1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 static DEFINE_MUTEX(proto_list_mutex
);
143 static LIST_HEAD(proto_list
);
145 static void sock_inuse_add(struct net
*net
, int val
);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock
*sk
,
158 struct user_namespace
*user_ns
, int cap
)
160 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
161 ns_capable(user_ns
, cap
);
163 EXPORT_SYMBOL(sk_ns_capable
);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock
*sk
, int cap
)
176 return sk_ns_capable(sk
, &init_user_ns
, cap
);
178 EXPORT_SYMBOL(sk_capable
);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock
*sk
, int cap
)
191 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
193 EXPORT_SYMBOL(sk_net_capable
);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
200 static struct lock_class_key af_family_keys
[AF_MAX
];
201 static struct lock_class_key af_family_kern_keys
[AF_MAX
];
202 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
203 static struct lock_class_key af_family_kern_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
211 #define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
229 static const char *const af_family_key_strings
[AF_MAX
+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings
[AF_MAX
+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings
[AF_MAX
+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings
[AF_MAX
+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings
[AF_MAX
+1] = {
249 _sock_locks("rlock-")
251 static const char *const af_family_wlock_key_strings
[AF_MAX
+1] = {
252 _sock_locks("wlock-")
254 static const char *const af_family_elock_key_strings
[AF_MAX
+1] = {
255 _sock_locks("elock-")
259 * sk_callback_lock and sk queues locking rules are per-address-family,
260 * so split the lock classes by using a per-AF key:
262 static struct lock_class_key af_callback_keys
[AF_MAX
];
263 static struct lock_class_key af_rlock_keys
[AF_MAX
];
264 static struct lock_class_key af_wlock_keys
[AF_MAX
];
265 static struct lock_class_key af_elock_keys
[AF_MAX
];
266 static struct lock_class_key af_kern_callback_keys
[AF_MAX
];
268 /* Run time adjustable parameters. */
269 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
270 EXPORT_SYMBOL(sysctl_wmem_max
);
271 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
272 EXPORT_SYMBOL(sysctl_rmem_max
);
273 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
274 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
276 /* Maximal space eaten by iovec or ancillary data plus some space */
277 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
278 EXPORT_SYMBOL(sysctl_optmem_max
);
280 int sysctl_tstamp_allow_data __read_mostly
= 1;
282 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key
);
283 EXPORT_SYMBOL_GPL(memalloc_socks_key
);
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
293 void sk_set_memalloc(struct sock
*sk
)
295 sock_set_flag(sk
, SOCK_MEMALLOC
);
296 sk
->sk_allocation
|= __GFP_MEMALLOC
;
297 static_branch_inc(&memalloc_socks_key
);
299 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
301 void sk_clear_memalloc(struct sock
*sk
)
303 sock_reset_flag(sk
, SOCK_MEMALLOC
);
304 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
305 static_branch_dec(&memalloc_socks_key
);
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
318 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
321 unsigned int noreclaim_flag
;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
326 noreclaim_flag
= memalloc_noreclaim_save();
327 ret
= sk
->sk_backlog_rcv(sk
, skb
);
328 memalloc_noreclaim_restore(noreclaim_flag
);
332 EXPORT_SYMBOL(__sk_backlog_rcv
);
334 static int sock_get_timeout(long timeo
, void *optval
, bool old_timeval
)
336 struct __kernel_sock_timeval tv
;
338 if (timeo
== MAX_SCHEDULE_TIMEOUT
) {
342 tv
.tv_sec
= timeo
/ HZ
;
343 tv
.tv_usec
= ((timeo
% HZ
) * USEC_PER_SEC
) / HZ
;
346 if (old_timeval
&& in_compat_syscall() && !COMPAT_USE_64BIT_TIME
) {
347 struct old_timeval32 tv32
= { tv
.tv_sec
, tv
.tv_usec
};
348 *(struct old_timeval32
*)optval
= tv32
;
353 struct __kernel_old_timeval old_tv
;
354 old_tv
.tv_sec
= tv
.tv_sec
;
355 old_tv
.tv_usec
= tv
.tv_usec
;
356 *(struct __kernel_old_timeval
*)optval
= old_tv
;
357 return sizeof(old_tv
);
360 *(struct __kernel_sock_timeval
*)optval
= tv
;
364 static int sock_set_timeout(long *timeo_p
, sockptr_t optval
, int optlen
,
367 struct __kernel_sock_timeval tv
;
369 if (old_timeval
&& in_compat_syscall() && !COMPAT_USE_64BIT_TIME
) {
370 struct old_timeval32 tv32
;
372 if (optlen
< sizeof(tv32
))
375 if (copy_from_sockptr(&tv32
, optval
, sizeof(tv32
)))
377 tv
.tv_sec
= tv32
.tv_sec
;
378 tv
.tv_usec
= tv32
.tv_usec
;
379 } else if (old_timeval
) {
380 struct __kernel_old_timeval old_tv
;
382 if (optlen
< sizeof(old_tv
))
384 if (copy_from_sockptr(&old_tv
, optval
, sizeof(old_tv
)))
386 tv
.tv_sec
= old_tv
.tv_sec
;
387 tv
.tv_usec
= old_tv
.tv_usec
;
389 if (optlen
< sizeof(tv
))
391 if (copy_from_sockptr(&tv
, optval
, sizeof(tv
)))
394 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
398 static int warned __read_mostly
;
401 if (warned
< 10 && net_ratelimit()) {
403 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404 __func__
, current
->comm
, task_pid_nr(current
));
408 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
409 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
411 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/ HZ
- 1))
412 *timeo_p
= tv
.tv_sec
* HZ
+ DIV_ROUND_UP((unsigned long)tv
.tv_usec
, USEC_PER_SEC
/ HZ
);
416 static void sock_warn_obsolete_bsdism(const char *name
)
419 static char warncomm
[TASK_COMM_LEN
];
420 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
421 strcpy(warncomm
, current
->comm
);
422 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
428 static bool sock_needs_netstamp(const struct sock
*sk
)
430 switch (sk
->sk_family
) {
439 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
441 if (sk
->sk_flags
& flags
) {
442 sk
->sk_flags
&= ~flags
;
443 if (sock_needs_netstamp(sk
) &&
444 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
445 net_disable_timestamp();
450 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
453 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
455 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
456 atomic_inc(&sk
->sk_drops
);
457 trace_sock_rcvqueue_full(sk
, skb
);
461 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
462 atomic_inc(&sk
->sk_drops
);
467 skb_set_owner_r(skb
, sk
);
469 /* we escape from rcu protected region, make sure we dont leak
474 spin_lock_irqsave(&list
->lock
, flags
);
475 sock_skb_set_dropcount(sk
, skb
);
476 __skb_queue_tail(list
, skb
);
477 spin_unlock_irqrestore(&list
->lock
, flags
);
479 if (!sock_flag(sk
, SOCK_DEAD
))
480 sk
->sk_data_ready(sk
);
483 EXPORT_SYMBOL(__sock_queue_rcv_skb
);
485 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
489 err
= sk_filter(sk
, skb
);
493 return __sock_queue_rcv_skb(sk
, skb
);
495 EXPORT_SYMBOL(sock_queue_rcv_skb
);
497 int __sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
,
498 const int nested
, unsigned int trim_cap
, bool refcounted
)
500 int rc
= NET_RX_SUCCESS
;
502 if (sk_filter_trim_cap(sk
, skb
, trim_cap
))
503 goto discard_and_relse
;
507 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
508 atomic_inc(&sk
->sk_drops
);
509 goto discard_and_relse
;
512 bh_lock_sock_nested(sk
);
515 if (!sock_owned_by_user(sk
)) {
517 * trylock + unlock semantics:
519 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
521 rc
= sk_backlog_rcv(sk
, skb
);
523 mutex_release(&sk
->sk_lock
.dep_map
, _RET_IP_
);
524 } else if (sk_add_backlog(sk
, skb
, READ_ONCE(sk
->sk_rcvbuf
))) {
526 atomic_inc(&sk
->sk_drops
);
527 goto discard_and_relse
;
539 EXPORT_SYMBOL(__sk_receive_skb
);
541 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
543 struct dst_entry
*dst
= __sk_dst_get(sk
);
545 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
546 sk_tx_queue_clear(sk
);
547 sk
->sk_dst_pending_confirm
= 0;
548 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
555 EXPORT_SYMBOL(__sk_dst_check
);
557 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
559 struct dst_entry
*dst
= sk_dst_get(sk
);
561 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
569 EXPORT_SYMBOL(sk_dst_check
);
571 static int sock_bindtoindex_locked(struct sock
*sk
, int ifindex
)
573 int ret
= -ENOPROTOOPT
;
574 #ifdef CONFIG_NETDEVICES
575 struct net
*net
= sock_net(sk
);
579 if (sk
->sk_bound_dev_if
&& !ns_capable(net
->user_ns
, CAP_NET_RAW
))
586 sk
->sk_bound_dev_if
= ifindex
;
587 if (sk
->sk_prot
->rehash
)
588 sk
->sk_prot
->rehash(sk
);
599 int sock_bindtoindex(struct sock
*sk
, int ifindex
, bool lock_sk
)
605 ret
= sock_bindtoindex_locked(sk
, ifindex
);
611 EXPORT_SYMBOL(sock_bindtoindex
);
613 static int sock_setbindtodevice(struct sock
*sk
, sockptr_t optval
, int optlen
)
615 int ret
= -ENOPROTOOPT
;
616 #ifdef CONFIG_NETDEVICES
617 struct net
*net
= sock_net(sk
);
618 char devname
[IFNAMSIZ
];
625 /* Bind this socket to a particular device like "eth0",
626 * as specified in the passed interface name. If the
627 * name is "" or the option length is zero the socket
630 if (optlen
> IFNAMSIZ
- 1)
631 optlen
= IFNAMSIZ
- 1;
632 memset(devname
, 0, sizeof(devname
));
635 if (copy_from_sockptr(devname
, optval
, optlen
))
639 if (devname
[0] != '\0') {
640 struct net_device
*dev
;
643 dev
= dev_get_by_name_rcu(net
, devname
);
645 index
= dev
->ifindex
;
652 return sock_bindtoindex(sk
, index
, true);
659 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
660 int __user
*optlen
, int len
)
662 int ret
= -ENOPROTOOPT
;
663 #ifdef CONFIG_NETDEVICES
664 struct net
*net
= sock_net(sk
);
665 char devname
[IFNAMSIZ
];
667 if (sk
->sk_bound_dev_if
== 0) {
676 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
680 len
= strlen(devname
) + 1;
683 if (copy_to_user(optval
, devname
, len
))
688 if (put_user(len
, optlen
))
699 bool sk_mc_loop(struct sock
*sk
)
701 if (dev_recursion_level())
705 switch (sk
->sk_family
) {
707 return inet_sk(sk
)->mc_loop
;
708 #if IS_ENABLED(CONFIG_IPV6)
710 return inet6_sk(sk
)->mc_loop
;
716 EXPORT_SYMBOL(sk_mc_loop
);
718 void sock_set_reuseaddr(struct sock
*sk
)
721 sk
->sk_reuse
= SK_CAN_REUSE
;
724 EXPORT_SYMBOL(sock_set_reuseaddr
);
726 void sock_set_reuseport(struct sock
*sk
)
729 sk
->sk_reuseport
= true;
732 EXPORT_SYMBOL(sock_set_reuseport
);
734 void sock_no_linger(struct sock
*sk
)
737 sk
->sk_lingertime
= 0;
738 sock_set_flag(sk
, SOCK_LINGER
);
741 EXPORT_SYMBOL(sock_no_linger
);
743 void sock_set_priority(struct sock
*sk
, u32 priority
)
746 sk
->sk_priority
= priority
;
749 EXPORT_SYMBOL(sock_set_priority
);
751 void sock_set_sndtimeo(struct sock
*sk
, s64 secs
)
754 if (secs
&& secs
< MAX_SCHEDULE_TIMEOUT
/ HZ
- 1)
755 sk
->sk_sndtimeo
= secs
* HZ
;
757 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
760 EXPORT_SYMBOL(sock_set_sndtimeo
);
762 static void __sock_set_timestamps(struct sock
*sk
, bool val
, bool new, bool ns
)
765 sock_valbool_flag(sk
, SOCK_TSTAMP_NEW
, new);
766 sock_valbool_flag(sk
, SOCK_RCVTSTAMPNS
, ns
);
767 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
768 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
770 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
771 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
772 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
776 void sock_enable_timestamps(struct sock
*sk
)
779 __sock_set_timestamps(sk
, true, false, true);
782 EXPORT_SYMBOL(sock_enable_timestamps
);
784 void sock_set_keepalive(struct sock
*sk
)
787 if (sk
->sk_prot
->keepalive
)
788 sk
->sk_prot
->keepalive(sk
, true);
789 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, true);
792 EXPORT_SYMBOL(sock_set_keepalive
);
794 static void __sock_set_rcvbuf(struct sock
*sk
, int val
)
796 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
797 * as a negative value.
