2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <asm/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex
);
148 static LIST_HEAD(proto_list
);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock
*sk
,
161 struct user_namespace
*user_ns
, int cap
)
163 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
164 ns_capable(user_ns
, cap
);
166 EXPORT_SYMBOL(sk_ns_capable
);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock
*sk
, int cap
)
179 return sk_ns_capable(sk
, &init_user_ns
, cap
);
181 EXPORT_SYMBOL(sk_capable
);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock
*sk
, int cap
)
194 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
196 EXPORT_SYMBOL(sk_net_capable
);
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
202 static struct lock_class_key af_family_keys
[AF_MAX
];
203 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
210 static const char *const af_family_key_strings
[AF_MAX
+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
227 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
244 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
266 static struct lock_class_key af_callback_keys
[AF_MAX
];
268 /* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
273 #define _SK_MEM_PACKETS 256
274 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
275 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
280 EXPORT_SYMBOL(sysctl_wmem_max
);
281 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
282 EXPORT_SYMBOL(sysctl_rmem_max
);
283 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
284 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
288 EXPORT_SYMBOL(sysctl_optmem_max
);
290 int sysctl_tstamp_allow_data __read_mostly
= 1;
292 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
293 EXPORT_SYMBOL_GPL(memalloc_socks
);
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
303 void sk_set_memalloc(struct sock
*sk
)
305 sock_set_flag(sk
, SOCK_MEMALLOC
);
306 sk
->sk_allocation
|= __GFP_MEMALLOC
;
307 static_key_slow_inc(&memalloc_socks
);
309 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
311 void sk_clear_memalloc(struct sock
*sk
)
313 sock_reset_flag(sk
, SOCK_MEMALLOC
);
314 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
315 static_key_slow_dec(&memalloc_socks
);
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
328 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
331 unsigned long pflags
= current
->flags
;
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
336 current
->flags
|= PF_MEMALLOC
;
337 ret
= sk
->sk_backlog_rcv(sk
, skb
);
338 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
342 EXPORT_SYMBOL(__sk_backlog_rcv
);
344 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
348 if (optlen
< sizeof(tv
))
350 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
352 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
356 static int warned __read_mostly
;
359 if (warned
< 10 && net_ratelimit()) {
361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__
, current
->comm
, task_pid_nr(current
));
366 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
367 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
369 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
370 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
374 static void sock_warn_obsolete_bsdism(const char *name
)
377 static char warncomm
[TASK_COMM_LEN
];
378 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
379 strcpy(warncomm
, current
->comm
);
380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
386 static bool sock_needs_netstamp(const struct sock
*sk
)
388 switch (sk
->sk_family
) {
397 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
399 if (sk
->sk_flags
& flags
) {
400 sk
->sk_flags
&= ~flags
;
401 if (sock_needs_netstamp(sk
) &&
402 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
403 net_disable_timestamp();
408 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
411 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
413 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
414 atomic_inc(&sk
->sk_drops
);
415 trace_sock_rcvqueue_full(sk
, skb
);
419 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
420 atomic_inc(&sk
->sk_drops
);
425 skb_set_owner_r(skb
, sk
);
427 /* we escape from rcu protected region, make sure we dont leak
432 spin_lock_irqsave(&list
->lock
, flags
);
433 sock_skb_set_dropcount(sk
, skb
);
434 __skb_queue_tail(list
, skb
);
435 spin_unlock_irqrestore(&list
->lock
, flags
);
437 if (!sock_flag(sk
, SOCK_DEAD
))
438 sk
->sk_data_ready(sk
);
441 EXPORT_SYMBOL(__sock_queue_rcv_skb
);
443 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
447 err
= sk_filter(sk
, skb
);
451 return __sock_queue_rcv_skb(sk
, skb
);
453 EXPORT_SYMBOL(sock_queue_rcv_skb
);
455 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
457 int rc
= NET_RX_SUCCESS
;
459 if (sk_filter(sk
, skb
))
460 goto discard_and_relse
;
464 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
465 atomic_inc(&sk
->sk_drops
);
466 goto discard_and_relse
;
469 bh_lock_sock_nested(sk
);
472 if (!sock_owned_by_user(sk
)) {
474 * trylock + unlock semantics:
476 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
478 rc
= sk_backlog_rcv(sk
, skb
);
480 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
481 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
483 atomic_inc(&sk
->sk_drops
);
484 goto discard_and_relse
;
495 EXPORT_SYMBOL(sk_receive_skb
);
497 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
499 struct dst_entry
*dst
= __sk_dst_get(sk
);
501 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
502 sk_tx_queue_clear(sk
);
503 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
510 EXPORT_SYMBOL(__sk_dst_check
);
512 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
514 struct dst_entry
*dst
= sk_dst_get(sk
);
516 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
524 EXPORT_SYMBOL(sk_dst_check
);
526 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
529 int ret
= -ENOPROTOOPT
;
530 #ifdef CONFIG_NETDEVICES
531 struct net
*net
= sock_net(sk
);
532 char devname
[IFNAMSIZ
];
537 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
544 /* Bind this socket to a particular device like "eth0",
545 * as specified in the passed interface name. If the
546 * name is "" or the option length is zero the socket
549 if (optlen
> IFNAMSIZ
- 1)
550 optlen
= IFNAMSIZ
- 1;
551 memset(devname
, 0, sizeof(devname
));
554 if (copy_from_user(devname
, optval
, optlen
))
558 if (devname
[0] != '\0') {
559 struct net_device
*dev
;
562 dev
= dev_get_by_name_rcu(net
, devname
);
564 index
= dev
->ifindex
;
572 sk
->sk_bound_dev_if
= index
;
584 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
585 int __user
*optlen
, int len
)
587 int ret
= -ENOPROTOOPT
;
588 #ifdef CONFIG_NETDEVICES
589 struct net
*net
= sock_net(sk
);
590 char devname
[IFNAMSIZ
];
592 if (sk
->sk_bound_dev_if
== 0) {
601 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
605 len
= strlen(devname
) + 1;
608 if (copy_to_user(optval
, devname
, len
))
613 if (put_user(len
, optlen
))
624 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
627 sock_set_flag(sk
, bit
);
629 sock_reset_flag(sk
, bit
);
632 bool sk_mc_loop(struct sock
*sk
)
634 if (dev_recursion_level())
638 switch (sk
->sk_family
) {
640 return inet_sk(sk
)->mc_loop
;
641 #if IS_ENABLED(CONFIG_IPV6)
643 return inet6_sk(sk
)->mc_loop
;
649 EXPORT_SYMBOL(sk_mc_loop
);
652 * This is meant for all protocols to use and covers goings on
653 * at the socket level. Everything here is generic.
