xref: /linux/net/core/sock.c (revision 5c458073553f0ef74f5c8db1bd459c87c722a299)
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 <linux/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/udp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
116 #include <linux/prefetch.h>
117 #include <linux/compat.h>
118 #include <linux/mroute.h>
119 #include <linux/mroute6.h>
120 #include <linux/icmpv6.h>
121 
122 #include <linux/uaccess.h>
123 
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <linux/skbuff_ref.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <net/proto_memory.h>
132 #include <linux/net_tstamp.h>
133 #include <net/xfrm.h>
134 #include <linux/ipsec.h>
135 #include <net/cls_cgroup.h>
136 #include <net/netprio_cgroup.h>
137 #include <linux/sock_diag.h>
138 
139 #include <linux/filter.h>
140 #include <net/sock_reuseport.h>
141 #include <net/bpf_sk_storage.h>
142 
143 #include <trace/events/sock.h>
144 
145 #include <net/tcp.h>
146 #include <net/busy_poll.h>
147 #include <net/phonet/phonet.h>
148 
149 #include <linux/ethtool.h>
150 
151 #include <uapi/linux/pidfd.h>
152 
153 #include "dev.h"
154 
155 static DEFINE_MUTEX(proto_list_mutex);
156 static LIST_HEAD(proto_list);
157 
158 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc);
159 static void sock_def_write_space(struct sock *sk);
160 
161 /**
162  * sk_ns_capable - General socket capability test
163  * @sk: Socket to use a capability on or through
164  * @user_ns: The user namespace of the capability to use
165  * @cap: The capability to use
166  *
167  * Test to see if the opener of the socket had when the socket was
168  * created and the current process has the capability @cap in the user
169  * namespace @user_ns.
170  */
171 bool sk_ns_capable(const struct sock *sk,
172 		   struct user_namespace *user_ns, int cap)
173 {
174 	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
175 		ns_capable(user_ns, cap);
176 }
177 EXPORT_SYMBOL(sk_ns_capable);
178 
179 /**
180  * sk_capable - Socket global capability test
181  * @sk: Socket to use a capability on or through
182  * @cap: The global capability to use
183  *
184  * Test to see if the opener of the socket had when the socket was
185  * created and the current process has the capability @cap in all user
186  * namespaces.
187  */
188 bool sk_capable(const struct sock *sk, int cap)
189 {
190 	return sk_ns_capable(sk, &init_user_ns, cap);
191 }
192 EXPORT_SYMBOL(sk_capable);
193 
194 /**
195  * sk_net_capable - Network namespace socket capability test
196  * @sk: Socket to use a capability on or through
197  * @cap: The capability to use
198  *
199  * Test to see if the opener of the socket had when the socket was created
200  * and the current process has the capability @cap over the network namespace
201  * the socket is a member of.
202  */
203 bool sk_net_capable(const struct sock *sk, int cap)
204 {
205 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
206 }
207 EXPORT_SYMBOL(sk_net_capable);
208 
209 /*
210  * Each address family might have different locking rules, so we have
211  * one slock key per address family and separate keys for internal and
212  * userspace sockets.
213  */
214 static struct lock_class_key af_family_keys[AF_MAX];
215 static struct lock_class_key af_family_kern_keys[AF_MAX];
216 static struct lock_class_key af_family_slock_keys[AF_MAX];
217 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
218 
219 /*
220  * Make lock validator output more readable. (we pre-construct these
221  * strings build-time, so that runtime initialization of socket
222  * locks is fast):
223  */
224 
225 #define _sock_locks(x)						  \
226   x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
227   x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
228   x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
229   x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
230   x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
231   x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
232   x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
233   x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
234   x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
235   x "27"       ,	x "28"          ,	x "AF_CAN"      , \
236   x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
237   x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
238   x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
239   x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
240   x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
241   x "AF_MCTP"  , \
242   x "AF_MAX"
243 
244 static const char *const af_family_key_strings[AF_MAX+1] = {
245 	_sock_locks("sk_lock-")
246 };
247 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
248 	_sock_locks("slock-")
249 };
250 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
251 	_sock_locks("clock-")
252 };
253 
254 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
255 	_sock_locks("k-sk_lock-")
256 };
257 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
258 	_sock_locks("k-slock-")
259 };
260 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
261 	_sock_locks("k-clock-")
262 };
263 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
264 	_sock_locks("rlock-")
265 };
266 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
267 	_sock_locks("wlock-")
268 };
269 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
270 	_sock_locks("elock-")
271 };
272 
273 /*
274  * sk_callback_lock and sk queues locking rules are per-address-family,
275  * so split the lock classes by using a per-AF key:
276  */
277 static struct lock_class_key af_callback_keys[AF_MAX];
278 static struct lock_class_key af_rlock_keys[AF_MAX];
279 static struct lock_class_key af_wlock_keys[AF_MAX];
280 static struct lock_class_key af_elock_keys[AF_MAX];
281 static struct lock_class_key af_kern_callback_keys[AF_MAX];
282 
283 /* Run time adjustable parameters. */
284 __u32 sysctl_wmem_max __read_mostly = 4 << 20;
285 EXPORT_SYMBOL(sysctl_wmem_max);
286 __u32 sysctl_rmem_max __read_mostly = 4 << 20;
287 EXPORT_SYMBOL(sysctl_rmem_max);
288 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_DEFAULT;
289 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_DEFAULT;
290 
291 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
292 EXPORT_SYMBOL_GPL(memalloc_socks_key);
293 
294 /**
295  * sk_set_memalloc - sets %SOCK_MEMALLOC
296  * @sk: socket to set it on
297  *
298  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
299  * It's the responsibility of the admin to adjust min_free_kbytes
300  * to meet the requirements
301  */
302 void sk_set_memalloc(struct sock *sk)
303 {
304 	sock_set_flag(sk, SOCK_MEMALLOC);
305 	sk->sk_allocation |= __GFP_MEMALLOC;
306 	static_branch_inc(&memalloc_socks_key);
307 }
308 EXPORT_SYMBOL_GPL(sk_set_memalloc);
309 
310 void sk_clear_memalloc(struct sock *sk)
311 {
312 	sock_reset_flag(sk, SOCK_MEMALLOC);
313 	sk->sk_allocation &= ~__GFP_MEMALLOC;
314 	static_branch_dec(&memalloc_socks_key);
315 
316 	/*
317 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
318 	 * progress of swapping. SOCK_MEMALLOC may be cleared while
319 	 * it has rmem allocations due to the last swapfile being deactivated
320 	 * but there is a risk that the socket is unusable due to exceeding
321 	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
322 	 */
323 	sk_mem_reclaim(sk);
324 }
325 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
326 
327 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
328 {
329 	int ret;
330 	unsigned int noreclaim_flag;
331 
332 	/* these should have been dropped before queueing */
333 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
334 
335 	noreclaim_flag = memalloc_noreclaim_save();
336 	ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
337 				 tcp_v6_do_rcv,
338 				 tcp_v4_do_rcv,
339 				 sk, skb);
340 	memalloc_noreclaim_restore(noreclaim_flag);
341 
342 	return ret;
343 }
344 EXPORT_SYMBOL(__sk_backlog_rcv);
345 
346 void sk_error_report(struct sock *sk)
347 {
348 	sk->sk_error_report(sk);
349 
350 	switch (sk->sk_family) {
351 	case AF_INET:
352 		fallthrough;
353 	case AF_INET6:
354 		trace_inet_sk_error_report(sk);
355 		break;
356 	default:
357 		break;
358 	}
359 }
360 EXPORT_SYMBOL(sk_error_report);
361 
362 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
363 {
364 	struct __kernel_sock_timeval tv;
365 
366 	if (timeo == MAX_SCHEDULE_TIMEOUT) {
367 		tv.tv_sec = 0;
368 		tv.tv_usec = 0;
369 	} else {
370 		tv.tv_sec = timeo / HZ;
371 		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
372 	}
373 
374 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
375 		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
376 		*(struct old_timeval32 *)optval = tv32;
377 		return sizeof(tv32);
378 	}
379 
380 	if (old_timeval) {
381 		struct __kernel_old_timeval old_tv;
382 		old_tv.tv_sec = tv.tv_sec;
383 		old_tv.tv_usec = tv.tv_usec;
384 		*(struct __kernel_old_timeval *)optval = old_tv;
385 		return sizeof(old_tv);
386 	}
387 
388 	*(struct __kernel_sock_timeval *)optval = tv;
389 	return sizeof(tv);
390 }
391 EXPORT_SYMBOL(sock_get_timeout);
392 
393 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
394 			   sockptr_t optval, int optlen, bool old_timeval)
395 {
396 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
397 		struct old_timeval32 tv32;
398 
399 		if (optlen < sizeof(tv32))
400 			return -EINVAL;
401 
402 		if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
403 			return -EFAULT;
404 		tv->tv_sec = tv32.tv_sec;
405 		tv->tv_usec = tv32.tv_usec;
406 	} else if (old_timeval) {
407 		struct __kernel_old_timeval old_tv;
408 
409 		if (optlen < sizeof(old_tv))
410 			return -EINVAL;
411 		if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
412 			return -EFAULT;
413 		tv->tv_sec = old_tv.tv_sec;
414 		tv->tv_usec = old_tv.tv_usec;
415 	} else {
416 		if (optlen < sizeof(*tv))
417 			return -EINVAL;
418 		if (copy_from_sockptr(tv, optval, sizeof(*tv)))
419 			return -EFAULT;
420 	}
421 
422 	return 0;
423 }
424 EXPORT_SYMBOL(sock_copy_user_timeval);
425 
426 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
427 			    bool old_timeval)
428 {
429 	struct __kernel_sock_timeval tv;
430 	int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
431 	long val;
432 
433 	if (err)
434 		return err;
435 
436 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
437 		return -EDOM;
438 
439 	if (tv.tv_sec < 0) {
440 		static int warned __read_mostly;
441 
442 		WRITE_ONCE(*timeo_p, 0);
443 		if (warned < 10 && net_ratelimit()) {
444 			warned++;
445 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
446 				__func__, current->comm, task_pid_nr(current));
447 		}
448 		return 0;
449 	}
450 	val = MAX_SCHEDULE_TIMEOUT;
451 	if ((tv.tv_sec || tv.tv_usec) &&
452 	    (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
453 		val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
454 						    USEC_PER_SEC / HZ);
455 	WRITE_ONCE(*timeo_p, val);
456 	return 0;
457 }
458 
459 static bool sk_set_prio_allowed(const struct sock *sk, int val)
460 {
461 	return ((val >= TC_PRIO_BESTEFFORT && val <= TC_PRIO_INTERACTIVE) ||
462 		sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
463 		sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN));
464 }
465 
466 static bool sock_needs_netstamp(const struct sock *sk)
467 {
468 	switch (sk->sk_family) {
469 	case AF_UNSPEC:
470 	case AF_UNIX:
471 		return false;
472 	default:
473 		return true;
474 	}
475 }
476 
477 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
478 {
479 	if (sk->sk_flags & flags) {
480 		sk->sk_flags &= ~flags;
481 		if (sock_needs_netstamp(sk) &&
482 		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
483 			net_disable_timestamp();
484 	}
485 }
486 
487 
488 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
489 {
490 	unsigned long flags;
491 	struct sk_buff_head *list = &sk->sk_receive_queue;
492 
493 	if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
494 		sk_drops_inc(sk);
495 		trace_sock_rcvqueue_full(sk, skb);
496 		return -ENOMEM;
497 	}
498 
499 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
500 		sk_drops_inc(sk);
501 		return -ENOBUFS;
502 	}
503 
504 	skb->dev = NULL;
505 	skb_set_owner_r(skb, sk);
506 
507 	/* we escape from rcu protected region, make sure we dont leak
508 	 * a norefcounted dst
509 	 */
510 	skb_dst_force(skb);
511 
512 	spin_lock_irqsave(&list->lock, flags);
513 	sock_skb_set_dropcount(sk, skb);
514 	__skb_queue_tail(list, skb);
515 	spin_unlock_irqrestore(&list->lock, flags);
516 
517 	if (!sock_flag(sk, SOCK_DEAD))
518 		sk->sk_data_ready(sk);
519 	return 0;
520 }
521 EXPORT_SYMBOL(__sock_queue_rcv_skb);
522 
523 enum skb_drop_reason
524 sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb)
525 {
526 	enum skb_drop_reason drop_reason;
527 	int err;
528 
529 	drop_reason = sk_filter_reason(sk, skb);
530 	if (drop_reason)
531 		return drop_reason;
532 
533 	err = __sock_queue_rcv_skb(sk, skb);
534 	switch (err) {
535 	case -ENOMEM:
536 		return SKB_DROP_REASON_SOCKET_RCVBUFF;
537 	case -ENOBUFS:
538 		return SKB_DROP_REASON_PROTO_MEM;
539 	}
540 	return SKB_NOT_DROPPED_YET;
541 }
542 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
543 
544 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
545 		     const int nested, unsigned int trim_cap, bool refcounted)
546 {
547 	enum skb_drop_reason reason;
548 	int rc = NET_RX_SUCCESS;
549 	int err;
550 
551 	reason = sk_filter_trim_cap(sk, skb, trim_cap);
552 	if (reason)
553 		goto discard_and_relse;
554 
555 	skb->dev = NULL;
556 
557 	if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
558 		sk_drops_inc(sk);
559 		reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
560 		goto discard_and_relse;
561 	}
562 	if (nested)
563 		bh_lock_sock_nested(sk);
564 	else
565 		bh_lock_sock(sk);
566 	if (!sock_owned_by_user(sk)) {
567 		/*
568 		 * trylock + unlock semantics:
569 		 */
570 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
571 
572 		rc = sk_backlog_rcv(sk, skb);
573 
574 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
575 	} else if ((err = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))) {
576 		bh_unlock_sock(sk);
577 		if (err == -ENOMEM)
578 			reason = SKB_DROP_REASON_PFMEMALLOC;
579 		if (err == -ENOBUFS)
580 			reason = SKB_DROP_REASON_SOCKET_BACKLOG;
581 		sk_drops_inc(sk);
582 		goto discard_and_relse;
583 	}
584 
585 	bh_unlock_sock(sk);
586 out:
587 	if (refcounted)
588 		sock_put(sk);
589 	return rc;
590 discard_and_relse:
591 	sk_skb_reason_drop(sk, skb, reason);
592 	goto out;
593 }
594 EXPORT_SYMBOL(__sk_receive_skb);
595 
596 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
597 							  u32));
598 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
599 							   u32));
600 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
601 {
602 	struct dst_entry *dst = __sk_dst_get(sk);
603 
604 	if (dst && READ_ONCE(dst->obsolete) &&
605 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
606 			       dst, cookie) == NULL) {
607 		sk_tx_queue_clear(sk);
608 		WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
609 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
610 		dst_release(dst);
611 		return NULL;
612 	}
613 
614 	return dst;
615 }
616 EXPORT_SYMBOL(__sk_dst_check);
617 
618 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
619 {
620 	struct dst_entry *dst = sk_dst_get(sk);
621 
622 	if (dst && READ_ONCE(dst->obsolete) &&
623 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
624 			       dst, cookie) == NULL) {
625 		sk_dst_reset(sk);
626 		dst_release(dst);
627 		return NULL;
628 	}
629 
630 	return dst;
631 }
632 EXPORT_SYMBOL(sk_dst_check);
633 
634 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
635 {
636 	int ret = -ENOPROTOOPT;
637 #ifdef CONFIG_NETDEVICES
638 	struct net *net = sock_net(sk);
639 
640 	/* Sorry... */
641 	ret = -EPERM;
642 	if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
643 		goto out;
644 
645 	ret = -EINVAL;
646 	if (ifindex < 0)
647 		goto out;
648 
649 	/* Paired with all READ_ONCE() done locklessly. */
650 	WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
651 
652 	if (sk->sk_prot->rehash)
653 		sk->sk_prot->rehash(sk);
654 	sk_dst_reset(sk);
655 
656 	ret = 0;
657 
658 out:
659 #endif
660 
661 	return ret;
662 }
663 
664 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
665 {
666 	int ret;
667 
668 	if (lock_sk)
669 		lock_sock(sk);
670 	ret = sock_bindtoindex_locked(sk, ifindex);
671 	if (lock_sk)
672 		release_sock(sk);
673 
674 	return ret;
675 }
676 EXPORT_SYMBOL(sock_bindtoindex);
677 
678 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
679 {
680 	int ret = -ENOPROTOOPT;
681 #ifdef CONFIG_NETDEVICES
682 	struct net *net = sock_net(sk);
683 	char devname[IFNAMSIZ];
684 	int index;
685 
686 	ret = -EINVAL;
687 	if (optlen < 0)
688 		goto out;
689 
690 	/* Bind this socket to a particular device like "eth0",
691 	 * as specified in the passed interface name. If the
692 	 * name is "" or the option length is zero the socket
693 	 * is not bound.
