2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/sched/mm.h>
106 #include <linux/timer.h>
107 #include <linux/string.h>
108 #include <linux/sockios.h>
109 #include <linux/net.h>
110 #include <linux/mm.h>
111 #include <linux/slab.h>
112 #include <linux/interrupt.h>
113 #include <linux/poll.h>
114 #include <linux/tcp.h>
115 #include <linux/init.h>
116 #include <linux/highmem.h>
117 #include <linux/user_namespace.h>
118 #include <linux/static_key.h>
119 #include <linux/memcontrol.h>
120 #include <linux/prefetch.h>
122 #include <linux/uaccess.h>
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
140 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
149 * sk_ns_capable - General socket capability test
150 * @sk: Socket to use a capability on or through
151 * @user_ns: The user namespace of the capability to use
152 * @cap: The capability to use
154 * Test to see if the opener of the socket had when the socket was
155 * created and the current process has the capability @cap in the user
156 * namespace @user_ns.
158 bool sk_ns_capable(const struct sock *sk,
159 struct user_namespace *user_ns, int cap)
161 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162 ns_capable(user_ns, cap);
164 EXPORT_SYMBOL(sk_ns_capable);
167 * sk_capable - Socket global capability test
168 * @sk: Socket to use a capability on or through
169 * @cap: The global capability to use
171 * Test to see if the opener of the socket had when the socket was
172 * created and the current process has the capability @cap in all user
175 bool sk_capable(const struct sock *sk, int cap)
177 return sk_ns_capable(sk, &init_user_ns, cap);
179 EXPORT_SYMBOL(sk_capable);
182 * sk_net_capable - Network namespace socket capability test
183 * @sk: Socket to use a capability on or through
184 * @cap: The capability to use
186 * Test to see if the opener of the socket had when the socket was created
187 * and the current process has the capability @cap over the network namespace
188 * the socket is a member of.
190 bool sk_net_capable(const struct sock *sk, int cap)
192 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 EXPORT_SYMBOL(sk_net_capable);
197 * Each address family might have different locking rules, so we have
198 * one slock key per address family and separate keys for internal and
201 static struct lock_class_key af_family_keys[AF_MAX];
202 static struct lock_class_key af_family_kern_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
207 * Make lock validator output more readable. (we pre-construct these
208 * strings build-time, so that runtime initialization of socket
212 #define _sock_locks(x) \
213 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
214 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
215 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
216 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
217 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
218 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
219 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
220 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
221 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
222 x "27" , x "28" , x "AF_CAN" , \
223 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
224 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
225 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
226 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
227 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
229 static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
250 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
251 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
252 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
253 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
254 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
255 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
256 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
257 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
258 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
259 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
260 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
261 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
262 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
263 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
265 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
266 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
267 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
268 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
269 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
270 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
271 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
272 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
273 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
274 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
275 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
276 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
277 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
278 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
279 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
280 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
282 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
283 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
284 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
285 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
286 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
287 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
288 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
289 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
290 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
291 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
292 "elock-27" , "elock-28" , "elock-AF_CAN" ,
293 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
294 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
295 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
296 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
297 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
301 * sk_callback_lock and sk queues locking rules are per-address-family,
302 * so split the lock classes by using a per-AF key:
304 static struct lock_class_key af_callback_keys[AF_MAX];
305 static struct lock_class_key af_rlock_keys[AF_MAX];
306 static struct lock_class_key af_wlock_keys[AF_MAX];
307 static struct lock_class_key af_elock_keys[AF_MAX];
308 static struct lock_class_key af_kern_callback_keys[AF_MAX];
310 /* Take into consideration the size of the struct sk_buff overhead in the
311 * determination of these values, since that is non-constant across
312 * platforms. This makes socket queueing behavior and performance
313 * not depend upon such differences.
315 #define _SK_MEM_PACKETS 256
316 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
317 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
318 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
320 /* Run time adjustable parameters. */
321 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
322 EXPORT_SYMBOL(sysctl_wmem_max);
323 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
324 EXPORT_SYMBOL(sysctl_rmem_max);
325 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
326 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
328 /* Maximal space eaten by iovec or ancillary data plus some space */
329 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
330 EXPORT_SYMBOL(sysctl_optmem_max);
332 int sysctl_tstamp_allow_data __read_mostly = 1;
334 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
335 EXPORT_SYMBOL_GPL(memalloc_socks);
338 * sk_set_memalloc - sets %SOCK_MEMALLOC
339 * @sk: socket to set it on
341 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
342 * It's the responsibility of the admin to adjust min_free_kbytes
343 * to meet the requirements
345 void sk_set_memalloc(struct sock *sk)
347 sock_set_flag(sk, SOCK_MEMALLOC);
348 sk->sk_allocation |= __GFP_MEMALLOC;
349 static_key_slow_inc(&memalloc_socks);
351 EXPORT_SYMBOL_GPL(sk_set_memalloc);
353 void sk_clear_memalloc(struct sock *sk)
355 sock_reset_flag(sk, SOCK_MEMALLOC);
356 sk->sk_allocation &= ~__GFP_MEMALLOC;
357 static_key_slow_dec(&memalloc_socks);
360 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
361 * progress of swapping. SOCK_MEMALLOC may be cleared while
362 * it has rmem allocations due to the last swapfile being deactivated
363 * but there is a risk that the socket is unusable due to exceeding
364 * the rmem limits. Reclaim the reserves and obey rmem limits again.
368 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
370 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
373 unsigned int noreclaim_flag;
375 /* these should have been dropped before queueing */
376 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
378 noreclaim_flag = memalloc_noreclaim_save();
