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:
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
210 static const char *const af_family_key_strings[AF_MAX+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
225 "sk_lock-AF_QIPCRTR", "sk_lock-AF_SMC" , "sk_lock-AF_MAX"
227 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
242 "slock-AF_QIPCRTR", "slock-AF_SMC" , "slock-AF_MAX"
244 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
259 "clock-AF_QIPCRTR", "clock-AF_SMC" , "clock-AF_MAX"
261 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
262 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
263 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
264 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
265 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
266 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
267 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
268 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
269 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
270 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
271 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
272 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
273 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
274 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
275 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
276 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
278 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
279 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
280 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
281 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
282 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
283 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
284 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
285 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
286 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
287 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
288 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
289 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
290 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
291 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
292 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
293 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
295 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
296 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
297 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
298 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
299 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
300 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
301 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
302 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
303 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
304 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
305 "elock-27" , "elock-28" , "elock-AF_CAN" ,
306 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
307 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
308 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
309 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
310 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
314 * sk_callback_lock and sk queues locking rules are per-address-family,
315 * so split the lock classes by using a per-AF key:
317 static struct lock_class_key af_callback_keys[AF_MAX];
318 static struct lock_class_key af_rlock_keys[AF_MAX];
319 static struct lock_class_key af_wlock_keys[AF_MAX];
320 static struct lock_class_key af_elock_keys[AF_MAX];
322 /* Take into consideration the size of the struct sk_buff overhead in the
323 * determination of these values, since that is non-constant across
324 * platforms. This makes socket queueing behavior and performance
325 * not depend upon such differences.
327 #define _SK_MEM_PACKETS 256
328 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
329 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
330 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
332 /* Run time adjustable parameters. */
333 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
334 EXPORT_SYMBOL(sysctl_wmem_max);
335 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
336 EXPORT_SYMBOL(sysctl_rmem_max);
337 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
338 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
340 /* Maximal space eaten by iovec or ancillary data plus some space */
341 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
342 EXPORT_SYMBOL(sysctl_optmem_max);
344 int sysctl_tstamp_allow_data __read_mostly = 1;
346 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
347 EXPORT_SYMBOL_GPL(memalloc_socks);
350 * sk_set_memalloc - sets %SOCK_MEMALLOC
351 * @sk: socket to set it on
353 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
354 * It's the responsibility of the admin to adjust min_free_kbytes
355 * to meet the requirements
357 void sk_set_memalloc(struct sock *sk)
359 sock_set_flag(sk, SOCK_MEMALLOC);
360 sk->sk_allocation |= __GFP_MEMALLOC;
361 static_key_slow_inc(&memalloc_socks);
363 EXPORT_SYMBOL_GPL(sk_set_memalloc);
365 void sk_clear_memalloc(struct sock *sk)
367 sock_reset_flag(sk, SOCK_MEMALLOC);
368 sk->sk_allocation &= ~__GFP_MEMALLOC;
369 static_key_slow_dec(&memalloc_socks);
372 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
373 * progress of swapping. SOCK_MEMALLOC may be cleared while
374 * it has rmem allocations due to the last swapfile being deactivated
375 * but there is a risk that the socket is unusable due to exceeding
376 * the rmem limits. Reclaim the reserves and obey rmem limits again.
380 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
382 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
385 unsigned long pflags = current->flags;
387 /* these should have been dropped before queueing */
388 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
390 current->flags |= PF_MEMALLOC;
391 ret = sk->sk_backlog_rcv(sk, skb);
392 tsk_restore_flags(current, pflags, PF_MEMALLOC);
396 EXPORT_SYMBOL(__sk_backlog_rcv);
398 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
402 if (optlen < sizeof(tv))
404 if (copy_from_user(&tv, optval, sizeof(tv)))
406 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
410 static int warned __read_mostly;
413 if (warned < 10 && net_ratelimit()) {
