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/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 kmemcheck_annotate_bitfield(sk, flags);
1474 if (security_sk_alloc(sk, family, priority))
1477 if (!try_module_get(prot->owner))
1479 sk_tx_queue_clear(sk);
1485 security_sk_free(sk);
1488 kmem_cache_free(slab, sk);
1494 static void sk_prot_free(struct proto *prot, struct sock *sk)
1496 struct kmem_cache *slab;
1497 struct module *owner;
1499 owner = prot->owner;
1502 cgroup_sk_free(&sk->sk_cgrp_data);
1503 mem_cgroup_sk_free(sk);
1504 security_sk_free(sk);
1506 kmem_cache_free(slab, sk);
1513 * sk_alloc - All socket objects are allocated here
1514 * @net: the applicable net namespace
1515 * @family: protocol family
1516 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1517 * @prot: struct proto associated with this new sock instance
1518 * @kern: is this to be a kernel socket?
1520 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1521 struct proto *prot, int kern)
1525 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1527 sk->sk_family = family;
1529 * See comment in struct sock definition to understand
1530 * why we need sk_prot_creator -acme
1532 sk->sk_prot = sk->sk_prot_creator = prot;
1533 sk->sk_kern_sock = kern;
1535 sk->sk_net_refcnt = kern ? 0 : 1;
1536 if (likely(sk->sk_net_refcnt))
1538 sock_net_set(sk, net);
1539 refcount_set(&sk->sk_wmem_alloc, 1);
1541 mem_cgroup_sk_alloc(sk);
1542 cgroup_sk_alloc(&sk->sk_cgrp_data);
1543 sock_update_classid(&sk->sk_cgrp_data);
1544 sock_update_netprioidx(&sk->sk_cgrp_data);
1549 EXPORT_SYMBOL(sk_alloc);
1551 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1552 * grace period. This is the case for UDP sockets and TCP listeners.
1554 static void __sk_destruct(struct rcu_head *head)
1556 struct sock *sk = container_of(head, struct sock, sk_rcu);
1557 struct sk_filter *filter;
1559 if (sk->sk_destruct)
1560 sk->sk_destruct(sk);
1562 filter = rcu_dereference_check(sk->sk_filter,
1563 refcount_read(&sk->sk_wmem_alloc) == 0);
1565 sk_filter_uncharge(sk, filter);
1566 RCU_INIT_POINTER(sk->sk_filter, NULL);
1568 if (rcu_access_pointer(sk->sk_reuseport_cb))
1569 reuseport_detach_sock(sk);
1571 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1573 if (atomic_read(&sk->sk_omem_alloc))
1574 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1575 __func__, atomic_read(&sk->sk_omem_alloc));
1577 if (sk->sk_frag.page) {
1578 put_page(sk->sk_frag.page);
1579 sk->sk_frag.page = NULL;
1582 if (sk->sk_peer_cred)
1583 put_cred(sk->sk_peer_cred);
1584 put_pid(sk->sk_peer_pid);
1585 if (likely(sk->sk_net_refcnt))
1586 put_net(sock_net(sk));
1587 sk_prot_free(sk->sk_prot_creator, sk);
1590 void sk_destruct(struct sock *sk)
1592 if (sock_flag(sk, SOCK_RCU_FREE))
1593 call_rcu(&sk->sk_rcu, __sk_destruct);
1595 __sk_destruct(&sk->sk_rcu);
1598 static void __sk_free(struct sock *sk)
1600 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1601 sock_diag_broadcast_destroy(sk);
1606 void sk_free(struct sock *sk)
1609 * We subtract one from sk_wmem_alloc and can know if
1610 * some packets are still in some tx queue.
1611 * If not null, sock_wfree() will call __sk_free(sk) later
1613 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1616 EXPORT_SYMBOL(sk_free);
1618 static void sk_init_common(struct sock *sk)
1620 skb_queue_head_init(&sk->sk_receive_queue);
1621 skb_queue_head_init(&sk->sk_write_queue);
1622 skb_queue_head_init(&sk->sk_error_queue);
1624 rwlock_init(&sk->sk_callback_lock);
1625 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1626 af_rlock_keys + sk->sk_family,
1627 af_family_rlock_key_strings[sk->sk_family]);
1628 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1629 af_wlock_keys + sk->sk_family,
1630 af_family_wlock_key_strings[sk->sk_family]);
1631 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1632 af_elock_keys + sk->sk_family,
1633 af_family_elock_key_strings[sk->sk_family]);
1634 lockdep_set_class_and_name(&sk->sk_callback_lock,
1635 af_callback_keys + sk->sk_family,
1636 af_family_clock_key_strings[sk->sk_family]);
1640 * sk_clone_lock - clone a socket, and lock its clone
1641 * @sk: the socket to clone
1642 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1644 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1646 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1649 bool is_charged = true;
1651 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1652 if (newsk != NULL) {
1653 struct sk_filter *filter;
1655 sock_copy(newsk, sk);
1657 newsk->sk_prot_creator = sk->sk_prot;
1660 if (likely(newsk->sk_net_refcnt))
1661 get_net(sock_net(newsk));
1662 sk_node_init(&newsk->sk_node);
1663 sock_lock_init(newsk);
1664 bh_lock_sock(newsk);
1665 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1666 newsk->sk_backlog.len = 0;
1668 atomic_set(&newsk->sk_rmem_alloc, 0);
1670 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1672 refcount_set(&newsk->sk_wmem_alloc, 1);
1673 atomic_set(&newsk->sk_omem_alloc, 0);
1674 sk_init_common(newsk);
1676 newsk->sk_dst_cache = NULL;
1677 newsk->sk_dst_pending_confirm = 0;
1678 newsk->sk_wmem_queued = 0;
1679 newsk->sk_forward_alloc = 0;
1681 /* sk->sk_memcg will be populated at accept() time */
1682 newsk->sk_memcg = NULL;
1684 atomic_set(&newsk->sk_drops, 0);
1685 newsk->sk_send_head = NULL;
1686 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1687 atomic_set(&newsk->sk_zckey, 0);
1689 sock_reset_flag(newsk, SOCK_DONE);
1690 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1693 filter = rcu_dereference(sk->sk_filter);
1695 /* though it's an empty new sock, the charging may fail
1696 * if sysctl_optmem_max was changed between creation of
1697 * original socket and cloning
1699 is_charged = sk_filter_charge(newsk, filter);
1700 RCU_INIT_POINTER(newsk->sk_filter, filter);
1703 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1704 /* We need to make sure that we don't uncharge the new
1705 * socket if we couldn't charge it in the first place
1706 * as otherwise we uncharge the parent's filter.