379 ret = sk->sk_backlog_rcv(sk, skb);
380 memalloc_noreclaim_restore(noreclaim_flag);
384 EXPORT_SYMBOL(__sk_backlog_rcv);
386 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
390 if (optlen < sizeof(tv))
392 if (copy_from_user(&tv, optval, sizeof(tv)))
394 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
398 static int warned __read_mostly;
401 if (warned < 10 && net_ratelimit()) {
403 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404 __func__, current->comm, task_pid_nr(current));
408 *timeo_p = MAX_SCHEDULE_TIMEOUT;
409 if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
412 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
416 static void sock_warn_obsolete_bsdism(const char *name)
419 static char warncomm[TASK_COMM_LEN];
420 if (strcmp(warncomm, current->comm) && warned < 5) {
421 strcpy(warncomm, current->comm);
422 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
428 static bool sock_needs_netstamp(const struct sock *sk)
430 switch (sk->sk_family) {
439 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
441 if (sk->sk_flags & flags) {
442 sk->sk_flags &= ~flags;
443 if (sock_needs_netstamp(sk) &&
444 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
445 net_disable_timestamp();
450 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
453 struct sk_buff_head *list = &sk->sk_receive_queue;
455 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
456 atomic_inc(&sk->sk_drops);
457 trace_sock_rcvqueue_full(sk, skb);
461 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
462 atomic_inc(&sk->sk_drops);
467 skb_set_owner_r(skb, sk);
469 /* we escape from rcu protected region, make sure we dont leak
474 spin_lock_irqsave(&list->lock, flags);
475 sock_skb_set_dropcount(sk, skb);
476 __skb_queue_tail(list, skb);
477 spin_unlock_irqrestore(&list->lock, flags);
479 if (!sock_flag(sk, SOCK_DEAD))
480 sk->sk_data_ready(sk);
483 EXPORT_SYMBOL(__sock_queue_rcv_skb);
485 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
489 err = sk_filter(sk, skb);
493 return __sock_queue_rcv_skb(sk, skb);
495 EXPORT_SYMBOL(sock_queue_rcv_skb);
497 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
498 const int nested, unsigned int trim_cap, bool refcounted)
500 int rc = NET_RX_SUCCESS;
502 if (sk_filter_trim_cap(sk, skb, trim_cap))
503 goto discard_and_relse;
507 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
508 atomic_inc(&sk->sk_drops);
509 goto discard_and_relse;
512 bh_lock_sock_nested(sk);
515 if (!sock_owned_by_user(sk)) {
517 * trylock + unlock semantics:
519 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
521 rc = sk_backlog_rcv(sk, skb);
523 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
524 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
526 atomic_inc(&sk->sk_drops);
527 goto discard_and_relse;
539 EXPORT_SYMBOL(__sk_receive_skb);
541 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
543 struct dst_entry *dst = __sk_dst_get(sk);
545 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
546 sk_tx_queue_clear(sk);
547 sk->sk_dst_pending_confirm = 0;
548 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
555 EXPORT_SYMBOL(__sk_dst_check);
557 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
559 struct dst_entry *dst = sk_dst_get(sk);
561 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
569 EXPORT_SYMBOL(sk_dst_check);
571 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
574 int ret = -ENOPROTOOPT;
575 #ifdef CONFIG_NETDEVICES
576 struct net *net = sock_net(sk);
577 char devname[IFNAMSIZ];
582 if (!ns_capable(net->user_ns, CAP_NET_RAW))
589 /* Bind this socket to a particular device like "eth0",
590 * as specified in the passed interface name. If the
591 * name is "" or the option length is zero the socket
594 if (optlen > IFNAMSIZ - 1)
595 optlen = IFNAMSIZ - 1;
596 memset(devname, 0, sizeof(devname));
599 if (copy_from_user(devname, optval, optlen))
603 if (devname[0] != '\0') {
604 struct net_device *dev;
607 dev = dev_get_by_name_rcu(net, devname);
609 index = dev->ifindex;
617 sk->sk_bound_dev_if = index;
629 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
630 int __user *optlen, int len)
632 int ret = -ENOPROTOOPT;
633 #ifdef CONFIG_NETDEVICES
634 struct net *net = sock_net(sk);
635 char devname[IFNAMSIZ];
637 if (sk->sk_bound_dev_if == 0) {
646 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
650 len = strlen(devname) + 1;
653 if (copy_to_user(optval, devname, len))
658 if (put_user(len, optlen))
669 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
672 sock_set_flag(sk, bit);
674 sock_reset_flag(sk, bit);
677 bool sk_mc_loop(struct sock *sk)
679 if (dev_recursion_level())
683 switch (sk->sk_family) {
685 return inet_sk(sk)->mc_loop;
686 #if IS_ENABLED(CONFIG_IPV6)
688 return inet6_sk(sk)->mc_loop;
694 EXPORT_SYMBOL(sk_mc_loop);
697 * This is meant for all protocols to use and covers goings on
698 * at the socket level. Everything here is generic.
701 int sock_setsockopt(struct socket *sock, int level, int optname,
702 char __user *optval, unsigned int optlen)
704 struct sock *sk = sock->sk;
711 * Options without arguments
714 if (optname == SO_BINDTODEVICE)
715 return sock_setbindtodevice(sk, optval, optlen);
717 if (optlen < sizeof(int))
720 if (get_user(val, (int __user *)optval))
723 valbool = val ? 1 : 0;
729 if (val && !capable(CAP_NET_ADMIN))
732 sock_valbool_flag(sk, SOCK_DBG, valbool);
735 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
738 sk->sk_reuseport = valbool;
747 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
750 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
753 /* Don't error on this BSD doesn't and if you think
754 * about it this is right. Otherwise apps have to
755 * play 'guess the biggest size' games. RCVBUF/SNDBUF
756 * are treated in BSD as hints
758 val = min_t(u32, val, sysctl_wmem_max);
760 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
761 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
762 /* Wake up sending tasks if we upped the value. */
763 sk->sk_write_space(sk);
767 if (!capable(CAP_NET_ADMIN)) {
774 /* Don't error on this BSD doesn't and if you think
775 * about it this is right. Otherwise apps have to
776 * play 'guess the biggest size' games. RCVBUF/SNDBUF
777 * are treated in BSD as hints
779 val = min_t(u32, val, sysctl_rmem_max);
781 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
783 * We double it on the way in to account for
784 * "struct sk_buff" etc. overhead. Applications
785 * assume that the SO_RCVBUF setting they make will
786 * allow that much actual data to be received on that
789 * Applications are unaware that "struct sk_buff" and
790 * other overheads allocate from the receive buffer
791 * during socket buffer allocation.
793 * And after considering the possible alternatives,
794 * returning the value we actually used in getsockopt
795 * is the most desirable behavior.
797 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
801 if (!capable(CAP_NET_ADMIN)) {
808 if (sk->sk_prot->keepalive)
809 sk->sk_prot->keepalive(sk, valbool);
810 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
814 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
818 sk->sk_no_check_tx = valbool;
822 if ((val >= 0 && val <= 6) ||
823 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
824 sk->sk_priority = val;
830 if (optlen < sizeof(ling)) {
831 ret = -EINVAL; /* 1003.1g */
834 if (copy_from_user(&ling, optval, sizeof(ling))) {
839 sock_reset_flag(sk, SOCK_LINGER);
841 #if (BITS_PER_LONG == 32)
842 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
843 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
846 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
847 sock_set_flag(sk, SOCK_LINGER);
852 sock_warn_obsolete_bsdism("setsockopt");
857 set_bit(SOCK_PASSCRED, &sock->flags);
859 clear_bit(SOCK_PASSCRED, &sock->flags);
865 if (optname == SO_TIMESTAMP)
866 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
868 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
869 sock_set_flag(sk, SOCK_RCVTSTAMP);
870 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
872 sock_reset_flag(sk, SOCK_RCVTSTAMP);
873 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
877 case SO_TIMESTAMPING:
878 if (val & ~SOF_TIMESTAMPING_MASK) {
883 if (val & SOF_TIMESTAMPING_OPT_ID &&
884 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
885 if (sk->sk_protocol == IPPROTO_TCP &&
886 sk->sk_type == SOCK_STREAM) {
887 if ((1 << sk->sk_state) &
888 (TCPF_CLOSE | TCPF_LISTEN)) {
892 sk->sk_tskey = tcp_sk(sk)->snd_una;
898 if (val & SOF_TIMESTAMPING_OPT_STATS &&
899 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
904 sk->sk_tsflags = val;
905 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
906 sock_enable_timestamp(sk,
907 SOCK_TIMESTAMPING_RX_SOFTWARE);
909 sock_disable_timestamp(sk,
910 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
916 sk->sk_rcvlowat = val ? : 1;
920 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
924 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
