() semantics here:
2147 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2150 EXPORT_SYMBOL(lock_sock_nested
);
2152 void release_sock(struct sock
*sk
)
2155 * The sk_lock has mutex_unlock() semantics:
2157 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2159 spin_lock_bh(&sk
->sk_lock
.slock
);
2160 if (sk
->sk_backlog
.tail
)
2163 if (sk
->sk_prot
->release_cb
)
2164 sk
->sk_prot
->release_cb(sk
);
2166 sk
->sk_lock
.owned
= 0;
2167 if (waitqueue_active(&sk
->sk_lock
.wq
))
2168 wake_up(&sk
->sk_lock
.wq
);
2169 spin_unlock_bh(&sk
->sk_lock
.slock
);
2171 EXPORT_SYMBOL(release_sock
);
2174 * lock_sock_fast - fast version of lock_sock
2177 * This version should be used for very small section, where process wont block
2178 * return false if fast path is taken
2179 * sk_lock.slock locked, owned = 0, BH disabled
2180 * return true if slow path is taken
2181 * sk_lock.slock unlocked, owned = 1, BH enabled
2183 bool lock_sock_fast(struct sock
*sk
)
2186 spin_lock_bh(&sk
->sk_lock
.slock
);
2188 if (!sk
->sk_lock
.owned
)
2190 * Note : We must disable BH
2195 sk
->sk_lock
.owned
= 1;
2196 spin_unlock(&sk
->sk_lock
.slock
);
2198 * The sk_lock has mutex_lock() semantics here:
2200 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2204 EXPORT_SYMBOL(lock_sock_fast
);
2206 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2209 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2210 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2211 tv
= ktime_to_timeval(sk
->sk_stamp
);
2212 if (tv
.tv_sec
== -1)
2214 if (tv
.tv_sec
== 0) {
2215 sk
->sk_stamp
= ktime_get_real();
2216 tv
= ktime_to_timeval(sk
->sk_stamp
);
2218 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2220 EXPORT_SYMBOL(sock_get_timestamp
);
2222 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2225 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2226 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2227 ts
= ktime_to_timespec(sk
->sk_stamp
);
2228 if (ts
.tv_sec
== -1)
2230 if (ts
.tv_sec
== 0) {
2231 sk
->sk_stamp
= ktime_get_real();
2232 ts
= ktime_to_timespec(sk
->sk_stamp
);
2234 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2236 EXPORT_SYMBOL(sock_get_timestampns
);
2238 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2240 if (!sock_flag(sk
, flag
)) {
2241 unsigned long previous_flags
= sk
->sk_flags
;
2243 sock_set_flag(sk
, flag
);
2245 * we just set one of the two flags which require net
2246 * time stamping, but time stamping might have been on
2247 * already because of the other one
2249 if (!(previous_flags
& SK_FLAGS_TIMESTAMP
))
2250 net_enable_timestamp();
2255 * Get a socket option on an socket.
2257 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2258 * asynchronous errors should be reported by getsockopt. We assume
2259 * this means if you specify SO_ERROR (otherwise whats the point of it).
2261 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2262 char __user
*optval
, int __user
*optlen
)
2264 struct sock
*sk
= sock
->sk
;
2266 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2268 EXPORT_SYMBOL(sock_common_getsockopt
);
2270 #ifdef CONFIG_COMPAT
2271 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2272 char __user
*optval
, int __user
*optlen
)
2274 struct sock
*sk
= sock
->sk
;
2276 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2277 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2279 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2281 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2284 int sock_common_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2285 struct msghdr
*msg
, size_t size
, int flags
)
2287 struct sock
*sk
= sock
->sk
;
2291 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2292 flags
& ~MSG_DONTWAIT
, &addr_len
);
2294 msg
->msg_namelen
= addr_len
;
2297 EXPORT_SYMBOL(sock_common_recvmsg
);
2300 * Set socket options on an inet socket.
2302 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2303 char __user
*optval
, unsigned int optlen
)
2305 struct sock
*sk
= sock
->sk
;
2307 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2309 EXPORT_SYMBOL(sock_common_setsockopt
);
2311 #ifdef CONFIG_COMPAT
2312 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2313 char __user
*optval
, unsigned int optlen
)
2315 struct sock
*sk
= sock
->sk
;
2317 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2318 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2320 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2322 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2325 void sk_common_release(struct sock
*sk
)
2327 if (sk
->sk_prot
->destroy
)
2328 sk
->sk_prot
->destroy(sk
);
2331 * Observation: when sock_common_release is called, processes have
2332 * no access to socket. But net still has.