415 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
416 __func__, current->comm, task_pid_nr(current));
420 *timeo_p = MAX_SCHEDULE_TIMEOUT;
421 if (tv.tv_sec == 0 && tv.tv_usec == 0)
423 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
424 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
428 static void sock_warn_obsolete_bsdism(const char *name)
431 static char warncomm[TASK_COMM_LEN];
432 if (strcmp(warncomm, current->comm) && warned < 5) {
433 strcpy(warncomm, current->comm);
434 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
440 static bool sock_needs_netstamp(const struct sock *sk)
442 switch (sk->sk_family) {
451 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
453 if (sk->sk_flags & flags) {
454 sk->sk_flags &= ~flags;
455 if (sock_needs_netstamp(sk) &&
456 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
457 net_disable_timestamp();
462 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
465 struct sk_buff_head *list = &sk->sk_receive_queue;
467 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
468 atomic_inc(&sk->sk_drops);
469 trace_sock_rcvqueue_full(sk, skb);
473 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
474 atomic_inc(&sk->sk_drops);
479 skb_set_owner_r(skb, sk);
481 /* we escape from rcu protected region, make sure we dont leak
486 spin_lock_irqsave(&list->lock, flags);
487 sock_skb_set_dropcount(sk, skb);
488 __skb_queue_tail(list, skb);
489 spin_unlock_irqrestore(&list->lock, flags);
491 if (!sock_flag(sk, SOCK_DEAD))
492 sk->sk_data_ready(sk);
495 EXPORT_SYMBOL(__sock_queue_rcv_skb);
497 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
501 err = sk_filter(sk, skb);
505 return __sock_queue_rcv_skb(sk, skb);
507 EXPORT_SYMBOL(sock_queue_rcv_skb);
509 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
510 const int nested, unsigned int trim_cap, bool refcounted)
512 int rc = NET_RX_SUCCESS;
514 if (sk_filter_trim_cap(sk, skb, trim_cap))
515 goto discard_and_relse;
519 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
520 atomic_inc(&sk->sk_drops);
521 goto discard_and_relse;
524 bh_lock_sock_nested(sk);
527 if (!sock_owned_by_user(sk)) {
529 * trylock + unlock semantics:
531 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
533 rc = sk_backlog_rcv(sk, skb);
535 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
536 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
538 atomic_inc(&sk->sk_drops);
539 goto discard_and_relse;
551 EXPORT_SYMBOL(__sk_receive_skb);
553 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
555 struct dst_entry *dst = __sk_dst_get(sk);
557 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
558 sk_tx_queue_clear(sk);
559 sk->sk_dst_pending_confirm = 0;
560 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
567 EXPORT_SYMBOL(__sk_dst_check);
569 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
571 struct dst_entry *dst = sk_dst_get(sk);
573 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
581 EXPORT_SYMBOL(sk_dst_check);
583 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
586 int ret = -ENOPROTOOPT;
587 #ifdef CONFIG_NETDEVICES
588 struct net *net = sock_net(sk);
589 char devname[IFNAMSIZ];
594 if (!ns_capable(net->user_ns, CAP_NET_RAW))
601 /* Bind this socket to a particular device like "eth0",
602 * as specified in the passed interface name. If the
603 * name is "" or the option length is zero the socket
606 if (optlen > IFNAMSIZ - 1)
607 optlen = IFNAMSIZ - 1;
608 memset(devname, 0, sizeof(devname));
611 if (copy_from_user(devname, optval, optlen))
615 if (devname[0] != '\0') {
616 struct net_device *dev;
619 dev = dev_get_by_name_rcu(net, devname);
621 index = dev->ifindex;
629 sk->sk_bound_dev_if = index;
641 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
642 int __user *optlen, int len)
644 int ret = -ENOPROTOOPT;
645 #ifdef CONFIG_NETDEVICES
646 struct net *net = sock_net(sk);
647 char devname[IFNAMSIZ];
649 if (sk->sk_bound_dev_if == 0) {
658 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
662 len = strlen(devname) + 1;
665 if (copy_to_user(optval, devname, len))
670 if (put_user(len, optlen))
681 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
684 sock_set_flag(sk, bit);
686 sock_reset_flag(sk, bit);
689 bool sk_mc_loop(struct sock *sk)
691 if (dev_recursion_level())
695 switch (sk->sk_family) {
697 return inet_sk(sk)->mc_loop;
698 #if IS_ENABLED(CONFIG_IPV6)
700 return inet6_sk(sk)->mc_loop;
706 EXPORT_SYMBOL(sk_mc_loop);
709 * This is meant for all protocols to use and covers goings on
710 * at the socket level. Everything here is generic.
713 int sock_setsockopt(struct socket *sock, int level, int optname,
714 char __user *optval, unsigned int optlen)
716 struct sock *sk = sock->sk;
723 * Options without arguments
726 if (optname == SO_BINDTODEVICE)
727 return sock_setbindtodevice(sk, optval, optlen);
729 if (optlen < sizeof(int))
732 if (get_user(val, (int __user *)optval))
735 valbool = val ? 1 : 0;
741 if (val && !capable(CAP_NET_ADMIN))
744 sock_valbool_flag(sk, SOCK_DBG, valbool);
747 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
750 sk->sk_reuseport = valbool;
759 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
762 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
765 /* Don't error on this BSD doesn't and if you think
766 * about it this is right. Otherwise apps have to
767 * play 'guess the biggest size' games. RCVBUF/SNDBUF
768 * are treated in BSD as hints
770 val = min_t(u32, val, sysctl_wmem_max);
772 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
773 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
774 /* Wake up sending tasks if we upped the value. */
775 sk->sk_write_space(sk);
779 if (!capable(CAP_NET_ADMIN)) {
786 /* Don't error on this BSD doesn't and if you think
787 * about it this is right. Otherwise apps have to
788 * play 'guess the biggest size' games. RCVBUF/SNDBUF
789 * are treated in BSD as hints
791 val = min_t(u32, val, sysctl_rmem_max);
793 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
795 * We double it on the way in to account for
796 * "struct sk_buff" etc. overhead. Applications
797 * assume that the SO_RCVBUF setting they make will
798 * allow that much actual data to be received on that
801 * Applications are unaware that "struct sk_buff" and
802 * other overheads allocate from the receive buffer
803 * during socket buffer allocation.
