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
,
456 const int nested
, unsigned int trim_cap
)
458 int rc
= NET_RX_SUCCESS
;
460 if (sk_filter_trim_cap(sk
, skb
, trim_cap
))
461 goto discard_and_relse
;
465 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
466 atomic_inc(&sk
->sk_drops
);
467 goto discard_and_relse
;
470 bh_lock_sock_nested(sk
);
473 if (!sock_owned_by_user(sk
)) {
475 * trylock + unlock semantics:
477 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
479 rc
= sk_backlog_rcv(sk
, skb
);
481 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
482 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
484 atomic_inc(&sk
->sk_drops
);
485 goto discard_and_relse
;
496 EXPORT_SYMBOL(__sk_receive_skb
);
498 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
500 struct dst_entry
*dst
= __sk_dst_get(sk
);
502 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
503 sk_tx_queue_clear(sk
);
504 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
511 EXPORT_SYMBOL(__sk_dst_check
);
513 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
515 struct dst_entry
*dst
= sk_dst_get(sk
);
517 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
525 EXPORT_SYMBOL(sk_dst_check
);
527 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
530 int ret
= -ENOPROTOOPT
;
531 #ifdef CONFIG_NETDEVICES
532 struct net
*net
= sock_net(sk
);
533 char devname
[IFNAMSIZ
];
538 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
545 /* Bind this socket to a particular device like "eth0",
546 * as specified in the passed interface name. If the
547 * name is "" or the option length is zero the socket
550 if (optlen
> IFNAMSIZ
- 1)
551 optlen
= IFNAMSIZ
- 1;
552 memset(devname
, 0, sizeof(devname
));
555 if (copy_from_user(devname
, optval
, optlen
))
559 if (devname
[0] != '\0') {
560 struct net_device
*dev
;
563 dev
= dev_get_by_name_rcu(net
, devname
);
565 index
= dev
->ifindex
;
573 sk
->sk_bound_dev_if
= index
;
585 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
586 int __user
*optlen
, int len
)
588 int ret
= -ENOPROTOOPT
;
589 #ifdef CONFIG_NETDEVICES
590 struct net
*net
= sock_net(sk
);
591 char devname
[IFNAMSIZ
];
593 if (sk
->sk_bound_dev_if
== 0) {
602 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
606 len
= strlen(devname
) + 1;
609 if (copy_to_user(optval
, devname
, len
))
614 if (put_user(len
, optlen
))
625 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
628 sock_set_flag(sk
, bit
);
630 sock_reset_flag(sk
, bit
);
633 bool sk_mc_loop(struct sock
*sk
)
635 if (dev_recursion_level())
639 switch (sk
->sk_family
) {
641 return inet_sk(sk
)->mc_loop
;
642 #if IS_ENABLED(CONFIG_IPV6)
644 return inet6_sk(sk
)->mc_loop
;
650 EXPORT_SYMBOL(sk_mc_loop
);
653 * This is meant for all protocols to use and covers goings on
654 * at the socket level. Everything here is generic.
657 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
658 char __user
*optval
, unsigned int optlen
)
660 struct sock
*sk
= sock
->sk
;
667 * Options without arguments
670 if (optname
== SO_BINDTODEVICE
)
671 return sock_setbindtodevice(sk
, optval
, optlen
);
673 if (optlen
< sizeof(int))
676 if (get_user(val
, (int __user
*)optval
))
679 valbool
= val
? 1 : 0;
685 if (val
&& !capable(CAP_NET_ADMIN
))
688 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
691 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
694 sk
->sk_reuseport
= valbool
;
703 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
706 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
709 /* Don't error on this BSD doesn't and if you think
710 * about it this is right. Otherwise apps have to
711 * play 'guess the biggest size' games. RCVBUF/SNDBUF
712 * are treated in BSD as hints
714 val
= min_t(u32
, val
, sysctl_wmem_max
);
716 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
717 sk
->sk_sndbuf
= max_t(u32
, val
* 2, SOCK_MIN_SNDBUF
);
718 /* Wake up sending tasks if we upped the value. */
719 sk
->sk_write_space(sk
);
723 if (!capable(CAP_NET_ADMIN
)) {
730 /* Don't error on this BSD doesn't and if you think
731 * about it this is right. Otherwise apps have to
732 * play 'guess the biggest size' games. RCVBUF/SNDBUF
733 * are treated in BSD as hints
735 val
= min_t(u32
, val
, sysctl_rmem_max
);
737 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
739 * We double it on the way in to account for
740 * "struct sk_buff" etc. overhead. Applications
741 * assume that the SO_RCVBUF setting they make will
742 * allow that much actual data to be received on that
745 * Applications are unaware that "struct sk_buff" and
746 * other overheads allocate from the receive buffer
747 * during socket buffer allocation.
749 * And after considering the possible alternatives,
750 * returning the value we actually used in getsockopt
751 * is the most desirable behavior.
