xref: /openbmc/linux/net/core/sock.c (revision b9f40e21ef4298650ab33e35740fa85bd57706d5)
1 /*
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.
5  *
6  *		Generic socket support routines. Memory allocators, socket lock/release
7  *		handler for protocols to use and generic option handler.
8  *
9  *
10  * Authors:	Ross Biro
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>
14  *
15  * Fixes:
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
36  *					TCP layer surgery.
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
64  *					(compatibility fix)
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
82  *
83  * To Fix:
84  *
85  *
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.
90  */
91 
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93 
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>
99 #include <linux/in.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>
120 
121 #include <asm/uaccess.h>
122 
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 
135 #include <linux/filter.h>
136 
137 #include <trace/events/sock.h>
138 
139 #ifdef CONFIG_INET
140 #include <net/tcp.h>
141 #endif
142 
143 #include <net/busy_poll.h>
144 
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
147 
148 /**
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
153  *
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.
157  */
158 bool sk_ns_capable(const struct sock *sk,
159 		   struct user_namespace *user_ns, int cap)
160 {
161 	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162 		ns_capable(user_ns, cap);
163 }
164 EXPORT_SYMBOL(sk_ns_capable);
165 
166 /**
167  * sk_capable - Socket global capability test
168  * @sk: Socket to use a capability on or through
169  * @cap: The global capbility to use
170  *
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
173  * namespaces.
174  */
175 bool sk_capable(const struct sock *sk, int cap)
176 {
177 	return sk_ns_capable(sk, &init_user_ns, cap);
178 }
179 EXPORT_SYMBOL(sk_capable);
180 
181 /**
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
185  *
186  * Test to see if the opener of the socket had when the socke was created
187  * and the current process has the capability @cap over the network namespace
188  * the socket is a member of.
189  */
190 bool sk_net_capable(const struct sock *sk, int cap)
191 {
192 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 }
194 EXPORT_SYMBOL(sk_net_capable);
195 
196 
197 #ifdef CONFIG_MEMCG_KMEM
198 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
199 {
200 	struct proto *proto;
201 	int ret = 0;
202 
203 	mutex_lock(&proto_list_mutex);
204 	list_for_each_entry(proto, &proto_list, node) {
205 		if (proto->init_cgroup) {
206 			ret = proto->init_cgroup(memcg, ss);
207 			if (ret)
208 				goto out;
209 		}
210 	}
211 
212 	mutex_unlock(&proto_list_mutex);
213 	return ret;
214 out:
215 	list_for_each_entry_continue_reverse(proto, &proto_list, node)
216 		if (proto->destroy_cgroup)
217 			proto->destroy_cgroup(memcg);
218 	mutex_unlock(&proto_list_mutex);
219 	return ret;
220 }
221 
222 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
223 {
224 	struct proto *proto;
225 
226 	mutex_lock(&proto_list_mutex);
227 	list_for_each_entry_reverse(proto, &proto_list, node)
228 		if (proto->destroy_cgroup)
229 			proto->destroy_cgroup(memcg);
230 	mutex_unlock(&proto_list_mutex);
231 }
232 #endif
233 
234 /*
235  * Each address family might have different locking rules, so we have
236  * one slock key per address family:
237  */
238 static struct lock_class_key af_family_keys[AF_MAX];
239 static struct lock_class_key af_family_slock_keys[AF_MAX];
240 
241 #if defined(CONFIG_MEMCG_KMEM)
242 struct static_key memcg_socket_limit_enabled;
243 EXPORT_SYMBOL(memcg_socket_limit_enabled);
244 #endif
245 
246 /*
247  * Make lock validator output more readable. (we pre-construct these
248  * strings build-time, so that runtime initialization of socket
249  * locks is fast):
250  */
251 static const char *const af_family_key_strings[AF_MAX+1] = {
252   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
253   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
254   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
255   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
256   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
257   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
258   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
259   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
260   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
261   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
262   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
263   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
264   "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
265   "sk_lock-AF_NFC"   , "sk_lock-AF_VSOCK"    , "sk_lock-AF_MAX"
266 };
267 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
268   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
269   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
270   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
271   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
272   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
273   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
274   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
275   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
276   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
277   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
278   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
279   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
280   "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
281   "slock-AF_NFC"   , "slock-AF_VSOCK"    ,"slock-AF_MAX"
282 };
283 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
284   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
285   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
286   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
287   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
288   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
289   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
290   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
291   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
292   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
293   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
294   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
295   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
296   "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
297   "clock-AF_NFC"   , "clock-AF_VSOCK"    , "clock-AF_MAX"
298 };
299 
300 /*
301  * sk_callback_lock locking rules are per-address-family,
302  * so split the lock classes by using a per-AF key:
303  */
304 static struct lock_class_key af_callback_keys[AF_MAX];
305 
306 /* Take into consideration the size of the struct sk_buff overhead in the
307  * determination of these values, since that is non-constant across
308  * platforms.  This makes socket queueing behavior and performance
309  * not depend upon such differences.
310  */
311 #define _SK_MEM_PACKETS		256
312 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
313 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
314 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 
316 /* Run time adjustable parameters. */
317 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
318 EXPORT_SYMBOL(sysctl_wmem_max);
319 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
320 EXPORT_SYMBOL(sysctl_rmem_max);
321 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
322 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
323 
324 /* Maximal space eaten by iovec or ancillary data plus some space */
325 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
326 EXPORT_SYMBOL(sysctl_optmem_max);
327 
328 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
329 EXPORT_SYMBOL_GPL(memalloc_socks);
330 
331 /**
332  * sk_set_memalloc - sets %SOCK_MEMALLOC
333  * @sk: socket to set it on
334  *
335  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
336  * It's the responsibility of the admin to adjust min_free_kbytes
337  * to meet the requirements
338  */
339 void sk_set_memalloc(struct sock *sk)
340 {
341 	sock_set_flag(sk, SOCK_MEMALLOC);
342 	sk->sk_allocation |= __GFP_MEMALLOC;
343 	static_key_slow_inc(&memalloc_socks);
344 }
345 EXPORT_SYMBOL_GPL(sk_set_memalloc);
346 
347 void sk_clear_memalloc(struct sock *sk)
348 {
349 	sock_reset_flag(sk, SOCK_MEMALLOC);
350 	sk->sk_allocation &= ~__GFP_MEMALLOC;
351 	static_key_slow_dec(&memalloc_socks);
352 
353 	/*
354 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
355 	 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
356 	 * it has rmem allocations there is a risk that the user of the
357 	 * socket cannot make forward progress due to exceeding the rmem
358 	 * limits. By rights, sk_clear_memalloc() should only be called
359 	 * on sockets being torn down but warn and reset the accounting if
360 	 * that assumption breaks.
361 	 */
362 	if (WARN_ON(sk->sk_forward_alloc))
363 		sk_mem_reclaim(sk);
364 }
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
366 
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
368 {
369 	int ret;
370 	unsigned long pflags = current->flags;
371 
372 	/* these should have been dropped before queueing */
373 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
374 
375 	current->flags |= PF_MEMALLOC;
376 	ret = sk->sk_backlog_rcv(sk, skb);
377 	tsk_restore_flags(current, pflags, PF_MEMALLOC);
378 
379 	return ret;
380 }
381 EXPORT_SYMBOL(__sk_backlog_rcv);
382 
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
384 {
385 	struct timeval tv;
386 
387 	if (optlen < sizeof(tv))
388 		return -EINVAL;
389 	if (copy_from_user(&tv, optval, sizeof(tv)))
390 		return -EFAULT;
391 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
392 		return -EDOM;
393 
394 	if (tv.tv_sec < 0) {
395 		static int warned __read_mostly;
396 
397 		*timeo_p = 0;
398 		if (warned < 10 && net_ratelimit()) {
399 			warned++;
400 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 				__func__, current->comm, task_pid_nr(current));
402 		}
403 		return 0;
404 	}
405 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
406 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
407 		return 0;
408 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 		*timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
410 	return 0;
411 }
412 
413 static void sock_warn_obsolete_bsdism(const char *name)
414 {
415 	static int warned;
416 	static char warncomm[TASK_COMM_LEN];
417 	if (strcmp(warncomm, current->comm) && warned < 5) {
418 		strcpy(warncomm,  current->comm);
419 		pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
420 			warncomm, name);
421 		warned++;
422 	}
423 }
424 
425 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
426 
427 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
428 {
429 	if (sk->sk_flags & flags) {
430 		sk->sk_flags &= ~flags;
431 		if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
432 			net_disable_timestamp();
433 	}
434 }
435 
436 
437 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
438 {
439 	int err;
440 	int skb_len;
441 	unsigned long flags;
442 	struct sk_buff_head *list = &sk->sk_receive_queue;
443 
444 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
445 		atomic_inc(&sk->sk_drops);
446 		trace_sock_rcvqueue_full(sk, skb);
447 		return -ENOMEM;
448 	}
449 
450 	err = sk_filter(sk, skb);
451 	if (err)
452 		return err;
453 
454 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
455 		atomic_inc(&sk->sk_drops);
456 		return -ENOBUFS;
457 	}
458 
459 	skb->dev = NULL;
460 	skb_set_owner_r(skb, sk);
461 
462 	/* Cache the SKB length before we tack it onto the receive
463 	 * queue.  Once it is added it no longer belongs to us and
464 	 * may be freed by other threads of control pulling packets
465 	 * from the queue.
466 	 */
467 	skb_len = skb->len;
468 
469 	/* we escape from rcu protected region, make sure we dont leak
470 	 * a norefcounted dst
471 	 */
472 	skb_dst_force(skb);
473 
474 	spin_lock_irqsave(&list->lock, flags);
475 	skb->dropcount = atomic_read(&sk->sk_drops);
476 	__skb_queue_tail(list, skb);
477 	spin_unlock_irqrestore(&list->lock, flags);
478 
479 	if (!sock_flag(sk, SOCK_DEAD))
480 		sk->sk_data_ready(sk);
481 	return 0;
482 }
483 EXPORT_SYMBOL(sock_queue_rcv_skb);
484 
485 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
486 {
487 	int rc = NET_RX_SUCCESS;
488 
489 	if (sk_filter(sk, skb))
490 		goto discard_and_relse;
491 
492 	skb->dev = NULL;
493 
494 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
495 		atomic_inc(&sk->sk_drops);
496 		goto discard_and_relse;
497 	}
498 	if (nested)
499 		bh_lock_sock_nested(sk);
500 	else
501 		bh_lock_sock(sk);
502 	if (!sock_owned_by_user(sk)) {
503 		/*
504 		 * trylock + unlock semantics:
505 		 */
506 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
507 
508 		rc = sk_backlog_rcv(sk, skb);
509 
510 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
511 	} else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
512 		bh_unlock_sock(sk);
513 		atomic_inc(&sk->sk_drops);
514 		goto discard_and_relse;
515 	}
516 
517 	bh_unlock_sock(sk);
518 out:
519 	sock_put(sk);
520 	return rc;
521 discard_and_relse:
522 	kfree_skb(skb);
523 	goto out;
524 }
525 EXPORT_SYMBOL(sk_receive_skb);
526 
527 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
528 {
529 	struct dst_entry *dst = __sk_dst_get(sk);
530 
531 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
532 		sk_tx_queue_clear(sk);
533 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
534 		dst_release(dst);
535 		return NULL;
536 	}
537 
538 	return dst;
539 }
540 EXPORT_SYMBOL(__sk_dst_check);
541 
542 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
543 {
544 	struct dst_entry *dst = sk_dst_get(sk);
545 
546 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
547 		sk_dst_reset(sk);
548 		dst_release(dst);
549 		return NULL;
550 	}
551 
552 	return dst;
553 }
554 EXPORT_SYMBOL(sk_dst_check);
555 
556 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
557 				int optlen)
558 {
559 	int ret = -ENOPROTOOPT;
560 #ifdef CONFIG_NETDEVICES
561 	struct net *net = sock_net(sk);
562 	char devname[IFNAMSIZ];
563 	int index;
564 
565 	/* Sorry... */
566 	ret = -EPERM;
567 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
568 		goto out;
569 
570 	ret = -EINVAL;
571 	if (optlen < 0)
572 		goto out;
573 
574 	/* Bind this socket to a particular device like "eth0",
575 	 * as specified in the passed interface name. If the
576 	 * name is "" or the option length is zero the socket
577 	 * is not bound.
