xref: /linux/net/core/sock.c (revision 6d05d3cb44c52bcd1739575b55e02dc8bdfcf41a)
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 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1469 		if (!ret)
1470 			ret = sk_attach_filter(&fprog, sk);
1471 		break;
1472 	}
1473 	case SO_ATTACH_BPF:
1474 		ret = -EINVAL;
1475 		if (optlen == sizeof(u32)) {
1476 			u32 ufd;
1477 
1478 			ret = -EFAULT;
1479 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1480 				break;
1481 
1482 			ret = sk_attach_bpf(ufd, sk);
1483 		}
1484 		break;
1485 
1486 	case SO_ATTACH_REUSEPORT_CBPF: {
1487 		struct sock_fprog fprog;
1488 
1489 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1490 		if (!ret)
1491 			ret = sk_reuseport_attach_filter(&fprog, sk);
1492 		break;
1493 	}
1494 	case SO_ATTACH_REUSEPORT_EBPF:
1495 		ret = -EINVAL;
1496 		if (optlen == sizeof(u32)) {
1497 			u32 ufd;
1498 
1499 			ret = -EFAULT;
1500 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1501 				break;
1502 
1503 			ret = sk_reuseport_attach_bpf(ufd, sk);
1504 		}
1505 		break;
1506 
1507 	case SO_DETACH_REUSEPORT_BPF:
1508 		ret = reuseport_detach_prog(sk);
1509 		break;
1510 
1511 	case SO_DETACH_FILTER:
1512 		ret = sk_detach_filter(sk);
1513 		break;
1514 
1515 	case SO_LOCK_FILTER:
1516 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1517 			ret = -EPERM;
1518 		else
1519 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1520 		break;
1521 
1522 	case SO_MARK:
1523 		if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1524 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1525 			ret = -EPERM;
1526 			break;
1527 		}
1528 
1529 		__sock_set_mark(sk, val);
1530 		break;
1531 	case SO_RCVMARK:
1532 		sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1533 		break;
1534 
1535 	case SO_RCVPRIORITY:
1536 		sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool);
1537 		break;
1538 
1539 	case SO_RXQ_OVFL:
1540 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1541 		break;
1542 
1543 	case SO_WIFI_STATUS:
1544 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1545 		break;
1546 
1547 	case SO_NOFCS:
1548 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1549 		break;
1550 
1551 	case SO_SELECT_ERR_QUEUE:
1552 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1553 		break;
1554 
1555 	case SO_PASSCRED:
1556 		if (sk_may_scm_recv(sk))
1557 			sk->sk_scm_credentials = valbool;
1558 		else
1559 			ret = -EOPNOTSUPP;
1560 		break;
1561 
1562 	case SO_PASSSEC:
1563 		if (IS_ENABLED(CONFIG_SECURITY_NETWORK) && sk_may_scm_recv(sk))
1564 			sk->sk_scm_security = valbool;
1565 		else
1566 			ret = -EOPNOTSUPP;
1567 		break;
1568 
1569 	case SO_PASSPIDFD:
1570 		if (sk_is_unix(sk))
1571 			sk->sk_scm_pidfd = valbool;
1572 		else
1573 			ret = -EOPNOTSUPP;
1574 		break;
1575 
1576 	case SO_PASSRIGHTS:
1577 		if (sk_is_unix(sk))
1578 			sk->sk_scm_rights = valbool;
1579 		else
1580 			ret = -EOPNOTSUPP;
1581 		break;
1582 
1583 	case SO_INCOMING_CPU:
1584 		reuseport_update_incoming_cpu(sk, val);
1585 		break;
1586 
1587 	case SO_CNX_ADVICE:
1588 		if (val == 1)
1589 			dst_negative_advice(sk);
1590 		break;
1591 
1592 	case SO_ZEROCOPY:
1593 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1594 			if (!(sk_is_tcp(sk) ||
1595 			      (sk->sk_type == SOCK_DGRAM &&
1596 			       sk->sk_protocol == IPPROTO_UDP)))
1597 				ret = -EOPNOTSUPP;
1598 		} else if (sk->sk_family != PF_RDS) {
1599 			ret = -EOPNOTSUPP;
1600 		}
1601 		if (!ret) {
1602 			if (val < 0 || val > 1)
1603 				ret = -EINVAL;
1604 			else
1605 				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1606 		}
1607 		break;
1608 
1609 	case SO_TXTIME:
1610 		if (optlen != sizeof(struct sock_txtime)) {
1611 			ret = -EINVAL;
1612 			break;
1613 		} else if (copy_from_sockptr(&sk_txtime, optval,
1614 			   sizeof(struct sock_txtime))) {
1615 			ret = -EFAULT;
1616 			break;
1617 		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1618 			ret = -EINVAL;
1619 			break;
1620 		}
1621 		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1622 		 * scheduler has enough safe guards.
1623 		 */
1624 		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1625 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1626 			ret = -EPERM;
1627 			break;
1628 		}
1629 
1630 		ret = sockopt_validate_clockid(sk_txtime.clockid);
1631 		if (ret)
1632 			break;
1633 
1634 		sock_valbool_flag(sk, SOCK_TXTIME, true);
1635 		sk->sk_clockid = sk_txtime.clockid;
1636 		sk->sk_txtime_deadline_mode =
1637 			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1638 		sk->sk_txtime_report_errors =
1639 			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1640 		break;
1641 
1642 	case SO_BINDTOIFINDEX:
1643 		ret = sock_bindtoindex_locked(sk, val);
1644 		break;
1645 
1646 	case SO_BUF_LOCK:
1647 		if (val & ~SOCK_BUF_LOCK_MASK) {
1648 			ret = -EINVAL;
1649 			break;
1650 		}
1651 		sk->sk_userlocks = val | (sk->sk_userlocks &
1652 					  ~SOCK_BUF_LOCK_MASK);
1653 		break;
1654 
1655 	case SO_RESERVE_MEM:
1656 	{
1657 		int delta;
1658 
1659 		if (val < 0) {
1660 			ret = -EINVAL;
1661 			break;
1662 		}
1663 
1664 		delta = val - sk->sk_reserved_mem;
1665 		if (delta < 0)
1666 			sock_release_reserved_memory(sk, -delta);
1667 		else
1668 			ret = sock_reserve_memory(sk, delta);
1669 		break;
1670 	}
1671 
1672 	default:
1673 		ret = -ENOPROTOOPT;
1674 		break;
1675 	}
1676 	sockopt_release_sock(sk);
1677 	return ret;
1678 }
1679 
1680 int sock_setsockopt(struct socket *sock, int level, int optname,
1681 		    sockptr_t optval, unsigned int optlen)
1682 {
1683 	return sk_setsockopt(sock->sk, level, optname,
1684 			     optval, optlen);
1685 }
1686 EXPORT_SYMBOL(sock_setsockopt);
1687 
1688 static const struct cred *sk_get_peer_cred(struct sock *sk)
1689 {
1690 	const struct cred *cred;
1691 
1692 	spin_lock(&sk->sk_peer_lock);
1693 	cred = get_cred(sk->sk_peer_cred);
1694 	spin_unlock(&sk->sk_peer_lock);
1695 
1696 	return cred;
1697 }
1698 
1699 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1700 			  struct ucred *ucred)
1701 {
1702 	ucred->pid = pid_vnr(pid);
1703 	ucred->uid = ucred->gid = -1;
1704 	if (cred) {
1705 		struct user_namespace *current_ns = current_user_ns();
1706 
1707 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1708 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1709 	}
1710 }
1711 
1712 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1713 {
1714 	struct user_namespace *user_ns = current_user_ns();
1715 	int i;
1716 
1717 	for (i = 0; i < src->ngroups; i++) {
1718 		gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1719 
1720 		if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1721 			return -EFAULT;
1722 	}
1723 
1724 	return 0;
1725 }
1726 
1727 int sk_getsockopt(struct sock *sk, int level, int optname,
1728 		  sockptr_t optval, sockptr_t optlen)
1729 {
1730 	struct socket *sock = sk->sk_socket;
1731 
1732 	union {
1733 		int val;
1734 		u64 val64;
1735 		unsigned long ulval;
1736 		struct linger ling;
1737 		struct old_timeval32 tm32;
1738 		struct __kernel_old_timeval tm;
1739 		struct  __kernel_sock_timeval stm;
1740 		struct sock_txtime txtime;
1741 		struct so_timestamping timestamping;
1742 	} v;
1743 
1744 	int lv = sizeof(int);
1745 	int len;
1746 
1747 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
1748 		return -EFAULT;
1749 	if (len < 0)
1750 		return -EINVAL;
1751 
1752 	memset(&v, 0, sizeof(v));
1753 
1754 	switch (optname) {
1755 	case SO_DEBUG:
1756 		v.val = sock_flag(sk, SOCK_DBG);
1757 		break;
1758 
1759 	case SO_DONTROUTE:
1760 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1761 		break;
1762 
1763 	case SO_BROADCAST:
1764 		v.val = sock_flag(sk, SOCK_BROADCAST);
1765 		break;
1766 
1767 	case SO_SNDBUF:
1768 		v.val = READ_ONCE(sk->sk_sndbuf);
1769 		break;
1770 
1771 	case SO_RCVBUF:
1772 		v.val = READ_ONCE(sk->sk_rcvbuf);
1773 		break;
1774 
1775 	case SO_REUSEADDR:
1776 		v.val = sk->sk_reuse;
1777 		break;
1778 
1779 	case SO_REUSEPORT:
1780 		v.val = sk->sk_reuseport;
1781 		break;
1782 
1783 	case SO_KEEPALIVE:
1784 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1785 		break;
1786 
1787 	case SO_TYPE:
1788 		v.val = sk->sk_type;
1789 		break;
1790 
1791 	case SO_PROTOCOL:
1792 		v.val = sk->sk_protocol;
1793 		break;
1794 
1795 	case SO_DOMAIN:
1796 		v.val = sk->sk_family;
1797 		break;
1798 
1799 	case SO_ERROR:
1800 		v.val = -sock_error(sk);
1801 		if (v.val == 0)
1802 			v.val = xchg(&sk->sk_err_soft, 0);
1803 		break;
1804 
1805 	case SO_OOBINLINE:
1806 		v.val = sock_flag(sk, SOCK_URGINLINE);
1807 		break;
1808 
1809 	case SO_NO_CHECK:
1810 		v.val = sk->sk_no_check_tx;
1811 		break;
1812 
1813 	case SO_PRIORITY:
1814 		v.val = READ_ONCE(sk->sk_priority);
1815 		break;
1816 
1817 	case SO_LINGER:
1818 		lv		= sizeof(v.ling);
1819 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1820 		v.ling.l_linger	= READ_ONCE(sk->sk_lingertime) / HZ;
1821 		break;
1822 
1823 	case SO_BSDCOMPAT:
1824 		break;
1825 
1826 	case SO_TIMESTAMP_OLD:
1827 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1828 				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1829 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1830 		break;
1831 
1832 	case SO_TIMESTAMPNS_OLD:
1833 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1834 		break;
1835 
1836 	case SO_TIMESTAMP_NEW:
1837 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1838 		break;
1839 
1840 	case SO_TIMESTAMPNS_NEW:
1841 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1842 		break;
1843 
1844 	case SO_TIMESTAMPING_OLD:
1845 	case SO_TIMESTAMPING_NEW:
1846 		lv = sizeof(v.timestamping);
1847 		/* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1848 		 * returning the flags when they were set through the same option.
