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.
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/sched/mm.h>
106 #include <linux/timer.h>
107 #include <linux/string.h>
108 #include <linux/sockios.h>
109 #include <linux/net.h>
110 #include <linux/mm.h>
111 #include <linux/slab.h>
112 #include <linux/interrupt.h>
113 #include <linux/poll.h>
114 #include <linux/tcp.h>
115 #include <linux/init.h>
116 #include <linux/highmem.h>
117 #include <linux/user_namespace.h>
118 #include <linux/static_key.h>
119 #include <linux/memcontrol.h>
120 #include <linux/prefetch.h>
122 #include <linux/uaccess.h>
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
140 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
149 * sk_ns_capable - General socket capability test
150 * @sk: Socket to use a capability on or through
151 * @user_ns: The user namespace of the capability to use
152 * @cap: The capability to use
154 * Test to see if the opener of the socket had when the socket was
155 * created and the current process has the capability @cap in the user
156 * namespace @user_ns.
158 bool sk_ns_capable(const struct sock *sk,
159 struct user_namespace *user_ns, int cap)
161 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162 ns_capable(user_ns, cap);
164 EXPORT_SYMBOL(sk_ns_capable);
167 * sk_capable - Socket global capability test
168 * @sk: Socket to use a capability on or through
169 * @cap: The global capability to use
171 * Test to see if the opener of the socket had when the socket was
172 * created and the current process has the capability @cap in all user
175 bool sk_capable(const struct sock *sk, int cap)
177 return sk_ns_capable(sk, &init_user_ns, cap);
179 EXPORT_SYMBOL(sk_capable);
182 * sk_net_capable - Network namespace socket capability test
183 * @sk: Socket to use a capability on or through
184 * @cap: The capability to use
186 * Test to see if the opener of the socket had when the socket was created
187 * and the current process has the capability @cap over the network namespace
188 * the socket is a member of.
190 bool sk_net_capable(const struct sock *sk, int cap)
192 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 EXPORT_SYMBOL(sk_net_capable);
197 * Each address family might have different locking rules, so we have
198 * one slock key per address family and separate keys for internal and
201 static struct lock_class_key af_family_keys[AF_MAX];
202 static struct lock_class_key af_family_kern_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
207 * Make lock validator output more readable. (we pre-construct these
208 * strings build-time, so that runtime initialization of socket
212 #define _sock_locks(x) \
213 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
214 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
215 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
216 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
217 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
218 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
219 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
220 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
221 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
222 x "27" , x "28" , x "AF_CAN" , \
223 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
224 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
225 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
226 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
227 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
229 static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
250 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
251 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
252 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
253 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
254 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
255 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
256 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
257 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
258 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
259 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
260 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
261 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
262 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
263 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
265 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
266 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
267 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
268 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
269 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
270 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
271 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
272 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
273 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
274 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
275 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
276 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
277 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
278 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
279 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
280 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
282 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
283 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
284 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
285 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
286 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
287 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
288 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
289 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
290 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
291 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
292 "elock-27" , "elock-28" , "elock-AF_CAN" ,
293 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
294 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
295 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
296 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
297 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
301 * sk_callback_lock and sk queues locking rules are per-address-family,
302 * so split the lock classes by using a per-AF key:
304 static struct lock_class_key af_callback_keys[AF_MAX];
305 static struct lock_class_key af_rlock_keys[AF_MAX];
306 static struct lock_class_key af_wlock_keys[AF_MAX];
307 static struct lock_class_key af_elock_keys[AF_MAX];
308 static struct lock_class_key af_kern_callback_keys[AF_MAX];
310 /* Run time adjustable parameters. */
311 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
312 EXPORT_SYMBOL(sysctl_wmem_max);
313 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
314 EXPORT_SYMBOL(sysctl_rmem_max);
315 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
316 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
318 /* Maximal space eaten by iovec or ancillary data plus some space */
319 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
320 EXPORT_SYMBOL(sysctl_optmem_max);
322 int sysctl_tstamp_allow_data __read_mostly = 1;
324 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
325 EXPORT_SYMBOL_GPL(memalloc_socks);
328 * sk_set_memalloc - sets %SOCK_MEMALLOC
329 * @sk: socket to set it on
331 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
332 * It's the responsibility of the admin to adjust min_free_kbytes
333 * to meet the requirements
335 void sk_set_memalloc(struct sock *sk)
337 sock_set_flag(sk, SOCK_MEMALLOC);
338 sk->sk_allocation |= __GFP_MEMALLOC;
339 static_key_slow_inc(&memalloc_socks);
341 EXPORT_SYMBOL_GPL(sk_set_memalloc);
343 void sk_clear_memalloc(struct sock *sk)
345 sock_reset_flag(sk, SOCK_MEMALLOC);
346 sk->sk_allocation &= ~__GFP_MEMALLOC;
347 static_key_slow_dec(&memalloc_socks);
350 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
351 * progress of swapping. SOCK_MEMALLOC may be cleared while
352 * it has rmem allocations due to the last swapfile being deactivated
353 * but there is a risk that the socket is unusable due to exceeding
354 * the rmem limits. Reclaim the reserves and obey rmem limits again.
358 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
360 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
363 unsigned int noreclaim_flag;
365 /* these should have been dropped before queueing */
366 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
368 noreclaim_flag = memalloc_noreclaim_save();
369 ret = sk->sk_backlog_rcv(sk, skb);
370 memalloc_noreclaim_restore(noreclaim_flag);
374 EXPORT_SYMBOL(__sk_backlog_rcv);
376 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
380 if (optlen < sizeof(tv))
382 if (copy_from_user(&tv, optval, sizeof(tv)))
384 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
388 static int warned __read_mostly;
391 if (warned < 10 && net_ratelimit()) {
393 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
394 __func__, current->comm, task_pid_nr(current));
398 *timeo_p = MAX_SCHEDULE_TIMEOUT;
399 if (tv.tv_sec == 0 && tv.tv_usec == 0)
401 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
402 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
406 static void sock_warn_obsolete_bsdism(const char *name)
409 static char warncomm[TASK_COMM_LEN];
410 if (strcmp(warncomm, current->comm) && warned < 5) {
411 strcpy(warncomm, current->comm);
412 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
418 static bool sock_needs_netstamp(const struct sock *sk)
420 switch (sk->sk_family) {
429 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
431 if (sk->sk_flags & flags) {
432 sk->sk_flags &= ~flags;
433 if (sock_needs_netstamp(sk) &&
434 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
435 net_disable_timestamp();
440 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
443 struct sk_buff_head *list = &sk->sk_receive_queue;
445 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
446 atomic_inc(&sk->sk_drops);
447 trace_sock_rcvqueue_full(sk, skb);
451 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
452 atomic_inc(&sk->sk_drops);
457 skb_set_owner_r(skb, sk);
459 /* we escape from rcu protected region, make sure we dont leak
464 spin_lock_irqsave(&list->lock, flags);
465 sock_skb_set_dropcount(sk, skb);
466 __skb_queue_tail(list, skb);
467 spin_unlock_irqrestore(&list->lock, flags);
469 if (!sock_flag(sk, SOCK_DEAD))
470 sk->sk_data_ready(sk);
473 EXPORT_SYMBOL(__sock_queue_rcv_skb);
475 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
479 err = sk_filter(sk, skb);
483 return __sock_queue_rcv_skb(sk, skb);
485 EXPORT_SYMBOL(sock_queue_rcv_skb);
487 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
488 const int nested, unsigned int trim_cap, bool refcounted)
490 int rc = NET_RX_SUCCESS;
492 if (sk_filter_trim_cap(sk, skb, trim_cap))
493 goto discard_and_relse;
497 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
498 atomic_inc(&sk->sk_drops);
499 goto discard_and_relse;
502 bh_lock_sock_nested(sk);
505 if (!sock_owned_by_user(sk)) {
507 * trylock + unlock semantics:
509 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
511 rc = sk_backlog_rcv(sk, skb);
513 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
514 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
516 atomic_inc(&sk->sk_drops);
517 goto discard_and_relse;
529 EXPORT_SYMBOL(__sk_receive_skb);
531 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
533 struct dst_entry *dst = __sk_dst_get(sk);
535 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
536 sk_tx_queue_clear(sk);
537 sk->sk_dst_pending_confirm = 0;
538 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
545 EXPORT_SYMBOL(__sk_dst_check);
547 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
549 struct dst_entry *dst = sk_dst_get(sk);
551 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
559 EXPORT_SYMBOL(sk_dst_check);
561 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
564 int ret = -ENOPROTOOPT;
565 #ifdef CONFIG_NETDEVICES
566 struct net *net = sock_net(sk);
567 char devname[IFNAMSIZ];
572 if (!ns_capable(net->user_ns, CAP_NET_RAW))
579 /* Bind this socket to a particular device like "eth0",
580 * as specified in the passed interface name. If the
581 * name is "" or the option length is zero the socket
584 if (optlen > IFNAMSIZ - 1)
585 optlen = IFNAMSIZ - 1;
586 memset(devname, 0, sizeof(devname));
589 if (copy_from_user(devname, optval, optlen))
593 if (devname[0] != '\0') {
594 struct net_device *dev;
597 dev = dev_get_by_name_rcu(net, devname);
599 index = dev->ifindex;
607 sk->sk_bound_dev_if = index;
619 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
620 int __user *optlen, int len)
622 int ret = -ENOPROTOOPT;
623 #ifdef CONFIG_NETDEVICES
624 struct net *net = sock_net(sk);
625 char devname[IFNAMSIZ];
627 if (sk->sk_bound_dev_if == 0) {
636 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
640 len = strlen(devname) + 1;
643 if (copy_to_user(optval, devname, len))
648 if (put_user(len, optlen))
659 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
662 sock_set_flag(sk, bit);
664 sock_reset_flag(sk, bit);
667 bool sk_mc_loop(struct sock *sk)
669 if (dev_recursion_level())
673 switch (sk->sk_family) {
675 return inet_sk(sk)->mc_loop;
676 #if IS_ENABLED(CONFIG_IPV6)
678 return inet6_sk(sk)->mc_loop;
684 EXPORT_SYMBOL(sk_mc_loop);
687 * This is meant for all protocols to use and covers goings on
688 * at the socket level. Everything here is generic.