578 	 */
579 	if (optlen > IFNAMSIZ - 1)
580 		optlen = IFNAMSIZ - 1;
581 	memset(devname, 0, sizeof(devname));
582 
583 	ret = -EFAULT;
584 	if (copy_from_user(devname, optval, optlen))
585 		goto out;
586 
587 	index = 0;
588 	if (devname[0] != '\0') {
589 		struct net_device *dev;
590 
591 		rcu_read_lock();
592 		dev = dev_get_by_name_rcu(net, devname);
593 		if (dev)
594 			index = dev->ifindex;
595 		rcu_read_unlock();
596 		ret = -ENODEV;
597 		if (!dev)
598 			goto out;
599 	}
600 
601 	lock_sock(sk);
602 	sk->sk_bound_dev_if = index;
603 	sk_dst_reset(sk);
604 	release_sock(sk);
605 
606 	ret = 0;
607 
608 out:
609 #endif
610 
611 	return ret;
612 }
613 
614 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
615 				int __user *optlen, int len)
616 {
617 	int ret = -ENOPROTOOPT;
618 #ifdef CONFIG_NETDEVICES
619 	struct net *net = sock_net(sk);
620 	char devname[IFNAMSIZ];
621 
622 	if (sk->sk_bound_dev_if == 0) {
623 		len = 0;
624 		goto zero;
625 	}
626 
627 	ret = -EINVAL;
628 	if (len < IFNAMSIZ)
629 		goto out;
630 
631 	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
632 	if (ret)
633 		goto out;
634 
635 	len = strlen(devname) + 1;
636 
637 	ret = -EFAULT;
638 	if (copy_to_user(optval, devname, len))
639 		goto out;
640 
641 zero:
642 	ret = -EFAULT;
643 	if (put_user(len, optlen))
644 		goto out;
645 
646 	ret = 0;
647 
648 out:
649 #endif
650 
651 	return ret;
652 }
653 
654 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
655 {
656 	if (valbool)
657 		sock_set_flag(sk, bit);
658 	else
659 		sock_reset_flag(sk, bit);
660 }
661 
662 /*
663  *	This is meant for all protocols to use and covers goings on
664  *	at the socket level. Everything here is generic.
665  */
666 
667 int sock_setsockopt(struct socket *sock, int level, int optname,
668 		    char __user *optval, unsigned int optlen)
669 {
670 	struct sock *sk = sock->sk;
671 	int val;
672 	int valbool;
673 	struct linger ling;
674 	int ret = 0;
675 
676 	/*
677 	 *	Options without arguments
678 	 */
679 
680 	if (optname == SO_BINDTODEVICE)
681 		return sock_setbindtodevice(sk, optval, optlen);
682 
683 	if (optlen < sizeof(int))
684 		return -EINVAL;
685 
686 	if (get_user(val, (int __user *)optval))
687 		return -EFAULT;
688 
689 	valbool = val ? 1 : 0;
690 
691 	lock_sock(sk);
692 
693 	switch (optname) {
694 	case SO_DEBUG:
695 		if (val && !capable(CAP_NET_ADMIN))
696 			ret = -EACCES;
697 		else
698 			sock_valbool_flag(sk, SOCK_DBG, valbool);
699 		break;
700 	case SO_REUSEADDR:
701 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
702 		break;
703 	case SO_REUSEPORT:
704 		sk->sk_reuseport = valbool;
705 		break;
706 	case SO_TYPE:
707 	case SO_PROTOCOL:
708 	case SO_DOMAIN:
709 	case SO_ERROR:
710 		ret = -ENOPROTOOPT;
711 		break;
712 	case SO_DONTROUTE:
713 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
714 		break;
715 	case SO_BROADCAST:
716 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
717 		break;
718 	case SO_SNDBUF:
719 		/* Don't error on this BSD doesn't and if you think
720 		 * about it this is right. Otherwise apps have to
721 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
722 		 * are treated in BSD as hints
723 		 */
724 		val = min_t(u32, val, sysctl_wmem_max);
725 set_sndbuf:
726 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
727 		sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
728 		/* Wake up sending tasks if we upped the value. */
729 		sk->sk_write_space(sk);
730 		break;
731 
732 	case SO_SNDBUFFORCE:
733 		if (!capable(CAP_NET_ADMIN)) {
734 			ret = -EPERM;
735 			break;
736 		}
737 		goto set_sndbuf;
738 
739 	case SO_RCVBUF:
740 		/* Don't error on this BSD doesn't and if you think
741 		 * about it this is right. Otherwise apps have to
742 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
743 		 * are treated in BSD as hints
744 		 */
745 		val = min_t(u32, val, sysctl_rmem_max);
746 set_rcvbuf:
747 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
748 		/*
749 		 * We double it on the way in to account for
750 		 * "struct sk_buff" etc. overhead.   Applications
751 		 * assume that the SO_RCVBUF setting they make will
752 		 * allow that much actual data to be received on that
753 		 * socket.
754 		 *
755 		 * Applications are unaware that "struct sk_buff" and
756 		 * other overheads allocate from the receive buffer
757 		 * during socket buffer allocation.
758 		 *
759 		 * And after considering the possible alternatives,
760 		 * returning the value we actually used in getsockopt
761 		 * is the most desirable behavior.
762 		 */
763 		sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
764 		break;
765 
766 	case SO_RCVBUFFORCE:
767 		if (!capable(CAP_NET_ADMIN)) {
768 			ret = -EPERM;
769 			break;
770 		}
771 		goto set_rcvbuf;
772 
773 	case SO_KEEPALIVE:
774 #ifdef CONFIG_INET
775 		if (sk->sk_protocol == IPPROTO_TCP &&
776 		    sk->sk_type == SOCK_STREAM)
777 			tcp_set_keepalive(sk, valbool);
778 #endif
779 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
780 		break;
781 
782 	case SO_OOBINLINE:
783 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
784 		break;
785 
786 	case SO_NO_CHECK:
787 		sk->sk_no_check_tx = valbool;
788 		break;
789 
790 	case SO_PRIORITY:
791 		if ((val >= 0 && val <= 6) ||
792 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
793 			sk->sk_priority = val;
794 		else
795 			ret = -EPERM;
796 		break;
797 
798 	case SO_LINGER:
799 		if (optlen < sizeof(ling)) {
800 			ret = -EINVAL;	/* 1003.1g */
801 			break;
802 		}
803 		if (copy_from_user(&ling, optval, sizeof(ling))) {
804 			ret = -EFAULT;
805 			break;
806 		}
807 		if (!ling.l_onoff)
808 			sock_reset_flag(sk, SOCK_LINGER);
809 		else {
810 #if (BITS_PER_LONG == 32)
811 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
812 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
813 			else
814 #endif
815 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
816 			sock_set_flag(sk, SOCK_LINGER);
817 		}
818 		break;
819 
820 	case SO_BSDCOMPAT:
821 		sock_warn_obsolete_bsdism("setsockopt");
822 		break;
823 
824 	case SO_PASSCRED:
825 		if (valbool)
826 			set_bit(SOCK_PASSCRED, &sock->flags);
827 		else
828 			clear_bit(SOCK_PASSCRED, &sock->flags);
829 		break;
830 
831 	case SO_TIMESTAMP:
832 	case SO_TIMESTAMPNS:
833 		if (valbool)  {
834 			if (optname == SO_TIMESTAMP)
835 				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
836 			else
837 				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
838 			sock_set_flag(sk, SOCK_RCVTSTAMP);
839 			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
840 		} else {
841 			sock_reset_flag(sk, SOCK_RCVTSTAMP);
842 			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
843 		}
844 		break;
845 
846 	case SO_TIMESTAMPING:
847 		if (val & ~SOF_TIMESTAMPING_MASK) {
848 			ret = -EINVAL;
849 			break;
850 		}
851 		sk->sk_tsflags = val;
852 		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
853 			sock_enable_timestamp(sk,
854 					      SOCK_TIMESTAMPING_RX_SOFTWARE);
855 		else
856 			sock_disable_timestamp(sk,
857 					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
858 		break;
859 
860 	case SO_RCVLOWAT:
861 		if (val < 0)
862 			val = INT_MAX;
863 		sk->sk_rcvlowat = val ? : 1;
864 		break;
865 
866 	case SO_RCVTIMEO:
867 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
868 		break;
869 
870 	case SO_SNDTIMEO:
871 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
872 		break;
873 
874 	case SO_ATTACH_FILTER:
875 		ret = -EINVAL;
876 		if (optlen == sizeof(struct sock_fprog)) {
877 			struct sock_fprog fprog;
878 
879 			ret = -EFAULT;
880 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
881 				break;
882 
883 			ret = sk_attach_filter(&fprog, sk);
884 		}
885 		break;
886 
887 	case SO_DETACH_FILTER:
888 		ret = sk_detach_filter(sk);
889 		break;
890 
891 	case SO_LOCK_FILTER:
892 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
893 			ret = -EPERM;
894 		else
