2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
166 EXPORT_SYMBOL(sk_ns_capable);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock *sk, int cap)
179 return sk_ns_capable(sk, &init_user_ns, cap);
181 EXPORT_SYMBOL(sk_capable);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock *sk, int cap)
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
196 EXPORT_SYMBOL(sk_net_capable);
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family and separate keys for internal and
203 static struct lock_class_key af_family_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_keys[AF_MAX];
205 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209 * Make lock validator output more readable. (we pre-construct these
210 * strings build-time, so that runtime initialization of socket
214 #define _sock_locks(x) \
215 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
216 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
217 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
218 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
219 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
220 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
221 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
222 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
223 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
224 x "27" , x "28" , x "AF_CAN" , \
225 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
226 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
227 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
228 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
229 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
231 static const char *const af_family_key_strings[AF_MAX+1] = {
232 _sock_locks("sk_lock-")
234 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
235 _sock_locks("slock-")
237 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
238 _sock_locks("clock-")
241 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
242 _sock_locks("k-sk_lock-")
244 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-slock-")
247 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
248 _sock_locks("k-clock-")
250 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
251 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
252 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
253 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
254 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
255 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
256 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
257 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
258 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
259 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
260 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
261 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
262 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
263 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
264 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
265 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
267 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
268 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
269 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
270 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
271 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
272 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
273 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
274 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
275 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
276 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
277 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
278 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
279 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
280 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
281 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
282 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
284 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
285 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
286 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
287 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
288 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
289 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
290 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
291 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
292 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
293 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
294 "elock-27" , "elock-28" , "elock-AF_CAN" ,
295 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
296 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
297 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
298 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
299 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
303 * sk_callback_lock and sk queues locking rules are per-address-family,
304 * so split the lock classes by using a per-AF key:
306 static struct lock_class_key af_callback_keys[AF_MAX];
307 static struct lock_class_key af_rlock_keys[AF_MAX];
308 static struct lock_class_key af_wlock_keys[AF_MAX];
309 static struct lock_class_key af_elock_keys[AF_MAX];
310 static struct lock_class_key af_kern_callback_keys[AF_MAX];
312 /* Take into consideration the size of the struct sk_buff overhead in the
313 * determination of these values, since that is non-constant across
314 * platforms. This makes socket queueing behavior and performance
315 * not depend upon such differences.
317 #define _SK_MEM_PACKETS 256
318 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
319 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
320 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
322 /* Run time adjustable parameters. */
323 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
324 EXPORT_SYMBOL(sysctl_wmem_max);
325 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
326 EXPORT_SYMBOL(sysctl_rmem_max);
327 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
328 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
330 /* Maximal space eaten by iovec or ancillary data plus some space */
331 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
332 EXPORT_SYMBOL(sysctl_optmem_max);
334 int sysctl_tstamp_allow_data __read_mostly = 1;
336 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
337 EXPORT_SYMBOL_GPL(memalloc_socks);
340 * sk_set_memalloc - sets %SOCK_MEMALLOC
341 * @sk: socket to set it on
343 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
344 * It's the responsibility of the admin to adjust min_free_kbytes
345 * to meet the requirements
347 void sk_set_memalloc(struct sock *sk)
349 sock_set_flag(sk, SOCK_MEMALLOC);
350 sk->sk_allocation |= __GFP_MEMALLOC;
351 static_key_slow_inc(&memalloc_socks);
353 EXPORT_SYMBOL_GPL(sk_set_memalloc);
355 void sk_clear_memalloc(struct sock *sk)
357 sock_reset_flag(sk, SOCK_MEMALLOC);
358 sk->sk_allocation &= ~__GFP_MEMALLOC;
359 static_key_slow_dec(&memalloc_socks);
362 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
363 * progress of swapping. SOCK_MEMALLOC may be cleared while
364 * it has rmem allocations due to the last swapfile being deactivated
365 * but there is a risk that the socket is unusable due to exceeding
366 * the rmem limits. Reclaim the reserves and obey rmem limits again.
370 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
372 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
375 unsigned long pflags = current->flags;
377 /* these should have been dropped before queueing */
378 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
380 current->flags |= PF_MEMALLOC;
381 ret = sk->sk_backlog_rcv(sk, skb);
382 tsk_restore_flags(current, pflags, PF_MEMALLOC);
386 EXPORT_SYMBOL(__sk_backlog_rcv);
388 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
392 if (optlen < sizeof(tv))
394 if (copy_from_user(&tv, optval, sizeof(tv)))
396 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
400 static int warned __read_mostly;
403 if (warned < 10 && net_ratelimit()) {
405 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
406 __func__, current->comm, task_pid_nr(current));
410 *timeo_p = MAX_SCHEDULE_TIMEOUT;
411 if (tv.tv_sec == 0 && tv.tv_usec == 0)
413 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
414 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
418 static void sock_warn_obsolete_bsdism(const char *name)
421 static char warncomm[TASK_COMM_LEN];
422 if (strcmp(warncomm, current->comm) && warned < 5) {
423 strcpy(warncomm, current->comm);
424 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
430 static bool sock_needs_netstamp(const struct sock *sk)
432 switch (sk->sk_family) {
441 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
443 if (sk->sk_flags & flags) {
444 sk->sk_flags &= ~flags;
445 if (sock_needs_netstamp(sk) &&
446 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
447 net_disable_timestamp();
452 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
455 struct sk_buff_head *list = &sk->sk_receive_queue;
457 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
458 atomic_inc(&sk->sk_drops);
459 trace_sock_rcvqueue_full(sk, skb);
463 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
464 atomic_inc(&sk->sk_drops);
469 skb_set_owner_r(skb, sk);
471 /* we escape from rcu protected region, make sure we dont leak
476 spin_lock_irqsave(&list->lock, flags);
477 sock_skb_set_dropcount(sk, skb);
478 __skb_queue_tail(list, skb);
479 spin_unlock_irqrestore(&list->lock, flags);
481 if (!sock_flag(sk, SOCK_DEAD))
482 sk->sk_data_ready(sk);
485 EXPORT_SYMBOL(__sock_queue_rcv_skb);
487 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
491 err = sk_filter(sk, skb);
495 return __sock_queue_rcv_skb(sk, skb);
497 EXPORT_SYMBOL(sock_queue_rcv_skb);
499 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
500 const int nested, unsigned int trim_cap, bool refcounted)
502 int rc = NET_RX_SUCCESS;
504 if (sk_filter_trim_cap(sk, skb, trim_cap))
505 goto discard_and_relse;
509 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
510 atomic_inc(&sk->sk_drops);
511 goto discard_and_relse;
514 bh_lock_sock_nested(sk);
517 if (!sock_owned_by_user(sk)) {
519 * trylock + unlock semantics:
521 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
523 rc = sk_backlog_rcv(sk, skb);
525 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
526 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
528 atomic_inc(&sk->sk_drops);
529 goto discard_and_relse;
541 EXPORT_SYMBOL(__sk_receive_skb);
543 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
545 struct dst_entry *dst = __sk_dst_get(sk);
547 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
548 sk_tx_queue_clear(sk);
549 sk->sk_dst_pending_confirm = 0;
550 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
557 EXPORT_SYMBOL(__sk_dst_check);
559 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
561 struct dst_entry *dst = sk_dst_get(sk);
563 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
571 EXPORT_SYMBOL(sk_dst_check);
573 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
576 int ret = -ENOPROTOOPT;
577 #ifdef CONFIG_NETDEVICES
578 struct net *net = sock_net(sk);
579 char devname[IFNAMSIZ];
584 if (!ns_capable(net->user_ns, CAP_NET_RAW))
591 /* Bind this socket to a particular device like "eth0",
592 * as specified in the passed interface name. If the
593 * name is "" or the option length is zero the socket
596 if (optlen > IFNAMSIZ - 1)
597 optlen = IFNAMSIZ - 1;
598 memset(devname, 0, sizeof(devname));
601 if (copy_from_user(devname, optval, optlen))
605 if (devname[0] != '\0') {
606 struct net_device *dev;
609 dev = dev_get_by_name_rcu(net, devname);
611 index = dev->ifindex;
619 sk->sk_bound_dev_if = index;
631 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
632 int __user *optlen, int len)
634 int ret = -ENOPROTOOPT;
635 #ifdef CONFIG_NETDEVICES
636 struct net *net = sock_net(sk);
637 char devname[IFNAMSIZ];
639 if (sk->sk_bound_dev_if == 0) {
648 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
652 len = strlen(devname) + 1;
655 if (copy_to_user(optval, devname, len))
660 if (put_user(len, optlen))
671 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
674 sock_set_flag(sk, bit);
676 sock_reset_flag(sk, bit);
679 bool sk_mc_loop(struct sock *sk)
681 if (dev_recursion_level())
685 switch (sk->sk_family) {
687 return inet_sk(sk)->mc_loop;
688 #if IS_ENABLED(CONFIG_IPV6)
690 return inet6_sk(sk)->mc_loop;
696 EXPORT_SYMBOL(sk_mc_loop);
699 * This is meant for all protocols to use and covers goings on
700 * at the socket level. Everything here is generic.
