2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
166 EXPORT_SYMBOL(sk_ns_capable);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock *sk, int cap)
179 return sk_ns_capable(sk, &init_user_ns, cap);
181 EXPORT_SYMBOL(sk_capable);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock *sk, int cap)
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
196 EXPORT_SYMBOL(sk_net_capable);
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family and separate keys for internal and
203 static struct lock_class_key af_family_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_keys[AF_MAX];
205 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209 * Make lock validator output more readable. (we pre-construct these
210 * strings build-time, so that runtime initialization of socket
214 #define _sock_locks(x) \
215 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
216 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
217 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
218 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
219 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
220 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
221 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
222 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
223 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
224 x "27" , x "28" , x "AF_CAN" , \
225 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
226 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
227 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
228 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
229 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
231 static const char *const af_family_key_strings[AF_MAX+1] = {
232 _sock_locks("sk_lock-")
234 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
235 _sock_locks("slock-")
237 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
238 _sock_locks("clock-")
241 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
242 _sock_locks("k-sk_lock-")
244 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-slock-")
247 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
248 _sock_locks("k-clock-")
250 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
251 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
252 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
253 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
254 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
255 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
256 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
257 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
258 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
259 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
260 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
261 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
262 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
263 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
264 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
265 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
267 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
268 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
269 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
270 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
271 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
272 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
273 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
274 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
275 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
276 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
277 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
278 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
279 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
280 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
281 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
282 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
284 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
285 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
286 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
287 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
288 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
289 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
290 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
291 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
292 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
293 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
294 "elock-27" , "elock-28" , "elock-AF_CAN" ,
295 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
296 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
297 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
298 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
299 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
303 * sk_callback_lock and sk queues locking rules are per-address-family,
304 * so split the lock classes by using a per-AF key:
306 static struct lock_class_key af_callback_keys[AF_MAX];
307 static struct lock_class_key af_rlock_keys[AF_MAX];
308 static struct lock_class_key af_wlock_keys[AF_MAX];
309 static struct lock_class_key af_elock_keys[AF_MAX];
310 static struct lock_class_key af_kern_callback_keys[AF_MAX];
312 /* Take into consideration the size of the struct sk_buff overhead in the
313 * determination of these values, since that is non-constant across
314 * platforms. This makes socket queueing behavior and performance
315 * not depend upon such differences.
317 #define _SK_MEM_PACKETS 256
318 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
319 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
320 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
322 /* Run time adjustable parameters. */
323 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
324 EXPORT_SYMBOL(sysctl_wmem_max);
325 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
326 EXPORT_SYMBOL(sysctl_rmem_max);
327 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
328 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
330 /* Maximal space eaten by iovec or ancillary data plus some space */
331 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
332 EXPORT_SYMBOL(sysctl_optmem_max);
334 int sysctl_tstamp_allow_data __read_mostly = 1;
336 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
337 EXPORT_SYMBOL_GPL(memalloc_socks);
340 * sk_set_memalloc - sets %SOCK_MEMALLOC
341 * @sk: socket to set it on
343 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
344 * It's the responsibility of the admin to adjust min_free_kbytes
345 * to meet the requirements
347 void sk_set_memalloc(struct sock *sk)
349 sock_set_flag(sk, SOCK_MEMALLOC);
350 sk->sk_allocation |= __GFP_MEMALLOC;
351 static_key_slow_inc(&memalloc_socks);
353 EXPORT_SYMBOL_GPL(sk_set_memalloc);
355 void sk_clear_memalloc(struct sock *sk)
357 sock_reset_flag(sk, SOCK_MEMALLOC);
358 sk->sk_allocation &= ~__GFP_MEMALLOC;
359 static_key_slow_dec(&memalloc_socks);
362 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
363 * progress of swapping. SOCK_MEMALLOC may be cleared while
364 * it has rmem allocations due to the last swapfile being deactivated
365 * but there is a risk that the socket is unusable due to exceeding
366 * the rmem limits. Reclaim the reserves and obey rmem limits again.
370 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
372 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
375 unsigned long pflags = current->flags;
377 /* these should have been dropped before queueing */
378 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
380 current->flags |= PF_MEMALLOC;
381 ret = sk->sk_backlog_rcv(sk, skb);
382 current_restore_flags(pflags, PF_MEMALLOC);
386 EXPORT_SYMBOL(__sk_backlog_rcv);
388 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
392 if (optlen < sizeof(tv))
394 if (copy_from_user(&tv, optval, sizeof(tv)))
396 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
400 static int warned __read_mostly;
403 if (warned < 10 && net_ratelimit()) {
405 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
406 __func__, current->comm, task_pid_nr(current));
410 *timeo_p = MAX_SCHEDULE_TIMEOUT;
411 if (tv.tv_sec == 0 && tv.tv_usec == 0)
413 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
414 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
418 static void sock_warn_obsolete_bsdism(const char *name)
421 static char warncomm[TASK_COMM_LEN];
422 if (strcmp(warncomm, current->comm) && warned < 5) {
423 strcpy(warncomm, current->comm);
424 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
430 static bool sock_needs_netstamp(const struct sock *sk)
432 switch (sk->sk_family) {
441 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
443 if (sk->sk_flags & flags) {
444 sk->sk_flags &= ~flags;
445 if (sock_needs_netstamp(sk) &&
446 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
447 net_disable_timestamp();
452 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
455 struct sk_buff_head *list = &sk->sk_receive_queue;
457 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
458 atomic_inc(&sk->sk_drops);
459 trace_sock_rcvqueue_full(sk, skb);
463 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
464 atomic_inc(&sk->sk_drops);
469 skb_set_owner_r(skb, sk);
471 /* we escape from rcu protected region, make sure we dont leak
476 spin_lock_irqsave(&list->lock, flags);
477 sock_skb_set_dropcount(sk, skb);
478 __skb_queue_tail(list, skb);
479 spin_unlock_irqrestore(&list->lock, flags);
481 if (!sock_flag(sk, SOCK_DEAD))
482 sk->sk_data_ready(sk);
485 EXPORT_SYMBOL(__sock_queue_rcv_skb);
487 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
491 err = sk_filter(sk, skb);
495 return __sock_queue_rcv_skb(sk, skb);
497 EXPORT_SYMBOL(sock_queue_rcv_skb);
499 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
500 const int nested, unsigned int trim_cap, bool refcounted)
502 int rc = NET_RX_SUCCESS;
504 if (sk_filter_trim_cap(sk, skb, trim_cap))
505 goto discard_and_relse;
509 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
510 atomic_inc(&sk->sk_drops);
511 goto discard_and_relse;
514 bh_lock_sock_nested(sk);
517 if (!sock_owned_by_user(sk)) {
519 * trylock + unlock semantics:
521 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
523 rc = sk_backlog_rcv(sk, skb);
525 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
526 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
528 atomic_inc(&sk->sk_drops);
529 goto discard_and_relse;
541 EXPORT_SYMBOL(__sk_receive_skb);
543 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
545 struct dst_entry *dst = __sk_dst_get(sk);
547 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
548 sk_tx_queue_clear(sk);
549 sk->sk_dst_pending_confirm = 0;
550 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
557 EXPORT_SYMBOL(__sk_dst_check);
559 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
561 struct dst_entry *dst = sk_dst_get(sk);
563 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
571 EXPORT_SYMBOL(sk_dst_check);
573 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
576 int ret = -ENOPROTOOPT;
577 #ifdef CONFIG_NETDEVICES
578 struct net *net = sock_net(sk);
579 char devname[IFNAMSIZ];
584 if (!ns_capable(net->user_ns, CAP_NET_RAW))
591 /* Bind this socket to a particular device like "eth0",
592 * as specified in the passed interface name. If the
593 * name is "" or the option length is zero the socket
596 if (optlen > IFNAMSIZ - 1)
597 optlen = IFNAMSIZ - 1;
598 memset(devname, 0, sizeof(devname));
601 if (copy_from_user(devname, optval, optlen))
605 if (devname[0] != '\0') {
606 struct net_device *dev;
609 dev = dev_get_by_name_rcu(net, devname);
611 index = dev->ifindex;
619 sk->sk_bound_dev_if = index;
631 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
632 int __user *optlen, int len)
634 int ret = -ENOPROTOOPT;
635 #ifdef CONFIG_NETDEVICES
636 struct net *net = sock_net(sk);
637 char devname[IFNAMSIZ];
639 if (sk->sk_bound_dev_if == 0) {
648 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
652 len = strlen(devname) + 1;
655 if (copy_to_user(optval, devname, len))
660 if (put_user(len, optlen))
671 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
674 sock_set_flag(sk, bit);
676 sock_reset_flag(sk, bit);
679 bool sk_mc_loop(struct sock *sk)
681 if (dev_recursion_level())
685 switch (sk->sk_family) {
687 return inet_sk(sk)->mc_loop;
688 #if IS_ENABLED(CONFIG_IPV6)
690 return inet6_sk(sk)->mc_loop;
696 EXPORT_SYMBOL(sk_mc_loop);
699 * This is meant for all protocols to use and covers goings on
700 * at the socket level. Everything here is generic.
