xref: /linux/net/core/sock.c (revision 723d2904a2a47339cae7cb1c2a6a7582c130ee66)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Generic socket support routines. Memory allocators, socket lock/release
8  *		handler for protocols to use and generic option handler.
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87 
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
94 #include <linux/in.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 
117 #include <linux/uaccess.h>
118 
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <linux/net_tstamp.h>
126 #include <net/xfrm.h>
127 #include <linux/ipsec.h>
128 #include <net/cls_cgroup.h>
129 #include <net/netprio_cgroup.h>
130 #include <linux/sock_diag.h>
131 
132 #include <linux/filter.h>
133 #include <net/sock_reuseport.h>
134 #include <net/bpf_sk_storage.h>
135 
136 #include <trace/events/sock.h>
137 
138 #include <net/tcp.h>
139 #include <net/busy_poll.h>
140 
141 static DEFINE_MUTEX(proto_list_mutex);
142 static LIST_HEAD(proto_list);
143 
144 static void sock_inuse_add(struct net *net, int val);
145 
146 /**
147  * sk_ns_capable - General socket capability test
148  * @sk: Socket to use a capability on or through
149  * @user_ns: The user namespace of the capability to use
150  * @cap: The capability to use
151  *
152  * Test to see if the opener of the socket had when the socket was
153  * created and the current process has the capability @cap in the user
154  * namespace @user_ns.
155  */
156 bool sk_ns_capable(const struct sock *sk,
157 		   struct user_namespace *user_ns, int cap)
158 {
159 	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
160 		ns_capable(user_ns, cap);
161 }
162 EXPORT_SYMBOL(sk_ns_capable);
163 
164 /**
165  * sk_capable - Socket global capability test
166  * @sk: Socket to use a capability on or through
167  * @cap: The global capability to use
168  *
169  * Test to see if the opener of the socket had when the socket was
170  * created and the current process has the capability @cap in all user
171  * namespaces.
172  */
173 bool sk_capable(const struct sock *sk, int cap)
174 {
175 	return sk_ns_capable(sk, &init_user_ns, cap);
176 }
177 EXPORT_SYMBOL(sk_capable);
178 
179 /**
180  * sk_net_capable - Network namespace socket capability test
181  * @sk: Socket to use a capability on or through
182  * @cap: The capability to use
183  *
184  * Test to see if the opener of the socket had when the socket was created
185  * and the current process has the capability @cap over the network namespace
186  * the socket is a member of.
187  */
188 bool sk_net_capable(const struct sock *sk, int cap)
189 {
190 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
191 }
192 EXPORT_SYMBOL(sk_net_capable);
193 
194 /*
195  * Each address family might have different locking rules, so we have
196  * one slock key per address family and separate keys for internal and
197  * userspace sockets.
198  */
199 static struct lock_class_key af_family_keys[AF_MAX];
200 static struct lock_class_key af_family_kern_keys[AF_MAX];
201 static struct lock_class_key af_family_slock_keys[AF_MAX];
202 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
203 
204 /*
205  * Make lock validator output more readable. (we pre-construct these
206  * strings build-time, so that runtime initialization of socket
207  * locks is fast):
208  */
209 
210 #define _sock_locks(x)						  \
211   x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
212   x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
213   x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
214   x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
215   x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
216   x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
217   x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
218   x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
219   x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
220   x "27"       ,	x "28"          ,	x "AF_CAN"      , \
221   x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
222   x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
223   x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
224   x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
225   x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
226   x "AF_MAX"
227 
228 static const char *const af_family_key_strings[AF_MAX+1] = {
229 	_sock_locks("sk_lock-")
230 };
231 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
232 	_sock_locks("slock-")
233 };
234 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
235 	_sock_locks("clock-")
236 };
237 
238 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
239 	_sock_locks("k-sk_lock-")
240 };
241 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
242 	_sock_locks("k-slock-")
243 };
244 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
245 	_sock_locks("k-clock-")
246 };
247 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
248 	_sock_locks("rlock-")
249 };
250 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
251 	_sock_locks("wlock-")
252 };
253 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
254 	_sock_locks("elock-")
255 };
256 
257 /*
258  * sk_callback_lock and sk queues locking rules are per-address-family,
259  * so split the lock classes by using a per-AF key:
260  */
261 static struct lock_class_key af_callback_keys[AF_MAX];
262 static struct lock_class_key af_rlock_keys[AF_MAX];
263 static struct lock_class_key af_wlock_keys[AF_MAX];
264 static struct lock_class_key af_elock_keys[AF_MAX];
265 static struct lock_class_key af_kern_callback_keys[AF_MAX];
266 
267 /* Run time adjustable parameters. */
268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
269 EXPORT_SYMBOL(sysctl_wmem_max);
270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
271 EXPORT_SYMBOL(sysctl_rmem_max);
272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
274 
275 /* Maximal space eaten by iovec or ancillary data plus some space */
276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
277 EXPORT_SYMBOL(sysctl_optmem_max);
278 
279 int sysctl_tstamp_allow_data __read_mostly = 1;
280 
281 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
282 EXPORT_SYMBOL_GPL(memalloc_socks_key);
283 
284 /**
285  * sk_set_memalloc - sets %SOCK_MEMALLOC
286  * @sk: socket to set it on
287  *
288  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
289  * It's the responsibility of the admin to adjust min_free_kbytes
290  * to meet the requirements
291  */
292 void sk_set_memalloc(struct sock *sk)
293 {
294 	sock_set_flag(sk, SOCK_MEMALLOC);
295 	sk->sk_allocation |= __GFP_MEMALLOC;
296 	static_branch_inc(&memalloc_socks_key);
297 }
298 EXPORT_SYMBOL_GPL(sk_set_memalloc);
299 
300 void sk_clear_memalloc(struct sock *sk)
301 {
302 	sock_reset_flag(sk, SOCK_MEMALLOC);
303 	sk->sk_allocation &= ~__GFP_MEMALLOC;
304 	static_branch_dec(&memalloc_socks_key);
305 
306 	/*
307 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
308 	 * progress of swapping. SOCK_MEMALLOC may be cleared while
309 	 * it has rmem allocations due to the last swapfile being deactivated
310 	 * but there is a risk that the socket is unusable due to exceeding
311 	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
312 	 */
313 	sk_mem_reclaim(sk);
314 }
315 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
316 
317 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
318 {
319 	int ret;
320 	unsigned int noreclaim_flag;
321 
322 	/* these should have been dropped before queueing */
323 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
324 
325 	noreclaim_flag = memalloc_noreclaim_save();
326 	ret = sk->sk_backlog_rcv(sk, skb);
327 	memalloc_noreclaim_restore(noreclaim_flag);
328 
329 	return ret;
330 }
331 EXPORT_SYMBOL(__sk_backlog_rcv);
332 
333 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
334 {
335 	struct __kernel_sock_timeval tv;
336 	int size;
337 
338 	if (timeo == MAX_SCHEDULE_TIMEOUT) {
339 		tv.tv_sec = 0;
340 		tv.tv_usec = 0;
341 	} else {
342 		tv.tv_sec = timeo / HZ;
343 		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
344 	}
345 
346 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 		*(struct old_timeval32 *)optval = tv32;
349 		return sizeof(tv32);
350 	}
351 
352 	if (old_timeval) {
353 		struct __kernel_old_timeval old_tv;
354 		old_tv.tv_sec = tv.tv_sec;
355 		old_tv.tv_usec = tv.tv_usec;
356 		*(struct __kernel_old_timeval *)optval = old_tv;
357 		size = sizeof(old_tv);
358 	} else {
359 		*(struct __kernel_sock_timeval *)optval = tv;
360 		size = sizeof(tv);
361 	}
362 
363 	return size;
364 }
365 
366 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
367 {
368 	struct __kernel_sock_timeval tv;
369 
370 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
371 		struct old_timeval32 tv32;
372 
373 		if (optlen < sizeof(tv32))
374 			return -EINVAL;
375 
376 		if (copy_from_user(&tv32, optval, sizeof(tv32)))
377 			return -EFAULT;
378 		tv.tv_sec = tv32.tv_sec;
379 		tv.tv_usec = tv32.tv_usec;
380 	} else if (old_timeval) {
381 		struct __kernel_old_timeval old_tv;
382 
383 		if (optlen < sizeof(old_tv))
384 			return -EINVAL;
385 		if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
386 			return -EFAULT;
387 		tv.tv_sec = old_tv.tv_sec;
388 		tv.tv_usec = old_tv.tv_usec;
389 	} else {
390 		if (optlen < sizeof(tv))
391 			return -EINVAL;
392 		if (copy_from_user(&tv, optval, sizeof(tv)))
393 			return -EFAULT;
394 	}
395 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
396 		return -EDOM;
397 
398 	if (tv.tv_sec < 0) {
399 		static int warned __read_mostly;
400 
401 		*timeo_p = 0;
402 		if (warned < 10 && net_ratelimit()) {
403 			warned++;
404 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
405 				__func__, current->comm, task_pid_nr(current));
406 		}
407 		return 0;
408 	}
409 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
410 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 		return 0;
412 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
413 		*timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
414 	return 0;
415 }
416 
417 static void sock_warn_obsolete_bsdism(const char *name)
418 {
419 	static int warned;
420 	static char warncomm[TASK_COMM_LEN];
421 	if (strcmp(warncomm, current->comm) && warned < 5) {
422 		strcpy(warncomm,  current->comm);
423 		pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
424 			warncomm, name);
425 		warned++;
426 	}
427 }
428 
429 static bool sock_needs_netstamp(const struct sock *sk)
430 {
431 	switch (sk->sk_family) {
432 	case AF_UNSPEC:
433 	case AF_UNIX:
434 		return false;
435 	default:
436 		return true;
437 	}
438 }
439 
440 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
441 {
442 	if (sk->sk_flags & flags) {
443 		sk->sk_flags &= ~flags;
444 		if (sock_needs_netstamp(sk) &&
445 		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
446 			net_disable_timestamp();
447 	}
448 }
449 
450 
451 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
452 {
453 	unsigned long flags;
454 	struct sk_buff_head *list = &sk->sk_receive_queue;
455 
456 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
457 		atomic_inc(&sk->sk_drops);
458 		trace_sock_rcvqueue_full(sk, skb);
459 		return -ENOMEM;
460 	}
461 
462 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
463 		atomic_inc(&sk->sk_drops);
464 		return -ENOBUFS;
465 	}
466 
467 	skb->dev = NULL;
468 	skb_set_owner_r(skb, sk);
469 
470 	/* we escape from rcu protected region, make sure we dont leak
471 	 * a norefcounted dst
472 	 */
473 	skb_dst_force(skb);
474 
475 	spin_lock_irqsave(&list->lock, flags);
476 	sock_skb_set_dropcount(sk, skb);
477 	__skb_queue_tail(list, skb);
478 	spin_unlock_irqrestore(&list->lock, flags);
479 
480 	if (!sock_flag(sk, SOCK_DEAD))
481 		sk->sk_data_ready(sk);
482 	return 0;
483 }
484 EXPORT_SYMBOL(__sock_queue_rcv_skb);
485 
486 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
487 {
488 	int err;
489 
490 	err = sk_filter(sk, skb);
491 	if (err)
492 		return err;
493 
494 	return __sock_queue_rcv_skb(sk, skb);
495 }
496 EXPORT_SYMBOL(sock_queue_rcv_skb);
497 
498 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
499 		     const int nested, unsigned int trim_cap, bool refcounted)
500 {
501 	int rc = NET_RX_SUCCESS;
502 
503 	if (sk_filter_trim_cap(sk, skb, trim_cap))
504 		goto discard_and_relse;
505 
506 	skb->dev = NULL;
507 
508 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
509 		atomic_inc(&sk->sk_drops);
510 		goto discard_and_relse;
511 	}
512 	if (nested)
513 		bh_lock_sock_nested(sk);
514 	else
515 		bh_lock_sock(sk);
516 	if (!sock_owned_by_user(sk)) {
517 		/*
518 		 * trylock + unlock semantics:
519 		 */
520 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
521 
522 		rc = sk_backlog_rcv(sk, skb);
523 
524 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
525 	} else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
526 		bh_unlock_sock(sk);
527 		atomic_inc(&sk->sk_drops);
528 		goto discard_and_relse;
529 	}
530 
531 	bh_unlock_sock(sk);
532 out:
533 	if (refcounted)
534 		sock_put(sk);
535 	return rc;
536 discard_and_relse:
537 	kfree_skb(skb);
538 	goto out;
539 }
540 EXPORT_SYMBOL(__sk_receive_skb);
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_tx_queue_clear(sk);
548 		sk->sk_dst_pending_confirm = 0;
549 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
550 		dst_release(dst);
551 		return NULL;
552 	}
553 
554 	return dst;
555 }
556 EXPORT_SYMBOL(__sk_dst_check);
557 
558 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
559 {
560 	struct dst_entry *dst = sk_dst_get(sk);
561 
562 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
563 		sk_dst_reset(sk);
564 		dst_release(dst);
565 		return NULL;
566 	}
567 
568 	return dst;
569 }
570 EXPORT_SYMBOL(sk_dst_check);
571 
572 static int sock_setbindtodevice_locked(struct sock *sk, int ifindex)
573 {
574 	int ret = -ENOPROTOOPT;
575 #ifdef CONFIG_NETDEVICES
576 	struct net *net = sock_net(sk);
577 
578 	/* Sorry... */
579 	ret = -EPERM;
580 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
581 		goto out;
582 
583 	ret = -EINVAL;
584 	if (ifindex < 0)
585 		goto out;
586 
587 	sk->sk_bound_dev_if = ifindex;
588 	if (sk->sk_prot->rehash)
589 		sk->sk_prot->rehash(sk);
590 	sk_dst_reset(sk);
591 
592 	ret = 0;
593 
594 out:
595 #endif
596 
597 	return ret;
598 }
599 
600 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
601 				int optlen)
602 {
603 	int ret = -ENOPROTOOPT;
604 #ifdef CONFIG_NETDEVICES
605 	struct net *net = sock_net(sk);
606 	char devname[IFNAMSIZ];
607 	int index;
608 
609 	ret = -EINVAL;
610 	if (optlen < 0)
611 		goto out;
612 
613 	/* Bind this socket to a particular device like "eth0",
614 	 * as specified in the passed interface name. If the
615 	 * name is "" or the option length is zero the socket
616 	 * is not bound.