703 int sock_setsockopt(struct socket *sock, int level, int optname,
704 char __user *optval, unsigned int optlen)
706 struct sock *sk = sock->sk;
713 * Options without arguments
716 if (optname == SO_BINDTODEVICE)
717 return sock_setbindtodevice(sk, optval, optlen);
719 if (optlen < sizeof(int))
722 if (get_user(val, (int __user *)optval))
725 valbool = val ? 1 : 0;
731 if (val && !capable(CAP_NET_ADMIN))
734 sock_valbool_flag(sk, SOCK_DBG, valbool);
737 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
740 sk->sk_reuseport = valbool;
749 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
752 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
755 /* Don't error on this BSD doesn't and if you think
756 * about it this is right. Otherwise apps have to
757 * play 'guess the biggest size' games. RCVBUF/SNDBUF
758 * are treated in BSD as hints
760 val = min_t(u32, val, sysctl_wmem_max);
762 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
763 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
764 /* Wake up sending tasks if we upped the value. */
765 sk->sk_write_space(sk);
769 if (!capable(CAP_NET_ADMIN)) {
776 /* Don't error on this BSD doesn't and if you think
777 * about it this is right. Otherwise apps have to
778 * play 'guess the biggest size' games. RCVBUF/SNDBUF
779 * are treated in BSD as hints
781 val = min_t(u32, val, sysctl_rmem_max);
783 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
785 * We double it on the way in to account for
786 * "struct sk_buff" etc. overhead. Applications
787 * assume that the SO_RCVBUF setting they make will
788 * allow that much actual data to be received on that
791 * Applications are unaware that "struct sk_buff" and
792 * other overheads allocate from the receive buffer
793 * during socket buffer allocation.
795 * And after considering the possible alternatives,
796 * returning the value we actually used in getsockopt
797 * is the most desirable behavior.
799 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
803 if (!capable(CAP_NET_ADMIN)) {
810 if (sk->sk_prot->keepalive)
811 sk->sk_prot->keepalive(sk, valbool);
812 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
816 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
820 sk->sk_no_check_tx = valbool;
824 if ((val >= 0 && val <= 6) ||
825 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
826 sk->sk_priority = val;
832 if (optlen < sizeof(ling)) {
833 ret = -EINVAL; /* 1003.1g */
836 if (copy_from_user(&ling, optval, sizeof(ling))) {
841 sock_reset_flag(sk, SOCK_LINGER);
843 #if (BITS_PER_LONG == 32)
844 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
845 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
848 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
849 sock_set_flag(sk, SOCK_LINGER);
854 sock_warn_obsolete_bsdism("setsockopt");
859 set_bit(SOCK_PASSCRED, &sock->flags);
861 clear_bit(SOCK_PASSCRED, &sock->flags);
867 if (optname == SO_TIMESTAMP)
868 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
870 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
871 sock_set_flag(sk, SOCK_RCVTSTAMP);
872 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
874 sock_reset_flag(sk, SOCK_RCVTSTAMP);
875 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
879 case SO_TIMESTAMPING:
880 if (val & ~SOF_TIMESTAMPING_MASK) {
885 if (val & SOF_TIMESTAMPING_OPT_ID &&
886 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
887 if (sk->sk_protocol == IPPROTO_TCP &&
888 sk->sk_type == SOCK_STREAM) {
889 if ((1 << sk->sk_state) &
890 (TCPF_CLOSE | TCPF_LISTEN)) {
894 sk->sk_tskey = tcp_sk(sk)->snd_una;
900 if (val & SOF_TIMESTAMPING_OPT_STATS &&
901 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
906 sk->sk_tsflags = val;
907 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
908 sock_enable_timestamp(sk,
909 SOCK_TIMESTAMPING_RX_SOFTWARE);
911 sock_disable_timestamp(sk,
912 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
918 sk->sk_rcvlowat = val ? : 1;
922 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
926 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
929 case SO_ATTACH_FILTER:
931 if (optlen == sizeof(struct sock_fprog)) {
932 struct sock_fprog fprog;
935 if (copy_from_user(&fprog, optval, sizeof(fprog)))
938 ret = sk_attach_filter(&fprog, sk);
944 if (optlen == sizeof(u32)) {
948 if (copy_from_user(&ufd, optval, sizeof(ufd)))
951 ret = sk_attach_bpf(ufd, sk);
955 case SO_ATTACH_REUSEPORT_CBPF:
957 if (optlen == sizeof(struct sock_fprog)) {
958 struct sock_fprog fprog;
961 if (copy_from_user(&fprog, optval, sizeof(fprog)))
964 ret = sk_reuseport_attach_filter(&fprog, sk);
968 case SO_ATTACH_REUSEPORT_EBPF:
970 if (optlen == sizeof(u32)) {
974 if (copy_from_user(&ufd, optval, sizeof(ufd)))
977 ret = sk_reuseport_attach_bpf(ufd, sk);
981 case SO_DETACH_FILTER:
982 ret = sk_detach_filter(sk);
986 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
989 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
994 set_bit(SOCK_PASSSEC, &sock->flags);
996 clear_bit(SOCK_PASSSEC, &sock->flags);
999 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1006 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1009 case SO_WIFI_STATUS:
1010 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1014 if (sock->ops->set_peek_off)
1015 ret = sock->ops->set_peek_off(sk, val);
1021 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1024 case SO_SELECT_ERR_QUEUE:
1025 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1028 #ifdef CONFIG_NET_RX_BUSY_POLL
1030 /* allow unprivileged users to decrease the value */
1031 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1037 sk->sk_ll_usec = val;
1042 case SO_MAX_PACING_RATE:
1043 sk->sk_max_pacing_rate = val;
1044 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1045 sk->sk_max_pacing_rate);
1048 case SO_INCOMING_CPU:
1049 sk->sk_incoming_cpu = val;
1054 dst_negative_advice(sk);
1063 EXPORT_SYMBOL(sock_setsockopt);
1066 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1067 struct ucred *ucred)
1069 ucred->pid = pid_vnr(pid);
1070 ucred->uid = ucred->gid = -1;
1072 struct user_namespace *current_ns = current_user_ns();
1074 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1075 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1079 int sock_getsockopt(struct socket *sock, int level, int optname,
1080 char __user *optval, int __user *optlen)
1082 struct sock *sk = sock->sk;
1090 int lv = sizeof(int);
1093 if (get_user(len, optlen))
1098 memset(&v, 0, sizeof(v));
1102 v.val = sock_flag(sk, SOCK_DBG);
1106 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1110 v.val = sock_flag(sk, SOCK_BROADCAST);
1114 v.val = sk->sk_sndbuf;
1118 v.val = sk->sk_rcvbuf;
1122 v.val = sk->sk_reuse;
1126 v.val = sk->sk_reuseport;
1130 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1134 v.val = sk->sk_type;
1138 v.val = sk->sk_protocol;
1142 v.val = sk->sk_family;
1146 v.val = -sock_error(sk);
1148 v.val = xchg(&sk->sk_err_soft, 0);
1152 v.val = sock_flag(sk, SOCK_URGINLINE);