895 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
896 		break;
897 
898 	case SO_PASSSEC:
899 		if (valbool)
900 			set_bit(SOCK_PASSSEC, &sock->flags);
901 		else
902 			clear_bit(SOCK_PASSSEC, &sock->flags);
903 		break;
904 	case SO_MARK:
905 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
906 			ret = -EPERM;
907 		else
908 			sk->sk_mark = val;
909 		break;
910 
911 		/* We implement the SO_SNDLOWAT etc to
912 		   not be settable (1003.1g 5.3) */
913 	case SO_RXQ_OVFL:
914 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
915 		break;
916 
917 	case SO_WIFI_STATUS:
918 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
919 		break;
920 
921 	case SO_PEEK_OFF:
922 		if (sock->ops->set_peek_off)
923 			ret = sock->ops->set_peek_off(sk, val);
924 		else
925 			ret = -EOPNOTSUPP;
926 		break;
927 
928 	case SO_NOFCS:
929 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
930 		break;
931 
932 	case SO_SELECT_ERR_QUEUE:
933 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
934 		break;
935 
936 #ifdef CONFIG_NET_RX_BUSY_POLL
937 	case SO_BUSY_POLL:
938 		/* allow unprivileged users to decrease the value */
939 		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
940 			ret = -EPERM;
941 		else {
942 			if (val < 0)
943 				ret = -EINVAL;
944 			else
945 				sk->sk_ll_usec = val;
946 		}
947 		break;
948 #endif
949 
950 	case SO_MAX_PACING_RATE:
951 		sk->sk_max_pacing_rate = val;
952 		sk->sk_pacing_rate = min(sk->sk_pacing_rate,
953 					 sk->sk_max_pacing_rate);
954 		break;
955 
956 	default:
957 		ret = -ENOPROTOOPT;
958 		break;
959 	}
960 	release_sock(sk);
961 	return ret;
962 }
963 EXPORT_SYMBOL(sock_setsockopt);
964 
965 
966 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
967 			  struct ucred *ucred)
968 {
969 	ucred->pid = pid_vnr(pid);
970 	ucred->uid = ucred->gid = -1;
971 	if (cred) {
972 		struct user_namespace *current_ns = current_user_ns();
973 
974 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
975 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
976 	}
977 }
978 
979 int sock_getsockopt(struct socket *sock, int level, int optname,
980 		    char __user *optval, int __user *optlen)
981 {
982 	struct sock *sk = sock->sk;
983 
984 	union {
985 		int val;
986 		struct linger ling;
987 		struct timeval tm;
988 	} v;
989 
990 	int lv = sizeof(int);
991 	int len;
992 
993 	if (get_user(len, optlen))
994 		return -EFAULT;
995 	if (len < 0)
996 		return -EINVAL;
997 
998 	memset(&v, 0, sizeof(v));
999 
1000 	switch (optname) {
1001 	case SO_DEBUG:
1002 		v.val = sock_flag(sk, SOCK_DBG);
1003 		break;
1004 
1005 	case SO_DONTROUTE:
1006 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1007 		break;
1008 
1009 	case SO_BROADCAST:
1010 		v.val = sock_flag(sk, SOCK_BROADCAST);
1011 		break;
1012 
1013 	case SO_SNDBUF:
1014 		v.val = sk->sk_sndbuf;
1015 		break;
1016 
1017 	case SO_RCVBUF:
1018 		v.val = sk->sk_rcvbuf;
1019 		break;
1020 
1021 	case SO_REUSEADDR:
1022 		v.val = sk->sk_reuse;
1023 		break;
1024 
1025 	case SO_REUSEPORT:
1026 		v.val = sk->sk_reuseport;
1027 		break;
1028 
1029 	case SO_KEEPALIVE:
1030 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1031 		break;
1032 
1033 	case SO_TYPE:
1034 		v.val = sk->sk_type;
1035 		break;
1036 
1037 	case SO_PROTOCOL:
1038 		v.val = sk->sk_protocol;
1039 		break;
1040 
1041 	case SO_DOMAIN:
1042 		v.val = sk->sk_family;
1043 		break;
1044 
1045 	case SO_ERROR:
1046 		v.val = -sock_error(sk);
1047 		if (v.val == 0)
1048 			v.val = xchg(&sk->sk_err_soft, 0);
1049 		break;
1050 
1051 	case SO_OOBINLINE:
1052 		v.val = sock_flag(sk, SOCK_URGINLINE);
1053 		break;
1054 
1055 	case SO_NO_CHECK:
1056 		v.val = sk->sk_no_check_tx;
1057 		break;
1058 
1059 	case SO_PRIORITY:
1060 		v.val = sk->sk_priority;
1061 		break;
1062 
1063 	case SO_LINGER:
1064 		lv		= sizeof(v.ling);
1065 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1066 		v.ling.l_linger	= sk->sk_lingertime / HZ;
1067 		break;
1068 
1069 	case SO_BSDCOMPAT:
1070 		sock_warn_obsolete_bsdism("getsockopt");
1071 		break;
1072 
1073 	case SO_TIMESTAMP:
1074 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1075 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1076 		break;
1077 
1078 	case SO_TIMESTAMPNS:
1079 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1080 		break;
1081 
1082 	case SO_TIMESTAMPING:
1083 		v.val = sk->sk_tsflags;
1084 		break;
1085 
1086 	case SO_RCVTIMEO:
1087 		lv = sizeof(struct timeval);
1088 		if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1089 			v.tm.tv_sec = 0;
1090 			v.tm.tv_usec = 0;
1091 		} else {
1092 			v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1093 			v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1094 		}
1095 		break;
1096 
1097 	case SO_SNDTIMEO:
1098 		lv = sizeof(struct timeval);
1099 		if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1100 			v.tm.tv_sec = 0;
1101 			v.tm.tv_usec = 0;
1102 		} else {
1103 			v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1104 			v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1105 		}
1106 		break;
1107 
1108 	case SO_RCVLOWAT:
1109 		v.val = sk->sk_rcvlowat;
1110 		break;
1111 
1112 	case SO_SNDLOWAT:
1113 		v.val = 1;
1114 		break;
1115 
1116 	case SO_PASSCRED:
1117 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1118 		break;
1119 
1120 	case SO_PEERCRED:
1121 	{
1122 		struct ucred peercred;
1123 		if (len > sizeof(peercred))
1124 			len = sizeof(peercred);
1125 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1126 		if (copy_to_user(optval, &peercred, len))
1127 			return -EFAULT;
1128 		goto lenout;
1129 	}
1130 
1131 	case SO_PEERNAME:
1132 	{
1133 		char address[128];
1134 
1135 		if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1136 			return -ENOTCONN;
1137 		if (lv < len)
1138 			return -EINVAL;
1139 		if (copy_to_user(optval, address, len))
1140 			return -EFAULT;
1141 		goto lenout;
1142 	}
1143 
1144 	/* Dubious BSD thing... Probably nobody even uses it, but
1145 	 * the UNIX standard wants it for whatever reason... -DaveM
1146 	 */
1147 	case SO_ACCEPTCONN:
1148 		v.val = sk->sk_state == TCP_LISTEN;
1149 		break;
1150 
1151 	case SO_PASSSEC:
1152 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1153 		break;
1154 
1155 	case SO_PEERSEC:
1156 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1157 
1158 	case SO_MARK:
1159 		v.val = sk->sk_mark;
1160 		break;
1161 
1162 	case SO_RXQ_OVFL:
1163 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1164 		break;
1165 
1166 	case SO_WIFI_STATUS:
1167 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1168 		break;
1169 
1170 	case SO_PEEK_OFF:
1171 		if (!sock->ops->set_peek_off)
1172 			return -EOPNOTSUPP;
1173 
1174 		v.val = sk->sk_peek_off;
1175 		break;
1176 	case SO_NOFCS:
1177 		v.val = sock_flag(sk, SOCK_NOFCS);
1178 		break;
1179 
1180 	case SO_BINDTODEVICE:
1181 		return sock_getbindtodevice(sk, optval, optlen, len);
1182 
1183 	case SO_GET_FILTER:
1184 		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1185 		if (len < 0)
1186 			return len;
1187 
1188 		goto lenout;
1189 
1190 	case SO_LOCK_FILTER:
1191 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1192 		break;
1193 
1194 	case SO_BPF_EXTENSIONS:
1195 		v.val = bpf_tell_extensions();
1196 		break;
1197 
1198 	case SO_SELECT_ERR_QUEUE:
1199 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1200 		break;
1201 
1202 #ifdef CONFIG_NET_RX_BUSY_POLL
1203 	case SO_BUSY_POLL:
1204 		v.val = sk->sk_ll_usec;
1205 		break;
1206 #endif
1207 
1208 	case SO_MAX_PACING_RATE:
1209 		v.val = sk->sk_max_pacing_rate;
1210 		break;
1211 
1212 	default:
1213 		return -ENOPROTOOPT;
1214 	}
1215 
1216 	if (len > lv)
1217 		len = lv;
1218 	if (copy_to_user(optval, &v, len))
1219 		return -EFAULT;
1220 lenout:
1221 	if (put_user(len, optlen))
1222 		return -EFAULT;
1223 	return 0;
1224 }
1225 
1226 /*
1227  * Initialize an sk_lock.
1228  *
1229  * (We also register the sk_lock with the lock validator.)
