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_MAX"
226 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
227 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
228 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
229 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
230 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
231 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
232 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
233 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
234 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
235 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
236 "slock-27" , "slock-28" , "slock-AF_CAN" ,
237 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
238 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
239 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
240 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
242 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
243 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
244 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
245 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
246 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
247 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
248 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
249 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
250 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
251 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
252 "clock-27" , "clock-28" , "clock-AF_CAN" ,
253 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
254 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
255 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
256 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
260 * sk_callback_lock locking rules are per-address-family,
261 * so split the lock classes by using a per-AF key:
263 static struct lock_class_key af_callback_keys
[AF_MAX
];
265 /* Take into consideration the size of the struct sk_buff overhead in the
266 * determination of these values, since that is non-constant across
267 * platforms. This makes socket queueing behavior and performance
268 * not depend upon such differences.
270 #define _SK_MEM_PACKETS 256
271 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
272 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
273 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
275 /* Run time adjustable parameters. */
276 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
277 EXPORT_SYMBOL(sysctl_wmem_max
);
278 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
279 EXPORT_SYMBOL(sysctl_rmem_max
);
280 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
281 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
283 /* Maximal space eaten by iovec or ancillary data plus some space */
284 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
285 EXPORT_SYMBOL(sysctl_optmem_max
);
287 int sysctl_tstamp_allow_data __read_mostly
= 1;
289 struct static_key memalloc_socks
= STATIC_KEY_INIT_FALSE
;
290 EXPORT_SYMBOL_GPL(memalloc_socks
);
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
300 void sk_set_memalloc(struct sock
*sk
)
302 sock_set_flag(sk
, SOCK_MEMALLOC
);
303 sk
->sk_allocation
|= __GFP_MEMALLOC
;
304 static_key_slow_inc(&memalloc_socks
);
306 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
308 void sk_clear_memalloc(struct sock
*sk
)
310 sock_reset_flag(sk
, SOCK_MEMALLOC
);
311 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
312 static_key_slow_dec(&memalloc_socks
);
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
325 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
328 unsigned long pflags
= current
->flags
;
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
333 current
->flags
|= PF_MEMALLOC
;
334 ret
= sk
->sk_backlog_rcv(sk
, skb
);
335 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
339 EXPORT_SYMBOL(__sk_backlog_rcv
);
341 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
)
345 if (optlen
< sizeof(tv
))
347 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
349 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
353 static int warned __read_mostly
;
356 if (warned
< 10 && net_ratelimit()) {
358 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
359 __func__
, current
->comm
, task_pid_nr(current
));
363 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
364 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
366 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/HZ
- 1))
367 *timeo_p
= tv
.tv_sec
*HZ
+ (tv
.tv_usec
+(1000000/HZ
-1))/(1000000/HZ
);
371 static void sock_warn_obsolete_bsdism(const char *name
)
374 static char warncomm
[TASK_COMM_LEN
];
375 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
376 strcpy(warncomm
, current
->comm
);
377 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
383 static bool sock_needs_netstamp(const struct sock
*sk
)
385 switch (sk
->sk_family
) {
394 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
396 if (sk
->sk_flags
& flags
) {
397 sk
->sk_flags
&= ~flags
;
398 if (sock_needs_netstamp(sk
) &&
399 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
400 net_disable_timestamp();
405 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
409 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
411 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
412 atomic_inc(&sk
->sk_drops
);
413 trace_sock_rcvqueue_full(sk
, skb
);
417 err
= sk_filter(sk
, skb
);
421 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
422 atomic_inc(&sk
->sk_drops
);
427 skb_set_owner_r(skb
, sk
);
429 /* we escape from rcu protected region, make sure we dont leak
434 spin_lock_irqsave(&list
->lock
, flags
);
435 sock_skb_set_dropcount(sk
, skb
);
436 __skb_queue_tail(list
, skb
);
437 spin_unlock_irqrestore(&list
->lock
, flags
);
439 if (!sock_flag(sk
, SOCK_DEAD
))
440 sk
->sk_data_ready(sk
);
443 EXPORT_SYMBOL(sock_queue_rcv_skb
);
445 int sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
, const int nested
)
447 int rc
= NET_RX_SUCCESS
;
449 if (sk_filter(sk
, skb
))
450 goto discard_and_relse
;
454 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
455 atomic_inc(&sk
->sk_drops
);
456 goto discard_and_relse
;
459 bh_lock_sock_nested(sk
);
462 if (!sock_owned_by_user(sk
)) {
464 * trylock + unlock semantics:
466 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
468 rc
= sk_backlog_rcv(sk
, skb
);
470 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
471 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
473 atomic_inc(&sk
->sk_drops
);
474 goto discard_and_relse
;
485 EXPORT_SYMBOL(sk_receive_skb
);
487 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
489 struct dst_entry
*dst
= __sk_dst_get(sk
);
491 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
492 sk_tx_queue_clear(sk
);
493 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
500 EXPORT_SYMBOL(__sk_dst_check
);
502 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
504 struct dst_entry
*dst
= sk_dst_get(sk
);
506 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
514 EXPORT_SYMBOL(sk_dst_check
);
516 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
519 int ret
= -ENOPROTOOPT
;
520 #ifdef CONFIG_NETDEVICES
521 struct net
*net
= sock_net(sk
);
522 char devname
[IFNAMSIZ
];
527 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
534 /* Bind this socket to a particular device like "eth0",
535 * as specified in the passed interface name. If the
536 * name is "" or the option length is zero the socket
539 if (optlen
> IFNAMSIZ
- 1)
540 optlen
= IFNAMSIZ
- 1;
541 memset(devname
, 0, sizeof(devname
));
544 if (copy_from_user(devname
, optval
, optlen
))
548 if (devname
[0] != '\0') {
549 struct net_device
*dev
;
552 dev
= dev_get_by_name_rcu(net
, devname
);
554 index
= dev
->ifindex
;
562 sk
->sk_bound_dev_if
= index
;
574 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
575 int __user
*optlen
, int len
)
577 int ret
= -ENOPROTOOPT
;
578 #ifdef CONFIG_NETDEVICES
579 struct net
*net
= sock_net(sk
);
580 char devname
[IFNAMSIZ
];
582 if (sk
->sk_bound_dev_if
== 0) {
591 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
595 len
= strlen(devname
) + 1;
598 if (copy_to_user(optval
, devname
, len
))
603 if (put_user(len
, optlen
))
614 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
617 sock_set_flag(sk
, bit
);
619 sock_reset_flag(sk
, bit
);
622 bool sk_mc_loop(struct sock
*sk
)
624 if (dev_recursion_level())
628 switch (sk
->sk_family
) {
630 return inet_sk(sk
)->mc_loop
;
631 #if IS_ENABLED(CONFIG_IPV6)
633 return inet6_sk(sk
)->mc_loop
;
639 EXPORT_SYMBOL(sk_mc_loop
);
642 * This is meant for all protocols to use and covers goings on
643 * at the socket level. Everything here is generic.