753 sk
->sk_rcvbuf
= max_t(u32
, val
* 2, SOCK_MIN_RCVBUF
);
757 if (!capable(CAP_NET_ADMIN
)) {
765 if (sk
->sk_protocol
== IPPROTO_TCP
&&
766 sk
->sk_type
== SOCK_STREAM
)
767 tcp_set_keepalive(sk
, valbool
);
769 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
773 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
777 sk
->sk_no_check_tx
= valbool
;
781 if ((val
>= 0 && val
<= 6) ||
782 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
783 sk
->sk_priority
= val
;
789 if (optlen
< sizeof(ling
)) {
790 ret
= -EINVAL
; /* 1003.1g */
793 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
798 sock_reset_flag(sk
, SOCK_LINGER
);
800 #if (BITS_PER_LONG == 32)
801 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
802 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
805 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
806 sock_set_flag(sk
, SOCK_LINGER
);
811 sock_warn_obsolete_bsdism("setsockopt");
816 set_bit(SOCK_PASSCRED
, &sock
->flags
);
818 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
824 if (optname
== SO_TIMESTAMP
)
825 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
827 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
828 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
829 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
831 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
832 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
836 case SO_TIMESTAMPING
:
837 if (val
& ~SOF_TIMESTAMPING_MASK
) {
842 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
843 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
844 if (sk
->sk_protocol
== IPPROTO_TCP
&&
845 sk
->sk_type
== SOCK_STREAM
) {
846 if ((1 << sk
->sk_state
) &
847 (TCPF_CLOSE
| TCPF_LISTEN
)) {
851 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
856 sk
->sk_tsflags
= val
;
857 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
858 sock_enable_timestamp(sk
,
859 SOCK_TIMESTAMPING_RX_SOFTWARE
);
861 sock_disable_timestamp(sk
,
862 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
868 sk
->sk_rcvlowat
= val
? : 1;
872 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
);
876 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
);
879 case SO_ATTACH_FILTER
:
881 if (optlen
== sizeof(struct sock_fprog
)) {
882 struct sock_fprog fprog
;
885 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
888 ret
= sk_attach_filter(&fprog
, sk
);
894 if (optlen
== sizeof(u32
)) {
898 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
901 ret
= sk_attach_bpf(ufd
, sk
);
905 case SO_ATTACH_REUSEPORT_CBPF
:
907 if (optlen
== sizeof(struct sock_fprog
)) {
908 struct sock_fprog fprog
;
911 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
914 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
918 case SO_ATTACH_REUSEPORT_EBPF
:
920 if (optlen
== sizeof(u32
)) {
924 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
927 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
931 case SO_DETACH_FILTER
:
932 ret
= sk_detach_filter(sk
);
936 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
939 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
944 set_bit(SOCK_PASSSEC
, &sock
->flags
);
946 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
949 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
956 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
960 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
964 if (sock
->ops
->set_peek_off
)
965 ret
= sock
->ops
->set_peek_off(sk
, val
);
971 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
974 case SO_SELECT_ERR_QUEUE
:
975 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
978 #ifdef CONFIG_NET_RX_BUSY_POLL
980 /* allow unprivileged users to decrease the value */
981 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
987 sk
->sk_ll_usec
= val
;
992 case SO_MAX_PACING_RATE
:
993 sk
->sk_max_pacing_rate
= val
;
994 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
,
995 sk
->sk_max_pacing_rate
);
998 case SO_INCOMING_CPU
:
999 sk
->sk_incoming_cpu
= val
;
1004 dst_negative_advice(sk
);
1013 EXPORT_SYMBOL(sock_setsockopt
);
1016 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1017 struct ucred
*ucred
)
1019 ucred
->pid
= pid_vnr(pid
);
1020 ucred
->uid
= ucred
->gid
= -1;
1022 struct user_namespace
*current_ns
= current_user_ns();
1024 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1025 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1029 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1030 char __user
*optval
, int __user
*optlen
)
1032 struct sock
*sk
= sock
->sk
;
1040 int lv
= sizeof(int);
1043 if (get_user(len
, optlen
))
1048 memset(&v
, 0, sizeof(v
));
1052 v
.val
= sock_flag(sk
, SOCK_DBG
);
1056 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1060 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1064 v
.val
= sk
->sk_sndbuf
;
1068 v
.val
= sk
->sk_rcvbuf
;
1072 v
.val
= sk
->sk_reuse
;
1076 v
.val
= sk
->sk_reuseport
;
1080 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1084 v
.val
= sk
->sk_type
;
1088 v
.val
= sk
->sk_protocol
;
1092 v
.val
= sk
->sk_family
;
1096 v
.val
= -sock_error(sk
);
1098 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1102 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1106 v
.val
= sk
->sk_no_check_tx
;
1110 v
.val
= sk
->sk_priority
;
1114 lv
= sizeof(v
.ling
);
1115 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1116 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1120 sock_warn_obsolete_bsdism("getsockopt");
1124 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1125 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1128 case SO_TIMESTAMPNS
:
1129 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1132 case SO_TIMESTAMPING
:
1133 v
.val
= sk
->sk_tsflags
;
1137 lv
= sizeof(struct timeval
);
1138 if (sk
->sk_rcvtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1142 v
.tm
.tv_sec
= sk
->sk_rcvtimeo
/ HZ
;
1143 v
.tm
.tv_usec
= ((sk
->sk_rcvtimeo
% HZ
) * 1000000) / HZ
;
1148 lv
= sizeof(struct timeval
);
1149 if (sk
->sk_sndtimeo
== MAX_SCHEDULE_TIMEOUT
) {
1153 v
.tm
.tv_sec
= sk
->sk_sndtimeo
/ HZ
;
1154 v
.tm
.tv_usec
= ((sk
->sk_sndtimeo
% HZ
) * 1000000) / HZ
;
1159 v
.val
= sk
->sk_rcvlowat
;
1167 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1172 struct ucred peercred
;
1173 if (len
> sizeof(peercred
))
1174 len
= sizeof(peercred
);
1175 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1176 if (copy_to_user(optval
, &peercred
, len
))
1185 if (sock
->ops
->getname(sock
, (struct sockaddr
*)address
, &lv
, 2))
1189 if (copy_to_user(optval
, address
, len
))
1194 /* Dubious BSD thing... Probably nobody even uses it, but
1195 * the UNIX standard wants it for whatever reason... -DaveM
1198 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1202 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1206 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1209 v
.val
= sk
->sk_mark
;
1213 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1216 case SO_WIFI_STATUS
:
1217 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1221 if (!sock
->ops
->set_peek_off
)
1224 v
.val
= sk
->sk_peek_off
;
1227 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1230 case SO_BINDTODEVICE
:
1231 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1234 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1240 case SO_LOCK_FILTER
:
1241 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1244 case SO_BPF_EXTENSIONS
:
1245 v
.val
= bpf_tell_extensions();
1248 case SO_SELECT_ERR_QUEUE
:
1249 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1252 #ifdef CONFIG_NET_RX_BUSY_POLL
1254 v
.val
= sk
->sk_ll_usec
;
1258 case SO_MAX_PACING_RATE
:
1259 v
.val
= sk
->sk_max_pacing_rate
;
1262 case SO_INCOMING_CPU
:
1263 v
.val
= sk
->sk_incoming_cpu
;
1267 /* We implement the SO_SNDLOWAT etc to not be settable
1270 return -ENOPROTOOPT
;
1275 if (copy_to_user(optval
, &v
, len
))
1278 if (put_user(len
, optlen
))
1284 * Initialize an sk_lock.
1286 * (We also register the sk_lock with the lock validator.)
1288 static inline void sock_lock_init(struct sock
*sk
)
1290 sock_lock_init_class_and_name(sk
,
1291 af_family_slock_key_strings
[sk
->sk_family
],
1292 af_family_slock_keys
+ sk
->sk_family
,
1293 af_family_key_strings
[sk
->sk_family
],
1294 af_family_keys
+ sk
->sk_family
);
1298 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1299 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1300 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1302 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1304 #ifdef CONFIG_SECURITY_NETWORK
1305 void *sptr
= nsk
->sk_security
;
1307 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1309 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1310 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1312 #ifdef CONFIG_SECURITY_NETWORK
1313 nsk
->sk_security
= sptr
;
1314 security_sk_clone(osk
, nsk
);
1318 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1322 struct kmem_cache
*slab
;
1326 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1329 if (priority
& __GFP_ZERO
)
1330 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1332 sk
= kmalloc(prot
->obj_size
, priority
);
1335 kmemcheck_annotate_bitfield(sk
, flags
);
1337 if (security_sk_alloc(sk
, family
, priority
))
1340 if (!try_module_get(prot
->owner
))
1342 sk_tx_queue_clear(sk
);
1343 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1349 security_sk_free(sk
);
1352 kmem_cache_free(slab
, sk
);
1358 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1360 struct kmem_cache
*slab
;
1361 struct module
*owner
;
1363 owner
= prot
->owner
;
1366 cgroup_sk_free(&sk
->sk_cgrp_data
);
1367 security_sk_free(sk
);
1369 kmem_cache_free(slab
, sk
);
1376 * sk_alloc - All socket objects are allocated here
1377 * @net: the applicable net namespace
1378 * @family: protocol family
1379 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1380 * @prot: struct proto associated with this new sock instance
1381 * @kern: is this to be a kernel socket?
