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 <asm/unaligned.h>
95 #include <linux/capability.h>
96 #include <linux/errno.h>
97 #include <linux/errqueue.h>
98 #include <linux/types.h>
99 #include <linux/socket.h>
100 #include <linux/in.h>
101 #include <linux/kernel.h>
102 #include <linux/module.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <linux/sched.h>
106 #include <linux/sched/mm.h>
107 #include <linux/timer.h>
108 #include <linux/string.h>
109 #include <linux/sockios.h>
110 #include <linux/net.h>
111 #include <linux/mm.h>
112 #include <linux/slab.h>
113 #include <linux/interrupt.h>
114 #include <linux/poll.h>
115 #include <linux/tcp.h>
116 #include <linux/init.h>
117 #include <linux/highmem.h>
118 #include <linux/user_namespace.h>
119 #include <linux/static_key.h>
120 #include <linux/memcontrol.h>
121 #include <linux/prefetch.h>
123 #include <linux/uaccess.h>
125 #include <linux/netdevice.h>
126 #include <net/protocol.h>
127 #include <linux/skbuff.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <linux/net_tstamp.h>
132 #include <net/xfrm.h>
133 #include <linux/ipsec.h>
134 #include <net/cls_cgroup.h>
135 #include <net/netprio_cgroup.h>
136 #include <linux/sock_diag.h>
138 #include <linux/filter.h>
139 #include <net/sock_reuseport.h>
141 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
146 static DEFINE_MUTEX(proto_list_mutex
);
147 static LIST_HEAD(proto_list
);
149 static void sock_inuse_add(struct net
*net
, int val
);
152 * sk_ns_capable - General socket capability test
153 * @sk: Socket to use a capability on or through
154 * @user_ns: The user namespace of the capability to use
155 * @cap: The capability to use
157 * Test to see if the opener of the socket had when the socket was
158 * created and the current process has the capability @cap in the user
159 * namespace @user_ns.
161 bool sk_ns_capable(const struct sock
*sk
,
162 struct user_namespace
*user_ns
, int cap
)
164 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
165 ns_capable(user_ns
, cap
);
167 EXPORT_SYMBOL(sk_ns_capable
);
170 * sk_capable - Socket global capability test
171 * @sk: Socket to use a capability on or through
172 * @cap: The global capability to use
174 * Test to see if the opener of the socket had when the socket was
175 * created and the current process has the capability @cap in all user
178 bool sk_capable(const struct sock
*sk
, int cap
)
180 return sk_ns_capable(sk
, &init_user_ns
, cap
);
182 EXPORT_SYMBOL(sk_capable
);
185 * sk_net_capable - Network namespace socket capability test
186 * @sk: Socket to use a capability on or through
187 * @cap: The capability to use
189 * Test to see if the opener of the socket had when the socket was created
190 * and the current process has the capability @cap over the network namespace
191 * the socket is a member of.
193 bool sk_net_capable(const struct sock
*sk
, int cap
)
195 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
197 EXPORT_SYMBOL(sk_net_capable
);
200 * Each address family might have different locking rules, so we have
201 * one slock key per address family and separate keys for internal and
204 static struct lock_class_key af_family_keys
[AF_MAX
];
205 static struct lock_class_key af_family_kern_keys
[AF_MAX
];
206 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
207 static struct lock_class_key af_family_kern_slock_keys
[AF_MAX
];
210 * Make lock validator output more readable. (we pre-construct these
211 * strings build-time, so that runtime initialization of socket
215 #define _sock_locks(x) \
216 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
217 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
218 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
219 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
220 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
221 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
222 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
223 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
224 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
225 x "27" , x "28" , x "AF_CAN" , \
226 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
227 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
228 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
229 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
230 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
233 static const char *const af_family_key_strings
[AF_MAX
+1] = {
234 _sock_locks("sk_lock-")
236 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
237 _sock_locks("slock-")
239 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
240 _sock_locks("clock-")
243 static const char *const af_family_kern_key_strings
[AF_MAX
+1] = {
244 _sock_locks("k-sk_lock-")
246 static const char *const af_family_kern_slock_key_strings
[AF_MAX
+1] = {
247 _sock_locks("k-slock-")
249 static const char *const af_family_kern_clock_key_strings
[AF_MAX
+1] = {
250 _sock_locks("k-clock-")
252 static const char *const af_family_rlock_key_strings
[AF_MAX
+1] = {
253 _sock_locks("rlock-")
255 static const char *const af_family_wlock_key_strings
[AF_MAX
+1] = {
256 _sock_locks("wlock-")
258 static const char *const af_family_elock_key_strings
[AF_MAX
+1] = {
259 _sock_locks("elock-")
263 * sk_callback_lock and sk queues 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
];
267 static struct lock_class_key af_rlock_keys
[AF_MAX
];
268 static struct lock_class_key af_wlock_keys
[AF_MAX
];
269 static struct lock_class_key af_elock_keys
[AF_MAX
];
270 static struct lock_class_key af_kern_callback_keys
[AF_MAX
];
272 /* Run time adjustable parameters. */
273 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
274 EXPORT_SYMBOL(sysctl_wmem_max
);
275 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
276 EXPORT_SYMBOL(sysctl_rmem_max
);
277 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
278 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
280 /* Maximal space eaten by iovec or ancillary data plus some space */
281 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
282 EXPORT_SYMBOL(sysctl_optmem_max
);
284 int sysctl_tstamp_allow_data __read_mostly
= 1;
286 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key
);
287 EXPORT_SYMBOL_GPL(memalloc_socks_key
);
290 * sk_set_memalloc - sets %SOCK_MEMALLOC
291 * @sk: socket to set it on
293 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
294 * It's the responsibility of the admin to adjust min_free_kbytes
295 * to meet the requirements
297 void sk_set_memalloc(struct sock
*sk
)
299 sock_set_flag(sk
, SOCK_MEMALLOC
);
300 sk
->sk_allocation
|= __GFP_MEMALLOC
;
301 static_branch_inc(&memalloc_socks_key
);
303 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
305 void sk_clear_memalloc(struct sock
*sk
)
307 sock_reset_flag(sk
, SOCK_MEMALLOC
);
308 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
309 static_branch_dec(&memalloc_socks_key
);
312 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
313 * progress of swapping. SOCK_MEMALLOC may be cleared while
314 * it has rmem allocations due to the last swapfile being deactivated
315 * but there is a risk that the socket is unusable due to exceeding
316 * the rmem limits. Reclaim the reserves and obey rmem limits again.
320 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
322 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
325 unsigned int noreclaim_flag
;
327 /* these should have been dropped before queueing */
328 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
330 noreclaim_flag
= memalloc_noreclaim_save();
331 ret
= sk
->sk_backlog_rcv(sk
, skb
);
332 memalloc_noreclaim_restore(noreclaim_flag
);
336 EXPORT_SYMBOL(__sk_backlog_rcv
);
338 static int sock_get_timeout(long timeo
, void *optval
, bool old_timeval
)
340 struct __kernel_sock_timeval tv
;
343 if (timeo
== MAX_SCHEDULE_TIMEOUT
) {
347 tv
.tv_sec
= timeo
/ HZ
;
348 tv
.tv_usec
= ((timeo
% HZ
) * USEC_PER_SEC
) / HZ
;
351 if (in_compat_syscall() && !COMPAT_USE_64BIT_TIME
) {
352 struct old_timeval32 tv32
= { tv
.tv_sec
, tv
.tv_usec
};
353 *(struct old_timeval32
*)optval
= tv32
;
358 struct __kernel_old_timeval old_tv
;
359 old_tv
.tv_sec
= tv
.tv_sec
;
360 old_tv
.tv_usec
= tv
.tv_usec
;
361 *(struct __kernel_old_timeval
*)optval
= old_tv
;
362 size
= sizeof(old_tv
);
364 *(struct __kernel_sock_timeval
*)optval
= tv
;
371 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
, bool old_timeval
)
373 struct __kernel_sock_timeval tv
;
375 if (in_compat_syscall() && !COMPAT_USE_64BIT_TIME
) {
376 struct old_timeval32 tv32
;
378 if (optlen
< sizeof(tv32
))
381 if (copy_from_user(&tv32
, optval
, sizeof(tv32
)))
383 tv
.tv_sec
= tv32
.tv_sec
;
384 tv
.tv_usec
= tv32
.tv_usec
;
385 } else if (old_timeval
) {
386 struct __kernel_old_timeval old_tv
;
388 if (optlen
< sizeof(old_tv
))
390 if (copy_from_user(&old_tv
, optval
, sizeof(old_tv
)))
392 tv
.tv_sec
= old_tv
.tv_sec
;
393 tv
.tv_usec
= old_tv
.tv_usec
;
395 if (optlen
< sizeof(tv
))
397 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
400 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
404 static int warned __read_mostly
;
407 if (warned
< 10 && net_ratelimit()) {
409 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
410 __func__
, current
->comm
, task_pid_nr(current
));
414 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
415 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
417 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/ HZ
- 1))
418 *timeo_p
= tv
.tv_sec
* HZ
+ DIV_ROUND_UP((unsigned long)tv
.tv_usec
, USEC_PER_SEC
/ HZ
);
422 static void sock_warn_obsolete_bsdism(const char *name
)