1709 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1710 sk_free_unlock_clone(newsk);
1714 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1717 newsk->sk_err_soft = 0;
1718 newsk->sk_priority = 0;
1719 newsk->sk_incoming_cpu = raw_smp_processor_id();
1720 atomic64_set(&newsk->sk_cookie, 0);
1723 * Before updating sk_refcnt, we must commit prior changes to memory
1724 * (Documentation/RCU/rculist_nulls.txt for details)
1727 refcount_set(&newsk->sk_refcnt, 2);
1730 * Increment the counter in the same struct proto as the master
1731 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1732 * is the same as sk->sk_prot->socks, as this field was copied
1735 * This _changes_ the previous behaviour, where
1736 * tcp_create_openreq_child always was incrementing the
1737 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1738 * to be taken into account in all callers. -acme
1740 sk_refcnt_debug_inc(newsk);
1741 sk_set_socket(newsk, NULL);
1742 newsk->sk_wq = NULL;
1744 if (newsk->sk_prot->sockets_allocated)
1745 sk_sockets_allocated_inc(newsk);
1747 if (sock_needs_netstamp(sk) &&
1748 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1749 net_enable_timestamp();
1754 EXPORT_SYMBOL_GPL(sk_clone_lock);
1756 void sk_free_unlock_clone(struct sock *sk)
1758 /* It is still raw copy of parent, so invalidate
1759 * destructor and make plain sk_free() */
1760 sk->sk_destruct = NULL;
1764 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1766 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1770 sk_dst_set(sk, dst);
1771 sk->sk_route_caps = dst->dev->features;
1772 if (sk->sk_route_caps & NETIF_F_GSO)
1773 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1774 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1775 if (sk_can_gso(sk)) {
1776 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1777 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1779 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1780 sk->sk_gso_max_size = dst->dev->gso_max_size;
1781 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1784 sk->sk_gso_max_segs = max_segs;
1786 EXPORT_SYMBOL_GPL(sk_setup_caps);
1789 * Simple resource managers for sockets.
1794 * Write buffer destructor automatically called from kfree_skb.
1796 void sock_wfree(struct sk_buff *skb)
1798 struct sock *sk = skb->sk;
1799 unsigned int len = skb->truesize;
1801 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1803 * Keep a reference on sk_wmem_alloc, this will be released
1804 * after sk_write_space() call
1806 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1807 sk->sk_write_space(sk);
1811 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1812 * could not do because of in-flight packets
1814 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1817 EXPORT_SYMBOL(sock_wfree);
1819 /* This variant of sock_wfree() is used by TCP,
1820 * since it sets SOCK_USE_WRITE_QUEUE.
1822 void __sock_wfree(struct sk_buff *skb)
1824 struct sock *sk = skb->sk;
1826 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1830 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1835 if (unlikely(!sk_fullsock(sk))) {
1836 skb->destructor = sock_edemux;
1841 skb->destructor = sock_wfree;
1842 skb_set_hash_from_sk(skb, sk);
1844 * We used to take a refcount on sk, but following operation
1845 * is enough to guarantee sk_free() wont free this sock until
1846 * all in-flight packets are completed
1848 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1850 EXPORT_SYMBOL(skb_set_owner_w);
1852 /* This helper is used by netem, as it can hold packets in its
1853 * delay queue. We want to allow the owner socket to send more
1854 * packets, as if they were already TX completed by a typical driver.
1855 * But we also want to keep skb->sk set because some packet schedulers
1856 * rely on it (sch_fq for example).
1858 void skb_orphan_partial(struct sk_buff *skb)
1860 if (skb_is_tcp_pure_ack(skb))
1863 if (skb->destructor == sock_wfree
1865 || skb->destructor == tcp_wfree
1868 struct sock *sk = skb->sk;
1870 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1871 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1872 skb->destructor = sock_efree;
1878 EXPORT_SYMBOL(skb_orphan_partial);
1881 * Read buffer destructor automatically called from kfree_skb.