617 	 */
618 	if (optlen > IFNAMSIZ - 1)
619 		optlen = IFNAMSIZ - 1;
620 	memset(devname, 0, sizeof(devname));
621 
622 	ret = -EFAULT;
623 	if (copy_from_user(devname, optval, optlen))
624 		goto out;
625 
626 	index = 0;
627 	if (devname[0] != '\0') {
628 		struct net_device *dev;
629 
630 		rcu_read_lock();
631 		dev = dev_get_by_name_rcu(net, devname);
632 		if (dev)
633 			index = dev->ifindex;
634 		rcu_read_unlock();
635 		ret = -ENODEV;
636 		if (!dev)
637 			goto out;
638 	}
639 
640 	lock_sock(sk);
641 	ret = sock_setbindtodevice_locked(sk, index);
642 	release_sock(sk);
643 
644 out:
645 #endif
646 
647 	return ret;
648 }
649 
650 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
651 				int __user *optlen, int len)
652 {
653 	int ret = -ENOPROTOOPT;
654 #ifdef CONFIG_NETDEVICES
655 	struct net *net = sock_net(sk);
656 	char devname[IFNAMSIZ];
657 
658 	if (sk->sk_bound_dev_if == 0) {
659 		len = 0;
660 		goto zero;
661 	}
662 
663 	ret = -EINVAL;
664 	if (len < IFNAMSIZ)
665 		goto out;
666 
667 	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
668 	if (ret)
669 		goto out;
670 
671 	len = strlen(devname) + 1;
672 
673 	ret = -EFAULT;
674 	if (copy_to_user(optval, devname, len))
675 		goto out;
676 
677 zero:
678 	ret = -EFAULT;
679 	if (put_user(len, optlen))
680 		goto out;
681 
682 	ret = 0;
683 
684 out:
685 #endif
686 
687 	return ret;
688 }
689 
690 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
691 				     int valbool)
692 {
693 	if (valbool)
694 		sock_set_flag(sk, bit);
695 	else
696 		sock_reset_flag(sk, bit);
697 }
698 
699 bool sk_mc_loop(struct sock *sk)
700 {
701 	if (dev_recursion_level())
702 		return false;
703 	if (!sk)
704 		return true;
705 	switch (sk->sk_family) {
706 	case AF_INET:
707 		return inet_sk(sk)->mc_loop;
708 #if IS_ENABLED(CONFIG_IPV6)
709 	case AF_INET6:
710 		return inet6_sk(sk)->mc_loop;
711 #endif
712 	}
713 	WARN_ON(1);
714 	return true;
715 }
716 EXPORT_SYMBOL(sk_mc_loop);
717 
718 /*
719  *	This is meant for all protocols to use and covers goings on
720  *	at the socket level. Everything here is generic.
721  */
722 
723 int sock_setsockopt(struct socket *sock, int level, int optname,
724 		    char __user *optval, unsigned int optlen)
725 {
726 	struct sock_txtime sk_txtime;
727 	struct sock *sk = sock->sk;
728 	int val;
729 	int valbool;
730 	struct linger ling;
731 	int ret = 0;
732 
733 	/*
734 	 *	Options without arguments
735 	 */
736 
737 	if (optname == SO_BINDTODEVICE)
738 		return sock_setbindtodevice(sk, optval, optlen);
739 
740 	if (optlen < sizeof(int))
741 		return -EINVAL;
742 
743 	if (get_user(val, (int __user *)optval))
744 		return -EFAULT;
745 
746 	valbool = val ? 1 : 0;
747 
748 	lock_sock(sk);
749 
750 	switch (optname) {
751 	case SO_DEBUG:
752 		if (val && !capable(CAP_NET_ADMIN))
753 			ret = -EACCES;
754 		else
755 			sock_valbool_flag(sk, SOCK_DBG, valbool);
756 		break;
757 	case SO_REUSEADDR:
758 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
759 		break;
760 	case SO_REUSEPORT:
761 		sk->sk_reuseport = valbool;
762 		break;
763 	case SO_TYPE:
764 	case SO_PROTOCOL:
765 	case SO_DOMAIN:
766 	case SO_ERROR:
767 		ret = -ENOPROTOOPT;
768 		break;
769 	case SO_DONTROUTE:
770 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
771 		sk_dst_reset(sk);
772 		break;
773 	case SO_BROADCAST:
774 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
775 		break;
776 	case SO_SNDBUF:
777 		/* Don't error on this BSD doesn't and if you think
778 		 * about it this is right. Otherwise apps have to
779 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
780 		 * are treated in BSD as hints
781 		 */
782 		val = min_t(u32, val, sysctl_wmem_max);
783 set_sndbuf:
784 		/* Ensure val * 2 fits into an int, to prevent max_t()
785 		 * from treating it as a negative value.
786 		 */
787 		val = min_t(int, val, INT_MAX / 2);
788 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
789 		sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
790 		/* Wake up sending tasks if we upped the value. */
791 		sk->sk_write_space(sk);
792 		break;
793 
794 	case SO_SNDBUFFORCE:
795 		if (!capable(CAP_NET_ADMIN)) {
796 			ret = -EPERM;
797 			break;
798 		}
799 
800 		/* No negative values (to prevent underflow, as val will be
801 		 * multiplied by 2).
802 		 */
803 		if (val < 0)
804 			val = 0;
805 		goto set_sndbuf;
806 
807 	case SO_RCVBUF:
808 		/* Don't error on this BSD doesn't and if you think
809 		 * about it this is right. Otherwise apps have to
810 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
811 		 * are treated in BSD as hints
812 		 */
813 		val = min_t(u32, val, sysctl_rmem_max);
814 set_rcvbuf:
815 		/* Ensure val * 2 fits into an int, to prevent max_t()
816 		 * from treating it as a negative value.
817 		 */
818 		val = min_t(int, val, INT_MAX / 2);
819 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
820 		/*
821 		 * We double it on the way in to account for
822 		 * "struct sk_buff" etc. overhead.   Applications
823 		 * assume that the SO_RCVBUF setting they make will
824 		 * allow that much actual data to be received on that
825 		 * socket.
826 		 *
827 		 * Applications are unaware that "struct sk_buff" and
828 		 * other overheads allocate from the receive buffer
829 		 * during socket buffer allocation.
830 		 *
831 		 * And after considering the possible alternatives,
832 		 * returning the value we actually used in getsockopt
833 		 * is the most desirable behavior.
834 		 */
835 		sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
836 		break;
837 
838 	case SO_RCVBUFFORCE:
839 		if (!capable(CAP_NET_ADMIN)) {
840 			ret = -EPERM;
841 			break;
842 		}
843 
844 		/* No negative values (to prevent underflow, as val will be
845 		 * multiplied by 2).
846 		 */
847 		if (val < 0)
848 			val = 0;
849 		goto set_rcvbuf;
850 
851 	case SO_KEEPALIVE:
852 		if (sk->sk_prot->keepalive)
853 			sk->sk_prot->keepalive(sk, valbool);
854 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
855 		break;
856 
857 	case SO_OOBINLINE:
858 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
859 		break;
860 
861 	case SO_NO_CHECK:
862 		sk->sk_no_check_tx = valbool;
863 		break;
864 
865 	case SO_PRIORITY:
866 		if ((val >= 0 && val <= 6) ||
867 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
868 			sk->sk_priority = val;
869 		else
870 			ret = -EPERM;
871 		break;
872 
873 	case SO_LINGER:
874 		if (optlen < sizeof(ling)) {
875 			ret = -EINVAL;	/* 1003.1g */
876 			break;
877 		}
878 		if (copy_from_user(&ling, optval, sizeof(ling))) {
879 			ret = -EFAULT;
880 			break;
881 		}
882 		if (!ling.l_onoff)
883 			sock_reset_flag(sk, SOCK_LINGER);
884 		else {
885 #if (BITS_PER_LONG == 32)
886 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
887 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
888 			else
889 #endif
890 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
891 			sock_set_flag(sk, SOCK_LINGER);
892 		}
893 		break;
894 
895 	case SO_BSDCOMPAT:
896 		sock_warn_obsolete_bsdism("setsockopt");
897 		break;
898 
899 	case SO_PASSCRED:
900 		if (valbool)
901 			set_bit(SOCK_PASSCRED, &sock->flags);
902 		else
903 			clear_bit(SOCK_PASSCRED, &sock->flags);
904 		break;
905 
906 	case SO_TIMESTAMP_OLD:
907 	case SO_TIMESTAMP_NEW:
908 	case SO_TIMESTAMPNS_OLD:
909 	case SO_TIMESTAMPNS_NEW:
910 		if (valbool)  {
911 			if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
912 				sock_set_flag(sk, SOCK_TSTAMP_NEW);
913 			else
914 				sock_reset_flag(sk, SOCK_TSTAMP_NEW);
915 
916 			if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
917 				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
918 			else
919 				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
920 			sock_set_flag(sk, SOCK_RCVTSTAMP);
921 			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
922 		} else {
923 			sock_reset_flag(sk, SOCK_RCVTSTAMP);
924 			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
925 			sock_reset_flag(sk, SOCK_TSTAMP_NEW);
926 		}
927 		break;
928 
929 	case SO_TIMESTAMPING_NEW:
930 		sock_set_flag(sk, SOCK_TSTAMP_NEW);
931 		/* fall through */
932 	case SO_TIMESTAMPING_OLD:
933 		if (val & ~SOF_TIMESTAMPING_MASK) {
934 			ret = -EINVAL;
935 			break;
936 		}
937 
938 		if (val & SOF_TIMESTAMPING_OPT_ID &&
939 		    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
940 			if (sk->sk_protocol == IPPROTO_TCP &&
941 			    sk->sk_type == SOCK_STREAM) {
942 				if ((1 << sk->sk_state) &
943 				    (TCPF_CLOSE | TCPF_LISTEN)) {
944 					ret = -EINVAL;
945 					break;
946 				}
947 				sk->sk_tskey = tcp_sk(sk)->snd_una;
948 			} else {
949 				sk->sk_tskey = 0;
950 			}
951 		}
952 
953 		if (val & SOF_TIMESTAMPING_OPT_STATS &&
954 		    !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
955 			ret = -EINVAL;
956 			break;
957 		}
958 
959 		sk->sk_tsflags = val;
960 		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
961 			sock_enable_timestamp(sk,
962 					      SOCK_TIMESTAMPING_RX_SOFTWARE);
963 		else {
964 			if (optname == SO_TIMESTAMPING_NEW)
965 				sock_reset_flag(sk, SOCK_TSTAMP_NEW);
966 
967 			sock_disable_timestamp(sk,
968 					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
969 		}
970 		break;
971 
972 	case SO_RCVLOWAT:
973 		if (val < 0)
974 			val = INT_MAX;
975 		if (sock->ops->set_rcvlowat)
976 			ret = sock->ops->set_rcvlowat(sk, val);
977 		else
978 			sk->sk_rcvlowat = val ? : 1;
979 		break;
980 
981 	case SO_RCVTIMEO_OLD:
982 	case SO_RCVTIMEO_NEW:
983 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
984 		break;
985 
986 	case SO_SNDTIMEO_OLD:
987 	case SO_SNDTIMEO_NEW:
988 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
989 		break;
990 
991 	case SO_ATTACH_FILTER:
992 		ret = -EINVAL;
993 		if (optlen == sizeof(struct sock_fprog)) {
994 			struct sock_fprog fprog;
995 
996 			ret = -EFAULT;
997 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
998 				break;
999 
1000 			ret = sk_attach_filter(&fprog, sk);
1001 		}
1002 		break;
1003 
1004 	case SO_ATTACH_BPF:
1005 		ret = -EINVAL;
1006 		if (optlen == sizeof(u32)) {
1007 			u32 ufd;
1008 
1009 			ret = -EFAULT;
1010 			if (copy_from_user(&ufd, optval, sizeof(ufd)))
1011 				break;
1012 
1013 			ret = sk_attach_bpf(ufd, sk);
1014 		}
1015 		break;
1016 
1017 	case SO_ATTACH_REUSEPORT_CBPF:
1018 		ret = -EINVAL;
1019 		if (optlen == sizeof(struct sock_fprog)) {
1020 			struct sock_fprog fprog;
1021 
1022 			ret = -EFAULT;
1023 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
1024 				break;
1025 
1026 			ret = sk_reuseport_attach_filter(&fprog, sk);
1027 		}
1028 		break;
1029 
1030 	case SO_ATTACH_REUSEPORT_EBPF:
1031 		ret = -EINVAL;
1032 		if (optlen == sizeof(u32)) {
1033 			u32 ufd;
1034 
1035 			ret = -EFAULT;
1036 			if (copy_from_user(&ufd, optval, sizeof(ufd)))
1037 				break;
1038 
1039 			ret = sk_reuseport_attach_bpf(ufd, sk);
1040 		}
1041 		break;
1042 
1043 	case SO_DETACH_REUSEPORT_BPF:
1044 		ret = reuseport_detach_prog(sk);
1045 		break;
1046 
1047 	case SO_DETACH_FILTER:
1048 		ret = sk_detach_filter(sk);
1049 		break;
1050 
1051 	case SO_LOCK_FILTER:
1052 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1053 			ret = -EPERM;
1054 		else
1055 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1056 		break;
1057 
1058 	case SO_PASSSEC:
1059 		if (valbool)
1060 			set_bit(SOCK_PASSSEC, &sock->flags);
1061 		else
1062 			clear_bit(SOCK_PASSSEC, &sock->flags);
1063 		break;
1064 	case SO_MARK:
