xref: /openbmc/linux/net/core/sock.c (revision c9933d49)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Generic socket support routines. Memory allocators, socket lock/release
8  *		handler for protocols to use and generic option handler.
9  *
10  * Authors:	Ross Biro
11  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *		Florian La Roche, <flla@stud.uni-sb.de>
13  *		Alan Cox, <A.Cox@swansea.ac.uk>
14  *
15  * Fixes:
16  *		Alan Cox	: 	Numerous verify_area() problems
17  *		Alan Cox	:	Connecting on a connecting socket
18  *					now returns an error for tcp.
19  *		Alan Cox	:	sock->protocol is set correctly.
20  *					and is not sometimes left as 0.
21  *		Alan Cox	:	connect handles icmp errors on a
22  *					connect properly. Unfortunately there
23  *					is a restart syscall nasty there. I
24  *					can't match BSD without hacking the C
25  *					library. Ideas urgently sought!
26  *		Alan Cox	:	Disallow bind() to addresses that are
27  *					not ours - especially broadcast ones!!
28  *		Alan Cox	:	Socket 1024 _IS_ ok for users. (fencepost)
29  *		Alan Cox	:	sock_wfree/sock_rfree don't destroy sockets,
30  *					instead they leave that for the DESTROY timer.
31  *		Alan Cox	:	Clean up error flag in accept
32  *		Alan Cox	:	TCP ack handling is buggy, the DESTROY timer
33  *					was buggy. Put a remove_sock() in the handler
34  *					for memory when we hit 0. Also altered the timer
35  *					code. The ACK stuff can wait and needs major
36  *					TCP layer surgery.
37  *		Alan Cox	:	Fixed TCP ack bug, removed remove sock
38  *					and fixed timer/inet_bh race.
39  *		Alan Cox	:	Added zapped flag for TCP
40  *		Alan Cox	:	Move kfree_skb into skbuff.c and tidied up surplus code
41  *		Alan Cox	:	for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42  *		Alan Cox	:	kfree_s calls now are kfree_skbmem so we can track skb resources
43  *		Alan Cox	:	Supports socket option broadcast now as does udp. Packet and raw need fixing.
44  *		Alan Cox	:	Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45  *		Rick Sladkey	:	Relaxed UDP rules for matching packets.
46  *		C.E.Hawkins	:	IFF_PROMISC/SIOCGHWADDR support
47  *	Pauline Middelink	:	identd support
48  *		Alan Cox	:	Fixed connect() taking signals I think.
49  *		Alan Cox	:	SO_LINGER supported
50  *		Alan Cox	:	Error reporting fixes
51  *		Anonymous	:	inet_create tidied up (sk->reuse setting)
52  *		Alan Cox	:	inet sockets don't set sk->type!
53  *		Alan Cox	:	Split socket option code
54  *		Alan Cox	:	Callbacks
55  *		Alan Cox	:	Nagle flag for Charles & Johannes stuff
56  *		Alex		:	Removed restriction on inet fioctl
57  *		Alan Cox	:	Splitting INET from NET core
58  *		Alan Cox	:	Fixed bogus SO_TYPE handling in getsockopt()
59  *		Adam Caldwell	:	Missing return in SO_DONTROUTE/SO_DEBUG code
60  *		Alan Cox	:	Split IP from generic code
61  *		Alan Cox	:	New kfree_skbmem()
62  *		Alan Cox	:	Make SO_DEBUG superuser only.
63  *		Alan Cox	:	Allow anyone to clear SO_DEBUG
64  *					(compatibility fix)
65  *		Alan Cox	:	Added optimistic memory grabbing for AF_UNIX throughput.
66  *		Alan Cox	:	Allocator for a socket is settable.
67  *		Alan Cox	:	SO_ERROR includes soft errors.
68  *		Alan Cox	:	Allow NULL arguments on some SO_ opts
69  *		Alan Cox	: 	Generic socket allocation to make hooks
70  *					easier (suggested by Craig Metz).
71  *		Michael Pall	:	SO_ERROR returns positive errno again
72  *              Steve Whitehouse:       Added default destructor to free
73  *                                      protocol private data.
74  *              Steve Whitehouse:       Added various other default routines
75  *                                      common to several socket families.
76  *              Chris Evans     :       Call suser() check last on F_SETOWN
77  *		Jay Schulist	:	Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78  *		Andi Kleen	:	Add sock_kmalloc()/sock_kfree_s()
79  *		Andi Kleen	:	Fix write_space callback
80  *		Chris Evans	:	Security fixes - signedness again
81  *		Arnaldo C. Melo :       cleanups, use skb_queue_purge
82  *
83  * To Fix:
84  */
85 
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87 
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
94 #include <linux/in.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
117 
118 #include <linux/uaccess.h>
119 
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
132 
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
136 
137 #include <trace/events/sock.h>
138 
139 #include <net/tcp.h>
140 #include <net/busy_poll.h>
141 
142 #include <linux/ethtool.h>
143 
144 #include "dev.h"
145 
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
148 
149 static void sock_def_write_space_wfree(struct sock *sk);
150 static void sock_def_write_space(struct sock *sk);
151 
152 /**
153  * sk_ns_capable - General socket capability test
154  * @sk: Socket to use a capability on or through
155  * @user_ns: The user namespace of the capability to use
156  * @cap: The capability to use
157  *
158  * Test to see if the opener of the socket had when the socket was
159  * created and the current process has the capability @cap in the user
160  * namespace @user_ns.
161  */
162 bool sk_ns_capable(const struct sock *sk,
163 		   struct user_namespace *user_ns, int cap)
164 {
165 	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
166 		ns_capable(user_ns, cap);
167 }
168 EXPORT_SYMBOL(sk_ns_capable);
169 
170 /**
171  * sk_capable - Socket global capability test
172  * @sk: Socket to use a capability on or through
173  * @cap: The global capability to use
174  *
175  * Test to see if the opener of the socket had when the socket was
176  * created and the current process has the capability @cap in all user
177  * namespaces.
178  */
179 bool sk_capable(const struct sock *sk, int cap)
180 {
181 	return sk_ns_capable(sk, &init_user_ns, cap);
182 }
183 EXPORT_SYMBOL(sk_capable);
184 
185 /**
186  * sk_net_capable - Network namespace socket capability test
187  * @sk: Socket to use a capability on or through
188  * @cap: The capability to use
189  *
190  * Test to see if the opener of the socket had when the socket was created
191  * and the current process has the capability @cap over the network namespace
192  * the socket is a member of.
193  */
194 bool sk_net_capable(const struct sock *sk, int cap)
195 {
196 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
197 }
198 EXPORT_SYMBOL(sk_net_capable);
199 
200 /*
201  * Each address family might have different locking rules, so we have
202  * one slock key per address family and separate keys for internal and
203  * userspace sockets.
204  */
205 static struct lock_class_key af_family_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_keys[AF_MAX];
207 static struct lock_class_key af_family_slock_keys[AF_MAX];
208 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209 
210 /*
211  * Make lock validator output more readable. (we pre-construct these
212  * strings build-time, so that runtime initialization of socket
213  * locks is fast):
214  */
215 
216 #define _sock_locks(x)						  \
217   x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
218   x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
219   x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
220   x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
221   x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
222   x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
223   x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
224   x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
225   x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
226   x "27"       ,	x "28"          ,	x "AF_CAN"      , \
227   x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
228   x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
229   x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
230   x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
231   x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
232   x "AF_MCTP"  , \
233   x "AF_MAX"
234 
235 static const char *const af_family_key_strings[AF_MAX+1] = {
236 	_sock_locks("sk_lock-")
237 };
238 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
239 	_sock_locks("slock-")
240 };
241 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
242 	_sock_locks("clock-")
243 };
244 
245 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
246 	_sock_locks("k-sk_lock-")
247 };
248 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
249 	_sock_locks("k-slock-")
250 };
251 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
252 	_sock_locks("k-clock-")
253 };
254 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
255 	_sock_locks("rlock-")
256 };
257 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
258 	_sock_locks("wlock-")
259 };
260 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
261 	_sock_locks("elock-")
262 };
263 
264 /*
265  * sk_callback_lock and sk queues locking rules are per-address-family,
266  * so split the lock classes by using a per-AF key:
267  */
268 static struct lock_class_key af_callback_keys[AF_MAX];
269 static struct lock_class_key af_rlock_keys[AF_MAX];
270 static struct lock_class_key af_wlock_keys[AF_MAX];
271 static struct lock_class_key af_elock_keys[AF_MAX];
272 static struct lock_class_key af_kern_callback_keys[AF_MAX];
273 
274 /* Run time adjustable parameters. */
275 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
276 EXPORT_SYMBOL(sysctl_wmem_max);
277 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
278 EXPORT_SYMBOL(sysctl_rmem_max);
279 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
280 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
281 
282 /* Maximal space eaten by iovec or ancillary data plus some space */
283 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
284 EXPORT_SYMBOL(sysctl_optmem_max);
285 
286 int sysctl_tstamp_allow_data __read_mostly = 1;
287 
288 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
289 EXPORT_SYMBOL_GPL(memalloc_socks_key);
290 
291 /**
292  * sk_set_memalloc - sets %SOCK_MEMALLOC
293  * @sk: socket to set it on
294  *
295  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
296  * It's the responsibility of the admin to adjust min_free_kbytes
297  * to meet the requirements
298  */
299 void sk_set_memalloc(struct sock *sk)
300 {
301 	sock_set_flag(sk, SOCK_MEMALLOC);
302 	sk->sk_allocation |= __GFP_MEMALLOC;
303 	static_branch_inc(&memalloc_socks_key);
304 }
305 EXPORT_SYMBOL_GPL(sk_set_memalloc);
306 
307 void sk_clear_memalloc(struct sock *sk)
308 {
309 	sock_reset_flag(sk, SOCK_MEMALLOC);
310 	sk->sk_allocation &= ~__GFP_MEMALLOC;
311 	static_branch_dec(&memalloc_socks_key);
312 
313 	/*
314 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
315 	 * progress of swapping. SOCK_MEMALLOC may be cleared while
316 	 * it has rmem allocations due to the last swapfile being deactivated
317 	 * but there is a risk that the socket is unusable due to exceeding
318 	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
319 	 */
320 	sk_mem_reclaim(sk);
321 }
322 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
323 
324 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
325 {
326 	int ret;
327 	unsigned int noreclaim_flag;
328 
329 	/* these should have been dropped before queueing */
330 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
331 
332 	noreclaim_flag = memalloc_noreclaim_save();
333 	ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
334 				 tcp_v6_do_rcv,
335 				 tcp_v4_do_rcv,
336 				 sk, skb);
337 	memalloc_noreclaim_restore(noreclaim_flag);
338 
339 	return ret;
340 }
341 EXPORT_SYMBOL(__sk_backlog_rcv);
342 
343 void sk_error_report(struct sock *sk)
344 {
345 	sk->sk_error_report(sk);
346 
347 	switch (sk->sk_family) {
348 	case AF_INET:
349 		fallthrough;
350 	case AF_INET6:
351 		trace_inet_sk_error_report(sk);
352 		break;
353 	default:
354 		break;
355 	}
356 }
357 EXPORT_SYMBOL(sk_error_report);
358 
359 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
360 {
361 	struct __kernel_sock_timeval tv;
362 
363 	if (timeo == MAX_SCHEDULE_TIMEOUT) {
364 		tv.tv_sec = 0;
365 		tv.tv_usec = 0;
366 	} else {
367 		tv.tv_sec = timeo / HZ;
368 		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
369 	}
370 
371 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
372 		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
373 		*(struct old_timeval32 *)optval = tv32;
374 		return sizeof(tv32);
375 	}
376 
377 	if (old_timeval) {
378 		struct __kernel_old_timeval old_tv;
379 		old_tv.tv_sec = tv.tv_sec;
380 		old_tv.tv_usec = tv.tv_usec;
381 		*(struct __kernel_old_timeval *)optval = old_tv;
382 		return sizeof(old_tv);
383 	}
384 
385 	*(struct __kernel_sock_timeval *)optval = tv;
386 	return sizeof(tv);
387 }
388 EXPORT_SYMBOL(sock_get_timeout);
389 
390 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
391 			   sockptr_t optval, int optlen, bool old_timeval)
392 {
393 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
394 		struct old_timeval32 tv32;
395 
396 		if (optlen < sizeof(tv32))
397 			return -EINVAL;
398 
399 		if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
400 			return -EFAULT;
401 		tv->tv_sec = tv32.tv_sec;
402 		tv->tv_usec = tv32.tv_usec;
403 	} else if (old_timeval) {
404 		struct __kernel_old_timeval old_tv;
405 
406 		if (optlen < sizeof(old_tv))
407 			return -EINVAL;
408 		if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
409 			return -EFAULT;
410 		tv->tv_sec = old_tv.tv_sec;
411 		tv->tv_usec = old_tv.tv_usec;
412 	} else {
413 		if (optlen < sizeof(*tv))
414 			return -EINVAL;
415 		if (copy_from_sockptr(tv, optval, sizeof(*tv)))
416 			return -EFAULT;
417 	}
418 
419 	return 0;
420 }
421 EXPORT_SYMBOL(sock_copy_user_timeval);
422 
423 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
424 			    bool old_timeval)
425 {
426 	struct __kernel_sock_timeval tv;
427 	int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
428 
429 	if (err)
430 		return err;
431 
432 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
433 		return -EDOM;
434 
435 	if (tv.tv_sec < 0) {
436 		static int warned __read_mostly;
437 
438 		*timeo_p = 0;
439 		if (warned < 10 && net_ratelimit()) {
440 			warned++;
441 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
442 				__func__, current->comm, task_pid_nr(current));
443 		}
444 		return 0;
445 	}
446 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
447 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
448 		return 0;
449 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
450 		*timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
451 	return 0;
452 }
453 
454 static bool sock_needs_netstamp(const struct sock *sk)
455 {
456 	switch (sk->sk_family) {
457 	case AF_UNSPEC:
458 	case AF_UNIX:
459 		return false;
460 	default:
461 		return true;
462 	}
463 }
464 
465 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
466 {
467 	if (sk->sk_flags & flags) {
468 		sk->sk_flags &= ~flags;
469 		if (sock_needs_netstamp(sk) &&
470 		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
471 			net_disable_timestamp();
472 	}
473 }
474 
475 
476 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
477 {
478 	unsigned long flags;
479 	struct sk_buff_head *list = &sk->sk_receive_queue;
480 
481 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
482 		atomic_inc(&sk->sk_drops);
483 		trace_sock_rcvqueue_full(sk, skb);
484 		return -ENOMEM;
485 	}
486 
487 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
488 		atomic_inc(&sk->sk_drops);
489 		return -ENOBUFS;
490 	}
491 
492 	skb->dev = NULL;
493 	skb_set_owner_r(skb, sk);
494 
495 	/* we escape from rcu protected region, make sure we dont leak
496 	 * a norefcounted dst
497 	 */
498 	skb_dst_force(skb);
499 
500 	spin_lock_irqsave(&list->lock, flags);
501 	sock_skb_set_dropcount(sk, skb);
502 	__skb_queue_tail(list, skb);
503 	spin_unlock_irqrestore(&list->lock, flags);
504 
505 	if (!sock_flag(sk, SOCK_DEAD))
506 		sk->sk_data_ready(sk);
507 	return 0;
508 }
509 EXPORT_SYMBOL(__sock_queue_rcv_skb);
510 
511 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
512 			      enum skb_drop_reason *reason)
513 {
514 	enum skb_drop_reason drop_reason;
515 	int err;
516 
517 	err = sk_filter(sk, skb);
518 	if (err) {
519 		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
520 		goto out;
521 	}
522 	err = __sock_queue_rcv_skb(sk, skb);
523 	switch (err) {
524 	case -ENOMEM:
525 		drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
526 		break;
527 	case -ENOBUFS:
528 		drop_reason = SKB_DROP_REASON_PROTO_MEM;
529 		break;
530 	default:
531 		drop_reason = SKB_NOT_DROPPED_YET;
532 		break;
533 	}
534 out:
535 	if (reason)
536 		*reason = drop_reason;
537 	return err;
538 }
539 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
540 
541 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
542 		     const int nested, unsigned int trim_cap, bool refcounted)
543 {
544 	int rc = NET_RX_SUCCESS;
545 
546 	if (sk_filter_trim_cap(sk, skb, trim_cap))
547 		goto discard_and_relse;
548 
549 	skb->dev = NULL;
550 
551 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
552 		atomic_inc(&sk->sk_drops);
553 		goto discard_and_relse;
554 	}
555 	if (nested)
556 		bh_lock_sock_nested(sk);
557 	else
558 		bh_lock_sock(sk);
559 	if (!sock_owned_by_user(sk)) {
560 		/*
561 		 * trylock + unlock semantics:
562 		 */
563 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
564 
565 		rc = sk_backlog_rcv(sk, skb);
566 
567 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
568 	} else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
569 		bh_unlock_sock(sk);
570 		atomic_inc(&sk->sk_drops);
571 		goto discard_and_relse;
572 	}
573 
574 	bh_unlock_sock(sk);
575 out:
576 	if (refcounted)
577 		sock_put(sk);
578 	return rc;
579 discard_and_relse:
580 	kfree_skb(skb);
581 	goto out;
582 }
583 EXPORT_SYMBOL(__sk_receive_skb);
584 
585 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
586 							  u32));
587 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
588 							   u32));
589 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
590 {
591 	struct dst_entry *dst = __sk_dst_get(sk);
592 
593 	if (dst && dst->obsolete &&
594 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
595 			       dst, cookie) == NULL) {
596 		sk_tx_queue_clear(sk);
597 		sk->sk_dst_pending_confirm = 0;
598 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
599 		dst_release(dst);
600 		return NULL;
601 	}
602 
603 	return dst;
604 }
605 EXPORT_SYMBOL(__sk_dst_check);
606 
607 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
608 {
609 	struct dst_entry *dst = sk_dst_get(sk);
610 
611 	if (dst && dst->obsolete &&
612 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
613 			       dst, cookie) == NULL) {
614 		sk_dst_reset(sk);
615 		dst_release(dst);
616 		return NULL;
617 	}
618 
619 	return dst;
620 }
621 EXPORT_SYMBOL(sk_dst_check);
622 
623 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
624 {
625 	int ret = -ENOPROTOOPT;
626 #ifdef CONFIG_NETDEVICES
627 	struct net *net = sock_net(sk);
628 
629 	/* Sorry... */
630 	ret = -EPERM;
631 	if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
632 		goto out;
633 
634 	ret = -EINVAL;
635 	if (ifindex < 0)
636 		goto out;
637 
638 	sk->sk_bound_dev_if = ifindex;
639 	if (sk->sk_prot->rehash)
640 		sk->sk_prot->rehash(sk);
641 	sk_dst_reset(sk);
642 
643 	ret = 0;
644 
645 out:
646 #endif
647 
648 	return ret;
649 }
650 
651 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
652 {
653 	int ret;
654 
655 	if (lock_sk)
656 		lock_sock(sk);
657 	ret = sock_bindtoindex_locked(sk, ifindex);
658 	if (lock_sk)
659 		release_sock(sk);
660 
661 	return ret;
662 }
663 EXPORT_SYMBOL(sock_bindtoindex);
664 
665 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
666 {
667 	int ret = -ENOPROTOOPT;
668 #ifdef CONFIG_NETDEVICES
669 	struct net *net = sock_net(sk);
670 	char devname[IFNAMSIZ];
671 	int index;
672 
673 	ret = -EINVAL;
674 	if (optlen < 0)
675 		goto out;
676 
677 	/* Bind this socket to a particular device like "eth0",
678 	 * as specified in the passed interface name. If the
679 	 * name is "" or the option length is zero the socket
680 	 * is not bound.