805 * And after considering the possible alternatives,
806 * returning the value we actually used in getsockopt
807 * is the most desirable behavior.
809 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
813 if (!capable(CAP_NET_ADMIN)) {
820 if (sk->sk_prot->keepalive)
821 sk->sk_prot->keepalive(sk, valbool);
822 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
826 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
830 sk->sk_no_check_tx = valbool;
834 if ((val >= 0 && val <= 6) ||
835 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
836 sk->sk_priority = val;
842 if (optlen < sizeof(ling)) {
843 ret = -EINVAL; /* 1003.1g */
846 if (copy_from_user(&ling, optval, sizeof(ling))) {
851 sock_reset_flag(sk, SOCK_LINGER);
853 #if (BITS_PER_LONG == 32)
854 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
855 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
858 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
859 sock_set_flag(sk, SOCK_LINGER);
864 sock_warn_obsolete_bsdism("setsockopt");
869 set_bit(SOCK_PASSCRED, &sock->flags);
871 clear_bit(SOCK_PASSCRED, &sock->flags);
877 if (optname == SO_TIMESTAMP)
878 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
880 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
881 sock_set_flag(sk, SOCK_RCVTSTAMP);
882 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
884 sock_reset_flag(sk, SOCK_RCVTSTAMP);
885 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
889 case SO_TIMESTAMPING:
890 if (val & ~SOF_TIMESTAMPING_MASK) {
895 if (val & SOF_TIMESTAMPING_OPT_ID &&
896 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
897 if (sk->sk_protocol == IPPROTO_TCP &&
898 sk->sk_type == SOCK_STREAM) {
899 if ((1 << sk->sk_state) &
900 (TCPF_CLOSE | TCPF_LISTEN)) {
904 sk->sk_tskey = tcp_sk(sk)->snd_una;
910 if (val & SOF_TIMESTAMPING_OPT_STATS &&
911 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
916 sk->sk_tsflags = val;
917 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
918 sock_enable_timestamp(sk,
919 SOCK_TIMESTAMPING_RX_SOFTWARE);
921 sock_disable_timestamp(sk,
922 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
928 sk->sk_rcvlowat = val ? : 1;
932 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
936 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
939 case SO_ATTACH_FILTER:
941 if (optlen == sizeof(struct sock_fprog)) {
942 struct sock_fprog fprog;
945 if (copy_from_user(&fprog, optval, sizeof(fprog)))
948 ret = sk_attach_filter(&fprog, sk);
954 if (optlen == sizeof(u32)) {
958 if (copy_from_user(&ufd, optval, sizeof(ufd)))
961 ret = sk_attach_bpf(ufd, sk);
965 case SO_ATTACH_REUSEPORT_CBPF:
967 if (optlen == sizeof(struct sock_fprog)) {
968 struct sock_fprog fprog;
971 if (copy_from_user(&fprog, optval, sizeof(fprog)))
974 ret = sk_reuseport_attach_filter(&fprog, sk);
978 case SO_ATTACH_REUSEPORT_EBPF:
980 if (optlen == sizeof(u32)) {
984 if (copy_from_user(&ufd, optval, sizeof(ufd)))
987 ret = sk_reuseport_attach_bpf(ufd, sk);
991 case SO_DETACH_FILTER:
992 ret = sk_detach_filter(sk);
996 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
999 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1004 set_bit(SOCK_PASSSEC, &sock->flags);
1006 clear_bit(SOCK_PASSSEC, &sock->flags);
1009 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1016 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1019 case SO_WIFI_STATUS:
1020 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1024 if (sock->ops->set_peek_off)
1025 ret = sock->ops->set_peek_off(sk, val);
1031 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1034 case SO_SELECT_ERR_QUEUE:
1035 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1038 #ifdef CONFIG_NET_RX_BUSY_POLL
1040 /* allow unprivileged users to decrease the value */
1041 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1047 sk->sk_ll_usec = val;
1052 case SO_MAX_PACING_RATE:
1053 sk->sk_max_pacing_rate = val;
1054 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1055 sk->sk_max_pacing_rate);
1058 case SO_INCOMING_CPU:
1059 sk->sk_incoming_cpu = val;
1064 dst_negative_advice(sk);
1073 EXPORT_SYMBOL(sock_setsockopt);
1076 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1077 struct ucred *ucred)
1079 ucred->pid = pid_vnr(pid);
1080 ucred->uid = ucred->gid = -1;
1082 struct user_namespace *current_ns = current_user_ns();
1084 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1085 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1089 int sock_getsockopt(struct socket *sock, int level, int optname,
1090 char __user *optval, int __user *optlen)
1092 struct sock *sk = sock->sk;
1100 int lv = sizeof(int);
1103 if (get_user(len, optlen))
1108 memset(&v, 0, sizeof(v));
1112 v.val = sock_flag(sk, SOCK_DBG);
1116 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1120 v.val = sock_flag(sk, SOCK_BROADCAST);
1124 v.val = sk->sk_sndbuf;
1128 v.val = sk->sk_rcvbuf;