656 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
657 char __user
*optval
, unsigned int optlen
)
659 struct sock
*sk
= sock
->sk
;
666 * Options without arguments
669 if (optname
== SO_BINDTODEVICE
)
670 return sock_setbindtodevice(sk
, optval
, optlen
);
672 if (optlen
< sizeof(int))
675 if (get_user(val
, (int __user
*)optval
))
678 valbool
= val
? 1 : 0;
684 if (val
&& !capable(CAP_NET_ADMIN
))
687 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
690 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
693 sk
->sk_reuseport
= valbool
;
702 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
705 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
708 /* Don't error on this BSD doesn't and if you think
709 * about it this is right. Otherwise apps have to
710 * play 'guess the biggest size' games. RCVBUF/SNDBUF
711 * are treated in BSD as hints
713 val
= min_t(u32
, val
, sysctl_wmem_max
);
715 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
716 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
717 /* Wake up sending tasks if we upped the value. */
718 sk
->sk_write_space(sk
);
722 if (!capable(CAP_NET_ADMIN
)) {
729 /* Don't error on this BSD doesn't and if you think
730 * about it this is right. Otherwise apps have to
731 * play 'guess the biggest size' games. RCVBUF/SNDBUF
732 * are treated in BSD as hints
734 val
= min_t(u32
, val
, sysctl_rmem_max
);
736 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
738 * We double it on the way in to account for
739 * "struct sk_buff" etc. overhead. Applications
740 * assume that the SO_RCVBUF setting they make will
741 * allow that much actual data to be received on that
744 * Applications are unaware that "struct sk_buff" and
745 * other overheads allocate from the receive buffer
746 * during socket buffer allocation.
748 * And after considering the possible alternatives,
749 * returning the value we actually used in getsockopt
750 * is the most desirable behavior.
752 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
756 if (!capable(CAP_NET_ADMIN
)) {
764 if (sk
->sk_protocol
== IPPROTO_TCP
&&
765 sk
->sk_type
== SOCK_STREAM
)
766 tcp_set_keepalive(sk
, valbool
);
768 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
772 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
776 sk
->sk_no_check_tx
= valbool
;
780 if ((val
>= 0 && val
<= 6) ||
781 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
782 sk
->sk_priority
= val
;
788 if (optlen
< sizeof(ling
)) {
789 ret
= -EINVAL
; /* 1003.1g */
792 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
797 sock_reset_flag(sk
, SOCK_LINGER
);
799 #if (BITS_PER_LONG == 32)
800 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
801 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
804 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
805 sock_set_flag(sk
, SOCK_LINGER
);
810 sock_warn_obsolete_bsdism("setsockopt");
815 set_bit(SOCK_PASSCRED
, &sock
->flags
);
817 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
823 if (optname
== SO_TIMESTAMP
)
824 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
826 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
827 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
828 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
830 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
831 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
835 case SO_TIMESTAMPING
:
836 if (val
& ~SOF_TIMESTAMPING_MASK
) {
841 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
842 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
843 if (sk
->sk_protocol
== IPPROTO_TCP
&&
844 sk
->sk_type
== SOCK_STREAM
) {
845 if ((1 << sk
->sk_state
) &
846 (TCPF_CLOSE
| TCPF_LISTEN
)) {
850 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
855 sk
->sk_tsflags
= val
;
856 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
857 sock_enable_timestamp(sk
,
858 SOCK_TIMESTAMPING_RX_SOFTWARE
);
860 sock_disable_timestamp(sk
,
861 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
867 sk
->sk_rcvlowat
= val
? : 1;
871 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
875 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
878 case SO_ATTACH_FILTER
:
880 if (optlen
== sizeof(struct sock_fprog
)) {
881 struct sock_fprog fprog
;
884 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
887 ret
= sk_attach_filter(&fprog
, sk
);
893 if (optlen
== sizeof(u32
)) {
897 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
900 ret
= sk_attach_bpf(ufd
, sk
);
904 case SO_ATTACH_REUSEPORT_CBPF
:
906 if (optlen
== sizeof(struct sock_fprog
)) {
907 struct sock_fprog fprog
;
910 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
913 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
917 case SO_ATTACH_REUSEPORT_EBPF
:
919 if (optlen
== sizeof(u32
)) {
923 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
926 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
930 case SO_DETACH_FILTER
:
931 ret
= sk_detach_filter(sk
);
935 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
938 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
943 set_bit(SOCK_PASSSEC
, &sock
->flags
);
945 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
948 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
955 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
959 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
963 if (sock
->ops
->set_peek_off
)
964 ret
= sock
->ops
->set_peek_off(sk
, val
);
970 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
973 case SO_SELECT_ERR_QUEUE
:
974 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
977 #ifdef CONFIG_NET_RX_BUSY_POLL
979 /* allow unprivileged users to decrease the value */
980 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
986 sk
->sk_ll_usec
= val
;
991 case SO_MAX_PACING_RATE
:
992 sk
->sk_max_pacing_rate
= val
;
993 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
994 sk
->sk_max_pacing_rate
);
997 case SO_INCOMING_CPU
:
998 sk
->sk_incoming_cpu
= val
;
1003 dst_negative_advice(sk
);
1012 EXPORT_SYMBOL(sock_setsockopt
);
1015 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1016 struct ucred
*ucred
)
1018 ucred
->pid
= pid_vnr(pid
);
1019 ucred
->uid
= ucred
->gid
= -1;
1021 struct user_namespace
*current_ns
= current_user_ns();
1023 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1024 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1028 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1029 char __user
*optval
, int __user
*optlen
)
1031 struct sock
*sk
= sock
->sk
;
1039 int lv
= sizeof(int);
1042 if (get_user(len
, optlen
))
1047 memset(&v
, 0, sizeof(v
));
1051 v
.val
= sock_flag(sk
, SOCK_DBG
);
1055 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1059 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1063 v
.val
= sk
->sk_sndbuf
;
1067 v
.val
= sk
->sk_rcvbuf
;
1071 v
.val
= sk
->sk_reuse
;
1075 v
.val
= sk
->sk_reuseport
;
1079 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1083 v
.val
= sk
->sk_type
;
1087 v
.val
= sk
->sk_protocol
;
1091 v
.val
= sk
->sk_family
;
1095 v
.val
= -sock_error(sk
);
1097 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1101 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1105 v
.val
= sk
->sk_no_check_tx
;
1109 v
.val
= sk
->sk_priority
;
1113 lv
= sizeof(v
.ling
);
1114 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1115 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1119 sock_warn_obsolete_bsdism("getsockopt");
1123 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1124 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1127 case SO_TIMESTAMPNS
:
1128 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1131 case SO_TIMESTAMPING
:
1132 v
.val
= sk
->sk_tsflags
;
1136 lv
= sizeof(struct timeval
);
1137 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1141 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1142 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1147 lv
= sizeof(struct timeval
);
1148 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1152 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1153 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1158 v
.val
= sk
->sk_rcvlowat
;
1166 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1171 struct ucred peercred
;
1172 if (len
> sizeof(peercred
))
1173 len
= sizeof(peercred
);
1174 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1175 if (copy_to_user(optval
, &peercred
, len
))
1184 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1188 if (copy_to_user(optval
, address
, len
))
1193 /* Dubious BSD thing... Probably nobody even uses it, but
1194 * the UNIX standard wants it for whatever reason... -DaveM
1197 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1201 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1205 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1208 v
.val
= sk
->sk_mark
;
1212 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1215 case SO_WIFI_STATUS
:
1216 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1220 if (!sock
->ops
->set_peek_off
)
1223 v
.val
= sk
->sk_peek_off
;
1226 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1229 case SO_BINDTODEVICE
:
1230 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1233 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1239 case SO_LOCK_FILTER
:
1240 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1243 case SO_BPF_EXTENSIONS
:
1244 v
.val
= bpf_tell_extensions();
1247 case SO_SELECT_ERR_QUEUE
:
1248 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1251 #ifdef CONFIG_NET_RX_BUSY_POLL
1253 v
.val
= sk
->sk_ll_usec
;
1257 case SO_MAX_PACING_RATE
:
1258 v
.val
= sk
->sk_max_pacing_rate
;
1261 case SO_INCOMING_CPU
:
1262 v
.val
= sk
->sk_incoming_cpu
;
1266 /* We implement the SO_SNDLOWAT etc to not be settable
1269 return -ENOPROTOOPT
;
1274 if (copy_to_user(optval
, &v
, len
))
1277 if (put_user(len
, optlen
))
1283 * Initialize an sk_lock.