927 case SO_ATTACH_FILTER:
929 if (optlen == sizeof(struct sock_fprog)) {
930 struct sock_fprog fprog;
933 if (copy_from_user(&fprog, optval, sizeof(fprog)))
936 ret = sk_attach_filter(&fprog, sk);
942 if (optlen == sizeof(u32)) {
946 if (copy_from_user(&ufd, optval, sizeof(ufd)))
949 ret = sk_attach_bpf(ufd, sk);
953 case SO_ATTACH_REUSEPORT_CBPF:
955 if (optlen == sizeof(struct sock_fprog)) {
956 struct sock_fprog fprog;
959 if (copy_from_user(&fprog, optval, sizeof(fprog)))
962 ret = sk_reuseport_attach_filter(&fprog, sk);
966 case SO_ATTACH_REUSEPORT_EBPF:
968 if (optlen == sizeof(u32)) {
972 if (copy_from_user(&ufd, optval, sizeof(ufd)))
975 ret = sk_reuseport_attach_bpf(ufd, sk);
979 case SO_DETACH_FILTER:
980 ret = sk_detach_filter(sk);
984 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
987 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
992 set_bit(SOCK_PASSSEC, &sock->flags);
994 clear_bit(SOCK_PASSSEC, &sock->flags);
997 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1004 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1007 case SO_WIFI_STATUS:
1008 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1012 if (sock->ops->set_peek_off)
1013 ret = sock->ops->set_peek_off(sk, val);
1019 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1022 case SO_SELECT_ERR_QUEUE:
1023 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1026 #ifdef CONFIG_NET_RX_BUSY_POLL
1028 /* allow unprivileged users to decrease the value */
1029 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1035 sk->sk_ll_usec = val;
1040 case SO_MAX_PACING_RATE:
1041 sk->sk_max_pacing_rate = val;
1042 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1043 sk->sk_max_pacing_rate);
1046 case SO_INCOMING_CPU:
1047 sk->sk_incoming_cpu = val;
1052 dst_negative_advice(sk);
1061 EXPORT_SYMBOL(sock_setsockopt);
1064 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1065 struct ucred *ucred)
1067 ucred->pid = pid_vnr(pid);
1068 ucred->uid = ucred->gid = -1;
1070 struct user_namespace *current_ns = current_user_ns();
1072 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1073 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1077 int sock_getsockopt(struct socket *sock, int level, int optname,
1078 char __user *optval, int __user *optlen)
1080 struct sock *sk = sock->sk;
1089 int lv = sizeof(int);
1092 if (get_user(len, optlen))
1097 memset(&v, 0, sizeof(v));
1101 v.val = sock_flag(sk, SOCK_DBG);
1105 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1109 v.val = sock_flag(sk, SOCK_BROADCAST);
1113 v.val = sk->sk_sndbuf;
1117 v.val = sk->sk_rcvbuf;
1121 v.val = sk->sk_reuse;
1125 v.val = sk->sk_reuseport;
1129 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1133 v.val = sk->sk_type;
1137 v.val = sk->sk_protocol;
1141 v.val = sk->sk_family;
1145 v.val = -sock_error(sk);
1147 v.val = xchg(&sk->sk_err_soft, 0);
1151 v.val = sock_flag(sk, SOCK_URGINLINE);
1155 v.val = sk->sk_no_check_tx;
1159 v.val = sk->sk_priority;
1163 lv = sizeof(v.ling);
1164 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1165 v.ling.l_linger = sk->sk_lingertime / HZ;
1169 sock_warn_obsolete_bsdism("getsockopt");
1173 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1174 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1177 case SO_TIMESTAMPNS:
1178 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1181 case SO_TIMESTAMPING:
1182 v.val = sk->sk_tsflags;
1186 lv = sizeof(struct timeval);
1187 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1191 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1192 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1197 lv = sizeof(struct timeval);
1198 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1202 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1203 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1208 v.val = sk->sk_rcvlowat;
1216 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1221 struct ucred peercred;
1222 if (len > sizeof(peercred))
1223 len = sizeof(peercred);
1224 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1225 if (copy_to_user(optval, &peercred, len))
1234 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1238 if (copy_to_user(optval, address, len))
1243 /* Dubious BSD thing... Probably nobody even uses it, but
1244 * the UNIX standard wants it for whatever reason... -DaveM
1247 v.val = sk->sk_state == TCP_LISTEN;
1251 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1255 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1258 v.val = sk->sk_mark;
1262 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1265 case SO_WIFI_STATUS:
1266 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1270 if (!sock->ops->set_peek_off)
1273 v.val = sk->sk_peek_off;
1276 v.val = sock_flag(sk, SOCK_NOFCS);
1279 case SO_BINDTODEVICE:
1280 return sock_getbindtodevice(sk, optval, optlen, len);
1283 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1289 case SO_LOCK_FILTER:
1290 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1293 case SO_BPF_EXTENSIONS:
1294 v.val = bpf_tell_extensions();
1297 case SO_SELECT_ERR_QUEUE:
1298 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1301 #ifdef CONFIG_NET_RX_BUSY_POLL
1303 v.val = sk->sk_ll_usec;
1307 case SO_MAX_PACING_RATE:
1308 v.val = sk->sk_max_pacing_rate;
1311 case SO_INCOMING_CPU:
1312 v.val = sk->sk_incoming_cpu;
1317 u32 meminfo[SK_MEMINFO_VARS];
1319 if (get_user(len, optlen))
1322 sk_get_meminfo(sk, meminfo);
1324 len = min_t(unsigned int, len, sizeof(meminfo));
1325 if (copy_to_user(optval, &meminfo, len))
1331 #ifdef CONFIG_NET_RX_BUSY_POLL
1332 case SO_INCOMING_NAPI_ID:
1333 v.val = READ_ONCE(sk->sk_napi_id);
1335 /* aggregate non-NAPI IDs down to 0 */
1336 if (v.val < MIN_NAPI_ID)
1346 v.val64 = sock_gen_cookie(sk);
1350 /* We implement the SO_SNDLOWAT etc to not be settable
1353 return -ENOPROTOOPT;
1358 if (copy_to_user(optval, &v, len))
1361 if (put_user(len, optlen))
1367 * Initialize an sk_lock.
1369 * (We also register the sk_lock with the lock validator.)
1371 static inline void sock_lock_init(struct sock *sk)
1373 if (sk->sk_kern_sock)
1374 sock_lock_init_class_and_name(
1376 af_family_kern_slock_key_strings[sk->sk_family],
1377 af_family_kern_slock_keys + sk->sk_family,
1378 af_family_kern_key_strings[sk->sk_family],
1379 af_family_kern_keys + sk->sk_family);
1381 sock_lock_init_class_and_name(
1383 af_family_slock_key_strings[sk->sk_family],
1384 af_family_slock_keys + sk->sk_family,
1385 af_family_key_strings[sk->sk_family],
1386 af_family_keys + sk->sk_family);
1390 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1391 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1392 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1394 static void sock_copy(struct sock *nsk, const struct sock *osk)
1396 #ifdef CONFIG_SECURITY_NETWORK
1397 void *sptr = nsk->sk_security;
1399 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1401 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1402 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1404 #ifdef CONFIG_SECURITY_NETWORK
1405 nsk->sk_security = sptr;
1406 security_sk_clone(osk, nsk);
1410 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1414 struct kmem_cache *slab;
1418 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1421 if (priority & __GFP_ZERO)
1422 sk_prot_clear_nulls(sk, prot->obj_size);
1424 sk = kmalloc(prot->obj_size, priority);
1427 kmemcheck_annotate_bitfield(sk, flags);
1429 if (security_sk_alloc(sk, family, priority))
1432 if (!try_module_get(prot->owner))
1434 sk_tx_queue_clear(sk);
1440 security_sk_free(sk);
1443 kmem_cache_free(slab, sk);
1449 static void sk_prot_free(struct proto *prot, struct sock *sk)
1451 struct kmem_cache *slab;
1452 struct module *owner;
1454 owner = prot->owner;
1457 cgroup_sk_free(&sk->sk_cgrp_data);
1458 mem_cgroup_sk_free(sk);
1459 security_sk_free(sk);
1461 kmem_cache_free(slab, sk);
1468 * sk_alloc - All socket objects are allocated here
1469 * @net: the applicable net namespace
1470 * @family: protocol family
1471 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1472 * @prot: struct proto associated with this new sock instance
1473 * @kern: is this to be a kernel socket?
1475 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1476 struct proto *prot, int kern)
1480 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1482 sk->sk_family = family;
1484 * See comment in struct sock definition to understand
1485 * why we need sk_prot_creator -acme
1487 sk->sk_prot = sk->sk_prot_creator = prot;
1488 sk->sk_kern_sock = kern;
1490 sk->sk_net_refcnt = kern ? 0 : 1;
1491 if (likely(sk->sk_net_refcnt))
1493 sock_net_set(sk, net);
1494 atomic_set(&sk->sk_wmem_alloc, 1);
1496 mem_cgroup_sk_alloc(sk);
1497 cgroup_sk_alloc(&sk->sk_cgrp_data);
1498 sock_update_classid(&sk->sk_cgrp_data);
1499 sock_update_netprioidx(&sk->sk_cgrp_data);
1504 EXPORT_SYMBOL(sk_alloc);
1506 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1507 * grace period. This is the case for UDP sockets and TCP listeners.
