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>
117 #include <linux/uaccess.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <linux/net_tstamp.h>
126 #include <net/xfrm.h>
127 #include <linux/ipsec.h>
128 #include <net/cls_cgroup.h>
129 #include <net/netprio_cgroup.h>
130 #include <linux/sock_diag.h>
132 #include <linux/filter.h>
133 #include <net/sock_reuseport.h>
134 #include <net/bpf_sk_storage.h>
136 #include <trace/events/sock.h>
139 #include <net/busy_poll.h>
141 static DEFINE_MUTEX(proto_list_mutex
);
142 static LIST_HEAD(proto_list
);
144 static void sock_inuse_add(struct net
*net
, int val
);
147 * sk_ns_capable - General socket capability test
148 * @sk: Socket to use a capability on or through
149 * @user_ns: The user namespace of the capability to use
150 * @cap: The capability to use
152 * Test to see if the opener of the socket had when the socket was
153 * created and the current process has the capability @cap in the user
154 * namespace @user_ns.
156 bool sk_ns_capable(const struct sock
*sk
,
157 struct user_namespace
*user_ns
, int cap
)
159 return file_ns_capable(sk
->sk_socket
->file
, user_ns
, cap
) &&
160 ns_capable(user_ns
, cap
);
162 EXPORT_SYMBOL(sk_ns_capable
);
165 * sk_capable - Socket global capability test
166 * @sk: Socket to use a capability on or through
167 * @cap: The global capability to use
169 * Test to see if the opener of the socket had when the socket was
170 * created and the current process has the capability @cap in all user
173 bool sk_capable(const struct sock
*sk
, int cap
)
175 return sk_ns_capable(sk
, &init_user_ns
, cap
);
177 EXPORT_SYMBOL(sk_capable
);
180 * sk_net_capable - Network namespace socket capability test
181 * @sk: Socket to use a capability on or through
182 * @cap: The capability to use
184 * Test to see if the opener of the socket had when the socket was created
185 * and the current process has the capability @cap over the network namespace
186 * the socket is a member of.
188 bool sk_net_capable(const struct sock
*sk
, int cap
)
190 return sk_ns_capable(sk
, sock_net(sk
)->user_ns
, cap
);
192 EXPORT_SYMBOL(sk_net_capable
);
195 * Each address family might have different locking rules, so we have
196 * one slock key per address family and separate keys for internal and
199 static struct lock_class_key af_family_keys
[AF_MAX
];
200 static struct lock_class_key af_family_kern_keys
[AF_MAX
];
201 static struct lock_class_key af_family_slock_keys
[AF_MAX
];
202 static struct lock_class_key af_family_kern_slock_keys
[AF_MAX
];
205 * Make lock validator output more readable. (we pre-construct these
206 * strings build-time, so that runtime initialization of socket
210 #define _sock_locks(x) \
211 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
212 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
213 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
214 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
215 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
216 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
217 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
218 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
219 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
220 x "27" , x "28" , x "AF_CAN" , \
221 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
222 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
223 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
224 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
225 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
228 static const char *const af_family_key_strings
[AF_MAX
+1] = {
229 _sock_locks("sk_lock-")
231 static const char *const af_family_slock_key_strings
[AF_MAX
+1] = {
232 _sock_locks("slock-")
234 static const char *const af_family_clock_key_strings
[AF_MAX
+1] = {
235 _sock_locks("clock-")
238 static const char *const af_family_kern_key_strings
[AF_MAX
+1] = {
239 _sock_locks("k-sk_lock-")
241 static const char *const af_family_kern_slock_key_strings
[AF_MAX
+1] = {
242 _sock_locks("k-slock-")
244 static const char *const af_family_kern_clock_key_strings
[AF_MAX
+1] = {
245 _sock_locks("k-clock-")
247 static const char *const af_family_rlock_key_strings
[AF_MAX
+1] = {
248 _sock_locks("rlock-")
250 static const char *const af_family_wlock_key_strings
[AF_MAX
+1] = {
251 _sock_locks("wlock-")
253 static const char *const af_family_elock_key_strings
[AF_MAX
+1] = {
254 _sock_locks("elock-")
258 * sk_callback_lock and sk queues locking rules are per-address-family,
259 * so split the lock classes by using a per-AF key:
261 static struct lock_class_key af_callback_keys
[AF_MAX
];
262 static struct lock_class_key af_rlock_keys
[AF_MAX
];
263 static struct lock_class_key af_wlock_keys
[AF_MAX
];
264 static struct lock_class_key af_elock_keys
[AF_MAX
];
265 static struct lock_class_key af_kern_callback_keys
[AF_MAX
];
267 /* Run time adjustable parameters. */
268 __u32 sysctl_wmem_max __read_mostly
= SK_WMEM_MAX
;
269 EXPORT_SYMBOL(sysctl_wmem_max
);
270 __u32 sysctl_rmem_max __read_mostly
= SK_RMEM_MAX
;
271 EXPORT_SYMBOL(sysctl_rmem_max
);
272 __u32 sysctl_wmem_default __read_mostly
= SK_WMEM_MAX
;
273 __u32 sysctl_rmem_default __read_mostly
= SK_RMEM_MAX
;
275 /* Maximal space eaten by iovec or ancillary data plus some space */
276 int sysctl_optmem_max __read_mostly
= sizeof(unsigned long)*(2*UIO_MAXIOV
+512);
277 EXPORT_SYMBOL(sysctl_optmem_max
);
279 int sysctl_tstamp_allow_data __read_mostly
= 1;
281 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key
);
282 EXPORT_SYMBOL_GPL(memalloc_socks_key
);
285 * sk_set_memalloc - sets %SOCK_MEMALLOC
286 * @sk: socket to set it on
288 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
289 * It's the responsibility of the admin to adjust min_free_kbytes
290 * to meet the requirements
292 void sk_set_memalloc(struct sock
*sk
)
294 sock_set_flag(sk
, SOCK_MEMALLOC
);
295 sk
->sk_allocation
|= __GFP_MEMALLOC
;
296 static_branch_inc(&memalloc_socks_key
);
298 EXPORT_SYMBOL_GPL(sk_set_memalloc
);
300 void sk_clear_memalloc(struct sock
*sk
)
302 sock_reset_flag(sk
, SOCK_MEMALLOC
);
303 sk
->sk_allocation
&= ~__GFP_MEMALLOC
;
304 static_branch_dec(&memalloc_socks_key
);
307 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
308 * progress of swapping. SOCK_MEMALLOC may be cleared while
309 * it has rmem allocations due to the last swapfile being deactivated
310 * but there is a risk that the socket is unusable due to exceeding
311 * the rmem limits. Reclaim the reserves and obey rmem limits again.
315 EXPORT_SYMBOL_GPL(sk_clear_memalloc
);
317 int __sk_backlog_rcv(struct sock
*sk
, struct sk_buff
*skb
)
320 unsigned int noreclaim_flag
;
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk
, SOCK_MEMALLOC
));
325 noreclaim_flag
= memalloc_noreclaim_save();
326 ret
= sk
->sk_backlog_rcv(sk
, skb
);
327 memalloc_noreclaim_restore(noreclaim_flag
);
331 EXPORT_SYMBOL(__sk_backlog_rcv
);
333 static int sock_get_timeout(long timeo
, void *optval
, bool old_timeval
)
335 struct __kernel_sock_timeval tv
;
337 if (timeo
== MAX_SCHEDULE_TIMEOUT
) {
341 tv
.tv_sec
= timeo
/ HZ
;
342 tv
.tv_usec
= ((timeo
% HZ
) * USEC_PER_SEC
) / HZ
;
345 if (old_timeval
&& in_compat_syscall() && !COMPAT_USE_64BIT_TIME
) {
346 struct old_timeval32 tv32
= { tv
.tv_sec
, tv
.tv_usec
};
347 *(struct old_timeval32
*)optval
= tv32
;
352 struct __kernel_old_timeval old_tv
;
353 old_tv
.tv_sec
= tv
.tv_sec
;
354 old_tv
.tv_usec
= tv
.tv_usec
;
355 *(struct __kernel_old_timeval
*)optval
= old_tv
;
356 return sizeof(old_tv
);
359 *(struct __kernel_sock_timeval
*)optval
= tv
;
363 static int sock_set_timeout(long *timeo_p
, char __user
*optval
, int optlen
, bool old_timeval
)
365 struct __kernel_sock_timeval tv
;
367 if (old_timeval
&& in_compat_syscall() && !COMPAT_USE_64BIT_TIME
) {
368 struct old_timeval32 tv32
;
370 if (optlen
< sizeof(tv32
))
373 if (copy_from_user(&tv32
, optval
, sizeof(tv32
)))
375 tv
.tv_sec
= tv32
.tv_sec
;
376 tv
.tv_usec
= tv32
.tv_usec
;
377 } else if (old_timeval
) {
378 struct __kernel_old_timeval old_tv
;
380 if (optlen
< sizeof(old_tv
))
382 if (copy_from_user(&old_tv
, optval
, sizeof(old_tv
)))
384 tv
.tv_sec
= old_tv
.tv_sec
;
385 tv
.tv_usec
= old_tv
.tv_usec
;
387 if (optlen
< sizeof(tv
))
389 if (copy_from_user(&tv
, optval
, sizeof(tv
)))
392 if (tv
.tv_usec
< 0 || tv
.tv_usec
>= USEC_PER_SEC
)
396 static int warned __read_mostly
;
399 if (warned
< 10 && net_ratelimit()) {
401 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
402 __func__
, current
->comm
, task_pid_nr(current
));
406 *timeo_p
= MAX_SCHEDULE_TIMEOUT
;
407 if (tv
.tv_sec
== 0 && tv
.tv_usec
== 0)
409 if (tv
.tv_sec
< (MAX_SCHEDULE_TIMEOUT
/ HZ
- 1))
410 *timeo_p
= tv
.tv_sec
* HZ
+ DIV_ROUND_UP((unsigned long)tv
.tv_usec
, USEC_PER_SEC
/ HZ
);
414 static void sock_warn_obsolete_bsdism(const char *name
)
417 static char warncomm
[TASK_COMM_LEN
];
418 if (strcmp(warncomm
, current
->comm
) && warned
< 5) {
419 strcpy(warncomm
, current
->comm
);
420 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
426 static bool sock_needs_netstamp(const struct sock
*sk
)
428 switch (sk
->sk_family
) {
437 static void sock_disable_timestamp(struct sock
*sk
, unsigned long flags
)
439 if (sk
->sk_flags
& flags
) {
440 sk
->sk_flags
&= ~flags
;
441 if (sock_needs_netstamp(sk
) &&
442 !(sk
->sk_flags
& SK_FLAGS_TIMESTAMP
))
443 net_disable_timestamp();
448 int __sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
451 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
453 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
) {
454 atomic_inc(&sk
->sk_drops
);
455 trace_sock_rcvqueue_full(sk
, skb
);
459 if (!sk_rmem_schedule(sk
, skb
, skb
->truesize
)) {
460 atomic_inc(&sk
->sk_drops
);
465 skb_set_owner_r(skb
, sk
);
467 /* we escape from rcu protected region, make sure we dont leak
472 spin_lock_irqsave(&list
->lock
, flags
);
473 sock_skb_set_dropcount(sk
, skb
);
474 __skb_queue_tail(list
, skb
);
475 spin_unlock_irqrestore(&list
->lock
, flags
);
477 if (!sock_flag(sk
, SOCK_DEAD
))
478 sk
->sk_data_ready(sk
);
481 EXPORT_SYMBOL(__sock_queue_rcv_skb
);
483 int sock_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
487 err
= sk_filter(sk
, skb
);
491 return __sock_queue_rcv_skb(sk
, skb
);
493 EXPORT_SYMBOL(sock_queue_rcv_skb
);
495 int __sk_receive_skb(struct sock
*sk
, struct sk_buff
*skb
,
496 const int nested
, unsigned int trim_cap
, bool refcounted
)
498 int rc
= NET_RX_SUCCESS
;
500 if (sk_filter_trim_cap(sk
, skb
, trim_cap
))
501 goto discard_and_relse
;
505 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
506 atomic_inc(&sk
->sk_drops
);
507 goto discard_and_relse
;
510 bh_lock_sock_nested(sk
);
513 if (!sock_owned_by_user(sk
)) {
515 * trylock + unlock semantics:
517 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 1, _RET_IP_
);
519 rc
= sk_backlog_rcv(sk
, skb
);
521 mutex_release(&sk
->sk_lock
.dep_map
, _RET_IP_
);
522 } else if (sk_add_backlog(sk
, skb
, READ_ONCE(sk
->sk_rcvbuf
))) {
524 atomic_inc(&sk
->sk_drops
);
525 goto discard_and_relse
;
537 EXPORT_SYMBOL(__sk_receive_skb
);
539 struct dst_entry
*__sk_dst_check(struct sock
*sk
, u32 cookie
)
541 struct dst_entry
*dst
= __sk_dst_get(sk
);
543 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
544 sk_tx_queue_clear(sk
);
545 sk
->sk_dst_pending_confirm
= 0;
546 RCU_INIT_POINTER(sk
->sk_dst_cache
, NULL
);
553 EXPORT_SYMBOL(__sk_dst_check
);
555 struct dst_entry
*sk_dst_check(struct sock
*sk
, u32 cookie
)
557 struct dst_entry
*dst
= sk_dst_get(sk
);
559 if (dst
&& dst
->obsolete
&& dst
->ops
->check(dst
, cookie
) == NULL
) {
567 EXPORT_SYMBOL(sk_dst_check
);
569 static int sock_setbindtodevice_locked(struct sock
*sk
, int ifindex
)
571 int ret
= -ENOPROTOOPT
;
572 #ifdef CONFIG_NETDEVICES
573 struct net
*net
= sock_net(sk
);
577 if (sk
->sk_bound_dev_if
&& !ns_capable(net
->user_ns
, CAP_NET_RAW
))
584 sk
->sk_bound_dev_if
= ifindex
;
585 if (sk
->sk_prot
->rehash
)
586 sk
->sk_prot
->rehash(sk
);
597 static int sock_setbindtodevice(struct sock
*sk
, char __user
*optval
,
600 int ret
= -ENOPROTOOPT
;
601 #ifdef CONFIG_NETDEVICES
602 struct net
*net
= sock_net(sk
);
603 char devname
[IFNAMSIZ
];
610 /* Bind this socket to a particular device like "eth0",
611 * as specified in the passed interface name. If the
612 * name is "" or the option length is zero the socket
615 if (optlen
> IFNAMSIZ
- 1)
616 optlen
= IFNAMSIZ
- 1;
617 memset(devname
, 0, sizeof(devname
));
620 if (copy_from_user(devname
, optval
, optlen
))
624 if (devname
[0] != '\0') {
625 struct net_device
*dev
;
628 dev
= dev_get_by_name_rcu(net
, devname
);
630 index
= dev
->ifindex
;
638 ret
= sock_setbindtodevice_locked(sk
, index
);
647 static int sock_getbindtodevice(struct sock
*sk
, char __user
*optval
,
648 int __user
*optlen
, int len
)
650 int ret
= -ENOPROTOOPT
;
651 #ifdef CONFIG_NETDEVICES
652 struct net
*net
= sock_net(sk
);
653 char devname
[IFNAMSIZ
];
655 if (sk
->sk_bound_dev_if
== 0) {
664 ret
= netdev_get_name(net
, devname
, sk
->sk_bound_dev_if
);
668 len
= strlen(devname
) + 1;
671 if (copy_to_user(optval
, devname
, len
))
676 if (put_user(len
, optlen
))
687 static inline void sock_valbool_flag(struct sock
*sk
, enum sock_flags bit
,
691 sock_set_flag(sk
, bit
);
693 sock_reset_flag(sk
, bit
);
696 bool sk_mc_loop(struct sock
*sk
)
698 if (dev_recursion_level())
702 switch (sk
->sk_family
) {
704 return inet_sk(sk
)->mc_loop
;
705 #if IS_ENABLED(CONFIG_IPV6)
707 return inet6_sk(sk
)->mc_loop
;
713 EXPORT_SYMBOL(sk_mc_loop
);
716 * This is meant for all protocols to use and covers goings on
717 * at the socket level. Everything here is generic.