1849 		 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1850 		 */
1851 		if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1852 			v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1853 			v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1854 		}
1855 		break;
1856 
1857 	case SO_RCVTIMEO_OLD:
1858 	case SO_RCVTIMEO_NEW:
1859 		lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1860 				      SO_RCVTIMEO_OLD == optname);
1861 		break;
1862 
1863 	case SO_SNDTIMEO_OLD:
1864 	case SO_SNDTIMEO_NEW:
1865 		lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1866 				      SO_SNDTIMEO_OLD == optname);
1867 		break;
1868 
1869 	case SO_RCVLOWAT:
1870 		v.val = READ_ONCE(sk->sk_rcvlowat);
1871 		break;
1872 
1873 	case SO_SNDLOWAT:
1874 		v.val = 1;
1875 		break;
1876 
1877 	case SO_PASSCRED:
1878 		if (!sk_may_scm_recv(sk))
1879 			return -EOPNOTSUPP;
1880 
1881 		v.val = sk->sk_scm_credentials;
1882 		break;
1883 
1884 	case SO_PASSPIDFD:
1885 		if (!sk_is_unix(sk))
1886 			return -EOPNOTSUPP;
1887 
1888 		v.val = sk->sk_scm_pidfd;
1889 		break;
1890 
1891 	case SO_PASSRIGHTS:
1892 		if (!sk_is_unix(sk))
1893 			return -EOPNOTSUPP;
1894 
1895 		v.val = sk->sk_scm_rights;
1896 		break;
1897 
1898 	case SO_PEERCRED:
1899 	{
1900 		struct ucred peercred;
1901 		if (len > sizeof(peercred))
1902 			len = sizeof(peercred);
1903 
1904 		spin_lock(&sk->sk_peer_lock);
1905 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1906 		spin_unlock(&sk->sk_peer_lock);
1907 
1908 		if (copy_to_sockptr(optval, &peercred, len))
1909 			return -EFAULT;
1910 		goto lenout;
1911 	}
1912 
1913 	case SO_PEERPIDFD:
1914 	{
1915 		struct pid *peer_pid;
1916 		struct file *pidfd_file = NULL;
1917 		unsigned int flags = 0;
1918 		int pidfd;
1919 
1920 		if (len > sizeof(pidfd))
1921 			len = sizeof(pidfd);
1922 
1923 		spin_lock(&sk->sk_peer_lock);
1924 		peer_pid = get_pid(sk->sk_peer_pid);
1925 		spin_unlock(&sk->sk_peer_lock);
1926 
1927 		if (!peer_pid)
1928 			return -ENODATA;
1929 
1930 		/* The use of PIDFD_STALE requires stashing of struct pid
1931 		 * on pidfs with pidfs_register_pid() and only AF_UNIX
1932 		 * were prepared for this.
1933 		 */
1934 		if (sk->sk_family == AF_UNIX)
1935 			flags = PIDFD_STALE;
1936 
1937 		pidfd = pidfd_prepare(peer_pid, flags, &pidfd_file);
1938 		put_pid(peer_pid);
1939 		if (pidfd < 0)
1940 			return pidfd;
1941 
1942 		if (copy_to_sockptr(optval, &pidfd, len) ||
1943 		    copy_to_sockptr(optlen, &len, sizeof(int))) {
1944 			put_unused_fd(pidfd);
1945 			fput(pidfd_file);
1946 
1947 			return -EFAULT;
1948 		}
1949 
1950 		fd_install(pidfd, pidfd_file);
1951 		return 0;
1952 	}
1953 
1954 	case SO_PEERGROUPS:
1955 	{
1956 		const struct cred *cred;
1957 		int ret, n;
1958 
1959 		cred = sk_get_peer_cred(sk);
1960 		if (!cred)
1961 			return -ENODATA;
1962 
1963 		n = cred->group_info->ngroups;
1964 		if (len < n * sizeof(gid_t)) {
1965 			len = n * sizeof(gid_t);
1966 			put_cred(cred);
1967 			return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1968 		}
1969 		len = n * sizeof(gid_t);
1970 
1971 		ret = groups_to_user(optval, cred->group_info);
1972 		put_cred(cred);
1973 		if (ret)
1974 			return ret;
1975 		goto lenout;
1976 	}
1977 
1978 	case SO_PEERNAME:
1979 	{
1980 		struct sockaddr_storage address;
1981 
1982 		lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1983 		if (lv < 0)
1984 			return -ENOTCONN;
1985 		if (lv < len)
1986 			return -EINVAL;
1987 		if (copy_to_sockptr(optval, &address, len))
1988 			return -EFAULT;
1989 		goto lenout;
1990 	}
1991 
1992 	/* Dubious BSD thing... Probably nobody even uses it, but
1993 	 * the UNIX standard wants it for whatever reason... -DaveM
1994 	 */
1995 	case SO_ACCEPTCONN:
1996 		v.val = sk->sk_state == TCP_LISTEN;
1997 		break;
1998 
1999 	case SO_PASSSEC:
2000 		if (!IS_ENABLED(CONFIG_SECURITY_NETWORK) || !sk_may_scm_recv(sk))
2001 			return -EOPNOTSUPP;
2002 
2003 		v.val = sk->sk_scm_security;
2004 		break;
2005 
2006 	case SO_PEERSEC:
2007 		return security_socket_getpeersec_stream(sock,
2008 							 optval, optlen, len);
2009 
2010 	case SO_MARK:
2011 		v.val = READ_ONCE(sk->sk_mark);
2012 		break;
2013 
2014 	case SO_RCVMARK:
2015 		v.val = sock_flag(sk, SOCK_RCVMARK);
2016 		break;
2017 
2018 	case SO_RCVPRIORITY:
2019 		v.val = sock_flag(sk, SOCK_RCVPRIORITY);
2020 		break;
2021 
2022 	case SO_RXQ_OVFL:
2023 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
2024 		break;
2025 
2026 	case SO_WIFI_STATUS:
2027 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
2028 		break;
2029 
2030 	case SO_PEEK_OFF:
2031 		if (!READ_ONCE(sock->ops)->set_peek_off)
2032 			return -EOPNOTSUPP;
2033 
2034 		v.val = READ_ONCE(sk->sk_peek_off);
2035 		break;
2036 	case SO_NOFCS:
2037 		v.val = sock_flag(sk, SOCK_NOFCS);
2038 		break;
2039 
2040 	case SO_BINDTODEVICE:
2041 		return sock_getbindtodevice(sk, optval, optlen, len);
2042 
2043 	case SO_GET_FILTER:
2044 		len = sk_get_filter(sk, optval, len);
2045 		if (len < 0)
2046 			return len;
2047 
2048 		goto lenout;
2049 
2050 	case SO_LOCK_FILTER:
2051 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
2052 		break;
2053 
2054 	case SO_BPF_EXTENSIONS:
2055 		v.val = bpf_tell_extensions();
2056 		break;
2057 
2058 	case SO_SELECT_ERR_QUEUE:
2059 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
2060 		break;
2061 
2062 #ifdef CONFIG_NET_RX_BUSY_POLL
2063 	case SO_BUSY_POLL:
2064 		v.val = READ_ONCE(sk->sk_ll_usec);
2065 		break;
2066 	case SO_PREFER_BUSY_POLL:
2067 		v.val = READ_ONCE(sk->sk_prefer_busy_poll);
2068 		break;
2069 #endif
2070 
2071 	case SO_MAX_PACING_RATE:
2072 		/* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
2073 		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2074 			lv = sizeof(v.ulval);
2075 			v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2076 		} else {
2077 			/* 32bit version */
2078 			v.val = min_t(unsigned long, ~0U,
2079 				      READ_ONCE(sk->sk_max_pacing_rate));
2080 		}
2081 		break;
2082 
2083 	case SO_INCOMING_CPU:
2084 		v.val = READ_ONCE(sk->sk_incoming_cpu);
2085 		break;
2086 
2087 	case SO_MEMINFO:
2088 	{
2089 		u32 meminfo[SK_MEMINFO_VARS];
2090 
2091 		sk_get_meminfo(sk, meminfo);
2092 
2093 		len = min_t(unsigned int, len, sizeof(meminfo));
2094 		if (copy_to_sockptr(optval, &meminfo, len))
2095 			return -EFAULT;
2096 
2097 		goto lenout;
2098 	}
2099 
2100 #ifdef CONFIG_NET_RX_BUSY_POLL
2101 	case SO_INCOMING_NAPI_ID:
2102 		v.val = READ_ONCE(sk->sk_napi_id);
2103 
2104 		/* aggregate non-NAPI IDs down to 0 */
2105 		if (!napi_id_valid(v.val))
2106 			v.val = 0;
2107 
2108 		break;
2109 #endif
2110 
2111 	case SO_COOKIE:
2112 		lv = sizeof(u64);
2113 		if (len < lv)
2114 			return -EINVAL;
2115 		v.val64 = sock_gen_cookie(sk);
2116 		break;
2117 
2118 	case SO_ZEROCOPY:
2119 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
2120 		break;
2121 
2122 	case SO_TXTIME:
2123 		lv = sizeof(v.txtime);
2124 		v.txtime.clockid = sk->sk_clockid;
2125 		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2126 				  SOF_TXTIME_DEADLINE_MODE : 0;
2127 		v.txtime.flags |= sk->sk_txtime_report_errors ?
2128 				  SOF_TXTIME_REPORT_ERRORS : 0;
2129 		break;
2130 
2131 	case SO_BINDTOIFINDEX:
2132 		v.val = READ_ONCE(sk->sk_bound_dev_if);
2133 		break;
2134 
2135 	case SO_NETNS_COOKIE:
2136 		lv = sizeof(u64);
2137 		if (len != lv)
2138 			return -EINVAL;
2139 		v.val64 = sock_net(sk)->net_cookie;
2140 		break;
2141 
2142 	case SO_BUF_LOCK:
2143 		v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2144 		break;
2145 
2146 	case SO_RESERVE_MEM:
2147 		v.val = READ_ONCE(sk->sk_reserved_mem);
2148 		break;
2149 
2150 	case SO_TXREHASH:
2151 		if (!sk_is_tcp(sk))
2152 			return -EOPNOTSUPP;
2153 
2154 		/* Paired with WRITE_ONCE() in sk_setsockopt() */
2155 		v.val = READ_ONCE(sk->sk_txrehash);
2156 		break;
2157 
2158 	default:
2159 		/* We implement the SO_SNDLOWAT etc to not be settable
2160 		 * (1003.1g 7).
2161 		 */
2162 		return -ENOPROTOOPT;
2163 	}
2164 
2165 	if (len > lv)
2166 		len = lv;
2167 	if (copy_to_sockptr(optval, &v, len))
2168 		return -EFAULT;
2169 lenout:
2170 	if (copy_to_sockptr(optlen, &len, sizeof(int)))
2171 		return -EFAULT;
2172 	return 0;
2173 }
2174 
2175 /*
2176  * Initialize an sk_lock.
2177  *
2178  * (We also register the sk_lock with the lock validator.)
2179  */
2180 static inline void sock_lock_init(struct sock *sk)
2181 {
2182 	sk_owner_clear(sk);
2183 
2184 	if (sk->sk_kern_sock)
2185 		sock_lock_init_class_and_name(
2186 			sk,
2187 			af_family_kern_slock_key_strings[sk->sk_family],
2188 			af_family_kern_slock_keys + sk->sk_family,
2189 			af_family_kern_key_strings[sk->sk_family],
2190 			af_family_kern_keys + sk->sk_family);
2191 	else
2192 		sock_lock_init_class_and_name(
2193 			sk,
2194 			af_family_slock_key_strings[sk->sk_family],
2195 			af_family_slock_keys + sk->sk_family,
2196 			af_family_key_strings[sk->sk_family],
2197 			af_family_keys + sk->sk_family);
2198 }
2199 
2200 /*
2201  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2202  * even temporarily, because of RCU lookups. sk_node should also be left as is.
2203  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2204  */
2205 static void sock_copy(struct sock *nsk, const struct sock *osk)
2206 {
2207 	const struct proto *prot = READ_ONCE(osk->sk_prot);
2208 #ifdef CONFIG_SECURITY_NETWORK
2209 	void *sptr = nsk->sk_security;
2210 #endif
2211 
2212 	/* If we move sk_tx_queue_mapping out of the private section,
2213 	 * we must check if sk_tx_queue_clear() is called after
2214 	 * sock_copy() in sk_clone_lock().
2215 	 */
2216 	BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2217 		     offsetof(struct sock, sk_dontcopy_begin) ||
2218 		     offsetof(struct sock, sk_tx_queue_mapping) >=
2219 		     offsetof(struct sock, sk_dontcopy_end));
2220 
2221 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2222 
2223 	unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2224 		      prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2225 		      /* alloc is larger than struct, see sk_prot_alloc() */);
2226 
2227 #ifdef CONFIG_SECURITY_NETWORK
2228 	nsk->sk_security = sptr;
2229 	security_sk_clone(osk, nsk);
2230 #endif
2231 }
2232 
2233 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2234 		int family)
2235 {
2236 	struct sock *sk;
2237 	struct kmem_cache *slab;
2238 
2239 	slab = prot->slab;
2240 	if (slab != NULL) {
2241 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2242 		if (!sk)
2243 			return sk;
2244 		if (want_init_on_alloc(priority))
2245 			sk_prot_clear_nulls(sk, prot->obj_size);
2246 	} else
2247 		sk = kmalloc(prot->obj_size, priority);
2248 
2249 	if (sk != NULL) {
2250 		if (security_sk_alloc(sk, family, priority))
2251 			goto out_free;
2252 
2253 		if (!try_module_get(prot->owner))
2254 			goto out_free_sec;
2255 	}
2256 
2257 	return sk;
2258 
2259 out_free_sec:
2260 	security_sk_free(sk);
2261 out_free:
2262 	if (slab != NULL)
2263 		kmem_cache_free(slab, sk);
2264 	else
2265 		kfree(sk);
2266 	return NULL;
2267 }
2268 
2269 static void sk_prot_free(struct proto *prot, struct sock *sk)
2270 {
2271 	struct kmem_cache *slab;
2272 	struct module *owner;
2273 
2274 	owner = prot->owner;
2275 	slab = prot->slab;
2276 
2277 	cgroup_sk_free(&sk->sk_cgrp_data);
2278 	mem_cgroup_sk_free(sk);
2279 	security_sk_free(sk);
2280 
2281 	sk_owner_put(sk);
2282 
2283 	if (slab != NULL)
2284 		kmem_cache_free(slab, sk);
2285 	else
2286 		kfree(sk);
2287 	module_put(owner);
2288 }
2289 
2290 /**
2291  *	sk_alloc - All socket objects are allocated here
2292  *	@net: the applicable net namespace
2293  *	@family: protocol family
2294  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2295  *	@prot: struct proto associated with this new sock instance
2296  *	@kern: is this to be a kernel socket?