425 static char warncomm
[TASK_COMM_LEN
];
426 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
427 strcpy(warncomm
, current
->comm
);
428 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
434 static bool sock_needs_netstamp(const struct sock
*sk
)
436 switch (sk
->sk_family
) {
445 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
447 if (sk
->sk_flags
& flags
) {
448 sk
->sk_flags
&= ~flags
;
449 if (sock_needs_netstamp(sk
) &&
450 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
451 net_disable_timestamp();
456 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
459 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
461 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
462 atomic_inc(&sk
->sk_drops
);
463 trace_sock_rcvqueue_full(sk
, skb
);
467 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
468 atomic_inc(&sk
->sk_drops
);
473 skb_set_owner_r(skb
, sk
);
475 /* we escape from rcu protected region, make sure we dont leak
480 spin_lock_irqsave(&list
->lock
, flags
);
481 sock_skb_set_dropcount(sk
, skb
);
482 __skb_queue_tail(list
, skb
);
483 spin_unlock_irqrestore(&list
->lock
, flags
);
485 if (!sock_flag(sk
, SOCK_DEAD
))
486 sk
->sk_data_ready(sk
);
489 EXPORT_SYMBOL(__sock_queue_rcv_skb
);
491 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
495 err
= sk_filter(sk
, skb
);
499 return __sock_queue_rcv_skb(sk
, skb
);
501 EXPORT_SYMBOL(sock_queue_rcv_skb
);
503 int __sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
,
504 const int nested
, unsigned int trim_cap
, bool refcounted
)
506 int rc
= NET_RX_SUCCESS
;
508 if (sk_filter_trim_cap(sk
, skb
, trim_cap
))
509 goto discard_and_relse
;
513 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
514 atomic_inc(&sk
->sk_drops
);
515 goto discard_and_relse
;
518 bh_lock_sock_nested(sk
);
521 if (!sock_owned_by_user(sk
)) {
523 * trylock + unlock semantics:
525 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
527 rc
= sk_backlog_rcv(sk
, skb
);
529 mutex_release(&sk
->sk_lock
.dep_map
, 1, _RET_IP_
);
530 } else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
532 atomic_inc(&sk
->sk_drops
);
533 goto discard_and_relse
;
545 EXPORT_SYMBOL(__sk_receive_skb
);
547 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
549 struct dst_entry
*dst
= __sk_dst_get(sk
);
551 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
552 sk_tx_queue_clear(sk
);
553 sk
->sk_dst_pending_confirm
= 0;
554 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
561 EXPORT_SYMBOL(__sk_dst_check
);
563 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
565 struct dst_entry
*dst
= sk_dst_get(sk
);
567 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
575 EXPORT_SYMBOL(sk_dst_check
);
577 static int sock_setbindtodevice_locked(struct sock
*sk
, int ifindex
)
579 int ret
= -ENOPROTOOPT
;
580 #ifdef CONFIG_NETDEVICES
581 struct net
*net
= sock_net(sk
);
585 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
592 sk
->sk_bound_dev_if
= ifindex
;
593 if (sk
->sk_prot
->rehash
)
594 sk
->sk_prot
->rehash(sk
);
605 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
608 int ret
= -ENOPROTOOPT
;
609 #ifdef CONFIG_NETDEVICES
610 struct net
*net
= sock_net(sk
);
611 char devname
[IFNAMSIZ
];
618 /* Bind this socket to a particular device like "eth0",
619 * as specified in the passed interface name. If the
620 * name is "" or the option length is zero the socket
623 if (optlen
> IFNAMSIZ
- 1)
624 optlen
= IFNAMSIZ
- 1;
625 memset(devname
, 0, sizeof(devname
));
628 if (copy_from_user(devname
, optval
, optlen
))
632 if (devname
[0] != '\0') {
633 struct net_device
*dev
;
636 dev
= dev_get_by_name_rcu(net
, devname
);
638 index
= dev
->ifindex
;
646 ret
= sock_setbindtodevice_locked(sk
, index
);
655 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
656 int __user
*optlen
, int len
)
658 int ret
= -ENOPROTOOPT
;
659 #ifdef CONFIG_NETDEVICES
660 struct net
*net
= sock_net(sk
);
661 char devname
[IFNAMSIZ
];
663 if (sk
->sk_bound_dev_if
== 0) {
672 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
676 len
= strlen(devname
) + 1;
679 if (copy_to_user(optval
, devname
, len
))
684 if (put_user(len
, optlen
))
695 static inline void sock_valbool_flag(struct sock
*sk
, int bit
, int valbool
)
698 sock_set_flag(sk
, bit
);
700 sock_reset_flag(sk
, bit
);
703 bool sk_mc_loop(struct sock
*sk
)
705 if (dev_recursion_level())
709 switch (sk
->sk_family
) {
711 return inet_sk(sk
)->mc_loop
;
712 #if IS_ENABLED(CONFIG_IPV6)
714 return inet6_sk(sk
)->mc_loop
;
720 EXPORT_SYMBOL(sk_mc_loop
);
723 * This is meant for all protocols to use and covers goings on
724 * at the socket level. Everything here is generic.
727 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
728 char __user
*optval
, unsigned int optlen
)
730 struct sock_txtime sk_txtime
;
731 struct sock
*sk
= sock
->sk
;
738 * Options without arguments
741 if (optname
== SO_BINDTODEVICE
)
742 return sock_setbindtodevice(sk
, optval
, optlen
);
744 if (optlen
< sizeof(int))
747 if (get_user(val
, (int __user
*)optval
))
750 valbool
= val
? 1 : 0;
756 if (val
&& !capable(CAP_NET_ADMIN
))
759 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
762 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
765 sk
->sk_reuseport
= valbool
;
774 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
778 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
781 /* Don't error on this BSD doesn't and if you think
782 * about it this is right. Otherwise apps have to
783 * play 'guess the biggest size' games. RCVBUF/SNDBUF
784 * are treated in BSD as hints
786 val
= min_t(u32
, val
, sysctl_wmem_max
);
788 /* Ensure val * 2 fits into an int, to prevent max_t()
789 * from treating it as a negative value.
791 val
= min_t(int, val
, INT_MAX
/ 2);
792 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
793 sk
->sk_sndbuf
= max_t(int, val
* 2, SOCK_MIN_SNDBUF
);
794 /* Wake up sending tasks if we upped the value. */
795 sk
->sk_write_space(sk
);
799 if (!capable(CAP_NET_ADMIN
)) {
804 /* No negative values (to prevent underflow, as val will be
812 /* Don't error on this BSD doesn't and if you think
813 * about it this is right. Otherwise apps have to
814 * play 'guess the biggest size' games. RCVBUF/SNDBUF
815 * are treated in BSD as hints
817 val
= min_t(u32
, val
, sysctl_rmem_max
);
819 /* Ensure val * 2 fits into an int, to prevent max_t()
820 * from treating it as a negative value.
822 val
= min_t(int, val
, INT_MAX
/ 2);
823 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
825 * We double it on the way in to account for
826 * "struct sk_buff" etc. overhead. Applications
827 * assume that the SO_RCVBUF setting they make will
828 * allow that much actual data to be received on that
831 * Applications are unaware that "struct sk_buff" and
832 * other overheads allocate from the receive buffer
833 * during socket buffer allocation.
835 * And after considering the possible alternatives,
836 * returning the value we actually used in getsockopt
837 * is the most desirable behavior.
839 sk
->sk_rcvbuf
= max_t(int, val
* 2, SOCK_MIN_RCVBUF
);
843 if (!capable(CAP_NET_ADMIN
)) {
848 /* No negative values (to prevent underflow, as val will be
856 if (sk
->sk_prot
->keepalive
)
857 sk
->sk_prot
->keepalive(sk
, valbool
);
858 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
862 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
866 sk
->sk_no_check_tx
= valbool
;
870 if ((val
>= 0 && val
<= 6) ||
871 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
872 sk
->sk_priority
= val
;
878 if (optlen
< sizeof(ling
)) {
879 ret
= -EINVAL
; /* 1003.1g */
882 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
887 sock_reset_flag(sk
, SOCK_LINGER
);
889 #if (BITS_PER_LONG == 32)
890 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
891 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
894 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
895 sock_set_flag(sk
, SOCK_LINGER
);
900 sock_warn_obsolete_bsdism("setsockopt");
905 set_bit(SOCK_PASSCRED
, &sock
->flags
);
907 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
910 case SO_TIMESTAMP_OLD
:
911 case SO_TIMESTAMP_NEW
:
912 case SO_TIMESTAMPNS_OLD
:
913 case SO_TIMESTAMPNS_NEW
:
915 if (optname
== SO_TIMESTAMP_NEW
|| optname
== SO_TIMESTAMPNS_NEW
)
916 sock_set_flag(sk
, SOCK_TSTAMP_NEW
);
918 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
920 if (optname
== SO_TIMESTAMP_OLD
|| optname
== SO_TIMESTAMP_NEW
)
921 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
923 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
924 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
925 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
927 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
928 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
929 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
933 case SO_TIMESTAMPING_NEW
:
934 sock_set_flag(sk
, SOCK_TSTAMP_NEW
);
936 case SO_TIMESTAMPING_OLD
:
937 if (val
& ~SOF_TIMESTAMPING_MASK
) {
942 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
943 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
944 if (sk
->sk_protocol
== IPPROTO_TCP
&&
945 sk
->sk_type
== SOCK_STREAM
) {
946 if ((1 << sk
->sk_state
) &
947 (TCPF_CLOSE
| TCPF_LISTEN
)) {
951 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
957 if (val
& SOF_TIMESTAMPING_OPT_STATS
&&
958 !(val
& SOF_TIMESTAMPING_OPT_TSONLY
)) {
963 sk
->sk_tsflags
= val
;
964 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
965 sock_enable_timestamp(sk
,
966 SOCK_TIMESTAMPING_RX_SOFTWARE
);
968 if (optname
== SO_TIMESTAMPING_NEW
)
969 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