646 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
647 char __user
*optval
, unsigned int optlen
)
649 struct sock
*sk
= sock
->sk
;
656 * Options without arguments
659 if (optname
== SO_BINDTODEVICE
)
660 return sock_setbindtodevice(sk
, optval
, optlen
);
662 if (optlen
< sizeof(int))
665 if (get_user(val
, (int __user
*)optval
))
668 valbool
= val
? 1 : 0;
674 if (val
&& !capable(CAP_NET_ADMIN
))
677 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
680 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
683 sk
->sk_reuseport
= valbool
;
692 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
695 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
698 /* Don't error on this BSD doesn't and if you think
699 * about it this is right. Otherwise apps have to
700 * play 'guess the biggest size' games. RCVBUF/SNDBUF
701 * are treated in BSD as hints
703 val
= min_t(u32
, val
, sysctl_wmem_max
);
705 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
706 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
707 /* Wake up sending tasks if we upped the value. */
708 sk
->sk_write_space(sk
);
712 if (!capable(CAP_NET_ADMIN
)) {
719 /* Don't error on this BSD doesn't and if you think
720 * about it this is right. Otherwise apps have to
721 * play 'guess the biggest size' games. RCVBUF/SNDBUF
722 * are treated in BSD as hints
724 val
= min_t(u32
, val
, sysctl_rmem_max
);
726 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
728 * We double it on the way in to account for
729 * "struct sk_buff" etc. overhead. Applications
730 * assume that the SO_RCVBUF setting they make will
731 * allow that much actual data to be received on that
734 * Applications are unaware that "struct sk_buff" and
735 * other overheads allocate from the receive buffer
736 * during socket buffer allocation.
738 * And after considering the possible alternatives,
739 * returning the value we actually used in getsockopt
740 * is the most desirable behavior.
742 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
746 if (!capable(CAP_NET_ADMIN
)) {
754 if (sk
->sk_protocol
== IPPROTO_TCP
&&
755 sk
->sk_type
== SOCK_STREAM
)
756 tcp_set_keepalive(sk
, valbool
);
758 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
762 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
766 sk
->sk_no_check_tx
= valbool
;
770 if ((val
>= 0 && val
<= 6) ||
771 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
772 sk
->sk_priority
= val
;
778 if (optlen
< sizeof(ling
)) {
779 ret
= -EINVAL
; /* 1003.1g */
782 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
787 sock_reset_flag(sk
, SOCK_LINGER
);
789 #if (BITS_PER_LONG == 32)
790 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
791 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
794 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
795 sock_set_flag(sk
, SOCK_LINGER
);
800 sock_warn_obsolete_bsdism("setsockopt");
805 set_bit(SOCK_PASSCRED
, &sock
->flags
);
807 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
813 if (optname
== SO_TIMESTAMP
)
814 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
816 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
817 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
818 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
820 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
821 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
825 case SO_TIMESTAMPING
:
826 if (val
& ~SOF_TIMESTAMPING_MASK
) {
831 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
832 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
833 if (sk
->sk_protocol
== IPPROTO_TCP
&&
834 sk
->sk_type
== SOCK_STREAM
) {
835 if (sk
->sk_state
!= TCP_ESTABLISHED
) {
839 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
844 sk
->sk_tsflags
= val
;
845 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
846 sock_enable_timestamp(sk
,
847 SOCK_TIMESTAMPING_RX_SOFTWARE
);
849 sock_disable_timestamp(sk
,
850 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
856 sk
->sk_rcvlowat
= val
? : 1;
860 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
864 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
867 case SO_ATTACH_FILTER
:
869 if (optlen
== sizeof(struct sock_fprog
)) {
870 struct sock_fprog fprog
;
873 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
876 ret
= sk_attach_filter(&fprog
, sk
);
882 if (optlen
== sizeof(u32
)) {
886 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
889 ret
= sk_attach_bpf(ufd
, sk
);
893 case SO_ATTACH_REUSEPORT_CBPF
:
895 if (optlen
== sizeof(struct sock_fprog
)) {
896 struct sock_fprog fprog
;
899 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
902 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
906 case SO_ATTACH_REUSEPORT_EBPF
:
908 if (optlen
== sizeof(u32
)) {
912 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
915 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
919 case SO_DETACH_FILTER
:
920 ret
= sk_detach_filter(sk
);
924 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
927 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
932 set_bit(SOCK_PASSSEC
, &sock
->flags
);
934 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
937 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
944 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
948 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
952 if (sock
->ops
->set_peek_off
)
953 ret
= sock
->ops
->set_peek_off(sk
, val
);
959 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
962 case SO_SELECT_ERR_QUEUE
:
963 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
966 #ifdef CONFIG_NET_RX_BUSY_POLL
968 /* allow unprivileged users to decrease the value */
969 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
975 sk
->sk_ll_usec
= val
;
980 case SO_MAX_PACING_RATE
:
981 sk
->sk_max_pacing_rate
= val
;
982 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
983 sk
->sk_max_pacing_rate
);
986 case SO_INCOMING_CPU
:
987 sk
->sk_incoming_cpu
= val
;
997 EXPORT_SYMBOL(sock_setsockopt
);
1000 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1001 struct ucred
*ucred
)
1003 ucred
->pid
= pid_vnr(pid
);
1004 ucred
->uid
= ucred
->gid
= -1;
1006 struct user_namespace
*current_ns
= current_user_ns();
1008 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1009 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1013 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1014 char __user
*optval
, int __user
*optlen
)
1016 struct sock
*sk
= sock
->sk
;
1024 int lv
= sizeof(int);
1027 if (get_user(len
, optlen
))
1032 memset(&v
, 0, sizeof(v
));
1036 v
.val
= sock_flag(sk
, SOCK_DBG
);
1040 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1044 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1048 v
.val
= sk
->sk_sndbuf
;
1052 v
.val
= sk
->sk_rcvbuf
;
1056 v
.val
= sk
->sk_reuse
;
1060 v
.val
= sk
->sk_reuseport
;
1064 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1068 v
.val
= sk
->sk_type
;
1072 v
.val
= sk
->sk_protocol
;
1076 v
.val
= sk
->sk_family
;
1080 v
.val
= -sock_error(sk
);
1082 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1086 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1090 v
.val
= sk
->sk_no_check_tx
;
1094 v
.val
= sk
->sk_priority
;
1098 lv
= sizeof(v
.ling
);
1099 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1100 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1104 sock_warn_obsolete_bsdism("getsockopt");
1108 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1109 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1112 case SO_TIMESTAMPNS
:
1113 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1116 case SO_TIMESTAMPING
:
1117 v
.val
= sk
->sk_tsflags
;
1121 lv
= sizeof(struct timeval
);
1122 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1126 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1127 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1132 lv
= sizeof(struct timeval
);
1133 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1137 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1138 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1143 v
.val
= sk
->sk_rcvlowat
;
1151 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1156 struct ucred peercred
;
1157 if (len
> sizeof(peercred
))
1158 len
= sizeof(peercred
);
1159 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1160 if (copy_to_user(optval
, &peercred
, len
))
1169 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1173 if (copy_to_user(optval
, address
, len
))
1178 /* Dubious BSD thing... Probably nobody even uses it, but
1179 * the UNIX standard wants it for whatever reason... -DaveM
1182 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1186 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1190 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1193 v
.val
= sk
->sk_mark
;
1197 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1200 case SO_WIFI_STATUS
:
1201 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1205 if (!sock
->ops
->set_peek_off
)
1208 v
.val
= sk
->sk_peek_off
;
1211 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1214 case SO_BINDTODEVICE
:
1215 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1218 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1224 case SO_LOCK_FILTER
:
1225 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1228 case SO_BPF_EXTENSIONS
:
1229 v
.val
= bpf_tell_extensions();
1232 case SO_SELECT_ERR_QUEUE
:
1233 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1236 #ifdef CONFIG_NET_RX_BUSY_POLL
1238 v
.val
= sk
->sk_ll_usec
;
1242 case SO_MAX_PACING_RATE
:
1243 v
.val
= sk
->sk_max_pacing_rate
;
1246 case SO_INCOMING_CPU
:
1247 v
.val
= sk
->sk_incoming_cpu
;
1251 /* We implement the SO_SNDLOWAT etc to not be settable
1254 return -ENOPROTOOPT
;
1259 if (copy_to_user(optval
, &v
, len
))
1262 if (put_user(len
, optlen
))
1268 * Initialize an sk_lock.
