xref: /openbmc/linux/net/core/sock.c (revision ae213c44)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Generic socket support routines. Memory allocators, socket lock/release
7  *		handler for protocols to use and generic option handler.
8  *
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  *		This program is free software; you can redistribute it and/or
87  *		modify it under the terms of the GNU General Public License
88  *		as published by the Free Software Foundation; either version
89  *		2 of the License, or (at your option) any later version.
90  */
91 
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93 
94 #include <asm/unaligned.h>
95 #include <linux/capability.h>
96 #include <linux/errno.h>
97 #include <linux/errqueue.h>
98 #include <linux/types.h>
99 #include <linux/socket.h>
100 #include <linux/in.h>
101 #include <linux/kernel.h>
102 #include <linux/module.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <linux/sched.h>
106 #include <linux/sched/mm.h>
107 #include <linux/timer.h>
108 #include <linux/string.h>
109 #include <linux/sockios.h>
110 #include <linux/net.h>
111 #include <linux/mm.h>
112 #include <linux/slab.h>
113 #include <linux/interrupt.h>
114 #include <linux/poll.h>
115 #include <linux/tcp.h>
116 #include <linux/init.h>
117 #include <linux/highmem.h>
118 #include <linux/user_namespace.h>
119 #include <linux/static_key.h>
120 #include <linux/memcontrol.h>
121 #include <linux/prefetch.h>
122 
123 #include <linux/uaccess.h>
124 
125 #include <linux/netdevice.h>
126 #include <net/protocol.h>
127 #include <linux/skbuff.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <linux/net_tstamp.h>
132 #include <net/xfrm.h>
133 #include <linux/ipsec.h>
134 #include <net/cls_cgroup.h>
135 #include <net/netprio_cgroup.h>
136 #include <linux/sock_diag.h>
137 
138 #include <linux/filter.h>
139 #include <net/sock_reuseport.h>
140 #include <net/bpf_sk_storage.h>
141 
142 #include <trace/events/sock.h>
143 
144 #include <net/tcp.h>
145 #include <net/busy_poll.h>
146 
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
149 
150 static void sock_inuse_add(struct net *net, int val);
151 
152 /**
153  * sk_ns_capable - General socket capability test
154  * @sk: Socket to use a capability on or through
155  * @user_ns: The user namespace of the capability to use
156  * @cap: The capability to use
157  *
158  * Test to see if the opener of the socket had when the socket was
159  * created and the current process has the capability @cap in the user
160  * namespace @user_ns.
161  */
162 bool sk_ns_capable(const struct sock *sk,
163 		   struct user_namespace *user_ns, int cap)
164 {
165 	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
166 		ns_capable(user_ns, cap);
167 }
168 EXPORT_SYMBOL(sk_ns_capable);
169 
170 /**
171  * sk_capable - Socket global capability test
172  * @sk: Socket to use a capability on or through
173  * @cap: The global capability to use
174  *
175  * Test to see if the opener of the socket had when the socket was
176  * created and the current process has the capability @cap in all user
177  * namespaces.
178  */
179 bool sk_capable(const struct sock *sk, int cap)
180 {
181 	return sk_ns_capable(sk, &init_user_ns, cap);
182 }
183 EXPORT_SYMBOL(sk_capable);
184 
185 /**
186  * sk_net_capable - Network namespace socket capability test
187  * @sk: Socket to use a capability on or through
188  * @cap: The capability to use
189  *
190  * Test to see if the opener of the socket had when the socket was created
191  * and the current process has the capability @cap over the network namespace
192  * the socket is a member of.
193  */
194 bool sk_net_capable(const struct sock *sk, int cap)
195 {
196 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
197 }
198 EXPORT_SYMBOL(sk_net_capable);
199 
200 /*
201  * Each address family might have different locking rules, so we have
202  * one slock key per address family and separate keys for internal and
203  * userspace sockets.
204  */
205 static struct lock_class_key af_family_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_keys[AF_MAX];
207 static struct lock_class_key af_family_slock_keys[AF_MAX];
208 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209 
210 /*
211  * Make lock validator output more readable. (we pre-construct these
212  * strings build-time, so that runtime initialization of socket
213  * locks is fast):
214  */
215 
216 #define _sock_locks(x)						  \
217   x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
218   x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
219   x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
220   x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
221   x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
222   x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
223   x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
224   x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
225   x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
226   x "27"       ,	x "28"          ,	x "AF_CAN"      , \
227   x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
228   x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
229   x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
230   x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
231   x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
232   x "AF_MAX"
233 
234 static const char *const af_family_key_strings[AF_MAX+1] = {
235 	_sock_locks("sk_lock-")
236 };
237 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
238 	_sock_locks("slock-")
239 };
240 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
241 	_sock_locks("clock-")
242 };
243 
244 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
245 	_sock_locks("k-sk_lock-")
246 };
247 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
248 	_sock_locks("k-slock-")
249 };
250 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
251 	_sock_locks("k-clock-")
252 };
253 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
254 	_sock_locks("rlock-")
255 };
256 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
257 	_sock_locks("wlock-")
258 };
259 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
260 	_sock_locks("elock-")
261 };
262 
263 /*
264  * sk_callback_lock and sk queues locking rules are per-address-family,
265  * so split the lock classes by using a per-AF key:
266  */
267 static struct lock_class_key af_callback_keys[AF_MAX];
268 static struct lock_class_key af_rlock_keys[AF_MAX];
269 static struct lock_class_key af_wlock_keys[AF_MAX];
270 static struct lock_class_key af_elock_keys[AF_MAX];
271 static struct lock_class_key af_kern_callback_keys[AF_MAX];
272 
273 /* Run time adjustable parameters. */
274 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
275 EXPORT_SYMBOL(sysctl_wmem_max);
276 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
277 EXPORT_SYMBOL(sysctl_rmem_max);
278 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
279 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
280 
281 /* Maximal space eaten by iovec or ancillary data plus some space */
282 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
283 EXPORT_SYMBOL(sysctl_optmem_max);
284 
285 int sysctl_tstamp_allow_data __read_mostly = 1;
286 
287 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
288 EXPORT_SYMBOL_GPL(memalloc_socks_key);
289 
290 /**
291  * sk_set_memalloc - sets %SOCK_MEMALLOC
292  * @sk: socket to set it on
293  *
294  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
295  * It's the responsibility of the admin to adjust min_free_kbytes
296  * to meet the requirements
297  */
298 void sk_set_memalloc(struct sock *sk)
299 {
300 	sock_set_flag(sk, SOCK_MEMALLOC);
301 	sk->sk_allocation |= __GFP_MEMALLOC;
302 	static_branch_inc(&memalloc_socks_key);
303 }
304 EXPORT_SYMBOL_GPL(sk_set_memalloc);
305 
306 void sk_clear_memalloc(struct sock *sk)
307 {
308 	sock_reset_flag(sk, SOCK_MEMALLOC);
309 	sk->sk_allocation &= ~__GFP_MEMALLOC;
310 	static_branch_dec(&memalloc_socks_key);
311 
312 	/*
313 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