1132 v.val = sk->sk_reuse;
1136 v.val = sk->sk_reuseport;
1140 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1144 v.val = sk->sk_type;
1148 v.val = sk->sk_protocol;
1152 v.val = sk->sk_family;
1156 v.val = -sock_error(sk);
1158 v.val = xchg(&sk->sk_err_soft, 0);
1162 v.val = sock_flag(sk, SOCK_URGINLINE);
1166 v.val = sk->sk_no_check_tx;
1170 v.val = sk->sk_priority;
1174 lv = sizeof(v.ling);
1175 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1176 v.ling.l_linger = sk->sk_lingertime / HZ;
1180 sock_warn_obsolete_bsdism("getsockopt");
1184 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1185 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1188 case SO_TIMESTAMPNS:
1189 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1192 case SO_TIMESTAMPING:
1193 v.val = sk->sk_tsflags;
1197 lv = sizeof(struct timeval);
1198 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1202 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1203 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1208 lv = sizeof(struct timeval);
1209 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1213 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1214 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1219 v.val = sk->sk_rcvlowat;
1227 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1232 struct ucred peercred;
1233 if (len > sizeof(peercred))
1234 len = sizeof(peercred);
1235 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1236 if (copy_to_user(optval, &peercred, len))
1245 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1249 if (copy_to_user(optval, address, len))
1254 /* Dubious BSD thing... Probably nobody even uses it, but
1255 * the UNIX standard wants it for whatever reason... -DaveM
1258 v.val = sk->sk_state == TCP_LISTEN;
1262 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1266 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1269 v.val = sk->sk_mark;
1273 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1276 case SO_WIFI_STATUS:
1277 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1281 if (!sock->ops->set_peek_off)
1284 v.val = sk->sk_peek_off;
1287 v.val = sock_flag(sk, SOCK_NOFCS);
1290 case SO_BINDTODEVICE:
1291 return sock_getbindtodevice(sk, optval, optlen, len);
1294 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1300 case SO_LOCK_FILTER:
1301 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1304 case SO_BPF_EXTENSIONS:
1305 v.val = bpf_tell_extensions();
1308 case SO_SELECT_ERR_QUEUE:
1309 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1312 #ifdef CONFIG_NET_RX_BUSY_POLL
1314 v.val = sk->sk_ll_usec;
1318 case SO_MAX_PACING_RATE:
1319 v.val = sk->sk_max_pacing_rate;
1322 case SO_INCOMING_CPU:
1323 v.val = sk->sk_incoming_cpu;
1327 /* We implement the SO_SNDLOWAT etc to not be settable
1330 return -ENOPROTOOPT;
1335 if (copy_to_user(optval, &v, len))
1338 if (put_user(len, optlen))
1344 * Initialize an sk_lock.
1346 * (We also register the sk_lock with the lock validator.)
1348 static inline void sock_lock_init(struct sock *sk)
1350 sock_lock_init_class_and_name(sk,
1351 af_family_slock_key_strings[sk->sk_family],
1352 af_family_slock_keys + sk->sk_family,
1353 af_family_key_strings[sk->sk_family],
1354 af_family_keys + sk->sk_family);
1358 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1359 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1360 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1362 static void sock_copy(struct sock *nsk, const struct sock *osk)
1364 #ifdef CONFIG_SECURITY_NETWORK
1365 void *sptr = nsk->sk_security;
1367 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1369 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1370 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1372 #ifdef CONFIG_SECURITY_NETWORK
1373 nsk->sk_security = sptr;
1374 security_sk_clone(osk, nsk);
1378 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1382 struct kmem_cache *slab;
1386 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1389 if (priority & __GFP_ZERO)
1390 sk_prot_clear_nulls(sk, prot->obj_size);
1392 sk = kmalloc(prot->obj_size, priority);
1395 kmemcheck_annotate_bitfield(sk, flags);
1397 if (security_sk_alloc(sk, family, priority))
1400 if (!try_module_get(prot->owner))
1402 sk_tx_queue_clear(sk);
1408 security_sk_free(sk);
1411 kmem_cache_free(slab, sk);
1417 static void sk_prot_free(struct proto *prot, struct sock *sk)
1419 struct kmem_cache *slab;
1420 struct module *owner;
1422 owner = prot->owner;
1425 cgroup_sk_free(&sk->sk_cgrp_data);
1426 mem_cgroup_sk_free(sk);
1427 security_sk_free(sk);
1429 kmem_cache_free(slab, sk);
1436 * sk_alloc - All socket objects are allocated here
1437 * @net: the applicable net namespace
1438 * @family: protocol family
1439 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1440 * @prot: struct proto associated with this new sock instance
1441 * @kern: is this to be a kernel socket?