1270 * (We also register the sk_lock with the lock validator.)
1272 static inline void sock_lock_init(struct sock
*sk
)
1274 sock_lock_init_class_and_name(sk
,
1275 af_family_slock_key_strings
[sk
->sk_family
],
1276 af_family_slock_keys
+ sk
->sk_family
,
1277 af_family_key_strings
[sk
->sk_family
],
1278 af_family_keys
+ sk
->sk_family
);
1282 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1283 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1284 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1286 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1288 #ifdef CONFIG_SECURITY_NETWORK
1289 void *sptr
= nsk
->sk_security
;
1291 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1293 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1294 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1296 #ifdef CONFIG_SECURITY_NETWORK
1297 nsk
->sk_security
= sptr
;
1298 security_sk_clone(osk
, nsk
);
1302 void sk_prot_clear_portaddr_nulls(struct sock
*sk
, int size
)
1304 unsigned long nulls1
, nulls2
;
1306 nulls1
= offsetof(struct sock
, __sk_common
.skc_node
.next
);
1307 nulls2
= offsetof(struct sock
, __sk_common
.skc_portaddr_node
.next
);
1308 if (nulls1
> nulls2
)
1309 swap(nulls1
, nulls2
);
1312 memset((char *)sk
, 0, nulls1
);
1313 memset((char *)sk
+ nulls1
+ sizeof(void *), 0,
1314 nulls2
- nulls1
- sizeof(void *));
1315 memset((char *)sk
+ nulls2
+ sizeof(void *), 0,
1316 size
- nulls2
- sizeof(void *));
1318 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls
);
1320 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1324 struct kmem_cache
*slab
;
1328 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1331 if (priority
& __GFP_ZERO
) {
1333 prot
->clear_sk(sk
, prot
->obj_size
);
1335 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1338 sk
= kmalloc(prot
->obj_size
, priority
);
1341 kmemcheck_annotate_bitfield(sk
, flags
);
1343 if (security_sk_alloc(sk
, family
, priority
))
1346 if (!try_module_get(prot
->owner
))
1348 sk_tx_queue_clear(sk
);
1349 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1355 security_sk_free(sk
);
1358 kmem_cache_free(slab
, sk
);
1364 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1366 struct kmem_cache
*slab
;
1367 struct module
*owner
;
1369 owner
= prot
->owner
;
1372 cgroup_sk_free(&sk
->sk_cgrp_data
);
1373 security_sk_free(sk
);
1375 kmem_cache_free(slab
, sk
);
1382 * sk_alloc - All socket objects are allocated here
1383 * @net: the applicable net namespace
1384 * @family: protocol family
1385 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1386 * @prot: struct proto associated with this new sock instance
1387 * @kern: is this to be a kernel socket?
1389 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1390 struct proto
*prot
, int kern
)
1394 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1396 sk
->sk_family
= family
;
1398 * See comment in struct sock definition to understand
1399 * why we need sk_prot_creator -acme
1401 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1403 sk
->sk_net_refcnt
= kern
? 0 : 1;
1404 if (likely(sk
->sk_net_refcnt
))
1406 sock_net_set(sk
, net
);
1407 atomic_set(&sk
->sk_wmem_alloc
, 1);
1409 sock_update_classid(&sk
->sk_cgrp_data
);
1410 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1415 EXPORT_SYMBOL(sk_alloc
);
1417 void sk_destruct(struct sock
*sk
)
1419 struct sk_filter
*filter
;
1421 if (sk
->sk_destruct
)
1422 sk
->sk_destruct(sk
);
1424 filter
= rcu_dereference_check(sk
->sk_filter
,
1425 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1427 sk_filter_uncharge(sk
, filter
);
1428 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1430 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1431 reuseport_detach_sock(sk
);
1433 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1435 if (atomic_read(&sk
->sk_omem_alloc
))
1436 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1437 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1439 if (sk
->sk_peer_cred
)
1440 put_cred(sk
->sk_peer_cred
);
1441 put_pid(sk
->sk_peer_pid
);
1442 if (likely(sk
->sk_net_refcnt
))
1443 put_net(sock_net(sk
));
1444 sk_prot_free(sk
->sk_prot_creator
, sk
);
1447 static void __sk_free(struct sock
*sk
)
1449 if (unlikely(sock_diag_has_destroy_listeners(sk
) && sk
->sk_net_refcnt
))
1450 sock_diag_broadcast_destroy(sk
);
1455 void sk_free(struct sock
*sk
)
1458 * We subtract one from sk_wmem_alloc and can know if
1459 * some packets are still in some tx queue.