703 int sock_setsockopt(struct socket *sock, int level, int optname,
704 char __user *optval, unsigned int optlen)
706 struct sock *sk = sock->sk;
713 * Options without arguments
716 if (optname == SO_BINDTODEVICE)
717 return sock_setbindtodevice(sk, optval, optlen);
719 if (optlen < sizeof(int))
722 if (get_user(val, (int __user *)optval))
725 valbool = val ? 1 : 0;
731 if (val && !capable(CAP_NET_ADMIN))
734 sock_valbool_flag(sk, SOCK_DBG, valbool);
737 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
740 sk->sk_reuseport = valbool;
749 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
752 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
755 /* Don't error on this BSD doesn't and if you think
756 * about it this is right. Otherwise apps have to
757 * play 'guess the biggest size' games. RCVBUF/SNDBUF
758 * are treated in BSD as hints
760 val = min_t(u32, val, sysctl_wmem_max);
762 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
763 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
764 /* Wake up sending tasks if we upped the value. */
765 sk->sk_write_space(sk);
769 if (!capable(CAP_NET_ADMIN)) {
776 /* Don't error on this BSD doesn't and if you think
777 * about it this is right. Otherwise apps have to
778 * play 'guess the biggest size' games. RCVBUF/SNDBUF
779 * are treated in BSD as hints
781 val = min_t(u32, val, sysctl_rmem_max);
783 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
785 * We double it on the way in to account for
786 * "struct sk_buff" etc. overhead. Applications
787 * assume that the SO_RCVBUF setting they make will
788 * allow that much actual data to be received on that
791 * Applications are unaware that "struct sk_buff" and
792 * other overheads allocate from the receive buffer
793 * during socket buffer allocation.
795 * And after considering the possible alternatives,
796 * returning the value we actually used in getsockopt
797 * is the most desirable behavior.
799 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
803 if (!capable(CAP_NET_ADMIN)) {
810 if (sk->sk_prot->keepalive)
811 sk->sk_prot->keepalive(sk, valbool);
812 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
816 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
820 sk->sk_no_check_tx = valbool;
824 if ((val >= 0 && val <= 6) ||
825 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
826 sk->sk_priority = val;
832 if (optlen < sizeof(ling)) {
833 ret = -EINVAL; /* 1003.1g */
836 if (copy_from_user(&ling, optval, sizeof(ling))) {
841 sock_reset_flag(sk, SOCK_LINGER);
843 #if (BITS_PER_LONG == 32)
844 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
845 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
848 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
849 sock_set_flag(sk, SOCK_LINGER);
854 sock_warn_obsolete_bsdism("setsockopt");
859 set_bit(SOCK_PASSCRED, &sock->flags);
861 clear_bit(SOCK_PASSCRED, &sock->flags);
867 if (optname == SO_TIMESTAMP)
868 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
870 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
871 sock_set_flag(sk, SOCK_RCVTSTAMP);
872 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
874 sock_reset_flag(sk, SOCK_RCVTSTAMP);
875 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
879 case SO_TIMESTAMPING:
880 if (val & ~SOF_TIMESTAMPING_MASK) {
885 if (val & SOF_TIMESTAMPING_OPT_ID &&
886 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
887 if (sk->sk_protocol == IPPROTO_TCP &&
888 sk->sk_type == SOCK_STREAM) {
889 if ((1 << sk->sk_state) &
890 (TCPF_CLOSE | TCPF_LISTEN)) {
894 sk->sk_tskey = tcp_sk(sk)->snd_una;
900 if (val & SOF_TIMESTAMPING_OPT_STATS &&
901 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
906 sk->sk_tsflags = val;
907 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
908 sock_enable_timestamp(sk,
909 SOCK_TIMESTAMPING_RX_SOFTWARE);
911 sock_disable_timestamp(sk,
912 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
918 sk->sk_rcvlowat = val ? : 1;
922 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
926 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
929 case SO_ATTACH_FILTER:
931 if (optlen == sizeof(struct sock_fprog)) {
932 struct sock_fprog fprog;
935 if (copy_from_user(&fprog, optval, sizeof(fprog)))
938 ret = sk_attach_filter(&fprog, sk);
944 if (optlen == sizeof(u32)) {
948 if (copy_from_user(&ufd, optval, sizeof(ufd)))
951 ret = sk_attach_bpf(ufd, sk);
955 case SO_ATTACH_REUSEPORT_CBPF:
957 if (optlen == sizeof(struct sock_fprog)) {
958 struct sock_fprog fprog;
961 if (copy_from_user(&fprog, optval, sizeof(fprog)))
964 ret = sk_reuseport_attach_filter(&fprog, sk);
968 case SO_ATTACH_REUSEPORT_EBPF:
970 if (optlen == sizeof(u32)) {
974 if (copy_from_user(&ufd, optval, sizeof(ufd)))
977 ret = sk_reuseport_attach_bpf(ufd, sk);
981 case SO_DETACH_FILTER:
982 ret = sk_detach_filter(sk);
986 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
989 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
994 set_bit(SOCK_PASSSEC, &sock->flags);
996 clear_bit(SOCK_PASSSEC, &sock->flags);
999 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1006 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1009 case SO_WIFI_STATUS:
1010 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1014 if (sock->ops->set_peek_off)
1015 ret = sock->ops->set_peek_off(sk, val);
1021 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1024 case SO_SELECT_ERR_QUEUE:
1025 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1028 #ifdef CONFIG_NET_RX_BUSY_POLL
1030 /* allow unprivileged users to decrease the value */
1031 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1037 sk->sk_ll_usec = val;
1042 case SO_MAX_PACING_RATE:
1043 sk->sk_max_pacing_rate = val;
1044 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1045 sk->sk_max_pacing_rate);
1048 case SO_INCOMING_CPU:
1049 sk->sk_incoming_cpu = val;
1054 dst_negative_advice(sk);
1063 EXPORT_SYMBOL(sock_setsockopt);
1066 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1067 struct ucred *ucred)
1069 ucred->pid = pid_vnr(pid);
1070 ucred->uid = ucred->gid = -1;
1072 struct user_namespace *current_ns = current_user_ns();
1074 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1075 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1079 int sock_getsockopt(struct socket *sock, int level, int optname,
1080 char __user *optval, int __user *optlen)
1082 struct sock *sk = sock->sk;
1091 int lv = sizeof(int);
1094 if (get_user(len, optlen))
1099 memset(&v, 0, sizeof(v));
1103 v.val = sock_flag(sk, SOCK_DBG);
1107 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1111 v.val = sock_flag(sk, SOCK_BROADCAST);
1115 v.val = sk->sk_sndbuf;
1119 v.val = sk->sk_rcvbuf;
1123 v.val = sk->sk_reuse;
1127 v.val = sk->sk_reuseport;
1131 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1135 v.val = sk->sk_type;
1139 v.val = sk->sk_protocol;
1143 v.val = sk->sk_family;
1147 v.val = -sock_error(sk);
1149 v.val = xchg(&sk->sk_err_soft, 0);
1153 v.val = sock_flag(sk, SOCK_URGINLINE);
1157 v.val = sk->sk_no_check_tx;
1161 v.val = sk->sk_priority;
1165 lv = sizeof(v.ling);
1166 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1167 v.ling.l_linger = sk->sk_lingertime / HZ;
1171 sock_warn_obsolete_bsdism("getsockopt");