2297  */
2298 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2299 		      struct proto *prot, int kern)
2300 {
2301 	struct sock *sk;
2302 
2303 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2304 	if (sk) {
2305 		sk->sk_family = family;
2306 		/*
2307 		 * See comment in struct sock definition to understand
2308 		 * why we need sk_prot_creator -acme
2309 		 */
2310 		sk->sk_prot = sk->sk_prot_creator = prot;
2311 
2312 		if (READ_ONCE(net->core.sysctl_bypass_prot_mem))
2313 			sk->sk_bypass_prot_mem = 1;
2314 
2315 		sk->sk_kern_sock = kern;
2316 		sock_lock_init(sk);
2317 
2318 		sk->sk_net_refcnt = kern ? 0 : 1;
2319 		if (likely(sk->sk_net_refcnt)) {
2320 			get_net_track(net, &sk->ns_tracker, priority);
2321 			sock_inuse_add(net, 1);
2322 		} else {
2323 			net_passive_inc(net);
2324 			__netns_tracker_alloc(net, &sk->ns_tracker,
2325 					      false, priority);
2326 		}
2327 
2328 		sock_net_set(sk, net);
2329 		refcount_set(&sk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS);
2330 
2331 		mem_cgroup_sk_alloc(sk);
2332 		cgroup_sk_alloc(&sk->sk_cgrp_data);
2333 		sock_update_classid(&sk->sk_cgrp_data);
2334 		sock_update_netprioidx(&sk->sk_cgrp_data);
2335 		sk_tx_queue_clear(sk);
2336 	}
2337 
2338 	return sk;
2339 }
2340 EXPORT_SYMBOL(sk_alloc);
2341 
2342 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2343  * grace period. This is the case for UDP sockets and TCP listeners.
2344  */
2345 static void __sk_destruct(struct rcu_head *head)
2346 {
2347 	struct sock *sk = container_of(head, struct sock, sk_rcu);
2348 	struct net *net = sock_net(sk);
2349 	struct sk_filter *filter;
2350 
2351 	if (sk->sk_destruct)
2352 		sk->sk_destruct(sk);
2353 
2354 	filter = rcu_dereference_check(sk->sk_filter,
2355 				       refcount_read(&sk->sk_wmem_alloc) == 0);
2356 	if (filter) {
2357 		sk_filter_uncharge(sk, filter);
2358 		RCU_INIT_POINTER(sk->sk_filter, NULL);
2359 	}
2360 
2361 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2362 
2363 #ifdef CONFIG_BPF_SYSCALL
2364 	bpf_sk_storage_free(sk);
2365 #endif
2366 
2367 	if (atomic_read(&sk->sk_omem_alloc))
2368 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
2369 			 __func__, atomic_read(&sk->sk_omem_alloc));
2370 
2371 	if (sk->sk_frag.page) {
2372 		put_page(sk->sk_frag.page);
2373 		sk->sk_frag.page = NULL;
2374 	}
2375 
2376 	/* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2377 	put_cred(sk->sk_peer_cred);
2378 	put_pid(sk->sk_peer_pid);
2379 
2380 	if (likely(sk->sk_net_refcnt)) {
2381 		put_net_track(net, &sk->ns_tracker);
2382 	} else {
2383 		__netns_tracker_free(net, &sk->ns_tracker, false);
2384 		net_passive_dec(net);
2385 	}
2386 	sk_prot_free(sk->sk_prot_creator, sk);
2387 }
2388 
2389 void sk_net_refcnt_upgrade(struct sock *sk)
2390 {
2391 	struct net *net = sock_net(sk);
2392 
2393 	WARN_ON_ONCE(sk->sk_net_refcnt);
2394 	__netns_tracker_free(net, &sk->ns_tracker, false);
2395 	net_passive_dec(net);
2396 	sk->sk_net_refcnt = 1;
2397 	get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
2398 	sock_inuse_add(net, 1);
2399 }
2400 EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade);
2401 
2402 void sk_destruct(struct sock *sk)
2403 {
2404 	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2405 
2406 	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2407 		reuseport_detach_sock(sk);
2408 		use_call_rcu = true;
2409 	}
2410 
2411 	if (use_call_rcu)
2412 		call_rcu(&sk->sk_rcu, __sk_destruct);
2413 	else
2414 		__sk_destruct(&sk->sk_rcu);
2415 }
2416 
2417 static void __sk_free(struct sock *sk)
2418 {
2419 	if (likely(sk->sk_net_refcnt))
2420 		sock_inuse_add(sock_net(sk), -1);
2421 
2422 	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2423 		sock_diag_broadcast_destroy(sk);
2424 	else
2425 		sk_destruct(sk);
2426 }
2427 
2428 void sk_free(struct sock *sk)
2429 {
2430 	/*
2431 	 * We subtract one from sk_wmem_alloc and can know if
2432 	 * some packets are still in some tx queue.
2433 	 * If not null, sock_wfree() will call __sk_free(sk) later
2434 	 */
2435 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2436 		__sk_free(sk);
2437 }
2438 EXPORT_SYMBOL(sk_free);
2439 
2440 static void sk_init_common(struct sock *sk)
2441 {
2442 	skb_queue_head_init(&sk->sk_receive_queue);
2443 	skb_queue_head_init(&sk->sk_write_queue);
2444 	skb_queue_head_init(&sk->sk_error_queue);
2445 
2446 	rwlock_init(&sk->sk_callback_lock);
2447 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2448 			af_rlock_keys + sk->sk_family,
2449 			af_family_rlock_key_strings[sk->sk_family]);
2450 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2451 			af_wlock_keys + sk->sk_family,
2452 			af_family_wlock_key_strings[sk->sk_family]);
2453 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2454 			af_elock_keys + sk->sk_family,
2455 			af_family_elock_key_strings[sk->sk_family]);
2456 	if (sk->sk_kern_sock)
2457 		lockdep_set_class_and_name(&sk->sk_callback_lock,
2458 			af_kern_callback_keys + sk->sk_family,
2459 			af_family_kern_clock_key_strings[sk->sk_family]);
2460 	else
2461 		lockdep_set_class_and_name(&sk->sk_callback_lock,
2462 			af_callback_keys + sk->sk_family,
2463 			af_family_clock_key_strings[sk->sk_family]);
2464 }
2465 
2466 /**
2467  * sk_clone - clone a socket
2468  * @sk: the socket to clone
2469  * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2470  * @lock: if true, lock the cloned sk
2471  *
2472  * If @lock is true, the clone is locked by bh_lock_sock(), and
2473  * caller must unlock socket even in error path by bh_unlock_sock().
2474  */
2475 struct sock *sk_clone(const struct sock *sk, const gfp_t priority,
2476 		      bool lock)
2477 {
2478 	struct proto *prot = READ_ONCE(sk->sk_prot);
2479 	struct sk_filter *filter;
2480 	bool is_charged = true;
2481 	struct sock *newsk;
2482 
2483 	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2484 	if (!newsk)
2485 		goto out;
2486 
2487 	sock_copy(newsk, sk);
2488 
2489 	newsk->sk_prot_creator = prot;
2490 
2491 	/* SANITY */
2492 	if (likely(newsk->sk_net_refcnt)) {
2493 		get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2494 		sock_inuse_add(sock_net(newsk), 1);
2495 	} else {
2496 		/* Kernel sockets are not elevating the struct net refcount.
2497 		 * Instead, use a tracker to more easily detect if a layer
2498 		 * is not properly dismantling its kernel sockets at netns
2499 		 * destroy time.
2500 		 */
2501 		net_passive_inc(sock_net(newsk));
2502 		__netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2503 				      false, priority);
2504 	}
2505 
2506 	sk_node_init(&newsk->sk_node);
2507 	sock_lock_init(newsk);
2508 
2509 	if (lock)
2510 		bh_lock_sock(newsk);
2511 
2512 	newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
2513 	newsk->sk_backlog.len = 0;
2514 
2515 	atomic_set(&newsk->sk_rmem_alloc, 0);
2516 
2517 	refcount_set(&newsk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS);
2518 
2519 	atomic_set(&newsk->sk_omem_alloc, 0);
2520 	sk_init_common(newsk);
2521 
2522 	newsk->sk_dst_cache	= NULL;
2523 	newsk->sk_dst_pending_confirm = 0;
2524 	newsk->sk_wmem_queued	= 0;
2525 	newsk->sk_forward_alloc = 0;
2526 	newsk->sk_reserved_mem  = 0;
2527 	DEBUG_NET_WARN_ON_ONCE(newsk->sk_drop_counters);
2528 	sk_drops_reset(newsk);
2529 	newsk->sk_send_head	= NULL;
2530 	newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2531 	atomic_set(&newsk->sk_zckey, 0);
2532 
2533 	sock_reset_flag(newsk, SOCK_DONE);
2534 
2535 #ifdef CONFIG_MEMCG
2536 	/* sk->sk_memcg will be populated at accept() time */
2537 	newsk->sk_memcg = NULL;
2538 #endif
2539 
2540 	cgroup_sk_clone(&newsk->sk_cgrp_data);
2541 
2542 	rcu_read_lock();
2543 	filter = rcu_dereference(sk->sk_filter);
2544 	if (filter != NULL)
2545 		/* though it's an empty new sock, the charging may fail
2546 		 * if sysctl_optmem_max was changed between creation of
2547 		 * original socket and cloning
2548 		 */
2549 		is_charged = sk_filter_charge(newsk, filter);
2550 	RCU_INIT_POINTER(newsk->sk_filter, filter);
2551 	rcu_read_unlock();
2552 
2553 	if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2554 		/* We need to make sure that we don't uncharge the new
2555 		 * socket if we couldn't charge it in the first place
2556 		 * as otherwise we uncharge the parent's filter.
2557 		 */
2558 		if (!is_charged)
2559 			RCU_INIT_POINTER(newsk->sk_filter, NULL);
2560 
2561 		goto free;
2562 	}
2563 
2564 	RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2565 
2566 	if (bpf_sk_storage_clone(sk, newsk))
2567 		goto free;
2568 
2569 	/* Clear sk_user_data if parent had the pointer tagged
2570 	 * as not suitable for copying when cloning.