681 	 */
682 	if (optlen > IFNAMSIZ - 1)
683 		optlen = IFNAMSIZ - 1;
684 	memset(devname, 0, sizeof(devname));
685 
686 	ret = -EFAULT;
687 	if (copy_from_sockptr(devname, optval, optlen))
688 		goto out;
689 
690 	index = 0;
691 	if (devname[0] != '\0') {
692 		struct net_device *dev;
693 
694 		rcu_read_lock();
695 		dev = dev_get_by_name_rcu(net, devname);
696 		if (dev)
697 			index = dev->ifindex;
698 		rcu_read_unlock();
699 		ret = -ENODEV;
700 		if (!dev)
701 			goto out;
702 	}
703 
704 	return sock_bindtoindex(sk, index, true);
705 out:
706 #endif
707 
708 	return ret;
709 }
710 
711 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
712 				int __user *optlen, int len)
713 {
714 	int ret = -ENOPROTOOPT;
715 #ifdef CONFIG_NETDEVICES
716 	struct net *net = sock_net(sk);
717 	char devname[IFNAMSIZ];
718 
719 	if (sk->sk_bound_dev_if == 0) {
720 		len = 0;
721 		goto zero;
722 	}
723 
724 	ret = -EINVAL;
725 	if (len < IFNAMSIZ)
726 		goto out;
727 
728 	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
729 	if (ret)
730 		goto out;
731 
732 	len = strlen(devname) + 1;
733 
734 	ret = -EFAULT;
735 	if (copy_to_user(optval, devname, len))
736 		goto out;
737 
738 zero:
739 	ret = -EFAULT;
740 	if (put_user(len, optlen))
741 		goto out;
742 
743 	ret = 0;
744 
745 out:
746 #endif
747 
748 	return ret;
749 }
750 
751 bool sk_mc_loop(struct sock *sk)
752 {
753 	if (dev_recursion_level())
754 		return false;
755 	if (!sk)
756 		return true;
757 	switch (sk->sk_family) {
758 	case AF_INET:
759 		return inet_sk(sk)->mc_loop;
760 #if IS_ENABLED(CONFIG_IPV6)
761 	case AF_INET6:
762 		return inet6_sk(sk)->mc_loop;
763 #endif
764 	}
765 	WARN_ON_ONCE(1);
766 	return true;
767 }
768 EXPORT_SYMBOL(sk_mc_loop);
769 
770 void sock_set_reuseaddr(struct sock *sk)
771 {
772 	lock_sock(sk);
773 	sk->sk_reuse = SK_CAN_REUSE;
774 	release_sock(sk);
775 }
776 EXPORT_SYMBOL(sock_set_reuseaddr);
777 
778 void sock_set_reuseport(struct sock *sk)
779 {
780 	lock_sock(sk);
781 	sk->sk_reuseport = true;
782 	release_sock(sk);
783 }
784 EXPORT_SYMBOL(sock_set_reuseport);
785 
786 void sock_no_linger(struct sock *sk)
787 {
788 	lock_sock(sk);
789 	sk->sk_lingertime = 0;
790 	sock_set_flag(sk, SOCK_LINGER);
791 	release_sock(sk);
792 }
793 EXPORT_SYMBOL(sock_no_linger);
794 
795 void sock_set_priority(struct sock *sk, u32 priority)
796 {
797 	lock_sock(sk);
798 	sk->sk_priority = priority;
799 	release_sock(sk);
800 }
801 EXPORT_SYMBOL(sock_set_priority);
802 
803 void sock_set_sndtimeo(struct sock *sk, s64 secs)
804 {
805 	lock_sock(sk);
806 	if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
807 		sk->sk_sndtimeo = secs * HZ;
808 	else
809 		sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
810 	release_sock(sk);
811 }
812 EXPORT_SYMBOL(sock_set_sndtimeo);
813 
814 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
815 {
816 	if (val)  {
817 		sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
818 		sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
819 		sock_set_flag(sk, SOCK_RCVTSTAMP);
820 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
821 	} else {
822 		sock_reset_flag(sk, SOCK_RCVTSTAMP);
823 		sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
824 	}
825 }
826 
827 void sock_enable_timestamps(struct sock *sk)
828 {
829 	lock_sock(sk);
830 	__sock_set_timestamps(sk, true, false, true);
831 	release_sock(sk);
832 }
833 EXPORT_SYMBOL(sock_enable_timestamps);
834 
835 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
836 {
837 	switch (optname) {
838 	case SO_TIMESTAMP_OLD:
839 		__sock_set_timestamps(sk, valbool, false, false);
840 		break;
841 	case SO_TIMESTAMP_NEW:
842 		__sock_set_timestamps(sk, valbool, true, false);
843 		break;
844 	case SO_TIMESTAMPNS_OLD:
845 		__sock_set_timestamps(sk, valbool, false, true);
846 		break;
847 	case SO_TIMESTAMPNS_NEW:
848 		__sock_set_timestamps(sk, valbool, true, true);
849 		break;
850 	}
851 }
852 
853 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
854 {
855 	struct net *net = sock_net(sk);
856 	struct net_device *dev = NULL;
857 	bool match = false;
858 	int *vclock_index;
859 	int i, num;
860 
861 	if (sk->sk_bound_dev_if)
862 		dev = dev_get_by_index(net, sk->sk_bound_dev_if);
863 
864 	if (!dev) {
865 		pr_err("%s: sock not bind to device\n", __func__);
866 		return -EOPNOTSUPP;
867 	}
868 
869 	num = ethtool_get_phc_vclocks(dev, &vclock_index);
870 	dev_put(dev);
871 
872 	for (i = 0; i < num; i++) {
873 		if (*(vclock_index + i) == phc_index) {
874 			match = true;
875 			break;
876 		}
877 	}
878 
879 	if (num > 0)
880 		kfree(vclock_index);
881 
882 	if (!match)
883 		return -EINVAL;
884 
885 	sk->sk_bind_phc = phc_index;
886 
887 	return 0;
888 }
889 
890 int sock_set_timestamping(struct sock *sk, int optname,
891 			  struct so_timestamping timestamping)
892 {
893 	int val = timestamping.flags;
894 	int ret;
895 
896 	if (val & ~SOF_TIMESTAMPING_MASK)
897 		return -EINVAL;
898 
899 	if (val & SOF_TIMESTAMPING_OPT_ID &&
900 	    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
901 		if (sk_is_tcp(sk)) {
902 			if ((1 << sk->sk_state) &
903 			    (TCPF_CLOSE | TCPF_LISTEN))
904 				return -EINVAL;
905 			atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
906 		} else {
907 			atomic_set(&sk->sk_tskey, 0);
908 		}
909 	}
910 
911 	if (val & SOF_TIMESTAMPING_OPT_STATS &&
912 	    !(val & SOF_TIMESTAMPING_OPT_TSONLY))
913 		return -EINVAL;
914 
915 	if (val & SOF_TIMESTAMPING_BIND_PHC) {
916 		ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
917 		if (ret)
918 			return ret;
919 	}
920 
921 	sk->sk_tsflags = val;
922 	sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
923 
924 	if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
925 		sock_enable_timestamp(sk,
926 				      SOCK_TIMESTAMPING_RX_SOFTWARE);
927 	else
928 		sock_disable_timestamp(sk,
929 				       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
930 	return 0;
931 }
932 
933 void sock_set_keepalive(struct sock *sk)
934 {
935 	lock_sock(sk);
936 	if (sk->sk_prot->keepalive)
937 		sk->sk_prot->keepalive(sk, true);
938 	sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
939 	release_sock(sk);
940 }
941 EXPORT_SYMBOL(sock_set_keepalive);
942 
943 static void __sock_set_rcvbuf(struct sock *sk, int val)
944 {
945 	/* Ensure val * 2 fits into an int, to prevent max_t() from treating it
946 	 * as a negative value.
947 	 */
948 	val = min_t(int, val, INT_MAX / 2);
949 	sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
950 
951 	/* We double it on the way in to account for "struct sk_buff" etc.
952 	 * overhead.   Applications assume that the SO_RCVBUF setting they make
953 	 * will allow that much actual data to be received on that socket.
954 	 *
955 	 * Applications are unaware that "struct sk_buff" and other overheads
956 	 * allocate from the receive buffer during socket buffer allocation.
957 	 *
958 	 * And after considering the possible alternatives, returning the value
959 	 * we actually used in getsockopt is the most desirable behavior.
960 	 */
961 	WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
962 }
963 
964 void sock_set_rcvbuf(struct sock *sk, int val)
965 {
966 	lock_sock(sk);
967 	__sock_set_rcvbuf(sk, val);
968 	release_sock(sk);
969 }
970 EXPORT_SYMBOL(sock_set_rcvbuf);
971 
972 static void __sock_set_mark(struct sock *sk, u32 val)
973 {
974 	if (val != sk->sk_mark) {
975 		sk->sk_mark = val;
976 		sk_dst_reset(sk);
977 	}
978 }
979 
980 void sock_set_mark(struct sock *sk, u32 val)
981 {
982 	lock_sock(sk);
983 	__sock_set_mark(sk, val);
984 	release_sock(sk);
985 }
986 EXPORT_SYMBOL(sock_set_mark);
987 
988 static void sock_release_reserved_memory(struct sock *sk, int bytes)
989 {
990 	/* Round down bytes to multiple of pages */
991 	bytes &= ~(SK_MEM_QUANTUM - 1);
992 
993 	WARN_ON(bytes > sk->sk_reserved_mem);
994 	sk->sk_reserved_mem -= bytes;
995 	sk_mem_reclaim(sk);
996 }
997 
998 static int sock_reserve_memory(struct sock *sk, int bytes)
999 {
1000 	long allocated;
1001 	bool charged;
1002 	int pages;
1003 
1004 	if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1005 		return -EOPNOTSUPP;
1006 
1007 	if (!bytes)
1008 		return 0;
1009 
1010 	pages = sk_mem_pages(bytes);
1011 
1012 	/* pre-charge to memcg */
1013 	charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1014 					  GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1015 	if (!charged)
1016 		return -ENOMEM;
1017 
1018 	/* pre-charge to forward_alloc */
1019 	allocated = sk_memory_allocated_add(sk, pages);
1020 	/* If the system goes into memory pressure with this
1021 	 * precharge, give up and return error.
1022 	 */
1023 	if (allocated > sk_prot_mem_limits(sk, 1)) {
1024 		sk_memory_allocated_sub(sk, pages);
1025 		mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1026 		return -ENOMEM;
1027 	}
1028 	sk->sk_forward_alloc += pages << SK_MEM_QUANTUM_SHIFT;
1029 
1030 	sk->sk_reserved_mem += pages << SK_MEM_QUANTUM_SHIFT;
1031 
1032 	return 0;
1033 }
1034 
1035 /*
1036  *	This is meant for all protocols to use and covers goings on
1037  *	at the socket level. Everything here is generic.