971 sock_disable_timestamp(sk
,
972 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
979 if (sock
->ops
->set_rcvlowat
)
980 ret
= sock
->ops
->set_rcvlowat(sk
, val
);
982 sk
->sk_rcvlowat
= val
? : 1;
985 case SO_RCVTIMEO_OLD
:
986 case SO_RCVTIMEO_NEW
:
987 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
, optname
== SO_RCVTIMEO_OLD
);
990 case SO_SNDTIMEO_OLD
:
991 case SO_SNDTIMEO_NEW
:
992 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
, optname
== SO_SNDTIMEO_OLD
);
995 case SO_ATTACH_FILTER
:
997 if (optlen
== sizeof(struct sock_fprog
)) {
998 struct sock_fprog fprog
;
1001 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
1004 ret
= sk_attach_filter(&fprog
, sk
);
1010 if (optlen
== sizeof(u32
)) {
1014 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
1017 ret
= sk_attach_bpf(ufd
, sk
);
1021 case SO_ATTACH_REUSEPORT_CBPF
:
1023 if (optlen
== sizeof(struct sock_fprog
)) {
1024 struct sock_fprog fprog
;
1027 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
1030 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
1034 case SO_ATTACH_REUSEPORT_EBPF
:
1036 if (optlen
== sizeof(u32
)) {
1040 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
1043 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
1047 case SO_DETACH_FILTER
:
1048 ret
= sk_detach_filter(sk
);
1051 case SO_LOCK_FILTER
:
1052 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
1055 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
1060 set_bit(SOCK_PASSSEC
, &sock
->flags
);
1062 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
1065 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
)) {
1067 } else if (val
!= sk
->sk_mark
) {
1074 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
1077 case SO_WIFI_STATUS
:
1078 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
1082 if (sock
->ops
->set_peek_off
)
1083 ret
= sock
->ops
->set_peek_off(sk
, val
);
1089 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
1092 case SO_SELECT_ERR_QUEUE
:
1093 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
1096 #ifdef CONFIG_NET_RX_BUSY_POLL
1098 /* allow unprivileged users to decrease the value */
1099 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
1105 sk
->sk_ll_usec
= val
;
1110 case SO_MAX_PACING_RATE
:
1112 unsigned long ulval
= (val
== ~0U) ? ~0UL : val
;
1114 if (sizeof(ulval
) != sizeof(val
) &&
1115 optlen
>= sizeof(ulval
) &&
1116 get_user(ulval
, (unsigned long __user
*)optval
)) {
1121 cmpxchg(&sk
->sk_pacing_status
,
1124 sk
->sk_max_pacing_rate
= ulval
;
1125 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
, ulval
);
1128 case SO_INCOMING_CPU
:
1129 sk
->sk_incoming_cpu
= val
;
1134 dst_negative_advice(sk
);
1138 if (sk
->sk_family
== PF_INET
|| sk
->sk_family
== PF_INET6
) {
1139 if (!((sk
->sk_type
== SOCK_STREAM
&&
1140 sk
->sk_protocol
== IPPROTO_TCP
) ||
1141 (sk
->sk_type
== SOCK_DGRAM
&&
1142 sk
->sk_protocol
== IPPROTO_UDP
)))
1144 } else if (sk
->sk_family
!= PF_RDS
) {
1148 if (val
< 0 || val
> 1)
1151 sock_valbool_flag(sk
, SOCK_ZEROCOPY
, valbool
);
1156 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
)) {
1158 } else if (optlen
!= sizeof(struct sock_txtime
)) {
1160 } else if (copy_from_user(&sk_txtime
, optval
,
1161 sizeof(struct sock_txtime
))) {
1163 } else if (sk_txtime
.flags
& ~SOF_TXTIME_FLAGS_MASK
) {
1166 sock_valbool_flag(sk
, SOCK_TXTIME
, true);
1167 sk
->sk_clockid
= sk_txtime
.clockid
;
1168 sk
->sk_txtime_deadline_mode
=
1169 !!(sk_txtime
.flags
& SOF_TXTIME_DEADLINE_MODE
);
1170 sk
->sk_txtime_report_errors
=
1171 !!(sk_txtime
.flags
& SOF_TXTIME_REPORT_ERRORS
);
1175 case SO_BINDTOIFINDEX
:
1176 ret
= sock_setbindtodevice_locked(sk
, val
);
1186 EXPORT_SYMBOL(sock_setsockopt
);
1189 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1190 struct ucred
*ucred
)
1192 ucred
->pid
= pid_vnr(pid
);
1193 ucred
->uid
= ucred
->gid
= -1;
1195 struct user_namespace
*current_ns
= current_user_ns();
1197 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1198 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1202 static int groups_to_user(gid_t __user
*dst
, const struct group_info
*src
)
1204 struct user_namespace
*user_ns
= current_user_ns();
1207 for (i
= 0; i
< src
->ngroups
; i
++)
1208 if (put_user(from_kgid_munged(user_ns
, src
->gid
[i
]), dst
+ i
))
1214 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1215 char __user
*optval
, int __user
*optlen
)
1217 struct sock
*sk
= sock
->sk
;
1222 unsigned long ulval
;
1224 struct old_timeval32 tm32
;
1225 struct __kernel_old_timeval tm
;
1226 struct __kernel_sock_timeval stm
;
1227 struct sock_txtime txtime
;
1230 int lv
= sizeof(int);
1233 if (get_user(len
, optlen
))
1238 memset(&v
, 0, sizeof(v
));
1242 v
.val
= sock_flag(sk
, SOCK_DBG
);
1246 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1250 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1254 v
.val
= sk
->sk_sndbuf
;
1258 v
.val
= sk
->sk_rcvbuf
;
1262 v
.val
= sk
->sk_reuse
;
1266 v
.val
= sk
->sk_reuseport
;
1270 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1274 v
.val
= sk
->sk_type
;
1278 v
.val
= sk
->sk_protocol
;
1282 v
.val
= sk
->sk_family
;
1286 v
.val
= -sock_error(sk
);
1288 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1292 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1296 v
.val
= sk
->sk_no_check_tx
;
1300 v
.val
= sk
->sk_priority
;
1304 lv
= sizeof(v
.ling
);
1305 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1306 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1310 sock_warn_obsolete_bsdism("getsockopt");
1313 case SO_TIMESTAMP_OLD
:
1314 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1315 !sock_flag(sk
, SOCK_TSTAMP_NEW
) &&
1316 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1319 case SO_TIMESTAMPNS_OLD
:
1320 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
) && !sock_flag(sk
, SOCK_TSTAMP_NEW
);
1323 case SO_TIMESTAMP_NEW
:
1324 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) && sock_flag(sk
, SOCK_TSTAMP_NEW
);
1327 case SO_TIMESTAMPNS_NEW
:
1328 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
) && sock_flag(sk
, SOCK_TSTAMP_NEW
);
1331 case SO_TIMESTAMPING_OLD
:
1332 v
.val
= sk
->sk_tsflags
;
1335 case SO_RCVTIMEO_OLD
:
1336 case SO_RCVTIMEO_NEW
:
1337 lv
= sock_get_timeout(sk
->sk_rcvtimeo
, &v
, SO_RCVTIMEO_OLD
== optname
);
1340 case SO_SNDTIMEO_OLD
:
1341 case SO_SNDTIMEO_NEW
:
1342 lv
= sock_get_timeout(sk
->sk_sndtimeo
, &v
, SO_SNDTIMEO_OLD
== optname
);
1346 v
.val
= sk
->sk_rcvlowat
;
1354 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1359 struct ucred peercred
;
1360 if (len
> sizeof(peercred
))
1361 len
= sizeof(peercred
);
1362 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1363 if (copy_to_user(optval
, &peercred
, len
))
1372 if (!sk
->sk_peer_cred
)
1375 n
= sk
->sk_peer_cred
->group_info
->ngroups
;
1376 if (len
< n
* sizeof(gid_t
)) {
1377 len
= n
* sizeof(gid_t
);
1378 return put_user(len
, optlen
) ? -EFAULT
: -ERANGE
;
1380 len
= n
* sizeof(gid_t
);
1382 ret
= groups_to_user((gid_t __user
*)optval
,
1383 sk
->sk_peer_cred
->group_info
);
1393 lv
= sock
->ops
->getname(sock
, (struct sockaddr
*)address
, 2);
1398 if (copy_to_user(optval
, address
, len
))
1403 /* Dubious BSD thing... Probably nobody even uses it, but
1404 * the UNIX standard wants it for whatever reason... -DaveM
1407 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1411 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1415 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1418 v
.val
= sk
->sk_mark
;
1422 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1425 case SO_WIFI_STATUS
:
1426 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1430 if (!sock
->ops
->set_peek_off
)
1433 v
.val
= sk
->sk_peek_off
;
1436 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1439 case SO_BINDTODEVICE
:
1440 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1443 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1449 case SO_LOCK_FILTER
:
1450 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1453 case SO_BPF_EXTENSIONS
:
1454 v
.val
= bpf_tell_extensions();
1457 case SO_SELECT_ERR_QUEUE
:
1458 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1461 #ifdef CONFIG_NET_RX_BUSY_POLL
1463 v
.val
= sk
->sk_ll_usec
;
1467 case SO_MAX_PACING_RATE
:
1468 if (sizeof(v
.ulval
) != sizeof(v
.val
) && len
>= sizeof(v
.ulval
)) {
1469 lv
= sizeof(v
.ulval
);
1470 v
.ulval
= sk
->sk_max_pacing_rate
;
1473 v
.val
= min_t(unsigned long, sk
->sk_max_pacing_rate
, ~0U);
1477 case SO_INCOMING_CPU
:
1478 v
.val
= sk
->sk_incoming_cpu
;
1483 u32 meminfo
[SK_MEMINFO_VARS
];
1485 if (get_user(len
, optlen
))
1488 sk_get_meminfo(sk
, meminfo
);
1490 len
= min_t(unsigned int, len
, sizeof(meminfo
));
1491 if (copy_to_user(optval
, &meminfo
, len
))
1497 #ifdef CONFIG_NET_RX_BUSY_POLL
1498 case SO_INCOMING_NAPI_ID
:
1499 v
.val
= READ_ONCE(sk
->sk_napi_id
);
1501 /* aggregate non-NAPI IDs down to 0 */
1502 if (v
.val
< MIN_NAPI_ID
)
1512 v
.val64
= sock_gen_cookie(sk
);
1516 v
.val
= sock_flag(sk
, SOCK_ZEROCOPY
);
1520 lv
= sizeof(v
.txtime
);
1521 v
.txtime
.clockid
= sk
->sk_clockid
;
1522 v
.txtime
.flags
|= sk
->sk_txtime_deadline_mode
?
1523 SOF_TXTIME_DEADLINE_MODE
: 0;
1524 v
.txtime
.flags
|= sk
->sk_txtime_report_errors
?
1525 SOF_TXTIME_REPORT_ERRORS
: 0;
1528 case SO_BINDTOIFINDEX
:
1529 v
.val
= sk
->sk_bound_dev_if
;
1533 /* We implement the SO_SNDLOWAT etc to not be settable
1536 return -ENOPROTOOPT
;
1541 if (copy_to_user(optval
, &v
, len
))
1544 if (put_user(len
, optlen
))
1550 * Initialize an sk_lock.
1552 * (We also register the sk_lock with the lock validator.)