1230  */
1231 static inline void sock_lock_init(struct sock *sk)
1232 {
1233 	sock_lock_init_class_and_name(sk,
1234 			af_family_slock_key_strings[sk->sk_family],
1235 			af_family_slock_keys + sk->sk_family,
1236 			af_family_key_strings[sk->sk_family],
1237 			af_family_keys + sk->sk_family);
1238 }
1239 
1240 /*
1241  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1242  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1243  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1244  */
1245 static void sock_copy(struct sock *nsk, const struct sock *osk)
1246 {
1247 #ifdef CONFIG_SECURITY_NETWORK
1248 	void *sptr = nsk->sk_security;
1249 #endif
1250 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1251 
1252 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1253 	       osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1254 
1255 #ifdef CONFIG_SECURITY_NETWORK
1256 	nsk->sk_security = sptr;
1257 	security_sk_clone(osk, nsk);
1258 #endif
1259 }
1260 
1261 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1262 {
1263 	unsigned long nulls1, nulls2;
1264 
1265 	nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1266 	nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1267 	if (nulls1 > nulls2)
1268 		swap(nulls1, nulls2);
1269 
1270 	if (nulls1 != 0)
1271 		memset((char *)sk, 0, nulls1);
1272 	memset((char *)sk + nulls1 + sizeof(void *), 0,
1273 	       nulls2 - nulls1 - sizeof(void *));
1274 	memset((char *)sk + nulls2 + sizeof(void *), 0,
1275 	       size - nulls2 - sizeof(void *));
1276 }
1277 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1278 
1279 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1280 		int family)
1281 {
1282 	struct sock *sk;
1283 	struct kmem_cache *slab;
1284 
1285 	slab = prot->slab;
1286 	if (slab != NULL) {
1287 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1288 		if (!sk)
1289 			return sk;
1290 		if (priority & __GFP_ZERO) {
1291 			if (prot->clear_sk)
1292 				prot->clear_sk(sk, prot->obj_size);
1293 			else
1294 				sk_prot_clear_nulls(sk, prot->obj_size);
1295 		}
1296 	} else
1297 		sk = kmalloc(prot->obj_size, priority);
1298 
1299 	if (sk != NULL) {
1300 		kmemcheck_annotate_bitfield(sk, flags);
1301 
1302 		if (security_sk_alloc(sk, family, priority))
1303 			goto out_free;
1304 
1305 		if (!try_module_get(prot->owner))
1306 			goto out_free_sec;
1307 		sk_tx_queue_clear(sk);
1308 	}
1309 
1310 	return sk;
1311 
1312 out_free_sec:
1313 	security_sk_free(sk);
1314 out_free:
1315 	if (slab != NULL)
1316 		kmem_cache_free(slab, sk);
1317 	else
1318 		kfree(sk);
1319 	return NULL;
1320 }
1321 
1322 static void sk_prot_free(struct proto *prot, struct sock *sk)
1323 {
1324 	struct kmem_cache *slab;
1325 	struct module *owner;
1326 
1327 	owner = prot->owner;
1328 	slab = prot->slab;
1329 
1330 	security_sk_free(sk);
1331 	if (slab != NULL)
1332 		kmem_cache_free(slab, sk);
1333 	else
1334 		kfree(sk);
1335 	module_put(owner);
1336 }
1337 
1338 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1339 void sock_update_netprioidx(struct sock *sk)
1340 {
1341 	if (in_interrupt())
1342 		return;
1343 
1344 	sk->sk_cgrp_prioidx = task_netprioidx(current);
1345 }
1346 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1347 #endif
1348 
1349 /**
1350  *	sk_alloc - All socket objects are allocated here
1351  *	@net: the applicable net namespace
1352  *	@family: protocol family
1353  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1354  *	@prot: struct proto associated with this new sock instance
1355  */
1356 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1357 		      struct proto *prot)
1358 {
1359 	struct sock *sk;
1360 
1361 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1362 	if (sk) {
1363 		sk->sk_family = family;
1364 		/*
1365 		 * See comment in struct sock definition to understand
1366 		 * why we need sk_prot_creator -acme
1367 		 */
1368 		sk->sk_prot = sk->sk_prot_creator = prot;
1369 		sock_lock_init(sk);
1370 		sock_net_set(sk, get_net(net));
1371 		atomic_set(&sk->sk_wmem_alloc, 1);
1372 
1373 		sock_update_classid(sk);
1374 		sock_update_netprioidx(sk);
1375 	}
1376 
1377 	return sk;
1378 }
1379 EXPORT_SYMBOL(sk_alloc);
1380 
1381 static void __sk_free(struct sock *sk)
1382 {
1383 	struct sk_filter *filter;
1384 
1385 	if (sk->sk_destruct)
1386 		sk->sk_destruct(sk);
1387 
1388 	filter = rcu_dereference_check(sk->sk_filter,
1389 				       atomic_read(&sk->sk_wmem_alloc) == 0);
1390 	if (filter) {
1391 		sk_filter_uncharge(sk, filter);
1392 		RCU_INIT_POINTER(sk->sk_filter, NULL);
1393 	}
1394 
1395 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1396 
1397 	if (atomic_read(&sk->sk_omem_alloc))
1398 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1399 			 __func__, atomic_read(&sk->sk_omem_alloc));
1400 
1401 	if (sk->sk_peer_cred)
1402 		put_cred(sk->sk_peer_cred);
1403 	put_pid(sk->sk_peer_pid);
1404 	put_net(sock_net(sk));
1405 	sk_prot_free(sk->sk_prot_creator, sk);
1406 }
1407 
1408 void sk_free(struct sock *sk)
1409 {
1410 	/*
1411 	 * We subtract one from sk_wmem_alloc and can know if
1412 	 * some packets are still in some tx queue.
1413 	 * If not null, sock_wfree() will call __sk_free(sk) later
1414 	 */
1415 	if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1416 		__sk_free(sk);
1417 }
1418 EXPORT_SYMBOL(sk_free);
1419 
1420 /*
1421  * Last sock_put should drop reference to sk->sk_net. It has already
1422  * been dropped in sk_change_net. Taking reference to stopping namespace
1423  * is not an option.
1424  * Take reference to a socket to remove it from hash _alive_ and after that
1425  * destroy it in the context of init_net.
1426  */
1427 void sk_release_kernel(struct sock *sk)
1428 {
1429 	if (sk == NULL || sk->sk_socket == NULL)
1430 		return;
1431 
1432 	sock_hold(sk);
1433 	sock_release(sk->sk_socket);
1434 	release_net(sock_net(sk));
1435 	sock_net_set(sk, get_net(&init_net));
1436 	sock_put(sk);
1437 }
1438 EXPORT_SYMBOL(sk_release_kernel);
1439 
1440 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1441 {
1442 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1443 		sock_update_memcg(newsk);
1444 }
1445 
1446 /**
1447  *	sk_clone_lock - clone a socket, and lock its clone
1448  *	@sk: the socket to clone
1449  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1450  *
1451  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1452  */
1453 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1454 {
1455 	struct sock *newsk;
1456 	bool is_charged = true;
1457 
1458 	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1459 	if (newsk != NULL) {
1460 		struct sk_filter *filter;
1461 
1462 		sock_copy(newsk, sk);
1463 
1464 		/* SANITY */
1465 		get_net(sock_net(newsk));
1466 		sk_node_init(&newsk->sk_node);
1467 		sock_lock_init(newsk);
1468 		bh_lock_sock(newsk);
1469 		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1470 		newsk->sk_backlog.len = 0;
1471 
1472 		atomic_set(&newsk->sk_rmem_alloc, 0);
1473 		/*
1474 		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1475 		 */
1476 		atomic_set(&newsk->sk_wmem_alloc, 1);
1477 		atomic_set(&newsk->sk_omem_alloc, 0);
1478 		skb_queue_head_init(&newsk->sk_receive_queue);
1479 		skb_queue_head_init(&newsk->sk_write_queue);
1480 #ifdef CONFIG_NET_DMA
1481 		skb_queue_head_init(&newsk->sk_async_wait_queue);
1482 #endif
1483 
1484 		spin_lock_init(&newsk->sk_dst_lock);
1485 		rwlock_init(&newsk->sk_callback_lock);
1486 		lockdep_set_class_and_name(&newsk->sk_callback_lock,
1487 				af_callback_keys + newsk->sk_family,
1488 				af_family_clock_key_strings[newsk->sk_family]);
1489 
1490 		newsk->sk_dst_cache	= NULL;
1491 		newsk->sk_wmem_queued	= 0;
1492 		newsk->sk_forward_alloc = 0;
1493 		newsk->sk_send_head	= NULL;
1494 		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1495 
1496 		sock_reset_flag(newsk, SOCK_DONE);
1497 		skb_queue_head_init(&newsk->sk_error_queue);
1498 
1499 		filter = rcu_dereference_protected(newsk->sk_filter, 1);
1500 		if (filter != NULL)
1501 			/* though it's an empty new sock, the charging may fail
1502 			 * if sysctl_optmem_max was changed between creation of
1503 			 * original socket and cloning
1504 			 */
1505 			is_charged = sk_filter_charge(newsk, filter);
1506 
1507 		if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1508 			/* It is still raw copy of parent, so invalidate
1509 			 * destructor and make plain sk_free() */
1510 			newsk->sk_destruct = NULL;
1511 			bh_unlock_sock(newsk);
1512 			sk_free(newsk);
1513 			newsk = NULL;
1514 			goto out;
1515 		}
1516 
1517 		newsk->sk_err	   = 0;
1518 		newsk->sk_priority = 0;
1519 		/*
1520 		 * Before updating sk_refcnt, we must commit prior changes to memory
1521 		 * (Documentation/RCU/rculist_nulls.txt for details)
1522 		 */
1523 		smp_wmb();
1524 		atomic_set(&newsk->sk_refcnt, 2);
1525 
1526 		/*
1527 		 * Increment the counter in the same struct proto as the master
1528 		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1529 		 * is the same as sk->sk_prot->socks, as this field was copied
1530 		 * with memcpy).
1531 		 *
1532 		 * This _changes_ the previous behaviour, where
1533 		 * tcp_create_openreq_child always was incrementing the
1534 		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1535 		 * to be taken into account in all callers. -acme
1536 		 */
1537 		sk_refcnt_debug_inc(newsk);
1538 		sk_set_socket(newsk, NULL);
1539 		newsk->sk_wq = NULL;
1540 
1541 		sk_update_clone(sk, newsk);
1542 
1543 		if (newsk->sk_prot->sockets_allocated)
1544 			sk_sockets_allocated_inc(newsk);
1545 
1546 		if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1547 			net_enable_timestamp();
1548 	}
1549 out:
1550 	return newsk;
1551 }
1552 EXPORT_SYMBOL_GPL(sk_clone_lock);
1553 
1554 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1555 {
1556 	__sk_dst_set(sk, dst);
1557 	sk->sk_route_caps = dst->dev->features;
1558 	if (sk->sk_route_caps & NETIF_F_GSO)
1559 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1560 	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1561 	if (sk_can_gso(sk)) {
1562 		if (dst->header_len) {
1563 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1564 		} else {
1565 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1566 			sk->sk_gso_max_size = dst->dev->gso_max_size;
1567 			sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1568 		}
1569 	}
1570 }
1571 EXPORT_SYMBOL_GPL(sk_setup_caps);
1572 
1573 /*
1574  *	Simple resource managers for sockets.
1575  */
1576 
1577 
1578 /*
1579  * Write buffer destructor automatically called from kfree_skb.
1580  */
1581 void sock_wfree(struct sk_buff *skb)
1582 {
1583 	struct sock *sk = skb->sk;
1584 	unsigned int len = skb->truesize;
1585 
1586 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1587 		/*
1588 		 * Keep a reference on sk_wmem_alloc, this will be released
1589 		 * after sk_write_space() call
1590 		 */
1591 		atomic_sub(len - 1, &sk->sk_wmem_alloc);
1592 		sk->sk_write_space(sk);
1593 		len = 1;
1594 	}
1595 	/*
1596 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1597 	 * could not do because of in-flight packets
1598 	 */
1599 	if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1600 		__sk_free(sk);
1601 }
1602 EXPORT_SYMBOL(sock_wfree);
1603 
1604 void skb_orphan_partial(struct sk_buff *skb)
1605 {
1606 	/* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1607 	 * so we do not completely orphan skb, but transfert all
1608 	 * accounted bytes but one, to avoid unexpected reorders.
1609 	 */
1610 	if (skb->destructor == sock_wfree
1611 #ifdef CONFIG_INET
1612 	    || skb->destructor == tcp_wfree
1613 #endif
1614 		) {
1615 		atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1616 		skb->truesize = 1;
1617 	} else {
1618 		skb_orphan(skb);
1619 	}
1620 }
1621 EXPORT_SYMBOL(skb_orphan_partial);
1622 
1623 /*
1624  * Read buffer destructor automatically called from kfree_skb.