1065 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1066 			ret = -EPERM;
1067 		} else if (val != sk->sk_mark) {
1068 			sk->sk_mark = val;
1069 			sk_dst_reset(sk);
1070 		}
1071 		break;
1072 
1073 	case SO_RXQ_OVFL:
1074 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1075 		break;
1076 
1077 	case SO_WIFI_STATUS:
1078 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1079 		break;
1080 
1081 	case SO_PEEK_OFF:
1082 		if (sock->ops->set_peek_off)
1083 			ret = sock->ops->set_peek_off(sk, val);
1084 		else
1085 			ret = -EOPNOTSUPP;
1086 		break;
1087 
1088 	case SO_NOFCS:
1089 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1090 		break;
1091 
1092 	case SO_SELECT_ERR_QUEUE:
1093 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1094 		break;
1095 
1096 #ifdef CONFIG_NET_RX_BUSY_POLL
1097 	case SO_BUSY_POLL:
1098 		/* allow unprivileged users to decrease the value */
1099 		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1100 			ret = -EPERM;
1101 		else {
1102 			if (val < 0)
1103 				ret = -EINVAL;
1104 			else
1105 				sk->sk_ll_usec = val;
1106 		}
1107 		break;
1108 #endif
1109 
1110 	case SO_MAX_PACING_RATE:
1111 		{
1112 		unsigned long ulval = (val == ~0U) ? ~0UL : val;
1113 
1114 		if (sizeof(ulval) != sizeof(val) &&
1115 		    optlen >= sizeof(ulval) &&
1116 		    get_user(ulval, (unsigned long __user *)optval)) {
1117 			ret = -EFAULT;
1118 			break;
1119 		}
1120 		if (ulval != ~0UL)
1121 			cmpxchg(&sk->sk_pacing_status,
1122 				SK_PACING_NONE,
1123 				SK_PACING_NEEDED);
1124 		sk->sk_max_pacing_rate = ulval;
1125 		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1126 		break;
1127 		}
1128 	case SO_INCOMING_CPU:
1129 		sk->sk_incoming_cpu = val;
1130 		break;
1131 
1132 	case SO_CNX_ADVICE:
1133 		if (val == 1)
1134 			dst_negative_advice(sk);
1135 		break;
1136 
1137 	case SO_ZEROCOPY:
1138 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1139 			if (!((sk->sk_type == SOCK_STREAM &&
1140 			       sk->sk_protocol == IPPROTO_TCP) ||
1141 			      (sk->sk_type == SOCK_DGRAM &&
1142 			       sk->sk_protocol == IPPROTO_UDP)))
1143 				ret = -ENOTSUPP;
1144 		} else if (sk->sk_family != PF_RDS) {
1145 			ret = -ENOTSUPP;
1146 		}
1147 		if (!ret) {
1148 			if (val < 0 || val > 1)
1149 				ret = -EINVAL;
1150 			else
1151 				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1152 		}
1153 		break;
1154 
1155 	case SO_TXTIME:
1156 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1157 			ret = -EPERM;
1158 		} else if (optlen != sizeof(struct sock_txtime)) {
1159 			ret = -EINVAL;
1160 		} else if (copy_from_user(&sk_txtime, optval,
1161 			   sizeof(struct sock_txtime))) {
1162 			ret = -EFAULT;
1163 		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1164 			ret = -EINVAL;
1165 		} else {
1166 			sock_valbool_flag(sk, SOCK_TXTIME, true);
1167 			sk->sk_clockid = sk_txtime.clockid;
1168 			sk->sk_txtime_deadline_mode =
1169 				!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1170 			sk->sk_txtime_report_errors =
1171 				!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1172 		}
1173 		break;
1174 
1175 	case SO_BINDTOIFINDEX:
1176 		ret = sock_setbindtodevice_locked(sk, val);
1177 		break;
1178 
1179 	default:
1180 		ret = -ENOPROTOOPT;
1181 		break;
1182 	}
1183 	release_sock(sk);
1184 	return ret;
1185 }
1186 EXPORT_SYMBOL(sock_setsockopt);
1187 
1188 
1189 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1190 			  struct ucred *ucred)
1191 {
1192 	ucred->pid = pid_vnr(pid);
1193 	ucred->uid = ucred->gid = -1;
1194 	if (cred) {
1195 		struct user_namespace *current_ns = current_user_ns();
1196 
1197 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1198 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1199 	}
1200 }
1201 
1202 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1203 {
1204 	struct user_namespace *user_ns = current_user_ns();
1205 	int i;
1206 
1207 	for (i = 0; i < src->ngroups; i++)
1208 		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1209 			return -EFAULT;
1210 
1211 	return 0;
1212 }
1213 
1214 int sock_getsockopt(struct socket *sock, int level, int optname,
1215 		    char __user *optval, int __user *optlen)
1216 {
1217 	struct sock *sk = sock->sk;
1218 
1219 	union {
1220 		int val;
1221 		u64 val64;
1222 		unsigned long ulval;
1223 		struct linger ling;
1224 		struct old_timeval32 tm32;
1225 		struct __kernel_old_timeval tm;
1226 		struct  __kernel_sock_timeval stm;
1227 		struct sock_txtime txtime;
1228 	} v;
1229 
1230 	int lv = sizeof(int);
1231 	int len;
1232 
1233 	if (get_user(len, optlen))
1234 		return -EFAULT;
1235 	if (len < 0)
1236 		return -EINVAL;
1237 
1238 	memset(&v, 0, sizeof(v));
1239 
1240 	switch (optname) {
1241 	case SO_DEBUG:
1242 		v.val = sock_flag(sk, SOCK_DBG);
1243 		break;
1244 
1245 	case SO_DONTROUTE:
1246 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1247 		break;
1248 
1249 	case SO_BROADCAST:
1250 		v.val = sock_flag(sk, SOCK_BROADCAST);
1251 		break;
1252 
1253 	case SO_SNDBUF:
1254 		v.val = sk->sk_sndbuf;
1255 		break;
1256 
1257 	case SO_RCVBUF:
1258 		v.val = sk->sk_rcvbuf;
1259 		break;
1260 
1261 	case SO_REUSEADDR:
1262 		v.val = sk->sk_reuse;
1263 		break;
1264 
1265 	case SO_REUSEPORT:
1266 		v.val = sk->sk_reuseport;
1267 		break;
1268 
1269 	case SO_KEEPALIVE:
1270 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1271 		break;
1272 
1273 	case SO_TYPE:
1274 		v.val = sk->sk_type;
1275 		break;
1276 
1277 	case SO_PROTOCOL:
1278 		v.val = sk->sk_protocol;
1279 		break;
1280 
1281 	case SO_DOMAIN:
1282 		v.val = sk->sk_family;
1283 		break;
1284 
1285 	case SO_ERROR:
1286 		v.val = -sock_error(sk);
1287 		if (v.val == 0)
1288 			v.val = xchg(&sk->sk_err_soft, 0);
1289 		break;
1290 
1291 	case SO_OOBINLINE:
1292 		v.val = sock_flag(sk, SOCK_URGINLINE);
1293 		break;
1294 
1295 	case SO_NO_CHECK:
1296 		v.val = sk->sk_no_check_tx;
1297 		break;
1298 
1299 	case SO_PRIORITY:
1300 		v.val = sk->sk_priority;
1301 		break;
1302 
1303 	case SO_LINGER:
1304 		lv		= sizeof(v.ling);
1305 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1306 		v.ling.l_linger	= sk->sk_lingertime / HZ;
1307 		break;
1308 
1309 	case SO_BSDCOMPAT:
1310 		sock_warn_obsolete_bsdism("getsockopt");
1311 		break;
1312 
1313 	case SO_TIMESTAMP_OLD:
1314 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1315 				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1316 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1317 		break;
1318 
1319 	case SO_TIMESTAMPNS_OLD:
1320 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1321 		break;
1322 
1323 	case SO_TIMESTAMP_NEW:
1324 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1325 		break;
1326 
1327 	case SO_TIMESTAMPNS_NEW:
1328 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1329 		break;
1330 
1331 	case SO_TIMESTAMPING_OLD:
1332 		v.val = sk->sk_tsflags;
1333 		break;
1334 
1335 	case SO_RCVTIMEO_OLD:
1336 	case SO_RCVTIMEO_NEW:
1337 		lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1338 		break;
1339 
1340 	case SO_SNDTIMEO_OLD:
1341 	case SO_SNDTIMEO_NEW:
1342 		lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1343 		break;
1344 
1345 	case SO_RCVLOWAT:
1346 		v.val = sk->sk_rcvlowat;
1347 		break;
1348 
1349 	case SO_SNDLOWAT:
1350 		v.val = 1;
1351 		break;
1352 
1353 	case SO_PASSCRED:
1354 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1355 		break;
1356 
1357 	case SO_PEERCRED:
1358 	{
1359 		struct ucred peercred;
1360 		if (len > sizeof(peercred))
1361 			len = sizeof(peercred);
1362 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1363 		if (copy_to_user(optval, &peercred, len))
1364 			return -EFAULT;
1365 		goto lenout;
1366 	}
1367 
1368 	case SO_PEERGROUPS:
1369 	{
1370 		int ret, n;
1371 
1372 		if (!sk->sk_peer_cred)
1373 			return -ENODATA;
1374 
1375 		n = sk->sk_peer_cred->group_info->ngroups;
1376 		if (len < n * sizeof(gid_t)) {
1377 			len = n * sizeof(gid_t);
1378 			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1379 		}
1380 		len = n * sizeof(gid_t);
1381 
1382 		ret = groups_to_user((gid_t __user *)optval,
1383 				     sk->sk_peer_cred->group_info);
1384 		if (ret)
1385 			return ret;
1386 		goto lenout;
1387 	}
1388 
1389 	case SO_PEERNAME:
1390 	{
1391 		char address[128];
1392 
1393 		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1394 		if (lv < 0)
1395 			return -ENOTCONN;
1396 		if (lv < len)
1397 			return -EINVAL;
1398 		if (copy_to_user(optval, address, len))
1399 			return -EFAULT;
1400 		goto lenout;
1401 	}
1402 
1403 	/* Dubious BSD thing... Probably nobody even uses it, but
1404 	 * the UNIX standard wants it for whatever reason... -DaveM
1405 	 */
1406 	case SO_ACCEPTCONN:
1407 		v.val = sk->sk_state == TCP_LISTEN;
1408 		break;
1409 
1410 	case SO_PASSSEC:
1411 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1412 		break;
1413 
1414 	case SO_PEERSEC:
1415 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1416 
1417 	case SO_MARK:
1418 		v.val = sk->sk_mark;
1419 		break;
1420 
1421 	case SO_RXQ_OVFL:
1422 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1423 		break;
1424 
1425 	case SO_WIFI_STATUS:
1426 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1427 		break;
1428 
1429 	case SO_PEEK_OFF:
1430 		if (!sock->ops->set_peek_off)
1431 			return -EOPNOTSUPP;
1432 
1433 		v.val = sk->sk_peek_off;
1434 		break;
1435 	case SO_NOFCS:
1436 		v.val = sock_flag(sk, SOCK_NOFCS);
1437 		break;
1438 
1439 	case SO_BINDTODEVICE:
1440 		return sock_getbindtodevice(sk, optval, optlen, len);
1441 
1442 	case SO_GET_FILTER:
1443 		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1444 		if (len < 0)
1445 			return len;
1446 
1447 		goto lenout;
1448 
1449 	case SO_LOCK_FILTER:
1450 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1451 		break;
1452 
1453 	case SO_BPF_EXTENSIONS:
1454 		v.val = bpf_tell_extensions();
1455 		break;
1456 
1457 	case SO_SELECT_ERR_QUEUE:
1458 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1459 		break;
1460 
1461 #ifdef CONFIG_NET_RX_BUSY_POLL
1462 	case SO_BUSY_POLL:
1463 		v.val = sk->sk_ll_usec;
1464 		break;
1465 #endif
1466 
1467 	case SO_MAX_PACING_RATE:
1468 		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1469 			lv = sizeof(v.ulval);
1470 			v.ulval = sk->sk_max_pacing_rate;
1471 		} else {
1472 			/* 32bit version */
1473 			v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1474 		}
1475 		break;
1476 
1477 	case SO_INCOMING_CPU:
1478 		v.val = sk->sk_incoming_cpu;
1479 		break;
1480 
1481 	case SO_MEMINFO:
1482 	{
1483 		u32 meminfo[SK_MEMINFO_VARS];
1484 
1485 		sk_get_meminfo(sk, meminfo);
1486 
1487 		len = min_t(unsigned int, len, sizeof(meminfo));
1488 		if (copy_to_user(optval, &meminfo, len))
1489 			return -EFAULT;
1490 
1491 		goto lenout;
1492 	}
1493 
1494 #ifdef CONFIG_NET_RX_BUSY_POLL
1495 	case SO_INCOMING_NAPI_ID:
1496 		v.val = READ_ONCE(sk->sk_napi_id);
1497 
1498 		/* aggregate non-NAPI IDs down to 0 */
1499 		if (v.val < MIN_NAPI_ID)
1500 			v.val = 0;
1501 
1502 		break;
1503 #endif
1504 
1505 	case SO_COOKIE:
1506 		lv = sizeof(u64);
1507 		if (len < lv)
1508 			return -EINVAL;
1509 		v.val64 = sock_gen_cookie(sk);
1510 		break;
1511 
1512 	case SO_ZEROCOPY:
1513 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1514 		break;
1515 
1516 	case SO_TXTIME:
1517 		lv = sizeof(v.txtime);
1518 		v.txtime.clockid = sk->sk_clockid;
1519 		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1520 				  SOF_TXTIME_DEADLINE_MODE : 0;
1521 		v.txtime.flags |= sk->sk_txtime_report_errors ?