1554 static inline void sock_lock_init(struct sock
*sk
)
1556 if (sk
->sk_kern_sock
)
1557 sock_lock_init_class_and_name(
1559 af_family_kern_slock_key_strings
[sk
->sk_family
],
1560 af_family_kern_slock_keys
+ sk
->sk_family
,
1561 af_family_kern_key_strings
[sk
->sk_family
],
1562 af_family_kern_keys
+ sk
->sk_family
);
1564 sock_lock_init_class_and_name(
1566 af_family_slock_key_strings
[sk
->sk_family
],
1567 af_family_slock_keys
+ sk
->sk_family
,
1568 af_family_key_strings
[sk
->sk_family
],
1569 af_family_keys
+ sk
->sk_family
);
1573 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1574 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1575 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1577 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1579 #ifdef CONFIG_SECURITY_NETWORK
1580 void *sptr
= nsk
->sk_security
;
1582 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1584 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1585 osk
->sk_prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1587 #ifdef CONFIG_SECURITY_NETWORK
1588 nsk
->sk_security
= sptr
;
1589 security_sk_clone(osk
, nsk
);
1593 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1597 struct kmem_cache
*slab
;
1601 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1604 if (priority
& __GFP_ZERO
)
1605 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1607 sk
= kmalloc(prot
->obj_size
, priority
);
1610 if (security_sk_alloc(sk
, family
, priority
))
1613 if (!try_module_get(prot
->owner
))
1615 sk_tx_queue_clear(sk
);
1621 security_sk_free(sk
);
1624 kmem_cache_free(slab
, sk
);
1630 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1632 struct kmem_cache
*slab
;
1633 struct module
*owner
;
1635 owner
= prot
->owner
;
1638 cgroup_sk_free(&sk
->sk_cgrp_data
);
1639 mem_cgroup_sk_free(sk
);
1640 security_sk_free(sk
);
1642 kmem_cache_free(slab
, sk
);
1649 * sk_alloc - All socket objects are allocated here
1650 * @net: the applicable net namespace
1651 * @family: protocol family
1652 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1653 * @prot: struct proto associated with this new sock instance
1654 * @kern: is this to be a kernel socket?
1656 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1657 struct proto
*prot
, int kern
)
1661 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1663 sk
->sk_family
= family
;
1665 * See comment in struct sock definition to understand
1666 * why we need sk_prot_creator -acme
1668 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1669 sk
->sk_kern_sock
= kern
;
1671 sk
->sk_net_refcnt
= kern
? 0 : 1;
1672 if (likely(sk
->sk_net_refcnt
)) {
1674 sock_inuse_add(net
, 1);
1677 sock_net_set(sk
, net
);
1678 refcount_set(&sk
->sk_wmem_alloc
, 1);
1680 mem_cgroup_sk_alloc(sk
);
1681 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1682 sock_update_classid(&sk
->sk_cgrp_data
);
1683 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1688 EXPORT_SYMBOL(sk_alloc
);
1690 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1691 * grace period. This is the case for UDP sockets and TCP listeners.
1693 static void __sk_destruct(struct rcu_head
*head
)
1695 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1696 struct sk_filter
*filter
;
1698 if (sk
->sk_destruct
)
1699 sk
->sk_destruct(sk
);
1701 filter
= rcu_dereference_check(sk
->sk_filter
,
1702 refcount_read(&sk
->sk_wmem_alloc
) == 0);
1704 sk_filter_uncharge(sk
, filter
);
1705 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1707 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1708 reuseport_detach_sock(sk
);
1710 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1712 if (atomic_read(&sk
->sk_omem_alloc
))
1713 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1714 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1716 if (sk
->sk_frag
.page
) {
1717 put_page(sk
->sk_frag
.page
);
1718 sk
->sk_frag
.page
= NULL
;
1721 if (sk
->sk_peer_cred
)
1722 put_cred(sk
->sk_peer_cred
);
1723 put_pid(sk
->sk_peer_pid
);
1724 if (likely(sk
->sk_net_refcnt
))
1725 put_net(sock_net(sk
));
1726 sk_prot_free(sk
->sk_prot_creator
, sk
);
1729 void sk_destruct(struct sock
*sk
)
1731 if (sock_flag(sk
, SOCK_RCU_FREE
))
1732 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1734 __sk_destruct(&sk
->sk_rcu
);
1737 static void __sk_free(struct sock
*sk
)
1739 if (likely(sk
->sk_net_refcnt
))
1740 sock_inuse_add(sock_net(sk
), -1);
1742 if (unlikely(sk
->sk_net_refcnt
&& sock_diag_has_destroy_listeners(sk
)))
1743 sock_diag_broadcast_destroy(sk
);
1748 void sk_free(struct sock
*sk
)
1751 * We subtract one from sk_wmem_alloc and can know if
1752 * some packets are still in some tx queue.
1753 * If not null, sock_wfree() will call __sk_free(sk) later
1755 if (refcount_dec_and_test(&sk
->sk_wmem_alloc
))
1758 EXPORT_SYMBOL(sk_free
);
1760 static void sk_init_common(struct sock
*sk
)
1762 skb_queue_head_init(&sk
->sk_receive_queue
);
1763 skb_queue_head_init(&sk
->sk_write_queue
);
1764 skb_queue_head_init(&sk
->sk_error_queue
);
1766 rwlock_init(&sk
->sk_callback_lock
);
1767 lockdep_set_class_and_name(&sk
->sk_receive_queue
.lock
,
1768 af_rlock_keys
+ sk
->sk_family
,
1769 af_family_rlock_key_strings
[sk
->sk_family
]);
1770 lockdep_set_class_and_name(&sk
->sk_write_queue
.lock
,
1771 af_wlock_keys
+ sk
->sk_family
,
1772 af_family_wlock_key_strings
[sk
->sk_family
]);
1773 lockdep_set_class_and_name(&sk
->sk_error_queue
.lock
,
1774 af_elock_keys
+ sk
->sk_family
,
1775 af_family_elock_key_strings
[sk
->sk_family
]);
1776 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
1777 af_callback_keys
+ sk
->sk_family
,
1778 af_family_clock_key_strings
[sk
->sk_family
]);
1782 * sk_clone_lock - clone a socket, and lock its clone
1783 * @sk: the socket to clone
1784 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1786 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1788 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1791 bool is_charged
= true;
1793 newsk
= sk_prot_alloc(sk
->sk_prot
, priority
, sk
->sk_family
);
1794 if (newsk
!= NULL
) {
1795 struct sk_filter
*filter
;
1797 sock_copy(newsk
, sk
);
1799 newsk
->sk_prot_creator
= sk
->sk_prot
;
1802 if (likely(newsk
->sk_net_refcnt
))
1803 get_net(sock_net(newsk
));
1804 sk_node_init(&newsk
->sk_node
);
1805 sock_lock_init(newsk
);
1806 bh_lock_sock(newsk
);
1807 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1808 newsk
->sk_backlog
.len
= 0;
1810 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1812 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1814 refcount_set(&newsk
->sk_wmem_alloc
, 1);
1815 atomic_set(&newsk
->sk_omem_alloc
, 0);
1816 sk_init_common(newsk
);
1818 newsk
->sk_dst_cache
= NULL
;
1819 newsk
->sk_dst_pending_confirm
= 0;
1820 newsk
->sk_wmem_queued
= 0;
1821 newsk
->sk_forward_alloc
= 0;
1822 atomic_set(&newsk
->sk_drops
, 0);
1823 newsk
->sk_send_head
= NULL
;
1824 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1825 atomic_set(&newsk
->sk_zckey
, 0);
1827 sock_reset_flag(newsk
, SOCK_DONE
);
1828 mem_cgroup_sk_alloc(newsk
);
1829 cgroup_sk_alloc(&newsk
->sk_cgrp_data
);
1832 filter
= rcu_dereference(sk
->sk_filter
);
1834 /* though it's an empty new sock, the charging may fail
1835 * if sysctl_optmem_max was changed between creation of
1836 * original socket and cloning
1838 is_charged
= sk_filter_charge(newsk
, filter
);
1839 RCU_INIT_POINTER(newsk
->sk_filter
, filter
);
1842 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1843 /* We need to make sure that we don't uncharge the new
1844 * socket if we couldn't charge it in the first place
1845 * as otherwise we uncharge the parent's filter.
1848 RCU_INIT_POINTER(newsk
->sk_filter
, NULL
);
1849 sk_free_unlock_clone(newsk
);
1853 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1856 newsk
->sk_err_soft
= 0;
1857 newsk
->sk_priority
= 0;
1858 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1859 if (likely(newsk
->sk_net_refcnt
))
1860 sock_inuse_add(sock_net(newsk
), 1);
1863 * Before updating sk_refcnt, we must commit prior changes to memory
1864 * (Documentation/RCU/rculist_nulls.txt for details)
1867 refcount_set(&newsk
->sk_refcnt
, 2);
1870 * Increment the counter in the same struct proto as the master
1871 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1872 * is the same as sk->sk_prot->socks, as this field was copied
1875 * This _changes_ the previous behaviour, where
1876 * tcp_create_openreq_child always was incrementing the
1877 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1878 * to be taken into account in all callers. -acme
1880 sk_refcnt_debug_inc(newsk
);
1881 sk_set_socket(newsk
, NULL
);
1882 RCU_INIT_POINTER(newsk
->sk_wq
, NULL
);
1884 if (newsk
->sk_prot
->sockets_allocated
)
1885 sk_sockets_allocated_inc(newsk
);
1887 if (sock_needs_netstamp(sk
) &&
1888 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1889 net_enable_timestamp();
1894 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1896 void sk_free_unlock_clone(struct sock
*sk
)
1898 /* It is still raw copy of parent, so invalidate
1899 * destructor and make plain sk_free() */
1900 sk
->sk_destruct
= NULL
;
1904 EXPORT_SYMBOL_GPL(sk_free_unlock_clone
);
1906 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1910 sk_dst_set(sk
, dst
);
1911 sk
->sk_route_caps
= dst
->dev
->features
| sk
->sk_route_forced_caps
;
1912 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1913 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1914 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1915 if (sk_can_gso(sk
)) {
1916 if (dst
->header_len
&& !xfrm_dst_offload_ok(dst
)) {
1917 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1919 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1920 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1921 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1924 sk
->sk_gso_max_segs
= max_segs
;
1926 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1929 * Simple resource managers for sockets.
