xref: /linux/net/core/sock.c (revision 0738599856542bab0ebcd73cab9d8f15bddedcee)
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 /**
150  * sk_ns_capable - General socket capability test
151  * @sk: Socket to use a capability on or through
152  * @user_ns: The user namespace of the capability to use
153  * @cap: The capability to use
154  *
155  * Test to see if the opener of the socket had when the socket was
156  * created and the current process has the capability @cap in the user
157  * namespace @user_ns.
158  */
159 bool sk_ns_capable(const struct sock *sk,
160 		   struct user_namespace *user_ns, int cap)
161 {
162 	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 		ns_capable(user_ns, cap);
164 }
165 EXPORT_SYMBOL(sk_ns_capable);
166 
167 /**
168  * sk_capable - Socket global capability test
169  * @sk: Socket to use a capability on or through
170  * @cap: The global capability to use
171  *
172  * Test to see if the opener of the socket had when the socket was
173  * created and the current process has the capability @cap in all user
174  * namespaces.
175  */
176 bool sk_capable(const struct sock *sk, int cap)
177 {
178 	return sk_ns_capable(sk, &init_user_ns, cap);
179 }
180 EXPORT_SYMBOL(sk_capable);
181 
182 /**
183  * sk_net_capable - Network namespace socket capability test
184  * @sk: Socket to use a capability on or through
185  * @cap: The capability to use
186  *
187  * Test to see if the opener of the socket had when the socket was created
188  * and the current process has the capability @cap over the network namespace
189  * the socket is a member of.
190  */
191 bool sk_net_capable(const struct sock *sk, int cap)
192 {
193 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 }
195 EXPORT_SYMBOL(sk_net_capable);
196 
197 /*
198  * Each address family might have different locking rules, so we have
199  * one slock key per address family and separate keys for internal and
200  * userspace sockets.
201  */
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_keys[AF_MAX];
204 static struct lock_class_key af_family_slock_keys[AF_MAX];
205 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
206 
207 /*
208  * Make lock validator output more readable. (we pre-construct these
209  * strings build-time, so that runtime initialization of socket
210  * locks is fast):
211  */
212 
213 #define _sock_locks(x)						  \
214   x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
215   x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
216   x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
217   x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
218   x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
219   x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
220   x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
221   x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
222   x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
223   x "27"       ,	x "28"          ,	x "AF_CAN"      , \
224   x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
225   x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
226   x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
227   x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
228   x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
229   x "AF_MCTP"  , \
230   x "AF_MAX"
231 
232 static const char *const af_family_key_strings[AF_MAX+1] = {
233 	_sock_locks("sk_lock-")
234 };
235 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
236 	_sock_locks("slock-")
237 };
238 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
239 	_sock_locks("clock-")
240 };
241 
242 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
243 	_sock_locks("k-sk_lock-")
244 };
245 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
246 	_sock_locks("k-slock-")
247 };
248 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
249 	_sock_locks("k-clock-")
250 };
251 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
252 	_sock_locks("rlock-")
253 };
254 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
255 	_sock_locks("wlock-")
256 };
257 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
258 	_sock_locks("elock-")
259 };
260 
261 /*
262  * sk_callback_lock and sk queues locking rules are per-address-family,
263  * so split the lock classes by using a per-AF key:
264  */
265 static struct lock_class_key af_callback_keys[AF_MAX];
266 static struct lock_class_key af_rlock_keys[AF_MAX];
267 static struct lock_class_key af_wlock_keys[AF_MAX];
268 static struct lock_class_key af_elock_keys[AF_MAX];
269 static struct lock_class_key af_kern_callback_keys[AF_MAX];
270 
271 /* Run time adjustable parameters. */
272 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
273 EXPORT_SYMBOL(sysctl_wmem_max);
274 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
275 EXPORT_SYMBOL(sysctl_rmem_max);
276 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
277 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
278 
279 /* Maximal space eaten by iovec or ancillary data plus some space */
280 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
281 EXPORT_SYMBOL(sysctl_optmem_max);
282 
283 int sysctl_tstamp_allow_data __read_mostly = 1;
284 
285 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
286 EXPORT_SYMBOL_GPL(memalloc_socks_key);
287 
288 /**
289  * sk_set_memalloc - sets %SOCK_MEMALLOC
290  * @sk: socket to set it on
291  *
292  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
293  * It's the responsibility of the admin to adjust min_free_kbytes
294  * to meet the requirements
295  */
296 void sk_set_memalloc(struct sock *sk)
297 {
298 	sock_set_flag(sk, SOCK_MEMALLOC);
299 	sk->sk_allocation |= __GFP_MEMALLOC;
300 	static_branch_inc(&memalloc_socks_key);
301 }
302 EXPORT_SYMBOL_GPL(sk_set_memalloc);
303 
304 void sk_clear_memalloc(struct sock *sk)
305 {
306 	sock_reset_flag(sk, SOCK_MEMALLOC);
307 	sk->sk_allocation &= ~__GFP_MEMALLOC;
308 	static_branch_dec(&memalloc_socks_key);
309 
310 	/*
311 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
312 	 * progress of swapping. SOCK_MEMALLOC may be cleared while
313 	 * it has rmem allocations due to the last swapfile being deactivated
314 	 * but there is a risk that the socket is unusable due to exceeding
315 	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
316 	 */
317 	sk_mem_reclaim(sk);
318 }
319 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
320 
321 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
322 {
323 	int ret;
324 	unsigned int noreclaim_flag;
325 
326 	/* these should have been dropped before queueing */
327 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
328 
329 	noreclaim_flag = memalloc_noreclaim_save();
330 	ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
331 				 tcp_v6_do_rcv,
332 				 tcp_v4_do_rcv,
333 				 sk, skb);
334 	memalloc_noreclaim_restore(noreclaim_flag);
335 
336 	return ret;
337 }
338 EXPORT_SYMBOL(__sk_backlog_rcv);
339 
340 void sk_error_report(struct sock *sk)
341 {
342 	sk->sk_error_report(sk);
343 
344 	switch (sk->sk_family) {
345 	case AF_INET:
346 		fallthrough;
347 	case AF_INET6:
348 		trace_inet_sk_error_report(sk);
349 		break;
350 	default:
351 		break;
352 	}
353 }
354 EXPORT_SYMBOL(sk_error_report);
355 
356 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
357 {
358 	struct __kernel_sock_timeval tv;
359 
360 	if (timeo == MAX_SCHEDULE_TIMEOUT) {
361 		tv.tv_sec = 0;
362 		tv.tv_usec = 0;
363 	} else {
364 		tv.tv_sec = timeo / HZ;
365 		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
366 	}
367 
368 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
369 		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
370 		*(struct old_timeval32 *)optval = tv32;
371 		return sizeof(tv32);
372 	}
373 
374 	if (old_timeval) {
375 		struct __kernel_old_timeval old_tv;
376 		old_tv.tv_sec = tv.tv_sec;
377 		old_tv.tv_usec = tv.tv_usec;
378 		*(struct __kernel_old_timeval *)optval = old_tv;
379 		return sizeof(old_tv);
380 	}
381 
382 	*(struct __kernel_sock_timeval *)optval = tv;
383 	return sizeof(tv);
384 }
385 EXPORT_SYMBOL(sock_get_timeout);
386 
387 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
388 			   sockptr_t optval, int optlen, bool old_timeval)
389 {
390 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
391 		struct old_timeval32 tv32;
392 
393 		if (optlen < sizeof(tv32))
394 			return -EINVAL;
395 
396 		if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
397 			return -EFAULT;
398 		tv->tv_sec = tv32.tv_sec;
399 		tv->tv_usec = tv32.tv_usec;
400 	} else if (old_timeval) {
401 		struct __kernel_old_timeval old_tv;
402 
403 		if (optlen < sizeof(old_tv))
404 			return -EINVAL;
405 		if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
406 			return -EFAULT;
407 		tv->tv_sec = old_tv.tv_sec;
408 		tv->tv_usec = old_tv.tv_usec;
409 	} else {
410 		if (optlen < sizeof(*tv))
411 			return -EINVAL;
412 		if (copy_from_sockptr(tv, optval, sizeof(*tv)))
413 			return -EFAULT;
414 	}
415 
416 	return 0;
417 }
418 EXPORT_SYMBOL(sock_copy_user_timeval);
419 
420 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
421 			    bool old_timeval)
422 {
423 	struct __kernel_sock_timeval tv;
424 	int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
425 
426 	if (err)
427 		return err;
428 
429 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
430 		return -EDOM;
431 
432 	if (tv.tv_sec < 0) {
433 		static int warned __read_mostly;
434 
435 		*timeo_p = 0;
436 		if (warned < 10 && net_ratelimit()) {
437 			warned++;
438 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
439 				__func__, current->comm, task_pid_nr(current));
440 		}
441 		return 0;
442 	}
443 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
444 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
445 		return 0;
446 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
447 		*timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
448 	return 0;
449 }
450 
451 static bool sock_needs_netstamp(const struct sock *sk)
452 {
453 	switch (sk->sk_family) {
454 	case AF_UNSPEC:
455 	case AF_UNIX:
456 		return false;
457 	default:
458 		return true;
459 	}
460 }
461 
462 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
463 {
464 	if (sk->sk_flags & flags) {
465 		sk->sk_flags &= ~flags;
466 		if (sock_needs_netstamp(sk) &&
467 		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
468 			net_disable_timestamp();
469 	}
470 }
471 
472 
473 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
474 {
475 	unsigned long flags;
476 	struct sk_buff_head *list = &sk->sk_receive_queue;
477 
478 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
479 		atomic_inc(&sk->sk_drops);
480 		trace_sock_rcvqueue_full(sk, skb);
481 		return -ENOMEM;
482 	}
483 
484 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
485 		atomic_inc(&sk->sk_drops);
486 		return -ENOBUFS;
487 	}
488 
489 	skb->dev = NULL;
490 	skb_set_owner_r(skb, sk);
491 
492 	/* we escape from rcu protected region, make sure we dont leak
493 	 * a norefcounted dst
494 	 */
495 	skb_dst_force(skb);
496 
497 	spin_lock_irqsave(&list->lock, flags);
498 	sock_skb_set_dropcount(sk, skb);
499 	__skb_queue_tail(list, skb);
500 	spin_unlock_irqrestore(&list->lock, flags);
501 
502 	if (!sock_flag(sk, SOCK_DEAD))
503 		sk->sk_data_ready(sk);
504 	return 0;
505 }
506 EXPORT_SYMBOL(__sock_queue_rcv_skb);
507 
508 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
509 {
510 	int err;
511 
512 	err = sk_filter(sk, skb);
513 	if (err)
514 		return err;
515 
516 	return __sock_queue_rcv_skb(sk, skb);
517 }
518 EXPORT_SYMBOL(sock_queue_rcv_skb);
519 
520 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
521 		     const int nested, unsigned int trim_cap, bool refcounted)
522 {
523 	int rc = NET_RX_SUCCESS;
524 
525 	if (sk_filter_trim_cap(sk, skb, trim_cap))
526 		goto discard_and_relse;
527 
528 	skb->dev = NULL;
529 
530 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
531 		atomic_inc(&sk->sk_drops);
532 		goto discard_and_relse;
533 	}
534 	if (nested)
535 		bh_lock_sock_nested(sk);
536 	else
537 		bh_lock_sock(sk);
538 	if (!sock_owned_by_user(sk)) {
539 		/*
540 		 * trylock + unlock semantics:
541 		 */
542 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
543 
544 		rc = sk_backlog_rcv(sk, skb);
545 
546 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
547 	} else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
548 		bh_unlock_sock(sk);
549 		atomic_inc(&sk->sk_drops);
550 		goto discard_and_relse;
551 	}
552 
553 	bh_unlock_sock(sk);
554 out:
555 	if (refcounted)
556 		sock_put(sk);
557 	return rc;
558 discard_and_relse:
559 	kfree_skb(skb);
560 	goto out;
561 }
562 EXPORT_SYMBOL(__sk_receive_skb);
563 
564 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
565 							  u32));
566 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
567 							   u32));
568 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
569 {
570 	struct dst_entry *dst = __sk_dst_get(sk);
571 
572 	if (dst && dst->obsolete &&
573 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
574 			       dst, cookie) == NULL) {
575 		sk_tx_queue_clear(sk);
576 		sk->sk_dst_pending_confirm = 0;
577 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
578 		dst_release(dst);
579 		return NULL;
580 	}
581 
582 	return dst;
583 }
584 EXPORT_SYMBOL(__sk_dst_check);
585 
586 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
587 {
588 	struct dst_entry *dst = sk_dst_get(sk);
589 
590 	if (dst && dst->obsolete &&
591 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
592 			       dst, cookie) == NULL) {
593 		sk_dst_reset(sk);
594 		dst_release(dst);
595 		return NULL;
596 	}
597 
598 	return dst;
599 }
600 EXPORT_SYMBOL(sk_dst_check);
601 
602 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
603 {
604 	int ret = -ENOPROTOOPT;
605 #ifdef CONFIG_NETDEVICES
606 	struct net *net = sock_net(sk);
607 
608 	/* Sorry... */
609 	ret = -EPERM;
610 	if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
611 		goto out;
612 
613 	ret = -EINVAL;
614 	if (ifindex < 0)
615 		goto out;
616 
617 	sk->sk_bound_dev_if = ifindex;
618 	if (sk->sk_prot->rehash)
619 		sk->sk_prot->rehash(sk);
620 	sk_dst_reset(sk);
621 
622 	ret = 0;
623 
624 out:
625 #endif
626 
627 	return ret;
628 }
629 
630 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
631 {
632 	int ret;
633 
634 	if (lock_sk)
635 		lock_sock(sk);
636 	ret = sock_bindtoindex_locked(sk, ifindex);
637 	if (lock_sk)
638 		release_sock(sk);
639 
640 	return ret;
641 }
642 EXPORT_SYMBOL(sock_bindtoindex);
643 
644 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
645 {
646 	int ret = -ENOPROTOOPT;
647 #ifdef CONFIG_NETDEVICES
648 	struct net *net = sock_net(sk);
649 	char devname[IFNAMSIZ];
650 	int index;
651 
652 	ret = -EINVAL;
653 	if (optlen < 0)
654 		goto out;
655 
656 	/* Bind this socket to a particular device like "eth0",
657 	 * as specified in the passed interface name. If the
658 	 * name is "" or the option length is zero the socket
659 	 * is not bound.