1509 static void __sk_destruct(struct rcu_head *head)
1511 struct sock *sk = container_of(head, struct sock, sk_rcu);
1512 struct sk_filter *filter;
1514 if (sk->sk_destruct)
1515 sk->sk_destruct(sk);
1517 filter = rcu_dereference_check(sk->sk_filter,
1518 atomic_read(&sk->sk_wmem_alloc) == 0);
1520 sk_filter_uncharge(sk, filter);
1521 RCU_INIT_POINTER(sk->sk_filter, NULL);
1523 if (rcu_access_pointer(sk->sk_reuseport_cb))
1524 reuseport_detach_sock(sk);
1526 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1528 if (atomic_read(&sk->sk_omem_alloc))
1529 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1530 __func__, atomic_read(&sk->sk_omem_alloc));
1532 if (sk->sk_frag.page) {
1533 put_page(sk->sk_frag.page);
1534 sk->sk_frag.page = NULL;
1537 if (sk->sk_peer_cred)
1538 put_cred(sk->sk_peer_cred);
1539 put_pid(sk->sk_peer_pid);
1540 if (likely(sk->sk_net_refcnt))
1541 put_net(sock_net(sk));
1542 sk_prot_free(sk->sk_prot_creator, sk);
1545 void sk_destruct(struct sock *sk)
1547 if (sock_flag(sk, SOCK_RCU_FREE))
1548 call_rcu(&sk->sk_rcu, __sk_destruct);
1550 __sk_destruct(&sk->sk_rcu);
1553 static void __sk_free(struct sock *sk)
1555 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1556 sock_diag_broadcast_destroy(sk);
1561 void sk_free(struct sock *sk)
1564 * We subtract one from sk_wmem_alloc and can know if
1565 * some packets are still in some tx queue.
1566 * If not null, sock_wfree() will call __sk_free(sk) later
1568 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1571 EXPORT_SYMBOL(sk_free);
1573 static void sk_init_common(struct sock *sk)
1575 skb_queue_head_init(&sk->sk_receive_queue);
1576 skb_queue_head_init(&sk->sk_write_queue);
1577 skb_queue_head_init(&sk->sk_error_queue);
1579 rwlock_init(&sk->sk_callback_lock);
1580 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1581 af_rlock_keys + sk->sk_family,
1582 af_family_rlock_key_strings[sk->sk_family]);
1583 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1584 af_wlock_keys + sk->sk_family,
1585 af_family_wlock_key_strings[sk->sk_family]);
1586 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1587 af_elock_keys + sk->sk_family,
1588 af_family_elock_key_strings[sk->sk_family]);
1589 lockdep_set_class_and_name(&sk->sk_callback_lock,
1590 af_callback_keys + sk->sk_family,
1591 af_family_clock_key_strings[sk->sk_family]);
1595 * sk_clone_lock - clone a socket, and lock its clone
1596 * @sk: the socket to clone
1597 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1599 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1601 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1604 bool is_charged = true;
1606 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1607 if (newsk != NULL) {
1608 struct sk_filter *filter;
1610 sock_copy(newsk, sk);
1613 if (likely(newsk->sk_net_refcnt))
1614 get_net(sock_net(newsk));
1615 sk_node_init(&newsk->sk_node);
1616 sock_lock_init(newsk);
1617 bh_lock_sock(newsk);
1618 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1619 newsk->sk_backlog.len = 0;
1621 atomic_set(&newsk->sk_rmem_alloc, 0);
1623 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1625 atomic_set(&newsk->sk_wmem_alloc, 1);
1626 atomic_set(&newsk->sk_omem_alloc, 0);
1627 sk_init_common(newsk);
1629 newsk->sk_dst_cache = NULL;
1630 newsk->sk_dst_pending_confirm = 0;
1631 newsk->sk_wmem_queued = 0;
1632 newsk->sk_forward_alloc = 0;
1633 atomic_set(&newsk->sk_drops, 0);
1634 newsk->sk_send_head = NULL;
1635 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1637 sock_reset_flag(newsk, SOCK_DONE);
1639 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1641 /* though it's an empty new sock, the charging may fail
1642 * if sysctl_optmem_max was changed between creation of
1643 * original socket and cloning
1645 is_charged = sk_filter_charge(newsk, filter);
1647 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1648 /* We need to make sure that we don't uncharge the new
1649 * socket if we couldn't charge it in the first place
1650 * as otherwise we uncharge the parent's filter.
1653 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1654 sk_free_unlock_clone(newsk);
1658 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1661 newsk->sk_err_soft = 0;
1662 newsk->sk_priority = 0;
1663 newsk->sk_incoming_cpu = raw_smp_processor_id();
1664 atomic64_set(&newsk->sk_cookie, 0);
1666 mem_cgroup_sk_alloc(newsk);
1667 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1670 * Before updating sk_refcnt, we must commit prior changes to memory
1671 * (Documentation/RCU/rculist_nulls.txt for details)
1674 atomic_set(&newsk->sk_refcnt, 2);
1677 * Increment the counter in the same struct proto as the master
1678 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1679 * is the same as sk->sk_prot->socks, as this field was copied
1682 * This _changes_ the previous behaviour, where
1683 * tcp_create_openreq_child always was incrementing the
1684 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1685 * to be taken into account in all callers. -acme
1687 sk_refcnt_debug_inc(newsk);
1688 sk_set_socket(newsk, NULL);
1689 newsk->sk_wq = NULL;
1691 if (newsk->sk_prot->sockets_allocated)
1692 sk_sockets_allocated_inc(newsk);
1694 if (sock_needs_netstamp(sk) &&
1695 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1696 net_enable_timestamp();
1701 EXPORT_SYMBOL_GPL(sk_clone_lock);
1703 void sk_free_unlock_clone(struct sock *sk)
1705 /* It is still raw copy of parent, so invalidate
1706 * destructor and make plain sk_free() */
1707 sk->sk_destruct = NULL;
1711 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1713 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1717 sk_dst_set(sk, dst);
1718 sk->sk_route_caps = dst->dev->features;
1719 if (sk->sk_route_caps & NETIF_F_GSO)
1720 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1721 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1722 if (sk_can_gso(sk)) {
1723 if (dst->header_len) {
1724 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1726 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1727 sk->sk_gso_max_size = dst->dev->gso_max_size;
1728 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1731 sk->sk_gso_max_segs = max_segs;
1733 EXPORT_SYMBOL_GPL(sk_setup_caps);
1736 * Simple resource managers for sockets.
1741 * Write buffer destructor automatically called from kfree_skb.
1743 void sock_wfree(struct sk_buff *skb)
1745 struct sock *sk = skb->sk;
1746 unsigned int len = skb->truesize;
1748 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1750 * Keep a reference on sk_wmem_alloc, this will be released
1751 * after sk_write_space() call
1753 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1754 sk->sk_write_space(sk);
1758 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1759 * could not do because of in-flight packets
1761 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1764 EXPORT_SYMBOL(sock_wfree);
1766 /* This variant of sock_wfree() is used by TCP,
1767 * since it sets SOCK_USE_WRITE_QUEUE.
1769 void __sock_wfree(struct sk_buff *skb)
1771 struct sock *sk = skb->sk;
1773 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1777 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1782 if (unlikely(!sk_fullsock(sk))) {
1783 skb->destructor = sock_edemux;
1788 skb->destructor = sock_wfree;
1789 skb_set_hash_from_sk(skb, sk);
1791 * We used to take a refcount on sk, but following operation
1792 * is enough to guarantee sk_free() wont free this sock until
1793 * all in-flight packets are completed
1795 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1797 EXPORT_SYMBOL(skb_set_owner_w);
1799 /* This helper is used by netem, as it can hold packets in its
1800 * delay queue. We want to allow the owner socket to send more
1801 * packets, as if they were already TX completed by a typical driver.
1802 * But we also want to keep skb->sk set because some packet schedulers
1803 * rely on it (sch_fq for example).