1038  */
1039 
1040 int sock_setsockopt(struct socket *sock, int level, int optname,
1041 		    sockptr_t optval, unsigned int optlen)
1042 {
1043 	struct so_timestamping timestamping;
1044 	struct sock_txtime sk_txtime;
1045 	struct sock *sk = sock->sk;
1046 	int val;
1047 	int valbool;
1048 	struct linger ling;
1049 	int ret = 0;
1050 
1051 	/*
1052 	 *	Options without arguments
1053 	 */
1054 
1055 	if (optname == SO_BINDTODEVICE)
1056 		return sock_setbindtodevice(sk, optval, optlen);
1057 
1058 	if (optlen < sizeof(int))
1059 		return -EINVAL;
1060 
1061 	if (copy_from_sockptr(&val, optval, sizeof(val)))
1062 		return -EFAULT;
1063 
1064 	valbool = val ? 1 : 0;
1065 
1066 	lock_sock(sk);
1067 
1068 	switch (optname) {
1069 	case SO_DEBUG:
1070 		if (val && !capable(CAP_NET_ADMIN))
1071 			ret = -EACCES;
1072 		else
1073 			sock_valbool_flag(sk, SOCK_DBG, valbool);
1074 		break;
1075 	case SO_REUSEADDR:
1076 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1077 		break;
1078 	case SO_REUSEPORT:
1079 		sk->sk_reuseport = valbool;
1080 		break;
1081 	case SO_TYPE:
1082 	case SO_PROTOCOL:
1083 	case SO_DOMAIN:
1084 	case SO_ERROR:
1085 		ret = -ENOPROTOOPT;
1086 		break;
1087 	case SO_DONTROUTE:
1088 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1089 		sk_dst_reset(sk);
1090 		break;
1091 	case SO_BROADCAST:
1092 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1093 		break;
1094 	case SO_SNDBUF:
1095 		/* Don't error on this BSD doesn't and if you think
1096 		 * about it this is right. Otherwise apps have to
1097 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1098 		 * are treated in BSD as hints
1099 		 */
1100 		val = min_t(u32, val, sysctl_wmem_max);
1101 set_sndbuf:
1102 		/* Ensure val * 2 fits into an int, to prevent max_t()
1103 		 * from treating it as a negative value.
1104 		 */
1105 		val = min_t(int, val, INT_MAX / 2);
1106 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1107 		WRITE_ONCE(sk->sk_sndbuf,
1108 			   max_t(int, val * 2, SOCK_MIN_SNDBUF));
1109 		/* Wake up sending tasks if we upped the value. */
1110 		sk->sk_write_space(sk);
1111 		break;
1112 
1113 	case SO_SNDBUFFORCE:
1114 		if (!capable(CAP_NET_ADMIN)) {
1115 			ret = -EPERM;
1116 			break;
1117 		}
1118 
1119 		/* No negative values (to prevent underflow, as val will be
1120 		 * multiplied by 2).
1121 		 */
1122 		if (val < 0)
1123 			val = 0;
1124 		goto set_sndbuf;
1125 
1126 	case SO_RCVBUF:
1127 		/* Don't error on this BSD doesn't and if you think
1128 		 * about it this is right. Otherwise apps have to
1129 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1130 		 * are treated in BSD as hints
1131 		 */
1132 		__sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1133 		break;
1134 
1135 	case SO_RCVBUFFORCE:
1136 		if (!capable(CAP_NET_ADMIN)) {
1137 			ret = -EPERM;
1138 			break;
1139 		}
1140 
1141 		/* No negative values (to prevent underflow, as val will be
1142 		 * multiplied by 2).
1143 		 */
1144 		__sock_set_rcvbuf(sk, max(val, 0));
1145 		break;
1146 
1147 	case SO_KEEPALIVE:
1148 		if (sk->sk_prot->keepalive)
1149 			sk->sk_prot->keepalive(sk, valbool);
1150 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1151 		break;
1152 
1153 	case SO_OOBINLINE:
1154 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1155 		break;
1156 
1157 	case SO_NO_CHECK:
1158 		sk->sk_no_check_tx = valbool;
1159 		break;
1160 
1161 	case SO_PRIORITY:
1162 		if ((val >= 0 && val <= 6) ||
1163 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1164 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1165 			sk->sk_priority = val;
1166 		else
1167 			ret = -EPERM;
1168 		break;
1169 
1170 	case SO_LINGER:
1171 		if (optlen < sizeof(ling)) {
1172 			ret = -EINVAL;	/* 1003.1g */
1173 			break;
1174 		}
1175 		if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1176 			ret = -EFAULT;
1177 			break;
1178 		}
1179 		if (!ling.l_onoff)
1180 			sock_reset_flag(sk, SOCK_LINGER);
1181 		else {
1182 #if (BITS_PER_LONG == 32)
1183 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1184 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1185 			else
1186 #endif
1187 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1188 			sock_set_flag(sk, SOCK_LINGER);
1189 		}
1190 		break;
1191 
1192 	case SO_BSDCOMPAT:
1193 		break;
1194 
1195 	case SO_PASSCRED:
1196 		if (valbool)
1197 			set_bit(SOCK_PASSCRED, &sock->flags);
1198 		else
1199 			clear_bit(SOCK_PASSCRED, &sock->flags);
1200 		break;
1201 
1202 	case SO_TIMESTAMP_OLD:
1203 	case SO_TIMESTAMP_NEW:
1204 	case SO_TIMESTAMPNS_OLD:
1205 	case SO_TIMESTAMPNS_NEW:
1206 		sock_set_timestamp(sk, optname, valbool);
1207 		break;
1208 
1209 	case SO_TIMESTAMPING_NEW:
1210 	case SO_TIMESTAMPING_OLD:
1211 		if (optlen == sizeof(timestamping)) {
1212 			if (copy_from_sockptr(&timestamping, optval,
1213 					      sizeof(timestamping))) {
1214 				ret = -EFAULT;
1215 				break;
1216 			}
1217 		} else {
1218 			memset(&timestamping, 0, sizeof(timestamping));
1219 			timestamping.flags = val;
1220 		}
1221 		ret = sock_set_timestamping(sk, optname, timestamping);
1222 		break;
1223 
1224 	case SO_RCVLOWAT:
1225 		if (val < 0)
1226 			val = INT_MAX;
1227 		if (sock->ops->set_rcvlowat)
1228 			ret = sock->ops->set_rcvlowat(sk, val);
1229 		else
1230 			WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1231 		break;
1232 
1233 	case SO_RCVTIMEO_OLD:
1234 	case SO_RCVTIMEO_NEW:
1235 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1236 				       optlen, optname == SO_RCVTIMEO_OLD);
1237 		break;
1238 
1239 	case SO_SNDTIMEO_OLD:
1240 	case SO_SNDTIMEO_NEW:
1241 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1242 				       optlen, optname == SO_SNDTIMEO_OLD);
1243 		break;
1244 
1245 	case SO_ATTACH_FILTER: {
1246 		struct sock_fprog fprog;
1247 
1248 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1249 		if (!ret)
1250 			ret = sk_attach_filter(&fprog, sk);
1251 		break;
1252 	}
1253 	case SO_ATTACH_BPF:
1254 		ret = -EINVAL;
1255 		if (optlen == sizeof(u32)) {
1256 			u32 ufd;
1257 
1258 			ret = -EFAULT;
1259 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1260 				break;
1261 
1262 			ret = sk_attach_bpf(ufd, sk);
1263 		}
1264 		break;
1265 
1266 	case SO_ATTACH_REUSEPORT_CBPF: {
1267 		struct sock_fprog fprog;
1268 
1269 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1270 		if (!ret)
1271 			ret = sk_reuseport_attach_filter(&fprog, sk);
1272 		break;
1273 	}
1274 	case SO_ATTACH_REUSEPORT_EBPF:
1275 		ret = -EINVAL;
1276 		if (optlen == sizeof(u32)) {
1277 			u32 ufd;
1278 
1279 			ret = -EFAULT;
1280 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1281 				break;
1282 
1283 			ret = sk_reuseport_attach_bpf(ufd, sk);
1284 		}
1285 		break;
1286 
1287 	case SO_DETACH_REUSEPORT_BPF:
1288 		ret = reuseport_detach_prog(sk);
1289 		break;
1290 
1291 	case SO_DETACH_FILTER:
1292 		ret = sk_detach_filter(sk);
1293 		break;
1294 
1295 	case SO_LOCK_FILTER:
1296 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1297 			ret = -EPERM;
1298 		else
1299 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1300 		break;
1301 
1302 	case SO_PASSSEC:
1303 		if (valbool)
1304 			set_bit(SOCK_PASSSEC, &sock->flags);
1305 		else
1306 			clear_bit(SOCK_PASSSEC, &sock->flags);
1307 		break;
1308 	case SO_MARK:
1309 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1310 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1311 			ret = -EPERM;
1312 			break;
1313 		}
1314 
1315 		__sock_set_mark(sk, val);
1316 		break;
1317 	case SO_RCVMARK:
1318 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1319 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1320 			ret = -EPERM;
1321 			break;
1322 		}
1323 
1324 		sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1325 		break;
1326 
1327 	case SO_RXQ_OVFL:
1328 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1329 		break;
1330 
1331 	case SO_WIFI_STATUS:
1332 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1333 		break;
1334 
1335 	case SO_PEEK_OFF:
1336 		if (sock->ops->set_peek_off)
1337 			ret = sock->ops->set_peek_off(sk, val);
1338 		else
1339 			ret = -EOPNOTSUPP;
1340 		break;
1341 
1342 	case SO_NOFCS:
1343 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1344 		break;
1345 
1346 	case SO_SELECT_ERR_QUEUE:
1347 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1348 		break;
1349 
1350 #ifdef CONFIG_NET_RX_BUSY_POLL
1351 	case SO_BUSY_POLL:
1352 		/* allow unprivileged users to decrease the value */
1353 		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1354 			ret = -EPERM;
1355 		else {
1356 			if (val < 0)
1357 				ret = -EINVAL;
1358 			else
1359 				WRITE_ONCE(sk->sk_ll_usec, val);
1360 		}
1361 		break;
1362 	case SO_PREFER_BUSY_POLL:
1363 		if (valbool && !capable(CAP_NET_ADMIN))
1364 			ret = -EPERM;
1365 		else
1366 			WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1367 		break;
1368 	case SO_BUSY_POLL_BUDGET:
1369 		if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1370 			ret = -EPERM;
1371 		} else {
1372 			if (val < 0 || val > U16_MAX)
1373 				ret = -EINVAL;
1374 			else
1375 				WRITE_ONCE(sk->sk_busy_poll_budget, val);
1376 		}
1377 		break;
1378 #endif
1379 
1380 	case SO_MAX_PACING_RATE:
1381 		{
1382 		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1383 
1384 		if (sizeof(ulval) != sizeof(val) &&
1385 		    optlen >= sizeof(ulval) &&
1386 		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1387 			ret = -EFAULT;
1388 			break;
1389 		}
1390 		if (ulval != ~0UL)
1391 			cmpxchg(&sk->sk_pacing_status,
1392 				SK_PACING_NONE,
1393 				SK_PACING_NEEDED);
1394 		sk->sk_max_pacing_rate = ulval;
1395 		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1396 		break;
1397 		}
1398 	case SO_INCOMING_CPU:
1399 		WRITE_ONCE(sk->sk_incoming_cpu, val);
1400 		break;
1401 
1402 	case SO_CNX_ADVICE:
1403 		if (val == 1)
1404 			dst_negative_advice(sk);
1405 		break;
1406 
1407 	case SO_ZEROCOPY:
1408 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1409 			if (!(sk_is_tcp(sk) ||
1410 			      (sk->sk_type == SOCK_DGRAM &&
1411 			       sk->sk_protocol == IPPROTO_UDP)))
1412 				ret = -EOPNOTSUPP;
1413 		} else if (sk->sk_family != PF_RDS) {
1414 			ret = -EOPNOTSUPP;
1415 		}
1416 		if (!ret) {
1417 			if (val < 0 || val > 1)
1418 				ret = -EINVAL;
1419 			else
1420 				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1421 		}
1422 		break;
1423 
1424 	case SO_TXTIME:
1425 		if (optlen != sizeof(struct sock_txtime)) {
1426 			ret = -EINVAL;
1427 			break;
1428 		} else if (copy_from_sockptr(&sk_txtime, optval,
1429 			   sizeof(struct sock_txtime))) {
1430 			ret = -EFAULT;
1431 			break;
1432 		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1433 			ret = -EINVAL;
1434 			break;
1435 		}
1436 		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1437 		 * scheduler has enough safe guards.