660 	 */
661 	if (optlen > IFNAMSIZ - 1)
662 		optlen = IFNAMSIZ - 1;
663 	memset(devname, 0, sizeof(devname));
664 
665 	ret = -EFAULT;
666 	if (copy_from_sockptr(devname, optval, optlen))
667 		goto out;
668 
669 	index = 0;
670 	if (devname[0] != '\0') {
671 		struct net_device *dev;
672 
673 		rcu_read_lock();
674 		dev = dev_get_by_name_rcu(net, devname);
675 		if (dev)
676 			index = dev->ifindex;
677 		rcu_read_unlock();
678 		ret = -ENODEV;
679 		if (!dev)
680 			goto out;
681 	}
682 
683 	return sock_bindtoindex(sk, index, true);
684 out:
685 #endif
686 
687 	return ret;
688 }
689 
690 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
691 				int __user *optlen, int len)
692 {
693 	int ret = -ENOPROTOOPT;
694 #ifdef CONFIG_NETDEVICES
695 	struct net *net = sock_net(sk);
696 	char devname[IFNAMSIZ];
697 
698 	if (sk->sk_bound_dev_if == 0) {
699 		len = 0;
700 		goto zero;
701 	}
702 
703 	ret = -EINVAL;
704 	if (len < IFNAMSIZ)
705 		goto out;
706 
707 	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
708 	if (ret)
709 		goto out;
710 
711 	len = strlen(devname) + 1;
712 
713 	ret = -EFAULT;
714 	if (copy_to_user(optval, devname, len))
715 		goto out;
716 
717 zero:
718 	ret = -EFAULT;
719 	if (put_user(len, optlen))
720 		goto out;
721 
722 	ret = 0;
723 
724 out:
725 #endif
726 
727 	return ret;
728 }
729 
730 bool sk_mc_loop(struct sock *sk)
731 {
732 	if (dev_recursion_level())
733 		return false;
734 	if (!sk)
735 		return true;
736 	switch (sk->sk_family) {
737 	case AF_INET:
738 		return inet_sk(sk)->mc_loop;
739 #if IS_ENABLED(CONFIG_IPV6)
740 	case AF_INET6:
741 		return inet6_sk(sk)->mc_loop;
742 #endif
743 	}
744 	WARN_ON_ONCE(1);
745 	return true;
746 }
747 EXPORT_SYMBOL(sk_mc_loop);
748 
749 void sock_set_reuseaddr(struct sock *sk)
750 {
751 	lock_sock(sk);
752 	sk->sk_reuse = SK_CAN_REUSE;
753 	release_sock(sk);
754 }
755 EXPORT_SYMBOL(sock_set_reuseaddr);
756 
757 void sock_set_reuseport(struct sock *sk)
758 {
759 	lock_sock(sk);
760 	sk->sk_reuseport = true;
761 	release_sock(sk);
762 }
763 EXPORT_SYMBOL(sock_set_reuseport);
764 
765 void sock_no_linger(struct sock *sk)
766 {
767 	lock_sock(sk);
768 	sk->sk_lingertime = 0;
769 	sock_set_flag(sk, SOCK_LINGER);
770 	release_sock(sk);
771 }
772 EXPORT_SYMBOL(sock_no_linger);
773 
774 void sock_set_priority(struct sock *sk, u32 priority)
775 {
776 	lock_sock(sk);
777 	sk->sk_priority = priority;
778 	release_sock(sk);
779 }
780 EXPORT_SYMBOL(sock_set_priority);
781 
782 void sock_set_sndtimeo(struct sock *sk, s64 secs)
783 {
784 	lock_sock(sk);
785 	if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
786 		sk->sk_sndtimeo = secs * HZ;
787 	else
788 		sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
789 	release_sock(sk);
790 }
791 EXPORT_SYMBOL(sock_set_sndtimeo);
792 
793 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
794 {
795 	if (val)  {
796 		sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
797 		sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
798 		sock_set_flag(sk, SOCK_RCVTSTAMP);
799 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
800 	} else {
801 		sock_reset_flag(sk, SOCK_RCVTSTAMP);
802 		sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
803 	}
804 }
805 
806 void sock_enable_timestamps(struct sock *sk)
807 {
808 	lock_sock(sk);
809 	__sock_set_timestamps(sk, true, false, true);
810 	release_sock(sk);
811 }
812 EXPORT_SYMBOL(sock_enable_timestamps);
813 
814 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
815 {
816 	switch (optname) {
817 	case SO_TIMESTAMP_OLD:
818 		__sock_set_timestamps(sk, valbool, false, false);
819 		break;
820 	case SO_TIMESTAMP_NEW:
821 		__sock_set_timestamps(sk, valbool, true, false);
822 		break;
823 	case SO_TIMESTAMPNS_OLD:
824 		__sock_set_timestamps(sk, valbool, false, true);
825 		break;
826 	case SO_TIMESTAMPNS_NEW:
827 		__sock_set_timestamps(sk, valbool, true, true);
828 		break;
829 	}
830 }
831 
832 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
833 {
834 	struct net *net = sock_net(sk);
835 	struct net_device *dev = NULL;
836 	bool match = false;
837 	int *vclock_index;
838 	int i, num;
839 
840 	if (sk->sk_bound_dev_if)
841 		dev = dev_get_by_index(net, sk->sk_bound_dev_if);
842 
843 	if (!dev) {
844 		pr_err("%s: sock not bind to device\n", __func__);
845 		return -EOPNOTSUPP;
846 	}
847 
848 	num = ethtool_get_phc_vclocks(dev, &vclock_index);
849 	dev_put(dev);
850 
851 	for (i = 0; i < num; i++) {
852 		if (*(vclock_index + i) == phc_index) {
853 			match = true;
854 			break;
855 		}
856 	}
857 
858 	if (num > 0)
859 		kfree(vclock_index);
860 
861 	if (!match)
862 		return -EINVAL;
863 
864 	sk->sk_bind_phc = phc_index;
865 
866 	return 0;
867 }
868 
869 int sock_set_timestamping(struct sock *sk, int optname,
870 			  struct so_timestamping timestamping)
871 {
872 	int val = timestamping.flags;
873 	int ret;
874 
875 	if (val & ~SOF_TIMESTAMPING_MASK)
876 		return -EINVAL;
877 
878 	if (val & SOF_TIMESTAMPING_OPT_ID &&
879 	    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
880 		if (sk_is_tcp(sk)) {
881 			if ((1 << sk->sk_state) &
882 			    (TCPF_CLOSE | TCPF_LISTEN))
883 				return -EINVAL;
884 			atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
885 		} else {
886 			atomic_set(&sk->sk_tskey, 0);
887 		}
888 	}
889 
890 	if (val & SOF_TIMESTAMPING_OPT_STATS &&
891 	    !(val & SOF_TIMESTAMPING_OPT_TSONLY))
892 		return -EINVAL;
893 
894 	if (val & SOF_TIMESTAMPING_BIND_PHC) {
895 		ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
896 		if (ret)
897 			return ret;
898 	}
899 
900 	sk->sk_tsflags = val;
901 	sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
902 
903 	if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
904 		sock_enable_timestamp(sk,
905 				      SOCK_TIMESTAMPING_RX_SOFTWARE);
906 	else
907 		sock_disable_timestamp(sk,
908 				       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
909 	return 0;
910 }
911 
912 void sock_set_keepalive(struct sock *sk)
913 {
914 	lock_sock(sk);
915 	if (sk->sk_prot->keepalive)
916 		sk->sk_prot->keepalive(sk, true);
917 	sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
918 	release_sock(sk);
919 }
920 EXPORT_SYMBOL(sock_set_keepalive);
921 
922 static void __sock_set_rcvbuf(struct sock *sk, int val)
923 {
924 	/* Ensure val * 2 fits into an int, to prevent max_t() from treating it
925 	 * as a negative value.
926 	 */
927 	val = min_t(int, val, INT_MAX / 2);
928 	sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
929 
930 	/* We double it on the way in to account for "struct sk_buff" etc.
931 	 * overhead.   Applications assume that the SO_RCVBUF setting they make
932 	 * will allow that much actual data to be received on that socket.
933 	 *
934 	 * Applications are unaware that "struct sk_buff" and other overheads
935 	 * allocate from the receive buffer during socket buffer allocation.
936 	 *
937 	 * And after considering the possible alternatives, returning the value
938 	 * we actually used in getsockopt is the most desirable behavior.
939 	 */
940 	WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
941 }
942 
943 void sock_set_rcvbuf(struct sock *sk, int val)
944 {
945 	lock_sock(sk);
946 	__sock_set_rcvbuf(sk, val);
947 	release_sock(sk);
948 }
949 EXPORT_SYMBOL(sock_set_rcvbuf);
950 
951 static void __sock_set_mark(struct sock *sk, u32 val)
952 {
953 	if (val != sk->sk_mark) {
954 		sk->sk_mark = val;
955 		sk_dst_reset(sk);
956 	}
957 }
958 
959 void sock_set_mark(struct sock *sk, u32 val)
960 {
961 	lock_sock(sk);
962 	__sock_set_mark(sk, val);
963 	release_sock(sk);
964 }
965 EXPORT_SYMBOL(sock_set_mark);
966 
967 static void sock_release_reserved_memory(struct sock *sk, int bytes)
968 {
969 	/* Round down bytes to multiple of pages */
970 	bytes &= ~(SK_MEM_QUANTUM - 1);
971 
972 	WARN_ON(bytes > sk->sk_reserved_mem);
973 	sk->sk_reserved_mem -= bytes;
974 	sk_mem_reclaim(sk);
975 }
976 
977 static int sock_reserve_memory(struct sock *sk, int bytes)
978 {
979 	long allocated;
980 	bool charged;
981 	int pages;
982 
983 	if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
984 		return -EOPNOTSUPP;
985 
986 	if (!bytes)
987 		return 0;
988 
989 	pages = sk_mem_pages(bytes);
990 
991 	/* pre-charge to memcg */
992 	charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
993 					  GFP_KERNEL | __GFP_RETRY_MAYFAIL);
994 	if (!charged)
995 		return -ENOMEM;
996 
997 	/* pre-charge to forward_alloc */
998 	allocated = sk_memory_allocated_add(sk, pages);
999 	/* If the system goes into memory pressure with this
1000 	 * precharge, give up and return error.
1001 	 */
1002 	if (allocated > sk_prot_mem_limits(sk, 1)) {
1003 		sk_memory_allocated_sub(sk, pages);
1004 		mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1005 		return -ENOMEM;
1006 	}
1007 	sk->sk_forward_alloc += pages << SK_MEM_QUANTUM_SHIFT;
1008 
1009 	sk->sk_reserved_mem += pages << SK_MEM_QUANTUM_SHIFT;
1010 
1011 	return 0;
1012 }
1013 
1014 /*
1015  *	This is meant for all protocols to use and covers goings on
1016  *	at the socket level. Everything here is generic.
1017  */
1018 
1019 int sock_setsockopt(struct socket *sock, int level, int optname,
1020 		    sockptr_t optval, unsigned int optlen)
1021 {
1022 	struct so_timestamping timestamping;
1023 	struct sock_txtime sk_txtime;
1024 	struct sock *sk = sock->sk;
1025 	int val;
1026 	int valbool;
1027 	struct linger ling;
1028 	int ret = 0;
1029 
1030 	/*
1031 	 *	Options without arguments
1032 	 */
1033 
1034 	if (optname == SO_BINDTODEVICE)
1035 		return sock_setbindtodevice(sk, optval, optlen);
1036 
1037 	if (optlen < sizeof(int))
1038 		return -EINVAL;
1039 
1040 	if (copy_from_sockptr(&val, optval, sizeof(val)))
1041 		return -EFAULT;
1042 
1043 	valbool = val ? 1 : 0;
1044 
1045 	lock_sock(sk);
1046 
1047 	switch (optname) {
1048 	case SO_DEBUG:
1049 		if (val && !capable(CAP_NET_ADMIN))
1050 			ret = -EACCES;
1051 		else
1052 			sock_valbool_flag(sk, SOCK_DBG, valbool);
1053 		break;
1054 	case SO_REUSEADDR:
1055 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1056 		break;
1057 	case SO_REUSEPORT:
1058 		sk->sk_reuseport = valbool;
1059 		break;
1060 	case SO_TYPE:
1061 	case SO_PROTOCOL:
1062 	case SO_DOMAIN:
1063 	case SO_ERROR:
1064 		ret = -ENOPROTOOPT;
1065 		break;
1066 	case SO_DONTROUTE:
1067 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1068 		sk_dst_reset(sk);
1069 		break;
1070 	case SO_BROADCAST:
1071 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1072 		break;
1073 	case SO_SNDBUF:
1074 		/* Don't error on this BSD doesn't and if you think
1075 		 * about it this is right. Otherwise apps have to
1076 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1077 		 * are treated in BSD as hints
1078 		 */
1079 		val = min_t(u32, val, sysctl_wmem_max);
1080 set_sndbuf:
1081 		/* Ensure val * 2 fits into an int, to prevent max_t()
1082 		 * from treating it as a negative value.
1083 		 */
1084 		val = min_t(int, val, INT_MAX / 2);
1085 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1086 		WRITE_ONCE(sk->sk_sndbuf,
1087 			   max_t(int, val * 2, SOCK_MIN_SNDBUF));
1088 		/* Wake up sending tasks if we upped the value. */
1089 		sk->sk_write_space(sk);
1090 		break;
1091 
1092 	case SO_SNDBUFFORCE:
1093 		if (!capable(CAP_NET_ADMIN)) {
1094 			ret = -EPERM;
1095 			break;
1096 		}
1097 
1098 		/* No negative values (to prevent underflow, as val will be
1099 		 * multiplied by 2).
1100 		 */
1101 		if (val < 0)
1102 			val = 0;
1103 		goto set_sndbuf;
1104 
1105 	case SO_RCVBUF:
1106 		/* Don't error on this BSD doesn't and if you think
1107 		 * about it this is right. Otherwise apps have to
1108 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1109 		 * are treated in BSD as hints
1110 		 */
1111 		__sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1112 		break;
1113 
1114 	case SO_RCVBUFFORCE:
1115 		if (!capable(CAP_NET_ADMIN)) {
1116 			ret = -EPERM;
1117 			break;
1118 		}
1119 
1120 		/* No negative values (to prevent underflow, as val will be
1121 		 * multiplied by 2).