799 val
= min_t(int, val
, INT_MAX
/ 2);
800 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
802 /* We double it on the way in to account for "struct sk_buff" etc.
803 * overhead. Applications assume that the SO_RCVBUF setting they make
804 * will allow that much actual data to be received on that socket.
806 * Applications are unaware that "struct sk_buff" and other overheads
807 * allocate from the receive buffer during socket buffer allocation.
809 * And after considering the possible alternatives, returning the value
810 * we actually used in getsockopt is the most desirable behavior.
812 WRITE_ONCE(sk
->sk_rcvbuf
, max_t(int, val
* 2, SOCK_MIN_RCVBUF
));
815 void sock_set_rcvbuf(struct sock
*sk
, int val
)
818 __sock_set_rcvbuf(sk
, val
);
821 EXPORT_SYMBOL(sock_set_rcvbuf
);
823 void sock_set_mark(struct sock
*sk
, u32 val
)
829 EXPORT_SYMBOL(sock_set_mark
);
832 * This is meant for all protocols to use and covers goings on
833 * at the socket level. Everything here is generic.
836 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
837 sockptr_t optval
, unsigned int optlen
)
839 struct sock_txtime sk_txtime
;
840 struct sock
*sk
= sock
->sk
;
847 * Options without arguments
850 if (optname
== SO_BINDTODEVICE
)
851 return sock_setbindtodevice(sk
, optval
, optlen
);
853 if (optlen
< sizeof(int))
856 if (copy_from_sockptr(&val
, optval
, sizeof(val
)))
859 valbool
= val
? 1 : 0;
865 if (val
&& !capable(CAP_NET_ADMIN
))
868 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
871 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
874 sk
->sk_reuseport
= valbool
;
883 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
887 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
890 /* Don't error on this BSD doesn't and if you think
891 * about it this is right. Otherwise apps have to
892 * play 'guess the biggest size' games. RCVBUF/SNDBUF
893 * are treated in BSD as hints
895 val
= min_t(u32
, val
, sysctl_wmem_max
);
897 /* Ensure val * 2 fits into an int, to prevent max_t()
898 * from treating it as a negative value.
900 val
= min_t(int, val
, INT_MAX
/ 2);
901 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
902 WRITE_ONCE(sk
->sk_sndbuf
,
903 max_t(int, val
* 2, SOCK_MIN_SNDBUF
));
904 /* Wake up sending tasks if we upped the value. */
905 sk
->sk_write_space(sk
);
909 if (!capable(CAP_NET_ADMIN
)) {
914 /* No negative values (to prevent underflow, as val will be
922 /* Don't error on this BSD doesn't and if you think
923 * about it this is right. Otherwise apps have to
924 * play 'guess the biggest size' games. RCVBUF/SNDBUF
925 * are treated in BSD as hints
927 __sock_set_rcvbuf(sk
, min_t(u32
, val
, sysctl_rmem_max
));
931 if (!capable(CAP_NET_ADMIN
)) {
936 /* No negative values (to prevent underflow, as val will be
939 __sock_set_rcvbuf(sk
, max(val
, 0));
943 if (sk
->sk_prot
->keepalive
)
944 sk
->sk_prot
->keepalive(sk
, valbool
);
945 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
949 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
953 sk
->sk_no_check_tx
= valbool
;
957 if ((val
>= 0 && val
<= 6) ||
958 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
959 sk
->sk_priority
= val
;
965 if (optlen
< sizeof(ling
)) {
966 ret
= -EINVAL
; /* 1003.1g */
969 if (copy_from_sockptr(&ling
, optval
, sizeof(ling
))) {
974 sock_reset_flag(sk
, SOCK_LINGER
);
976 #if (BITS_PER_LONG == 32)
977 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
978 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
981 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
982 sock_set_flag(sk
, SOCK_LINGER
);
987 sock_warn_obsolete_bsdism("setsockopt");
992 set_bit(SOCK_PASSCRED
, &sock
->flags
);
994 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
997 case SO_TIMESTAMP_OLD
:
998 __sock_set_timestamps(sk
, valbool
, false, false);
1000 case SO_TIMESTAMP_NEW
:
1001 __sock_set_timestamps(sk
, valbool
, true, false);
1003 case SO_TIMESTAMPNS_OLD
:
1004 __sock_set_timestamps(sk
, valbool
, false, true);
1006 case SO_TIMESTAMPNS_NEW
:
1007 __sock_set_timestamps(sk
, valbool
, true, true);
1009 case SO_TIMESTAMPING_NEW
:
1010 sock_set_flag(sk
, SOCK_TSTAMP_NEW
);
1012 case SO_TIMESTAMPING_OLD
:
1013 if (val
& ~SOF_TIMESTAMPING_MASK
) {
1018 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
1019 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
1020 if (sk
->sk_protocol
== IPPROTO_TCP
&&
1021 sk
->sk_type
== SOCK_STREAM
) {
1022 if ((1 << sk
->sk_state
) &
1023 (TCPF_CLOSE
| TCPF_LISTEN
)) {
1027 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
1033 if (val
& SOF_TIMESTAMPING_OPT_STATS
&&
1034 !(val
& SOF_TIMESTAMPING_OPT_TSONLY
)) {
1039 sk
->sk_tsflags
= val
;
1040 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
1041 sock_enable_timestamp(sk
,
1042 SOCK_TIMESTAMPING_RX_SOFTWARE
);
1044 if (optname
== SO_TIMESTAMPING_NEW
)
1045 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
1047 sock_disable_timestamp(sk
,
1048 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
1055 if (sock
->ops
->set_rcvlowat
)
1056 ret
= sock
->ops
->set_rcvlowat(sk
, val
);
1058 WRITE_ONCE(sk
->sk_rcvlowat
, val
? : 1);
1061 case SO_RCVTIMEO_OLD
:
1062 case SO_RCVTIMEO_NEW
:
1063 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
,
1064 optlen
, optname
== SO_RCVTIMEO_OLD
);
1067 case SO_SNDTIMEO_OLD
:
1068 case SO_SNDTIMEO_NEW
:
1069 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
,
1070 optlen
, optname
== SO_SNDTIMEO_OLD
);
1073 case SO_ATTACH_FILTER
: {
1074 struct sock_fprog fprog
;
1076 ret
= copy_bpf_fprog_from_user(&fprog
, optval
, optlen
);
1078 ret
= sk_attach_filter(&fprog
, sk
);
1083 if (optlen
== sizeof(u32
)) {
1087 if (copy_from_sockptr(&ufd
, optval
, sizeof(ufd
)))
1090 ret
= sk_attach_bpf(ufd
, sk
);
1094 case SO_ATTACH_REUSEPORT_CBPF
: {
1095 struct sock_fprog fprog
;
1097 ret
= copy_bpf_fprog_from_user(&fprog
, optval
, optlen
);
1099 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
1102 case SO_ATTACH_REUSEPORT_EBPF
:
1104 if (optlen
== sizeof(u32
)) {
1108 if (copy_from_sockptr(&ufd
, optval
, sizeof(ufd
)))
1111 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
1115 case SO_DETACH_REUSEPORT_BPF
:
1116 ret
= reuseport_detach_prog(sk
);
1119 case SO_DETACH_FILTER
:
1120 ret
= sk_detach_filter(sk
);
1123 case SO_LOCK_FILTER
:
1124 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
1127 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
1132 set_bit(SOCK_PASSSEC
, &sock
->flags
);
1134 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
1137 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
)) {
1139 } else if (val
!= sk
->sk_mark
) {
1146 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
1149 case SO_WIFI_STATUS
:
1150 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
1154 if (sock
->ops
->set_peek_off
)
1155 ret
= sock
->ops
->set_peek_off(sk
, val
);
1161 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
1164 case SO_SELECT_ERR_QUEUE
:
1165 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
1168 #ifdef CONFIG_NET_RX_BUSY_POLL
1170 /* allow unprivileged users to decrease the value */
1171 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
1177 sk
->sk_ll_usec
= val
;
1182 case SO_MAX_PACING_RATE
:
1184 unsigned long ulval
= (val
== ~0U) ? ~0UL : val
;
1186 if (sizeof(ulval
) != sizeof(val
) &&
1187 optlen
>= sizeof(ulval
) &&
1188 copy_from_sockptr(&ulval
, optval
, sizeof(ulval
))) {
1193 cmpxchg(&sk
->sk_pacing_status
,
1196 sk
->sk_max_pacing_rate
= ulval
;
1197 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
, ulval
);
1200 case SO_INCOMING_CPU
:
1201 WRITE_ONCE(sk
->sk_incoming_cpu
, val
);
1206 dst_negative_advice(sk
);
1210 if (sk
->sk_family
== PF_INET
|| sk
->sk_family
== PF_INET6
) {
1211 if (!((sk
->sk_type
== SOCK_STREAM
&&
1212 sk
->sk_protocol
== IPPROTO_TCP
) ||
1213 (sk
->sk_type
== SOCK_DGRAM
&&
1214 sk
->sk_protocol
== IPPROTO_UDP
)))
1216 } else if (sk
->sk_family
!= PF_RDS
) {
1220 if (val
< 0 || val
> 1)
1223 sock_valbool_flag(sk
, SOCK_ZEROCOPY
, valbool
);
1228 if (optlen
!= sizeof(struct sock_txtime
)) {
1231 } else if (copy_from_sockptr(&sk_txtime
, optval
,
1232 sizeof(struct sock_txtime
))) {
1235 } else if (sk_txtime
.flags
& ~SOF_TXTIME_FLAGS_MASK
) {
1239 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1240 * scheduler has enough safe guards.