1625  */
1626 void sock_rfree(struct sk_buff *skb)
1627 {
1628 	struct sock *sk = skb->sk;
1629 	unsigned int len = skb->truesize;
1630 
1631 	atomic_sub(len, &sk->sk_rmem_alloc);
1632 	sk_mem_uncharge(sk, len);
1633 }
1634 EXPORT_SYMBOL(sock_rfree);
1635 
1636 void sock_edemux(struct sk_buff *skb)
1637 {
1638 	struct sock *sk = skb->sk;
1639 
1640 #ifdef CONFIG_INET
1641 	if (sk->sk_state == TCP_TIME_WAIT)
1642 		inet_twsk_put(inet_twsk(sk));
1643 	else
1644 #endif
1645 		sock_put(sk);
1646 }
1647 EXPORT_SYMBOL(sock_edemux);
1648 
1649 kuid_t sock_i_uid(struct sock *sk)
1650 {
1651 	kuid_t uid;
1652 
1653 	read_lock_bh(&sk->sk_callback_lock);
1654 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1655 	read_unlock_bh(&sk->sk_callback_lock);
1656 	return uid;
1657 }
1658 EXPORT_SYMBOL(sock_i_uid);
1659 
1660 unsigned long sock_i_ino(struct sock *sk)
1661 {
1662 	unsigned long ino;
1663 
1664 	read_lock_bh(&sk->sk_callback_lock);
1665 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1666 	read_unlock_bh(&sk->sk_callback_lock);
1667 	return ino;
1668 }
1669 EXPORT_SYMBOL(sock_i_ino);
1670 
1671 /*
1672  * Allocate a skb from the socket's send buffer.
1673  */
1674 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1675 			     gfp_t priority)
1676 {
1677 	if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1678 		struct sk_buff *skb = alloc_skb(size, priority);
1679 		if (skb) {
1680 			skb_set_owner_w(skb, sk);
1681 			return skb;
1682 		}
1683 	}
1684 	return NULL;
1685 }
1686 EXPORT_SYMBOL(sock_wmalloc);
1687 
1688 /*
1689  * Allocate a memory block from the socket's option memory buffer.
1690  */
1691 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1692 {
1693 	if ((unsigned int)size <= sysctl_optmem_max &&
1694 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1695 		void *mem;
1696 		/* First do the add, to avoid the race if kmalloc
1697 		 * might sleep.
1698 		 */
1699 		atomic_add(size, &sk->sk_omem_alloc);
1700 		mem = kmalloc(size, priority);
1701 		if (mem)
1702 			return mem;
1703 		atomic_sub(size, &sk->sk_omem_alloc);
1704 	}
1705 	return NULL;
1706 }
1707 EXPORT_SYMBOL(sock_kmalloc);
1708 
1709 /*
1710  * Free an option memory block.
1711  */
1712 void sock_kfree_s(struct sock *sk, void *mem, int size)
1713 {
1714 	kfree(mem);
1715 	atomic_sub(size, &sk->sk_omem_alloc);
1716 }
1717 EXPORT_SYMBOL(sock_kfree_s);
1718 
1719 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1720    I think, these locks should be removed for datagram sockets.
1721  */
1722 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1723 {
1724 	DEFINE_WAIT(wait);
1725 
1726 	clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1727 	for (;;) {
1728 		if (!timeo)
1729 			break;
1730 		if (signal_pending(current))
1731 			break;
1732 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1733 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1734 		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1735 			break;
1736 		if (sk->sk_shutdown & SEND_SHUTDOWN)
1737 			break;
1738 		if (sk->sk_err)
1739 			break;
1740 		timeo = schedule_timeout(timeo);
1741 	}
1742 	finish_wait(sk_sleep(sk), &wait);
1743 	return timeo;
1744 }
1745 
1746 
1747 /*
1748  *	Generic send/receive buffer handlers
1749  */
1750 
1751 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1752 				     unsigned long data_len, int noblock,
1753 				     int *errcode, int max_page_order)
1754 {
1755 	struct sk_buff *skb = NULL;
1756 	unsigned long chunk;
1757 	gfp_t gfp_mask;
1758 	long timeo;
1759 	int err;
1760 	int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1761 	struct page *page;
1762 	int i;
1763 
1764 	err = -EMSGSIZE;
1765 	if (npages > MAX_SKB_FRAGS)
1766 		goto failure;
1767 
1768 	timeo = sock_sndtimeo(sk, noblock);
1769 	while (!skb) {
1770 		err = sock_error(sk);
1771 		if (err != 0)
1772 			goto failure;
1773 
1774 		err = -EPIPE;
1775 		if (sk->sk_shutdown & SEND_SHUTDOWN)
1776 			goto failure;
1777 
1778 		if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) {
1779 			set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1780 			set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1781 			err = -EAGAIN;
1782 			if (!timeo)
1783 				goto failure;
1784 			if (signal_pending(current))
1785 				goto interrupted;
1786 			timeo = sock_wait_for_wmem(sk, timeo);
1787 			continue;
1788 		}
1789 
1790 		err = -ENOBUFS;
1791 		gfp_mask = sk->sk_allocation;
1792 		if (gfp_mask & __GFP_WAIT)
1793 			gfp_mask |= __GFP_REPEAT;
1794 
1795 		skb = alloc_skb(header_len, gfp_mask);
1796 		if (!skb)
1797 			goto failure;
1798 
1799 		skb->truesize += data_len;
1800 
1801 		for (i = 0; npages > 0; i++) {
1802 			int order = max_page_order;
1803 
1804 			while (order) {
1805 				if (npages >= 1 << order) {
1806 					page = alloc_pages(sk->sk_allocation |
1807 							   __GFP_COMP |
1808 							   __GFP_NOWARN |
1809 							   __GFP_NORETRY,
1810 							   order);
1811 					if (page)
1812 						goto fill_page;
1813 				}
1814 				order--;
1815 			}
1816 			page = alloc_page(sk->sk_allocation);
1817 			if (!page)
1818 				goto failure;
1819 fill_page:
1820 			chunk = min_t(unsigned long, data_len,
1821 				      PAGE_SIZE << order);
1822 			skb_fill_page_desc(skb, i, page, 0, chunk);
1823 			data_len -= chunk;
1824 			npages -= 1 << order;
1825 		}
1826 	}
1827 
1828 	skb_set_owner_w(skb, sk);
1829 	return skb;
1830 
1831 interrupted:
1832 	err = sock_intr_errno(timeo);
1833 failure:
1834 	kfree_skb(skb);
1835 	*errcode = err;
1836 	return NULL;
1837 }
1838 EXPORT_SYMBOL(sock_alloc_send_pskb);
1839 
1840 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1841 				    int noblock, int *errcode)
1842 {
1843 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1844 }
1845 EXPORT_SYMBOL(sock_alloc_send_skb);
1846 
1847 /* On 32bit arches, an skb frag is limited to 2^15 */
1848 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
1849 
1850 /**
1851  * skb_page_frag_refill - check that a page_frag contains enough room
1852  * @sz: minimum size of the fragment we want to get
1853  * @pfrag: pointer to page_frag
1854  * @prio: priority for memory allocation
1855  *
1856  * Note: While this allocator tries to use high order pages, there is
1857  * no guarantee that allocations succeed. Therefore, @sz MUST be
1858  * less or equal than PAGE_SIZE.
1859  */
1860 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio)
1861 {
1862 	int order;
1863 
1864 	if (pfrag->page) {
1865 		if (atomic_read(&pfrag->page->_count) == 1) {
1866 			pfrag->offset = 0;
1867 			return true;
1868 		}
1869 		if (pfrag->offset + sz <= pfrag->size)
1870 			return true;
1871 		put_page(pfrag->page);
1872 	}
1873 
1874 	order = SKB_FRAG_PAGE_ORDER;
1875 	do {
1876 		gfp_t gfp = prio;
1877 
1878 		if (order)
1879 			gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY;
1880 		pfrag->page = alloc_pages(gfp, order);
1881 		if (likely(pfrag->page)) {
1882 			pfrag->offset = 0;
1883 			pfrag->size = PAGE_SIZE << order;
1884 			return true;
1885 		}
1886 	} while (--order >= 0);
1887 
1888 	return false;
1889 }
1890 EXPORT_SYMBOL(skb_page_frag_refill);
1891 
1892 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1893 {
1894 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1895 		return true;
1896 
1897 	sk_enter_memory_pressure(sk);
1898 	sk_stream_moderate_sndbuf(sk);
1899 	return false;
1900 }
1901 EXPORT_SYMBOL(sk_page_frag_refill);
1902 
1903 static void __lock_sock(struct sock *sk)
1904 	__releases(&sk->sk_lock.slock)
1905 	__acquires(&sk->sk_lock.slock)
1906 {
1907 	DEFINE_WAIT(wait);
1908 
1909 	for (;;) {
1910 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1911 					TASK_UNINTERRUPTIBLE);
1912 		spin_unlock_bh(&sk->sk_lock.slock);
1913 		schedule();
1914 		spin_lock_bh(&sk->sk_lock.slock);
1915 		if (!sock_owned_by_user(sk))
1916 			break;
1917 	}
1918 	finish_wait(&sk->sk_lock.wq, &wait);
1919 }
1920 
1921 static void __release_sock(struct sock *sk)
1922 	__releases(&sk->sk_lock.slock)
1923 	__acquires(&sk->sk_lock.slock)
1924 {
1925 	struct sk_buff *skb = sk->sk_backlog.head;
1926 
1927 	do {
1928 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1929 		bh_unlock_sock(sk);
1930 
1931 		do {
1932 			struct sk_buff *next = skb->next;
1933 
1934 			prefetch(next);
1935 			WARN_ON_ONCE(skb_dst_is_noref(skb));
1936 			skb->next = NULL;
1937 			sk_backlog_rcv(sk, skb);
1938 
1939 			/*
1940 			 * We are in process context here with softirqs
1941 			 * disabled, use cond_resched_softirq() to preempt.
1942 			 * This is safe to do because we've taken the backlog
1943 			 * queue private:
1944 			 */
1945 			cond_resched_softirq();
1946 
1947 			skb = next;
1948 		} while (skb != NULL);
1949 
1950 		bh_lock_sock(sk);
1951 	} while ((skb = sk->sk_backlog.head) != NULL);
1952 
1953 	/*
1954 	 * Doing the zeroing here guarantee we can not loop forever
1955 	 * while a wild producer attempts to flood us.
1956 	 */
1957 	sk->sk_backlog.len = 0;
1958 }
1959 
1960 /**
1961  * sk_wait_data - wait for data to arrive at sk_receive_queue
1962  * @sk:    sock to wait on
1963  * @timeo: for how long
1964  *
1965  * Now socket state including sk->sk_err is changed only under lock,
1966  * hence we may omit checks after joining wait queue.
1967  * We check receive queue before schedule() only as optimization;
1968  * it is very likely that release_sock() added new data.