694 	 */
695 	if (optlen > IFNAMSIZ - 1)
696 		optlen = IFNAMSIZ - 1;
697 	memset(devname, 0, sizeof(devname));
698 
699 	ret = -EFAULT;
700 	if (copy_from_sockptr(devname, optval, optlen))
701 		goto out;
702 
703 	index = 0;
704 	if (devname[0] != '\0') {
705 		struct net_device *dev;
706 
707 		rcu_read_lock();
708 		dev = dev_get_by_name_rcu(net, devname);
709 		if (dev)
710 			index = dev->ifindex;
711 		rcu_read_unlock();
712 		ret = -ENODEV;
713 		if (!dev)
714 			goto out;
715 	}
716 
717 	sockopt_lock_sock(sk);
718 	ret = sock_bindtoindex_locked(sk, index);
719 	sockopt_release_sock(sk);
720 out:
721 #endif
722 
723 	return ret;
724 }
725 
726 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
727 				sockptr_t optlen, int len)
728 {
729 	int ret = -ENOPROTOOPT;
730 #ifdef CONFIG_NETDEVICES
731 	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
732 	struct net *net = sock_net(sk);
733 	char devname[IFNAMSIZ];
734 
735 	if (bound_dev_if == 0) {
736 		len = 0;
737 		goto zero;
738 	}
739 
740 	ret = -EINVAL;
741 	if (len < IFNAMSIZ)
742 		goto out;
743 
744 	ret = netdev_get_name(net, devname, bound_dev_if);
745 	if (ret)
746 		goto out;
747 
748 	len = strlen(devname) + 1;
749 
750 	ret = -EFAULT;
751 	if (copy_to_sockptr(optval, devname, len))
752 		goto out;
753 
754 zero:
755 	ret = -EFAULT;
756 	if (copy_to_sockptr(optlen, &len, sizeof(int)))
757 		goto out;
758 
759 	ret = 0;
760 
761 out:
762 #endif
763 
764 	return ret;
765 }
766 
767 bool sk_mc_loop(const struct sock *sk)
768 {
769 	if (dev_recursion_level())
770 		return false;
771 	if (!sk)
772 		return true;
773 	/* IPV6_ADDRFORM can change sk->sk_family under us. */
774 	switch (READ_ONCE(sk->sk_family)) {
775 	case AF_INET:
776 		return inet_test_bit(MC_LOOP, sk);
777 #if IS_ENABLED(CONFIG_IPV6)
778 	case AF_INET6:
779 		return inet6_test_bit(MC6_LOOP, sk);
780 #endif
781 	}
782 	WARN_ON_ONCE(1);
783 	return true;
784 }
785 EXPORT_SYMBOL(sk_mc_loop);
786 
787 void sock_set_reuseaddr(struct sock *sk)
788 {
789 	lock_sock(sk);
790 	sk->sk_reuse = SK_CAN_REUSE;
791 	release_sock(sk);
792 }
793 EXPORT_SYMBOL(sock_set_reuseaddr);
794 
795 void sock_set_reuseport(struct sock *sk)
796 {
797 	lock_sock(sk);
798 	sk->sk_reuseport = true;
799 	release_sock(sk);
800 }
801 EXPORT_SYMBOL(sock_set_reuseport);
802 
803 void sock_no_linger(struct sock *sk)
804 {
805 	lock_sock(sk);
806 	WRITE_ONCE(sk->sk_lingertime, 0);
807 	sock_set_flag(sk, SOCK_LINGER);
808 	release_sock(sk);
809 }
810 EXPORT_SYMBOL(sock_no_linger);
811 
812 void sock_set_priority(struct sock *sk, u32 priority)
813 {
814 	WRITE_ONCE(sk->sk_priority, priority);
815 }
816 EXPORT_SYMBOL(sock_set_priority);
817 
818 void sock_set_sndtimeo(struct sock *sk, s64 secs)
819 {
820 	if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
821 		WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
822 	else
823 		WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
824 }
825 EXPORT_SYMBOL(sock_set_sndtimeo);
826 
827 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
828 {
829 	sock_valbool_flag(sk, SOCK_RCVTSTAMP, val);
830 	sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns);
831 	if (val)  {
832 		sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
833 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
834 	}
835 }
836 
837 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
838 {
839 	switch (optname) {
840 	case SO_TIMESTAMP_OLD:
841 		__sock_set_timestamps(sk, valbool, false, false);
842 		break;
843 	case SO_TIMESTAMP_NEW:
844 		__sock_set_timestamps(sk, valbool, true, false);
845 		break;
846 	case SO_TIMESTAMPNS_OLD:
847 		__sock_set_timestamps(sk, valbool, false, true);
848 		break;
849 	case SO_TIMESTAMPNS_NEW:
850 		__sock_set_timestamps(sk, valbool, true, true);
851 		break;
852 	}
853 }
854 
855 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
856 {
857 	struct net *net = sock_net(sk);
858 	struct net_device *dev = NULL;
859 	bool match = false;
860 	int *vclock_index;
861 	int i, num;
862 
863 	if (sk->sk_bound_dev_if)
864 		dev = dev_get_by_index(net, sk->sk_bound_dev_if);
865 
866 	if (!dev) {
867 		pr_err("%s: sock not bind to device\n", __func__);
868 		return -EOPNOTSUPP;
869 	}
870 
871 	num = ethtool_get_phc_vclocks(dev, &vclock_index);
872 	dev_put(dev);
873 
874 	for (i = 0; i < num; i++) {
875 		if (*(vclock_index + i) == phc_index) {
876 			match = true;
877 			break;
878 		}
879 	}
880 
881 	if (num > 0)
882 		kfree(vclock_index);
883 
884 	if (!match)
885 		return -EINVAL;
886 
887 	WRITE_ONCE(sk->sk_bind_phc, phc_index);
888 
889 	return 0;
890 }
891 
892 int sock_set_timestamping(struct sock *sk, int optname,
893 			  struct so_timestamping timestamping)
894 {
895 	int val = timestamping.flags;
896 	int ret;
897 
898 	if (val & ~SOF_TIMESTAMPING_MASK)
899 		return -EINVAL;
900 
901 	if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
902 	    !(val & SOF_TIMESTAMPING_OPT_ID))
903 		return -EINVAL;
904 
905 	if (val & SOF_TIMESTAMPING_OPT_ID &&
906 	    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
907 		if (sk_is_tcp(sk)) {
908 			if ((1 << sk->sk_state) &
909 			    (TCPF_CLOSE | TCPF_LISTEN))
910 				return -EINVAL;
911 			if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
912 				atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
913 			else
914 				atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
915 		} else {
916 			atomic_set(&sk->sk_tskey, 0);
917 		}
918 	}
919 
920 	if (val & SOF_TIMESTAMPING_OPT_STATS &&
921 	    !(val & SOF_TIMESTAMPING_OPT_TSONLY))
922 		return -EINVAL;
923 
924 	if (val & SOF_TIMESTAMPING_BIND_PHC) {
925 		ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
926 		if (ret)
927 			return ret;
928 	}
929 
930 	WRITE_ONCE(sk->sk_tsflags, val);
931 	sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
932 	sock_valbool_flag(sk, SOCK_TIMESTAMPING_ANY, !!(val & TSFLAGS_ANY));
933 
934 	if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
935 		sock_enable_timestamp(sk,
936 				      SOCK_TIMESTAMPING_RX_SOFTWARE);
937 	else
938 		sock_disable_timestamp(sk,
939 				       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
940 	return 0;
941 }
942 
943 #if defined(CONFIG_CGROUP_BPF)
944 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
945 {
946 	struct bpf_sock_ops_kern sock_ops;
947 
948 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
949 	sock_ops.op = op;
950 	sock_ops.is_fullsock = 1;
951 	sock_ops.sk = sk;
952 	bpf_skops_init_skb(&sock_ops, skb, 0);
953 	__cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS);
954 }
955 #endif
956 
957 void sock_set_keepalive(struct sock *sk)
958 {
959 	lock_sock(sk);
960 	if (sk->sk_prot->keepalive)
961 		sk->sk_prot->keepalive(sk, true);
962 	sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
963 	release_sock(sk);
964 }
965 EXPORT_SYMBOL(sock_set_keepalive);
966 
967 static void __sock_set_rcvbuf(struct sock *sk, int val)
968 {
969 	struct socket *sock = sk->sk_socket;
970 
971 	/* Ensure val * 2 fits into an int, to prevent max_t() from treating it
972 	 * as a negative value.
973 	 */
974 	val = min_t(int, val, INT_MAX / 2);
975 	sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
976 
977 	/* We double it on the way in to account for "struct sk_buff" etc.
978 	 * overhead.   Applications assume that the SO_RCVBUF setting they make
979 	 * will allow that much actual data to be received on that socket.
980 	 *
981 	 * Applications are unaware that "struct sk_buff" and other overheads
982 	 * allocate from the receive buffer during socket buffer allocation.
983 	 *
984 	 * And after considering the possible alternatives, returning the value
985 	 * we actually used in getsockopt is the most desirable behavior.
986 	 */
987 	WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
988 
989 	if (sock) {
990 		const struct proto_ops *ops = READ_ONCE(sock->ops);
991 
992 		if (ops->set_rcvbuf)
993 			ops->set_rcvbuf(sk, sk->sk_rcvbuf);
994 	}
995 }
996 
997 void sock_set_rcvbuf(struct sock *sk, int val)
998 {
999 	lock_sock(sk);
1000 	__sock_set_rcvbuf(sk, val);
1001 	release_sock(sk);
1002 }
1003 EXPORT_SYMBOL(sock_set_rcvbuf);
1004 
1005 static void __sock_set_mark(struct sock *sk, u32 val)
1006 {
1007 	if (val != sk->sk_mark) {
1008 		WRITE_ONCE(sk->sk_mark, val);
1009 		sk_dst_reset(sk);
1010 	}
1011 }
1012 
1013 void sock_set_mark(struct sock *sk, u32 val)
1014 {
1015 	lock_sock(sk);
1016 	__sock_set_mark(sk, val);
1017 	release_sock(sk);
1018 }
1019 EXPORT_SYMBOL(sock_set_mark);
1020 
1021 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1022 {
1023 	/* Round down bytes to multiple of pages */
1024 	bytes = round_down(bytes, PAGE_SIZE);
1025 
1026 	WARN_ON(bytes > sk->sk_reserved_mem);
1027 	WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1028 	sk_mem_reclaim(sk);
1029 }
1030 
1031 static int sock_reserve_memory(struct sock *sk, int bytes)
1032 {
1033 	long allocated;
1034 	bool charged;
1035 	int pages;
1036 
1037 	if (!mem_cgroup_sk_enabled(sk) || !sk_has_account(sk))
1038 		return -EOPNOTSUPP;
1039 
1040 	if (!bytes)
1041 		return 0;
1042 
1043 	pages = sk_mem_pages(bytes);
1044 
1045 	/* pre-charge to memcg */
1046 	charged = mem_cgroup_sk_charge(sk, pages,
1047 				       GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1048 	if (!charged)
1049 		return -ENOMEM;
1050 
1051 	if (sk->sk_bypass_prot_mem)
1052 		goto success;
1053 
1054 	/* pre-charge to forward_alloc */
1055 	sk_memory_allocated_add(sk, pages);
1056 	allocated = sk_memory_allocated(sk);
1057 
1058 	/* If the system goes into memory pressure with this
1059 	 * precharge, give up and return error.
1060 	 */
1061 	if (allocated > sk_prot_mem_limits(sk, 1)) {
1062 		sk_memory_allocated_sub(sk, pages);
1063 		mem_cgroup_sk_uncharge(sk, pages);
1064 		return -ENOMEM;
1065 	}
1066 
1067 success:
1068 	sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1069 
1070 	WRITE_ONCE(sk->sk_reserved_mem,
1071 		   sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1072 
1073 	return 0;
1074 }
1075 
1076 #ifdef CONFIG_PAGE_POOL
1077 
1078 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED
1079  * in 1 syscall. The limit exists to limit the amount of memory the kernel
1080  * allocates to copy these tokens, and to prevent looping over the frags for
1081  * too long.
1082  */
1083 #define MAX_DONTNEED_TOKENS 128
1084 #define MAX_DONTNEED_FRAGS 1024
1085 
1086 static noinline_for_stack int
1087 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1088 {
1089 	unsigned int num_tokens, i, j, k, netmem_num = 0;
1090 	struct dmabuf_token *tokens;
1091 	int ret = 0, num_frags = 0;
1092 	netmem_ref netmems[16];
1093 
1094 	if (!sk_is_tcp(sk))
1095 		return -EBADF;
1096 
1097 	if (optlen % sizeof(*tokens) ||
1098 	    optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1099 		return -EINVAL;
1100 
1101 	num_tokens = optlen / sizeof(*tokens);
1102 	tokens = kvmalloc_objs(*tokens, num_tokens);
1103 	if (!tokens)
1104 		return -ENOMEM;
1105 
1106 	if (copy_from_sockptr(tokens, optval, optlen)) {
1107 		kvfree(tokens);
1108 		return -EFAULT;
1109 	}
1110 
1111 	xa_lock_bh(&sk->sk_user_frags);
1112 	for (i = 0; i < num_tokens; i++) {
1113 		for (j = 0; j < tokens[i].token_count; j++) {
1114 			if (++num_frags > MAX_DONTNEED_FRAGS)
1115 				goto frag_limit_reached;
1116 
1117 			netmem_ref netmem = (__force netmem_ref)__xa_erase(
1118 				&sk->sk_user_frags, tokens[i].token_start + j);
1119 
1120 			if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
1121 				continue;
1122 
1123 			netmems[netmem_num++] = netmem;
1124 			if (netmem_num == ARRAY_SIZE(netmems)) {
1125 				xa_unlock_bh(&sk->sk_user_frags);
1126 				for (k = 0; k < netmem_num; k++)
1127 					WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1128 				netmem_num = 0;
1129 				xa_lock_bh(&sk->sk_user_frags);
1130 			}
1131 			ret++;
1132 		}
1133 	}
1134 
1135 frag_limit_reached:
1136 	xa_unlock_bh(&sk->sk_user_frags);
1137 	for (k = 0; k < netmem_num; k++)
1138 		WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1139 
1140 	kvfree(tokens);
1141 	return ret;
1142 }
1143 #endif
1144 
1145 void sockopt_lock_sock(struct sock *sk)
1146 {
1147 	/* When current->bpf_ctx is set, the setsockopt is called from
1148 	 * a bpf prog.  bpf has ensured the sk lock has been
1149 	 * acquired before calling setsockopt().
1150 	 */
1151 	if (has_current_bpf_ctx())
1152 		return;
1153 
1154 	lock_sock(sk);
1155 }
1156 EXPORT_SYMBOL(sockopt_lock_sock);
1157 
1158 void sockopt_release_sock(struct sock *sk)
1159 {
1160 	if (has_current_bpf_ctx())
1161 		return;
1162 
1163 	release_sock(sk);
1164 }
1165 EXPORT_SYMBOL(sockopt_release_sock);
1166 
1167 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1168 {
1169 	return has_current_bpf_ctx() || ns_capable(ns, cap);
1170 }
1171 EXPORT_SYMBOL(sockopt_ns_capable);
1172 
1173 bool sockopt_capable(int cap)
1174 {
1175 	return has_current_bpf_ctx() || capable(cap);
1176 }
1177 EXPORT_SYMBOL(sockopt_capable);
1178 
1179 static int sockopt_validate_clockid(__kernel_clockid_t value)
1180 {
1181 	switch (value) {
1182 	case CLOCK_REALTIME:
1183 	case CLOCK_MONOTONIC:
1184 	case CLOCK_TAI:
1185 		return 0;
1186 	}
1187 	return -EINVAL;
1188 }
1189 
1190 /*
1191  *	This is meant for all protocols to use and covers goings on
1192  *	at the socket level. Everything here is generic.
1193  */
1194 
1195 int sk_setsockopt(struct sock *sk, int level, int optname,
1196 		  sockptr_t optval, unsigned int optlen)
1197 {
1198 	struct so_timestamping timestamping;
1199 	struct socket *sock = sk->sk_socket;
1200 	struct sock_txtime sk_txtime;
1201 	int val;
1202 	int valbool;
1203 	struct linger ling;
1204 	int ret = 0;
1205 
1206 	/*
1207 	 *	Options without arguments
1208 	 */
1209 
1210 	if (optname == SO_BINDTODEVICE)
1211 		return sock_setbindtodevice(sk, optval, optlen);
1212 
1213 	if (optlen < sizeof(int))
1214 		return -EINVAL;
1215 
1216 	if (copy_from_sockptr(&val, optval, sizeof(val)))
1217 		return -EFAULT;
1218 
1219 	valbool = val ? 1 : 0;
1220 
1221 	/* handle options which do not require locking the socket. */
1222 	switch (optname) {
1223 	case SO_PRIORITY:
1224 		if (sk_set_prio_allowed(sk, val)) {
1225 			sock_set_priority(sk, val);
1226 			return 0;
1227 		}
1228 		return -EPERM;
1229 	case SO_TYPE:
1230 	case SO_PROTOCOL:
1231 	case SO_DOMAIN:
1232 	case SO_ERROR:
1233 		return -ENOPROTOOPT;
1234 #ifdef CONFIG_NET_RX_BUSY_POLL
1235 	case SO_BUSY_POLL:
1236 		if (val < 0)
1237 			return -EINVAL;
1238 		WRITE_ONCE(sk->sk_ll_usec, val);
1239 		return 0;
1240 	case SO_PREFER_BUSY_POLL:
1241 		if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1242 			return -EPERM;
1243 		WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1244 		return 0;
1245 	case SO_BUSY_POLL_BUDGET:
1246 		if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1247 		    !sockopt_capable(CAP_NET_ADMIN))
1248 			return -EPERM;
1249 		if (val < 0 || val > U16_MAX)
1250 			return -EINVAL;
1251 		WRITE_ONCE(sk->sk_busy_poll_budget, val);
1252 		return 0;
1253 #endif
1254 	case SO_MAX_PACING_RATE:
1255 		{
1256 		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1257 		unsigned long pacing_rate;
1258 
1259 		if (sizeof(ulval) != sizeof(val) &&
1260 		    optlen >= sizeof(ulval) &&
1261 		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1262 			return -EFAULT;
1263 		}
1264 		if (ulval != ~0UL)
1265 			cmpxchg(&sk->sk_pacing_status,
1266 				SK_PACING_NONE,
1267 				SK_PACING_NEEDED);
1268 		/* Pairs with READ_ONCE() from sk_getsockopt() */
1269 		WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1270 		pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1271 		if (ulval < pacing_rate)
1272 			WRITE_ONCE(sk->sk_pacing_rate, ulval);
1273 		return 0;
1274 		}
1275 	case SO_TXREHASH:
1276 		if (!sk_is_tcp(sk))
1277 			return -EOPNOTSUPP;
1278 		if (val < -1 || val > 1)
1279 			return -EINVAL;
1280 		if ((u8)val == SOCK_TXREHASH_DEFAULT)
1281 			val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1282 		/* Paired with READ_ONCE() in tcp_rtx_synack()
1283 		 * and sk_getsockopt().