1805 void skb_orphan_partial(struct sk_buff *skb)
1807 if (skb_is_tcp_pure_ack(skb))
1810 if (skb->destructor == sock_wfree
1812 || skb->destructor == tcp_wfree
1815 struct sock *sk = skb->sk;
1817 if (atomic_inc_not_zero(&sk->sk_refcnt)) {
1818 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1819 skb->destructor = sock_efree;
1825 EXPORT_SYMBOL(skb_orphan_partial);
1828 * Read buffer destructor automatically called from kfree_skb.
1830 void sock_rfree(struct sk_buff *skb)
1832 struct sock *sk = skb->sk;
1833 unsigned int len = skb->truesize;
1835 atomic_sub(len, &sk->sk_rmem_alloc);
1836 sk_mem_uncharge(sk, len);
1838 EXPORT_SYMBOL(sock_rfree);
1841 * Buffer destructor for skbs that are not used directly in read or write
1842 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1844 void sock_efree(struct sk_buff *skb)
1848 EXPORT_SYMBOL(sock_efree);
1850 kuid_t sock_i_uid(struct sock *sk)
1854 read_lock_bh(&sk->sk_callback_lock);
1855 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1856 read_unlock_bh(&sk->sk_callback_lock);
1859 EXPORT_SYMBOL(sock_i_uid);
1861 unsigned long sock_i_ino(struct sock *sk)
1865 read_lock_bh(&sk->sk_callback_lock);
1866 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1867 read_unlock_bh(&sk->sk_callback_lock);
1870 EXPORT_SYMBOL(sock_i_ino);
1873 * Allocate a skb from the socket's send buffer.
1875 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1878 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1879 struct sk_buff *skb = alloc_skb(size, priority);
1881 skb_set_owner_w(skb, sk);
1887 EXPORT_SYMBOL(sock_wmalloc);
1890 * Allocate a memory block from the socket's option memory buffer.
1892 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1894 if ((unsigned int)size <= sysctl_optmem_max &&
1895 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1897 /* First do the add, to avoid the race if kmalloc
1900 atomic_add(size, &sk->sk_omem_alloc);
1901 mem = kmalloc(size, priority);
1904 atomic_sub(size, &sk->sk_omem_alloc);
1908 EXPORT_SYMBOL(sock_kmalloc);
1910 /* Free an option memory block. Note, we actually want the inline
1911 * here as this allows gcc to detect the nullify and fold away the
1912 * condition entirely.
1914 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1917 if (WARN_ON_ONCE(!mem))
1923 atomic_sub(size, &sk->sk_omem_alloc);
1926 void sock_kfree_s(struct sock *sk, void *mem, int size)
1928 __sock_kfree_s(sk, mem, size, false);
1930 EXPORT_SYMBOL(sock_kfree_s);
1932 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1934 __sock_kfree_s(sk, mem, size, true);
1936 EXPORT_SYMBOL(sock_kzfree_s);
1938 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1939 I think, these locks should be removed for datagram sockets.
1941 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1945 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1949 if (signal_pending(current))
1951 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1952 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1953 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1955 if (sk->sk_shutdown & SEND_SHUTDOWN)
1959 timeo = schedule_timeout(timeo);
1961 finish_wait(sk_sleep(sk), &wait);
1967 * Generic send/receive buffer handlers
1970 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1971 unsigned long data_len, int noblock,
1972 int *errcode, int max_page_order)
1974 struct sk_buff *skb;
1978 timeo = sock_sndtimeo(sk, noblock);
1980 err = sock_error(sk);
1985 if (sk->sk_shutdown & SEND_SHUTDOWN)
1988 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1991 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1992 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1996 if (signal_pending(current))
1998 timeo = sock_wait_for_wmem(sk, timeo);
2000 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2001 errcode, sk->sk_allocation);
2003 skb_set_owner_w(skb, sk);
2007 err = sock_intr_errno(timeo);
2012 EXPORT_SYMBOL(sock_alloc_send_pskb);
2014 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2015 int noblock, int *errcode)
2017 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2019 EXPORT_SYMBOL(sock_alloc_send_skb);
2021 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2022 struct sockcm_cookie *sockc)
2026 switch (cmsg->cmsg_type) {
2028 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2030 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2032 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2034 case SO_TIMESTAMPING:
2035 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2038 tsflags = *(u32 *)CMSG_DATA(cmsg);
2039 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2042 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2043 sockc->tsflags |= tsflags;
2045 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2047 case SCM_CREDENTIALS:
2054 EXPORT_SYMBOL(__sock_cmsg_send);
2056 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2057 struct sockcm_cookie *sockc)
2059 struct cmsghdr *cmsg;
2062 for_each_cmsghdr(cmsg, msg) {
2063 if (!CMSG_OK(msg, cmsg))
2065 if (cmsg->cmsg_level != SOL_SOCKET)
2067 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2073 EXPORT_SYMBOL(sock_cmsg_send);
2075 /* On 32bit arches, an skb frag is limited to 2^15 */
2076 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2079 * skb_page_frag_refill - check that a page_frag contains enough room
2080 * @sz: minimum size of the fragment we want to get
2081 * @pfrag: pointer to page_frag
2082 * @gfp: priority for memory allocation
2084 * Note: While this allocator tries to use high order pages, there is
2085 * no guarantee that allocations succeed. Therefore, @sz MUST be
2086 * less or equal than PAGE_SIZE.
2088 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2091 if (page_ref_count(pfrag->page) == 1) {
2095 if (pfrag->offset + sz <= pfrag->size)
2097 put_page(pfrag->page);
2101 if (SKB_FRAG_PAGE_ORDER) {
2102 /* Avoid direct reclaim but allow kswapd to wake */
2103 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2104 __GFP_COMP | __GFP_NOWARN |
2106 SKB_FRAG_PAGE_ORDER);
2107 if (likely(pfrag->page)) {
2108 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2112 pfrag->page = alloc_page(gfp);
2113 if (likely(pfrag->page)) {
2114 pfrag->size = PAGE_SIZE;
2119 EXPORT_SYMBOL(skb_page_frag_refill);
2121 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2123 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2126 sk_enter_memory_pressure(sk);
2127 sk_stream_moderate_sndbuf(sk);
2130 EXPORT_SYMBOL(sk_page_frag_refill);
2132 static void __lock_sock(struct sock *sk)
2133 __releases(&sk->sk_lock.slock)
2134 __acquires(&sk->sk_lock.slock)
2139 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2140 TASK_UNINTERRUPTIBLE);
2141 spin_unlock_bh(&sk->sk_lock.slock);
2143 spin_lock_bh(&sk->sk_lock.slock);
2144 if (!sock_owned_by_user(sk))
2147 finish_wait(&sk->sk_lock.wq, &wait);
2150 static void __release_sock(struct sock *sk)
2151 __releases(&sk->sk_lock.slock)
2152 __acquires(&sk->sk_lock.slock)
2154 struct sk_buff *skb, *next;
2156 while ((skb = sk->sk_backlog.head) != NULL) {
2157 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2159 spin_unlock_bh(&sk->sk_lock.slock);
2164 WARN_ON_ONCE(skb_dst_is_noref(skb));
2166 sk_backlog_rcv(sk, skb);
2171 } while (skb != NULL);
2173 spin_lock_bh(&sk->sk_lock.slock);
2177 * Doing the zeroing here guarantee we can not loop forever
2178 * while a wild producer attempts to flood us.
2180 sk->sk_backlog.len = 0;
2183 void __sk_flush_backlog(struct sock *sk)
2185 spin_lock_bh(&sk->sk_lock.slock);
2187 spin_unlock_bh(&sk->sk_lock.slock);
2191 * sk_wait_data - wait for data to arrive at sk_receive_queue
2192 * @sk: sock to wait on
2193 * @timeo: for how long
2194 * @skb: last skb seen on sk_receive_queue
2196 * Now socket state including sk->sk_err is changed only under lock,
2197 * hence we may omit checks after joining wait queue.