1383 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1384 struct proto
*prot
, int kern
)
1388 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1390 sk
->sk_family
= family
;
1392 * See comment in struct sock definition to understand
1393 * why we need sk_prot_creator -acme
1395 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1397 sk
->sk_net_refcnt
= kern
? 0 : 1;
1398 if (likely(sk
->sk_net_refcnt
))
1400 sock_net_set(sk
, net
);
1401 atomic_set(&sk
->sk_wmem_alloc
, 1);
1403 sock_update_classid(&sk
->sk_cgrp_data
);
1404 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1409 EXPORT_SYMBOL(sk_alloc
);
1411 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1412 * grace period. This is the case for UDP sockets and TCP listeners.
1414 static void __sk_destruct(struct rcu_head
*head
)
1416 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1417 struct sk_filter
*filter
;
1419 if (sk
->sk_destruct
)
1420 sk
->sk_destruct(sk
);
1422 filter
= rcu_dereference_check(sk
->sk_filter
,
1423 atomic_read(&sk
->sk_wmem_alloc
) == 0);
1425 sk_filter_uncharge(sk
, filter
);
1426 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1428 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1429 reuseport_detach_sock(sk
);
1431 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1433 if (atomic_read(&sk
->sk_omem_alloc
))
1434 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1435 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1437 if (sk
->sk_peer_cred
)
1438 put_cred(sk
->sk_peer_cred
);
1439 put_pid(sk
->sk_peer_pid
);
1440 if (likely(sk
->sk_net_refcnt
))
1441 put_net(sock_net(sk
));
1442 sk_prot_free(sk
->sk_prot_creator
, sk
);
1445 void sk_destruct(struct sock
*sk
)
1447 if (sock_flag(sk
, SOCK_RCU_FREE
))
1448 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1450 __sk_destruct(&sk
->sk_rcu
);
1453 static void __sk_free(struct sock
*sk
)
1455 if (unlikely(sock_diag_has_destroy_listeners(sk
) && sk
->sk_net_refcnt
))
1456 sock_diag_broadcast_destroy(sk
);
1461 void sk_free(struct sock
*sk
)
1464 * We subtract one from sk_wmem_alloc and can know if
1465 * some packets are still in some tx queue.
1466 * If not null, sock_wfree() will call __sk_free(sk) later
1468 if (atomic_dec_and_test(&sk
->sk_wmem_alloc
))
1471 EXPORT_SYMBOL(sk_free
);
1474 * sk_clone_lock - clone a socket, and lock its clone
1475 * @sk: the socket to clone
1476 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1478 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1480 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1483 bool is_charged
= true;
1485 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1486 if (newsk
!= NULL
) {
1487 struct sk_filter
*filter
;
1489 sock_copy(newsk
, sk
);
1492 if (likely(newsk
->sk_net_refcnt
))
1493 get_net(sock_net(newsk
));
1494 sk_node_init(&newsk
->sk_node
);
1495 sock_lock_init(newsk
);
1496 bh_lock_sock(newsk
);
1497 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1498 newsk
->sk_backlog
.len
= 0;
1500 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1502 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1504 atomic_set(&newsk
->sk_wmem_alloc
, 1);
1505 atomic_set(&newsk
->sk_omem_alloc
, 0);
1506 skb_queue_head_init(&newsk
->sk_receive_queue
);
1507 skb_queue_head_init(&newsk
->sk_write_queue
);
1509 rwlock_init(&newsk
->sk_callback_lock
);
1510 lockdep_set_class_and_name(&newsk
->sk_callback_lock
,
1511 af_callback_keys
+ newsk
->sk_family
,
1512 af_family_clock_key_strings
[newsk
->sk_family
]);
1514 newsk
->sk_dst_cache
= NULL
;
1515 newsk
->sk_wmem_queued
= 0;
1516 newsk
->sk_forward_alloc
= 0;
1517 atomic_set(&newsk
->sk_drops
, 0);
1518 newsk
->sk_send_head
= NULL
;
1519 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1521 sock_reset_flag(newsk
, SOCK_DONE
);
1522 skb_queue_head_init(&newsk
->sk_error_queue
);
1524 filter
= rcu_dereference_protected(newsk
->sk_filter
, 1);
1526 /* though it's an empty new sock, the charging may fail
1527 * if sysctl_optmem_max was changed between creation of
1528 * original socket and cloning
1530 is_charged
= sk_filter_charge(newsk
, filter
);
1532 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1533 /* It is still raw copy of parent, so invalidate
1534 * destructor and make plain sk_free() */
1535 newsk
->sk_destruct
= NULL
;
1536 bh_unlock_sock(newsk
);
1541 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1544 newsk
->sk_priority
= 0;
1545 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1546 atomic64_set(&newsk
->sk_cookie
, 0);
1548 * Before updating sk_refcnt, we must commit prior changes to memory
1549 * (Documentation/RCU/rculist_nulls.txt for details)
1552 atomic_set(&newsk
->sk_refcnt
, 2);
1555 * Increment the counter in the same struct proto as the master
1556 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1557 * is the same as sk->sk_prot->socks, as this field was copied
1560 * This _changes_ the previous behaviour, where
1561 * tcp_create_openreq_child always was incrementing the
1562 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1563 * to be taken into account in all callers. -acme
1565 sk_refcnt_debug_inc(newsk
);
1566 sk_set_socket(newsk
, NULL
);
1567 newsk
->sk_wq
= NULL
;
1569 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
1570 sock_update_memcg(newsk
);
1572 if (newsk
->sk_prot
->sockets_allocated
)
1573 sk_sockets_allocated_inc(newsk
);
1575 if (sock_needs_netstamp(sk
) &&
1576 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1577 net_enable_timestamp();
1582 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1584 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1588 sk_dst_set(sk
, dst
);
1589 sk
->sk_route_caps
= dst
->dev
->features
;
1590 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1591 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1592 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1593 if (sk_can_gso(sk
)) {
1594 if (dst
->header_len
) {
1595 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1597 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1598 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1599 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1602 sk
->sk_gso_max_segs
= max_segs
;
1604 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1607 * Simple resource managers for sockets.
1612 * Write buffer destructor automatically called from kfree_skb.