2333 * Step one, detach it from networking:
2335 * A. Remove from hash tables.
2338 sk
->sk_prot
->unhash(sk
);
2341 * In this point socket cannot receive new packets, but it is possible
2342 * that some packets are in flight because some CPU runs receiver and
2343 * did hash table lookup before we unhashed socket. They will achieve
2344 * receive queue and will be purged by socket destructor.
2346 * Also we still have packets pending on receive queue and probably,
2347 * our own packets waiting in device queues. sock_destroy will drain
2348 * receive queue, but transmitted packets will delay socket destruction
2349 * until the last reference will be released.
2354 xfrm_sk_free_policy(sk
);
2356 sk_refcnt_debug_release(sk
);
2359 EXPORT_SYMBOL(sk_common_release
);
2361 #ifdef CONFIG_PROC_FS
2362 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2364 int val
[PROTO_INUSE_NR
];
2367 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2369 #ifdef CONFIG_NET_NS
2370 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2372 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2374 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2376 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2378 int cpu
, idx
= prot
->inuse_idx
;
2381 for_each_possible_cpu(cpu
)
2382 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2384 return res
>= 0 ? res
: 0;
2386 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2388 static int __net_init
sock_inuse_init_net(struct net
*net
)
2390 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2391 return net
->core
.inuse
? 0 : -ENOMEM
;
2394 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2396 free_percpu(net
->core
.inuse
);
2399 static struct pernet_operations net_inuse_ops
= {
2400 .init
= sock_inuse_init_net
,
2401 .exit
= sock_inuse_exit_net
,
2404 static __init
int net_inuse_init(void)
2406 if (register_pernet_subsys(&net_inuse_ops
))
2407 panic("Cannot initialize net inuse counters");
2412 core_initcall(net_inuse_init
);
2414 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2416 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2418 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2420 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2422 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2424 int cpu
, idx
= prot
->inuse_idx
;
2427 for_each_possible_cpu(cpu
)
2428 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2430 return res
>= 0 ? res
: 0;
2432 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2435 static void assign_proto_idx(struct proto
*prot
)
2437 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2439 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2440 pr_err("PROTO_INUSE_NR exhausted\n");
2444 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2447 static void release_proto_idx(struct proto
*prot
)
2449 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2450 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2453 static inline void assign_proto_idx(struct proto
*prot
)
2457 static inline void release_proto_idx(struct proto
*prot
)
2462 int proto_register(struct proto
*prot
, int alloc_slab
)
2465 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2466 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2469 if (prot
->slab
== NULL
) {
2470 pr_crit("%s: Can't create sock SLAB cache!\n",
2475 if (prot
->rsk_prot
!= NULL
) {
2476 prot
->rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s", prot
->name
);
2477 if (prot
->rsk_prot
->slab_name
== NULL
)
2478 goto out_free_sock_slab
;
2480 prot
->rsk_prot
->slab
= kmem_cache_create(prot
->rsk_prot
->slab_name
,
2481 prot
->rsk_prot
->obj_size
, 0,
2482 SLAB_HWCACHE_ALIGN
, NULL
);
2484 if (prot
->rsk_prot
->slab
== NULL
) {
2485 pr_crit("%s: Can't create request sock SLAB cache!\n",
2487 goto out_free_request_sock_slab_name
;
2491 if (prot
->twsk_prot
!= NULL
) {
2492 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2494 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2495 goto out_free_request_sock_slab
;
2497 prot
->twsk_prot
->twsk_slab
=
2498 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2499 prot
->twsk_prot
->twsk_obj_size
,
2501 SLAB_HWCACHE_ALIGN
|
2504 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2505 goto out_free_timewait_sock_slab_name
;
2509 mutex_lock(&proto_list_mutex
);
2510 list_add(&prot
->node
, &proto_list
);
2511 assign_proto_idx(prot
);
2512 mutex_unlock(&proto_list_mutex
);
2515 out_free_timewait_sock_slab_name
:
2516 kfree(prot
->twsk_prot
->twsk_slab_name