314 	 * progress of swapping. SOCK_MEMALLOC may be cleared while
315 	 * it has rmem allocations due to the last swapfile being deactivated
316 	 * but there is a risk that the socket is unusable due to exceeding
317 	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
318 	 */
319 	sk_mem_reclaim(sk);
320 }
321 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
322 
323 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
324 {
325 	int ret;
326 	unsigned int noreclaim_flag;
327 
328 	/* these should have been dropped before queueing */
329 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
330 
331 	noreclaim_flag = memalloc_noreclaim_save();
332 	ret = sk->sk_backlog_rcv(sk, skb);
333 	memalloc_noreclaim_restore(noreclaim_flag);
334 
335 	return ret;
336 }
337 EXPORT_SYMBOL(__sk_backlog_rcv);
338 
339 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
340 {
341 	struct __kernel_sock_timeval tv;
342 	int size;
343 
344 	if (timeo == MAX_SCHEDULE_TIMEOUT) {
345 		tv.tv_sec = 0;
346 		tv.tv_usec = 0;
347 	} else {
348 		tv.tv_sec = timeo / HZ;
349 		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
350 	}
351 
352 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
353 		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
354 		*(struct old_timeval32 *)optval = tv32;
355 		return sizeof(tv32);
356 	}
357 
358 	if (old_timeval) {
359 		struct __kernel_old_timeval old_tv;
360 		old_tv.tv_sec = tv.tv_sec;
361 		old_tv.tv_usec = tv.tv_usec;
362 		*(struct __kernel_old_timeval *)optval = old_tv;
363 		size = sizeof(old_tv);
364 	} else {
365 		*(struct __kernel_sock_timeval *)optval = tv;
366 		size = sizeof(tv);
367 	}
368 
369 	return size;
370 }
371 
372 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
373 {
374 	struct __kernel_sock_timeval tv;
375 
376 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
377 		struct old_timeval32 tv32;
378 
379 		if (optlen < sizeof(tv32))
380 			return -EINVAL;
381 
382 		if (copy_from_user(&tv32, optval, sizeof(tv32)))
383 			return -EFAULT;
384 		tv.tv_sec = tv32.tv_sec;
385 		tv.tv_usec = tv32.tv_usec;
386 	} else if (old_timeval) {
387 		struct __kernel_old_timeval old_tv;
388 
389 		if (optlen < sizeof(old_tv))
390 			return -EINVAL;
391 		if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
392 			return -EFAULT;
393 		tv.tv_sec = old_tv.tv_sec;
394 		tv.tv_usec = old_tv.tv_usec;
395 	} else {
396 		if (optlen < sizeof(tv))
397 			return -EINVAL;
398 		if (copy_from_user(&tv, optval, sizeof(tv)))
399 			return -EFAULT;
400 	}
401 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
402 		return -EDOM;
403 
404 	if (tv.tv_sec < 0) {
405 		static int warned __read_mostly;
406 
407 		*timeo_p = 0;
408 		if (warned < 10 && net_ratelimit()) {
409 			warned++;
410 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
411 				__func__, current->comm, task_pid_nr(current));
412 		}
413 		return 0;
414 	}
415 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
416 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
417 		return 0;
418 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
419 		*timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
420 	return 0;
421 }
422 
423 static void sock_warn_obsolete_bsdism(const char *name)
424 {
425 	static int warned;
426 	static char warncomm[TASK_COMM_LEN];
427 	if (strcmp(warncomm, current->comm) && warned < 5) {
428 		strcpy(warncomm,  current->comm);
429 		pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
430 			warncomm, name);
431 		warned++;
432 	}
433 }
434 
435 static bool sock_needs_netstamp(const struct sock *sk)
436 {
437 	switch (sk->sk_family) {
438 	case AF_UNSPEC:
439 	case AF_UNIX:
440 		return false;
441 	default:
442 		return true;
443 	}
444 }
445 
446 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
447 {
448 	if (sk->sk_flags & flags) {
449 		sk->sk_flags &= ~flags;
450 		if (sock_needs_netstamp(sk) &&
451 		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
452 			net_disable_timestamp();
453 	}
454 }
455 
456 
457 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
458 {
459 	unsigned long flags;
460 	struct sk_buff_head *list = &sk->sk_receive_queue;
461 
462 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
463 		atomic_inc(&sk->sk_drops);
464 		trace_sock_rcvqueue_full(sk, skb);
465 		return -ENOMEM;
466 	}
467 
468 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
469 		atomic_inc(&sk->sk_drops);
470 		return -ENOBUFS;
471 	}
472 
473 	skb->dev = NULL;
474 	skb_set_owner_r(skb, sk);
475 
476 	/* we escape from rcu protected region, make sure we dont leak
477 	 * a norefcounted dst
478 	 */
479 	skb_dst_force(skb);
480 
481 	spin_lock_irqsave(&list->lock, flags);
482 	sock_skb_set_dropcount(sk, skb);
483 	__skb_queue_tail(list, skb);
484 	spin_unlock_irqrestore(&list->lock, flags);
485 
486 	if (!sock_flag(sk, SOCK_DEAD))
487 		sk->sk_data_ready(sk);
488 	return 0;
489 }
490 EXPORT_SYMBOL(__sock_queue_rcv_skb);
491 
492 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
493 {
494 	int err;
495 
496 	err = sk_filter(sk, skb);
497 	if (err)
498 		return err;
499 
500 	return __sock_queue_rcv_skb(sk, skb);
501 }
502 EXPORT_SYMBOL(sock_queue_rcv_skb);
503 
504 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
505 		     const int nested, unsigned int trim_cap, bool refcounted)
506 {
507 	int rc = NET_RX_SUCCESS;
508 
509 	if (sk_filter_trim_cap(sk, skb, trim_cap))
510 		goto discard_and_relse;
511 
512 	skb->dev = NULL;
513 
514 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
515 		atomic_inc(&sk->sk_drops);
516 		goto discard_and_relse;
517 	}
518 	if (nested)
519 		bh_lock_sock_nested(sk);
520 	else
521 		bh_lock_sock(sk);
522 	if (!sock_owned_by_user(sk)) {
523 		/*
524 		 * trylock + unlock semantics:
525 		 */
526 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
527 
528 		rc = sk_backlog_rcv(sk, skb);
529 
530 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
531 	} else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
532 		bh_unlock_sock(sk);
533 		atomic_inc(&sk->sk_drops);
534 		goto discard_and_relse;
535 	}
536 
537 	bh_unlock_sock(sk);
538 out:
539 	if (refcounted)
540 		sock_put(sk);
541 	return rc;
542 discard_and_relse:
543 	kfree_skb(skb);
544 	goto out;
545 }
546 EXPORT_SYMBOL(__sk_receive_skb);
547 
548 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
549 {
550 	struct dst_entry *dst = __sk_dst_get(sk);
551 
552 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
553 		sk_tx_queue_clear(sk);
554 		sk->sk_dst_pending_confirm = 0;
555 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
556 		dst_release(dst);
557 		return NULL;
558 	}
559 
560 	return dst;
561 }
562 EXPORT_SYMBOL(__sk_dst_check);
563 
564 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
565 {
566 	struct dst_entry *dst = sk_dst_get(sk);
567 
568 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
569 		sk_dst_reset(sk);
570 		dst_release(dst);
571 		return NULL;
572 	}
573 
574 	return dst;
575 }
576 EXPORT_SYMBOL(sk_dst_check);
577 
578 static int sock_setbindtodevice_locked(struct sock *sk, int ifindex)
579 {
580 	int ret = -ENOPROTOOPT;
581 #ifdef CONFIG_NETDEVICES
582 	struct net *net = sock_net(sk);
583 
584 	/* Sorry... */
585 	ret = -EPERM;
586 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
587 		goto out;
588 
589 	ret = -EINVAL;
590 	if (ifindex < 0)
591 		goto out;
592 
593 	sk->sk_bound_dev_if = ifindex;
594 	if (sk->sk_prot->rehash)
595 		sk->sk_prot->rehash(sk);
596 	sk_dst_reset(sk);
597 
598 	ret = 0;
599 
600 out:
601 #endif
602 
603 	return ret;
604 }
605 
606 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
607 				int optlen)
608 {
609 	int ret = -ENOPROTOOPT;
610 #ifdef CONFIG_NETDEVICES
611 	struct net *net = sock_net(sk);
612 	char devname[IFNAMSIZ];
613 	int index;
614 
615 	ret = -EINVAL;
616 	if (optlen < 0)
617 		goto out;
618 
619 	/* Bind this socket to a particular device like "eth0",
620 	 * as specified in the passed interface name. If the
621 	 * name is "" or the option length is zero the socket
622 	 * is not bound.