1614 void sock_wfree(struct sk_buff
*skb
)
1616 struct sock
*sk
= skb
->sk
;
1617 unsigned int len
= skb
->truesize
;
1619 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1621 * Keep a reference on sk_wmem_alloc, this will be released
1622 * after sk_write_space() call
1624 atomic_sub(len
- 1, &sk
->sk_wmem_alloc
);
1625 sk
->sk_write_space(sk
);
1629 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1630 * could not do because of in-flight packets
1632 if (atomic_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1635 EXPORT_SYMBOL(sock_wfree
);
1637 /* This variant of sock_wfree() is used by TCP,
1638 * since it sets SOCK_USE_WRITE_QUEUE.
1640 void __sock_wfree(struct sk_buff
*skb
)
1642 struct sock
*sk
= skb
->sk
;
1644 if (atomic_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
))
1648 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1653 if (unlikely(!sk_fullsock(sk
))) {
1654 skb
->destructor
= sock_edemux
;
1659 skb
->destructor
= sock_wfree
;
1660 skb_set_hash_from_sk(skb
, sk
);
1662 * We used to take a refcount on sk, but following operation
1663 * is enough to guarantee sk_free() wont free this sock until
1664 * all in-flight packets are completed
1666 atomic_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1668 EXPORT_SYMBOL(skb_set_owner_w
);
1670 /* This helper is used by netem, as it can hold packets in its
1671 * delay queue. We want to allow the owner socket to send more
1672 * packets, as if they were already TX completed by a typical driver.
1673 * But we also want to keep skb->sk set because some packet schedulers
1674 * rely on it (sch_fq for example). So we set skb->truesize to a small
1675 * amount (1) and decrease sk_wmem_alloc accordingly.
1677 void skb_orphan_partial(struct sk_buff
*skb
)
1679 /* If this skb is a TCP pure ACK or already went here,
1680 * we have nothing to do. 2 is already a very small truesize.
1682 if (skb
->truesize
<= 2)
1685 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1686 * so we do not completely orphan skb, but transfert all
1687 * accounted bytes but one, to avoid unexpected reorders.
1689 if (skb
->destructor
== sock_wfree
1691 || skb
->destructor
== tcp_wfree
1694 atomic_sub(skb
->truesize
- 1, &skb
->sk
->sk_wmem_alloc
);
1700 EXPORT_SYMBOL(skb_orphan_partial
);
1703 * Read buffer destructor automatically called from kfree_skb.
1705 void sock_rfree(struct sk_buff
*skb
)
1707 struct sock
*sk
= skb
->sk
;
1708 unsigned int len
= skb
->truesize
;
1710 atomic_sub(len
, &sk
->sk_rmem_alloc
);
1711 sk_mem_uncharge(sk
, len
);
1713 EXPORT_SYMBOL(sock_rfree
);
1716 * Buffer destructor for skbs that are not used directly in read or write
1717 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1719 void sock_efree(struct sk_buff
*skb
)
1723 EXPORT_SYMBOL(sock_efree
);
1725 kuid_t
sock_i_uid(struct sock
*sk
)
1729 read_lock_bh(&sk
->sk_callback_lock
);
1730 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
1731 read_unlock_bh(&sk
->sk_callback_lock
);
1734 EXPORT_SYMBOL(sock_i_uid
);
1736 unsigned long sock_i_ino(struct sock
*sk
)
1740 read_lock_bh(&sk
->sk_callback_lock
);
1741 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
1742 read_unlock_bh(&sk
->sk_callback_lock
);
1745 EXPORT_SYMBOL(sock_i_ino
);
1748 * Allocate a skb from the socket's send buffer.
1750 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
1753 if (force
|| atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
1754 struct sk_buff
*skb
= alloc_skb(size
, priority
);
1756 skb_set_owner_w(skb
, sk
);
1762 EXPORT_SYMBOL(sock_wmalloc
);
1765 * Allocate a memory block from the socket's option memory buffer.
1767 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
1769 if ((unsigned int)size
<= sysctl_optmem_max
&&
1770 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
1772 /* First do the add, to avoid the race if kmalloc
1775 atomic_add(size
, &sk
->sk_omem_alloc
);
1776 mem
= kmalloc(size
, priority
);
1779 atomic_sub(size
, &sk
->sk_omem_alloc
);
1783 EXPORT_SYMBOL(sock_kmalloc
);
1785 /* Free an option memory block. Note, we actually want the inline
1786 * here as this allows gcc to detect the nullify and fold away the
1787 * condition entirely.
1789 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
1792 if (WARN_ON_ONCE(!mem
))
1798 atomic_sub(size
, &sk
->sk_omem_alloc
);
1801 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
1803 __sock_kfree_s(sk
, mem
, size
, false);
1805 EXPORT_SYMBOL(sock_kfree_s
);
1807 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
1809 __sock_kfree_s(sk
, mem
, size
, true);
1811 EXPORT_SYMBOL(sock_kzfree_s
);
1813 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1814 I think, these locks should be removed for datagram sockets.
1816 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
1820 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1824 if (signal_pending(current
))
1826 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1827 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1828 if (atomic_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
1830 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1834 timeo
= schedule_timeout(timeo
);
1836 finish_wait(sk_sleep(sk
), &wait
);
1842 * Generic send/receive buffer handlers
1845 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
1846 unsigned long data_len
, int noblock
,
1847 int *errcode
, int max_page_order
)
1849 struct sk_buff
*skb
;
1853 timeo
= sock_sndtimeo(sk
, noblock
);
1855 err
= sock_error(sk
);
1860 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1863 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
1866 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
1867 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1871 if (signal_pending(current
))
1873 timeo
= sock_wait_for_wmem(sk
, timeo
);
1875 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
1876 errcode
, sk
->sk_allocation
);
1878 skb_set_owner_w(skb
, sk
);
1882 err
= sock_intr_errno(timeo
);
1887 EXPORT_SYMBOL(sock_alloc_send_pskb
);
1889 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
1890 int noblock
, int *errcode
)
1892 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
1894 EXPORT_SYMBOL(sock_alloc_send_skb
);
1896 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
1897 struct sockcm_cookie
*sockc
)
1901 switch (cmsg
->cmsg_type
) {
1903 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
1905 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1907 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
1909 case SO_TIMESTAMPING
:
1910 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
1913 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
1914 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
1917 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
1918 sockc
->tsflags
|= tsflags
;
1920 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1922 case SCM_CREDENTIALS
:
1929 EXPORT_SYMBOL(__sock_cmsg_send
);
1931 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
1932 struct sockcm_cookie
*sockc
)
1934 struct cmsghdr
*cmsg
;
1937 for_each_cmsghdr(cmsg
, msg
) {
1938 if (!CMSG_OK(msg
, cmsg
))
1940 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
1942 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
1948 EXPORT_SYMBOL(sock_cmsg_send
);
1950 /* On 32bit arches, an skb frag is limited to 2^15 */
1951 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1954 * skb_page_frag_refill - check that a page_frag contains enough room
1955 * @sz: minimum size of the fragment we want to get
1956 * @pfrag: pointer to page_frag
1957 * @gfp: priority for memory allocation
1959 * Note: While this allocator tries to use high order pages, there is
1960 * no guarantee that allocations succeed. Therefore, @sz MUST be
1961 * less or equal than PAGE_SIZE.