720 int sock_setsockopt(struct socket
*sock
, int level
, int optname
,
721 char __user
*optval
, unsigned int optlen
)
723 struct sock_txtime sk_txtime
;
724 struct sock
*sk
= sock
->sk
;
731 * Options without arguments
734 if (optname
== SO_BINDTODEVICE
)
735 return sock_setbindtodevice(sk
, optval
, optlen
);
737 if (optlen
< sizeof(int))
740 if (get_user(val
, (int __user
*)optval
))
743 valbool
= val
? 1 : 0;
749 if (val
&& !capable(CAP_NET_ADMIN
))
752 sock_valbool_flag(sk
, SOCK_DBG
, valbool
);
755 sk
->sk_reuse
= (valbool
? SK_CAN_REUSE
: SK_NO_REUSE
);
758 sk
->sk_reuseport
= valbool
;
767 sock_valbool_flag(sk
, SOCK_LOCALROUTE
, valbool
);
771 sock_valbool_flag(sk
, SOCK_BROADCAST
, valbool
);
774 /* Don't error on this BSD doesn't and if you think
775 * about it this is right. Otherwise apps have to
776 * play 'guess the biggest size' games. RCVBUF/SNDBUF
777 * are treated in BSD as hints
779 val
= min_t(u32
, val
, sysctl_wmem_max
);
781 /* Ensure val * 2 fits into an int, to prevent max_t()
782 * from treating it as a negative value.
784 val
= min_t(int, val
, INT_MAX
/ 2);
785 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
786 WRITE_ONCE(sk
->sk_sndbuf
,
787 max_t(int, val
* 2, SOCK_MIN_SNDBUF
));
788 /* Wake up sending tasks if we upped the value. */
789 sk
->sk_write_space(sk
);
793 if (!capable(CAP_NET_ADMIN
)) {
798 /* No negative values (to prevent underflow, as val will be
806 /* Don't error on this BSD doesn't and if you think
807 * about it this is right. Otherwise apps have to
808 * play 'guess the biggest size' games. RCVBUF/SNDBUF
809 * are treated in BSD as hints
811 val
= min_t(u32
, val
, sysctl_rmem_max
);
813 /* Ensure val * 2 fits into an int, to prevent max_t()
814 * from treating it as a negative value.
816 val
= min_t(int, val
, INT_MAX
/ 2);
817 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
819 * We double it on the way in to account for
820 * "struct sk_buff" etc. overhead. Applications
821 * assume that the SO_RCVBUF setting they make will
822 * allow that much actual data to be received on that
825 * Applications are unaware that "struct sk_buff" and
826 * other overheads allocate from the receive buffer
827 * during socket buffer allocation.
829 * And after considering the possible alternatives,
830 * returning the value we actually used in getsockopt
831 * is the most desirable behavior.
833 WRITE_ONCE(sk
->sk_rcvbuf
,
834 max_t(int, val
* 2, SOCK_MIN_RCVBUF
));
838 if (!capable(CAP_NET_ADMIN
)) {
843 /* No negative values (to prevent underflow, as val will be
851 if (sk
->sk_prot
->keepalive
)
852 sk
->sk_prot
->keepalive(sk
, valbool
);
853 sock_valbool_flag(sk
, SOCK_KEEPOPEN
, valbool
);
857 sock_valbool_flag(sk
, SOCK_URGINLINE
, valbool
);
861 sk
->sk_no_check_tx
= valbool
;
865 if ((val
>= 0 && val
<= 6) ||
866 ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
867 sk
->sk_priority
= val
;
873 if (optlen
< sizeof(ling
)) {
874 ret
= -EINVAL
; /* 1003.1g */
877 if (copy_from_user(&ling
, optval
, sizeof(ling
))) {
882 sock_reset_flag(sk
, SOCK_LINGER
);
884 #if (BITS_PER_LONG == 32)
885 if ((unsigned int)ling
.l_linger
>= MAX_SCHEDULE_TIMEOUT
/HZ
)
886 sk
->sk_lingertime
= MAX_SCHEDULE_TIMEOUT
;
889 sk
->sk_lingertime
= (unsigned int)ling
.l_linger
* HZ
;
890 sock_set_flag(sk
, SOCK_LINGER
);
895 sock_warn_obsolete_bsdism("setsockopt");
900 set_bit(SOCK_PASSCRED
, &sock
->flags
);
902 clear_bit(SOCK_PASSCRED
, &sock
->flags
);
905 case SO_TIMESTAMP_OLD
:
906 case SO_TIMESTAMP_NEW
:
907 case SO_TIMESTAMPNS_OLD
:
908 case SO_TIMESTAMPNS_NEW
:
910 if (optname
== SO_TIMESTAMP_NEW
|| optname
== SO_TIMESTAMPNS_NEW
)
911 sock_set_flag(sk
, SOCK_TSTAMP_NEW
);
913 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
915 if (optname
== SO_TIMESTAMP_OLD
|| optname
== SO_TIMESTAMP_NEW
)
916 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
918 sock_set_flag(sk
, SOCK_RCVTSTAMPNS
);
919 sock_set_flag(sk
, SOCK_RCVTSTAMP
);
920 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
922 sock_reset_flag(sk
, SOCK_RCVTSTAMP
);
923 sock_reset_flag(sk
, SOCK_RCVTSTAMPNS
);
924 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
928 case SO_TIMESTAMPING_NEW
:
929 sock_set_flag(sk
, SOCK_TSTAMP_NEW
);
931 case SO_TIMESTAMPING_OLD
:
932 if (val
& ~SOF_TIMESTAMPING_MASK
) {
937 if (val
& SOF_TIMESTAMPING_OPT_ID
&&
938 !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)) {
939 if (sk
->sk_protocol
== IPPROTO_TCP
&&
940 sk
->sk_type
== SOCK_STREAM
) {
941 if ((1 << sk
->sk_state
) &
942 (TCPF_CLOSE
| TCPF_LISTEN
)) {
946 sk
->sk_tskey
= tcp_sk(sk
)->snd_una
;
952 if (val
& SOF_TIMESTAMPING_OPT_STATS
&&
953 !(val
& SOF_TIMESTAMPING_OPT_TSONLY
)) {
958 sk
->sk_tsflags
= val
;
959 if (val
& SOF_TIMESTAMPING_RX_SOFTWARE
)
960 sock_enable_timestamp(sk
,
961 SOCK_TIMESTAMPING_RX_SOFTWARE
);
963 if (optname
== SO_TIMESTAMPING_NEW
)
964 sock_reset_flag(sk
, SOCK_TSTAMP_NEW
);
966 sock_disable_timestamp(sk
,
967 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE
));
974 if (sock
->ops
->set_rcvlowat
)
975 ret
= sock
->ops
->set_rcvlowat(sk
, val
);
977 WRITE_ONCE(sk
->sk_rcvlowat
, val
? : 1);
980 case SO_RCVTIMEO_OLD
:
981 case SO_RCVTIMEO_NEW
:
982 ret
= sock_set_timeout(&sk
->sk_rcvtimeo
, optval
, optlen
, optname
== SO_RCVTIMEO_OLD
);
985 case SO_SNDTIMEO_OLD
:
986 case SO_SNDTIMEO_NEW
:
987 ret
= sock_set_timeout(&sk
->sk_sndtimeo
, optval
, optlen
, optname
== SO_SNDTIMEO_OLD
);
990 case SO_ATTACH_FILTER
:
992 if (optlen
== sizeof(struct sock_fprog
)) {
993 struct sock_fprog fprog
;
996 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
999 ret
= sk_attach_filter(&fprog
, sk
);
1005 if (optlen
== sizeof(u32
)) {
1009 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
1012 ret
= sk_attach_bpf(ufd
, sk
);
1016 case SO_ATTACH_REUSEPORT_CBPF
:
1018 if (optlen
== sizeof(struct sock_fprog
)) {
1019 struct sock_fprog fprog
;
1022 if (copy_from_user(&fprog
, optval
, sizeof(fprog
)))
1025 ret
= sk_reuseport_attach_filter(&fprog
, sk
);
1029 case SO_ATTACH_REUSEPORT_EBPF
:
1031 if (optlen
== sizeof(u32
)) {
1035 if (copy_from_user(&ufd
, optval
, sizeof(ufd
)))
1038 ret
= sk_reuseport_attach_bpf(ufd
, sk
);
1042 case SO_DETACH_REUSEPORT_BPF
:
1043 ret
= reuseport_detach_prog(sk
);
1046 case SO_DETACH_FILTER
:
1047 ret
= sk_detach_filter(sk
);
1050 case SO_LOCK_FILTER
:
1051 if (sock_flag(sk
, SOCK_FILTER_LOCKED
) && !valbool
)
1054 sock_valbool_flag(sk
, SOCK_FILTER_LOCKED
, valbool
);
1059 set_bit(SOCK_PASSSEC
, &sock
->flags
);
1061 clear_bit(SOCK_PASSSEC
, &sock
->flags
);
1064 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
)) {
1066 } else if (val
!= sk
->sk_mark
) {
1073 sock_valbool_flag(sk
, SOCK_RXQ_OVFL
, valbool
);
1076 case SO_WIFI_STATUS
:
1077 sock_valbool_flag(sk
, SOCK_WIFI_STATUS
, valbool
);
1081 if (sock
->ops
->set_peek_off
)
1082 ret
= sock
->ops
->set_peek_off(sk
, val
);
1088 sock_valbool_flag(sk
, SOCK_NOFCS
, valbool
);
1091 case SO_SELECT_ERR_QUEUE
:
1092 sock_valbool_flag(sk
, SOCK_SELECT_ERR_QUEUE
, valbool
);
1095 #ifdef CONFIG_NET_RX_BUSY_POLL
1097 /* allow unprivileged users to decrease the value */
1098 if ((val
> sk
->sk_ll_usec
) && !capable(CAP_NET_ADMIN
))
1104 sk
->sk_ll_usec
= val
;
1109 case SO_MAX_PACING_RATE
:
1111 unsigned long ulval
= (val