623 	 */
624 	if (optlen > IFNAMSIZ - 1)
625 		optlen = IFNAMSIZ - 1;
626 	memset(devname, 0, sizeof(devname));
627 
628 	ret = -EFAULT;
629 	if (copy_from_user(devname, optval, optlen))
630 		goto out;
631 
632 	index = 0;
633 	if (devname[0] != '\0') {
634 		struct net_device *dev;
635 
636 		rcu_read_lock();
637 		dev = dev_get_by_name_rcu(net, devname);
638 		if (dev)
639 			index = dev->ifindex;
640 		rcu_read_unlock();
641 		ret = -ENODEV;
642 		if (!dev)
643 			goto out;
644 	}
645 
646 	lock_sock(sk);
647 	ret = sock_setbindtodevice_locked(sk, index);
648 	release_sock(sk);
649 
650 out:
651 #endif
652 
653 	return ret;
654 }
655 
656 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
657 				int __user *optlen, int len)
658 {
659 	int ret = -ENOPROTOOPT;
660 #ifdef CONFIG_NETDEVICES
661 	struct net *net = sock_net(sk);
662 	char devname[IFNAMSIZ];
663 
664 	if (sk->sk_bound_dev_if == 0) {
665 		len = 0;
666 		goto zero;
667 	}
668 
669 	ret = -EINVAL;
670 	if (len < IFNAMSIZ)
671 		goto out;
672 
673 	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
674 	if (ret)
675 		goto out;
676 
677 	len = strlen(devname) + 1;
678 
679 	ret = -EFAULT;
680 	if (copy_to_user(optval, devname, len))
681 		goto out;
682 
683 zero:
684 	ret = -EFAULT;
685 	if (put_user(len, optlen))
686 		goto out;
687 
688 	ret = 0;
689 
690 out:
691 #endif
692 
693 	return ret;
694 }
695 
696 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
697 {
698 	if (valbool)
699 		sock_set_flag(sk, bit);
700 	else
701 		sock_reset_flag(sk, bit);
702 }
703 
704 bool sk_mc_loop(struct sock *sk)
705 {
706 	if (dev_recursion_level())
707 		return false;
708 	if (!sk)
709 		return true;
710 	switch (sk->sk_family) {
711 	case AF_INET:
712 		return inet_sk(sk)->mc_loop;
713 #if IS_ENABLED(CONFIG_IPV6)
714 	case AF_INET6:
715 		return inet6_sk(sk)->mc_loop;
716 #endif
717 	}
718 	WARN_ON(1);
719 	return true;
720 }
721 EXPORT_SYMBOL(sk_mc_loop);
722 
723 /*
724  *	This is meant for all protocols to use and covers goings on
725  *	at the socket level. Everything here is generic.
726  */
727 
728 int sock_setsockopt(struct socket *sock, int level, int optname,
729 		    char __user *optval, unsigned int optlen)
730 {
731 	struct sock_txtime sk_txtime;
732 	struct sock *sk = sock->sk;
733 	int val;
734 	int valbool;
735 	struct linger ling;
736 	int ret = 0;
737 
738 	/*
739 	 *	Options without arguments
740 	 */
741 
742 	if (optname == SO_BINDTODEVICE)
743 		return sock_setbindtodevice(sk, optval, optlen);
744 
745 	if (optlen < sizeof(int))
746 		return -EINVAL;
747 
748 	if (get_user(val, (int __user *)optval))
749 		return -EFAULT;
750 
751 	valbool = val ? 1 : 0;
752 
753 	lock_sock(sk);
754 
755 	switch (optname) {
756 	case SO_DEBUG:
757 		if (val && !capable(CAP_NET_ADMIN))
758 			ret = -EACCES;
759 		else
760 			sock_valbool_flag(sk, SOCK_DBG, valbool);
761 		break;
762 	case SO_REUSEADDR:
763 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
764 		break;
765 	case SO_REUSEPORT:
766 		sk->sk_reuseport = valbool;
767 		break;
768 	case SO_TYPE:
769 	case SO_PROTOCOL:
770 	case SO_DOMAIN:
771 	case SO_ERROR:
772 		ret = -ENOPROTOOPT;
773 		break;
774 	case SO_DONTROUTE:
775 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
776 		sk_dst_reset(sk);
777 		break;
778 	case SO_BROADCAST:
779 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
780 		break;
781 	case SO_SNDBUF:
782 		/* Don't error on this BSD doesn't and if you think
783 		 * about it this is right. Otherwise apps have to
784 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
785 		 * are treated in BSD as hints
786 		 */
787 		val = min_t(u32, val, sysctl_wmem_max);
788 set_sndbuf:
789 		/* Ensure val * 2 fits into an int, to prevent max_t()
790 		 * from treating it as a negative value.
791 		 */
792 		val = min_t(int, val, INT_MAX / 2);
793 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
794 		sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
795 		/* Wake up sending tasks if we upped the value. */
796 		sk->sk_write_space(sk);
797 		break;
798 
799 	case SO_SNDBUFFORCE:
800 		if (!capable(CAP_NET_ADMIN)) {
801 			ret = -EPERM;
802 			break;
803 		}
804 
805 		/* No negative values (to prevent underflow, as val will be
806 		 * multiplied by 2).
807 		 */
808 		if (val < 0)
809 			val = 0;
810 		goto set_sndbuf;
811 
812 	case SO_RCVBUF:
813 		/* Don't error on this BSD doesn't and if you think
814 		 * about it this is right. Otherwise apps have to
815 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
816 		 * are treated in BSD as hints
817 		 */
818 		val = min_t(u32, val, sysctl_rmem_max);
819 set_rcvbuf:
820 		/* Ensure val * 2 fits into an int, to prevent max_t()
821 		 * from treating it as a negative value.
822 		 */
823 		val = min_t(int, val, INT_MAX / 2);
824 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
825 		/*
826 		 * We double it on the way in to account for
827 		 * "struct sk_buff" etc. overhead.   Applications
828 		 * assume that the SO_RCVBUF setting they make will
829 		 * allow that much actual data to be received on that
830 		 * socket.