1156 v.val = sk->sk_no_check_tx;
1160 v.val = sk->sk_priority;
1164 lv = sizeof(v.ling);
1165 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1166 v.ling.l_linger = sk->sk_lingertime / HZ;
1170 sock_warn_obsolete_bsdism("getsockopt");
1174 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1175 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1178 case SO_TIMESTAMPNS:
1179 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1182 case SO_TIMESTAMPING:
1183 v.val = sk->sk_tsflags;
1187 lv = sizeof(struct timeval);
1188 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1192 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1193 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1198 lv = sizeof(struct timeval);
1199 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1203 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1204 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1209 v.val = sk->sk_rcvlowat;
1217 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1222 struct ucred peercred;
1223 if (len > sizeof(peercred))
1224 len = sizeof(peercred);
1225 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1226 if (copy_to_user(optval, &peercred, len))
1235 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1239 if (copy_to_user(optval, address, len))
1244 /* Dubious BSD thing... Probably nobody even uses it, but
1245 * the UNIX standard wants it for whatever reason... -DaveM
1248 v.val = sk->sk_state == TCP_LISTEN;
1252 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1256 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1259 v.val = sk->sk_mark;
1263 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1266 case SO_WIFI_STATUS:
1267 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1271 if (!sock->ops->set_peek_off)
1274 v.val = sk->sk_peek_off;
1277 v.val = sock_flag(sk, SOCK_NOFCS);
1280 case SO_BINDTODEVICE:
1281 return sock_getbindtodevice(sk, optval, optlen, len);
1284 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1290 case SO_LOCK_FILTER:
1291 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1294 case SO_BPF_EXTENSIONS:
1295 v.val = bpf_tell_extensions();
1298 case SO_SELECT_ERR_QUEUE:
1299 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1302 #ifdef CONFIG_NET_RX_BUSY_POLL
1304 v.val = sk->sk_ll_usec;
1308 case SO_MAX_PACING_RATE:
1309 v.val = sk->sk_max_pacing_rate;
1312 case SO_INCOMING_CPU:
1313 v.val = sk->sk_incoming_cpu;
1317 /* We implement the SO_SNDLOWAT etc to not be settable
1320 return -ENOPROTOOPT;
1325 if (copy_to_user(optval, &v, len))
1328 if (put_user(len, optlen))
1334 * Initialize an sk_lock.
1336 * (We also register the sk_lock with the lock validator.)
1338 static inline void sock_lock_init(struct sock *sk)
1340 if (sk->sk_kern_sock)
1341 sock_lock_init_class_and_name(
1343 af_family_kern_slock_key_strings[sk->sk_family],
1344 af_family_kern_slock_keys + sk->sk_family,
1345 af_family_kern_key_strings[sk->sk_family],
1346 af_family_kern_keys + sk->sk_family);
1348 sock_lock_init_class_and_name(
1350 af_family_slock_key_strings[sk->sk_family],
1351 af_family_slock_keys + sk->sk_family,
1352 af_family_key_strings[sk->sk_family],
1353 af_family_keys + sk->sk_family);
1357 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1358 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1359 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1361 static void sock_copy(struct sock *nsk, const struct sock *osk)
1363 #ifdef CONFIG_SECURITY_NETWORK
1364 void *sptr = nsk->sk_security;
1366 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1368 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1369 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1371 #ifdef CONFIG_SECURITY_NETWORK
1372 nsk->sk_security = sptr;
1373 security_sk_clone(osk, nsk);
1377 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1381 struct kmem_cache *slab;
1385 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1388 if (priority & __GFP_ZERO)
1389 sk_prot_clear_nulls(sk, prot->obj_size);
1391 sk = kmalloc(prot->obj_size, priority);
1394 kmemcheck_annotate_bitfield(sk, flags);
1396 if (security_sk_alloc(sk, family, priority))
1399 if (!try_module_get(prot->owner))
1401 sk_tx_queue_clear(sk);
1407 security_sk_free(sk);
1410 kmem_cache_free(slab, sk);
1416 static void sk_prot_free(struct proto *prot, struct sock *sk)
1418 struct kmem_cache *slab;
1419 struct module *owner;
1421 owner = prot->owner;
1424 cgroup_sk_free(&sk->sk_cgrp_data);
1425 mem_cgroup_sk_free(sk);
1426 security_sk_free(sk);
1428 kmem_cache_free(slab, sk);
1435 * sk_alloc - All socket objects are allocated here
1436 * @net: the applicable net namespace
1437 * @family: protocol family
1438 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1439 * @prot: struct proto associated with this new sock instance
1440 * @kern: is this to be a kernel socket?
1442 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1443 struct proto *prot, int kern)
1447 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1449 sk->sk_family = family;
1451 * See comment in struct sock definition to understand
1452 * why we need sk_prot_creator -acme
1454 sk->sk_prot = sk->sk_prot_creator = prot;
1455 sk->sk_kern_sock = kern;
1457 sk->sk_net_refcnt = kern ? 0 : 1;
1458 if (likely(sk->sk_net_refcnt))
1460 sock_net_set(sk, net);
1461 atomic_set(&sk->sk_wmem_alloc, 1);
1463 mem_cgroup_sk_alloc(sk);
1464 cgroup_sk_alloc(&sk->sk_cgrp_data);
1465 sock_update_classid(&sk->sk_cgrp_data);
1466 sock_update_netprioidx(&sk->sk_cgrp_data);
1471 EXPORT_SYMBOL(sk_alloc);
1473 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1474 * grace period. This is the case for UDP sockets and TCP listeners.
1476 static void __sk_destruct(struct rcu_head *head)
1478 struct sock *sk = container_of(head, struct sock, sk_rcu);
1479 struct sk_filter *filter;
1481 if (sk->sk_destruct)
1482 sk->sk_destruct(sk);
1484 filter = rcu_dereference_check(sk->sk_filter,
1485 atomic_read(&sk->sk_wmem_alloc) == 0);
1487 sk_filter_uncharge(sk, filter);
1488 RCU_INIT_POINTER(sk->sk_filter, NULL);
1490 if (rcu_access_pointer(sk->sk_reuseport_cb))
1491 reuseport_detach_sock(sk);
1493 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1495 if (atomic_read(&sk->sk_omem_alloc))
1496 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1497 __func__, atomic_read(&sk->sk_omem_alloc));
1499 if (sk->sk_peer_cred)
1500 put_cred(sk->sk_peer_cred);
1501 put_pid(sk->sk_peer_pid);
1502 if (likely(sk->sk_net_refcnt))
1503 put_net(sock_net(sk));
1504 sk_prot_free(sk->sk_prot_creator, sk);
1507 void sk_destruct(struct sock *sk)
1509 if (sock_flag(sk, SOCK_RCU_FREE))
1510 call_rcu(&sk->sk_rcu, __sk_destruct);
1512 __sk_destruct(&sk->sk_rcu);
1515 static void __sk_free(struct sock *sk)
1517 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1518 sock_diag_broadcast_destroy(sk);
1523 void sk_free(struct sock *sk)
1526 * We subtract one from sk_wmem_alloc and can know if
1527 * some packets are still in some tx queue.