1883 void sock_rfree(struct sk_buff *skb)
1885 struct sock *sk = skb->sk;
1886 unsigned int len = skb->truesize;
1888 atomic_sub(len, &sk->sk_rmem_alloc);
1889 sk_mem_uncharge(sk, len);
1891 EXPORT_SYMBOL(sock_rfree);
1894 * Buffer destructor for skbs that are not used directly in read or write
1895 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1897 void sock_efree(struct sk_buff *skb)
1901 EXPORT_SYMBOL(sock_efree);
1903 kuid_t sock_i_uid(struct sock *sk)
1907 read_lock_bh(&sk->sk_callback_lock);
1908 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1909 read_unlock_bh(&sk->sk_callback_lock);
1912 EXPORT_SYMBOL(sock_i_uid);
1914 unsigned long sock_i_ino(struct sock *sk)
1918 read_lock_bh(&sk->sk_callback_lock);
1919 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1920 read_unlock_bh(&sk->sk_callback_lock);
1923 EXPORT_SYMBOL(sock_i_ino);
1926 * Allocate a skb from the socket's send buffer.
1928 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1931 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1932 struct sk_buff *skb = alloc_skb(size, priority);
1934 skb_set_owner_w(skb, sk);
1940 EXPORT_SYMBOL(sock_wmalloc);
1942 static void sock_ofree(struct sk_buff *skb)
1944 struct sock *sk = skb->sk;
1946 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1949 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1952 struct sk_buff *skb;
1954 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1955 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1959 skb = alloc_skb(size, priority);
1963 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1965 skb->destructor = sock_ofree;
1970 * Allocate a memory block from the socket's option memory buffer.
1972 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1974 if ((unsigned int)size <= sysctl_optmem_max &&
1975 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1977 /* First do the add, to avoid the race if kmalloc
1980 atomic_add(size, &sk->sk_omem_alloc);
1981 mem = kmalloc(size, priority);
1984 atomic_sub(size, &sk->sk_omem_alloc);
1988 EXPORT_SYMBOL(sock_kmalloc);
1990 /* Free an option memory block. Note, we actually want the inline
1991 * here as this allows gcc to detect the nullify and fold away the
1992 * condition entirely.
1994 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1997 if (WARN_ON_ONCE(!mem))
2003 atomic_sub(size, &sk->sk_omem_alloc);
2006 void sock_kfree_s(struct sock *sk, void *mem, int size)
2008 __sock_kfree_s(sk, mem, size, false);
2010 EXPORT_SYMBOL(sock_kfree_s);
2012 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2014 __sock_kfree_s(sk, mem, size, true);
2016 EXPORT_SYMBOL(sock_kzfree_s);
2018 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2019 I think, these locks should be removed for datagram sockets.
2021 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2025 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2029 if (signal_pending(current))
2031 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2032 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2033 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2035 if (sk->sk_shutdown & SEND_SHUTDOWN)
2039 timeo = schedule_timeout(timeo);
2041 finish_wait(sk_sleep(sk), &wait);
2047 * Generic send/receive buffer handlers
2050 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2051 unsigned long data_len, int noblock,
2052 int *errcode, int max_page_order)
2054 struct sk_buff *skb;
2058 timeo = sock_sndtimeo(sk, noblock);
2060 err = sock_error(sk);
2065 if (sk->sk_shutdown & SEND_SHUTDOWN)
2068 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2071 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2072 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2076 if (signal_pending(current))
2078 timeo = sock_wait_for_wmem(sk, timeo);
2080 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2081 errcode, sk->sk_allocation);
2083 skb_set_owner_w(skb, sk);
2087 err = sock_intr_errno(timeo);
2092 EXPORT_SYMBOL(sock_alloc_send_pskb);
2094 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2095 int noblock, int *errcode)
2097 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2099 EXPORT_SYMBOL(sock_alloc_send_skb);
2101 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2102 struct sockcm_cookie *sockc)
2106 switch (cmsg->cmsg_type) {
2108 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2110 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2112 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2114 case SO_TIMESTAMPING:
2115 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2118 tsflags = *(u32 *)CMSG_DATA(cmsg);
2119 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2122 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2123 sockc->tsflags |= tsflags;
2125 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2127 case SCM_CREDENTIALS:
2134 EXPORT_SYMBOL(__sock_cmsg_send);
2136 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2137 struct sockcm_cookie *sockc)
2139 struct cmsghdr *cmsg;
2142 for_each_cmsghdr(cmsg, msg) {
2143 if (!CMSG_OK(msg, cmsg))
2145 if (cmsg->cmsg_level != SOL_SOCKET)
2147 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2153 EXPORT_SYMBOL(sock_cmsg_send);
2155 static void sk_enter_memory_pressure(struct sock *sk)
2157 if (!sk->sk_prot->enter_memory_pressure)
2160 sk->sk_prot->enter_memory_pressure(sk);
2163 static void sk_leave_memory_pressure(struct sock *sk)
2165 if (sk->sk_prot->leave_memory_pressure) {
2166 sk->sk_prot->leave_memory_pressure(sk);
2168 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2170 if (memory_pressure && *memory_pressure)
2171 *memory_pressure = 0;
2175 /* On 32bit arches, an skb frag is limited to 2^15 */
2176 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2179 * skb_page_frag_refill - check that a page_frag contains enough room
2180 * @sz: minimum size of the fragment we want to get
2181 * @pfrag: pointer to page_frag
2182 * @gfp: priority for memory allocation
2184 * Note: While this allocator tries to use high order pages, there is
2185 * no guarantee that allocations succeed. Therefore, @sz MUST be
2186 * less or equal than PAGE_SIZE.