1122 		 */
1123 		__sock_set_rcvbuf(sk, max(val, 0));
1124 		break;
1125 
1126 	case SO_KEEPALIVE:
1127 		if (sk->sk_prot->keepalive)
1128 			sk->sk_prot->keepalive(sk, valbool);
1129 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1130 		break;
1131 
1132 	case SO_OOBINLINE:
1133 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1134 		break;
1135 
1136 	case SO_NO_CHECK:
1137 		sk->sk_no_check_tx = valbool;
1138 		break;
1139 
1140 	case SO_PRIORITY:
1141 		if ((val >= 0 && val <= 6) ||
1142 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1143 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1144 			sk->sk_priority = val;
1145 		else
1146 			ret = -EPERM;
1147 		break;
1148 
1149 	case SO_LINGER:
1150 		if (optlen < sizeof(ling)) {
1151 			ret = -EINVAL;	/* 1003.1g */
1152 			break;
1153 		}
1154 		if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1155 			ret = -EFAULT;
1156 			break;
1157 		}
1158 		if (!ling.l_onoff)
1159 			sock_reset_flag(sk, SOCK_LINGER);
1160 		else {
1161 #if (BITS_PER_LONG == 32)
1162 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1163 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1164 			else
1165 #endif
1166 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1167 			sock_set_flag(sk, SOCK_LINGER);
1168 		}
1169 		break;
1170 
1171 	case SO_BSDCOMPAT:
1172 		break;
1173 
1174 	case SO_PASSCRED:
1175 		if (valbool)
1176 			set_bit(SOCK_PASSCRED, &sock->flags);
1177 		else
1178 			clear_bit(SOCK_PASSCRED, &sock->flags);
1179 		break;
1180 
1181 	case SO_TIMESTAMP_OLD:
1182 	case SO_TIMESTAMP_NEW:
1183 	case SO_TIMESTAMPNS_OLD:
1184 	case SO_TIMESTAMPNS_NEW:
1185 		sock_set_timestamp(sk, optname, valbool);
1186 		break;
1187 
1188 	case SO_TIMESTAMPING_NEW:
1189 	case SO_TIMESTAMPING_OLD:
1190 		if (optlen == sizeof(timestamping)) {
1191 			if (copy_from_sockptr(&timestamping, optval,
1192 					      sizeof(timestamping))) {
1193 				ret = -EFAULT;
1194 				break;
1195 			}
1196 		} else {
1197 			memset(&timestamping, 0, sizeof(timestamping));
1198 			timestamping.flags = val;
1199 		}
1200 		ret = sock_set_timestamping(sk, optname, timestamping);
1201 		break;
1202 
1203 	case SO_RCVLOWAT:
1204 		if (val < 0)
1205 			val = INT_MAX;
1206 		if (sock->ops->set_rcvlowat)
1207 			ret = sock->ops->set_rcvlowat(sk, val);
1208 		else
1209 			WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1210 		break;
1211 
1212 	case SO_RCVTIMEO_OLD:
1213 	case SO_RCVTIMEO_NEW:
1214 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1215 				       optlen, optname == SO_RCVTIMEO_OLD);
1216 		break;
1217 
1218 	case SO_SNDTIMEO_OLD:
1219 	case SO_SNDTIMEO_NEW:
1220 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1221 				       optlen, optname == SO_SNDTIMEO_OLD);
1222 		break;
1223 
1224 	case SO_ATTACH_FILTER: {
1225 		struct sock_fprog fprog;
1226 
1227 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1228 		if (!ret)
1229 			ret = sk_attach_filter(&fprog, sk);
1230 		break;
1231 	}
1232 	case SO_ATTACH_BPF:
1233 		ret = -EINVAL;
1234 		if (optlen == sizeof(u32)) {
1235 			u32 ufd;
1236 
1237 			ret = -EFAULT;
1238 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1239 				break;
1240 
1241 			ret = sk_attach_bpf(ufd, sk);
1242 		}
1243 		break;
1244 
1245 	case SO_ATTACH_REUSEPORT_CBPF: {
1246 		struct sock_fprog fprog;
1247 
1248 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1249 		if (!ret)
1250 			ret = sk_reuseport_attach_filter(&fprog, sk);
1251 		break;
1252 	}
1253 	case SO_ATTACH_REUSEPORT_EBPF:
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_reuseport_attach_bpf(ufd, sk);
1263 		}
1264 		break;
1265 
1266 	case SO_DETACH_REUSEPORT_BPF:
1267 		ret = reuseport_detach_prog(sk);
1268 		break;
1269 
1270 	case SO_DETACH_FILTER:
1271 		ret = sk_detach_filter(sk);
1272 		break;
1273 
1274 	case SO_LOCK_FILTER:
1275 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1276 			ret = -EPERM;
1277 		else
1278 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1279 		break;
1280 
1281 	case SO_PASSSEC:
1282 		if (valbool)
1283 			set_bit(SOCK_PASSSEC, &sock->flags);
1284 		else
1285 			clear_bit(SOCK_PASSSEC, &sock->flags);
1286 		break;
1287 	case SO_MARK:
1288 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1289 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1290 			ret = -EPERM;
1291 			break;
1292 		}
1293 
1294 		__sock_set_mark(sk, val);
1295 		break;
1296 
1297 	case SO_RXQ_OVFL:
1298 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1299 		break;
1300 
1301 	case SO_WIFI_STATUS:
1302 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1303 		break;
1304 
1305 	case SO_PEEK_OFF:
1306 		if (sock->ops->set_peek_off)
1307 			ret = sock->ops->set_peek_off(sk, val);
1308 		else
1309 			ret = -EOPNOTSUPP;
1310 		break;
1311 
1312 	case SO_NOFCS:
1313 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1314 		break;
1315 
1316 	case SO_SELECT_ERR_QUEUE:
1317 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1318 		break;
1319 
1320 #ifdef CONFIG_NET_RX_BUSY_POLL
1321 	case SO_BUSY_POLL:
1322 		/* allow unprivileged users to decrease the value */
1323 		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1324 			ret = -EPERM;
1325 		else {
1326 			if (val < 0)
1327 				ret = -EINVAL;
1328 			else
1329 				WRITE_ONCE(sk->sk_ll_usec, val);
1330 		}
1331 		break;
1332 	case SO_PREFER_BUSY_POLL:
1333 		if (valbool && !capable(CAP_NET_ADMIN))
1334 			ret = -EPERM;
1335 		else
1336 			WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1337 		break;
1338 	case SO_BUSY_POLL_BUDGET:
1339 		if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1340 			ret = -EPERM;
1341 		} else {
1342 			if (val < 0 || val > U16_MAX)
1343 				ret = -EINVAL;
1344 			else
1345 				WRITE_ONCE(sk->sk_busy_poll_budget, val);
1346 		}
1347 		break;
1348 #endif
1349 
1350 	case SO_MAX_PACING_RATE:
1351 		{
1352 		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1353 
1354 		if (sizeof(ulval) != sizeof(val) &&
1355 		    optlen >= sizeof(ulval) &&
1356 		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1357 			ret = -EFAULT;
1358 			break;
1359 		}
1360 		if (ulval != ~0UL)
1361 			cmpxchg(&sk->sk_pacing_status,
1362 				SK_PACING_NONE,
1363 				SK_PACING_NEEDED);
1364 		sk->sk_max_pacing_rate = ulval;
1365 		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1366 		break;
1367 		}
1368 	case SO_INCOMING_CPU:
1369 		WRITE_ONCE(sk->sk_incoming_cpu, val);
1370 		break;
1371 
1372 	case SO_CNX_ADVICE:
1373 		if (val == 1)
1374 			dst_negative_advice(sk);
1375 		break;
1376 
1377 	case SO_ZEROCOPY:
1378 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1379 			if (!(sk_is_tcp(sk) ||
1380 			      (sk->sk_type == SOCK_DGRAM &&
1381 			       sk->sk_protocol == IPPROTO_UDP)))
1382 				ret = -EOPNOTSUPP;
1383 		} else if (sk->sk_family != PF_RDS) {
1384 			ret = -EOPNOTSUPP;
1385 		}
1386 		if (!ret) {
1387 			if (val < 0 || val > 1)
1388 				ret = -EINVAL;
1389 			else
1390 				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1391 		}
1392 		break;
1393 
1394 	case SO_TXTIME:
1395 		if (optlen != sizeof(struct sock_txtime)) {
1396 			ret = -EINVAL;
1397 			break;
1398 		} else if (copy_from_sockptr(&sk_txtime, optval,
1399 			   sizeof(struct sock_txtime))) {
1400 			ret = -EFAULT;
1401 			break;
1402 		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1403 			ret = -EINVAL;
1404 			break;
1405 		}
1406 		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1407 		 * scheduler has enough safe guards.
1408 		 */
1409 		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1410 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1411 			ret = -EPERM;
1412 			break;
1413 		}
1414 		sock_valbool_flag(sk, SOCK_TXTIME, true);
1415 		sk->sk_clockid = sk_txtime.clockid;
1416 		sk->sk_txtime_deadline_mode =
1417 			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1418 		sk->sk_txtime_report_errors =
1419 			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1420 		break;
1421 
1422 	case SO_BINDTOIFINDEX:
1423 		ret = sock_bindtoindex_locked(sk, val);
1424 		break;
1425 
1426 	case SO_BUF_LOCK:
1427 		if (val & ~SOCK_BUF_LOCK_MASK) {
1428 			ret = -EINVAL;
1429 			break;
1430 		}
1431 		sk->sk_userlocks = val | (sk->sk_userlocks &
1432 					  ~SOCK_BUF_LOCK_MASK);
1433 		break;
1434 
1435 	case SO_RESERVE_MEM:
1436 	{
1437 		int delta;
1438 
1439 		if (val < 0) {
1440 			ret = -EINVAL;
1441 			break;
1442 		}
1443 
1444 		delta = val - sk->sk_reserved_mem;
1445 		if (delta < 0)
1446 			sock_release_reserved_memory(sk, -delta);
1447 		else
1448 			ret = sock_reserve_memory(sk, delta);
1449 		break;
1450 	}
1451 
1452 	case SO_TXREHASH:
1453 		if (val < -1 || val > 1) {
1454 			ret = -EINVAL;
1455 			break;
1456 		}
1457 		/* Paired with READ_ONCE() in tcp_rtx_synack() */
1458 		WRITE_ONCE(sk->sk_txrehash, (u8)val);
1459 		break;
1460 
1461 	default:
1462 		ret = -ENOPROTOOPT;
1463 		break;
1464 	}
1465 	release_sock(sk);
1466 	return ret;
1467 }
1468 EXPORT_SYMBOL(sock_setsockopt);
1469 
1470 static const struct cred *sk_get_peer_cred(struct sock *sk)
1471 {
1472 	const struct cred *cred;
1473 
1474 	spin_lock(&sk->sk_peer_lock);
1475 	cred = get_cred(sk->sk_peer_cred);
1476 	spin_unlock(&sk->sk_peer_lock);
1477 
1478 	return cred;
1479 }
1480 
1481 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1482 			  struct ucred *ucred)
1483 {
1484 	ucred->pid = pid_vnr(pid);
1485 	ucred->uid = ucred->gid = -1;
1486 	if (cred) {
1487 		struct user_namespace *current_ns = current_user_ns();
1488 
1489 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1490 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1491 	}
1492 }
1493 
1494 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1495 {
1496 	struct user_namespace *user_ns = current_user_ns();
1497 	int i;
1498 
1499 	for (i = 0; i < src->ngroups; i++)
1500 		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1501 			return -EFAULT;
1502 
1503 	return 0;
1504 }
1505 
1506 int sock_getsockopt(struct socket *sock, int level, int optname,
1507 		    char __user *optval, int __user *optlen)
1508 {
1509 	struct sock *sk = sock->sk;
1510 
1511 	union {
1512 		int val;
1513 		u64 val64;
1514 		unsigned long ulval;
1515 		struct linger ling;
1516 		struct old_timeval32 tm32;
1517 		struct __kernel_old_timeval tm;
1518 		struct  __kernel_sock_timeval stm;
1519 		struct sock_txtime txtime;
1520 		struct so_timestamping timestamping;
1521 	} v;
1522 
1523 	int lv = sizeof(int);
1524 	int len;
1525 
1526 	if (get_user(len, optlen))
1527 		return -EFAULT;
1528 	if (len < 0)
1529 		return -EINVAL;
1530 
1531 	memset(&v, 0, sizeof(v));
1532 
1533 	switch (optname) {
1534 	case SO_DEBUG:
1535 		v.val = sock_flag(sk, SOCK_DBG);
1536 		break;
1537 
1538 	case SO_DONTROUTE:
1539 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1540 		break;
1541 
1542 	case SO_BROADCAST:
1543 		v.val = sock_flag(sk, SOCK_BROADCAST);
1544 		break;
1545 
1546 	case SO_SNDBUF:
1547 		v.val = sk->sk_sndbuf;
1548 		break;
1549 
1550 	case SO_RCVBUF:
1551 		v.val = sk->sk_rcvbuf;
1552 		break;
1553 
1554 	case SO_REUSEADDR:
1555 		v.val = sk->sk_reuse;
1556 		break;
1557 
1558 	case SO_REUSEPORT:
1559 		v.val = sk->sk_reuseport;
1560 		break;
1561 
1562 	case SO_KEEPALIVE:
1563 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1564 		break;
1565 
1566 	case SO_TYPE:
1567 		v.val = sk->sk_type;
1568 		break;
1569 
1570 	case SO_PROTOCOL:
1571 		v.val = sk->sk_protocol;
1572 		break;
1573 
1574 	case SO_DOMAIN:
1575 		v.val = sk->sk_family;
1576 		break;
1577 
1578 	case SO_ERROR:
1579 		v.val = -sock_error(sk);
1580 		if (v.val == 0)
1581 			v.val = xchg(&sk->sk_err_soft, 0);
1582 		break;
1583 
1584 	case SO_OOBINLINE:
1585 		v.val = sock_flag(sk, SOCK_URGINLINE);
1586 		break;
1587 
1588 	case SO_NO_CHECK:
1589 		v.val = sk->sk_no_check_tx;
1590 		break;
1591 
1592 	case SO_PRIORITY:
1593 		v.val = sk->sk_priority;
1594 		break;
1595 
1596 	case SO_LINGER:
1597 		lv		= sizeof(v.ling);
1598 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1599 		v.ling.l_linger	= sk->sk_lingertime / HZ;
1600 		break;
1601 
1602 	case SO_BSDCOMPAT:
1603 		break;
1604 
1605 	case SO_TIMESTAMP_OLD:
1606 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1607 				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1608 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1609 		break;
1610 
1611 	case SO_TIMESTAMPNS_OLD:
1612 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1613 		break;
1614 
1615 	case SO_TIMESTAMP_NEW:
1616 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1617 		break;
1618 
1619 	case SO_TIMESTAMPNS_NEW:
1620 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1621 		break;