1963 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
1966 if (page_ref_count(pfrag
->page
) == 1) {
1970 if (pfrag
->offset
+ sz
<= pfrag
->size
)
1972 put_page(pfrag
->page
);
1976 if (SKB_FRAG_PAGE_ORDER
) {
1977 /* Avoid direct reclaim but allow kswapd to wake */
1978 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
1979 __GFP_COMP
| __GFP_NOWARN
|
1981 SKB_FRAG_PAGE_ORDER
);
1982 if (likely(pfrag
->page
)) {
1983 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
1987 pfrag
->page
= alloc_page(gfp
);
1988 if (likely(pfrag
->page
)) {
1989 pfrag
->size
= PAGE_SIZE
;
1994 EXPORT_SYMBOL(skb_page_frag_refill
);
1996 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
1998 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
2001 sk_enter_memory_pressure(sk
);
2002 sk_stream_moderate_sndbuf(sk
);
2005 EXPORT_SYMBOL(sk_page_frag_refill
);
2007 static void __lock_sock(struct sock
*sk
)
2008 __releases(&sk
->sk_lock
.slock
)
2009 __acquires(&sk
->sk_lock
.slock
)
2014 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2015 TASK_UNINTERRUPTIBLE
);
2016 spin_unlock_bh(&sk
->sk_lock
.slock
);
2018 spin_lock_bh(&sk
->sk_lock
.slock
);
2019 if (!sock_owned_by_user(sk
))
2022 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2025 static void __release_sock(struct sock
*sk
)
2026 __releases(&sk
->sk_lock
.slock
)
2027 __acquires(&sk
->sk_lock
.slock
)
2029 struct sk_buff
*skb
, *next
;
2031 while ((skb
= sk
->sk_backlog
.head
) != NULL
) {
2032 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2034 spin_unlock_bh(&sk
->sk_lock
.slock
);
2039 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2041 sk_backlog_rcv(sk
, skb
);
2046 } while (skb
!= NULL
);
2048 spin_lock_bh(&sk
->sk_lock
.slock
);
2052 * Doing the zeroing here guarantee we can not loop forever
2053 * while a wild producer attempts to flood us.
2055 sk
->sk_backlog
.len
= 0;
2058 void __sk_flush_backlog(struct sock
*sk
)
2060 spin_lock_bh(&sk
->sk_lock
.slock
);
2062 spin_unlock_bh(&sk
->sk_lock
.slock
);
2066 * sk_wait_data - wait for data to arrive at sk_receive_queue
2067 * @sk: sock to wait on
2068 * @timeo: for how long
2069 * @skb: last skb seen on sk_receive_queue
2071 * Now socket state including sk->sk_err is changed only under lock,
2072 * hence we may omit checks after joining wait queue.
2073 * We check receive queue before schedule() only as optimization;
2074 * it is very likely that release_sock() added new data.
2076 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2081 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2082 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2083 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
);
2084 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2085 finish_wait(sk_sleep(sk
), &wait
);
2088 EXPORT_SYMBOL(sk_wait_data
);
2091 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2093 * @size: memory size to allocate
2094 * @kind: allocation type
2096 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2097 * rmem allocation. This function assumes that protocols which have
2098 * memory_pressure use sk_wmem_queued as write buffer accounting.
2100 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2102 struct proto
*prot
= sk
->sk_prot
;
2103 int amt
= sk_mem_pages(size
);
2106 sk
->sk_forward_alloc
+= amt
* SK_MEM_QUANTUM
;
2108 allocated
= sk_memory_allocated_add(sk
, amt
);
2110 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2111 !mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
))
2112 goto suppress_allocation
;
2115 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2116 sk_leave_memory_pressure(sk
);
2120 /* Under pressure. */
2121 if (allocated
> sk_prot_mem_limits(sk
, 1))
2122 sk_enter_memory_pressure(sk
);
2124 /* Over hard limit. */
2125 if (allocated
> sk_prot_mem_limits(sk
, 2))
2126 goto suppress_allocation
;
2128 /* guarantee minimum buffer size under pressure */
2129 if (kind
== SK_MEM_RECV
) {
2130 if (atomic_read(&sk
->sk_rmem_alloc
) < prot
->sysctl_rmem
[0])
2133 } else { /* SK_MEM_SEND */
2134 if (sk
->sk_type
== SOCK_STREAM
) {
2135 if (sk
->sk_wmem_queued
< prot
->sysctl_wmem
[0])
2137 } else if (atomic_read(&sk
->sk_wmem_alloc
) <
2138 prot
->sysctl_wmem
[0])
2142 if (sk_has_memory_pressure(sk
)) {
2145 if (!sk_under_memory_pressure(sk
))
2147 alloc
= sk_sockets_allocated_read_positive(sk
);
2148 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2149 sk_mem_pages(sk
->sk_wmem_queued
+
2150 atomic_read(&sk
->sk_rmem_alloc
) +
2151 sk
->sk_forward_alloc
))
2155 suppress_allocation
:
2157 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2158 sk_stream_moderate_sndbuf(sk
);
2160 /* Fail only if socket is _under_ its sndbuf.
2161 * In this case we cannot block, so that we have to fail.
2163 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2167 trace_sock_exceed_buf_limit(sk
, prot
, allocated
);
2169 /* Alas. Undo changes. */
2170 sk
->sk_forward_alloc
-= amt
* SK_MEM_QUANTUM
;
2172 sk_memory_allocated_sub(sk
, amt
);
2174 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2175 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2179 EXPORT_SYMBOL(__sk_mem_schedule
);
2182 * __sk_mem_reclaim - reclaim memory_allocated
2184 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2186 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2188 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2189 sk_memory_allocated_sub(sk
, amount
);
2190 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2192 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2193 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2195 if (sk_under_memory_pressure(sk
) &&
2196 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2197 sk_leave_memory_pressure(sk
);
2199 EXPORT_SYMBOL(__sk_mem_reclaim
);
2201 int sk_set_peek_off(struct sock
*sk
, int val
)
2206 sk
->sk_peek_off
= val
;
2209 EXPORT_SYMBOL_GPL(sk_set_peek_off
);
2212 * Set of default routines for initialising struct proto_ops when
2213 * the protocol does not support a particular function. In certain
2214 * cases where it makes no sense for a protocol to have a "do nothing"
2215 * function, some default processing is provided.