2188 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2191 if (page_ref_count(pfrag->page) == 1) {
2195 if (pfrag->offset + sz <= pfrag->size)
2197 put_page(pfrag->page);
2201 if (SKB_FRAG_PAGE_ORDER) {
2202 /* Avoid direct reclaim but allow kswapd to wake */
2203 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2204 __GFP_COMP | __GFP_NOWARN |
2206 SKB_FRAG_PAGE_ORDER);
2207 if (likely(pfrag->page)) {
2208 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2212 pfrag->page = alloc_page(gfp);
2213 if (likely(pfrag->page)) {
2214 pfrag->size = PAGE_SIZE;
2219 EXPORT_SYMBOL(skb_page_frag_refill);
2221 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2223 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2226 sk_enter_memory_pressure(sk);
2227 sk_stream_moderate_sndbuf(sk);
2230 EXPORT_SYMBOL(sk_page_frag_refill);
2232 static void __lock_sock(struct sock *sk)
2233 __releases(&sk->sk_lock.slock)
2234 __acquires(&sk->sk_lock.slock)
2239 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2240 TASK_UNINTERRUPTIBLE);
2241 spin_unlock_bh(&sk->sk_lock.slock);
2243 spin_lock_bh(&sk->sk_lock.slock);
2244 if (!sock_owned_by_user(sk))
2247 finish_wait(&sk->sk_lock.wq, &wait);
2250 static void __release_sock(struct sock *sk)
2251 __releases(&sk->sk_lock.slock)
2252 __acquires(&sk->sk_lock.slock)
2254 struct sk_buff *skb, *next;
2256 while ((skb = sk->sk_backlog.head) != NULL) {
2257 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2259 spin_unlock_bh(&sk->sk_lock.slock);
2264 WARN_ON_ONCE(skb_dst_is_noref(skb));
2266 sk_backlog_rcv(sk, skb);
2271 } while (skb != NULL);
2273 spin_lock_bh(&sk->sk_lock.slock);
2277 * Doing the zeroing here guarantee we can not loop forever
2278 * while a wild producer attempts to flood us.
2280 sk->sk_backlog.len = 0;
2283 void __sk_flush_backlog(struct sock *sk)
2285 spin_lock_bh(&sk->sk_lock.slock);
2287 spin_unlock_bh(&sk->sk_lock.slock);
2291 * sk_wait_data - wait for data to arrive at sk_receive_queue
2292 * @sk: sock to wait on
2293 * @timeo: for how long
2294 * @skb: last skb seen on sk_receive_queue
2296 * Now socket state including sk->sk_err is changed only under lock,
2297 * hence we may omit checks after joining wait queue.
2298 * We check receive queue before schedule() only as optimization;
2299 * it is very likely that release_sock() added new data.
2301 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2303 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2306 add_wait_queue(sk_sleep(sk), &wait);
2307 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2308 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2309 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2310 remove_wait_queue(sk_sleep(sk), &wait);
2313 EXPORT_SYMBOL(sk_wait_data);
2316 * __sk_mem_raise_allocated - increase memory_allocated
2318 * @size: memory size to allocate
2319 * @amt: pages to allocate
2320 * @kind: allocation type
2322 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2324 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2326 struct proto *prot = sk->sk_prot;
2327 long allocated = sk_memory_allocated_add(sk, amt);
2329 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2330 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2331 goto suppress_allocation;
2334 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2335 sk_leave_memory_pressure(sk);
2339 /* Under pressure. */
2340 if (allocated > sk_prot_mem_limits(sk, 1))
2341 sk_enter_memory_pressure(sk);
2343 /* Over hard limit. */
2344 if (allocated > sk_prot_mem_limits(sk, 2))
2345 goto suppress_allocation;
2347 /* guarantee minimum buffer size under pressure */
2348 if (kind == SK_MEM_RECV) {
2349 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2352 } else { /* SK_MEM_SEND */
2353 int wmem0 = sk_get_wmem0(sk, prot);
2355 if (sk->sk_type == SOCK_STREAM) {
2356 if (sk->sk_wmem_queued < wmem0)
2358 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2363 if (sk_has_memory_pressure(sk)) {
2366 if (!sk_under_memory_pressure(sk))
2368 alloc = sk_sockets_allocated_read_positive(sk);
2369 if (sk_prot_mem_limits(sk, 2) > alloc *
2370 sk_mem_pages(sk->sk_wmem_queued +
2371 atomic_read(&sk->sk_rmem_alloc) +
2372 sk->sk_forward_alloc))
2376 suppress_allocation:
2378 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2379 sk_stream_moderate_sndbuf(sk);
2381 /* Fail only if socket is _under_ its sndbuf.
2382 * In this case we cannot block, so that we have to fail.
2384 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2388 trace_sock_exceed_buf_limit(sk, prot, allocated);
2390 sk_memory_allocated_sub(sk, amt);
2392 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2393 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2397 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2400 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2402 * @size: memory size to allocate
2403 * @kind: allocation type
2405 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2406 * rmem allocation. This function assumes that protocols which have
2407 * memory_pressure use sk_wmem_queued as write buffer accounting.