1438 		 */
1439 		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1440 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1441 			ret = -EPERM;
1442 			break;
1443 		}
1444 		sock_valbool_flag(sk, SOCK_TXTIME, true);
1445 		sk->sk_clockid = sk_txtime.clockid;
1446 		sk->sk_txtime_deadline_mode =
1447 			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1448 		sk->sk_txtime_report_errors =
1449 			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1450 		break;
1451 
1452 	case SO_BINDTOIFINDEX:
1453 		ret = sock_bindtoindex_locked(sk, val);
1454 		break;
1455 
1456 	case SO_BUF_LOCK:
1457 		if (val & ~SOCK_BUF_LOCK_MASK) {
1458 			ret = -EINVAL;
1459 			break;
1460 		}
1461 		sk->sk_userlocks = val | (sk->sk_userlocks &
1462 					  ~SOCK_BUF_LOCK_MASK);
1463 		break;
1464 
1465 	case SO_RESERVE_MEM:
1466 	{
1467 		int delta;
1468 
1469 		if (val < 0) {
1470 			ret = -EINVAL;
1471 			break;
1472 		}
1473 
1474 		delta = val - sk->sk_reserved_mem;
1475 		if (delta < 0)
1476 			sock_release_reserved_memory(sk, -delta);
1477 		else
1478 			ret = sock_reserve_memory(sk, delta);
1479 		break;
1480 	}
1481 
1482 	case SO_TXREHASH:
1483 		if (val < -1 || val > 1) {
1484 			ret = -EINVAL;
1485 			break;
1486 		}
1487 		/* Paired with READ_ONCE() in tcp_rtx_synack() */
1488 		WRITE_ONCE(sk->sk_txrehash, (u8)val);
1489 		break;
1490 
1491 	default:
1492 		ret = -ENOPROTOOPT;
1493 		break;
1494 	}
1495 	release_sock(sk);
1496 	return ret;
1497 }
1498 EXPORT_SYMBOL(sock_setsockopt);
1499 
1500 static const struct cred *sk_get_peer_cred(struct sock *sk)
1501 {
1502 	const struct cred *cred;
1503 
1504 	spin_lock(&sk->sk_peer_lock);
1505 	cred = get_cred(sk->sk_peer_cred);
1506 	spin_unlock(&sk->sk_peer_lock);
1507 
1508 	return cred;
1509 }
1510 
1511 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1512 			  struct ucred *ucred)
1513 {
1514 	ucred->pid = pid_vnr(pid);
1515 	ucred->uid = ucred->gid = -1;
1516 	if (cred) {
1517 		struct user_namespace *current_ns = current_user_ns();
1518 
1519 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1520 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1521 	}
1522 }
1523 
1524 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1525 {
1526 	struct user_namespace *user_ns = current_user_ns();
1527 	int i;
1528 
1529 	for (i = 0; i < src->ngroups; i++)
1530 		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1531 			return -EFAULT;
1532 
1533 	return 0;
1534 }
1535 
1536 int sock_getsockopt(struct socket *sock, int level, int optname,
1537 		    char __user *optval, int __user *optlen)
1538 {
1539 	struct sock *sk = sock->sk;
1540 
1541 	union {
1542 		int val;
1543 		u64 val64;
1544 		unsigned long ulval;
1545 		struct linger ling;
1546 		struct old_timeval32 tm32;
1547 		struct __kernel_old_timeval tm;
1548 		struct  __kernel_sock_timeval stm;
1549 		struct sock_txtime txtime;
1550 		struct so_timestamping timestamping;
1551 	} v;
1552 
1553 	int lv = sizeof(int);
1554 	int len;
1555 
1556 	if (get_user(len, optlen))
1557 		return -EFAULT;
1558 	if (len < 0)
1559 		return -EINVAL;
1560 
1561 	memset(&v, 0, sizeof(v));
1562 
1563 	switch (optname) {
1564 	case SO_DEBUG:
1565 		v.val = sock_flag(sk, SOCK_DBG);
1566 		break;
1567 
1568 	case SO_DONTROUTE:
1569 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1570 		break;
1571 
1572 	case SO_BROADCAST:
1573 		v.val = sock_flag(sk, SOCK_BROADCAST);
1574 		break;
1575 
1576 	case SO_SNDBUF:
1577 		v.val = sk->sk_sndbuf;
1578 		break;
1579 
1580 	case SO_RCVBUF:
1581 		v.val = sk->sk_rcvbuf;
1582 		break;
1583 
1584 	case SO_REUSEADDR:
1585 		v.val = sk->sk_reuse;
1586 		break;
1587 
1588 	case SO_REUSEPORT:
1589 		v.val = sk->sk_reuseport;
1590 		break;
1591 
1592 	case SO_KEEPALIVE:
1593 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1594 		break;
1595 
1596 	case SO_TYPE:
1597 		v.val = sk->sk_type;
1598 		break;
1599 
1600 	case SO_PROTOCOL:
1601 		v.val = sk->sk_protocol;
1602 		break;
1603 
1604 	case SO_DOMAIN:
1605 		v.val = sk->sk_family;
1606 		break;
1607 
1608 	case SO_ERROR:
1609 		v.val = -sock_error(sk);
1610 		if (v.val == 0)
1611 			v.val = xchg(&sk->sk_err_soft, 0);
1612 		break;
1613 
1614 	case SO_OOBINLINE:
1615 		v.val = sock_flag(sk, SOCK_URGINLINE);
1616 		break;
1617 
1618 	case SO_NO_CHECK:
1619 		v.val = sk->sk_no_check_tx;
1620 		break;
1621 
1622 	case SO_PRIORITY:
1623 		v.val = sk->sk_priority;
1624 		break;
1625 
1626 	case SO_LINGER:
1627 		lv		= sizeof(v.ling);
1628 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1629 		v.ling.l_linger	= sk->sk_lingertime / HZ;
1630 		break;
1631 
1632 	case SO_BSDCOMPAT:
1633 		break;
1634 
1635 	case SO_TIMESTAMP_OLD:
1636 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1637 				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1638 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1639 		break;
1640 
1641 	case SO_TIMESTAMPNS_OLD:
1642 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1643 		break;
1644 
1645 	case SO_TIMESTAMP_NEW:
1646 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1647 		break;
1648 
1649 	case SO_TIMESTAMPNS_NEW:
1650 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1651 		break;
1652 
1653 	case SO_TIMESTAMPING_OLD:
1654 		lv = sizeof(v.timestamping);
1655 		v.timestamping.flags = sk->sk_tsflags;
1656 		v.timestamping.bind_phc = sk->sk_bind_phc;
1657 		break;
1658 
1659 	case SO_RCVTIMEO_OLD:
1660 	case SO_RCVTIMEO_NEW:
1661 		lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1662 		break;
1663 
1664 	case SO_SNDTIMEO_OLD:
1665 	case SO_SNDTIMEO_NEW:
1666 		lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1667 		break;
1668 
1669 	case SO_RCVLOWAT:
1670 		v.val = sk->sk_rcvlowat;
1671 		break;
1672 
1673 	case SO_SNDLOWAT:
1674 		v.val = 1;
1675 		break;
1676 
1677 	case SO_PASSCRED:
1678 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1679 		break;
1680 
1681 	case SO_PEERCRED:
1682 	{
1683 		struct ucred peercred;
1684 		if (len > sizeof(peercred))
1685 			len = sizeof(peercred);
1686 
1687 		spin_lock(&sk->sk_peer_lock);
1688 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1689 		spin_unlock(&sk->sk_peer_lock);
1690 
1691 		if (copy_to_user(optval, &peercred, len))
1692 			return -EFAULT;
1693 		goto lenout;
1694 	}
1695 
1696 	case SO_PEERGROUPS:
1697 	{
1698 		const struct cred *cred;
1699 		int ret, n;
1700 
1701 		cred = sk_get_peer_cred(sk);
1702 		if (!cred)
1703 			return -ENODATA;
1704 
1705 		n = cred->group_info->ngroups;
1706 		if (len < n * sizeof(gid_t)) {
1707 			len = n * sizeof(gid_t);
1708 			put_cred(cred);
1709 			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1710 		}
1711 		len = n * sizeof(gid_t);
1712 
1713 		ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1714 		put_cred(cred);
1715 		if (ret)
1716 			return ret;
1717 		goto lenout;
1718 	}
1719 
1720 	case SO_PEERNAME:
1721 	{
1722 		char address[128];
1723 
1724 		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1725 		if (lv < 0)
1726 			return -ENOTCONN;
1727 		if (lv < len)
1728 			return -EINVAL;
1729 		if (copy_to_user(optval, address, len))
1730 			return -EFAULT;
1731 		goto lenout;
1732 	}
1733 
1734 	/* Dubious BSD thing... Probably nobody even uses it, but
1735 	 * the UNIX standard wants it for whatever reason... -DaveM
1736 	 */
1737 	case SO_ACCEPTCONN:
1738 		v.val = sk->sk_state == TCP_LISTEN;
1739 		break;
1740 
1741 	case SO_PASSSEC:
1742 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1743 		break;
1744 
1745 	case SO_PEERSEC:
1746 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1747 
1748 	case SO_MARK:
1749 		v.val = sk->sk_mark;
1750 		break;
1751 
1752 	case SO_RCVMARK:
1753 		v.val = sock_flag(sk, SOCK_RCVMARK);
1754 		break;
1755 
1756 	case SO_RXQ_OVFL:
1757 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1758 		break;
1759 
1760 	case SO_WIFI_STATUS:
1761 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1762 		break;
1763 
1764 	case SO_PEEK_OFF:
1765 		if (!sock->ops->set_peek_off)
1766 			return -EOPNOTSUPP;
1767 
1768 		v.val = sk->sk_peek_off;
1769 		break;
1770 	case SO_NOFCS:
1771 		v.val = sock_flag(sk, SOCK_NOFCS);
1772 		break;
1773 
1774 	case SO_BINDTODEVICE:
1775 		return sock_getbindtodevice(sk, optval, optlen, len);
1776 
1777 	case SO_GET_FILTER:
1778 		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1779 		if (len < 0)
1780 			return len;
1781 
1782 		goto lenout;
1783 
1784 	case SO_LOCK_FILTER:
1785 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1786 		break;
1787 
1788 	case SO_BPF_EXTENSIONS:
1789 		v.val = bpf_tell_extensions();
1790 		break;
1791 
1792 	case SO_SELECT_ERR_QUEUE:
1793 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1794 		break;
1795 
1796 #ifdef CONFIG_NET_RX_BUSY_POLL
1797 	case SO_BUSY_POLL:
1798 		v.val = sk->sk_ll_usec;
1799 		break;
1800 	case SO_PREFER_BUSY_POLL:
1801 		v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1802 		break;
1803 #endif
1804 
1805 	case SO_MAX_PACING_RATE:
1806 		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1807 			lv = sizeof(v.ulval);
1808 			v.ulval = sk->sk_max_pacing_rate;
1809 		} else {
1810 			/* 32bit version */
1811 			v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1812 		}
1813 		break;
1814 
1815 	case SO_INCOMING_CPU:
1816 		v.val = READ_ONCE(sk->sk_incoming_cpu);
1817 		break;
1818 
1819 	case SO_MEMINFO:
1820 	{
1821 		u32 meminfo[SK_MEMINFO_VARS];
1822 
1823 		sk_get_meminfo(sk, meminfo);
1824 
1825 		len = min_t(unsigned int, len, sizeof(meminfo));
1826 		if (copy_to_user(optval, &meminfo, len))
1827 			return -EFAULT;
1828 
1829 		goto lenout;
1830 	}
1831 
1832 #ifdef CONFIG_NET_RX_BUSY_POLL
1833 	case SO_INCOMING_NAPI_ID:
1834 		v.val = READ_ONCE(sk->sk_napi_id);
1835 
1836 		/* aggregate non-NAPI IDs down to 0 */
1837 		if (v.val < MIN_NAPI_ID)
1838 			v.val = 0;
1839 
1840 		break;
1841 #endif
1842 
1843 	case SO_COOKIE:
1844 		lv = sizeof(u64);
1845 		if (len < lv)
1846 			return -EINVAL;
1847 		v.val64 = sock_gen_cookie(sk);
1848 		break;
1849 
1850 	case SO_ZEROCOPY:
1851 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1852 		break;
1853 
1854 	case SO_TXTIME:
1855 		lv = sizeof(v.txtime);
1856 		v.txtime.clockid = sk->sk_clockid;
1857 		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1858 				  SOF_TXTIME_DEADLINE_MODE : 0;
1859 		v.txtime.flags |= sk->sk_txtime_report_errors ?
1860 				  SOF_TXTIME_REPORT_ERRORS : 0;
1861 		break;
1862 
1863 	case SO_BINDTOIFINDEX:
1864 		v.val = sk->sk_bound_dev_if;
1865 		break;
1866 
1867 	case SO_NETNS_COOKIE:
1868 		lv = sizeof(u64);
1869 		if (len != lv)
1870 			return -EINVAL;
1871 		v.val64 = sock_net(sk)->net_cookie;
1872 		break;
1873 
1874 	case SO_BUF_LOCK:
1875 		v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1876 		break;
1877 
1878 	case SO_RESERVE_MEM:
1879 		v.val = sk->sk_reserved_mem;
1880 		break;
1881 
1882 	case SO_TXREHASH:
1883 		v.val = sk->sk_txrehash;
1884 		break;
1885 
1886 	default:
1887 		/* We implement the SO_SNDLOWAT etc to not be settable
1888 		 * (1003.1g 7).
1889 		 */
1890 		return -ENOPROTOOPT;
1891 	}
1892 
1893 	if (len > lv)
1894 		len = lv;
1895 	if (copy_to_user(optval, &v, len))
1896 		return -EFAULT;
1897 lenout:
1898 	if (put_user(len, optlen))
1899 		return -EFAULT;
1900 	return 0;
1901 }
1902 
1903 /*
1904  * Initialize an sk_lock.
1905  *
1906  * (We also register the sk_lock with the lock validator.)
1907  */
1908 static inline void sock_lock_init(struct sock *sk)
1909 {
1910 	if (sk->sk_kern_sock)
1911 		sock_lock_init_class_and_name(
1912 			sk,
1913 			af_family_kern_slock_key_strings[sk->sk_family],
1914 			af_family_kern_slock_keys + sk->sk_family,
1915 			af_family_kern_key_strings[sk->sk_family],
1916 			af_family_kern_keys + sk->sk_family);
1917 	else
1918 		sock_lock_init_class_and_name(
1919 			sk,
1920 			af_family_slock_key_strings[sk->sk_family],
1921 			af_family_slock_keys + sk->sk_family,
1922 			af_family_key_strings[sk->sk_family],
1923 			af_family_keys + sk->sk_family);
1924 }
1925 
1926 /*
1927  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1928  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1929  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1930  */
1931 static void sock_copy(struct sock *nsk, const struct sock *osk)
1932 {
1933 	const struct proto *prot = READ_ONCE(osk->sk_prot);
1934 #ifdef CONFIG_SECURITY_NETWORK
1935 	void *sptr = nsk->sk_security;
1936 #endif
1937 
1938 	/* If we move sk_tx_queue_mapping out of the private section,
1939 	 * we must check if sk_tx_queue_clear() is called after
1940 	 * sock_copy() in sk_clone_lock().
1941 	 */
1942 	BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1943 		     offsetof(struct sock, sk_dontcopy_begin) ||
1944 		     offsetof(struct sock, sk_tx_queue_mapping) >=
1945 		     offsetof(struct sock, sk_dontcopy_end));
1946 
1947 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1948 
1949 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1950 	       prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1951 
1952 #ifdef CONFIG_SECURITY_NETWORK
1953 	nsk->sk_security = sptr;
1954 	security_sk_clone(osk, nsk);
1955 #endif
1956 }
1957 
1958 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1959 		int family)
1960 {
1961 	struct sock *sk;
1962 	struct kmem_cache *slab;
1963 
1964 	slab = prot->slab;
1965 	if (slab != NULL) {
1966 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1967 		if (!sk)
1968 			return sk;
1969 		if (want_init_on_alloc(priority))
1970 			sk_prot_clear_nulls(sk, prot->obj_size);
1971 	} else
1972 		sk = kmalloc(prot->obj_size, priority);
1973 
1974 	if (sk != NULL) {
1975 		if (security_sk_alloc(sk, family, priority))
1976 			goto out_free;
1977 
1978 		if (!try_module_get(prot->owner))
1979 			goto out_free_sec;
1980 	}
1981 
1982 	return sk;
1983 
1984 out_free_sec:
1985 	security_sk_free(sk);
1986 out_free:
1987 	if (slab != NULL)
1988 		kmem_cache_free(slab, sk);
1989 	else
1990 		kfree(sk);
1991 	return NULL;
1992 }
1993 
1994 static void sk_prot_free(struct proto *prot, struct sock *sk)
1995 {
1996 	struct kmem_cache *slab;
1997 	struct module *owner;
1998 
1999 	owner = prot->owner;
2000 	slab = prot->slab;
2001 
2002 	cgroup_sk_free(&sk->sk_cgrp_data);
2003 	mem_cgroup_sk_free(sk);
2004 	security_sk_free(sk);
2005 	if (slab != NULL)
2006 		kmem_cache_free(slab, sk);
2007 	else
2008 		kfree(sk);
2009 	module_put(owner);
2010 }
2011 
2012 /**
2013  *	sk_alloc - All socket objects are allocated here
2014  *	@net: the applicable net namespace
2015  *	@family: protocol family
2016  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2017  *	@prot: struct proto associated with this new sock instance
2018  *	@kern: is this to be a kernel socket?
2019  */
2020 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2021 		      struct proto *prot, int kern)
2022 {
2023 	struct sock *sk;
2024 
2025 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2026 	if (sk) {
2027 		sk->sk_family = family;
2028 		/*
2029 		 * See comment in struct sock definition to understand
2030 		 * why we need sk_prot_creator -acme
2031 		 */
2032 		sk->sk_prot = sk->sk_prot_creator = prot;
2033 		sk->sk_kern_sock = kern;
2034 		sock_lock_init(sk);
2035 		sk->sk_net_refcnt = kern ? 0 : 1;
2036 		if (likely(sk->sk_net_refcnt)) {
2037 			get_net_track(net, &sk->ns_tracker, priority);
2038 			sock_inuse_add(net, 1);
2039 		}
2040 
2041 		sock_net_set(sk, net);
2042 		refcount_set(&sk->sk_wmem_alloc, 1);
2043 
2044 		mem_cgroup_sk_alloc(sk);
2045 		cgroup_sk_alloc(&sk->sk_cgrp_data);
2046 		sock_update_classid(&sk->sk_cgrp_data);
2047 		sock_update_netprioidx(&sk->sk_cgrp_data);
2048 		sk_tx_queue_clear(sk);
2049 	}
2050 
2051 	return sk;
2052 }
2053 EXPORT_SYMBOL(sk_alloc);
2054 
2055 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2056  * grace period. This is the case for UDP sockets and TCP listeners.