1522 				  SOF_TXTIME_REPORT_ERRORS : 0;
1523 		break;
1524 
1525 	case SO_BINDTOIFINDEX:
1526 		v.val = sk->sk_bound_dev_if;
1527 		break;
1528 
1529 	default:
1530 		/* We implement the SO_SNDLOWAT etc to not be settable
1531 		 * (1003.1g 7).
1532 		 */
1533 		return -ENOPROTOOPT;
1534 	}
1535 
1536 	if (len > lv)
1537 		len = lv;
1538 	if (copy_to_user(optval, &v, len))
1539 		return -EFAULT;
1540 lenout:
1541 	if (put_user(len, optlen))
1542 		return -EFAULT;
1543 	return 0;
1544 }
1545 
1546 /*
1547  * Initialize an sk_lock.
1548  *
1549  * (We also register the sk_lock with the lock validator.)
1550  */
1551 static inline void sock_lock_init(struct sock *sk)
1552 {
1553 	if (sk->sk_kern_sock)
1554 		sock_lock_init_class_and_name(
1555 			sk,
1556 			af_family_kern_slock_key_strings[sk->sk_family],
1557 			af_family_kern_slock_keys + sk->sk_family,
1558 			af_family_kern_key_strings[sk->sk_family],
1559 			af_family_kern_keys + sk->sk_family);
1560 	else
1561 		sock_lock_init_class_and_name(
1562 			sk,
1563 			af_family_slock_key_strings[sk->sk_family],
1564 			af_family_slock_keys + sk->sk_family,
1565 			af_family_key_strings[sk->sk_family],
1566 			af_family_keys + sk->sk_family);
1567 }
1568 
1569 /*
1570  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1571  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1572  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1573  */
1574 static void sock_copy(struct sock *nsk, const struct sock *osk)
1575 {
1576 #ifdef CONFIG_SECURITY_NETWORK
1577 	void *sptr = nsk->sk_security;
1578 #endif
1579 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1580 
1581 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1582 	       osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1583 
1584 #ifdef CONFIG_SECURITY_NETWORK
1585 	nsk->sk_security = sptr;
1586 	security_sk_clone(osk, nsk);
1587 #endif
1588 }
1589 
1590 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1591 		int family)
1592 {
1593 	struct sock *sk;
1594 	struct kmem_cache *slab;
1595 
1596 	slab = prot->slab;
1597 	if (slab != NULL) {
1598 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1599 		if (!sk)
1600 			return sk;
1601 		if (want_init_on_alloc(priority))
1602 			sk_prot_clear_nulls(sk, prot->obj_size);
1603 	} else
1604 		sk = kmalloc(prot->obj_size, priority);
1605 
1606 	if (sk != NULL) {
1607 		if (security_sk_alloc(sk, family, priority))
1608 			goto out_free;
1609 
1610 		if (!try_module_get(prot->owner))
1611 			goto out_free_sec;
1612 		sk_tx_queue_clear(sk);
1613 	}
1614 
1615 	return sk;
1616 
1617 out_free_sec:
1618 	security_sk_free(sk);
1619 out_free:
1620 	if (slab != NULL)
1621 		kmem_cache_free(slab, sk);
1622 	else
1623 		kfree(sk);
1624 	return NULL;
1625 }
1626 
1627 static void sk_prot_free(struct proto *prot, struct sock *sk)
1628 {
1629 	struct kmem_cache *slab;
1630 	struct module *owner;
1631 
1632 	owner = prot->owner;
1633 	slab = prot->slab;
1634 
1635 	cgroup_sk_free(&sk->sk_cgrp_data);
1636 	mem_cgroup_sk_free(sk);
1637 	security_sk_free(sk);
1638 	if (slab != NULL)
1639 		kmem_cache_free(slab, sk);
1640 	else
1641 		kfree(sk);
1642 	module_put(owner);
1643 }
1644 
1645 /**
1646  *	sk_alloc - All socket objects are allocated here
1647  *	@net: the applicable net namespace
1648  *	@family: protocol family
1649  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1650  *	@prot: struct proto associated with this new sock instance
1651  *	@kern: is this to be a kernel socket?
1652  */
1653 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1654 		      struct proto *prot, int kern)
1655 {
1656 	struct sock *sk;
1657 
1658 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1659 	if (sk) {
1660 		sk->sk_family = family;
1661 		/*
1662 		 * See comment in struct sock definition to understand
1663 		 * why we need sk_prot_creator -acme
1664 		 */
1665 		sk->sk_prot = sk->sk_prot_creator = prot;
1666 		sk->sk_kern_sock = kern;
1667 		sock_lock_init(sk);
1668 		sk->sk_net_refcnt = kern ? 0 : 1;
1669 		if (likely(sk->sk_net_refcnt)) {
1670 			get_net(net);
1671 			sock_inuse_add(net, 1);
1672 		}
1673 
1674 		sock_net_set(sk, net);
1675 		refcount_set(&sk->sk_wmem_alloc, 1);
1676 
1677 		mem_cgroup_sk_alloc(sk);
1678 		cgroup_sk_alloc(&sk->sk_cgrp_data);
1679 		sock_update_classid(&sk->sk_cgrp_data);
1680 		sock_update_netprioidx(&sk->sk_cgrp_data);
1681 	}
1682 
1683 	return sk;
1684 }
1685 EXPORT_SYMBOL(sk_alloc);
1686 
1687 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1688  * grace period. This is the case for UDP sockets and TCP listeners.
1689  */
1690 static void __sk_destruct(struct rcu_head *head)
1691 {
1692 	struct sock *sk = container_of(head, struct sock, sk_rcu);
1693 	struct sk_filter *filter;
1694 
1695 	if (sk->sk_destruct)
1696 		sk->sk_destruct(sk);
1697 
1698 	filter = rcu_dereference_check(sk->sk_filter,
1699 				       refcount_read(&sk->sk_wmem_alloc) == 0);
1700 	if (filter) {
1701 		sk_filter_uncharge(sk, filter);
1702 		RCU_INIT_POINTER(sk->sk_filter, NULL);
1703 	}
1704 
1705 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1706 
1707 #ifdef CONFIG_BPF_SYSCALL
1708 	bpf_sk_storage_free(sk);
1709 #endif
1710 
1711 	if (atomic_read(&sk->sk_omem_alloc))
1712 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1713 			 __func__, atomic_read(&sk->sk_omem_alloc));
1714 
1715 	if (sk->sk_frag.page) {
1716 		put_page(sk->sk_frag.page);
1717 		sk->sk_frag.page = NULL;
1718 	}
1719 
1720 	if (sk->sk_peer_cred)
1721 		put_cred(sk->sk_peer_cred);
1722 	put_pid(sk->sk_peer_pid);
1723 	if (likely(sk->sk_net_refcnt))
1724 		put_net(sock_net(sk));
1725 	sk_prot_free(sk->sk_prot_creator, sk);
1726 }
1727 
1728 void sk_destruct(struct sock *sk)
1729 {
1730 	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1731 
1732 	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1733 		reuseport_detach_sock(sk);
1734 		use_call_rcu = true;
1735 	}
1736 
1737 	if (use_call_rcu)
1738 		call_rcu(&sk->sk_rcu, __sk_destruct);
1739 	else
1740 		__sk_destruct(&sk->sk_rcu);
1741 }
1742 
1743 static void __sk_free(struct sock *sk)
1744 {
1745 	if (likely(sk->sk_net_refcnt))
1746 		sock_inuse_add(sock_net(sk), -1);
1747 
1748 	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1749 		sock_diag_broadcast_destroy(sk);
1750 	else
1751 		sk_destruct(sk);
1752 }
1753 
1754 void sk_free(struct sock *sk)
1755 {
1756 	/*
1757 	 * We subtract one from sk_wmem_alloc and can know if
1758 	 * some packets are still in some tx queue.
1759 	 * If not null, sock_wfree() will call __sk_free(sk) later
1760 	 */
1761 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1762 		__sk_free(sk);
1763 }
1764 EXPORT_SYMBOL(sk_free);
1765 
1766 static void sk_init_common(struct sock *sk)
1767 {
1768 	skb_queue_head_init(&sk->sk_receive_queue);
1769 	skb_queue_head_init(&sk->sk_write_queue);
1770 	skb_queue_head_init(&sk->sk_error_queue);
1771 
1772 	rwlock_init(&sk->sk_callback_lock);
1773 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1774 			af_rlock_keys + sk->sk_family,
1775 			af_family_rlock_key_strings[sk->sk_family]);
1776 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1777 			af_wlock_keys + sk->sk_family,
1778 			af_family_wlock_key_strings[sk->sk_family]);
1779 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1780 			af_elock_keys + sk->sk_family,
1781 			af_family_elock_key_strings[sk->sk_family]);
1782 	lockdep_set_class_and_name(&sk->sk_callback_lock,
1783 			af_callback_keys + sk->sk_family,
1784 			af_family_clock_key_strings[sk->sk_family]);
1785 }
1786 
1787 /**
1788  *	sk_clone_lock - clone a socket, and lock its clone
1789  *	@sk: the socket to clone
1790  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1791  *
1792  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1793  */
1794 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1795 {
1796 	struct sock *newsk;
1797 	bool is_charged = true;
1798 
1799 	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1800 	if (newsk != NULL) {
1801 		struct sk_filter *filter;
1802 
1803 		sock_copy(newsk, sk);
1804 
1805 		newsk->sk_prot_creator = sk->sk_prot;
1806 
1807 		/* SANITY */
1808 		if (likely(newsk->sk_net_refcnt))
1809 			get_net(sock_net(newsk));
1810 		sk_node_init(&newsk->sk_node);
1811 		sock_lock_init(newsk);
1812 		bh_lock_sock(newsk);
1813 		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1814 		newsk->sk_backlog.len = 0;
1815 
1816 		atomic_set(&newsk->sk_rmem_alloc, 0);
1817 		/*
1818 		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1819 		 */
1820 		refcount_set(&newsk->sk_wmem_alloc, 1);
1821 		atomic_set(&newsk->sk_omem_alloc, 0);
1822 		sk_init_common(newsk);
1823 
1824 		newsk->sk_dst_cache	= NULL;
1825 		newsk->sk_dst_pending_confirm = 0;
1826 		newsk->sk_wmem_queued	= 0;
1827 		newsk->sk_forward_alloc = 0;
1828 		atomic_set(&newsk->sk_drops, 0);
1829 		newsk->sk_send_head	= NULL;
1830 		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1831 		atomic_set(&newsk->sk_zckey, 0);
1832 
1833 		sock_reset_flag(newsk, SOCK_DONE);
1834 		mem_cgroup_sk_alloc(newsk);
1835 		cgroup_sk_alloc(&newsk->sk_cgrp_data);
1836 
1837 		rcu_read_lock();
1838 		filter = rcu_dereference(sk->sk_filter);
1839 		if (filter != NULL)
1840 			/* though it's an empty new sock, the charging may fail
1841 			 * if sysctl_optmem_max was changed between creation of
1842 			 * original socket and cloning
1843 			 */
1844 			is_charged = sk_filter_charge(newsk, filter);
1845 		RCU_INIT_POINTER(newsk->sk_filter, filter);
1846 		rcu_read_unlock();
1847 
1848 		if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1849 			/* We need to make sure that we don't uncharge the new
1850 			 * socket if we couldn't charge it in the first place
1851 			 * as otherwise we uncharge the parent's filter.
1852 			 */
1853 			if (!is_charged)
1854 				RCU_INIT_POINTER(newsk->sk_filter, NULL);
1855 			sk_free_unlock_clone(newsk);
1856 			newsk = NULL;
1857 			goto out;
1858 		}
1859 		RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1860 
1861 		if (bpf_sk_storage_clone(sk, newsk)) {
1862 			sk_free_unlock_clone(newsk);
1863 			newsk = NULL;
1864 			goto out;
1865 		}
1866 
1867 		newsk->sk_err	   = 0;
1868 		newsk->sk_err_soft = 0;
1869 		newsk->sk_priority = 0;
1870 		newsk->sk_incoming_cpu = raw_smp_processor_id();
1871 		if (likely(newsk->sk_net_refcnt))
1872 			sock_inuse_add(sock_net(newsk), 1);
1873 
1874 		/*
1875 		 * Before updating sk_refcnt, we must commit prior changes to memory
1876 		 * (Documentation/RCU/rculist_nulls.txt for details)
1877 		 */
1878 		smp_wmb();
1879 		refcount_set(&newsk->sk_refcnt, 2);
1880 
1881 		/*
1882 		 * Increment the counter in the same struct proto as the master
1883 		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1884 		 * is the same as sk->sk_prot->socks, as this field was copied
1885 		 * with memcpy).