2409 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2411 int ret, amt = sk_mem_pages(size);
2413 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2414 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2416 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2419 EXPORT_SYMBOL(__sk_mem_schedule);
2422 * __sk_mem_reduce_allocated - reclaim memory_allocated
2424 * @amount: number of quanta
2426 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2428 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2430 sk_memory_allocated_sub(sk, amount);
2432 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2433 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2435 if (sk_under_memory_pressure(sk) &&
2436 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2437 sk_leave_memory_pressure(sk);
2439 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2442 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2444 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2446 void __sk_mem_reclaim(struct sock *sk, int amount)
2448 amount >>= SK_MEM_QUANTUM_SHIFT;
2449 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2450 __sk_mem_reduce_allocated(sk, amount);
2452 EXPORT_SYMBOL(__sk_mem_reclaim);
2454 int sk_set_peek_off(struct sock *sk, int val)
2456 sk->sk_peek_off = val;
2459 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2462 * Set of default routines for initialising struct proto_ops when
2463 * the protocol does not support a particular function. In certain
2464 * cases where it makes no sense for a protocol to have a "do nothing"
2465 * function, some default processing is provided.
2468 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2472 EXPORT_SYMBOL(sock_no_bind);
2474 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2479 EXPORT_SYMBOL(sock_no_connect);
2481 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2485 EXPORT_SYMBOL(sock_no_socketpair);
2487 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2492 EXPORT_SYMBOL(sock_no_accept);
2494 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2499 EXPORT_SYMBOL(sock_no_getname);
2501 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2505 EXPORT_SYMBOL(sock_no_poll);
2507 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2511 EXPORT_SYMBOL(sock_no_ioctl);
2513 int sock_no_listen(struct socket *sock, int backlog)
2517 EXPORT_SYMBOL(sock_no_listen);
2519 int sock_no_shutdown(struct socket *sock, int how)
2523 EXPORT_SYMBOL(sock_no_shutdown);
2525 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2526 char __user *optval, unsigned int optlen)
2530 EXPORT_SYMBOL(sock_no_setsockopt);
2532 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2533 char __user *optval, int __user *optlen)
2537 EXPORT_SYMBOL(sock_no_getsockopt);
2539 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2543 EXPORT_SYMBOL(sock_no_sendmsg);
2545 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2549 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2551 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2556 EXPORT_SYMBOL(sock_no_recvmsg);
2558 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2560 /* Mirror missing mmap method error code */
2563 EXPORT_SYMBOL(sock_no_mmap);
2565 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2568 struct msghdr msg = {.msg_flags = flags};
2570 char *kaddr = kmap(page);
2571 iov.iov_base = kaddr + offset;
2573 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2577 EXPORT_SYMBOL(sock_no_sendpage);
2579 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2580 int offset, size_t size, int flags)
2583 struct msghdr msg = {.msg_flags = flags};
2585 char *kaddr = kmap(page);
2587 iov.iov_base = kaddr + offset;
2589 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2593 EXPORT_SYMBOL(sock_no_sendpage_locked);
2596 * Default Socket Callbacks
2599 static void sock_def_wakeup(struct sock *sk)
2601 struct socket_wq *wq;
2604 wq = rcu_dereference(sk->sk_wq);
2605 if (skwq_has_sleeper(wq))
2606 wake_up_interruptible_all(&wq->wait);
2610 static void sock_def_error_report(struct sock *sk)
2612 struct socket_wq *wq;
2615 wq = rcu_dereference(sk->sk_wq);
2616 if (skwq_has_sleeper(wq))
2617 wake_up_interruptible_poll(&wq->wait, POLLERR);
2618 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2622 static void sock_def_readable(struct sock *sk)
2624 struct socket_wq *wq;
2627 wq = rcu_dereference(sk->sk_wq);
2628 if (skwq_has_sleeper(wq))
2629 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2630 POLLRDNORM | POLLRDBAND);
2631 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2635 static void sock_def_write_space(struct sock *sk)
2637 struct socket_wq *wq;
2641 /* Do not wake up a writer until he can make "significant"
2644 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2645 wq = rcu_dereference(sk->sk_wq);
2646 if (skwq_has_sleeper(wq))
2647 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2648 POLLWRNORM | POLLWRBAND);
2650 /* Should agree with poll, otherwise some programs break */
2651 if (sock_writeable(sk))
2652 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2658 static void sock_def_destruct(struct sock *sk)
2662 void sk_send_sigurg(struct sock *sk)
2664 if (sk->sk_socket && sk->sk_socket->file)
2665 if (send_sigurg(&sk->sk_socket->file->f_owner))
2666 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2668 EXPORT_SYMBOL(sk_send_sigurg);
2670 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2671 unsigned long expires)