== ~0U) ? ~0UL : val
;
1113 if (sizeof(ulval
) != sizeof(val
) &&
1114 optlen
>= sizeof(ulval
) &&
1115 get_user(ulval
, (unsigned long __user
*)optval
)) {
1120 cmpxchg(&sk
->sk_pacing_status
,
1123 sk
->sk_max_pacing_rate
= ulval
;
1124 sk
->sk_pacing_rate
= min(sk
->sk_pacing_rate
, ulval
);
1127 case SO_INCOMING_CPU
:
1128 WRITE_ONCE(sk
->sk_incoming_cpu
, val
);
1133 dst_negative_advice(sk
);
1137 if (sk
->sk_family
== PF_INET
|| sk
->sk_family
== PF_INET6
) {
1138 if (!((sk
->sk_type
== SOCK_STREAM
&&
1139 sk
->sk_protocol
== IPPROTO_TCP
) ||
1140 (sk
->sk_type
== SOCK_DGRAM
&&
1141 sk
->sk_protocol
== IPPROTO_UDP
)))
1143 } else if (sk
->sk_family
!= PF_RDS
) {
1147 if (val
< 0 || val
> 1)
1150 sock_valbool_flag(sk
, SOCK_ZEROCOPY
, valbool
);
1155 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
)) {
1157 } else if (optlen
!= sizeof(struct sock_txtime
)) {
1159 } else if (copy_from_user(&sk_txtime
, optval
,
1160 sizeof(struct sock_txtime
))) {
1162 } else if (sk_txtime
.flags
& ~SOF_TXTIME_FLAGS_MASK
) {
1165 sock_valbool_flag(sk
, SOCK_TXTIME
, true);
1166 sk
->sk_clockid
= sk_txtime
.clockid
;
1167 sk
->sk_txtime_deadline_mode
=
1168 !!(sk_txtime
.flags
& SOF_TXTIME_DEADLINE_MODE
);
1169 sk
->sk_txtime_report_errors
=
1170 !!(sk_txtime
.flags
& SOF_TXTIME_REPORT_ERRORS
);
1174 case SO_BINDTOIFINDEX
:
1175 ret
= sock_setbindtodevice_locked(sk
, val
);
1185 EXPORT_SYMBOL(sock_setsockopt
);
1188 static void cred_to_ucred(struct pid
*pid
, const struct cred
*cred
,
1189 struct ucred
*ucred
)
1191 ucred
->pid
= pid_vnr(pid
);
1192 ucred
->uid
= ucred
->gid
= -1;
1194 struct user_namespace
*current_ns
= current_user_ns();
1196 ucred
->uid
= from_kuid_munged(current_ns
, cred
->euid
);
1197 ucred
->gid
= from_kgid_munged(current_ns
, cred
->egid
);
1201 static int groups_to_user(gid_t __user
*dst
, const struct group_info
*src
)
1203 struct user_namespace
*user_ns
= current_user_ns();
1206 for (i
= 0; i
< src
->ngroups
; i
++)
1207 if (put_user(from_kgid_munged(user_ns
, src
->gid
[i
]), dst
+ i
))
1213 int sock_getsockopt(struct socket
*sock
, int level
, int optname
,
1214 char __user
*optval
, int __user
*optlen
)
1216 struct sock
*sk
= sock
->sk
;
1221 unsigned long ulval
;
1223 struct old_timeval32 tm32
;
1224 struct __kernel_old_timeval tm
;
1225 struct __kernel_sock_timeval stm
;
1226 struct sock_txtime txtime
;
1229 int lv
= sizeof(int);
1232 if (get_user(len
, optlen
))
1237 memset(&v
, 0, sizeof(v
));
1241 v
.val
= sock_flag(sk
, SOCK_DBG
);
1245 v
.val
= sock_flag(sk
, SOCK_LOCALROUTE
);
1249 v
.val
= sock_flag(sk
, SOCK_BROADCAST
);
1253 v
.val
= sk
->sk_sndbuf
;
1257 v
.val
= sk
->sk_rcvbuf
;
1261 v
.val
= sk
->sk_reuse
;
1265 v
.val
= sk
->sk_reuseport
;
1269 v
.val
= sock_flag(sk
, SOCK_KEEPOPEN
);
1273 v
.val
= sk
->sk_type
;
1277 v
.val
= sk
->sk_protocol
;
1281 v
.val
= sk
->sk_family
;
1285 v
.val
= -sock_error(sk
);
1287 v
.val
= xchg(&sk
->sk_err_soft
, 0);
1291 v
.val
= sock_flag(sk
, SOCK_URGINLINE
);
1295 v
.val
= sk
->sk_no_check_tx
;
1299 v
.val
= sk
->sk_priority
;
1303 lv
= sizeof(v
.ling
);
1304 v
.ling
.l_onoff
= sock_flag(sk
, SOCK_LINGER
);
1305 v
.ling
.l_linger
= sk
->sk_lingertime
/ HZ
;
1309 sock_warn_obsolete_bsdism("getsockopt");
1312 case SO_TIMESTAMP_OLD
:
1313 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) &&
1314 !sock_flag(sk
, SOCK_TSTAMP_NEW
) &&
1315 !sock_flag(sk
, SOCK_RCVTSTAMPNS
);
1318 case SO_TIMESTAMPNS_OLD
:
1319 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
) && !sock_flag(sk
, SOCK_TSTAMP_NEW
);
1322 case SO_TIMESTAMP_NEW
:
1323 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMP
) && sock_flag(sk
, SOCK_TSTAMP_NEW
);
1326 case SO_TIMESTAMPNS_NEW
:
1327 v
.val
= sock_flag(sk
, SOCK_RCVTSTAMPNS
) && sock_flag(sk
, SOCK_TSTAMP_NEW
);
1330 case SO_TIMESTAMPING_OLD
:
1331 v
.val
= sk
->sk_tsflags
;
1334 case SO_RCVTIMEO_OLD
:
1335 case SO_RCVTIMEO_NEW
:
1336 lv
= sock_get_timeout(sk
->sk_rcvtimeo
, &v
, SO_RCVTIMEO_OLD
== optname
);
1339 case SO_SNDTIMEO_OLD
:
1340 case SO_SNDTIMEO_NEW
:
1341 lv
= sock_get_timeout(sk
->sk_sndtimeo
, &v
, SO_SNDTIMEO_OLD
== optname
);
1345 v
.val
= sk
->sk_rcvlowat
;
1353 v
.val
= !!test_bit(SOCK_PASSCRED
, &sock
->flags
);
1358 struct ucred peercred
;
1359 if (len
> sizeof(peercred
))
1360 len
= sizeof(peercred
);
1361 cred_to_ucred(sk
->sk_peer_pid
, sk
->sk_peer_cred
, &peercred
);
1362 if (copy_to_user(optval
, &peercred
, len
))
1371 if (!sk
->sk_peer_cred
)
1374 n
= sk
->sk_peer_cred
->group_info
->ngroups
;
1375 if (len
< n
* sizeof(gid_t
)) {
1376 len
= n
* sizeof(gid_t
);
1377 return put_user(len
, optlen
) ? -EFAULT
: -ERANGE
;
1379 len
= n
* sizeof(gid_t
);
1381 ret
= groups_to_user((gid_t __user
*)optval
,
1382 sk
->sk_peer_cred
->group_info
);
1392 lv
= sock
->ops
->getname(sock
, (struct sockaddr
*)address
, 2);
1397 if (copy_to_user(optval
, address
, len
))
1402 /* Dubious BSD thing... Probably nobody even uses it, but
1403 * the UNIX standard wants it for whatever reason... -DaveM
1406 v
.val
= sk
->sk_state
== TCP_LISTEN
;
1410 v
.val
= !!test_bit(SOCK_PASSSEC
, &sock
->flags
);
1414 return security_socket_getpeersec_stream(sock
, optval
, optlen
, len
);
1417 v
.val
= sk
->sk_mark
;
1421 v
.val
= sock_flag(sk
, SOCK_RXQ_OVFL
);
1424 case SO_WIFI_STATUS
:
1425 v
.val
= sock_flag(sk
, SOCK_WIFI_STATUS
);
1429 if (!sock
->ops
->set_peek_off
)
1432 v
.val
= sk
->sk_peek_off
;
1435 v
.val
= sock_flag(sk
, SOCK_NOFCS
);
1438 case SO_BINDTODEVICE
:
1439 return sock_getbindtodevice(sk
, optval
, optlen
, len
);
1442 len
= sk_get_filter(sk
, (struct sock_filter __user
*)optval
, len
);
1448 case SO_LOCK_FILTER
:
1449 v
.val
= sock_flag(sk
, SOCK_FILTER_LOCKED
);
1452 case SO_BPF_EXTENSIONS
:
1453 v
.val
= bpf_tell_extensions();
1456 case SO_SELECT_ERR_QUEUE
:
1457 v
.val
= sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
);
1460 #ifdef CONFIG_NET_RX_BUSY_POLL
1462 v
.val
= sk
->sk_ll_usec
;
1466 case SO_MAX_PACING_RATE
:
1467 if (sizeof(v
.ulval
) != sizeof(v
.val
) && len
>= sizeof(v
.ulval
)) {
1468 lv
= sizeof(v
.ulval
);
1469 v
.ulval
= sk
->sk_max_pacing_rate
;
1472 v
.val
= min_t(unsigned long, sk
->sk_max_pacing_rate
, ~0U);
1476 case SO_INCOMING_CPU
:
1477 v
.val
= READ_ONCE(sk
->sk_incoming_cpu
);
1482 u32 meminfo
[SK_MEMINFO_VARS
];
1484 sk_get_meminfo(sk
, meminfo
);
1486 len
= min_t(unsigned int, len
, sizeof(meminfo
));
1487 if (copy_to_user(optval
, &meminfo
, len
))
1493 #ifdef CONFIG_NET_RX_BUSY_POLL
1494 case SO_INCOMING_NAPI_ID
:
1495 v
.val
= READ_ONCE(sk
->sk_napi_id
);
1497 /* aggregate non-NAPI IDs down to 0 */
1498 if (v
.val
< MIN_NAPI_ID
)
1508 v
.val64
= sock_gen_cookie(sk
);
1512 v
.val
= sock_flag(sk
, SOCK_ZEROCOPY
);
1516 lv
= sizeof(v
.txtime
);
1517 v
.txtime
.clockid
= sk
->sk_clockid
;
1518 v
.txtime
.flags
|= sk
->sk_txtime_deadline_mode
?
1519 SOF_TXTIME_DEADLINE_MODE
: 0;
1520 v
.txtime
.flags
|= sk
->sk_txtime_report_errors
?
1521 SOF_TXTIME_REPORT_ERRORS
: 0;
1524 case SO_BINDTOIFINDEX
:
1525 v
.val
= sk
->sk_bound_dev_if
;
1529 /* We implement the SO_SNDLOWAT etc to not be settable
1532 return -ENOPROTOOPT
;
1537 if (copy_to_user(optval
, &v
, len
))
1540 if (put_user(len
, optlen
))
1546 * Initialize an sk_lock.
1548 * (We also register the sk_lock with the lock validator.)