1242 if (sk_txtime
.clockid
!= CLOCK_MONOTONIC
&&
1243 !ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
)) {
1247 sock_valbool_flag(sk
, SOCK_TXTIME
, true);
1248 sk
->sk_clockid
= sk_txtime
.clockid
;
1249 sk
->sk_txtime_deadline_mode
=
1250 !!(sk_txtime
.flags
& SOF_TXTIME_DEADLINE_MODE
);
1251 sk
->sk_txtime_report_errors
=
1252 !!(sk_txtime
.flags
& SOF_TXTIME_REPORT_ERRORS
);
1255 case SO_BINDTOIFINDEX
:
1256 ret
= sock_bindtoindex_locked(sk
, val
);
1266 EXPORT_SYMBOL(sock_setsockopt
);
1269 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1270 struct ucred
*ucred
)
1272 ucred
->pid
= pid_vnr(pid
);
1273 ucred
->uid
= ucred
->gid
= -1;
1275 struct user_namespace
*current_ns
= current_user_ns();
1277 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1278 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1282 static int groups_to_user(gid_t __user
*dst
, const struct group_info
*src
)
1284 struct user_namespace
*user_ns
= current_user_ns();
1287 for (i
= 0; i
< src
->ngroups
; i
++)
1288 if (put_user(from_kgid_munged(user_ns
, src
->gid
[i
]), dst
+ i
))
1294 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1295 char __user
*optval
, int __user
*optlen
)
1297 struct sock
*sk
= sock
->sk
;
1302 unsigned long ulval
;
1304 struct old_timeval32 tm32
;
1305 struct __kernel_old_timeval tm
;
1306 struct __kernel_sock_timeval stm
;
1307 struct sock_txtime txtime
;
1310 int lv
= sizeof(int);
1313 if (get_user(len
, optlen
))
1318 memset(&v
, 0, sizeof(v
));
1322 v
.val
= sock_flag(sk
, SOCK_DBG
);
1326 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1330 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1334 v
.val
= sk
->sk_sndbuf
;
1338 v
.val
= sk
->sk_rcvbuf
;
1342 v
.val
= sk
->sk_reuse
;
1346 v
.val
= sk
->sk_reuseport
;
1350 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1354 v
.val
= sk
->sk_type
;
1358 v
.val
= sk
->sk_protocol
;
1362 v
.val
= sk
->sk_family
;
1366 v
.val
= -sock_error(sk
);
1368 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1372 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1376 v
.val
= sk
->sk_no_check_tx
;
1380 v
.val
= sk
->sk_priority
;
1384 lv
= sizeof(v
.ling
);
1385 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1386 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1390 sock_warn_obsolete_bsdism("getsockopt");
1393 case SO_TIMESTAMP_OLD
:
1394 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1395 !sock_flag(sk
, SOCK_TSTAMP_NEW
) &&
1396 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1399 case SO_TIMESTAMPNS_OLD
:
1400 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
) && !sock_flag(sk
, SOCK_TSTAMP_NEW
);
1403 case SO_TIMESTAMP_NEW
:
1404 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) && sock_flag(sk
, SOCK_TSTAMP_NEW
);
1407 case SO_TIMESTAMPNS_NEW
:
1408 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
) && sock_flag(sk
, SOCK_TSTAMP_NEW
);
1411 case SO_TIMESTAMPING_OLD
:
1412 v
.val
= sk
->sk_tsflags
;
1415 case SO_RCVTIMEO_OLD
:
1416 case SO_RCVTIMEO_NEW
:
1417 lv
= sock_get_timeout(sk
->sk_rcvtimeo
, &v
, SO_RCVTIMEO_OLD
== optname
);
1420 case SO_SNDTIMEO_OLD
:
1421 case SO_SNDTIMEO_NEW
:
1422 lv
= sock_get_timeout(sk
->sk_sndtimeo
, &v
, SO_SNDTIMEO_OLD
== optname
);
1426 v
.val
= sk
->sk_rcvlowat
;
1434 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1439 struct ucred peercred
;
1440 if (len
> sizeof(peercred
))
1441 len
= sizeof(peercred
);
1442 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1443 if (copy_to_user(optval
, &peercred
, len
))
1452 if (!sk
->sk_peer_cred
)
1455 n
= sk
->sk_peer_cred
->group_info
->ngroups
;
1456 if (len
< n
* sizeof(gid_t
)) {
1457 len
= n
* sizeof(gid_t
);
1458 return put_user(len
, optlen
) ? -EFAULT
: -ERANGE
;
1460 len
= n
* sizeof(gid_t
);
1462 ret
= groups_to_user((gid_t __user
*)optval
,
1463 sk
->sk_peer_cred
->group_info
);
1473 lv
= sock
->ops
->getname(sock
, (struct sockaddr
*)address
, 2);
1478 if (copy_to_user(optval
, address
, len
))
1483 /* Dubious BSD thing... Probably nobody even uses it, but
1484 * the UNIX standard wants it for whatever reason... -DaveM
1487 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1491 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1495 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1498 v
.val
= sk
->sk_mark
;
1502 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1505 case SO_WIFI_STATUS
:
1506 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1510 if (!sock
->ops
->set_peek_off
)
1513 v
.val
= sk
->sk_peek_off
;
1516 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1519 case SO_BINDTODEVICE
:
1520 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1523 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1529 case SO_LOCK_FILTER
:
1530 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1533 case SO_BPF_EXTENSIONS
:
1534 v
.val
= bpf_tell_extensions();
1537 case SO_SELECT_ERR_QUEUE
:
1538 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1541 #ifdef CONFIG_NET_RX_BUSY_POLL
1543 v
.val
= sk
->sk_ll_usec
;
1547 case SO_MAX_PACING_RATE
:
1548 if (sizeof(v
.ulval
) != sizeof(v
.val
) && len
>= sizeof(v
.ulval
)) {
1549 lv
= sizeof(v
.ulval
);
1550 v
.ulval
= sk
->sk_max_pacing_rate
;
1553 v
.val
= min_t(unsigned long, sk
->sk_max_pacing_rate
, ~0U);
1557 case SO_INCOMING_CPU
:
1558 v
.val
= READ_ONCE(sk
->sk_incoming_cpu
);
1563 u32 meminfo
[SK_MEMINFO_VARS
];
1565 sk_get_meminfo(sk
, meminfo
);
1567 len
= min_t(unsigned int, len
, sizeof(meminfo
));
1568 if (copy_to_user(optval
, &meminfo
, len
))
1574 #ifdef CONFIG_NET_RX_BUSY_POLL
1575 case SO_INCOMING_NAPI_ID
:
1576 v
.val
= READ_ONCE(sk
->sk_napi_id
);
1578 /* aggregate non-NAPI IDs down to 0 */
1579 if (v
.val
< MIN_NAPI_ID
)
1589 v
.val64
= sock_gen_cookie(sk
);
1593 v
.val
= sock_flag(sk
, SOCK_ZEROCOPY
);
1597 lv
= sizeof(v
.txtime
);
1598 v
.txtime
.clockid
= sk
->sk_clockid
;
1599 v
.txtime
.flags
|= sk
->sk_txtime_deadline_mode
?
1600 SOF_TXTIME_DEADLINE_MODE
: 0;
1601 v
.txtime
.flags
|= sk
->sk_txtime_report_errors
?
1602 SOF_TXTIME_REPORT_ERRORS
: 0;
1605 case SO_BINDTOIFINDEX
:
1606 v
.val
= sk
->sk_bound_dev_if
;
1610 /* We implement the SO_SNDLOWAT etc to not be settable
1613 return -ENOPROTOOPT
;
1618 if (copy_to_user(optval
, &v
, len
))
1621 if (put_user(len
, optlen
))
1627 * Initialize an sk_lock.
1629 * (We also register the sk_lock with the lock validator.)
1631 static inline void sock_lock_init(struct sock
*sk
)
1633 if (sk
->sk_kern_sock
)
1634 sock_lock_init_class_and_name(
1636 af_family_kern_slock_key_strings
[sk
->sk_family
],
1637 af_family_kern_slock_keys
+ sk
->sk_family
,
1638 af_family_kern_key_strings
[sk
->sk_family
],
1639 af_family_kern_keys
+ sk
->sk_family
);
1641 sock_lock_init_class_and_name(
1643 af_family_slock_key_strings
[sk
->sk_family
],
1644 af_family_slock_keys
+ sk
->sk_family
,
1645 af_family_key_strings
[sk
->sk_family
],
1646 af_family_keys
+ sk
->sk_family
);
1650 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1651 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1652 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1654 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1656 const struct proto
*prot
= READ_ONCE(osk
->sk_prot
);
1657 #ifdef CONFIG_SECURITY_NETWORK
1658 void *sptr
= nsk
->sk_security
;
1660 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1662 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1663 prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1665 #ifdef CONFIG_SECURITY_NETWORK
1666 nsk
->sk_security
= sptr
;
1667 security_sk_clone(osk
, nsk
);
1671 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1675 struct kmem_cache
*slab
;
1679 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1682 if (want_init_on_alloc(priority
))
1683 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1685 sk
= kmalloc(prot
->obj_size
, priority
);
1688 if (security_sk_alloc(sk
, family
, priority
))
1691 if (!try_module_get(prot
->owner
))
1693 sk_tx_queue_clear(sk
);
1699 security_sk_free(sk
);
1702 kmem_cache_free(slab
, sk
);
1708 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1710 struct kmem_cache
*slab
;
1711 struct module
*owner
;
1713 owner
= prot
->owner
;
1716 cgroup_sk_free(&sk
->sk_cgrp_data
);
1717 mem_cgroup_sk_free(sk
);
1718 security_sk_free(sk
);
1720 kmem_cache_free(slab
, sk
);
1727 * sk_alloc - All socket objects are allocated here
1728 * @net: the applicable net namespace
1729 * @family: protocol family
1730 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1731 * @prot: struct proto associated with this new sock instance
1732 * @kern: is this to be a kernel socket?