2198 * We check receive queue before schedule() only as optimization;
2199 * it is very likely that release_sock() added new data.
2201 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2203 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2206 add_wait_queue(sk_sleep(sk), &wait);
2207 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2208 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2209 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2210 remove_wait_queue(sk_sleep(sk), &wait);
2213 EXPORT_SYMBOL(sk_wait_data);
2216 * __sk_mem_raise_allocated - increase memory_allocated
2218 * @size: memory size to allocate
2219 * @amt: pages to allocate
2220 * @kind: allocation type
2222 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2224 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2226 struct proto *prot = sk->sk_prot;
2227 long allocated = sk_memory_allocated_add(sk, amt);
2229 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2230 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2231 goto suppress_allocation;
2234 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2235 sk_leave_memory_pressure(sk);
2239 /* Under pressure. */
2240 if (allocated > sk_prot_mem_limits(sk, 1))
2241 sk_enter_memory_pressure(sk);
2243 /* Over hard limit. */
2244 if (allocated > sk_prot_mem_limits(sk, 2))
2245 goto suppress_allocation;
2247 /* guarantee minimum buffer size under pressure */
2248 if (kind == SK_MEM_RECV) {
2249 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2252 } else { /* SK_MEM_SEND */
2253 if (sk->sk_type == SOCK_STREAM) {
2254 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2256 } else if (atomic_read(&sk->sk_wmem_alloc) <
2257 prot->sysctl_wmem[0])
2261 if (sk_has_memory_pressure(sk)) {
2264 if (!sk_under_memory_pressure(sk))
2266 alloc = sk_sockets_allocated_read_positive(sk);
2267 if (sk_prot_mem_limits(sk, 2) > alloc *
2268 sk_mem_pages(sk->sk_wmem_queued +
2269 atomic_read(&sk->sk_rmem_alloc) +
2270 sk->sk_forward_alloc))
2274 suppress_allocation:
2276 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2277 sk_stream_moderate_sndbuf(sk);
2279 /* Fail only if socket is _under_ its sndbuf.
2280 * In this case we cannot block, so that we have to fail.
2282 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2286 trace_sock_exceed_buf_limit(sk, prot, allocated);
2288 sk_memory_allocated_sub(sk, amt);
2290 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2291 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2295 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2298 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2300 * @size: memory size to allocate
2301 * @kind: allocation type
2303 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2304 * rmem allocation. This function assumes that protocols which have
2305 * memory_pressure use sk_wmem_queued as write buffer accounting.
2307 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2309 int ret, amt = sk_mem_pages(size);
2311 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2312 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2314 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2317 EXPORT_SYMBOL(__sk_mem_schedule);
2320 * __sk_mem_reduce_allocated - reclaim memory_allocated
2322 * @amount: number of quanta
2324 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2326 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2328 sk_memory_allocated_sub(sk, amount);
2330 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2331 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2333 if (sk_under_memory_pressure(sk) &&
2334 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2335 sk_leave_memory_pressure(sk);
2337 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2340 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2342 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2344 void __sk_mem_reclaim(struct sock *sk, int amount)
2346 amount >>= SK_MEM_QUANTUM_SHIFT;
2347 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2348 __sk_mem_reduce_allocated(sk, amount);
2350 EXPORT_SYMBOL(__sk_mem_reclaim);
2352 int sk_set_peek_off(struct sock *sk, int val)
2357 sk->sk_peek_off = val;
2360 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2363 * Set of default routines for initialising struct proto_ops when
2364 * the protocol does not support a particular function. In certain
2365 * cases where it makes no sense for a protocol to have a "do nothing"
2366 * function, some default processing is provided.
2369 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2373 EXPORT_SYMBOL(sock_no_bind);
2375 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2380 EXPORT_SYMBOL(sock_no_connect);
2382 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2386 EXPORT_SYMBOL(sock_no_socketpair);
2388 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2393 EXPORT_SYMBOL(sock_no_accept);
2395 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2400 EXPORT_SYMBOL(sock_no_getname);
2402 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2406 EXPORT_SYMBOL(sock_no_poll);
2408 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2412 EXPORT_SYMBOL(sock_no_ioctl);
2414 int sock_no_listen(struct socket *sock, int backlog)
2418 EXPORT_SYMBOL(sock_no_listen);
2420 int sock_no_shutdown(struct socket *sock, int how)
2424 EXPORT_SYMBOL(sock_no_shutdown);
2426 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2427 char __user *optval, unsigned int optlen)
2431 EXPORT_SYMBOL(sock_no_setsockopt);
2433 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2434 char __user *optval, int __user *optlen)
2438 EXPORT_SYMBOL(sock_no_getsockopt);
2440 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2444 EXPORT_SYMBOL(sock_no_sendmsg);
2446 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2451 EXPORT_SYMBOL(sock_no_recvmsg);
2453 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2455 /* Mirror missing mmap method error code */
2458 EXPORT_SYMBOL(sock_no_mmap);
2460 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2463 struct msghdr msg = {.msg_flags = flags};
2465 char *kaddr = kmap(page);
2466 iov.iov_base = kaddr + offset;
2468 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2472 EXPORT_SYMBOL(sock_no_sendpage);
2475 * Default Socket Callbacks
2478 static void sock_def_wakeup(struct sock *sk)
2480 struct socket_wq *wq;
2483 wq = rcu_dereference(sk->sk_wq);
2484 if (skwq_has_sleeper(wq))
2485 wake_up_interruptible_all(&wq->wait);
2489 static void sock_def_error_report(struct sock *sk)
2491 struct socket_wq *wq;
2494 wq = rcu_dereference(sk->sk_wq);
2495 if (skwq_has_sleeper(wq))
2496 wake_up_interruptible_poll(&wq->wait, POLLERR);
2497 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2501 static void sock_def_readable(struct sock *sk)
2503 struct socket_wq *wq;
2506 wq = rcu_dereference(sk->sk_wq);
2507 if (skwq_has_sleeper(wq))
2508 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2509 POLLRDNORM | POLLRDBAND);
2510 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2514 static void sock_def_write_space(struct sock *sk)
2516 struct socket_wq *wq;
2520 /* Do not wake up a writer until he can make "significant"
2523 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2524 wq = rcu_dereference(sk->sk_wq);
2525 if (skwq_has_sleeper(wq))
2526 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2527 POLLWRNORM | POLLWRBAND);
2529 /* Should agree with poll, otherwise some programs break */
2530 if (sock_writeable(sk))
2531 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2537 static void sock_def_destruct(struct sock *sk)
2541 void sk_send_sigurg(struct sock *sk)
2543 if (sk->sk_socket && sk->sk_socket->file)
2544 if (send_sigurg(&sk->sk_socket->file->f_owner))
2545 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2547 EXPORT_SYMBOL(sk_send_sigurg);
2549 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2550 unsigned long expires)
2552 if (!mod_timer(timer, expires))
2555 EXPORT_SYMBOL(sk_reset_timer);
2557 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2559 if (del_timer(timer))
2562 EXPORT_SYMBOL(sk_stop_timer);
2564 void sock_init_data(struct socket *sock, struct sock *sk)
2567 sk->sk_send_head = NULL;
2569 init_timer(&sk->sk_timer);
2571 sk->sk_allocation = GFP_KERNEL;
2572 sk->sk_rcvbuf = sysctl_rmem_default;
2573 sk->sk_sndbuf = sysctl_wmem_default;
2574 sk->sk_state = TCP_CLOSE;
2575 sk_set_socket(sk, sock);
2577 sock_set_flag(sk, SOCK_ZAPPED);
2580 sk->sk_type = sock->type;
2581 sk->sk_wq = sock->wq;
2583 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2586 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2589 rwlock_init(&sk->sk_callback_lock);
2590 if (sk->sk_kern_sock)
2591 lockdep_set_class_and_name(
2592 &sk->sk_callback_lock,
2593 af_kern_callback_keys + sk->sk_family,
2594 af_family_kern_clock_key_strings[sk->sk_family]);
2596 lockdep_set_class_and_name(
2597 &sk->sk_callback_lock,
2598 af_callback_keys + sk->sk_family,
2599 af_family_clock_key_strings[sk->sk_family]);
2601 sk->sk_state_change = sock_def_wakeup;
2602 sk->sk_data_ready = sock_def_readable;
2603 sk->sk_write_space = sock_def_write_space;
2604 sk->sk_error_report = sock_def_error_report;
2605 sk->sk_destruct = sock_def_destruct;
2607 sk->sk_frag.page = NULL;
2608 sk->sk_frag.offset = 0;
2609 sk->sk_peek_off = -1;
2611 sk->sk_peer_pid = NULL;
2612 sk->sk_peer_cred = NULL;
2613 sk->sk_write_pending = 0;
2614 sk->sk_rcvlowat = 1;
2615 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2616 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2618 sk->sk_stamp = SK_DEFAULT_STAMP;
2620 #ifdef CONFIG_NET_RX_BUSY_POLL
2622 sk->sk_ll_usec = sysctl_net_busy_read;
2625 sk->sk_max_pacing_rate = ~0U;
2626 sk->sk_pacing_rate = ~0U;
2627 sk->sk_incoming_cpu = -1;
2629 * Before updating sk_refcnt, we must commit prior changes to memory
2630 * (Documentation/RCU/rculist_nulls.txt for details)
2633 atomic_set(&sk->sk_refcnt, 1);
2634 atomic_set(&sk->sk_drops, 0);
2636 EXPORT_SYMBOL(sock_init_data);
2638 void lock_sock_nested(struct sock *sk, int subclass)
2641 spin_lock_bh(&sk->sk_lock.slock);
2642 if (sk->sk_lock.owned)
2644 sk->sk_lock.owned = 1;
2645 spin_unlock(&sk->sk_lock.slock);
2647 * The sk_lock has mutex_lock() semantics here:
2649 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2652 EXPORT_SYMBOL(lock_sock_nested);
2654 void release_sock(struct sock *sk)
2656 spin_lock_bh(&sk->sk_lock.slock);
2657 if (sk->sk_backlog.tail)
2660 /* Warning : release_cb() might need to release sk ownership,
2661 * ie call sock_release_ownership(sk) before us.