2673 if (!mod_timer(timer, expires))
2676 EXPORT_SYMBOL(sk_reset_timer);
2678 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2680 if (del_timer(timer))
2683 EXPORT_SYMBOL(sk_stop_timer);
2685 void sock_init_data(struct socket *sock, struct sock *sk)
2688 sk->sk_send_head = NULL;
2690 timer_setup(&sk->sk_timer, NULL, 0);
2692 sk->sk_allocation = GFP_KERNEL;
2693 sk->sk_rcvbuf = sysctl_rmem_default;
2694 sk->sk_sndbuf = sysctl_wmem_default;
2695 sk->sk_state = TCP_CLOSE;
2696 sk_set_socket(sk, sock);
2698 sock_set_flag(sk, SOCK_ZAPPED);
2701 sk->sk_type = sock->type;
2702 sk->sk_wq = sock->wq;
2704 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2707 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2710 rwlock_init(&sk->sk_callback_lock);
2711 if (sk->sk_kern_sock)
2712 lockdep_set_class_and_name(
2713 &sk->sk_callback_lock,
2714 af_kern_callback_keys + sk->sk_family,
2715 af_family_kern_clock_key_strings[sk->sk_family]);
2717 lockdep_set_class_and_name(
2718 &sk->sk_callback_lock,
2719 af_callback_keys + sk->sk_family,
2720 af_family_clock_key_strings[sk->sk_family]);
2722 sk->sk_state_change = sock_def_wakeup;
2723 sk->sk_data_ready = sock_def_readable;
2724 sk->sk_write_space = sock_def_write_space;
2725 sk->sk_error_report = sock_def_error_report;
2726 sk->sk_destruct = sock_def_destruct;
2728 sk->sk_frag.page = NULL;
2729 sk->sk_frag.offset = 0;
2730 sk->sk_peek_off = -1;
2732 sk->sk_peer_pid = NULL;
2733 sk->sk_peer_cred = NULL;
2734 sk->sk_write_pending = 0;
2735 sk->sk_rcvlowat = 1;
2736 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2737 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2739 sk->sk_stamp = SK_DEFAULT_STAMP;
2740 atomic_set(&sk->sk_zckey, 0);
2742 #ifdef CONFIG_NET_RX_BUSY_POLL
2744 sk->sk_ll_usec = sysctl_net_busy_read;
2747 sk->sk_max_pacing_rate = ~0U;
2748 sk->sk_pacing_rate = ~0U;
2749 sk->sk_pacing_shift = 10;
2750 sk->sk_incoming_cpu = -1;
2752 * Before updating sk_refcnt, we must commit prior changes to memory
2753 * (Documentation/RCU/rculist_nulls.txt for details)
2756 refcount_set(&sk->sk_refcnt, 1);
2757 atomic_set(&sk->sk_drops, 0);
2759 EXPORT_SYMBOL(sock_init_data);
2761 void lock_sock_nested(struct sock *sk, int subclass)
2764 spin_lock_bh(&sk->sk_lock.slock);
2765 if (sk->sk_lock.owned)
2767 sk->sk_lock.owned = 1;
2768 spin_unlock(&sk->sk_lock.slock);
2770 * The sk_lock has mutex_lock() semantics here:
2772 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2775 EXPORT_SYMBOL(lock_sock_nested);
2777 void release_sock(struct sock *sk)
2779 spin_lock_bh(&sk->sk_lock.slock);
2780 if (sk->sk_backlog.tail)
2783 /* Warning : release_cb() might need to release sk ownership,
2784 * ie call sock_release_ownership(sk) before us.
2786 if (sk->sk_prot->release_cb)
2787 sk->sk_prot->release_cb(sk);
2789 sock_release_ownership(sk);
2790 if (waitqueue_active(&sk->sk_lock.wq))
2791 wake_up(&sk->sk_lock.wq);
2792 spin_unlock_bh(&sk->sk_lock.slock);
2794 EXPORT_SYMBOL(release_sock);
2797 * lock_sock_fast - fast version of lock_sock
2800 * This version should be used for very small section, where process wont block
2801 * return false if fast path is taken:
2803 * sk_lock.slock locked, owned = 0, BH disabled
2805 * return true if slow path is taken:
2807 * sk_lock.slock unlocked, owned = 1, BH enabled
2809 bool lock_sock_fast(struct sock *sk)
2812 spin_lock_bh(&sk->sk_lock.slock);
2814 if (!sk->sk_lock.owned)
2816 * Note : We must disable BH
2821 sk->sk_lock.owned = 1;
2822 spin_unlock(&sk->sk_lock.slock);
2824 * The sk_lock has mutex_lock() semantics here:
2826 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2830 EXPORT_SYMBOL(lock_sock_fast);
2832 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2835 if (!sock_flag(sk, SOCK_TIMESTAMP))
2836 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2837 tv = ktime_to_timeval(sk->sk_stamp);
2838 if (tv.tv_sec == -1)
2840 if (tv.tv_sec == 0) {
2841 sk->sk_stamp = ktime_get_real();
2842 tv = ktime_to_timeval(sk->sk_stamp);
2844 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2846 EXPORT_SYMBOL(sock_get_timestamp);
2848 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2851 if (!sock_flag(sk, SOCK_TIMESTAMP))
2852 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2853 ts = ktime_to_timespec(sk->sk_stamp);
2854 if (ts.tv_sec == -1)
2856 if (ts.tv_sec == 0) {
2857 sk->sk_stamp = ktime_get_real();
2858 ts = ktime_to_timespec(sk->sk_stamp);
2860 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2862 EXPORT_SYMBOL(sock_get_timestampns);
2864 void sock_enable_timestamp(struct sock *sk, int flag)
2866 if (!sock_flag(sk, flag)) {
2867 unsigned long previous_flags = sk->sk_flags;
2869 sock_set_flag(sk, flag);
2871 * we just set one of the two flags which require net
2872 * time stamping, but time stamping might have been on
2873 * already because of the other one
2875 if (sock_needs_netstamp(sk) &&
2876 !(previous_flags & SK_FLAGS_TIMESTAMP))
2877 net_enable_timestamp();
2881 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2882 int level, int type)
2884 struct sock_exterr_skb *serr;
2885 struct sk_buff *skb;
2889 skb = sock_dequeue_err_skb(sk);
2895 msg->msg_flags |= MSG_TRUNC;
2898 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2902 sock_recv_timestamp(msg, sk, skb);
2904 serr = SKB_EXT_ERR(skb);
2905 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2907 msg->msg_flags |= MSG_ERRQUEUE;
2915 EXPORT_SYMBOL(sock_recv_errqueue);
2918 * Get a socket option on an socket.