2057  */
2058 static void __sk_destruct(struct rcu_head *head)
2059 {
2060 	struct sock *sk = container_of(head, struct sock, sk_rcu);
2061 	struct sk_filter *filter;
2062 
2063 	if (sk->sk_destruct)
2064 		sk->sk_destruct(sk);
2065 
2066 	filter = rcu_dereference_check(sk->sk_filter,
2067 				       refcount_read(&sk->sk_wmem_alloc) == 0);
2068 	if (filter) {
2069 		sk_filter_uncharge(sk, filter);
2070 		RCU_INIT_POINTER(sk->sk_filter, NULL);
2071 	}
2072 
2073 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2074 
2075 #ifdef CONFIG_BPF_SYSCALL
2076 	bpf_sk_storage_free(sk);
2077 #endif
2078 
2079 	if (atomic_read(&sk->sk_omem_alloc))
2080 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
2081 			 __func__, atomic_read(&sk->sk_omem_alloc));
2082 
2083 	if (sk->sk_frag.page) {
2084 		put_page(sk->sk_frag.page);
2085 		sk->sk_frag.page = NULL;
2086 	}
2087 
2088 	/* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2089 	put_cred(sk->sk_peer_cred);
2090 	put_pid(sk->sk_peer_pid);
2091 
2092 	if (likely(sk->sk_net_refcnt))
2093 		put_net_track(sock_net(sk), &sk->ns_tracker);
2094 	sk_prot_free(sk->sk_prot_creator, sk);
2095 }
2096 
2097 void sk_destruct(struct sock *sk)
2098 {
2099 	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2100 
2101 	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2102 		reuseport_detach_sock(sk);
2103 		use_call_rcu = true;
2104 	}
2105 
2106 	if (use_call_rcu)
2107 		call_rcu(&sk->sk_rcu, __sk_destruct);
2108 	else
2109 		__sk_destruct(&sk->sk_rcu);
2110 }
2111 
2112 static void __sk_free(struct sock *sk)
2113 {
2114 	if (likely(sk->sk_net_refcnt))
2115 		sock_inuse_add(sock_net(sk), -1);
2116 
2117 	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2118 		sock_diag_broadcast_destroy(sk);
2119 	else
2120 		sk_destruct(sk);
2121 }
2122 
2123 void sk_free(struct sock *sk)
2124 {
2125 	/*
2126 	 * We subtract one from sk_wmem_alloc and can know if
2127 	 * some packets are still in some tx queue.
2128 	 * If not null, sock_wfree() will call __sk_free(sk) later
2129 	 */
2130 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2131 		__sk_free(sk);
2132 }
2133 EXPORT_SYMBOL(sk_free);
2134 
2135 static void sk_init_common(struct sock *sk)
2136 {
2137 	skb_queue_head_init(&sk->sk_receive_queue);
2138 	skb_queue_head_init(&sk->sk_write_queue);
2139 	skb_queue_head_init(&sk->sk_error_queue);
2140 
2141 	rwlock_init(&sk->sk_callback_lock);
2142 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2143 			af_rlock_keys + sk->sk_family,
2144 			af_family_rlock_key_strings[sk->sk_family]);
2145 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2146 			af_wlock_keys + sk->sk_family,
2147 			af_family_wlock_key_strings[sk->sk_family]);
2148 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2149 			af_elock_keys + sk->sk_family,
2150 			af_family_elock_key_strings[sk->sk_family]);
2151 	lockdep_set_class_and_name(&sk->sk_callback_lock,
2152 			af_callback_keys + sk->sk_family,
2153 			af_family_clock_key_strings[sk->sk_family]);
2154 }
2155 
2156 /**
2157  *	sk_clone_lock - clone a socket, and lock its clone
2158  *	@sk: the socket to clone
2159  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2160  *
2161  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2162  */
2163 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2164 {
2165 	struct proto *prot = READ_ONCE(sk->sk_prot);
2166 	struct sk_filter *filter;
2167 	bool is_charged = true;
2168 	struct sock *newsk;
2169 
2170 	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2171 	if (!newsk)
2172 		goto out;
2173 
2174 	sock_copy(newsk, sk);
2175 
2176 	newsk->sk_prot_creator = prot;
2177 
2178 	/* SANITY */
2179 	if (likely(newsk->sk_net_refcnt)) {
2180 		get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2181 		sock_inuse_add(sock_net(newsk), 1);
2182 	}
2183 	sk_node_init(&newsk->sk_node);
2184 	sock_lock_init(newsk);
2185 	bh_lock_sock(newsk);
2186 	newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
2187 	newsk->sk_backlog.len = 0;
2188 
2189 	atomic_set(&newsk->sk_rmem_alloc, 0);
2190 
2191 	/* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2192 	refcount_set(&newsk->sk_wmem_alloc, 1);
2193 
2194 	atomic_set(&newsk->sk_omem_alloc, 0);
2195 	sk_init_common(newsk);
2196 
2197 	newsk->sk_dst_cache	= NULL;
2198 	newsk->sk_dst_pending_confirm = 0;
2199 	newsk->sk_wmem_queued	= 0;
2200 	newsk->sk_forward_alloc = 0;
2201 	newsk->sk_reserved_mem  = 0;
2202 	atomic_set(&newsk->sk_drops, 0);
2203 	newsk->sk_send_head	= NULL;
2204 	newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2205 	atomic_set(&newsk->sk_zckey, 0);
2206 
2207 	sock_reset_flag(newsk, SOCK_DONE);
2208 
2209 	/* sk->sk_memcg will be populated at accept() time */
2210 	newsk->sk_memcg = NULL;
2211 
2212 	cgroup_sk_clone(&newsk->sk_cgrp_data);
2213 
2214 	rcu_read_lock();
2215 	filter = rcu_dereference(sk->sk_filter);
2216 	if (filter != NULL)
2217 		/* though it's an empty new sock, the charging may fail
2218 		 * if sysctl_optmem_max was changed between creation of
2219 		 * original socket and cloning
2220 		 */
2221 		is_charged = sk_filter_charge(newsk, filter);
2222 	RCU_INIT_POINTER(newsk->sk_filter, filter);
2223 	rcu_read_unlock();
2224 
2225 	if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2226 		/* We need to make sure that we don't uncharge the new
2227 		 * socket if we couldn't charge it in the first place
2228 		 * as otherwise we uncharge the parent's filter.
2229 		 */
2230 		if (!is_charged)
2231 			RCU_INIT_POINTER(newsk->sk_filter, NULL);
2232 		sk_free_unlock_clone(newsk);
2233 		newsk = NULL;
2234 		goto out;
2235 	}
2236 	RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2237 
2238 	if (bpf_sk_storage_clone(sk, newsk)) {
2239 		sk_free_unlock_clone(newsk);
2240 		newsk = NULL;
2241 		goto out;
2242 	}
2243 
2244 	/* Clear sk_user_data if parent had the pointer tagged
2245 	 * as not suitable for copying when cloning.
2246 	 */
2247 	if (sk_user_data_is_nocopy(newsk))
2248 		newsk->sk_user_data = NULL;
2249 
2250 	newsk->sk_err	   = 0;
2251 	newsk->sk_err_soft = 0;
2252 	newsk->sk_priority = 0;
2253 	newsk->sk_incoming_cpu = raw_smp_processor_id();
2254 
2255 	/* Before updating sk_refcnt, we must commit prior changes to memory
2256 	 * (Documentation/RCU/rculist_nulls.rst for details)
2257 	 */
2258 	smp_wmb();
2259 	refcount_set(&newsk->sk_refcnt, 2);
2260 
2261 	/* Increment the counter in the same struct proto as the master
2262 	 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2263 	 * is the same as sk->sk_prot->socks, as this field was copied
2264 	 * with memcpy).
2265 	 *
2266 	 * This _changes_ the previous behaviour, where
2267 	 * tcp_create_openreq_child always was incrementing the
2268 	 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2269 	 * to be taken into account in all callers. -acme
2270 	 */
2271 	sk_refcnt_debug_inc(newsk);
2272 	sk_set_socket(newsk, NULL);
2273 	sk_tx_queue_clear(newsk);
2274 	RCU_INIT_POINTER(newsk->sk_wq, NULL);
2275 
2276 	if (newsk->sk_prot->sockets_allocated)
2277 		sk_sockets_allocated_inc(newsk);
2278 
2279 	if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2280 		net_enable_timestamp();
2281 out:
2282 	return newsk;
2283 }
2284 EXPORT_SYMBOL_GPL(sk_clone_lock);
2285 
2286 void sk_free_unlock_clone(struct sock *sk)
2287 {
2288 	/* It is still raw copy of parent, so invalidate
2289 	 * destructor and make plain sk_free() */
2290 	sk->sk_destruct = NULL;
2291 	bh_unlock_sock(sk);
2292 	sk_free(sk);
2293 }
2294 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2295 
2296 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2297 {
2298 	u32 max_segs = 1;
2299 
2300 	sk_dst_set(sk, dst);
2301 	sk->sk_route_caps = dst->dev->features;
2302 	if (sk_is_tcp(sk))
2303 		sk->sk_route_caps |= NETIF_F_GSO;
2304 	if (sk->sk_route_caps & NETIF_F_GSO)
2305 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2306 	if (unlikely(sk->sk_gso_disabled))
2307 		sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2308 	if (sk_can_gso(sk)) {
2309 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2310 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2311 		} else {
2312 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2313 			/* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2314 			sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2315 			sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1);
2316 			/* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2317 			max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2318 		}
2319 	}
2320 	sk->sk_gso_max_segs = max_segs;
2321 }
2322 EXPORT_SYMBOL_GPL(sk_setup_caps);
2323 
2324 /*
2325  *	Simple resource managers for sockets.
2326  */
2327 
2328 
2329 /*
2330  * Write buffer destructor automatically called from kfree_skb.
2331  */
2332 void sock_wfree(struct sk_buff *skb)
2333 {
2334 	struct sock *sk = skb->sk;
2335 	unsigned int len = skb->truesize;
2336 	bool free;
2337 
2338 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2339 		if (sock_flag(sk, SOCK_RCU_FREE) &&
2340 		    sk->sk_write_space == sock_def_write_space) {
2341 			rcu_read_lock();
2342 			free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2343 			sock_def_write_space_wfree(sk);
2344 			rcu_read_unlock();
2345 			if (unlikely(free))
2346 				__sk_free(sk);
2347 			return;
2348 		}
2349 
2350 		/*
2351 		 * Keep a reference on sk_wmem_alloc, this will be released
2352 		 * after sk_write_space() call
2353 		 */
2354 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2355 		sk->sk_write_space(sk);
2356 		len = 1;
2357 	}
2358 	/*
2359 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2360 	 * could not do because of in-flight packets
2361 	 */
2362 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2363 		__sk_free(sk);
2364 }
2365 EXPORT_SYMBOL(sock_wfree);
2366 
2367 /* This variant of sock_wfree() is used by TCP,
2368  * since it sets SOCK_USE_WRITE_QUEUE.
2369  */
2370 void __sock_wfree(struct sk_buff *skb)
2371 {
2372 	struct sock *sk = skb->sk;
2373 
2374 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2375 		__sk_free(sk);
2376 }
2377 
2378 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2379 {
2380 	skb_orphan(skb);
2381 	skb->sk = sk;
2382 #ifdef CONFIG_INET
2383 	if (unlikely(!sk_fullsock(sk))) {
2384 		skb->destructor = sock_edemux;
2385 		sock_hold(sk);
2386 		return;
2387 	}
2388 #endif
2389 	skb->destructor = sock_wfree;
2390 	skb_set_hash_from_sk(skb, sk);
2391 	/*
2392 	 * We used to take a refcount on sk, but following operation
2393 	 * is enough to guarantee sk_free() wont free this sock until
2394 	 * all in-flight packets are completed
2395 	 */
2396 	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2397 }
2398 EXPORT_SYMBOL(skb_set_owner_w);
2399 
2400 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2401 {
2402 #ifdef CONFIG_TLS_DEVICE
2403 	/* Drivers depend on in-order delivery for crypto offload,
2404 	 * partial orphan breaks out-of-order-OK logic.
2405 	 */
2406 	if (skb->decrypted)
2407 		return false;
2408 #endif
2409 	return (skb->destructor == sock_wfree ||
2410 		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2411 }
2412 
2413 /* This helper is used by netem, as it can hold packets in its
2414  * delay queue. We want to allow the owner socket to send more
2415  * packets, as if they were already TX completed by a typical driver.
2416  * But we also want to keep skb->sk set because some packet schedulers
2417  * rely on it (sch_fq for example).
2418  */
2419 void skb_orphan_partial(struct sk_buff *skb)
2420 {
2421 	if (skb_is_tcp_pure_ack(skb))
2422 		return;
2423 
2424 	if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2425 		return;
2426 
2427 	skb_orphan(skb);
2428 }
2429 EXPORT_SYMBOL(skb_orphan_partial);
2430 
2431 /*
2432  * Read buffer destructor automatically called from kfree_skb.
2433  */
2434 void sock_rfree(struct sk_buff *skb)
2435 {
2436 	struct sock *sk = skb->sk;
2437 	unsigned int len = skb->truesize;
2438 
2439 	atomic_sub(len, &sk->sk_rmem_alloc);
2440 	sk_mem_uncharge(sk, len);
2441 }
2442 EXPORT_SYMBOL(sock_rfree);
2443 
2444 /*
2445  * Buffer destructor for skbs that are not used directly in read or write
2446  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2447  */
2448 void sock_efree(struct sk_buff *skb)
2449 {
2450 	sock_put(skb->sk);
2451 }
2452 EXPORT_SYMBOL(sock_efree);
2453 
2454 /* Buffer destructor for prefetch/receive path where reference count may
2455  * not be held, e.g. for listen sockets.
2456  */
2457 #ifdef CONFIG_INET
2458 void sock_pfree(struct sk_buff *skb)
2459 {
2460 	if (sk_is_refcounted(skb->sk))
2461 		sock_gen_put(skb->sk);
2462 }
2463 EXPORT_SYMBOL(sock_pfree);
2464 #endif /* CONFIG_INET */
2465 
2466 kuid_t sock_i_uid(struct sock *sk)
2467 {
2468 	kuid_t uid;
2469 
2470 	read_lock_bh(&sk->sk_callback_lock);
2471 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2472 	read_unlock_bh(&sk->sk_callback_lock);
2473 	return uid;
2474 }
2475 EXPORT_SYMBOL(sock_i_uid);
2476 
2477 unsigned long sock_i_ino(struct sock *sk)
2478 {
2479 	unsigned long ino;
2480 
2481 	read_lock_bh(&sk->sk_callback_lock);
2482 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2483 	read_unlock_bh(&sk->sk_callback_lock);
2484 	return ino;
2485 }
2486 EXPORT_SYMBOL(sock_i_ino);
2487 
2488 /*
2489  * Allocate a skb from the socket's send buffer.
2490  */
2491 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2492 			     gfp_t priority)
2493 {
2494 	if (force ||
2495 	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2496 		struct sk_buff *skb = alloc_skb(size, priority);
2497 
2498 		if (skb) {
2499 			skb_set_owner_w(skb, sk);
2500 			return skb;
2501 		}
2502 	}
2503 	return NULL;
2504 }
2505 EXPORT_SYMBOL(sock_wmalloc);
2506 
2507 static void sock_ofree(struct sk_buff *skb)
2508 {
2509 	struct sock *sk = skb->sk;
2510 
2511 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2512 }
2513 
2514 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2515 			     gfp_t priority)
2516 {
2517 	struct sk_buff *skb;
2518 
2519 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2520 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2521 	    sysctl_optmem_max)
2522 		return NULL;
2523 
2524 	skb = alloc_skb(size, priority);
2525 	if (!skb)
2526 		return NULL;
2527 
2528 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2529 	skb->sk = sk;
2530 	skb->destructor = sock_ofree;
2531 	return skb;
2532 }
2533 
2534 /*
2535  * Allocate a memory block from the socket's option memory buffer.