1175 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1176 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1179 case SO_TIMESTAMPNS:
1180 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1183 case SO_TIMESTAMPING:
1184 v.val = sk->sk_tsflags;
1188 lv = sizeof(struct timeval);
1189 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1193 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1194 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1199 lv = sizeof(struct timeval);
1200 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1204 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1205 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1210 v.val = sk->sk_rcvlowat;
1218 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1223 struct ucred peercred;
1224 if (len > sizeof(peercred))
1225 len = sizeof(peercred);
1226 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1227 if (copy_to_user(optval, &peercred, len))
1236 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1240 if (copy_to_user(optval, address, len))
1245 /* Dubious BSD thing... Probably nobody even uses it, but
1246 * the UNIX standard wants it for whatever reason... -DaveM
1249 v.val = sk->sk_state == TCP_LISTEN;
1253 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1257 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1260 v.val = sk->sk_mark;
1264 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1267 case SO_WIFI_STATUS:
1268 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1272 if (!sock->ops->set_peek_off)
1275 v.val = sk->sk_peek_off;
1278 v.val = sock_flag(sk, SOCK_NOFCS);
1281 case SO_BINDTODEVICE:
1282 return sock_getbindtodevice(sk, optval, optlen, len);
1285 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1291 case SO_LOCK_FILTER:
1292 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1295 case SO_BPF_EXTENSIONS:
1296 v.val = bpf_tell_extensions();
1299 case SO_SELECT_ERR_QUEUE:
1300 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1303 #ifdef CONFIG_NET_RX_BUSY_POLL
1305 v.val = sk->sk_ll_usec;
1309 case SO_MAX_PACING_RATE:
1310 v.val = sk->sk_max_pacing_rate;
1313 case SO_INCOMING_CPU:
1314 v.val = sk->sk_incoming_cpu;
1319 u32 meminfo[SK_MEMINFO_VARS];
1321 if (get_user(len, optlen))
1324 sk_get_meminfo(sk, meminfo);
1326 len = min_t(unsigned int, len, sizeof(meminfo));
1327 if (copy_to_user(optval, &meminfo, len))
1333 #ifdef CONFIG_NET_RX_BUSY_POLL
1334 case SO_INCOMING_NAPI_ID:
1335 v.val = READ_ONCE(sk->sk_napi_id);
1337 /* aggregate non-NAPI IDs down to 0 */
1338 if (v.val < MIN_NAPI_ID)
1348 v.val64 = sock_gen_cookie(sk);
1352 /* We implement the SO_SNDLOWAT etc to not be settable
1355 return -ENOPROTOOPT;
1360 if (copy_to_user(optval, &v, len))
1363 if (put_user(len, optlen))
1369 * Initialize an sk_lock.
1371 * (We also register the sk_lock with the lock validator.)
1373 static inline void sock_lock_init(struct sock *sk)
1375 if (sk->sk_kern_sock)
1376 sock_lock_init_class_and_name(
1378 af_family_kern_slock_key_strings[sk->sk_family],
1379 af_family_kern_slock_keys + sk->sk_family,
1380 af_family_kern_key_strings[sk->sk_family],
1381 af_family_kern_keys + sk->sk_family);
1383 sock_lock_init_class_and_name(
1385 af_family_slock_key_strings[sk->sk_family],
1386 af_family_slock_keys + sk->sk_family,
1387 af_family_key_strings[sk->sk_family],
1388 af_family_keys + sk->sk_family);
1392 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1393 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1394 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1396 static void sock_copy(struct sock *nsk, const struct sock *osk)
1398 #ifdef CONFIG_SECURITY_NETWORK
1399 void *sptr = nsk->sk_security;
1401 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1403 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1404 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1406 #ifdef CONFIG_SECURITY_NETWORK
1407 nsk->sk_security = sptr;
1408 security_sk_clone(osk, nsk);
1412 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1416 struct kmem_cache *slab;
1420 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1423 if (priority & __GFP_ZERO)
1424 sk_prot_clear_nulls(sk, prot->obj_size);
1426 sk = kmalloc(prot->obj_size, priority);
1429 kmemcheck_annotate_bitfield(sk, flags);
1431 if (security_sk_alloc(sk, family, priority))
1434 if (!try_module_get(prot->owner))
1436 sk_tx_queue_clear(sk);
1442 security_sk_free(sk);
1445 kmem_cache_free(slab, sk);
1451 static void sk_prot_free(struct proto *prot, struct sock *sk)
1453 struct kmem_cache *slab;
1454 struct module *owner;
1456 owner = prot->owner;
1459 cgroup_sk_free(&sk->sk_cgrp_data);
1460 mem_cgroup_sk_free(sk);
1461 security_sk_free(sk);
1463 kmem_cache_free(slab, sk);
1470 * sk_alloc - All socket objects are allocated here
1471 * @net: the applicable net namespace
1472 * @family: protocol family
1473 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1474 * @prot: struct proto associated with this new sock instance
1475 * @kern: is this to be a kernel socket?
1477 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1478 struct proto *prot, int kern)
1482 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1484 sk->sk_family = family;
1486 * See comment in struct sock definition to understand
1487 * why we need sk_prot_creator -acme
1489 sk->sk_prot = sk->sk_prot_creator = prot;
1490 sk->sk_kern_sock = kern;
1492 sk->sk_net_refcnt = kern ? 0 : 1;
1493 if (likely(sk->sk_net_refcnt))
1495 sock_net_set(sk, net);
1496 atomic_set(&sk->sk_wmem_alloc, 1);
1498 mem_cgroup_sk_alloc(sk);
1499 cgroup_sk_alloc(&sk->sk_cgrp_data);
1500 sock_update_classid(&sk->sk_cgrp_data);
1501 sock_update_netprioidx(&sk->sk_cgrp_data);
1506 EXPORT_SYMBOL(sk_alloc);
1508 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1509 * grace period. This is the case for UDP sockets and TCP listeners.
1511 static void __sk_destruct(struct rcu_head *head)
1513 struct sock *sk = container_of(head, struct sock, sk_rcu);
1514 struct sk_filter *filter;
1516 if (sk->sk_destruct)
1517 sk->sk_destruct(sk);
1519 filter = rcu_dereference_check(sk->sk_filter,
1520 atomic_read(&sk->sk_wmem_alloc) == 0);
1522 sk_filter_uncharge(sk, filter);
1523 RCU_INIT_POINTER(sk->sk_filter, NULL);
1525 if (rcu_access_pointer(sk->sk_reuseport_cb))
1526 reuseport_detach_sock(sk);
1528 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1530 if (atomic_read(&sk->sk_omem_alloc))
1531 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1532 __func__, atomic_read(&sk->sk_omem_alloc));
1534 if (sk->sk_frag.page) {
1535 put_page(sk->sk_frag.page);
1536 sk->sk_frag.page = NULL;
1539 if (sk->sk_peer_cred)
1540 put_cred(sk->sk_peer_cred);
1541 put_pid(sk->sk_peer_pid);
1542 if (likely(sk->sk_net_refcnt))
1543 put_net(sock_net(sk));
1544 sk_prot_free(sk->sk_prot_creator, sk);
1547 void sk_destruct(struct sock *sk)
1549 if (sock_flag(sk, SOCK_RCU_FREE))
1550 call_rcu(&sk->sk_rcu, __sk_destruct);
1552 __sk_destruct(&sk->sk_rcu);
1555 static void __sk_free(struct sock *sk)
1557 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1558 sock_diag_broadcast_destroy(sk);
1563 void sk_free(struct sock *sk)
1566 * We subtract one from sk_wmem_alloc and can know if
1567 * some packets are still in some tx queue.