2571 	 */
2572 	if (sk_user_data_is_nocopy(newsk))
2573 		newsk->sk_user_data = NULL;
2574 
2575 	newsk->sk_err	   = 0;
2576 	newsk->sk_err_soft = 0;
2577 	newsk->sk_priority = 0;
2578 	newsk->sk_incoming_cpu = raw_smp_processor_id();
2579 
2580 	/* Before updating sk_refcnt, we must commit prior changes to memory
2581 	 * (Documentation/RCU/rculist_nulls.rst for details)
2582 	 */
2583 	smp_wmb();
2584 	refcount_set(&newsk->sk_refcnt, 2);
2585 
2586 	sk_set_socket(newsk, NULL);
2587 	sk_tx_queue_clear(newsk);
2588 	sk_rx_queue_clear(newsk);
2589 	RCU_INIT_POINTER(newsk->sk_wq, NULL);
2590 
2591 	if (newsk->sk_prot->sockets_allocated)
2592 		sk_sockets_allocated_inc(newsk);
2593 
2594 	if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2595 		net_enable_timestamp();
2596 out:
2597 	return newsk;
2598 free:
2599 	/* It is still raw copy of parent, so invalidate
2600 	 * destructor and make plain sk_free()
2601 	 */
2602 	newsk->sk_destruct = NULL;
2603 	if (lock)
2604 		bh_unlock_sock(newsk);
2605 	sk_free(newsk);
2606 	newsk = NULL;
2607 	goto out;
2608 }
2609 EXPORT_SYMBOL_GPL(sk_clone);
2610 
2611 static u32 sk_dst_gso_max_size(struct sock *sk, const struct net_device *dev)
2612 {
2613 	bool is_ipv6 = false;
2614 	u32 max_size;
2615 
2616 #if IS_ENABLED(CONFIG_IPV6)
2617 	is_ipv6 = (sk->sk_family == AF_INET6 &&
2618 		   !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2619 #endif
2620 	/* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2621 	max_size = is_ipv6 ? READ_ONCE(dev->gso_max_size) :
2622 			READ_ONCE(dev->gso_ipv4_max_size);
2623 	if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2624 		max_size = GSO_LEGACY_MAX_SIZE;
2625 
2626 	return max_size - (MAX_TCP_HEADER + 1);
2627 }
2628 
2629 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2630 {
2631 	const struct net_device *dev;
2632 	u32 max_segs = 1;
2633 
2634 	rcu_read_lock();
2635 	dev = dst_dev_rcu(dst);
2636 	sk->sk_route_caps = dev->features;
2637 	if (sk_is_tcp(sk)) {
2638 		struct inet_connection_sock *icsk = inet_csk(sk);
2639 
2640 		sk->sk_route_caps |= NETIF_F_GSO;
2641 		icsk->icsk_ack.dst_quick_ack = dst_metric(dst, RTAX_QUICKACK);
2642 	}
2643 	if (sk->sk_route_caps & NETIF_F_GSO)
2644 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2645 	if (unlikely(sk->sk_gso_disabled))
2646 		sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2647 	if (sk_can_gso(sk)) {
2648 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2649 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2650 		} else {
2651 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2652 			sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dev);
2653 			/* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2654 			max_segs = max_t(u32, READ_ONCE(dev->gso_max_segs), 1);
2655 		}
2656 	}
2657 	sk->sk_gso_max_segs = max_segs;
2658 	sk_dst_set(sk, dst);
2659 	rcu_read_unlock();
2660 }
2661 EXPORT_SYMBOL_GPL(sk_setup_caps);
2662 
2663 /*
2664  *	Simple resource managers for sockets.
2665  */
2666 
2667 
2668 /*
2669  * Write buffer destructor automatically called from kfree_skb.
2670  */
2671 void sock_wfree(struct sk_buff *skb)
2672 {
2673 	unsigned int len = skb->truesize;
2674 	struct sock *sk = skb->sk;
2675 	bool free;
2676 	int old;
2677 
2678 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2679 		void (*sk_write_space)(struct sock *sk);
2680 
2681 		sk_write_space = READ_ONCE(sk->sk_write_space);
2682 
2683 		if (sock_flag(sk, SOCK_RCU_FREE) &&
2684 		    sk_write_space == sock_def_write_space) {
2685 			rcu_read_lock();
2686 			free = __refcount_sub_and_test(len, &sk->sk_wmem_alloc,
2687 						       &old);
2688 			sock_def_write_space_wfree(sk, old - len);
2689 			rcu_read_unlock();
2690 			if (unlikely(free))
2691 				__sk_free(sk);
2692 			return;
2693 		}
2694 
2695 		/*
2696 		 * Keep a reference on sk_wmem_alloc, this will be released
2697 		 * after sk_write_space() call
2698 		 */
2699 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2700 		sk_write_space(sk);
2701 		len = 1;
2702 	}
2703 	/*
2704 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2705 	 * could not do because of in-flight packets
2706 	 */
2707 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2708 		__sk_free(sk);
2709 }
2710 EXPORT_SYMBOL(sock_wfree);
2711 
2712 /* This variant of sock_wfree() is used by TCP,
2713  * since it sets SOCK_USE_WRITE_QUEUE.
2714  */
2715 #ifdef CONFIG_INET
2716 void __sock_wfree(struct sk_buff *skb)
2717 {
2718 	struct sock *sk = skb->sk;
2719 
2720 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2721 		__sk_free(sk);
2722 }
2723 EXPORT_SYMBOL_GPL(__sock_wfree);
2724 #endif
2725 
2726 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2727 {
2728 	int old_wmem;
2729 
2730 	skb_orphan(skb);
2731 #ifdef CONFIG_INET
2732 	if (unlikely(!sk_fullsock(sk)))
2733 		return skb_set_owner_edemux(skb, sk);
2734 #endif
2735 	skb->sk = sk;
2736 	skb->destructor = sock_wfree;
2737 	skb_set_hash_from_sk(skb, sk);
2738 	/*
2739 	 * We used to take a refcount on sk, but following operation
2740 	 * is enough to guarantee sk_free() won't free this sock until
2741 	 * all in-flight packets are completed
2742 	 */
2743 	__refcount_add(skb->truesize, &sk->sk_wmem_alloc, &old_wmem);
2744 
2745 	/* (old_wmem == SK_WMEM_ALLOC_BIAS) if no other TX packet for this socket
2746 	 * is in a host queue (qdisc, NIC queue).
2747 	 * Set skb->ooo_okay so that netdev_pick_tx() can choose a TX queue
2748 	 * based on XPS for better performance.
2749 	 * Otherwise clear ooo_okay to not risk Out Of Order delivery.
2750 	 */
2751 	skb->ooo_okay = (old_wmem == SK_WMEM_ALLOC_BIAS);
2752 }
2753 EXPORT_SYMBOL(skb_set_owner_w);
2754 
2755 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2756 {
2757 	/* Drivers depend on in-order delivery for crypto offload,
2758 	 * partial orphan breaks out-of-order-OK logic.
2759 	 */
2760 	if (skb_is_decrypted(skb))
2761 		return false;
2762 
2763 	return (skb->destructor == sock_wfree ||
2764 		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2765 }
2766 
2767 /* This helper is used by netem, as it can hold packets in its
2768  * delay queue. We want to allow the owner socket to send more
2769  * packets, as if they were already TX completed by a typical driver.
2770  * But we also want to keep skb->sk set because some packet schedulers
2771  * rely on it (sch_fq for example).
2772  */
2773 void skb_orphan_partial(struct sk_buff *skb)
2774 {
2775 	if (skb_is_tcp_pure_ack(skb))
2776 		return;
2777 
2778 	if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2779 		return;
2780 
2781 	skb_orphan(skb);
2782 }
2783 EXPORT_SYMBOL(skb_orphan_partial);
2784 
2785 /*
2786  * Read buffer destructor automatically called from kfree_skb.
2787  */
2788 void sock_rfree(struct sk_buff *skb)
2789 {
2790 	struct sock *sk = skb->sk;
2791 	unsigned int len = skb->truesize;
2792 
2793 	atomic_sub(len, &sk->sk_rmem_alloc);
2794 	sk_mem_uncharge(sk, len);
2795 }
2796 EXPORT_SYMBOL(sock_rfree);
2797 
2798 /*
2799  * Buffer destructor for skbs that are not used directly in read or write
2800  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2801  */
2802 void sock_efree(struct sk_buff *skb)
2803 {
2804 	sock_put(skb->sk);
2805 }
2806 EXPORT_SYMBOL(sock_efree);
2807 
2808 /* Buffer destructor for prefetch/receive path where reference count may
2809  * not be held, e.g. for listen sockets.
2810  */
2811 #ifdef CONFIG_INET
2812 void sock_pfree(struct sk_buff *skb)
2813 {
2814 	struct sock *sk = skb->sk;
2815 
2816 	if (!sk_is_refcounted(sk))
2817 		return;
2818 
2819 	if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2820 		inet_reqsk(sk)->rsk_listener = NULL;
2821 		reqsk_free(inet_reqsk(sk));
2822 		return;
2823 	}
2824 
2825 	sock_gen_put(sk);
2826 }
2827 EXPORT_SYMBOL(sock_pfree);
2828 #endif /* CONFIG_INET */
2829 
2830 /*
2831  * Allocate a skb from the socket's send buffer.
2832  */
2833 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2834 			     gfp_t priority)
2835 {
2836 	if (force ||
2837 	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2838 		struct sk_buff *skb = alloc_skb(size, priority);
2839 
2840 		if (skb) {
2841 			skb_set_owner_w(skb, sk);
2842 			return skb;
2843 		}
2844 	}
2845 	return NULL;
2846 }
2847 EXPORT_SYMBOL(sock_wmalloc);
2848 
2849 static void sock_ofree(struct sk_buff *skb)
2850 {
2851 	struct sock *sk = skb->sk;
2852 
2853 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2854 }
2855 
2856 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2857 			     gfp_t priority)
2858 {
2859 	struct sk_buff *skb;
2860 
2861 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2862 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2863 	    READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2864 		return NULL;
2865 
2866 	skb = alloc_skb(size, priority);
2867 	if (!skb)
2868 		return NULL;
2869 
2870 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2871 	skb->sk = sk;
2872 	skb->destructor = sock_ofree;
2873 	return skb;
2874 }
2875 
2876 /*
2877  * Allocate a memory block from the socket's option memory buffer.
2878  */
2879 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2880 {
2881 	int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2882 
2883 	if ((unsigned int)size <= optmem_max &&
2884 	    atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2885 		void *mem;
2886 		/* First do the add, to avoid the race if kmalloc
2887 		 * might sleep.
2888 		 */
2889 		atomic_add(size, &sk->sk_omem_alloc);
2890 		mem = kmalloc(size, priority);
2891 		if (mem)
2892 			return mem;
2893 		atomic_sub(size, &sk->sk_omem_alloc);
2894 	}
2895 	return NULL;
2896 }
2897 EXPORT_SYMBOL(sock_kmalloc);
2898 
2899 /*
2900  * Duplicate the input "src" memory block using the socket's
2901  * option memory buffer.
2902  */
2903 void *sock_kmemdup(struct sock *sk, const void *src,
2904 		   int size, gfp_t priority)
2905 {
2906 	void *mem;
2907 
2908 	mem = sock_kmalloc(sk, size, priority);
2909 	if (mem)
2910 		memcpy(mem, src, size);
2911 	return mem;
2912 }
2913 EXPORT_SYMBOL(sock_kmemdup);
2914 
2915 /* Free an option memory block. Note, we actually want the inline
2916  * here as this allows gcc to detect the nullify and fold away the
2917  * condition entirely.
2918  */
2919 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2920 				  const bool nullify)
2921 {
2922 	if (WARN_ON_ONCE(!mem))
2923 		return;
2924 	if (nullify)
2925 		kfree_sensitive(mem);
2926 	else
2927 		kfree(mem);
2928 	atomic_sub(size, &sk->sk_omem_alloc);
2929 }
2930 
2931 void sock_kfree_s(struct sock *sk, void *mem, int size)
2932 {
2933 	__sock_kfree_s(sk, mem, size, false);
2934 }
2935 EXPORT_SYMBOL(sock_kfree_s);
2936 
2937 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2938 {
2939 	__sock_kfree_s(sk, mem, size, true);
2940 }
2941 EXPORT_SYMBOL(sock_kzfree_s);
2942 
2943 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2944    I think, these locks should be removed for datagram sockets.