1284 		 */
1285 		WRITE_ONCE(sk->sk_txrehash, (u8)val);
1286 		return 0;
1287 	case SO_PEEK_OFF:
1288 		{
1289 		int (*set_peek_off)(struct sock *sk, int val);
1290 
1291 		set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1292 		if (set_peek_off)
1293 			ret = set_peek_off(sk, val);
1294 		else
1295 			ret = -EOPNOTSUPP;
1296 		return ret;
1297 		}
1298 #ifdef CONFIG_PAGE_POOL
1299 	case SO_DEVMEM_DONTNEED:
1300 		return sock_devmem_dontneed(sk, optval, optlen);
1301 #endif
1302 	case SO_SNDTIMEO_OLD:
1303 	case SO_SNDTIMEO_NEW:
1304 		return sock_set_timeout(&sk->sk_sndtimeo, optval,
1305 					optlen, optname == SO_SNDTIMEO_OLD);
1306 	case SO_RCVTIMEO_OLD:
1307 	case SO_RCVTIMEO_NEW:
1308 		return sock_set_timeout(&sk->sk_rcvtimeo, optval,
1309 					optlen, optname == SO_RCVTIMEO_OLD);
1310 	}
1311 
1312 	sockopt_lock_sock(sk);
1313 
1314 	switch (optname) {
1315 	case SO_DEBUG:
1316 		if (val && !sockopt_capable(CAP_NET_ADMIN))
1317 			ret = -EACCES;
1318 		else
1319 			sock_valbool_flag(sk, SOCK_DBG, valbool);
1320 		break;
1321 	case SO_REUSEADDR:
1322 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1323 		break;
1324 	case SO_REUSEPORT:
1325 		if (valbool && !sk_is_inet(sk))
1326 			ret = -EOPNOTSUPP;
1327 		else
1328 			sk->sk_reuseport = valbool;
1329 		break;
1330 	case SO_DONTROUTE:
1331 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1332 		sk_dst_reset(sk);
1333 		break;
1334 	case SO_BROADCAST:
1335 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1336 		break;
1337 	case SO_SNDBUF:
1338 		/* Don't error on this BSD doesn't and if you think
1339 		 * about it this is right. Otherwise apps have to
1340 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1341 		 * are treated in BSD as hints
1342 		 */
1343 		val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1344 set_sndbuf:
1345 		/* Ensure val * 2 fits into an int, to prevent max_t()
1346 		 * from treating it as a negative value.
1347 		 */
1348 		val = min_t(int, val, INT_MAX / 2);
1349 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1350 		WRITE_ONCE(sk->sk_sndbuf,
1351 			   max_t(int, val * 2, SOCK_MIN_SNDBUF));
1352 		/* Wake up sending tasks if we upped the value. */
1353 		sk->sk_write_space(sk);
1354 		break;
1355 
1356 	case SO_SNDBUFFORCE:
1357 		if (!sockopt_capable(CAP_NET_ADMIN)) {
1358 			ret = -EPERM;
1359 			break;
1360 		}
1361 
1362 		/* No negative values (to prevent underflow, as val will be
1363 		 * multiplied by 2).
1364 		 */
1365 		if (val < 0)
1366 			val = 0;
1367 		goto set_sndbuf;
1368 
1369 	case SO_RCVBUF:
1370 		/* Don't error on this BSD doesn't and if you think
1371 		 * about it this is right. Otherwise apps have to
1372 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1373 		 * are treated in BSD as hints
1374 		 */
1375 		__sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1376 		break;
1377 
1378 	case SO_RCVBUFFORCE:
1379 		if (!sockopt_capable(CAP_NET_ADMIN)) {
1380 			ret = -EPERM;
1381 			break;
1382 		}
1383 
1384 		/* No negative values (to prevent underflow, as val will be
1385 		 * multiplied by 2).
1386 		 */
1387 		__sock_set_rcvbuf(sk, max(val, 0));
1388 		break;
1389 
1390 	case SO_KEEPALIVE:
1391 		if (sk->sk_prot->keepalive)
1392 			sk->sk_prot->keepalive(sk, valbool);
1393 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1394 		break;
1395 
1396 	case SO_OOBINLINE:
1397 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1398 		break;
1399 
1400 	case SO_NO_CHECK:
1401 		sk->sk_no_check_tx = valbool;
1402 		break;
1403 
1404 	case SO_LINGER:
1405 		if (optlen < sizeof(ling)) {
1406 			ret = -EINVAL;	/* 1003.1g */
1407 			break;
1408 		}
1409 		if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1410 			ret = -EFAULT;
1411 			break;
1412 		}
1413 		if (!ling.l_onoff) {
1414 			sock_reset_flag(sk, SOCK_LINGER);
1415 		} else {
1416 			unsigned long t_sec = ling.l_linger;
1417 
1418 			if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1419 				WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1420 			else
1421 				WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1422 			sock_set_flag(sk, SOCK_LINGER);
1423 		}
1424 		break;
1425 
1426 	case SO_BSDCOMPAT:
1427 		break;
1428 
1429 	case SO_TIMESTAMP_OLD:
1430 	case SO_TIMESTAMP_NEW:
1431 	case SO_TIMESTAMPNS_OLD:
1432 	case SO_TIMESTAMPNS_NEW:
1433 		sock_set_timestamp(sk, optname, valbool);
1434 		break;
1435 
1436 	case SO_TIMESTAMPING_NEW:
1437 	case SO_TIMESTAMPING_OLD:
1438 		if (optlen == sizeof(timestamping)) {
1439 			if (copy_from_sockptr(&timestamping, optval,
1440 					      sizeof(timestamping))) {
1441 				ret = -EFAULT;
1442 				break;
1443 			}
1444 		} else {
1445 			memset(&timestamping, 0, sizeof(timestamping));
1446 			timestamping.flags = val;
1447 		}
1448 		ret = sock_set_timestamping(sk, optname, timestamping);
1449 		break;
1450 
1451 	case SO_RCVLOWAT:
1452 		{
1453 		int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1454 
1455 		if (val < 0)
1456 			val = INT_MAX;
1457 		if (sock)
1458 			set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1459 		if (set_rcvlowat)
1460 			ret = set_rcvlowat(sk, val);
1461 		else
1462 			WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1463 		break;
1464 		}
1465 	case SO_ATTACH_FILTER: {
1466 		struct sock_fprog fprog;
1467 
1468 		if (sk_is_tcp(sk) &&
1469 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1470 			ret = -EPERM;
1471 			break;
1472 		}
1473 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1474 		if (!ret)
1475 			ret = sk_attach_filter(&fprog, sk);
1476 		break;
1477 	}
1478 	case SO_ATTACH_BPF:
1479 		ret = -EINVAL;
1480 		if (optlen == sizeof(u32)) {
1481 			u32 ufd;
1482 
1483 			ret = -EFAULT;
1484 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1485 				break;
1486 
1487 			ret = sk_attach_bpf(ufd, sk);
1488 		}
1489 		break;
1490 
1491 	case SO_ATTACH_REUSEPORT_CBPF: {
1492 		struct sock_fprog fprog;
1493 
1494 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1495 		if (!ret)
1496 			ret = sk_reuseport_attach_filter(&fprog, sk);
1497 		break;
1498 	}
1499 	case SO_ATTACH_REUSEPORT_EBPF:
1500 		ret = -EINVAL;
1501 		if (optlen == sizeof(u32)) {
1502 			u32 ufd;
1503 
1504 			ret = -EFAULT;
1505 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1506 				break;
1507 
1508 			ret = sk_reuseport_attach_bpf(ufd, sk);
1509 		}
1510 		break;
1511 
1512 	case SO_DETACH_REUSEPORT_BPF:
1513 		ret = reuseport_detach_prog(sk);
1514 		break;
1515 
1516 	case SO_DETACH_FILTER:
1517 		ret = sk_detach_filter(sk);
1518 		break;
1519 
1520 	case SO_LOCK_FILTER:
1521 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1522 			ret = -EPERM;
1523 		else
1524 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1525 		break;
1526 
1527 	case SO_MARK:
1528 		if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1529 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1530 			ret = -EPERM;
1531 			break;
1532 		}
1533 
1534 		__sock_set_mark(sk, val);
1535 		break;
1536 	case SO_RCVMARK:
1537 		sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1538 		break;
1539 
1540 	case SO_RCVPRIORITY:
1541 		sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool);
1542 		break;
1543 
1544 	case SO_RXQ_OVFL:
1545 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1546 		break;
1547 
1548 	case SO_WIFI_STATUS:
1549 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1550 		break;
1551 
1552 	case SO_NOFCS:
1553 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1554 		break;
1555 
1556 	case SO_SELECT_ERR_QUEUE:
1557 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1558 		break;
1559 
1560 	case SO_PASSCRED:
1561 		if (sk_may_scm_recv(sk))
1562 			sk->sk_scm_credentials = valbool;
1563 		else
1564 			ret = -EOPNOTSUPP;
1565 		break;
1566 
1567 	case SO_PASSSEC:
1568 		if (IS_ENABLED(CONFIG_SECURITY_NETWORK) && sk_may_scm_recv(sk))
1569 			sk->sk_scm_security = valbool;
1570 		else
1571 			ret = -EOPNOTSUPP;
1572 		break;
1573 
1574 	case SO_PASSPIDFD:
1575 		if (sk_is_unix(sk))
1576 			sk->sk_scm_pidfd = valbool;
1577 		else
1578 			ret = -EOPNOTSUPP;
1579 		break;
1580 
1581 	case SO_PASSRIGHTS:
1582 		if (sk_is_unix(sk))
1583 			sk->sk_scm_rights = valbool;
1584 		else
1585 			ret = -EOPNOTSUPP;
1586 		break;
1587 
1588 	case SO_INCOMING_CPU:
1589 		reuseport_update_incoming_cpu(sk, val);
1590 		break;
1591 
1592 	case SO_CNX_ADVICE:
1593 		if (val == 1)
1594 			dst_negative_advice(sk);
1595 		break;
1596 
1597 	case SO_ZEROCOPY:
1598 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1599 			if (!(sk_is_tcp(sk) ||
1600 			      (sk->sk_type == SOCK_DGRAM &&
1601 			       sk->sk_protocol == IPPROTO_UDP)))
1602 				ret = -EOPNOTSUPP;
1603 		} else if (sk->sk_family != PF_RDS) {
1604 			ret = -EOPNOTSUPP;
1605 		}
1606 		if (!ret) {
1607 			if (val < 0 || val > 1)
1608 				ret = -EINVAL;
1609 			else
1610 				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1611 		}
1612 		break;
1613 
1614 	case SO_TXTIME:
1615 		if (optlen != sizeof(struct sock_txtime)) {
1616 			ret = -EINVAL;
1617 			break;
1618 		} else if (copy_from_sockptr(&sk_txtime, optval,
1619 			   sizeof(struct sock_txtime))) {
1620 			ret = -EFAULT;
1621 			break;
1622 		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1623 			ret = -EINVAL;
1624 			break;
1625 		}
1626 		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1627 		 * scheduler has enough safe guards.
1628 		 */
1629 		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1630 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1631 			ret = -EPERM;
1632 			break;
1633 		}
1634 
1635 		ret = sockopt_validate_clockid(sk_txtime.clockid);
1636 		if (ret)
1637 			break;
1638 
1639 		sock_valbool_flag(sk, SOCK_TXTIME, true);
1640 		sk->sk_clockid = sk_txtime.clockid;
1641 		sk->sk_txtime_deadline_mode =
1642 			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1643 		sk->sk_txtime_report_errors =
1644 			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1645 		break;
1646 
1647 	case SO_BINDTOIFINDEX:
1648 		ret = sock_bindtoindex_locked(sk, val);
1649 		break;
1650 
1651 	case SO_BUF_LOCK:
1652 		if (val & ~SOCK_BUF_LOCK_MASK) {
1653 			ret = -EINVAL;
1654 			break;
1655 		}
1656 		sk->sk_userlocks = val | (sk->sk_userlocks &
1657 					  ~SOCK_BUF_LOCK_MASK);
1658 		break;
1659 
1660 	case SO_RESERVE_MEM:
1661 	{
1662 		int delta;
1663 
1664 		if (val < 0) {
1665 			ret = -EINVAL;
1666 			break;
1667 		}
1668 
1669 		delta = val - sk->sk_reserved_mem;
1670 		if (delta < 0)
1671 			sock_release_reserved_memory(sk, -delta);
1672 		else
1673 			ret = sock_reserve_memory(sk, delta);
1674 		break;
1675 	}
1676 
1677 	default:
1678 		ret = -ENOPROTOOPT;
1679 		break;
1680 	}
1681 	sockopt_release_sock(sk);
1682 	return ret;
1683 }
1684 
1685 int sock_setsockopt(struct socket *sock, int level, int optname,
1686 		    sockptr_t optval, unsigned int optlen)
1687 {
1688 	return sk_setsockopt(sock->sk, level, optname,
1689 			     optval, optlen);
1690 }
1691 EXPORT_SYMBOL(sock_setsockopt);
1692 
1693 static const struct cred *sk_get_peer_cred(struct sock *sk)
1694 {
1695 	const struct cred *cred;
1696 
1697 	spin_lock(&sk->sk_peer_lock);
1698 	cred = get_cred(sk->sk_peer_cred);
1699 	spin_unlock(&sk->sk_peer_lock);
1700 
1701 	return cred;
1702 }
1703 
1704 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1705 			  struct ucred *ucred)
1706 {
1707 	ucred->pid = pid_vnr(pid);
1708 	ucred->uid = ucred->gid = -1;
1709 	if (cred) {
1710 		struct user_namespace *current_ns = current_user_ns();
1711 
1712 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1713 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1714 	}
1715 }
1716 
1717 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1718 {
1719 	struct user_namespace *user_ns = current_user_ns();
1720 	int i;
1721 
1722 	for (i = 0; i < src->ngroups; i++) {
1723 		gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1724 
1725 		if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1726 			return -EFAULT;
1727 	}
1728 
1729 	return 0;
1730 }
1731 
1732 int sk_getsockopt(struct sock *sk, int level, int optname,
1733 		  sockptr_t optval, sockptr_t optlen)
1734 {
1735 	struct socket *sock = sk->sk_socket;
1736 
1737 	union {
1738 		int val;
1739 		u64 val64;
1740 		unsigned long ulval;
1741 		struct linger ling;
1742 		struct old_timeval32 tm32;
1743 		struct __kernel_old_timeval tm;
1744 		struct  __kernel_sock_timeval stm;
1745 		struct sock_txtime txtime;
1746 		struct so_timestamping timestamping;
1747 	} v;
1748 
1749 	int lv = sizeof(int);
1750 	int len;
1751 
1752 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
1753 		return -EFAULT;
1754 	if (len < 0)
1755 		return -EINVAL;
1756 
1757 	memset(&v, 0, sizeof(v));
1758 
1759 	switch (optname) {
1760 	case SO_DEBUG:
1761 		v.val = sock_flag(sk, SOCK_DBG);
1762 		break;
1763 
1764 	case SO_DONTROUTE:
1765 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1766 		break;
1767 
1768 	case SO_BROADCAST:
1769 		v.val = sock_flag(sk, SOCK_BROADCAST);
1770 		break;
1771 
1772 	case SO_SNDBUF:
1773 		v.val = READ_ONCE(sk->sk_sndbuf);
1774 		break;
1775 
1776 	case SO_RCVBUF:
1777 		v.val = READ_ONCE(sk->sk_rcvbuf);
1778 		break;
1779 
1780 	case SO_REUSEADDR:
1781 		v.val = sk->sk_reuse;
1782 		break;
1783 
1784 	case SO_REUSEPORT:
1785 		v.val = sk->sk_reuseport;
1786 		break;
1787 
1788 	case SO_KEEPALIVE:
1789 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1790 		break;
1791 
1792 	case SO_TYPE:
1793 		v.val = sk->sk_type;
1794 		break;
1795 
1796 	case SO_PROTOCOL:
1797 		v.val = sk->sk_protocol;
1798 		break;
1799 
1800 	case SO_DOMAIN:
1801 		v.val = sk->sk_family;
1802 		break;
1803 
1804 	case SO_ERROR:
1805 		v.val = -sock_error(sk);
1806 		if (v.val == 0)
1807 			v.val = xchg(&sk->sk_err_soft, 0);
1808 		break;
1809 
1810 	case SO_OOBINLINE:
1811 		v.val = sock_flag(sk, SOCK_URGINLINE);
1812 		break;
1813 
1814 	case SO_NO_CHECK:
1815 		v.val = sk->sk_no_check_tx;
1816 		break;
1817 
1818 	case SO_PRIORITY:
1819 		v.val = READ_ONCE(sk->sk_priority);
1820 		break;
1821 
1822 	case SO_LINGER:
1823 		lv		= sizeof(v.ling);
1824 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1825 		v.ling.l_linger	= READ_ONCE(sk->sk_lingertime) / HZ;
1826 		break;
1827 
1828 	case SO_BSDCOMPAT:
1829 		break;
1830 
1831 	case SO_TIMESTAMP_OLD:
1832 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1833 				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1834 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1835 		break;
1836 
1837 	case SO_TIMESTAMPNS_OLD:
1838 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1839 		break;
1840 
1841 	case SO_TIMESTAMP_NEW:
1842 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1843 		break;
1844 
1845 	case SO_TIMESTAMPNS_NEW:
1846 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1847 		break;
1848 
1849 	case SO_TIMESTAMPING_OLD:
1850 	case SO_TIMESTAMPING_NEW:
1851 		lv = sizeof(v.timestamping);
1852 		/* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1853 		 * returning the flags when they were set through the same option.