1934 * Write buffer destructor automatically called from kfree_skb.
1936 void sock_wfree(struct sk_buff
*skb
)
1938 struct sock
*sk
= skb
->sk
;
1939 unsigned int len
= skb
->truesize
;
1941 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1943 * Keep a reference on sk_wmem_alloc, this will be released
1944 * after sk_write_space() call
1946 WARN_ON(refcount_sub_and_test(len
- 1, &sk
->sk_wmem_alloc
));
1947 sk
->sk_write_space(sk
);
1951 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1952 * could not do because of in-flight packets
1954 if (refcount_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1957 EXPORT_SYMBOL(sock_wfree
);
1959 /* This variant of sock_wfree() is used by TCP,
1960 * since it sets SOCK_USE_WRITE_QUEUE.
1962 void __sock_wfree(struct sk_buff
*skb
)
1964 struct sock
*sk
= skb
->sk
;
1966 if (refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
))
1970 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1975 if (unlikely(!sk_fullsock(sk
))) {
1976 skb
->destructor
= sock_edemux
;
1981 skb
->destructor
= sock_wfree
;
1982 skb_set_hash_from_sk(skb
, sk
);
1984 * We used to take a refcount on sk, but following operation
1985 * is enough to guarantee sk_free() wont free this sock until
1986 * all in-flight packets are completed
1988 refcount_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
1990 EXPORT_SYMBOL(skb_set_owner_w
);
1992 /* This helper is used by netem, as it can hold packets in its
1993 * delay queue. We want to allow the owner socket to send more
1994 * packets, as if they were already TX completed by a typical driver.
1995 * But we also want to keep skb->sk set because some packet schedulers
1996 * rely on it (sch_fq for example).
1998 void skb_orphan_partial(struct sk_buff
*skb
)
2000 if (skb_is_tcp_pure_ack(skb
))
2003 if (skb
->destructor
== sock_wfree
2005 || skb
->destructor
== tcp_wfree
2008 struct sock
*sk
= skb
->sk
;
2010 if (refcount_inc_not_zero(&sk
->sk_refcnt
)) {
2011 WARN_ON(refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
));
2012 skb
->destructor
= sock_efree
;
2018 EXPORT_SYMBOL(skb_orphan_partial
);
2021 * Read buffer destructor automatically called from kfree_skb.
2023 void sock_rfree(struct sk_buff
*skb
)
2025 struct sock
*sk
= skb
->sk
;
2026 unsigned int len
= skb
->truesize
;
2028 atomic_sub(len
, &sk
->sk_rmem_alloc
);
2029 sk_mem_uncharge(sk
, len
);
2031 EXPORT_SYMBOL(sock_rfree
);
2034 * Buffer destructor for skbs that are not used directly in read or write
2035 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2037 void sock_efree(struct sk_buff
*skb
)
2041 EXPORT_SYMBOL(sock_efree
);
2043 kuid_t
sock_i_uid(struct sock
*sk
)
2047 read_lock_bh(&sk
->sk_callback_lock
);
2048 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
2049 read_unlock_bh(&sk
->sk_callback_lock
);
2052 EXPORT_SYMBOL(sock_i_uid
);
2054 unsigned long sock_i_ino(struct sock
*sk
)
2058 read_lock_bh(&sk
->sk_callback_lock
);
2059 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
2060 read_unlock_bh(&sk
->sk_callback_lock
);
2063 EXPORT_SYMBOL(sock_i_ino
);
2066 * Allocate a skb from the socket's send buffer.
2068 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
2071 if (force
|| refcount_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
) {
2072 struct sk_buff
*skb
= alloc_skb(size
, priority
);
2074 skb_set_owner_w(skb
, sk
);
2080 EXPORT_SYMBOL(sock_wmalloc
);
2082 static void sock_ofree(struct sk_buff
*skb
)
2084 struct sock
*sk
= skb
->sk
;
2086 atomic_sub(skb
->truesize
, &sk
->sk_omem_alloc
);
2089 struct sk_buff
*sock_omalloc(struct sock
*sk
, unsigned long size
,
2092 struct sk_buff
*skb
;
2094 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2095 if (atomic_read(&sk
->sk_omem_alloc
) + SKB_TRUESIZE(size
) >
2099 skb
= alloc_skb(size
, priority
);
2103 atomic_add(skb
->truesize
, &sk
->sk_omem_alloc
);
2105 skb
->destructor
= sock_ofree
;
2110 * Allocate a memory block from the socket's option memory buffer.
2112 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
2114 if ((unsigned int)size
<= sysctl_optmem_max
&&
2115 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
2117 /* First do the add, to avoid the race if kmalloc
2120 atomic_add(size
, &sk
->sk_omem_alloc
);
2121 mem
= kmalloc(size
, priority
);
2124 atomic_sub(size
, &sk
->sk_omem_alloc
);
2128 EXPORT_SYMBOL(sock_kmalloc
);
2130 /* Free an option memory block. Note, we actually want the inline
2131 * here as this allows gcc to detect the nullify and fold away the
2132 * condition entirely.
2134 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
2137 if (WARN_ON_ONCE(!mem
))
2143 atomic_sub(size
, &sk
->sk_omem_alloc
);
2146 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
2148 __sock_kfree_s(sk
, mem
, size
, false);
2150 EXPORT_SYMBOL(sock_kfree_s
);
2152 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
2154 __sock_kfree_s(sk
, mem
, size
, true);
2156 EXPORT_SYMBOL(sock_kzfree_s
);
2158 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2159 I think, these locks should be removed for datagram sockets.
2161 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
2165 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2169 if (signal_pending(current
))
2171 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2172 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2173 if (refcount_read(&sk
->sk_wmem_alloc
) < sk
->sk_sndbuf
)
2175 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2179 timeo
= schedule_timeout(timeo
);
2181 finish_wait(sk_sleep(sk
), &wait
);
2187 * Generic send/receive buffer handlers
2190 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
2191 unsigned long data_len
, int noblock
,
2192 int *errcode
, int max_page_order
)
2194 struct sk_buff
*skb
;
2198 timeo
= sock_sndtimeo(sk
, noblock
);
2200 err
= sock_error(sk
);
2205 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2208 if (sk_wmem_alloc_get(sk
) < sk
->sk_sndbuf
)
2211 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2212 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2216 if (signal_pending(current
))
2218 timeo
= sock_wait_for_wmem(sk
, timeo
);
2220 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
2221 errcode
, sk
->sk_allocation
);
2223 skb_set_owner_w(skb
, sk
);
2227 err
= sock_intr_errno(timeo
);
2232 EXPORT_SYMBOL(sock_alloc_send_pskb
);
2234 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
2235 int noblock
, int *errcode
)
2237 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
2239 EXPORT_SYMBOL(sock_alloc_send_skb
);
2241 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
2242 struct sockcm_cookie
*sockc
)
2246 switch (cmsg
->cmsg_type
) {
2248 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
2250 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2252 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
2254 case SO_TIMESTAMPING_OLD
:
2255 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2258 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
2259 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
2262 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
2263 sockc
->tsflags
|= tsflags
;
2266 if (!sock_flag(sk
, SOCK_TXTIME
))
2268 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u64
)))
2270 sockc
->transmit_time
= get_unaligned((u64
*)CMSG_DATA(cmsg
));
2272 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2274 case SCM_CREDENTIALS
:
2281 EXPORT_SYMBOL(__sock_cmsg_send
);
2283 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
2284 struct sockcm_cookie
*sockc
)
2286 struct cmsghdr
*cmsg
;
2289 for_each_cmsghdr(cmsg
, msg
) {
2290 if (!CMSG_OK(msg
, cmsg
))
2292 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
2294 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
2300 EXPORT_SYMBOL(sock_cmsg_send
);
2302 static void sk_enter_memory_pressure(struct sock
*sk
)
2304 if (!sk
->sk_prot
->enter_memory_pressure
)
2307 sk
->sk_prot
->enter_memory_pressure(sk
);
2310 static void sk_leave_memory_pressure(struct sock
*sk
)
2312 if (sk
->sk_prot
->leave_memory_pressure
) {
2313 sk
->sk_prot
->leave_memory_pressure(sk
);
2315 unsigned long *memory_pressure
= sk
->sk_prot
->memory_pressure
;
2317 if (memory_pressure
&& *memory_pressure
)
2318 *memory_pressure
= 0;
2322 /* On 32bit arches, an skb frag is limited to 2^15 */
2323 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2326 * skb_page_frag_refill - check that a page_frag contains enough room
2327 * @sz: minimum size of the fragment we want to get
2328 * @pfrag: pointer to page_frag
2329 * @gfp: priority for memory allocation
2331 * Note: While this allocator tries to use high order pages, there is
2332 * no guarantee that allocations succeed. Therefore, @sz MUST be
2333 * less or equal than PAGE_SIZE.
2335 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
2338 if (page_ref_count(pfrag
->page
) == 1) {
2342 if (pfrag
->offset
+ sz
<= pfrag
->size
)
2344 put_page(pfrag
->page
);
2348 if (SKB_FRAG_PAGE_ORDER
) {
2349 /* Avoid direct reclaim but allow kswapd to wake */
2350 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
2351 __GFP_COMP
| __GFP_NOWARN
|
2353 SKB_FRAG_PAGE_ORDER
);
2354 if (likely(pfrag
->page
)) {
2355 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
2359 pfrag
->page
= alloc_page(gfp
);
2360 if (likely(pfrag
->page
)) {
2361 pfrag
->size
= PAGE_SIZE
;
2366 EXPORT_SYMBOL(skb_page_frag_refill
);
2368 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
2370 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
2373 sk_enter_memory_pressure(sk
);
2374 sk_stream_moderate_sndbuf(sk
);
2377 EXPORT_SYMBOL(sk_page_frag_refill
);
2379 static void __lock_sock(struct sock
*sk
)
2380 __releases(&sk
->sk_lock
.slock
)
2381 __acquires(&sk
->sk_lock
.slock
)
2386 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2387 TASK_UNINTERRUPTIBLE
);
2388 spin_unlock_bh(&sk
->sk_lock
.slock
);
2390 spin_lock_bh(&sk
->sk_lock
.slock
);
2391 if (!sock_owned_by_user(sk
))
2394 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2397 void __release_sock(struct sock
*sk
)
2398 __releases(&sk
->sk_lock
.slock
)
2399 __acquires(&sk
->sk_lock
.slock
)
2401 struct sk_buff
*skb
, *next
;
2403 while ((skb
= sk
->sk_backlog
.head
) != NULL
) {
2404 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2406 spin_unlock_bh(&sk
->sk_lock
.slock
);
2411 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2412 skb_mark_not_on_list(skb
);
2413 sk_backlog_rcv(sk
, skb
);
2418 } while (skb
!= NULL
);
2420 spin_lock_bh(&sk
->sk_lock
.slock
);
2424 * Doing the zeroing here guarantee we can not loop forever
2425 * while a wild producer attempts to flood us.