1460 * If not null, sock_wfree() will call __sk_free(sk) later
1462 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1465 EXPORT_SYMBOL(sk_free
);
1468 * sk_clone_lock - clone a socket, and lock its clone
1469 * @sk: the socket to clone
1470 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1472 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1474 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1477 bool is_charged
= true;
1479 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1480 if (newsk
!= NULL
) {
1481 struct sk_filter
*filter
;
1483 sock_copy(newsk
, sk
);
1486 if (likely(newsk
->sk_net_refcnt
))
1487 get_net(sock_net(newsk
));
1488 sk_node_init(&newsk
->sk_node
);
1489 sock_lock_init(newsk
);
1490 bh_lock_sock(newsk
);
1491 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1492 newsk
->sk_backlog
.len
= 0;
1494 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1496 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1498 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1499 atomic_set(&newsk
->sk_omem_alloc
, 0);
1500 skb_queue_head_init(&newsk
->sk_receive_queue
);
1501 skb_queue_head_init(&newsk
->sk_write_queue
);
1503 rwlock_init(&newsk
->sk_callback_lock
);
1504 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1505 af_callback_keys
+ newsk
->sk_family
,
1506 af_family_clock_key_strings
[newsk
->sk_family
]);
1508 newsk
->sk_dst_cache
= NULL
;
1509 newsk
->sk_wmem_queued
= 0;
1510 newsk
->sk_forward_alloc
= 0;
1511 newsk
->sk_send_head
= NULL
;
1512 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1514 sock_reset_flag(newsk
, SOCK_DONE
);
1515 skb_queue_head_init(&newsk
->sk_error_queue
);
1517 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1519 /* though it's an empty new sock, the charging may fail
1520 * if sysctl_optmem_max was changed between creation of
1521 * original socket and cloning
1523 is_charged
= sk_filter_charge(newsk
, filter
);
1525 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1526 /* It is still raw copy of parent, so invalidate
1527 * destructor and make plain sk_free() */
1528 newsk
->sk_destruct
= NULL
;
1529 bh_unlock_sock(newsk
);
1536 newsk
->sk_priority
= 0;
1537 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1538 atomic64_set(&newsk
->sk_cookie
, 0);
1540 * Before updating sk_refcnt, we must commit prior changes to memory
1541 * (Documentation/RCU/rculist_nulls.txt for details)
1544 atomic_set(&newsk
->sk_refcnt
, 2);
1547 * Increment the counter in the same struct proto as the master
1548 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1549 * is the same as sk->sk_prot->socks, as this field was copied
1552 * This _changes_ the previous behaviour, where
1553 * tcp_create_openreq_child always was incrementing the
1554 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1555 * to be taken into account in all callers. -acme
1557 sk_refcnt_debug_inc(newsk
);
1558 sk_set_socket(newsk
, NULL
);
1559 newsk
->sk_wq
= NULL
;
1561 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
1562 sock_update_memcg(newsk
);
1564 if (newsk
->sk_prot
->sockets_allocated
)
1565 sk_sockets_allocated_inc(newsk
);
1567 if (sock_needs_netstamp(sk
) &&
1568 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1569 net_enable_timestamp();
1574 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1576 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1580 sk_dst_set(sk
, dst
);
1581 sk
->sk_route_caps
= dst
->dev
->features
;
1582 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1583 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1584 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1585 if (sk_can_gso(sk
)) {
1586 if (dst
->header_len
) {
1587 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1589 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1590 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1591 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1594 sk
->sk_gso_max_segs
= max_segs
;
1596 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1599 * Simple resource managers for sockets.
1604 * Write buffer destructor automatically called from kfree_skb.
1606 void sock_wfree(struct sk_buff
*skb
)
1608 struct sock
*sk
= skb
->sk
;
1609 unsigned int len
= skb
->truesize
;
1611 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1613 * Keep a reference on sk_wmem_alloc, this will be released
1614 * after sk_write_space() call
1616 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1617 sk
->sk_write_space(sk
);
1621 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1622 * could not do because of in-flight packets
1624 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1627 EXPORT_SYMBOL(sock_wfree
);
1629 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1634 if (unlikely(!sk_fullsock(sk
))) {
1635 skb
->destructor
= sock_edemux
;
1640 skb
->destructor
= sock_wfree
;
1641 skb_set_hash_from_sk(skb
, sk
);
1643 * We used to take a refcount on sk, but following operation
1644 * is enough to guarantee sk_free() wont free this sock until
1645 * all in-flight packets are completed
1647 atomic_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1649 EXPORT_SYMBOL(skb_set_owner_w
);
1651 void skb_orphan_partial(struct sk_buff
*skb
)
1653 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1654 * so we do not completely orphan skb, but transfert all
1655 * accounted bytes but one, to avoid unexpected reorders.
1657 if (skb
->destructor
== sock_wfree
1659 || skb
->destructor
== tcp_wfree
1662 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1668 EXPORT_SYMBOL(skb_orphan_partial
);
1671 * Read buffer destructor automatically called from kfree_skb.
1673 void sock_rfree(struct sk_buff
*skb
)
1675 struct sock
*sk
= skb
->sk
;
1676 unsigned int len
= skb
->truesize
;
1678 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1679 sk_mem_uncharge(sk
, len
);
1681 EXPORT_SYMBOL(sock_rfree
);
1684 * Buffer destructor for skbs that are not used directly in read or write
1685 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1687 void sock_efree(struct sk_buff
*skb
)
1691 EXPORT_SYMBOL(sock_efree
);
1693 kuid_t
sock_i_uid(struct sock
*sk
)
1697 read_lock_bh(&sk
->sk_callback_lock
);
1698 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1699 read_unlock_bh(&sk
->sk_callback_lock
);
1702 EXPORT_SYMBOL(sock_i_uid
);
1704 unsigned long sock_i_ino(struct sock
*sk
)
1708 read_lock_bh(&sk
->sk_callback_lock
);
1709 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1710 read_unlock_bh(&sk
->sk_callback_lock
);
1713 EXPORT_SYMBOL(sock_i_ino
);
1716 * Allocate a skb from the socket's send buffer.
1718 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1721 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1722 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1724 skb_set_owner_w(skb
, sk
);
1730 EXPORT_SYMBOL(sock_wmalloc
);
1733 * Allocate a memory block from the socket's option memory buffer.
1735 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1737 if ((unsigned int)size
<= sysctl_optmem_max
&&
1738 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1740 /* First do the add, to avoid the race if kmalloc
1743 atomic_add(size
, &sk
->sk_omem_alloc
);
1744 mem
= kmalloc(size
, priority
);
1747 atomic_sub(size
, &sk
->sk_omem_alloc
);
1751 EXPORT_SYMBOL(sock_kmalloc
);
1753 /* Free an option memory block. Note, we actually want the inline
1754 * here as this allows gcc to detect the nullify and fold away the
1755 * condition entirely.
1757 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1760 if (WARN_ON_ONCE(!mem
))
1766 atomic_sub(size
, &sk
->sk_omem_alloc
);
1769 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1771 __sock_kfree_s(sk
, mem
, size
, false);
1773 EXPORT_SYMBOL(sock_kfree_s
);
1775 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1777 __sock_kfree_s(sk
, mem
, size
, true);
1779 EXPORT_SYMBOL(sock_kzfree_s
);
1781 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1782 I think, these locks should be removed for datagram sockets.