1854 		 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1855 		 */
1856 		if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1857 			v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1858 			v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1859 		}
1860 		break;
1861 
1862 	case SO_RCVTIMEO_OLD:
1863 	case SO_RCVTIMEO_NEW:
1864 		lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1865 				      SO_RCVTIMEO_OLD == optname);
1866 		break;
1867 
1868 	case SO_SNDTIMEO_OLD:
1869 	case SO_SNDTIMEO_NEW:
1870 		lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1871 				      SO_SNDTIMEO_OLD == optname);
1872 		break;
1873 
1874 	case SO_RCVLOWAT:
1875 		v.val = READ_ONCE(sk->sk_rcvlowat);
1876 		break;
1877 
1878 	case SO_SNDLOWAT:
1879 		v.val = 1;
1880 		break;
1881 
1882 	case SO_PASSCRED:
1883 		if (!sk_may_scm_recv(sk))
1884 			return -EOPNOTSUPP;
1885 
1886 		v.val = sk->sk_scm_credentials;
1887 		break;
1888 
1889 	case SO_PASSPIDFD:
1890 		if (!sk_is_unix(sk))
1891 			return -EOPNOTSUPP;
1892 
1893 		v.val = sk->sk_scm_pidfd;
1894 		break;
1895 
1896 	case SO_PASSRIGHTS:
1897 		if (!sk_is_unix(sk))
1898 			return -EOPNOTSUPP;
1899 
1900 		v.val = sk->sk_scm_rights;
1901 		break;
1902 
1903 	case SO_PEERCRED:
1904 	{
1905 		struct ucred peercred;
1906 		if (len > sizeof(peercred))
1907 			len = sizeof(peercred);
1908 
1909 		spin_lock(&sk->sk_peer_lock);
1910 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1911 		spin_unlock(&sk->sk_peer_lock);
1912 
1913 		if (copy_to_sockptr(optval, &peercred, len))
1914 			return -EFAULT;
1915 		goto lenout;
1916 	}
1917 
1918 	case SO_PEERPIDFD:
1919 	{
1920 		struct pid *peer_pid;
1921 		struct file *pidfd_file = NULL;
1922 		unsigned int flags = 0;
1923 		int pidfd;
1924 
1925 		if (len > sizeof(pidfd))
1926 			len = sizeof(pidfd);
1927 
1928 		spin_lock(&sk->sk_peer_lock);
1929 		peer_pid = get_pid(sk->sk_peer_pid);
1930 		spin_unlock(&sk->sk_peer_lock);
1931 
1932 		if (!peer_pid)
1933 			return -ENODATA;
1934 
1935 		/* The use of PIDFD_STALE requires stashing of struct pid
1936 		 * on pidfs with pidfs_register_pid() and only AF_UNIX
1937 		 * were prepared for this.
1938 		 */
1939 		if (sk->sk_family == AF_UNIX)
1940 			flags = PIDFD_STALE;
1941 
1942 		pidfd = pidfd_prepare(peer_pid, flags, &pidfd_file);
1943 		put_pid(peer_pid);
1944 		if (pidfd < 0)
1945 			return pidfd;
1946 
1947 		if (copy_to_sockptr(optval, &pidfd, len) ||
1948 		    copy_to_sockptr(optlen, &len, sizeof(int))) {
1949 			put_unused_fd(pidfd);
1950 			fput(pidfd_file);
1951 
1952 			return -EFAULT;
1953 		}
1954 
1955 		fd_install(pidfd, pidfd_file);
1956 		return 0;
1957 	}
1958 
1959 	case SO_PEERGROUPS:
1960 	{
1961 		const struct cred *cred;
1962 		int ret, n;
1963 
1964 		cred = sk_get_peer_cred(sk);
1965 		if (!cred)
1966 			return -ENODATA;
1967 
1968 		n = cred->group_info->ngroups;
1969 		if (len < n * sizeof(gid_t)) {
1970 			len = n * sizeof(gid_t);
1971 			put_cred(cred);
1972 			return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1973 		}
1974 		len = n * sizeof(gid_t);
1975 
1976 		ret = groups_to_user(optval, cred->group_info);
1977 		put_cred(cred);
1978 		if (ret)
1979 			return ret;
1980 		goto lenout;
1981 	}
1982 
1983 	case SO_PEERNAME:
1984 	{
1985 		struct sockaddr_storage address;
1986 
1987 		lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1988 		if (lv < 0)
1989 			return -ENOTCONN;
1990 		if (lv < len)
1991 			return -EINVAL;
1992 		if (copy_to_sockptr(optval, &address, len))
1993 			return -EFAULT;
1994 		goto lenout;
1995 	}
1996 
1997 	/* Dubious BSD thing... Probably nobody even uses it, but
1998 	 * the UNIX standard wants it for whatever reason... -DaveM
1999 	 */
2000 	case SO_ACCEPTCONN:
2001 		v.val = sk->sk_state == TCP_LISTEN;
2002 		break;
2003 
2004 	case SO_PASSSEC:
2005 		if (!IS_ENABLED(CONFIG_SECURITY_NETWORK) || !sk_may_scm_recv(sk))
2006 			return -EOPNOTSUPP;
2007 
2008 		v.val = sk->sk_scm_security;
2009 		break;
2010 
2011 	case SO_PEERSEC:
2012 		return security_socket_getpeersec_stream(sock,
2013 							 optval, optlen, len);
2014 
2015 	case SO_MARK:
2016 		v.val = READ_ONCE(sk->sk_mark);
2017 		break;
2018 
2019 	case SO_RCVMARK:
2020 		v.val = sock_flag(sk, SOCK_RCVMARK);
2021 		break;
2022 
2023 	case SO_RCVPRIORITY:
2024 		v.val = sock_flag(sk, SOCK_RCVPRIORITY);
2025 		break;
2026 
2027 	case SO_RXQ_OVFL:
2028 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
2029 		break;
2030 
2031 	case SO_WIFI_STATUS:
2032 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
2033 		break;
2034 
2035 	case SO_PEEK_OFF:
2036 		if (!READ_ONCE(sock->ops)->set_peek_off)
2037 			return -EOPNOTSUPP;
2038 
2039 		v.val = READ_ONCE(sk->sk_peek_off);
2040 		break;
2041 	case SO_NOFCS:
2042 		v.val = sock_flag(sk, SOCK_NOFCS);
2043 		break;
2044 
2045 	case SO_BINDTODEVICE:
2046 		return sock_getbindtodevice(sk, optval, optlen, len);
2047 
2048 	case SO_GET_FILTER:
2049 		len = sk_get_filter(sk, optval, len);
2050 		if (len < 0)
2051 			return len;
2052 
2053 		goto lenout;
2054 
2055 	case SO_LOCK_FILTER:
2056 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
2057 		break;
2058 
2059 	case SO_BPF_EXTENSIONS:
2060 		v.val = bpf_tell_extensions();
2061 		break;
2062 
2063 	case SO_SELECT_ERR_QUEUE:
2064 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
2065 		break;
2066 
2067 #ifdef CONFIG_NET_RX_BUSY_POLL
2068 	case SO_BUSY_POLL:
2069 		v.val = READ_ONCE(sk->sk_ll_usec);
2070 		break;
2071 	case SO_PREFER_BUSY_POLL:
2072 		v.val = READ_ONCE(sk->sk_prefer_busy_poll);
2073 		break;
2074 #endif
2075 
2076 	case SO_MAX_PACING_RATE:
2077 		/* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
2078 		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2079 			lv = sizeof(v.ulval);
2080 			v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2081 		} else {
2082 			/* 32bit version */
2083 			v.val = min_t(unsigned long, ~0U,
2084 				      READ_ONCE(sk->sk_max_pacing_rate));
2085 		}
2086 		break;
2087 
2088 	case SO_INCOMING_CPU:
2089 		v.val = READ_ONCE(sk->sk_incoming_cpu);
2090 		break;
2091 
2092 	case SO_MEMINFO:
2093 	{
2094 		u32 meminfo[SK_MEMINFO_VARS];
2095 
2096 		sk_get_meminfo(sk, meminfo);
2097 
2098 		len = min_t(unsigned int, len, sizeof(meminfo));
2099 		if (copy_to_sockptr(optval, &meminfo, len))
2100 			return -EFAULT;
2101 
2102 		goto lenout;
2103 	}
2104 
2105 #ifdef CONFIG_NET_RX_BUSY_POLL
2106 	case SO_INCOMING_NAPI_ID:
2107 		v.val = READ_ONCE(sk->sk_napi_id);
2108 
2109 		/* aggregate non-NAPI IDs down to 0 */
2110 		if (!napi_id_valid(v.val))
2111 			v.val = 0;
2112 
2113 		break;
2114 #endif
2115 
2116 	case SO_COOKIE:
2117 		lv = sizeof(u64);
2118 		if (len < lv)
2119 			return -EINVAL;
2120 		v.val64 = sock_gen_cookie(sk);
2121 		break;
2122 
2123 	case SO_ZEROCOPY:
2124 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
2125 		break;
2126 
2127 	case SO_TXTIME:
2128 		lv = sizeof(v.txtime);
2129 		v.txtime.clockid = sk->sk_clockid;
2130 		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2131 				  SOF_TXTIME_DEADLINE_MODE : 0;
2132 		v.txtime.flags |= sk->sk_txtime_report_errors ?
2133 				  SOF_TXTIME_REPORT_ERRORS : 0;
2134 		break;
2135 
2136 	case SO_BINDTOIFINDEX:
2137 		v.val = READ_ONCE(sk->sk_bound_dev_if);
2138 		break;
2139 
2140 	case SO_NETNS_COOKIE:
2141 		lv = sizeof(u64);
2142 		if (len != lv)
2143 			return -EINVAL;
2144 		v.val64 = sock_net(sk)->net_cookie;
2145 		break;
2146 
2147 	case SO_BUF_LOCK:
2148 		v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2149 		break;
2150 
2151 	case SO_RESERVE_MEM:
2152 		v.val = READ_ONCE(sk->sk_reserved_mem);
2153 		break;
2154 
2155 	case SO_TXREHASH:
2156 		if (!sk_is_tcp(sk))
2157 			return -EOPNOTSUPP;
2158 
2159 		/* Paired with WRITE_ONCE() in sk_setsockopt() */
2160 		v.val = READ_ONCE(sk->sk_txrehash);
2161 		break;
2162 
2163 	default:
2164 		/* We implement the SO_SNDLOWAT etc to not be settable
2165 		 * (1003.1g 7).
2166 		 */
2167 		return -ENOPROTOOPT;
2168 	}
2169 
2170 	if (len > lv)
2171 		len = lv;
2172 	if (copy_to_sockptr(optval, &v, len))
2173 		return -EFAULT;
2174 lenout:
2175 	if (copy_to_sockptr(optlen, &len, sizeof(int)))
2176 		return -EFAULT;
2177 	return 0;
2178 }
2179 
2180 /*
2181  * Initialize an sk_lock.
2182  *
2183  * (We also register the sk_lock with the lock validator.)
2184  */
2185 static inline void sock_lock_init(struct sock *sk)
2186 {
2187 	sk_owner_clear(sk);
2188 
2189 	if (sk->sk_kern_sock)
2190 		sock_lock_init_class_and_name(
2191 			sk,
2192 			af_family_kern_slock_key_strings[sk->sk_family],
2193 			af_family_kern_slock_keys + sk->sk_family,
2194 			af_family_kern_key_strings[sk->sk_family],
2195 			af_family_kern_keys + sk->sk_family);
2196 	else
2197 		sock_lock_init_class_and_name(
2198 			sk,
2199 			af_family_slock_key_strings[sk->sk_family],
2200 			af_family_slock_keys + sk->sk_family,
2201 			af_family_key_strings[sk->sk_family],
2202 			af_family_keys + sk->sk_family);
2203 }
2204 
2205 /*
2206  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2207  * even temporarily, because of RCU lookups. sk_node should also be left as is.
2208  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2209  */
2210 static void sock_copy(struct sock *nsk, const struct sock *osk)
2211 {
2212 	const struct proto *prot = READ_ONCE(osk->sk_prot);
2213 #ifdef CONFIG_SECURITY_NETWORK
2214 	void *sptr = nsk->sk_security;
2215 #endif
2216 
2217 	/* If we move sk_tx_queue_mapping out of the private section,
2218 	 * we must check if sk_tx_queue_clear() is called after
2219 	 * sock_copy() in sk_clone_lock().
2220 	 */
2221 	BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2222 		     offsetof(struct sock, sk_dontcopy_begin) ||
2223 		     offsetof(struct sock, sk_tx_queue_mapping) >=
2224 		     offsetof(struct sock, sk_dontcopy_end));
2225 
2226 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2227 
2228 	unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2229 		      prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2230 		      /* alloc is larger than struct, see sk_prot_alloc() */);
2231 
2232 #ifdef CONFIG_SECURITY_NETWORK
2233 	nsk->sk_security = sptr;
2234 	security_sk_clone(osk, nsk);
2235 #endif
2236 }
2237 
2238 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2239 		int family)
2240 {
2241 	struct sock *sk;
2242 	struct kmem_cache *slab;
2243 
2244 	slab = prot->slab;
2245 	if (slab != NULL) {
2246 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2247 		if (!sk)
2248 			return sk;
2249 		if (want_init_on_alloc(priority))
2250 			sk_prot_clear_nulls(sk, prot->obj_size);
2251 	} else
2252 		sk = kmalloc(prot->obj_size, priority);
2253 
2254 	if (sk != NULL) {
2255 		if (security_sk_alloc(sk, family, priority))
2256 			goto out_free;
2257 
2258 		if (!try_module_get(prot->owner))
2259 			goto out_free_sec;
2260 	}
2261 
2262 	return sk;
2263 
2264 out_free_sec:
2265 	security_sk_free(sk);
2266 out_free:
2267 	if (slab != NULL)
2268 		kmem_cache_free(slab, sk);
2269 	else
2270 		kfree(sk);
2271 	return NULL;
2272 }
2273 
2274 static void sk_prot_free(struct proto *prot, struct sock *sk)
2275 {
2276 	struct kmem_cache *slab;
2277 	struct module *owner;
2278 
2279 	owner = prot->owner;
2280 	slab = prot->slab;
2281 
2282 	cgroup_sk_free(&sk->sk_cgrp_data);
2283 	mem_cgroup_sk_free(sk);
2284 	security_sk_free(sk);
2285 
2286 	sk_owner_put(sk);
2287 
2288 	if (slab != NULL)
2289 		kmem_cache_free(slab, sk);
2290 	else
2291 		kfree(sk);
2292 	module_put(owner);
2293 }
2294 
2295 /**
2296  *	sk_alloc - All socket objects are allocated here
2297  *	@net: the applicable net namespace
2298  *	@family: protocol family
2299  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2300  *	@prot: struct proto associated with this new sock instance
2301  *	@kern: is this to be a kernel socket?
2302  */
2303 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2304 		      struct proto *prot, int kern)
2305 {
2306 	struct sock *sk;
2307 
2308 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2309 	if (sk) {
2310 		sk->sk_family = family;
2311 		/*
2312 		 * See comment in struct sock definition to understand
2313 		 * why we need sk_prot_creator -acme
2314 		 */
2315 		sk->sk_prot = sk->sk_prot_creator = prot;
2316 
2317 		if (READ_ONCE(net->core.sysctl_bypass_prot_mem))
2318 			sk->sk_bypass_prot_mem = 1;
2319 
2320 		sk->sk_kern_sock = kern;
2321 		sock_lock_init(sk);
2322 
2323 		sk->sk_net_refcnt = kern ? 0 : 1;
2324 		if (likely(sk->sk_net_refcnt)) {
2325 			get_net_track(net, &sk->ns_tracker, priority);
2326 			sock_inuse_add(net, 1);
2327 		} else {
2328 			net_passive_inc(net);
2329 			__netns_tracker_alloc(net, &sk->ns_tracker,
2330 					      false, priority);
2331 		}
2332 
2333 		sock_net_set(sk, net);
2334 		refcount_set(&sk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS);
2335 
2336 		mem_cgroup_sk_alloc(sk);
2337 		cgroup_sk_alloc(&sk->sk_cgrp_data);
2338 		sock_update_classid(&sk->sk_cgrp_data);
2339 		sock_update_netprioidx(&sk->sk_cgrp_data);
2340 		sk_tx_queue_clear(sk);
2341 	}
2342 
2343 	return sk;
2344 }
2345 EXPORT_SYMBOL(sk_alloc);
2346 
2347 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2348  * grace period. This is the case for UDP sockets and TCP listeners.
2349  */
2350 static void __sk_destruct(struct rcu_head *head)
2351 {
2352 	struct sock *sk = container_of(head, struct sock, sk_rcu);
2353 	struct net *net = sock_net(sk);
2354 	struct sk_filter *filter;
2355 
2356 	if (sk->sk_destruct)
2357 		sk->sk_destruct(sk);
2358 
2359 	filter = rcu_dereference_check(sk->sk_filter,
2360 				       refcount_read(&sk->sk_wmem_alloc) == 0);
2361 	if (filter) {
2362 		sk_filter_uncharge(sk, filter);
2363 		RCU_INIT_POINTER(sk->sk_filter, NULL);
2364 	}
2365 
2366 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2367 
2368 #ifdef CONFIG_BPF_SYSCALL
2369 	bpf_sk_storage_free(sk);
2370 #endif
2371 
2372 	if (atomic_read(&sk->sk_omem_alloc))
2373 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
2374 			 __func__, atomic_read(&sk->sk_omem_alloc));
2375 
2376 	if (sk->sk_frag.page) {
2377 		put_page(sk->sk_frag.page);
2378 		sk->sk_frag.page = NULL;
2379 	}
2380 
2381 	/* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2382 	put_cred(sk->sk_peer_cred);
2383 	put_pid(sk->sk_peer_pid);
2384 
2385 	if (likely(sk->sk_net_refcnt)) {
2386 		put_net_track(net, &sk->ns_tracker);
2387 	} else {
2388 		__netns_tracker_free(net, &sk->ns_tracker, false);
2389 		net_passive_dec(net);
2390 	}
2391 	sk_prot_free(sk->sk_prot_creator, sk);
2392 }
2393 
2394 void sk_net_refcnt_upgrade(struct sock *sk)
2395 {
2396 	struct net *net = sock_net(sk);
2397 
2398 	WARN_ON_ONCE(sk->sk_net_refcnt);
2399 	__netns_tracker_free(net, &sk->ns_tracker, false);
2400 	net_passive_dec(net);
2401 	sk->sk_net_refcnt = 1;
2402 	get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
2403 	sock_inuse_add(net, 1);
2404 }
2405 EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade);
2406 
2407 void sk_destruct(struct sock *sk)
2408 {
2409 	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2410 
2411 	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2412 		reuseport_detach_sock(sk);
2413 		use_call_rcu = true;
2414 	}
2415 
2416 	if (use_call_rcu)
2417 		call_rcu(&sk->sk_rcu, __sk_destruct);
2418 	else
2419 		__sk_destruct(&sk->sk_rcu);
2420 }
2421 
2422 static void __sk_free(struct sock *sk)
2423 {
2424 	if (likely(sk->sk_net_refcnt))
2425 		sock_inuse_add(sock_net(sk), -1);
2426 
2427 	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2428 		sock_diag_broadcast_destroy(sk);
2429 	else
2430 		sk_destruct(sk);
2431 }
2432 
2433 void sk_free(struct sock *sk)
2434 {
2435 	/*
2436 	 * We subtract one from sk_wmem_alloc and can know if
2437 	 * some packets are still in some tx queue.