691 int sock_setsockopt(struct socket *sock, int level, int optname,
692 char __user *optval, unsigned int optlen)
694 struct sock *sk = sock->sk;
701 * Options without arguments
704 if (optname == SO_BINDTODEVICE)
705 return sock_setbindtodevice(sk, optval, optlen);
707 if (optlen < sizeof(int))
710 if (get_user(val, (int __user *)optval))
713 valbool = val ? 1 : 0;
719 if (val && !capable(CAP_NET_ADMIN))
722 sock_valbool_flag(sk, SOCK_DBG, valbool);
725 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
728 sk->sk_reuseport = valbool;
737 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
740 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
743 /* Don't error on this BSD doesn't and if you think
744 * about it this is right. Otherwise apps have to
745 * play 'guess the biggest size' games. RCVBUF/SNDBUF
746 * are treated in BSD as hints
748 val = min_t(u32, val, sysctl_wmem_max);
750 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
751 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
752 /* Wake up sending tasks if we upped the value. */
753 sk->sk_write_space(sk);
757 if (!capable(CAP_NET_ADMIN)) {
764 /* Don't error on this BSD doesn't and if you think
765 * about it this is right. Otherwise apps have to
766 * play 'guess the biggest size' games. RCVBUF/SNDBUF
767 * are treated in BSD as hints
769 val = min_t(u32, val, sysctl_rmem_max);
771 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
773 * We double it on the way in to account for
774 * "struct sk_buff" etc. overhead. Applications
775 * assume that the SO_RCVBUF setting they make will
776 * allow that much actual data to be received on that
779 * Applications are unaware that "struct sk_buff" and
780 * other overheads allocate from the receive buffer
781 * during socket buffer allocation.
783 * And after considering the possible alternatives,
784 * returning the value we actually used in getsockopt
785 * is the most desirable behavior.
787 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
791 if (!capable(CAP_NET_ADMIN)) {
798 if (sk->sk_prot->keepalive)
799 sk->sk_prot->keepalive(sk, valbool);
800 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
804 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
808 sk->sk_no_check_tx = valbool;
812 if ((val >= 0 && val <= 6) ||
813 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
814 sk->sk_priority = val;
820 if (optlen < sizeof(ling)) {
821 ret = -EINVAL; /* 1003.1g */
824 if (copy_from_user(&ling, optval, sizeof(ling))) {
829 sock_reset_flag(sk, SOCK_LINGER);
831 #if (BITS_PER_LONG == 32)
832 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
833 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
836 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
837 sock_set_flag(sk, SOCK_LINGER);
842 sock_warn_obsolete_bsdism("setsockopt");
847 set_bit(SOCK_PASSCRED, &sock->flags);
849 clear_bit(SOCK_PASSCRED, &sock->flags);
855 if (optname == SO_TIMESTAMP)
856 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
858 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
859 sock_set_flag(sk, SOCK_RCVTSTAMP);
860 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
862 sock_reset_flag(sk, SOCK_RCVTSTAMP);
863 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
867 case SO_TIMESTAMPING:
868 if (val & ~SOF_TIMESTAMPING_MASK) {
873 if (val & SOF_TIMESTAMPING_OPT_ID &&
874 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
875 if (sk->sk_protocol == IPPROTO_TCP &&
876 sk->sk_type == SOCK_STREAM) {
877 if ((1 << sk->sk_state) &
878 (TCPF_CLOSE | TCPF_LISTEN)) {
882 sk->sk_tskey = tcp_sk(sk)->snd_una;
888 if (val & SOF_TIMESTAMPING_OPT_STATS &&
889 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
894 sk->sk_tsflags = val;
895 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
896 sock_enable_timestamp(sk,
897 SOCK_TIMESTAMPING_RX_SOFTWARE);
899 sock_disable_timestamp(sk,
900 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
906 sk->sk_rcvlowat = val ? : 1;
910 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
914 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
917 case SO_ATTACH_FILTER:
919 if (optlen == sizeof(struct sock_fprog)) {
920 struct sock_fprog fprog;
923 if (copy_from_user(&fprog, optval, sizeof(fprog)))
926 ret = sk_attach_filter(&fprog, sk);
932 if (optlen == sizeof(u32)) {
936 if (copy_from_user(&ufd, optval, sizeof(ufd)))
939 ret = sk_attach_bpf(ufd, sk);
943 case SO_ATTACH_REUSEPORT_CBPF:
945 if (optlen == sizeof(struct sock_fprog)) {
946 struct sock_fprog fprog;
949 if (copy_from_user(&fprog, optval, sizeof(fprog)))
952 ret = sk_reuseport_attach_filter(&fprog, sk);
956 case SO_ATTACH_REUSEPORT_EBPF:
958 if (optlen == sizeof(u32)) {
962 if (copy_from_user(&ufd, optval, sizeof(ufd)))
965 ret = sk_reuseport_attach_bpf(ufd, sk);
969 case SO_DETACH_FILTER:
970 ret = sk_detach_filter(sk);
974 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
977 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
982 set_bit(SOCK_PASSSEC, &sock->flags);
984 clear_bit(SOCK_PASSSEC, &sock->flags);
987 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
994 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
998 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1002 if (sock->ops->set_peek_off)
1003 ret = sock->ops->set_peek_off(sk, val);
1009 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1012 case SO_SELECT_ERR_QUEUE:
1013 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1016 #ifdef CONFIG_NET_RX_BUSY_POLL
1018 /* allow unprivileged users to decrease the value */
1019 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1025 sk->sk_ll_usec = val;
1030 case SO_MAX_PACING_RATE:
1032 cmpxchg(&sk->sk_pacing_status,
1035 sk->sk_max_pacing_rate = val;
1036 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1037 sk->sk_max_pacing_rate);
1040 case SO_INCOMING_CPU:
1041 sk->sk_incoming_cpu = val;
1046 dst_negative_advice(sk);
1050 if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1052 else if (sk->sk_protocol != IPPROTO_TCP)
1054 else if (sk->sk_state != TCP_CLOSE)
1056 else if (val < 0 || val > 1)
1059 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1069 EXPORT_SYMBOL(sock_setsockopt);
1072 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1073 struct ucred *ucred)
1075 ucred->pid = pid_vnr(pid);
1076 ucred->uid = ucred->gid = -1;
1078 struct user_namespace *current_ns = current_user_ns();
1080 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1081 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1085 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1087 struct user_namespace *user_ns = current_user_ns();
1090 for (i = 0; i < src->ngroups; i++)
1091 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1097 int sock_getsockopt(struct socket *sock, int level, int optname,
1098 char __user *optval, int __user *optlen)
1100 struct sock *sk = sock->sk;
1109 int lv = sizeof(int);
1112 if (get_user(len, optlen))
1117 memset(&v, 0, sizeof(v));