1784 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1788 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1792 if (signal_pending(current
))
1794 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1795 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1796 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1798 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1802 timeo
= schedule_timeout(timeo
);
1804 finish_wait(sk_sleep(sk
), &wait
);
1810 * Generic send/receive buffer handlers
1813 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1814 unsigned long data_len
, int noblock
,
1815 int *errcode
, int max_page_order
)
1817 struct sk_buff
*skb
;
1821 timeo
= sock_sndtimeo(sk
, noblock
);
1823 err
= sock_error(sk
);
1828 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1831 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1834 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1835 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1839 if (signal_pending(current
))
1841 timeo
= sock_wait_for_wmem(sk
, timeo
);
1843 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
1844 errcode
, sk
->sk_allocation
);
1846 skb_set_owner_w(skb
, sk
);
1850 err
= sock_intr_errno(timeo
);
1855 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1857 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1858 int noblock
, int *errcode
)
1860 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1862 EXPORT_SYMBOL(sock_alloc_send_skb
);
1864 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
1865 struct sockcm_cookie
*sockc
)
1867 struct cmsghdr
*cmsg
;
1869 for_each_cmsghdr(cmsg
, msg
) {
1870 if (!CMSG_OK(msg
, cmsg
))
1872 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
1874 switch (cmsg
->cmsg_type
) {
1876 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
1878 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1880 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
1888 EXPORT_SYMBOL(sock_cmsg_send
);
1890 /* On 32bit arches, an skb frag is limited to 2^15 */
1891 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1894 * skb_page_frag_refill - check that a page_frag contains enough room
1895 * @sz: minimum size of the fragment we want to get
1896 * @pfrag: pointer to page_frag
1897 * @gfp: priority for memory allocation
1899 * Note: While this allocator tries to use high order pages, there is
1900 * no guarantee that allocations succeed. Therefore, @sz MUST be
1901 * less or equal than PAGE_SIZE.
1903 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
1906 if (atomic_read(&pfrag
->page
->_count
) == 1) {
1910 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1912 put_page(pfrag
->page
);
1916 if (SKB_FRAG_PAGE_ORDER
) {
1917 /* Avoid direct reclaim but allow kswapd to wake */
1918 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
1919 __GFP_COMP
| __GFP_NOWARN
|
1921 SKB_FRAG_PAGE_ORDER
);
1922 if (likely(pfrag
->page
)) {
1923 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
1927 pfrag
->page
= alloc_page(gfp
);
1928 if (likely(pfrag
->page
)) {
1929 pfrag
->size
= PAGE_SIZE
;
1934 EXPORT_SYMBOL(skb_page_frag_refill
);
1936 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1938 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
1941 sk_enter_memory_pressure(sk
);
1942 sk_stream_moderate_sndbuf(sk
);
1945 EXPORT_SYMBOL(sk_page_frag_refill
);
1947 static void __lock_sock(struct sock
*sk
)
1948 __releases(&sk
->sk_lock
.slock
)
1949 __acquires(&sk
->sk_lock
.slock
)
1954 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
1955 TASK_UNINTERRUPTIBLE
);
1956 spin_unlock_bh(&sk
->sk_lock
.slock
);
1958 spin_lock_bh(&sk
->sk_lock
.slock
);
1959 if (!sock_owned_by_user(sk
))
1962 finish_wait(&sk
->sk_lock
.wq
, &wait
);
1965 static void __release_sock(struct sock
*sk
)
1966 __releases(&sk
->sk_lock
.slock
)
1967 __acquires(&sk
->sk_lock
.slock
)
1969 struct sk_buff
*skb
= sk
->sk_backlog
.head
;
1972 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
1976 struct sk_buff
*next
= skb
->next
;
1979 WARN_ON_ONCE(skb_dst_is_noref(skb
));
1981 sk_backlog_rcv(sk
, skb
);
1984 * We are in process context here with softirqs
1985 * disabled, use cond_resched_softirq() to preempt.
1986 * This is safe to do because we've taken the backlog
1989 cond_resched_softirq();
1992 } while (skb
!= NULL
);
1995 } while ((skb
= sk
->sk_backlog
.head
) != NULL
);
1998 * Doing the zeroing here guarantee we can not loop forever
1999 * while a wild producer attempts to flood us.
2001 sk
->sk_backlog
.len
= 0;
2005 * sk_wait_data - wait for data to arrive at sk_receive_queue
2006 * @sk: sock to wait on
2007 * @timeo: for how long
2008 * @skb: last skb seen on sk_receive_queue
2010 * Now socket state including sk->sk_err is changed only under lock,
2011 * hence we may omit checks after joining wait queue.
2012 * We check receive queue before schedule() only as optimization;
2013 * it is very likely that release_sock() added new data.
2015 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2020 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2021 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2022 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
);
2023 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2024 finish_wait(sk_sleep(sk
), &wait
);
2027 EXPORT_SYMBOL(sk_wait_data
);
2030 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2032 * @size: memory size to allocate
2033 * @kind: allocation type
2035 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2036 * rmem allocation. This function assumes that protocols which have
2037 * memory_pressure use sk_wmem_queued as write buffer accounting.
2039 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2041 struct proto
*prot
= sk
->sk_prot
;
2042 int amt
= sk_mem_pages(size
);
2045 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2047 allocated
= sk_memory_allocated_add(sk
, amt
);
2049 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2050 !mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
))
2051 goto suppress_allocation
;
2054 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2055 sk_leave_memory_pressure(sk
);
2059 /* Under pressure. */
2060 if (allocated
> sk_prot_mem_limits(sk
, 1))
2061 sk_enter_memory_pressure(sk
);
2063 /* Over hard limit. */
2064 if (allocated
> sk_prot_mem_limits(sk
, 2))
2065 goto suppress_allocation
;
2067 /* guarantee minimum buffer size under pressure */
2068 if (kind
== SK_MEM_RECV
) {
2069 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2072 } else { /* SK_MEM_SEND */
2073 if (sk
->sk_type
== SOCK_STREAM
) {
2074 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2076 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2077 prot
->sysctl_wmem
[0])
2081 if (sk_has_memory_pressure(sk
)) {
2084 if (!sk_under_memory_pressure(sk
))
2086 alloc
= sk_sockets_allocated_read_positive(sk
);
2087 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2088 sk_mem_pages(sk
->sk_wmem_queued
+
2089 atomic_read(&sk
->sk_rmem_alloc
) +
2090 sk
->sk_forward_alloc
))
2094 suppress_allocation
:
2096 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2097 sk_stream_moderate_sndbuf(sk
);
2099 /* Fail only if socket is _under_ its sndbuf.
2100 * In this case we cannot block, so that we have to fail.