1285 * (We also register the sk_lock with the lock validator.)
1287 static inline void sock_lock_init(struct sock
*sk
)
1289 sock_lock_init_class_and_name(sk
,
1290 af_family_slock_key_strings
[sk
->sk_family
],
1291 af_family_slock_keys
+ sk
->sk_family
,
1292 af_family_key_strings
[sk
->sk_family
],
1293 af_family_keys
+ sk
->sk_family
);
1297 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1298 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1299 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1301 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1303 #ifdef CONFIG_SECURITY_NETWORK
1304 void *sptr
= nsk
->sk_security
;
1306 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1308 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1309 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1311 #ifdef CONFIG_SECURITY_NETWORK
1312 nsk
->sk_security
= sptr
;
1313 security_sk_clone(osk
, nsk
);
1317 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1319 unsigned long nulls1
, nulls2
;
1321 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1322 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1323 if (nulls1
> nulls2
)
1324 swap(nulls1
, nulls2
);
1327 memset((char *)sk
, 0, nulls1
);
1328 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1329 nulls2
- nulls1
- sizeof(void *));
1330 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1331 size
- nulls2
- sizeof(void *));
1333 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1335 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1339 struct kmem_cache
*slab
;
1343 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1346 if (priority
& __GFP_ZERO
) {
1348 prot
->clear_sk(sk
, prot
->obj_size
);
1350 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1353 sk
= kmalloc(prot
->obj_size
, priority
);
1356 kmemcheck_annotate_bitfield(sk
, flags
);
1358 if (security_sk_alloc(sk
, family
, priority
))
1361 if (!try_module_get(prot
->owner
))
1363 sk_tx_queue_clear(sk
);
1364 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1370 security_sk_free(sk
);
1373 kmem_cache_free(slab
, sk
);
1379 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1381 struct kmem_cache
*slab
;
1382 struct module
*owner
;
1384 owner
= prot
->owner
;
1387 cgroup_sk_free(&sk
->sk_cgrp_data
);
1388 security_sk_free(sk
);
1390 kmem_cache_free(slab
, sk
);
1397 * sk_alloc - All socket objects are allocated here
1398 * @net: the applicable net namespace
1399 * @family: protocol family
1400 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1401 * @prot: struct proto associated with this new sock instance
1402 * @kern: is this to be a kernel socket?
1404 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1405 struct proto
*prot
, int kern
)
1409 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1411 sk
->sk_family
= family
;
1413 * See comment in struct sock definition to understand
1414 * why we need sk_prot_creator -acme
1416 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1418 sk
->sk_net_refcnt
= kern
? 0 : 1;
1419 if (likely(sk
->sk_net_refcnt
))
1421 sock_net_set(sk
, net
);
1422 atomic_set(&sk
->sk_wmem_alloc
, 1);
1424 sock_update_classid(&sk
->sk_cgrp_data
);
1425 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1430 EXPORT_SYMBOL(sk_alloc
);
1432 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1433 * grace period. This is the case for UDP sockets and TCP listeners.
1435 static void __sk_destruct(struct rcu_head
*head
)
1437 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1438 struct sk_filter
*filter
;
1440 if (sk
->sk_destruct
)
1441 sk
->sk_destruct(sk
);
1443 filter
= rcu_dereference_check(sk
->sk_filter
,
1444 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1446 sk_filter_uncharge(sk
, filter
);
1447 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1449 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1450 reuseport_detach_sock(sk
);
1452 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1454 if (atomic_read(&sk
->sk_omem_alloc
))
1455 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1456 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1458 if (sk
->sk_peer_cred
)
1459 put_cred(sk
->sk_peer_cred
);
1460 put_pid(sk
->sk_peer_pid
);
1461 if (likely(sk
->sk_net_refcnt
))
1462 put_net(sock_net(sk
));
1463 sk_prot_free(sk
->sk_prot_creator
, sk
);
1466 void sk_destruct(struct sock
*sk
)
1468 if (sock_flag(sk
, SOCK_RCU_FREE
))
1469 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1471 __sk_destruct(&sk
->sk_rcu
);
1474 static void __sk_free(struct sock
*sk
)
1476 if (unlikely(sock_diag_has_destroy_listeners(sk
) && sk
->sk_net_refcnt
))
1477 sock_diag_broadcast_destroy(sk
);
1482 void sk_free(struct sock
*sk
)
1485 * We subtract one from sk_wmem_alloc and can know if
1486 * some packets are still in some tx queue.