1886 		 *
1887 		 * This _changes_ the previous behaviour, where
1888 		 * tcp_create_openreq_child always was incrementing the
1889 		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1890 		 * to be taken into account in all callers. -acme
1891 		 */
1892 		sk_refcnt_debug_inc(newsk);
1893 		sk_set_socket(newsk, NULL);
1894 		RCU_INIT_POINTER(newsk->sk_wq, NULL);
1895 
1896 		if (newsk->sk_prot->sockets_allocated)
1897 			sk_sockets_allocated_inc(newsk);
1898 
1899 		if (sock_needs_netstamp(sk) &&
1900 		    newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1901 			net_enable_timestamp();
1902 	}
1903 out:
1904 	return newsk;
1905 }
1906 EXPORT_SYMBOL_GPL(sk_clone_lock);
1907 
1908 void sk_free_unlock_clone(struct sock *sk)
1909 {
1910 	/* It is still raw copy of parent, so invalidate
1911 	 * destructor and make plain sk_free() */
1912 	sk->sk_destruct = NULL;
1913 	bh_unlock_sock(sk);
1914 	sk_free(sk);
1915 }
1916 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1917 
1918 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1919 {
1920 	u32 max_segs = 1;
1921 
1922 	sk_dst_set(sk, dst);
1923 	sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1924 	if (sk->sk_route_caps & NETIF_F_GSO)
1925 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1926 	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1927 	if (sk_can_gso(sk)) {
1928 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1929 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1930 		} else {
1931 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1932 			sk->sk_gso_max_size = dst->dev->gso_max_size;
1933 			max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1934 		}
1935 	}
1936 	sk->sk_gso_max_segs = max_segs;
1937 }
1938 EXPORT_SYMBOL_GPL(sk_setup_caps);
1939 
1940 /*
1941  *	Simple resource managers for sockets.
1942  */
1943 
1944 
1945 /*
1946  * Write buffer destructor automatically called from kfree_skb.
1947  */
1948 void sock_wfree(struct sk_buff *skb)
1949 {
1950 	struct sock *sk = skb->sk;
1951 	unsigned int len = skb->truesize;
1952 
1953 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1954 		/*
1955 		 * Keep a reference on sk_wmem_alloc, this will be released
1956 		 * after sk_write_space() call
1957 		 */
1958 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1959 		sk->sk_write_space(sk);
1960 		len = 1;
1961 	}
1962 	/*
1963 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1964 	 * could not do because of in-flight packets
1965 	 */
1966 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1967 		__sk_free(sk);
1968 }
1969 EXPORT_SYMBOL(sock_wfree);
1970 
1971 /* This variant of sock_wfree() is used by TCP,
1972  * since it sets SOCK_USE_WRITE_QUEUE.
1973  */
1974 void __sock_wfree(struct sk_buff *skb)
1975 {
1976 	struct sock *sk = skb->sk;
1977 
1978 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1979 		__sk_free(sk);
1980 }
1981 
1982 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1983 {
1984 	skb_orphan(skb);
1985 	skb->sk = sk;
1986 #ifdef CONFIG_INET
1987 	if (unlikely(!sk_fullsock(sk))) {
1988 		skb->destructor = sock_edemux;
1989 		sock_hold(sk);
1990 		return;
1991 	}
1992 #endif
1993 	skb->destructor = sock_wfree;
1994 	skb_set_hash_from_sk(skb, sk);
1995 	/*
1996 	 * We used to take a refcount on sk, but following operation
1997 	 * is enough to guarantee sk_free() wont free this sock until
1998 	 * all in-flight packets are completed
1999 	 */
2000 	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2001 }
2002 EXPORT_SYMBOL(skb_set_owner_w);
2003 
2004 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2005 {
2006 #ifdef CONFIG_TLS_DEVICE
2007 	/* Drivers depend on in-order delivery for crypto offload,
2008 	 * partial orphan breaks out-of-order-OK logic.
2009 	 */
2010 	if (skb->decrypted)
2011 		return false;
2012 #endif
2013 	return (skb->destructor == sock_wfree ||
2014 		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2015 }
2016 
2017 /* This helper is used by netem, as it can hold packets in its
2018  * delay queue. We want to allow the owner socket to send more
2019  * packets, as if they were already TX completed by a typical driver.
2020  * But we also want to keep skb->sk set because some packet schedulers
2021  * rely on it (sch_fq for example).
2022  */
2023 void skb_orphan_partial(struct sk_buff *skb)
2024 {
2025 	if (skb_is_tcp_pure_ack(skb))
2026 		return;
2027 
2028 	if (can_skb_orphan_partial(skb)) {
2029 		struct sock *sk = skb->sk;
2030 
2031 		if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2032 			WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2033 			skb->destructor = sock_efree;
2034 		}
2035 	} else {
2036 		skb_orphan(skb);
2037 	}
2038 }
2039 EXPORT_SYMBOL(skb_orphan_partial);
2040 
2041 /*
2042  * Read buffer destructor automatically called from kfree_skb.
2043  */
2044 void sock_rfree(struct sk_buff *skb)
2045 {
2046 	struct sock *sk = skb->sk;
2047 	unsigned int len = skb->truesize;
2048 
2049 	atomic_sub(len, &sk->sk_rmem_alloc);
2050 	sk_mem_uncharge(sk, len);
2051 }
2052 EXPORT_SYMBOL(sock_rfree);
2053 
2054 /*
2055  * Buffer destructor for skbs that are not used directly in read or write
2056  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2057  */
2058 void sock_efree(struct sk_buff *skb)
2059 {
2060 	sock_put(skb->sk);
2061 }
2062 EXPORT_SYMBOL(sock_efree);
2063 
2064 kuid_t sock_i_uid(struct sock *sk)
2065 {
2066 	kuid_t uid;
2067 
2068 	read_lock_bh(&sk->sk_callback_lock);
2069 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2070 	read_unlock_bh(&sk->sk_callback_lock);
2071 	return uid;
2072 }
2073 EXPORT_SYMBOL(sock_i_uid);
2074 
2075 unsigned long sock_i_ino(struct sock *sk)
2076 {
2077 	unsigned long ino;
2078 
2079 	read_lock_bh(&sk->sk_callback_lock);
2080 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2081 	read_unlock_bh(&sk->sk_callback_lock);
2082 	return ino;
2083 }
2084 EXPORT_SYMBOL(sock_i_ino);
2085 
2086 /*
2087  * Allocate a skb from the socket's send buffer.
2088  */
2089 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2090 			     gfp_t priority)
2091 {
2092 	if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2093 		struct sk_buff *skb = alloc_skb(size, priority);
2094 		if (skb) {
2095 			skb_set_owner_w(skb, sk);
2096 			return skb;
2097 		}
2098 	}
2099 	return NULL;
2100 }
2101 EXPORT_SYMBOL(sock_wmalloc);
2102 
2103 static void sock_ofree(struct sk_buff *skb)
2104 {
2105 	struct sock *sk = skb->sk;
2106 
2107 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2108 }
2109 
2110 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2111 			     gfp_t priority)
2112 {
2113 	struct sk_buff *skb;
2114 
2115 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2116 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2117 	    sysctl_optmem_max)
2118 		return NULL;
2119 
2120 	skb = alloc_skb(size, priority);
2121 	if (!skb)
2122 		return NULL;
2123 
2124 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2125 	skb->sk = sk;
2126 	skb->destructor = sock_ofree;
2127 	return skb;
2128 }
2129 
2130 /*
2131  * Allocate a memory block from the socket's option memory buffer.
2132  */
2133 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2134 {
2135 	if ((unsigned int)size <= sysctl_optmem_max &&
2136 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2137 		void *mem;
2138 		/* First do the add, to avoid the race if kmalloc
2139 		 * might sleep.
2140 		 */
2141 		atomic_add(size, &sk->sk_omem_alloc);
2142 		mem = kmalloc(size, priority);
2143 		if (mem)
2144 			return mem;
2145 		atomic_sub(size, &sk->sk_omem_alloc);
2146 	}
2147 	return NULL;
2148 }
2149 EXPORT_SYMBOL(sock_kmalloc);
2150 
2151 /* Free an option memory block. Note, we actually want the inline
2152  * here as this allows gcc to detect the nullify and fold away the
2153  * condition entirely.
2154  */
2155 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2156 				  const bool nullify)
2157 {
2158 	if (WARN_ON_ONCE(!mem))
2159 		return;
2160 	if (nullify)
2161 		kzfree(mem);
2162 	else
2163 		kfree(mem);
2164 	atomic_sub(size, &sk->sk_omem_alloc);
2165 }
2166 
2167 void sock_kfree_s(struct sock *sk, void *mem, int size)
2168 {
2169 	__sock_kfree_s(sk, mem, size, false);
2170 }
2171 EXPORT_SYMBOL(sock_kfree_s);
2172 
2173 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2174 {
2175 	__sock_kfree_s(sk, mem, size, true);
2176 }
2177 EXPORT_SYMBOL(sock_kzfree_s);
2178 
2179 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2180    I think, these locks should be removed for datagram sockets.
2181  */
2182 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2183 {
2184 	DEFINE_WAIT(wait);
2185 
2186 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2187 	for (;;) {
2188 		if (!timeo)
2189 			break;
2190 		if (signal_pending(current))
2191 			break;
2192 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2193 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2194 		if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2195 			break;
2196 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2197 			break;
2198 		if (sk->sk_err)
2199 			break;
2200 		timeo = schedule_timeout(timeo);
2201 	}
2202 	finish_wait(sk_sleep(sk), &wait);
2203 	return timeo;
2204 }
2205 
2206 
2207 /*
2208  *	Generic send/receive buffer handlers
2209  */
2210 
2211 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2212 				     unsigned long data_len, int noblock,
2213 				     int *errcode, int max_page_order)
2214 {
2215 	struct sk_buff *skb;
2216 	long timeo;
2217 	int err;
2218 
2219 	timeo = sock_sndtimeo(sk, noblock);
2220 	for (;;) {
2221 		err = sock_error(sk);
2222 		if (err != 0)
2223 			goto failure;
2224 
2225 		err = -EPIPE;
2226 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2227 			goto failure;
2228 
2229 		if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2230 			break;
2231 
2232 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2233 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2234 		err = -EAGAIN;
2235 		if (!timeo)
2236 			goto failure;
2237 		if (signal_pending(current))
2238 			goto interrupted;
2239 		timeo = sock_wait_for_wmem(sk, timeo);
2240 	}
2241 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2242 				   errcode, sk->sk_allocation);
2243 	if (skb)
2244 		skb_set_owner_w(skb, sk);
2245 	return skb;
2246 
2247 interrupted:
2248 	err = sock_intr_errno(timeo);
2249 failure:
2250 	*errcode = err;
2251 	return NULL;
2252 }
2253 EXPORT_SYMBOL(sock_alloc_send_pskb);
2254 
2255 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2256 				    int noblock, int *errcode)
2257 {
2258 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2259 }
2260 EXPORT_SYMBOL(sock_alloc_send_skb);
2261 
2262 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2263 		     struct sockcm_cookie *sockc)
2264 {
2265 	u32 tsflags;
2266 
2267 	switch (cmsg->cmsg_type) {
2268 	case SO_MARK:
2269 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2270 			return -EPERM;
2271 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2272 			return -EINVAL;
2273 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2274 		break;
2275 	case SO_TIMESTAMPING_OLD:
2276 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2277 			return -EINVAL;
2278 
2279 		tsflags = *(u32 *)CMSG_DATA(cmsg);
2280 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2281 			return -EINVAL;
2282 
2283 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2284 		sockc->tsflags |= tsflags;
2285 		break;
2286 	case SCM_TXTIME:
2287 		if (!sock_flag(sk, SOCK_TXTIME))
2288 			return -EINVAL;
2289 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2290 			return -EINVAL;
2291 		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2292 		break;
2293 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2294 	case SCM_RIGHTS:
2295 	case SCM_CREDENTIALS:
2296 		break;
2297 	default:
2298 		return -EINVAL;
2299 	}
2300 	return 0;
2301 }
2302 EXPORT_SYMBOL(__sock_cmsg_send);
2303 
2304 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2305 		   struct sockcm_cookie *sockc)
2306 {
2307 	struct cmsghdr *cmsg;
2308 	int ret;
2309 
2310 	for_each_cmsghdr(cmsg, msg) {
2311 		if (!CMSG_OK(msg, cmsg))
2312 			return -EINVAL;
2313 		if (cmsg->cmsg_level != SOL_SOCKET)
2314 			continue;
2315 		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2316 		if (ret)
2317 			return ret;
2318 	}
2319 	return 0;
2320 }
2321 EXPORT_SYMBOL(sock_cmsg_send);
2322 
2323 static void sk_enter_memory_pressure(struct sock *sk)
2324 {
2325 	if (!sk->sk_prot->enter_memory_pressure)
2326 		return;
2327 
2328 	sk->sk_prot->enter_memory_pressure(sk);
2329 }
2330 
2331 static void sk_leave_memory_pressure(struct sock *sk)
2332 {
2333 	if (sk->sk_prot->leave_memory_pressure) {
2334 		sk->sk_prot->leave_memory_pressure(sk);
2335 	} else {
2336 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2337 
2338 		if (memory_pressure && *memory_pressure)
2339 			*memory_pressure = 0;
2340 	}
2341 }
2342 
2343 /* On 32bit arches, an skb frag is limited to 2^15 */
2344 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
2345 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2346 
2347 /**
2348  * skb_page_frag_refill - check that a page_frag contains enough room
2349  * @sz: minimum size of the fragment we want to get
2350  * @pfrag: pointer to page_frag
2351  * @gfp: priority for memory allocation
2352  *
2353  * Note: While this allocator tries to use high order pages, there is
2354  * no guarantee that allocations succeed. Therefore, @sz MUST be
2355  * less or equal than PAGE_SIZE.