1443 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1444 struct proto *prot, int kern)
1448 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1450 sk->sk_family = family;
1452 * See comment in struct sock definition to understand
1453 * why we need sk_prot_creator -acme
1455 sk->sk_prot = sk->sk_prot_creator = prot;
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)
2352 EXPORT_SYMBOL(sock_no_accept);
2354 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2359 EXPORT_SYMBOL(sock_no_getname);
2361 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2365 EXPORT_SYMBOL(sock_no_poll);
2367 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2371 EXPORT_SYMBOL(sock_no_ioctl);
2373 int sock_no_listen(struct socket *sock, int backlog)
2377 EXPORT_SYMBOL(sock_no_listen);
2379 int sock_no_shutdown(struct socket *sock, int how)
2383 EXPORT_SYMBOL(sock_no_shutdown);
2385 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2386 char __user *optval, unsigned int optlen)
2390 EXPORT_SYMBOL(sock_no_setsockopt);
2392 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2393 char __user *optval, int __user *optlen)
2397 EXPORT_SYMBOL(sock_no_getsockopt);
2399 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2403 EXPORT_SYMBOL(sock_no_sendmsg);
2405 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2410 EXPORT_SYMBOL(sock_no_recvmsg);
2412 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2414 /* Mirror missing mmap method error code */
2417 EXPORT_SYMBOL(sock_no_mmap);
2419 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2422 struct msghdr msg = {.msg_flags = flags};
2424 char *kaddr = kmap(page);
2425 iov.iov_base = kaddr + offset;
2427 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2431 EXPORT_SYMBOL(sock_no_sendpage);
2434 * Default Socket Callbacks
2437 static void sock_def_wakeup(struct sock *sk)
2439 struct socket_wq *wq;
2442 wq = rcu_dereference(sk->sk_wq);
2443 if (skwq_has_sleeper(wq))
2444 wake_up_interruptible_all(&wq->wait);
2448 static void sock_def_error_report(struct sock *sk)
2450 struct socket_wq *wq;
2453 wq = rcu_dereference(sk->sk_wq);
2454 if (skwq_has_sleeper(wq))
2455 wake_up_interruptible_poll(&wq->wait, POLLERR);
2456 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2460 static void sock_def_readable(struct sock *sk)
2462 struct socket_wq *wq;
2465 wq = rcu_dereference(sk->sk_wq);
2466 if (skwq_has_sleeper(wq))
2467 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2468 POLLRDNORM | POLLRDBAND);
2469 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2473 static void sock_def_write_space(struct sock *sk)
2475 struct socket_wq *wq;
2479 /* Do not wake up a writer until he can make "significant"
2482 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2483 wq = rcu_dereference(sk->sk_wq);
2484 if (skwq_has_sleeper(wq))
2485 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2486 POLLWRNORM | POLLWRBAND);
2488 /* Should agree with poll, otherwise some programs break */
2489 if (sock_writeable(sk))
2490 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2496 static void sock_def_destruct(struct sock *sk)
2500 void sk_send_sigurg(struct sock *sk)
2502 if (sk->sk_socket && sk->sk_socket->file)
2503 if (send_sigurg(&sk->sk_socket->file->f_owner))
2504 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2506 EXPORT_SYMBOL(sk_send_sigurg);
2508 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2509 unsigned long expires)
2511 if (!mod_timer(timer, expires))
2514 EXPORT_SYMBOL(sk_reset_timer);
2516 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2518 if (del_timer(timer))
2521 EXPORT_SYMBOL(sk_stop_timer);
2523 void sock_init_data(struct socket *sock, struct sock *sk)
2526 sk->sk_send_head = NULL;
2528 init_timer(&sk->sk_timer);
2530 sk->sk_allocation = GFP_KERNEL;
2531 sk->sk_rcvbuf = sysctl_rmem_default;
2532 sk->sk_sndbuf = sysctl_wmem_default;
2533 sk->sk_state = TCP_CLOSE;
2534 sk_set_socket(sk, sock);
2536 sock_set_flag(sk, SOCK_ZAPPED);
2539 sk->sk_type = sock->type;
2540 sk->sk_wq = sock->wq;
2542 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2545 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2548 sk->sk_state_change = sock_def_wakeup;
2549 sk->sk_data_ready = sock_def_readable;
2550 sk->sk_write_space = sock_def_write_space;
2551 sk->sk_error_report = sock_def_error_report;