1734 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1735 struct proto
*prot
, int kern
)
1739 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1741 sk
->sk_family
= family
;
1743 * See comment in struct sock definition to understand
1744 * why we need sk_prot_creator -acme
1746 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1747 sk
->sk_kern_sock
= kern
;
1749 sk
->sk_net_refcnt
= kern
? 0 : 1;
1750 if (likely(sk
->sk_net_refcnt
)) {
1752 sock_inuse_add(net
, 1);
1755 sock_net_set(sk
, net
);
1756 refcount_set(&sk
->sk_wmem_alloc
, 1);
1758 mem_cgroup_sk_alloc(sk
);
1759 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1760 sock_update_classid(&sk
->sk_cgrp_data
);
1761 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1762 sk_tx_queue_clear(sk
);
1767 EXPORT_SYMBOL(sk_alloc
);
1769 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1770 * grace period. This is the case for UDP sockets and TCP listeners.
1772 static void __sk_destruct(struct rcu_head
*head
)
1774 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1775 struct sk_filter
*filter
;
1777 if (sk
->sk_destruct
)
1778 sk
->sk_destruct(sk
);
1780 filter
= rcu_dereference_check(sk
->sk_filter
,
1781 refcount_read(&sk
->sk_wmem_alloc
) == 0);
1783 sk_filter_uncharge(sk
, filter
);
1784 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1787 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1789 #ifdef CONFIG_BPF_SYSCALL
1790 bpf_sk_storage_free(sk
);
1793 if (atomic_read(&sk
->sk_omem_alloc
))
1794 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1795 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1797 if (sk
->sk_frag
.page
) {
1798 put_page(sk
->sk_frag
.page
);
1799 sk
->sk_frag
.page
= NULL
;
1802 if (sk
->sk_peer_cred
)
1803 put_cred(sk
->sk_peer_cred
);
1804 put_pid(sk
->sk_peer_pid
);
1805 if (likely(sk
->sk_net_refcnt
))
1806 put_net(sock_net(sk
));
1807 sk_prot_free(sk
->sk_prot_creator
, sk
);
1810 void sk_destruct(struct sock
*sk
)
1812 bool use_call_rcu
= sock_flag(sk
, SOCK_RCU_FREE
);
1814 if (rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1815 reuseport_detach_sock(sk
);
1816 use_call_rcu
= true;
1820 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1822 __sk_destruct(&sk
->sk_rcu
);
1825 static void __sk_free(struct sock
*sk
)
1827 if (likely(sk
->sk_net_refcnt
))
1828 sock_inuse_add(sock_net(sk
), -1);
1830 if (unlikely(sk
->sk_net_refcnt
&& sock_diag_has_destroy_listeners(sk
)))
1831 sock_diag_broadcast_destroy(sk
);
1836 void sk_free(struct sock
*sk
)
1839 * We subtract one from sk_wmem_alloc and can know if
1840 * some packets are still in some tx queue.
1841 * If not null, sock_wfree() will call __sk_free(sk) later
1843 if (refcount_dec_and_test(&sk
->sk_wmem_alloc
))
1846 EXPORT_SYMBOL(sk_free
);
1848 static void sk_init_common(struct sock
*sk
)
1850 skb_queue_head_init(&sk
->sk_receive_queue
);
1851 skb_queue_head_init(&sk
->sk_write_queue
);
1852 skb_queue_head_init(&sk
->sk_error_queue
);
1854 rwlock_init(&sk
->sk_callback_lock
);
1855 lockdep_set_class_and_name(&sk
->sk_receive_queue
.lock
,
1856 af_rlock_keys
+ sk
->sk_family
,
1857 af_family_rlock_key_strings
[sk
->sk_family
]);
1858 lockdep_set_class_and_name(&sk
->sk_write_queue
.lock
,
1859 af_wlock_keys
+ sk
->sk_family
,
1860 af_family_wlock_key_strings
[sk
->sk_family
]);
1861 lockdep_set_class_and_name(&sk
->sk_error_queue
.lock
,
1862 af_elock_keys
+ sk
->sk_family
,
1863 af_family_elock_key_strings
[sk
->sk_family
]);
1864 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
1865 af_callback_keys
+ sk
->sk_family
,
1866 af_family_clock_key_strings
[sk
->sk_family
]);
1870 * sk_clone_lock - clone a socket, and lock its clone
1871 * @sk: the socket to clone
1872 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1874 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1876 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1878 struct proto
*prot
= READ_ONCE(sk
->sk_prot
);
1880 bool is_charged
= true;
1882 newsk
= sk_prot_alloc(prot
, priority
, sk
->sk_family
);
1883 if (newsk
!= NULL
) {
1884 struct sk_filter
*filter
;
1886 sock_copy(newsk
, sk
);
1888 newsk
->sk_prot_creator
= prot
;
1891 if (likely(newsk
->sk_net_refcnt
))
1892 get_net(sock_net(newsk
));
1893 sk_node_init(&newsk
->sk_node
);
1894 sock_lock_init(newsk
);
1895 bh_lock_sock(newsk
);
1896 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1897 newsk
->sk_backlog
.len
= 0;
1899 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1901 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1903 refcount_set(&newsk
->sk_wmem_alloc
, 1);
1904 atomic_set(&newsk
->sk_omem_alloc
, 0);
1905 sk_init_common(newsk
);
1907 newsk
->sk_dst_cache
= NULL
;
1908 newsk
->sk_dst_pending_confirm
= 0;
1909 newsk
->sk_wmem_queued
= 0;
1910 newsk
->sk_forward_alloc
= 0;
1911 atomic_set(&newsk
->sk_drops
, 0);
1912 newsk
->sk_send_head
= NULL
;
1913 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1914 atomic_set(&newsk
->sk_zckey
, 0);
1916 sock_reset_flag(newsk
, SOCK_DONE
);
1918 /* sk->sk_memcg will be populated at accept() time */
1919 newsk
->sk_memcg
= NULL
;
1921 cgroup_sk_clone(&newsk
->sk_cgrp_data
);
1924 filter
= rcu_dereference(sk
->sk_filter
);
1926 /* though it's an empty new sock, the charging may fail
1927 * if sysctl_optmem_max was changed between creation of
1928 * original socket and cloning
1930 is_charged
= sk_filter_charge(newsk
, filter
);
1931 RCU_INIT_POINTER(newsk
->sk_filter
, filter
);
1934 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1935 /* We need to make sure that we don't uncharge the new
1936 * socket if we couldn't charge it in the first place
1937 * as otherwise we uncharge the parent's filter.
1940 RCU_INIT_POINTER(newsk
->sk_filter
, NULL
);
1941 sk_free_unlock_clone(newsk
);
1945 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1947 if (bpf_sk_storage_clone(sk
, newsk
)) {
1948 sk_free_unlock_clone(newsk
);
1953 /* Clear sk_user_data if parent had the pointer tagged
1954 * as not suitable for copying when cloning.
1956 if (sk_user_data_is_nocopy(newsk
))
1957 newsk
->sk_user_data
= NULL
;
1960 newsk
->sk_err_soft
= 0;
1961 newsk
->sk_priority
= 0;
1962 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1963 if (likely(newsk
->sk_net_refcnt
))
1964 sock_inuse_add(sock_net(newsk
), 1);
1967 * Before updating sk_refcnt, we must commit prior changes to memory
1968 * (Documentation/RCU/rculist_nulls.rst for details)
1971 refcount_set(&newsk
->sk_refcnt
, 2);
1974 * Increment the counter in the same struct proto as the master
1975 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1976 * is the same as sk->sk_prot->socks, as this field was copied
1979 * This _changes_ the previous behaviour, where
1980 * tcp_create_openreq_child always was incrementing the
1981 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1982 * to be taken into account in all callers. -acme
1984 sk_refcnt_debug_inc(newsk
);
1985 sk_set_socket(newsk
, NULL
);
1986 sk_tx_queue_clear(newsk
);
1987 RCU_INIT_POINTER(newsk
->sk_wq
, NULL
);
1989 if (newsk
->sk_prot
->sockets_allocated
)
1990 sk_sockets_allocated_inc(newsk
);
1992 if (sock_needs_netstamp(sk
) &&
1993 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1994 net_enable_timestamp();
1999 EXPORT_SYMBOL_GPL(sk_clone_lock
);
2001 void sk_free_unlock_clone(struct sock
*sk
)
2003 /* It is still raw copy of parent, so invalidate
2004 * destructor and make plain sk_free() */
2005 sk
->sk_destruct
= NULL
;
2009 EXPORT_SYMBOL_GPL(sk_free_unlock_clone
);
2011 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
2015 sk_dst_set(sk
, dst
);
2016 sk
->sk_route_caps
= dst
->dev
->features
| sk
->sk_route_forced_caps
;
2017 if (sk
->sk_route_caps
& NETIF_F_GSO
)
2018 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
2019 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
2020 if (sk_can_gso(sk
)) {
2021 if (dst
->header_len
&& !xfrm_dst_offload_ok(dst
)) {
2022 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
2024 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
2025 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
2026 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
2029 sk
->sk_gso_max_segs
= max_segs
;
2031 EXPORT_SYMBOL_GPL(sk_setup_caps
);
2034 * Simple resource managers for sockets.
2039 * Write buffer destructor automatically called from kfree_skb.