1969  */
1970 int sk_wait_data(struct sock *sk, long *timeo)
1971 {
1972 	int rc;
1973 	DEFINE_WAIT(wait);
1974 
1975 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1976 	set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1977 	rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1978 	clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1979 	finish_wait(sk_sleep(sk), &wait);
1980 	return rc;
1981 }
1982 EXPORT_SYMBOL(sk_wait_data);
1983 
1984 /**
1985  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1986  *	@sk: socket
1987  *	@size: memory size to allocate
1988  *	@kind: allocation type
1989  *
1990  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1991  *	rmem allocation. This function assumes that protocols which have
1992  *	memory_pressure use sk_wmem_queued as write buffer accounting.
1993  */
1994 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1995 {
1996 	struct proto *prot = sk->sk_prot;
1997 	int amt = sk_mem_pages(size);
1998 	long allocated;
1999 	int parent_status = UNDER_LIMIT;
2000 
2001 	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2002 
2003 	allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2004 
2005 	/* Under limit. */
2006 	if (parent_status == UNDER_LIMIT &&
2007 			allocated <= sk_prot_mem_limits(sk, 0)) {
2008 		sk_leave_memory_pressure(sk);
2009 		return 1;
2010 	}
2011 
2012 	/* Under pressure. (we or our parents) */
2013 	if ((parent_status > SOFT_LIMIT) ||
2014 			allocated > sk_prot_mem_limits(sk, 1))
2015 		sk_enter_memory_pressure(sk);
2016 
2017 	/* Over hard limit (we or our parents) */
2018 	if ((parent_status == OVER_LIMIT) ||
2019 			(allocated > sk_prot_mem_limits(sk, 2)))
2020 		goto suppress_allocation;
2021 
2022 	/* guarantee minimum buffer size under pressure */
2023 	if (kind == SK_MEM_RECV) {
2024 		if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2025 			return 1;
2026 
2027 	} else { /* SK_MEM_SEND */
2028 		if (sk->sk_type == SOCK_STREAM) {
2029 			if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2030 				return 1;
2031 		} else if (atomic_read(&sk->sk_wmem_alloc) <
2032 			   prot->sysctl_wmem[0])
2033 				return 1;
2034 	}
2035 
2036 	if (sk_has_memory_pressure(sk)) {
2037 		int alloc;
2038 
2039 		if (!sk_under_memory_pressure(sk))
2040 			return 1;
2041 		alloc = sk_sockets_allocated_read_positive(sk);
2042 		if (sk_prot_mem_limits(sk, 2) > alloc *
2043 		    sk_mem_pages(sk->sk_wmem_queued +
2044 				 atomic_read(&sk->sk_rmem_alloc) +
2045 				 sk->sk_forward_alloc))
2046 			return 1;
2047 	}
2048 
2049 suppress_allocation:
2050 
2051 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2052 		sk_stream_moderate_sndbuf(sk);
2053 
2054 		/* Fail only if socket is _under_ its sndbuf.
2055 		 * In this case we cannot block, so that we have to fail.
2056 		 */
2057 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2058 			return 1;
2059 	}
2060 
2061 	trace_sock_exceed_buf_limit(sk, prot, allocated);
2062 
2063 	/* Alas. Undo changes. */
2064 	sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2065 
2066 	sk_memory_allocated_sub(sk, amt);
2067 
2068 	return 0;
2069 }
2070 EXPORT_SYMBOL(__sk_mem_schedule);
2071 
2072 /**
2073  *	__sk_reclaim - reclaim memory_allocated
2074  *	@sk: socket
2075  */
2076 void __sk_mem_reclaim(struct sock *sk)
2077 {
2078 	sk_memory_allocated_sub(sk,
2079 				sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2080 	sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2081 
2082 	if (sk_under_memory_pressure(sk) &&
2083 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2084 		sk_leave_memory_pressure(sk);
2085 }
2086 EXPORT_SYMBOL(__sk_mem_reclaim);
2087 
2088 
2089 /*
2090  * Set of default routines for initialising struct proto_ops when
2091  * the protocol does not support a particular function. In certain
2092  * cases where it makes no sense for a protocol to have a "do nothing"
2093  * function, some default processing is provided.
2094  */
2095 
2096 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2097 {
2098 	return -EOPNOTSUPP;
2099 }
2100 EXPORT_SYMBOL(sock_no_bind);
2101 
2102 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2103 		    int len, int flags)
2104 {
2105 	return -EOPNOTSUPP;
2106 }
2107 EXPORT_SYMBOL(sock_no_connect);
2108 
2109 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2110 {
2111 	return -EOPNOTSUPP;
2112 }
2113 EXPORT_SYMBOL(sock_no_socketpair);
2114 
2115 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2116 {
2117 	return -EOPNOTSUPP;
2118 }
2119 EXPORT_SYMBOL(sock_no_accept);
2120 
2121 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2122 		    int *len, int peer)
2123 {
2124 	return -EOPNOTSUPP;
2125 }
2126 EXPORT_SYMBOL(sock_no_getname);
2127 
2128 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2129 {
2130 	return 0;
2131 }
2132 EXPORT_SYMBOL(sock_no_poll);
2133 
2134 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2135 {
2136 	return -EOPNOTSUPP;
2137 }
2138 EXPORT_SYMBOL(sock_no_ioctl);
2139 
2140 int sock_no_listen(struct socket *sock, int backlog)
2141 {
2142 	return -EOPNOTSUPP;
2143 }
2144 EXPORT_SYMBOL(sock_no_listen);
2145 
2146 int sock_no_shutdown(struct socket *sock, int how)
2147 {
2148 	return -EOPNOTSUPP;
2149 }
2150 EXPORT_SYMBOL(sock_no_shutdown);
2151 
2152 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2153 		    char __user *optval, unsigned int optlen)
2154 {
2155 	return -EOPNOTSUPP;
2156 }
2157 EXPORT_SYMBOL(sock_no_setsockopt);
2158 
2159 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2160 		    char __user *optval, int __user *optlen)
2161 {
2162 	return -EOPNOTSUPP;
2163 }
2164 EXPORT_SYMBOL(sock_no_getsockopt);
2165 
2166 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2167 		    size_t len)
2168 {
2169 	return -EOPNOTSUPP;
2170 }
2171 EXPORT_SYMBOL(sock_no_sendmsg);
2172 
2173 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2174 		    size_t len, int flags)
2175 {
2176 	return -EOPNOTSUPP;
2177 }
2178 EXPORT_SYMBOL(sock_no_recvmsg);
2179 
2180 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2181 {
2182 	/* Mirror missing mmap method error code */
2183 	return -ENODEV;
2184 }
2185 EXPORT_SYMBOL(sock_no_mmap);
2186 
2187 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2188 {
2189 	ssize_t res;
2190 	struct msghdr msg = {.msg_flags = flags};
2191 	struct kvec iov;
2192 	char *kaddr = kmap(page);
2193 	iov.iov_base = kaddr + offset;
2194 	iov.iov_len = size;
2195 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2196 	kunmap(page);
2197 	return res;
2198 }
2199 EXPORT_SYMBOL(sock_no_sendpage);
2200 
2201 /*
2202  *	Default Socket Callbacks
2203  */
2204 
2205 static void sock_def_wakeup(struct sock *sk)
2206 {
2207 	struct socket_wq *wq;
2208 
2209 	rcu_read_lock();
2210 	wq = rcu_dereference(sk->sk_wq);
2211 	if (wq_has_sleeper(wq))
2212 		wake_up_interruptible_all(&wq->wait);
2213 	rcu_read_unlock();
2214 }
2215 
2216 static void sock_def_error_report(struct sock *sk)
2217 {
2218 	struct socket_wq *wq;
2219 
2220 	rcu_read_lock();
2221 	wq = rcu_dereference(sk->sk_wq);
2222 	if (wq_has_sleeper(wq))
2223 		wake_up_interruptible_poll(&wq->wait, POLLERR);
2224 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2225 	rcu_read_unlock();
2226 }
2227 
2228 static void sock_def_readable(struct sock *sk)
2229 {
2230 	struct socket_wq *wq;
2231 
2232 	rcu_read_lock();
2233 	wq = rcu_dereference(sk->sk_wq);
2234 	if (wq_has_sleeper(wq))
2235 		wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2236 						POLLRDNORM | POLLRDBAND);
2237 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2238 	rcu_read_unlock();
2239 }
2240 
2241 static void sock_def_write_space(struct sock *sk)
2242 {
2243 	struct socket_wq *wq;
2244 
2245 	rcu_read_lock();
2246 
2247 	/* Do not wake up a writer until he can make "significant"
2248 	 * progress.  --DaveM
2249 	 */
2250 	if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2251 		wq = rcu_dereference(sk->sk_wq);
2252 		if (wq_has_sleeper(wq))
2253 			wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2254 						POLLWRNORM | POLLWRBAND);
2255 
2256 		/* Should agree with poll, otherwise some programs break */
2257 		if (sock_writeable(sk))
2258 			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2259 	}
2260 
2261 	rcu_read_unlock();
2262 }
2263 
2264 static void sock_def_destruct(struct sock *sk)
2265 {
2266 	kfree(sk->sk_protinfo);
2267 }
2268 
2269 void sk_send_sigurg(struct sock *sk)
2270 {
2271 	if (sk->sk_socket && sk->sk_socket->file)
2272 		if (send_sigurg(&sk->sk_socket->file->f_owner))
2273 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2274 }
2275 EXPORT_SYMBOL(sk_send_sigurg);
2276 
2277 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2278 		    unsigned long expires)
2279 {
2280 	if (!mod_timer(timer, expires))
2281 		sock_hold(sk);
2282 }
2283 EXPORT_SYMBOL(sk_reset_timer);
2284 
2285 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2286 {
2287 	if (del_timer(timer))
2288 		__sock_put(sk);
2289 }
2290 EXPORT_SYMBOL(sk_stop_timer);
2291 
2292 void sock_init_data(struct socket *sock, struct sock *sk)
2293 {
2294 	skb_queue_head_init(&sk->sk_receive_queue);
2295 	skb_queue_head_init(&sk->sk_write_queue);
2296 	skb_queue_head_init(&sk->sk_error_queue);
2297 #ifdef CONFIG_NET_DMA
2298 	skb_queue_head_init(&sk->sk_async_wait_queue);
2299 #endif
2300 
2301 	sk->sk_send_head	=	NULL;
2302 
2303 	init_timer(&sk->sk_timer);
2304 
2305 	sk->sk_allocation	=	GFP_KERNEL;
2306 	sk->sk_rcvbuf		=	sysctl_rmem_default;
2307 	sk->sk_sndbuf		=	sysctl_wmem_default;
2308 	sk->sk_state		=	TCP_CLOSE;
2309 	sk_set_socket(sk, sock);
2310 
2311 	sock_set_flag(sk, SOCK_ZAPPED);
2312 
2313 	if (sock) {
2314 		sk->sk_type	=	sock->type;
2315 		sk->sk_wq	=	sock->wq;
2316 		sock->sk	=	sk;
2317 	} else
2318 		sk->sk_wq	=	NULL;
2319 
2320 	spin_lock_init(&sk->sk_dst_lock);
2321 	rwlock_init(&sk->sk_callback_lock);
2322 	lockdep_set_class_and_name(&sk->sk_callback_lock,
2323 			af_callback_keys + sk->sk_family,
2324 			af_family_clock_key_strings[sk->sk_family]);
2325 
2326 	sk->sk_state_change	=	sock_def_wakeup;
2327 	sk->sk_data_ready	=	sock_def_readable;
2328 	sk->sk_write_space	=	sock_def_write_space;
2329 	sk->sk_error_report	=	sock_def_error_report;
2330 	sk->sk_destruct		=	sock_def_destruct;
2331 
2332 	sk->sk_frag.page	=	NULL;
2333 	sk->sk_frag.offset	=	0;
2334 	sk->sk_peek_off		=	-1;
2335 
2336 	sk->sk_peer_pid 	=	NULL;
2337 	sk->sk_peer_cred	=	NULL;
2338 	sk->sk_write_pending	=	0;
2339 	sk->sk_rcvlowat		=	1;
2340 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
2341 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
2342 
2343 	sk->sk_stamp = ktime_set(-1L, 0);
2344 
2345 #ifdef CONFIG_NET_RX_BUSY_POLL
2346 	sk->sk_napi_id		=	0;
2347 	sk->sk_ll_usec		=	sysctl_net_busy_read;
2348 #endif
2349 
2350 	sk->sk_max_pacing_rate = ~0U;
2351 	sk->sk_pacing_rate = ~0U;
2352 	/*
2353 	 * Before updating sk_refcnt, we must commit prior changes to memory
2354 	 * (Documentation/RCU/rculist_nulls.txt for details)
2355 	 */
2356 	smp_wmb();
2357 	atomic_set(&sk->sk_refcnt, 1);
2358 	atomic_set(&sk->sk_drops, 0);
2359 }
2360 EXPORT_SYMBOL(sock_init_data);
2361 
2362 void lock_sock_nested(struct sock *sk, int subclass)
2363 {
2364 	might_sleep();
2365 	spin_lock_bh(&sk->sk_lock.slock);
2366 	if (sk->sk_lock.owned)
2367 		__lock_sock(sk);
2368 	sk->sk_lock.owned = 1;
2369 	spin_unlock(&sk->sk_lock.slock);
2370 	/*
2371 	 * The sk_lock has mutex_lock() semantics here:
2372 	 */
2373 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2374 	local_bh_enable();
2375 }
2376 EXPORT_SYMBOL(lock_sock_nested);
2377 
2378 void release_sock(struct sock *sk)
2379 {
2380 	/*
2381 	 * The sk_lock has mutex_unlock() semantics:
2382 	 */
2383 	mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2384 
2385 	spin_lock_bh(&sk->sk_lock.slock);
2386 	if (sk->sk_backlog.tail)
2387 		__release_sock(sk);
2388 
2389 	/* Warning : release_cb() might need to release sk ownership,
2390 	 * ie call sock_release_ownership(sk) before us.