2438 	 * If not null, sock_wfree() will call __sk_free(sk) later
2439 	 */
2440 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2441 		__sk_free(sk);
2442 }
2443 EXPORT_SYMBOL(sk_free);
2444 
2445 static void sk_init_common(struct sock *sk)
2446 {
2447 	skb_queue_head_init(&sk->sk_receive_queue);
2448 	skb_queue_head_init(&sk->sk_write_queue);
2449 	skb_queue_head_init(&sk->sk_error_queue);
2450 
2451 	rwlock_init(&sk->sk_callback_lock);
2452 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2453 			af_rlock_keys + sk->sk_family,
2454 			af_family_rlock_key_strings[sk->sk_family]);
2455 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2456 			af_wlock_keys + sk->sk_family,
2457 			af_family_wlock_key_strings[sk->sk_family]);
2458 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2459 			af_elock_keys + sk->sk_family,
2460 			af_family_elock_key_strings[sk->sk_family]);
2461 	if (sk->sk_kern_sock)
2462 		lockdep_set_class_and_name(&sk->sk_callback_lock,
2463 			af_kern_callback_keys + sk->sk_family,
2464 			af_family_kern_clock_key_strings[sk->sk_family]);
2465 	else
2466 		lockdep_set_class_and_name(&sk->sk_callback_lock,
2467 			af_callback_keys + sk->sk_family,
2468 			af_family_clock_key_strings[sk->sk_family]);
2469 }
2470 
2471 /**
2472  * sk_clone - clone a socket
2473  * @sk: the socket to clone
2474  * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2475  * @lock: if true, lock the cloned sk
2476  *
2477  * If @lock is true, the clone is locked by bh_lock_sock(), and
2478  * caller must unlock socket even in error path by bh_unlock_sock().
2479  */
2480 struct sock *sk_clone(const struct sock *sk, const gfp_t priority,
2481 		      bool lock)
2482 {
2483 	struct proto *prot = READ_ONCE(sk->sk_prot);
2484 	struct sk_filter *filter;
2485 	bool is_charged = true;
2486 	struct sock *newsk;
2487 
2488 	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2489 	if (!newsk)
2490 		goto out;
2491 
2492 	sock_copy(newsk, sk);
2493 
2494 	newsk->sk_prot_creator = prot;
2495 #ifdef CONFIG_BPF_SYSCALL
2496 	RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
2497 #endif
2498 
2499 	/* SANITY */
2500 	if (likely(newsk->sk_net_refcnt)) {
2501 		get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2502 		sock_inuse_add(sock_net(newsk), 1);
2503 	} else {
2504 		/* Kernel sockets are not elevating the struct net refcount.
2505 		 * Instead, use a tracker to more easily detect if a layer
2506 		 * is not properly dismantling its kernel sockets at netns
2507 		 * destroy time.
2508 		 */
2509 		net_passive_inc(sock_net(newsk));
2510 		__netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2511 				      false, priority);
2512 	}
2513 
2514 	sk_node_init(&newsk->sk_node);
2515 	sock_lock_init(newsk);
2516 
2517 	if (lock)
2518 		bh_lock_sock(newsk);
2519 
2520 	newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
2521 	newsk->sk_backlog.len = 0;
2522 
2523 	atomic_set(&newsk->sk_rmem_alloc, 0);
2524 
2525 	refcount_set(&newsk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS);
2526 
2527 	atomic_set(&newsk->sk_omem_alloc, 0);
2528 	sk_init_common(newsk);
2529 
2530 	newsk->sk_dst_cache	= NULL;
2531 	newsk->sk_dst_pending_confirm = 0;
2532 	newsk->sk_wmem_queued	= 0;
2533 	newsk->sk_forward_alloc = 0;
2534 	newsk->sk_reserved_mem  = 0;
2535 	DEBUG_NET_WARN_ON_ONCE(newsk->sk_drop_counters);
2536 	sk_drops_reset(newsk);
2537 	newsk->sk_send_head	= NULL;
2538 	newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2539 	atomic_set(&newsk->sk_zckey, 0);
2540 
2541 	sock_reset_flag(newsk, SOCK_DONE);
2542 
2543 #ifdef CONFIG_MEMCG
2544 	/* sk->sk_memcg will be populated at accept() time */
2545 	newsk->sk_memcg = NULL;
2546 #endif
2547 
2548 	cgroup_sk_clone(&newsk->sk_cgrp_data);
2549 
2550 	RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2551 
2552 	if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2553 		net_enable_timestamp();
2554 
2555 	rcu_read_lock();
2556 	filter = rcu_dereference(sk->sk_filter);
2557 	if (filter != NULL)
2558 		/* though it's an empty new sock, the charging may fail
2559 		 * if sysctl_optmem_max was changed between creation of
2560 		 * original socket and cloning
2561 		 */
2562 		is_charged = sk_filter_charge(newsk, filter);
2563 	RCU_INIT_POINTER(newsk->sk_filter, filter);
2564 	rcu_read_unlock();
2565 
2566 	if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2567 		/* We need to make sure that we don't uncharge the new
2568 		 * socket if we couldn't charge it in the first place
2569 		 * as otherwise we uncharge the parent's filter.
2570 		 */
2571 		if (!is_charged)
2572 			RCU_INIT_POINTER(newsk->sk_filter, NULL);
2573 
2574 		goto free;
2575 	}
2576 
2577 	if (bpf_sk_storage_clone(sk, newsk))
2578 		goto free;
2579 
2580 	/* Clear sk_user_data if parent had the pointer tagged
2581 	 * as not suitable for copying when cloning.
2582 	 */
2583 	if (sk_user_data_is_nocopy(newsk))
2584 		newsk->sk_user_data = NULL;
2585 
2586 	newsk->sk_err	   = 0;
2587 	newsk->sk_err_soft = 0;
2588 	newsk->sk_priority = 0;
2589 	newsk->sk_incoming_cpu = raw_smp_processor_id();
2590 
2591 	/* Before updating sk_refcnt, we must commit prior changes to memory
2592 	 * (Documentation/RCU/rculist_nulls.rst for details)
2593 	 */
2594 	smp_wmb();
2595 	refcount_set(&newsk->sk_refcnt, 2);
2596 
2597 	sk_set_socket(newsk, NULL);
2598 	sk_tx_queue_clear(newsk);
2599 	sk_rx_queue_clear(newsk);
2600 	RCU_INIT_POINTER(newsk->sk_wq, NULL);
2601 
2602 	if (newsk->sk_prot->sockets_allocated)
2603 		sk_sockets_allocated_inc(newsk);
2604 out:
2605 	return newsk;
2606 free:
2607 	/* It is still raw copy of parent, so invalidate
2608 	 * destructor and make plain sk_free()
2609 	 */
2610 	newsk->sk_destruct = NULL;
2611 	if (lock)
2612 		bh_unlock_sock(newsk);
2613 	sk_free(newsk);
2614 	newsk = NULL;
2615 	goto out;
2616 }
2617 EXPORT_SYMBOL_GPL(sk_clone);
2618 
2619 static u32 sk_dst_gso_max_size(struct sock *sk, const struct net_device *dev)
2620 {
2621 	bool is_ipv6 = false;
2622 	u32 max_size;
2623 
2624 #if IS_ENABLED(CONFIG_IPV6)
2625 	is_ipv6 = (sk->sk_family == AF_INET6 &&
2626 		   !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2627 #endif
2628 	/* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2629 	max_size = is_ipv6 ? READ_ONCE(dev->gso_max_size) :
2630 			READ_ONCE(dev->gso_ipv4_max_size);
2631 	if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2632 		max_size = GSO_LEGACY_MAX_SIZE;
2633 
2634 	return max_size - (MAX_TCP_HEADER + 1);
2635 }
2636 
2637 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2638 {
2639 	const struct net_device *dev;
2640 	u32 max_segs = 1;
2641 
2642 	rcu_read_lock();
2643 	dev = dst_dev_rcu(dst);
2644 	sk->sk_route_caps = dev->features;
2645 	if (sk_is_tcp(sk)) {
2646 		struct inet_connection_sock *icsk = inet_csk(sk);
2647 
2648 		sk->sk_route_caps |= NETIF_F_GSO;
2649 		icsk->icsk_ack.dst_quick_ack = dst_metric(dst, RTAX_QUICKACK);
2650 	}
2651 	if (sk->sk_route_caps & NETIF_F_GSO)
2652 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2653 	if (unlikely(sk->sk_gso_disabled))
2654 		sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2655 	if (sk_can_gso(sk)) {
2656 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2657 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2658 		} else {
2659 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2660 			sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dev);
2661 			/* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2662 			max_segs = max_t(u32, READ_ONCE(dev->gso_max_segs), 1);
2663 		}
2664 	}
2665 	sk->sk_gso_max_segs = max_segs;
2666 	sk_dst_set(sk, dst);
2667 	rcu_read_unlock();
2668 }
2669 EXPORT_SYMBOL_GPL(sk_setup_caps);
2670 
2671 /*
2672  *	Simple resource managers for sockets.
2673  */
2674 
2675 
2676 /*
2677  * Write buffer destructor automatically called from kfree_skb.
2678  */
2679 void sock_wfree(struct sk_buff *skb)
2680 {
2681 	unsigned int len = skb->truesize;
2682 	struct sock *sk = skb->sk;
2683 	bool free;
2684 	int old;
2685 
2686 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2687 		void (*sk_write_space)(struct sock *sk);
2688 
2689 		sk_write_space = READ_ONCE(sk->sk_write_space);
2690 
2691 		if (sock_flag(sk, SOCK_RCU_FREE) &&
2692 		    sk_write_space == sock_def_write_space) {
2693 			rcu_read_lock();
2694 			free = __refcount_sub_and_test(len, &sk->sk_wmem_alloc,
2695 						       &old);
2696 			sock_def_write_space_wfree(sk, old - len);
2697 			rcu_read_unlock();
2698 			if (unlikely(free))
2699 				__sk_free(sk);
2700 			return;
2701 		}
2702 
2703 		/*
2704 		 * Keep a reference on sk_wmem_alloc, this will be released
2705 		 * after sk_write_space() call
2706 		 */
2707 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2708 		sk_write_space(sk);
2709 		len = 1;
2710 	}
2711 	/*
2712 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2713 	 * could not do because of in-flight packets
2714 	 */
2715 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2716 		__sk_free(sk);
2717 }
2718 EXPORT_SYMBOL(sock_wfree);
2719 
2720 /* This variant of sock_wfree() is used by TCP,
2721  * since it sets SOCK_USE_WRITE_QUEUE.
2722  */
2723 #ifdef CONFIG_INET
2724 void __sock_wfree(struct sk_buff *skb)
2725 {
2726 	struct sock *sk = skb->sk;
2727 
2728 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2729 		__sk_free(sk);
2730 }
2731 EXPORT_SYMBOL_GPL(__sock_wfree);
2732 #endif
2733 
2734 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2735 {
2736 	int old_wmem;
2737 
2738 	skb_orphan(skb);
2739 #ifdef CONFIG_INET
2740 	if (unlikely(!sk_fullsock(sk)))
2741 		return skb_set_owner_edemux(skb, sk);
2742 #endif
2743 	skb->sk = sk;
2744 	skb->destructor = sock_wfree;
2745 	skb_set_hash_from_sk(skb, sk);
2746 	/*
2747 	 * We used to take a refcount on sk, but following operation
2748 	 * is enough to guarantee sk_free() won't free this sock until
2749 	 * all in-flight packets are completed
2750 	 */
2751 	__refcount_add(skb->truesize, &sk->sk_wmem_alloc, &old_wmem);
2752 
2753 	/* (old_wmem == SK_WMEM_ALLOC_BIAS) if no other TX packet for this socket
2754 	 * is in a host queue (qdisc, NIC queue).
2755 	 * Set skb->ooo_okay so that netdev_pick_tx() can choose a TX queue
2756 	 * based on XPS for better performance.
2757 	 * Otherwise clear ooo_okay to not risk Out Of Order delivery.
2758 	 */
2759 	skb->ooo_okay = (old_wmem == SK_WMEM_ALLOC_BIAS);
2760 }
2761 EXPORT_SYMBOL(skb_set_owner_w);
2762 
2763 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2764 {
2765 	/* Drivers depend on in-order delivery for crypto offload,
2766 	 * partial orphan breaks out-of-order-OK logic.
2767 	 */
2768 	if (skb_is_decrypted(skb))
2769 		return false;
2770 
2771 	return (skb->destructor == sock_wfree ||
2772 		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2773 }
2774 
2775 /* This helper is used by netem, as it can hold packets in its
2776  * delay queue. We want to allow the owner socket to send more
2777  * packets, as if they were already TX completed by a typical driver.
2778  * But we also want to keep skb->sk set because some packet schedulers
2779  * rely on it (sch_fq for example).
2780  */
2781 void skb_orphan_partial(struct sk_buff *skb)
2782 {
2783 	if (skb_is_tcp_pure_ack(skb))
2784 		return;
2785 
2786 	if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2787 		return;
2788 
2789 	skb_orphan(skb);
2790 }
2791 EXPORT_SYMBOL(skb_orphan_partial);
2792 
2793 /*
2794  * Read buffer destructor automatically called from kfree_skb.
2795  */
2796 void sock_rfree(struct sk_buff *skb)
2797 {
2798 	struct sock *sk = skb->sk;
2799 	unsigned int len = skb->truesize;
2800 
2801 	atomic_sub(len, &sk->sk_rmem_alloc);
2802 	sk_mem_uncharge(sk, len);
2803 }
2804 EXPORT_SYMBOL(sock_rfree);
2805 
2806 /*
2807  * Buffer destructor for skbs that are not used directly in read or write
2808  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2809  */
2810 void sock_efree(struct sk_buff *skb)
2811 {
2812 	sock_put(skb->sk);
2813 }
2814 EXPORT_SYMBOL(sock_efree);
2815 
2816 /* Buffer destructor for prefetch/receive path where reference count may
2817  * not be held, e.g. for listen sockets.
2818  */
2819 #ifdef CONFIG_INET
2820 void sock_pfree(struct sk_buff *skb)
2821 {
2822 	struct sock *sk = skb->sk;
2823 
2824 	if (!sk_is_refcounted(sk))
2825 		return;
2826 
2827 	if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2828 		inet_reqsk(sk)->rsk_listener = NULL;
2829 		reqsk_free(inet_reqsk(sk));
2830 		return;
2831 	}
2832 
2833 	sock_gen_put(sk);
2834 }
2835 EXPORT_SYMBOL(sock_pfree);
2836 #endif /* CONFIG_INET */
2837 
2838 /*
2839  * Allocate a skb from the socket's send buffer.
2840  */
2841 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2842 			     gfp_t priority)
2843 {
2844 	if (force ||
2845 	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2846 		struct sk_buff *skb = alloc_skb(size, priority);
2847 
2848 		if (skb) {
2849 			skb_set_owner_w(skb, sk);
2850 			return skb;
2851 		}
2852 	}
2853 	return NULL;
2854 }
2855 EXPORT_SYMBOL(sock_wmalloc);
2856 
2857 static void sock_ofree(struct sk_buff *skb)
2858 {
2859 	struct sock *sk = skb->sk;
2860 
2861 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2862 }
2863 
2864 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2865 			     gfp_t priority)
2866 {
2867 	struct sk_buff *skb;
2868 
2869 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2870 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2871 	    READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2872 		return NULL;
2873 
2874 	skb = alloc_skb(size, priority);
2875 	if (!skb)
2876 		return NULL;
2877 
2878 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2879 	skb->sk = sk;
2880 	skb->destructor = sock_ofree;
2881 	return skb;
2882 }
2883 
2884 /*
2885  * Allocate a memory block from the socket's option memory buffer.
2886  */
2887 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2888 {
2889 	int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2890 
2891 	if ((unsigned int)size <= optmem_max &&
2892 	    atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2893 		void *mem;
2894 		/* First do the add, to avoid the race if kmalloc
2895 		 * might sleep.
2896 		 */
2897 		atomic_add(size, &sk->sk_omem_alloc);
2898 		mem = kmalloc(size, priority);
2899 		if (mem)
2900 			return mem;
2901 		atomic_sub(size, &sk->sk_omem_alloc);
2902 	}
2903 	return NULL;
2904 }
2905 EXPORT_SYMBOL(sock_kmalloc);
2906 
2907 /*
2908  * Duplicate the input "src" memory block using the socket's
2909  * option memory buffer.