2041 void sock_wfree(struct sk_buff
*skb
)
2043 struct sock
*sk
= skb
->sk
;
2044 unsigned int len
= skb
->truesize
;
2046 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
2048 * Keep a reference on sk_wmem_alloc, this will be released
2049 * after sk_write_space() call
2051 WARN_ON(refcount_sub_and_test(len
- 1, &sk
->sk_wmem_alloc
));
2052 sk
->sk_write_space(sk
);
2056 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2057 * could not do because of in-flight packets
2059 if (refcount_sub_and_test(len
, &sk
->sk_wmem_alloc
))
2062 EXPORT_SYMBOL(sock_wfree
);
2064 /* This variant of sock_wfree() is used by TCP,
2065 * since it sets SOCK_USE_WRITE_QUEUE.
2067 void __sock_wfree(struct sk_buff
*skb
)
2069 struct sock
*sk
= skb
->sk
;
2071 if (refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
))
2075 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
2080 if (unlikely(!sk_fullsock(sk
))) {
2081 skb
->destructor
= sock_edemux
;
2086 skb
->destructor
= sock_wfree
;
2087 skb_set_hash_from_sk(skb
, sk
);
2089 * We used to take a refcount on sk, but following operation
2090 * is enough to guarantee sk_free() wont free this sock until
2091 * all in-flight packets are completed
2093 refcount_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
2095 EXPORT_SYMBOL(skb_set_owner_w
);
2097 static bool can_skb_orphan_partial(const struct sk_buff
*skb
)
2099 #ifdef CONFIG_TLS_DEVICE
2100 /* Drivers depend on in-order delivery for crypto offload,
2101 * partial orphan breaks out-of-order-OK logic.
2106 return (skb
->destructor
== sock_wfree
||
2107 (IS_ENABLED(CONFIG_INET
) && skb
->destructor
== tcp_wfree
));
2110 /* This helper is used by netem, as it can hold packets in its
2111 * delay queue. We want to allow the owner socket to send more
2112 * packets, as if they were already TX completed by a typical driver.
2113 * But we also want to keep skb->sk set because some packet schedulers
2114 * rely on it (sch_fq for example).
2116 void skb_orphan_partial(struct sk_buff
*skb
)
2118 if (skb_is_tcp_pure_ack(skb
))
2121 if (can_skb_orphan_partial(skb
)) {
2122 struct sock
*sk
= skb
->sk
;
2124 if (refcount_inc_not_zero(&sk
->sk_refcnt
)) {
2125 WARN_ON(refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
));
2126 skb
->destructor
= sock_efree
;
2132 EXPORT_SYMBOL(skb_orphan_partial
);
2135 * Read buffer destructor automatically called from kfree_skb.
2137 void sock_rfree(struct sk_buff
*skb
)
2139 struct sock
*sk
= skb
->sk
;
2140 unsigned int len
= skb
->truesize
;
2142 atomic_sub(len
, &sk
->sk_rmem_alloc
);
2143 sk_mem_uncharge(sk
, len
);
2145 EXPORT_SYMBOL(sock_rfree
);
2148 * Buffer destructor for skbs that are not used directly in read or write
2149 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2151 void sock_efree(struct sk_buff
*skb
)
2155 EXPORT_SYMBOL(sock_efree
);
2157 /* Buffer destructor for prefetch/receive path where reference count may
2158 * not be held, e.g. for listen sockets.
2161 void sock_pfree(struct sk_buff
*skb
)
2163 if (sk_is_refcounted(skb
->sk
))
2164 sock_gen_put(skb
->sk
);
2166 EXPORT_SYMBOL(sock_pfree
);
2167 #endif /* CONFIG_INET */
2169 kuid_t
sock_i_uid(struct sock
*sk
)
2173 read_lock_bh(&sk
->sk_callback_lock
);
2174 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
2175 read_unlock_bh(&sk
->sk_callback_lock
);
2178 EXPORT_SYMBOL(sock_i_uid
);
2180 unsigned long sock_i_ino(struct sock
*sk
)
2184 read_lock_bh(&sk
->sk_callback_lock
);
2185 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
2186 read_unlock_bh(&sk
->sk_callback_lock
);
2189 EXPORT_SYMBOL(sock_i_ino
);
2192 * Allocate a skb from the socket's send buffer.
2194 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
2198 refcount_read(&sk
->sk_wmem_alloc
) < READ_ONCE(sk
->sk_sndbuf
)) {
2199 struct sk_buff
*skb
= alloc_skb(size
, priority
);
2202 skb_set_owner_w(skb
, sk
);
2208 EXPORT_SYMBOL(sock_wmalloc
);
2210 static void sock_ofree(struct sk_buff
*skb
)
2212 struct sock
*sk
= skb
->sk
;
2214 atomic_sub(skb
->truesize
, &sk
->sk_omem_alloc
);
2217 struct sk_buff
*sock_omalloc(struct sock
*sk
, unsigned long size
,
2220 struct sk_buff
*skb
;
2222 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2223 if (atomic_read(&sk
->sk_omem_alloc
) + SKB_TRUESIZE(size
) >
2227 skb
= alloc_skb(size
, priority
);
2231 atomic_add(skb
->truesize
, &sk
->sk_omem_alloc
);
2233 skb
->destructor
= sock_ofree
;
2238 * Allocate a memory block from the socket's option memory buffer.
2240 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
2242 if ((unsigned int)size
<= sysctl_optmem_max
&&
2243 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
2245 /* First do the add, to avoid the race if kmalloc
2248 atomic_add(size
, &sk
->sk_omem_alloc
);
2249 mem
= kmalloc(size
, priority
);
2252 atomic_sub(size
, &sk
->sk_omem_alloc
);
2256 EXPORT_SYMBOL(sock_kmalloc
);
2258 /* Free an option memory block. Note, we actually want the inline
2259 * here as this allows gcc to detect the nullify and fold away the
2260 * condition entirely.
2262 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
2265 if (WARN_ON_ONCE(!mem
))
2268 kfree_sensitive(mem
);
2271 atomic_sub(size
, &sk
->sk_omem_alloc
);
2274 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
2276 __sock_kfree_s(sk
, mem
, size
, false);
2278 EXPORT_SYMBOL(sock_kfree_s
);
2280 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
2282 __sock_kfree_s(sk
, mem
, size
, true);
2284 EXPORT_SYMBOL(sock_kzfree_s
);
2286 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2287 I think, these locks should be removed for datagram sockets.
2289 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
2293 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2297 if (signal_pending(current
))
2299 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2300 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2301 if (refcount_read(&sk
->sk_wmem_alloc
) < READ_ONCE(sk
->sk_sndbuf
))
2303 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2307 timeo
= schedule_timeout(timeo
);
2309 finish_wait(sk_sleep(sk
), &wait
);
2315 * Generic send/receive buffer handlers
2318 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
2319 unsigned long data_len
, int noblock
,
2320 int *errcode
, int max_page_order
)
2322 struct sk_buff
*skb
;
2326 timeo
= sock_sndtimeo(sk
, noblock
);
2328 err
= sock_error(sk
);
2333 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2336 if (sk_wmem_alloc_get(sk
) < READ_ONCE(sk
->sk_sndbuf
))
2339 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2340 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2344 if (signal_pending(current
))
2346 timeo
= sock_wait_for_wmem(sk
, timeo
);
2348 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
2349 errcode
, sk
->sk_allocation
);
2351 skb_set_owner_w(skb
, sk
);
2355 err
= sock_intr_errno(timeo
);
2360 EXPORT_SYMBOL(sock_alloc_send_pskb
);
2362 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
2363 int noblock
, int *errcode
)
2365 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
2367 EXPORT_SYMBOL(sock_alloc_send_skb
);
2369 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
2370 struct sockcm_cookie
*sockc
)
2374 switch (cmsg
->cmsg_type
) {
2376 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
2378 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2380 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
2382 case SO_TIMESTAMPING_OLD
:
2383 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2386 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
2387 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
2390 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
2391 sockc
->tsflags
|= tsflags
;
2394 if (!sock_flag(sk
, SOCK_TXTIME
))
2396 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u64
)))
2398 sockc
->transmit_time
= get_unaligned((u64
*)CMSG_DATA(cmsg
));
2400 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2402 case SCM_CREDENTIALS
:
2409 EXPORT_SYMBOL(__sock_cmsg_send
);
2411 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
2412 struct sockcm_cookie
*sockc
)
2414 struct cmsghdr
*cmsg
;
2417 for_each_cmsghdr(cmsg
, msg
) {
2418 if (!CMSG_OK(msg
, cmsg
))
2420 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
2422 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
2428 EXPORT_SYMBOL(sock_cmsg_send
);
2430 static void sk_enter_memory_pressure(struct sock
*sk
)
2432 if (!sk
->sk_prot
->enter_memory_pressure
)
2435 sk
->sk_prot
->enter_memory_pressure(sk
);
2438 static void sk_leave_memory_pressure(struct sock
*sk
)
2440 if (sk
->sk_prot
->leave_memory_pressure
) {
2441 sk
->sk_prot
->leave_memory_pressure(sk
);
2443 unsigned long *memory_pressure
= sk
->sk_prot
->memory_pressure
;
2445 if (memory_pressure
&& READ_ONCE(*memory_pressure
))
2446 WRITE_ONCE(*memory_pressure
, 0);
2450 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2451 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key
);
2454 * skb_page_frag_refill - check that a page_frag contains enough room
2455 * @sz: minimum size of the fragment we want to get
2456 * @pfrag: pointer to page_frag
2457 * @gfp: priority for memory allocation
2459 * Note: While this allocator tries to use high order pages, there is
2460 * no guarantee that allocations succeed. Therefore, @sz MUST be
2461 * less or equal than PAGE_SIZE.
2463 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
2466 if (page_ref_count(pfrag
->page
) == 1) {
2470 if (pfrag
->offset
+ sz
<= pfrag
->size
)
2472 put_page(pfrag
->page
);
2476 if (SKB_FRAG_PAGE_ORDER
&&
2477 !static_branch_unlikely(&net_high_order_alloc_disable_key
)) {
2478 /* Avoid direct reclaim but allow kswapd to wake */
2479 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
2480 __GFP_COMP
| __GFP_NOWARN
|
2482 SKB_FRAG_PAGE_ORDER
);
2483 if (likely(pfrag
->page
)) {
2484 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
2488 pfrag
->page
= alloc_page(gfp
);
2489 if (likely(pfrag
->page
)) {
2490 pfrag
->size
= PAGE_SIZE
;
2495 EXPORT_SYMBOL(skb_page_frag_refill
);
2497 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
2499 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
2502 sk_enter_memory_pressure(sk
);
2503 sk_stream_moderate_sndbuf(sk
);
2506 EXPORT_SYMBOL(sk_page_frag_refill
);
2508 static void __lock_sock(struct sock
*sk
)
2509 __releases(&sk
->sk_lock
.slock
)
2510 __acquires(&sk
->sk_lock
.slock
)
2515 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2516 TASK_UNINTERRUPTIBLE
);
2517 spin_unlock_bh(&sk
->sk_lock
.slock
);
2519 spin_lock_bh(&sk
->sk_lock
.slock
);
2520 if (!sock_owned_by_user(sk
))
2523 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2526 void __release_sock(struct sock
*sk
)
2527 __releases(&sk
->sk_lock
.slock
)
2528 __acquires(&sk
->sk_lock
.slock
)
2530 struct sk_buff
*skb
, *next
;
2532 while ((skb
= sk
->sk_backlog
.head
) != NULL
) {
2533 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2535 spin_unlock_bh(&sk
->sk_lock
.slock
);
2540 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2541 skb_mark_not_on_list(skb
);
2542 sk_backlog_rcv(sk
, skb
);
2547 } while (skb
!= NULL
);
2549 spin_lock_bh(&sk
->sk_lock
.slock
);
2553 * Doing the zeroing here guarantee we can not loop forever
2554 * while a wild producer attempts to flood us.