1568 * If not null, sock_wfree() will call __sk_free(sk) later
1570 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1573 EXPORT_SYMBOL(sk_free);
1575 static void sk_init_common(struct sock *sk)
1577 skb_queue_head_init(&sk->sk_receive_queue);
1578 skb_queue_head_init(&sk->sk_write_queue);
1579 skb_queue_head_init(&sk->sk_error_queue);
1581 rwlock_init(&sk->sk_callback_lock);
1582 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1583 af_rlock_keys + sk->sk_family,
1584 af_family_rlock_key_strings[sk->sk_family]);
1585 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1586 af_wlock_keys + sk->sk_family,
1587 af_family_wlock_key_strings[sk->sk_family]);
1588 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1589 af_elock_keys + sk->sk_family,
1590 af_family_elock_key_strings[sk->sk_family]);
1591 lockdep_set_class_and_name(&sk->sk_callback_lock,
1592 af_callback_keys + sk->sk_family,
1593 af_family_clock_key_strings[sk->sk_family]);
1597 * sk_clone_lock - clone a socket, and lock its clone
1598 * @sk: the socket to clone
1599 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1601 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1603 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1606 bool is_charged = true;
1608 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1609 if (newsk != NULL) {
1610 struct sk_filter *filter;
1612 sock_copy(newsk, sk);
1615 if (likely(newsk->sk_net_refcnt))
1616 get_net(sock_net(newsk));
1617 sk_node_init(&newsk->sk_node);
1618 sock_lock_init(newsk);
1619 bh_lock_sock(newsk);
1620 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1621 newsk->sk_backlog.len = 0;
1623 atomic_set(&newsk->sk_rmem_alloc, 0);
1625 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1627 atomic_set(&newsk->sk_wmem_alloc, 1);
1628 atomic_set(&newsk->sk_omem_alloc, 0);
1629 sk_init_common(newsk);
1631 newsk->sk_dst_cache = NULL;
1632 newsk->sk_dst_pending_confirm = 0;
1633 newsk->sk_wmem_queued = 0;
1634 newsk->sk_forward_alloc = 0;
1635 atomic_set(&newsk->sk_drops, 0);
1636 newsk->sk_send_head = NULL;
1637 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1639 sock_reset_flag(newsk, SOCK_DONE);
1641 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1643 /* though it's an empty new sock, the charging may fail
1644 * if sysctl_optmem_max was changed between creation of
1645 * original socket and cloning
1647 is_charged = sk_filter_charge(newsk, filter);
1649 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1650 /* We need to make sure that we don't uncharge the new
1651 * socket if we couldn't charge it in the first place
1652 * as otherwise we uncharge the parent's filter.
1655 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1656 sk_free_unlock_clone(newsk);
1660 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1663 newsk->sk_err_soft = 0;
1664 newsk->sk_priority = 0;
1665 newsk->sk_incoming_cpu = raw_smp_processor_id();
1666 atomic64_set(&newsk->sk_cookie, 0);
1668 mem_cgroup_sk_alloc(newsk);
1669 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1672 * Before updating sk_refcnt, we must commit prior changes to memory
1673 * (Documentation/RCU/rculist_nulls.txt for details)
1676 atomic_set(&newsk->sk_refcnt, 2);
1679 * Increment the counter in the same struct proto as the master
1680 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1681 * is the same as sk->sk_prot->socks, as this field was copied
1684 * This _changes_ the previous behaviour, where
1685 * tcp_create_openreq_child always was incrementing the
1686 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1687 * to be taken into account in all callers. -acme
1689 sk_refcnt_debug_inc(newsk);
1690 sk_set_socket(newsk, NULL);
1691 newsk->sk_wq = NULL;
1693 if (newsk->sk_prot->sockets_allocated)
1694 sk_sockets_allocated_inc(newsk);
1696 if (sock_needs_netstamp(sk) &&
1697 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1698 net_enable_timestamp();
1703 EXPORT_SYMBOL_GPL(sk_clone_lock);
1705 void sk_free_unlock_clone(struct sock *sk)
1707 /* It is still raw copy of parent, so invalidate
1708 * destructor and make plain sk_free() */
1709 sk->sk_destruct = NULL;
1713 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1715 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1719 sk_dst_set(sk, dst);
1720 sk->sk_route_caps = dst->dev->features;
1721 if (sk->sk_route_caps & NETIF_F_GSO)
1722 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1723 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1724 if (sk_can_gso(sk)) {
1725 if (dst->header_len) {
1726 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1728 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1729 sk->sk_gso_max_size = dst->dev->gso_max_size;
1730 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1733 sk->sk_gso_max_segs = max_segs;
1735 EXPORT_SYMBOL_GPL(sk_setup_caps);
1738 * Simple resource managers for sockets.
1743 * Write buffer destructor automatically called from kfree_skb.
1745 void sock_wfree(struct sk_buff *skb)
1747 struct sock *sk = skb->sk;
1748 unsigned int len = skb->truesize;
1750 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1752 * Keep a reference on sk_wmem_alloc, this will be released
1753 * after sk_write_space() call
1755 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1756 sk->sk_write_space(sk);
1760 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1761 * could not do because of in-flight packets
1763 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1766 EXPORT_SYMBOL(sock_wfree);
1768 /* This variant of sock_wfree() is used by TCP,
1769 * since it sets SOCK_USE_WRITE_QUEUE.
1771 void __sock_wfree(struct sk_buff *skb)
1773 struct sock *sk = skb->sk;
1775 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1779 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1784 if (unlikely(!sk_fullsock(sk))) {
1785 skb->destructor = sock_edemux;
1790 skb->destructor = sock_wfree;
1791 skb_set_hash_from_sk(skb, sk);
1793 * We used to take a refcount on sk, but following operation
1794 * is enough to guarantee sk_free() wont free this sock until
1795 * all in-flight packets are completed
1797 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1799 EXPORT_SYMBOL(skb_set_owner_w);
1801 /* This helper is used by netem, as it can hold packets in its
1802 * delay queue. We want to allow the owner socket to send more
1803 * packets, as if they were already TX completed by a typical driver.
1804 * But we also want to keep skb->sk set because some packet schedulers
1805 * rely on it (sch_fq for example). So we set skb->truesize to a small
1806 * amount (1) and decrease sk_wmem_alloc accordingly.
1808 void skb_orphan_partial(struct sk_buff *skb)
1810 /* If this skb is a TCP pure ACK or already went here,
1811 * we have nothing to do. 2 is already a very small truesize.
1813 if (skb->truesize <= 2)
1816 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1817 * so we do not completely orphan skb, but transfert all
1818 * accounted bytes but one, to avoid unexpected reorders.
1820 if (skb->destructor == sock_wfree
1822 || skb->destructor == tcp_wfree
1825 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1831 EXPORT_SYMBOL(skb_orphan_partial);
1834 * Read buffer destructor automatically called from kfree_skb.