2663 if (sk->sk_prot->release_cb)
2664 sk->sk_prot->release_cb(sk);
2666 sock_release_ownership(sk);
2667 if (waitqueue_active(&sk->sk_lock.wq))
2668 wake_up(&sk->sk_lock.wq);
2669 spin_unlock_bh(&sk->sk_lock.slock);
2671 EXPORT_SYMBOL(release_sock);
2674 * lock_sock_fast - fast version of lock_sock
2677 * This version should be used for very small section, where process wont block
2678 * return false if fast path is taken
2679 * sk_lock.slock locked, owned = 0, BH disabled
2680 * return true if slow path is taken
2681 * sk_lock.slock unlocked, owned = 1, BH enabled
2683 bool lock_sock_fast(struct sock *sk)
2686 spin_lock_bh(&sk->sk_lock.slock);
2688 if (!sk->sk_lock.owned)
2690 * Note : We must disable BH
2695 sk->sk_lock.owned = 1;
2696 spin_unlock(&sk->sk_lock.slock);
2698 * The sk_lock has mutex_lock() semantics here:
2700 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2704 EXPORT_SYMBOL(lock_sock_fast);
2706 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2709 if (!sock_flag(sk, SOCK_TIMESTAMP))
2710 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2711 tv = ktime_to_timeval(sk->sk_stamp);
2712 if (tv.tv_sec == -1)
2714 if (tv.tv_sec == 0) {
2715 sk->sk_stamp = ktime_get_real();
2716 tv = ktime_to_timeval(sk->sk_stamp);
2718 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2720 EXPORT_SYMBOL(sock_get_timestamp);
2722 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2725 if (!sock_flag(sk, SOCK_TIMESTAMP))
2726 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2727 ts = ktime_to_timespec(sk->sk_stamp);
2728 if (ts.tv_sec == -1)
2730 if (ts.tv_sec == 0) {
2731 sk->sk_stamp = ktime_get_real();
2732 ts = ktime_to_timespec(sk->sk_stamp);
2734 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2736 EXPORT_SYMBOL(sock_get_timestampns);
2738 void sock_enable_timestamp(struct sock *sk, int flag)
2740 if (!sock_flag(sk, flag)) {
2741 unsigned long previous_flags = sk->sk_flags;
2743 sock_set_flag(sk, flag);
2745 * we just set one of the two flags which require net
2746 * time stamping, but time stamping might have been on
2747 * already because of the other one
2749 if (sock_needs_netstamp(sk) &&
2750 !(previous_flags & SK_FLAGS_TIMESTAMP))
2751 net_enable_timestamp();
2755 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2756 int level, int type)
2758 struct sock_exterr_skb *serr;
2759 struct sk_buff *skb;
2763 skb = sock_dequeue_err_skb(sk);
2769 msg->msg_flags |= MSG_TRUNC;
2772 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2776 sock_recv_timestamp(msg, sk, skb);
2778 serr = SKB_EXT_ERR(skb);
2779 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2781 msg->msg_flags |= MSG_ERRQUEUE;
2789 EXPORT_SYMBOL(sock_recv_errqueue);
2792 * Get a socket option on an socket.
2794 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2795 * asynchronous errors should be reported by getsockopt. We assume
2796 * this means if you specify SO_ERROR (otherwise whats the point of it).
2798 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2799 char __user *optval, int __user *optlen)
2801 struct sock *sk = sock->sk;
2803 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2805 EXPORT_SYMBOL(sock_common_getsockopt);
2807 #ifdef CONFIG_COMPAT
2808 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2809 char __user *optval, int __user *optlen)
2811 struct sock *sk = sock->sk;
2813 if (sk->sk_prot->compat_getsockopt != NULL)
2814 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2816 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2818 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2821 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2824 struct sock *sk = sock->sk;
2828 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2829 flags & ~MSG_DONTWAIT, &addr_len);
2831 msg->msg_namelen = addr_len;
2834 EXPORT_SYMBOL(sock_common_recvmsg);
2837 * Set socket options on an inet socket.
2839 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2840 char __user *optval, unsigned int optlen)
2842 struct sock *sk = sock->sk;
2844 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2846 EXPORT_SYMBOL(sock_common_setsockopt);
2848 #ifdef CONFIG_COMPAT
2849 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2850 char __user *optval, unsigned int optlen)
2852 struct sock *sk = sock->sk;
2854 if (sk->sk_prot->compat_setsockopt != NULL)
2855 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2857 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2859 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2862 void sk_common_release(struct sock *sk)
2864 if (sk->sk_prot->destroy)
2865 sk->sk_prot->destroy(sk);
2868 * Observation: when sock_common_release is called, processes have
2869 * no access to socket. But net still has.
2870 * Step one, detach it from networking:
2872 * A. Remove from hash tables.
2875 sk->sk_prot->unhash(sk);
2878 * In this point socket cannot receive new packets, but it is possible
2879 * that some packets are in flight because some CPU runs receiver and
2880 * did hash table lookup before we unhashed socket. They will achieve
2881 * receive queue and will be purged by socket destructor.
2883 * Also we still have packets pending on receive queue and probably,
2884 * our own packets waiting in device queues. sock_destroy will drain
2885 * receive queue, but transmitted packets will delay socket destruction
2886 * until the last reference will be released.