2556 sk
->sk_backlog
.len
= 0;
2559 void __sk_flush_backlog(struct sock
*sk
)
2561 spin_lock_bh(&sk
->sk_lock
.slock
);
2563 spin_unlock_bh(&sk
->sk_lock
.slock
);
2567 * sk_wait_data - wait for data to arrive at sk_receive_queue
2568 * @sk: sock to wait on
2569 * @timeo: for how long
2570 * @skb: last skb seen on sk_receive_queue
2572 * Now socket state including sk->sk_err is changed only under lock,
2573 * hence we may omit checks after joining wait queue.
2574 * We check receive queue before schedule() only as optimization;
2575 * it is very likely that release_sock() added new data.
2577 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2579 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
2582 add_wait_queue(sk_sleep(sk
), &wait
);
2583 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2584 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
, &wait
);
2585 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2586 remove_wait_queue(sk_sleep(sk
), &wait
);
2589 EXPORT_SYMBOL(sk_wait_data
);
2592 * __sk_mem_raise_allocated - increase memory_allocated
2594 * @size: memory size to allocate
2595 * @amt: pages to allocate
2596 * @kind: allocation type
2598 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2600 int __sk_mem_raise_allocated(struct sock
*sk
, int size
, int amt
, int kind
)
2602 struct proto
*prot
= sk
->sk_prot
;
2603 long allocated
= sk_memory_allocated_add(sk
, amt
);
2604 bool charged
= true;
2606 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2607 !(charged
= mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
)))
2608 goto suppress_allocation
;
2611 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2612 sk_leave_memory_pressure(sk
);
2616 /* Under pressure. */
2617 if (allocated
> sk_prot_mem_limits(sk
, 1))
2618 sk_enter_memory_pressure(sk
);
2620 /* Over hard limit. */
2621 if (allocated
> sk_prot_mem_limits(sk
, 2))
2622 goto suppress_allocation
;
2624 /* guarantee minimum buffer size under pressure */
2625 if (kind
== SK_MEM_RECV
) {
2626 if (atomic_read(&sk
->sk_rmem_alloc
) < sk_get_rmem0(sk
, prot
))
2629 } else { /* SK_MEM_SEND */
2630 int wmem0
= sk_get_wmem0(sk
, prot
);
2632 if (sk
->sk_type
== SOCK_STREAM
) {
2633 if (sk
->sk_wmem_queued
< wmem0
)
2635 } else if (refcount_read(&sk
->sk_wmem_alloc
) < wmem0
) {
2640 if (sk_has_memory_pressure(sk
)) {
2643 if (!sk_under_memory_pressure(sk
))
2645 alloc
= sk_sockets_allocated_read_positive(sk
);
2646 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2647 sk_mem_pages(sk
->sk_wmem_queued
+
2648 atomic_read(&sk
->sk_rmem_alloc
) +
2649 sk
->sk_forward_alloc
))
2653 suppress_allocation
:
2655 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2656 sk_stream_moderate_sndbuf(sk
);
2658 /* Fail only if socket is _under_ its sndbuf.
2659 * In this case we cannot block, so that we have to fail.
2661 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2665 if (kind
== SK_MEM_SEND
|| (kind
== SK_MEM_RECV
&& charged
))
2666 trace_sock_exceed_buf_limit(sk
, prot
, allocated
, kind
);
2668 sk_memory_allocated_sub(sk
, amt
);
2670 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2671 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2675 EXPORT_SYMBOL(__sk_mem_raise_allocated
);
2678 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2680 * @size: memory size to allocate
2681 * @kind: allocation type
2683 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2684 * rmem allocation. This function assumes that protocols which have
2685 * memory_pressure use sk_wmem_queued as write buffer accounting.
2687 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2689 int ret
, amt
= sk_mem_pages(size
);
2691 sk
->sk_forward_alloc
+= amt
<< SK_MEM_QUANTUM_SHIFT
;
2692 ret
= __sk_mem_raise_allocated(sk
, size
, amt
, kind
);
2694 sk
->sk_forward_alloc
-= amt
<< SK_MEM_QUANTUM_SHIFT
;
2697 EXPORT_SYMBOL(__sk_mem_schedule
);
2700 * __sk_mem_reduce_allocated - reclaim memory_allocated
2702 * @amount: number of quanta
2704 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2706 void __sk_mem_reduce_allocated(struct sock
*sk
, int amount
)
2708 sk_memory_allocated_sub(sk
, amount
);
2710 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2711 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2713 if (sk_under_memory_pressure(sk
) &&
2714 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2715 sk_leave_memory_pressure(sk
);
2717 EXPORT_SYMBOL(__sk_mem_reduce_allocated
);
2720 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2722 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2724 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2726 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2727 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2728 __sk_mem_reduce_allocated(sk
, amount
);
2730 EXPORT_SYMBOL(__sk_mem_reclaim
);
2732 int sk_set_peek_off(struct sock
*sk
, int val
)
2734 sk
->sk_peek_off
= val
;
2737 EXPORT_SYMBOL_GPL(sk_set_peek_off
);
2740 * Set of default routines for initialising struct proto_ops when
2741 * the protocol does not support a particular function. In certain
2742 * cases where it makes no sense for a protocol to have a "do nothing"
2743 * function, some default processing is provided.
2746 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2750 EXPORT_SYMBOL(sock_no_bind
);
2752 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2757 EXPORT_SYMBOL(sock_no_connect
);
2759 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2763 EXPORT_SYMBOL(sock_no_socketpair
);
2765 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
,
2770 EXPORT_SYMBOL(sock_no_accept
);
2772 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2777 EXPORT_SYMBOL(sock_no_getname
);
2779 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2783 EXPORT_SYMBOL(sock_no_ioctl
);
2785 int sock_no_listen(struct socket
*sock
, int backlog
)
2789 EXPORT_SYMBOL(sock_no_listen
);
2791 int sock_no_shutdown(struct socket
*sock
, int how
)
2795 EXPORT_SYMBOL(sock_no_shutdown
);
2797 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2801 EXPORT_SYMBOL(sock_no_sendmsg
);
2803 int sock_no_sendmsg_locked(struct sock
*sk
, struct msghdr
*m
, size_t len
)
2807 EXPORT_SYMBOL(sock_no_sendmsg_locked
);
2809 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2814 EXPORT_SYMBOL(sock_no_recvmsg
);
2816 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2818 /* Mirror missing mmap method error code */
2821 EXPORT_SYMBOL(sock_no_mmap
);
2824 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2825 * various sock-based usage counts.
2827 void __receive_sock(struct file
*file
)
2829 struct socket
*sock
;
2833 * The resulting value of "error" is ignored here since we only
2834 * need to take action when the file is a socket and testing
2835 * "sock" for NULL is sufficient.