1487 * If not null, sock_wfree() will call __sk_free(sk) later
1489 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1492 EXPORT_SYMBOL(sk_free
);
1495 * sk_clone_lock - clone a socket, and lock its clone
1496 * @sk: the socket to clone
1497 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1499 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1501 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1504 bool is_charged
= true;
1506 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1507 if (newsk
!= NULL
) {
1508 struct sk_filter
*filter
;
1510 sock_copy(newsk
, sk
);
1513 if (likely(newsk
->sk_net_refcnt
))
1514 get_net(sock_net(newsk
));
1515 sk_node_init(&newsk
->sk_node
);
1516 sock_lock_init(newsk
);
1517 bh_lock_sock(newsk
);
1518 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1519 newsk
->sk_backlog
.len
= 0;
1521 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1523 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1525 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1526 atomic_set(&newsk
->sk_omem_alloc
, 0);
1527 skb_queue_head_init(&newsk
->sk_receive_queue
);
1528 skb_queue_head_init(&newsk
->sk_write_queue
);
1530 rwlock_init(&newsk
->sk_callback_lock
);
1531 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1532 af_callback_keys
+ newsk
->sk_family
,
1533 af_family_clock_key_strings
[newsk
->sk_family
]);
1535 newsk
->sk_dst_cache
= NULL
;
1536 newsk
->sk_wmem_queued
= 0;
1537 newsk
->sk_forward_alloc
= 0;
1538 atomic_set(&newsk
->sk_drops
, 0);
1539 newsk
->sk_send_head
= NULL
;
1540 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1542 sock_reset_flag(newsk
, SOCK_DONE
);
1543 skb_queue_head_init(&newsk
->sk_error_queue
);
1545 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1547 /* though it's an empty new sock, the charging may fail
1548 * if sysctl_optmem_max was changed between creation of
1549 * original socket and cloning
1551 is_charged
= sk_filter_charge(newsk
, filter
);
1553 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1554 /* It is still raw copy of parent, so invalidate
1555 * destructor and make plain sk_free() */
1556 newsk
->sk_destruct
= NULL
;
1557 bh_unlock_sock(newsk
);
1562 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1565 newsk
->sk_priority
= 0;
1566 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1567 atomic64_set(&newsk
->sk_cookie
, 0);
1569 * Before updating sk_refcnt, we must commit prior changes to memory
1570 * (Documentation/RCU/rculist_nulls.txt for details)
1573 atomic_set(&newsk
->sk_refcnt
, 2);
1576 * Increment the counter in the same struct proto as the master
1577 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1578 * is the same as sk->sk_prot->socks, as this field was copied
1581 * This _changes_ the previous behaviour, where
1582 * tcp_create_openreq_child always was incrementing the
1583 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1584 * to be taken into account in all callers. -acme
1586 sk_refcnt_debug_inc(newsk
);
1587 sk_set_socket(newsk
, NULL
);
1588 newsk
->sk_wq
= NULL
;
1590 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
1591 sock_update_memcg(newsk
);
1593 if (newsk
->sk_prot
->sockets_allocated
)
1594 sk_sockets_allocated_inc(newsk
);
1596 if (sock_needs_netstamp(sk
) &&
1597 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1598 net_enable_timestamp();
1603 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1605 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1609 sk_dst_set(sk
, dst
);
1610 sk
->sk_route_caps
= dst
->dev
->features
;
1611 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1612 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1613 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1614 if (sk_can_gso(sk
)) {
1615 if (dst
->header_len
) {
1616 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1618 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1619 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1620 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1623 sk
->sk_gso_max_segs
= max_segs
;
1625 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1628 * Simple resource managers for sockets.
1633 * Write buffer destructor automatically called from kfree_skb.
1635 void sock_wfree(struct sk_buff
*skb
)
1637 struct sock
*sk
= skb
->sk
;
1638 unsigned int len
= skb
->truesize
;
1640 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1642 * Keep a reference on sk_wmem_alloc, this will be released
1643 * after sk_write_space() call
1645 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1646 sk
->sk_write_space(sk
);
1650 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1651 * could not do because of in-flight packets
1653 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1656 EXPORT_SYMBOL(sock_wfree
);
1658 /* This variant of sock_wfree() is used by TCP,
1659 * since it sets SOCK_USE_WRITE_QUEUE.
1661 void __sock_wfree(struct sk_buff
*skb
)
1663 struct sock
*sk
= skb
->sk
;
1665 if (atomic_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
))
1669 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1674 if (unlikely(!sk_fullsock(sk
))) {
1675 skb
->destructor
= sock_edemux
;
1680 skb
->destructor
= sock_wfree
;
1681 skb_set_hash_from_sk(skb
, sk
);
1683 * We used to take a refcount on sk, but following operation
1684 * is enough to guarantee sk_free() wont free this sock until
1685 * all in-flight packets are completed
1687 atomic_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1689 EXPORT_SYMBOL(skb_set_owner_w
);
1691 /* This helper is used by netem, as it can hold packets in its
1692 * delay queue. We want to allow the owner socket to send more
1693 * packets, as if they were already TX completed by a typical driver.
1694 * But we also want to keep skb->sk set because some packet schedulers
1695 * rely on it (sch_fq for example). So we set skb->truesize to a small
1696 * amount (1) and decrease sk_wmem_alloc accordingly.
1698 void skb_orphan_partial(struct sk_buff
*skb
)
1700 /* If this skb is a TCP pure ACK or already went here,
1701 * we have nothing to do. 2 is already a very small truesize.
1703 if (skb
->truesize
<= 2)
1706 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1707 * so we do not completely orphan skb, but transfert all
1708 * accounted bytes but one, to avoid unexpected reorders.
1710 if (skb
->destructor
== sock_wfree
1712 || skb
->destructor
== tcp_wfree
1715 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1721 EXPORT_SYMBOL(skb_orphan_partial
);
1724 * Read buffer destructor automatically called from kfree_skb.
1726 void sock_rfree(struct sk_buff
*skb
)
1728 struct sock
*sk
= skb
->sk
;
1729 unsigned int len
= skb
->truesize
;
1731 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1732 sk_mem_uncharge(sk
, len
);
1734 EXPORT_SYMBOL(sock_rfree
);
1737 * Buffer destructor for skbs that are not used directly in read or write
1738 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1740 void sock_efree(struct sk_buff
*skb
)
1744 EXPORT_SYMBOL(sock_efree
);
1746 kuid_t
sock_i_uid(struct sock
*sk
)
1750 read_lock_bh(&sk
->sk_callback_lock
);
1751 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1752 read_unlock_bh(&sk
->sk_callback_lock
);
1755 EXPORT_SYMBOL(sock_i_uid
);
1757 unsigned long sock_i_ino(struct sock
*sk
)
1761 read_lock_bh(&sk
->sk_callback_lock
);
1762 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1763 read_unlock_bh(&sk
->sk_callback_lock
);
1766 EXPORT_SYMBOL(sock_i_ino
);
1769 * Allocate a skb from the socket's send buffer.
1771 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1774 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1775 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1777 skb_set_owner_w(skb
, sk
);
1783 EXPORT_SYMBOL(sock_wmalloc
);
1786 * Allocate a memory block from the socket's option memory buffer.
1788 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1790 if ((unsigned int)size
<= sysctl_optmem_max
&&
1791 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1793 /* First do the add, to avoid the race if kmalloc
1796 atomic_add(size
, &sk
->sk_omem_alloc
);
1797 mem
= kmalloc(size
, priority
);
1800 atomic_sub(size
, &sk
->sk_omem_alloc
);
1804 EXPORT_SYMBOL(sock_kmalloc
);
1806 /* Free an option memory block. Note, we actually want the inline
1807 * here as this allows gcc to detect the nullify and fold away the
1808 * condition entirely.
1810 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1813 if (WARN_ON_ONCE(!mem
))
1819 atomic_sub(size
, &sk
->sk_omem_alloc
);
1822 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1824 __sock_kfree_s(sk
, mem
, size
, false);
1826 EXPORT_SYMBOL(sock_kfree_s
);
1828 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1830 __sock_kfree_s(sk
, mem
, size
, true);
1832 EXPORT_SYMBOL(sock_kzfree_s
);
1834 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1835 I think, these locks should be removed for datagram sockets.