1622 
1623 	case SO_TIMESTAMPING_OLD:
1624 		lv = sizeof(v.timestamping);
1625 		v.timestamping.flags = sk->sk_tsflags;
1626 		v.timestamping.bind_phc = sk->sk_bind_phc;
1627 		break;
1628 
1629 	case SO_RCVTIMEO_OLD:
1630 	case SO_RCVTIMEO_NEW:
1631 		lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1632 		break;
1633 
1634 	case SO_SNDTIMEO_OLD:
1635 	case SO_SNDTIMEO_NEW:
1636 		lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1637 		break;
1638 
1639 	case SO_RCVLOWAT:
1640 		v.val = sk->sk_rcvlowat;
1641 		break;
1642 
1643 	case SO_SNDLOWAT:
1644 		v.val = 1;
1645 		break;
1646 
1647 	case SO_PASSCRED:
1648 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1649 		break;
1650 
1651 	case SO_PEERCRED:
1652 	{
1653 		struct ucred peercred;
1654 		if (len > sizeof(peercred))
1655 			len = sizeof(peercred);
1656 
1657 		spin_lock(&sk->sk_peer_lock);
1658 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1659 		spin_unlock(&sk->sk_peer_lock);
1660 
1661 		if (copy_to_user(optval, &peercred, len))
1662 			return -EFAULT;
1663 		goto lenout;
1664 	}
1665 
1666 	case SO_PEERGROUPS:
1667 	{
1668 		const struct cred *cred;
1669 		int ret, n;
1670 
1671 		cred = sk_get_peer_cred(sk);
1672 		if (!cred)
1673 			return -ENODATA;
1674 
1675 		n = cred->group_info->ngroups;
1676 		if (len < n * sizeof(gid_t)) {
1677 			len = n * sizeof(gid_t);
1678 			put_cred(cred);
1679 			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1680 		}
1681 		len = n * sizeof(gid_t);
1682 
1683 		ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1684 		put_cred(cred);
1685 		if (ret)
1686 			return ret;
1687 		goto lenout;
1688 	}
1689 
1690 	case SO_PEERNAME:
1691 	{
1692 		char address[128];
1693 
1694 		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1695 		if (lv < 0)
1696 			return -ENOTCONN;
1697 		if (lv < len)
1698 			return -EINVAL;
1699 		if (copy_to_user(optval, address, len))
1700 			return -EFAULT;
1701 		goto lenout;
1702 	}
1703 
1704 	/* Dubious BSD thing... Probably nobody even uses it, but
1705 	 * the UNIX standard wants it for whatever reason... -DaveM
1706 	 */
1707 	case SO_ACCEPTCONN:
1708 		v.val = sk->sk_state == TCP_LISTEN;
1709 		break;
1710 
1711 	case SO_PASSSEC:
1712 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1713 		break;
1714 
1715 	case SO_PEERSEC:
1716 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1717 
1718 	case SO_MARK:
1719 		v.val = sk->sk_mark;
1720 		break;
1721 
1722 	case SO_RXQ_OVFL:
1723 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1724 		break;
1725 
1726 	case SO_WIFI_STATUS:
1727 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1728 		break;
1729 
1730 	case SO_PEEK_OFF:
1731 		if (!sock->ops->set_peek_off)
1732 			return -EOPNOTSUPP;
1733 
1734 		v.val = sk->sk_peek_off;
1735 		break;
1736 	case SO_NOFCS:
1737 		v.val = sock_flag(sk, SOCK_NOFCS);
1738 		break;
1739 
1740 	case SO_BINDTODEVICE:
1741 		return sock_getbindtodevice(sk, optval, optlen, len);
1742 
1743 	case SO_GET_FILTER:
1744 		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1745 		if (len < 0)
1746 			return len;
1747 
1748 		goto lenout;
1749 
1750 	case SO_LOCK_FILTER:
1751 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1752 		break;
1753 
1754 	case SO_BPF_EXTENSIONS:
1755 		v.val = bpf_tell_extensions();
1756 		break;
1757 
1758 	case SO_SELECT_ERR_QUEUE:
1759 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1760 		break;
1761 
1762 #ifdef CONFIG_NET_RX_BUSY_POLL
1763 	case SO_BUSY_POLL:
1764 		v.val = sk->sk_ll_usec;
1765 		break;
1766 	case SO_PREFER_BUSY_POLL:
1767 		v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1768 		break;
1769 #endif
1770 
1771 	case SO_MAX_PACING_RATE:
1772 		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1773 			lv = sizeof(v.ulval);
1774 			v.ulval = sk->sk_max_pacing_rate;
1775 		} else {
1776 			/* 32bit version */
1777 			v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1778 		}
1779 		break;
1780 
1781 	case SO_INCOMING_CPU:
1782 		v.val = READ_ONCE(sk->sk_incoming_cpu);
1783 		break;
1784 
1785 	case SO_MEMINFO:
1786 	{
1787 		u32 meminfo[SK_MEMINFO_VARS];
1788 
1789 		sk_get_meminfo(sk, meminfo);
1790 
1791 		len = min_t(unsigned int, len, sizeof(meminfo));
1792 		if (copy_to_user(optval, &meminfo, len))
1793 			return -EFAULT;
1794 
1795 		goto lenout;
1796 	}
1797 
1798 #ifdef CONFIG_NET_RX_BUSY_POLL
1799 	case SO_INCOMING_NAPI_ID:
1800 		v.val = READ_ONCE(sk->sk_napi_id);
1801 
1802 		/* aggregate non-NAPI IDs down to 0 */
1803 		if (v.val < MIN_NAPI_ID)
1804 			v.val = 0;
1805 
1806 		break;
1807 #endif
1808 
1809 	case SO_COOKIE:
1810 		lv = sizeof(u64);
1811 		if (len < lv)
1812 			return -EINVAL;
1813 		v.val64 = sock_gen_cookie(sk);
1814 		break;
1815 
1816 	case SO_ZEROCOPY:
1817 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1818 		break;
1819 
1820 	case SO_TXTIME:
1821 		lv = sizeof(v.txtime);
1822 		v.txtime.clockid = sk->sk_clockid;
1823 		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1824 				  SOF_TXTIME_DEADLINE_MODE : 0;
1825 		v.txtime.flags |= sk->sk_txtime_report_errors ?
1826 				  SOF_TXTIME_REPORT_ERRORS : 0;
1827 		break;
1828 
1829 	case SO_BINDTOIFINDEX:
1830 		v.val = sk->sk_bound_dev_if;
1831 		break;
1832 
1833 	case SO_NETNS_COOKIE:
1834 		lv = sizeof(u64);
1835 		if (len != lv)
1836 			return -EINVAL;
1837 		v.val64 = sock_net(sk)->net_cookie;
1838 		break;
1839 
1840 	case SO_BUF_LOCK:
1841 		v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1842 		break;
1843 
1844 	case SO_RESERVE_MEM:
1845 		v.val = sk->sk_reserved_mem;
1846 		break;
1847 
1848 	case SO_TXREHASH:
1849 		v.val = sk->sk_txrehash;
1850 		break;
1851 
1852 	default:
1853 		/* We implement the SO_SNDLOWAT etc to not be settable
1854 		 * (1003.1g 7).
1855 		 */
1856 		return -ENOPROTOOPT;
1857 	}
1858 
1859 	if (len > lv)
1860 		len = lv;
1861 	if (copy_to_user(optval, &v, len))
1862 		return -EFAULT;
1863 lenout:
1864 	if (put_user(len, optlen))
1865 		return -EFAULT;
1866 	return 0;
1867 }
1868 
1869 /*
1870  * Initialize an sk_lock.
1871  *
1872  * (We also register the sk_lock with the lock validator.)
1873  */
1874 static inline void sock_lock_init(struct sock *sk)
1875 {
1876 	if (sk->sk_kern_sock)
1877 		sock_lock_init_class_and_name(
1878 			sk,
1879 			af_family_kern_slock_key_strings[sk->sk_family],
1880 			af_family_kern_slock_keys + sk->sk_family,
1881 			af_family_kern_key_strings[sk->sk_family],
1882 			af_family_kern_keys + sk->sk_family);
1883 	else
1884 		sock_lock_init_class_and_name(
1885 			sk,
1886 			af_family_slock_key_strings[sk->sk_family],
1887 			af_family_slock_keys + sk->sk_family,
1888 			af_family_key_strings[sk->sk_family],
1889 			af_family_keys + sk->sk_family);
1890 }
1891 
1892 /*
1893  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1894  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1895  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1896  */
1897 static void sock_copy(struct sock *nsk, const struct sock *osk)
1898 {
1899 	const struct proto *prot = READ_ONCE(osk->sk_prot);
1900 #ifdef CONFIG_SECURITY_NETWORK
1901 	void *sptr = nsk->sk_security;
1902 #endif
1903 
1904 	/* If we move sk_tx_queue_mapping out of the private section,
1905 	 * we must check if sk_tx_queue_clear() is called after
1906 	 * sock_copy() in sk_clone_lock().
1907 	 */
1908 	BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1909 		     offsetof(struct sock, sk_dontcopy_begin) ||
1910 		     offsetof(struct sock, sk_tx_queue_mapping) >=
1911 		     offsetof(struct sock, sk_dontcopy_end));
1912 
1913 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1914 
1915 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1916 	       prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1917 
1918 #ifdef CONFIG_SECURITY_NETWORK
1919 	nsk->sk_security = sptr;
1920 	security_sk_clone(osk, nsk);
1921 #endif
1922 }
1923 
1924 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1925 		int family)
1926 {
1927 	struct sock *sk;
1928 	struct kmem_cache *slab;
1929 
1930 	slab = prot->slab;
1931 	if (slab != NULL) {
1932 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1933 		if (!sk)
1934 			return sk;
1935 		if (want_init_on_alloc(priority))
1936 			sk_prot_clear_nulls(sk, prot->obj_size);
1937 	} else
1938 		sk = kmalloc(prot->obj_size, priority);
1939 
1940 	if (sk != NULL) {
1941 		if (security_sk_alloc(sk, family, priority))
1942 			goto out_free;
1943 
1944 		if (!try_module_get(prot->owner))
1945 			goto out_free_sec;
1946 	}
1947 
1948 	return sk;
1949 
1950 out_free_sec:
1951 	security_sk_free(sk);
1952 out_free:
1953 	if (slab != NULL)
1954 		kmem_cache_free(slab, sk);
1955 	else
1956 		kfree(sk);
1957 	return NULL;
1958 }
1959 
1960 static void sk_prot_free(struct proto *prot, struct sock *sk)
1961 {
1962 	struct kmem_cache *slab;
1963 	struct module *owner;
1964 
1965 	owner = prot->owner;
1966 	slab = prot->slab;
1967 
1968 	cgroup_sk_free(&sk->sk_cgrp_data);
1969 	mem_cgroup_sk_free(sk);
1970 	security_sk_free(sk);
1971 	if (slab != NULL)
1972 		kmem_cache_free(slab, sk);
1973 	else
1974 		kfree(sk);
1975 	module_put(owner);
1976 }
1977 
1978 /**
1979  *	sk_alloc - All socket objects are allocated here
1980  *	@net: the applicable net namespace
1981  *	@family: protocol family
1982  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1983  *	@prot: struct proto associated with this new sock instance
1984  *	@kern: is this to be a kernel socket?
1985  */
1986 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1987 		      struct proto *prot, int kern)
1988 {
1989 	struct sock *sk;
1990 
1991 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1992 	if (sk) {
1993 		sk->sk_family = family;
1994 		/*
1995 		 * See comment in struct sock definition to understand
1996 		 * why we need sk_prot_creator -acme
1997 		 */
1998 		sk->sk_prot = sk->sk_prot_creator = prot;
1999 		sk->sk_kern_sock = kern;
2000 		sock_lock_init(sk);
2001 		sk->sk_net_refcnt = kern ? 0 : 1;
2002 		if (likely(sk->sk_net_refcnt)) {
2003 			get_net_track(net, &sk->ns_tracker, priority);
2004 			sock_inuse_add(net, 1);
2005 		}
2006 
2007 		sock_net_set(sk, net);
2008 		refcount_set(&sk->sk_wmem_alloc, 1);
2009 
2010 		mem_cgroup_sk_alloc(sk);
2011 		cgroup_sk_alloc(&sk->sk_cgrp_data);
2012 		sock_update_classid(&sk->sk_cgrp_data);
2013 		sock_update_netprioidx(&sk->sk_cgrp_data);
2014 		sk_tx_queue_clear(sk);
2015 	}
2016 
2017 	return sk;
2018 }
2019 EXPORT_SYMBOL(sk_alloc);
2020 
2021 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2022  * grace period. This is the case for UDP sockets and TCP listeners.
2023  */
2024 static void __sk_destruct(struct rcu_head *head)
2025 {
2026 	struct sock *sk = container_of(head, struct sock, sk_rcu);
2027 	struct sk_filter *filter;
2028 
2029 	if (sk->sk_destruct)
2030 		sk->sk_destruct(sk);
2031 
2032 	filter = rcu_dereference_check(sk->sk_filter,
2033 				       refcount_read(&sk->sk_wmem_alloc) == 0);
2034 	if (filter) {
2035 		sk_filter_uncharge(sk, filter);
2036 		RCU_INIT_POINTER(sk->sk_filter, NULL);
2037 	}
2038 
2039 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2040 
2041 #ifdef CONFIG_BPF_SYSCALL
2042 	bpf_sk_storage_free(sk);
2043 #endif
2044 
2045 	if (atomic_read(&sk->sk_omem_alloc))
2046 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
2047 			 __func__, atomic_read(&sk->sk_omem_alloc));
2048 
2049 	if (sk->sk_frag.page) {
2050 		put_page(sk->sk_frag.page);
2051 		sk->sk_frag.page = NULL;
2052 	}
2053 
2054 	/* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2055 	put_cred(sk->sk_peer_cred);
2056 	put_pid(sk->sk_peer_pid);
2057 
2058 	if (likely(sk->sk_net_refcnt))
2059 		put_net_track(sock_net(sk), &sk->ns_tracker);
2060 	sk_prot_free(sk->sk_prot_creator, sk);
2061 }
2062 
2063 void sk_destruct(struct sock *sk)
2064 {
2065 	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2066 
2067 	WARN_ON_ONCE(!llist_empty(&sk->defer_list));
2068 	sk_defer_free_flush(sk);
2069 
2070 	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2071 		reuseport_detach_sock(sk);
2072 		use_call_rcu = true;
2073 	}
2074 
2075 	if (use_call_rcu)
2076 		call_rcu(&sk->sk_rcu, __sk_destruct);
2077 	else
2078 		__sk_destruct(&sk->sk_rcu);
2079 }
2080 
2081 static void __sk_free(struct sock *sk)
2082 {
2083 	if (likely(sk->sk_net_refcnt))
2084 		sock_inuse_add(sock_net(sk), -1);
2085 
2086 	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2087 		sock_diag_broadcast_destroy(sk);
2088 	else
2089 		sk_destruct(sk);
2090 }
2091 
2092 void sk_free(struct sock *sk)
2093 {
2094 	/*
2095 	 * We subtract one from sk_wmem_alloc and can know if
2096 	 * some packets are still in some tx queue.