1528 * If not null, sock_wfree() will call __sk_free(sk) later
1530 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1533 EXPORT_SYMBOL(sk_free);
1535 static void sk_init_common(struct sock *sk)
1537 skb_queue_head_init(&sk->sk_receive_queue);
1538 skb_queue_head_init(&sk->sk_write_queue);
1539 skb_queue_head_init(&sk->sk_error_queue);
1541 rwlock_init(&sk->sk_callback_lock);
1542 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1543 af_rlock_keys + sk->sk_family,
1544 af_family_rlock_key_strings[sk->sk_family]);
1545 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1546 af_wlock_keys + sk->sk_family,
1547 af_family_wlock_key_strings[sk->sk_family]);
1548 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1549 af_elock_keys + sk->sk_family,
1550 af_family_elock_key_strings[sk->sk_family]);
1551 lockdep_set_class_and_name(&sk->sk_callback_lock,
1552 af_callback_keys + sk->sk_family,
1553 af_family_clock_key_strings[sk->sk_family]);
1557 * sk_clone_lock - clone a socket, and lock its clone
1558 * @sk: the socket to clone
1559 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1561 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1563 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1566 bool is_charged = true;
1568 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1569 if (newsk != NULL) {
1570 struct sk_filter *filter;
1572 sock_copy(newsk, sk);
1575 if (likely(newsk->sk_net_refcnt))
1576 get_net(sock_net(newsk));
1577 sk_node_init(&newsk->sk_node);
1578 sock_lock_init(newsk);
1579 bh_lock_sock(newsk);
1580 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1581 newsk->sk_backlog.len = 0;
1583 atomic_set(&newsk->sk_rmem_alloc, 0);
1585 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1587 atomic_set(&newsk->sk_wmem_alloc, 1);
1588 atomic_set(&newsk->sk_omem_alloc, 0);
1589 sk_init_common(newsk);
1591 newsk->sk_dst_cache = NULL;
1592 newsk->sk_dst_pending_confirm = 0;
1593 newsk->sk_wmem_queued = 0;
1594 newsk->sk_forward_alloc = 0;
1595 atomic_set(&newsk->sk_drops, 0);
1596 newsk->sk_send_head = NULL;
1597 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1599 sock_reset_flag(newsk, SOCK_DONE);
1601 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1603 /* though it's an empty new sock, the charging may fail
1604 * if sysctl_optmem_max was changed between creation of
1605 * original socket and cloning
1607 is_charged = sk_filter_charge(newsk, filter);
1609 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1610 sk_free_unlock_clone(newsk);
1614 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1617 newsk->sk_err_soft = 0;
1618 newsk->sk_priority = 0;
1619 newsk->sk_incoming_cpu = raw_smp_processor_id();
1620 atomic64_set(&newsk->sk_cookie, 0);
1622 mem_cgroup_sk_alloc(newsk);
1623 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1626 * Before updating sk_refcnt, we must commit prior changes to memory
1627 * (Documentation/RCU/rculist_nulls.txt for details)
1630 atomic_set(&newsk->sk_refcnt, 2);
1633 * Increment the counter in the same struct proto as the master
1634 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1635 * is the same as sk->sk_prot->socks, as this field was copied
1638 * This _changes_ the previous behaviour, where
1639 * tcp_create_openreq_child always was incrementing the
1640 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1641 * to be taken into account in all callers. -acme
1643 sk_refcnt_debug_inc(newsk);
1644 sk_set_socket(newsk, NULL);
1645 newsk->sk_wq = NULL;
1647 if (newsk->sk_prot->sockets_allocated)
1648 sk_sockets_allocated_inc(newsk);
1650 if (sock_needs_netstamp(sk) &&
1651 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1652 net_enable_timestamp();
1657 EXPORT_SYMBOL_GPL(sk_clone_lock);
1659 void sk_free_unlock_clone(struct sock *sk)
1661 /* It is still raw copy of parent, so invalidate
1662 * destructor and make plain sk_free() */
1663 sk->sk_destruct = NULL;
1667 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1669 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1673 sk_dst_set(sk, dst);
1674 sk->sk_route_caps = dst->dev->features;
1675 if (sk->sk_route_caps & NETIF_F_GSO)
1676 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1677 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1678 if (sk_can_gso(sk)) {
1679 if (dst->header_len) {
1680 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1682 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1683 sk->sk_gso_max_size = dst->dev->gso_max_size;
1684 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1687 sk->sk_gso_max_segs = max_segs;
1689 EXPORT_SYMBOL_GPL(sk_setup_caps);
1692 * Simple resource managers for sockets.
1697 * Write buffer destructor automatically called from kfree_skb.
1699 void sock_wfree(struct sk_buff *skb)
1701 struct sock *sk = skb->sk;
1702 unsigned int len = skb->truesize;
1704 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1706 * Keep a reference on sk_wmem_alloc, this will be released
1707 * after sk_write_space() call
1709 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1710 sk->sk_write_space(sk);
1714 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1715 * could not do because of in-flight packets
1717 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1720 EXPORT_SYMBOL(sock_wfree);
1722 /* This variant of sock_wfree() is used by TCP,
1723 * since it sets SOCK_USE_WRITE_QUEUE.
1725 void __sock_wfree(struct sk_buff *skb)
1727 struct sock *sk = skb->sk;
1729 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1733 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1738 if (unlikely(!sk_fullsock(sk))) {
1739 skb->destructor = sock_edemux;
1744 skb->destructor = sock_wfree;
1745 skb_set_hash_from_sk(skb, sk);
1747 * We used to take a refcount on sk, but following operation
1748 * is enough to guarantee sk_free() wont free this sock until
1749 * all in-flight packets are completed
1751 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1753 EXPORT_SYMBOL(skb_set_owner_w);
1755 /* This helper is used by netem, as it can hold packets in its
1756 * delay queue. We want to allow the owner socket to send more
1757 * packets, as if they were already TX completed by a typical driver.
1758 * But we also want to keep skb->sk set because some packet schedulers
1759 * rely on it (sch_fq for example). So we set skb->truesize to a small
1760 * amount (1) and decrease sk_wmem_alloc accordingly.
1762 void skb_orphan_partial(struct sk_buff *skb)
1764 /* If this skb is a TCP pure ACK or already went here,
1765 * we have nothing to do. 2 is already a very small truesize.
1767 if (skb->truesize <= 2)
1770 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1771 * so we do not completely orphan skb, but transfert all
1772 * accounted bytes but one, to avoid unexpected reorders.
1774 if (skb->destructor == sock_wfree
1776 || skb->destructor == tcp_wfree
1779 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1785 EXPORT_SYMBOL(skb_orphan_partial);
1788 * Read buffer destructor automatically called from kfree_skb.