831 		 *
832 		 * Applications are unaware that "struct sk_buff" and
833 		 * other overheads allocate from the receive buffer
834 		 * during socket buffer allocation.
835 		 *
836 		 * And after considering the possible alternatives,
837 		 * returning the value we actually used in getsockopt
838 		 * is the most desirable behavior.
839 		 */
840 		sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
841 		break;
842 
843 	case SO_RCVBUFFORCE:
844 		if (!capable(CAP_NET_ADMIN)) {
845 			ret = -EPERM;
846 			break;
847 		}
848 
849 		/* No negative values (to prevent underflow, as val will be
850 		 * multiplied by 2).
851 		 */
852 		if (val < 0)
853 			val = 0;
854 		goto set_rcvbuf;
855 
856 	case SO_KEEPALIVE:
857 		if (sk->sk_prot->keepalive)
858 			sk->sk_prot->keepalive(sk, valbool);
859 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
860 		break;
861 
862 	case SO_OOBINLINE:
863 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
864 		break;
865 
866 	case SO_NO_CHECK:
867 		sk->sk_no_check_tx = valbool;
868 		break;
869 
870 	case SO_PRIORITY:
871 		if ((val >= 0 && val <= 6) ||
872 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
873 			sk->sk_priority = val;
874 		else
875 			ret = -EPERM;
876 		break;
877 
878 	case SO_LINGER:
879 		if (optlen < sizeof(ling)) {
880 			ret = -EINVAL;	/* 1003.1g */
881 			break;
882 		}
883 		if (copy_from_user(&ling, optval, sizeof(ling))) {
884 			ret = -EFAULT;
885 			break;
886 		}
887 		if (!ling.l_onoff)
888 			sock_reset_flag(sk, SOCK_LINGER);
889 		else {
890 #if (BITS_PER_LONG == 32)
891 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
892 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
893 			else
894 #endif
895 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
896 			sock_set_flag(sk, SOCK_LINGER);
897 		}
898 		break;
899 
900 	case SO_BSDCOMPAT:
901 		sock_warn_obsolete_bsdism("setsockopt");
902 		break;
903 
904 	case SO_PASSCRED:
905 		if (valbool)
906 			set_bit(SOCK_PASSCRED, &sock->flags);
907 		else
908 			clear_bit(SOCK_PASSCRED, &sock->flags);
909 		break;
910 
911 	case SO_TIMESTAMP_OLD:
912 	case SO_TIMESTAMP_NEW:
913 	case SO_TIMESTAMPNS_OLD:
914 	case SO_TIMESTAMPNS_NEW:
915 		if (valbool)  {
916 			if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
917 				sock_set_flag(sk, SOCK_TSTAMP_NEW);
918 			else
919 				sock_reset_flag(sk, SOCK_TSTAMP_NEW);
920 
921 			if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
922 				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
923 			else
924 				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
925 			sock_set_flag(sk, SOCK_RCVTSTAMP);
926 			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
927 		} else {
928 			sock_reset_flag(sk, SOCK_RCVTSTAMP);
929 			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
930 			sock_reset_flag(sk, SOCK_TSTAMP_NEW);
931 		}
932 		break;
933 
934 	case SO_TIMESTAMPING_NEW:
935 		sock_set_flag(sk, SOCK_TSTAMP_NEW);
936 		/* fall through */
937 	case SO_TIMESTAMPING_OLD:
938 		if (val & ~SOF_TIMESTAMPING_MASK) {
939 			ret = -EINVAL;
940 			break;
941 		}
942 
943 		if (val & SOF_TIMESTAMPING_OPT_ID &&
944 		    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
945 			if (sk->sk_protocol == IPPROTO_TCP &&
946 			    sk->sk_type == SOCK_STREAM) {
947 				if ((1 << sk->sk_state) &
948 				    (TCPF_CLOSE | TCPF_LISTEN)) {
949 					ret = -EINVAL;
950 					break;
951 				}
952 				sk->sk_tskey = tcp_sk(sk)->snd_una;
953 			} else {
954 				sk->sk_tskey = 0;
955 			}
956 		}
957 
958 		if (val & SOF_TIMESTAMPING_OPT_STATS &&
959 		    !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
960 			ret = -EINVAL;
961 			break;
962 		}
963 
964 		sk->sk_tsflags = val;
965 		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
966 			sock_enable_timestamp(sk,
967 					      SOCK_TIMESTAMPING_RX_SOFTWARE);
968 		else {
969 			if (optname == SO_TIMESTAMPING_NEW)
970 				sock_reset_flag(sk, SOCK_TSTAMP_NEW);
971 
972 			sock_disable_timestamp(sk,
973 					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
974 		}
975 		break;
976 
977 	case SO_RCVLOWAT:
978 		if (val < 0)
979 			val = INT_MAX;
980 		if (sock->ops->set_rcvlowat)
981 			ret = sock->ops->set_rcvlowat(sk, val);
982 		else
983 			sk->sk_rcvlowat = val ? : 1;
984 		break;
985 
986 	case SO_RCVTIMEO_OLD:
987 	case SO_RCVTIMEO_NEW:
988 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
989 		break;
990 
991 	case SO_SNDTIMEO_OLD:
992 	case SO_SNDTIMEO_NEW:
993 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
994 		break;
995 
996 	case SO_ATTACH_FILTER:
997 		ret = -EINVAL;
998 		if (optlen == sizeof(struct sock_fprog)) {
999 			struct sock_fprog fprog;
1000 
1001 			ret = -EFAULT;
1002 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
1003 				break;
1004 
1005 			ret = sk_attach_filter(&fprog, sk);
1006 		}
1007 		break;
1008 
1009 	case SO_ATTACH_BPF:
1010 		ret = -EINVAL;
1011 		if (optlen == sizeof(u32)) {
1012 			u32 ufd;
1013 
1014 			ret = -EFAULT;
1015 			if (copy_from_user(&ufd, optval, sizeof(ufd)))
1016 				break;
1017 
1018 			ret = sk_attach_bpf(ufd, sk);
1019 		}
1020 		break;
1021 
1022 	case SO_ATTACH_REUSEPORT_CBPF:
1023 		ret = -EINVAL;
1024 		if (optlen == sizeof(struct sock_fprog)) {
1025 			struct sock_fprog fprog;
1026 
1027 			ret = -EFAULT;
1028 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
1029 				break;
1030 
1031 			ret = sk_reuseport_attach_filter(&fprog, sk);
1032 		}
1033 		break;
1034 
1035 	case SO_ATTACH_REUSEPORT_EBPF:
1036 		ret = -EINVAL;
1037 		if (optlen == sizeof(u32)) {
1038 			u32 ufd;
1039 
1040 			ret = -EFAULT;
1041 			if (copy_from_user(&ufd, optval, sizeof(ufd)))
1042 				break;
1043 
1044 			ret = sk_reuseport_attach_bpf(ufd, sk);
1045 		}
1046 		break;
1047 
1048 	case SO_DETACH_FILTER:
1049 		ret = sk_detach_filter(sk);
1050 		break;
1051 
1052 	case SO_LOCK_FILTER:
1053 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1054 			ret = -EPERM;
1055 		else
1056 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1057 		break;