2910  */
2911 void *sock_kmemdup(struct sock *sk, const void *src,
2912 		   int size, gfp_t priority)
2913 {
2914 	void *mem;
2915 
2916 	mem = sock_kmalloc(sk, size, priority);
2917 	if (mem)
2918 		memcpy(mem, src, size);
2919 	return mem;
2920 }
2921 EXPORT_SYMBOL(sock_kmemdup);
2922 
2923 /* Free an option memory block. Note, we actually want the inline
2924  * here as this allows gcc to detect the nullify and fold away the
2925  * condition entirely.
2926  */
2927 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2928 				  const bool nullify)
2929 {
2930 	if (WARN_ON_ONCE(!mem))
2931 		return;
2932 	if (nullify)
2933 		kfree_sensitive(mem);
2934 	else
2935 		kfree(mem);
2936 	atomic_sub(size, &sk->sk_omem_alloc);
2937 }
2938 
2939 void sock_kfree_s(struct sock *sk, void *mem, int size)
2940 {
2941 	__sock_kfree_s(sk, mem, size, false);
2942 }
2943 EXPORT_SYMBOL(sock_kfree_s);
2944 
2945 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2946 {
2947 	__sock_kfree_s(sk, mem, size, true);
2948 }
2949 EXPORT_SYMBOL(sock_kzfree_s);
2950 
2951 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2952    I think, these locks should be removed for datagram sockets.
2953  */
2954 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2955 {
2956 	DEFINE_WAIT(wait);
2957 
2958 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2959 	for (;;) {
2960 		if (!timeo)
2961 			break;
2962 		if (signal_pending(current))
2963 			break;
2964 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2965 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2966 		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2967 			break;
2968 		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2969 			break;
2970 		if (READ_ONCE(sk->sk_err))
2971 			break;
2972 		timeo = schedule_timeout(timeo);
2973 	}
2974 	finish_wait(sk_sleep(sk), &wait);
2975 	return timeo;
2976 }
2977 
2978 
2979 /*
2980  *	Generic send/receive buffer handlers
2981  */
2982 
2983 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2984 				     unsigned long data_len, int noblock,
2985 				     int *errcode, int max_page_order)
2986 {
2987 	struct sk_buff *skb;
2988 	long timeo;
2989 	int err;
2990 
2991 	timeo = sock_sndtimeo(sk, noblock);
2992 	for (;;) {
2993 		err = sock_error(sk);
2994 		if (err != 0)
2995 			goto failure;
2996 
2997 		err = -EPIPE;
2998 		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2999 			goto failure;
3000 
3001 		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
3002 			break;
3003 
3004 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
3005 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
3006 		err = -EAGAIN;
3007 		if (!timeo)
3008 			goto failure;
3009 		if (signal_pending(current))
3010 			goto interrupted;
3011 		timeo = sock_wait_for_wmem(sk, timeo);
3012 	}
3013 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
3014 				   errcode, sk->sk_allocation);
3015 	if (skb)
3016 		skb_set_owner_w(skb, sk);
3017 	return skb;
3018 
3019 interrupted:
3020 	err = sock_intr_errno(timeo);
3021 failure:
3022 	*errcode = err;
3023 	return NULL;
3024 }
3025 EXPORT_SYMBOL(sock_alloc_send_pskb);
3026 
3027 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
3028 		     struct sockcm_cookie *sockc)
3029 {
3030 	u32 tsflags;
3031 
3032 	BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
3033 
3034 	switch (cmsg->cmsg_type) {
3035 	case SO_MARK:
3036 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
3037 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
3038 			return -EPERM;
3039 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3040 			return -EINVAL;
3041 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
3042 		break;
3043 	case SO_TIMESTAMPING_OLD:
3044 	case SO_TIMESTAMPING_NEW:
3045 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3046 			return -EINVAL;
3047 
3048 		tsflags = *(u32 *)CMSG_DATA(cmsg);
3049 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
3050 			return -EINVAL;
3051 
3052 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
3053 		sockc->tsflags |= tsflags;
3054 		break;
3055 	case SCM_TXTIME:
3056 	{
3057 		ktime_t tmin;
3058 		u64 txtime;
3059 
3060 		if (!sock_flag(sk, SOCK_TXTIME))
3061 			return -EINVAL;
3062 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
3063 			return -EINVAL;
3064 
3065 		txtime = get_unaligned((u64 *)CMSG_DATA(cmsg));
3066 
3067 		/* Allow sending without a delivery time: zero special case */
3068 		if (!txtime) {
3069 			sockc->transmit_time = 0;
3070 			break;
3071 		}
3072 
3073 		switch (sk->sk_clockid) {
3074 		case CLOCK_MONOTONIC:
3075 			tmin = 1;
3076 			break;
3077 		case CLOCK_REALTIME:
3078 			tmin = max(ktime_mono_to_real(0), 1);
3079 			break;
3080 		case CLOCK_TAI:
3081 			tmin = max(ktime_mono_to_any(0, TK_OFFS_TAI), 1);
3082 			break;
3083 		default:
3084 			tmin = 1;
3085 			WARN_ON_ONCE(1);
3086 			break;
3087 		}
3088 
3089 		sockc->transmit_time = max_t(ktime_t, txtime, tmin);
3090 		break;
3091 	}
3092 	case SCM_TS_OPT_ID:
3093 		if (sk_is_tcp(sk))
3094 			return -EINVAL;
3095 		tsflags = READ_ONCE(sk->sk_tsflags);
3096 		if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
3097 			return -EINVAL;
3098 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3099 			return -EINVAL;
3100 		sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
3101 		sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
3102 		break;
3103 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
3104 	case SCM_RIGHTS:
3105 	case SCM_CREDENTIALS:
3106 		break;
3107 	case SO_PRIORITY:
3108 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3109 			return -EINVAL;
3110 		if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg)))
3111 			return -EPERM;
3112 		sockc->priority = *(u32 *)CMSG_DATA(cmsg);
3113 		break;
3114 	case SCM_DEVMEM_DMABUF:
3115 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3116 			return -EINVAL;
3117 		sockc->dmabuf_id = *(u32 *)CMSG_DATA(cmsg);
3118 		break;
3119 	default:
3120 		return -EINVAL;
3121 	}
3122 	return 0;
3123 }
3124 EXPORT_SYMBOL(__sock_cmsg_send);
3125 
3126 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
3127 		   struct sockcm_cookie *sockc)
3128 {
3129 	struct cmsghdr *cmsg;
3130 	int ret;
3131 
3132 	for_each_cmsghdr(cmsg, msg) {
3133 		if (!CMSG_OK(msg, cmsg))
3134 			return -EINVAL;
3135 		if (cmsg->cmsg_level != SOL_SOCKET)
3136 			continue;
3137 		ret = __sock_cmsg_send(sk, cmsg, sockc);
3138 		if (ret)
3139 			return ret;
3140 	}
3141 	return 0;
3142 }
3143 EXPORT_SYMBOL(sock_cmsg_send);
3144 
3145 static void sk_enter_memory_pressure(struct sock *sk)
3146 {
3147 	if (!sk->sk_prot->enter_memory_pressure)
3148 		return;
3149 
3150 	sk->sk_prot->enter_memory_pressure(sk);
3151 }
3152 
3153 static void sk_leave_memory_pressure(struct sock *sk)
3154 {
3155 	if (sk->sk_prot->leave_memory_pressure) {
3156 		INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
3157 				     tcp_leave_memory_pressure, sk);
3158 	} else {
3159 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
3160 
3161 		if (memory_pressure && READ_ONCE(*memory_pressure))
3162 			WRITE_ONCE(*memory_pressure, 0);
3163 	}
3164 }
3165 
3166 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3167 
3168 /**
3169  * skb_page_frag_refill - check that a page_frag contains enough room
3170  * @sz: minimum size of the fragment we want to get
3171  * @pfrag: pointer to page_frag
3172  * @gfp: priority for memory allocation
3173  *
3174  * Note: While this allocator tries to use high order pages, there is
3175  * no guarantee that allocations succeed. Therefore, @sz MUST be
3176  * less or equal than PAGE_SIZE.
3177  */
3178 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3179 {
3180 	if (pfrag->page) {
3181 		if (page_ref_count(pfrag->page) == 1) {
3182 			pfrag->offset = 0;
3183 			return true;
3184 		}
3185 		if (pfrag->offset + sz <= pfrag->size)
3186 			return true;
3187 		put_page(pfrag->page);
3188 	}
3189 
3190 	pfrag->offset = 0;
3191 	if (SKB_FRAG_PAGE_ORDER &&
3192 	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3193 		/* Avoid direct reclaim but allow kswapd to wake */
3194 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3195 					  __GFP_COMP | __GFP_NOWARN |
3196 					  __GFP_NORETRY,
3197 					  SKB_FRAG_PAGE_ORDER);
3198 		if (likely(pfrag->page)) {
3199 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3200 			return true;
3201 		}
3202 	}
3203 	pfrag->page = alloc_page(gfp);
3204 	if (likely(pfrag->page)) {
3205 		pfrag->size = PAGE_SIZE;
3206 		return true;
3207 	}
3208 	return false;
3209 }
3210 EXPORT_SYMBOL(skb_page_frag_refill);
3211 
3212 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3213 {
3214 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3215 		return true;
3216 
3217 	if (!sk->sk_bypass_prot_mem)
3218 		sk_enter_memory_pressure(sk);
3219 
3220 	sk_stream_moderate_sndbuf(sk);
3221 
3222 	return false;
3223 }
3224 EXPORT_SYMBOL(sk_page_frag_refill);
3225 
3226 static void __lock_sock(struct sock *sk)
3227 	__releases(&sk->sk_lock.slock)
3228 	__acquires(&sk->sk_lock.slock)
3229 {
3230 	DEFINE_WAIT(wait);
3231 
3232 	for (;;) {
3233 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3234 					TASK_UNINTERRUPTIBLE);
3235 		spin_unlock_bh(&sk->sk_lock.slock);
3236 		schedule();
3237 		spin_lock_bh(&sk->sk_lock.slock);
3238 		if (!sock_owned_by_user(sk))
3239 			break;
3240 	}
3241 	finish_wait(&sk->sk_lock.wq, &wait);
3242 }
3243 
3244 void __release_sock(struct sock *sk)
3245 	__releases(&sk->sk_lock.slock)
3246 	__acquires(&sk->sk_lock.slock)
3247 {
3248 	struct sk_buff *skb, *next;
3249 	int nb = 0;
3250 
3251 	while ((skb = sk->sk_backlog.head) != NULL) {
3252 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3253 
3254 		spin_unlock_bh(&sk->sk_lock.slock);
3255 
3256 		while (1) {
3257 			next = skb->next;
3258 			prefetch(next);
3259 			DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3260 			skb_mark_not_on_list(skb);
3261 			sk_backlog_rcv(sk, skb);
3262 
3263 			skb = next;
3264 			if (!skb)
3265 				break;
3266 
3267 			if (!(++nb & 15))
3268 				cond_resched();
3269 		}
3270 
3271 		spin_lock_bh(&sk->sk_lock.slock);
3272 	}
3273 
3274 	/*
3275 	 * Doing the zeroing here guarantee we can not loop forever
3276 	 * while a wild producer attempts to flood us.
3277 	 */
3278 	sk->sk_backlog.len = 0;
3279 }
3280 
3281 void __sk_flush_backlog(struct sock *sk)
3282 {
3283 	spin_lock_bh(&sk->sk_lock.slock);
3284 	__release_sock(sk);
3285 
3286 	if (sk->sk_prot->release_cb)
3287 		INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3288 				     tcp_release_cb, sk);
3289 
3290 	spin_unlock_bh(&sk->sk_lock.slock);
3291 }
3292 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3293 
3294 /**
3295  * sk_wait_data - wait for data to arrive at sk_receive_queue
3296  * @sk:    sock to wait on
3297  * @timeo: for how long
3298  * @skb:   last skb seen on sk_receive_queue
3299  *
3300  * Now socket state including sk->sk_err is changed only under lock,
3301  * hence we may omit checks after joining wait queue.
3302  * We check receive queue before schedule() only as optimization;
3303  * it is very likely that release_sock() added new data.
3304  */
3305 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3306 {
3307 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
3308 	int rc;
3309 
3310 	add_wait_queue(sk_sleep(sk), &wait);
3311 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3312 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3313 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3314 	remove_wait_queue(sk_sleep(sk), &wait);
3315 	return rc;
3316 }
3317 EXPORT_SYMBOL(sk_wait_data);
3318 
3319 /**
3320  *	__sk_mem_raise_allocated - increase memory_allocated
3321  *	@sk: socket
3322  *	@size: memory size to allocate
3323  *	@amt: pages to allocate
3324  *	@kind: allocation type
3325  *
3326  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3327  *
3328  *	Unlike the globally shared limits among the sockets under same protocol,
3329  *	consuming the budget of a memcg won't have direct effect on other ones.
3330  *	So be optimistic about memcg's tolerance, and leave the callers to decide
3331  *	whether or not to raise allocated through sk_under_memory_pressure() or
3332  *	its variants.
3333  */
3334 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3335 {
3336 	bool memcg_enabled = false, charged = false;
3337 	struct proto *prot = sk->sk_prot;
3338 	long allocated = 0;
3339 
3340 	if (!sk->sk_bypass_prot_mem) {
3341 		sk_memory_allocated_add(sk, amt);
3342 		allocated = sk_memory_allocated(sk);
3343 	}
3344 
3345 	if (mem_cgroup_sk_enabled(sk)) {
3346 		memcg_enabled = true;
3347 		charged = mem_cgroup_sk_charge(sk, amt, gfp_memcg_charge());
3348 		if (!charged)
3349 			goto suppress_allocation;
3350 	}
3351 
3352 	if (!allocated)
3353 		return 1;
3354 
3355 	/* Under limit. */
3356 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
3357 		sk_leave_memory_pressure(sk);
3358 		return 1;
3359 	}
3360 
3361 	/* Under pressure. */
3362 	if (allocated > sk_prot_mem_limits(sk, 1))
3363 		sk_enter_memory_pressure(sk);
3364 
3365 	/* Over hard limit. */
3366 	if (allocated > sk_prot_mem_limits(sk, 2))
3367 		goto suppress_allocation;
3368 
3369 	/* Guarantee minimum buffer size under pressure (either global
3370 	 * or memcg) to make sure features described in RFC 7323 (TCP
3371 	 * Extensions for High Performance) work properly.
3372 	 *
3373 	 * This rule does NOT stand when exceeds global or memcg's hard
3374 	 * limit, or else a DoS attack can be taken place by spawning
3375 	 * lots of sockets whose usage are under minimum buffer size.
3376 	 */
3377 	if (kind == SK_MEM_RECV) {
3378 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3379 			return 1;
3380 
3381 	} else { /* SK_MEM_SEND */
3382 		int wmem0 = sk_get_wmem0(sk, prot);
3383 
3384 		if (sk->sk_type == SOCK_STREAM) {
3385 			if (sk->sk_wmem_queued < wmem0)
3386 				return 1;
3387 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3388 				return 1;
3389 		}
3390 	}
3391 
3392 	if (sk_has_memory_pressure(sk)) {
3393 		u64 alloc;
3394 
3395 		/* The following 'average' heuristic is within the
3396 		 * scope of global accounting, so it only makes
3397 		 * sense for global memory pressure.
3398 		 */
3399 		if (!sk_under_global_memory_pressure(sk))
3400 			return 1;
3401 
3402 		/* Try to be fair among all the sockets under global
3403 		 * pressure by allowing the ones that below average
3404 		 * usage to raise.
3405 		 */
3406 		alloc = sk_sockets_allocated_read_positive(sk);
3407 		if (sk_prot_mem_limits(sk, 2) > alloc *
3408 		    sk_mem_pages(sk->sk_wmem_queued +
3409 				 atomic_read(&sk->sk_rmem_alloc) +
3410 				 sk->sk_forward_alloc))
3411 			return 1;
3412 	}
3413 
3414 suppress_allocation:
3415 
3416 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3417 		sk_stream_moderate_sndbuf(sk);
3418 
3419 		/* Fail only if socket is _under_ its sndbuf.
3420 		 * In this case we cannot block, so that we have to fail.
3421 		 */
3422 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3423 			/* Force charge with __GFP_NOFAIL */
3424 			if (memcg_enabled && !charged)
3425 				mem_cgroup_sk_charge(sk, amt,
3426 						     gfp_memcg_charge() | __GFP_NOFAIL);
3427 			return 1;
3428 		}
3429 	}
3430 
3431 	trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3432 
3433 	if (allocated)
3434 		sk_memory_allocated_sub(sk, amt);
3435 
3436 	if (charged)
3437 		mem_cgroup_sk_uncharge(sk, amt);
3438 
3439 	return 0;
3440 }
3441 
3442 /**
3443  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3444  *	@sk: socket
3445  *	@size: memory size to allocate
3446  *	@kind: allocation type
3447  *
3448  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3449  *	rmem allocation. This function assumes that protocols which have
3450  *	memory_pressure use sk_wmem_queued as write buffer accounting.
3451  */
3452 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3453 {
3454 	int ret, amt = sk_mem_pages(size);
3455 
3456 	sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3457 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3458 	if (!ret)
3459 		sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3460 	return ret;
3461 }
3462 EXPORT_SYMBOL(__sk_mem_schedule);
3463 
3464 /**
3465  *	__sk_mem_reduce_allocated - reclaim memory_allocated
3466  *	@sk: socket
3467  *	@amount: number of quanta
3468  *
3469  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3470  */
3471 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3472 {
3473 	if (mem_cgroup_sk_enabled(sk))
3474 		mem_cgroup_sk_uncharge(sk, amount);
3475 
3476 	if (sk->sk_bypass_prot_mem)
3477 		return;
3478 
3479 	sk_memory_allocated_sub(sk, amount);
3480 
3481 	if (sk_under_global_memory_pressure(sk) &&
3482 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3483 		sk_leave_memory_pressure(sk);
3484 }
3485 
3486 /**
3487  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3488  *	@sk: socket
3489  *	@amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3490  */
3491 void __sk_mem_reclaim(struct sock *sk, int amount)
3492 {
3493 	amount >>= PAGE_SHIFT;
3494 	sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3495 	__sk_mem_reduce_allocated(sk, amount);
3496 }
3497 EXPORT_SYMBOL(__sk_mem_reclaim);
3498 
3499 void __sk_charge(struct sock *sk, gfp_t gfp)
3500 {
3501 	int amt;
3502 
3503 	gfp |= __GFP_NOFAIL;
3504 	if (mem_cgroup_from_sk(sk)) {
3505 		/* The socket has not been accepted yet, no need
3506 		 * to look at newsk->sk_wmem_queued.