2536  */
2537 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2538 {
2539 	if ((unsigned int)size <= sysctl_optmem_max &&
2540 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2541 		void *mem;
2542 		/* First do the add, to avoid the race if kmalloc
2543 		 * might sleep.
2544 		 */
2545 		atomic_add(size, &sk->sk_omem_alloc);
2546 		mem = kmalloc(size, priority);
2547 		if (mem)
2548 			return mem;
2549 		atomic_sub(size, &sk->sk_omem_alloc);
2550 	}
2551 	return NULL;
2552 }
2553 EXPORT_SYMBOL(sock_kmalloc);
2554 
2555 /* Free an option memory block. Note, we actually want the inline
2556  * here as this allows gcc to detect the nullify and fold away the
2557  * condition entirely.
2558  */
2559 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2560 				  const bool nullify)
2561 {
2562 	if (WARN_ON_ONCE(!mem))
2563 		return;
2564 	if (nullify)
2565 		kfree_sensitive(mem);
2566 	else
2567 		kfree(mem);
2568 	atomic_sub(size, &sk->sk_omem_alloc);
2569 }
2570 
2571 void sock_kfree_s(struct sock *sk, void *mem, int size)
2572 {
2573 	__sock_kfree_s(sk, mem, size, false);
2574 }
2575 EXPORT_SYMBOL(sock_kfree_s);
2576 
2577 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2578 {
2579 	__sock_kfree_s(sk, mem, size, true);
2580 }
2581 EXPORT_SYMBOL(sock_kzfree_s);
2582 
2583 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2584    I think, these locks should be removed for datagram sockets.
2585  */
2586 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2587 {
2588 	DEFINE_WAIT(wait);
2589 
2590 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2591 	for (;;) {
2592 		if (!timeo)
2593 			break;
2594 		if (signal_pending(current))
2595 			break;
2596 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2597 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2598 		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2599 			break;
2600 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2601 			break;
2602 		if (sk->sk_err)
2603 			break;
2604 		timeo = schedule_timeout(timeo);
2605 	}
2606 	finish_wait(sk_sleep(sk), &wait);
2607 	return timeo;
2608 }
2609 
2610 
2611 /*
2612  *	Generic send/receive buffer handlers
2613  */
2614 
2615 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2616 				     unsigned long data_len, int noblock,
2617 				     int *errcode, int max_page_order)
2618 {
2619 	struct sk_buff *skb;
2620 	long timeo;
2621 	int err;
2622 
2623 	timeo = sock_sndtimeo(sk, noblock);
2624 	for (;;) {
2625 		err = sock_error(sk);
2626 		if (err != 0)
2627 			goto failure;
2628 
2629 		err = -EPIPE;
2630 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2631 			goto failure;
2632 
2633 		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2634 			break;
2635 
2636 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2637 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2638 		err = -EAGAIN;
2639 		if (!timeo)
2640 			goto failure;
2641 		if (signal_pending(current))
2642 			goto interrupted;
2643 		timeo = sock_wait_for_wmem(sk, timeo);
2644 	}
2645 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2646 				   errcode, sk->sk_allocation);
2647 	if (skb)
2648 		skb_set_owner_w(skb, sk);
2649 	return skb;
2650 
2651 interrupted:
2652 	err = sock_intr_errno(timeo);
2653 failure:
2654 	*errcode = err;
2655 	return NULL;
2656 }
2657 EXPORT_SYMBOL(sock_alloc_send_pskb);
2658 
2659 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2660 		     struct sockcm_cookie *sockc)
2661 {
2662 	u32 tsflags;
2663 
2664 	switch (cmsg->cmsg_type) {
2665 	case SO_MARK:
2666 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2667 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2668 			return -EPERM;
2669 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2670 			return -EINVAL;
2671 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2672 		break;
2673 	case SO_TIMESTAMPING_OLD:
2674 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2675 			return -EINVAL;
2676 
2677 		tsflags = *(u32 *)CMSG_DATA(cmsg);
2678 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2679 			return -EINVAL;
2680 
2681 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2682 		sockc->tsflags |= tsflags;
2683 		break;
2684 	case SCM_TXTIME:
2685 		if (!sock_flag(sk, SOCK_TXTIME))
2686 			return -EINVAL;
2687 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2688 			return -EINVAL;
2689 		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2690 		break;
2691 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2692 	case SCM_RIGHTS:
2693 	case SCM_CREDENTIALS:
2694 		break;
2695 	default:
2696 		return -EINVAL;
2697 	}
2698 	return 0;
2699 }
2700 EXPORT_SYMBOL(__sock_cmsg_send);
2701 
2702 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2703 		   struct sockcm_cookie *sockc)
2704 {
2705 	struct cmsghdr *cmsg;
2706 	int ret;
2707 
2708 	for_each_cmsghdr(cmsg, msg) {
2709 		if (!CMSG_OK(msg, cmsg))
2710 			return -EINVAL;
2711 		if (cmsg->cmsg_level != SOL_SOCKET)
2712 			continue;
2713 		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2714 		if (ret)
2715 			return ret;
2716 	}
2717 	return 0;
2718 }
2719 EXPORT_SYMBOL(sock_cmsg_send);
2720 
2721 static void sk_enter_memory_pressure(struct sock *sk)
2722 {
2723 	if (!sk->sk_prot->enter_memory_pressure)
2724 		return;
2725 
2726 	sk->sk_prot->enter_memory_pressure(sk);
2727 }
2728 
2729 static void sk_leave_memory_pressure(struct sock *sk)
2730 {
2731 	if (sk->sk_prot->leave_memory_pressure) {
2732 		sk->sk_prot->leave_memory_pressure(sk);
2733 	} else {
2734 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2735 
2736 		if (memory_pressure && READ_ONCE(*memory_pressure))
2737 			WRITE_ONCE(*memory_pressure, 0);
2738 	}
2739 }
2740 
2741 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2742 
2743 /**
2744  * skb_page_frag_refill - check that a page_frag contains enough room
2745  * @sz: minimum size of the fragment we want to get
2746  * @pfrag: pointer to page_frag
2747  * @gfp: priority for memory allocation
2748  *
2749  * Note: While this allocator tries to use high order pages, there is
2750  * no guarantee that allocations succeed. Therefore, @sz MUST be
2751  * less or equal than PAGE_SIZE.
2752  */
2753 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2754 {
2755 	if (pfrag->page) {
2756 		if (page_ref_count(pfrag->page) == 1) {
2757 			pfrag->offset = 0;
2758 			return true;
2759 		}
2760 		if (pfrag->offset + sz <= pfrag->size)
2761 			return true;
2762 		put_page(pfrag->page);
2763 	}
2764 
2765 	pfrag->offset = 0;
2766 	if (SKB_FRAG_PAGE_ORDER &&
2767 	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2768 		/* Avoid direct reclaim but allow kswapd to wake */
2769 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2770 					  __GFP_COMP | __GFP_NOWARN |
2771 					  __GFP_NORETRY,
2772 					  SKB_FRAG_PAGE_ORDER);
2773 		if (likely(pfrag->page)) {
2774 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2775 			return true;
2776 		}
2777 	}
2778 	pfrag->page = alloc_page(gfp);
2779 	if (likely(pfrag->page)) {
2780 		pfrag->size = PAGE_SIZE;
2781 		return true;
2782 	}
2783 	return false;
2784 }
2785 EXPORT_SYMBOL(skb_page_frag_refill);
2786 
2787 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2788 {
2789 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2790 		return true;
2791 
2792 	sk_enter_memory_pressure(sk);
2793 	sk_stream_moderate_sndbuf(sk);
2794 	return false;
2795 }
2796 EXPORT_SYMBOL(sk_page_frag_refill);
2797 
2798 void __lock_sock(struct sock *sk)
2799 	__releases(&sk->sk_lock.slock)
2800 	__acquires(&sk->sk_lock.slock)
2801 {
2802 	DEFINE_WAIT(wait);
2803 
2804 	for (;;) {
2805 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2806 					TASK_UNINTERRUPTIBLE);
2807 		spin_unlock_bh(&sk->sk_lock.slock);
2808 		schedule();
2809 		spin_lock_bh(&sk->sk_lock.slock);
2810 		if (!sock_owned_by_user(sk))
2811 			break;
2812 	}
2813 	finish_wait(&sk->sk_lock.wq, &wait);
2814 }
2815 
2816 void __release_sock(struct sock *sk)
2817 	__releases(&sk->sk_lock.slock)
2818 	__acquires(&sk->sk_lock.slock)
2819 {
2820 	struct sk_buff *skb, *next;
2821 
2822 	while ((skb = sk->sk_backlog.head) != NULL) {
2823 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2824 
2825 		spin_unlock_bh(&sk->sk_lock.slock);
2826 
2827 		do {
2828 			next = skb->next;
2829 			prefetch(next);
2830 			WARN_ON_ONCE(skb_dst_is_noref(skb));
2831 			skb_mark_not_on_list(skb);
2832 			sk_backlog_rcv(sk, skb);
2833 
2834 			cond_resched();
2835 
2836 			skb = next;
2837 		} while (skb != NULL);
2838 
2839 		spin_lock_bh(&sk->sk_lock.slock);
2840 	}
2841 
2842 	/*
2843 	 * Doing the zeroing here guarantee we can not loop forever
2844 	 * while a wild producer attempts to flood us.
2845 	 */
2846 	sk->sk_backlog.len = 0;
2847 }
2848 
2849 void __sk_flush_backlog(struct sock *sk)
2850 {
2851 	spin_lock_bh(&sk->sk_lock.slock);
2852 	__release_sock(sk);
2853 	spin_unlock_bh(&sk->sk_lock.slock);
2854 }
2855 
2856 /**
2857  * sk_wait_data - wait for data to arrive at sk_receive_queue
2858  * @sk:    sock to wait on
2859  * @timeo: for how long
2860  * @skb:   last skb seen on sk_receive_queue
2861  *
2862  * Now socket state including sk->sk_err is changed only under lock,
2863  * hence we may omit checks after joining wait queue.
2864  * We check receive queue before schedule() only as optimization;
2865  * it is very likely that release_sock() added new data.
2866  */
2867 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2868 {
2869 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2870 	int rc;
2871 
2872 	add_wait_queue(sk_sleep(sk), &wait);
2873 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2874 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2875 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2876 	remove_wait_queue(sk_sleep(sk), &wait);
2877 	return rc;
2878 }
2879 EXPORT_SYMBOL(sk_wait_data);
2880 
2881 /**
2882  *	__sk_mem_raise_allocated - increase memory_allocated
2883  *	@sk: socket
2884  *	@size: memory size to allocate
2885  *	@amt: pages to allocate
2886  *	@kind: allocation type
2887  *
2888  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2889  */
2890 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2891 {
2892 	struct proto *prot = sk->sk_prot;
2893 	long allocated = sk_memory_allocated_add(sk, amt);
2894 	bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2895 	bool charged = true;
2896 
2897 	if (memcg_charge &&
2898 	    !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2899 						gfp_memcg_charge())))
2900 		goto suppress_allocation;
2901 
2902 	/* Under limit. */
2903 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2904 		sk_leave_memory_pressure(sk);
2905 		return 1;
2906 	}
2907 
2908 	/* Under pressure. */
2909 	if (allocated > sk_prot_mem_limits(sk, 1))
2910 		sk_enter_memory_pressure(sk);
2911 
2912 	/* Over hard limit. */
2913 	if (allocated > sk_prot_mem_limits(sk, 2))
2914 		goto suppress_allocation;
2915 
2916 	/* guarantee minimum buffer size under pressure */
2917 	if (kind == SK_MEM_RECV) {
2918 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2919 			return 1;
2920 
2921 	} else { /* SK_MEM_SEND */
2922 		int wmem0 = sk_get_wmem0(sk, prot);
2923 
2924 		if (sk->sk_type == SOCK_STREAM) {
2925 			if (sk->sk_wmem_queued < wmem0)
2926 				return 1;
2927 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2928 				return 1;
2929 		}
2930 	}
2931 
2932 	if (sk_has_memory_pressure(sk)) {
2933 		u64 alloc;
2934 
2935 		if (!sk_under_memory_pressure(sk))
2936 			return 1;
2937 		alloc = sk_sockets_allocated_read_positive(sk);
2938 		if (sk_prot_mem_limits(sk, 2) > alloc *
2939 		    sk_mem_pages(sk->sk_wmem_queued +
2940 				 atomic_read(&sk->sk_rmem_alloc) +
2941 				 sk->sk_forward_alloc))
2942 			return 1;
2943 	}
2944 
2945 suppress_allocation:
2946 
2947 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2948 		sk_stream_moderate_sndbuf(sk);
2949 
2950 		/* Fail only if socket is _under_ its sndbuf.
2951 		 * In this case we cannot block, so that we have to fail.
2952 		 */
2953 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2954 			/* Force charge with __GFP_NOFAIL */
2955 			if (memcg_charge && !charged) {
2956 				mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2957 					gfp_memcg_charge() | __GFP_NOFAIL);
2958 			}
2959 			return 1;
2960 		}
2961 	}
2962 
2963 	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2964 		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2965 
2966 	sk_memory_allocated_sub(sk, amt);
2967 
2968 	if (memcg_charge && charged)
2969 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2970 
2971 	return 0;
2972 }
2973 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2974 
2975 /**
2976  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2977  *	@sk: socket
2978  *	@size: memory size to allocate
2979  *	@kind: allocation type
2980  *
2981  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2982  *	rmem allocation. This function assumes that protocols which have
2983  *	memory_pressure use sk_wmem_queued as write buffer accounting.
2984  */
2985 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2986 {
2987 	int ret, amt = sk_mem_pages(size);
2988 
2989 	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2990 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2991 	if (!ret)
2992 		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2993 	return ret;
2994 }
2995 EXPORT_SYMBOL(__sk_mem_schedule);
2996 
2997 /**
2998  *	__sk_mem_reduce_allocated - reclaim memory_allocated
2999  *	@sk: socket
3000  *	@amount: number of quanta
3001  *
3002  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3003  */
3004 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3005 {
3006 	sk_memory_allocated_sub(sk, amount);
3007 
3008 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3009 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3010 
3011 	if (sk_under_memory_pressure(sk) &&
3012 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3013 		sk_leave_memory_pressure(sk);
3014 }
3015 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
3016 
3017 /**
3018  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3019  *	@sk: socket
3020  *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
3021  */
3022 void __sk_mem_reclaim(struct sock *sk, int amount)
3023 {
3024 	amount >>= SK_MEM_QUANTUM_SHIFT;
3025 	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
3026 	__sk_mem_reduce_allocated(sk, amount);
3027 }
3028 EXPORT_SYMBOL(__sk_mem_reclaim);
3029 
3030 int sk_set_peek_off(struct sock *sk, int val)
3031 {
3032 	sk->sk_peek_off = val;
3033 	return 0;
3034 }
3035 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3036 
3037 /*
3038  * Set of default routines for initialising struct proto_ops when
3039  * the protocol does not support a particular function. In certain
3040  * cases where it makes no sense for a protocol to have a "do nothing"
3041  * function, some default processing is provided.