2945  */
2946 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2947 {
2948 	DEFINE_WAIT(wait);
2949 
2950 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2951 	for (;;) {
2952 		if (!timeo)
2953 			break;
2954 		if (signal_pending(current))
2955 			break;
2956 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2957 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2958 		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2959 			break;
2960 		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2961 			break;
2962 		if (READ_ONCE(sk->sk_err))
2963 			break;
2964 		timeo = schedule_timeout(timeo);
2965 	}
2966 	finish_wait(sk_sleep(sk), &wait);
2967 	return timeo;
2968 }
2969 
2970 
2971 /*
2972  *	Generic send/receive buffer handlers
2973  */
2974 
2975 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2976 				     unsigned long data_len, int noblock,
2977 				     int *errcode, int max_page_order)
2978 {
2979 	struct sk_buff *skb;
2980 	long timeo;
2981 	int err;
2982 
2983 	timeo = sock_sndtimeo(sk, noblock);
2984 	for (;;) {
2985 		err = sock_error(sk);
2986 		if (err != 0)
2987 			goto failure;
2988 
2989 		err = -EPIPE;
2990 		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2991 			goto failure;
2992 
2993 		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2994 			break;
2995 
2996 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2997 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2998 		err = -EAGAIN;
2999 		if (!timeo)
3000 			goto failure;
3001 		if (signal_pending(current))
3002 			goto interrupted;
3003 		timeo = sock_wait_for_wmem(sk, timeo);
3004 	}
3005 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
3006 				   errcode, sk->sk_allocation);
3007 	if (skb)
3008 		skb_set_owner_w(skb, sk);
3009 	return skb;
3010 
3011 interrupted:
3012 	err = sock_intr_errno(timeo);
3013 failure:
3014 	*errcode = err;
3015 	return NULL;
3016 }
3017 EXPORT_SYMBOL(sock_alloc_send_pskb);
3018 
3019 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
3020 		     struct sockcm_cookie *sockc)
3021 {
3022 	u32 tsflags;
3023 
3024 	BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
3025 
3026 	switch (cmsg->cmsg_type) {
3027 	case SO_MARK:
3028 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
3029 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
3030 			return -EPERM;
3031 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3032 			return -EINVAL;
3033 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
3034 		break;
3035 	case SO_TIMESTAMPING_OLD:
3036 	case SO_TIMESTAMPING_NEW:
3037 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3038 			return -EINVAL;
3039 
3040 		tsflags = *(u32 *)CMSG_DATA(cmsg);
3041 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
3042 			return -EINVAL;
3043 
3044 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
3045 		sockc->tsflags |= tsflags;
3046 		break;
3047 	case SCM_TXTIME:
3048 		if (!sock_flag(sk, SOCK_TXTIME))
3049 			return -EINVAL;
3050 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
3051 			return -EINVAL;
3052 		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
3053 		break;
3054 	case SCM_TS_OPT_ID:
3055 		if (sk_is_tcp(sk))
3056 			return -EINVAL;
3057 		tsflags = READ_ONCE(sk->sk_tsflags);
3058 		if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
3059 			return -EINVAL;
3060 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3061 			return -EINVAL;
3062 		sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
3063 		sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
3064 		break;
3065 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
3066 	case SCM_RIGHTS:
3067 	case SCM_CREDENTIALS:
3068 		break;
3069 	case SO_PRIORITY:
3070 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3071 			return -EINVAL;
3072 		if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg)))
3073 			return -EPERM;
3074 		sockc->priority = *(u32 *)CMSG_DATA(cmsg);
3075 		break;
3076 	case SCM_DEVMEM_DMABUF:
3077 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3078 			return -EINVAL;
3079 		sockc->dmabuf_id = *(u32 *)CMSG_DATA(cmsg);
3080 		break;
3081 	default:
3082 		return -EINVAL;
3083 	}
3084 	return 0;
3085 }
3086 EXPORT_SYMBOL(__sock_cmsg_send);
3087 
3088 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
3089 		   struct sockcm_cookie *sockc)
3090 {
3091 	struct cmsghdr *cmsg;
3092 	int ret;
3093 
3094 	for_each_cmsghdr(cmsg, msg) {
3095 		if (!CMSG_OK(msg, cmsg))
3096 			return -EINVAL;
3097 		if (cmsg->cmsg_level != SOL_SOCKET)
3098 			continue;
3099 		ret = __sock_cmsg_send(sk, cmsg, sockc);
3100 		if (ret)
3101 			return ret;
3102 	}
3103 	return 0;
3104 }
3105 EXPORT_SYMBOL(sock_cmsg_send);
3106 
3107 static void sk_enter_memory_pressure(struct sock *sk)
3108 {
3109 	if (!sk->sk_prot->enter_memory_pressure)
3110 		return;
3111 
3112 	sk->sk_prot->enter_memory_pressure(sk);
3113 }
3114 
3115 static void sk_leave_memory_pressure(struct sock *sk)
3116 {
3117 	if (sk->sk_prot->leave_memory_pressure) {
3118 		INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
3119 				     tcp_leave_memory_pressure, sk);
3120 	} else {
3121 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
3122 
3123 		if (memory_pressure && READ_ONCE(*memory_pressure))
3124 			WRITE_ONCE(*memory_pressure, 0);
3125 	}
3126 }
3127 
3128 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3129 
3130 /**
3131  * skb_page_frag_refill - check that a page_frag contains enough room
3132  * @sz: minimum size of the fragment we want to get
3133  * @pfrag: pointer to page_frag
3134  * @gfp: priority for memory allocation
3135  *
3136  * Note: While this allocator tries to use high order pages, there is
3137  * no guarantee that allocations succeed. Therefore, @sz MUST be
3138  * less or equal than PAGE_SIZE.
3139  */
3140 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3141 {
3142 	if (pfrag->page) {
3143 		if (page_ref_count(pfrag->page) == 1) {
3144 			pfrag->offset = 0;
3145 			return true;
3146 		}
3147 		if (pfrag->offset + sz <= pfrag->size)
3148 			return true;
3149 		put_page(pfrag->page);
3150 	}
3151 
3152 	pfrag->offset = 0;
3153 	if (SKB_FRAG_PAGE_ORDER &&
3154 	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3155 		/* Avoid direct reclaim but allow kswapd to wake */
3156 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3157 					  __GFP_COMP | __GFP_NOWARN |
3158 					  __GFP_NORETRY,
3159 					  SKB_FRAG_PAGE_ORDER);
3160 		if (likely(pfrag->page)) {
3161 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3162 			return true;
3163 		}
3164 	}
3165 	pfrag->page = alloc_page(gfp);
3166 	if (likely(pfrag->page)) {
3167 		pfrag->size = PAGE_SIZE;
3168 		return true;
3169 	}
3170 	return false;
3171 }
3172 EXPORT_SYMBOL(skb_page_frag_refill);
3173 
3174 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3175 {
3176 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3177 		return true;
3178 
3179 	if (!sk->sk_bypass_prot_mem)
3180 		sk_enter_memory_pressure(sk);
3181 
3182 	sk_stream_moderate_sndbuf(sk);
3183 
3184 	return false;
3185 }
3186 EXPORT_SYMBOL(sk_page_frag_refill);
3187 
3188 static void __lock_sock(struct sock *sk)
3189 	__releases(&sk->sk_lock.slock)
3190 	__acquires(&sk->sk_lock.slock)
3191 {
3192 	DEFINE_WAIT(wait);
3193 
3194 	for (;;) {
3195 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3196 					TASK_UNINTERRUPTIBLE);
3197 		spin_unlock_bh(&sk->sk_lock.slock);
3198 		schedule();
3199 		spin_lock_bh(&sk->sk_lock.slock);
3200 		if (!sock_owned_by_user(sk))
3201 			break;
3202 	}
3203 	finish_wait(&sk->sk_lock.wq, &wait);
3204 }
3205 
3206 void __release_sock(struct sock *sk)
3207 	__releases(&sk->sk_lock.slock)
3208 	__acquires(&sk->sk_lock.slock)
3209 {
3210 	struct sk_buff *skb, *next;
3211 	int nb = 0;
3212 
3213 	while ((skb = sk->sk_backlog.head) != NULL) {
3214 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3215 
3216 		spin_unlock_bh(&sk->sk_lock.slock);
3217 
3218 		while (1) {
3219 			next = skb->next;
3220 			prefetch(next);
3221 			DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3222 			skb_mark_not_on_list(skb);
3223 			sk_backlog_rcv(sk, skb);
3224 
3225 			skb = next;
3226 			if (!skb)
3227 				break;
3228 
3229 			if (!(++nb & 15))
3230 				cond_resched();
3231 		}
3232 
3233 		spin_lock_bh(&sk->sk_lock.slock);
3234 	}
3235 
3236 	/*
3237 	 * Doing the zeroing here guarantee we can not loop forever
3238 	 * while a wild producer attempts to flood us.
3239 	 */
3240 	sk->sk_backlog.len = 0;
3241 }
3242 
3243 void __sk_flush_backlog(struct sock *sk)
3244 {
3245 	spin_lock_bh(&sk->sk_lock.slock);
3246 	__release_sock(sk);
3247 
3248 	if (sk->sk_prot->release_cb)
3249 		INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3250 				     tcp_release_cb, sk);
3251 
3252 	spin_unlock_bh(&sk->sk_lock.slock);
3253 }
3254 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3255 
3256 /**
3257  * sk_wait_data - wait for data to arrive at sk_receive_queue
3258  * @sk:    sock to wait on
3259  * @timeo: for how long
3260  * @skb:   last skb seen on sk_receive_queue
3261  *
3262  * Now socket state including sk->sk_err is changed only under lock,
3263  * hence we may omit checks after joining wait queue.
3264  * We check receive queue before schedule() only as optimization;
3265  * it is very likely that release_sock() added new data.
3266  */
3267 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3268 {
3269 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
3270 	int rc;
3271 
3272 	add_wait_queue(sk_sleep(sk), &wait);
3273 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3274 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3275 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3276 	remove_wait_queue(sk_sleep(sk), &wait);
3277 	return rc;
3278 }
3279 EXPORT_SYMBOL(sk_wait_data);
3280 
3281 /**
3282  *	__sk_mem_raise_allocated - increase memory_allocated
3283  *	@sk: socket
3284  *	@size: memory size to allocate
3285  *	@amt: pages to allocate
3286  *	@kind: allocation type
3287  *
3288  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3289  *
3290  *	Unlike the globally shared limits among the sockets under same protocol,
3291  *	consuming the budget of a memcg won't have direct effect on other ones.
3292  *	So be optimistic about memcg's tolerance, and leave the callers to decide
3293  *	whether or not to raise allocated through sk_under_memory_pressure() or
3294  *	its variants.
3295  */
3296 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3297 {
3298 	bool memcg_enabled = false, charged = false;
3299 	struct proto *prot = sk->sk_prot;
3300 	long allocated = 0;
3301 
3302 	if (!sk->sk_bypass_prot_mem) {
3303 		sk_memory_allocated_add(sk, amt);
3304 		allocated = sk_memory_allocated(sk);
3305 	}
3306 
3307 	if (mem_cgroup_sk_enabled(sk)) {
3308 		memcg_enabled = true;
3309 		charged = mem_cgroup_sk_charge(sk, amt, gfp_memcg_charge());
3310 		if (!charged)
3311 			goto suppress_allocation;
3312 	}
3313 
3314 	if (!allocated)
3315 		return 1;
3316 
3317 	/* Under limit. */
3318 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
3319 		sk_leave_memory_pressure(sk);
3320 		return 1;
3321 	}
3322 
3323 	/* Under pressure. */
3324 	if (allocated > sk_prot_mem_limits(sk, 1))
3325 		sk_enter_memory_pressure(sk);
3326 
3327 	/* Over hard limit. */
3328 	if (allocated > sk_prot_mem_limits(sk, 2))
3329 		goto suppress_allocation;
3330 
3331 	/* Guarantee minimum buffer size under pressure (either global
3332 	 * or memcg) to make sure features described in RFC 7323 (TCP
3333 	 * Extensions for High Performance) work properly.
3334 	 *
3335 	 * This rule does NOT stand when exceeds global or memcg's hard
3336 	 * limit, or else a DoS attack can be taken place by spawning
3337 	 * lots of sockets whose usage are under minimum buffer size.
3338 	 */
3339 	if (kind == SK_MEM_RECV) {
3340 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3341 			return 1;
3342 
3343 	} else { /* SK_MEM_SEND */
3344 		int wmem0 = sk_get_wmem0(sk, prot);
3345 
3346 		if (sk->sk_type == SOCK_STREAM) {
3347 			if (sk->sk_wmem_queued < wmem0)
3348 				return 1;
3349 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3350 				return 1;
3351 		}
3352 	}
3353 
3354 	if (sk_has_memory_pressure(sk)) {
3355 		u64 alloc;
3356 
3357 		/* The following 'average' heuristic is within the
3358 		 * scope of global accounting, so it only makes
3359 		 * sense for global memory pressure.
3360 		 */
3361 		if (!sk_under_global_memory_pressure(sk))
3362 			return 1;
3363 
3364 		/* Try to be fair among all the sockets under global
3365 		 * pressure by allowing the ones that below average
3366 		 * usage to raise.
3367 		 */
3368 		alloc = sk_sockets_allocated_read_positive(sk);
3369 		if (sk_prot_mem_limits(sk, 2) > alloc *
3370 		    sk_mem_pages(sk->sk_wmem_queued +
3371 				 atomic_read(&sk->sk_rmem_alloc) +
3372 				 sk->sk_forward_alloc))
3373 			return 1;
3374 	}
3375 
3376 suppress_allocation:
3377 
3378 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3379 		sk_stream_moderate_sndbuf(sk);
3380 
3381 		/* Fail only if socket is _under_ its sndbuf.
3382 		 * In this case we cannot block, so that we have to fail.