1121 v.val = sock_flag(sk, SOCK_DBG);
1125 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1129 v.val = sock_flag(sk, SOCK_BROADCAST);
1133 v.val = sk->sk_sndbuf;
1137 v.val = sk->sk_rcvbuf;
1141 v.val = sk->sk_reuse;
1145 v.val = sk->sk_reuseport;
1149 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1153 v.val = sk->sk_type;
1157 v.val = sk->sk_protocol;
1161 v.val = sk->sk_family;
1165 v.val = -sock_error(sk);
1167 v.val = xchg(&sk->sk_err_soft, 0);
1171 v.val = sock_flag(sk, SOCK_URGINLINE);
1175 v.val = sk->sk_no_check_tx;
1179 v.val = sk->sk_priority;
1183 lv = sizeof(v.ling);
1184 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1185 v.ling.l_linger = sk->sk_lingertime / HZ;
1189 sock_warn_obsolete_bsdism("getsockopt");
1193 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1194 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1197 case SO_TIMESTAMPNS:
1198 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1201 case SO_TIMESTAMPING:
1202 v.val = sk->sk_tsflags;
1206 lv = sizeof(struct timeval);
1207 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1211 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1212 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1217 lv = sizeof(struct timeval);
1218 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1222 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1223 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1228 v.val = sk->sk_rcvlowat;
1236 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1241 struct ucred peercred;
1242 if (len > sizeof(peercred))
1243 len = sizeof(peercred);
1244 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1245 if (copy_to_user(optval, &peercred, len))
1254 if (!sk->sk_peer_cred)
1257 n = sk->sk_peer_cred->group_info->ngroups;
1258 if (len < n * sizeof(gid_t)) {
1259 len = n * sizeof(gid_t);
1260 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1262 len = n * sizeof(gid_t);
1264 ret = groups_to_user((gid_t __user *)optval,
1265 sk->sk_peer_cred->group_info);
1275 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1279 if (copy_to_user(optval, address, len))
1284 /* Dubious BSD thing... Probably nobody even uses it, but
1285 * the UNIX standard wants it for whatever reason... -DaveM
1288 v.val = sk->sk_state == TCP_LISTEN;
1292 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1296 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1299 v.val = sk->sk_mark;
1303 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1306 case SO_WIFI_STATUS:
1307 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1311 if (!sock->ops->set_peek_off)
1314 v.val = sk->sk_peek_off;
1317 v.val = sock_flag(sk, SOCK_NOFCS);
1320 case SO_BINDTODEVICE:
1321 return sock_getbindtodevice(sk, optval, optlen, len);
1324 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1330 case SO_LOCK_FILTER:
1331 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1334 case SO_BPF_EXTENSIONS:
1335 v.val = bpf_tell_extensions();
1338 case SO_SELECT_ERR_QUEUE:
1339 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1342 #ifdef CONFIG_NET_RX_BUSY_POLL
1344 v.val = sk->sk_ll_usec;
1348 case SO_MAX_PACING_RATE:
1349 v.val = sk->sk_max_pacing_rate;
1352 case SO_INCOMING_CPU:
1353 v.val = sk->sk_incoming_cpu;
1358 u32 meminfo[SK_MEMINFO_VARS];
1360 if (get_user(len, optlen))
1363 sk_get_meminfo(sk, meminfo);
1365 len = min_t(unsigned int, len, sizeof(meminfo));
1366 if (copy_to_user(optval, &meminfo, len))
1372 #ifdef CONFIG_NET_RX_BUSY_POLL
1373 case SO_INCOMING_NAPI_ID:
1374 v.val = READ_ONCE(sk->sk_napi_id);
1376 /* aggregate non-NAPI IDs down to 0 */
1377 if (v.val < MIN_NAPI_ID)
1387 v.val64 = sock_gen_cookie(sk);
1391 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1395 /* We implement the SO_SNDLOWAT etc to not be settable
1398 return -ENOPROTOOPT;
1403 if (copy_to_user(optval, &v, len))
1406 if (put_user(len, optlen))
1412 * Initialize an sk_lock.
1414 * (We also register the sk_lock with the lock validator.)
1416 static inline void sock_lock_init(struct sock *sk)
1418 if (sk->sk_kern_sock)
1419 sock_lock_init_class_and_name(
1421 af_family_kern_slock_key_strings[sk->sk_family],
1422 af_family_kern_slock_keys + sk->sk_family,
1423 af_family_kern_key_strings[sk->sk_family],
1424 af_family_kern_keys + sk->sk_family);
1426 sock_lock_init_class_and_name(
1428 af_family_slock_key_strings[sk->sk_family],
1429 af_family_slock_keys + sk->sk_family,
1430 af_family_key_strings[sk->sk_family],
1431 af_family_keys + sk->sk_family);
1435 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1436 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1437 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1439 static void sock_copy(struct sock *nsk, const struct sock *osk)
1441 #ifdef CONFIG_SECURITY_NETWORK
1442 void *sptr = nsk->sk_security;
1444 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1446 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1447 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1449 #ifdef CONFIG_SECURITY_NETWORK
1450 nsk->sk_security = sptr;
1451 security_sk_clone(osk, nsk);
1455 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1459 struct kmem_cache *slab;
1463 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1466 if (priority & __GFP_ZERO)
1467 sk_prot_clear_nulls(sk, prot->obj_size);
1469 sk = kmalloc(prot->obj_size, priority);
1472 if (security_sk_alloc(sk, family, priority))
1475 if (!try_module_get(prot->owner))
1477 sk_tx_queue_clear(sk);
1483 security_sk_free(sk);
1486 kmem_cache_free(slab, sk);
1492 static void sk_prot_free(struct proto *prot, struct sock *sk)
1494 struct kmem_cache *slab;
1495 struct module *owner;
1497 owner = prot->owner;
1500 cgroup_sk_free(&sk->sk_cgrp_data);
1501 mem_cgroup_sk_free(sk);
1502 security_sk_free(sk);
1504 kmem_cache_free(slab, sk);
1511 * sk_alloc - All socket objects are allocated here
1512 * @net: the applicable net namespace
1513 * @family: protocol family
1514 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1515 * @prot: struct proto associated with this new sock instance
1516 * @kern: is this to be a kernel socket?
1518 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1519 struct proto *prot, int kern)
1523 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1525 sk->sk_family = family;
1527 * See comment in struct sock definition to understand
1528 * why we need sk_prot_creator -acme
1530 sk->sk_prot = sk->sk_prot_creator = prot;
1531 sk->sk_kern_sock = kern;
1533 sk->sk_net_refcnt = kern ? 0 : 1;
1534 if (likely(sk->sk_net_refcnt))
1536 sock_net_set(sk, net);
1537 refcount_set(&sk->sk_wmem_alloc, 1);
1539 mem_cgroup_sk_alloc(sk);
1540 cgroup_sk_alloc(&sk->sk_cgrp_data);
1541 sock_update_classid(&sk->sk_cgrp_data);
1542 sock_update_netprioidx(&sk->sk_cgrp_data);
1547 EXPORT_SYMBOL(sk_alloc);
1549 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1550 * grace period. This is the case for UDP sockets and TCP listeners.