1836 void sock_rfree(struct sk_buff *skb)
1838 struct sock *sk = skb->sk;
1839 unsigned int len = skb->truesize;
1841 atomic_sub(len, &sk->sk_rmem_alloc);
1842 sk_mem_uncharge(sk, len);
1844 EXPORT_SYMBOL(sock_rfree);
1847 * Buffer destructor for skbs that are not used directly in read or write
1848 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1850 void sock_efree(struct sk_buff *skb)
1854 EXPORT_SYMBOL(sock_efree);
1856 kuid_t sock_i_uid(struct sock *sk)
1860 read_lock_bh(&sk->sk_callback_lock);
1861 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1862 read_unlock_bh(&sk->sk_callback_lock);
1865 EXPORT_SYMBOL(sock_i_uid);
1867 unsigned long sock_i_ino(struct sock *sk)
1871 read_lock_bh(&sk->sk_callback_lock);
1872 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1873 read_unlock_bh(&sk->sk_callback_lock);
1876 EXPORT_SYMBOL(sock_i_ino);
1879 * Allocate a skb from the socket's send buffer.
1881 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1884 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1885 struct sk_buff *skb = alloc_skb(size, priority);
1887 skb_set_owner_w(skb, sk);
1893 EXPORT_SYMBOL(sock_wmalloc);
1896 * Allocate a memory block from the socket's option memory buffer.
1898 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1900 if ((unsigned int)size <= sysctl_optmem_max &&
1901 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1903 /* First do the add, to avoid the race if kmalloc
1906 atomic_add(size, &sk->sk_omem_alloc);
1907 mem = kmalloc(size, priority);
1910 atomic_sub(size, &sk->sk_omem_alloc);
1914 EXPORT_SYMBOL(sock_kmalloc);
1916 /* Free an option memory block. Note, we actually want the inline
1917 * here as this allows gcc to detect the nullify and fold away the
1918 * condition entirely.
1920 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1923 if (WARN_ON_ONCE(!mem))
1929 atomic_sub(size, &sk->sk_omem_alloc);
1932 void sock_kfree_s(struct sock *sk, void *mem, int size)
1934 __sock_kfree_s(sk, mem, size, false);
1936 EXPORT_SYMBOL(sock_kfree_s);
1938 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1940 __sock_kfree_s(sk, mem, size, true);
1942 EXPORT_SYMBOL(sock_kzfree_s);
1944 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1945 I think, these locks should be removed for datagram sockets.
1947 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1951 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1955 if (signal_pending(current))
1957 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1958 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1959 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1961 if (sk->sk_shutdown & SEND_SHUTDOWN)
1965 timeo = schedule_timeout(timeo);
1967 finish_wait(sk_sleep(sk), &wait);
1973 * Generic send/receive buffer handlers
1976 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1977 unsigned long data_len, int noblock,
1978 int *errcode, int max_page_order)
1980 struct sk_buff *skb;
1984 timeo = sock_sndtimeo(sk, noblock);
1986 err = sock_error(sk);
1991 if (sk->sk_shutdown & SEND_SHUTDOWN)
1994 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1997 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1998 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2002 if (signal_pending(current))
2004 timeo = sock_wait_for_wmem(sk, timeo);
2006 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2007 errcode, sk->sk_allocation);
2009 skb_set_owner_w(skb, sk);
2013 err = sock_intr_errno(timeo);
2018 EXPORT_SYMBOL(sock_alloc_send_pskb);
2020 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2021 int noblock, int *errcode)
2023 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2025 EXPORT_SYMBOL(sock_alloc_send_skb);
2027 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2028 struct sockcm_cookie *sockc)
2032 switch (cmsg->cmsg_type) {
2034 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2036 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2038 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2040 case SO_TIMESTAMPING:
2041 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2044 tsflags = *(u32 *)CMSG_DATA(cmsg);
2045 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2048 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2049 sockc->tsflags |= tsflags;
2051 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2053 case SCM_CREDENTIALS:
2060 EXPORT_SYMBOL(__sock_cmsg_send);
2062 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2063 struct sockcm_cookie *sockc)
2065 struct cmsghdr *cmsg;
2068 for_each_cmsghdr(cmsg, msg) {
2069 if (!CMSG_OK(msg, cmsg))
2071 if (cmsg->cmsg_level != SOL_SOCKET)
2073 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2079 EXPORT_SYMBOL(sock_cmsg_send);
2081 /* On 32bit arches, an skb frag is limited to 2^15 */
2082 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2085 * skb_page_frag_refill - check that a page_frag contains enough room
2086 * @sz: minimum size of the fragment we want to get
2087 * @pfrag: pointer to page_frag
2088 * @gfp: priority for memory allocation
2090 * Note: While this allocator tries to use high order pages, there is
2091 * no guarantee that allocations succeed. Therefore, @sz MUST be
2092 * less or equal than PAGE_SIZE.
2094 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2097 if (page_ref_count(pfrag->page) == 1) {
2101 if (pfrag->offset + sz <= pfrag->size)
2103 put_page(pfrag->page);
2107 if (SKB_FRAG_PAGE_ORDER) {
2108 /* Avoid direct reclaim but allow kswapd to wake */
2109 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2110 __GFP_COMP | __GFP_NOWARN |
2112 SKB_FRAG_PAGE_ORDER);
2113 if (likely(pfrag->page)) {
2114 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2118 pfrag->page = alloc_page(gfp);
2119 if (likely(pfrag->page)) {
2120 pfrag->size = PAGE_SIZE;
2125 EXPORT_SYMBOL(skb_page_frag_refill);
2127 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2129 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2132 sk_enter_memory_pressure(sk);
2133 sk_stream_moderate_sndbuf(sk);
2136 EXPORT_SYMBOL(sk_page_frag_refill);
2138 static void __lock_sock(struct sock *sk)
2139 __releases(&sk->sk_lock.slock)
2140 __acquires(&sk->sk_lock.slock)
2145 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2146 TASK_UNINTERRUPTIBLE);
2147 spin_unlock_bh(&sk->sk_lock.slock);
2149 spin_lock_bh(&sk->sk_lock.slock);
2150 if (!sock_owned_by_user(sk))
2153 finish_wait(&sk->sk_lock.wq, &wait);
2156 static void __release_sock(struct sock *sk)
2157 __releases(&sk->sk_lock.slock)
2158 __acquires(&sk->sk_lock.slock)
2160 struct sk_buff *skb, *next;
2162 while ((skb = sk->sk_backlog.head) != NULL) {
2163 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2165 spin_unlock_bh(&sk->sk_lock.slock);
2170 WARN_ON_ONCE(skb_dst_is_noref(skb));
2172 sk_backlog_rcv(sk, skb);
2177 } while (skb != NULL);
2179 spin_lock_bh(&sk->sk_lock.slock);
2183 * Doing the zeroing here guarantee we can not loop forever
2184 * while a wild producer attempts to flood us.
2186 sk->sk_backlog.len = 0;
2189 void __sk_flush_backlog(struct sock *sk)
2191 spin_lock_bh(&sk->sk_lock.slock);
2193 spin_unlock_bh(&sk->sk_lock.slock);
2197 * sk_wait_data - wait for data to arrive at sk_receive_queue
2198 * @sk: sock to wait on
2199 * @timeo: for how long
2200 * @skb: last skb seen on sk_receive_queue
2202 * Now socket state including sk->sk_err is changed only under lock,
2203 * hence we may omit checks after joining wait queue.
2204 * We check receive queue before schedule() only as optimization;
2205 * it is very likely that release_sock() added new data.