1058 
1059 	case SO_PASSSEC:
1060 		if (valbool)
1061 			set_bit(SOCK_PASSSEC, &sock->flags);
1062 		else
1063 			clear_bit(SOCK_PASSSEC, &sock->flags);
1064 		break;
1065 	case SO_MARK:
1066 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1067 			ret = -EPERM;
1068 		} else if (val != sk->sk_mark) {
1069 			sk->sk_mark = val;
1070 			sk_dst_reset(sk);
1071 		}
1072 		break;
1073 
1074 	case SO_RXQ_OVFL:
1075 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1076 		break;
1077 
1078 	case SO_WIFI_STATUS:
1079 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1080 		break;
1081 
1082 	case SO_PEEK_OFF:
1083 		if (sock->ops->set_peek_off)
1084 			ret = sock->ops->set_peek_off(sk, val);
1085 		else
1086 			ret = -EOPNOTSUPP;
1087 		break;
1088 
1089 	case SO_NOFCS:
1090 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1091 		break;
1092 
1093 	case SO_SELECT_ERR_QUEUE:
1094 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1095 		break;
1096 
1097 #ifdef CONFIG_NET_RX_BUSY_POLL
1098 	case SO_BUSY_POLL:
1099 		/* allow unprivileged users to decrease the value */
1100 		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1101 			ret = -EPERM;
1102 		else {
1103 			if (val < 0)
1104 				ret = -EINVAL;
1105 			else
1106 				sk->sk_ll_usec = val;
1107 		}
1108 		break;
1109 #endif
1110 
1111 	case SO_MAX_PACING_RATE:
1112 		{
1113 		unsigned long ulval = (val == ~0U) ? ~0UL : val;
1114 
1115 		if (sizeof(ulval) != sizeof(val) &&
1116 		    optlen >= sizeof(ulval) &&
1117 		    get_user(ulval, (unsigned long __user *)optval)) {
1118 			ret = -EFAULT;
1119 			break;
1120 		}
1121 		if (ulval != ~0UL)
1122 			cmpxchg(&sk->sk_pacing_status,
1123 				SK_PACING_NONE,
1124 				SK_PACING_NEEDED);
1125 		sk->sk_max_pacing_rate = ulval;
1126 		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1127 		break;
1128 		}
1129 	case SO_INCOMING_CPU:
1130 		sk->sk_incoming_cpu = val;
1131 		break;
1132 
1133 	case SO_CNX_ADVICE:
1134 		if (val == 1)
1135 			dst_negative_advice(sk);
1136 		break;
1137 
1138 	case SO_ZEROCOPY:
1139 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1140 			if (!((sk->sk_type == SOCK_STREAM &&
1141 			       sk->sk_protocol == IPPROTO_TCP) ||
1142 			      (sk->sk_type == SOCK_DGRAM &&
1143 			       sk->sk_protocol == IPPROTO_UDP)))
1144 				ret = -ENOTSUPP;
1145 		} else if (sk->sk_family != PF_RDS) {
1146 			ret = -ENOTSUPP;
1147 		}
1148 		if (!ret) {
1149 			if (val < 0 || val > 1)
1150 				ret = -EINVAL;
1151 			else
1152 				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1153 		}
1154 		break;
1155 
1156 	case SO_TXTIME:
1157 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1158 			ret = -EPERM;
1159 		} else if (optlen != sizeof(struct sock_txtime)) {
1160 			ret = -EINVAL;
1161 		} else if (copy_from_user(&sk_txtime, optval,
1162 			   sizeof(struct sock_txtime))) {
1163 			ret = -EFAULT;
1164 		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1165 			ret = -EINVAL;
1166 		} else {
1167 			sock_valbool_flag(sk, SOCK_TXTIME, true);
1168 			sk->sk_clockid = sk_txtime.clockid;
1169 			sk->sk_txtime_deadline_mode =
1170 				!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1171 			sk->sk_txtime_report_errors =
1172 				!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1173 		}
1174 		break;
1175 
1176 	case SO_BINDTOIFINDEX:
1177 		ret = sock_setbindtodevice_locked(sk, val);
1178 		break;
1179 
1180 	default:
1181 		ret = -ENOPROTOOPT;
1182 		break;
1183 	}
1184 	release_sock(sk);
1185 	return ret;
1186 }
1187 EXPORT_SYMBOL(sock_setsockopt);
1188 
1189 
1190 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1191 			  struct ucred *ucred)
1192 {
1193 	ucred->pid = pid_vnr(pid);
1194 	ucred->uid = ucred->gid = -1;
1195 	if (cred) {
1196 		struct user_namespace *current_ns = current_user_ns();
1197 
1198 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1199 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1200 	}
1201 }
1202 
1203 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1204 {
1205 	struct user_namespace *user_ns = current_user_ns();
1206 	int i;
1207 
1208 	for (i = 0; i < src->ngroups; i++)
1209 		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1210 			return -EFAULT;
1211 
1212 	return 0;
1213 }
1214 
1215 int sock_getsockopt(struct socket *sock, int level, int optname,
1216 		    char __user *optval, int __user *optlen)
1217 {
1218 	struct sock *sk = sock->sk;
1219 
1220 	union {
1221 		int val;
1222 		u64 val64;
1223 		unsigned long ulval;
1224 		struct linger ling;
1225 		struct old_timeval32 tm32;
1226 		struct __kernel_old_timeval tm;
1227 		struct  __kernel_sock_timeval stm;
1228 		struct sock_txtime txtime;
1229 	} v;
1230 
1231 	int lv = sizeof(int);
1232 	int len;
1233 
1234 	if (get_user(len, optlen))
1235 		return -EFAULT;
1236 	if (len < 0)
1237 		return -EINVAL;
1238 
1239 	memset(&v, 0, sizeof(v));
1240 
1241 	switch (optname) {
1242 	case SO_DEBUG:
1243 		v.val = sock_flag(sk, SOCK_DBG);
1244 		break;
1245 
1246 	case SO_DONTROUTE:
1247 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1248 		break;
1249 
1250 	case SO_BROADCAST:
1251 		v.val = sock_flag(sk, SOCK_BROADCAST);
1252 		break;
1253 
1254 	case SO_SNDBUF:
1255 		v.val = sk->sk_sndbuf;
1256 		break;
1257 
1258 	case SO_RCVBUF:
1259 		v.val = sk->sk_rcvbuf;
1260 		break;
1261 
1262 	case SO_REUSEADDR:
1263 		v.val = sk->sk_reuse;
1264 		break;
1265 
1266 	case SO_REUSEPORT:
1267 		v.val = sk->sk_reuseport;
1268 		break;
1269 
1270 	case SO_KEEPALIVE:
1271 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1272 		break;
1273 
1274 	case SO_TYPE:
1275 		v.val = sk->sk_type;
1276 		break;
1277 
1278 	case SO_PROTOCOL:
1279 		v.val = sk->sk_protocol;
1280 		break;
1281 
1282 	case SO_DOMAIN:
1283 		v.val = sk->sk_family;
1284 		break;
1285 
1286 	case SO_ERROR:
1287 		v.val = -sock_error(sk);
1288 		if (v.val == 0)
1289 			v.val = xchg(&sk->sk_err_soft, 0);
1290 		break;
1291 
1292 	case SO_OOBINLINE:
1293 		v.val = sock_flag(sk, SOCK_URGINLINE);