1837 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1841 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1845 if (signal_pending(current
))
1847 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1848 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1849 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1851 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1855 timeo
= schedule_timeout(timeo
);
1857 finish_wait(sk_sleep(sk
), &wait
);
1863 * Generic send/receive buffer handlers
1866 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1867 unsigned long data_len
, int noblock
,
1868 int *errcode
, int max_page_order
)
1870 struct sk_buff
*skb
;
1874 timeo
= sock_sndtimeo(sk
, noblock
);
1876 err
= sock_error(sk
);
1881 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1884 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1887 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1888 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1892 if (signal_pending(current
))
1894 timeo
= sock_wait_for_wmem(sk
, timeo
);
1896 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
1897 errcode
, sk
->sk_allocation
);
1899 skb_set_owner_w(skb
, sk
);
1903 err
= sock_intr_errno(timeo
);
1908 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1910 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1911 int noblock
, int *errcode
)
1913 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1915 EXPORT_SYMBOL(sock_alloc_send_skb
);
1917 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
1918 struct sockcm_cookie
*sockc
)
1922 switch (cmsg
->cmsg_type
) {
1924 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
1926 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1928 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
1930 case SO_TIMESTAMPING
:
1931 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1934 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
1935 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
1938 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
1939 sockc
->tsflags
|= tsflags
;
1946 EXPORT_SYMBOL(__sock_cmsg_send
);
1948 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
1949 struct sockcm_cookie
*sockc
)
1951 struct cmsghdr
*cmsg
;
1954 for_each_cmsghdr(cmsg
, msg
) {
1955 if (!CMSG_OK(msg
, cmsg
))
1957 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
1959 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
1965 EXPORT_SYMBOL(sock_cmsg_send
);
1967 /* On 32bit arches, an skb frag is limited to 2^15 */
1968 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1971 * skb_page_frag_refill - check that a page_frag contains enough room
1972 * @sz: minimum size of the fragment we want to get
1973 * @pfrag: pointer to page_frag
1974 * @gfp: priority for memory allocation
1976 * Note: While this allocator tries to use high order pages, there is
1977 * no guarantee that allocations succeed. Therefore, @sz MUST be
1978 * less or equal than PAGE_SIZE.
1980 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
1983 if (page_ref_count(pfrag
->page
) == 1) {
1987 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1989 put_page(pfrag
->page
);
1993 if (SKB_FRAG_PAGE_ORDER
) {
1994 /* Avoid direct reclaim but allow kswapd to wake */
1995 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
1996 __GFP_COMP
| __GFP_NOWARN
|
1998 SKB_FRAG_PAGE_ORDER
);
1999 if (likely(pfrag
->page
)) {
2000 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
2004 pfrag
->page
= alloc_page(gfp
);
2005 if (likely(pfrag
->page
)) {
2006 pfrag
->size
= PAGE_SIZE
;
2011 EXPORT_SYMBOL(skb_page_frag_refill
);
2013 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
2015 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
2018 sk_enter_memory_pressure(sk
);
2019 sk_stream_moderate_sndbuf(sk
);
2022 EXPORT_SYMBOL(sk_page_frag_refill
);
2024 static void __lock_sock(struct sock
*sk
)
2025 __releases(&sk
->sk_lock
.slock
)
2026 __acquires(&sk
->sk_lock
.slock
)
2031 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2032 TASK_UNINTERRUPTIBLE
);
2033 spin_unlock_bh(&sk
->sk_lock
.slock
);
2035 spin_lock_bh(&sk
->sk_lock
.slock
);
2036 if (!sock_owned_by_user(sk
))
2039 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2042 static void __release_sock(struct sock
*sk
)
2043 __releases(&sk
->sk_lock
.slock
)
2044 __acquires(&sk
->sk_lock
.slock
)
2046 struct sk_buff
*skb
, *next
;
2048 while ((skb
= sk
->sk_backlog
.head
) != NULL
) {
2049 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2051 spin_unlock_bh(&sk
->sk_lock
.slock
);
2056 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2058 sk_backlog_rcv(sk
, skb
);
2063 } while (skb
!= NULL
);
2065 spin_lock_bh(&sk
->sk_lock
.slock
);
2069 * Doing the zeroing here guarantee we can not loop forever
2070 * while a wild producer attempts to flood us.
2072 sk
->sk_backlog
.len
= 0;
2075 void __sk_flush_backlog(struct sock
*sk
)
2077 spin_lock_bh(&sk
->sk_lock
.slock
);
2079 spin_unlock_bh(&sk
->sk_lock
.slock
);
2083 * sk_wait_data - wait for data to arrive at sk_receive_queue
2084 * @sk: sock to wait on
2085 * @timeo: for how long
2086 * @skb: last skb seen on sk_receive_queue
2088 * Now socket state including sk->sk_err is changed only under lock,
2089 * hence we may omit checks after joining wait queue.
2090 * We check receive queue before schedule() only as optimization;
2091 * it is very likely that release_sock() added new data.
2093 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2098 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2099 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2100 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
);
2101 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2102 finish_wait(sk_sleep(sk
), &wait
);
2105 EXPORT_SYMBOL(sk_wait_data
);
2108 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2110 * @size: memory size to allocate
2111 * @kind: allocation type
2113 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2114 * rmem allocation. This function assumes that protocols which have
2115 * memory_pressure use sk_wmem_queued as write buffer accounting.
2117 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2119 struct proto
*prot
= sk
->sk_prot
;
2120 int amt
= sk_mem_pages(size
);
2123 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2125 allocated
= sk_memory_allocated_add(sk
, amt
);
2127 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2128 !mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
))
2129 goto suppress_allocation
;
2132 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2133 sk_leave_memory_pressure(sk
);
2137 /* Under pressure. */
2138 if (allocated
> sk_prot_mem_limits(sk
, 1))
2139 sk_enter_memory_pressure(sk
);
2141 /* Over hard limit. */
2142 if (allocated
> sk_prot_mem_limits(sk
, 2))
2143 goto suppress_allocation
;
2145 /* guarantee minimum buffer size under pressure */
2146 if (kind
== SK_MEM_RECV
) {
2147 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2150 } else { /* SK_MEM_SEND */
2151 if (sk
->sk_type
== SOCK_STREAM
) {
2152 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2154 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2155 prot
->sysctl_wmem
[0])
2159 if (sk_has_memory_pressure(sk
)) {
2162 if (!sk_under_memory_pressure(sk
))
2164 alloc
= sk_sockets_allocated_read_positive(sk
);
2165 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2166 sk_mem_pages(sk
->sk_wmem_queued
+
2167 atomic_read(&sk
->sk_rmem_alloc
) +
2168 sk
->sk_forward_alloc
))
2172 suppress_allocation
:
2174 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2175 sk_stream_moderate_sndbuf(sk
);
2177 /* Fail only if socket is _under_ its sndbuf.