2837 sock
= sock_from_file(file
, &error
);
2839 sock_update_netprioidx(&sock
->sk
->sk_cgrp_data
);
2840 sock_update_classid(&sock
->sk
->sk_cgrp_data
);
2844 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2847 struct msghdr msg
= {.msg_flags
= flags
};
2849 char *kaddr
= kmap(page
);
2850 iov
.iov_base
= kaddr
+ offset
;
2852 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2856 EXPORT_SYMBOL(sock_no_sendpage
);
2858 ssize_t
sock_no_sendpage_locked(struct sock
*sk
, struct page
*page
,
2859 int offset
, size_t size
, int flags
)
2862 struct msghdr msg
= {.msg_flags
= flags
};
2864 char *kaddr
= kmap(page
);
2866 iov
.iov_base
= kaddr
+ offset
;
2868 res
= kernel_sendmsg_locked(sk
, &msg
, &iov
, 1, size
);
2872 EXPORT_SYMBOL(sock_no_sendpage_locked
);
2875 * Default Socket Callbacks
2878 static void sock_def_wakeup(struct sock
*sk
)
2880 struct socket_wq
*wq
;
2883 wq
= rcu_dereference(sk
->sk_wq
);
2884 if (skwq_has_sleeper(wq
))
2885 wake_up_interruptible_all(&wq
->wait
);
2889 static void sock_def_error_report(struct sock
*sk
)
2891 struct socket_wq
*wq
;
2894 wq
= rcu_dereference(sk
->sk_wq
);
2895 if (skwq_has_sleeper(wq
))
2896 wake_up_interruptible_poll(&wq
->wait
, EPOLLERR
);
2897 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2901 void sock_def_readable(struct sock
*sk
)
2903 struct socket_wq
*wq
;
2906 wq
= rcu_dereference(sk
->sk_wq
);
2907 if (skwq_has_sleeper(wq
))
2908 wake_up_interruptible_sync_poll(&wq
->wait
, EPOLLIN
| EPOLLPRI
|
2909 EPOLLRDNORM
| EPOLLRDBAND
);
2910 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2914 static void sock_def_write_space(struct sock
*sk
)
2916 struct socket_wq
*wq
;
2920 /* Do not wake up a writer until he can make "significant"
2923 if ((refcount_read(&sk
->sk_wmem_alloc
) << 1) <= READ_ONCE(sk
->sk_sndbuf
)) {
2924 wq
= rcu_dereference(sk
->sk_wq
);
2925 if (skwq_has_sleeper(wq
))
2926 wake_up_interruptible_sync_poll(&wq
->wait
, EPOLLOUT
|
2927 EPOLLWRNORM
| EPOLLWRBAND
);
2929 /* Should agree with poll, otherwise some programs break */
2930 if (sock_writeable(sk
))
2931 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2937 static void sock_def_destruct(struct sock
*sk
)
2941 void sk_send_sigurg(struct sock
*sk
)
2943 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2944 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2945 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2947 EXPORT_SYMBOL(sk_send_sigurg
);
2949 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2950 unsigned long expires
)
2952 if (!mod_timer(timer
, expires
))
2955 EXPORT_SYMBOL(sk_reset_timer
);
2957 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2959 if (del_timer(timer
))
2962 EXPORT_SYMBOL(sk_stop_timer
);
2964 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2967 sk
->sk_send_head
= NULL
;
2969 timer_setup(&sk
->sk_timer
, NULL
, 0);
2971 sk
->sk_allocation
= GFP_KERNEL
;
2972 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2973 sk
->sk_sndbuf
= sysctl_wmem_default
;
2974 sk
->sk_state
= TCP_CLOSE
;
2975 sk_set_socket(sk
, sock
);
2977 sock_set_flag(sk
, SOCK_ZAPPED
);
2980 sk
->sk_type
= sock
->type
;
2981 RCU_INIT_POINTER(sk
->sk_wq
, &sock
->wq
);
2983 sk
->sk_uid
= SOCK_INODE(sock
)->i_uid
;
2985 RCU_INIT_POINTER(sk
->sk_wq
, NULL
);
2986 sk
->sk_uid
= make_kuid(sock_net(sk
)->user_ns
, 0);
2989 rwlock_init(&sk
->sk_callback_lock
);
2990 if (sk
->sk_kern_sock
)
2991 lockdep_set_class_and_name(
2992 &sk
->sk_callback_lock
,
2993 af_kern_callback_keys
+ sk
->sk_family
,
2994 af_family_kern_clock_key_strings
[sk
->sk_family
]);
2996 lockdep_set_class_and_name(
2997 &sk
->sk_callback_lock
,
2998 af_callback_keys
+ sk
->sk_family
,
2999 af_family_clock_key_strings
[sk
->sk_family
]);
3001 sk
->sk_state_change
= sock_def_wakeup
;
3002 sk
->sk_data_ready
= sock_def_readable
;
3003 sk
->sk_write_space
= sock_def_write_space
;
3004 sk
->sk_error_report
= sock_def_error_report
;
3005 sk
->sk_destruct
= sock_def_destruct
;
3007 sk
->sk_frag
.page
= NULL
;
3008 sk
->sk_frag
.offset
= 0;
3009 sk
->sk_peek_off
= -1;
3011 sk
->sk_peer_pid
= NULL
;
3012 sk
->sk_peer_cred
= NULL
;
3013 sk
->sk_write_pending
= 0;
3014 sk
->sk_rcvlowat
= 1;
3015 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
3016 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
3018 sk
->sk_stamp
= SK_DEFAULT_STAMP
;
3019 #if BITS_PER_LONG==32
3020 seqlock_init(&sk
->sk_stamp_seq
);
3022 atomic_set(&sk
->sk_zckey
, 0);
3024 #ifdef CONFIG_NET_RX_BUSY_POLL
3026 sk
->sk_ll_usec
= sysctl_net_busy_read
;
3029 sk
->sk_max_pacing_rate
= ~0UL;
3030 sk
->sk_pacing_rate
= ~0UL;
3031 WRITE_ONCE(sk
->sk_pacing_shift
, 10);
3032 sk
->sk_incoming_cpu
= -1;
3034 sk_rx_queue_clear(sk
);
3036 * Before updating sk_refcnt, we must commit prior changes to memory
3037 * (Documentation/RCU/rculist_nulls.rst for details)
3040 refcount_set(&sk
->sk_refcnt
, 1);
3041 atomic_set(&sk
->sk_drops
, 0);
3043 EXPORT_SYMBOL(sock_init_data
);
3045 void lock_sock_nested(struct sock
*sk
, int subclass
)
3048 spin_lock_bh(&sk
->sk_lock
.slock
);
3049 if (sk
->sk_lock
.owned
)
3051 sk
->sk_lock
.owned
= 1;
3052 spin_unlock(&sk
->sk_lock
.slock
);
3054 * The sk_lock has mutex_lock() semantics here:
3056 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
3059 EXPORT_SYMBOL(lock_sock_nested
);
3061 void release_sock(struct sock
*sk
)
3063 spin_lock_bh(&sk
->sk_lock
.slock
);
3064 if (sk
->sk_backlog
.tail
)
3067 /* Warning : release_cb() might need to release sk ownership,
3068 * ie call sock_release_ownership(sk) before us.
3070 if (sk
->sk_prot
->release_cb
)
3071 sk
->sk_prot
->release_cb(sk
);
3073 sock_release_ownership(sk
);
3074 if (waitqueue_active(&sk
->sk_lock
.wq
))
3075 wake_up(&sk
->sk_lock
.wq
);
3076 spin_unlock_bh(&sk
->sk_lock
.slock
);
3078 EXPORT_SYMBOL(release_sock
);
3081 * lock_sock_fast - fast version of lock_sock
3084 * This version should be used for very small section, where process wont block
3085 * return false if fast path is taken:
3087 * sk_lock.slock locked, owned = 0, BH disabled
3089 * return true if slow path is taken:
3091 * sk_lock.slock unlocked, owned = 1, BH enabled
3093 bool lock_sock_fast(struct sock
*sk
)
3096 spin_lock_bh(&sk
->sk_lock
.slock
);
3098 if (!sk
->sk_lock
.owned
)
3100 * Note : We must disable BH
3105 sk
->sk_lock
.owned
= 1;
3106 spin_unlock(&sk
->sk_lock
.slock
);
3108 * The sk_lock has mutex_lock() semantics here:
3110 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
3114 EXPORT_SYMBOL(lock_sock_fast
);
3116 int sock_gettstamp(struct socket
*sock
, void __user
*userstamp
,
3117 bool timeval
, bool time32
)
3119 struct sock
*sk
= sock
->sk
;
3120 struct timespec64 ts
;
3122 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
3123 ts
= ktime_to_timespec64(sock_read_timestamp(sk
));
3124 if (ts
.tv_sec
== -1)
3126 if (ts
.tv_sec
== 0) {
3127 ktime_t kt
= ktime_get_real();
3128 sock_write_timestamp(sk
, kt
);
3129 ts
= ktime_to_timespec64(kt
);
3135 #ifdef CONFIG_COMPAT_32BIT_TIME
3137 return put_old_timespec32(&ts
, userstamp
);
3139 #ifdef CONFIG_SPARC64
3140 /* beware of padding in sparc64 timeval */
3141 if (timeval
&& !in_compat_syscall()) {
3142 struct __kernel_old_timeval __user tv
= {
3143 .tv_sec
= ts
.tv_sec
,
3144 .tv_usec
= ts
.tv_nsec
,
3146 if (copy_to_user(userstamp
, &tv
, sizeof(tv
)))
3151 return put_timespec64(&ts
, userstamp
);
3153 EXPORT_SYMBOL(sock_gettstamp
);
3155 void sock_enable_timestamp(struct sock
*sk
, enum sock_flags flag
)
3157 if (!sock_flag(sk
, flag
)) {
3158 unsigned long previous_flags
= sk
->sk_flags
;
3160 sock_set_flag(sk
, flag
);
3162 * we just set one of the two flags which require net
3163 * time stamping, but time stamping might have been on
3164 * already because of the other one
3166 if (sock_needs_netstamp(sk
) &&
3167 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
3168 net_enable_timestamp();
3172 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
3173 int level
, int type
)
3175 struct sock_exterr_skb
*serr
;
3176 struct sk_buff
*skb
;
3180 skb
= sock_dequeue_err_skb(sk
);
3186 msg
->msg_flags
|= MSG_TRUNC
;
3189 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
3193 sock_recv_timestamp(msg
, sk
, skb
);
3195 serr
= SKB_EXT_ERR(skb
);
3196 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
3198 msg
->msg_flags
|= MSG_ERRQUEUE
;
3206 EXPORT_SYMBOL(sock_recv_errqueue
);
3209 * Get a socket option on an socket.
3211 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3212 * asynchronous errors should be reported by getsockopt. We assume
3213 * this means if you specify SO_ERROR (otherwise whats the point of it).
3215 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
3216 char __user
*optval
, int __user
*optlen
)
3218 struct sock
*sk
= sock
->sk
;
3220 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
3222 EXPORT_SYMBOL(sock_common_getsockopt
);
3224 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
3227 struct sock
*sk
= sock
->sk
;
3231 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
3232 flags
& ~MSG_DONTWAIT
, &addr_len
);
3234 msg
->msg_namelen
= addr_len
;
3237 EXPORT_SYMBOL(sock_common_recvmsg
);
3240 * Set socket options on an inet socket.
3242 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
3243 sockptr_t optval
, unsigned int optlen
)
3245 struct sock
*sk
= sock
->sk
;
3247 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
3249 EXPORT_SYMBOL(sock_common_setsockopt
);
3251 void sk_common_release(struct sock
*sk
)
3253 if (sk
->sk_prot
->destroy
)
3254 sk
->sk_prot
->destroy(sk
);
3257 * Observation: when sk_common_release is called, processes have
3258 * no access to socket. But net still has.
3259 * Step one, detach it from networking:
3261 * A. Remove from hash tables.
3264 sk
->sk_prot
->unhash(sk
);
3267 * In this point socket cannot receive new packets, but it is possible
3268 * that some packets are in flight because some CPU runs receiver and
3269 * did hash table lookup before we unhashed socket. They will achieve
3270 * receive queue and will be purged by socket destructor.
3272 * Also we still have packets pending on receive queue and probably,
3273 * our own packets waiting in device queues. sock_destroy will drain
3274 * receive queue, but transmitted packets will delay socket destruction
3275 * until the last reference will be released.