1294 		break;
1295 
1296 	case SO_NO_CHECK:
1297 		v.val = sk->sk_no_check_tx;
1298 		break;
1299 
1300 	case SO_PRIORITY:
1301 		v.val = sk->sk_priority;
1302 		break;
1303 
1304 	case SO_LINGER:
1305 		lv		= sizeof(v.ling);
1306 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1307 		v.ling.l_linger	= sk->sk_lingertime / HZ;
1308 		break;
1309 
1310 	case SO_BSDCOMPAT:
1311 		sock_warn_obsolete_bsdism("getsockopt");
1312 		break;
1313 
1314 	case SO_TIMESTAMP_OLD:
1315 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1316 				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1317 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1318 		break;
1319 
1320 	case SO_TIMESTAMPNS_OLD:
1321 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1322 		break;
1323 
1324 	case SO_TIMESTAMP_NEW:
1325 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1326 		break;
1327 
1328 	case SO_TIMESTAMPNS_NEW:
1329 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1330 		break;
1331 
1332 	case SO_TIMESTAMPING_OLD:
1333 		v.val = sk->sk_tsflags;
1334 		break;
1335 
1336 	case SO_RCVTIMEO_OLD:
1337 	case SO_RCVTIMEO_NEW:
1338 		lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1339 		break;
1340 
1341 	case SO_SNDTIMEO_OLD:
1342 	case SO_SNDTIMEO_NEW:
1343 		lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1344 		break;
1345 
1346 	case SO_RCVLOWAT:
1347 		v.val = sk->sk_rcvlowat;
1348 		break;
1349 
1350 	case SO_SNDLOWAT:
1351 		v.val = 1;
1352 		break;
1353 
1354 	case SO_PASSCRED:
1355 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1356 		break;
1357 
1358 	case SO_PEERCRED:
1359 	{
1360 		struct ucred peercred;
1361 		if (len > sizeof(peercred))
1362 			len = sizeof(peercred);
1363 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1364 		if (copy_to_user(optval, &peercred, len))
1365 			return -EFAULT;
1366 		goto lenout;
1367 	}
1368 
1369 	case SO_PEERGROUPS:
1370 	{
1371 		int ret, n;
1372 
1373 		if (!sk->sk_peer_cred)
1374 			return -ENODATA;
1375 
1376 		n = sk->sk_peer_cred->group_info->ngroups;
1377 		if (len < n * sizeof(gid_t)) {
1378 			len = n * sizeof(gid_t);
1379 			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1380 		}
1381 		len = n * sizeof(gid_t);
1382 
1383 		ret = groups_to_user((gid_t __user *)optval,
1384 				     sk->sk_peer_cred->group_info);
1385 		if (ret)
1386 			return ret;
1387 		goto lenout;
1388 	}
1389 
1390 	case SO_PEERNAME:
1391 	{
1392 		char address[128];
1393 
1394 		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1395 		if (lv < 0)
1396 			return -ENOTCONN;
1397 		if (lv < len)
1398 			return -EINVAL;
1399 		if (copy_to_user(optval, address, len))
1400 			return -EFAULT;
1401 		goto lenout;
1402 	}
1403 
1404 	/* Dubious BSD thing... Probably nobody even uses it, but
1405 	 * the UNIX standard wants it for whatever reason... -DaveM
1406 	 */
1407 	case SO_ACCEPTCONN:
1408 		v.val = sk->sk_state == TCP_LISTEN;
1409 		break;
1410 
1411 	case SO_PASSSEC:
1412 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1413 		break;
1414 
1415 	case SO_PEERSEC:
1416 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1417 
1418 	case SO_MARK:
1419 		v.val = sk->sk_mark;
1420 		break;
1421 
1422 	case SO_RXQ_OVFL:
1423 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1424 		break;
1425 
1426 	case SO_WIFI_STATUS:
1427 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1428 		break;
1429 
1430 	case SO_PEEK_OFF:
1431 		if (!sock->ops->set_peek_off)
1432 			return -EOPNOTSUPP;
1433 
1434 		v.val = sk->sk_peek_off;
1435 		break;
1436 	case SO_NOFCS:
1437 		v.val = sock_flag(sk, SOCK_NOFCS);
1438 		break;
1439 
1440 	case SO_BINDTODEVICE:
1441 		return sock_getbindtodevice(sk, optval, optlen, len);
1442 
1443 	case SO_GET_FILTER:
1444 		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1445 		if (len < 0)
1446 			return len;
1447 
1448 		goto lenout;
1449 
1450 	case SO_LOCK_FILTER:
1451 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1452 		break;
1453 
1454 	case SO_BPF_EXTENSIONS:
1455 		v.val = bpf_tell_extensions();
1456 		break;
1457 
1458 	case SO_SELECT_ERR_QUEUE:
1459 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1460 		break;
1461 
1462 #ifdef CONFIG_NET_RX_BUSY_POLL
1463 	case SO_BUSY_POLL:
1464 		v.val = sk->sk_ll_usec;
1465 		break;
1466 #endif
1467 
1468 	case SO_MAX_PACING_RATE:
1469 		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1470 			lv = sizeof(v.ulval);
1471 			v.ulval = sk->sk_max_pacing_rate;
1472 		} else {
1473 			/* 32bit version */
1474 			v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1475 		}
1476 		break;
1477 
1478 	case SO_INCOMING_CPU:
1479 		v.val = sk->sk_incoming_cpu;
1480 		break;
1481 
1482 	case SO_MEMINFO:
1483 	{
1484 		u32 meminfo[SK_MEMINFO_VARS];
1485 
1486 		if (get_user(len, optlen))
1487 			return -EFAULT;
1488 
1489 		sk_get_meminfo(sk, meminfo);
1490 
1491 		len = min_t(unsigned int, len, sizeof(meminfo));
1492 		if (copy_to_user(optval, &meminfo, len))
1493 			return -EFAULT;
1494 
1495 		goto lenout;
1496 	}
1497 
1498 #ifdef CONFIG_NET_RX_BUSY_POLL
1499 	case SO_INCOMING_NAPI_ID:
1500 		v.val = READ_ONCE(sk->sk_napi_id);
1501 
1502 		/* aggregate non-NAPI IDs down to 0 */
1503 		if (v.val < MIN_NAPI_ID)
1504 			v.val = 0;
1505 
1506 		break;
1507 #endif
1508 
1509 	case SO_COOKIE:
1510 		lv = sizeof(u64);
1511 		if (len < lv)
1512 			return -EINVAL;
1513 		v.val64 = sock_gen_cookie(sk);
1514 		break;
1515 
1516 	case SO_ZEROCOPY:
1517 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1518 		break;
1519 
1520 	case SO_TXTIME:
1521 		lv = sizeof(v.txtime);
1522 		v.txtime.clockid = sk->sk_clockid;
1523 		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1524 				  SOF_TXTIME_DEADLINE_MODE : 0;
1525 		v.txtime.flags |= sk->sk_txtime_report_errors ?
1526 				  SOF_TXTIME_REPORT_ERRORS : 0;
1527 		break;
1528 
1529 	case SO_BINDTOIFINDEX:
1530 		v.val = sk->sk_bound_dev_if;
1531 		break;
1532 
1533 	default:
1534 		/* We implement the SO_SNDLOWAT etc to not be settable
1535 		 * (1003.1g 7).