3507 		 */
3508 		amt = sk_mem_pages(sk->sk_forward_alloc +
3509 				   atomic_read(&sk->sk_rmem_alloc));
3510 		if (amt)
3511 			mem_cgroup_sk_charge(sk, amt, gfp);
3512 	}
3513 
3514 	kmem_cache_charge(sk, gfp);
3515 }
3516 
3517 int sk_set_peek_off(struct sock *sk, int val)
3518 {
3519 	WRITE_ONCE(sk->sk_peek_off, val);
3520 	return 0;
3521 }
3522 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3523 
3524 /*
3525  * Set of default routines for initialising struct proto_ops when
3526  * the protocol does not support a particular function. In certain
3527  * cases where it makes no sense for a protocol to have a "do nothing"
3528  * function, some default processing is provided.
3529  */
3530 
3531 int sock_no_bind(struct socket *sock, struct sockaddr_unsized *saddr, int len)
3532 {
3533 	return -EOPNOTSUPP;
3534 }
3535 EXPORT_SYMBOL(sock_no_bind);
3536 
3537 int sock_no_connect(struct socket *sock, struct sockaddr_unsized *saddr,
3538 		    int len, int flags)
3539 {
3540 	return -EOPNOTSUPP;
3541 }
3542 EXPORT_SYMBOL(sock_no_connect);
3543 
3544 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3545 {
3546 	return -EOPNOTSUPP;
3547 }
3548 EXPORT_SYMBOL(sock_no_socketpair);
3549 
3550 int sock_no_accept(struct socket *sock, struct socket *newsock,
3551 		   struct proto_accept_arg *arg)
3552 {
3553 	return -EOPNOTSUPP;
3554 }
3555 EXPORT_SYMBOL(sock_no_accept);
3556 
3557 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3558 		    int peer)
3559 {
3560 	return -EOPNOTSUPP;
3561 }
3562 EXPORT_SYMBOL(sock_no_getname);
3563 
3564 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3565 {
3566 	return -EOPNOTSUPP;
3567 }
3568 EXPORT_SYMBOL(sock_no_ioctl);
3569 
3570 int sock_no_listen(struct socket *sock, int backlog)
3571 {
3572 	return -EOPNOTSUPP;
3573 }
3574 EXPORT_SYMBOL(sock_no_listen);
3575 
3576 int sock_no_shutdown(struct socket *sock, int how)
3577 {
3578 	return -EOPNOTSUPP;
3579 }
3580 EXPORT_SYMBOL(sock_no_shutdown);
3581 
3582 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3583 {
3584 	return -EOPNOTSUPP;
3585 }
3586 EXPORT_SYMBOL(sock_no_sendmsg);
3587 
3588 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3589 {
3590 	return -EOPNOTSUPP;
3591 }
3592 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3593 
3594 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3595 		    int flags)
3596 {
3597 	return -EOPNOTSUPP;
3598 }
3599 EXPORT_SYMBOL(sock_no_recvmsg);
3600 
3601 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3602 {
3603 	/* Mirror missing mmap method error code */
3604 	return -ENODEV;
3605 }
3606 EXPORT_SYMBOL(sock_no_mmap);
3607 
3608 /*
3609  * When a file is received (via SCM_RIGHTS, etc), we must bump the
3610  * various sock-based usage counts.
3611  */
3612 void __receive_sock(struct file *file)
3613 {
3614 	struct socket *sock;
3615 
3616 	sock = sock_from_file(file);
3617 	if (sock) {
3618 		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3619 		sock_update_classid(&sock->sk->sk_cgrp_data);
3620 	}
3621 }
3622 
3623 /*
3624  *	Default Socket Callbacks
3625  */
3626 
3627 static void sock_def_wakeup(struct sock *sk)
3628 {
3629 	struct socket_wq *wq;
3630 
3631 	rcu_read_lock();
3632 	wq = rcu_dereference(sk->sk_wq);
3633 	if (skwq_has_sleeper(wq))
3634 		wake_up_interruptible_all(&wq->wait);
3635 	rcu_read_unlock();
3636 }
3637 
3638 static void sock_def_error_report(struct sock *sk)
3639 {
3640 	struct socket_wq *wq;
3641 
3642 	rcu_read_lock();
3643 	wq = rcu_dereference(sk->sk_wq);
3644 	if (skwq_has_sleeper(wq))
3645 		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3646 	sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3647 	rcu_read_unlock();
3648 }
3649 
3650 void sock_def_readable(struct sock *sk)
3651 {
3652 	struct socket_wq *wq;
3653 
3654 	trace_sk_data_ready(sk);
3655 
3656 	rcu_read_lock();
3657 	wq = rcu_dereference(sk->sk_wq);
3658 	if (skwq_has_sleeper(wq))
3659 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3660 						EPOLLRDNORM | EPOLLRDBAND);
3661 	sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3662 	rcu_read_unlock();
3663 }
3664 
3665 static void sock_def_write_space(struct sock *sk)
3666 {
3667 	struct socket_wq *wq;
3668 
3669 	rcu_read_lock();
3670 
3671 	/* Do not wake up a writer until he can make "significant"
3672 	 * progress.  --DaveM
3673 	 */
3674 	if (sock_writeable(sk)) {
3675 		wq = rcu_dereference(sk->sk_wq);
3676 		if (skwq_has_sleeper(wq))
3677 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3678 						EPOLLWRNORM | EPOLLWRBAND);
3679 
3680 		/* Should agree with poll, otherwise some programs break */
3681 		sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3682 	}
3683 
3684 	rcu_read_unlock();
3685 }
3686 
3687 /* An optimised version of sock_def_write_space(), should only be called
3688  * for SOCK_RCU_FREE sockets under RCU read section and after putting
3689  * ->sk_wmem_alloc.
3690  */
3691 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc)
3692 {
3693 	/* Do not wake up a writer until he can make "significant"
3694 	 * progress.  --DaveM
3695 	 */
3696 	if (__sock_writeable(sk, wmem_alloc)) {
3697 		struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3698 
3699 		/* rely on refcount_sub from sock_wfree() */
3700 		smp_mb__after_atomic();
3701 		if (wq && waitqueue_active(&wq->wait))
3702 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3703 						EPOLLWRNORM | EPOLLWRBAND);
3704 
3705 		/* Should agree with poll, otherwise some programs break */
3706 		sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3707 	}
3708 }
3709 
3710 static void sock_def_destruct(struct sock *sk)
3711 {
3712 }
3713 
3714 void sk_send_sigurg(struct sock *sk)
3715 {
3716 	if (sk->sk_socket && sk->sk_socket->file)
3717 		if (send_sigurg(sk->sk_socket->file))
3718 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3719 }
3720 EXPORT_SYMBOL(sk_send_sigurg);
3721 
3722 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3723 		    unsigned long expires)
3724 {
3725 	if (!mod_timer(timer, expires))
3726 		sock_hold(sk);
3727 }
3728 EXPORT_SYMBOL(sk_reset_timer);
3729 
3730 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3731 {
3732 	if (timer_delete(timer))
3733 		__sock_put(sk);
3734 }
3735 EXPORT_SYMBOL(sk_stop_timer);
3736 
3737 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3738 {
3739 	if (timer_delete_sync(timer))
3740 		__sock_put(sk);
3741 }
3742 EXPORT_SYMBOL(sk_stop_timer_sync);
3743 
3744 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3745 {
3746 	sk_init_common(sk);
3747 	sk->sk_send_head	=	NULL;
3748 
3749 	timer_setup(&sk->sk_timer, NULL, 0);
3750 
3751 	sk->sk_allocation	=	GFP_KERNEL;
3752 	sk->sk_rcvbuf		=	READ_ONCE(sysctl_rmem_default);
3753 	sk->sk_sndbuf		=	READ_ONCE(sysctl_wmem_default);
3754 	sk->sk_state		=	TCP_CLOSE;
3755 	sk->sk_use_task_frag	=	true;
3756 	sk_set_socket(sk, sock);
3757 
3758 	sock_set_flag(sk, SOCK_ZAPPED);
3759 
3760 	if (sock) {
3761 		sk->sk_type	=	sock->type;
3762 		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3763 		sock->sk	=	sk;
3764 	} else {
3765 		RCU_INIT_POINTER(sk->sk_wq, NULL);
3766 	}
3767 	sk->sk_uid	=	uid;
3768 
3769 	sk->sk_state_change	=	sock_def_wakeup;
3770 	sk->sk_data_ready	=	sock_def_readable;
3771 	sk->sk_write_space	=	sock_def_write_space;
3772 	sk->sk_error_report	=	sock_def_error_report;
3773 	sk->sk_destruct		=	sock_def_destruct;
3774 
3775 	sk->sk_frag.page	=	NULL;
3776 	sk->sk_frag.offset	=	0;
3777 	sk->sk_peek_off		=	-1;
3778 
3779 	sk->sk_peer_pid 	=	NULL;
3780 	sk->sk_peer_cred	=	NULL;
3781 	spin_lock_init(&sk->sk_peer_lock);
3782 
3783 	sk->sk_write_pending	=	0;
3784 	sk->sk_rcvlowat		=	1;
3785 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3786 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3787 
3788 	sk->sk_stamp = SK_DEFAULT_STAMP;
3789 #if BITS_PER_LONG==32
3790 	seqlock_init(&sk->sk_stamp_seq);
3791 #endif
3792 	atomic_set(&sk->sk_zckey, 0);
3793 
3794 #ifdef CONFIG_NET_RX_BUSY_POLL
3795 	sk->sk_napi_id		=	0;
3796 	sk->sk_ll_usec		=	READ_ONCE(sysctl_net_busy_read);
3797 #endif
3798 
3799 	sk->sk_max_pacing_rate = ~0UL;
3800 	sk->sk_pacing_rate = ~0UL;
3801 	WRITE_ONCE(sk->sk_pacing_shift, 10);
3802 	sk->sk_incoming_cpu = -1;
3803 
3804 	sk_rx_queue_clear(sk);
3805 	/*
3806 	 * Before updating sk_refcnt, we must commit prior changes to memory
3807 	 * (Documentation/RCU/rculist_nulls.rst for details)
3808 	 */
3809 	smp_wmb();
3810 	refcount_set(&sk->sk_refcnt, 1);
3811 	sk_drops_reset(sk);
3812 }
3813 EXPORT_SYMBOL(sock_init_data_uid);
3814 
3815 void sock_init_data(struct socket *sock, struct sock *sk)
3816 {
3817 	kuid_t uid = sock ?
3818 		SOCK_INODE(sock)->i_uid :
3819 		make_kuid(sock_net(sk)->user_ns, 0);
3820 
3821 	sock_init_data_uid(sock, sk, uid);
3822 }
3823 EXPORT_SYMBOL(sock_init_data);
3824 
3825 void noinline lock_sock_nested(struct sock *sk, int subclass)
3826 {
3827 	/* The sk_lock has mutex_lock() semantics here. */
3828 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3829 
3830 	might_sleep();
3831 #ifdef CONFIG_64BIT
3832 	if (sizeof(struct slock_owned) == sizeof(long)) {
3833 		socket_lock_t tmp = {
3834 			.slock = __SPIN_LOCK_UNLOCKED(tmp.slock),
3835 			.owned = 1,
3836 		};
3837 		socket_lock_t old = {
3838 			.slock = __SPIN_LOCK_UNLOCKED(old.slock),
3839 			.owned = 0,
3840 		};
3841 
3842 		if (likely(try_cmpxchg(&sk->sk_lock.combined,
3843 				       &old.combined, tmp.combined)))
3844 			return;
3845 	}
3846 #endif
3847 	spin_lock_bh(&sk->sk_lock.slock);
3848 	if (unlikely(sock_owned_by_user_nocheck(sk)))
3849 		__lock_sock(sk);
3850 	sk->sk_lock.owned = 1;
3851 	spin_unlock_bh(&sk->sk_lock.slock);
3852 }
3853 EXPORT_SYMBOL(lock_sock_nested);
3854 
3855 void release_sock(struct sock *sk)
3856 {
3857 	spin_lock_bh(&sk->sk_lock.slock);
3858 
3859 	if (unlikely(sk->sk_backlog.tail))
3860 		__release_sock(sk);
3861 
3862 	if (sk->sk_prot->release_cb) {
3863 		if (!tcp_release_cb_cond(sk))
3864 			sk->sk_prot->release_cb(sk);
3865 	}
3866 	sock_release_ownership(sk);
3867 	if (unlikely(waitqueue_active(&sk->sk_lock.wq)))
3868 		wake_up(&sk->sk_lock.wq);
3869 
3870 	spin_unlock_bh(&sk->sk_lock.slock);
3871 }
3872 EXPORT_SYMBOL(release_sock);
3873 
3874 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3875 {
3876 	might_sleep();
3877 	spin_lock_bh(&sk->sk_lock.slock);
3878 
3879 	if (likely(!sock_owned_by_user_nocheck(sk))) {
3880 		/*
3881 		 * Fast path return with bottom halves disabled and
3882 		 * sock::sk_lock.slock held.
3883 		 *
3884 		 * The 'mutex' is not contended and holding
3885 		 * sock::sk_lock.slock prevents all other lockers to
3886 		 * proceed so the corresponding unlock_sock_fast() can
3887 		 * avoid the slow path of release_sock() completely and
3888 		 * just release slock.
3889 		 *
3890 		 * From a semantical POV this is equivalent to 'acquiring'
3891 		 * the 'mutex', hence the corresponding lockdep
3892 		 * mutex_release() has to happen in the fast path of
3893 		 * unlock_sock_fast().
3894 		 */
3895 		return false;
3896 	}
3897 
3898 	__lock_sock(sk);
3899 	sk->sk_lock.owned = 1;
3900 	__acquire(&sk->sk_lock.slock);
3901 	spin_unlock_bh(&sk->sk_lock.slock);
3902 	return true;
3903 }
3904 EXPORT_SYMBOL(__lock_sock_fast);
3905 
3906 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3907 		   bool timeval, bool time32)
3908 {
3909 	struct sock *sk = sock->sk;
3910 	struct timespec64 ts;
3911 
3912 	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3913 	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3914 	if (ts.tv_sec == -1)
3915 		return -ENOENT;
3916 	if (ts.tv_sec == 0) {
3917 		ktime_t kt = ktime_get_real();
3918 		sock_write_timestamp(sk, kt);
3919 		ts = ktime_to_timespec64(kt);
3920 	}
3921 
3922 	if (timeval)
3923 		ts.tv_nsec /= 1000;
3924 
3925 #ifdef CONFIG_COMPAT_32BIT_TIME
3926 	if (time32)
3927 		return put_old_timespec32(&ts, userstamp);
3928 #endif
3929 #ifdef CONFIG_SPARC64
3930 	/* beware of padding in sparc64 timeval */
3931 	if (timeval && !in_compat_syscall()) {
3932 		struct __kernel_old_timeval __user tv = {
3933 			.tv_sec = ts.tv_sec,
3934 			.tv_usec = ts.tv_nsec,
3935 		};
3936 		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3937 			return -EFAULT;
3938 		return 0;
3939 	}
3940 #endif
3941 	return put_timespec64(&ts, userstamp);
3942 }
3943 EXPORT_SYMBOL(sock_gettstamp);
3944 
3945 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3946 {
3947 	if (!sock_flag(sk, flag)) {
3948 		unsigned long previous_flags = sk->sk_flags;
3949 
3950 		sock_set_flag(sk, flag);
3951 		/*
3952 		 * we just set one of the two flags which require net
3953 		 * time stamping, but time stamping might have been on
3954 		 * already because of the other one
3955 		 */
3956 		if (sock_needs_netstamp(sk) &&
3957 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3958 			net_enable_timestamp();
3959 	}
3960 }
3961 
3962 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3963 		       int level, int type)
3964 {
3965 	struct sock_extended_err ee;
3966 	struct sk_buff *skb;
3967 	int copied, err;
3968 
3969 	err = -EAGAIN;
3970 	skb = sock_dequeue_err_skb(sk);
3971 	if (skb == NULL)
3972 		goto out;
3973 
3974 	copied = skb->len;
3975 	if (copied > len) {
3976 		msg->msg_flags |= MSG_TRUNC;
3977 		copied = len;
3978 	}
3979 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3980 	if (err)
3981 		goto out_free_skb;
3982 
3983 	sock_recv_timestamp(msg, sk, skb);
3984 
3985 	/* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
3986 	ee = SKB_EXT_ERR(skb)->ee;
3987 	put_cmsg(msg, level, type, sizeof(ee), &ee);
3988 
3989 	msg->msg_flags |= MSG_ERRQUEUE;
3990 	err = copied;
3991 
3992 out_free_skb:
3993 	kfree_skb(skb);
3994 out:
3995 	return err;
3996 }
3997 EXPORT_SYMBOL(sock_recv_errqueue);
3998 
3999 /*
4000  *	Get a socket option on an socket.
4001  *
4002  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
4003  *	asynchronous errors should be reported by getsockopt. We assume
4004  *	this means if you specify SO_ERROR (otherwise what is the point of it).
4005  */
4006 int sock_common_getsockopt(struct socket *sock, int level, int optname,
4007 			   char __user *optval, int __user *optlen)
4008 {
4009 	struct sock *sk = sock->sk;
4010 
4011 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
4012 	return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
4013 }
4014 EXPORT_SYMBOL(sock_common_getsockopt);
4015 
4016 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
4017 			int flags)
4018 {
4019 	struct sock *sk = sock->sk;
4020 
4021 	return sk->sk_prot->recvmsg(sk, msg, size, flags);
4022 }
4023 EXPORT_SYMBOL(sock_common_recvmsg);
4024 
4025 /*
4026  *	Set socket options on an inet socket.