2207 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2209 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2212 add_wait_queue(sk_sleep(sk), &wait);
2213 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2214 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2215 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2216 remove_wait_queue(sk_sleep(sk), &wait);
2219 EXPORT_SYMBOL(sk_wait_data);
2222 * __sk_mem_raise_allocated - increase memory_allocated
2224 * @size: memory size to allocate
2225 * @amt: pages to allocate
2226 * @kind: allocation type
2228 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2230 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2232 struct proto *prot = sk->sk_prot;
2233 long allocated = sk_memory_allocated_add(sk, amt);
2235 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2236 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2237 goto suppress_allocation;
2240 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2241 sk_leave_memory_pressure(sk);
2245 /* Under pressure. */
2246 if (allocated > sk_prot_mem_limits(sk, 1))
2247 sk_enter_memory_pressure(sk);
2249 /* Over hard limit. */
2250 if (allocated > sk_prot_mem_limits(sk, 2))
2251 goto suppress_allocation;
2253 /* guarantee minimum buffer size under pressure */
2254 if (kind == SK_MEM_RECV) {
2255 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2258 } else { /* SK_MEM_SEND */
2259 if (sk->sk_type == SOCK_STREAM) {
2260 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2262 } else if (atomic_read(&sk->sk_wmem_alloc) <
2263 prot->sysctl_wmem[0])
2267 if (sk_has_memory_pressure(sk)) {
2270 if (!sk_under_memory_pressure(sk))
2272 alloc = sk_sockets_allocated_read_positive(sk);
2273 if (sk_prot_mem_limits(sk, 2) > alloc *
2274 sk_mem_pages(sk->sk_wmem_queued +
2275 atomic_read(&sk->sk_rmem_alloc) +
2276 sk->sk_forward_alloc))
2280 suppress_allocation:
2282 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2283 sk_stream_moderate_sndbuf(sk);
2285 /* Fail only if socket is _under_ its sndbuf.
2286 * In this case we cannot block, so that we have to fail.
2288 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2292 trace_sock_exceed_buf_limit(sk, prot, allocated);
2294 sk_memory_allocated_sub(sk, amt);
2296 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2297 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2301 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2304 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2306 * @size: memory size to allocate
2307 * @kind: allocation type
2309 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2310 * rmem allocation. This function assumes that protocols which have
2311 * memory_pressure use sk_wmem_queued as write buffer accounting.
2313 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2315 int ret, amt = sk_mem_pages(size);
2317 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2318 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2320 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2323 EXPORT_SYMBOL(__sk_mem_schedule);
2326 * __sk_mem_reduce_allocated - reclaim memory_allocated
2328 * @amount: number of quanta
2330 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2332 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2334 sk_memory_allocated_sub(sk, amount);
2336 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2337 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2339 if (sk_under_memory_pressure(sk) &&
2340 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2341 sk_leave_memory_pressure(sk);
2343 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2346 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2348 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2350 void __sk_mem_reclaim(struct sock *sk, int amount)
2352 amount >>= SK_MEM_QUANTUM_SHIFT;
2353 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2354 __sk_mem_reduce_allocated(sk, amount);
2356 EXPORT_SYMBOL(__sk_mem_reclaim);
2358 int sk_set_peek_off(struct sock *sk, int val)
2363 sk->sk_peek_off = val;
2366 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2369 * Set of default routines for initialising struct proto_ops when
2370 * the protocol does not support a particular function. In certain
2371 * cases where it makes no sense for a protocol to have a "do nothing"
2372 * function, some default processing is provided.
2375 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2379 EXPORT_SYMBOL(sock_no_bind);
2381 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2386 EXPORT_SYMBOL(sock_no_connect);
2388 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2392 EXPORT_SYMBOL(sock_no_socketpair);
2394 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2399 EXPORT_SYMBOL(sock_no_accept);
2401 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2406 EXPORT_SYMBOL(sock_no_getname);
2408 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2412 EXPORT_SYMBOL(sock_no_poll);
2414 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2418 EXPORT_SYMBOL(sock_no_ioctl);
2420 int sock_no_listen(struct socket *sock, int backlog)
2424 EXPORT_SYMBOL(sock_no_listen);
2426 int sock_no_shutdown(struct socket *sock, int how)
2430 EXPORT_SYMBOL(sock_no_shutdown);
2432 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2433 char __user *optval, unsigned int optlen)
2437 EXPORT_SYMBOL(sock_no_setsockopt);
2439 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2440 char __user *optval, int __user *optlen)
2444 EXPORT_SYMBOL(sock_no_getsockopt);
2446 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2450 EXPORT_SYMBOL(sock_no_sendmsg);
2452 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2457 EXPORT_SYMBOL(sock_no_recvmsg);
2459 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2461 /* Mirror missing mmap method error code */
2464 EXPORT_SYMBOL(sock_no_mmap);
2466 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2469 struct msghdr msg = {.msg_flags = flags};
2471 char *kaddr = kmap(page);
2472 iov.iov_base = kaddr + offset;
2474 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2478 EXPORT_SYMBOL(sock_no_sendpage);
2481 * Default Socket Callbacks
2484 static void sock_def_wakeup(struct sock *sk)
2486 struct socket_wq *wq;
2489 wq = rcu_dereference(sk->sk_wq);
2490 if (skwq_has_sleeper(wq))
2491 wake_up_interruptible_all(&wq->wait);
2495 static void sock_def_error_report(struct sock *sk)
2497 struct socket_wq *wq;
2500 wq = rcu_dereference(sk->sk_wq);
2501 if (skwq_has_sleeper(wq))
2502 wake_up_interruptible_poll(&wq->wait, POLLERR);
2503 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2507 static void sock_def_readable(struct sock *sk)
2509 struct socket_wq *wq;
2512 wq = rcu_dereference(sk->sk_wq);
2513 if (skwq_has_sleeper(wq))
2514 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2515 POLLRDNORM | POLLRDBAND);
2516 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2520 static void sock_def_write_space(struct sock *sk)
2522 struct socket_wq *wq;
2526 /* Do not wake up a writer until he can make "significant"
2529 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2530 wq = rcu_dereference(sk->sk_wq);
2531 if (skwq_has_sleeper(wq))
2532 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2533 POLLWRNORM | POLLWRBAND);
2535 /* Should agree with poll, otherwise some programs break */
2536 if (sock_writeable(sk))
2537 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2543 static void sock_def_destruct(struct sock *sk)
2547 void sk_send_sigurg(struct sock *sk)
2549 if (sk->sk_socket && sk->sk_socket->file)
2550 if (send_sigurg(&sk->sk_socket->file->f_owner))
2551 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2553 EXPORT_SYMBOL(sk_send_sigurg);
2555 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2556 unsigned long expires)
2558 if (!mod_timer(timer, expires))
2561 EXPORT_SYMBOL(sk_reset_timer);
2563 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2565 if (del_timer(timer))
2568 EXPORT_SYMBOL(sk_stop_timer);
2570 void sock_init_data(struct socket *sock, struct sock *sk)
2573 sk->sk_send_head = NULL;
2575 init_timer(&sk->sk_timer);
2577 sk->sk_allocation = GFP_KERNEL;
2578 sk->sk_rcvbuf = sysctl_rmem_default;
2579 sk->sk_sndbuf = sysctl_wmem_default;
2580 sk->sk_state = TCP_CLOSE;
2581 sk_set_socket(sk, sock);
2583 sock_set_flag(sk, SOCK_ZAPPED);
2586 sk->sk_type = sock->type;
2587 sk->sk_wq = sock->wq;
2589 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2592 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2595 rwlock_init(&sk->sk_callback_lock);
2596 if (sk->sk_kern_sock)
2597 lockdep_set_class_and_name(
2598 &sk->sk_callback_lock,
2599 af_kern_callback_keys + sk->sk_family,
2600 af_family_kern_clock_key_strings[sk->sk_family]);
2602 lockdep_set_class_and_name(
2603 &sk->sk_callback_lock,
2604 af_callback_keys + sk->sk_family,
2605 af_family_clock_key_strings[sk->sk_family]);
2607 sk->sk_state_change = sock_def_wakeup;
2608 sk->sk_data_ready = sock_def_readable;
2609 sk->sk_write_space = sock_def_write_space;
2610 sk->sk_error_report = sock_def_error_report;
2611 sk->sk_destruct = sock_def_destruct;
2613 sk->sk_frag.page = NULL;
2614 sk->sk_frag.offset = 0;
2615 sk->sk_peek_off = -1;
2617 sk->sk_peer_pid = NULL;
2618 sk->sk_peer_cred = NULL;
2619 sk->sk_write_pending = 0;
2620 sk->sk_rcvlowat = 1;
2621 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2622 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2624 sk->sk_stamp = SK_DEFAULT_STAMP;
2626 #ifdef CONFIG_NET_RX_BUSY_POLL
2628 sk->sk_ll_usec = sysctl_net_busy_read;
2631 sk->sk_max_pacing_rate = ~0U;
2632 sk->sk_pacing_rate = ~0U;
2633 sk->sk_incoming_cpu = -1;
2635 * Before updating sk_refcnt, we must commit prior changes to memory
2636 * (Documentation/RCU/rculist_nulls.txt for details)
2639 atomic_set(&sk->sk_refcnt, 1);
2640 atomic_set(&sk->sk_drops, 0);
2642 EXPORT_SYMBOL(sock_init_data);
2644 void lock_sock_nested(struct sock *sk, int subclass)
2647 spin_lock_bh(&sk->sk_lock.slock);
2648 if (sk->sk_lock.owned)
2650 sk->sk_lock.owned = 1;
2651 spin_unlock(&sk->sk_lock.slock);
2653 * The sk_lock has mutex_lock() semantics here:
2655 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2658 EXPORT_SYMBOL(lock_sock_nested);
2660 void release_sock(struct sock *sk)
2662 spin_lock_bh(&sk->sk_lock.slock);
2663 if (sk->sk_backlog.tail)
2666 /* Warning : release_cb() might need to release sk ownership,
2667 * ie call sock_release_ownership(sk) before us.