3042  */
3043 
3044 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3045 {
3046 	return -EOPNOTSUPP;
3047 }
3048 EXPORT_SYMBOL(sock_no_bind);
3049 
3050 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3051 		    int len, int flags)
3052 {
3053 	return -EOPNOTSUPP;
3054 }
3055 EXPORT_SYMBOL(sock_no_connect);
3056 
3057 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3058 {
3059 	return -EOPNOTSUPP;
3060 }
3061 EXPORT_SYMBOL(sock_no_socketpair);
3062 
3063 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3064 		   bool kern)
3065 {
3066 	return -EOPNOTSUPP;
3067 }
3068 EXPORT_SYMBOL(sock_no_accept);
3069 
3070 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3071 		    int peer)
3072 {
3073 	return -EOPNOTSUPP;
3074 }
3075 EXPORT_SYMBOL(sock_no_getname);
3076 
3077 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3078 {
3079 	return -EOPNOTSUPP;
3080 }
3081 EXPORT_SYMBOL(sock_no_ioctl);
3082 
3083 int sock_no_listen(struct socket *sock, int backlog)
3084 {
3085 	return -EOPNOTSUPP;
3086 }
3087 EXPORT_SYMBOL(sock_no_listen);
3088 
3089 int sock_no_shutdown(struct socket *sock, int how)
3090 {
3091 	return -EOPNOTSUPP;
3092 }
3093 EXPORT_SYMBOL(sock_no_shutdown);
3094 
3095 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3096 {
3097 	return -EOPNOTSUPP;
3098 }
3099 EXPORT_SYMBOL(sock_no_sendmsg);
3100 
3101 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3102 {
3103 	return -EOPNOTSUPP;
3104 }
3105 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3106 
3107 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3108 		    int flags)
3109 {
3110 	return -EOPNOTSUPP;
3111 }
3112 EXPORT_SYMBOL(sock_no_recvmsg);
3113 
3114 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3115 {
3116 	/* Mirror missing mmap method error code */
3117 	return -ENODEV;
3118 }
3119 EXPORT_SYMBOL(sock_no_mmap);
3120 
3121 /*
3122  * When a file is received (via SCM_RIGHTS, etc), we must bump the
3123  * various sock-based usage counts.
3124  */
3125 void __receive_sock(struct file *file)
3126 {
3127 	struct socket *sock;
3128 
3129 	sock = sock_from_file(file);
3130 	if (sock) {
3131 		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3132 		sock_update_classid(&sock->sk->sk_cgrp_data);
3133 	}
3134 }
3135 
3136 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3137 {
3138 	ssize_t res;
3139 	struct msghdr msg = {.msg_flags = flags};
3140 	struct kvec iov;
3141 	char *kaddr = kmap(page);
3142 	iov.iov_base = kaddr + offset;
3143 	iov.iov_len = size;
3144 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3145 	kunmap(page);
3146 	return res;
3147 }
3148 EXPORT_SYMBOL(sock_no_sendpage);
3149 
3150 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3151 				int offset, size_t size, int flags)
3152 {
3153 	ssize_t res;
3154 	struct msghdr msg = {.msg_flags = flags};
3155 	struct kvec iov;
3156 	char *kaddr = kmap(page);
3157 
3158 	iov.iov_base = kaddr + offset;
3159 	iov.iov_len = size;
3160 	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3161 	kunmap(page);
3162 	return res;
3163 }
3164 EXPORT_SYMBOL(sock_no_sendpage_locked);
3165 
3166 /*
3167  *	Default Socket Callbacks
3168  */
3169 
3170 static void sock_def_wakeup(struct sock *sk)
3171 {
3172 	struct socket_wq *wq;
3173 
3174 	rcu_read_lock();
3175 	wq = rcu_dereference(sk->sk_wq);
3176 	if (skwq_has_sleeper(wq))
3177 		wake_up_interruptible_all(&wq->wait);
3178 	rcu_read_unlock();
3179 }
3180 
3181 static void sock_def_error_report(struct sock *sk)
3182 {
3183 	struct socket_wq *wq;
3184 
3185 	rcu_read_lock();
3186 	wq = rcu_dereference(sk->sk_wq);
3187 	if (skwq_has_sleeper(wq))
3188 		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3189 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3190 	rcu_read_unlock();
3191 }
3192 
3193 void sock_def_readable(struct sock *sk)
3194 {
3195 	struct socket_wq *wq;
3196 
3197 	rcu_read_lock();
3198 	wq = rcu_dereference(sk->sk_wq);
3199 	if (skwq_has_sleeper(wq))
3200 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3201 						EPOLLRDNORM | EPOLLRDBAND);
3202 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3203 	rcu_read_unlock();
3204 }
3205 
3206 static void sock_def_write_space(struct sock *sk)
3207 {
3208 	struct socket_wq *wq;
3209 
3210 	rcu_read_lock();
3211 
3212 	/* Do not wake up a writer until he can make "significant"
3213 	 * progress.  --DaveM
3214 	 */
3215 	if (sock_writeable(sk)) {
3216 		wq = rcu_dereference(sk->sk_wq);
3217 		if (skwq_has_sleeper(wq))
3218 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3219 						EPOLLWRNORM | EPOLLWRBAND);
3220 
3221 		/* Should agree with poll, otherwise some programs break */
3222 		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3223 	}
3224 
3225 	rcu_read_unlock();
3226 }
3227 
3228 /* An optimised version of sock_def_write_space(), should only be called
3229  * for SOCK_RCU_FREE sockets under RCU read section and after putting
3230  * ->sk_wmem_alloc.
3231  */
3232 static void sock_def_write_space_wfree(struct sock *sk)
3233 {
3234 	/* Do not wake up a writer until he can make "significant"
3235 	 * progress.  --DaveM
3236 	 */
3237 	if (sock_writeable(sk)) {
3238 		struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3239 
3240 		/* rely on refcount_sub from sock_wfree() */
3241 		smp_mb__after_atomic();
3242 		if (wq && waitqueue_active(&wq->wait))
3243 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3244 						EPOLLWRNORM | EPOLLWRBAND);
3245 
3246 		/* Should agree with poll, otherwise some programs break */
3247 		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3248 	}
3249 }
3250 
3251 static void sock_def_destruct(struct sock *sk)
3252 {
3253 }
3254 
3255 void sk_send_sigurg(struct sock *sk)
3256 {
3257 	if (sk->sk_socket && sk->sk_socket->file)
3258 		if (send_sigurg(&sk->sk_socket->file->f_owner))
3259 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3260 }
3261 EXPORT_SYMBOL(sk_send_sigurg);
3262 
3263 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3264 		    unsigned long expires)
3265 {
3266 	if (!mod_timer(timer, expires))
3267 		sock_hold(sk);
3268 }
3269 EXPORT_SYMBOL(sk_reset_timer);
3270 
3271 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3272 {
3273 	if (del_timer(timer))
3274 		__sock_put(sk);
3275 }
3276 EXPORT_SYMBOL(sk_stop_timer);
3277 
3278 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3279 {
3280 	if (del_timer_sync(timer))
3281 		__sock_put(sk);
3282 }
3283 EXPORT_SYMBOL(sk_stop_timer_sync);
3284 
3285 void sock_init_data(struct socket *sock, struct sock *sk)
3286 {
3287 	sk_init_common(sk);
3288 	sk->sk_send_head	=	NULL;
3289 
3290 	timer_setup(&sk->sk_timer, NULL, 0);
3291 
3292 	sk->sk_allocation	=	GFP_KERNEL;
3293 	sk->sk_rcvbuf		=	sysctl_rmem_default;
3294 	sk->sk_sndbuf		=	sysctl_wmem_default;
3295 	sk->sk_state		=	TCP_CLOSE;
3296 	sk_set_socket(sk, sock);
3297 
3298 	sock_set_flag(sk, SOCK_ZAPPED);
3299 
3300 	if (sock) {
3301 		sk->sk_type	=	sock->type;
3302 		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3303 		sock->sk	=	sk;
3304 		sk->sk_uid	=	SOCK_INODE(sock)->i_uid;
3305 	} else {
3306 		RCU_INIT_POINTER(sk->sk_wq, NULL);
3307 		sk->sk_uid	=	make_kuid(sock_net(sk)->user_ns, 0);
3308 	}
3309 
3310 	rwlock_init(&sk->sk_callback_lock);
3311 	if (sk->sk_kern_sock)
3312 		lockdep_set_class_and_name(
3313 			&sk->sk_callback_lock,
3314 			af_kern_callback_keys + sk->sk_family,
3315 			af_family_kern_clock_key_strings[sk->sk_family]);
3316 	else
3317 		lockdep_set_class_and_name(
3318 			&sk->sk_callback_lock,
3319 			af_callback_keys + sk->sk_family,
3320 			af_family_clock_key_strings[sk->sk_family]);
3321 
3322 	sk->sk_state_change	=	sock_def_wakeup;
3323 	sk->sk_data_ready	=	sock_def_readable;
3324 	sk->sk_write_space	=	sock_def_write_space;
3325 	sk->sk_error_report	=	sock_def_error_report;
3326 	sk->sk_destruct		=	sock_def_destruct;
3327 
3328 	sk->sk_frag.page	=	NULL;
3329 	sk->sk_frag.offset	=	0;
3330 	sk->sk_peek_off		=	-1;
3331 
3332 	sk->sk_peer_pid 	=	NULL;
3333 	sk->sk_peer_cred	=	NULL;
3334 	spin_lock_init(&sk->sk_peer_lock);
3335 
3336 	sk->sk_write_pending	=	0;
3337 	sk->sk_rcvlowat		=	1;
3338 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3339 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3340 
3341 	sk->sk_stamp = SK_DEFAULT_STAMP;
3342 #if BITS_PER_LONG==32
3343 	seqlock_init(&sk->sk_stamp_seq);
3344 #endif
3345 	atomic_set(&sk->sk_zckey, 0);
3346 
3347 #ifdef CONFIG_NET_RX_BUSY_POLL
3348 	sk->sk_napi_id		=	0;
3349 	sk->sk_ll_usec		=	sysctl_net_busy_read;
3350 #endif
3351 
3352 	sk->sk_max_pacing_rate = ~0UL;
3353 	sk->sk_pacing_rate = ~0UL;
3354 	WRITE_ONCE(sk->sk_pacing_shift, 10);
3355 	sk->sk_incoming_cpu = -1;
3356 	sk->sk_txrehash = SOCK_TXREHASH_DEFAULT;
3357 
3358 	sk_rx_queue_clear(sk);
3359 	/*
3360 	 * Before updating sk_refcnt, we must commit prior changes to memory
3361 	 * (Documentation/RCU/rculist_nulls.rst for details)
3362 	 */
3363 	smp_wmb();
3364 	refcount_set(&sk->sk_refcnt, 1);
3365 	atomic_set(&sk->sk_drops, 0);
3366 }
3367 EXPORT_SYMBOL(sock_init_data);
3368 
3369 void lock_sock_nested(struct sock *sk, int subclass)
3370 {
3371 	/* The sk_lock has mutex_lock() semantics here. */
3372 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3373 
3374 	might_sleep();
3375 	spin_lock_bh(&sk->sk_lock.slock);
3376 	if (sock_owned_by_user_nocheck(sk))
3377 		__lock_sock(sk);
3378 	sk->sk_lock.owned = 1;
3379 	spin_unlock_bh(&sk->sk_lock.slock);
3380 }
3381 EXPORT_SYMBOL(lock_sock_nested);
3382 
3383 void release_sock(struct sock *sk)
3384 {
3385 	spin_lock_bh(&sk->sk_lock.slock);
3386 	if (sk->sk_backlog.tail)
3387 		__release_sock(sk);
3388 
3389 	/* Warning : release_cb() might need to release sk ownership,
3390 	 * ie call sock_release_ownership(sk) before us.
3391 	 */
3392 	if (sk->sk_prot->release_cb)
3393 		sk->sk_prot->release_cb(sk);
3394 
3395 	sock_release_ownership(sk);
3396 	if (waitqueue_active(&sk->sk_lock.wq))
3397 		wake_up(&sk->sk_lock.wq);
3398 	spin_unlock_bh(&sk->sk_lock.slock);
3399 }
3400 EXPORT_SYMBOL(release_sock);
3401 
3402 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3403 {
3404 	might_sleep();
3405 	spin_lock_bh(&sk->sk_lock.slock);
3406 
3407 	if (!sock_owned_by_user_nocheck(sk)) {
3408 		/*
3409 		 * Fast path return with bottom halves disabled and
3410 		 * sock::sk_lock.slock held.
3411 		 *
3412 		 * The 'mutex' is not contended and holding
3413 		 * sock::sk_lock.slock prevents all other lockers to
3414 		 * proceed so the corresponding unlock_sock_fast() can
3415 		 * avoid the slow path of release_sock() completely and
3416 		 * just release slock.
3417 		 *
3418 		 * From a semantical POV this is equivalent to 'acquiring'
3419 		 * the 'mutex', hence the corresponding lockdep
3420 		 * mutex_release() has to happen in the fast path of
3421 		 * unlock_sock_fast().
3422 		 */
3423 		return false;
3424 	}
3425 
3426 	__lock_sock(sk);
3427 	sk->sk_lock.owned = 1;
3428 	__acquire(&sk->sk_lock.slock);
3429 	spin_unlock_bh(&sk->sk_lock.slock);
3430 	return true;
3431 }
3432 EXPORT_SYMBOL(__lock_sock_fast);
3433 
3434 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3435 		   bool timeval, bool time32)
3436 {
3437 	struct sock *sk = sock->sk;
3438 	struct timespec64 ts;
3439 
3440 	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3441 	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3442 	if (ts.tv_sec == -1)
3443 		return -ENOENT;
3444 	if (ts.tv_sec == 0) {
3445 		ktime_t kt = ktime_get_real();
3446 		sock_write_timestamp(sk, kt);
3447 		ts = ktime_to_timespec64(kt);
3448 	}
3449 
3450 	if (timeval)
3451 		ts.tv_nsec /= 1000;
3452 
3453 #ifdef CONFIG_COMPAT_32BIT_TIME
3454 	if (time32)
3455 		return put_old_timespec32(&ts, userstamp);
3456 #endif
3457 #ifdef CONFIG_SPARC64
3458 	/* beware of padding in sparc64 timeval */
3459 	if (timeval && !in_compat_syscall()) {
3460 		struct __kernel_old_timeval __user tv = {
3461 			.tv_sec = ts.tv_sec,
3462 			.tv_usec = ts.tv_nsec,
3463 		};
3464 		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3465 			return -EFAULT;
3466 		return 0;
3467 	}
3468 #endif
3469 	return put_timespec64(&ts, userstamp);
3470 }
3471 EXPORT_SYMBOL(sock_gettstamp);
3472 
3473 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3474 {
3475 	if (!sock_flag(sk, flag)) {
3476 		unsigned long previous_flags = sk->sk_flags;
3477 
3478 		sock_set_flag(sk, flag);
3479 		/*
3480 		 * we just set one of the two flags which require net
3481 		 * time stamping, but time stamping might have been on
3482 		 * already because of the other one
3483 		 */
3484 		if (sock_needs_netstamp(sk) &&
3485 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3486 			net_enable_timestamp();
3487 	}
3488 }
3489 
3490 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3491 		       int level, int type)
3492 {
3493 	struct sock_exterr_skb *serr;
3494 	struct sk_buff *skb;
3495 	int copied, err;
3496 
3497 	err = -EAGAIN;
3498 	skb = sock_dequeue_err_skb(sk);
3499 	if (skb == NULL)
3500 		goto out;
3501 
3502 	copied = skb->len;
3503 	if (copied > len) {
3504 		msg->msg_flags |= MSG_TRUNC;
3505 		copied = len;
3506 	}
3507 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3508 	if (err)
3509 		goto out_free_skb;
3510 
3511 	sock_recv_timestamp(msg, sk, skb);
3512 
3513 	serr = SKB_EXT_ERR(skb);
3514 	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3515 
3516 	msg->msg_flags |= MSG_ERRQUEUE;
3517 	err = copied;
3518 
3519 out_free_skb:
3520 	kfree_skb(skb);
3521 out:
3522 	return err;
3523 }
3524 EXPORT_SYMBOL(sock_recv_errqueue);
3525 
3526 /*
3527  *	Get a socket option on an socket.