2102 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2106 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2108 /* Alas. Undo changes. */
2109 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2111 sk_memory_allocated_sub(sk
, amt
);
2113 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2114 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2118 EXPORT_SYMBOL(__sk_mem_schedule
);
2121 * __sk_mem_reclaim - reclaim memory_allocated
2123 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2125 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2127 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2128 sk_memory_allocated_sub(sk
, amount
);
2129 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2131 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2132 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2134 if (sk_under_memory_pressure(sk
) &&
2135 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2136 sk_leave_memory_pressure(sk
);
2138 EXPORT_SYMBOL(__sk_mem_reclaim
);
2142 * Set of default routines for initialising struct proto_ops when
2143 * the protocol does not support a particular function. In certain
2144 * cases where it makes no sense for a protocol to have a "do nothing"
2145 * function, some default processing is provided.
2148 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2152 EXPORT_SYMBOL(sock_no_bind
);
2154 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2159 EXPORT_SYMBOL(sock_no_connect
);
2161 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2165 EXPORT_SYMBOL(sock_no_socketpair
);
2167 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2171 EXPORT_SYMBOL(sock_no_accept
);
2173 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2178 EXPORT_SYMBOL(sock_no_getname
);
2180 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2184 EXPORT_SYMBOL(sock_no_poll
);
2186 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2190 EXPORT_SYMBOL(sock_no_ioctl
);
2192 int sock_no_listen(struct socket
*sock
, int backlog
)
2196 EXPORT_SYMBOL(sock_no_listen
);
2198 int sock_no_shutdown(struct socket
*sock
, int how
)
2202 EXPORT_SYMBOL(sock_no_shutdown
);
2204 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2205 char __user
*optval
, unsigned int optlen
)
2209 EXPORT_SYMBOL(sock_no_setsockopt
);
2211 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2212 char __user
*optval
, int __user
*optlen
)
2216 EXPORT_SYMBOL(sock_no_getsockopt
);
2218 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2222 EXPORT_SYMBOL(sock_no_sendmsg
);
2224 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2229 EXPORT_SYMBOL(sock_no_recvmsg
);
2231 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2233 /* Mirror missing mmap method error code */
2236 EXPORT_SYMBOL(sock_no_mmap
);
2238 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2241 struct msghdr msg
= {.msg_flags
= flags
};
2243 char *kaddr
= kmap(page
);
2244 iov
.iov_base
= kaddr
+ offset
;
2246 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2250 EXPORT_SYMBOL(sock_no_sendpage
);
2253 * Default Socket Callbacks
2256 static void sock_def_wakeup(struct sock
*sk
)
2258 struct socket_wq
*wq
;
2261 wq
= rcu_dereference(sk
->sk_wq
);
2262 if (skwq_has_sleeper(wq
))
2263 wake_up_interruptible_all(&wq
->wait
);
2267 static void sock_def_error_report(struct sock
*sk
)
2269 struct socket_wq
*wq
;
2272 wq
= rcu_dereference(sk
->sk_wq
);
2273 if (skwq_has_sleeper(wq
))
2274 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2275 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2279 static void sock_def_readable(struct sock
*sk
)
2281 struct socket_wq
*wq
;
2284 wq
= rcu_dereference(sk
->sk_wq
);
2285 if (skwq_has_sleeper(wq
))
2286 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2287 POLLRDNORM
| POLLRDBAND
);
2288 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2292 static void sock_def_write_space(struct sock
*sk
)
2294 struct socket_wq
*wq
;
2298 /* Do not wake up a writer until he can make "significant"
2301 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2302 wq
= rcu_dereference(sk
->sk_wq
);
2303 if (skwq_has_sleeper(wq
))
2304 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2305 POLLWRNORM
| POLLWRBAND
);
2307 /* Should agree with poll, otherwise some programs break */
2308 if (sock_writeable(sk
))
2309 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2315 static void sock_def_destruct(struct sock
*sk
)
2319 void sk_send_sigurg(struct sock
*sk
)
2321 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2322 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2323 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2325 EXPORT_SYMBOL(sk_send_sigurg
);
2327 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2328 unsigned long expires
)
2330 if (!mod_timer(timer
, expires
))
2333 EXPORT_SYMBOL(sk_reset_timer
);
2335 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2337 if (del_timer(timer
))
2340 EXPORT_SYMBOL(sk_stop_timer
);
2342 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2344 skb_queue_head_init(&sk
->sk_receive_queue
);
2345 skb_queue_head_init(&sk
->sk_write_queue
);
2346 skb_queue_head_init(&sk
->sk_error_queue
);
2348 sk
->sk_send_head
= NULL
;
2350 init_timer(&sk
->sk_timer
);
2352 sk
->sk_allocation
= GFP_KERNEL
;
2353 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2354 sk
->sk_sndbuf
= sysctl_wmem_default
;
2355 sk
->sk_state
= TCP_CLOSE
;
2356 sk_set_socket(sk
, sock
);
2358 sock_set_flag(sk
, SOCK_ZAPPED
);
2361 sk
->sk_type
= sock
->type
;
2362 sk
->sk_wq
= sock
->wq
;
2367 rwlock_init(&sk
->sk_callback_lock
);
2368 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2369 af_callback_keys
+ sk
->sk_family
,
2370 af_family_clock_key_strings
[sk
->sk_family
]);
2372 sk
->sk_state_change
= sock_def_wakeup
;
2373 sk
->sk_data_ready
= sock_def_readable
;
2374 sk
->sk_write_space
= sock_def_write_space
;
2375 sk
->sk_error_report
= sock_def_error_report
;
2376 sk
->sk_destruct
= sock_def_destruct
;
2378 sk
->sk_frag
.page
= NULL
;
2379 sk
->sk_frag
.offset
= 0;
2380 sk
->sk_peek_off
= -1;
2382 sk
->sk_peer_pid
= NULL
;
2383 sk
->sk_peer_cred
= NULL
;
2384 sk
->sk_write_pending
= 0;
2385 sk
->sk_rcvlowat
= 1;
2386 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2387 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2389 sk
->sk_stamp
= ktime_set(-1L, 0);
2391 #ifdef CONFIG_NET_RX_BUSY_POLL
2393 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2396 sk
->sk_max_pacing_rate
= ~0U;
2397 sk
->sk_pacing_rate
= ~0U;
2398 sk
->sk_incoming_cpu
= -1;
2400 * Before updating sk_refcnt, we must commit prior changes to memory
2401 * (Documentation/RCU/rculist_nulls.txt for details)
2404 atomic_set(&sk
->sk_refcnt
, 1);
2405 atomic_set(&sk
->sk_drops
, 0);
2407 EXPORT_SYMBOL(sock_init_data
);
2409 void lock_sock_nested(struct sock
*sk
, int subclass
)
2412 spin_lock_bh(&sk
->sk_lock
.slock
);
2413 if (sk
->sk_lock
.owned
)
2415 sk
->sk_lock
.owned
= 1;
2416 spin_unlock(&sk
->sk_lock
.slock
);
2418 * The sk_lock has mutex_lock() semantics here:
2420 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2423 EXPORT_SYMBOL(lock_sock_nested
);
2425 void release_sock(struct sock
*sk
)
2428 * The sk_lock has mutex_unlock() semantics:
2430 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
2432 spin_lock_bh(&sk
->sk_lock
.slock
);
2433 if (sk
->sk_backlog
.tail
)
2436 /* Warning : release_cb() might need to release sk ownership,
2437 * ie call sock_release_ownership(sk) before us.