1550 static inline void sock_lock_init(struct sock
*sk
)
1552 if (sk
->sk_kern_sock
)
1553 sock_lock_init_class_and_name(
1555 af_family_kern_slock_key_strings
[sk
->sk_family
],
1556 af_family_kern_slock_keys
+ sk
->sk_family
,
1557 af_family_kern_key_strings
[sk
->sk_family
],
1558 af_family_kern_keys
+ sk
->sk_family
);
1560 sock_lock_init_class_and_name(
1562 af_family_slock_key_strings
[sk
->sk_family
],
1563 af_family_slock_keys
+ sk
->sk_family
,
1564 af_family_key_strings
[sk
->sk_family
],
1565 af_family_keys
+ sk
->sk_family
);
1569 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1570 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1571 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1573 static void sock_copy(struct sock
*nsk
, const struct sock
*osk
)
1575 const struct proto
*prot
= READ_ONCE(osk
->sk_prot
);
1576 #ifdef CONFIG_SECURITY_NETWORK
1577 void *sptr
= nsk
->sk_security
;
1579 memcpy(nsk
, osk
, offsetof(struct sock
, sk_dontcopy_begin
));
1581 memcpy(&nsk
->sk_dontcopy_end
, &osk
->sk_dontcopy_end
,
1582 prot
->obj_size
- offsetof(struct sock
, sk_dontcopy_end
));
1584 #ifdef CONFIG_SECURITY_NETWORK
1585 nsk
->sk_security
= sptr
;
1586 security_sk_clone(osk
, nsk
);
1590 static struct sock
*sk_prot_alloc(struct proto
*prot
, gfp_t priority
,
1594 struct kmem_cache
*slab
;
1598 sk
= kmem_cache_alloc(slab
, priority
& ~__GFP_ZERO
);
1601 if (want_init_on_alloc(priority
))
1602 sk_prot_clear_nulls(sk
, prot
->obj_size
);
1604 sk
= kmalloc(prot
->obj_size
, priority
);
1607 if (security_sk_alloc(sk
, family
, priority
))
1610 if (!try_module_get(prot
->owner
))
1612 sk_tx_queue_clear(sk
);
1618 security_sk_free(sk
);
1621 kmem_cache_free(slab
, sk
);
1627 static void sk_prot_free(struct proto
*prot
, struct sock
*sk
)
1629 struct kmem_cache
*slab
;
1630 struct module
*owner
;
1632 owner
= prot
->owner
;
1635 cgroup_sk_free(&sk
->sk_cgrp_data
);
1636 mem_cgroup_sk_free(sk
);
1637 security_sk_free(sk
);
1639 kmem_cache_free(slab
, sk
);
1646 * sk_alloc - All socket objects are allocated here
1647 * @net: the applicable net namespace
1648 * @family: protocol family
1649 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1650 * @prot: struct proto associated with this new sock instance
1651 * @kern: is this to be a kernel socket?
1653 struct sock
*sk_alloc(struct net
*net
, int family
, gfp_t priority
,
1654 struct proto
*prot
, int kern
)
1658 sk
= sk_prot_alloc(prot
, priority
| __GFP_ZERO
, family
);
1660 sk
->sk_family
= family
;
1662 * See comment in struct sock definition to understand
1663 * why we need sk_prot_creator -acme
1665 sk
->sk_prot
= sk
->sk_prot_creator
= prot
;
1666 sk
->sk_kern_sock
= kern
;
1668 sk
->sk_net_refcnt
= kern
? 0 : 1;
1669 if (likely(sk
->sk_net_refcnt
)) {
1671 sock_inuse_add(net
, 1);
1674 sock_net_set(sk
, net
);
1675 refcount_set(&sk
->sk_wmem_alloc
, 1);
1677 mem_cgroup_sk_alloc(sk
);
1678 cgroup_sk_alloc(&sk
->sk_cgrp_data
);
1679 sock_update_classid(&sk
->sk_cgrp_data
);
1680 sock_update_netprioidx(&sk
->sk_cgrp_data
);
1685 EXPORT_SYMBOL(sk_alloc
);
1687 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1688 * grace period. This is the case for UDP sockets and TCP listeners.
1690 static void __sk_destruct(struct rcu_head
*head
)
1692 struct sock
*sk
= container_of(head
, struct sock
, sk_rcu
);
1693 struct sk_filter
*filter
;
1695 if (sk
->sk_destruct
)
1696 sk
->sk_destruct(sk
);
1698 filter
= rcu_dereference_check(sk
->sk_filter
,
1699 refcount_read(&sk
->sk_wmem_alloc
) == 0);
1701 sk_filter_uncharge(sk
, filter
);
1702 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1705 sock_disable_timestamp(sk
, SK_FLAGS_TIMESTAMP
);
1707 #ifdef CONFIG_BPF_SYSCALL
1708 bpf_sk_storage_free(sk
);
1711 if (atomic_read(&sk
->sk_omem_alloc
))
1712 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1713 __func__
, atomic_read(&sk
->sk_omem_alloc
));
1715 if (sk
->sk_frag
.page
) {
1716 put_page(sk
->sk_frag
.page
);
1717 sk
->sk_frag
.page
= NULL
;
1720 if (sk
->sk_peer_cred
)
1721 put_cred(sk
->sk_peer_cred
);
1722 put_pid(sk
->sk_peer_pid
);
1723 if (likely(sk
->sk_net_refcnt
))
1724 put_net(sock_net(sk
));
1725 sk_prot_free(sk
->sk_prot_creator
, sk
);
1728 void sk_destruct(struct sock
*sk
)
1730 bool use_call_rcu
= sock_flag(sk
, SOCK_RCU_FREE
);
1732 if (rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1733 reuseport_detach_sock(sk
);
1734 use_call_rcu
= true;
1738 call_rcu(&sk
->sk_rcu
, __sk_destruct
);
1740 __sk_destruct(&sk
->sk_rcu
);
1743 static void __sk_free(struct sock
*sk
)
1745 if (likely(sk
->sk_net_refcnt
))
1746 sock_inuse_add(sock_net(sk
), -1);
1748 if (unlikely(sk
->sk_net_refcnt
&& sock_diag_has_destroy_listeners(sk
)))
1749 sock_diag_broadcast_destroy(sk
);
1754 void sk_free(struct sock
*sk
)
1757 * We subtract one from sk_wmem_alloc and can know if
1758 * some packets are still in some tx queue.
1759 * If not null, sock_wfree() will call __sk_free(sk) later
1761 if (refcount_dec_and_test(&sk
->sk_wmem_alloc
))
1764 EXPORT_SYMBOL(sk_free
);
1766 static void sk_init_common(struct sock
*sk
)
1768 skb_queue_head_init(&sk
->sk_receive_queue
);
1769 skb_queue_head_init(&sk
->sk_write_queue
);
1770 skb_queue_head_init(&sk
->sk_error_queue
);
1772 rwlock_init(&sk
->sk_callback_lock
);
1773 lockdep_set_class_and_name(&sk
->sk_receive_queue
.lock
,
1774 af_rlock_keys
+ sk
->sk_family
,
1775 af_family_rlock_key_strings
[sk
->sk_family
]);
1776 lockdep_set_class_and_name(&sk
->sk_write_queue
.lock
,
1777 af_wlock_keys
+ sk
->sk_family
,
1778 af_family_wlock_key_strings
[sk
->sk_family
]);
1779 lockdep_set_class_and_name(&sk
->sk_error_queue
.lock
,
1780 af_elock_keys
+ sk
->sk_family
,
1781 af_family_elock_key_strings
[sk
->sk_family
]);
1782 lockdep_set_class_and_name(&sk
->sk_callback_lock
,
1783 af_callback_keys
+ sk
->sk_family
,
1784 af_family_clock_key_strings
[sk
->sk_family
]);
1788 * sk_clone_lock - clone a socket, and lock its clone
1789 * @sk: the socket to clone
1790 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1792 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1794 struct sock
*sk_clone_lock(const struct sock
*sk
, const gfp_t priority
)
1796 struct proto
*prot
= READ_ONCE(sk
->sk_prot
);
1798 bool is_charged
= true;
1800 newsk
= sk_prot_alloc(prot
, priority
, sk
->sk_family
);
1801 if (newsk
!= NULL
) {
1802 struct sk_filter
*filter
;
1804 sock_copy(newsk
, sk
);
1806 newsk
->sk_prot_creator
= prot
;
1809 if (likely(newsk
->sk_net_refcnt
))
1810 get_net(sock_net(newsk
));
1811 sk_node_init(&newsk
->sk_node
);
1812 sock_lock_init(newsk
);
1813 bh_lock_sock(newsk
);
1814 newsk
->sk_backlog
.head
= newsk
->sk_backlog
.tail
= NULL
;
1815 newsk
->sk_backlog
.len
= 0;
1817 atomic_set(&newsk
->sk_rmem_alloc
, 0);
1819 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1821 refcount_set(&newsk
->sk_wmem_alloc
, 1);
1822 atomic_set(&newsk
->sk_omem_alloc
, 0);
1823 sk_init_common(newsk
);
1825 newsk
->sk_dst_cache
= NULL
;
1826 newsk
->sk_dst_pending_confirm
= 0;
1827 newsk
->sk_wmem_queued
= 0;
1828 newsk
->sk_forward_alloc
= 0;
1829 atomic_set(&newsk
->sk_drops
, 0);
1830 newsk
->sk_send_head
= NULL
;
1831 newsk
->sk_userlocks
= sk
->sk_userlocks
& ~SOCK_BINDPORT_LOCK
;
1832 atomic_set(&newsk
->sk_zckey
, 0);
1834 sock_reset_flag(newsk
, SOCK_DONE
);
1836 /* sk->sk_memcg will be populated at accept() time */
1837 newsk
->sk_memcg
= NULL
;
1839 cgroup_sk_alloc(&newsk
->sk_cgrp_data
);
1842 filter
= rcu_dereference(sk
->sk_filter
);
1844 /* though it's an empty new sock, the charging may fail
1845 * if sysctl_optmem_max was changed between creation of
1846 * original socket and cloning
1848 is_charged
= sk_filter_charge(newsk
, filter
);
1849 RCU_INIT_POINTER(newsk
->sk_filter
, filter
);
1852 if (unlikely(!is_charged
|| xfrm_sk_clone_policy(newsk
, sk
))) {
1853 /* We need to make sure that we don't uncharge the new
1854 * socket if we couldn't charge it in the first place
1855 * as otherwise we uncharge the parent's filter.
1858 RCU_INIT_POINTER(newsk
->sk_filter
, NULL
);
1859 sk_free_unlock_clone(newsk
);
1863 RCU_INIT_POINTER(newsk
->sk_reuseport_cb
, NULL
);
1865 if (bpf_sk_storage_clone(sk
, newsk
)) {
1866 sk_free_unlock_clone(newsk
);
1871 /* Clear sk_user_data if parent had the pointer tagged
1872 * as not suitable for copying when cloning.
1874 if (sk_user_data_is_nocopy(newsk
))
1875 newsk
->sk_user_data
= NULL
;
1878 newsk
->sk_err_soft
= 0;
1879 newsk
->sk_priority
= 0;
1880 newsk
->sk_incoming_cpu
= raw_smp_processor_id();
1881 if (likely(newsk
->sk_net_refcnt
))
1882 sock_inuse_add(sock_net(newsk
), 1);
1885 * Before updating sk_refcnt, we must commit prior changes to memory
1886 * (Documentation/RCU/rculist_nulls.txt for details)
1889 refcount_set(&newsk
->sk_refcnt
, 2);
1892 * Increment the counter in the same struct proto as the master
1893 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1894 * is the same as sk->sk_prot->socks, as this field was copied
1897 * This _changes_ the previous behaviour, where
1898 * tcp_create_openreq_child always was incrementing the
1899 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1900 * to be taken into account in all callers. -acme
1902 sk_refcnt_debug_inc(newsk
);
1903 sk_set_socket(newsk
, NULL
);
1904 RCU_INIT_POINTER(newsk
->sk_wq
, NULL
);
1906 if (newsk
->sk_prot
->sockets_allocated
)
1907 sk_sockets_allocated_inc(newsk
);
1909 if (sock_needs_netstamp(sk
) &&
1910 newsk
->sk_flags
& SK_FLAGS_TIMESTAMP
)
1911 net_enable_timestamp();
1916 EXPORT_SYMBOL_GPL(sk_clone_lock
);
1918 void sk_free_unlock_clone(struct sock
*sk
)
1920 /* It is still raw copy of parent, so invalidate
1921 * destructor and make plain sk_free() */
1922 sk
->sk_destruct
= NULL
;
1926 EXPORT_SYMBOL_GPL(sk_free_unlock_clone
);
1928 void sk_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1932 sk_dst_set(sk
, dst
);
1933 sk
->sk_route_caps
= dst
->dev
->features
| sk
->sk_route_forced_caps
;
1934 if (sk
->sk_route_caps
& NETIF_F_GSO
)
1935 sk
->sk_route_caps
|= NETIF_F_GSO_SOFTWARE
;
1936 sk
->sk_route_caps
&= ~sk
->sk_route_nocaps
;
1937 if (sk_can_gso(sk
)) {
1938 if (dst
->header_len
&& !xfrm_dst_offload_ok(dst
)) {
1939 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1941 sk
->sk_route_caps
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1942 sk
->sk_gso_max_size
= dst
->dev
->gso_max_size
;
1943 max_segs
= max_t(u32
, dst
->dev
->gso_max_segs
, 1);
1946 sk
->sk_gso_max_segs
= max_segs
;
1948 EXPORT_SYMBOL_GPL(sk_setup_caps
);
1951 * Simple resource managers for sockets.
1956 * Write buffer destructor automatically called from kfree_skb.