3383 		 */
3384 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3385 			/* Force charge with __GFP_NOFAIL */
3386 			if (memcg_enabled && !charged)
3387 				mem_cgroup_sk_charge(sk, amt,
3388 						     gfp_memcg_charge() | __GFP_NOFAIL);
3389 			return 1;
3390 		}
3391 	}
3392 
3393 	trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3394 
3395 	if (allocated)
3396 		sk_memory_allocated_sub(sk, amt);
3397 
3398 	if (charged)
3399 		mem_cgroup_sk_uncharge(sk, amt);
3400 
3401 	return 0;
3402 }
3403 
3404 /**
3405  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3406  *	@sk: socket
3407  *	@size: memory size to allocate
3408  *	@kind: allocation type
3409  *
3410  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3411  *	rmem allocation. This function assumes that protocols which have
3412  *	memory_pressure use sk_wmem_queued as write buffer accounting.
3413  */
3414 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3415 {
3416 	int ret, amt = sk_mem_pages(size);
3417 
3418 	sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3419 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3420 	if (!ret)
3421 		sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3422 	return ret;
3423 }
3424 EXPORT_SYMBOL(__sk_mem_schedule);
3425 
3426 /**
3427  *	__sk_mem_reduce_allocated - reclaim memory_allocated
3428  *	@sk: socket
3429  *	@amount: number of quanta
3430  *
3431  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3432  */
3433 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3434 {
3435 	if (mem_cgroup_sk_enabled(sk))
3436 		mem_cgroup_sk_uncharge(sk, amount);
3437 
3438 	if (sk->sk_bypass_prot_mem)
3439 		return;
3440 
3441 	sk_memory_allocated_sub(sk, amount);
3442 
3443 	if (sk_under_global_memory_pressure(sk) &&
3444 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3445 		sk_leave_memory_pressure(sk);
3446 }
3447 
3448 /**
3449  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3450  *	@sk: socket
3451  *	@amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3452  */
3453 void __sk_mem_reclaim(struct sock *sk, int amount)
3454 {
3455 	amount >>= PAGE_SHIFT;
3456 	sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3457 	__sk_mem_reduce_allocated(sk, amount);
3458 }
3459 EXPORT_SYMBOL(__sk_mem_reclaim);
3460 
3461 void __sk_charge(struct sock *sk, gfp_t gfp)
3462 {
3463 	int amt;
3464 
3465 	gfp |= __GFP_NOFAIL;
3466 	if (mem_cgroup_from_sk(sk)) {
3467 		/* The socket has not been accepted yet, no need
3468 		 * to look at newsk->sk_wmem_queued.
3469 		 */
3470 		amt = sk_mem_pages(sk->sk_forward_alloc +
3471 				   atomic_read(&sk->sk_rmem_alloc));
3472 		if (amt)
3473 			mem_cgroup_sk_charge(sk, amt, gfp);
3474 	}
3475 
3476 	kmem_cache_charge(sk, gfp);
3477 }
3478 
3479 int sk_set_peek_off(struct sock *sk, int val)
3480 {
3481 	WRITE_ONCE(sk->sk_peek_off, val);
3482 	return 0;
3483 }
3484 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3485 
3486 /*
3487  * Set of default routines for initialising struct proto_ops when
3488  * the protocol does not support a particular function. In certain
3489  * cases where it makes no sense for a protocol to have a "do nothing"
3490  * function, some default processing is provided.
3491  */
3492 
3493 int sock_no_bind(struct socket *sock, struct sockaddr_unsized *saddr, int len)
3494 {
3495 	return -EOPNOTSUPP;
3496 }
3497 EXPORT_SYMBOL(sock_no_bind);
3498 
3499 int sock_no_connect(struct socket *sock, struct sockaddr_unsized *saddr,
3500 		    int len, int flags)
3501 {
3502 	return -EOPNOTSUPP;
3503 }
3504 EXPORT_SYMBOL(sock_no_connect);
3505 
3506 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3507 {
3508 	return -EOPNOTSUPP;
3509 }
3510 EXPORT_SYMBOL(sock_no_socketpair);
3511 
3512 int sock_no_accept(struct socket *sock, struct socket *newsock,
3513 		   struct proto_accept_arg *arg)
3514 {
3515 	return -EOPNOTSUPP;
3516 }
3517 EXPORT_SYMBOL(sock_no_accept);
3518 
3519 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3520 		    int peer)
3521 {
3522 	return -EOPNOTSUPP;
3523 }
3524 EXPORT_SYMBOL(sock_no_getname);
3525 
3526 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3527 {
3528 	return -EOPNOTSUPP;
3529 }
3530 EXPORT_SYMBOL(sock_no_ioctl);
3531 
3532 int sock_no_listen(struct socket *sock, int backlog)
3533 {
3534 	return -EOPNOTSUPP;
3535 }
3536 EXPORT_SYMBOL(sock_no_listen);
3537 
3538 int sock_no_shutdown(struct socket *sock, int how)
3539 {
3540 	return -EOPNOTSUPP;
3541 }
3542 EXPORT_SYMBOL(sock_no_shutdown);
3543 
3544 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3545 {
3546 	return -EOPNOTSUPP;
3547 }
3548 EXPORT_SYMBOL(sock_no_sendmsg);
3549 
3550 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3551 {
3552 	return -EOPNOTSUPP;
3553 }
3554 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3555 
3556 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3557 		    int flags)
3558 {
3559 	return -EOPNOTSUPP;
3560 }
3561 EXPORT_SYMBOL(sock_no_recvmsg);
3562 
3563 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3564 {
3565 	/* Mirror missing mmap method error code */
3566 	return -ENODEV;
3567 }
3568 EXPORT_SYMBOL(sock_no_mmap);
3569 
3570 /*
3571  * When a file is received (via SCM_RIGHTS, etc), we must bump the
3572  * various sock-based usage counts.
3573  */
3574 void __receive_sock(struct file *file)
3575 {
3576 	struct socket *sock;
3577 
3578 	sock = sock_from_file(file);
3579 	if (sock) {
3580 		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3581 		sock_update_classid(&sock->sk->sk_cgrp_data);
3582 	}
3583 }
3584 
3585 /*
3586  *	Default Socket Callbacks
3587  */
3588 
3589 static void sock_def_wakeup(struct sock *sk)
3590 {
3591 	struct socket_wq *wq;
3592 
3593 	rcu_read_lock();
3594 	wq = rcu_dereference(sk->sk_wq);
3595 	if (skwq_has_sleeper(wq))
3596 		wake_up_interruptible_all(&wq->wait);
3597 	rcu_read_unlock();
3598 }
3599 
3600 static void sock_def_error_report(struct sock *sk)
3601 {
3602 	struct socket_wq *wq;
3603 
3604 	rcu_read_lock();
3605 	wq = rcu_dereference(sk->sk_wq);
3606 	if (skwq_has_sleeper(wq))
3607 		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3608 	sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3609 	rcu_read_unlock();
3610 }
3611 
3612 void sock_def_readable(struct sock *sk)
3613 {
3614 	struct socket_wq *wq;
3615 
3616 	trace_sk_data_ready(sk);
3617 
3618 	rcu_read_lock();
3619 	wq = rcu_dereference(sk->sk_wq);
3620 	if (skwq_has_sleeper(wq))
3621 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3622 						EPOLLRDNORM | EPOLLRDBAND);
3623 	sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3624 	rcu_read_unlock();
3625 }
3626 
3627 static void sock_def_write_space(struct sock *sk)
3628 {
3629 	struct socket_wq *wq;
3630 
3631 	rcu_read_lock();
3632 
3633 	/* Do not wake up a writer until he can make "significant"
3634 	 * progress.  --DaveM
3635 	 */
3636 	if (sock_writeable(sk)) {
3637 		wq = rcu_dereference(sk->sk_wq);
3638 		if (skwq_has_sleeper(wq))
3639 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3640 						EPOLLWRNORM | EPOLLWRBAND);
3641 
3642 		/* Should agree with poll, otherwise some programs break */
3643 		sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3644 	}
3645 
3646 	rcu_read_unlock();
3647 }
3648 
3649 /* An optimised version of sock_def_write_space(), should only be called
3650  * for SOCK_RCU_FREE sockets under RCU read section and after putting
3651  * ->sk_wmem_alloc.
3652  */
3653 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc)
3654 {
3655 	/* Do not wake up a writer until he can make "significant"
3656 	 * progress.  --DaveM
3657 	 */
3658 	if (__sock_writeable(sk, wmem_alloc)) {
3659 		struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3660 
3661 		/* rely on refcount_sub from sock_wfree() */
3662 		smp_mb__after_atomic();
3663 		if (wq && waitqueue_active(&wq->wait))
3664 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3665 						EPOLLWRNORM | EPOLLWRBAND);
3666 
3667 		/* Should agree with poll, otherwise some programs break */
3668 		sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3669 	}
3670 }
3671 
3672 static void sock_def_destruct(struct sock *sk)
3673 {
3674 }
3675 
3676 void sk_send_sigurg(struct sock *sk)
3677 {
3678 	if (sk->sk_socket && sk->sk_socket->file)
3679 		if (send_sigurg(sk->sk_socket->file))
3680 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3681 }
3682 EXPORT_SYMBOL(sk_send_sigurg);
3683 
3684 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3685 		    unsigned long expires)
3686 {
3687 	if (!mod_timer(timer, expires))
3688 		sock_hold(sk);
3689 }
3690 EXPORT_SYMBOL(sk_reset_timer);
3691 
3692 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3693 {
3694 	if (timer_delete(timer))
3695 		__sock_put(sk);
3696 }
3697 EXPORT_SYMBOL(sk_stop_timer);
3698 
3699 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3700 {
3701 	if (timer_delete_sync(timer))
3702 		__sock_put(sk);
3703 }
3704 EXPORT_SYMBOL(sk_stop_timer_sync);
3705 
3706 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3707 {
3708 	sk_init_common(sk);
3709 	sk->sk_send_head	=	NULL;
3710 
3711 	timer_setup(&sk->sk_timer, NULL, 0);
3712 
3713 	sk->sk_allocation	=	GFP_KERNEL;
3714 	sk->sk_rcvbuf		=	READ_ONCE(sysctl_rmem_default);
3715 	sk->sk_sndbuf		=	READ_ONCE(sysctl_wmem_default);
3716 	sk->sk_state		=	TCP_CLOSE;
3717 	sk->sk_use_task_frag	=	true;
3718 	sk_set_socket(sk, sock);
3719 
3720 	sock_set_flag(sk, SOCK_ZAPPED);
3721 
3722 	if (sock) {
3723 		sk->sk_type	=	sock->type;
3724 		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3725 		sock->sk	=	sk;
3726 	} else {
3727 		RCU_INIT_POINTER(sk->sk_wq, NULL);
3728 	}
3729 	sk->sk_uid	=	uid;
3730 
3731 	sk->sk_state_change	=	sock_def_wakeup;
3732 	sk->sk_data_ready	=	sock_def_readable;
3733 	sk->sk_write_space	=	sock_def_write_space;
3734 	sk->sk_error_report	=	sock_def_error_report;
3735 	sk->sk_destruct		=	sock_def_destruct;
3736 
3737 	sk->sk_frag.page	=	NULL;
3738 	sk->sk_frag.offset	=	0;
3739 	sk->sk_peek_off		=	-1;
3740 
3741 	sk->sk_peer_pid 	=	NULL;
3742 	sk->sk_peer_cred	=	NULL;
3743 	spin_lock_init(&sk->sk_peer_lock);
3744 
3745 	sk->sk_write_pending	=	0;
3746 	sk->sk_rcvlowat		=	1;
3747 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3748 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3749 
3750 	sk->sk_stamp = SK_DEFAULT_STAMP;
3751 #if BITS_PER_LONG==32
3752 	seqlock_init(&sk->sk_stamp_seq);
3753 #endif
3754 	atomic_set(&sk->sk_zckey, 0);
3755 
3756 #ifdef CONFIG_NET_RX_BUSY_POLL
3757 	sk->sk_napi_id		=	0;
3758 	sk->sk_ll_usec		=	READ_ONCE(sysctl_net_busy_read);
3759 #endif
3760 
3761 	sk->sk_max_pacing_rate = ~0UL;
3762 	sk->sk_pacing_rate = ~0UL;
3763 	WRITE_ONCE(sk->sk_pacing_shift, 10);
3764 	sk->sk_incoming_cpu = -1;
3765 
3766 	sk_rx_queue_clear(sk);
3767 	/*
3768 	 * Before updating sk_refcnt, we must commit prior changes to memory
3769 	 * (Documentation/RCU/rculist_nulls.rst for details)
3770 	 */
3771 	smp_wmb();
3772 	refcount_set(&sk->sk_refcnt, 1);
3773 	sk_drops_reset(sk);
3774 }
3775 EXPORT_SYMBOL(sock_init_data_uid);
3776 
3777 void sock_init_data(struct socket *sock, struct sock *sk)
3778 {
3779 	kuid_t uid = sock ?