2920 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2921 * asynchronous errors should be reported by getsockopt. We assume
2922 * this means if you specify SO_ERROR (otherwise whats the point of it).
2924 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2925 char __user *optval, int __user *optlen)
2927 struct sock *sk = sock->sk;
2929 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2931 EXPORT_SYMBOL(sock_common_getsockopt);
2933 #ifdef CONFIG_COMPAT
2934 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2935 char __user *optval, int __user *optlen)
2937 struct sock *sk = sock->sk;
2939 if (sk->sk_prot->compat_getsockopt != NULL)
2940 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2942 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2944 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2947 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2950 struct sock *sk = sock->sk;
2954 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2955 flags & ~MSG_DONTWAIT, &addr_len);
2957 msg->msg_namelen = addr_len;
2960 EXPORT_SYMBOL(sock_common_recvmsg);
2963 * Set socket options on an inet socket.
2965 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2966 char __user *optval, unsigned int optlen)
2968 struct sock *sk = sock->sk;
2970 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2972 EXPORT_SYMBOL(sock_common_setsockopt);
2974 #ifdef CONFIG_COMPAT
2975 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2976 char __user *optval, unsigned int optlen)
2978 struct sock *sk = sock->sk;
2980 if (sk->sk_prot->compat_setsockopt != NULL)
2981 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2983 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2985 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2988 void sk_common_release(struct sock *sk)
2990 if (sk->sk_prot->destroy)
2991 sk->sk_prot->destroy(sk);
2994 * Observation: when sock_common_release is called, processes have
2995 * no access to socket. But net still has.
2996 * Step one, detach it from networking:
2998 * A. Remove from hash tables.
3001 sk->sk_prot->unhash(sk);
3004 * In this point socket cannot receive new packets, but it is possible
3005 * that some packets are in flight because some CPU runs receiver and
3006 * did hash table lookup before we unhashed socket. They will achieve
3007 * receive queue and will be purged by socket destructor.
3009 * Also we still have packets pending on receive queue and probably,
3010 * our own packets waiting in device queues. sock_destroy will drain
3011 * receive queue, but transmitted packets will delay socket destruction
3012 * until the last reference will be released.
3017 xfrm_sk_free_policy(sk);
3019 sk_refcnt_debug_release(sk);
3023 EXPORT_SYMBOL(sk_common_release);
3025 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3027 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3029 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3030 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3031 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3032 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3033 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3034 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3035 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3036 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3037 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3040 #ifdef CONFIG_PROC_FS
3041 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3043 int val[PROTO_INUSE_NR];
3046 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3048 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3050 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
3052 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3054 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3056 int cpu, idx = prot->inuse_idx;
3059 for_each_possible_cpu(cpu)
3060 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
3062 return res >= 0 ? res : 0;
3064 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3066 static int __net_init sock_inuse_init_net(struct net *net)
3068 net->core.inuse = alloc_percpu(struct prot_inuse);
3069 return net->core.inuse ? 0 : -ENOMEM;
3072 static void __net_exit sock_inuse_exit_net(struct net *net)
3074 free_percpu(net->core.inuse);
3077 static struct pernet_operations net_inuse_ops = {
3078 .init = sock_inuse_init_net,
3079 .exit = sock_inuse_exit_net,
3082 static __init int net_inuse_init(void)
3084 if (register_pernet_subsys(&net_inuse_ops))
3085 panic("Cannot initialize net inuse counters");
3090 core_initcall(net_inuse_init);
3092 static void assign_proto_idx(struct proto *prot)
3094 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3096 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3097 pr_err("PROTO_INUSE_NR exhausted\n");
3101 set_bit(prot->inuse_idx, proto_inuse_idx);
3104 static void release_proto_idx(struct proto *prot)
3106 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3107 clear_bit(prot->inuse_idx, proto_inuse_idx);
3110 static inline void assign_proto_idx(struct proto *prot)
3114 static inline void release_proto_idx(struct proto *prot)
3119 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3123 kfree(rsk_prot->slab_name);
3124 rsk_prot->slab_name = NULL;
3125 kmem_cache_destroy(rsk_prot->slab);
3126 rsk_prot->slab = NULL;
3129 static int req_prot_init(const struct proto *prot)
3131 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3136 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3138 if (!rsk_prot->slab_name)