2391 	 */
2392 	if (sk->sk_prot->release_cb)
2393 		sk->sk_prot->release_cb(sk);
2394 
2395 	sock_release_ownership(sk);
2396 	if (waitqueue_active(&sk->sk_lock.wq))
2397 		wake_up(&sk->sk_lock.wq);
2398 	spin_unlock_bh(&sk->sk_lock.slock);
2399 }
2400 EXPORT_SYMBOL(release_sock);
2401 
2402 /**
2403  * lock_sock_fast - fast version of lock_sock
2404  * @sk: socket
2405  *
2406  * This version should be used for very small section, where process wont block
2407  * return false if fast path is taken
2408  *   sk_lock.slock locked, owned = 0, BH disabled
2409  * return true if slow path is taken
2410  *   sk_lock.slock unlocked, owned = 1, BH enabled
2411  */
2412 bool lock_sock_fast(struct sock *sk)
2413 {
2414 	might_sleep();
2415 	spin_lock_bh(&sk->sk_lock.slock);
2416 
2417 	if (!sk->sk_lock.owned)
2418 		/*
2419 		 * Note : We must disable BH
2420 		 */
2421 		return false;
2422 
2423 	__lock_sock(sk);
2424 	sk->sk_lock.owned = 1;
2425 	spin_unlock(&sk->sk_lock.slock);
2426 	/*
2427 	 * The sk_lock has mutex_lock() semantics here:
2428 	 */
2429 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2430 	local_bh_enable();
2431 	return true;
2432 }
2433 EXPORT_SYMBOL(lock_sock_fast);
2434 
2435 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2436 {
2437 	struct timeval tv;
2438 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2439 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2440 	tv = ktime_to_timeval(sk->sk_stamp);
2441 	if (tv.tv_sec == -1)
2442 		return -ENOENT;
2443 	if (tv.tv_sec == 0) {
2444 		sk->sk_stamp = ktime_get_real();
2445 		tv = ktime_to_timeval(sk->sk_stamp);
2446 	}
2447 	return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2448 }
2449 EXPORT_SYMBOL(sock_get_timestamp);
2450 
2451 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2452 {
2453 	struct timespec ts;
2454 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2455 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2456 	ts = ktime_to_timespec(sk->sk_stamp);
2457 	if (ts.tv_sec == -1)
2458 		return -ENOENT;
2459 	if (ts.tv_sec == 0) {
2460 		sk->sk_stamp = ktime_get_real();
2461 		ts = ktime_to_timespec(sk->sk_stamp);
2462 	}
2463 	return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2464 }
2465 EXPORT_SYMBOL(sock_get_timestampns);
2466 
2467 void sock_enable_timestamp(struct sock *sk, int flag)
2468 {
2469 	if (!sock_flag(sk, flag)) {
2470 		unsigned long previous_flags = sk->sk_flags;
2471 
2472 		sock_set_flag(sk, flag);
2473 		/*
2474 		 * we just set one of the two flags which require net
2475 		 * time stamping, but time stamping might have been on
2476 		 * already because of the other one
2477 		 */
2478 		if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2479 			net_enable_timestamp();
2480 	}
2481 }
2482 
2483 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2484 		       int level, int type)
2485 {
2486 	struct sock_exterr_skb *serr;
2487 	struct sk_buff *skb, *skb2;
2488 	int copied, err;
2489 
2490 	err = -EAGAIN;
2491 	skb = skb_dequeue(&sk->sk_error_queue);
2492 	if (skb == NULL)
2493 		goto out;
2494 
2495 	copied = skb->len;
2496 	if (copied > len) {
2497 		msg->msg_flags |= MSG_TRUNC;
2498 		copied = len;
2499 	}
2500 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2501 	if (err)
2502 		goto out_free_skb;
2503 
2504 	sock_recv_timestamp(msg, sk, skb);
2505 
2506 	serr = SKB_EXT_ERR(skb);
2507 	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2508 
2509 	msg->msg_flags |= MSG_ERRQUEUE;
2510 	err = copied;
2511 
2512 	/* Reset and regenerate socket error */
2513 	spin_lock_bh(&sk->sk_error_queue.lock);
2514 	sk->sk_err = 0;
2515 	if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2516 		sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2517 		spin_unlock_bh(&sk->sk_error_queue.lock);
2518 		sk->sk_error_report(sk);
2519 	} else
2520 		spin_unlock_bh(&sk->sk_error_queue.lock);
2521 
2522 out_free_skb:
2523 	kfree_skb(skb);
2524 out:
2525 	return err;
2526 }
2527 EXPORT_SYMBOL(sock_recv_errqueue);
2528 
2529 /*
2530  *	Get a socket option on an socket.
2531  *
2532  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
2533  *	asynchronous errors should be reported by getsockopt. We assume
2534  *	this means if you specify SO_ERROR (otherwise whats the point of it).
2535  */
2536 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2537 			   char __user *optval, int __user *optlen)
2538 {
2539 	struct sock *sk = sock->sk;
2540 
2541 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2542 }
2543 EXPORT_SYMBOL(sock_common_getsockopt);
2544 
2545 #ifdef CONFIG_COMPAT
2546 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2547 				  char __user *optval, int __user *optlen)
2548 {
2549 	struct sock *sk = sock->sk;
2550 
2551 	if (sk->sk_prot->compat_getsockopt != NULL)
2552 		return sk->sk_prot->compat_getsockopt(sk, level, optname,
2553 						      optval, optlen);
2554 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2555 }
2556 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2557 #endif
2558 
2559 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2560 			struct msghdr *msg, size_t size, int flags)
2561 {
2562 	struct sock *sk = sock->sk;
2563 	int addr_len = 0;
2564 	int err;
2565 
2566 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2567 				   flags & ~MSG_DONTWAIT, &addr_len);
2568 	if (err >= 0)
2569 		msg->msg_namelen = addr_len;
2570 	return err;
2571 }
2572 EXPORT_SYMBOL(sock_common_recvmsg);
2573 
2574 /*
2575  *	Set socket options on an inet socket.
2576  */
2577 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2578 			   char __user *optval, unsigned int optlen)
2579 {
2580 	struct sock *sk = sock->sk;
2581 
2582 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2583 }
2584 EXPORT_SYMBOL(sock_common_setsockopt);
2585 
2586 #ifdef CONFIG_COMPAT
2587 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2588 				  char __user *optval, unsigned int optlen)
2589 {
2590 	struct sock *sk = sock->sk;
2591 
2592 	if (sk->sk_prot->compat_setsockopt != NULL)
2593 		return sk->sk_prot->compat_setsockopt(sk, level, optname,
2594 						      optval, optlen);
2595 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2596 }
2597 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2598 #endif
2599 
2600 void sk_common_release(struct sock *sk)
2601 {
2602 	if (sk->sk_prot->destroy)
2603 		sk->sk_prot->destroy(sk);
2604 
2605 	/*
2606 	 * Observation: when sock_common_release is called, processes have
2607 	 * no access to socket. But net still has.
2608 	 * Step one, detach it from networking:
2609 	 *
2610 	 * A. Remove from hash tables.
2611 	 */
2612 
2613 	sk->sk_prot->unhash(sk);
2614 
2615 	/*
2616 	 * In this point socket cannot receive new packets, but it is possible
2617 	 * that some packets are in flight because some CPU runs receiver and
2618 	 * did hash table lookup before we unhashed socket. They will achieve
2619 	 * receive queue and will be purged by socket destructor.
2620 	 *
2621 	 * Also we still have packets pending on receive queue and probably,
2622 	 * our own packets waiting in device queues. sock_destroy will drain
2623 	 * receive queue, but transmitted packets will delay socket destruction
2624 	 * until the last reference will be released.