2552 sk->sk_destruct = sock_def_destruct;
2554 sk->sk_frag.page = NULL;
2555 sk->sk_frag.offset = 0;
2556 sk->sk_peek_off = -1;
2558 sk->sk_peer_pid = NULL;
2559 sk->sk_peer_cred = NULL;
2560 sk->sk_write_pending = 0;
2561 sk->sk_rcvlowat = 1;
2562 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2563 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2565 sk->sk_stamp = ktime_set(-1L, 0);
2567 #ifdef CONFIG_NET_RX_BUSY_POLL
2569 sk->sk_ll_usec = sysctl_net_busy_read;
2572 sk->sk_max_pacing_rate = ~0U;
2573 sk->sk_pacing_rate = ~0U;
2574 sk->sk_incoming_cpu = -1;
2576 * Before updating sk_refcnt, we must commit prior changes to memory
2577 * (Documentation/RCU/rculist_nulls.txt for details)
2580 atomic_set(&sk->sk_refcnt, 1);
2581 atomic_set(&sk->sk_drops, 0);
2583 EXPORT_SYMBOL(sock_init_data);
2585 void lock_sock_nested(struct sock *sk, int subclass)
2588 spin_lock_bh(&sk->sk_lock.slock);
2589 if (sk->sk_lock.owned)
2591 sk->sk_lock.owned = 1;
2592 spin_unlock(&sk->sk_lock.slock);
2594 * The sk_lock has mutex_lock() semantics here:
2596 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2599 EXPORT_SYMBOL(lock_sock_nested);
2601 void release_sock(struct sock *sk)
2603 spin_lock_bh(&sk->sk_lock.slock);
2604 if (sk->sk_backlog.tail)
2607 /* Warning : release_cb() might need to release sk ownership,
2608 * ie call sock_release_ownership(sk) before us.
2610 if (sk->sk_prot->release_cb)
2611 sk->sk_prot->release_cb(sk);
2613 sock_release_ownership(sk);
2614 if (waitqueue_active(&sk->sk_lock.wq))
2615 wake_up(&sk->sk_lock.wq);
2616 spin_unlock_bh(&sk->sk_lock.slock);
2618 EXPORT_SYMBOL(release_sock);
2621 * lock_sock_fast - fast version of lock_sock
2624 * This version should be used for very small section, where process wont block
2625 * return false if fast path is taken
2626 * sk_lock.slock locked, owned = 0, BH disabled
2627 * return true if slow path is taken
2628 * sk_lock.slock unlocked, owned = 1, BH enabled
2630 bool lock_sock_fast(struct sock *sk)
2633 spin_lock_bh(&sk->sk_lock.slock);
2635 if (!sk->sk_lock.owned)
2637 * Note : We must disable BH
2642 sk->sk_lock.owned = 1;
2643 spin_unlock(&sk->sk_lock.slock);
2645 * The sk_lock has mutex_lock() semantics here:
2647 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2651 EXPORT_SYMBOL(lock_sock_fast);
2653 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2656 if (!sock_flag(sk, SOCK_TIMESTAMP))
2657 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2658 tv = ktime_to_timeval(sk->sk_stamp);
2659 if (tv.tv_sec == -1)
2661 if (tv.tv_sec == 0) {
2662 sk->sk_stamp = ktime_get_real();
2663 tv = ktime_to_timeval(sk->sk_stamp);
2665 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2667 EXPORT_SYMBOL(sock_get_timestamp);
2669 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2672 if (!sock_flag(sk, SOCK_TIMESTAMP))
2673 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2674 ts = ktime_to_timespec(sk->sk_stamp);
2675 if (ts.tv_sec == -1)
2677 if (ts.tv_sec == 0) {
2678 sk->sk_stamp = ktime_get_real();
2679 ts = ktime_to_timespec(sk->sk_stamp);
2681 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2683 EXPORT_SYMBOL(sock_get_timestampns);
2685 void sock_enable_timestamp(struct sock *sk, int flag)
2687 if (!sock_flag(sk, flag)) {
2688 unsigned long previous_flags = sk->sk_flags;
2690 sock_set_flag(sk, flag);
2692 * we just set one of the two flags which require net
2693 * time stamping, but time stamping might have been on
2694 * already because of the other one
2696 if (sock_needs_netstamp(sk) &&
2697 !(previous_flags & SK_FLAGS_TIMESTAMP))
2698 net_enable_timestamp();
2702 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2703 int level, int type)
2705 struct sock_exterr_skb *serr;
2706 struct sk_buff *skb;
2710 skb = sock_dequeue_err_skb(sk);
2716 msg->msg_flags |= MSG_TRUNC;
2719 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2723 sock_recv_timestamp(msg, sk, skb);
2725 serr = SKB_EXT_ERR(skb);
2726 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2728 msg->msg_flags |= MSG_ERRQUEUE;
2736 EXPORT_SYMBOL(sock_recv_errqueue);
2739 * Get a socket option on an socket.
2741 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2742 * asynchronous errors should be reported by getsockopt. We assume
2743 * this means if you specify SO_ERROR (otherwise whats the point of it).