2427 sk
->sk_backlog
.len
= 0;
2430 void __sk_flush_backlog(struct sock
*sk
)
2432 spin_lock_bh(&sk
->sk_lock
.slock
);
2434 spin_unlock_bh(&sk
->sk_lock
.slock
);
2438 * sk_wait_data - wait for data to arrive at sk_receive_queue
2439 * @sk: sock to wait on
2440 * @timeo: for how long
2441 * @skb: last skb seen on sk_receive_queue
2443 * Now socket state including sk->sk_err is changed only under lock,
2444 * hence we may omit checks after joining wait queue.
2445 * We check receive queue before schedule() only as optimization;
2446 * it is very likely that release_sock() added new data.
2448 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2450 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
2453 add_wait_queue(sk_sleep(sk
), &wait
);
2454 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2455 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
, &wait
);
2456 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2457 remove_wait_queue(sk_sleep(sk
), &wait
);
2460 EXPORT_SYMBOL(sk_wait_data
);
2463 * __sk_mem_raise_allocated - increase memory_allocated
2465 * @size: memory size to allocate
2466 * @amt: pages to allocate
2467 * @kind: allocation type
2469 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2471 int __sk_mem_raise_allocated(struct sock
*sk
, int size
, int amt
, int kind
)
2473 struct proto
*prot
= sk
->sk_prot
;
2474 long allocated
= sk_memory_allocated_add(sk
, amt
);
2475 bool charged
= true;
2477 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2478 !(charged
= mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
)))
2479 goto suppress_allocation
;
2482 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2483 sk_leave_memory_pressure(sk
);
2487 /* Under pressure. */
2488 if (allocated
> sk_prot_mem_limits(sk
, 1))
2489 sk_enter_memory_pressure(sk
);
2491 /* Over hard limit. */
2492 if (allocated
> sk_prot_mem_limits(sk
, 2))
2493 goto suppress_allocation
;
2495 /* guarantee minimum buffer size under pressure */
2496 if (kind
== SK_MEM_RECV
) {
2497 if (atomic_read(&sk
->sk_rmem_alloc
) < sk_get_rmem0(sk
, prot
))
2500 } else { /* SK_MEM_SEND */
2501 int wmem0
= sk_get_wmem0(sk
, prot
);
2503 if (sk
->sk_type
== SOCK_STREAM
) {
2504 if (sk
->sk_wmem_queued
< wmem0
)
2506 } else if (refcount_read(&sk
->sk_wmem_alloc
) < wmem0
) {
2511 if (sk_has_memory_pressure(sk
)) {
2514 if (!sk_under_memory_pressure(sk
))
2516 alloc
= sk_sockets_allocated_read_positive(sk
);
2517 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2518 sk_mem_pages(sk
->sk_wmem_queued
+
2519 atomic_read(&sk
->sk_rmem_alloc
) +
2520 sk
->sk_forward_alloc
))
2524 suppress_allocation
:
2526 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2527 sk_stream_moderate_sndbuf(sk
);
2529 /* Fail only if socket is _under_ its sndbuf.
2530 * In this case we cannot block, so that we have to fail.
2532 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2536 if (kind
== SK_MEM_SEND
|| (kind
== SK_MEM_RECV
&& charged
))
2537 trace_sock_exceed_buf_limit(sk
, prot
, allocated
, kind
);
2539 sk_memory_allocated_sub(sk
, amt
);
2541 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2542 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2546 EXPORT_SYMBOL(__sk_mem_raise_allocated
);
2549 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2551 * @size: memory size to allocate
2552 * @kind: allocation type
2554 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2555 * rmem allocation. This function assumes that protocols which have
2556 * memory_pressure use sk_wmem_queued as write buffer accounting.
2558 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2560 int ret
, amt
= sk_mem_pages(size
);
2562 sk
->sk_forward_alloc
+= amt
<< SK_MEM_QUANTUM_SHIFT
;
2563 ret
= __sk_mem_raise_allocated(sk
, size
, amt
, kind
);
2565 sk
->sk_forward_alloc
-= amt
<< SK_MEM_QUANTUM_SHIFT
;
2568 EXPORT_SYMBOL(__sk_mem_schedule
);
2571 * __sk_mem_reduce_allocated - reclaim memory_allocated
2573 * @amount: number of quanta
2575 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2577 void __sk_mem_reduce_allocated(struct sock
*sk
, int amount
)
2579 sk_memory_allocated_sub(sk
, amount
);
2581 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2582 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2584 if (sk_under_memory_pressure(sk
) &&
2585 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2586 sk_leave_memory_pressure(sk
);
2588 EXPORT_SYMBOL(__sk_mem_reduce_allocated
);
2591 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2593 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2595 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2597 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2598 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2599 __sk_mem_reduce_allocated(sk
, amount
);
2601 EXPORT_SYMBOL(__sk_mem_reclaim
);
2603 int sk_set_peek_off(struct sock
*sk
, int val
)
2605 sk
->sk_peek_off
= val
;
2608 EXPORT_SYMBOL_GPL(sk_set_peek_off
);
2611 * Set of default routines for initialising struct proto_ops when
2612 * the protocol does not support a particular function. In certain
2613 * cases where it makes no sense for a protocol to have a "do nothing"
2614 * function, some default processing is provided.
2617 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2621 EXPORT_SYMBOL(sock_no_bind
);
2623 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2628 EXPORT_SYMBOL(sock_no_connect
);
2630 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2634 EXPORT_SYMBOL(sock_no_socketpair
);
2636 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
,
2641 EXPORT_SYMBOL(sock_no_accept
);
2643 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2648 EXPORT_SYMBOL(sock_no_getname
);
2650 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2654 EXPORT_SYMBOL(sock_no_ioctl
);
2656 int sock_no_listen(struct socket
*sock
, int backlog
)
2660 EXPORT_SYMBOL(sock_no_listen
);
2662 int sock_no_shutdown(struct socket
*sock
, int how
)
2666 EXPORT_SYMBOL(sock_no_shutdown
);
2668 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2669 char __user
*optval
, unsigned int optlen
)
2673 EXPORT_SYMBOL(sock_no_setsockopt
);
2675 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2676 char __user
*optval
, int __user
*optlen
)
2680 EXPORT_SYMBOL(sock_no_getsockopt
);
2682 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2686 EXPORT_SYMBOL(sock_no_sendmsg
);
2688 int sock_no_sendmsg_locked(struct sock
*sk
, struct msghdr
*m
, size_t len
)
2692 EXPORT_SYMBOL(sock_no_sendmsg_locked
);
2694 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2699 EXPORT_SYMBOL(sock_no_recvmsg
);
2701 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2703 /* Mirror missing mmap method error code */
2706 EXPORT_SYMBOL(sock_no_mmap
);
2708 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2711 struct msghdr msg
= {.msg_flags
= flags
};
2713 char *kaddr
= kmap(page
);
2714 iov
.iov_base
= kaddr
+ offset
;
2716 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2720 EXPORT_SYMBOL(sock_no_sendpage
);
2722 ssize_t
sock_no_sendpage_locked(struct sock
*sk
, struct page
*page
,
2723 int offset
, size_t size
, int flags
)
2726 struct msghdr msg
= {.msg_flags
= flags
};
2728 char *kaddr
= kmap(page
);
2730 iov
.iov_base
= kaddr
+ offset
;
2732 res
= kernel_sendmsg_locked(sk
, &msg
, &iov
, 1, size
);
2736 EXPORT_SYMBOL(sock_no_sendpage_locked
);
2739 * Default Socket Callbacks
2742 static void sock_def_wakeup(struct sock
*sk
)
2744 struct socket_wq
*wq
;
2747 wq
= rcu_dereference(sk
->sk_wq
);
2748 if (skwq_has_sleeper(wq
))
2749 wake_up_interruptible_all(&wq
->wait
);
2753 static void sock_def_error_report(struct sock
*sk
)
2755 struct socket_wq
*wq
;
2758 wq
= rcu_dereference(sk
->sk_wq
);
2759 if (skwq_has_sleeper(wq
))
2760 wake_up_interruptible_poll(&wq
->wait
, EPOLLERR
);
2761 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2765 static void sock_def_readable(struct sock
*sk
)
2767 struct socket_wq
*wq
;
2770 wq
= rcu_dereference(sk
->sk_wq
);
2771 if (skwq_has_sleeper(wq
))
2772 wake_up_interruptible_sync_poll(&wq
->wait
, EPOLLIN
| EPOLLPRI
|
2773 EPOLLRDNORM
| EPOLLRDBAND
);
2774 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2778 static void sock_def_write_space(struct sock
*sk
)
2780 struct socket_wq
*wq
;
2784 /* Do not wake up a writer until he can make "significant"
2787 if ((refcount_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
) {
2788 wq
= rcu_dereference(sk
->sk_wq
);
2789 if (skwq_has_sleeper(wq
))
2790 wake_up_interruptible_sync_poll(&wq
->wait
, EPOLLOUT
|
2791 EPOLLWRNORM
| EPOLLWRBAND
);
2793 /* Should agree with poll, otherwise some programs break */
2794 if (sock_writeable(sk
))
2795 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2801 static void sock_def_destruct(struct sock
*sk
)
2805 void sk_send_sigurg(struct sock
*sk
)