3141 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3142 rsk_prot->obj_size, 0,
3143 prot->slab_flags, NULL);
3145 if (!rsk_prot->slab) {
3146 pr_crit("%s: Can't create request sock SLAB cache!\n",
3153 int proto_register(struct proto *prot, int alloc_slab)
3156 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
3157 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3160 if (prot->slab == NULL) {
3161 pr_crit("%s: Can't create sock SLAB cache!\n",
3166 if (req_prot_init(prot))
3167 goto out_free_request_sock_slab;
3169 if (prot->twsk_prot != NULL) {
3170 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3172 if (prot->twsk_prot->twsk_slab_name == NULL)
3173 goto out_free_request_sock_slab;
3175 prot->twsk_prot->twsk_slab =
3176 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3177 prot->twsk_prot->twsk_obj_size,
3181 if (prot->twsk_prot->twsk_slab == NULL)
3182 goto out_free_timewait_sock_slab_name;
3186 mutex_lock(&proto_list_mutex);
3187 list_add(&prot->node, &proto_list);
3188 assign_proto_idx(prot);
3189 mutex_unlock(&proto_list_mutex);
3192 out_free_timewait_sock_slab_name:
3193 kfree(prot->twsk_prot->twsk_slab_name);
3194 out_free_request_sock_slab:
3195 req_prot_cleanup(prot->rsk_prot);
3197 kmem_cache_destroy(prot->slab);
3202 EXPORT_SYMBOL(proto_register);
3204 void proto_unregister(struct proto *prot)
3206 mutex_lock(&proto_list_mutex);
3207 release_proto_idx(prot);
3208 list_del(&prot->node);
3209 mutex_unlock(&proto_list_mutex);
3211 kmem_cache_destroy(prot->slab);
3214 req_prot_cleanup(prot->rsk_prot);
3216 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3217 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3218 kfree(prot->twsk_prot->twsk_slab_name);
3219 prot->twsk_prot->twsk_slab = NULL;
3222 EXPORT_SYMBOL(proto_unregister);
3224 #ifdef CONFIG_PROC_FS
3225 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3226 __acquires(proto_list_mutex)
3228 mutex_lock(&proto_list_mutex);
3229 return seq_list_start_head(&proto_list, *pos);
3232 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3234 return seq_list_next(v, &proto_list, pos);
3237 static void proto_seq_stop(struct seq_file *seq, void *v)
3238 __releases(proto_list_mutex)
3240 mutex_unlock(&proto_list_mutex);
3243 static char proto_method_implemented(const void *method)
3245 return method == NULL ? 'n' : 'y';
3247 static long sock_prot_memory_allocated(struct proto *proto)
3249 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3252 static char *sock_prot_memory_pressure(struct proto *proto)
3254 return proto->memory_pressure != NULL ?
3255 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3258 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3261 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3262 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3265 sock_prot_inuse_get(seq_file_net(seq), proto),
3266 sock_prot_memory_allocated(proto),
3267 sock_prot_memory_pressure(proto),
3269 proto->slab == NULL ? "no" : "yes",
3270 module_name(proto->owner),
3271 proto_method_implemented(proto->close),
3272 proto_method_implemented(proto->connect),
3273 proto_method_implemented(proto->disconnect),
3274 proto_method_implemented(proto->accept),
3275 proto_method_implemented(proto->ioctl),
3276 proto_method_implemented(proto->init),
3277 proto_method_implemented(proto->destroy),
3278 proto_method_implemented(proto->shutdown),
3279 proto_method_implemented(proto->setsockopt),
3280 proto_method_implemented(proto->getsockopt),
3281 proto_method_implemented(proto->sendmsg),
3282 proto_method_implemented(proto->recvmsg),
3283 proto_method_implemented(proto->sendpage),
3284 proto_method_implemented(proto->bind),
3285 proto_method_implemented(proto->backlog_rcv),
3286 proto_method_implemented(proto->hash),
3287 proto_method_implemented(proto->unhash),
3288 proto_method_implemented(proto->get_port),
3289 proto_method_implemented(proto->enter_memory_pressure));
3292 static int proto_seq_show(struct seq_file *seq, void *v)
3294 if (v == &proto_list)
3295 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3304 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3306 proto_seq_printf(seq, list_entry(v, struct proto, node));
3310 static const struct seq_operations proto_seq_ops = {
3311 .start = proto_seq_start,
3312 .next = proto_seq_next,
3313 .stop = proto_seq_stop,
3314 .show = proto_seq_show,
3317 static int proto_seq_open(struct inode *inode, struct file *file)
3319 return seq_open_net(inode, file, &proto_seq_ops,
3320 sizeof(struct seq_net_private));
3323 static const struct file_operations proto_seq_fops = {
3324 .owner = THIS_MODULE,
3325 .open = proto_seq_open,
3327 .llseek = seq_lseek,
3328 .release = seq_release_net,
3331 static __net_init int proto_init_net(struct net *net)
3333 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3339 static __net_exit void proto_exit_net(struct net *net)
3341 remove_proc_entry("protocols", net->proc_net);
3345 static __net_initdata struct pernet_operations proto_net_ops = {
3346 .init = proto_init_net,
3347 .exit = proto_exit_net,
3350 static int __init proto_init(void)
3352 return register_pernet_subsys(&proto_net_ops);
3355 subsys_initcall(proto_init);
3357 #endif /* PROC_FS */
3359 #ifdef CONFIG_NET_RX_BUSY_POLL
3360 bool sk_busy_loop_end(void *p, unsigned long start_time)
3362 struct sock *sk = p;
3364 return !skb_queue_empty(&sk->sk_receive_queue) ||
3365 sk_busy_loop_timeout(sk, start_time);
3367 EXPORT_SYMBOL(sk_busy_loop_end);
3368 #endif /* CONFIG_NET_RX_BUSY_POLL */