1790 void sock_rfree(struct sk_buff *skb)
1792 struct sock *sk = skb->sk;
1793 unsigned int len = skb->truesize;
1795 atomic_sub(len, &sk->sk_rmem_alloc);
1796 sk_mem_uncharge(sk, len);
1798 EXPORT_SYMBOL(sock_rfree);
1801 * Buffer destructor for skbs that are not used directly in read or write
1802 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1804 void sock_efree(struct sk_buff *skb)
1808 EXPORT_SYMBOL(sock_efree);
1810 kuid_t sock_i_uid(struct sock *sk)
1814 read_lock_bh(&sk->sk_callback_lock);
1815 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1816 read_unlock_bh(&sk->sk_callback_lock);
1819 EXPORT_SYMBOL(sock_i_uid);
1821 unsigned long sock_i_ino(struct sock *sk)
1825 read_lock_bh(&sk->sk_callback_lock);
1826 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1827 read_unlock_bh(&sk->sk_callback_lock);
1830 EXPORT_SYMBOL(sock_i_ino);
1833 * Allocate a skb from the socket's send buffer.
1835 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1838 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1839 struct sk_buff *skb = alloc_skb(size, priority);
1841 skb_set_owner_w(skb, sk);
1847 EXPORT_SYMBOL(sock_wmalloc);
1850 * Allocate a memory block from the socket's option memory buffer.
1852 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1854 if ((unsigned int)size <= sysctl_optmem_max &&
1855 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1857 /* First do the add, to avoid the race if kmalloc
1860 atomic_add(size, &sk->sk_omem_alloc);
1861 mem = kmalloc(size, priority);
1864 atomic_sub(size, &sk->sk_omem_alloc);
1868 EXPORT_SYMBOL(sock_kmalloc);
1870 /* Free an option memory block. Note, we actually want the inline
1871 * here as this allows gcc to detect the nullify and fold away the
1872 * condition entirely.
1874 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1877 if (WARN_ON_ONCE(!mem))
1883 atomic_sub(size, &sk->sk_omem_alloc);
1886 void sock_kfree_s(struct sock *sk, void *mem, int size)
1888 __sock_kfree_s(sk, mem, size, false);
1890 EXPORT_SYMBOL(sock_kfree_s);
1892 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1894 __sock_kfree_s(sk, mem, size, true);
1896 EXPORT_SYMBOL(sock_kzfree_s);
1898 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1899 I think, these locks should be removed for datagram sockets.
1901 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1905 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1909 if (signal_pending(current))
1911 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1912 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1913 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1915 if (sk->sk_shutdown & SEND_SHUTDOWN)
1919 timeo = schedule_timeout(timeo);
1921 finish_wait(sk_sleep(sk), &wait);
1927 * Generic send/receive buffer handlers
1930 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1931 unsigned long data_len, int noblock,
1932 int *errcode, int max_page_order)
1934 struct sk_buff *skb;
1938 timeo = sock_sndtimeo(sk, noblock);
1940 err = sock_error(sk);
1945 if (sk->sk_shutdown & SEND_SHUTDOWN)
1948 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1951 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1952 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1956 if (signal_pending(current))
1958 timeo = sock_wait_for_wmem(sk, timeo);
1960 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1961 errcode, sk->sk_allocation);
1963 skb_set_owner_w(skb, sk);
1967 err = sock_intr_errno(timeo);
1972 EXPORT_SYMBOL(sock_alloc_send_pskb);
1974 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1975 int noblock, int *errcode)
1977 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1979 EXPORT_SYMBOL(sock_alloc_send_skb);
1981 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1982 struct sockcm_cookie *sockc)
1986 switch (cmsg->cmsg_type) {
1988 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1990 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1992 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1994 case SO_TIMESTAMPING:
1995 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1998 tsflags = *(u32 *)CMSG_DATA(cmsg);
1999 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2002 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2003 sockc->tsflags |= tsflags;
2005 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2007 case SCM_CREDENTIALS:
2014 EXPORT_SYMBOL(__sock_cmsg_send);
2016 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2017 struct sockcm_cookie *sockc)
2019 struct cmsghdr *cmsg;
2022 for_each_cmsghdr(cmsg, msg) {
2023 if (!CMSG_OK(msg, cmsg))
2025 if (cmsg->cmsg_level != SOL_SOCKET)
2027 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2033 EXPORT_SYMBOL(sock_cmsg_send);
2035 /* On 32bit arches, an skb frag is limited to 2^15 */
2036 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2039 * skb_page_frag_refill - check that a page_frag contains enough room
2040 * @sz: minimum size of the fragment we want to get
2041 * @pfrag: pointer to page_frag
2042 * @gfp: priority for memory allocation
2044 * Note: While this allocator tries to use high order pages, there is
2045 * no guarantee that allocations succeed. Therefore, @sz MUST be
2046 * less or equal than PAGE_SIZE.
2048 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2051 if (page_ref_count(pfrag->page) == 1) {
2055 if (pfrag->offset + sz <= pfrag->size)
2057 put_page(pfrag->page);
2061 if (SKB_FRAG_PAGE_ORDER) {
2062 /* Avoid direct reclaim but allow kswapd to wake */
2063 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2064 __GFP_COMP | __GFP_NOWARN |
2066 SKB_FRAG_PAGE_ORDER);
2067 if (likely(pfrag->page)) {
2068 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2072 pfrag->page = alloc_page(gfp);
2073 if (likely(pfrag->page)) {
2074 pfrag->size = PAGE_SIZE;
2079 EXPORT_SYMBOL(skb_page_frag_refill);
2081 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2083 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2086 sk_enter_memory_pressure(sk);
2087 sk_stream_moderate_sndbuf(sk);
2090 EXPORT_SYMBOL(sk_page_frag_refill);
2092 static void __lock_sock(struct sock *sk)
2093 __releases(&sk->sk_lock.slock)
2094 __acquires(&sk->sk_lock.slock)
2099 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2100 TASK_UNINTERRUPTIBLE);
2101 spin_unlock_bh(&sk->sk_lock.slock);
2103 spin_lock_bh(&sk->sk_lock.slock);
2104 if (!sock_owned_by_user(sk))
2107 finish_wait(&sk->sk_lock.wq, &wait);
2110 static void __release_sock(struct sock *sk)
2111 __releases(&sk->sk_lock.slock)
2112 __acquires(&sk->sk_lock.slock)
2114 struct sk_buff *skb, *next;
2116 while ((skb = sk->sk_backlog.head) != NULL) {
2117 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2119 spin_unlock_bh(&sk->sk_lock.slock);
2124 WARN_ON_ONCE(skb_dst_is_noref(skb));
2126 sk_backlog_rcv(sk, skb);
2131 } while (skb != NULL);
2133 spin_lock_bh(&sk->sk_lock.slock);
2137 * Doing the zeroing here guarantee we can not loop forever
2138 * while a wild producer attempts to flood us.
2140 sk->sk_backlog.len = 0;
2143 void __sk_flush_backlog(struct sock *sk)
2145 spin_lock_bh(&sk->sk_lock.slock);
2147 spin_unlock_bh(&sk->sk_lock.slock);
2151 * sk_wait_data - wait for data to arrive at sk_receive_queue
2152 * @sk: sock to wait on
2153 * @timeo: for how long
2154 * @skb: last skb seen on sk_receive_queue
2156 * Now socket state including sk->sk_err is changed only under lock,
2157 * hence we may omit checks after joining wait queue.
2158 * We check receive queue before schedule() only as optimization;
2159 * it is very likely that release_sock() added new data.