2669 if (sk->sk_prot->release_cb)
2670 sk->sk_prot->release_cb(sk);
2672 sock_release_ownership(sk);
2673 if (waitqueue_active(&sk->sk_lock.wq))
2674 wake_up(&sk->sk_lock.wq);
2675 spin_unlock_bh(&sk->sk_lock.slock);
2677 EXPORT_SYMBOL(release_sock);
2680 * lock_sock_fast - fast version of lock_sock
2683 * This version should be used for very small section, where process wont block
2684 * return false if fast path is taken
2685 * sk_lock.slock locked, owned = 0, BH disabled
2686 * return true if slow path is taken
2687 * sk_lock.slock unlocked, owned = 1, BH enabled
2689 bool lock_sock_fast(struct sock *sk)
2692 spin_lock_bh(&sk->sk_lock.slock);
2694 if (!sk->sk_lock.owned)
2696 * Note : We must disable BH
2701 sk->sk_lock.owned = 1;
2702 spin_unlock(&sk->sk_lock.slock);
2704 * The sk_lock has mutex_lock() semantics here:
2706 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2710 EXPORT_SYMBOL(lock_sock_fast);
2712 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2715 if (!sock_flag(sk, SOCK_TIMESTAMP))
2716 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2717 tv = ktime_to_timeval(sk->sk_stamp);
2718 if (tv.tv_sec == -1)
2720 if (tv.tv_sec == 0) {
2721 sk->sk_stamp = ktime_get_real();
2722 tv = ktime_to_timeval(sk->sk_stamp);
2724 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2726 EXPORT_SYMBOL(sock_get_timestamp);
2728 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2731 if (!sock_flag(sk, SOCK_TIMESTAMP))
2732 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2733 ts = ktime_to_timespec(sk->sk_stamp);
2734 if (ts.tv_sec == -1)
2736 if (ts.tv_sec == 0) {
2737 sk->sk_stamp = ktime_get_real();
2738 ts = ktime_to_timespec(sk->sk_stamp);
2740 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2742 EXPORT_SYMBOL(sock_get_timestampns);
2744 void sock_enable_timestamp(struct sock *sk, int flag)
2746 if (!sock_flag(sk, flag)) {
2747 unsigned long previous_flags = sk->sk_flags;
2749 sock_set_flag(sk, flag);
2751 * we just set one of the two flags which require net
2752 * time stamping, but time stamping might have been on
2753 * already because of the other one
2755 if (sock_needs_netstamp(sk) &&
2756 !(previous_flags & SK_FLAGS_TIMESTAMP))
2757 net_enable_timestamp();
2761 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2762 int level, int type)
2764 struct sock_exterr_skb *serr;
2765 struct sk_buff *skb;
2769 skb = sock_dequeue_err_skb(sk);
2775 msg->msg_flags |= MSG_TRUNC;
2778 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2782 sock_recv_timestamp(msg, sk, skb);
2784 serr = SKB_EXT_ERR(skb);
2785 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2787 msg->msg_flags |= MSG_ERRQUEUE;
2795 EXPORT_SYMBOL(sock_recv_errqueue);
2798 * Get a socket option on an socket.
2800 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2801 * asynchronous errors should be reported by getsockopt. We assume
2802 * this means if you specify SO_ERROR (otherwise whats the point of it).
2804 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2805 char __user *optval, int __user *optlen)
2807 struct sock *sk = sock->sk;
2809 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2811 EXPORT_SYMBOL(sock_common_getsockopt);
2813 #ifdef CONFIG_COMPAT
2814 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2815 char __user *optval, int __user *optlen)
2817 struct sock *sk = sock->sk;
2819 if (sk->sk_prot->compat_getsockopt != NULL)
2820 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2822 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2824 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2827 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2830 struct sock *sk = sock->sk;
2834 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2835 flags & ~MSG_DONTWAIT, &addr_len);
2837 msg->msg_namelen = addr_len;
2840 EXPORT_SYMBOL(sock_common_recvmsg);
2843 * Set socket options on an inet socket.
2845 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2846 char __user *optval, unsigned int optlen)
2848 struct sock *sk = sock->sk;
2850 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2852 EXPORT_SYMBOL(sock_common_setsockopt);
2854 #ifdef CONFIG_COMPAT
2855 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2856 char __user *optval, unsigned int optlen)
2858 struct sock *sk = sock->sk;
2860 if (sk->sk_prot->compat_setsockopt != NULL)
2861 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2863 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2865 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2868 void sk_common_release(struct sock *sk)
2870 if (sk->sk_prot->destroy)
2871 sk->sk_prot->destroy(sk);
2874 * Observation: when sock_common_release is called, processes have
2875 * no access to socket. But net still has.
2876 * Step one, detach it from networking:
2878 * A. Remove from hash tables.
2881 sk->sk_prot->unhash(sk);
2884 * In this point socket cannot receive new packets, but it is possible
2885 * that some packets are in flight because some CPU runs receiver and
2886 * did hash table lookup before we unhashed socket. They will achieve
2887 * receive queue and will be purged by socket destructor.
2889 * Also we still have packets pending on receive queue and probably,
2890 * our own packets waiting in device queues. sock_destroy will drain
2891 * receive queue, but transmitted packets will delay socket destruction
2892 * until the last reference will be released.