1552 static void __sk_destruct(struct rcu_head *head)
1554 struct sock *sk = container_of(head, struct sock, sk_rcu);
1555 struct sk_filter *filter;
1557 if (sk->sk_destruct)
1558 sk->sk_destruct(sk);
1560 filter = rcu_dereference_check(sk->sk_filter,
1561 refcount_read(&sk->sk_wmem_alloc) == 0);
1563 sk_filter_uncharge(sk, filter);
1564 RCU_INIT_POINTER(sk->sk_filter, NULL);
1566 if (rcu_access_pointer(sk->sk_reuseport_cb))
1567 reuseport_detach_sock(sk);
1569 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1571 if (atomic_read(&sk->sk_omem_alloc))
1572 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1573 __func__, atomic_read(&sk->sk_omem_alloc));
1575 if (sk->sk_frag.page) {
1576 put_page(sk->sk_frag.page);
1577 sk->sk_frag.page = NULL;
1580 if (sk->sk_peer_cred)
1581 put_cred(sk->sk_peer_cred);
1582 put_pid(sk->sk_peer_pid);
1583 if (likely(sk->sk_net_refcnt))
1584 put_net(sock_net(sk));
1585 sk_prot_free(sk->sk_prot_creator, sk);
1588 void sk_destruct(struct sock *sk)
1590 if (sock_flag(sk, SOCK_RCU_FREE))
1591 call_rcu(&sk->sk_rcu, __sk_destruct);
1593 __sk_destruct(&sk->sk_rcu);
1596 static void __sk_free(struct sock *sk)
1598 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1599 sock_diag_broadcast_destroy(sk);
1604 void sk_free(struct sock *sk)
1607 * We subtract one from sk_wmem_alloc and can know if
1608 * some packets are still in some tx queue.
1609 * If not null, sock_wfree() will call __sk_free(sk) later
1611 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1614 EXPORT_SYMBOL(sk_free);
1616 static void sk_init_common(struct sock *sk)
1618 skb_queue_head_init(&sk->sk_receive_queue);
1619 skb_queue_head_init(&sk->sk_write_queue);
1620 skb_queue_head_init(&sk->sk_error_queue);
1622 rwlock_init(&sk->sk_callback_lock);
1623 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1624 af_rlock_keys + sk->sk_family,
1625 af_family_rlock_key_strings[sk->sk_family]);
1626 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1627 af_wlock_keys + sk->sk_family,
1628 af_family_wlock_key_strings[sk->sk_family]);
1629 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1630 af_elock_keys + sk->sk_family,
1631 af_family_elock_key_strings[sk->sk_family]);
1632 lockdep_set_class_and_name(&sk->sk_callback_lock,
1633 af_callback_keys + sk->sk_family,
1634 af_family_clock_key_strings[sk->sk_family]);
1638 * sk_clone_lock - clone a socket, and lock its clone
1639 * @sk: the socket to clone
1640 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1642 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1644 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1647 bool is_charged = true;
1649 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1650 if (newsk != NULL) {
1651 struct sk_filter *filter;
1653 sock_copy(newsk, sk);
1655 newsk->sk_prot_creator = sk->sk_prot;
1658 if (likely(newsk->sk_net_refcnt))
1659 get_net(sock_net(newsk));
1660 sk_node_init(&newsk->sk_node);
1661 sock_lock_init(newsk);
1662 bh_lock_sock(newsk);
1663 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1664 newsk->sk_backlog.len = 0;
1666 atomic_set(&newsk->sk_rmem_alloc, 0);
1668 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1670 refcount_set(&newsk->sk_wmem_alloc, 1);
1671 atomic_set(&newsk->sk_omem_alloc, 0);
1672 sk_init_common(newsk);
1674 newsk->sk_dst_cache = NULL;
1675 newsk->sk_dst_pending_confirm = 0;
1676 newsk->sk_wmem_queued = 0;
1677 newsk->sk_forward_alloc = 0;
1679 /* sk->sk_memcg will be populated at accept() time */
1680 newsk->sk_memcg = NULL;
1682 atomic_set(&newsk->sk_drops, 0);
1683 newsk->sk_send_head = NULL;
1684 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1685 atomic_set(&newsk->sk_zckey, 0);
1687 sock_reset_flag(newsk, SOCK_DONE);
1688 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1691 filter = rcu_dereference(sk->sk_filter);
1693 /* though it's an empty new sock, the charging may fail
1694 * if sysctl_optmem_max was changed between creation of
1695 * original socket and cloning
1697 is_charged = sk_filter_charge(newsk, filter);
1698 RCU_INIT_POINTER(newsk->sk_filter, filter);
1701 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1702 /* We need to make sure that we don't uncharge the new
1703 * socket if we couldn't charge it in the first place
1704 * as otherwise we uncharge the parent's filter.
1707 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1708 sk_free_unlock_clone(newsk);
1712 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1715 newsk->sk_err_soft = 0;
1716 newsk->sk_priority = 0;
1717 newsk->sk_incoming_cpu = raw_smp_processor_id();
1718 atomic64_set(&newsk->sk_cookie, 0);
1721 * Before updating sk_refcnt, we must commit prior changes to memory
1722 * (Documentation/RCU/rculist_nulls.txt for details)
1725 refcount_set(&newsk->sk_refcnt, 2);
1728 * Increment the counter in the same struct proto as the master
1729 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1730 * is the same as sk->sk_prot->socks, as this field was copied
1733 * This _changes_ the previous behaviour, where
1734 * tcp_create_openreq_child always was incrementing the
1735 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1736 * to be taken into account in all callers. -acme
1738 sk_refcnt_debug_inc(newsk);
1739 sk_set_socket(newsk, NULL);
1740 newsk->sk_wq = NULL;
1742 if (newsk->sk_prot->sockets_allocated)
1743 sk_sockets_allocated_inc(newsk);
1745 if (sock_needs_netstamp(sk) &&
1746 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1747 net_enable_timestamp();
1752 EXPORT_SYMBOL_GPL(sk_clone_lock);
1754 void sk_free_unlock_clone(struct sock *sk)
1756 /* It is still raw copy of parent, so invalidate
1757 * destructor and make plain sk_free() */
1758 sk->sk_destruct = NULL;
1762 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1764 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1768 sk_dst_set(sk, dst);
1769 sk->sk_route_caps = dst->dev->features;
1770 if (sk->sk_route_caps & NETIF_F_GSO)
1771 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1772 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1773 if (sk_can_gso(sk)) {
1774 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1775 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1777 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1778 sk->sk_gso_max_size = dst->dev->gso_max_size;
1779 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1782 sk->sk_gso_max_segs = max_segs;
1784 EXPORT_SYMBOL_GPL(sk_setup_caps);
1787 * Simple resource managers for sockets.
1792 * Write buffer destructor automatically called from kfree_skb.
1794 void sock_wfree(struct sk_buff *skb)
1796 struct sock *sk = skb->sk;
1797 unsigned int len = skb->truesize;
1799 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1801 * Keep a reference on sk_wmem_alloc, this will be released
1802 * after sk_write_space() call
1804 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1805 sk->sk_write_space(sk);
1809 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1810 * could not do because of in-flight packets
1812 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1815 EXPORT_SYMBOL(sock_wfree);
1817 /* This variant of sock_wfree() is used by TCP,
1818 * since it sets SOCK_USE_WRITE_QUEUE.
1820 void __sock_wfree(struct sk_buff *skb)
1822 struct sock *sk = skb->sk;
1824 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1828 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1833 if (unlikely(!sk_fullsock(sk))) {
1834 skb->destructor = sock_edemux;
1839 skb->destructor = sock_wfree;
1840 skb_set_hash_from_sk(skb, sk);
1842 * We used to take a refcount on sk, but following operation
1843 * is enough to guarantee sk_free() wont free this sock until
1844 * all in-flight packets are completed
1846 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1848 EXPORT_SYMBOL(skb_set_owner_w);
1850 /* This helper is used by netem, as it can hold packets in its
1851 * delay queue. We want to allow the owner socket to send more
1852 * packets, as if they were already TX completed by a typical driver.
1853 * But we also want to keep skb->sk set because some packet schedulers
1854 * rely on it (sch_fq for example).