2745 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2746 char __user *optval, int __user *optlen)
2748 struct sock *sk = sock->sk;
2750 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2752 EXPORT_SYMBOL(sock_common_getsockopt);
2754 #ifdef CONFIG_COMPAT
2755 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2756 char __user *optval, int __user *optlen)
2758 struct sock *sk = sock->sk;
2760 if (sk->sk_prot->compat_getsockopt != NULL)
2761 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2763 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2765 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2768 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2771 struct sock *sk = sock->sk;
2775 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2776 flags & ~MSG_DONTWAIT, &addr_len);
2778 msg->msg_namelen = addr_len;
2781 EXPORT_SYMBOL(sock_common_recvmsg);
2784 * Set socket options on an inet socket.
2786 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2787 char __user *optval, unsigned int optlen)
2789 struct sock *sk = sock->sk;
2791 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2793 EXPORT_SYMBOL(sock_common_setsockopt);
2795 #ifdef CONFIG_COMPAT
2796 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2797 char __user *optval, unsigned int optlen)
2799 struct sock *sk = sock->sk;
2801 if (sk->sk_prot->compat_setsockopt != NULL)
2802 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2804 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2806 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2809 void sk_common_release(struct sock *sk)
2811 if (sk->sk_prot->destroy)
2812 sk->sk_prot->destroy(sk);
2815 * Observation: when sock_common_release is called, processes have
2816 * no access to socket. But net still has.
2817 * Step one, detach it from networking:
2819 * A. Remove from hash tables.
2822 sk->sk_prot->unhash(sk);
2825 * In this point socket cannot receive new packets, but it is possible
2826 * that some packets are in flight because some CPU runs receiver and
2827 * did hash table lookup before we unhashed socket. They will achieve
2828 * receive queue and will be purged by socket destructor.
2830 * Also we still have packets pending on receive queue and probably,
2831 * our own packets waiting in device queues. sock_destroy will drain
2832 * receive queue, but transmitted packets will delay socket destruction
2833 * until the last reference will be released.
2838 xfrm_sk_free_policy(sk);
2840 sk_refcnt_debug_release(sk);
2842 if (sk->sk_frag.page) {
2843 put_page(sk->sk_frag.page);
2844 sk->sk_frag.page = NULL;
2849 EXPORT_SYMBOL(sk_common_release);
2851 #ifdef CONFIG_PROC_FS
2852 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2854 int val[PROTO_INUSE_NR];
2857 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2859 #ifdef CONFIG_NET_NS
2860 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2862 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2864 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2866 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2868 int cpu, idx = prot->inuse_idx;
2871 for_each_possible_cpu(cpu)
2872 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2874 return res >= 0 ? res : 0;
2876 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2878 static int __net_init sock_inuse_init_net(struct net *net)
2880 net->core.inuse = alloc_percpu(struct prot_inuse);
2881 return net->core.inuse ? 0 : -ENOMEM;
2884 static void __net_exit sock_inuse_exit_net(struct net *net)
2886 free_percpu(net->core.inuse);
2889 static struct pernet_operations net_inuse_ops = {
2890 .init = sock_inuse_init_net,
2891 .exit = sock_inuse_exit_net,
2894 static __init int net_inuse_init(void)
2896 if (register_pernet_subsys(&net_inuse_ops))
2897 panic("Cannot initialize net inuse counters");
2902 core_initcall(net_inuse_init);
2904 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2906 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2908 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2910 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2912 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2914 int cpu, idx = prot->inuse_idx;
2917 for_each_possible_cpu(cpu)
2918 res += per_cpu(prot_inuse, cpu).val[idx];
2920 return res >= 0 ? res : 0;
2922 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2925 static void assign_proto_idx(struct proto *prot)
2927 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2929 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2930 pr_err("PROTO_INUSE_NR exhausted\n");
2934 set_bit(prot->inuse_idx, proto_inuse_idx);
2937 static void release_proto_idx(struct proto *prot)
2939 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2940 clear_bit(prot->inuse_idx, proto_inuse_idx);
2943 static inline void assign_proto_idx(struct proto *prot)
2947 static inline void release_proto_idx(struct proto *prot)
2952 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2956 kfree(rsk_prot->slab_name);
2957 rsk_prot->slab_name = NULL;
2958 kmem_cache_destroy(rsk_prot->slab);
2959 rsk_prot->slab = NULL;
2962 static int req_prot_init(const struct proto *prot)
2964 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2969 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2971 if (!rsk_prot->slab_name)