);
2517 out_free_request_sock_slab
:
2518 if (prot
->rsk_prot
&& prot
->rsk_prot
->slab
) {
2519 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2520 prot
->rsk_prot
->slab
= NULL
;
2522 out_free_request_sock_slab_name
:
2524 kfree(prot
->rsk_prot
->slab_name
);
2526 kmem_cache_destroy(prot
->slab
);
2531 EXPORT_SYMBOL(proto_register
);
2533 void proto_unregister(struct proto
*prot
)
2535 mutex_lock(&proto_list_mutex
);
2536 release_proto_idx(prot
);
2537 list_del(&prot
->node
);
2538 mutex_unlock(&proto_list_mutex
);
2540 if (prot
->slab
!= NULL
) {
2541 kmem_cache_destroy(prot
->slab
);
2545 if (prot
->rsk_prot
!= NULL
&& prot
->rsk_prot
->slab
!= NULL
) {
2546 kmem_cache_destroy(prot
->rsk_prot
->slab
);
2547 kfree(prot
->rsk_prot
->slab_name
);
2548 prot
->rsk_prot
->slab
= NULL
;
2551 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2552 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2553 kfree(prot
->twsk_prot
->twsk_slab_name
);
2554 prot
->twsk_prot
->twsk_slab
= NULL
;
2557 EXPORT_SYMBOL(proto_unregister
);
2559 #ifdef CONFIG_PROC_FS
2560 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2561 __acquires(proto_list_mutex
)
2563 mutex_lock(&proto_list_mutex
);
2564 return seq_list_start_head(&proto_list
, *pos
);
2567 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2569 return seq_list_next(v
, &proto_list
, pos
);
2572 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2573 __releases(proto_list_mutex
)
2575 mutex_unlock(&proto_list_mutex
);
2578 static char proto_method_implemented(const void *method
)
2580 return method
== NULL
? 'n' : 'y';
2582 static long sock_prot_memory_allocated(struct proto
*proto
)
2584 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2587 static char *sock_prot_memory_pressure(struct proto
*proto
)
2589 return proto
->memory_pressure
!= NULL
?
2590 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2593 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2596 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2597 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2600 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2601 sock_prot_memory_allocated(proto
),
2602 sock_prot_memory_pressure(proto
),
2604 proto
->slab
== NULL
? "no" : "yes",
2605 module_name(proto
->owner
),
2606 proto_method_implemented(proto
->close
),
2607 proto_method_implemented(proto
->connect
),
2608 proto_method_implemented(proto
->disconnect
),
2609 proto_method_implemented(proto
->accept
),
2610 proto_method_implemented(proto
->ioctl
),
2611 proto_method_implemented(proto
->init
),
2612 proto_method_implemented(proto
->destroy
),
2613 proto_method_implemented(proto
->shutdown
),
2614 proto_method_implemented(proto
->setsockopt
),
2615 proto_method_implemented(proto
->getsockopt
),
2616 proto_method_implemented(proto
->sendmsg
),
2617 proto_method_implemented(proto
->recvmsg
),
2618 proto_method_implemented(proto
->sendpage
),
2619 proto_method_implemented(proto
->bind
),
2620 proto_method_implemented(proto
->backlog_rcv
),
2621 proto_method_implemented(proto
->hash
),
2622 proto_method_implemented(proto
->unhash
),
2623 proto_method_implemented(proto
->get_port
),
2624 proto_method_implemented(proto
->enter_memory_pressure
));
2627 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2629 if (v
== &proto_list
)
2630 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2639 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2641 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2645 static const struct seq_operations proto_seq_ops
= {
2646 .start
= proto_seq_start
,
2647 .next
= proto_seq_next
,
2648 .stop
= proto_seq_stop
,
2649 .show
= proto_seq_show
,
2652 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2654 return seq_open_net(inode
, file
, &proto_seq_ops
,
2655 sizeof(struct seq_net_private
));
2658 static const struct file_operations proto_seq_fops
= {
2659 .owner
= THIS_MODULE
,
2660 .open
= proto_seq_open
,
2662 .llseek
= seq_lseek
,
2663 .release
= seq_release_net
,
2666 static __net_init
int proto_init_net(struct net
*net
)
2668 if (!proc_net_fops_create(net
, "protocols", S_IRUGO
, &proto_seq_fops
))
2674 static __net_exit
void proto_exit_net(struct net
*net
)
2676 proc_net_remove(net
, "protocols");
2680 static __net_initdata
struct pernet_operations proto_net_ops
= {
2681 .init
= proto_init_net
,
2682 .exit
= proto_exit_net
,
2685 static int __init
proto_init(void)
2687 return register_pernet_subsys(&proto_net_ops
);
2690 subsys_initcall(proto_init
);
2692 #endif /* PROC_FS */