1856 void skb_orphan_partial(struct sk_buff *skb)
1858 if (skb_is_tcp_pure_ack(skb))
1861 if (skb->destructor == sock_wfree
1863 || skb->destructor == tcp_wfree
1866 struct sock *sk = skb->sk;
1868 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1869 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1870 skb->destructor = sock_efree;
1876 EXPORT_SYMBOL(skb_orphan_partial);
1879 * Read buffer destructor automatically called from kfree_skb.
1881 void sock_rfree(struct sk_buff *skb)
1883 struct sock *sk = skb->sk;
1884 unsigned int len = skb->truesize;
1886 atomic_sub(len, &sk->sk_rmem_alloc);
1887 sk_mem_uncharge(sk, len);
1889 EXPORT_SYMBOL(sock_rfree);
1892 * Buffer destructor for skbs that are not used directly in read or write
1893 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1895 void sock_efree(struct sk_buff *skb)
1899 EXPORT_SYMBOL(sock_efree);
1901 kuid_t sock_i_uid(struct sock *sk)
1905 read_lock_bh(&sk->sk_callback_lock);
1906 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1907 read_unlock_bh(&sk->sk_callback_lock);
1910 EXPORT_SYMBOL(sock_i_uid);
1912 unsigned long sock_i_ino(struct sock *sk)
1916 read_lock_bh(&sk->sk_callback_lock);
1917 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1918 read_unlock_bh(&sk->sk_callback_lock);
1921 EXPORT_SYMBOL(sock_i_ino);
1924 * Allocate a skb from the socket's send buffer.
1926 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1929 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1930 struct sk_buff *skb = alloc_skb(size, priority);
1932 skb_set_owner_w(skb, sk);
1938 EXPORT_SYMBOL(sock_wmalloc);
1940 static void sock_ofree(struct sk_buff *skb)
1942 struct sock *sk = skb->sk;
1944 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1947 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1950 struct sk_buff *skb;
1952 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1953 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1957 skb = alloc_skb(size, priority);
1961 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1963 skb->destructor = sock_ofree;
1968 * Allocate a memory block from the socket's option memory buffer.
1970 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1972 if ((unsigned int)size <= sysctl_optmem_max &&
1973 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1975 /* First do the add, to avoid the race if kmalloc
1978 atomic_add(size, &sk->sk_omem_alloc);
1979 mem = kmalloc(size, priority);
1982 atomic_sub(size, &sk->sk_omem_alloc);
1986 EXPORT_SYMBOL(sock_kmalloc);
1988 /* Free an option memory block. Note, we actually want the inline
1989 * here as this allows gcc to detect the nullify and fold away the
1990 * condition entirely.
1992 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1995 if (WARN_ON_ONCE(!mem))
2001 atomic_sub(size, &sk->sk_omem_alloc);
2004 void sock_kfree_s(struct sock *sk, void *mem, int size)
2006 __sock_kfree_s(sk, mem, size, false);
2008 EXPORT_SYMBOL(sock_kfree_s);
2010 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2012 __sock_kfree_s(sk, mem, size, true);
2014 EXPORT_SYMBOL(sock_kzfree_s);
2016 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2017 I think, these locks should be removed for datagram sockets.
2019 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2023 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2027 if (signal_pending(current))
2029 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2030 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2031 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2033 if (sk->sk_shutdown & SEND_SHUTDOWN)
2037 timeo = schedule_timeout(timeo);
2039 finish_wait(sk_sleep(sk), &wait);
2045 * Generic send/receive buffer handlers
2048 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2049 unsigned long data_len, int noblock,
2050 int *errcode, int max_page_order)
2052 struct sk_buff *skb;
2056 timeo = sock_sndtimeo(sk, noblock);
2058 err = sock_error(sk);
2063 if (sk->sk_shutdown & SEND_SHUTDOWN)
2066 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2069 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2070 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2074 if (signal_pending(current))
2076 timeo = sock_wait_for_wmem(sk, timeo);
2078 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2079 errcode, sk->sk_allocation);
2081 skb_set_owner_w(skb, sk);
2085 err = sock_intr_errno(timeo);
2090 EXPORT_SYMBOL(sock_alloc_send_pskb);
2092 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2093 int noblock, int *errcode)
2095 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2097 EXPORT_SYMBOL(sock_alloc_send_skb);
2099 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2100 struct sockcm_cookie *sockc)
2104 switch (cmsg->cmsg_type) {
2106 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2108 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2110 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2112 case SO_TIMESTAMPING:
2113 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2116 tsflags = *(u32 *)CMSG_DATA(cmsg);
2117 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2120 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2121 sockc->tsflags |= tsflags;
2123 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2125 case SCM_CREDENTIALS:
2132 EXPORT_SYMBOL(__sock_cmsg_send);
2134 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2135 struct sockcm_cookie *sockc)
2137 struct cmsghdr *cmsg;
2140 for_each_cmsghdr(cmsg, msg) {
2141 if (!CMSG_OK(msg, cmsg))
2143 if (cmsg->cmsg_level != SOL_SOCKET)
2145 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2151 EXPORT_SYMBOL(sock_cmsg_send);
2153 static void sk_enter_memory_pressure(struct sock *sk)
2155 if (!sk->sk_prot->enter_memory_pressure)
2158 sk->sk_prot->enter_memory_pressure(sk);
2161 static void sk_leave_memory_pressure(struct sock *sk)
2163 if (sk->sk_prot->leave_memory_pressure) {
2164 sk->sk_prot->leave_memory_pressure(sk);
2166 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2168 if (memory_pressure && *memory_pressure)
2169 *memory_pressure = 0;
2173 /* On 32bit arches, an skb frag is limited to 2^15 */
2174 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2177 * skb_page_frag_refill - check that a page_frag contains enough room
2178 * @sz: minimum size of the fragment we want to get
2179 * @pfrag: pointer to page_frag
2180 * @gfp: priority for memory allocation
2182 * Note: While this allocator tries to use high order pages, there is
2183 * no guarantee that allocations succeed. Therefore, @sz MUST be
2184 * less or equal than PAGE_SIZE.
2186 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2189 if (page_ref_count(pfrag->page) == 1) {
2193 if (pfrag->offset + sz <= pfrag->size)
2195 put_page(pfrag->page);
2199 if (SKB_FRAG_PAGE_ORDER) {
2200 /* Avoid direct reclaim but allow kswapd to wake */
2201 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2202 __GFP_COMP | __GFP_NOWARN |
2204 SKB_FRAG_PAGE_ORDER);
2205 if (likely(pfrag->page)) {
2206 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2210 pfrag->page = alloc_page(gfp);
2211 if (likely(pfrag->page)) {
2212 pfrag->size = PAGE_SIZE;
2217 EXPORT_SYMBOL(skb_page_frag_refill);
2219 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2221 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2224 sk_enter_memory_pressure(sk);
2225 sk_stream_moderate_sndbuf(sk);
2228 EXPORT_SYMBOL(sk_page_frag_refill);
2230 static void __lock_sock(struct sock *sk)
2231 __releases(&sk->sk_lock.slock)
2232 __acquires(&sk->sk_lock.slock)
2237 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2238 TASK_UNINTERRUPTIBLE);
2239 spin_unlock_bh(&sk->sk_lock.slock);
2241 spin_lock_bh(&sk->sk_lock.slock);
2242 if (!sock_owned_by_user(sk))
2245 finish_wait(&sk->sk_lock.wq, &wait);
2248 static void __release_sock(struct sock *sk)
2249 __releases(&sk->sk_lock.slock)
2250 __acquires(&sk->sk_lock.slock)
2252 struct sk_buff *skb, *next;
2254 while ((skb = sk->sk_backlog.head) != NULL) {
2255 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2257 spin_unlock_bh(&sk->sk_lock.slock);
2262 WARN_ON_ONCE(skb_dst_is_noref(skb));
2264 sk_backlog_rcv(sk, skb);
2269 } while (skb != NULL);
2271 spin_lock_bh(&sk->sk_lock.slock);
2275 * Doing the zeroing here guarantee we can not loop forever
2276 * while a wild producer attempts to flood us.