2356  */
2357 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2358 {
2359 	if (pfrag->page) {
2360 		if (page_ref_count(pfrag->page) == 1) {
2361 			pfrag->offset = 0;
2362 			return true;
2363 		}
2364 		if (pfrag->offset + sz <= pfrag->size)
2365 			return true;
2366 		put_page(pfrag->page);
2367 	}
2368 
2369 	pfrag->offset = 0;
2370 	if (SKB_FRAG_PAGE_ORDER &&
2371 	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2372 		/* Avoid direct reclaim but allow kswapd to wake */
2373 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2374 					  __GFP_COMP | __GFP_NOWARN |
2375 					  __GFP_NORETRY,
2376 					  SKB_FRAG_PAGE_ORDER);
2377 		if (likely(pfrag->page)) {
2378 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2379 			return true;
2380 		}
2381 	}
2382 	pfrag->page = alloc_page(gfp);
2383 	if (likely(pfrag->page)) {
2384 		pfrag->size = PAGE_SIZE;
2385 		return true;
2386 	}
2387 	return false;
2388 }
2389 EXPORT_SYMBOL(skb_page_frag_refill);
2390 
2391 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2392 {
2393 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2394 		return true;
2395 
2396 	sk_enter_memory_pressure(sk);
2397 	sk_stream_moderate_sndbuf(sk);
2398 	return false;
2399 }
2400 EXPORT_SYMBOL(sk_page_frag_refill);
2401 
2402 static void __lock_sock(struct sock *sk)
2403 	__releases(&sk->sk_lock.slock)
2404 	__acquires(&sk->sk_lock.slock)
2405 {
2406 	DEFINE_WAIT(wait);
2407 
2408 	for (;;) {
2409 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2410 					TASK_UNINTERRUPTIBLE);
2411 		spin_unlock_bh(&sk->sk_lock.slock);
2412 		schedule();
2413 		spin_lock_bh(&sk->sk_lock.slock);
2414 		if (!sock_owned_by_user(sk))
2415 			break;
2416 	}
2417 	finish_wait(&sk->sk_lock.wq, &wait);
2418 }
2419 
2420 void __release_sock(struct sock *sk)
2421 	__releases(&sk->sk_lock.slock)
2422 	__acquires(&sk->sk_lock.slock)
2423 {
2424 	struct sk_buff *skb, *next;
2425 
2426 	while ((skb = sk->sk_backlog.head) != NULL) {
2427 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2428 
2429 		spin_unlock_bh(&sk->sk_lock.slock);
2430 
2431 		do {
2432 			next = skb->next;
2433 			prefetch(next);
2434 			WARN_ON_ONCE(skb_dst_is_noref(skb));
2435 			skb_mark_not_on_list(skb);
2436 			sk_backlog_rcv(sk, skb);
2437 
2438 			cond_resched();
2439 
2440 			skb = next;
2441 		} while (skb != NULL);
2442 
2443 		spin_lock_bh(&sk->sk_lock.slock);
2444 	}
2445 
2446 	/*
2447 	 * Doing the zeroing here guarantee we can not loop forever
2448 	 * while a wild producer attempts to flood us.
2449 	 */
2450 	sk->sk_backlog.len = 0;
2451 }
2452 
2453 void __sk_flush_backlog(struct sock *sk)
2454 {
2455 	spin_lock_bh(&sk->sk_lock.slock);
2456 	__release_sock(sk);
2457 	spin_unlock_bh(&sk->sk_lock.slock);
2458 }
2459 
2460 /**
2461  * sk_wait_data - wait for data to arrive at sk_receive_queue
2462  * @sk:    sock to wait on
2463  * @timeo: for how long
2464  * @skb:   last skb seen on sk_receive_queue
2465  *
2466  * Now socket state including sk->sk_err is changed only under lock,
2467  * hence we may omit checks after joining wait queue.
2468  * We check receive queue before schedule() only as optimization;
2469  * it is very likely that release_sock() added new data.
2470  */
2471 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2472 {
2473 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2474 	int rc;
2475 
2476 	add_wait_queue(sk_sleep(sk), &wait);
2477 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2478 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2479 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2480 	remove_wait_queue(sk_sleep(sk), &wait);
2481 	return rc;
2482 }
2483 EXPORT_SYMBOL(sk_wait_data);
2484 
2485 /**
2486  *	__sk_mem_raise_allocated - increase memory_allocated
2487  *	@sk: socket
2488  *	@size: memory size to allocate
2489  *	@amt: pages to allocate
2490  *	@kind: allocation type
2491  *
2492  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2493  */
2494 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2495 {
2496 	struct proto *prot = sk->sk_prot;
2497 	long allocated = sk_memory_allocated_add(sk, amt);
2498 	bool charged = true;
2499 
2500 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2501 	    !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2502 		goto suppress_allocation;
2503 
2504 	/* Under limit. */
2505 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2506 		sk_leave_memory_pressure(sk);
2507 		return 1;
2508 	}
2509 
2510 	/* Under pressure. */
2511 	if (allocated > sk_prot_mem_limits(sk, 1))
2512 		sk_enter_memory_pressure(sk);
2513 
2514 	/* Over hard limit. */
2515 	if (allocated > sk_prot_mem_limits(sk, 2))
2516 		goto suppress_allocation;
2517 
2518 	/* guarantee minimum buffer size under pressure */
2519 	if (kind == SK_MEM_RECV) {
2520 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2521 			return 1;
2522 
2523 	} else { /* SK_MEM_SEND */
2524 		int wmem0 = sk_get_wmem0(sk, prot);
2525 
2526 		if (sk->sk_type == SOCK_STREAM) {
2527 			if (sk->sk_wmem_queued < wmem0)
2528 				return 1;
2529 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2530 				return 1;
2531 		}
2532 	}
2533 
2534 	if (sk_has_memory_pressure(sk)) {
2535 		u64 alloc;
2536 
2537 		if (!sk_under_memory_pressure(sk))
2538 			return 1;
2539 		alloc = sk_sockets_allocated_read_positive(sk);
2540 		if (sk_prot_mem_limits(sk, 2) > alloc *
2541 		    sk_mem_pages(sk->sk_wmem_queued +
2542 				 atomic_read(&sk->sk_rmem_alloc) +
2543 				 sk->sk_forward_alloc))
2544 			return 1;
2545 	}
2546 
2547 suppress_allocation:
2548 
2549 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2550 		sk_stream_moderate_sndbuf(sk);
2551 
2552 		/* Fail only if socket is _under_ its sndbuf.
2553 		 * In this case we cannot block, so that we have to fail.
2554 		 */
2555 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2556 			return 1;
2557 	}
2558 
2559 	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2560 		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2561 
2562 	sk_memory_allocated_sub(sk, amt);
2563 
2564 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2565 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2566 
2567 	return 0;
2568 }
2569 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2570 
2571 /**
2572  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2573  *	@sk: socket
2574  *	@size: memory size to allocate
2575  *	@kind: allocation type
2576  *
2577  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2578  *	rmem allocation. This function assumes that protocols which have
2579  *	memory_pressure use sk_wmem_queued as write buffer accounting.
2580  */
2581 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2582 {
2583 	int ret, amt = sk_mem_pages(size);
2584 
2585 	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2586 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2587 	if (!ret)
2588 		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2589 	return ret;
2590 }
2591 EXPORT_SYMBOL(__sk_mem_schedule);
2592 
2593 /**
2594  *	__sk_mem_reduce_allocated - reclaim memory_allocated
2595  *	@sk: socket
2596  *	@amount: number of quanta
2597  *
2598  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2599  */
2600 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2601 {
2602 	sk_memory_allocated_sub(sk, amount);
2603 
2604 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2605 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2606 
2607 	if (sk_under_memory_pressure(sk) &&
2608 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2609 		sk_leave_memory_pressure(sk);
2610 }
2611 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2612 
2613 /**
2614  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2615  *	@sk: socket
2616  *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2617  */
2618 void __sk_mem_reclaim(struct sock *sk, int amount)
2619 {
2620 	amount >>= SK_MEM_QUANTUM_SHIFT;
2621 	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2622 	__sk_mem_reduce_allocated(sk, amount);
2623 }
2624 EXPORT_SYMBOL(__sk_mem_reclaim);
2625 
2626 int sk_set_peek_off(struct sock *sk, int val)
2627 {
2628 	sk->sk_peek_off = val;
2629 	return 0;
2630 }
2631 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2632 
2633 /*
2634  * Set of default routines for initialising struct proto_ops when
2635  * the protocol does not support a particular function. In certain
2636  * cases where it makes no sense for a protocol to have a "do nothing"
2637  * function, some default processing is provided.