3280 xfrm_sk_free_policy(sk
);
3282 sk_refcnt_debug_release(sk
);
3286 EXPORT_SYMBOL(sk_common_release
);
3288 void sk_get_meminfo(const struct sock
*sk
, u32
*mem
)
3290 memset(mem
, 0, sizeof(*mem
) * SK_MEMINFO_VARS
);
3292 mem
[SK_MEMINFO_RMEM_ALLOC
] = sk_rmem_alloc_get(sk
);
3293 mem
[SK_MEMINFO_RCVBUF
] = READ_ONCE(sk
->sk_rcvbuf
);
3294 mem
[SK_MEMINFO_WMEM_ALLOC
] = sk_wmem_alloc_get(sk
);
3295 mem
[SK_MEMINFO_SNDBUF
] = READ_ONCE(sk
->sk_sndbuf
);
3296 mem
[SK_MEMINFO_FWD_ALLOC
] = sk
->sk_forward_alloc
;
3297 mem
[SK_MEMINFO_WMEM_QUEUED
] = READ_ONCE(sk
->sk_wmem_queued
);
3298 mem
[SK_MEMINFO_OPTMEM
] = atomic_read(&sk
->sk_omem_alloc
);
3299 mem
[SK_MEMINFO_BACKLOG
] = READ_ONCE(sk
->sk_backlog
.len
);
3300 mem
[SK_MEMINFO_DROPS
] = atomic_read(&sk
->sk_drops
);
3303 #ifdef CONFIG_PROC_FS
3304 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3306 int val
[PROTO_INUSE_NR
];
3309 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
3311 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
3313 __this_cpu_add(net
->core
.prot_inuse
->val
[prot
->inuse_idx
], val
);
3315 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
3317 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
3319 int cpu
, idx
= prot
->inuse_idx
;
3322 for_each_possible_cpu(cpu
)
3323 res
+= per_cpu_ptr(net
->core
.prot_inuse
, cpu
)->val
[idx
];
3325 return res
>= 0 ? res
: 0;
3327 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
3329 static void sock_inuse_add(struct net
*net
, int val
)
3331 this_cpu_add(*net
->core
.sock_inuse
, val
);
3334 int sock_inuse_get(struct net
*net
)
3338 for_each_possible_cpu(cpu
)
3339 res
+= *per_cpu_ptr(net
->core
.sock_inuse
, cpu
);
3344 EXPORT_SYMBOL_GPL(sock_inuse_get
);
3346 static int __net_init
sock_inuse_init_net(struct net
*net
)
3348 net
->core
.prot_inuse
= alloc_percpu(struct prot_inuse
);
3349 if (net
->core
.prot_inuse
== NULL
)
3352 net
->core
.sock_inuse
= alloc_percpu(int);
3353 if (net
->core
.sock_inuse
== NULL
)
3359 free_percpu(net
->core
.prot_inuse
);
3363 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
3365 free_percpu(net
->core
.prot_inuse
);
3366 free_percpu(net
->core
.sock_inuse
);
3369 static struct pernet_operations net_inuse_ops
= {
3370 .init
= sock_inuse_init_net
,
3371 .exit
= sock_inuse_exit_net
,
3374 static __init
int net_inuse_init(void)
3376 if (register_pernet_subsys(&net_inuse_ops
))
3377 panic("Cannot initialize net inuse counters");
3382 core_initcall(net_inuse_init
);
3384 static int assign_proto_idx(struct proto
*prot
)
3386 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
3388 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
3389 pr_err("PROTO_INUSE_NR exhausted\n");
3393 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
3397 static void release_proto_idx(struct proto
*prot
)
3399 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
3400 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
3403 static inline int assign_proto_idx(struct proto
*prot
)
3408 static inline void release_proto_idx(struct proto
*prot
)
3412 static void sock_inuse_add(struct net
*net
, int val
)
3417 static void tw_prot_cleanup(struct timewait_sock_ops
*twsk_prot
)
3421 kfree(twsk_prot
->twsk_slab_name
);
3422 twsk_prot
->twsk_slab_name
= NULL
;
3423 kmem_cache_destroy(twsk_prot
->twsk_slab
);
3424 twsk_prot
->twsk_slab
= NULL
;
3427 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
3431 kfree(rsk_prot
->slab_name
);
3432 rsk_prot
->slab_name
= NULL
;
3433 kmem_cache_destroy(rsk_prot
->slab
);
3434 rsk_prot
->slab
= NULL
;
3437 static int req_prot_init(const struct proto
*prot
)
3439 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
3444 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
3446 if (!rsk_prot
->slab_name
)
3449 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
3450 rsk_prot
->obj_size
, 0,
3451 SLAB_ACCOUNT
| prot
->slab_flags
,
3454 if (!rsk_prot
->slab
) {
3455 pr_crit("%s: Can't create request sock SLAB cache!\n",
3462 int proto_register(struct proto
*prot
, int alloc_slab
)
3467 prot
->slab
= kmem_cache_create_usercopy(prot
->name
,
3469 SLAB_HWCACHE_ALIGN
| SLAB_ACCOUNT
|
3471 prot
->useroffset
, prot
->usersize
,
3474 if (prot
->slab
== NULL
) {
3475 pr_crit("%s: Can't create sock SLAB cache!\n",
3480 if (req_prot_init(prot
))
3481 goto out_free_request_sock_slab
;
3483 if (prot
->twsk_prot
!= NULL
) {
3484 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
3486 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
3487 goto out_free_request_sock_slab
;
3489 prot
->twsk_prot
->twsk_slab
=
3490 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
3491 prot
->twsk_prot
->twsk_obj_size
,
3496 if (prot
->twsk_prot
->twsk_slab
== NULL
)
3497 goto out_free_timewait_sock_slab
;
3501 mutex_lock(&proto_list_mutex
);
3502 ret
= assign_proto_idx(prot
);
3504 mutex_unlock(&proto_list_mutex
);
3505 goto out_free_timewait_sock_slab
;
3507 list_add(&prot
->node
, &proto_list
);
3508 mutex_unlock(&proto_list_mutex
);
3511 out_free_timewait_sock_slab
:
3512 if (alloc_slab
&& prot
->twsk_prot
)
3513 tw_prot_cleanup(prot
->twsk_prot
);
3514 out_free_request_sock_slab
:
3516 req_prot_cleanup(prot
->rsk_prot
);
3518 kmem_cache_destroy(prot
->slab
);
3524 EXPORT_SYMBOL(proto_register
);
3526 void proto_unregister(struct proto
*prot
)
3528 mutex_lock(&proto_list_mutex
);
3529 release_proto_idx(prot
);
3530 list_del(&prot
->node
);
3531 mutex_unlock(&proto_list_mutex
);
3533 kmem_cache_destroy(prot
->slab
);
3536 req_prot_cleanup(prot
->rsk_prot
);
3537 tw_prot_cleanup(prot
->twsk_prot
);
3539 EXPORT_SYMBOL(proto_unregister
);
3541 int sock_load_diag_module(int family
, int protocol
)
3544 if (!sock_is_registered(family
))
3547 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK
,
3548 NETLINK_SOCK_DIAG
, family
);
3552 if (family
== AF_INET
&&
3553 protocol
!= IPPROTO_RAW
&&
3554 protocol
< MAX_INET_PROTOS
&&
3555 !rcu_access_pointer(inet_protos
[protocol
]))
3559 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK
,
3560 NETLINK_SOCK_DIAG
, family
, protocol
);
3562 EXPORT_SYMBOL(sock_load_diag_module
);
3564 #ifdef CONFIG_PROC_FS
3565 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3566 __acquires(proto_list_mutex
)
3568 mutex_lock(&proto_list_mutex
);
3569 return seq_list_start_head(&proto_list
, *pos
);
3572 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3574 return seq_list_next(v
, &proto_list
, pos
);
3577 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
3578 __releases(proto_list_mutex
)
3580 mutex_unlock(&proto_list_mutex
);
3583 static char proto_method_implemented(const void *method
)
3585 return method
== NULL
? 'n' : 'y';
3587 static long sock_prot_memory_allocated(struct proto
*proto
)
3589 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
3592 static const char *sock_prot_memory_pressure(struct proto
*proto
)
3594 return proto
->memory_pressure
!= NULL
?
3595 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
3598 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
3601 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3602 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3605 sock_prot_inuse_get(seq_file_net(seq
), proto
),
3606 sock_prot_memory_allocated(proto
),
3607 sock_prot_memory_pressure(proto
),
3609 proto
->slab
== NULL
? "no" : "yes",
3610 module_name(proto
->owner
),
3611 proto_method_implemented(proto
->close
),
3612 proto_method_implemented(proto
->connect
),
3613 proto_method_implemented(proto
->disconnect
),
3614 proto_method_implemented(proto
->accept
),
3615 proto_method_implemented(proto
->ioctl
),
3616 proto_method_implemented(proto
->init
),
3617 proto_method_implemented(proto
->destroy
),
3618 proto_method_implemented(proto
->shutdown
),
3619 proto_method_implemented(proto
->setsockopt
),
3620 proto_method_implemented(proto
->getsockopt
),
3621 proto_method_implemented(proto
->sendmsg
),
3622 proto_method_implemented(proto
->recvmsg
),
3623 proto_method_implemented(proto
->sendpage
),
3624 proto_method_implemented(proto
->bind
),
3625 proto_method_implemented(proto
->backlog_rcv
),
3626 proto_method_implemented(proto
->hash
),
3627 proto_method_implemented(proto
->unhash
),
3628 proto_method_implemented(proto
->get_port
),
3629 proto_method_implemented(proto
->enter_memory_pressure
));
3632 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3634 if (v
== &proto_list
)
3635 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3644 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3646 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3650 static const struct seq_operations proto_seq_ops
= {
3651 .start
= proto_seq_start
,
3652 .next
= proto_seq_next
,
3653 .stop
= proto_seq_stop
,
3654 .show
= proto_seq_show
,
3657 static __net_init
int proto_init_net(struct net
*net
)
3659 if (!proc_create_net("protocols", 0444, net
->proc_net
, &proto_seq_ops
,
3660 sizeof(struct seq_net_private
)))
3666 static __net_exit
void proto_exit_net(struct net
*net
)
3668 remove_proc_entry("protocols", net
->proc_net
);
3672 static __net_initdata
struct pernet_operations proto_net_ops
= {
3673 .init
= proto_init_net
,
3674 .exit
= proto_exit_net
,
3677 static int __init
proto_init(void)
3679 return register_pernet_subsys(&proto_net_ops
);
3682 subsys_initcall(proto_init
);
3684 #endif /* PROC_FS */
3686 #ifdef CONFIG_NET_RX_BUSY_POLL
3687 bool sk_busy_loop_end(void *p
, unsigned long start_time
)
3689 struct sock
*sk
= p
;
3691 return !skb_queue_empty_lockless(&sk
->sk_receive_queue
) ||
3692 sk_busy_loop_timeout(sk
, start_time
);
3694 EXPORT_SYMBOL(sk_busy_loop_end
);
3695 #endif /* CONFIG_NET_RX_BUSY_POLL */
3697 int sock_bind_add(struct sock
*sk
, struct sockaddr
*addr
, int addr_len
)
3699 if (!sk
->sk_prot
->bind_add
)
3701 return sk
->sk_prot
->bind_add(sk
, addr
, addr_len
);
3703 EXPORT_SYMBOL(sock_bind_add
);