1958 void sock_wfree(struct sk_buff
*skb
)
1960 struct sock
*sk
= skb
->sk
;
1961 unsigned int len
= skb
->truesize
;
1963 if (!sock_flag(sk
, SOCK_USE_WRITE_QUEUE
)) {
1965 * Keep a reference on sk_wmem_alloc, this will be released
1966 * after sk_write_space() call
1968 WARN_ON(refcount_sub_and_test(len
- 1, &sk
->sk_wmem_alloc
));
1969 sk
->sk_write_space(sk
);
1973 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1974 * could not do because of in-flight packets
1976 if (refcount_sub_and_test(len
, &sk
->sk_wmem_alloc
))
1979 EXPORT_SYMBOL(sock_wfree
);
1981 /* This variant of sock_wfree() is used by TCP,
1982 * since it sets SOCK_USE_WRITE_QUEUE.
1984 void __sock_wfree(struct sk_buff
*skb
)
1986 struct sock
*sk
= skb
->sk
;
1988 if (refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
))
1992 void skb_set_owner_w(struct sk_buff
*skb
, struct sock
*sk
)
1997 if (unlikely(!sk_fullsock(sk
))) {
1998 skb
->destructor
= sock_edemux
;
2003 skb
->destructor
= sock_wfree
;
2004 skb_set_hash_from_sk(skb
, sk
);
2006 * We used to take a refcount on sk, but following operation
2007 * is enough to guarantee sk_free() wont free this sock until
2008 * all in-flight packets are completed
2010 refcount_add(skb
->truesize
, &sk
->sk_wmem_alloc
);
2012 EXPORT_SYMBOL(skb_set_owner_w
);
2014 static bool can_skb_orphan_partial(const struct sk_buff
*skb
)
2016 #ifdef CONFIG_TLS_DEVICE
2017 /* Drivers depend on in-order delivery for crypto offload,
2018 * partial orphan breaks out-of-order-OK logic.
2023 return (skb
->destructor
== sock_wfree
||
2024 (IS_ENABLED(CONFIG_INET
) && skb
->destructor
== tcp_wfree
));
2027 /* This helper is used by netem, as it can hold packets in its
2028 * delay queue. We want to allow the owner socket to send more
2029 * packets, as if they were already TX completed by a typical driver.
2030 * But we also want to keep skb->sk set because some packet schedulers
2031 * rely on it (sch_fq for example).
2033 void skb_orphan_partial(struct sk_buff
*skb
)
2035 if (skb_is_tcp_pure_ack(skb
))
2038 if (can_skb_orphan_partial(skb
)) {
2039 struct sock
*sk
= skb
->sk
;
2041 if (refcount_inc_not_zero(&sk
->sk_refcnt
)) {
2042 WARN_ON(refcount_sub_and_test(skb
->truesize
, &sk
->sk_wmem_alloc
));
2043 skb
->destructor
= sock_efree
;
2049 EXPORT_SYMBOL(skb_orphan_partial
);
2052 * Read buffer destructor automatically called from kfree_skb.
2054 void sock_rfree(struct sk_buff
*skb
)
2056 struct sock
*sk
= skb
->sk
;
2057 unsigned int len
= skb
->truesize
;
2059 atomic_sub(len
, &sk
->sk_rmem_alloc
);
2060 sk_mem_uncharge(sk
, len
);
2062 EXPORT_SYMBOL(sock_rfree
);
2065 * Buffer destructor for skbs that are not used directly in read or write
2066 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2068 void sock_efree(struct sk_buff
*skb
)
2072 EXPORT_SYMBOL(sock_efree
);
2074 /* Buffer destructor for prefetch/receive path where reference count may
2075 * not be held, e.g. for listen sockets.
2078 void sock_pfree(struct sk_buff
*skb
)
2080 if (sk_is_refcounted(skb
->sk
))
2081 sock_gen_put(skb
->sk
);
2083 EXPORT_SYMBOL(sock_pfree
);
2084 #endif /* CONFIG_INET */
2086 kuid_t
sock_i_uid(struct sock
*sk
)
2090 read_lock_bh(&sk
->sk_callback_lock
);
2091 uid
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_uid
: GLOBAL_ROOT_UID
;
2092 read_unlock_bh(&sk
->sk_callback_lock
);
2095 EXPORT_SYMBOL(sock_i_uid
);
2097 unsigned long sock_i_ino(struct sock
*sk
)
2101 read_lock_bh(&sk
->sk_callback_lock
);
2102 ino
= sk
->sk_socket
? SOCK_INODE(sk
->sk_socket
)->i_ino
: 0;
2103 read_unlock_bh(&sk
->sk_callback_lock
);
2106 EXPORT_SYMBOL(sock_i_ino
);
2109 * Allocate a skb from the socket's send buffer.
2111 struct sk_buff
*sock_wmalloc(struct sock
*sk
, unsigned long size
, int force
,
2115 refcount_read(&sk
->sk_wmem_alloc
) < READ_ONCE(sk
->sk_sndbuf
)) {
2116 struct sk_buff
*skb
= alloc_skb(size
, priority
);
2119 skb_set_owner_w(skb
, sk
);
2125 EXPORT_SYMBOL(sock_wmalloc
);
2127 static void sock_ofree(struct sk_buff
*skb
)
2129 struct sock
*sk
= skb
->sk
;
2131 atomic_sub(skb
->truesize
, &sk
->sk_omem_alloc
);
2134 struct sk_buff
*sock_omalloc(struct sock
*sk
, unsigned long size
,
2137 struct sk_buff
*skb
;
2139 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2140 if (atomic_read(&sk
->sk_omem_alloc
) + SKB_TRUESIZE(size
) >
2144 skb
= alloc_skb(size
, priority
);
2148 atomic_add(skb
->truesize
, &sk
->sk_omem_alloc
);
2150 skb
->destructor
= sock_ofree
;
2155 * Allocate a memory block from the socket's option memory buffer.
2157 void *sock_kmalloc(struct sock
*sk
, int size
, gfp_t priority
)
2159 if ((unsigned int)size
<= sysctl_optmem_max
&&
2160 atomic_read(&sk
->sk_omem_alloc
) + size
< sysctl_optmem_max
) {
2162 /* First do the add, to avoid the race if kmalloc
2165 atomic_add(size
, &sk
->sk_omem_alloc
);
2166 mem
= kmalloc(size
, priority
);
2169 atomic_sub(size
, &sk
->sk_omem_alloc
);
2173 EXPORT_SYMBOL(sock_kmalloc
);
2175 /* Free an option memory block. Note, we actually want the inline
2176 * here as this allows gcc to detect the nullify and fold away the
2177 * condition entirely.
2179 static inline void __sock_kfree_s(struct sock
*sk
, void *mem
, int size
,
2182 if (WARN_ON_ONCE(!mem
))
2188 atomic_sub(size
, &sk
->sk_omem_alloc
);
2191 void sock_kfree_s(struct sock
*sk
, void *mem
, int size
)
2193 __sock_kfree_s(sk
, mem
, size
, false);
2195 EXPORT_SYMBOL(sock_kfree_s
);
2197 void sock_kzfree_s(struct sock
*sk
, void *mem
, int size
)
2199 __sock_kfree_s(sk
, mem
, size
, true);
2201 EXPORT_SYMBOL(sock_kzfree_s
);
2203 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2204 I think, these locks should be removed for datagram sockets.
2206 static long sock_wait_for_wmem(struct sock
*sk
, long timeo
)
2210 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2214 if (signal_pending(current
))
2216 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2217 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2218 if (refcount_read(&sk
->sk_wmem_alloc
) < READ_ONCE(sk
->sk_sndbuf
))
2220 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2224 timeo
= schedule_timeout(timeo
);
2226 finish_wait(sk_sleep(sk
), &wait
);
2232 * Generic send/receive buffer handlers
2235 struct sk_buff
*sock_alloc_send_pskb(struct sock
*sk
, unsigned long header_len
,
2236 unsigned long data_len
, int noblock
,
2237 int *errcode
, int max_page_order
)
2239 struct sk_buff
*skb
;
2243 timeo
= sock_sndtimeo(sk
, noblock
);
2245 err
= sock_error(sk
);
2250 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
2253 if (sk_wmem_alloc_get(sk
) < READ_ONCE(sk
->sk_sndbuf
))
2256 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2257 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
2261 if (signal_pending(current
))
2263 timeo
= sock_wait_for_wmem(sk
, timeo
);
2265 skb
= alloc_skb_with_frags(header_len
, data_len
, max_page_order
,
2266 errcode
, sk
->sk_allocation
);
2268 skb_set_owner_w(skb
, sk
);
2272 err
= sock_intr_errno(timeo
);
2277 EXPORT_SYMBOL(sock_alloc_send_pskb
);
2279 struct sk_buff
*sock_alloc_send_skb(struct sock
*sk
, unsigned long size
,
2280 int noblock
, int *errcode
)
2282 return sock_alloc_send_pskb(sk
, size
, 0, noblock
, errcode
, 0);
2284 EXPORT_SYMBOL(sock_alloc_send_skb
);
2286 int __sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, struct cmsghdr
*cmsg
,
2287 struct sockcm_cookie
*sockc
)
2291 switch (cmsg
->cmsg_type
) {
2293 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
2295 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2297 sockc
->mark
= *(u32
*)CMSG_DATA(cmsg
);
2299 case SO_TIMESTAMPING_OLD
:
2300 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u32
)))
2303 tsflags
= *(u32
*)CMSG_DATA(cmsg
);
2304 if (tsflags
& ~SOF_TIMESTAMPING_TX_RECORD_MASK
)
2307 sockc
->tsflags
&= ~SOF_TIMESTAMPING_TX_RECORD_MASK
;
2308 sockc
->tsflags
|= tsflags
;
2311 if (!sock_flag(sk
, SOCK_TXTIME
))
2313 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(u64
)))
2315 sockc
->transmit_time
= get_unaligned((u64
*)CMSG_DATA(cmsg
));
2317 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2319 case SCM_CREDENTIALS
:
2326 EXPORT_SYMBOL(__sock_cmsg_send
);
2328 int sock_cmsg_send(struct sock
*sk
, struct msghdr
*msg
,
2329 struct sockcm_cookie
*sockc
)
2331 struct cmsghdr
*cmsg
;
2334 for_each_cmsghdr(cmsg
, msg
) {
2335 if (!CMSG_OK(msg
, cmsg
))
2337 if (cmsg
->cmsg_level
!= SOL_SOCKET
)
2339 ret
= __sock_cmsg_send(sk
, msg
, cmsg
, sockc
);
2345 EXPORT_SYMBOL(sock_cmsg_send
);
2347 static void sk_enter_memory_pressure(struct sock
*sk
)
2349 if (!sk
->sk_prot
->enter_memory_pressure
)
2352 sk
->sk_prot
->enter_memory_pressure(sk
);
2355 static void sk_leave_memory_pressure(struct sock
*sk
)
2357 if (sk
->sk_prot
->leave_memory_pressure
) {
2358 sk
->sk_prot
->leave_memory_pressure(sk
);
2360 unsigned long *memory_pressure
= sk
->sk_prot
->memory_pressure
;
2362 if (memory_pressure
&& READ_ONCE(*memory_pressure
))
2363 WRITE_ONCE(*memory_pressure
, 0);
2367 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2368 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key
);
2371 * skb_page_frag_refill - check that a page_frag contains enough room
2372 * @sz: minimum size of the fragment we want to get
2373 * @pfrag: pointer to page_frag
2374 * @gfp: priority for memory allocation
2376 * Note: While this allocator tries to use high order pages, there is
2377 * no guarantee that allocations succeed. Therefore, @sz MUST be
2378 * less or equal than PAGE_SIZE.