2097 	 * If not null, sock_wfree() will call __sk_free(sk) later
2098 	 */
2099 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2100 		__sk_free(sk);
2101 }
2102 EXPORT_SYMBOL(sk_free);
2103 
2104 static void sk_init_common(struct sock *sk)
2105 {
2106 	skb_queue_head_init(&sk->sk_receive_queue);
2107 	skb_queue_head_init(&sk->sk_write_queue);
2108 	skb_queue_head_init(&sk->sk_error_queue);
2109 
2110 	rwlock_init(&sk->sk_callback_lock);
2111 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2112 			af_rlock_keys + sk->sk_family,
2113 			af_family_rlock_key_strings[sk->sk_family]);
2114 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2115 			af_wlock_keys + sk->sk_family,
2116 			af_family_wlock_key_strings[sk->sk_family]);
2117 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2118 			af_elock_keys + sk->sk_family,
2119 			af_family_elock_key_strings[sk->sk_family]);
2120 	lockdep_set_class_and_name(&sk->sk_callback_lock,
2121 			af_callback_keys + sk->sk_family,
2122 			af_family_clock_key_strings[sk->sk_family]);
2123 }
2124 
2125 /**
2126  *	sk_clone_lock - clone a socket, and lock its clone
2127  *	@sk: the socket to clone
2128  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2129  *
2130  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2131  */
2132 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2133 {
2134 	struct proto *prot = READ_ONCE(sk->sk_prot);
2135 	struct sk_filter *filter;
2136 	bool is_charged = true;
2137 	struct sock *newsk;
2138 
2139 	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2140 	if (!newsk)
2141 		goto out;
2142 
2143 	sock_copy(newsk, sk);
2144 
2145 	newsk->sk_prot_creator = prot;
2146 
2147 	/* SANITY */
2148 	if (likely(newsk->sk_net_refcnt)) {
2149 		get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2150 		sock_inuse_add(sock_net(newsk), 1);
2151 	}
2152 	sk_node_init(&newsk->sk_node);
2153 	sock_lock_init(newsk);
2154 	bh_lock_sock(newsk);
2155 	newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
2156 	newsk->sk_backlog.len = 0;
2157 
2158 	atomic_set(&newsk->sk_rmem_alloc, 0);
2159 
2160 	/* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2161 	refcount_set(&newsk->sk_wmem_alloc, 1);
2162 
2163 	atomic_set(&newsk->sk_omem_alloc, 0);
2164 	sk_init_common(newsk);
2165 
2166 	newsk->sk_dst_cache	= NULL;
2167 	newsk->sk_dst_pending_confirm = 0;
2168 	newsk->sk_wmem_queued	= 0;
2169 	newsk->sk_forward_alloc = 0;
2170 	newsk->sk_reserved_mem  = 0;
2171 	atomic_set(&newsk->sk_drops, 0);
2172 	newsk->sk_send_head	= NULL;
2173 	newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2174 	atomic_set(&newsk->sk_zckey, 0);
2175 
2176 	sock_reset_flag(newsk, SOCK_DONE);
2177 
2178 	/* sk->sk_memcg will be populated at accept() time */
2179 	newsk->sk_memcg = NULL;
2180 
2181 	cgroup_sk_clone(&newsk->sk_cgrp_data);
2182 
2183 	rcu_read_lock();
2184 	filter = rcu_dereference(sk->sk_filter);
2185 	if (filter != NULL)
2186 		/* though it's an empty new sock, the charging may fail
2187 		 * if sysctl_optmem_max was changed between creation of
2188 		 * original socket and cloning
2189 		 */
2190 		is_charged = sk_filter_charge(newsk, filter);
2191 	RCU_INIT_POINTER(newsk->sk_filter, filter);
2192 	rcu_read_unlock();
2193 
2194 	if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2195 		/* We need to make sure that we don't uncharge the new
2196 		 * socket if we couldn't charge it in the first place
2197 		 * as otherwise we uncharge the parent's filter.
2198 		 */
2199 		if (!is_charged)
2200 			RCU_INIT_POINTER(newsk->sk_filter, NULL);
2201 		sk_free_unlock_clone(newsk);
2202 		newsk = NULL;
2203 		goto out;
2204 	}
2205 	RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2206 
2207 	if (bpf_sk_storage_clone(sk, newsk)) {
2208 		sk_free_unlock_clone(newsk);
2209 		newsk = NULL;
2210 		goto out;
2211 	}
2212 
2213 	/* Clear sk_user_data if parent had the pointer tagged
2214 	 * as not suitable for copying when cloning.
2215 	 */
2216 	if (sk_user_data_is_nocopy(newsk))
2217 		newsk->sk_user_data = NULL;
2218 
2219 	newsk->sk_err	   = 0;
2220 	newsk->sk_err_soft = 0;
2221 	newsk->sk_priority = 0;
2222 	newsk->sk_incoming_cpu = raw_smp_processor_id();
2223 
2224 	/* Before updating sk_refcnt, we must commit prior changes to memory
2225 	 * (Documentation/RCU/rculist_nulls.rst for details)
2226 	 */
2227 	smp_wmb();
2228 	refcount_set(&newsk->sk_refcnt, 2);
2229 
2230 	/* Increment the counter in the same struct proto as the master
2231 	 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2232 	 * is the same as sk->sk_prot->socks, as this field was copied
2233 	 * with memcpy).
2234 	 *
2235 	 * This _changes_ the previous behaviour, where
2236 	 * tcp_create_openreq_child always was incrementing the
2237 	 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2238 	 * to be taken into account in all callers. -acme
2239 	 */
2240 	sk_refcnt_debug_inc(newsk);
2241 	sk_set_socket(newsk, NULL);
2242 	sk_tx_queue_clear(newsk);
2243 	RCU_INIT_POINTER(newsk->sk_wq, NULL);
2244 
2245 	if (newsk->sk_prot->sockets_allocated)
2246 		sk_sockets_allocated_inc(newsk);
2247 
2248 	if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2249 		net_enable_timestamp();
2250 out:
2251 	return newsk;
2252 }
2253 EXPORT_SYMBOL_GPL(sk_clone_lock);
2254 
2255 void sk_free_unlock_clone(struct sock *sk)
2256 {
2257 	/* It is still raw copy of parent, so invalidate
2258 	 * destructor and make plain sk_free() */
2259 	sk->sk_destruct = NULL;
2260 	bh_unlock_sock(sk);
2261 	sk_free(sk);
2262 }
2263 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2264 
2265 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2266 {
2267 	u32 max_segs = 1;
2268 
2269 	sk_dst_set(sk, dst);
2270 	sk->sk_route_caps = dst->dev->features;
2271 	if (sk_is_tcp(sk))
2272 		sk->sk_route_caps |= NETIF_F_GSO;
2273 	if (sk->sk_route_caps & NETIF_F_GSO)
2274 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2275 	if (unlikely(sk->sk_gso_disabled))
2276 		sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2277 	if (sk_can_gso(sk)) {
2278 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2279 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2280 		} else {
2281 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2282 			/* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2283 			sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2284 			sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1);
2285 			/* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2286 			max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2287 		}
2288 	}
2289 	sk->sk_gso_max_segs = max_segs;
2290 }
2291 EXPORT_SYMBOL_GPL(sk_setup_caps);
2292 
2293 /*
2294  *	Simple resource managers for sockets.
2295  */
2296 
2297 
2298 /*
2299  * Write buffer destructor automatically called from kfree_skb.
2300  */
2301 void sock_wfree(struct sk_buff *skb)
2302 {
2303 	struct sock *sk = skb->sk;
2304 	unsigned int len = skb->truesize;
2305 
2306 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2307 		/*
2308 		 * Keep a reference on sk_wmem_alloc, this will be released
2309 		 * after sk_write_space() call
2310 		 */
2311 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2312 		sk->sk_write_space(sk);
2313 		len = 1;
2314 	}
2315 	/*
2316 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2317 	 * could not do because of in-flight packets
2318 	 */
2319 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2320 		__sk_free(sk);
2321 }
2322 EXPORT_SYMBOL(sock_wfree);
2323 
2324 /* This variant of sock_wfree() is used by TCP,
2325  * since it sets SOCK_USE_WRITE_QUEUE.
2326  */
2327 void __sock_wfree(struct sk_buff *skb)
2328 {
2329 	struct sock *sk = skb->sk;
2330 
2331 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2332 		__sk_free(sk);
2333 }
2334 
2335 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2336 {
2337 	skb_orphan(skb);
2338 	skb->sk = sk;
2339 #ifdef CONFIG_INET
2340 	if (unlikely(!sk_fullsock(sk))) {
2341 		skb->destructor = sock_edemux;
2342 		sock_hold(sk);
2343 		return;
2344 	}
2345 #endif
2346 	skb->destructor = sock_wfree;
2347 	skb_set_hash_from_sk(skb, sk);
2348 	/*
2349 	 * We used to take a refcount on sk, but following operation
2350 	 * is enough to guarantee sk_free() wont free this sock until
2351 	 * all in-flight packets are completed
2352 	 */
2353 	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2354 }
2355 EXPORT_SYMBOL(skb_set_owner_w);
2356 
2357 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2358 {
2359 #ifdef CONFIG_TLS_DEVICE
2360 	/* Drivers depend on in-order delivery for crypto offload,
2361 	 * partial orphan breaks out-of-order-OK logic.
2362 	 */
2363 	if (skb->decrypted)
2364 		return false;
2365 #endif
2366 	return (skb->destructor == sock_wfree ||
2367 		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2368 }
2369 
2370 /* This helper is used by netem, as it can hold packets in its
2371  * delay queue. We want to allow the owner socket to send more
2372  * packets, as if they were already TX completed by a typical driver.
2373  * But we also want to keep skb->sk set because some packet schedulers
2374  * rely on it (sch_fq for example).
2375  */
2376 void skb_orphan_partial(struct sk_buff *skb)
2377 {
2378 	if (skb_is_tcp_pure_ack(skb))
2379 		return;
2380 
2381 	if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2382 		return;
2383 
2384 	skb_orphan(skb);
2385 }
2386 EXPORT_SYMBOL(skb_orphan_partial);
2387 
2388 /*
2389  * Read buffer destructor automatically called from kfree_skb.
2390  */
2391 void sock_rfree(struct sk_buff *skb)
2392 {
2393 	struct sock *sk = skb->sk;
2394 	unsigned int len = skb->truesize;
2395 
2396 	atomic_sub(len, &sk->sk_rmem_alloc);
2397 	sk_mem_uncharge(sk, len);
2398 }
2399 EXPORT_SYMBOL(sock_rfree);
2400 
2401 /*
2402  * Buffer destructor for skbs that are not used directly in read or write
2403  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2404  */
2405 void sock_efree(struct sk_buff *skb)
2406 {
2407 	sock_put(skb->sk);
2408 }
2409 EXPORT_SYMBOL(sock_efree);
2410 
2411 /* Buffer destructor for prefetch/receive path where reference count may
2412  * not be held, e.g. for listen sockets.
2413  */
2414 #ifdef CONFIG_INET
2415 void sock_pfree(struct sk_buff *skb)
2416 {
2417 	if (sk_is_refcounted(skb->sk))
2418 		sock_gen_put(skb->sk);
2419 }
2420 EXPORT_SYMBOL(sock_pfree);
2421 #endif /* CONFIG_INET */
2422 
2423 kuid_t sock_i_uid(struct sock *sk)
2424 {
2425 	kuid_t uid;
2426 
2427 	read_lock_bh(&sk->sk_callback_lock);
2428 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2429 	read_unlock_bh(&sk->sk_callback_lock);
2430 	return uid;
2431 }
2432 EXPORT_SYMBOL(sock_i_uid);
2433 
2434 unsigned long sock_i_ino(struct sock *sk)
2435 {
2436 	unsigned long ino;
2437 
2438 	read_lock_bh(&sk->sk_callback_lock);
2439 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2440 	read_unlock_bh(&sk->sk_callback_lock);
2441 	return ino;
2442 }
2443 EXPORT_SYMBOL(sock_i_ino);
2444 
2445 /*
2446  * Allocate a skb from the socket's send buffer.
2447  */
2448 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2449 			     gfp_t priority)
2450 {
2451 	if (force ||
2452 	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2453 		struct sk_buff *skb = alloc_skb(size, priority);
2454 
2455 		if (skb) {
2456 			skb_set_owner_w(skb, sk);
2457 			return skb;
2458 		}
2459 	}
2460 	return NULL;
2461 }
2462 EXPORT_SYMBOL(sock_wmalloc);
2463 
2464 static void sock_ofree(struct sk_buff *skb)
2465 {
2466 	struct sock *sk = skb->sk;
2467 
2468 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2469 }
2470 
2471 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2472 			     gfp_t priority)
2473 {
2474 	struct sk_buff *skb;
2475 
2476 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2477 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2478 	    sysctl_optmem_max)
2479 		return NULL;
2480 
2481 	skb = alloc_skb(size, priority);
2482 	if (!skb)
2483 		return NULL;
2484 
2485 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2486 	skb->sk = sk;
2487 	skb->destructor = sock_ofree;
2488 	return skb;
2489 }
2490 
2491 /*
2492  * Allocate a memory block from the socket's option memory buffer.
2493  */
2494 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2495 {
2496 	if ((unsigned int)size <= sysctl_optmem_max &&
2497 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2498 		void *mem;
2499 		/* First do the add, to avoid the race if kmalloc
2500 		 * might sleep.
2501 		 */
2502 		atomic_add(size, &sk->sk_omem_alloc);
2503 		mem = kmalloc(size, priority);
2504 		if (mem)
2505 			return mem;
2506 		atomic_sub(size, &sk->sk_omem_alloc);
2507 	}
2508 	return NULL;
2509 }
2510 EXPORT_SYMBOL(sock_kmalloc);
2511 
2512 /* Free an option memory block. Note, we actually want the inline
2513  * here as this allows gcc to detect the nullify and fold away the
2514  * condition entirely.
2515  */
2516 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2517 				  const bool nullify)
2518 {
2519 	if (WARN_ON_ONCE(!mem))
2520 		return;
2521 	if (nullify)
2522 		kfree_sensitive(mem);
2523 	else
2524 		kfree(mem);
2525 	atomic_sub(size, &sk->sk_omem_alloc);
2526 }
2527 
2528 void sock_kfree_s(struct sock *sk, void *mem, int size)
2529 {
2530 	__sock_kfree_s(sk, mem, size, false);
2531 }
2532 EXPORT_SYMBOL(sock_kfree_s);
2533 
2534 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2535 {
2536 	__sock_kfree_s(sk, mem, size, true);
2537 }
2538 EXPORT_SYMBOL(sock_kzfree_s);
2539 
2540 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2541    I think, these locks should be removed for datagram sockets.