2897 xfrm_sk_free_policy(sk);
2899 sk_refcnt_debug_release(sk);
2903 EXPORT_SYMBOL(sk_common_release);
2905 void sk_get_meminfo(const struct sock *sk, u32 *mem)
2907 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
2909 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
2910 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
2911 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
2912 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
2913 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
2914 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
2915 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
2916 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
2917 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
2920 #ifdef CONFIG_PROC_FS
2921 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2923 int val[PROTO_INUSE_NR];
2926 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2928 #ifdef CONFIG_NET_NS
2929 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2931 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2933 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2935 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2937 int cpu, idx = prot->inuse_idx;
2940 for_each_possible_cpu(cpu)
2941 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2943 return res >= 0 ? res : 0;
2945 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2947 static int __net_init sock_inuse_init_net(struct net *net)
2949 net->core.inuse = alloc_percpu(struct prot_inuse);
2950 return net->core.inuse ? 0 : -ENOMEM;
2953 static void __net_exit sock_inuse_exit_net(struct net *net)
2955 free_percpu(net->core.inuse);
2958 static struct pernet_operations net_inuse_ops = {
2959 .init = sock_inuse_init_net,
2960 .exit = sock_inuse_exit_net,
2963 static __init int net_inuse_init(void)
2965 if (register_pernet_subsys(&net_inuse_ops))
2966 panic("Cannot initialize net inuse counters");
2971 core_initcall(net_inuse_init);
2973 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2975 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2977 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2979 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2981 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2983 int cpu, idx = prot->inuse_idx;
2986 for_each_possible_cpu(cpu)
2987 res += per_cpu(prot_inuse, cpu).val[idx];
2989 return res >= 0 ? res : 0;
2991 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2994 static void assign_proto_idx(struct proto *prot)
2996 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2998 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2999 pr_err("PROTO_INUSE_NR exhausted\n");
3003 set_bit(prot->inuse_idx, proto_inuse_idx);
3006 static void release_proto_idx(struct proto *prot)
3008 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3009 clear_bit(prot->inuse_idx, proto_inuse_idx);
3012 static inline void assign_proto_idx(struct proto *prot)
3016 static inline void release_proto_idx(struct proto *prot)
3021 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3025 kfree(rsk_prot->slab_name);
3026 rsk_prot->slab_name = NULL;
3027 kmem_cache_destroy(rsk_prot->slab);
3028 rsk_prot->slab = NULL;
3031 static int req_prot_init(const struct proto *prot)
3033 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3038 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3040 if (!rsk_prot->slab_name)
3043 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3044 rsk_prot->obj_size, 0,
3045 prot->slab_flags, NULL);
3047 if (!rsk_prot->slab) {
3048 pr_crit("%s: Can't create request sock SLAB cache!\n",
3055 int proto_register(struct proto *prot, int alloc_slab)
3058 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
3059 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3062 if (prot->slab == NULL) {
3063 pr_crit("%s: Can't create sock SLAB cache!\n",
3068 if (req_prot_init(prot))
3069 goto out_free_request_sock_slab;
3071 if (prot->twsk_prot != NULL) {
3072 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3074 if (prot->twsk_prot->twsk_slab_name == NULL)
3075 goto out_free_request_sock_slab;
3077 prot->twsk_prot->twsk_slab =
3078 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3079 prot->twsk_prot->twsk_obj_size,
3083 if (prot->twsk_prot->twsk_slab == NULL)
3084 goto out_free_timewait_sock_slab_name;
3088 mutex_lock(&proto_list_mutex);
3089 list_add(&prot->node, &proto_list);
3090 assign_proto_idx(prot);
3091 mutex_unlock(&proto_list_mutex);
3094 out_free_timewait_sock_slab_name:
3095 kfree(prot->twsk_prot->twsk_slab_name);
3096 out_free_request_sock_slab:
3097 req_prot_cleanup(prot->rsk_prot);
3099 kmem_cache_destroy(prot->slab);
3104 EXPORT_SYMBOL(proto_register);
3106 void proto_unregister(struct proto *prot)
3108 mutex_lock(&proto_list_mutex);
3109 release_proto_idx(prot);
3110 list_del(&prot->node);
3111 mutex_unlock(&proto_list_mutex);
3113 kmem_cache_destroy(prot->slab);
3116 req_prot_cleanup(prot->rsk_prot);
3118 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3119 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3120 kfree(prot->twsk_prot->twsk_slab_name);
3121 prot->twsk_prot->twsk_slab = NULL;
3124 EXPORT_SYMBOL(proto_unregister);
3126 #ifdef CONFIG_PROC_FS
3127 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3128 __acquires(proto_list_mutex)
3130 mutex_lock(&proto_list_mutex);
3131 return seq_list_start_head(&proto_list, *pos);
3134 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3136 return seq_list_next(v, &proto_list, pos);
3139 static void proto_seq_stop(struct seq_file *seq, void *v)
3140 __releases(proto_list_mutex)
3142 mutex_unlock(&proto_list_mutex);
3145 static char proto_method_implemented(const void *method)
3147 return method == NULL ? 'n' : 'y';
3149 static long sock_prot_memory_allocated(struct proto *proto)
3151 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3154 static char *sock_prot_memory_pressure(struct proto *proto)
3156 return proto->memory_pressure != NULL ?
3157 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3160 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3163 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3164 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3167 sock_prot_inuse_get(seq_file_net(seq), proto),
3168 sock_prot_memory_allocated(proto),
3169 sock_prot_memory_pressure(proto),
3171 proto->slab == NULL ? "no" : "yes",
3172 module_name(proto->owner),
3173 proto_method_implemented(proto->close),
3174 proto_method_implemented(proto->connect),
3175 proto_method_implemented(proto->disconnect),
3176 proto_method_implemented(proto->accept),
3177 proto_method_implemented(proto->ioctl),
3178 proto_method_implemented(proto->init),
3179 proto_method_implemented(proto->destroy),
3180 proto_method_implemented(proto->shutdown),
3181 proto_method_implemented(proto->setsockopt),
3182 proto_method_implemented(proto->getsockopt),
3183 proto_method_implemented(proto->sendmsg),
3184 proto_method_implemented(proto->recvmsg),
3185 proto_method_implemented(proto->sendpage),
3186 proto_method_implemented(proto->bind),
3187 proto_method_implemented(proto->backlog_rcv),
3188 proto_method_implemented(proto->hash),
3189 proto_method_implemented(proto->unhash),
3190 proto_method_implemented(proto->get_port),
3191 proto_method_implemented(proto->enter_memory_pressure));
3194 static int proto_seq_show(struct seq_file *seq, void *v)
3196 if (v == &proto_list)
3197 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3206 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3208 proto_seq_printf(seq, list_entry(v, struct proto, node));
3212 static const struct seq_operations proto_seq_ops = {
3213 .start = proto_seq_start,
3214 .next = proto_seq_next,
3215 .stop = proto_seq_stop,
3216 .show = proto_seq_show,
3219 static int proto_seq_open(struct inode *inode, struct file *file)
3221 return seq_open_net(inode, file, &proto_seq_ops,
3222 sizeof(struct seq_net_private));
3225 static const struct file_operations proto_seq_fops = {
3226 .owner = THIS_MODULE,
3227 .open = proto_seq_open,
3229 .llseek = seq_lseek,
3230 .release = seq_release_net,
3233 static __net_init int proto_init_net(struct net *net)
3235 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3241 static __net_exit void proto_exit_net(struct net *net)
3243 remove_proc_entry("protocols", net->proc_net);
3247 static __net_initdata struct pernet_operations proto_net_ops = {
3248 .init = proto_init_net,
3249 .exit = proto_exit_net,
3252 static int __init proto_init(void)
3254 return register_pernet_subsys(&proto_net_ops);
3257 subsys_initcall(proto_init);
3259 #endif /* PROC_FS */
3261 #ifdef CONFIG_NET_RX_BUSY_POLL
3262 bool sk_busy_loop_end(void *p, unsigned long start_time)
3264 struct sock *sk = p;
3266 return !skb_queue_empty(&sk->sk_receive_queue) ||
3267 sk_busy_loop_timeout(sk, start_time);
3269 EXPORT_SYMBOL(sk_busy_loop_end);
3270 #endif /* CONFIG_NET_RX_BUSY_POLL */