3528  *
3529  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3530  *	asynchronous errors should be reported by getsockopt. We assume
3531  *	this means if you specify SO_ERROR (otherwise whats the point of it).
3532  */
3533 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3534 			   char __user *optval, int __user *optlen)
3535 {
3536 	struct sock *sk = sock->sk;
3537 
3538 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3539 }
3540 EXPORT_SYMBOL(sock_common_getsockopt);
3541 
3542 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3543 			int flags)
3544 {
3545 	struct sock *sk = sock->sk;
3546 	int addr_len = 0;
3547 	int err;
3548 
3549 	err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3550 	if (err >= 0)
3551 		msg->msg_namelen = addr_len;
3552 	return err;
3553 }
3554 EXPORT_SYMBOL(sock_common_recvmsg);
3555 
3556 /*
3557  *	Set socket options on an inet socket.
3558  */
3559 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3560 			   sockptr_t optval, unsigned int optlen)
3561 {
3562 	struct sock *sk = sock->sk;
3563 
3564 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3565 }
3566 EXPORT_SYMBOL(sock_common_setsockopt);
3567 
3568 void sk_common_release(struct sock *sk)
3569 {
3570 	if (sk->sk_prot->destroy)
3571 		sk->sk_prot->destroy(sk);
3572 
3573 	/*
3574 	 * Observation: when sk_common_release is called, processes have
3575 	 * no access to socket. But net still has.
3576 	 * Step one, detach it from networking:
3577 	 *
3578 	 * A. Remove from hash tables.
3579 	 */
3580 
3581 	sk->sk_prot->unhash(sk);
3582 
3583 	/*
3584 	 * In this point socket cannot receive new packets, but it is possible
3585 	 * that some packets are in flight because some CPU runs receiver and
3586 	 * did hash table lookup before we unhashed socket. They will achieve
3587 	 * receive queue and will be purged by socket destructor.
3588 	 *
3589 	 * Also we still have packets pending on receive queue and probably,
3590 	 * our own packets waiting in device queues. sock_destroy will drain
3591 	 * receive queue, but transmitted packets will delay socket destruction
3592 	 * until the last reference will be released.
3593 	 */
3594 
3595 	sock_orphan(sk);
3596 
3597 	xfrm_sk_free_policy(sk);
3598 
3599 	sk_refcnt_debug_release(sk);
3600 
3601 	sock_put(sk);
3602 }
3603 EXPORT_SYMBOL(sk_common_release);
3604 
3605 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3606 {
3607 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3608 
3609 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3610 	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3611 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3612 	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3613 	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3614 	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3615 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3616 	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3617 	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3618 }
3619 
3620 #ifdef CONFIG_PROC_FS
3621 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3622 
3623 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3624 {
3625 	int cpu, idx = prot->inuse_idx;
3626 	int res = 0;
3627 
3628 	for_each_possible_cpu(cpu)
3629 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3630 
3631 	return res >= 0 ? res : 0;
3632 }
3633 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3634 
3635 int sock_inuse_get(struct net *net)
3636 {
3637 	int cpu, res = 0;
3638 
3639 	for_each_possible_cpu(cpu)
3640 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3641 
3642 	return res;
3643 }
3644 
3645 EXPORT_SYMBOL_GPL(sock_inuse_get);
3646 
3647 static int __net_init sock_inuse_init_net(struct net *net)
3648 {
3649 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3650 	if (net->core.prot_inuse == NULL)
3651 		return -ENOMEM;
3652 	return 0;
3653 }
3654 
3655 static void __net_exit sock_inuse_exit_net(struct net *net)
3656 {
3657 	free_percpu(net->core.prot_inuse);
3658 }
3659 
3660 static struct pernet_operations net_inuse_ops = {
3661 	.init = sock_inuse_init_net,
3662 	.exit = sock_inuse_exit_net,
3663 };
3664 
3665 static __init int net_inuse_init(void)
3666 {
3667 	if (register_pernet_subsys(&net_inuse_ops))
3668 		panic("Cannot initialize net inuse counters");
3669 
3670 	return 0;
3671 }
3672 
3673 core_initcall(net_inuse_init);
3674 
3675 static int assign_proto_idx(struct proto *prot)
3676 {
3677 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3678 
3679 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3680 		pr_err("PROTO_INUSE_NR exhausted\n");
3681 		return -ENOSPC;
3682 	}
3683 
3684 	set_bit(prot->inuse_idx, proto_inuse_idx);
3685 	return 0;
3686 }
3687 
3688 static void release_proto_idx(struct proto *prot)
3689 {
3690 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3691 		clear_bit(prot->inuse_idx, proto_inuse_idx);
3692 }
3693 #else
3694 static inline int assign_proto_idx(struct proto *prot)
3695 {
3696 	return 0;
3697 }
3698 
3699 static inline void release_proto_idx(struct proto *prot)
3700 {
3701 }
3702 
3703 #endif
3704 
3705 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3706 {
3707 	if (!twsk_prot)
3708 		return;
3709 	kfree(twsk_prot->twsk_slab_name);
3710 	twsk_prot->twsk_slab_name = NULL;
3711 	kmem_cache_destroy(twsk_prot->twsk_slab);
3712 	twsk_prot->twsk_slab = NULL;
3713 }
3714 
3715 static int tw_prot_init(const struct proto *prot)
3716 {
3717 	struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3718 
3719 	if (!twsk_prot)
3720 		return 0;
3721 
3722 	twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3723 					      prot->name);
3724 	if (!twsk_prot->twsk_slab_name)
3725 		return -ENOMEM;
3726 
3727 	twsk_prot->twsk_slab =
3728 		kmem_cache_create(twsk_prot->twsk_slab_name,
3729 				  twsk_prot->twsk_obj_size, 0,
3730 				  SLAB_ACCOUNT | prot->slab_flags,
3731 				  NULL);
3732 	if (!twsk_prot->twsk_slab) {
3733 		pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3734 			prot->name);
3735 		return -ENOMEM;
3736 	}
3737 
3738 	return 0;
3739 }
3740 
3741 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3742 {
3743 	if (!rsk_prot)
3744 		return;
3745 	kfree(rsk_prot->slab_name);
3746 	rsk_prot->slab_name = NULL;
3747 	kmem_cache_destroy(rsk_prot->slab);
3748 	rsk_prot->slab = NULL;
3749 }
3750 
3751 static int req_prot_init(const struct proto *prot)
3752 {
3753 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3754 
3755 	if (!rsk_prot)
3756 		return 0;
3757 
3758 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3759 					prot->name);
3760 	if (!rsk_prot->slab_name)
3761 		return -ENOMEM;
3762 
3763 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3764 					   rsk_prot->obj_size, 0,
3765 					   SLAB_ACCOUNT | prot->slab_flags,
3766 					   NULL);
3767 
3768 	if (!rsk_prot->slab) {
3769 		pr_crit("%s: Can't create request sock SLAB cache!\n",
3770 			prot->name);
3771 		return -ENOMEM;
3772 	}
3773 	return 0;
3774 }
3775 
3776 int proto_register(struct proto *prot, int alloc_slab)
3777 {
3778 	int ret = -ENOBUFS;
3779 
3780 	if (prot->memory_allocated && !prot->sysctl_mem) {
3781 		pr_err("%s: missing sysctl_mem\n", prot->name);
3782 		return -EINVAL;
3783 	}
3784 	if (alloc_slab) {
3785 		prot->slab = kmem_cache_create_usercopy(prot->name,
3786 					prot->obj_size, 0,
3787 					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3788 					prot->slab_flags,
3789 					prot->useroffset, prot->usersize,
3790 					NULL);
3791 
3792 		if (prot->slab == NULL) {
3793 			pr_crit("%s: Can't create sock SLAB cache!\n",
3794 				prot->name);
3795 			goto out;
3796 		}
3797 
3798 		if (req_prot_init(prot))
3799 			goto out_free_request_sock_slab;
3800 
3801 		if (tw_prot_init(prot))
3802 			goto out_free_timewait_sock_slab;
3803 	}
3804 
3805 	mutex_lock(&proto_list_mutex);
3806 	ret = assign_proto_idx(prot);
3807 	if (ret) {
3808 		mutex_unlock(&proto_list_mutex);
3809 		goto out_free_timewait_sock_slab;
3810 	}
3811 	list_add(&prot->node, &proto_list);
3812 	mutex_unlock(&proto_list_mutex);
3813 	return ret;
3814 
3815 out_free_timewait_sock_slab:
3816 	if (alloc_slab)
3817 		tw_prot_cleanup(prot->twsk_prot);
3818 out_free_request_sock_slab:
3819 	if (alloc_slab) {
3820 		req_prot_cleanup(prot->rsk_prot);
3821 
3822 		kmem_cache_destroy(prot->slab);
3823 		prot->slab = NULL;
3824 	}
3825 out:
3826 	return ret;
3827 }
3828 EXPORT_SYMBOL(proto_register);
3829 
3830 void proto_unregister(struct proto *prot)
3831 {
3832 	mutex_lock(&proto_list_mutex);
3833 	release_proto_idx(prot);
3834 	list_del(&prot->node);
3835 	mutex_unlock(&proto_list_mutex);
3836 
3837 	kmem_cache_destroy(prot->slab);
3838 	prot->slab = NULL;
3839 
3840 	req_prot_cleanup(prot->rsk_prot);
3841 	tw_prot_cleanup(prot->twsk_prot);
3842 }
3843 EXPORT_SYMBOL(proto_unregister);
3844 
3845 int sock_load_diag_module(int family, int protocol)
3846 {
3847 	if (!protocol) {
3848 		if (!sock_is_registered(family))
3849 			return -ENOENT;
3850 
3851 		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3852 				      NETLINK_SOCK_DIAG, family);
3853 	}
3854 
3855 #ifdef CONFIG_INET
3856 	if (family == AF_INET &&
3857 	    protocol != IPPROTO_RAW &&
3858 	    protocol < MAX_INET_PROTOS &&
3859 	    !rcu_access_pointer(inet_protos[protocol]))
3860 		return -ENOENT;
3861 #endif
3862 
3863 	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3864 			      NETLINK_SOCK_DIAG, family, protocol);
3865 }
3866 EXPORT_SYMBOL(sock_load_diag_module);
3867 
3868 #ifdef CONFIG_PROC_FS
3869 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3870 	__acquires(proto_list_mutex)
3871 {
3872 	mutex_lock(&proto_list_mutex);
3873 	return seq_list_start_head(&proto_list, *pos);
3874 }
3875 
3876 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3877 {
3878 	return seq_list_next(v, &proto_list, pos);
3879 }
3880 
3881 static void proto_seq_stop(struct seq_file *seq, void *v)
3882 	__releases(proto_list_mutex)
3883 {
3884 	mutex_unlock(&proto_list_mutex);
3885 }
3886 
3887 static char proto_method_implemented(const void *method)
3888 {
3889 	return method == NULL ? 'n' : 'y';
3890 }
3891 static long sock_prot_memory_allocated(struct proto *proto)
3892 {
3893 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3894 }
3895 
3896 static const char *sock_prot_memory_pressure(struct proto *proto)
3897 {
3898 	return proto->memory_pressure != NULL ?
3899 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3900 }
3901 
3902 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3903 {
3904 
3905 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3906 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3907 		   proto->name,
3908 		   proto->obj_size,
3909 		   sock_prot_inuse_get(seq_file_net(seq), proto),
3910 		   sock_prot_memory_allocated(proto),
3911 		   sock_prot_memory_pressure(proto),
3912 		   proto->max_header,
3913 		   proto->slab == NULL ? "no" : "yes",
3914 		   module_name(proto->owner),
3915 		   proto_method_implemented(proto->close),
3916 		   proto_method_implemented(proto->connect),
3917 		   proto_method_implemented(proto->disconnect),
3918 		   proto_method_implemented(proto->accept),
3919 		   proto_method_implemented(proto->ioctl),
3920 		   proto_method_implemented(proto->init),
3921 		   proto_method_implemented(proto->destroy),
3922 		   proto_method_implemented(proto->shutdown),
3923 		   proto_method_implemented(proto->setsockopt),
3924 		   proto_method_implemented(proto->getsockopt),
3925 		   proto_method_implemented(proto->sendmsg),
3926 		   proto_method_implemented(proto->recvmsg),
3927 		   proto_method_implemented(proto->sendpage),
3928 		   proto_method_implemented(proto->bind),
3929 		   proto_method_implemented(proto->backlog_rcv),
3930 		   proto_method_implemented(proto->hash),
3931 		   proto_method_implemented(proto->unhash),
3932 		   proto_method_implemented(proto->get_port),
3933 		   proto_method_implemented(proto->enter_memory_pressure));
3934 }
3935 
3936 static int proto_seq_show(struct seq_file *seq, void *v)
3937 {
3938 	if (v == &proto_list)
3939 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3940 			   "protocol",
3941 			   "size",
3942 			   "sockets",
3943 			   "memory",
3944 			   "press",
3945 			   "maxhdr",
3946 			   "slab",
3947 			   "module",
3948 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3949 	else
3950 		proto_seq_printf(seq, list_entry(v, struct proto, node));
3951 	return 0;
3952 }
3953 
3954 static const struct seq_operations proto_seq_ops = {
3955 	.start  = proto_seq_start,
3956 	.next   = proto_seq_next,
3957 	.stop   = proto_seq_stop,
3958 	.show   = proto_seq_show,
3959 };
3960 
3961 static __net_init int proto_init_net(struct net *net)
3962 {
3963 	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3964 			sizeof(struct seq_net_private)))
3965 		return -ENOMEM;
3966 
3967 	return 0;
3968 }
3969 
3970 static __net_exit void proto_exit_net(struct net *net)
3971 {
3972 	remove_proc_entry("protocols", net->proc_net);
3973 }
3974 
3975 
3976 static __net_initdata struct pernet_operations proto_net_ops = {
3977 	.init = proto_init_net,
3978 	.exit = proto_exit_net,
3979 };
3980 
3981 static int __init proto_init(void)
3982 {
3983 	return register_pernet_subsys(&proto_net_ops);
3984 }
3985 
3986 subsys_initcall(proto_init);
3987 
3988 #endif /* PROC_FS */
3989 
3990 #ifdef CONFIG_NET_RX_BUSY_POLL
3991 bool sk_busy_loop_end(void *p, unsigned long start_time)
3992 {
3993 	struct sock *sk = p;
3994 
3995 	return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3996 	       sk_busy_loop_timeout(sk, start_time);
3997 }
3998 EXPORT_SYMBOL(sk_busy_loop_end);
3999 #endif /* CONFIG_NET_RX_BUSY_POLL */
4000 
4001 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4002 {
4003 	if (!sk->sk_prot->bind_add)
4004 		return -EOPNOTSUPP;
4005 	return sk->sk_prot->bind_add(sk, addr, addr_len);
4006 }
4007 EXPORT_SYMBOL(sock_bind_add);
4008