2380 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t gfp
)
2383 if (page_ref_count(pfrag
->page
) == 1) {
2387 if (pfrag
->offset
+ sz
<= pfrag
->size
)
2389 put_page(pfrag
->page
);
2393 if (SKB_FRAG_PAGE_ORDER
&&
2394 !static_branch_unlikely(&net_high_order_alloc_disable_key
)) {
2395 /* Avoid direct reclaim but allow kswapd to wake */
2396 pfrag
->page
= alloc_pages((gfp
& ~__GFP_DIRECT_RECLAIM
) |
2397 __GFP_COMP
| __GFP_NOWARN
|
2399 SKB_FRAG_PAGE_ORDER
);
2400 if (likely(pfrag
->page
)) {
2401 pfrag
->size
= PAGE_SIZE
<< SKB_FRAG_PAGE_ORDER
;
2405 pfrag
->page
= alloc_page(gfp
);
2406 if (likely(pfrag
->page
)) {
2407 pfrag
->size
= PAGE_SIZE
;
2412 EXPORT_SYMBOL(skb_page_frag_refill
);
2414 bool sk_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
2416 if (likely(skb_page_frag_refill(32U, pfrag
, sk
->sk_allocation
)))
2419 sk_enter_memory_pressure(sk
);
2420 sk_stream_moderate_sndbuf(sk
);
2423 EXPORT_SYMBOL(sk_page_frag_refill
);
2425 static void __lock_sock(struct sock
*sk
)
2426 __releases(&sk
->sk_lock
.slock
)
2427 __acquires(&sk
->sk_lock
.slock
)
2432 prepare_to_wait_exclusive(&sk
->sk_lock
.wq
, &wait
,
2433 TASK_UNINTERRUPTIBLE
);
2434 spin_unlock_bh(&sk
->sk_lock
.slock
);
2436 spin_lock_bh(&sk
->sk_lock
.slock
);
2437 if (!sock_owned_by_user(sk
))
2440 finish_wait(&sk
->sk_lock
.wq
, &wait
);
2443 void __release_sock(struct sock
*sk
)
2444 __releases(&sk
->sk_lock
.slock
)
2445 __acquires(&sk
->sk_lock
.slock
)
2447 struct sk_buff
*skb
, *next
;
2449 while ((skb
= sk
->sk_backlog
.head
) != NULL
) {
2450 sk
->sk_backlog
.head
= sk
->sk_backlog
.tail
= NULL
;
2452 spin_unlock_bh(&sk
->sk_lock
.slock
);
2457 WARN_ON_ONCE(skb_dst_is_noref(skb
));
2458 skb_mark_not_on_list(skb
);
2459 sk_backlog_rcv(sk
, skb
);
2464 } while (skb
!= NULL
);
2466 spin_lock_bh(&sk
->sk_lock
.slock
);
2470 * Doing the zeroing here guarantee we can not loop forever
2471 * while a wild producer attempts to flood us.
2473 sk
->sk_backlog
.len
= 0;
2476 void __sk_flush_backlog(struct sock
*sk
)
2478 spin_lock_bh(&sk
->sk_lock
.slock
);
2480 spin_unlock_bh(&sk
->sk_lock
.slock
);
2484 * sk_wait_data - wait for data to arrive at sk_receive_queue
2485 * @sk: sock to wait on
2486 * @timeo: for how long
2487 * @skb: last skb seen on sk_receive_queue
2489 * Now socket state including sk->sk_err is changed only under lock,
2490 * hence we may omit checks after joining wait queue.
2491 * We check receive queue before schedule() only as optimization;
2492 * it is very likely that release_sock() added new data.
2494 int sk_wait_data(struct sock
*sk
, long *timeo
, const struct sk_buff
*skb
)
2496 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
2499 add_wait_queue(sk_sleep(sk
), &wait
);
2500 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2501 rc
= sk_wait_event(sk
, timeo
, skb_peek_tail(&sk
->sk_receive_queue
) != skb
, &wait
);
2502 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2503 remove_wait_queue(sk_sleep(sk
), &wait
);
2506 EXPORT_SYMBOL(sk_wait_data
);
2509 * __sk_mem_raise_allocated - increase memory_allocated
2511 * @size: memory size to allocate
2512 * @amt: pages to allocate
2513 * @kind: allocation type
2515 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2517 int __sk_mem_raise_allocated(struct sock
*sk
, int size
, int amt
, int kind
)
2519 struct proto
*prot
= sk
->sk_prot
;
2520 long allocated
= sk_memory_allocated_add(sk
, amt
);
2521 bool charged
= true;
2523 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
2524 !(charged
= mem_cgroup_charge_skmem(sk
->sk_memcg
, amt
)))
2525 goto suppress_allocation
;
2528 if (allocated
<= sk_prot_mem_limits(sk
, 0)) {
2529 sk_leave_memory_pressure(sk
);
2533 /* Under pressure. */
2534 if (allocated
> sk_prot_mem_limits(sk
, 1))
2535 sk_enter_memory_pressure(sk
);
2537 /* Over hard limit. */
2538 if (allocated
> sk_prot_mem_limits(sk
, 2))
2539 goto suppress_allocation
;
2541 /* guarantee minimum buffer size under pressure */
2542 if (kind
== SK_MEM_RECV
) {
2543 if (atomic_read(&sk
->sk_rmem_alloc
) < sk_get_rmem0(sk
, prot
))
2546 } else { /* SK_MEM_SEND */
2547 int wmem0
= sk_get_wmem0(sk
, prot
);
2549 if (sk
->sk_type
== SOCK_STREAM
) {
2550 if (sk
->sk_wmem_queued
< wmem0
)
2552 } else if (refcount_read(&sk
->sk_wmem_alloc
) < wmem0
) {
2557 if (sk_has_memory_pressure(sk
)) {
2560 if (!sk_under_memory_pressure(sk
))
2562 alloc
= sk_sockets_allocated_read_positive(sk
);
2563 if (sk_prot_mem_limits(sk
, 2) > alloc
*
2564 sk_mem_pages(sk
->sk_wmem_queued
+
2565 atomic_read(&sk
->sk_rmem_alloc
) +
2566 sk
->sk_forward_alloc
))
2570 suppress_allocation
:
2572 if (kind
== SK_MEM_SEND
&& sk
->sk_type
== SOCK_STREAM
) {
2573 sk_stream_moderate_sndbuf(sk
);
2575 /* Fail only if socket is _under_ its sndbuf.
2576 * In this case we cannot block, so that we have to fail.
2578 if (sk
->sk_wmem_queued
+ size
>= sk
->sk_sndbuf
)
2582 if (kind
== SK_MEM_SEND
|| (kind
== SK_MEM_RECV
&& charged
))
2583 trace_sock_exceed_buf_limit(sk
, prot
, allocated
, kind
);
2585 sk_memory_allocated_sub(sk
, amt
);
2587 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2588 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amt
);
2592 EXPORT_SYMBOL(__sk_mem_raise_allocated
);
2595 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2597 * @size: memory size to allocate
2598 * @kind: allocation type
2600 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2601 * rmem allocation. This function assumes that protocols which have
2602 * memory_pressure use sk_wmem_queued as write buffer accounting.
2604 int __sk_mem_schedule(struct sock
*sk
, int size
, int kind
)
2606 int ret
, amt
= sk_mem_pages(size
);
2608 sk
->sk_forward_alloc
+= amt
<< SK_MEM_QUANTUM_SHIFT
;
2609 ret
= __sk_mem_raise_allocated(sk
, size
, amt
, kind
);
2611 sk
->sk_forward_alloc
-= amt
<< SK_MEM_QUANTUM_SHIFT
;
2614 EXPORT_SYMBOL(__sk_mem_schedule
);
2617 * __sk_mem_reduce_allocated - reclaim memory_allocated
2619 * @amount: number of quanta
2621 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2623 void __sk_mem_reduce_allocated(struct sock
*sk
, int amount
)
2625 sk_memory_allocated_sub(sk
, amount
);
2627 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
)
2628 mem_cgroup_uncharge_skmem(sk
->sk_memcg
, amount
);
2630 if (sk_under_memory_pressure(sk
) &&
2631 (sk_memory_allocated(sk
) < sk_prot_mem_limits(sk
, 0)))
2632 sk_leave_memory_pressure(sk
);
2634 EXPORT_SYMBOL(__sk_mem_reduce_allocated
);
2637 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2639 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2641 void __sk_mem_reclaim(struct sock
*sk
, int amount
)
2643 amount
>>= SK_MEM_QUANTUM_SHIFT
;
2644 sk
->sk_forward_alloc
-= amount
<< SK_MEM_QUANTUM_SHIFT
;
2645 __sk_mem_reduce_allocated(sk
, amount
);
2647 EXPORT_SYMBOL(__sk_mem_reclaim
);
2649 int sk_set_peek_off(struct sock
*sk
, int val
)
2651 sk
->sk_peek_off
= val
;
2654 EXPORT_SYMBOL_GPL(sk_set_peek_off
);
2657 * Set of default routines for initialising struct proto_ops when
2658 * the protocol does not support a particular function. In certain
2659 * cases where it makes no sense for a protocol to have a "do nothing"
2660 * function, some default processing is provided.
2663 int sock_no_bind(struct socket
*sock
, struct sockaddr
*saddr
, int len
)
2667 EXPORT_SYMBOL(sock_no_bind
);
2669 int sock_no_connect(struct socket
*sock
, struct sockaddr
*saddr
,
2674 EXPORT_SYMBOL(sock_no_connect
);
2676 int sock_no_socketpair(struct socket
*sock1
, struct socket
*sock2
)
2680 EXPORT_SYMBOL(sock_no_socketpair
);
2682 int sock_no_accept(struct socket
*sock
, struct socket
*newsock
, int flags
,
2687 EXPORT_SYMBOL(sock_no_accept
);
2689 int sock_no_getname(struct socket
*sock
, struct sockaddr
*saddr
,
2694 EXPORT_SYMBOL(sock_no_getname
);
2696 int sock_no_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2700 EXPORT_SYMBOL(sock_no_ioctl
);
2702 int sock_no_listen(struct socket
*sock
, int backlog
)
2706 EXPORT_SYMBOL(sock_no_listen
);
2708 int sock_no_shutdown(struct socket
*sock
, int how
)
2712 EXPORT_SYMBOL(sock_no_shutdown
);
2714 int sock_no_setsockopt(struct socket
*sock
, int level
, int optname
,
2715 char __user
*optval
, unsigned int optlen
)
2719 EXPORT_SYMBOL(sock_no_setsockopt
);
2721 int sock_no_getsockopt(struct socket
*sock
, int level
, int optname
,
2722 char __user
*optval
, int __user
*optlen
)
2726 EXPORT_SYMBOL(sock_no_getsockopt
);
2728 int sock_no_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
)
2732 EXPORT_SYMBOL(sock_no_sendmsg
);
2734 int sock_no_sendmsg_locked(struct sock
*sk
, struct msghdr
*m
, size_t len
)
2738 EXPORT_SYMBOL(sock_no_sendmsg_locked
);
2740 int sock_no_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t len
,
2745 EXPORT_SYMBOL(sock_no_recvmsg
);
2747 int sock_no_mmap(struct file
*file
, struct socket
*sock
, struct vm_area_struct
*vma
)
2749 /* Mirror missing mmap method error code */
2752 EXPORT_SYMBOL(sock_no_mmap
);
2754 ssize_t
sock_no_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
2757 struct msghdr msg
= {.msg_flags
= flags
};
2759 char *kaddr
= kmap(page
);
2760 iov
.iov_base
= kaddr
+ offset
;
2762 res
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
2766 EXPORT_SYMBOL(sock_no_sendpage
);
2768 ssize_t
sock_no_sendpage_locked(struct sock
*sk
, struct page
*page
,
2769 int offset
, size_t size
, int flags
)
2772 struct msghdr msg
= {.msg_flags
= flags
};
2774 char *kaddr
= kmap(page
);
2776 iov
.iov_base
= kaddr
+ offset
;
2778 res
= kernel_sendmsg_locked(sk
, &msg
, &iov
, 1, size
);
2782 EXPORT_SYMBOL(sock_no_sendpage_locked
);
2785 * Default Socket Callbacks
2788 static void sock_def_wakeup(struct sock
*sk
)
2790 struct socket_wq
*wq
;
2793 wq
= rcu_dereference(sk
->sk_wq
);
2794 if (skwq_has_sleeper(wq
))
2795 wake_up_interruptible_all(&wq
->wait
);
2799 static void sock_def_error_report(struct sock
*sk
)
2801 struct socket_wq
*wq
;
2804 wq
= rcu_dereference(sk
->sk_wq
);
2805 if (skwq_has_sleeper(wq
))
2806 wake_up_interruptible_poll(&wq
->wait
, EPOLLERR
);