3780 		SOCK_INODE(sock)->i_uid :
3781 		make_kuid(sock_net(sk)->user_ns, 0);
3782 
3783 	sock_init_data_uid(sock, sk, uid);
3784 }
3785 EXPORT_SYMBOL(sock_init_data);
3786 
3787 void noinline lock_sock_nested(struct sock *sk, int subclass)
3788 {
3789 	/* The sk_lock has mutex_lock() semantics here. */
3790 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3791 
3792 	might_sleep();
3793 #ifdef CONFIG_64BIT
3794 	if (sizeof(struct slock_owned) == sizeof(long)) {
3795 		socket_lock_t tmp = {
3796 			.slock = __SPIN_LOCK_UNLOCKED(tmp.slock),
3797 			.owned = 1,
3798 		};
3799 		socket_lock_t old = {
3800 			.slock = __SPIN_LOCK_UNLOCKED(old.slock),
3801 			.owned = 0,
3802 		};
3803 
3804 		if (likely(try_cmpxchg(&sk->sk_lock.combined,
3805 				       &old.combined, tmp.combined)))
3806 			return;
3807 	}
3808 #endif
3809 	spin_lock_bh(&sk->sk_lock.slock);
3810 	if (unlikely(sock_owned_by_user_nocheck(sk)))
3811 		__lock_sock(sk);
3812 	sk->sk_lock.owned = 1;
3813 	spin_unlock_bh(&sk->sk_lock.slock);
3814 }
3815 EXPORT_SYMBOL(lock_sock_nested);
3816 
3817 void release_sock(struct sock *sk)
3818 {
3819 	spin_lock_bh(&sk->sk_lock.slock);
3820 
3821 	if (unlikely(sk->sk_backlog.tail))
3822 		__release_sock(sk);
3823 
3824 	if (sk->sk_prot->release_cb) {
3825 		if (!tcp_release_cb_cond(sk))
3826 			sk->sk_prot->release_cb(sk);
3827 	}
3828 	sock_release_ownership(sk);
3829 	if (unlikely(waitqueue_active(&sk->sk_lock.wq)))
3830 		wake_up(&sk->sk_lock.wq);
3831 
3832 	spin_unlock_bh(&sk->sk_lock.slock);
3833 }
3834 EXPORT_SYMBOL(release_sock);
3835 
3836 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3837 {
3838 	might_sleep();
3839 	spin_lock_bh(&sk->sk_lock.slock);
3840 
3841 	if (likely(!sock_owned_by_user_nocheck(sk))) {
3842 		/*
3843 		 * Fast path return with bottom halves disabled and
3844 		 * sock::sk_lock.slock held.
3845 		 *
3846 		 * The 'mutex' is not contended and holding
3847 		 * sock::sk_lock.slock prevents all other lockers to
3848 		 * proceed so the corresponding unlock_sock_fast() can
3849 		 * avoid the slow path of release_sock() completely and
3850 		 * just release slock.
3851 		 *
3852 		 * From a semantical POV this is equivalent to 'acquiring'
3853 		 * the 'mutex', hence the corresponding lockdep
3854 		 * mutex_release() has to happen in the fast path of
3855 		 * unlock_sock_fast().
3856 		 */
3857 		return false;
3858 	}
3859 
3860 	__lock_sock(sk);
3861 	sk->sk_lock.owned = 1;
3862 	__acquire(&sk->sk_lock.slock);
3863 	spin_unlock_bh(&sk->sk_lock.slock);
3864 	return true;
3865 }
3866 EXPORT_SYMBOL(__lock_sock_fast);
3867 
3868 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3869 		   bool timeval, bool time32)
3870 {
3871 	struct sock *sk = sock->sk;
3872 	struct timespec64 ts;
3873 
3874 	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3875 	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3876 	if (ts.tv_sec == -1)
3877 		return -ENOENT;
3878 	if (ts.tv_sec == 0) {
3879 		ktime_t kt = ktime_get_real();
3880 		sock_write_timestamp(sk, kt);
3881 		ts = ktime_to_timespec64(kt);
3882 	}
3883 
3884 	if (timeval)
3885 		ts.tv_nsec /= 1000;
3886 
3887 #ifdef CONFIG_COMPAT_32BIT_TIME
3888 	if (time32)
3889 		return put_old_timespec32(&ts, userstamp);
3890 #endif
3891 #ifdef CONFIG_SPARC64
3892 	/* beware of padding in sparc64 timeval */
3893 	if (timeval && !in_compat_syscall()) {
3894 		struct __kernel_old_timeval __user tv = {
3895 			.tv_sec = ts.tv_sec,
3896 			.tv_usec = ts.tv_nsec,
3897 		};
3898 		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3899 			return -EFAULT;
3900 		return 0;
3901 	}
3902 #endif
3903 	return put_timespec64(&ts, userstamp);
3904 }
3905 EXPORT_SYMBOL(sock_gettstamp);
3906 
3907 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3908 {
3909 	if (!sock_flag(sk, flag)) {
3910 		unsigned long previous_flags = sk->sk_flags;
3911 
3912 		sock_set_flag(sk, flag);
3913 		/*
3914 		 * we just set one of the two flags which require net
3915 		 * time stamping, but time stamping might have been on
3916 		 * already because of the other one
3917 		 */
3918 		if (sock_needs_netstamp(sk) &&
3919 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3920 			net_enable_timestamp();
3921 	}
3922 }
3923 
3924 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3925 		       int level, int type)
3926 {
3927 	struct sock_extended_err ee;
3928 	struct sk_buff *skb;
3929 	int copied, err;
3930 
3931 	err = -EAGAIN;
3932 	skb = sock_dequeue_err_skb(sk);
3933 	if (skb == NULL)
3934 		goto out;
3935 
3936 	copied = skb->len;
3937 	if (copied > len) {
3938 		msg->msg_flags |= MSG_TRUNC;
3939 		copied = len;
3940 	}
3941 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3942 	if (err)
3943 		goto out_free_skb;
3944 
3945 	sock_recv_timestamp(msg, sk, skb);
3946 
3947 	/* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
3948 	ee = SKB_EXT_ERR(skb)->ee;
3949 	put_cmsg(msg, level, type, sizeof(ee), &ee);
3950 
3951 	msg->msg_flags |= MSG_ERRQUEUE;
3952 	err = copied;
3953 
3954 out_free_skb:
3955 	kfree_skb(skb);
3956 out:
3957 	return err;
3958 }
3959 EXPORT_SYMBOL(sock_recv_errqueue);
3960 
3961 /*
3962  *	Get a socket option on an socket.
3963  *
3964  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3965  *	asynchronous errors should be reported by getsockopt. We assume
3966  *	this means if you specify SO_ERROR (otherwise what is the point of it).
3967  */
3968 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3969 			   char __user *optval, int __user *optlen)
3970 {
3971 	struct sock *sk = sock->sk;
3972 
3973 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
3974 	return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3975 }
3976 EXPORT_SYMBOL(sock_common_getsockopt);
3977 
3978 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3979 			int flags)
3980 {
3981 	struct sock *sk = sock->sk;
3982 
3983 	return sk->sk_prot->recvmsg(sk, msg, size, flags);
3984 }
3985 EXPORT_SYMBOL(sock_common_recvmsg);
3986 
3987 /*
3988  *	Set socket options on an inet socket.
3989  */
3990 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3991 			   sockptr_t optval, unsigned int optlen)
3992 {
3993 	struct sock *sk = sock->sk;
3994 
3995 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
3996 	return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3997 }
3998 EXPORT_SYMBOL(sock_common_setsockopt);
3999 
4000 void sk_common_release(struct sock *sk)
4001 {
4002 	if (sk->sk_prot->destroy)
4003 		sk->sk_prot->destroy(sk);
4004 
4005 	/*
4006 	 * Observation: when sk_common_release is called, processes have
4007 	 * no access to socket. But net still has.
4008 	 * Step one, detach it from networking:
4009 	 *
4010 	 * A. Remove from hash tables.
4011 	 */
4012 
4013 	sk->sk_prot->unhash(sk);
4014 
4015 	/*
4016 	 * In this point socket cannot receive new packets, but it is possible
4017 	 * that some packets are in flight because some CPU runs receiver and
4018 	 * did hash table lookup before we unhashed socket. They will achieve
4019 	 * receive queue and will be purged by socket destructor.
4020 	 *
4021 	 * Also we still have packets pending on receive queue and probably,
4022 	 * our own packets waiting in device queues. sock_destroy will drain
4023 	 * receive queue, but transmitted packets will delay socket destruction
4024 	 * until the last reference will be released.
4025 	 */
4026 
4027 	sock_orphan(sk);
4028 
4029 	xfrm_sk_free_policy(sk);
4030 
4031 	sock_put(sk);
4032 }
4033 EXPORT_SYMBOL(sk_common_release);
4034 
4035 void sk_get_meminfo(const struct sock *sk, u32 *mem)
4036 {
4037 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
4038 
4039 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
4040 	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
4041 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
4042 	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
4043 	mem[SK_MEMINFO_FWD_ALLOC] = READ_ONCE(sk->sk_forward_alloc);
4044 	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
4045 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
4046 	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
4047 	mem[SK_MEMINFO_DROPS] = sk_drops_read(sk);
4048 }
4049 
4050 #ifdef CONFIG_PROC_FS
4051 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
4052 
4053 int sock_prot_inuse_get(struct net *net, struct proto *prot)
4054 {
4055 	int cpu, idx = prot->inuse_idx;
4056 	int res = 0;
4057 
4058 	for_each_possible_cpu(cpu)
4059 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
4060 
4061 	return res >= 0 ? res : 0;
4062 }
4063 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
4064 
4065 int sock_inuse_get(struct net *net)
4066 {
4067 	int cpu, res = 0;
4068 
4069 	for_each_possible_cpu(cpu)
4070 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
4071 
4072 	return res;
4073 }
4074 
4075 EXPORT_SYMBOL_GPL(sock_inuse_get);
4076 
4077 static int __net_init sock_inuse_init_net(struct net *net)
4078 {
4079 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
4080 	if (net->core.prot_inuse == NULL)
4081 		return -ENOMEM;
4082 	return 0;
4083 }
4084 
4085 static void __net_exit sock_inuse_exit_net(struct net *net)
4086 {
4087 	free_percpu(net->core.prot_inuse);
4088 }
4089 
4090 static struct pernet_operations net_inuse_ops = {
4091 	.init = sock_inuse_init_net,
4092 	.exit = sock_inuse_exit_net,
4093 };
4094 
4095 static __init int net_inuse_init(void)
4096 {
4097 	if (register_pernet_subsys(&net_inuse_ops))
4098 		panic("Cannot initialize net inuse counters");
4099 
4100 	return 0;
4101 }
4102 
4103 core_initcall(net_inuse_init);
4104 
4105 static int assign_proto_idx(struct proto *prot)
4106 {
4107 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
4108 
4109 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR)) {
4110 		pr_err("PROTO_INUSE_NR exhausted\n");
4111 		return -ENOSPC;
4112 	}
4113 
4114 	set_bit(prot->inuse_idx, proto_inuse_idx);
4115 	return 0;
4116 }
4117 
4118 static void release_proto_idx(struct proto *prot)
4119 {
4120 	if (prot->inuse_idx != PROTO_INUSE_NR)
4121 		clear_bit(prot->inuse_idx, proto_inuse_idx);
4122 }
4123 #else
4124 static inline int assign_proto_idx(struct proto *prot)
4125 {
4126 	return 0;
4127 }
4128 
4129 static inline void release_proto_idx(struct proto *prot)
4130 {
4131 }
4132 
4133 #endif
4134 
4135 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
4136 {
4137 	if (!twsk_prot)
4138 		return;
4139 	kfree(twsk_prot->twsk_slab_name);
4140 	twsk_prot->twsk_slab_name = NULL;
4141 	kmem_cache_destroy(twsk_prot->twsk_slab);
4142 	twsk_prot->twsk_slab = NULL;
4143 }
4144 
4145 static int tw_prot_init(const struct proto *prot)
4146 {
4147 	struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
4148 
4149 	if (!twsk_prot)
4150 		return 0;
4151 
4152 	twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
4153 					      prot->name);
4154 	if (!twsk_prot->twsk_slab_name)
4155 		return -ENOMEM;
4156 
4157 	twsk_prot->twsk_slab =
4158 		kmem_cache_create(twsk_prot->twsk_slab_name,
4159 				  twsk_prot->twsk_obj_size, 0,
4160 				  SLAB_ACCOUNT | prot->slab_flags,
4161 				  NULL);
4162 	if (!twsk_prot->twsk_slab) {
4163 		pr_crit("%s: Can't create timewait sock SLAB cache!\n",
4164 			prot->name);
4165 		return -ENOMEM;
4166 	}
4167 
4168 	return 0;
4169 }
4170 
4171 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
4172 {
4173 	if (!rsk_prot)
4174 		return;
4175 	kfree(rsk_prot->slab_name);
4176 	rsk_prot->slab_name = NULL;
4177 	kmem_cache_destroy(rsk_prot->slab);
4178 	rsk_prot->slab = NULL;
4179 }
4180 
4181 static int req_prot_init(const struct proto *prot)
4182 {
4183 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
4184 
4185 	if (!rsk_prot)
4186 		return 0;
4187 
4188 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4189 					prot->name);
4190 	if (!rsk_prot->slab_name)
4191 		return -ENOMEM;
4192 
4193 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4194 					   rsk_prot->obj_size, 0,
4195 					   SLAB_ACCOUNT | prot->slab_flags,
4196 					   NULL);
4197 
4198 	if (!rsk_prot->slab) {
4199 		pr_crit("%s: Can't create request sock SLAB cache!\n",
4200 			prot->name);
4201 		return -ENOMEM;
4202 	}
4203 	return 0;
4204 }
4205 
4206 int proto_register(struct proto *prot, int alloc_slab)
4207 {
4208 	int ret = -ENOBUFS;
4209 
4210 	if (prot->memory_allocated && !prot->sysctl_mem) {
4211 		pr_err("%s: missing sysctl_mem\n", prot->name);