2218 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2222 EXPORT_SYMBOL(sock_no_bind
);
2224 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2229 EXPORT_SYMBOL(sock_no_connect
);
2231 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2235 EXPORT_SYMBOL(sock_no_socketpair
);
2237 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
2241 EXPORT_SYMBOL(sock_no_accept
);
2243 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2248 EXPORT_SYMBOL(sock_no_getname
);
2250 unsigned int sock_no_poll(struct file
*file
, struct socket
*sock
, poll_table
*pt
)
2254 EXPORT_SYMBOL(sock_no_poll
);
2256 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2260 EXPORT_SYMBOL(sock_no_ioctl
);
2262 int sock_no_listen(struct socket
*sock
, int backlog
)
2266 EXPORT_SYMBOL(sock_no_listen
);
2268 int sock_no_shutdown(struct socket
*sock
, int how
)
2272 EXPORT_SYMBOL(sock_no_shutdown
);
2274 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2275 char __user
*optval
, unsigned int optlen
)
2279 EXPORT_SYMBOL(sock_no_setsockopt
);
2281 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2282 char __user
*optval
, int __user
*optlen
)
2286 EXPORT_SYMBOL(sock_no_getsockopt
);
2288 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2292 EXPORT_SYMBOL(sock_no_sendmsg
);
2294 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2299 EXPORT_SYMBOL(sock_no_recvmsg
);
2301 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2303 /* Mirror missing mmap method error code */
2306 EXPORT_SYMBOL(sock_no_mmap
);
2308 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2311 struct msghdr msg
= {.msg_flags
= flags
};
2313 char *kaddr
= kmap(page
);
2314 iov
.iov_base
= kaddr
+ offset
;
2316 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2320 EXPORT_SYMBOL(sock_no_sendpage
);
2323 * Default Socket Callbacks
2326 static void sock_def_wakeup(struct sock
*sk
)
2328 struct socket_wq
*wq
;
2331 wq
= rcu_dereference(sk
->sk_wq
);
2332 if (skwq_has_sleeper(wq
))
2333 wake_up_interruptible_all(&wq
->wait
);
2337 static void sock_def_error_report(struct sock
*sk
)
2339 struct socket_wq
*wq
;
2342 wq
= rcu_dereference(sk
->sk_wq
);
2343 if (skwq_has_sleeper(wq
))
2344 wake_up_interruptible_poll(&wq
->wait
, POLLERR
);
2345 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2349 static void sock_def_readable(struct sock
*sk
)
2351 struct socket_wq
*wq
;
2354 wq
= rcu_dereference(sk
->sk_wq
);
2355 if (skwq_has_sleeper(wq
))
2356 wake_up_interruptible_sync_poll(&wq
->wait
, POLLIN
| POLLPRI
|
2357 POLLRDNORM
| POLLRDBAND
);
2358 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2362 static void sock_def_write_space(struct sock
*sk
)
2364 struct socket_wq
*wq
;
2368 /* Do not wake up a writer until he can make "significant"
2371 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2372 wq
= rcu_dereference(sk
->sk_wq
);
2373 if (skwq_has_sleeper(wq
))
2374 wake_up_interruptible_sync_poll(&wq
->wait
, POLLOUT
|
2375 POLLWRNORM
| POLLWRBAND
);
2377 /* Should agree with poll, otherwise some programs break */
2378 if (sock_writeable(sk
))
2379 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2385 static void sock_def_destruct(struct sock
*sk
)
2389 void sk_send_sigurg(struct sock
*sk
)
2391 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2392 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2393 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2395 EXPORT_SYMBOL(sk_send_sigurg
);
2397 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2398 unsigned long expires
)
2400 if (!mod_timer(timer
, expires
))
2403 EXPORT_SYMBOL(sk_reset_timer
);
2405 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2407 if (del_timer(timer
))
2410 EXPORT_SYMBOL(sk_stop_timer
);
2412 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2414 skb_queue_head_init(&sk
->sk_receive_queue
);
2415 skb_queue_head_init(&sk
->sk_write_queue
);
2416 skb_queue_head_init(&sk
->sk_error_queue
);
2418 sk
->sk_send_head
= NULL
;
2420 init_timer(&sk
->sk_timer
);
2422 sk
->sk_allocation
= GFP_KERNEL
;
2423 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2424 sk
->sk_sndbuf
= sysctl_wmem_default
;
2425 sk
->sk_state
= TCP_CLOSE
;
2426 sk_set_socket(sk
, sock
);
2428 sock_set_flag(sk
, SOCK_ZAPPED
);
2431 sk
->sk_type
= sock
->type
;
2432 sk
->sk_wq
= sock
->wq
;
2437 rwlock_init(&sk
->sk_callback_lock
);
2438 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
2439 af_callback_keys
+ sk
->sk_family
,
2440 af_family_clock_key_strings
[sk
->sk_family
]);
2442 sk
->sk_state_change
= sock_def_wakeup
;
2443 sk
->sk_data_ready
= sock_def_readable
;
2444 sk
->sk_write_space
= sock_def_write_space
;
2445 sk
->sk_error_report
= sock_def_error_report
;
2446 sk
->sk_destruct
= sock_def_destruct
;
2448 sk
->sk_frag
.page
= NULL
;
2449 sk
->sk_frag
.offset
= 0;
2450 sk
->sk_peek_off
= -1;
2452 sk
->sk_peer_pid
= NULL
;
2453 sk
->sk_peer_cred
= NULL
;
2454 sk
->sk_write_pending
= 0;
2455 sk
->sk_rcvlowat
= 1;
2456 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2457 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2459 sk
->sk_stamp
= ktime_set(-1L, 0);
2461 #ifdef CONFIG_NET_RX_BUSY_POLL
2463 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2466 sk
->sk_max_pacing_rate
= ~0U;
2467 sk
->sk_pacing_rate
= ~0U;
2468 sk
->sk_incoming_cpu
= -1;
2470 * Before updating sk_refcnt, we must commit prior changes to memory
2471 * (Documentation/RCU/rculist_nulls.txt for details)
2474 atomic_set(&sk
->sk_refcnt
, 1);
2475 atomic_set(&sk
->sk_drops
, 0);
2477 EXPORT_SYMBOL(sock_init_data
);
2479 void lock_sock_nested(struct sock
*sk
, int subclass
)
2482 spin_lock_bh(&sk
->sk_lock
.slock
);
2483 if (sk
->sk_lock
.owned
)
2485 sk
->sk_lock
.owned
= 1;
2486 spin_unlock(&sk
->sk_lock
.slock
);
2488 * The sk_lock has mutex_lock() semantics here:
2490 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2493 EXPORT_SYMBOL(lock_sock_nested
);
2495 void release_sock(struct sock
*sk
)
2497 spin_lock_bh(&sk
->sk_lock
.slock
);
2498 if (sk
->sk_backlog
.tail
)
2501 /* Warning : release_cb() might need to release sk ownership,
2502 * ie call sock_release_ownership(sk) before us.