2891 xfrm_sk_free_policy(sk);
2893 sk_refcnt_debug_release(sk);
2897 EXPORT_SYMBOL(sk_common_release);
2899 void sk_get_meminfo(const struct sock *sk, u32 *mem)
2901 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
2903 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
2904 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
2905 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
2906 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
2907 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
2908 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
2909 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
2910 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
2911 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
2914 #ifdef CONFIG_PROC_FS
2915 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2917 int val[PROTO_INUSE_NR];
2920 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2922 #ifdef CONFIG_NET_NS
2923 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2925 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2927 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2929 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2931 int cpu, idx = prot->inuse_idx;
2934 for_each_possible_cpu(cpu)
2935 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2937 return res >= 0 ? res : 0;
2939 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2941 static int __net_init sock_inuse_init_net(struct net *net)
2943 net->core.inuse = alloc_percpu(struct prot_inuse);
2944 return net->core.inuse ? 0 : -ENOMEM;
2947 static void __net_exit sock_inuse_exit_net(struct net *net)
2949 free_percpu(net->core.inuse);
2952 static struct pernet_operations net_inuse_ops = {
2953 .init = sock_inuse_init_net,
2954 .exit = sock_inuse_exit_net,
2957 static __init int net_inuse_init(void)
2959 if (register_pernet_subsys(&net_inuse_ops))
2960 panic("Cannot initialize net inuse counters");
2965 core_initcall(net_inuse_init);
2967 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2969 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2971 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2973 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2975 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2977 int cpu, idx = prot->inuse_idx;
2980 for_each_possible_cpu(cpu)
2981 res += per_cpu(prot_inuse, cpu).val[idx];
2983 return res >= 0 ? res : 0;
2985 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2988 static void assign_proto_idx(struct proto *prot)
2990 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2992 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2993 pr_err("PROTO_INUSE_NR exhausted\n");
2997 set_bit(prot->inuse_idx, proto_inuse_idx);
3000 static void release_proto_idx(struct proto *prot)
3002 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3003 clear_bit(prot->inuse_idx, proto_inuse_idx);
3006 static inline void assign_proto_idx(struct proto *prot)
3010 static inline void release_proto_idx(struct proto *prot)
3015 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3019 kfree(rsk_prot->slab_name);
3020 rsk_prot->slab_name = NULL;
3021 kmem_cache_destroy(rsk_prot->slab);
3022 rsk_prot->slab = NULL;
3025 static int req_prot_init(const struct proto *prot)
3027 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3032 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3034 if (!rsk_prot->slab_name)
3037 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3038 rsk_prot->obj_size, 0,
3039 prot->slab_flags, NULL);
3041 if (!rsk_prot->slab) {
3042 pr_crit("%s: Can't create request sock SLAB cache!\n",
3049 int proto_register(struct proto *prot, int alloc_slab)
3052 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
3053 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3056 if (prot->slab == NULL) {
3057 pr_crit("%s: Can't create sock SLAB cache!\n",
3062 if (req_prot_init(prot))
3063 goto out_free_request_sock_slab;
3065 if (prot->twsk_prot != NULL) {
3066 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3068 if (prot->twsk_prot->twsk_slab_name == NULL)
3069 goto out_free_request_sock_slab;
3071 prot->twsk_prot->twsk_slab =
3072 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3073 prot->twsk_prot->twsk_obj_size,
3077 if (prot->twsk_prot->twsk_slab == NULL)
3078 goto out_free_timewait_sock_slab_name;
3082 mutex_lock(&proto_list_mutex);
3083 list_add(&prot->node, &proto_list);
3084 assign_proto_idx(prot);
3085 mutex_unlock(&proto_list_mutex);
3088 out_free_timewait_sock_slab_name:
3089 kfree(prot->twsk_prot->twsk_slab_name);
3090 out_free_request_sock_slab:
3091 req_prot_cleanup(prot->rsk_prot);
3093 kmem_cache_destroy(prot->slab);
3098 EXPORT_SYMBOL(proto_register);
3100 void proto_unregister(struct proto *prot)
3102 mutex_lock(&proto_list_mutex);
3103 release_proto_idx(prot);
3104 list_del(&prot->node);
3105 mutex_unlock(&proto_list_mutex);
3107 kmem_cache_destroy(prot->slab);
3110 req_prot_cleanup(prot->rsk_prot);
3112 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3113 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3114 kfree(prot->twsk_prot->twsk_slab_name);
3115 prot->twsk_prot->twsk_slab = NULL;
3118 EXPORT_SYMBOL(proto_unregister);
3120 #ifdef CONFIG_PROC_FS
3121 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3122 __acquires(proto_list_mutex)
3124 mutex_lock(&proto_list_mutex);
3125 return seq_list_start_head(&proto_list, *pos);
3128 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3130 return seq_list_next(v, &proto_list, pos);
3133 static void proto_seq_stop(struct seq_file *seq, void *v)
3134 __releases(proto_list_mutex)
3136 mutex_unlock(&proto_list_mutex);
3139 static char proto_method_implemented(const void *method)
3141 return method == NULL ? 'n' : 'y';
3143 static long sock_prot_memory_allocated(struct proto *proto)
3145 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3148 static char *sock_prot_memory_pressure(struct proto *proto)
3150 return proto->memory_pressure != NULL ?
3151 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3154 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3157 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3158 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3161 sock_prot_inuse_get(seq_file_net(seq), proto),
3162 sock_prot_memory_allocated(proto),
3163 sock_prot_memory_pressure(proto),
3165 proto->slab == NULL ? "no" : "yes",
3166 module_name(proto->owner),
3167 proto_method_implemented(proto->close),
3168 proto_method_implemented(proto->connect),
3169 proto_method_implemented(proto->disconnect),
3170 proto_method_implemented(proto->accept),
3171 proto_method_implemented(proto->ioctl),
3172 proto_method_implemented(proto->init),
3173 proto_method_implemented(proto->destroy),
3174 proto_method_implemented(proto->shutdown),
3175 proto_method_implemented(proto->setsockopt),
3176 proto_method_implemented(proto->getsockopt),
3177 proto_method_implemented(proto->sendmsg),
3178 proto_method_implemented(proto->recvmsg),
3179 proto_method_implemented(proto->sendpage),
3180 proto_method_implemented(proto->bind),
3181 proto_method_implemented(proto->backlog_rcv),
3182 proto_method_implemented(proto->hash),
3183 proto_method_implemented(proto->unhash),
3184 proto_method_implemented(proto->get_port),
3185 proto_method_implemented(proto->enter_memory_pressure));
3188 static int proto_seq_show(struct seq_file *seq, void *v)
3190 if (v == &proto_list)
3191 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3200 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3202 proto_seq_printf(seq, list_entry(v, struct proto, node));
3206 static const struct seq_operations proto_seq_ops = {
3207 .start = proto_seq_start,
3208 .next = proto_seq_next,
3209 .stop = proto_seq_stop,
3210 .show = proto_seq_show,
3213 static int proto_seq_open(struct inode *inode, struct file *file)
3215 return seq_open_net(inode, file, &proto_seq_ops,
3216 sizeof(struct seq_net_private));
3219 static const struct file_operations proto_seq_fops = {
3220 .owner = THIS_MODULE,
3221 .open = proto_seq_open,
3223 .llseek = seq_lseek,
3224 .release = seq_release_net,
3227 static __net_init int proto_init_net(struct net *net)
3229 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3235 static __net_exit void proto_exit_net(struct net *net)
3237 remove_proc_entry("protocols", net->proc_net);
3241 static __net_initdata struct pernet_operations proto_net_ops = {
3242 .init = proto_init_net,
3243 .exit = proto_exit_net,
3246 static int __init proto_init(void)
3248 return register_pernet_subsys(&proto_net_ops);
3251 subsys_initcall(proto_init);
3253 #endif /* PROC_FS */
3255 #ifdef CONFIG_NET_RX_BUSY_POLL
3256 bool sk_busy_loop_end(void *p, unsigned long start_time)
3258 struct sock *sk = p;
3260 return !skb_queue_empty(&sk->sk_receive_queue) ||
3261 sk_busy_loop_timeout(sk, start_time);
3263 EXPORT_SYMBOL(sk_busy_loop_end);
3264 #endif /* CONFIG_NET_RX_BUSY_POLL */