4027  */
4028 int sock_common_setsockopt(struct socket *sock, int level, int optname,
4029 			   sockptr_t optval, unsigned int optlen)
4030 {
4031 	struct sock *sk = sock->sk;
4032 
4033 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
4034 	return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
4035 }
4036 EXPORT_SYMBOL(sock_common_setsockopt);
4037 
4038 void sk_common_release(struct sock *sk)
4039 {
4040 	if (sk->sk_prot->destroy)
4041 		sk->sk_prot->destroy(sk);
4042 
4043 	/*
4044 	 * Observation: when sk_common_release is called, processes have
4045 	 * no access to socket. But net still has.
4046 	 * Step one, detach it from networking:
4047 	 *
4048 	 * A. Remove from hash tables.
4049 	 */
4050 
4051 	sk->sk_prot->unhash(sk);
4052 
4053 	/*
4054 	 * In this point socket cannot receive new packets, but it is possible
4055 	 * that some packets are in flight because some CPU runs receiver and
4056 	 * did hash table lookup before we unhashed socket. They will achieve
4057 	 * receive queue and will be purged by socket destructor.
4058 	 *
4059 	 * Also we still have packets pending on receive queue and probably,
4060 	 * our own packets waiting in device queues. sock_destroy will drain
4061 	 * receive queue, but transmitted packets will delay socket destruction
4062 	 * until the last reference will be released.
4063 	 */
4064 
4065 	sock_orphan(sk);
4066 
4067 	xfrm_sk_free_policy(sk);
4068 
4069 	sock_put(sk);
4070 }
4071 EXPORT_SYMBOL(sk_common_release);
4072 
4073 void sk_get_meminfo(const struct sock *sk, u32 *mem)
4074 {
4075 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
4076 
4077 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
4078 	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
4079 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
4080 	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
4081 	mem[SK_MEMINFO_FWD_ALLOC] = READ_ONCE(sk->sk_forward_alloc);
4082 	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
4083 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
4084 	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
4085 	mem[SK_MEMINFO_DROPS] = sk_drops_read(sk);
4086 }
4087 
4088 #ifdef CONFIG_PROC_FS
4089 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
4090 
4091 int sock_prot_inuse_get(struct net *net, struct proto *prot)
4092 {
4093 	int cpu, idx = prot->inuse_idx;
4094 	int res = 0;
4095 
4096 	for_each_possible_cpu(cpu)
4097 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
4098 
4099 	return res >= 0 ? res : 0;
4100 }
4101 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
4102 
4103 int sock_inuse_get(struct net *net)
4104 {
4105 	int cpu, res = 0;
4106 
4107 	for_each_possible_cpu(cpu)
4108 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
4109 
4110 	return res;
4111 }
4112 
4113 EXPORT_SYMBOL_GPL(sock_inuse_get);
4114 
4115 static int __net_init sock_inuse_init_net(struct net *net)
4116 {
4117 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
4118 	if (net->core.prot_inuse == NULL)
4119 		return -ENOMEM;
4120 	return 0;
4121 }
4122 
4123 static void __net_exit sock_inuse_exit_net(struct net *net)
4124 {
4125 	free_percpu(net->core.prot_inuse);
4126 }
4127 
4128 static struct pernet_operations net_inuse_ops = {
4129 	.init = sock_inuse_init_net,
4130 	.exit = sock_inuse_exit_net,
4131 };
4132 
4133 static __init int net_inuse_init(void)
4134 {
4135 	if (register_pernet_subsys(&net_inuse_ops))
4136 		panic("Cannot initialize net inuse counters");
4137 
4138 	return 0;
4139 }
4140 
4141 core_initcall(net_inuse_init);
4142 
4143 static int assign_proto_idx(struct proto *prot)
4144 {
4145 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
4146 
4147 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR)) {
4148 		pr_err("PROTO_INUSE_NR exhausted\n");
4149 		return -ENOSPC;
4150 	}
4151 
4152 	set_bit(prot->inuse_idx, proto_inuse_idx);
4153 	return 0;
4154 }
4155 
4156 static void release_proto_idx(struct proto *prot)
4157 {
4158 	if (prot->inuse_idx != PROTO_INUSE_NR)
4159 		clear_bit(prot->inuse_idx, proto_inuse_idx);
4160 }
4161 #else
4162 static inline int assign_proto_idx(struct proto *prot)
4163 {
4164 	return 0;
4165 }
4166 
4167 static inline void release_proto_idx(struct proto *prot)
4168 {
4169 }
4170 
4171 #endif
4172 
4173 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
4174 {
4175 	if (!twsk_prot)
4176 		return;
4177 	kfree(twsk_prot->twsk_slab_name);
4178 	twsk_prot->twsk_slab_name = NULL;
4179 	kmem_cache_destroy(twsk_prot->twsk_slab);
4180 	twsk_prot->twsk_slab = NULL;
4181 }
4182 
4183 static int tw_prot_init(const struct proto *prot)
4184 {
4185 	struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
4186 
4187 	if (!twsk_prot)
4188 		return 0;
4189 
4190 	twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
4191 					      prot->name);
4192 	if (!twsk_prot->twsk_slab_name)
4193 		return -ENOMEM;
4194 
4195 	twsk_prot->twsk_slab =
4196 		kmem_cache_create(twsk_prot->twsk_slab_name,
4197 				  twsk_prot->twsk_obj_size, 0,
4198 				  SLAB_ACCOUNT | prot->slab_flags,
4199 				  NULL);
4200 	if (!twsk_prot->twsk_slab) {
4201 		pr_crit("%s: Can't create timewait sock SLAB cache!\n",
4202 			prot->name);
4203 		return -ENOMEM;
4204 	}
4205 
4206 	return 0;
4207 }
4208 
4209 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
4210 {
4211 	if (!rsk_prot)
4212 		return;
4213 	kfree(rsk_prot->slab_name);
4214 	rsk_prot->slab_name = NULL;
4215 	kmem_cache_destroy(rsk_prot->slab);
4216 	rsk_prot->slab = NULL;
4217 }
4218 
4219 static int req_prot_init(const struct proto *prot)
4220 {
4221 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
4222 
4223 	if (!rsk_prot)
4224 		return 0;
4225 
4226 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4227 					prot->name);
4228 	if (!rsk_prot->slab_name)
4229 		return -ENOMEM;
4230 
4231 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4232 					   rsk_prot->obj_size, 0,
4233 					   SLAB_ACCOUNT | prot->slab_flags,
4234 					   NULL);
4235 
4236 	if (!rsk_prot->slab) {
4237 		pr_crit("%s: Can't create request sock SLAB cache!\n",
4238 			prot->name);
4239 		return -ENOMEM;
4240 	}
4241 	return 0;
4242 }
4243 
4244 int proto_register(struct proto *prot, int alloc_slab)
4245 {
4246 	int ret = -ENOBUFS;
4247 
4248 	if (prot->memory_allocated && !prot->sysctl_mem) {
4249 		pr_err("%s: missing sysctl_mem\n", prot->name);
4250 		return -EINVAL;
4251 	}
4252 	if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4253 		pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4254 		return -EINVAL;
4255 	}
4256 	if (alloc_slab) {
4257 		struct kmem_cache_args args = {
4258 			.useroffset	= prot->useroffset,
4259 			.usersize	= prot->usersize,
4260 			.freeptr_offset = prot->freeptr_offset,
4261 			.use_freeptr_offset = !!prot->freeptr_offset,
4262 		};
4263 
4264 		prot->slab = kmem_cache_create(prot->name, prot->obj_size,
4265 					&args,
4266 					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4267 					prot->slab_flags);
4268 		if (prot->slab == NULL) {
4269 			pr_crit("%s: Can't create sock SLAB cache!\n",
4270 				prot->name);
4271 			goto out;
4272 		}
4273 
4274 		if (req_prot_init(prot))
4275 			goto out_free_request_sock_slab;
4276 
4277 		if (tw_prot_init(prot))
4278 			goto out_free_timewait_sock_slab;
4279 	}
4280 
4281 	mutex_lock(&proto_list_mutex);
4282 	ret = assign_proto_idx(prot);
4283 	if (ret) {
4284 		mutex_unlock(&proto_list_mutex);
4285 		goto out_free_timewait_sock_slab;
4286 	}
4287 	list_add(&prot->node, &proto_list);
4288 	mutex_unlock(&proto_list_mutex);
4289 	return ret;
4290 
4291 out_free_timewait_sock_slab:
4292 	if (alloc_slab)
4293 		tw_prot_cleanup(prot->twsk_prot);
4294 out_free_request_sock_slab:
4295 	if (alloc_slab) {
4296 		req_prot_cleanup(prot->rsk_prot);
4297 
4298 		kmem_cache_destroy(prot->slab);
4299 		prot->slab = NULL;
4300 	}
4301 out:
4302 	return ret;
4303 }
4304 EXPORT_SYMBOL(proto_register);
4305 
4306 void proto_unregister(struct proto *prot)
4307 {
4308 	mutex_lock(&proto_list_mutex);
4309 	release_proto_idx(prot);
4310 	list_del(&prot->node);
4311 	mutex_unlock(&proto_list_mutex);
4312 
4313 	kmem_cache_destroy(prot->slab);
4314 	prot->slab = NULL;
4315 
4316 	req_prot_cleanup(prot->rsk_prot);
4317 	tw_prot_cleanup(prot->twsk_prot);
4318 }
4319 EXPORT_SYMBOL(proto_unregister);
4320 
4321 int sock_load_diag_module(int family, int protocol)
4322 {
4323 	if (!protocol) {
4324 		if (!sock_is_registered(family))
4325 			return -ENOENT;
4326 
4327 		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4328 				      NETLINK_SOCK_DIAG, family);
4329 	}
4330 
4331 #ifdef CONFIG_INET
4332 	if (family == AF_INET &&
4333 	    protocol != IPPROTO_RAW &&
4334 	    protocol < MAX_INET_PROTOS &&
4335 	    !rcu_access_pointer(inet_protos[protocol]))
4336 		return -ENOENT;
4337 #endif
4338 
4339 	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4340 			      NETLINK_SOCK_DIAG, family, protocol);
4341 }
4342 EXPORT_SYMBOL(sock_load_diag_module);
4343 
4344 #ifdef CONFIG_PROC_FS
4345 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4346 	__acquires(proto_list_mutex)
4347 {
4348 	mutex_lock(&proto_list_mutex);
4349 	return seq_list_start_head(&proto_list, *pos);
4350 }
4351 
4352 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4353 {
4354 	return seq_list_next(v, &proto_list, pos);
4355 }
4356 
4357 static void proto_seq_stop(struct seq_file *seq, void *v)
4358 	__releases(proto_list_mutex)
4359 {
4360 	mutex_unlock(&proto_list_mutex);
4361 }
4362 
4363 static char proto_method_implemented(const void *method)
4364 {
4365 	return method == NULL ? 'n' : 'y';
4366 }
4367 static long sock_prot_memory_allocated(struct proto *proto)
4368 {
4369 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4370 }
4371 
4372 static const char *sock_prot_memory_pressure(struct proto *proto)
4373 {
4374 	return proto->memory_pressure != NULL ?
4375 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4376 }
4377 
4378 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4379 {
4380 
4381 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
4382 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4383 		   proto->name,
4384 		   proto->obj_size,
4385 		   sock_prot_inuse_get(seq_file_net(seq), proto),
4386 		   sock_prot_memory_allocated(proto),
4387 		   sock_prot_memory_pressure(proto),
4388 		   proto->max_header,
4389 		   proto->slab == NULL ? "no" : "yes",
4390 		   module_name(proto->owner),
4391 		   proto_method_implemented(proto->close),
4392 		   proto_method_implemented(proto->connect),
4393 		   proto_method_implemented(proto->disconnect),
4394 		   proto_method_implemented(proto->accept),
4395 		   proto_method_implemented(proto->ioctl),
4396 		   proto_method_implemented(proto->init),
4397 		   proto_method_implemented(proto->destroy),
4398 		   proto_method_implemented(proto->shutdown),
4399 		   proto_method_implemented(proto->setsockopt),
4400 		   proto_method_implemented(proto->getsockopt),
4401 		   proto_method_implemented(proto->sendmsg),
4402 		   proto_method_implemented(proto->recvmsg),
4403 		   proto_method_implemented(proto->bind),
4404 		   proto_method_implemented(proto->backlog_rcv),
4405 		   proto_method_implemented(proto->hash),
4406 		   proto_method_implemented(proto->unhash),
4407 		   proto_method_implemented(proto->get_port),
4408 		   proto_method_implemented(proto->enter_memory_pressure));
4409 }
4410 
4411 static int proto_seq_show(struct seq_file *seq, void *v)
4412 {
4413 	if (v == &proto_list)
4414 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4415 			   "protocol",
4416 			   "size",
4417 			   "sockets",
4418 			   "memory",
4419 			   "press",
4420 			   "maxhdr",
4421 			   "slab",
4422 			   "module",
4423 			   "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4424 	else
4425 		proto_seq_printf(seq, list_entry(v, struct proto, node));
4426 	return 0;
4427 }
4428 
4429 static const struct seq_operations proto_seq_ops = {
4430 	.start  = proto_seq_start,
4431 	.next   = proto_seq_next,
4432 	.stop   = proto_seq_stop,
4433 	.show   = proto_seq_show,
4434 };
4435 
4436 static __net_init int proto_init_net(struct net *net)
4437 {
4438 	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4439 			sizeof(struct seq_net_private)))
4440 		return -ENOMEM;
4441 
4442 	return 0;
4443 }
4444 
4445 static __net_exit void proto_exit_net(struct net *net)
4446 {
4447 	remove_proc_entry("protocols", net->proc_net);
4448 }
4449 
4450 
4451 static __net_initdata struct pernet_operations proto_net_ops = {
4452 	.init = proto_init_net,
4453 	.exit = proto_exit_net,
4454 };
4455 
4456 static int __init proto_init(void)
4457 {
4458 	return register_pernet_subsys(&proto_net_ops);
4459 }
4460 
4461 subsys_initcall(proto_init);
4462 
4463 #endif /* PROC_FS */
4464 
4465 #ifdef CONFIG_NET_RX_BUSY_POLL
4466 bool sk_busy_loop_end(void *p, unsigned long start_time)
4467 {
4468 	struct sock *sk = p;
4469 
4470 	if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4471 		return true;
4472 
4473 	if (sk_is_udp(sk) &&
4474 	    !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4475 		return true;
4476 
4477 	return sk_busy_loop_timeout(sk, start_time);
4478 }
4479 EXPORT_SYMBOL(sk_busy_loop_end);
4480 #endif /* CONFIG_NET_RX_BUSY_POLL */
4481 
4482 int sock_bind_add(struct sock *sk, struct sockaddr_unsized *addr, int addr_len)
4483 {
4484 	if (!sk->sk_prot->bind_add)
4485 		return -EOPNOTSUPP;
4486 	return sk->sk_prot->bind_add(sk, addr, addr_len);
4487 }
4488 EXPORT_SYMBOL(sock_bind_add);
4489 
4490 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4491 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4492 		     void __user *arg, void *karg, size_t size)
4493 {
4494 	int ret;
4495 
4496 	if (copy_from_user(karg, arg, size))
4497 		return -EFAULT;
4498 
4499 	ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4500 	if (ret)
4501 		return ret;
4502 
4503 	if (copy_to_user(arg, karg, size))
4504 		return -EFAULT;
4505 
4506 	return 0;
4507 }
4508 EXPORT_SYMBOL(sock_ioctl_inout);
4509 
4510 /* This is the most common ioctl prep function, where the result (4 bytes) is
4511  * copied back to userspace if the ioctl() returns successfully. No input is
4512  * copied from userspace as input argument.
4513  */
4514 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4515 {
4516 	int ret, karg = 0;
4517 
4518 	ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4519 	if (ret)
4520 		return ret;
4521 
4522 	return put_user(karg, (int __user *)arg);
4523 }
4524 
4525 /* A wrapper around sock ioctls, which copies the data from userspace
4526  * (depending on the protocol/ioctl), and copies back the result to userspace.
4527  * The main motivation for this function is to pass kernel memory to the
4528  * protocol ioctl callbacks, instead of userspace memory.
4529  */
4530 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4531 {
4532 	int rc = 1;
4533 
4534 	if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4535 		rc = ipmr_sk_ioctl(sk, cmd, arg);
4536 	else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4537 		rc = ip6mr_sk_ioctl(sk, cmd, arg);
4538 	else if (sk_is_phonet(sk))
4539 		rc = phonet_sk_ioctl(sk, cmd, arg);
4540 
4541 	/* If ioctl was processed, returns its value */
4542 	if (rc <= 0)
4543 		return rc;
4544 
4545 	/* Otherwise call the default handler */
4546 	return sock_ioctl_out(sk, cmd, arg);
4547 }
4548 EXPORT_SYMBOL(sk_ioctl);
4549 
4550 static int __init sock_struct_check(void)
4551 {
4552 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4553 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4554 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4555 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4556 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4557 
4558 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4559 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4560 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4561 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4562 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4563 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4564 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4565 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4566 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4567 
4568 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4569 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4570 #ifdef CONFIG_MEMCG
4571 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4572 #endif
4573 
4574 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4575 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4576 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4577 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4578 
4579 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4580 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4581 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_err_soft);
4582 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4583 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4584 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4585 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4586 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4587 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4588 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4589 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4590 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4591 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4592 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4593 
4594 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_pending_confirm);
4595 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_status);
4596 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4597 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4598 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4599 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4600 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_uid);
4601 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_protocol);
4602 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4603 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4604 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4605 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4606 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4607 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4608 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndbuf);
4609 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4610 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4611 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4612 	return 0;
4613 }
4614 
4615 core_initcall(sock_struct_check);
4616