2504 if (sk
->sk_prot
->release_cb
)
2505 sk
->sk_prot
->release_cb(sk
);
2507 sock_release_ownership(sk
);
2508 if (waitqueue_active(&sk
->sk_lock
.wq
))
2509 wake_up(&sk
->sk_lock
.wq
);
2510 spin_unlock_bh(&sk
->sk_lock
.slock
);
2512 EXPORT_SYMBOL(release_sock
);
2515 * lock_sock_fast - fast version of lock_sock
2518 * This version should be used for very small section, where process wont block
2519 * return false if fast path is taken
2520 * sk_lock.slock locked, owned = 0, BH disabled
2521 * return true if slow path is taken
2522 * sk_lock.slock unlocked, owned = 1, BH enabled
2524 bool lock_sock_fast(struct sock
*sk
)
2527 spin_lock_bh(&sk
->sk_lock
.slock
);
2529 if (!sk
->sk_lock
.owned
)
2531 * Note : We must disable BH
2536 sk
->sk_lock
.owned
= 1;
2537 spin_unlock(&sk
->sk_lock
.slock
);
2539 * The sk_lock has mutex_lock() semantics here:
2541 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2545 EXPORT_SYMBOL(lock_sock_fast
);
2547 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2550 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2551 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2552 tv
= ktime_to_timeval(sk
->sk_stamp
);
2553 if (tv
.tv_sec
== -1)
2555 if (tv
.tv_sec
== 0) {
2556 sk
->sk_stamp
= ktime_get_real();
2557 tv
= ktime_to_timeval(sk
->sk_stamp
);
2559 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2561 EXPORT_SYMBOL(sock_get_timestamp
);
2563 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
2566 if (!sock_flag(sk
, SOCK_TIMESTAMP
))
2567 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2568 ts
= ktime_to_timespec(sk
->sk_stamp
);
2569 if (ts
.tv_sec
== -1)
2571 if (ts
.tv_sec
== 0) {
2572 sk
->sk_stamp
= ktime_get_real();
2573 ts
= ktime_to_timespec(sk
->sk_stamp
);
2575 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
2577 EXPORT_SYMBOL(sock_get_timestampns
);
2579 void sock_enable_timestamp(struct sock
*sk
, int flag
)
2581 if (!sock_flag(sk
, flag
)) {
2582 unsigned long previous_flags
= sk
->sk_flags
;
2584 sock_set_flag(sk
, flag
);
2586 * we just set one of the two flags which require net
2587 * time stamping, but time stamping might have been on
2588 * already because of the other one
2590 if (sock_needs_netstamp(sk
) &&
2591 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
2592 net_enable_timestamp();
2596 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
2597 int level
, int type
)
2599 struct sock_exterr_skb
*serr
;
2600 struct sk_buff
*skb
;
2604 skb
= sock_dequeue_err_skb(sk
);
2610 msg
->msg_flags
|= MSG_TRUNC
;
2613 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
2617 sock_recv_timestamp(msg
, sk
, skb
);
2619 serr
= SKB_EXT_ERR(skb
);
2620 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
2622 msg
->msg_flags
|= MSG_ERRQUEUE
;
2630 EXPORT_SYMBOL(sock_recv_errqueue
);
2633 * Get a socket option on an socket.
2635 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2636 * asynchronous errors should be reported by getsockopt. We assume
2637 * this means if you specify SO_ERROR (otherwise whats the point of it).
2639 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2640 char __user
*optval
, int __user
*optlen
)
2642 struct sock
*sk
= sock
->sk
;
2644 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2646 EXPORT_SYMBOL(sock_common_getsockopt
);
2648 #ifdef CONFIG_COMPAT
2649 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
2650 char __user
*optval
, int __user
*optlen
)
2652 struct sock
*sk
= sock
->sk
;
2654 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
2655 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
2657 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
2659 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
2662 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
2665 struct sock
*sk
= sock
->sk
;
2669 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
2670 flags
& ~MSG_DONTWAIT
, &addr_len
);
2672 msg
->msg_namelen
= addr_len
;
2675 EXPORT_SYMBOL(sock_common_recvmsg
);
2678 * Set socket options on an inet socket.
2680 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2681 char __user
*optval
, unsigned int optlen
)
2683 struct sock
*sk
= sock
->sk
;
2685 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2687 EXPORT_SYMBOL(sock_common_setsockopt
);
2689 #ifdef CONFIG_COMPAT
2690 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
2691 char __user
*optval
, unsigned int optlen
)
2693 struct sock
*sk
= sock
->sk
;
2695 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
2696 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
2698 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
2700 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
2703 void sk_common_release(struct sock
*sk
)
2705 if (sk
->sk_prot
->destroy
)
2706 sk
->sk_prot
->destroy(sk
);
2709 * Observation: when sock_common_release is called, processes have
2710 * no access to socket. But net still has.
2711 * Step one, detach it from networking:
2713 * A. Remove from hash tables.
2716 sk
->sk_prot
->unhash(sk
);
2719 * In this point socket cannot receive new packets, but it is possible
2720 * that some packets are in flight because some CPU runs receiver and
2721 * did hash table lookup before we unhashed socket. They will achieve
2722 * receive queue and will be purged by socket destructor.
2724 * Also we still have packets pending on receive queue and probably,
2725 * our own packets waiting in device queues. sock_destroy will drain
2726 * receive queue, but transmitted packets will delay socket destruction
2727 * until the last reference will be released.