2278 sk->sk_backlog.len = 0;
2281 void __sk_flush_backlog(struct sock *sk)
2283 spin_lock_bh(&sk->sk_lock.slock);
2285 spin_unlock_bh(&sk->sk_lock.slock);
2289 * sk_wait_data - wait for data to arrive at sk_receive_queue
2290 * @sk: sock to wait on
2291 * @timeo: for how long
2292 * @skb: last skb seen on sk_receive_queue
2294 * Now socket state including sk->sk_err is changed only under lock,
2295 * hence we may omit checks after joining wait queue.
2296 * We check receive queue before schedule() only as optimization;
2297 * it is very likely that release_sock() added new data.
2299 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2301 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2304 add_wait_queue(sk_sleep(sk), &wait);
2305 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2306 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2307 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2308 remove_wait_queue(sk_sleep(sk), &wait);
2311 EXPORT_SYMBOL(sk_wait_data);
2314 * __sk_mem_raise_allocated - increase memory_allocated
2316 * @size: memory size to allocate
2317 * @amt: pages to allocate
2318 * @kind: allocation type
2320 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2322 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2324 struct proto *prot = sk->sk_prot;
2325 long allocated = sk_memory_allocated_add(sk, amt);
2327 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2328 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2329 goto suppress_allocation;
2332 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2333 sk_leave_memory_pressure(sk);
2337 /* Under pressure. */
2338 if (allocated > sk_prot_mem_limits(sk, 1))
2339 sk_enter_memory_pressure(sk);
2341 /* Over hard limit. */
2342 if (allocated > sk_prot_mem_limits(sk, 2))
2343 goto suppress_allocation;
2345 /* guarantee minimum buffer size under pressure */
2346 if (kind == SK_MEM_RECV) {
2347 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2350 } else { /* SK_MEM_SEND */
2351 int wmem0 = sk_get_wmem0(sk, prot);
2353 if (sk->sk_type == SOCK_STREAM) {
2354 if (sk->sk_wmem_queued < wmem0)
2356 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2361 if (sk_has_memory_pressure(sk)) {
2364 if (!sk_under_memory_pressure(sk))
2366 alloc = sk_sockets_allocated_read_positive(sk);
2367 if (sk_prot_mem_limits(sk, 2) > alloc *
2368 sk_mem_pages(sk->sk_wmem_queued +
2369 atomic_read(&sk->sk_rmem_alloc) +
2370 sk->sk_forward_alloc))
2374 suppress_allocation:
2376 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2377 sk_stream_moderate_sndbuf(sk);
2379 /* Fail only if socket is _under_ its sndbuf.
2380 * In this case we cannot block, so that we have to fail.
2382 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2386 trace_sock_exceed_buf_limit(sk, prot, allocated);
2388 sk_memory_allocated_sub(sk, amt);
2390 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2391 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2395 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2398 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2400 * @size: memory size to allocate
2401 * @kind: allocation type
2403 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2404 * rmem allocation. This function assumes that protocols which have
2405 * memory_pressure use sk_wmem_queued as write buffer accounting.
2407 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2409 int ret, amt = sk_mem_pages(size);
2411 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2412 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2414 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2417 EXPORT_SYMBOL(__sk_mem_schedule);
2420 * __sk_mem_reduce_allocated - reclaim memory_allocated
2422 * @amount: number of quanta
2424 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2426 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2428 sk_memory_allocated_sub(sk, amount);
2430 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2431 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2433 if (sk_under_memory_pressure(sk) &&
2434 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2435 sk_leave_memory_pressure(sk);
2437 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2440 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2442 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2444 void __sk_mem_reclaim(struct sock *sk, int amount)
2446 amount >>= SK_MEM_QUANTUM_SHIFT;
2447 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2448 __sk_mem_reduce_allocated(sk, amount);
2450 EXPORT_SYMBOL(__sk_mem_reclaim);
2452 int sk_set_peek_off(struct sock *sk, int val)
2454 sk->sk_peek_off = val;
2457 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2460 * Set of default routines for initialising struct proto_ops when
2461 * the protocol does not support a particular function. In certain
2462 * cases where it makes no sense for a protocol to have a "do nothing"
2463 * function, some default processing is provided.
2466 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2470 EXPORT_SYMBOL(sock_no_bind);
2472 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2477 EXPORT_SYMBOL(sock_no_connect);
2479 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2483 EXPORT_SYMBOL(sock_no_socketpair);
2485 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2490 EXPORT_SYMBOL(sock_no_accept);
2492 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2497 EXPORT_SYMBOL(sock_no_getname);
2499 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2503 EXPORT_SYMBOL(sock_no_poll);
2505 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2509 EXPORT_SYMBOL(sock_no_ioctl);
2511 int sock_no_listen(struct socket *sock, int backlog)
2515 EXPORT_SYMBOL(sock_no_listen);
2517 int sock_no_shutdown(struct socket *sock, int how)
2521 EXPORT_SYMBOL(sock_no_shutdown);
2523 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2524 char __user *optval, unsigned int optlen)
2528 EXPORT_SYMBOL(sock_no_setsockopt);
2530 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2531 char __user *optval, int __user *optlen)
2535 EXPORT_SYMBOL(sock_no_getsockopt);
2537 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2541 EXPORT_SYMBOL(sock_no_sendmsg);
2543 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2547 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2549 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2554 EXPORT_SYMBOL(sock_no_recvmsg);
2556 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2558 /* Mirror missing mmap method error code */
2561 EXPORT_SYMBOL(sock_no_mmap);
2563 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2566 struct msghdr msg = {.msg_flags = flags};
2568 char *kaddr = kmap(page);
2569 iov.iov_base = kaddr + offset;
2571 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2575 EXPORT_SYMBOL(sock_no_sendpage);
2577 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2578 int offset, size_t size, int flags)
2581 struct msghdr msg = {.msg_flags = flags};
2583 char *kaddr = kmap(page);
2585 iov.iov_base = kaddr + offset;
2587 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2591 EXPORT_SYMBOL(sock_no_sendpage_locked);
2594 * Default Socket Callbacks
2597 static void sock_def_wakeup(struct sock *sk)
2599 struct socket_wq *wq;
2602 wq = rcu_dereference(sk->sk_wq);
2603 if (skwq_has_sleeper(wq))
2604 wake_up_interruptible_all(&wq->wait);
2608 static void sock_def_error_report(struct sock *sk)
2610 struct socket_wq *wq;
2613 wq = rcu_dereference(sk->sk_wq);
2614 if (skwq_has_sleeper(wq))
2615 wake_up_interruptible_poll(&wq->wait, POLLERR);
2616 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2620 static void sock_def_readable(struct sock *sk)
2622 struct socket_wq *wq;
2625 wq = rcu_dereference(sk->sk_wq);
2626 if (skwq_has_sleeper(wq))
2627 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2628 POLLRDNORM | POLLRDBAND);
2629 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2633 static void sock_def_write_space(struct sock *sk)
2635 struct socket_wq *wq;
2639 /* Do not wake up a writer until he can make "significant"
2642 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2643 wq = rcu_dereference(sk->sk_wq);
2644 if (skwq_has_sleeper(wq))
2645 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2646 POLLWRNORM | POLLWRBAND);
2648 /* Should agree with poll, otherwise some programs break */