2625 	 */
2626 
2627 	sock_orphan(sk);
2628 
2629 	xfrm_sk_free_policy(sk);
2630 
2631 	sk_refcnt_debug_release(sk);
2632 
2633 	if (sk->sk_frag.page) {
2634 		put_page(sk->sk_frag.page);
2635 		sk->sk_frag.page = NULL;
2636 	}
2637 
2638 	sock_put(sk);
2639 }
2640 EXPORT_SYMBOL(sk_common_release);
2641 
2642 #ifdef CONFIG_PROC_FS
2643 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
2644 struct prot_inuse {
2645 	int val[PROTO_INUSE_NR];
2646 };
2647 
2648 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2649 
2650 #ifdef CONFIG_NET_NS
2651 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2652 {
2653 	__this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2654 }
2655 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2656 
2657 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2658 {
2659 	int cpu, idx = prot->inuse_idx;
2660 	int res = 0;
2661 
2662 	for_each_possible_cpu(cpu)
2663 		res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2664 
2665 	return res >= 0 ? res : 0;
2666 }
2667 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2668 
2669 static int __net_init sock_inuse_init_net(struct net *net)
2670 {
2671 	net->core.inuse = alloc_percpu(struct prot_inuse);
2672 	return net->core.inuse ? 0 : -ENOMEM;
2673 }
2674 
2675 static void __net_exit sock_inuse_exit_net(struct net *net)
2676 {
2677 	free_percpu(net->core.inuse);
2678 }
2679 
2680 static struct pernet_operations net_inuse_ops = {
2681 	.init = sock_inuse_init_net,
2682 	.exit = sock_inuse_exit_net,
2683 };
2684 
2685 static __init int net_inuse_init(void)
2686 {
2687 	if (register_pernet_subsys(&net_inuse_ops))
2688 		panic("Cannot initialize net inuse counters");
2689 
2690 	return 0;
2691 }
2692 
2693 core_initcall(net_inuse_init);
2694 #else
2695 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2696 
2697 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2698 {
2699 	__this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2700 }
2701 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2702 
2703 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2704 {
2705 	int cpu, idx = prot->inuse_idx;
2706 	int res = 0;
2707 
2708 	for_each_possible_cpu(cpu)
2709 		res += per_cpu(prot_inuse, cpu).val[idx];
2710 
2711 	return res >= 0 ? res : 0;
2712 }
2713 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2714 #endif
2715 
2716 static void assign_proto_idx(struct proto *prot)
2717 {
2718 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2719 
2720 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2721 		pr_err("PROTO_INUSE_NR exhausted\n");
2722 		return;
2723 	}
2724 
2725 	set_bit(prot->inuse_idx, proto_inuse_idx);
2726 }
2727 
2728 static void release_proto_idx(struct proto *prot)
2729 {
2730 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2731 		clear_bit(prot->inuse_idx, proto_inuse_idx);
2732 }
2733 #else
2734 static inline void assign_proto_idx(struct proto *prot)
2735 {
2736 }
2737 
2738 static inline void release_proto_idx(struct proto *prot)
2739 {
2740 }
2741 #endif
2742 
2743 int proto_register(struct proto *prot, int alloc_slab)
2744 {
2745 	if (alloc_slab) {
2746 		prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2747 					SLAB_HWCACHE_ALIGN | prot->slab_flags,
2748 					NULL);
2749 
2750 		if (prot->slab == NULL) {
2751 			pr_crit("%s: Can't create sock SLAB cache!\n",
2752 				prot->name);
2753 			goto out;
2754 		}
2755 
2756 		if (prot->rsk_prot != NULL) {
2757 			prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2758 			if (prot->rsk_prot->slab_name == NULL)
2759 				goto out_free_sock_slab;
2760 
2761 			prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2762 								 prot->rsk_prot->obj_size, 0,
2763 								 SLAB_HWCACHE_ALIGN, NULL);
2764 
2765 			if (prot->rsk_prot->slab == NULL) {
2766 				pr_crit("%s: Can't create request sock SLAB cache!\n",
2767 					prot->name);
2768 				goto out_free_request_sock_slab_name;
2769 			}
2770 		}
2771 
2772 		if (prot->twsk_prot != NULL) {
2773 			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2774 
2775 			if (prot->twsk_prot->twsk_slab_name == NULL)
2776 				goto out_free_request_sock_slab;
2777 
2778 			prot->twsk_prot->twsk_slab =
2779 				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2780 						  prot->twsk_prot->twsk_obj_size,
2781 						  0,
2782 						  SLAB_HWCACHE_ALIGN |
2783 							prot->slab_flags,
2784 						  NULL);
2785 			if (prot->twsk_prot->twsk_slab == NULL)
2786 				goto out_free_timewait_sock_slab_name;
2787 		}
2788 	}
2789 
2790 	mutex_lock(&proto_list_mutex);
2791 	list_add(&prot->node, &proto_list);
2792 	assign_proto_idx(prot);
2793 	mutex_unlock(&proto_list_mutex);
2794 	return 0;
2795 
2796 out_free_timewait_sock_slab_name:
2797 	kfree(prot->twsk_prot->twsk_slab_name);
2798 out_free_request_sock_slab:
2799 	if (prot->rsk_prot && prot->rsk_prot->slab) {
2800 		kmem_cache_destroy(prot->rsk_prot->slab);
2801 		prot->rsk_prot->slab = NULL;
2802 	}
2803 out_free_request_sock_slab_name:
2804 	if (prot->rsk_prot)
2805 		kfree(prot->rsk_prot->slab_name);
2806 out_free_sock_slab:
2807 	kmem_cache_destroy(prot->slab);
2808 	prot->slab = NULL;
2809 out:
2810 	return -ENOBUFS;
2811 }
2812 EXPORT_SYMBOL(proto_register);
2813 
2814 void proto_unregister(struct proto *prot)
2815 {
2816 	mutex_lock(&proto_list_mutex);
2817 	release_proto_idx(prot);
2818 	list_del(&prot->node);
2819 	mutex_unlock(&proto_list_mutex);
2820 
2821 	if (prot->slab != NULL) {
2822 		kmem_cache_destroy(prot->slab);
2823 		prot->slab = NULL;
2824 	}
2825 
2826 	if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2827 		kmem_cache_destroy(prot->rsk_prot->slab);
2828 		kfree(prot->rsk_prot->slab_name);
2829 		prot->rsk_prot->slab = NULL;
2830 	}
2831 
2832 	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2833 		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2834 		kfree(prot->twsk_prot->twsk_slab_name);
2835 		prot->twsk_prot->twsk_slab = NULL;
2836 	}
2837 }
2838 EXPORT_SYMBOL(proto_unregister);
2839 
2840 #ifdef CONFIG_PROC_FS
2841 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2842 	__acquires(proto_list_mutex)
2843 {
2844 	mutex_lock(&proto_list_mutex);
2845 	return seq_list_start_head(&proto_list, *pos);
2846 }
2847 
2848 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2849 {
2850 	return seq_list_next(v, &proto_list, pos);
2851 }
2852 
2853 static void proto_seq_stop(struct seq_file *seq, void *v)
2854 	__releases(proto_list_mutex)
2855 {
2856 	mutex_unlock(&proto_list_mutex);
2857 }
2858 
2859 static char proto_method_implemented(const void *method)
2860 {
2861 	return method == NULL ? 'n' : 'y';
2862 }
2863 static long sock_prot_memory_allocated(struct proto *proto)
2864 {
2865 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2866 }
2867 
2868 static char *sock_prot_memory_pressure(struct proto *proto)
2869 {
2870 	return proto->memory_pressure != NULL ?
2871 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2872 }
2873 
2874 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2875 {
2876 
2877 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2878 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2879 		   proto->name,
2880 		   proto->obj_size,
2881 		   sock_prot_inuse_get(seq_file_net(seq), proto),
2882 		   sock_prot_memory_allocated(proto),
2883 		   sock_prot_memory_pressure(proto),
2884 		   proto->max_header,
2885 		   proto->slab == NULL ? "no" : "yes",
2886 		   module_name(proto->owner),
2887 		   proto_method_implemented(proto->close),
2888 		   proto_method_implemented(proto->connect),
2889 		   proto_method_implemented(proto->disconnect),
2890 		   proto_method_implemented(proto->accept),
2891 		   proto_method_implemented(proto->ioctl),
2892 		   proto_method_implemented(proto->init),
2893 		   proto_method_implemented(proto->destroy),
2894 		   proto_method_implemented(proto->shutdown),
2895 		   proto_method_implemented(proto->setsockopt),
2896 		   proto_method_implemented(proto->getsockopt),
2897 		   proto_method_implemented(proto->sendmsg),
2898 		   proto_method_implemented(proto->recvmsg),
2899 		   proto_method_implemented(proto->sendpage),
2900 		   proto_method_implemented(proto->bind),
2901 		   proto_method_implemented(proto->backlog_rcv),
2902 		   proto_method_implemented(proto->hash),
2903 		   proto_method_implemented(proto->unhash),
2904 		   proto_method_implemented(proto->get_port),
2905 		   proto_method_implemented(proto->enter_memory_pressure));
2906 }
2907 
2908 static int proto_seq_show(struct seq_file *seq, void *v)
2909 {
2910 	if (v == &proto_list)
2911 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2912 			   "protocol",
2913 			   "size",
2914 			   "sockets",
2915 			   "memory",
2916 			   "press",
2917 			   "maxhdr",
2918 			   "slab",
2919 			   "module",
2920 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2921 	else
2922 		proto_seq_printf(seq, list_entry(v, struct proto, node));
2923 	return 0;
2924 }
2925 
2926 static const struct seq_operations proto_seq_ops = {
2927 	.start  = proto_seq_start,
2928 	.next   = proto_seq_next,
2929 	.stop   = proto_seq_stop,
2930 	.show   = proto_seq_show,
2931 };
2932 
2933 static int proto_seq_open(struct inode *inode, struct file *file)
2934 {
2935 	return seq_open_net(inode, file, &proto_seq_ops,
2936 			    sizeof(struct seq_net_private));
2937 }
2938 
2939 static const struct file_operations proto_seq_fops = {
2940 	.owner		= THIS_MODULE,
2941 	.open		= proto_seq_open,
2942 	.read		= seq_read,
2943 	.llseek		= seq_lseek,
2944 	.release	= seq_release_net,
2945 };
2946 
2947 static __net_init int proto_init_net(struct net *net)
2948 {
2949 	if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2950 		return -ENOMEM;
2951 
2952 	return 0;
2953 }
2954 
2955 static __net_exit void proto_exit_net(struct net *net)
2956 {
2957 	remove_proc_entry("protocols", net->proc_net);
2958 }
2959 
2960 
2961 static __net_initdata struct pernet_operations proto_net_ops = {
2962 	.init = proto_init_net,
2963 	.exit = proto_exit_net,
2964 };
2965 
2966 static int __init proto_init(void)
2967 {
2968 	return register_pernet_subsys(&proto_net_ops);
2969 }
2970 
2971 subsys_initcall(proto_init);
2972 
2973 #endif /* PROC_FS */
2974