2542  */
2543 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2544 {
2545 	DEFINE_WAIT(wait);
2546 
2547 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2548 	for (;;) {
2549 		if (!timeo)
2550 			break;
2551 		if (signal_pending(current))
2552 			break;
2553 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2554 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2555 		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2556 			break;
2557 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2558 			break;
2559 		if (sk->sk_err)
2560 			break;
2561 		timeo = schedule_timeout(timeo);
2562 	}
2563 	finish_wait(sk_sleep(sk), &wait);
2564 	return timeo;
2565 }
2566 
2567 
2568 /*
2569  *	Generic send/receive buffer handlers
2570  */
2571 
2572 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2573 				     unsigned long data_len, int noblock,
2574 				     int *errcode, int max_page_order)
2575 {
2576 	struct sk_buff *skb;
2577 	long timeo;
2578 	int err;
2579 
2580 	timeo = sock_sndtimeo(sk, noblock);
2581 	for (;;) {
2582 		err = sock_error(sk);
2583 		if (err != 0)
2584 			goto failure;
2585 
2586 		err = -EPIPE;
2587 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2588 			goto failure;
2589 
2590 		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2591 			break;
2592 
2593 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2594 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2595 		err = -EAGAIN;
2596 		if (!timeo)
2597 			goto failure;
2598 		if (signal_pending(current))
2599 			goto interrupted;
2600 		timeo = sock_wait_for_wmem(sk, timeo);
2601 	}
2602 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2603 				   errcode, sk->sk_allocation);
2604 	if (skb)
2605 		skb_set_owner_w(skb, sk);
2606 	return skb;
2607 
2608 interrupted:
2609 	err = sock_intr_errno(timeo);
2610 failure:
2611 	*errcode = err;
2612 	return NULL;
2613 }
2614 EXPORT_SYMBOL(sock_alloc_send_pskb);
2615 
2616 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2617 				    int noblock, int *errcode)
2618 {
2619 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2620 }
2621 EXPORT_SYMBOL(sock_alloc_send_skb);
2622 
2623 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2624 		     struct sockcm_cookie *sockc)
2625 {
2626 	u32 tsflags;
2627 
2628 	switch (cmsg->cmsg_type) {
2629 	case SO_MARK:
2630 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2631 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2632 			return -EPERM;
2633 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2634 			return -EINVAL;
2635 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2636 		break;
2637 	case SO_TIMESTAMPING_OLD:
2638 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2639 			return -EINVAL;
2640 
2641 		tsflags = *(u32 *)CMSG_DATA(cmsg);
2642 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2643 			return -EINVAL;
2644 
2645 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2646 		sockc->tsflags |= tsflags;
2647 		break;
2648 	case SCM_TXTIME:
2649 		if (!sock_flag(sk, SOCK_TXTIME))
2650 			return -EINVAL;
2651 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2652 			return -EINVAL;
2653 		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2654 		break;
2655 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2656 	case SCM_RIGHTS:
2657 	case SCM_CREDENTIALS:
2658 		break;
2659 	default:
2660 		return -EINVAL;
2661 	}
2662 	return 0;
2663 }
2664 EXPORT_SYMBOL(__sock_cmsg_send);
2665 
2666 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2667 		   struct sockcm_cookie *sockc)
2668 {
2669 	struct cmsghdr *cmsg;
2670 	int ret;
2671 
2672 	for_each_cmsghdr(cmsg, msg) {
2673 		if (!CMSG_OK(msg, cmsg))
2674 			return -EINVAL;
2675 		if (cmsg->cmsg_level != SOL_SOCKET)
2676 			continue;
2677 		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2678 		if (ret)
2679 			return ret;
2680 	}
2681 	return 0;
2682 }
2683 EXPORT_SYMBOL(sock_cmsg_send);
2684 
2685 static void sk_enter_memory_pressure(struct sock *sk)
2686 {
2687 	if (!sk->sk_prot->enter_memory_pressure)
2688 		return;
2689 
2690 	sk->sk_prot->enter_memory_pressure(sk);
2691 }
2692 
2693 static void sk_leave_memory_pressure(struct sock *sk)
2694 {
2695 	if (sk->sk_prot->leave_memory_pressure) {
2696 		sk->sk_prot->leave_memory_pressure(sk);
2697 	} else {
2698 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2699 
2700 		if (memory_pressure && READ_ONCE(*memory_pressure))
2701 			WRITE_ONCE(*memory_pressure, 0);
2702 	}
2703 }
2704 
2705 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2706 
2707 /**
2708  * skb_page_frag_refill - check that a page_frag contains enough room
2709  * @sz: minimum size of the fragment we want to get
2710  * @pfrag: pointer to page_frag
2711  * @gfp: priority for memory allocation
2712  *
2713  * Note: While this allocator tries to use high order pages, there is
2714  * no guarantee that allocations succeed. Therefore, @sz MUST be
2715  * less or equal than PAGE_SIZE.
2716  */
2717 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2718 {
2719 	if (pfrag->page) {
2720 		if (page_ref_count(pfrag->page) == 1) {
2721 			pfrag->offset = 0;
2722 			return true;
2723 		}
2724 		if (pfrag->offset + sz <= pfrag->size)
2725 			return true;
2726 		put_page(pfrag->page);
2727 	}
2728 
2729 	pfrag->offset = 0;
2730 	if (SKB_FRAG_PAGE_ORDER &&
2731 	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2732 		/* Avoid direct reclaim but allow kswapd to wake */
2733 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2734 					  __GFP_COMP | __GFP_NOWARN |
2735 					  __GFP_NORETRY,
2736 					  SKB_FRAG_PAGE_ORDER);
2737 		if (likely(pfrag->page)) {
2738 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2739 			return true;
2740 		}
2741 	}
2742 	pfrag->page = alloc_page(gfp);
2743 	if (likely(pfrag->page)) {
2744 		pfrag->size = PAGE_SIZE;
2745 		return true;
2746 	}
2747 	return false;
2748 }
2749 EXPORT_SYMBOL(skb_page_frag_refill);
2750 
2751 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2752 {
2753 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2754 		return true;
2755 
2756 	sk_enter_memory_pressure(sk);
2757 	sk_stream_moderate_sndbuf(sk);
2758 	return false;
2759 }
2760 EXPORT_SYMBOL(sk_page_frag_refill);
2761 
2762 void __lock_sock(struct sock *sk)
2763 	__releases(&sk->sk_lock.slock)
2764 	__acquires(&sk->sk_lock.slock)
2765 {
2766 	DEFINE_WAIT(wait);
2767 
2768 	for (;;) {
2769 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2770 					TASK_UNINTERRUPTIBLE);
2771 		spin_unlock_bh(&sk->sk_lock.slock);
2772 		schedule();
2773 		spin_lock_bh(&sk->sk_lock.slock);
2774 		if (!sock_owned_by_user(sk))
2775 			break;
2776 	}
2777 	finish_wait(&sk->sk_lock.wq, &wait);
2778 }
2779 
2780 void __release_sock(struct sock *sk)
2781 	__releases(&sk->sk_lock.slock)
2782 	__acquires(&sk->sk_lock.slock)
2783 {
2784 	struct sk_buff *skb, *next;
2785 
2786 	while ((skb = sk->sk_backlog.head) != NULL) {
2787 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2788 
2789 		spin_unlock_bh(&sk->sk_lock.slock);
2790 
2791 		do {
2792 			next = skb->next;
2793 			prefetch(next);
2794 			WARN_ON_ONCE(skb_dst_is_noref(skb));
2795 			skb_mark_not_on_list(skb);
2796 			sk_backlog_rcv(sk, skb);
2797 
2798 			cond_resched();
2799 
2800 			skb = next;
2801 		} while (skb != NULL);
2802 
2803 		spin_lock_bh(&sk->sk_lock.slock);
2804 	}
2805 
2806 	/*
2807 	 * Doing the zeroing here guarantee we can not loop forever
2808 	 * while a wild producer attempts to flood us.
2809 	 */
2810 	sk->sk_backlog.len = 0;
2811 }
2812 
2813 void __sk_flush_backlog(struct sock *sk)
2814 {
2815 	spin_lock_bh(&sk->sk_lock.slock);
2816 	__release_sock(sk);
2817 	spin_unlock_bh(&sk->sk_lock.slock);
2818 }
2819 
2820 /**
2821  * sk_wait_data - wait for data to arrive at sk_receive_queue
2822  * @sk:    sock to wait on
2823  * @timeo: for how long
2824  * @skb:   last skb seen on sk_receive_queue
2825  *
2826  * Now socket state including sk->sk_err is changed only under lock,
2827  * hence we may omit checks after joining wait queue.
2828  * We check receive queue before schedule() only as optimization;
2829  * it is very likely that release_sock() added new data.
2830  */
2831 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2832 {
2833 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2834 	int rc;
2835 
2836 	add_wait_queue(sk_sleep(sk), &wait);
2837 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2838 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2839 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2840 	remove_wait_queue(sk_sleep(sk), &wait);
2841 	return rc;
2842 }
2843 EXPORT_SYMBOL(sk_wait_data);
2844 
2845 /**
2846  *	__sk_mem_raise_allocated - increase memory_allocated
2847  *	@sk: socket
2848  *	@size: memory size to allocate
2849  *	@amt: pages to allocate
2850  *	@kind: allocation type
2851  *
2852  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2853  */
2854 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2855 {
2856 	struct proto *prot = sk->sk_prot;
2857 	long allocated = sk_memory_allocated_add(sk, amt);
2858 	bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2859 	bool charged = true;
2860 
2861 	if (memcg_charge &&
2862 	    !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2863 						gfp_memcg_charge())))
2864 		goto suppress_allocation;
2865 
2866 	/* Under limit. */
2867 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2868 		sk_leave_memory_pressure(sk);
2869 		return 1;
2870 	}
2871 
2872 	/* Under pressure. */
2873 	if (allocated > sk_prot_mem_limits(sk, 1))
2874 		sk_enter_memory_pressure(sk);
2875 
2876 	/* Over hard limit. */
2877 	if (allocated > sk_prot_mem_limits(sk, 2))
2878 		goto suppress_allocation;
2879 
2880 	/* guarantee minimum buffer size under pressure */
2881 	if (kind == SK_MEM_RECV) {
2882 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2883 			return 1;
2884 
2885 	} else { /* SK_MEM_SEND */
2886 		int wmem0 = sk_get_wmem0(sk, prot);
2887 
2888 		if (sk->sk_type == SOCK_STREAM) {
2889 			if (sk->sk_wmem_queued < wmem0)
2890 				return 1;
2891 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2892 				return 1;
2893 		}
2894 	}
2895 
2896 	if (sk_has_memory_pressure(sk)) {
2897 		u64 alloc;
2898 
2899 		if (!sk_under_memory_pressure(sk))
2900 			return 1;
2901 		alloc = sk_sockets_allocated_read_positive(sk);
2902 		if (sk_prot_mem_limits(sk, 2) > alloc *
2903 		    sk_mem_pages(sk->sk_wmem_queued +
2904 				 atomic_read(&sk->sk_rmem_alloc) +
2905 				 sk->sk_forward_alloc))
2906 			return 1;
2907 	}
2908 
2909 suppress_allocation:
2910 
2911 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2912 		sk_stream_moderate_sndbuf(sk);
2913 
2914 		/* Fail only if socket is _under_ its sndbuf.
2915 		 * In this case we cannot block, so that we have to fail.
2916 		 */
2917 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2918 			/* Force charge with __GFP_NOFAIL */
2919 			if (memcg_charge && !charged) {
2920 				mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2921 					gfp_memcg_charge() | __GFP_NOFAIL);
2922 			}
2923 			return 1;
2924 		}
2925 	}
2926 
2927 	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2928 		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2929 
2930 	sk_memory_allocated_sub(sk, amt);
2931 
2932 	if (memcg_charge && charged)
2933 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2934 
2935 	return 0;
2936 }
2937 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2938 
2939 /**
2940  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2941  *	@sk: socket
2942  *	@size: memory size to allocate
2943  *	@kind: allocation type
2944  *
2945  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2946  *	rmem allocation. This function assumes that protocols which have
2947  *	memory_pressure use sk_wmem_queued as write buffer accounting.
2948  */
2949 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2950 {
2951 	int ret, amt = sk_mem_pages(size);
2952 
2953 	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2954 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2955 	if (!ret)
2956 		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2957 	return ret;
2958 }
2959 EXPORT_SYMBOL(__sk_mem_schedule);
2960 
2961 /**
2962  *	__sk_mem_reduce_allocated - reclaim memory_allocated
2963  *	@sk: socket
2964  *	@amount: number of quanta
2965  *
2966  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2967  */
2968 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2969 {
2970 	sk_memory_allocated_sub(sk, amount);
2971 
2972 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2973 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2974 
2975 	if (sk_under_memory_pressure(sk) &&
2976 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2977 		sk_leave_memory_pressure(sk);
2978 }
2979 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2980 
2981 /**
2982  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2983  *	@sk: socket
2984  *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2985  */
2986 void __sk_mem_reclaim(struct sock *sk, int amount)
2987 {
2988 	amount >>= SK_MEM_QUANTUM_SHIFT;
2989 	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2990 	__sk_mem_reduce_allocated(sk, amount);
2991 }
2992 EXPORT_SYMBOL(__sk_mem_reclaim);
2993 
2994 int sk_set_peek_off(struct sock *sk, int val)
2995 {
2996 	sk->sk_peek_off = val;
2997 	return 0;
2998 }
2999 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3000 
3001 /*
3002  * Set of default routines for initialising struct proto_ops when
3003  * the protocol does not support a particular function. In certain
3004  * cases where it makes no sense for a protocol to have a "do nothing"
3005  * function, some default processing is provided.