2178 * In this case we cannot block, so that we have to fail.
2180 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2184 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2186 /* Alas. Undo changes. */
2187 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2189 sk_memory_allocated_sub(sk
, amt
);
2191 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2192 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2196 EXPORT_SYMBOL(__sk_mem_schedule
);
2199 * __sk_mem_reclaim - reclaim memory_allocated
2201 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2203 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2205 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2206 sk_memory_allocated_sub(sk
, amount
);
2207 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2209 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2210 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2212 if (sk_under_memory_pressure(sk
) &&
2213 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2214 sk_leave_memory_pressure(sk
);
2216 EXPORT_SYMBOL(__sk_mem_reclaim
);
2218 int sk_set_peek_off(struct sock
*sk
, int val
)
2223 sk
->sk_peek_off
= val
;
2226 EXPORT_SYMBOL_GPL(sk_set_peek_off
);
2229 * Set of default routines for initialising struct proto_ops when
2230 * the protocol does not support a particular function. In certain
2231 * cases where it makes no sense for a protocol to have a "do nothing"
2232 * function, some default processing is provided.
2235 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2239 EXPORT_SYMBOL(sock_no_bind
);
2241 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2246 EXPORT_SYMBOL(sock_no_connect
);
2248 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2252 EXPORT_SYMBOL(sock_no_socketpair
);
2254 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2258 EXPORT_SYMBOL(sock_no_accept
);
2260 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2265 EXPORT_SYMBOL(sock_no_getname
);
2267 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2271 EXPORT_SYMBOL(sock_no_poll
);
2273 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2277 EXPORT_SYMBOL(sock_no_ioctl
);
2279 int sock_no_listen(struct socket
*sock
, int backlog
)
2283 EXPORT_SYMBOL(sock_no_listen
);
2285 int sock_no_shutdown(struct socket
*sock
, int how
)
2289 EXPORT_SYMBOL(sock_no_shutdown
);
2291 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2292 char __user
*optval
, unsigned int optlen
)
2296 EXPORT_SYMBOL(sock_no_setsockopt
);
2298 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2299 char __user
*optval
, int __user
*optlen
)
2303 EXPORT_SYMBOL(sock_no_getsockopt
);
2305 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2309 EXPORT_SYMBOL(sock_no_sendmsg
);
2311 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2316 EXPORT_SYMBOL(sock_no_recvmsg
);
2318 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2320 /* Mirror missing mmap method error code */
2323 EXPORT_SYMBOL(sock_no_mmap
);
2325 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2328 struct msghdr msg
= {.msg_flags
= flags
};
2330 char *kaddr
= kmap(page
);
2331 iov
.iov_base
= kaddr
+ offset
;
2333 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2337 EXPORT_SYMBOL(sock_no_sendpage
);
2340 * Default Socket Callbacks
2343 static void sock_def_wakeup(struct sock
*sk
)
2345 struct socket_wq
*wq
;
2348 wq
= rcu_dereference(sk
->sk_wq
);
2349 if (skwq_has_sleeper(wq
))
2350 wake_up_interruptible_all(&wq
->wait
);
2354 static void sock_def_error_report(struct sock
*sk
)
2356 struct socket_wq
*wq
;
2359 wq
= rcu_dereference(sk
->sk_wq
);
2360 if (skwq_has_sleeper(wq
))
2361 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2362 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2366 static void sock_def_readable(struct sock
*sk
)
2368 struct socket_wq
*wq
;
2371 wq
= rcu_dereference(sk
->sk_wq
);
2372 if (skwq_has_sleeper(wq
))
2373 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2374 POLLRDNORM
| POLLRDBAND
);
2375 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2379 static void sock_def_write_space(struct sock
*sk
)
2381 struct socket_wq
*wq
;
2385 /* Do not wake up a writer until he can make "significant"
2388 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2389 wq
= rcu_dereference(sk
->sk_wq
);
2390 if (skwq_has_sleeper(wq
))
2391 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2392 POLLWRNORM
| POLLWRBAND
);
2394 /* Should agree with poll, otherwise some programs break */
2395 if (sock_writeable(sk
))
2396 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2402 static void sock_def_destruct(struct sock
*sk
)
2406 void sk_send_sigurg(struct sock
*sk
)
2408 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2409 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2410 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2412 EXPORT_SYMBOL(sk_send_sigurg
);
2414 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2415 unsigned long expires
)
2417 if (!mod_timer(timer
, expires
))
2420 EXPORT_SYMBOL(sk_reset_timer
);
2422 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2424 if (del_timer(timer
))
2427 EXPORT_SYMBOL(sk_stop_timer
);
2429 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2431 skb_queue_head_init(&sk
->sk_receive_queue
);
2432 skb_queue_head_init(&sk
->sk_write_queue
);
2433 skb_queue_head_init(&sk
->sk_error_queue
);
2435 sk
->sk_send_head
= NULL
;
2437 init_timer(&sk
->sk_timer
);
2439 sk
->sk_allocation
= GFP_KERNEL
;
2440 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2441 sk
->sk_sndbuf
= sysctl_wmem_default
;
2442 sk
->sk_state
= TCP_CLOSE
;
2443 sk_set_socket(sk
, sock
);
2445 sock_set_flag(sk
, SOCK_ZAPPED
);
2448 sk
->sk_type
= sock
->type
;
2449 sk
->sk_wq
= sock
->wq
;
2454 rwlock_init(&sk
->sk_callback_lock
);
2455 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2456 af_callback_keys
+ sk
->sk_family
,
2457 af_family_clock_key_strings
[sk
->sk_family
]);
2459 sk
->sk_state_change
= sock_def_wakeup
;
2460 sk
->sk_data_ready
= sock_def_readable
;
2461 sk
->sk_write_space
= sock_def_write_space
;
2462 sk
->sk_error_report
= sock_def_error_report
;
2463 sk
->sk_destruct
= sock_def_destruct
;
2465 sk
->sk_frag
.page
= NULL
;
2466 sk
->sk_frag
.offset
= 0;
2467 sk
->sk_peek_off
= -1;
2469 sk
->sk_peer_pid
= NULL
;
2470 sk
->sk_peer_cred
= NULL
;
2471 sk
->sk_write_pending
= 0;
2472 sk
->sk_rcvlowat
= 1;
2473 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2474 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2476 sk
->sk_stamp
= ktime_set(-1L, 0);
2478 #ifdef CONFIG_NET_RX_BUSY_POLL
2480 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2483 sk
->sk_max_pacing_rate
= ~0U;
2484 sk
->sk_pacing_rate
= ~0U;
2485 sk
->sk_incoming_cpu
= -1;
2487 * Before updating sk_refcnt, we must commit prior changes to memory
2488 * (Documentation/RCU/rculist_nulls.txt for details)
2491 atomic_set(&sk
->sk_refcnt
, 1);
2492 atomic_set(&sk
->sk_drops
, 0);
2494 EXPORT_SYMBOL(sock_init_data
);
2496 void lock_sock_nested(struct sock
*sk
, int subclass
)
2499 spin_lock_bh(&sk
->sk_lock
.slock
);
2500 if (sk
->sk_lock
.owned
)
2502 sk
->sk_lock
.owned
= 1;
2503 spin_unlock(&sk
->sk_lock
.slock
);
2505 * The sk_lock has mutex_lock() semantics here:
2507 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2510 EXPORT_SYMBOL(lock_sock_nested
);
2512 void release_sock(struct sock
*sk
)
2514 spin_lock_bh(&sk
->sk_lock
.slock
);
2515 if (sk
->sk_backlog
.tail
)
2518 /* Warning : release_cb() might need to release sk ownership,
2519 * ie call sock_release_ownership(sk) before us.