2439 if (sk
->sk_prot
->release_cb
)
2440 sk
->sk_prot
->release_cb(sk
);
2442 sock_release_ownership(sk
);
2443 if (waitqueue_active(&sk
->sk_lock
.wq
))
2444 wake_up(&sk
->sk_lock
.wq
);
2445 spin_unlock_bh(&sk
->sk_lock
.slock
);
2447 EXPORT_SYMBOL(release_sock
);
2450 * lock_sock_fast - fast version of lock_sock
2453 * This version should be used for very small section, where process wont block
2454 * return false if fast path is taken
2455 * sk_lock.slock locked, owned = 0, BH disabled
2456 * return true if slow path is taken
2457 * sk_lock.slock unlocked, owned = 1, BH enabled
2459 bool lock_sock_fast(struct sock
*sk
)
2462 spin_lock_bh(&sk
->sk_lock
.slock
);
2464 if (!sk
->sk_lock
.owned
)
2466 * Note : We must disable BH
2471 sk
->sk_lock
.owned
= 1;
2472 spin_unlock(&sk
->sk_lock
.slock
);
2474 * The sk_lock has mutex_lock() semantics here:
2476 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2480 EXPORT_SYMBOL(lock_sock_fast
);
2482 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2485 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2486 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2487 tv
= ktime_to_timeval(sk
->sk_stamp
);
2488 if (tv
.tv_sec
== -1)
2490 if (tv
.tv_sec
== 0) {
2491 sk
->sk_stamp
= ktime_get_real();
2492 tv
= ktime_to_timeval(sk
->sk_stamp
);
2494 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2496 EXPORT_SYMBOL(sock_get_timestamp
);
2498 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2501 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2502 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2503 ts
= ktime_to_timespec(sk
->sk_stamp
);
2504 if (ts
.tv_sec
== -1)
2506 if (ts
.tv_sec
== 0) {
2507 sk
->sk_stamp
= ktime_get_real();
2508 ts
= ktime_to_timespec(sk
->sk_stamp
);
2510 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2512 EXPORT_SYMBOL(sock_get_timestampns
);
2514 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2516 if (!sock_flag(sk
, flag
)) {
2517 unsigned long previous_flags
= sk
->sk_flags
;
2519 sock_set_flag(sk
, flag
);
2521 * we just set one of the two flags which require net
2522 * time stamping, but time stamping might have been on
2523 * already because of the other one
2525 if (sock_needs_netstamp(sk
) &&
2526 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
2527 net_enable_timestamp();
2531 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2532 int level
, int type
)
2534 struct sock_exterr_skb
*serr
;
2535 struct sk_buff
*skb
;
2539 skb
= sock_dequeue_err_skb(sk
);
2545 msg
->msg_flags
|= MSG_TRUNC
;
2548 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2552 sock_recv_timestamp(msg
, sk
, skb
);
2554 serr
= SKB_EXT_ERR(skb
);
2555 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2557 msg
->msg_flags
|= MSG_ERRQUEUE
;
2565 EXPORT_SYMBOL(sock_recv_errqueue
);
2568 * Get a socket option on an socket.
2570 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2571 * asynchronous errors should be reported by getsockopt. We assume
2572 * this means if you specify SO_ERROR (otherwise whats the point of it).
2574 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2575 char __user
*optval
, int __user
*optlen
)
2577 struct sock
*sk
= sock
->sk
;
2579 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2581 EXPORT_SYMBOL(sock_common_getsockopt
);
2583 #ifdef CONFIG_COMPAT
2584 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2585 char __user
*optval
, int __user
*optlen
)
2587 struct sock
*sk
= sock
->sk
;
2589 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2590 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2592 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2594 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2597 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2600 struct sock
*sk
= sock
->sk
;
2604 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2605 flags
& ~MSG_DONTWAIT
, &addr_len
);
2607 msg
->msg_namelen
= addr_len
;
2610 EXPORT_SYMBOL(sock_common_recvmsg
);
2613 * Set socket options on an inet socket.
2615 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2616 char __user
*optval
, unsigned int optlen
)
2618 struct sock
*sk
= sock
->sk
;
2620 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2622 EXPORT_SYMBOL(sock_common_setsockopt
);
2624 #ifdef CONFIG_COMPAT
2625 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2626 char __user
*optval
, unsigned int optlen
)
2628 struct sock
*sk
= sock
->sk
;
2630 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2631 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2633 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2635 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2638 void sk_common_release(struct sock
*sk
)
2640 if (sk
->sk_prot
->destroy
)
2641 sk
->sk_prot
->destroy(sk
);
2644 * Observation: when sock_common_release is called, processes have
2645 * no access to socket. But net still has.
2646 * Step one, detach it from networking:
2648 * A. Remove from hash tables.
2651 sk
->sk_prot
->unhash(sk
);
2654 * In this point socket cannot receive new packets, but it is possible
2655 * that some packets are in flight because some CPU runs receiver and
2656 * did hash table lookup before we unhashed socket. They will achieve
2657 * receive queue and will be purged by socket destructor.
2659 * Also we still have packets pending on receive queue and probably,
2660 * our own packets waiting in device queues. sock_destroy will drain
2661 * receive queue, but transmitted packets will delay socket destruction
2662 * until the last reference will be released.