2807 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2808 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2809 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2811 EXPORT_SYMBOL(sk_send_sigurg
);
2813 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2814 unsigned long expires
)
2816 if (!mod_timer(timer
, expires
))
2819 EXPORT_SYMBOL(sk_reset_timer
);
2821 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2823 if (del_timer(timer
))
2826 EXPORT_SYMBOL(sk_stop_timer
);
2828 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2831 sk
->sk_send_head
= NULL
;
2833 timer_setup(&sk
->sk_timer
, NULL
, 0);
2835 sk
->sk_allocation
= GFP_KERNEL
;
2836 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2837 sk
->sk_sndbuf
= sysctl_wmem_default
;
2838 sk
->sk_state
= TCP_CLOSE
;
2839 sk_set_socket(sk
, sock
);
2841 sock_set_flag(sk
, SOCK_ZAPPED
);
2844 sk
->sk_type
= sock
->type
;
2845 RCU_INIT_POINTER(sk
->sk_wq
, sock
->wq
);
2847 sk
->sk_uid
= SOCK_INODE(sock
)->i_uid
;
2849 RCU_INIT_POINTER(sk
->sk_wq
, NULL
);
2850 sk
->sk_uid
= make_kuid(sock_net(sk
)->user_ns
, 0);
2853 rwlock_init(&sk
->sk_callback_lock
);
2854 if (sk
->sk_kern_sock
)
2855 lockdep_set_class_and_name(
2856 &sk
->sk_callback_lock
,
2857 af_kern_callback_keys
+ sk
->sk_family
,
2858 af_family_kern_clock_key_strings
[sk
->sk_family
]);
2860 lockdep_set_class_and_name(
2861 &sk
->sk_callback_lock
,
2862 af_callback_keys
+ sk
->sk_family
,
2863 af_family_clock_key_strings
[sk
->sk_family
]);
2865 sk
->sk_state_change
= sock_def_wakeup
;
2866 sk
->sk_data_ready
= sock_def_readable
;
2867 sk
->sk_write_space
= sock_def_write_space
;
2868 sk
->sk_error_report
= sock_def_error_report
;
2869 sk
->sk_destruct
= sock_def_destruct
;
2871 sk
->sk_frag
.page
= NULL
;
2872 sk
->sk_frag
.offset
= 0;
2873 sk
->sk_peek_off
= -1;
2875 sk
->sk_peer_pid
= NULL
;
2876 sk
->sk_peer_cred
= NULL
;
2877 sk
->sk_write_pending
= 0;
2878 sk
->sk_rcvlowat
= 1;
2879 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2880 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2882 sk
->sk_stamp
= SK_DEFAULT_STAMP
;
2883 #if BITS_PER_LONG==32
2884 seqlock_init(&sk
->sk_stamp_seq
);
2886 atomic_set(&sk
->sk_zckey
, 0);
2888 #ifdef CONFIG_NET_RX_BUSY_POLL
2890 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2893 sk
->sk_max_pacing_rate
= ~0UL;
2894 sk
->sk_pacing_rate
= ~0UL;
2895 sk
->sk_pacing_shift
= 10;
2896 sk
->sk_incoming_cpu
= -1;
2898 sk_rx_queue_clear(sk
);
2900 * Before updating sk_refcnt, we must commit prior changes to memory
2901 * (Documentation/RCU/rculist_nulls.txt for details)
2904 refcount_set(&sk
->sk_refcnt
, 1);
2905 atomic_set(&sk
->sk_drops
, 0);
2907 EXPORT_SYMBOL(sock_init_data
);
2909 void lock_sock_nested(struct sock
*sk
, int subclass
)
2912 spin_lock_bh(&sk
->sk_lock
.slock
);
2913 if (sk
->sk_lock
.owned
)
2915 sk
->sk_lock
.owned
= 1;
2916 spin_unlock(&sk
->sk_lock
.slock
);
2918 * The sk_lock has mutex_lock() semantics here:
2920 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2923 EXPORT_SYMBOL(lock_sock_nested
);
2925 void release_sock(struct sock
*sk
)
2927 spin_lock_bh(&sk
->sk_lock
.slock
);
2928 if (sk
->sk_backlog
.tail
)
2931 /* Warning : release_cb() might need to release sk ownership,
2932 * ie call sock_release_ownership(sk) before us.
2934 if (sk
->sk_prot
->release_cb
)
2935 sk
->sk_prot
->release_cb(sk
);
2937 sock_release_ownership(sk
);
2938 if (waitqueue_active(&sk
->sk_lock
.wq
))
2939 wake_up(&sk
->sk_lock
.wq
);
2940 spin_unlock_bh(&sk
->sk_lock
.slock
);
2942 EXPORT_SYMBOL(release_sock
);
2945 * lock_sock_fast - fast version of lock_sock
2948 * This version should be used for very small section, where process wont block
2949 * return false if fast path is taken:
2951 * sk_lock.slock locked, owned = 0, BH disabled
2953 * return true if slow path is taken:
2955 * sk_lock.slock unlocked, owned = 1, BH enabled
2957 bool lock_sock_fast(struct sock
*sk
)
2960 spin_lock_bh(&sk
->sk_lock
.slock
);
2962 if (!sk
->sk_lock
.owned
)
2964 * Note : We must disable BH
2969 sk
->sk_lock
.owned
= 1;
2970 spin_unlock(&sk
->sk_lock
.slock
);
2972 * The sk_lock has mutex_lock() semantics here:
2974 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
2978 EXPORT_SYMBOL(lock_sock_fast
);
2980 int sock_get_timestamp(struct sock
*sk
, struct timeval __user
*userstamp
)
2984 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
2985 tv
= ktime_to_timeval(sock_read_timestamp(sk
));
2986 if (tv
.tv_sec
== -1)
2988 if (tv
.tv_sec
== 0) {
2989 ktime_t kt
= ktime_get_real();
2990 sock_write_timestamp(sk
, kt
);
2991 tv
= ktime_to_timeval(kt
);
2993 return copy_to_user(userstamp
, &tv
, sizeof(tv
)) ? -EFAULT
: 0;
2995 EXPORT_SYMBOL(sock_get_timestamp
);
2997 int sock_get_timestampns(struct sock
*sk
, struct timespec __user
*userstamp
)
3001 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
3002 ts
= ktime_to_timespec(sock_read_timestamp(sk
));
3003 if (ts
.tv_sec
== -1)
3005 if (ts
.tv_sec
== 0) {
3006 ktime_t kt
= ktime_get_real();
3007 sock_write_timestamp(sk
, kt
);
3008 ts
= ktime_to_timespec(sk
->sk_stamp
);
3010 return copy_to_user(userstamp
, &ts
, sizeof(ts
)) ? -EFAULT
: 0;
3012 EXPORT_SYMBOL(sock_get_timestampns
);
3014 void sock_enable_timestamp(struct sock
*sk
, int flag
)
3016 if (!sock_flag(sk
, flag
)) {
3017 unsigned long previous_flags
= sk
->sk_flags
;
3019 sock_set_flag(sk
, flag
);
3021 * we just set one of the two flags which require net
3022 * time stamping, but time stamping might have been on
3023 * already because of the other one
3025 if (sock_needs_netstamp(sk
) &&
3026 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
3027 net_enable_timestamp();
3031 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
3032 int level
, int type
)
3034 struct sock_exterr_skb
*serr
;
3035 struct sk_buff
*skb
;
3039 skb
= sock_dequeue_err_skb(sk
);
3045 msg
->msg_flags
|= MSG_TRUNC
;
3048 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
3052 sock_recv_timestamp(msg
, sk
, skb
);
3054 serr
= SKB_EXT_ERR(skb
);
3055 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
3057 msg
->msg_flags
|= MSG_ERRQUEUE
;
3065 EXPORT_SYMBOL(sock_recv_errqueue
);
3068 * Get a socket option on an socket.
3070 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3071 * asynchronous errors should be reported by getsockopt. We assume
3072 * this means if you specify SO_ERROR (otherwise whats the point of it).
3074 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
3075 char __user
*optval
, int __user
*optlen
)
3077 struct sock
*sk
= sock
->sk
;
3079 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
3081 EXPORT_SYMBOL(sock_common_getsockopt
);
3083 #ifdef CONFIG_COMPAT
3084 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
3085 char __user
*optval
, int __user
*optlen
)
3087 struct sock
*sk
= sock
->sk
;
3089 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
3090 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
3092 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
3094 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
3097 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
3100 struct sock
*sk
= sock
->sk
;
3104 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
3105 flags
& ~MSG_DONTWAIT
, &addr_len
);
3107 msg
->msg_namelen
= addr_len
;
3110 EXPORT_SYMBOL(sock_common_recvmsg
);
3113 * Set socket options on an inet socket.
3115 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
3116 char __user
*optval
, unsigned int optlen
)
3118 struct sock
*sk
= sock
->sk
;
3120 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
3122 EXPORT_SYMBOL(sock_common_setsockopt
);
3124 #ifdef CONFIG_COMPAT
3125 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
3126 char __user
*optval
, unsigned int optlen
)
3128 struct sock
*sk
= sock
->sk
;
3130 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
3131 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
3133 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
3135 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
3138 void sk_common_release(struct sock
*sk
)
3140 if (sk
->sk_prot
->destroy
)
3141 sk
->sk_prot
->destroy(sk
);
3144 * Observation: when sock_common_release is called, processes have
3145 * no access to socket. But net still has.
3146 * Step one, detach it from networking:
3148 * A. Remove from hash tables.
3151 sk
->sk_prot
->unhash(sk
);
3154 * In this point socket cannot receive new packets, but it is possible
3155 * that some packets are in flight because some CPU runs receiver and
3156 * did hash table lookup before we unhashed socket. They will achieve
3157 * receive queue and will be purged by socket destructor.
3159 * Also we still have packets pending on receive queue and probably,
3160 * our own packets waiting in device queues. sock_destroy will drain
3161 * receive queue, but transmitted packets will delay socket destruction
3162 * until the last reference will be released.