1536 		 */
1537 		return -ENOPROTOOPT;
1538 	}
1539 
1540 	if (len > lv)
1541 		len = lv;
1542 	if (copy_to_user(optval, &v, len))
1543 		return -EFAULT;
1544 lenout:
1545 	if (put_user(len, optlen))
1546 		return -EFAULT;
1547 	return 0;
1548 }
1549 
1550 /*
1551  * Initialize an sk_lock.
1552  *
1553  * (We also register the sk_lock with the lock validator.)
1554  */
1555 static inline void sock_lock_init(struct sock *sk)
1556 {
1557 	if (sk->sk_kern_sock)
1558 		sock_lock_init_class_and_name(
1559 			sk,
1560 			af_family_kern_slock_key_strings[sk->sk_family],
1561 			af_family_kern_slock_keys + sk->sk_family,
1562 			af_family_kern_key_strings[sk->sk_family],
1563 			af_family_kern_keys + sk->sk_family);
1564 	else
1565 		sock_lock_init_class_and_name(
1566 			sk,
1567 			af_family_slock_key_strings[sk->sk_family],
1568 			af_family_slock_keys + sk->sk_family,
1569 			af_family_key_strings[sk->sk_family],
1570 			af_family_keys + sk->sk_family);
1571 }
1572 
1573 /*
1574  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1575  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1576  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1577  */
1578 static void sock_copy(struct sock *nsk, const struct sock *osk)
1579 {
1580 #ifdef CONFIG_SECURITY_NETWORK
1581 	void *sptr = nsk->sk_security;
1582 #endif
1583 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1584 
1585 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1586 	       osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1587 
1588 #ifdef CONFIG_SECURITY_NETWORK
1589 	nsk->sk_security = sptr;
1590 	security_sk_clone(osk, nsk);
1591 #endif
1592 }
1593 
1594 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1595 		int family)
1596 {
1597 	struct sock *sk;
1598 	struct kmem_cache *slab;
1599 
1600 	slab = prot->slab;
1601 	if (slab != NULL) {
1602 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1603 		if (!sk)
1604 			return sk;
1605 		if (priority & __GFP_ZERO)
1606 			sk_prot_clear_nulls(sk, prot->obj_size);
1607 	} else
1608 		sk = kmalloc(prot->obj_size, priority);
1609 
1610 	if (sk != NULL) {
1611 		if (security_sk_alloc(sk, family, priority))
1612 			goto out_free;
1613 
1614 		if (!try_module_get(prot->owner))
1615 			goto out_free_sec;
1616 		sk_tx_queue_clear(sk);
1617 	}
1618 
1619 	return sk;
1620 
1621 out_free_sec:
1622 	security_sk_free(sk);
1623 out_free:
1624 	if (slab != NULL)
1625 		kmem_cache_free(slab, sk);
1626 	else
1627 		kfree(sk);
1628 	return NULL;
1629 }
1630 
1631 static void sk_prot_free(struct proto *prot, struct sock *sk)
1632 {
1633 	struct kmem_cache *slab;
1634 	struct module *owner;
1635 
1636 	owner = prot->owner;
1637 	slab = prot->slab;
1638 
1639 	cgroup_sk_free(&sk->sk_cgrp_data);
1640 	mem_cgroup_sk_free(sk);
1641 	security_sk_free(sk);
1642 	if (slab != NULL)
1643 		kmem_cache_free(slab, sk);
1644 	else
1645 		kfree(sk);
1646 	module_put(owner);
1647 }
1648 
1649 /**
1650  *	sk_alloc - All socket objects are allocated here
1651  *	@net: the applicable net namespace
1652  *	@family: protocol family
1653  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1654  *	@prot: struct proto associated with this new sock instance
1655  *	@kern: is this to be a kernel socket?
1656  */
1657 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1658 		      struct proto *prot, int kern)
1659 {
1660 	struct sock *sk;
1661 
1662 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1663 	if (sk) {
1664 		sk->sk_family = family;
1665 		/*
1666 		 * See comment in struct sock definition to understand
1667 		 * why we need sk_prot_creator -acme
1668 		 */
1669 		sk->sk_prot = sk->sk_prot_creator = prot;
1670 		sk->sk_kern_sock = kern;
1671 		sock_lock_init(sk);
1672 		sk->sk_net_refcnt = kern ? 0 : 1;
1673 		if (likely(sk->sk_net_refcnt)) {
1674 			get_net(net);
1675 			sock_inuse_add(net, 1);
1676 		}
1677 
1678 		sock_net_set(sk, net);
1679 		refcount_set(&sk->sk_wmem_alloc, 1);
1680 
1681 		mem_cgroup_sk_alloc(sk);
1682 		cgroup_sk_alloc(&sk->sk_cgrp_data);
1683 		sock_update_classid(&sk->sk_cgrp_data);
1684 		sock_update_netprioidx(&sk->sk_cgrp_data);
1685 	}
1686 
1687 	return sk;
1688 }
1689 EXPORT_SYMBOL(sk_alloc);
1690 
1691 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1692  * grace period. This is the case for UDP sockets and TCP listeners.
1693  */
1694 static void __sk_destruct(struct rcu_head *head)
1695 {
1696 	struct sock *sk = container_of(head, struct sock, sk_rcu);
1697 	struct sk_filter *filter;
1698 
1699 	if (sk->sk_destruct)
1700 		sk->sk_destruct(sk);
1701 
1702 	filter = rcu_dereference_check(sk->sk_filter,
1703 				       refcount_read(&sk->sk_wmem_alloc) == 0);
1704 	if (filter) {
1705 		sk_filter_uncharge(sk, filter);
1706 		RCU_INIT_POINTER(sk->sk_filter, NULL);
1707 	}
1708 	if (rcu_access_pointer(sk->sk_reuseport_cb))
1709 		reuseport_detach_sock(sk);
1710 
1711 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1712 
1713 #ifdef CONFIG_BPF_SYSCALL
1714 	bpf_sk_storage_free(sk);
1715 #endif
1716 
1717 	if (atomic_read(&sk->sk_omem_alloc))
1718 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1719 			 __func__, atomic_read(&sk->sk_omem_alloc));
1720 
1721 	if (sk->sk_frag.page) {
1722 		put_page(sk->sk_frag.page);
1723 		sk->sk_frag.page = NULL;
1724 	}
1725 
1726 	if (sk->sk_peer_cred)
1727 		put_cred(sk->sk_peer_cred);
1728 	put_pid(sk->sk_peer_pid);
1729 	if (likely(sk->sk_net_refcnt))
1730 		put_net(sock_net(sk));
1731 	sk_prot_free(sk->sk_prot_creator, sk);
1732 }
1733 
1734 void sk_destruct(struct sock *sk)
1735 {
1736 	if (sock_flag(sk, SOCK_RCU_FREE))
1737 		call_rcu(&sk->sk_rcu, __sk_destruct);
1738 	else
1739 		__sk_destruct(&sk->sk_rcu);
1740 }
1741 
1742 static void __sk_free(struct sock *sk)
1743 {
1744 	if (likely(sk->sk_net_refcnt))
1745 		sock_inuse_add(sock_net(sk), -1);
1746 
1747 	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1748 		sock_diag_broadcast_destroy(sk);
1749 	else
1750 		sk_destruct(sk);
1751 }
1752 
1753 void sk_free(struct sock *sk)
1754 {
1755 	/*
1756 	 * We subtract one from sk_wmem_alloc and can know if
1757 	 * some packets are still in some tx queue.