2649 if (sock_writeable(sk))
2650 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2656 static void sock_def_destruct(struct sock *sk)
2660 void sk_send_sigurg(struct sock *sk)
2662 if (sk->sk_socket && sk->sk_socket->file)
2663 if (send_sigurg(&sk->sk_socket->file->f_owner))
2664 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2666 EXPORT_SYMBOL(sk_send_sigurg);
2668 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2669 unsigned long expires)
2671 if (!mod_timer(timer, expires))
2674 EXPORT_SYMBOL(sk_reset_timer);
2676 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2678 if (del_timer(timer))
2681 EXPORT_SYMBOL(sk_stop_timer);
2683 void sock_init_data(struct socket *sock, struct sock *sk)
2686 sk->sk_send_head = NULL;
2688 timer_setup(&sk->sk_timer, NULL, 0);
2690 sk->sk_allocation = GFP_KERNEL;
2691 sk->sk_rcvbuf = sysctl_rmem_default;
2692 sk->sk_sndbuf = sysctl_wmem_default;
2693 sk->sk_state = TCP_CLOSE;
2694 sk_set_socket(sk, sock);
2696 sock_set_flag(sk, SOCK_ZAPPED);
2699 sk->sk_type = sock->type;
2700 sk->sk_wq = sock->wq;
2702 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2705 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2708 rwlock_init(&sk->sk_callback_lock);
2709 if (sk->sk_kern_sock)
2710 lockdep_set_class_and_name(
2711 &sk->sk_callback_lock,
2712 af_kern_callback_keys + sk->sk_family,
2713 af_family_kern_clock_key_strings[sk->sk_family]);
2715 lockdep_set_class_and_name(
2716 &sk->sk_callback_lock,
2717 af_callback_keys + sk->sk_family,
2718 af_family_clock_key_strings[sk->sk_family]);
2720 sk->sk_state_change = sock_def_wakeup;
2721 sk->sk_data_ready = sock_def_readable;
2722 sk->sk_write_space = sock_def_write_space;
2723 sk->sk_error_report = sock_def_error_report;
2724 sk->sk_destruct = sock_def_destruct;
2726 sk->sk_frag.page = NULL;
2727 sk->sk_frag.offset = 0;
2728 sk->sk_peek_off = -1;
2730 sk->sk_peer_pid = NULL;
2731 sk->sk_peer_cred = NULL;
2732 sk->sk_write_pending = 0;
2733 sk->sk_rcvlowat = 1;
2734 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2735 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2737 sk->sk_stamp = SK_DEFAULT_STAMP;
2738 atomic_set(&sk->sk_zckey, 0);
2740 #ifdef CONFIG_NET_RX_BUSY_POLL
2742 sk->sk_ll_usec = sysctl_net_busy_read;
2745 sk->sk_max_pacing_rate = ~0U;
2746 sk->sk_pacing_rate = ~0U;
2747 sk->sk_pacing_shift = 10;
2748 sk->sk_incoming_cpu = -1;
2750 * Before updating sk_refcnt, we must commit prior changes to memory
2751 * (Documentation/RCU/rculist_nulls.txt for details)
2754 refcount_set(&sk->sk_refcnt, 1);
2755 atomic_set(&sk->sk_drops, 0);
2757 EXPORT_SYMBOL(sock_init_data);
2759 void lock_sock_nested(struct sock *sk, int subclass)
2762 spin_lock_bh(&sk->sk_lock.slock);
2763 if (sk->sk_lock.owned)
2765 sk->sk_lock.owned = 1;
2766 spin_unlock(&sk->sk_lock.slock);
2768 * The sk_lock has mutex_lock() semantics here:
2770 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2773 EXPORT_SYMBOL(lock_sock_nested);
2775 void release_sock(struct sock *sk)
2777 spin_lock_bh(&sk->sk_lock.slock);
2778 if (sk->sk_backlog.tail)
2781 /* Warning : release_cb() might need to release sk ownership,
2782 * ie call sock_release_ownership(sk) before us.
2784 if (sk->sk_prot->release_cb)
2785 sk->sk_prot->release_cb(sk);
2787 sock_release_ownership(sk);
2788 if (waitqueue_active(&sk->sk_lock.wq))
2789 wake_up(&sk->sk_lock.wq);
2790 spin_unlock_bh(&sk->sk_lock.slock);
2792 EXPORT_SYMBOL(release_sock);
2795 * lock_sock_fast - fast version of lock_sock
2798 * This version should be used for very small section, where process wont block
2799 * return false if fast path is taken:
2801 * sk_lock.slock locked, owned = 0, BH disabled
2803 * return true if slow path is taken:
2805 * sk_lock.slock unlocked, owned = 1, BH enabled
2807 bool lock_sock_fast(struct sock *sk)
2810 spin_lock_bh(&sk->sk_lock.slock);
2812 if (!sk->sk_lock.owned)
2814 * Note : We must disable BH
2819 sk->sk_lock.owned = 1;
2820 spin_unlock(&sk->sk_lock.slock);
2822 * The sk_lock has mutex_lock() semantics here:
2824 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2828 EXPORT_SYMBOL(lock_sock_fast);
2830 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2833 if (!sock_flag(sk, SOCK_TIMESTAMP))
2834 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2835 tv = ktime_to_timeval(sk->sk_stamp);
2836 if (tv.tv_sec == -1)
2838 if (tv.tv_sec == 0) {
2839 sk->sk_stamp = ktime_get_real();
2840 tv = ktime_to_timeval(sk->sk_stamp);
2842 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2844 EXPORT_SYMBOL(sock_get_timestamp);
2846 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2849 if (!sock_flag(sk, SOCK_TIMESTAMP))
2850 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2851 ts = ktime_to_timespec(sk->sk_stamp);
2852 if (ts.tv_sec == -1)
2854 if (ts.tv_sec == 0) {
2855 sk->sk_stamp = ktime_get_real();
2856 ts = ktime_to_timespec(sk->sk_stamp);
2858 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2860 EXPORT_SYMBOL(sock_get_timestampns);
2862 void sock_enable_timestamp(struct sock *sk, int flag)
2864 if (!sock_flag(sk, flag)) {
2865 unsigned long previous_flags = sk->sk_flags;
2867 sock_set_flag(sk, flag);
2869 * we just set one of the two flags which require net
2870 * time stamping, but time stamping might have been on
2871 * already because of the other one
2873 if (sock_needs_netstamp(sk) &&
2874 !(previous_flags & SK_FLAGS_TIMESTAMP))
2875 net_enable_timestamp();
2879 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2880 int level, int type)
2882 struct sock_exterr_skb *serr;
2883 struct sk_buff *skb;
2887 skb = sock_dequeue_err_skb(sk);
2893 msg->msg_flags |= MSG_TRUNC;
2896 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2900 sock_recv_timestamp(msg, sk, skb);
2902 serr = SKB_EXT_ERR(skb);
2903 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2905 msg->msg_flags |= MSG_ERRQUEUE;
2913 EXPORT_SYMBOL(sock_recv_errqueue);
2916 * Get a socket option on an socket.
2918 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2919 * asynchronous errors should be reported by getsockopt. We assume
2920 * this means if you specify SO_ERROR (otherwise whats the point of it).
2922 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2923 char __user *optval, int __user *optlen)
2925 struct sock *sk = sock->sk;
2927 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2929 EXPORT_SYMBOL(sock_common_getsockopt);
2931 #ifdef CONFIG_COMPAT
2932 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2933 char __user *optval, int __user *optlen)
2935 struct sock *sk = sock->sk;
2937 if (sk->sk_prot->compat_getsockopt != NULL)
2938 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2940 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2942 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2945 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2948 struct sock *sk = sock->sk;
2952 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2953 flags & ~MSG_DONTWAIT, &addr_len);
2955 msg->msg_namelen = addr_len;
2958 EXPORT_SYMBOL(sock_common_recvmsg);
2961 * Set socket options on an inet socket.
2963 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2964 char __user *optval, unsigned int optlen)
2966 struct sock *sk = sock->sk;
2968 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2970 EXPORT_SYMBOL(sock_common_setsockopt);
2972 #ifdef CONFIG_COMPAT
2973 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2974 char __user *optval, unsigned int optlen)
2976 struct sock *sk = sock->sk;
2978 if (sk->sk_prot->compat_setsockopt != NULL)
2979 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2981 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2983 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2986 void sk_common_release(struct sock *sk)
2988 if (sk->sk_prot->destroy)
2989 sk->sk_prot->destroy(sk);
2992 * Observation: when sock_common_release is called, processes have
2993 * no access to socket. But net still has.
2994 * Step one, detach it from networking:
2996 * A. Remove from hash tables.
2999 sk->sk_prot->unhash(sk);
3002 * In this point socket cannot receive new packets, but it is possible
3003 * that some packets are in flight because some CPU runs receiver and
3004 * did hash table lookup before we unhashed socket. They will achieve
3005 * receive queue and will be purged by socket destructor.