3006  */
3007 
3008 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3009 {
3010 	return -EOPNOTSUPP;
3011 }
3012 EXPORT_SYMBOL(sock_no_bind);
3013 
3014 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3015 		    int len, int flags)
3016 {
3017 	return -EOPNOTSUPP;
3018 }
3019 EXPORT_SYMBOL(sock_no_connect);
3020 
3021 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3022 {
3023 	return -EOPNOTSUPP;
3024 }
3025 EXPORT_SYMBOL(sock_no_socketpair);
3026 
3027 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3028 		   bool kern)
3029 {
3030 	return -EOPNOTSUPP;
3031 }
3032 EXPORT_SYMBOL(sock_no_accept);
3033 
3034 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3035 		    int peer)
3036 {
3037 	return -EOPNOTSUPP;
3038 }
3039 EXPORT_SYMBOL(sock_no_getname);
3040 
3041 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3042 {
3043 	return -EOPNOTSUPP;
3044 }
3045 EXPORT_SYMBOL(sock_no_ioctl);
3046 
3047 int sock_no_listen(struct socket *sock, int backlog)
3048 {
3049 	return -EOPNOTSUPP;
3050 }
3051 EXPORT_SYMBOL(sock_no_listen);
3052 
3053 int sock_no_shutdown(struct socket *sock, int how)
3054 {
3055 	return -EOPNOTSUPP;
3056 }
3057 EXPORT_SYMBOL(sock_no_shutdown);
3058 
3059 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3060 {
3061 	return -EOPNOTSUPP;
3062 }
3063 EXPORT_SYMBOL(sock_no_sendmsg);
3064 
3065 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3066 {
3067 	return -EOPNOTSUPP;
3068 }
3069 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3070 
3071 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3072 		    int flags)
3073 {
3074 	return -EOPNOTSUPP;
3075 }
3076 EXPORT_SYMBOL(sock_no_recvmsg);
3077 
3078 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3079 {
3080 	/* Mirror missing mmap method error code */
3081 	return -ENODEV;
3082 }
3083 EXPORT_SYMBOL(sock_no_mmap);
3084 
3085 /*
3086  * When a file is received (via SCM_RIGHTS, etc), we must bump the
3087  * various sock-based usage counts.
3088  */
3089 void __receive_sock(struct file *file)
3090 {
3091 	struct socket *sock;
3092 
3093 	sock = sock_from_file(file);
3094 	if (sock) {
3095 		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3096 		sock_update_classid(&sock->sk->sk_cgrp_data);
3097 	}
3098 }
3099 
3100 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3101 {
3102 	ssize_t res;
3103 	struct msghdr msg = {.msg_flags = flags};
3104 	struct kvec iov;
3105 	char *kaddr = kmap(page);
3106 	iov.iov_base = kaddr + offset;
3107 	iov.iov_len = size;
3108 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3109 	kunmap(page);
3110 	return res;
3111 }
3112 EXPORT_SYMBOL(sock_no_sendpage);
3113 
3114 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3115 				int offset, size_t size, int flags)
3116 {
3117 	ssize_t res;
3118 	struct msghdr msg = {.msg_flags = flags};
3119 	struct kvec iov;
3120 	char *kaddr = kmap(page);
3121 
3122 	iov.iov_base = kaddr + offset;
3123 	iov.iov_len = size;
3124 	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3125 	kunmap(page);
3126 	return res;
3127 }
3128 EXPORT_SYMBOL(sock_no_sendpage_locked);
3129 
3130 /*
3131  *	Default Socket Callbacks
3132  */
3133 
3134 static void sock_def_wakeup(struct sock *sk)
3135 {
3136 	struct socket_wq *wq;
3137 
3138 	rcu_read_lock();
3139 	wq = rcu_dereference(sk->sk_wq);
3140 	if (skwq_has_sleeper(wq))
3141 		wake_up_interruptible_all(&wq->wait);
3142 	rcu_read_unlock();
3143 }
3144 
3145 static void sock_def_error_report(struct sock *sk)
3146 {
3147 	struct socket_wq *wq;
3148 
3149 	rcu_read_lock();
3150 	wq = rcu_dereference(sk->sk_wq);
3151 	if (skwq_has_sleeper(wq))
3152 		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3153 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3154 	rcu_read_unlock();
3155 }
3156 
3157 void sock_def_readable(struct sock *sk)
3158 {
3159 	struct socket_wq *wq;
3160 
3161 	rcu_read_lock();
3162 	wq = rcu_dereference(sk->sk_wq);
3163 	if (skwq_has_sleeper(wq))
3164 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3165 						EPOLLRDNORM | EPOLLRDBAND);
3166 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3167 	rcu_read_unlock();
3168 }
3169 
3170 static void sock_def_write_space(struct sock *sk)
3171 {
3172 	struct socket_wq *wq;
3173 
3174 	rcu_read_lock();
3175 
3176 	/* Do not wake up a writer until he can make "significant"
3177 	 * progress.  --DaveM
3178 	 */
3179 	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3180 		wq = rcu_dereference(sk->sk_wq);
3181 		if (skwq_has_sleeper(wq))
3182 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3183 						EPOLLWRNORM | EPOLLWRBAND);
3184 
3185 		/* Should agree with poll, otherwise some programs break */
3186 		if (sock_writeable(sk))
3187 			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3188 	}
3189 
3190 	rcu_read_unlock();
3191 }
3192 
3193 static void sock_def_destruct(struct sock *sk)
3194 {
3195 }
3196 
3197 void sk_send_sigurg(struct sock *sk)
3198 {
3199 	if (sk->sk_socket && sk->sk_socket->file)
3200 		if (send_sigurg(&sk->sk_socket->file->f_owner))
3201 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3202 }
3203 EXPORT_SYMBOL(sk_send_sigurg);
3204 
3205 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3206 		    unsigned long expires)
3207 {
3208 	if (!mod_timer(timer, expires))
3209 		sock_hold(sk);
3210 }
3211 EXPORT_SYMBOL(sk_reset_timer);
3212 
3213 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3214 {
3215 	if (del_timer(timer))
3216 		__sock_put(sk);
3217 }
3218 EXPORT_SYMBOL(sk_stop_timer);
3219 
3220 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3221 {
3222 	if (del_timer_sync(timer))
3223 		__sock_put(sk);
3224 }
3225 EXPORT_SYMBOL(sk_stop_timer_sync);
3226 
3227 void sock_init_data(struct socket *sock, struct sock *sk)
3228 {
3229 	sk_init_common(sk);
3230 	sk->sk_send_head	=	NULL;
3231 
3232 	timer_setup(&sk->sk_timer, NULL, 0);
3233 
3234 	sk->sk_allocation	=	GFP_KERNEL;
3235 	sk->sk_rcvbuf		=	sysctl_rmem_default;
3236 	sk->sk_sndbuf		=	sysctl_wmem_default;
3237 	sk->sk_state		=	TCP_CLOSE;
3238 	sk_set_socket(sk, sock);
3239 
3240 	sock_set_flag(sk, SOCK_ZAPPED);
3241 
3242 	if (sock) {
3243 		sk->sk_type	=	sock->type;
3244 		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3245 		sock->sk	=	sk;
3246 		sk->sk_uid	=	SOCK_INODE(sock)->i_uid;
3247 	} else {
3248 		RCU_INIT_POINTER(sk->sk_wq, NULL);
3249 		sk->sk_uid	=	make_kuid(sock_net(sk)->user_ns, 0);
3250 	}
3251 
3252 	rwlock_init(&sk->sk_callback_lock);
3253 	if (sk->sk_kern_sock)
3254 		lockdep_set_class_and_name(
3255 			&sk->sk_callback_lock,
3256 			af_kern_callback_keys + sk->sk_family,
3257 			af_family_kern_clock_key_strings[sk->sk_family]);
3258 	else
3259 		lockdep_set_class_and_name(
3260 			&sk->sk_callback_lock,
3261 			af_callback_keys + sk->sk_family,
3262 			af_family_clock_key_strings[sk->sk_family]);
3263 
3264 	sk->sk_state_change	=	sock_def_wakeup;
3265 	sk->sk_data_ready	=	sock_def_readable;
3266 	sk->sk_write_space	=	sock_def_write_space;
3267 	sk->sk_error_report	=	sock_def_error_report;
3268 	sk->sk_destruct		=	sock_def_destruct;
3269 
3270 	sk->sk_frag.page	=	NULL;
3271 	sk->sk_frag.offset	=	0;
3272 	sk->sk_peek_off		=	-1;
3273 
3274 	sk->sk_peer_pid 	=	NULL;
3275 	sk->sk_peer_cred	=	NULL;
3276 	spin_lock_init(&sk->sk_peer_lock);
3277 
3278 	sk->sk_write_pending	=	0;
3279 	sk->sk_rcvlowat		=	1;
3280 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3281 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3282 
3283 	sk->sk_stamp = SK_DEFAULT_STAMP;
3284 #if BITS_PER_LONG==32
3285 	seqlock_init(&sk->sk_stamp_seq);
3286 #endif
3287 	atomic_set(&sk->sk_zckey, 0);
3288 
3289 #ifdef CONFIG_NET_RX_BUSY_POLL
3290 	sk->sk_napi_id		=	0;
3291 	sk->sk_ll_usec		=	sysctl_net_busy_read;
3292 #endif
3293 
3294 	sk->sk_max_pacing_rate = ~0UL;
3295 	sk->sk_pacing_rate = ~0UL;
3296 	WRITE_ONCE(sk->sk_pacing_shift, 10);
3297 	sk->sk_incoming_cpu = -1;
3298 	sk->sk_txrehash = SOCK_TXREHASH_DEFAULT;
3299 
3300 	sk_rx_queue_clear(sk);
3301 	/*
3302 	 * Before updating sk_refcnt, we must commit prior changes to memory
3303 	 * (Documentation/RCU/rculist_nulls.rst for details)
3304 	 */
3305 	smp_wmb();
3306 	refcount_set(&sk->sk_refcnt, 1);
3307 	atomic_set(&sk->sk_drops, 0);
3308 }
3309 EXPORT_SYMBOL(sock_init_data);
3310 
3311 void lock_sock_nested(struct sock *sk, int subclass)
3312 {
3313 	/* The sk_lock has mutex_lock() semantics here. */
3314 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3315 
3316 	might_sleep();
3317 	spin_lock_bh(&sk->sk_lock.slock);
3318 	if (sock_owned_by_user_nocheck(sk))
3319 		__lock_sock(sk);
3320 	sk->sk_lock.owned = 1;
3321 	spin_unlock_bh(&sk->sk_lock.slock);
3322 }
3323 EXPORT_SYMBOL(lock_sock_nested);
3324 
3325 void release_sock(struct sock *sk)
3326 {
3327 	spin_lock_bh(&sk->sk_lock.slock);
3328 	if (sk->sk_backlog.tail)
3329 		__release_sock(sk);
3330 
3331 	/* Warning : release_cb() might need to release sk ownership,
3332 	 * ie call sock_release_ownership(sk) before us.
3333 	 */
3334 	if (sk->sk_prot->release_cb)
3335 		sk->sk_prot->release_cb(sk);
3336 
3337 	sock_release_ownership(sk);
3338 	if (waitqueue_active(&sk->sk_lock.wq))
3339 		wake_up(&sk->sk_lock.wq);
3340 	spin_unlock_bh(&sk->sk_lock.slock);
3341 }
3342 EXPORT_SYMBOL(release_sock);
3343 
3344 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3345 {
3346 	might_sleep();
3347 	spin_lock_bh(&sk->sk_lock.slock);
3348 
3349 	if (!sock_owned_by_user_nocheck(sk)) {
3350 		/*
3351 		 * Fast path return with bottom halves disabled and
3352 		 * sock::sk_lock.slock held.
3353 		 *
3354 		 * The 'mutex' is not contended and holding
3355 		 * sock::sk_lock.slock prevents all other lockers to
3356 		 * proceed so the corresponding unlock_sock_fast() can
3357 		 * avoid the slow path of release_sock() completely and
3358 		 * just release slock.
3359 		 *
3360 		 * From a semantical POV this is equivalent to 'acquiring'
3361 		 * the 'mutex', hence the corresponding lockdep
3362 		 * mutex_release() has to happen in the fast path of
3363 		 * unlock_sock_fast().
3364 		 */
3365 		return false;
3366 	}
3367 
3368 	__lock_sock(sk);
3369 	sk->sk_lock.owned = 1;
3370 	__acquire(&sk->sk_lock.slock);
3371 	spin_unlock_bh(&sk->sk_lock.slock);
3372 	return true;
3373 }
3374 EXPORT_SYMBOL(__lock_sock_fast);
3375 
3376 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3377 		   bool timeval, bool time32)
3378 {
3379 	struct sock *sk = sock->sk;
3380 	struct timespec64 ts;
3381 
3382 	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3383 	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3384 	if (ts.tv_sec == -1)
3385 		return -ENOENT;
3386 	if (ts.tv_sec == 0) {
3387 		ktime_t kt = ktime_get_real();
3388 		sock_write_timestamp(sk, kt);
3389 		ts = ktime_to_timespec64(kt);
3390 	}
3391 
3392 	if (timeval)
3393 		ts.tv_nsec /= 1000;
3394 
3395 #ifdef CONFIG_COMPAT_32BIT_TIME
3396 	if (time32)
3397 		return put_old_timespec32(&ts, userstamp);
3398 #endif
3399 #ifdef CONFIG_SPARC64
3400 	/* beware of padding in sparc64 timeval */
3401 	if (timeval && !in_compat_syscall()) {
3402 		struct __kernel_old_timeval __user tv = {
3403 			.tv_sec = ts.tv_sec,
3404 			.tv_usec = ts.tv_nsec,
3405 		};
3406 		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3407 			return -EFAULT;
3408 		return 0;
3409 	}
3410 #endif
3411 	return put_timespec64(&ts, userstamp);
3412 }
3413 EXPORT_SYMBOL(sock_gettstamp);
3414 
3415 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3416 {
3417 	if (!sock_flag(sk, flag)) {
3418 		unsigned long previous_flags = sk->sk_flags;
3419 
3420 		sock_set_flag(sk, flag);
3421 		/*
3422 		 * we just set one of the two flags which require net
3423 		 * time stamping, but time stamping might have been on
3424 		 * already because of the other one
3425 		 */
3426 		if (sock_needs_netstamp(sk) &&
3427 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3428 			net_enable_timestamp();
3429 	}
3430 }
3431 
3432 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3433 		       int level, int type)
3434 {
3435 	struct sock_exterr_skb *serr;
3436 	struct sk_buff *skb;
3437 	int copied, err;
3438 
3439 	err = -EAGAIN;
3440 	skb = sock_dequeue_err_skb(sk);
3441 	if (skb == NULL)
3442 		goto out;
3443 
3444 	copied = skb->len;
3445 	if (copied > len) {
3446 		msg->msg_flags |= MSG_TRUNC;
3447 		copied = len;
3448 	}
3449 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3450 	if (err)
3451 		goto out_free_skb;
3452 
3453 	sock_recv_timestamp(msg, sk, skb);
3454 
3455 	serr = SKB_EXT_ERR(skb);
3456 	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3457 
3458 	msg->msg_flags |= MSG_ERRQUEUE;
3459 	err = copied;
3460 
3461 out_free_skb:
3462 	kfree_skb(skb);
3463 out:
3464 	return err;
3465 }
3466 EXPORT_SYMBOL(sock_recv_errqueue);
3467 
3468 /*
3469  *	Get a socket option on an socket.
3470  *
3471  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3472  *	asynchronous errors should be reported by getsockopt. We assume
3473  *	this means if you specify SO_ERROR (otherwise whats the point of it).