2638  */
2639 
2640 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2641 {
2642 	return -EOPNOTSUPP;
2643 }
2644 EXPORT_SYMBOL(sock_no_bind);
2645 
2646 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2647 		    int len, int flags)
2648 {
2649 	return -EOPNOTSUPP;
2650 }
2651 EXPORT_SYMBOL(sock_no_connect);
2652 
2653 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2654 {
2655 	return -EOPNOTSUPP;
2656 }
2657 EXPORT_SYMBOL(sock_no_socketpair);
2658 
2659 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2660 		   bool kern)
2661 {
2662 	return -EOPNOTSUPP;
2663 }
2664 EXPORT_SYMBOL(sock_no_accept);
2665 
2666 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2667 		    int peer)
2668 {
2669 	return -EOPNOTSUPP;
2670 }
2671 EXPORT_SYMBOL(sock_no_getname);
2672 
2673 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2674 {
2675 	return -EOPNOTSUPP;
2676 }
2677 EXPORT_SYMBOL(sock_no_ioctl);
2678 
2679 int sock_no_listen(struct socket *sock, int backlog)
2680 {
2681 	return -EOPNOTSUPP;
2682 }
2683 EXPORT_SYMBOL(sock_no_listen);
2684 
2685 int sock_no_shutdown(struct socket *sock, int how)
2686 {
2687 	return -EOPNOTSUPP;
2688 }
2689 EXPORT_SYMBOL(sock_no_shutdown);
2690 
2691 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2692 		    char __user *optval, unsigned int optlen)
2693 {
2694 	return -EOPNOTSUPP;
2695 }
2696 EXPORT_SYMBOL(sock_no_setsockopt);
2697 
2698 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2699 		    char __user *optval, int __user *optlen)
2700 {
2701 	return -EOPNOTSUPP;
2702 }
2703 EXPORT_SYMBOL(sock_no_getsockopt);
2704 
2705 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2706 {
2707 	return -EOPNOTSUPP;
2708 }
2709 EXPORT_SYMBOL(sock_no_sendmsg);
2710 
2711 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2712 {
2713 	return -EOPNOTSUPP;
2714 }
2715 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2716 
2717 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2718 		    int flags)
2719 {
2720 	return -EOPNOTSUPP;
2721 }
2722 EXPORT_SYMBOL(sock_no_recvmsg);
2723 
2724 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2725 {
2726 	/* Mirror missing mmap method error code */
2727 	return -ENODEV;
2728 }
2729 EXPORT_SYMBOL(sock_no_mmap);
2730 
2731 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2732 {
2733 	ssize_t res;
2734 	struct msghdr msg = {.msg_flags = flags};
2735 	struct kvec iov;
2736 	char *kaddr = kmap(page);
2737 	iov.iov_base = kaddr + offset;
2738 	iov.iov_len = size;
2739 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2740 	kunmap(page);
2741 	return res;
2742 }
2743 EXPORT_SYMBOL(sock_no_sendpage);
2744 
2745 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2746 				int offset, size_t size, int flags)
2747 {
2748 	ssize_t res;
2749 	struct msghdr msg = {.msg_flags = flags};
2750 	struct kvec iov;
2751 	char *kaddr = kmap(page);
2752 
2753 	iov.iov_base = kaddr + offset;
2754 	iov.iov_len = size;
2755 	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2756 	kunmap(page);
2757 	return res;
2758 }
2759 EXPORT_SYMBOL(sock_no_sendpage_locked);
2760 
2761 /*
2762  *	Default Socket Callbacks
2763  */
2764 
2765 static void sock_def_wakeup(struct sock *sk)
2766 {
2767 	struct socket_wq *wq;
2768 
2769 	rcu_read_lock();
2770 	wq = rcu_dereference(sk->sk_wq);
2771 	if (skwq_has_sleeper(wq))
2772 		wake_up_interruptible_all(&wq->wait);
2773 	rcu_read_unlock();
2774 }
2775 
2776 static void sock_def_error_report(struct sock *sk)
2777 {
2778 	struct socket_wq *wq;
2779 
2780 	rcu_read_lock();
2781 	wq = rcu_dereference(sk->sk_wq);
2782 	if (skwq_has_sleeper(wq))
2783 		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2784 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2785 	rcu_read_unlock();
2786 }
2787 
2788 static void sock_def_readable(struct sock *sk)
2789 {
2790 	struct socket_wq *wq;
2791 
2792 	rcu_read_lock();
2793 	wq = rcu_dereference(sk->sk_wq);
2794 	if (skwq_has_sleeper(wq))
2795 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2796 						EPOLLRDNORM | EPOLLRDBAND);
2797 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2798 	rcu_read_unlock();
2799 }
2800 
2801 static void sock_def_write_space(struct sock *sk)
2802 {
2803 	struct socket_wq *wq;
2804 
2805 	rcu_read_lock();
2806 
2807 	/* Do not wake up a writer until he can make "significant"
2808 	 * progress.  --DaveM
2809 	 */
2810 	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2811 		wq = rcu_dereference(sk->sk_wq);
2812 		if (skwq_has_sleeper(wq))
2813 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2814 						EPOLLWRNORM | EPOLLWRBAND);
2815 
2816 		/* Should agree with poll, otherwise some programs break */
2817 		if (sock_writeable(sk))
2818 			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2819 	}
2820 
2821 	rcu_read_unlock();
2822 }
2823 
2824 static void sock_def_destruct(struct sock *sk)
2825 {
2826 }
2827 
2828 void sk_send_sigurg(struct sock *sk)
2829 {
2830 	if (sk->sk_socket && sk->sk_socket->file)
2831 		if (send_sigurg(&sk->sk_socket->file->f_owner))
2832 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2833 }
2834 EXPORT_SYMBOL(sk_send_sigurg);
2835 
2836 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2837 		    unsigned long expires)
2838 {
2839 	if (!mod_timer(timer, expires))
2840 		sock_hold(sk);
2841 }
2842 EXPORT_SYMBOL(sk_reset_timer);
2843 
2844 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2845 {
2846 	if (del_timer(timer))
2847 		__sock_put(sk);
2848 }
2849 EXPORT_SYMBOL(sk_stop_timer);
2850 
2851 void sock_init_data(struct socket *sock, struct sock *sk)
2852 {
2853 	sk_init_common(sk);
2854 	sk->sk_send_head	=	NULL;
2855 
2856 	timer_setup(&sk->sk_timer, NULL, 0);
2857 
2858 	sk->sk_allocation	=	GFP_KERNEL;
2859 	sk->sk_rcvbuf		=	sysctl_rmem_default;
2860 	sk->sk_sndbuf		=	sysctl_wmem_default;
2861 	sk->sk_state		=	TCP_CLOSE;
2862 	sk_set_socket(sk, sock);
2863 
2864 	sock_set_flag(sk, SOCK_ZAPPED);
2865 
2866 	if (sock) {
2867 		sk->sk_type	=	sock->type;
2868 		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2869 		sock->sk	=	sk;
2870 		sk->sk_uid	=	SOCK_INODE(sock)->i_uid;
2871 	} else {
2872 		RCU_INIT_POINTER(sk->sk_wq, NULL);
2873 		sk->sk_uid	=	make_kuid(sock_net(sk)->user_ns, 0);
2874 	}
2875 
2876 	rwlock_init(&sk->sk_callback_lock);
2877 	if (sk->sk_kern_sock)
2878 		lockdep_set_class_and_name(
2879 			&sk->sk_callback_lock,
2880 			af_kern_callback_keys + sk->sk_family,
2881 			af_family_kern_clock_key_strings[sk->sk_family]);
2882 	else
2883 		lockdep_set_class_and_name(
2884 			&sk->sk_callback_lock,
2885 			af_callback_keys + sk->sk_family,
2886 			af_family_clock_key_strings[sk->sk_family]);
2887 
2888 	sk->sk_state_change	=	sock_def_wakeup;
2889 	sk->sk_data_ready	=	sock_def_readable;
2890 	sk->sk_write_space	=	sock_def_write_space;
2891 	sk->sk_error_report	=	sock_def_error_report;
2892 	sk->sk_destruct		=	sock_def_destruct;
2893 
2894 	sk->sk_frag.page	=	NULL;
2895 	sk->sk_frag.offset	=	0;
2896 	sk->sk_peek_off		=	-1;
2897 
2898 	sk->sk_peer_pid 	=	NULL;
2899 	sk->sk_peer_cred	=	NULL;
2900 	sk->sk_write_pending	=	0;
2901 	sk->sk_rcvlowat		=	1;
2902 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
2903 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
2904 
2905 	sk->sk_stamp = SK_DEFAULT_STAMP;
2906 #if BITS_PER_LONG==32
2907 	seqlock_init(&sk->sk_stamp_seq);
2908 #endif
2909 	atomic_set(&sk->sk_zckey, 0);
2910 
2911 #ifdef CONFIG_NET_RX_BUSY_POLL
2912 	sk->sk_napi_id		=	0;
2913 	sk->sk_ll_usec		=	sysctl_net_busy_read;
2914 #endif
2915 
2916 	sk->sk_max_pacing_rate = ~0UL;
2917 	sk->sk_pacing_rate = ~0UL;
2918 	sk->sk_pacing_shift = 10;
2919 	sk->sk_incoming_cpu = -1;
2920 
2921 	sk_rx_queue_clear(sk);
2922 	/*
2923 	 * Before updating sk_refcnt, we must commit prior changes to memory
2924 	 * (Documentation/RCU/rculist_nulls.txt for details)
2925 	 */
2926 	smp_wmb();
2927 	refcount_set(&sk->sk_refcnt, 1);
2928 	atomic_set(&sk->sk_drops, 0);
2929 }
2930 EXPORT_SYMBOL(sock_init_data);
2931 
2932 void lock_sock_nested(struct sock *sk, int subclass)
2933 {
2934 	might_sleep();
2935 	spin_lock_bh(&sk->sk_lock.slock);
2936 	if (sk->sk_lock.owned)
2937 		__lock_sock(sk);
2938 	sk->sk_lock.owned = 1;
2939 	spin_unlock(&sk->sk_lock.slock);
2940 	/*
2941 	 * The sk_lock has mutex_lock() semantics here:
2942 	 */
2943 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2944 	local_bh_enable();
2945 }
2946 EXPORT_SYMBOL(lock_sock_nested);
2947 
2948 void release_sock(struct sock *sk)
2949 {
2950 	spin_lock_bh(&sk->sk_lock.slock);
2951 	if (sk->sk_backlog.tail)
2952 		__release_sock(sk);
2953 
2954 	/* Warning : release_cb() might need to release sk ownership,
2955 	 * ie call sock_release_ownership(sk) before us.
2956 	 */
2957 	if (sk->sk_prot->release_cb)
2958 		sk->sk_prot->release_cb(sk);
2959 
2960 	sock_release_ownership(sk);
2961 	if (waitqueue_active(&sk->sk_lock.wq))
2962 		wake_up(&sk->sk_lock.wq);
2963 	spin_unlock_bh(&sk->sk_lock.slock);
2964 }
2965 EXPORT_SYMBOL(release_sock);
2966 
2967 /**
2968  * lock_sock_fast - fast version of lock_sock
2969  * @sk: socket
2970  *
2971  * This version should be used for very small section, where process wont block
2972  * return false if fast path is taken:
2973  *
2974  *   sk_lock.slock locked, owned = 0, BH disabled
2975  *
2976  * return true if slow path is taken:
2977  *
2978  *   sk_lock.slock unlocked, owned = 1, BH enabled
2979  */
2980 bool lock_sock_fast(struct sock *sk)
2981 {
2982 	might_sleep();
2983 	spin_lock_bh(&sk->sk_lock.slock);
2984 
2985 	if (!sk->sk_lock.owned)
2986 		/*
2987 		 * Note : We must disable BH
2988 		 */
2989 		return false;
2990 
2991 	__lock_sock(sk);
2992 	sk->sk_lock.owned = 1;
2993 	spin_unlock(&sk->sk_lock.slock);
2994 	/*
2995 	 * The sk_lock has mutex_lock() semantics here:
2996 	 */
2997 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2998 	local_bh_enable();
2999 	return true;
3000 }
3001 EXPORT_SYMBOL(lock_sock_fast);
3002 
3003 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3004 		   bool timeval, bool time32)
3005 {
3006 	struct sock *sk = sock->sk;
3007 	struct timespec64 ts;
3008 
3009 	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3010 	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3011 	if (ts.tv_sec == -1)
3012 		return -ENOENT;
3013 	if (ts.tv_sec == 0) {
3014 		ktime_t kt = ktime_get_real();
3015 		sock_write_timestamp(sk, kt);;
3016 		ts = ktime_to_timespec64(kt);
3017 	}
3018 
3019 	if (timeval)
3020 		ts.tv_nsec /= 1000;
3021 
3022 #ifdef CONFIG_COMPAT_32BIT_TIME
3023 	if (time32)
3024 		return put_old_timespec32(&ts, userstamp);
3025 #endif
3026 #ifdef CONFIG_SPARC64
3027 	/* beware of padding in sparc64 timeval */
3028 	if (timeval && !in_compat_syscall()) {
3029 		struct __kernel_old_timeval __user tv = {
3030 			.tv_sec = ts.tv_sec,
3031 			.tv_usec = ts.tv_nsec,
3032 		};
3033 		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3034 			return -EFAULT;
3035 		return 0;
3036 	}
3037 #endif
3038 	return put_timespec64(&ts, userstamp);
3039 }
3040 EXPORT_SYMBOL(sock_gettstamp);
3041 
3042 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3043 {
3044 	if (!sock_flag(sk, flag)) {
3045 		unsigned long previous_flags = sk->sk_flags;
3046 
3047 		sock_set_flag(sk, flag);
3048 		/*
3049 		 * we just set one of the two flags which require net
3050 		 * time stamping, but time stamping might have been on
3051 		 * already because of the other one
3052 		 */
3053 		if (sock_needs_netstamp(sk) &&
3054 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3055 			net_enable_timestamp();
3056 	}
3057 }
3058 
3059 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3060 		       int level, int type)
3061 {
3062 	struct sock_exterr_skb *serr;
3063 	struct sk_buff *skb;
3064 	int copied, err;
3065 
3066 	err = -EAGAIN;
3067 	skb = sock_dequeue_err_skb(sk);
3068 	if (skb == NULL)
3069 		goto out;
3070 
3071 	copied = skb->len;
3072 	if (copied > len) {
3073 		msg->msg_flags |= MSG_TRUNC;
3074 		copied = len;
3075 	}
3076 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3077 	if (err)
3078 		goto out_free_skb;
3079 
3080 	sock_recv_timestamp(msg, sk, skb);
3081 
3082 	serr = SKB_EXT_ERR(skb);
3083 	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3084 
3085 	msg->msg_flags |= MSG_ERRQUEUE;
3086 	err = copied;
3087 
3088 out_free_skb:
3089 	kfree_skb(skb);
3090 out:
3091 	return err;
3092 }
3093 EXPORT_SYMBOL(sock_recv_errqueue);
3094 
3095 /*
3096  *	Get a socket option on an socket.
3097  *
3098  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3099  *	asynchronous errors should be reported by getsockopt. We assume
3100  *	this means if you specify SO_ERROR (otherwise whats the point of it).
3101  */
3102 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3103 			   char __user *optval, int __user *optlen)
3104 {
3105 	struct sock *sk = sock->sk;
3106 
3107 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3108 }
3109 EXPORT_SYMBOL(sock_common_getsockopt);
3110 
3111 #ifdef CONFIG_COMPAT
3112 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3113 				  char __user *optval, int __user *optlen)
3114 {
3115 	struct sock *sk = sock->sk;
3116 
3117 	if (sk->sk_prot->compat_getsockopt != NULL)
3118 		return sk->sk_prot->compat_getsockopt(sk, level, optname,
3119 						      optval, optlen);
3120 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3121 }
3122 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3123 #endif
3124 
3125 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3126 			int flags)
3127 {
3128 	struct sock *sk = sock->sk;
3129 	int addr_len = 0;
3130 	int err;
3131 
3132 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3133 				   flags & ~MSG_DONTWAIT, &addr_len);
3134 	if (err >= 0)
3135 		msg->msg_namelen = addr_len;
3136 	return err;
3137 }
3138 EXPORT_SYMBOL(sock_common_recvmsg);
3139 
3140 /*
3141  *	Set socket options on an inet socket.
3142  */
3143 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3144 			   char __user *optval, unsigned int optlen)
3145 {
3146 	struct sock *sk = sock->sk;
3147 
3148 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3149 }
3150 EXPORT_SYMBOL(sock_common_setsockopt);
3151 
3152 #ifdef CONFIG_COMPAT
3153 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3154 				  char __user *optval, unsigned int optlen)
3155 {
3156 	struct sock *sk = sock->sk;
3157 
3158 	if (sk->sk_prot->compat_setsockopt != NULL)
3159 		return sk->sk_prot->compat_setsockopt(sk, level, optname,
3160 						      optval, optlen);
3161 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3162 }
3163 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3164 #endif
3165 
3166 void sk_common_release(struct sock *sk)
3167 {
3168 	if (sk->sk_prot->destroy)
3169 		sk->sk_prot->destroy(sk);
3170 
3171 	/*
3172 	 * Observation: when sock_common_release is called, processes have
3173 	 * no access to socket. But net still has.