2807 sk_wake_async(sk
, SOCK_WAKE_IO
, POLL_ERR
);
2811 void sock_def_readable(struct sock
*sk
)
2813 struct socket_wq
*wq
;
2816 wq
= rcu_dereference(sk
->sk_wq
);
2817 if (skwq_has_sleeper(wq
))
2818 wake_up_interruptible_sync_poll(&wq
->wait
, EPOLLIN
| EPOLLPRI
|
2819 EPOLLRDNORM
| EPOLLRDBAND
);
2820 sk_wake_async(sk
, SOCK_WAKE_WAITD
, POLL_IN
);
2824 static void sock_def_write_space(struct sock
*sk
)
2826 struct socket_wq
*wq
;
2830 /* Do not wake up a writer until he can make "significant"
2833 if ((refcount_read(&sk
->sk_wmem_alloc
) << 1) <= READ_ONCE(sk
->sk_sndbuf
)) {
2834 wq
= rcu_dereference(sk
->sk_wq
);
2835 if (skwq_has_sleeper(wq
))
2836 wake_up_interruptible_sync_poll(&wq
->wait
, EPOLLOUT
|
2837 EPOLLWRNORM
| EPOLLWRBAND
);
2839 /* Should agree with poll, otherwise some programs break */
2840 if (sock_writeable(sk
))
2841 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
2847 static void sock_def_destruct(struct sock
*sk
)
2851 void sk_send_sigurg(struct sock
*sk
)
2853 if (sk
->sk_socket
&& sk
->sk_socket
->file
)
2854 if (send_sigurg(&sk
->sk_socket
->file
->f_owner
))
2855 sk_wake_async(sk
, SOCK_WAKE_URG
, POLL_PRI
);
2857 EXPORT_SYMBOL(sk_send_sigurg
);
2859 void sk_reset_timer(struct sock
*sk
, struct timer_list
* timer
,
2860 unsigned long expires
)
2862 if (!mod_timer(timer
, expires
))
2865 EXPORT_SYMBOL(sk_reset_timer
);
2867 void sk_stop_timer(struct sock
*sk
, struct timer_list
* timer
)
2869 if (del_timer(timer
))
2872 EXPORT_SYMBOL(sk_stop_timer
);
2874 void sock_init_data(struct socket
*sock
, struct sock
*sk
)
2877 sk
->sk_send_head
= NULL
;
2879 timer_setup(&sk
->sk_timer
, NULL
, 0);
2881 sk
->sk_allocation
= GFP_KERNEL
;
2882 sk
->sk_rcvbuf
= sysctl_rmem_default
;
2883 sk
->sk_sndbuf
= sysctl_wmem_default
;
2884 sk
->sk_state
= TCP_CLOSE
;
2885 sk_set_socket(sk
, sock
);
2887 sock_set_flag(sk
, SOCK_ZAPPED
);
2890 sk
->sk_type
= sock
->type
;
2891 RCU_INIT_POINTER(sk
->sk_wq
, &sock
->wq
);
2893 sk
->sk_uid
= SOCK_INODE(sock
)->i_uid
;
2895 RCU_INIT_POINTER(sk
->sk_wq
, NULL
);
2896 sk
->sk_uid
= make_kuid(sock_net(sk
)->user_ns
, 0);
2899 rwlock_init(&sk
->sk_callback_lock
);
2900 if (sk
->sk_kern_sock
)
2901 lockdep_set_class_and_name(
2902 &sk
->sk_callback_lock
,
2903 af_kern_callback_keys
+ sk
->sk_family
,
2904 af_family_kern_clock_key_strings
[sk
->sk_family
]);
2906 lockdep_set_class_and_name(
2907 &sk
->sk_callback_lock
,
2908 af_callback_keys
+ sk
->sk_family
,
2909 af_family_clock_key_strings
[sk
->sk_family
]);
2911 sk
->sk_state_change
= sock_def_wakeup
;
2912 sk
->sk_data_ready
= sock_def_readable
;
2913 sk
->sk_write_space
= sock_def_write_space
;
2914 sk
->sk_error_report
= sock_def_error_report
;
2915 sk
->sk_destruct
= sock_def_destruct
;
2917 sk
->sk_frag
.page
= NULL
;
2918 sk
->sk_frag
.offset
= 0;
2919 sk
->sk_peek_off
= -1;
2921 sk
->sk_peer_pid
= NULL
;
2922 sk
->sk_peer_cred
= NULL
;
2923 sk
->sk_write_pending
= 0;
2924 sk
->sk_rcvlowat
= 1;
2925 sk
->sk_rcvtimeo
= MAX_SCHEDULE_TIMEOUT
;
2926 sk
->sk_sndtimeo
= MAX_SCHEDULE_TIMEOUT
;
2928 sk
->sk_stamp
= SK_DEFAULT_STAMP
;
2929 #if BITS_PER_LONG==32
2930 seqlock_init(&sk
->sk_stamp_seq
);
2932 atomic_set(&sk
->sk_zckey
, 0);
2934 #ifdef CONFIG_NET_RX_BUSY_POLL
2936 sk
->sk_ll_usec
= sysctl_net_busy_read
;
2939 sk
->sk_max_pacing_rate
= ~0UL;
2940 sk
->sk_pacing_rate
= ~0UL;
2941 WRITE_ONCE(sk
->sk_pacing_shift
, 10);
2942 sk
->sk_incoming_cpu
= -1;
2944 sk_rx_queue_clear(sk
);
2946 * Before updating sk_refcnt, we must commit prior changes to memory
2947 * (Documentation/RCU/rculist_nulls.txt for details)
2950 refcount_set(&sk
->sk_refcnt
, 1);
2951 atomic_set(&sk
->sk_drops
, 0);
2953 EXPORT_SYMBOL(sock_init_data
);
2955 void lock_sock_nested(struct sock
*sk
, int subclass
)
2958 spin_lock_bh(&sk
->sk_lock
.slock
);
2959 if (sk
->sk_lock
.owned
)
2961 sk
->sk_lock
.owned
= 1;
2962 spin_unlock(&sk
->sk_lock
.slock
);
2964 * The sk_lock has mutex_lock() semantics here:
2966 mutex_acquire(&sk
->sk_lock
.dep_map
, subclass
, 0, _RET_IP_
);
2969 EXPORT_SYMBOL(lock_sock_nested
);
2971 void release_sock(struct sock
*sk
)
2973 spin_lock_bh(&sk
->sk_lock
.slock
);
2974 if (sk
->sk_backlog
.tail
)
2977 /* Warning : release_cb() might need to release sk ownership,
2978 * ie call sock_release_ownership(sk) before us.
2980 if (sk
->sk_prot
->release_cb
)
2981 sk
->sk_prot
->release_cb(sk
);
2983 sock_release_ownership(sk
);
2984 if (waitqueue_active(&sk
->sk_lock
.wq
))
2985 wake_up(&sk
->sk_lock
.wq
);
2986 spin_unlock_bh(&sk
->sk_lock
.slock
);
2988 EXPORT_SYMBOL(release_sock
);
2991 * lock_sock_fast - fast version of lock_sock
2994 * This version should be used for very small section, where process wont block
2995 * return false if fast path is taken:
2997 * sk_lock.slock locked, owned = 0, BH disabled
2999 * return true if slow path is taken:
3001 * sk_lock.slock unlocked, owned = 1, BH enabled
3003 bool lock_sock_fast(struct sock
*sk
)
3006 spin_lock_bh(&sk
->sk_lock
.slock
);
3008 if (!sk
->sk_lock
.owned
)
3010 * Note : We must disable BH
3015 sk
->sk_lock
.owned
= 1;
3016 spin_unlock(&sk
->sk_lock
.slock
);
3018 * The sk_lock has mutex_lock() semantics here:
3020 mutex_acquire(&sk
->sk_lock
.dep_map
, 0, 0, _RET_IP_
);
3024 EXPORT_SYMBOL(lock_sock_fast
);
3026 int sock_gettstamp(struct socket
*sock
, void __user
*userstamp
,
3027 bool timeval
, bool time32
)
3029 struct sock
*sk
= sock
->sk
;
3030 struct timespec64 ts
;
3032 sock_enable_timestamp(sk
, SOCK_TIMESTAMP
);
3033 ts
= ktime_to_timespec64(sock_read_timestamp(sk
));
3034 if (ts
.tv_sec
== -1)
3036 if (ts
.tv_sec
== 0) {
3037 ktime_t kt
= ktime_get_real();
3038 sock_write_timestamp(sk
, kt
);
3039 ts
= ktime_to_timespec64(kt
);
3045 #ifdef CONFIG_COMPAT_32BIT_TIME
3047 return put_old_timespec32(&ts
, userstamp
);
3049 #ifdef CONFIG_SPARC64
3050 /* beware of padding in sparc64 timeval */
3051 if (timeval
&& !in_compat_syscall()) {
3052 struct __kernel_old_timeval __user tv
= {
3053 .tv_sec
= ts
.tv_sec
,
3054 .tv_usec
= ts
.tv_nsec
,
3056 if (copy_to_user(userstamp
, &tv
, sizeof(tv
)))
3061 return put_timespec64(&ts
, userstamp
);
3063 EXPORT_SYMBOL(sock_gettstamp
);
3065 void sock_enable_timestamp(struct sock
*sk
, enum sock_flags flag
)
3067 if (!sock_flag(sk
, flag
)) {
3068 unsigned long previous_flags
= sk
->sk_flags
;
3070 sock_set_flag(sk
, flag
);
3072 * we just set one of the two flags which require net
3073 * time stamping, but time stamping might have been on
3074 * already because of the other one
3076 if (sock_needs_netstamp(sk
) &&
3077 !(previous_flags
& SK_FLAGS_TIMESTAMP
))
3078 net_enable_timestamp();
3082 int sock_recv_errqueue(struct sock
*sk
, struct msghdr
*msg
, int len
,
3083 int level
, int type
)
3085 struct sock_exterr_skb
*serr
;
3086 struct sk_buff
*skb
;
3090 skb
= sock_dequeue_err_skb(sk
);
3096 msg
->msg_flags
|= MSG_TRUNC
;
3099 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
3103 sock_recv_timestamp(msg
, sk
, skb
);
3105 serr
= SKB_EXT_ERR(skb
);
3106 put_cmsg(msg
, level
, type
, sizeof(serr
->ee
), &serr
->ee
);
3108 msg
->msg_flags
|= MSG_ERRQUEUE
;
3116 EXPORT_SYMBOL(sock_recv_errqueue
);
3119 * Get a socket option on an socket.
3121 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3122 * asynchronous errors should be reported by getsockopt. We assume
3123 * this means if you specify SO_ERROR (otherwise whats the point of it).
3125 int sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
3126 char __user
*optval
, int __user
*optlen
)
3128 struct sock
*sk
= sock
->sk
;
3130 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
3132 EXPORT_SYMBOL(sock_common_getsockopt
);
3134 #ifdef CONFIG_COMPAT
3135 int compat_sock_common_getsockopt(struct socket
*sock
, int level
, int optname
,
3136 char __user
*optval
, int __user
*optlen
)
3138 struct sock
*sk
= sock
->sk
;
3140 if (sk
->sk_prot
->compat_getsockopt
!= NULL
)
3141 return sk
->sk_prot
->compat_getsockopt(sk
, level
, optname
,
3143 return sk
->sk_prot
->getsockopt(sk
, level
, optname
, optval
, optlen
);
3145 EXPORT_SYMBOL(compat_sock_common_getsockopt
);
3148 int sock_common_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
3151 struct sock
*sk
= sock
->sk
;
3155 err
= sk
->sk_prot
->recvmsg(sk
, msg
, size
, flags
& MSG_DONTWAIT
,
3156 flags
& ~MSG_DONTWAIT
, &addr_len
);
3158 msg
->msg_namelen
= addr_len
;
3161 EXPORT_SYMBOL(sock_common_recvmsg
);
3164 * Set socket options on an inet socket.
3166 int sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
3167 char __user
*optval
, unsigned int optlen
)
3169 struct sock
*sk
= sock
->sk
;
3171 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
3173 EXPORT_SYMBOL(sock_common_setsockopt
);
3175 #ifdef CONFIG_COMPAT
3176 int compat_sock_common_setsockopt(struct socket
*sock
, int level
, int optname
,
3177 char __user
*optval
, unsigned int optlen
)
3179 struct sock
*sk
= sock
->sk
;
3181 if (sk
->sk_prot
->compat_setsockopt
!= NULL
)
3182 return sk
->sk_prot
->compat_setsockopt(sk
, level
, optname
,
3184 return sk
->sk_prot
->setsockopt(sk
, level
, optname
, optval
, optlen
);
3186 EXPORT_SYMBOL(compat_sock_common_setsockopt
);
3189 void sk_common_release(struct sock
*sk
)
3191 if (sk
->sk_prot
->destroy
)
3192 sk
->sk_prot
->destroy(sk
);
3195 * Observation: when sock_common_release is called, processes have
3196 * no access to socket. But net still has.
3197 * Step one, detach it from networking:
3199 * A. Remove from hash tables.
3202 sk
->sk_prot
->unhash(sk
);
3205 * In this point socket cannot receive new packets, but it is possible
3206 * that some packets are in flight because some CPU runs receiver and
3207 * did hash table lookup before we unhashed socket. They will achieve
3208 * receive queue and will be purged by socket destructor.
3210 * Also we still have packets pending on receive queue and probably,
3211 * our own packets waiting in device queues. sock_destroy will drain
3212 * receive queue, but transmitted packets will delay socket destruction
3213 * until the last reference will be released.