2161 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2163 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2166 add_wait_queue(sk_sleep(sk), &wait);
2167 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2168 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2169 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2170 remove_wait_queue(sk_sleep(sk), &wait);
2173 EXPORT_SYMBOL(sk_wait_data);
2176 * __sk_mem_raise_allocated - increase memory_allocated
2178 * @size: memory size to allocate
2179 * @amt: pages to allocate
2180 * @kind: allocation type
2182 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2184 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2186 struct proto *prot = sk->sk_prot;
2187 long allocated = sk_memory_allocated_add(sk, amt);
2189 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2190 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2191 goto suppress_allocation;
2194 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2195 sk_leave_memory_pressure(sk);
2199 /* Under pressure. */
2200 if (allocated > sk_prot_mem_limits(sk, 1))
2201 sk_enter_memory_pressure(sk);
2203 /* Over hard limit. */
2204 if (allocated > sk_prot_mem_limits(sk, 2))
2205 goto suppress_allocation;
2207 /* guarantee minimum buffer size under pressure */
2208 if (kind == SK_MEM_RECV) {
2209 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2212 } else { /* SK_MEM_SEND */
2213 if (sk->sk_type == SOCK_STREAM) {
2214 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2216 } else if (atomic_read(&sk->sk_wmem_alloc) <
2217 prot->sysctl_wmem[0])
2221 if (sk_has_memory_pressure(sk)) {
2224 if (!sk_under_memory_pressure(sk))
2226 alloc = sk_sockets_allocated_read_positive(sk);
2227 if (sk_prot_mem_limits(sk, 2) > alloc *
2228 sk_mem_pages(sk->sk_wmem_queued +
2229 atomic_read(&sk->sk_rmem_alloc) +
2230 sk->sk_forward_alloc))
2234 suppress_allocation:
2236 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2237 sk_stream_moderate_sndbuf(sk);
2239 /* Fail only if socket is _under_ its sndbuf.
2240 * In this case we cannot block, so that we have to fail.
2242 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2246 trace_sock_exceed_buf_limit(sk, prot, allocated);
2248 sk_memory_allocated_sub(sk, amt);
2250 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2251 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2255 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2258 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2260 * @size: memory size to allocate
2261 * @kind: allocation type
2263 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2264 * rmem allocation. This function assumes that protocols which have
2265 * memory_pressure use sk_wmem_queued as write buffer accounting.
2267 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2269 int ret, amt = sk_mem_pages(size);
2271 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2272 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2274 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2277 EXPORT_SYMBOL(__sk_mem_schedule);
2280 * __sk_mem_reduce_allocated - reclaim memory_allocated
2282 * @amount: number of quanta
2284 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2286 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2288 sk_memory_allocated_sub(sk, amount);
2290 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2291 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2293 if (sk_under_memory_pressure(sk) &&
2294 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2295 sk_leave_memory_pressure(sk);
2297 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2300 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2302 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2304 void __sk_mem_reclaim(struct sock *sk, int amount)
2306 amount >>= SK_MEM_QUANTUM_SHIFT;
2307 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2308 __sk_mem_reduce_allocated(sk, amount);
2310 EXPORT_SYMBOL(__sk_mem_reclaim);
2312 int sk_set_peek_off(struct sock *sk, int val)
2317 sk->sk_peek_off = val;
2320 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2323 * Set of default routines for initialising struct proto_ops when
2324 * the protocol does not support a particular function. In certain
2325 * cases where it makes no sense for a protocol to have a "do nothing"
2326 * function, some default processing is provided.
2329 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2333 EXPORT_SYMBOL(sock_no_bind);
2335 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2340 EXPORT_SYMBOL(sock_no_connect);
2342 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2346 EXPORT_SYMBOL(sock_no_socketpair);
2348 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2353 EXPORT_SYMBOL(sock_no_accept);
2355 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2360 EXPORT_SYMBOL(sock_no_getname);
2362 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2366 EXPORT_SYMBOL(sock_no_poll);
2368 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2372 EXPORT_SYMBOL(sock_no_ioctl);
2374 int sock_no_listen(struct socket *sock, int backlog)
2378 EXPORT_SYMBOL(sock_no_listen);
2380 int sock_no_shutdown(struct socket *sock, int how)
2384 EXPORT_SYMBOL(sock_no_shutdown);
2386 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2387 char __user *optval, unsigned int optlen)
2391 EXPORT_SYMBOL(sock_no_setsockopt);
2393 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2394 char __user *optval, int __user *optlen)
2398 EXPORT_SYMBOL(sock_no_getsockopt);
2400 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2404 EXPORT_SYMBOL(sock_no_sendmsg);
2406 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2411 EXPORT_SYMBOL(sock_no_recvmsg);
2413 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2415 /* Mirror missing mmap method error code */
2418 EXPORT_SYMBOL(sock_no_mmap);
2420 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2423 struct msghdr msg = {.msg_flags = flags};
2425 char *kaddr = kmap(page);
2426 iov.iov_base = kaddr + offset;
2428 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2432 EXPORT_SYMBOL(sock_no_sendpage);
2435 * Default Socket Callbacks
2438 static void sock_def_wakeup(struct sock *sk)
2440 struct socket_wq *wq;
2443 wq = rcu_dereference(sk->sk_wq);
2444 if (skwq_has_sleeper(wq))
2445 wake_up_interruptible_all(&wq->wait);
2449 static void sock_def_error_report(struct sock *sk)
2451 struct socket_wq *wq;
2454 wq = rcu_dereference(sk->sk_wq);
2455 if (skwq_has_sleeper(wq))
2456 wake_up_interruptible_poll(&wq->wait, POLLERR);
2457 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2461 static void sock_def_readable(struct sock *sk)
2463 struct socket_wq *wq;
2466 wq = rcu_dereference(sk->sk_wq);
2467 if (skwq_has_sleeper(wq))
2468 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2469 POLLRDNORM | POLLRDBAND);
2470 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2474 static void sock_def_write_space(struct sock *sk)
2476 struct socket_wq *wq;
2480 /* Do not wake up a writer until he can make "significant"
2483 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2484 wq = rcu_dereference(sk->sk_wq);
2485 if (skwq_has_sleeper(wq))
2486 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2487 POLLWRNORM | POLLWRBAND);
2489 /* Should agree with poll, otherwise some programs break */
2490 if (sock_writeable(sk))
2491 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2497 static void sock_def_destruct(struct sock *sk)
2501 void sk_send_sigurg(struct sock *sk)
2503 if (sk->sk_socket && sk->sk_socket->file)
2504 if (send_sigurg(&sk->sk_socket->file->f_owner))
2505 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2507 EXPORT_SYMBOL(sk_send_sigurg);
2509 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2510 unsigned long expires)
2512 if (!mod_timer(timer, expires))
2515 EXPORT_SYMBOL(sk_reset_timer);
2517 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2519 if (del_timer(timer))
2522 EXPORT_SYMBOL(sk_stop_timer);
2524 void sock_init_data(struct socket *sock, struct sock *sk)
2527 sk->sk_send_head = NULL;
2529 init_timer(&sk->sk_timer);
2531 sk->sk_allocation = GFP_KERNEL;
2532 sk->sk_rcvbuf = sysctl_rmem_default;
2533 sk->sk_sndbuf = sysctl_wmem_default;
2534 sk->sk_state = TCP_CLOSE;
2535 sk_set_socket(sk, sock);
2537 sock_set_flag(sk, SOCK_ZAPPED);
2540 sk->sk_type = sock->type;
2541 sk->sk_wq = sock->wq;
2543 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2546 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2549 rwlock_init(&sk->sk_callback_lock);
2550 if (sk->sk_kern_sock)
2551 lockdep_set_class_and_name(
2552 &sk->sk_callback_lock,
2553 af_kern_callback_keys + sk->sk_family,
2554 af_family_kern_clock_key_strings[sk->sk_family]);
2556 lockdep_set_class_and_name(
2557 &sk->sk_callback_lock,
2558 af_callback_keys + sk->sk_family,
2559 af_family_clock_key_strings[sk->sk_family]);
2561 sk->sk_state_change = sock_def_wakeup;
2562 sk->sk_data_ready = sock_def_readable;
2563 sk->sk_write_space = sock_def_write_space;
2564 sk->sk_error_report = sock_def_error_report;
2565 sk->sk_destruct = sock_def_destruct;
2567 sk->sk_frag.page = NULL;
2568 sk->sk_frag.offset = 0;
2569 sk->sk_peek_off = -1;
2571 sk->sk_peer_pid = NULL;
2572 sk->sk_peer_cred = NULL;
2573 sk->sk_write_pending = 0;
2574 sk->sk_rcvlowat = 1;
2575 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2576 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2578 sk->sk_stamp = ktime_set(-1L, 0);
2580 #ifdef CONFIG_NET_RX_BUSY_POLL
2582 sk->sk_ll_usec = sysctl_net_busy_read;
2585 sk->sk_max_pacing_rate = ~0U;
2586 sk->sk_pacing_rate = ~0U;
2587 sk->sk_incoming_cpu = -1;
2589 * Before updating sk_refcnt, we must commit prior changes to memory
2590 * (Documentation/RCU/rculist_nulls.txt for details)
2593 atomic_set(&sk->sk_refcnt, 1);
2594 atomic_set(&sk->sk_drops, 0);
2596 EXPORT_SYMBOL(sock_init_data);
2598 void lock_sock_nested(struct sock *sk, int subclass)
2601 spin_lock_bh(&sk->sk_lock.slock);
2602 if (sk->sk_lock.owned)
2604 sk->sk_lock.owned = 1;
2605 spin_unlock(&sk->sk_lock.slock);
2607 * The sk_lock has mutex_lock() semantics here:
2609 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2612 EXPORT_SYMBOL(lock_sock_nested);
2614 void release_sock(struct sock *sk)
2616 spin_lock_bh(&sk->sk_lock.slock);
2617 if (sk->sk_backlog.tail)
2620 /* Warning : release_cb() might need to release sk ownership,
2621 * ie call sock_release_ownership(sk) before us.