2521 if (sk
->sk_prot
->release_cb
)
2522 sk
->sk_prot
->release_cb(sk
);
2524 sock_release_ownership(sk
);
2525 if (waitqueue_active(&sk
->sk_lock
.wq
))
2526 wake_up(&sk
->sk_lock
.wq
);
2527 spin_unlock_bh(&sk
->sk_lock
.slock
);
2529 EXPORT_SYMBOL(release_sock
);
2532 * lock_sock_fast - fast version of lock_sock
2535 * This version should be used for very small section, where process wont block
2536 * return false if fast path is taken
2537 * sk_lock.slock locked, owned = 0, BH disabled
2538 * return true if slow path is taken
2539 * sk_lock.slock unlocked, owned = 1, BH enabled
2541 bool lock_sock_fast(struct sock
*sk
)
2544 spin_lock_bh(&sk
->sk_lock
.slock
);
2546 if (!sk
->sk_lock
.owned
)
2548 * Note : We must disable BH
2553 sk
->sk_lock
.owned
= 1;
2554 spin_unlock(&sk
->sk_lock
.slock
);
2556 * The sk_lock has mutex_lock() semantics here:
2558 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2562 EXPORT_SYMBOL(lock_sock_fast
);
2564 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2567 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2568 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2569 tv
= ktime_to_timeval(sk
->sk_stamp
);
2570 if (tv
.tv_sec
== -1)
2572 if (tv
.tv_sec
== 0) {
2573 sk
->sk_stamp
= ktime_get_real();
2574 tv
= ktime_to_timeval(sk
->sk_stamp
);
2576 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2578 EXPORT_SYMBOL(sock_get_timestamp
);
2580 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2583 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2584 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2585 ts
= ktime_to_timespec(sk
->sk_stamp
);
2586 if (ts
.tv_sec
== -1)
2588 if (ts
.tv_sec
== 0) {
2589 sk
->sk_stamp
= ktime_get_real();
2590 ts
= ktime_to_timespec(sk
->sk_stamp
);
2592 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2594 EXPORT_SYMBOL(sock_get_timestampns
);
2596 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2598 if (!sock_flag(sk
, flag
)) {
2599 unsigned long previous_flags
= sk
->sk_flags
;
2601 sock_set_flag(sk
, flag
);
2603 * we just set one of the two flags which require net
2604 * time stamping, but time stamping might have been on
2605 * already because of the other one
2607 if (sock_needs_netstamp(sk
) &&
2608 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
2609 net_enable_timestamp();
2613 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2614 int level
, int type
)
2616 struct sock_exterr_skb
*serr
;
2617 struct sk_buff
*skb
;
2621 skb
= sock_dequeue_err_skb(sk
);
2627 msg
->msg_flags
|= MSG_TRUNC
;
2630 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2634 sock_recv_timestamp(msg
, sk
, skb
);
2636 serr
= SKB_EXT_ERR(skb
);
2637 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2639 msg
->msg_flags
|= MSG_ERRQUEUE
;
2647 EXPORT_SYMBOL(sock_recv_errqueue
);
2650 * Get a socket option on an socket.
2652 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2653 * asynchronous errors should be reported by getsockopt. We assume
2654 * this means if you specify SO_ERROR (otherwise whats the point of it).
2656 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2657 char __user
*optval
, int __user
*optlen
)
2659 struct sock
*sk
= sock
->sk
;
2661 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2663 EXPORT_SYMBOL(sock_common_getsockopt
);
2665 #ifdef CONFIG_COMPAT
2666 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2667 char __user
*optval
, int __user
*optlen
)
2669 struct sock
*sk
= sock
->sk
;
2671 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2672 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2674 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2676 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2679 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2682 struct sock
*sk
= sock
->sk
;
2686 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2687 flags
& ~MSG_DONTWAIT
, &addr_len
);
2689 msg
->msg_namelen
= addr_len
;
2692 EXPORT_SYMBOL(sock_common_recvmsg
);
2695 * Set socket options on an inet socket.
2697 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2698 char __user
*optval
, unsigned int optlen
)
2700 struct sock
*sk
= sock
->sk
;
2702 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2704 EXPORT_SYMBOL(sock_common_setsockopt
);
2706 #ifdef CONFIG_COMPAT
2707 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2708 char __user
*optval
, unsigned int optlen
)
2710 struct sock
*sk
= sock
->sk
;
2712 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2713 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2715 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2717 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2720 void sk_common_release(struct sock
*sk
)
2722 if (sk
->sk_prot
->destroy
)
2723 sk
->sk_prot
->destroy(sk
);
2726 * Observation: when sock_common_release is called, processes have
2727 * no access to socket. But net still has.
2728 * Step one, detach it from networking:
2730 * A. Remove from hash tables.
2733 sk
->sk_prot
->unhash(sk
);
2736 * In this point socket cannot receive new packets, but it is possible
2737 * that some packets are in flight because some CPU runs receiver and
2738 * did hash table lookup before we unhashed socket. They will achieve
2739 * receive queue and will be purged by socket destructor.
2741 * Also we still have packets pending on receive queue and probably,
2742 * our own packets waiting in device queues. sock_destroy will drain
2743 * receive queue, but transmitted packets will delay socket destruction
2744 * until the last reference will be released.