3218 xfrm_sk_free_policy(sk
);
3220 sk_refcnt_debug_release(sk
);
3224 EXPORT_SYMBOL(sk_common_release
);
3226 void sk_get_meminfo(const struct sock
*sk
, u32
*mem
)
3228 memset(mem
, 0, sizeof(*mem
) * SK_MEMINFO_VARS
);
3230 mem
[SK_MEMINFO_RMEM_ALLOC
] = sk_rmem_alloc_get(sk
);
3231 mem
[SK_MEMINFO_RCVBUF
] = READ_ONCE(sk
->sk_rcvbuf
);
3232 mem
[SK_MEMINFO_WMEM_ALLOC
] = sk_wmem_alloc_get(sk
);
3233 mem
[SK_MEMINFO_SNDBUF
] = READ_ONCE(sk
->sk_sndbuf
);
3234 mem
[SK_MEMINFO_FWD_ALLOC
] = sk
->sk_forward_alloc
;
3235 mem
[SK_MEMINFO_WMEM_QUEUED
] = READ_ONCE(sk
->sk_wmem_queued
);
3236 mem
[SK_MEMINFO_OPTMEM
] = atomic_read(&sk
->sk_omem_alloc
);
3237 mem
[SK_MEMINFO_BACKLOG
] = READ_ONCE(sk
->sk_backlog
.len
);
3238 mem
[SK_MEMINFO_DROPS
] = atomic_read(&sk
->sk_drops
);
3241 #ifdef CONFIG_PROC_FS
3242 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3244 int val
[PROTO_INUSE_NR
];
3247 static DECLARE_BITMAP(proto_inuse_idx
, PROTO_INUSE_NR
);
3249 void sock_prot_inuse_add(struct net
*net
, struct proto
*prot
, int val
)
3251 __this_cpu_add(net
->core
.prot_inuse
->val
[prot
->inuse_idx
], val
);
3253 EXPORT_SYMBOL_GPL(sock_prot_inuse_add
);
3255 int sock_prot_inuse_get(struct net
*net
, struct proto
*prot
)
3257 int cpu
, idx
= prot
->inuse_idx
;
3260 for_each_possible_cpu(cpu
)
3261 res
+= per_cpu_ptr(net
->core
.prot_inuse
, cpu
)->val
[idx
];
3263 return res
>= 0 ? res
: 0;
3265 EXPORT_SYMBOL_GPL(sock_prot_inuse_get
);
3267 static void sock_inuse_add(struct net
*net
, int val
)
3269 this_cpu_add(*net
->core
.sock_inuse
, val
);
3272 int sock_inuse_get(struct net
*net
)
3276 for_each_possible_cpu(cpu
)
3277 res
+= *per_cpu_ptr(net
->core
.sock_inuse
, cpu
);
3282 EXPORT_SYMBOL_GPL(sock_inuse_get
);
3284 static int __net_init
sock_inuse_init_net(struct net
*net
)
3286 net
->core
.prot_inuse
= alloc_percpu(struct prot_inuse
);
3287 if (net
->core
.prot_inuse
== NULL
)
3290 net
->core
.sock_inuse
= alloc_percpu(int);
3291 if (net
->core
.sock_inuse
== NULL
)
3297 free_percpu(net
->core
.prot_inuse
);
3301 static void __net_exit
sock_inuse_exit_net(struct net
*net
)
3303 free_percpu(net
->core
.prot_inuse
);
3304 free_percpu(net
->core
.sock_inuse
);
3307 static struct pernet_operations net_inuse_ops
= {
3308 .init
= sock_inuse_init_net
,
3309 .exit
= sock_inuse_exit_net
,
3312 static __init
int net_inuse_init(void)
3314 if (register_pernet_subsys(&net_inuse_ops
))
3315 panic("Cannot initialize net inuse counters");
3320 core_initcall(net_inuse_init
);
3322 static int assign_proto_idx(struct proto
*prot
)
3324 prot
->inuse_idx
= find_first_zero_bit(proto_inuse_idx
, PROTO_INUSE_NR
);
3326 if (unlikely(prot
->inuse_idx
== PROTO_INUSE_NR
- 1)) {
3327 pr_err("PROTO_INUSE_NR exhausted\n");
3331 set_bit(prot
->inuse_idx
, proto_inuse_idx
);
3335 static void release_proto_idx(struct proto
*prot
)
3337 if (prot
->inuse_idx
!= PROTO_INUSE_NR
- 1)
3338 clear_bit(prot
->inuse_idx
, proto_inuse_idx
);
3341 static inline int assign_proto_idx(struct proto
*prot
)
3346 static inline void release_proto_idx(struct proto
*prot
)
3350 static void sock_inuse_add(struct net
*net
, int val
)
3355 static void req_prot_cleanup(struct request_sock_ops
*rsk_prot
)
3359 kfree(rsk_prot
->slab_name
);
3360 rsk_prot
->slab_name
= NULL
;
3361 kmem_cache_destroy(rsk_prot
->slab
);
3362 rsk_prot
->slab
= NULL
;
3365 static int req_prot_init(const struct proto
*prot
)
3367 struct request_sock_ops
*rsk_prot
= prot
->rsk_prot
;
3372 rsk_prot
->slab_name
= kasprintf(GFP_KERNEL
, "request_sock_%s",
3374 if (!rsk_prot
->slab_name
)
3377 rsk_prot
->slab
= kmem_cache_create(rsk_prot
->slab_name
,
3378 rsk_prot
->obj_size
, 0,
3379 SLAB_ACCOUNT
| prot
->slab_flags
,
3382 if (!rsk_prot
->slab
) {
3383 pr_crit("%s: Can't create request sock SLAB cache!\n",
3390 int proto_register(struct proto
*prot
, int alloc_slab
)
3395 prot
->slab
= kmem_cache_create_usercopy(prot
->name
,
3397 SLAB_HWCACHE_ALIGN
| SLAB_ACCOUNT
|
3399 prot
->useroffset
, prot
->usersize
,
3402 if (prot
->slab
== NULL
) {
3403 pr_crit("%s: Can't create sock SLAB cache!\n",
3408 if (req_prot_init(prot
))
3409 goto out_free_request_sock_slab
;
3411 if (prot
->twsk_prot
!= NULL
) {
3412 prot
->twsk_prot
->twsk_slab_name
= kasprintf(GFP_KERNEL
, "tw_sock_%s", prot
->name
);
3414 if (prot
->twsk_prot
->twsk_slab_name
== NULL
)
3415 goto out_free_request_sock_slab
;
3417 prot
->twsk_prot
->twsk_slab
=
3418 kmem_cache_create(prot
->twsk_prot
->twsk_slab_name
,
3419 prot
->twsk_prot
->twsk_obj_size
,
3424 if (prot
->twsk_prot
->twsk_slab
== NULL
)
3425 goto out_free_timewait_sock_slab_name
;
3429 mutex_lock(&proto_list_mutex
);
3430 ret
= assign_proto_idx(prot
);
3432 mutex_unlock(&proto_list_mutex
);
3433 goto out_free_timewait_sock_slab_name
;
3435 list_add(&prot
->node
, &proto_list
);
3436 mutex_unlock(&proto_list_mutex
);
3439 out_free_timewait_sock_slab_name
:
3440 if (alloc_slab
&& prot
->twsk_prot
)
3441 kfree(prot
->twsk_prot
->twsk_slab_name
);
3442 out_free_request_sock_slab
:
3444 req_prot_cleanup(prot
->rsk_prot
);
3446 kmem_cache_destroy(prot
->slab
);
3452 EXPORT_SYMBOL(proto_register
);
3454 void proto_unregister(struct proto
*prot
)
3456 mutex_lock(&proto_list_mutex
);
3457 release_proto_idx(prot
);
3458 list_del(&prot
->node
);
3459 mutex_unlock(&proto_list_mutex
);
3461 kmem_cache_destroy(prot
->slab
);
3464 req_prot_cleanup(prot
->rsk_prot
);
3466 if (prot
->twsk_prot
!= NULL
&& prot
->twsk_prot
->twsk_slab
!= NULL
) {
3467 kmem_cache_destroy(prot
->twsk_prot
->twsk_slab
);
3468 kfree(prot
->twsk_prot
->twsk_slab_name
);
3469 prot
->twsk_prot
->twsk_slab
= NULL
;
3472 EXPORT_SYMBOL(proto_unregister
);
3474 int sock_load_diag_module(int family
, int protocol
)
3477 if (!sock_is_registered(family
))
3480 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK
,
3481 NETLINK_SOCK_DIAG
, family
);
3485 if (family
== AF_INET
&&
3486 protocol
!= IPPROTO_RAW
&&
3487 !rcu_access_pointer(inet_protos
[protocol
]))
3491 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK
,
3492 NETLINK_SOCK_DIAG
, family
, protocol
);
3494 EXPORT_SYMBOL(sock_load_diag_module
);
3496 #ifdef CONFIG_PROC_FS
3497 static void *proto_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3498 __acquires(proto_list_mutex
)
3500 mutex_lock(&proto_list_mutex
);
3501 return seq_list_start_head(&proto_list
, *pos
);
3504 static void *proto_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3506 return seq_list_next(v
, &proto_list
, pos
);
3509 static void proto_seq_stop(struct seq_file
*seq
, void *v
)
3510 __releases(proto_list_mutex
)
3512 mutex_unlock(&proto_list_mutex
);
3515 static char proto_method_implemented(const void *method
)
3517 return method
== NULL
? 'n' : 'y';
3519 static long sock_prot_memory_allocated(struct proto
*proto
)
3521 return proto
->memory_allocated
!= NULL
? proto_memory_allocated(proto
) : -1L;
3524 static const char *sock_prot_memory_pressure(struct proto
*proto
)
3526 return proto
->memory_pressure
!= NULL
?
3527 proto_memory_pressure(proto
) ? "yes" : "no" : "NI";
3530 static void proto_seq_printf(struct seq_file
*seq
, struct proto
*proto
)
3533 seq_printf(seq
, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3534 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3537 sock_prot_inuse_get(seq_file_net(seq
), proto
),
3538 sock_prot_memory_allocated(proto
),
3539 sock_prot_memory_pressure(proto
),
3541 proto
->slab
== NULL
? "no" : "yes",
3542 module_name(proto
->owner
),
3543 proto_method_implemented(proto
->close
),
3544 proto_method_implemented(proto
->connect
),
3545 proto_method_implemented(proto
->disconnect
),
3546 proto_method_implemented(proto
->accept
),
3547 proto_method_implemented(proto
->ioctl
),
3548 proto_method_implemented(proto
->init
),
3549 proto_method_implemented(proto
->destroy
),
3550 proto_method_implemented(proto
->shutdown
),
3551 proto_method_implemented(proto
->setsockopt
),
3552 proto_method_implemented(proto
->getsockopt
),
3553 proto_method_implemented(proto
->sendmsg
),
3554 proto_method_implemented(proto
->recvmsg
),
3555 proto_method_implemented(proto
->sendpage
),
3556 proto_method_implemented(proto
->bind
),
3557 proto_method_implemented(proto
->backlog_rcv
),
3558 proto_method_implemented(proto
->hash
),
3559 proto_method_implemented(proto
->unhash
),
3560 proto_method_implemented(proto
->get_port
),
3561 proto_method_implemented(proto
->enter_memory_pressure
));
3564 static int proto_seq_show(struct seq_file
*seq
, void *v
)
3566 if (v
== &proto_list
)
3567 seq_printf(seq
, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3576 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3578 proto_seq_printf(seq
, list_entry(v
, struct proto
, node
));
3582 static const struct seq_operations proto_seq_ops
= {
3583 .start
= proto_seq_start
,
3584 .next
= proto_seq_next
,
3585 .stop
= proto_seq_stop
,
3586 .show
= proto_seq_show
,
3589 static __net_init
int proto_init_net(struct net
*net
)
3591 if (!proc_create_net("protocols", 0444, net
->proc_net
, &proto_seq_ops
,
3592 sizeof(struct seq_net_private
)))
3598 static __net_exit
void proto_exit_net(struct net
*net
)
3600 remove_proc_entry("protocols", net
->proc_net
);
3604 static __net_initdata
struct pernet_operations proto_net_ops
= {
3605 .init
= proto_init_net
,
3606 .exit
= proto_exit_net
,
3609 static int __init
proto_init(void)
3611 return register_pernet_subsys(&proto_net_ops
);
3614 subsys_initcall(proto_init
);
3616 #endif /* PROC_FS */
3618 #ifdef CONFIG_NET_RX_BUSY_POLL
3619 bool sk_busy_loop_end(void *p
, unsigned long start_time
)
3621 struct sock
*sk
= p
;
3623 return !skb_queue_empty_lockless(&sk
->sk_receive_queue
) ||
3624 sk_busy_loop_timeout(sk
, start_time
);
3626 EXPORT_SYMBOL(sk_busy_loop_end
);
3627 #endif /* CONFIG_NET_RX_BUSY_POLL */