4212 		return -EINVAL;
4213 	}
4214 	if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4215 		pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4216 		return -EINVAL;
4217 	}
4218 	if (alloc_slab) {
4219 		struct kmem_cache_args args = {
4220 			.useroffset	= prot->useroffset,
4221 			.usersize	= prot->usersize,
4222 			.freeptr_offset = prot->freeptr_offset,
4223 			.use_freeptr_offset = !!prot->freeptr_offset,
4224 		};
4225 
4226 		prot->slab = kmem_cache_create(prot->name, prot->obj_size,
4227 					&args,
4228 					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4229 					prot->slab_flags);
4230 		if (prot->slab == NULL) {
4231 			pr_crit("%s: Can't create sock SLAB cache!\n",
4232 				prot->name);
4233 			goto out;
4234 		}
4235 
4236 		if (req_prot_init(prot))
4237 			goto out_free_request_sock_slab;
4238 
4239 		if (tw_prot_init(prot))
4240 			goto out_free_timewait_sock_slab;
4241 	}
4242 
4243 	mutex_lock(&proto_list_mutex);
4244 	ret = assign_proto_idx(prot);
4245 	if (ret) {
4246 		mutex_unlock(&proto_list_mutex);
4247 		goto out_free_timewait_sock_slab;
4248 	}
4249 	list_add(&prot->node, &proto_list);
4250 	mutex_unlock(&proto_list_mutex);
4251 	return ret;
4252 
4253 out_free_timewait_sock_slab:
4254 	if (alloc_slab)
4255 		tw_prot_cleanup(prot->twsk_prot);
4256 out_free_request_sock_slab:
4257 	if (alloc_slab) {
4258 		req_prot_cleanup(prot->rsk_prot);
4259 
4260 		kmem_cache_destroy(prot->slab);
4261 		prot->slab = NULL;
4262 	}
4263 out:
4264 	return ret;
4265 }
4266 EXPORT_SYMBOL(proto_register);
4267 
4268 void proto_unregister(struct proto *prot)
4269 {
4270 	mutex_lock(&proto_list_mutex);
4271 	release_proto_idx(prot);
4272 	list_del(&prot->node);
4273 	mutex_unlock(&proto_list_mutex);
4274 
4275 	kmem_cache_destroy(prot->slab);
4276 	prot->slab = NULL;
4277 
4278 	req_prot_cleanup(prot->rsk_prot);
4279 	tw_prot_cleanup(prot->twsk_prot);
4280 }
4281 EXPORT_SYMBOL(proto_unregister);
4282 
4283 int sock_load_diag_module(int family, int protocol)
4284 {
4285 	if (!protocol) {
4286 		if (!sock_is_registered(family))
4287 			return -ENOENT;
4288 
4289 		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4290 				      NETLINK_SOCK_DIAG, family);
4291 	}
4292 
4293 #ifdef CONFIG_INET
4294 	if (family == AF_INET &&
4295 	    protocol != IPPROTO_RAW &&
4296 	    protocol < MAX_INET_PROTOS &&
4297 	    !rcu_access_pointer(inet_protos[protocol]))
4298 		return -ENOENT;
4299 #endif
4300 
4301 	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4302 			      NETLINK_SOCK_DIAG, family, protocol);
4303 }
4304 EXPORT_SYMBOL(sock_load_diag_module);
4305 
4306 #ifdef CONFIG_PROC_FS
4307 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4308 	__acquires(proto_list_mutex)
4309 {
4310 	mutex_lock(&proto_list_mutex);
4311 	return seq_list_start_head(&proto_list, *pos);
4312 }
4313 
4314 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4315 {
4316 	return seq_list_next(v, &proto_list, pos);
4317 }
4318 
4319 static void proto_seq_stop(struct seq_file *seq, void *v)
4320 	__releases(proto_list_mutex)
4321 {
4322 	mutex_unlock(&proto_list_mutex);
4323 }
4324 
4325 static char proto_method_implemented(const void *method)
4326 {
4327 	return method == NULL ? 'n' : 'y';
4328 }
4329 static long sock_prot_memory_allocated(struct proto *proto)
4330 {
4331 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4332 }
4333 
4334 static const char *sock_prot_memory_pressure(struct proto *proto)
4335 {
4336 	return proto->memory_pressure != NULL ?
4337 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4338 }
4339 
4340 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4341 {
4342 
4343 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
4344 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4345 		   proto->name,
4346 		   proto->obj_size,
4347 		   sock_prot_inuse_get(seq_file_net(seq), proto),
4348 		   sock_prot_memory_allocated(proto),
4349 		   sock_prot_memory_pressure(proto),
4350 		   proto->max_header,
4351 		   proto->slab == NULL ? "no" : "yes",
4352 		   module_name(proto->owner),
4353 		   proto_method_implemented(proto->close),
4354 		   proto_method_implemented(proto->connect),
4355 		   proto_method_implemented(proto->disconnect),
4356 		   proto_method_implemented(proto->accept),
4357 		   proto_method_implemented(proto->ioctl),
4358 		   proto_method_implemented(proto->init),
4359 		   proto_method_implemented(proto->destroy),
4360 		   proto_method_implemented(proto->shutdown),
4361 		   proto_method_implemented(proto->setsockopt),
4362 		   proto_method_implemented(proto->getsockopt),
4363 		   proto_method_implemented(proto->sendmsg),
4364 		   proto_method_implemented(proto->recvmsg),
4365 		   proto_method_implemented(proto->bind),
4366 		   proto_method_implemented(proto->backlog_rcv),
4367 		   proto_method_implemented(proto->hash),
4368 		   proto_method_implemented(proto->unhash),
4369 		   proto_method_implemented(proto->get_port),
4370 		   proto_method_implemented(proto->enter_memory_pressure));
4371 }
4372 
4373 static int proto_seq_show(struct seq_file *seq, void *v)
4374 {
4375 	if (v == &proto_list)
4376 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4377 			   "protocol",
4378 			   "size",
4379 			   "sockets",
4380 			   "memory",
4381 			   "press",
4382 			   "maxhdr",
4383 			   "slab",
4384 			   "module",
4385 			   "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4386 	else
4387 		proto_seq_printf(seq, list_entry(v, struct proto, node));
4388 	return 0;
4389 }
4390 
4391 static const struct seq_operations proto_seq_ops = {
4392 	.start  = proto_seq_start,
4393 	.next   = proto_seq_next,
4394 	.stop   = proto_seq_stop,
4395 	.show   = proto_seq_show,
4396 };
4397 
4398 static __net_init int proto_init_net(struct net *net)
4399 {
4400 	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4401 			sizeof(struct seq_net_private)))
4402 		return -ENOMEM;
4403 
4404 	return 0;
4405 }
4406 
4407 static __net_exit void proto_exit_net(struct net *net)
4408 {
4409 	remove_proc_entry("protocols", net->proc_net);
4410 }
4411 
4412 
4413 static __net_initdata struct pernet_operations proto_net_ops = {
4414 	.init = proto_init_net,
4415 	.exit = proto_exit_net,
4416 };
4417 
4418 static int __init proto_init(void)
4419 {
4420 	return register_pernet_subsys(&proto_net_ops);
4421 }
4422 
4423 subsys_initcall(proto_init);
4424 
4425 #endif /* PROC_FS */
4426 
4427 #ifdef CONFIG_NET_RX_BUSY_POLL
4428 bool sk_busy_loop_end(void *p, unsigned long start_time)
4429 {
4430 	struct sock *sk = p;
4431 
4432 	if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4433 		return true;
4434 
4435 	if (sk_is_udp(sk) &&
4436 	    !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4437 		return true;
4438 
4439 	return sk_busy_loop_timeout(sk, start_time);
4440 }
4441 EXPORT_SYMBOL(sk_busy_loop_end);
4442 #endif /* CONFIG_NET_RX_BUSY_POLL */
4443 
4444 int sock_bind_add(struct sock *sk, struct sockaddr_unsized *addr, int addr_len)
4445 {
4446 	if (!sk->sk_prot->bind_add)
4447 		return -EOPNOTSUPP;
4448 	return sk->sk_prot->bind_add(sk, addr, addr_len);
4449 }
4450 EXPORT_SYMBOL(sock_bind_add);
4451 
4452 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4453 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4454 		     void __user *arg, void *karg, size_t size)
4455 {
4456 	int ret;
4457 
4458 	if (copy_from_user(karg, arg, size))
4459 		return -EFAULT;
4460 
4461 	ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4462 	if (ret)
4463 		return ret;
4464 
4465 	if (copy_to_user(arg, karg, size))
4466 		return -EFAULT;
4467 
4468 	return 0;
4469 }
4470 EXPORT_SYMBOL(sock_ioctl_inout);
4471 
4472 /* This is the most common ioctl prep function, where the result (4 bytes) is
4473  * copied back to userspace if the ioctl() returns successfully. No input is
4474  * copied from userspace as input argument.
4475  */
4476 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4477 {
4478 	int ret, karg = 0;
4479 
4480 	ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4481 	if (ret)
4482 		return ret;
4483 
4484 	return put_user(karg, (int __user *)arg);
4485 }
4486 
4487 /* A wrapper around sock ioctls, which copies the data from userspace
4488  * (depending on the protocol/ioctl), and copies back the result to userspace.
4489  * The main motivation for this function is to pass kernel memory to the
4490  * protocol ioctl callbacks, instead of userspace memory.
4491  */
4492 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4493 {
4494 	int rc = 1;
4495 
4496 	if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4497 		rc = ipmr_sk_ioctl(sk, cmd, arg);
4498 	else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4499 		rc = ip6mr_sk_ioctl(sk, cmd, arg);
4500 	else if (sk_is_phonet(sk))
4501 		rc = phonet_sk_ioctl(sk, cmd, arg);
4502 
4503 	/* If ioctl was processed, returns its value */
4504 	if (rc <= 0)
4505 		return rc;
4506 
4507 	/* Otherwise call the default handler */
4508 	return sock_ioctl_out(sk, cmd, arg);
4509 }
4510 EXPORT_SYMBOL(sk_ioctl);
4511 
4512 static int __init sock_struct_check(void)
4513 {
4514 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4515 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4516 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4517 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4518 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4519 
4520 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4521 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4522 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4523 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4524 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4525 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4526 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4527 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4528 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4529 
4530 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4531 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4532 #ifdef CONFIG_MEMCG
4533 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4534 #endif
4535 
4536 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4537 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4538 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4539 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4540 
4541 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4542 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4543 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_err_soft);
4544 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4545 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4546 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4547 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4548 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4549 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4550 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4551 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4552 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4553 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4554 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4555 
4556 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_pending_confirm);
4557 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_status);
4558 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4559 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4560 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4561 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4562 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_uid);
4563 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_protocol);
4564 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4565 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4566 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4567 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4568 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4569 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4570 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndbuf);
4571 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4572 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4573 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4574 	return 0;
4575 }
4576 
4577 core_initcall(sock_struct_check);
4578