3167 xfrm_sk_free_policy(sk
);
3169 sk_refcnt_debug_release(sk
);
3173 EXPORT_SYMBOL(sk_common_release
);
3175 void sk_get_meminfo(const struct sock
*sk
, u32
*mem
)
3177 memset(mem
, 0, sizeof(*mem
) * SK_MEMINFO_VARS
);
3179 mem
[SK_MEMINFO_RMEM_ALLOC
] = sk_rmem_alloc_get(sk
);
3180 mem
[SK_MEMINFO_RCVBUF
] = sk
->sk_rcvbuf
;
3181 mem
[SK_MEMINFO_WMEM_ALLOC
] = sk_wmem_alloc_get(sk
);
3182 mem
[SK_MEMINFO_SNDBUF
] = sk
->sk_sndbuf
;
3183 mem
[SK_MEMINFO_FWD_ALLOC
] = sk
->sk_forward_alloc
;
3184 mem
[SK_MEMINFO_WMEM_QUEUED
] = sk
->sk_wmem_queued
;
3185 mem
[SK_MEMINFO_OPTMEM
] = atomic_read(&sk
->sk_omem_alloc
);
3186 mem
[SK_MEMINFO_BACKLOG
] = sk
->sk_backlog
.len
;
3187 mem
[SK_MEMINFO_DROPS
] = atomic_read(&sk
->sk_drops
);
3190 #ifdef CONFIG_PROC_FS
3191 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3193 int val
[PROTO_INUSE_NR
];
3196 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
3198 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
3200 __this_cpu_add(net
->core
.prot_inuse
->val
[prot
->inuse_idx
], val
);
3202 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
3204 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
3206 int cpu
, idx
= prot
->inuse_idx
;
3209 for_each_possible_cpu(cpu
)
3210 res
+= per_cpu_ptr(net
->core
.prot_inuse
, cpu
)->val
[idx
];
3212 return res
>= 0 ? res
: 0;
3214 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
3216 static void sock_inuse_add(struct net
*net
, int val
)
3218 this_cpu_add(*net
->core
.sock_inuse
, val
);
3221 int sock_inuse_get(struct net
*net
)
3225 for_each_possible_cpu(cpu
)
3226 res
+= *per_cpu_ptr(net
->core
.sock_inuse
, cpu
);
3231 EXPORT_SYMBOL_GPL(sock_inuse_get
);
3233 static int __net_init
sock_inuse_init_net(struct net
*net
)
3235 net
->core
.prot_inuse
= alloc_percpu(struct prot_inuse
);
3236 if (net
->core
.prot_inuse
== NULL
)
3239 net
->core
.sock_inuse
= alloc_percpu(int);
3240 if (net
->core
.sock_inuse
== NULL
)
3246 free_percpu(net
->core
.prot_inuse
);
3250 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
3252 free_percpu(net
->core
.prot_inuse
);
3253 free_percpu(net
->core
.sock_inuse
);
3256 static struct pernet_operations net_inuse_ops
= {
3257 .init
= sock_inuse_init_net
,
3258 .exit
= sock_inuse_exit_net
,
3261 static __init
int net_inuse_init(void)
3263 if (register_pernet_subsys(&net_inuse_ops
))
3264 panic("Cannot initialize net inuse counters");
3269 core_initcall(net_inuse_init
);
3271 static void assign_proto_idx(struct proto
*prot
)
3273 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
3275 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
3276 pr_err("PROTO_INUSE_NR exhausted\n");
3280 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
3283 static void release_proto_idx(struct proto
*prot
)
3285 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
3286 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
3289 static inline void assign_proto_idx(struct proto
*prot
)
3293 static inline void release_proto_idx(struct proto
*prot
)
3297 static void sock_inuse_add(struct net
*net
, int val
)
3302 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
3306 kfree(rsk_prot
->slab_name
);
3307 rsk_prot
->slab_name
= NULL
;
3308 kmem_cache_destroy(rsk_prot
->slab
);
3309 rsk_prot
->slab
= NULL
;
3312 static int req_prot_init(const struct proto
*prot
)
3314 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
3319 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
3321 if (!rsk_prot
->slab_name
)
3324 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
3325 rsk_prot
->obj_size
, 0,
3326 SLAB_ACCOUNT
| prot
->slab_flags
,
3329 if (!rsk_prot
->slab
) {
3330 pr_crit("%s: Can't create request sock SLAB cache!\n",
3337 int proto_register(struct proto
*prot
, int alloc_slab
)
3340 prot
->slab
= kmem_cache_create_usercopy(prot
->name
,
3342 SLAB_HWCACHE_ALIGN
| SLAB_ACCOUNT
|
3344 prot
->useroffset
, prot
->usersize
,
3347 if (prot
->slab
== NULL
) {
3348 pr_crit("%s: Can't create sock SLAB cache!\n",
3353 if (req_prot_init(prot
))
3354 goto out_free_request_sock_slab
;
3356 if (prot
->twsk_prot
!= NULL
) {
3357 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
3359 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
3360 goto out_free_request_sock_slab
;
3362 prot
->twsk_prot
->twsk_slab
=
3363 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
3364 prot
->twsk_prot
->twsk_obj_size
,
3369 if (prot
->twsk_prot
->twsk_slab
== NULL
)
3370 goto out_free_timewait_sock_slab_name
;
3374 mutex_lock(&proto_list_mutex
);
3375 list_add(&prot
->node
, &proto_list
);
3376 assign_proto_idx(prot
);
3377 mutex_unlock(&proto_list_mutex
);
3380 out_free_timewait_sock_slab_name
:
3381 kfree(prot
->twsk_prot
->twsk_slab_name
);
3382 out_free_request_sock_slab
:
3383 req_prot_cleanup(prot
->rsk_prot
);
3385 kmem_cache_destroy(prot
->slab
);
3390 EXPORT_SYMBOL(proto_register
);
3392 void proto_unregister(struct proto
*prot
)
3394 mutex_lock(&proto_list_mutex
);
3395 release_proto_idx(prot
);
3396 list_del(&prot
->node
);
3397 mutex_unlock(&proto_list_mutex
);
3399 kmem_cache_destroy(prot
->slab
);
3402 req_prot_cleanup(prot
->rsk_prot
);
3404 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
3405 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
3406 kfree(prot
->twsk_prot
->twsk_slab_name
);
3407 prot
->twsk_prot
->twsk_slab
= NULL
;
3410 EXPORT_SYMBOL(proto_unregister
);
3412 int sock_load_diag_module(int family
, int protocol
)
3415 if (!sock_is_registered(family
))
3418 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK
,
3419 NETLINK_SOCK_DIAG
, family
);
3423 if (family
== AF_INET
&&
3424 protocol
!= IPPROTO_RAW
&&
3425 !rcu_access_pointer(inet_protos
[protocol
]))
3429 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK
,
3430 NETLINK_SOCK_DIAG
, family
, protocol
);
3432 EXPORT_SYMBOL(sock_load_diag_module
);
3434 #ifdef CONFIG_PROC_FS
3435 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3436 __acquires(proto_list_mutex
)
3438 mutex_lock(&proto_list_mutex
);
3439 return seq_list_start_head(&proto_list
, *pos
);
3442 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3444 return seq_list_next(v
, &proto_list
, pos
);
3447 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
3448 __releases(proto_list_mutex
)
3450 mutex_unlock(&proto_list_mutex
);
3453 static char proto_method_implemented(const void *method
)
3455 return method
== NULL
? 'n' : 'y';
3457 static long sock_prot_memory_allocated(struct proto
*proto
)
3459 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
3462 static char *sock_prot_memory_pressure(struct proto
*proto
)
3464 return proto
->memory_pressure
!= NULL
?
3465 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
3468 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
3471 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3472 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3475 sock_prot_inuse_get(seq_file_net(seq
), proto
),
3476 sock_prot_memory_allocated(proto
),
3477 sock_prot_memory_pressure(proto
),
3479 proto
->slab
== NULL
? "no" : "yes",
3480 module_name(proto
->owner
),
3481 proto_method_implemented(proto
->close
),
3482 proto_method_implemented(proto
->connect
),
3483 proto_method_implemented(proto
->disconnect
),
3484 proto_method_implemented(proto
->accept
),
3485 proto_method_implemented(proto
->ioctl
),
3486 proto_method_implemented(proto
->init
),
3487 proto_method_implemented(proto
->destroy
),
3488 proto_method_implemented(proto
->shutdown
),
3489 proto_method_implemented(proto
->setsockopt
),
3490 proto_method_implemented(proto
->getsockopt
),
3491 proto_method_implemented(proto
->sendmsg
),
3492 proto_method_implemented(proto
->recvmsg
),
3493 proto_method_implemented(proto
->sendpage
),
3494 proto_method_implemented(proto
->bind
),
3495 proto_method_implemented(proto
->backlog_rcv
),
3496 proto_method_implemented(proto
->hash
),
3497 proto_method_implemented(proto
->unhash
),
3498 proto_method_implemented(proto
->get_port
),
3499 proto_method_implemented(proto
->enter_memory_pressure
));
3502 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3504 if (v
== &proto_list
)
3505 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3514 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3516 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3520 static const struct seq_operations proto_seq_ops
= {
3521 .start
= proto_seq_start
,
3522 .next
= proto_seq_next
,
3523 .stop
= proto_seq_stop
,
3524 .show
= proto_seq_show
,
3527 static __net_init
int proto_init_net(struct net
*net
)
3529 if (!proc_create_net("protocols", 0444, net
->proc_net
, &proto_seq_ops
,
3530 sizeof(struct seq_net_private
)))
3536 static __net_exit
void proto_exit_net(struct net
*net
)
3538 remove_proc_entry("protocols", net
->proc_net
);
3542 static __net_initdata
struct pernet_operations proto_net_ops
= {
3543 .init
= proto_init_net
,
3544 .exit
= proto_exit_net
,
3547 static int __init
proto_init(void)
3549 return register_pernet_subsys(&proto_net_ops
);
3552 subsys_initcall(proto_init
);
3554 #endif /* PROC_FS */
3556 #ifdef CONFIG_NET_RX_BUSY_POLL
3557 bool sk_busy_loop_end(void *p
, unsigned long start_time
)
3559 struct sock
*sk
= p
;
3561 return !skb_queue_empty(&sk
->sk_receive_queue
) ||
3562 sk_busy_loop_timeout(sk
, start_time
);
3564 EXPORT_SYMBOL(sk_busy_loop_end
);
3565 #endif /* CONFIG_NET_RX_BUSY_POLL */