3007 * Also we still have packets pending on receive queue and probably,
3008 * our own packets waiting in device queues. sock_destroy will drain
3009 * receive queue, but transmitted packets will delay socket destruction
3010 * until the last reference will be released.
3015 xfrm_sk_free_policy(sk);
3017 sk_refcnt_debug_release(sk);
3021 EXPORT_SYMBOL(sk_common_release);
3023 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3025 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3027 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3028 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3029 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3030 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3031 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3032 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3033 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3034 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3035 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3038 #ifdef CONFIG_PROC_FS
3039 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3041 int val[PROTO_INUSE_NR];
3044 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3046 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3048 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
3050 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3052 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3054 int cpu, idx = prot->inuse_idx;
3057 for_each_possible_cpu(cpu)
3058 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
3060 return res >= 0 ? res : 0;
3062 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3064 static int __net_init sock_inuse_init_net(struct net *net)
3066 net->core.inuse = alloc_percpu(struct prot_inuse);
3067 return net->core.inuse ? 0 : -ENOMEM;
3070 static void __net_exit sock_inuse_exit_net(struct net *net)
3072 free_percpu(net->core.inuse);
3075 static struct pernet_operations net_inuse_ops = {
3076 .init = sock_inuse_init_net,
3077 .exit = sock_inuse_exit_net,
3080 static __init int net_inuse_init(void)
3082 if (register_pernet_subsys(&net_inuse_ops))
3083 panic("Cannot initialize net inuse counters");
3088 core_initcall(net_inuse_init);
3090 static void assign_proto_idx(struct proto *prot)
3092 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3094 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3095 pr_err("PROTO_INUSE_NR exhausted\n");
3099 set_bit(prot->inuse_idx, proto_inuse_idx);
3102 static void release_proto_idx(struct proto *prot)
3104 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3105 clear_bit(prot->inuse_idx, proto_inuse_idx);
3108 static inline void assign_proto_idx(struct proto *prot)
3112 static inline void release_proto_idx(struct proto *prot)
3117 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3121 kfree(rsk_prot->slab_name);
3122 rsk_prot->slab_name = NULL;
3123 kmem_cache_destroy(rsk_prot->slab);
3124 rsk_prot->slab = NULL;
3127 static int req_prot_init(const struct proto *prot)
3129 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3134 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3136 if (!rsk_prot->slab_name)
3139 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3140 rsk_prot->obj_size, 0,
3141 prot->slab_flags, NULL);
3143 if (!rsk_prot->slab) {
3144 pr_crit("%s: Can't create request sock SLAB cache!\n",
3151 int proto_register(struct proto *prot, int alloc_slab)
3154 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
3155 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3158 if (prot->slab == NULL) {
3159 pr_crit("%s: Can't create sock SLAB cache!\n",
3164 if (req_prot_init(prot))
3165 goto out_free_request_sock_slab;
3167 if (prot->twsk_prot != NULL) {
3168 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3170 if (prot->twsk_prot->twsk_slab_name == NULL)
3171 goto out_free_request_sock_slab;
3173 prot->twsk_prot->twsk_slab =
3174 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3175 prot->twsk_prot->twsk_obj_size,
3179 if (prot->twsk_prot->twsk_slab == NULL)
3180 goto out_free_timewait_sock_slab_name;
3184 mutex_lock(&proto_list_mutex);
3185 list_add(&prot->node, &proto_list);
3186 assign_proto_idx(prot);
3187 mutex_unlock(&proto_list_mutex);
3190 out_free_timewait_sock_slab_name:
3191 kfree(prot->twsk_prot->twsk_slab_name);
3192 out_free_request_sock_slab:
3193 req_prot_cleanup(prot->rsk_prot);
3195 kmem_cache_destroy(prot->slab);
3200 EXPORT_SYMBOL(proto_register);
3202 void proto_unregister(struct proto *prot)
3204 mutex_lock(&proto_list_mutex);
3205 release_proto_idx(prot);
3206 list_del(&prot->node);
3207 mutex_unlock(&proto_list_mutex);
3209 kmem_cache_destroy(prot->slab);
3212 req_prot_cleanup(prot->rsk_prot);
3214 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3215 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3216 kfree(prot->twsk_prot->twsk_slab_name);
3217 prot->twsk_prot->twsk_slab = NULL;
3220 EXPORT_SYMBOL(proto_unregister);
3222 #ifdef CONFIG_PROC_FS
3223 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3224 __acquires(proto_list_mutex)
3226 mutex_lock(&proto_list_mutex);
3227 return seq_list_start_head(&proto_list, *pos);
3230 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3232 return seq_list_next(v, &proto_list, pos);
3235 static void proto_seq_stop(struct seq_file *seq, void *v)
3236 __releases(proto_list_mutex)
3238 mutex_unlock(&proto_list_mutex);
3241 static char proto_method_implemented(const void *method)
3243 return method == NULL ? 'n' : 'y';
3245 static long sock_prot_memory_allocated(struct proto *proto)
3247 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3250 static char *sock_prot_memory_pressure(struct proto *proto)
3252 return proto->memory_pressure != NULL ?
3253 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3256 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3259 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3260 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3263 sock_prot_inuse_get(seq_file_net(seq), proto),
3264 sock_prot_memory_allocated(proto),
3265 sock_prot_memory_pressure(proto),
3267 proto->slab == NULL ? "no" : "yes",
3268 module_name(proto->owner),
3269 proto_method_implemented(proto->close),
3270 proto_method_implemented(proto->connect),
3271 proto_method_implemented(proto->disconnect),
3272 proto_method_implemented(proto->accept),
3273 proto_method_implemented(proto->ioctl),
3274 proto_method_implemented(proto->init),
3275 proto_method_implemented(proto->destroy),
3276 proto_method_implemented(proto->shutdown),
3277 proto_method_implemented(proto->setsockopt),
3278 proto_method_implemented(proto->getsockopt),
3279 proto_method_implemented(proto->sendmsg),
3280 proto_method_implemented(proto->recvmsg),
3281 proto_method_implemented(proto->sendpage),
3282 proto_method_implemented(proto->bind),
3283 proto_method_implemented(proto->backlog_rcv),
3284 proto_method_implemented(proto->hash),
3285 proto_method_implemented(proto->unhash),
3286 proto_method_implemented(proto->get_port),
3287 proto_method_implemented(proto->enter_memory_pressure));
3290 static int proto_seq_show(struct seq_file *seq, void *v)
3292 if (v == &proto_list)
3293 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3302 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3304 proto_seq_printf(seq, list_entry(v, struct proto, node));
3308 static const struct seq_operations proto_seq_ops = {
3309 .start = proto_seq_start,
3310 .next = proto_seq_next,
3311 .stop = proto_seq_stop,
3312 .show = proto_seq_show,
3315 static int proto_seq_open(struct inode *inode, struct file *file)
3317 return seq_open_net(inode, file, &proto_seq_ops,
3318 sizeof(struct seq_net_private));
3321 static const struct file_operations proto_seq_fops = {
3322 .owner = THIS_MODULE,
3323 .open = proto_seq_open,
3325 .llseek = seq_lseek,
3326 .release = seq_release_net,
3329 static __net_init int proto_init_net(struct net *net)
3331 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3337 static __net_exit void proto_exit_net(struct net *net)
3339 remove_proc_entry("protocols", net->proc_net);
3343 static __net_initdata struct pernet_operations proto_net_ops = {
3344 .init = proto_init_net,
3345 .exit = proto_exit_net,
3348 static int __init proto_init(void)
3350 return register_pernet_subsys(&proto_net_ops);
3353 subsys_initcall(proto_init);
3355 #endif /* PROC_FS */
3357 #ifdef CONFIG_NET_RX_BUSY_POLL
3358 bool sk_busy_loop_end(void *p, unsigned long start_time)
3360 struct sock *sk = p;
3362 return !skb_queue_empty(&sk->sk_receive_queue) ||
3363 sk_busy_loop_timeout(sk, start_time);
3365 EXPORT_SYMBOL(sk_busy_loop_end);
3366 #endif /* CONFIG_NET_RX_BUSY_POLL */