3474  */
3475 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3476 			   char __user *optval, int __user *optlen)
3477 {
3478 	struct sock *sk = sock->sk;
3479 
3480 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3481 }
3482 EXPORT_SYMBOL(sock_common_getsockopt);
3483 
3484 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3485 			int flags)
3486 {
3487 	struct sock *sk = sock->sk;
3488 	int addr_len = 0;
3489 	int err;
3490 
3491 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3492 				   flags & ~MSG_DONTWAIT, &addr_len);
3493 	if (err >= 0)
3494 		msg->msg_namelen = addr_len;
3495 	return err;
3496 }
3497 EXPORT_SYMBOL(sock_common_recvmsg);
3498 
3499 /*
3500  *	Set socket options on an inet socket.
3501  */
3502 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3503 			   sockptr_t optval, unsigned int optlen)
3504 {
3505 	struct sock *sk = sock->sk;
3506 
3507 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3508 }
3509 EXPORT_SYMBOL(sock_common_setsockopt);
3510 
3511 void sk_common_release(struct sock *sk)
3512 {
3513 	if (sk->sk_prot->destroy)
3514 		sk->sk_prot->destroy(sk);
3515 
3516 	/*
3517 	 * Observation: when sk_common_release is called, processes have
3518 	 * no access to socket. But net still has.
3519 	 * Step one, detach it from networking:
3520 	 *
3521 	 * A. Remove from hash tables.
3522 	 */
3523 
3524 	sk->sk_prot->unhash(sk);
3525 
3526 	/*
3527 	 * In this point socket cannot receive new packets, but it is possible
3528 	 * that some packets are in flight because some CPU runs receiver and
3529 	 * did hash table lookup before we unhashed socket. They will achieve
3530 	 * receive queue and will be purged by socket destructor.
3531 	 *
3532 	 * Also we still have packets pending on receive queue and probably,
3533 	 * our own packets waiting in device queues. sock_destroy will drain
3534 	 * receive queue, but transmitted packets will delay socket destruction
3535 	 * until the last reference will be released.
3536 	 */
3537 
3538 	sock_orphan(sk);
3539 
3540 	xfrm_sk_free_policy(sk);
3541 
3542 	sk_refcnt_debug_release(sk);
3543 
3544 	sock_put(sk);
3545 }
3546 EXPORT_SYMBOL(sk_common_release);
3547 
3548 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3549 {
3550 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3551 
3552 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3553 	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3554 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3555 	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3556 	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3557 	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3558 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3559 	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3560 	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3561 }
3562 
3563 #ifdef CONFIG_PROC_FS
3564 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3565 
3566 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3567 {
3568 	int cpu, idx = prot->inuse_idx;
3569 	int res = 0;
3570 
3571 	for_each_possible_cpu(cpu)
3572 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3573 
3574 	return res >= 0 ? res : 0;
3575 }
3576 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3577 
3578 int sock_inuse_get(struct net *net)
3579 {
3580 	int cpu, res = 0;
3581 
3582 	for_each_possible_cpu(cpu)
3583 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3584 
3585 	return res;
3586 }
3587 
3588 EXPORT_SYMBOL_GPL(sock_inuse_get);
3589 
3590 static int __net_init sock_inuse_init_net(struct net *net)
3591 {
3592 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3593 	if (net->core.prot_inuse == NULL)
3594 		return -ENOMEM;
3595 	return 0;
3596 }
3597 
3598 static void __net_exit sock_inuse_exit_net(struct net *net)
3599 {
3600 	free_percpu(net->core.prot_inuse);
3601 }
3602 
3603 static struct pernet_operations net_inuse_ops = {
3604 	.init = sock_inuse_init_net,
3605 	.exit = sock_inuse_exit_net,
3606 };
3607 
3608 static __init int net_inuse_init(void)
3609 {
3610 	if (register_pernet_subsys(&net_inuse_ops))
3611 		panic("Cannot initialize net inuse counters");
3612 
3613 	return 0;
3614 }
3615 
3616 core_initcall(net_inuse_init);
3617 
3618 static int assign_proto_idx(struct proto *prot)
3619 {
3620 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3621 
3622 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3623 		pr_err("PROTO_INUSE_NR exhausted\n");
3624 		return -ENOSPC;
3625 	}
3626 
3627 	set_bit(prot->inuse_idx, proto_inuse_idx);
3628 	return 0;
3629 }
3630 
3631 static void release_proto_idx(struct proto *prot)
3632 {
3633 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3634 		clear_bit(prot->inuse_idx, proto_inuse_idx);
3635 }
3636 #else
3637 static inline int assign_proto_idx(struct proto *prot)
3638 {
3639 	return 0;
3640 }
3641 
3642 static inline void release_proto_idx(struct proto *prot)
3643 {
3644 }
3645 
3646 #endif
3647 
3648 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3649 {
3650 	if (!twsk_prot)
3651 		return;
3652 	kfree(twsk_prot->twsk_slab_name);
3653 	twsk_prot->twsk_slab_name = NULL;
3654 	kmem_cache_destroy(twsk_prot->twsk_slab);
3655 	twsk_prot->twsk_slab = NULL;
3656 }
3657 
3658 static int tw_prot_init(const struct proto *prot)
3659 {
3660 	struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3661 
3662 	if (!twsk_prot)
3663 		return 0;
3664 
3665 	twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3666 					      prot->name);
3667 	if (!twsk_prot->twsk_slab_name)
3668 		return -ENOMEM;
3669 
3670 	twsk_prot->twsk_slab =
3671 		kmem_cache_create(twsk_prot->twsk_slab_name,
3672 				  twsk_prot->twsk_obj_size, 0,
3673 				  SLAB_ACCOUNT | prot->slab_flags,
3674 				  NULL);
3675 	if (!twsk_prot->twsk_slab) {
3676 		pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3677 			prot->name);
3678 		return -ENOMEM;
3679 	}
3680 
3681 	return 0;
3682 }
3683 
3684 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3685 {
3686 	if (!rsk_prot)
3687 		return;
3688 	kfree(rsk_prot->slab_name);
3689 	rsk_prot->slab_name = NULL;
3690 	kmem_cache_destroy(rsk_prot->slab);
3691 	rsk_prot->slab = NULL;
3692 }
3693 
3694 static int req_prot_init(const struct proto *prot)
3695 {
3696 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3697 
3698 	if (!rsk_prot)
3699 		return 0;
3700 
3701 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3702 					prot->name);
3703 	if (!rsk_prot->slab_name)
3704 		return -ENOMEM;
3705 
3706 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3707 					   rsk_prot->obj_size, 0,
3708 					   SLAB_ACCOUNT | prot->slab_flags,
3709 					   NULL);
3710 
3711 	if (!rsk_prot->slab) {
3712 		pr_crit("%s: Can't create request sock SLAB cache!\n",
3713 			prot->name);
3714 		return -ENOMEM;
3715 	}
3716 	return 0;
3717 }
3718 
3719 int proto_register(struct proto *prot, int alloc_slab)
3720 {
3721 	int ret = -ENOBUFS;
3722 
3723 	if (prot->memory_allocated && !prot->sysctl_mem) {
3724 		pr_err("%s: missing sysctl_mem\n", prot->name);
3725 		return -EINVAL;
3726 	}
3727 	if (alloc_slab) {
3728 		prot->slab = kmem_cache_create_usercopy(prot->name,
3729 					prot->obj_size, 0,
3730 					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3731 					prot->slab_flags,
3732 					prot->useroffset, prot->usersize,
3733 					NULL);
3734 
3735 		if (prot->slab == NULL) {
3736 			pr_crit("%s: Can't create sock SLAB cache!\n",
3737 				prot->name);
3738 			goto out;
3739 		}
3740 
3741 		if (req_prot_init(prot))
3742 			goto out_free_request_sock_slab;
3743 
3744 		if (tw_prot_init(prot))
3745 			goto out_free_timewait_sock_slab;
3746 	}
3747 
3748 	mutex_lock(&proto_list_mutex);
3749 	ret = assign_proto_idx(prot);
3750 	if (ret) {
3751 		mutex_unlock(&proto_list_mutex);
3752 		goto out_free_timewait_sock_slab;
3753 	}
3754 	list_add(&prot->node, &proto_list);
3755 	mutex_unlock(&proto_list_mutex);
3756 	return ret;
3757 
3758 out_free_timewait_sock_slab:
3759 	if (alloc_slab)
3760 		tw_prot_cleanup(prot->twsk_prot);
3761 out_free_request_sock_slab:
3762 	if (alloc_slab) {
3763 		req_prot_cleanup(prot->rsk_prot);
3764 
3765 		kmem_cache_destroy(prot->slab);
3766 		prot->slab = NULL;
3767 	}
3768 out:
3769 	return ret;
3770 }
3771 EXPORT_SYMBOL(proto_register);
3772 
3773 void proto_unregister(struct proto *prot)
3774 {
3775 	mutex_lock(&proto_list_mutex);
3776 	release_proto_idx(prot);
3777 	list_del(&prot->node);
3778 	mutex_unlock(&proto_list_mutex);
3779 
3780 	kmem_cache_destroy(prot->slab);
3781 	prot->slab = NULL;
3782 
3783 	req_prot_cleanup(prot->rsk_prot);
3784 	tw_prot_cleanup(prot->twsk_prot);
3785 }
3786 EXPORT_SYMBOL(proto_unregister);
3787 
3788 int sock_load_diag_module(int family, int protocol)
3789 {
3790 	if (!protocol) {
3791 		if (!sock_is_registered(family))
3792 			return -ENOENT;
3793 
3794 		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3795 				      NETLINK_SOCK_DIAG, family);
3796 	}
3797 
3798 #ifdef CONFIG_INET
3799 	if (family == AF_INET &&
3800 	    protocol != IPPROTO_RAW &&
3801 	    protocol < MAX_INET_PROTOS &&
3802 	    !rcu_access_pointer(inet_protos[protocol]))
3803 		return -ENOENT;
3804 #endif
3805 
3806 	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3807 			      NETLINK_SOCK_DIAG, family, protocol);
3808 }
3809 EXPORT_SYMBOL(sock_load_diag_module);
3810 
3811 #ifdef CONFIG_PROC_FS
3812 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3813 	__acquires(proto_list_mutex)
3814 {
3815 	mutex_lock(&proto_list_mutex);
3816 	return seq_list_start_head(&proto_list, *pos);
3817 }
3818 
3819 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3820 {
3821 	return seq_list_next(v, &proto_list, pos);
3822 }
3823 
3824 static void proto_seq_stop(struct seq_file *seq, void *v)
3825 	__releases(proto_list_mutex)
3826 {
3827 	mutex_unlock(&proto_list_mutex);
3828 }
3829 
3830 static char proto_method_implemented(const void *method)
3831 {
3832 	return method == NULL ? 'n' : 'y';
3833 }
3834 static long sock_prot_memory_allocated(struct proto *proto)
3835 {
3836 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3837 }
3838 
3839 static const char *sock_prot_memory_pressure(struct proto *proto)
3840 {
3841 	return proto->memory_pressure != NULL ?
3842 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3843 }
3844 
3845 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3846 {
3847 
3848 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3849 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3850 		   proto->name,
3851 		   proto->obj_size,
3852 		   sock_prot_inuse_get(seq_file_net(seq), proto),
3853 		   sock_prot_memory_allocated(proto),
3854 		   sock_prot_memory_pressure(proto),
3855 		   proto->max_header,
3856 		   proto->slab == NULL ? "no" : "yes",
3857 		   module_name(proto->owner),
3858 		   proto_method_implemented(proto->close),
3859 		   proto_method_implemented(proto->connect),
3860 		   proto_method_implemented(proto->disconnect),
3861 		   proto_method_implemented(proto->accept),
3862 		   proto_method_implemented(proto->ioctl),
3863 		   proto_method_implemented(proto->init),
3864 		   proto_method_implemented(proto->destroy),
3865 		   proto_method_implemented(proto->shutdown),
3866 		   proto_method_implemented(proto->setsockopt),
3867 		   proto_method_implemented(proto->getsockopt),
3868 		   proto_method_implemented(proto->sendmsg),
3869 		   proto_method_implemented(proto->recvmsg),
3870 		   proto_method_implemented(proto->sendpage),
3871 		   proto_method_implemented(proto->bind),
3872 		   proto_method_implemented(proto->backlog_rcv),
3873 		   proto_method_implemented(proto->hash),
3874 		   proto_method_implemented(proto->unhash),
3875 		   proto_method_implemented(proto->get_port),
3876 		   proto_method_implemented(proto->enter_memory_pressure));
3877 }
3878 
3879 static int proto_seq_show(struct seq_file *seq, void *v)
3880 {
3881 	if (v == &proto_list)
3882 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3883 			   "protocol",
3884 			   "size",
3885 			   "sockets",
3886 			   "memory",
3887 			   "press",
3888 			   "maxhdr",
3889 			   "slab",
3890 			   "module",
3891 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3892 	else
3893 		proto_seq_printf(seq, list_entry(v, struct proto, node));
3894 	return 0;
3895 }
3896 
3897 static const struct seq_operations proto_seq_ops = {
3898 	.start  = proto_seq_start,
3899 	.next   = proto_seq_next,
3900 	.stop   = proto_seq_stop,
3901 	.show   = proto_seq_show,
3902 };
3903 
3904 static __net_init int proto_init_net(struct net *net)
3905 {
3906 	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3907 			sizeof(struct seq_net_private)))
3908 		return -ENOMEM;
3909 
3910 	return 0;
3911 }
3912 
3913 static __net_exit void proto_exit_net(struct net *net)
3914 {
3915 	remove_proc_entry("protocols", net->proc_net);
3916 }
3917 
3918 
3919 static __net_initdata struct pernet_operations proto_net_ops = {
3920 	.init = proto_init_net,
3921 	.exit = proto_exit_net,
3922 };
3923 
3924 static int __init proto_init(void)
3925 {
3926 	return register_pernet_subsys(&proto_net_ops);
3927 }
3928 
3929 subsys_initcall(proto_init);
3930 
3931 #endif /* PROC_FS */
3932 
3933 #ifdef CONFIG_NET_RX_BUSY_POLL
3934 bool sk_busy_loop_end(void *p, unsigned long start_time)
3935 {
3936 	struct sock *sk = p;
3937 
3938 	return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3939 	       sk_busy_loop_timeout(sk, start_time);
3940 }
3941 EXPORT_SYMBOL(sk_busy_loop_end);
3942 #endif /* CONFIG_NET_RX_BUSY_POLL */
3943 
3944 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3945 {
3946 	if (!sk->sk_prot->bind_add)
3947 		return -EOPNOTSUPP;
3948 	return sk->sk_prot->bind_add(sk, addr, addr_len);
3949 }
3950 EXPORT_SYMBOL(sock_bind_add);
3951