2732 xfrm_sk_free_policy(sk
);
2734 sk_refcnt_debug_release(sk
);
2736 if (sk
->sk_frag
.page
) {
2737 put_page(sk
->sk_frag
.page
);
2738 sk
->sk_frag
.page
= NULL
;
2743 EXPORT_SYMBOL(sk_common_release
);
2745 #ifdef CONFIG_PROC_FS
2746 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2748 int val
[PROTO_INUSE_NR
];
2751 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
2753 #ifdef CONFIG_NET_NS
2754 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2756 __this_cpu_add(net
->core
.inuse
->val
[prot
->inuse_idx
], val
);
2758 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2760 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2762 int cpu
, idx
= prot
->inuse_idx
;
2765 for_each_possible_cpu(cpu
)
2766 res
+= per_cpu_ptr(net
->core
.inuse
, cpu
)->val
[idx
];
2768 return res
>= 0 ? res
: 0;
2770 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2772 static int __net_init
sock_inuse_init_net(struct net
*net
)
2774 net
->core
.inuse
= alloc_percpu(struct prot_inuse
);
2775 return net
->core
.inuse
? 0 : -ENOMEM
;
2778 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
2780 free_percpu(net
->core
.inuse
);
2783 static struct pernet_operations net_inuse_ops
= {
2784 .init
= sock_inuse_init_net
,
2785 .exit
= sock_inuse_exit_net
,
2788 static __init
int net_inuse_init(void)
2790 if (register_pernet_subsys(&net_inuse_ops
))
2791 panic("Cannot initialize net inuse counters");
2796 core_initcall(net_inuse_init
);
2798 static DEFINE_PER_CPU(struct prot_inuse
, prot_inuse
);
2800 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
2802 __this_cpu_add(prot_inuse
.val
[prot
->inuse_idx
], val
);
2804 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
2806 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
2808 int cpu
, idx
= prot
->inuse_idx
;
2811 for_each_possible_cpu(cpu
)
2812 res
+= per_cpu(prot_inuse
, cpu
).val
[idx
];
2814 return res
>= 0 ? res
: 0;
2816 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
2819 static void assign_proto_idx(struct proto
*prot
)
2821 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
2823 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
2824 pr_err("PROTO_INUSE_NR exhausted\n");
2828 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
2831 static void release_proto_idx(struct proto
*prot
)
2833 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
2834 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
2837 static inline void assign_proto_idx(struct proto
*prot
)
2841 static inline void release_proto_idx(struct proto
*prot
)
2846 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
2850 kfree(rsk_prot
->slab_name
);
2851 rsk_prot
->slab_name
= NULL
;
2852 kmem_cache_destroy(rsk_prot
->slab
);
2853 rsk_prot
->slab
= NULL
;
2856 static int req_prot_init(const struct proto
*prot
)
2858 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
2863 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
2865 if (!rsk_prot
->slab_name
)
2868 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
2869 rsk_prot
->obj_size
, 0,
2870 prot
->slab_flags
, NULL
);
2872 if (!rsk_prot
->slab
) {
2873 pr_crit("%s: Can't create request sock SLAB cache!\n",
2880 int proto_register(struct proto
*prot
, int alloc_slab
)
2883 prot
->slab
= kmem_cache_create(prot
->name
, prot
->obj_size
, 0,
2884 SLAB_HWCACHE_ALIGN
| prot
->slab_flags
,
2887 if (prot
->slab
== NULL
) {
2888 pr_crit("%s: Can't create sock SLAB cache!\n",
2893 if (req_prot_init(prot
))
2894 goto out_free_request_sock_slab
;
2896 if (prot
->twsk_prot
!= NULL
) {
2897 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
2899 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
2900 goto out_free_request_sock_slab
;
2902 prot
->twsk_prot
->twsk_slab
=
2903 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
2904 prot
->twsk_prot
->twsk_obj_size
,
2908 if (prot
->twsk_prot
->twsk_slab
== NULL
)
2909 goto out_free_timewait_sock_slab_name
;
2913 mutex_lock(&proto_list_mutex
);
2914 list_add(&prot
->node
, &proto_list
);
2915 assign_proto_idx(prot
);
2916 mutex_unlock(&proto_list_mutex
);
2919 out_free_timewait_sock_slab_name
:
2920 kfree(prot
->twsk_prot
->twsk_slab_name
);
2921 out_free_request_sock_slab
:
2922 req_prot_cleanup(prot
->rsk_prot
);
2924 kmem_cache_destroy(prot
->slab
);
2929 EXPORT_SYMBOL(proto_register
);
2931 void proto_unregister(struct proto
*prot
)
2933 mutex_lock(&proto_list_mutex
);
2934 release_proto_idx(prot
);
2935 list_del(&prot
->node
);
2936 mutex_unlock(&proto_list_mutex
);
2938 kmem_cache_destroy(prot
->slab
);
2941 req_prot_cleanup(prot
->rsk_prot
);
2943 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
2944 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
2945 kfree(prot
->twsk_prot
->twsk_slab_name
);
2946 prot
->twsk_prot
->twsk_slab
= NULL
;
2949 EXPORT_SYMBOL(proto_unregister
);
2951 #ifdef CONFIG_PROC_FS
2952 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2953 __acquires(proto_list_mutex
)
2955 mutex_lock(&proto_list_mutex
);
2956 return seq_list_start_head(&proto_list
, *pos
);
2959 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2961 return seq_list_next(v
, &proto_list
, pos
);
2964 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
2965 __releases(proto_list_mutex
)
2967 mutex_unlock(&proto_list_mutex
);
2970 static char proto_method_implemented(const void *method
)
2972 return method
== NULL
? 'n' : 'y';
2974 static long sock_prot_memory_allocated(struct proto
*proto
)
2976 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
2979 static char *sock_prot_memory_pressure(struct proto
*proto
)
2981 return proto
->memory_pressure
!= NULL
?
2982 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
2985 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
2988 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2989 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2992 sock_prot_inuse_get(seq_file_net(seq
), proto
),
2993 sock_prot_memory_allocated(proto
),
2994 sock_prot_memory_pressure(proto
),
2996 proto
->slab
== NULL
? "no" : "yes",
2997 module_name(proto
->owner
),
2998 proto_method_implemented(proto
->close
),
2999 proto_method_implemented(proto
->connect
),
3000 proto_method_implemented(proto
->disconnect
),
3001 proto_method_implemented(proto
->accept
),
3002 proto_method_implemented(proto
->ioctl
),
3003 proto_method_implemented(proto
->init
),
3004 proto_method_implemented(proto
->destroy
),
3005 proto_method_implemented(proto
->shutdown
),
3006 proto_method_implemented(proto
->setsockopt
),
3007 proto_method_implemented(proto
->getsockopt
),
3008 proto_method_implemented(proto
->sendmsg
),
3009 proto_method_implemented(proto
->recvmsg
),
3010 proto_method_implemented(proto
->sendpage
),
3011 proto_method_implemented(proto
->bind
),
3012 proto_method_implemented(proto
->backlog_rcv
),
3013 proto_method_implemented(proto
->hash
),
3014 proto_method_implemented(proto
->unhash
),
3015 proto_method_implemented(proto
->get_port
),
3016 proto_method_implemented(proto
->enter_memory_pressure
));
3019 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3021 if (v
== &proto_list
)
3022 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3031 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3033 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3037 static const struct seq_operations proto_seq_ops
= {
3038 .start
= proto_seq_start
,
3039 .next
= proto_seq_next
,
3040 .stop
= proto_seq_stop
,
3041 .show
= proto_seq_show
,
3044 static int proto_seq_open(struct inode
*inode
, struct file
*file
)
3046 return seq_open_net(inode
, file
, &proto_seq_ops
,
3047 sizeof(struct seq_net_private
));
3050 static const struct file_operations proto_seq_fops
= {
3051 .owner
= THIS_MODULE
,
3052 .open
= proto_seq_open
,
3054 .llseek
= seq_lseek
,
3055 .release
= seq_release_net
,
3058 static __net_init
int proto_init_net(struct net
*net
)
3060 if (!proc_create("protocols", S_IRUGO
, net
->proc_net
, &proto_seq_fops
))
3066 static __net_exit
void proto_exit_net(struct net
*net
)
3068 remove_proc_entry("protocols", net
->proc_net
);
3072 static __net_initdata
struct pernet_operations proto_net_ops
= {
3073 .init
= proto_init_net
,
3074 .exit
= proto_exit_net
,
3077 static int __init
proto_init(void)
3079 return register_pernet_subsys(&proto_net_ops
);
3082 subsys_initcall(proto_init
);
3084 #endif /* PROC_FS */