2667 xfrm_sk_free_policy(sk
);
2669 sk_refcnt_debug_release(sk
);
2671 if (sk
->sk_frag
.page
) {
2672 put_page(sk
->sk_frag
.page
);
2673 sk
->sk_frag
.page
= NULL
;
2678 EXPORT_SYMBOL(sk_common_release
);
2680 #ifdef CONFIG_PROC_FS
2681 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2683 int val
[PROTO_INUSE_NR
];
2686 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2688 #ifdef CONFIG_NET_NS
2689 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2691 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2693 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2695 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2697 int cpu
, idx
= prot
->inuse_idx
;
2700 for_each_possible_cpu(cpu
)
2701 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2703 return res
>= 0 ? res
: 0;
2705 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2707 static int __net_init
sock_inuse_init_net(struct net
*net
)
2709 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2710 return net
->core
.inuse
? 0 : -ENOMEM
;
2713 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2715 free_percpu(net
->core
.inuse
);
2718 static struct pernet_operations net_inuse_ops
= {
2719 .init
= sock_inuse_init_net
,
2720 .exit
= sock_inuse_exit_net
,
2723 static __init
int net_inuse_init(void)
2725 if (register_pernet_subsys(&net_inuse_ops
))
2726 panic("Cannot initialize net inuse counters");
2731 core_initcall(net_inuse_init
);
2733 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2735 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2737 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2739 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2741 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2743 int cpu
, idx
= prot
->inuse_idx
;
2746 for_each_possible_cpu(cpu
)
2747 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2749 return res
>= 0 ? res
: 0;
2751 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2754 static void assign_proto_idx(struct proto
*prot
)
2756 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2758 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2759 pr_err("PROTO_INUSE_NR exhausted\n");
2763 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2766 static void release_proto_idx(struct proto
*prot
)
2768 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2769 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2772 static inline void assign_proto_idx(struct proto
*prot
)
2776 static inline void release_proto_idx(struct proto
*prot
)
2781 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
2785 kfree(rsk_prot
->slab_name
);
2786 rsk_prot
->slab_name
= NULL
;
2787 kmem_cache_destroy(rsk_prot
->slab
);
2788 rsk_prot
->slab
= NULL
;
2791 static int req_prot_init(const struct proto
*prot
)
2793 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
2798 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
2800 if (!rsk_prot
->slab_name
)
2803 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
2804 rsk_prot
->obj_size
, 0,
2805 prot
->slab_flags
, NULL
);
2807 if (!rsk_prot
->slab
) {
2808 pr_crit("%s: Can't create request sock SLAB cache!\n",
2815 int proto_register(struct proto
*prot
, int alloc_slab
)
2818 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2819 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2822 if (prot
->slab
== NULL
) {
2823 pr_crit("%s: Can't create sock SLAB cache!\n",
2828 if (req_prot_init(prot
))
2829 goto out_free_request_sock_slab
;
2831 if (prot
->twsk_prot
!= NULL
) {
2832 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2834 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2835 goto out_free_request_sock_slab
;
2837 prot
->twsk_prot
->twsk_slab
=
2838 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2839 prot
->twsk_prot
->twsk_obj_size
,
2843 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2844 goto out_free_timewait_sock_slab_name
;
2848 mutex_lock(&proto_list_mutex
);
2849 list_add(&prot
->node
, &proto_list
);
2850 assign_proto_idx(prot
);
2851 mutex_unlock(&proto_list_mutex
);
2854 out_free_timewait_sock_slab_name
:
2855 kfree(prot
->twsk_prot
->twsk_slab_name
);
2856 out_free_request_sock_slab
:
2857 req_prot_cleanup(prot
->rsk_prot
);
2859 kmem_cache_destroy(prot
->slab
);
2864 EXPORT_SYMBOL(proto_register
);
2866 void proto_unregister(struct proto
*prot
)
2868 mutex_lock(&proto_list_mutex
);
2869 release_proto_idx(prot
);
2870 list_del(&prot
->node
);
2871 mutex_unlock(&proto_list_mutex
);
2873 kmem_cache_destroy(prot
->slab
);
2876 req_prot_cleanup(prot
->rsk_prot
);
2878 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2879 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2880 kfree(prot
->twsk_prot
->twsk_slab_name
);
2881 prot
->twsk_prot
->twsk_slab
= NULL
;
2884 EXPORT_SYMBOL(proto_unregister
);
2886 #ifdef CONFIG_PROC_FS
2887 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2888 __acquires(proto_list_mutex
)
2890 mutex_lock(&proto_list_mutex
);
2891 return seq_list_start_head(&proto_list
, *pos
);
2894 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2896 return seq_list_next(v
, &proto_list
, pos
);
2899 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2900 __releases(proto_list_mutex
)
2902 mutex_unlock(&proto_list_mutex
);
2905 static char proto_method_implemented(const void *method
)
2907 return method
== NULL
? 'n' : 'y';
2909 static long sock_prot_memory_allocated(struct proto
*proto
)
2911 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2914 static char *sock_prot_memory_pressure(struct proto
*proto
)
2916 return proto
->memory_pressure
!= NULL
?
2917 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2920 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2923 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2924 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2927 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2928 sock_prot_memory_allocated(proto
),
2929 sock_prot_memory_pressure(proto
),
2931 proto
->slab
== NULL
? "no" : "yes",
2932 module_name(proto
->owner
),
2933 proto_method_implemented(proto
->close
),
2934 proto_method_implemented(proto
->connect
),
2935 proto_method_implemented(proto
->disconnect
),
2936 proto_method_implemented(proto
->accept
),
2937 proto_method_implemented(proto
->ioctl
),
2938 proto_method_implemented(proto
->init
),
2939 proto_method_implemented(proto
->destroy
),
2940 proto_method_implemented(proto
->shutdown
),
2941 proto_method_implemented(proto
->setsockopt
),
2942 proto_method_implemented(proto
->getsockopt
),
2943 proto_method_implemented(proto
->sendmsg
),
2944 proto_method_implemented(proto
->recvmsg
),
2945 proto_method_implemented(proto
->sendpage
),
2946 proto_method_implemented(proto
->bind
),
2947 proto_method_implemented(proto
->backlog_rcv
),
2948 proto_method_implemented(proto
->hash
),
2949 proto_method_implemented(proto
->unhash
),
2950 proto_method_implemented(proto
->get_port
),
2951 proto_method_implemented(proto
->enter_memory_pressure
));
2954 static int proto_seq_show(struct seq_file
*seq
, void *v
)
2956 if (v
== &proto_list
)
2957 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2966 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2968 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
2972 static const struct seq_operations proto_seq_ops
= {
2973 .start
= proto_seq_start
,
2974 .next
= proto_seq_next
,
2975 .stop
= proto_seq_stop
,
2976 .show
= proto_seq_show
,
2979 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
2981 return seq_open_net(inode
, file
, &proto_seq_ops
,
2982 sizeof(struct seq_net_private
));
2985 static const struct file_operations proto_seq_fops
= {
2986 .owner
= THIS_MODULE
,
2987 .open
= proto_seq_open
,
2989 .llseek
= seq_lseek
,
2990 .release
= seq_release_net
,
2993 static __net_init
int proto_init_net(struct net
*net
)
2995 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3001 static __net_exit
void proto_exit_net(struct net
*net
)
3003 remove_proc_entry("protocols", net
->proc_net
);
3007 static __net_initdata
struct pernet_operations proto_net_ops
= {
3008 .init
= proto_init_net
,
3009 .exit
= proto_exit_net
,
3012 static int __init
proto_init(void)
3014 return register_pernet_subsys(&proto_net_ops
);
3017 subsys_initcall(proto_init
);
3019 #endif /* PROC_FS */