3174 	 * Step one, detach it from networking:
3175 	 *
3176 	 * A. Remove from hash tables.
3177 	 */
3178 
3179 	sk->sk_prot->unhash(sk);
3180 
3181 	/*
3182 	 * In this point socket cannot receive new packets, but it is possible
3183 	 * that some packets are in flight because some CPU runs receiver and
3184 	 * did hash table lookup before we unhashed socket. They will achieve
3185 	 * receive queue and will be purged by socket destructor.
3186 	 *
3187 	 * Also we still have packets pending on receive queue and probably,
3188 	 * our own packets waiting in device queues. sock_destroy will drain
3189 	 * receive queue, but transmitted packets will delay socket destruction
3190 	 * until the last reference will be released.
3191 	 */
3192 
3193 	sock_orphan(sk);
3194 
3195 	xfrm_sk_free_policy(sk);
3196 
3197 	sk_refcnt_debug_release(sk);
3198 
3199 	sock_put(sk);
3200 }
3201 EXPORT_SYMBOL(sk_common_release);
3202 
3203 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3204 {
3205 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3206 
3207 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3208 	mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3209 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3210 	mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3211 	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3212 	mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3213 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3214 	mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3215 	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3216 }
3217 
3218 #ifdef CONFIG_PROC_FS
3219 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
3220 struct prot_inuse {
3221 	int val[PROTO_INUSE_NR];
3222 };
3223 
3224 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3225 
3226 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3227 {
3228 	__this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3229 }
3230 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3231 
3232 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3233 {
3234 	int cpu, idx = prot->inuse_idx;
3235 	int res = 0;
3236 
3237 	for_each_possible_cpu(cpu)
3238 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3239 
3240 	return res >= 0 ? res : 0;
3241 }
3242 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3243 
3244 static void sock_inuse_add(struct net *net, int val)
3245 {
3246 	this_cpu_add(*net->core.sock_inuse, val);
3247 }
3248 
3249 int sock_inuse_get(struct net *net)
3250 {
3251 	int cpu, res = 0;
3252 
3253 	for_each_possible_cpu(cpu)
3254 		res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3255 
3256 	return res;
3257 }
3258 
3259 EXPORT_SYMBOL_GPL(sock_inuse_get);
3260 
3261 static int __net_init sock_inuse_init_net(struct net *net)
3262 {
3263 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3264 	if (net->core.prot_inuse == NULL)
3265 		return -ENOMEM;
3266 
3267 	net->core.sock_inuse = alloc_percpu(int);
3268 	if (net->core.sock_inuse == NULL)
3269 		goto out;
3270 
3271 	return 0;
3272 
3273 out:
3274 	free_percpu(net->core.prot_inuse);
3275 	return -ENOMEM;
3276 }
3277 
3278 static void __net_exit sock_inuse_exit_net(struct net *net)
3279 {
3280 	free_percpu(net->core.prot_inuse);
3281 	free_percpu(net->core.sock_inuse);
3282 }
3283 
3284 static struct pernet_operations net_inuse_ops = {
3285 	.init = sock_inuse_init_net,
3286 	.exit = sock_inuse_exit_net,
3287 };
3288 
3289 static __init int net_inuse_init(void)
3290 {
3291 	if (register_pernet_subsys(&net_inuse_ops))
3292 		panic("Cannot initialize net inuse counters");
3293 
3294 	return 0;
3295 }
3296 
3297 core_initcall(net_inuse_init);
3298 
3299 static int assign_proto_idx(struct proto *prot)
3300 {
3301 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3302 
3303 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3304 		pr_err("PROTO_INUSE_NR exhausted\n");
3305 		return -ENOSPC;
3306 	}
3307 
3308 	set_bit(prot->inuse_idx, proto_inuse_idx);
3309 	return 0;
3310 }
3311 
3312 static void release_proto_idx(struct proto *prot)
3313 {
3314 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3315 		clear_bit(prot->inuse_idx, proto_inuse_idx);
3316 }
3317 #else
3318 static inline int assign_proto_idx(struct proto *prot)
3319 {
3320 	return 0;
3321 }
3322 
3323 static inline void release_proto_idx(struct proto *prot)
3324 {
3325 }
3326 
3327 static void sock_inuse_add(struct net *net, int val)
3328 {
3329 }
3330 #endif
3331 
3332 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3333 {
3334 	if (!rsk_prot)
3335 		return;
3336 	kfree(rsk_prot->slab_name);
3337 	rsk_prot->slab_name = NULL;
3338 	kmem_cache_destroy(rsk_prot->slab);
3339 	rsk_prot->slab = NULL;
3340 }
3341 
3342 static int req_prot_init(const struct proto *prot)
3343 {
3344 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3345 
3346 	if (!rsk_prot)
3347 		return 0;
3348 
3349 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3350 					prot->name);
3351 	if (!rsk_prot->slab_name)
3352 		return -ENOMEM;
3353 
3354 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3355 					   rsk_prot->obj_size, 0,
3356 					   SLAB_ACCOUNT | prot->slab_flags,
3357 					   NULL);
3358 
3359 	if (!rsk_prot->slab) {
3360 		pr_crit("%s: Can't create request sock SLAB cache!\n",
3361 			prot->name);
3362 		return -ENOMEM;
3363 	}
3364 	return 0;
3365 }
3366 
3367 int proto_register(struct proto *prot, int alloc_slab)
3368 {
3369 	int ret = -ENOBUFS;
3370 
3371 	if (alloc_slab) {
3372 		prot->slab = kmem_cache_create_usercopy(prot->name,
3373 					prot->obj_size, 0,
3374 					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3375 					prot->slab_flags,
3376 					prot->useroffset, prot->usersize,
3377 					NULL);
3378 
3379 		if (prot->slab == NULL) {
3380 			pr_crit("%s: Can't create sock SLAB cache!\n",
3381 				prot->name);
3382 			goto out;
3383 		}
3384 
3385 		if (req_prot_init(prot))
3386 			goto out_free_request_sock_slab;
3387 
3388 		if (prot->twsk_prot != NULL) {
3389 			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3390 
3391 			if (prot->twsk_prot->twsk_slab_name == NULL)
3392 				goto out_free_request_sock_slab;
3393 
3394 			prot->twsk_prot->twsk_slab =
3395 				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3396 						  prot->twsk_prot->twsk_obj_size,
3397 						  0,
3398 						  SLAB_ACCOUNT |
3399 						  prot->slab_flags,
3400 						  NULL);
3401 			if (prot->twsk_prot->twsk_slab == NULL)
3402 				goto out_free_timewait_sock_slab_name;
3403 		}
3404 	}
3405 
3406 	mutex_lock(&proto_list_mutex);
3407 	ret = assign_proto_idx(prot);
3408 	if (ret) {
3409 		mutex_unlock(&proto_list_mutex);
3410 		goto out_free_timewait_sock_slab_name;
3411 	}
3412 	list_add(&prot->node, &proto_list);
3413 	mutex_unlock(&proto_list_mutex);
3414 	return ret;
3415 
3416 out_free_timewait_sock_slab_name:
3417 	if (alloc_slab && prot->twsk_prot)
3418 		kfree(prot->twsk_prot->twsk_slab_name);
3419 out_free_request_sock_slab:
3420 	if (alloc_slab) {
3421 		req_prot_cleanup(prot->rsk_prot);
3422 
3423 		kmem_cache_destroy(prot->slab);
3424 		prot->slab = NULL;
3425 	}
3426 out:
3427 	return ret;
3428 }
3429 EXPORT_SYMBOL(proto_register);
3430 
3431 void proto_unregister(struct proto *prot)
3432 {
3433 	mutex_lock(&proto_list_mutex);
3434 	release_proto_idx(prot);
3435 	list_del(&prot->node);
3436 	mutex_unlock(&proto_list_mutex);
3437 
3438 	kmem_cache_destroy(prot->slab);
3439 	prot->slab = NULL;
3440 
3441 	req_prot_cleanup(prot->rsk_prot);
3442 
3443 	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3444 		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3445 		kfree(prot->twsk_prot->twsk_slab_name);
3446 		prot->twsk_prot->twsk_slab = NULL;
3447 	}
3448 }
3449 EXPORT_SYMBOL(proto_unregister);
3450 
3451 int sock_load_diag_module(int family, int protocol)
3452 {
3453 	if (!protocol) {
3454 		if (!sock_is_registered(family))
3455 			return -ENOENT;
3456 
3457 		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3458 				      NETLINK_SOCK_DIAG, family);
3459 	}
3460 
3461 #ifdef CONFIG_INET
3462 	if (family == AF_INET &&
3463 	    protocol != IPPROTO_RAW &&
3464 	    !rcu_access_pointer(inet_protos[protocol]))
3465 		return -ENOENT;
3466 #endif
3467 
3468 	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3469 			      NETLINK_SOCK_DIAG, family, protocol);
3470 }
3471 EXPORT_SYMBOL(sock_load_diag_module);
3472 
3473 #ifdef CONFIG_PROC_FS
3474 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3475 	__acquires(proto_list_mutex)
3476 {
3477 	mutex_lock(&proto_list_mutex);
3478 	return seq_list_start_head(&proto_list, *pos);
3479 }
3480 
3481 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3482 {
3483 	return seq_list_next(v, &proto_list, pos);
3484 }
3485 
3486 static void proto_seq_stop(struct seq_file *seq, void *v)
3487 	__releases(proto_list_mutex)
3488 {
3489 	mutex_unlock(&proto_list_mutex);
3490 }
3491 
3492 static char proto_method_implemented(const void *method)
3493 {
3494 	return method == NULL ? 'n' : 'y';
3495 }
3496 static long sock_prot_memory_allocated(struct proto *proto)
3497 {
3498 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3499 }
3500 
3501 static const char *sock_prot_memory_pressure(struct proto *proto)
3502 {
3503 	return proto->memory_pressure != NULL ?
3504 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3505 }
3506 
3507 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3508 {
3509 
3510 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3511 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3512 		   proto->name,
3513 		   proto->obj_size,
3514 		   sock_prot_inuse_get(seq_file_net(seq), proto),
3515 		   sock_prot_memory_allocated(proto),
3516 		   sock_prot_memory_pressure(proto),
3517 		   proto->max_header,
3518 		   proto->slab == NULL ? "no" : "yes",
3519 		   module_name(proto->owner),
3520 		   proto_method_implemented(proto->close),
3521 		   proto_method_implemented(proto->connect),
3522 		   proto_method_implemented(proto->disconnect),
3523 		   proto_method_implemented(proto->accept),
3524 		   proto_method_implemented(proto->ioctl),
3525 		   proto_method_implemented(proto->init),
3526 		   proto_method_implemented(proto->destroy),
3527 		   proto_method_implemented(proto->shutdown),
3528 		   proto_method_implemented(proto->setsockopt),
3529 		   proto_method_implemented(proto->getsockopt),
3530 		   proto_method_implemented(proto->sendmsg),
3531 		   proto_method_implemented(proto->recvmsg),
3532 		   proto_method_implemented(proto->sendpage),
3533 		   proto_method_implemented(proto->bind),
3534 		   proto_method_implemented(proto->backlog_rcv),
3535 		   proto_method_implemented(proto->hash),
3536 		   proto_method_implemented(proto->unhash),
3537 		   proto_method_implemented(proto->get_port),
3538 		   proto_method_implemented(proto->enter_memory_pressure));
3539 }
3540 
3541 static int proto_seq_show(struct seq_file *seq, void *v)
3542 {
3543 	if (v == &proto_list)
3544 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3545 			   "protocol",
3546 			   "size",
3547 			   "sockets",
3548 			   "memory",
3549 			   "press",
3550 			   "maxhdr",
3551 			   "slab",
3552 			   "module",
3553 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3554 	else
3555 		proto_seq_printf(seq, list_entry(v, struct proto, node));
3556 	return 0;
3557 }
3558 
3559 static const struct seq_operations proto_seq_ops = {
3560 	.start  = proto_seq_start,
3561 	.next   = proto_seq_next,
3562 	.stop   = proto_seq_stop,
3563 	.show   = proto_seq_show,
3564 };
3565 
3566 static __net_init int proto_init_net(struct net *net)
3567 {
3568 	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3569 			sizeof(struct seq_net_private)))
3570 		return -ENOMEM;
3571 
3572 	return 0;
3573 }
3574 
3575 static __net_exit void proto_exit_net(struct net *net)
3576 {
3577 	remove_proc_entry("protocols", net->proc_net);
3578 }
3579 
3580 
3581 static __net_initdata struct pernet_operations proto_net_ops = {
3582 	.init = proto_init_net,
3583 	.exit = proto_exit_net,
3584 };
3585 
3586 static int __init proto_init(void)
3587 {
3588 	return register_pernet_subsys(&proto_net_ops);
3589 }
3590 
3591 subsys_initcall(proto_init);
3592 
3593 #endif /* PROC_FS */
3594 
3595 #ifdef CONFIG_NET_RX_BUSY_POLL
3596 bool sk_busy_loop_end(void *p, unsigned long start_time)
3597 {
3598 	struct sock *sk = p;
3599 
3600 	return !skb_queue_empty(&sk->sk_receive_queue) ||
3601 	       sk_busy_loop_timeout(sk, start_time);
3602 }
3603 EXPORT_SYMBOL(sk_busy_loop_end);
3604 #endif /* CONFIG_NET_RX_BUSY_POLL */
3605