2749 xfrm_sk_free_policy(sk
);
2751 sk_refcnt_debug_release(sk
);
2753 if (sk
->sk_frag
.page
) {
2754 put_page(sk
->sk_frag
.page
);
2755 sk
->sk_frag
.page
= NULL
;
2760 EXPORT_SYMBOL(sk_common_release
);
2762 #ifdef CONFIG_PROC_FS
2763 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2765 int val
[PROTO_INUSE_NR
];
2768 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2770 #ifdef CONFIG_NET_NS
2771 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2773 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2775 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2777 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2779 int cpu
, idx
= prot
->inuse_idx
;
2782 for_each_possible_cpu(cpu
)
2783 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2785 return res
>= 0 ? res
: 0;
2787 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2789 static int __net_init
sock_inuse_init_net(struct net
*net
)
2791 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2792 return net
->core
.inuse
? 0 : -ENOMEM
;
2795 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2797 free_percpu(net
->core
.inuse
);
2800 static struct pernet_operations net_inuse_ops
= {
2801 .init
= sock_inuse_init_net
,
2802 .exit
= sock_inuse_exit_net
,
2805 static __init
int net_inuse_init(void)
2807 if (register_pernet_subsys(&net_inuse_ops
))
2808 panic("Cannot initialize net inuse counters");
2813 core_initcall(net_inuse_init
);
2815 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2817 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2819 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2821 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2823 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2825 int cpu
, idx
= prot
->inuse_idx
;
2828 for_each_possible_cpu(cpu
)
2829 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2831 return res
>= 0 ? res
: 0;
2833 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2836 static void assign_proto_idx(struct proto
*prot
)
2838 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2840 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2841 pr_err("PROTO_INUSE_NR exhausted\n");
2845 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2848 static void release_proto_idx(struct proto
*prot
)
2850 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2851 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2854 static inline void assign_proto_idx(struct proto
*prot
)
2858 static inline void release_proto_idx(struct proto
*prot
)
2863 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
2867 kfree(rsk_prot
->slab_name
);
2868 rsk_prot
->slab_name
= NULL
;
2869 kmem_cache_destroy(rsk_prot
->slab
);
2870 rsk_prot
->slab
= NULL
;
2873 static int req_prot_init(const struct proto
*prot
)
2875 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
2880 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
2882 if (!rsk_prot
->slab_name
)
2885 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
2886 rsk_prot
->obj_size
, 0,
2887 prot
->slab_flags
, NULL
);
2889 if (!rsk_prot
->slab
) {
2890 pr_crit("%s: Can't create request sock SLAB cache!\n",
2897 int proto_register(struct proto
*prot
, int alloc_slab
)
2900 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2901 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2904 if (prot
->slab
== NULL
) {
2905 pr_crit("%s: Can't create sock SLAB cache!\n",
2910 if (req_prot_init(prot
))
2911 goto out_free_request_sock_slab
;
2913 if (prot
->twsk_prot
!= NULL
) {
2914 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2916 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2917 goto out_free_request_sock_slab
;
2919 prot
->twsk_prot
->twsk_slab
=
2920 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2921 prot
->twsk_prot
->twsk_obj_size
,
2925 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2926 goto out_free_timewait_sock_slab_name
;
2930 mutex_lock(&proto_list_mutex
);
2931 list_add(&prot
->node
, &proto_list
);
2932 assign_proto_idx(prot
);
2933 mutex_unlock(&proto_list_mutex
);
2936 out_free_timewait_sock_slab_name
:
2937 kfree(prot
->twsk_prot
->twsk_slab_name
);
2938 out_free_request_sock_slab
:
2939 req_prot_cleanup(prot
->rsk_prot
);
2941 kmem_cache_destroy(prot
->slab
);
2946 EXPORT_SYMBOL(proto_register
);
2948 void proto_unregister(struct proto
*prot
)
2950 mutex_lock(&proto_list_mutex
);
2951 release_proto_idx(prot
);
2952 list_del(&prot
->node
);
2953 mutex_unlock(&proto_list_mutex
);
2955 kmem_cache_destroy(prot
->slab
);
2958 req_prot_cleanup(prot
->rsk_prot
);
2960 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2961 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2962 kfree(prot
->twsk_prot
->twsk_slab_name
);
2963 prot
->twsk_prot
->twsk_slab
= NULL
;
2966 EXPORT_SYMBOL(proto_unregister
);
2968 #ifdef CONFIG_PROC_FS
2969 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2970 __acquires(proto_list_mutex
)
2972 mutex_lock(&proto_list_mutex
);
2973 return seq_list_start_head(&proto_list
, *pos
);
2976 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2978 return seq_list_next(v
, &proto_list
, pos
);
2981 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2982 __releases(proto_list_mutex
)
2984 mutex_unlock(&proto_list_mutex
);
2987 static char proto_method_implemented(const void *method
)
2989 return method
== NULL
? 'n' : 'y';
2991 static long sock_prot_memory_allocated(struct proto
*proto
)
2993 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2996 static char *sock_prot_memory_pressure(struct proto
*proto
)
2998 return proto
->memory_pressure
!= NULL
?
2999 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
3002 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
3005 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3006 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3009 sock_prot_inuse_get(seq_file_net(seq
), proto
),
3010 sock_prot_memory_allocated(proto
),
3011 sock_prot_memory_pressure(proto
),
3013 proto
->slab
== NULL
? "no" : "yes",
3014 module_name(proto
->owner
),
3015 proto_method_implemented(proto
->close
),
3016 proto_method_implemented(proto
->connect
),
3017 proto_method_implemented(proto
->disconnect
),
3018 proto_method_implemented(proto
->accept
),
3019 proto_method_implemented(proto
->ioctl
),
3020 proto_method_implemented(proto
->init
),
3021 proto_method_implemented(proto
->destroy
),
3022 proto_method_implemented(proto
->shutdown
),
3023 proto_method_implemented(proto
->setsockopt
),
3024 proto_method_implemented(proto
->getsockopt
),
3025 proto_method_implemented(proto
->sendmsg
),
3026 proto_method_implemented(proto
->recvmsg
),
3027 proto_method_implemented(proto
->sendpage
),
3028 proto_method_implemented(proto
->bind
),
3029 proto_method_implemented(proto
->backlog_rcv
),
3030 proto_method_implemented(proto
->hash
),
3031 proto_method_implemented(proto
->unhash
),
3032 proto_method_implemented(proto
->get_port
),
3033 proto_method_implemented(proto
->enter_memory_pressure
));
3036 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3038 if (v
== &proto_list
)
3039 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3048 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3050 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3054 static const struct seq_operations proto_seq_ops
= {
3055 .start
= proto_seq_start
,
3056 .next
= proto_seq_next
,
3057 .stop
= proto_seq_stop
,
3058 .show
= proto_seq_show
,
3061 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
3063 return seq_open_net(inode
, file
, &proto_seq_ops
,
3064 sizeof(struct seq_net_private
));
3067 static const struct file_operations proto_seq_fops
= {
3068 .owner
= THIS_MODULE
,
3069 .open
= proto_seq_open
,
3071 .llseek
= seq_lseek
,
3072 .release
= seq_release_net
,
3075 static __net_init
int proto_init_net(struct net
*net
)
3077 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3083 static __net_exit
void proto_exit_net(struct net
*net
)
3085 remove_proc_entry("protocols", net
->proc_net
);
3089 static __net_initdata
struct pernet_operations proto_net_ops
= {
3090 .init
= proto_init_net
,
3091 .exit
= proto_exit_net
,
3094 static int __init
proto_init(void)
3096 return register_pernet_subsys(&proto_net_ops
);
3099 subsys_initcall(proto_init
);
3101 #endif /* PROC_FS */