2974 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2975 rsk_prot->obj_size, 0,
2976 prot->slab_flags, NULL);
2978 if (!rsk_prot->slab) {
2979 pr_crit("%s: Can't create request sock SLAB cache!\n",
2986 int proto_register(struct proto *prot, int alloc_slab)
2989 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2990 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2993 if (prot->slab == NULL) {
2994 pr_crit("%s: Can't create sock SLAB cache!\n",
2999 if (req_prot_init(prot))
3000 goto out_free_request_sock_slab;
3002 if (prot->twsk_prot != NULL) {
3003 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3005 if (prot->twsk_prot->twsk_slab_name == NULL)
3006 goto out_free_request_sock_slab;
3008 prot->twsk_prot->twsk_slab =
3009 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3010 prot->twsk_prot->twsk_obj_size,
3014 if (prot->twsk_prot->twsk_slab == NULL)
3015 goto out_free_timewait_sock_slab_name;
3019 mutex_lock(&proto_list_mutex);
3020 list_add(&prot->node, &proto_list);
3021 assign_proto_idx(prot);
3022 mutex_unlock(&proto_list_mutex);
3025 out_free_timewait_sock_slab_name:
3026 kfree(prot->twsk_prot->twsk_slab_name);
3027 out_free_request_sock_slab:
3028 req_prot_cleanup(prot->rsk_prot);
3030 kmem_cache_destroy(prot->slab);
3035 EXPORT_SYMBOL(proto_register);
3037 void proto_unregister(struct proto *prot)
3039 mutex_lock(&proto_list_mutex);
3040 release_proto_idx(prot);
3041 list_del(&prot->node);
3042 mutex_unlock(&proto_list_mutex);
3044 kmem_cache_destroy(prot->slab);
3047 req_prot_cleanup(prot->rsk_prot);
3049 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3050 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3051 kfree(prot->twsk_prot->twsk_slab_name);
3052 prot->twsk_prot->twsk_slab = NULL;
3055 EXPORT_SYMBOL(proto_unregister);
3057 #ifdef CONFIG_PROC_FS
3058 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3059 __acquires(proto_list_mutex)
3061 mutex_lock(&proto_list_mutex);
3062 return seq_list_start_head(&proto_list, *pos);
3065 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3067 return seq_list_next(v, &proto_list, pos);
3070 static void proto_seq_stop(struct seq_file *seq, void *v)
3071 __releases(proto_list_mutex)
3073 mutex_unlock(&proto_list_mutex);
3076 static char proto_method_implemented(const void *method)
3078 return method == NULL ? 'n' : 'y';
3080 static long sock_prot_memory_allocated(struct proto *proto)
3082 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3085 static char *sock_prot_memory_pressure(struct proto *proto)
3087 return proto->memory_pressure != NULL ?
3088 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3091 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3094 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3095 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3098 sock_prot_inuse_get(seq_file_net(seq), proto),
3099 sock_prot_memory_allocated(proto),
3100 sock_prot_memory_pressure(proto),
3102 proto->slab == NULL ? "no" : "yes",
3103 module_name(proto->owner),
3104 proto_method_implemented(proto->close),
3105 proto_method_implemented(proto->connect),
3106 proto_method_implemented(proto->disconnect),
3107 proto_method_implemented(proto->accept),
3108 proto_method_implemented(proto->ioctl),
3109 proto_method_implemented(proto->init),
3110 proto_method_implemented(proto->destroy),
3111 proto_method_implemented(proto->shutdown),
3112 proto_method_implemented(proto->setsockopt),
3113 proto_method_implemented(proto->getsockopt),
3114 proto_method_implemented(proto->sendmsg),
3115 proto_method_implemented(proto->recvmsg),
3116 proto_method_implemented(proto->sendpage),
3117 proto_method_implemented(proto->bind),
3118 proto_method_implemented(proto->backlog_rcv),
3119 proto_method_implemented(proto->hash),
3120 proto_method_implemented(proto->unhash),
3121 proto_method_implemented(proto->get_port),
3122 proto_method_implemented(proto->enter_memory_pressure));
3125 static int proto_seq_show(struct seq_file *seq, void *v)
3127 if (v == &proto_list)
3128 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3137 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3139 proto_seq_printf(seq, list_entry(v, struct proto, node));
3143 static const struct seq_operations proto_seq_ops = {
3144 .start = proto_seq_start,
3145 .next = proto_seq_next,
3146 .stop = proto_seq_stop,
3147 .show = proto_seq_show,
3150 static int proto_seq_open(struct inode *inode, struct file *file)
3152 return seq_open_net(inode, file, &proto_seq_ops,
3153 sizeof(struct seq_net_private));
3156 static const struct file_operations proto_seq_fops = {
3157 .owner = THIS_MODULE,
3158 .open = proto_seq_open,
3160 .llseek = seq_lseek,
3161 .release = seq_release_net,
3164 static __net_init int proto_init_net(struct net *net)
3166 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3172 static __net_exit void proto_exit_net(struct net *net)
3174 remove_proc_entry("protocols", net->proc_net);
3178 static __net_initdata struct pernet_operations proto_net_ops = {
3179 .init = proto_init_net,
3180 .exit = proto_exit_net,
3183 static int __init proto_init(void)
3185 return register_pernet_subsys(&proto_net_ops);
3188 subsys_initcall(proto_init);
3190 #endif /* PROC_FS */