2623 if (sk->sk_prot->release_cb)
2624 sk->sk_prot->release_cb(sk);
2626 sock_release_ownership(sk);
2627 if (waitqueue_active(&sk->sk_lock.wq))
2628 wake_up(&sk->sk_lock.wq);
2629 spin_unlock_bh(&sk->sk_lock.slock);
2631 EXPORT_SYMBOL(release_sock);
2634 * lock_sock_fast - fast version of lock_sock
2637 * This version should be used for very small section, where process wont block
2638 * return false if fast path is taken
2639 * sk_lock.slock locked, owned = 0, BH disabled
2640 * return true if slow path is taken
2641 * sk_lock.slock unlocked, owned = 1, BH enabled
2643 bool lock_sock_fast(struct sock *sk)
2646 spin_lock_bh(&sk->sk_lock.slock);
2648 if (!sk->sk_lock.owned)
2650 * Note : We must disable BH
2655 sk->sk_lock.owned = 1;
2656 spin_unlock(&sk->sk_lock.slock);
2658 * The sk_lock has mutex_lock() semantics here:
2660 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2664 EXPORT_SYMBOL(lock_sock_fast);
2666 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2669 if (!sock_flag(sk, SOCK_TIMESTAMP))
2670 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2671 tv = ktime_to_timeval(sk->sk_stamp);
2672 if (tv.tv_sec == -1)
2674 if (tv.tv_sec == 0) {
2675 sk->sk_stamp = ktime_get_real();
2676 tv = ktime_to_timeval(sk->sk_stamp);
2678 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2680 EXPORT_SYMBOL(sock_get_timestamp);
2682 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2685 if (!sock_flag(sk, SOCK_TIMESTAMP))
2686 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2687 ts = ktime_to_timespec(sk->sk_stamp);
2688 if (ts.tv_sec == -1)
2690 if (ts.tv_sec == 0) {
2691 sk->sk_stamp = ktime_get_real();
2692 ts = ktime_to_timespec(sk->sk_stamp);
2694 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2696 EXPORT_SYMBOL(sock_get_timestampns);
2698 void sock_enable_timestamp(struct sock *sk, int flag)
2700 if (!sock_flag(sk, flag)) {
2701 unsigned long previous_flags = sk->sk_flags;
2703 sock_set_flag(sk, flag);
2705 * we just set one of the two flags which require net
2706 * time stamping, but time stamping might have been on
2707 * already because of the other one
2709 if (sock_needs_netstamp(sk) &&
2710 !(previous_flags & SK_FLAGS_TIMESTAMP))
2711 net_enable_timestamp();
2715 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2716 int level, int type)
2718 struct sock_exterr_skb *serr;
2719 struct sk_buff *skb;
2723 skb = sock_dequeue_err_skb(sk);
2729 msg->msg_flags |= MSG_TRUNC;
2732 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2736 sock_recv_timestamp(msg, sk, skb);
2738 serr = SKB_EXT_ERR(skb);
2739 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2741 msg->msg_flags |= MSG_ERRQUEUE;
2749 EXPORT_SYMBOL(sock_recv_errqueue);
2752 * Get a socket option on an socket.
2754 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2755 * asynchronous errors should be reported by getsockopt. We assume
2756 * this means if you specify SO_ERROR (otherwise whats the point of it).
2758 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2759 char __user *optval, int __user *optlen)
2761 struct sock *sk = sock->sk;
2763 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2765 EXPORT_SYMBOL(sock_common_getsockopt);
2767 #ifdef CONFIG_COMPAT
2768 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2769 char __user *optval, int __user *optlen)
2771 struct sock *sk = sock->sk;
2773 if (sk->sk_prot->compat_getsockopt != NULL)
2774 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2776 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2778 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2781 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2784 struct sock *sk = sock->sk;
2788 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2789 flags & ~MSG_DONTWAIT, &addr_len);
2791 msg->msg_namelen = addr_len;
2794 EXPORT_SYMBOL(sock_common_recvmsg);
2797 * Set socket options on an inet socket.
2799 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2800 char __user *optval, unsigned int optlen)
2802 struct sock *sk = sock->sk;
2804 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2806 EXPORT_SYMBOL(sock_common_setsockopt);
2808 #ifdef CONFIG_COMPAT
2809 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2810 char __user *optval, unsigned int optlen)
2812 struct sock *sk = sock->sk;
2814 if (sk->sk_prot->compat_setsockopt != NULL)
2815 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2817 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2819 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2822 void sk_common_release(struct sock *sk)
2824 if (sk->sk_prot->destroy)
2825 sk->sk_prot->destroy(sk);
2828 * Observation: when sock_common_release is called, processes have
2829 * no access to socket. But net still has.
2830 * Step one, detach it from networking:
2832 * A. Remove from hash tables.
2835 sk->sk_prot->unhash(sk);
2838 * In this point socket cannot receive new packets, but it is possible
2839 * that some packets are in flight because some CPU runs receiver and
2840 * did hash table lookup before we unhashed socket. They will achieve
2841 * receive queue and will be purged by socket destructor.
2843 * Also we still have packets pending on receive queue and probably,
2844 * our own packets waiting in device queues. sock_destroy will drain
2845 * receive queue, but transmitted packets will delay socket destruction
2846 * until the last reference will be released.
2851 xfrm_sk_free_policy(sk);
2853 sk_refcnt_debug_release(sk);
2855 if (sk->sk_frag.page) {
2856 put_page(sk->sk_frag.page);
2857 sk->sk_frag.page = NULL;
2862 EXPORT_SYMBOL(sk_common_release);
2864 #ifdef CONFIG_PROC_FS
2865 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2867 int val[PROTO_INUSE_NR];
2870 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2872 #ifdef CONFIG_NET_NS
2873 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2875 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2877 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2879 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2881 int cpu, idx = prot->inuse_idx;
2884 for_each_possible_cpu(cpu)
2885 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2887 return res >= 0 ? res : 0;
2889 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2891 static int __net_init sock_inuse_init_net(struct net *net)
2893 net->core.inuse = alloc_percpu(struct prot_inuse);
2894 return net->core.inuse ? 0 : -ENOMEM;
2897 static void __net_exit sock_inuse_exit_net(struct net *net)
2899 free_percpu(net->core.inuse);
2902 static struct pernet_operations net_inuse_ops = {
2903 .init = sock_inuse_init_net,
2904 .exit = sock_inuse_exit_net,
2907 static __init int net_inuse_init(void)
2909 if (register_pernet_subsys(&net_inuse_ops))
2910 panic("Cannot initialize net inuse counters");
2915 core_initcall(net_inuse_init);
2917 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2919 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2921 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2923 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2925 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2927 int cpu, idx = prot->inuse_idx;
2930 for_each_possible_cpu(cpu)
2931 res += per_cpu(prot_inuse, cpu).val[idx];
2933 return res >= 0 ? res : 0;
2935 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2938 static void assign_proto_idx(struct proto *prot)
2940 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2942 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2943 pr_err("PROTO_INUSE_NR exhausted\n");
2947 set_bit(prot->inuse_idx, proto_inuse_idx);
2950 static void release_proto_idx(struct proto *prot)
2952 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2953 clear_bit(prot->inuse_idx, proto_inuse_idx);
2956 static inline void assign_proto_idx(struct proto *prot)
2960 static inline void release_proto_idx(struct proto *prot)
2965 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2969 kfree(rsk_prot->slab_name);
2970 rsk_prot->slab_name = NULL;
2971 kmem_cache_destroy(rsk_prot->slab);
2972 rsk_prot->slab = NULL;
2975 static int req_prot_init(const struct proto *prot)
2977 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2982 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2984 if (!rsk_prot->slab_name)
2987 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2988 rsk_prot->obj_size, 0,
2989 prot->slab_flags, NULL);
2991 if (!rsk_prot->slab) {
2992 pr_crit("%s: Can't create request sock SLAB cache!\n",
2999 int proto_register(struct proto *prot, int alloc_slab)
3002 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
3003 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3006 if (prot->slab == NULL) {
3007 pr_crit("%s: Can't create sock SLAB cache!\n",
3012 if (req_prot_init(prot))
3013 goto out_free_request_sock_slab;
3015 if (prot->twsk_prot != NULL) {
3016 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3018 if (prot->twsk_prot->twsk_slab_name == NULL)
3019 goto out_free_request_sock_slab;
3021 prot->twsk_prot->twsk_slab =
3022 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3023 prot->twsk_prot->twsk_obj_size,
3027 if (prot->twsk_prot->twsk_slab == NULL)
3028 goto out_free_timewait_sock_slab_name;
3032 mutex_lock(&proto_list_mutex);
3033 list_add(&prot->node, &proto_list);
3034 assign_proto_idx(prot);
3035 mutex_unlock(&proto_list_mutex);
3038 out_free_timewait_sock_slab_name:
3039 kfree(prot->twsk_prot->twsk_slab_name);
3040 out_free_request_sock_slab:
3041 req_prot_cleanup(prot->rsk_prot);
3043 kmem_cache_destroy(prot->slab);
3048 EXPORT_SYMBOL(proto_register);
3050 void proto_unregister(struct proto *prot)
3052 mutex_lock(&proto_list_mutex);
3053 release_proto_idx(prot);
3054 list_del(&prot->node);
3055 mutex_unlock(&proto_list_mutex);
3057 kmem_cache_destroy(prot->slab);
3060 req_prot_cleanup(prot->rsk_prot);
3062 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3063 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3064 kfree(prot->twsk_prot->twsk_slab_name);
3065 prot->twsk_prot->twsk_slab = NULL;
3068 EXPORT_SYMBOL(proto_unregister);
3070 #ifdef CONFIG_PROC_FS
3071 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3072 __acquires(proto_list_mutex)
3074 mutex_lock(&proto_list_mutex);
3075 return seq_list_start_head(&proto_list, *pos);
3078 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3080 return seq_list_next(v, &proto_list, pos);
3083 static void proto_seq_stop(struct seq_file *seq, void *v)
3084 __releases(proto_list_mutex)
3086 mutex_unlock(&proto_list_mutex);
3089 static char proto_method_implemented(const void *method)
3091 return method == NULL ? 'n' : 'y';
3093 static long sock_prot_memory_allocated(struct proto *proto)
3095 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3098 static char *sock_prot_memory_pressure(struct proto *proto)
3100 return proto->memory_pressure != NULL ?
3101 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3104 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3107 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3108 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3111 sock_prot_inuse_get(seq_file_net(seq), proto),
3112 sock_prot_memory_allocated(proto),
3113 sock_prot_memory_pressure(proto),
3115 proto->slab == NULL ? "no" : "yes",
3116 module_name(proto->owner),
3117 proto_method_implemented(proto->close),
3118 proto_method_implemented(proto->connect),
3119 proto_method_implemented(proto->disconnect),
3120 proto_method_implemented(proto->accept),
3121 proto_method_implemented(proto->ioctl),
3122 proto_method_implemented(proto->init),
3123 proto_method_implemented(proto->destroy),
3124 proto_method_implemented(proto->shutdown),
3125 proto_method_implemented(proto->setsockopt),
3126 proto_method_implemented(proto->getsockopt),
3127 proto_method_implemented(proto->sendmsg),
3128 proto_method_implemented(proto->recvmsg),
3129 proto_method_implemented(proto->sendpage),
3130 proto_method_implemented(proto->bind),
3131 proto_method_implemented(proto->backlog_rcv),
3132 proto_method_implemented(proto->hash),
3133 proto_method_implemented(proto->unhash),
3134 proto_method_implemented(proto->get_port),
3135 proto_method_implemented(proto->enter_memory_pressure));
3138 static int proto_seq_show(struct seq_file *seq, void *v)
3140 if (v == &proto_list)
3141 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3150 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3152 proto_seq_printf(seq, list_entry(v, struct proto, node));
3156 static const struct seq_operations proto_seq_ops = {
3157 .start = proto_seq_start,
3158 .next = proto_seq_next,
3159 .stop = proto_seq_stop,
3160 .show = proto_seq_show,
3163 static int proto_seq_open(struct inode *inode, struct file *file)
3165 return seq_open_net(inode, file, &proto_seq_ops,
3166 sizeof(struct seq_net_private));
3169 static const struct file_operations proto_seq_fops = {
3170 .owner = THIS_MODULE,
3171 .open = proto_seq_open,
3173 .llseek = seq_lseek,
3174 .release = seq_release_net,
3177 static __net_init int proto_init_net(struct net *net)
3179 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3185 static __net_exit void proto_exit_net(struct net *net)
3187 remove_proc_entry("protocols", net->proc_net);
3191 static __net_initdata struct pernet_operations proto_net_ops = {
3192 .init = proto_init_net,
3193 .exit = proto_exit_net,
3196 static int __init proto_init(void)
3198 return register_pernet_subsys(&proto_net_ops);
3201 subsys_initcall(proto_init);
3203 #endif /* PROC_FS */