1758 	 * If not null, sock_wfree() will call __sk_free(sk) later
1759 	 */
1760 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1761 		__sk_free(sk);
1762 }
1763 EXPORT_SYMBOL(sk_free);
1764 
1765 static void sk_init_common(struct sock *sk)
1766 {
1767 	skb_queue_head_init(&sk->sk_receive_queue);
1768 	skb_queue_head_init(&sk->sk_write_queue);
1769 	skb_queue_head_init(&sk->sk_error_queue);
1770 
1771 	rwlock_init(&sk->sk_callback_lock);
1772 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1773 			af_rlock_keys + sk->sk_family,
1774 			af_family_rlock_key_strings[sk->sk_family]);
1775 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1776 			af_wlock_keys + sk->sk_family,
1777 			af_family_wlock_key_strings[sk->sk_family]);
1778 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1779 			af_elock_keys + sk->sk_family,
1780 			af_family_elock_key_strings[sk->sk_family]);
1781 	lockdep_set_class_and_name(&sk->sk_callback_lock,
1782 			af_callback_keys + sk->sk_family,
1783 			af_family_clock_key_strings[sk->sk_family]);
1784 }
1785 
1786 /**
1787  *	sk_clone_lock - clone a socket, and lock its clone
1788  *	@sk: the socket to clone
1789  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1790  *
1791  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1792  */
1793 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1794 {
1795 	struct sock *newsk;
1796 	bool is_charged = true;
1797 
1798 	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1799 	if (newsk != NULL) {
1800 		struct sk_filter *filter;
1801 
1802 		sock_copy(newsk, sk);
1803 
1804 		newsk->sk_prot_creator = sk->sk_prot;
1805 
1806 		/* SANITY */
1807 		if (likely(newsk->sk_net_refcnt))
1808 			get_net(sock_net(newsk));
1809 		sk_node_init(&newsk->sk_node);
1810 		sock_lock_init(newsk);
1811 		bh_lock_sock(newsk);
1812 		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1813 		newsk->sk_backlog.len = 0;
1814 
1815 		atomic_set(&newsk->sk_rmem_alloc, 0);
1816 		/*
1817 		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1818 		 */
1819 		refcount_set(&newsk->sk_wmem_alloc, 1);
1820 		atomic_set(&newsk->sk_omem_alloc, 0);
1821 		sk_init_common(newsk);
1822 
1823 		newsk->sk_dst_cache	= NULL;
1824 		newsk->sk_dst_pending_confirm = 0;
1825 		newsk->sk_wmem_queued	= 0;
1826 		newsk->sk_forward_alloc = 0;
1827 		atomic_set(&newsk->sk_drops, 0);
1828 		newsk->sk_send_head	= NULL;
1829 		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1830 		atomic_set(&newsk->sk_zckey, 0);
1831 
1832 		sock_reset_flag(newsk, SOCK_DONE);
1833 		mem_cgroup_sk_alloc(newsk);
1834 		cgroup_sk_alloc(&newsk->sk_cgrp_data);
1835 
1836 		rcu_read_lock();
1837 		filter = rcu_dereference(sk->sk_filter);
1838 		if (filter != NULL)
1839 			/* though it's an empty new sock, the charging may fail
1840 			 * if sysctl_optmem_max was changed between creation of
1841 			 * original socket and cloning
1842 			 */
1843 			is_charged = sk_filter_charge(newsk, filter);
1844 		RCU_INIT_POINTER(newsk->sk_filter, filter);
1845 		rcu_read_unlock();
1846 
1847 		if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1848 			/* We need to make sure that we don't uncharge the new
1849 			 * socket if we couldn't charge it in the first place
1850 			 * as otherwise we uncharge the parent's filter.
1851 			 */
1852 			if (!is_charged)
1853 				RCU_INIT_POINTER(newsk->sk_filter, NULL);
1854 			sk_free_unlock_clone(newsk);
1855 			newsk = NULL;
1856 			goto out;
1857 		}
1858 		RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1859 
1860 		newsk->sk_err	   = 0;
1861 		newsk->sk_err_soft = 0;
1862 		newsk->sk_priority = 0;
1863 		newsk->sk_incoming_cpu = raw_smp_processor_id();
1864 		if (likely(newsk->sk_net_refcnt))
1865 			sock_inuse_add(sock_net(newsk), 1);
1866 
1867 		/*
1868 		 * Before updating sk_refcnt, we must commit prior changes to memory
1869 		 * (Documentation/RCU/rculist_nulls.txt for details)
1870 		 */
1871 		smp_wmb();
1872 		refcount_set(&newsk->sk_refcnt, 2);
1873 
1874 		/*
1875 		 * Increment the counter in the same struct proto as the master
1876 		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1877 		 * is the same as sk->sk_prot->socks, as this field was copied
1878 		 * with memcpy).
1879 		 *
1880 		 * This _changes_ the previous behaviour, where
1881 		 * tcp_create_openreq_child always was incrementing the
1882 		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1883 		 * to be taken into account in all callers. -acme
1884 		 */
1885 		sk_refcnt_debug_inc(newsk);
1886 		sk_set_socket(newsk, NULL);
1887 		RCU_INIT_POINTER(newsk->sk_wq, NULL);
1888 
1889 		if (newsk->sk_prot->sockets_allocated)
1890 			sk_sockets_allocated_inc(newsk);
1891 
1892 		if (sock_needs_netstamp(sk) &&
1893 		    newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1894 			net_enable_timestamp();
1895 	}
1896 out:
1897 	return newsk;
1898 }
1899 EXPORT_SYMBOL_GPL(sk_clone_lock);
1900 
1901 void sk_free_unlock_clone(struct sock *sk)
1902 {
1903 	/* It is still raw copy of parent, so invalidate
1904 	 * destructor and make plain sk_free() */
1905 	sk->sk_destruct = NULL;
1906 	bh_unlock_sock(sk);
1907 	sk_free(sk);
1908 }
1909 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1910 
1911 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1912 {
1913 	u32 max_segs = 1;
1914 
1915 	sk_dst_set(sk, dst);
1916 	sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1917 	if (sk->sk_route_caps & NETIF_F_GSO)
1918 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1919 	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1920 	if (sk_can_gso(sk)) {
1921 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1922 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1923 		} else {
1924 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1925 			sk->sk_gso_max_size = dst->dev->gso_max_size;
1926 			max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1927 		}
1928 	}
1929 	sk->sk_gso_max_segs = max_segs;
1930 }
1931 EXPORT_SYMBOL_GPL(sk_setup_caps);
1932 
1933 /*
1934  *	Simple resource managers for sockets.
1935  */
1936 
1937 
1938 /*
1939  * Write buffer destructor automatically called from kfree_skb.
1940  */
1941 void sock_wfree(struct sk_buff *skb)
1942 {
1943 	struct sock *sk = skb->sk;
1944 	unsigned int len = skb->truesize;
1945 
1946 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1947 		/*
1948 		 * Keep a reference on sk_wmem_alloc, this will be released
1949 		 * after sk_write_space() call
1950 		 */
1951 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1952 		sk->sk_write_space(sk);
1953 		len = 1;
1954 	}
1955 	/*
1956 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1957 	 * could not do because of in-flight packets
1958 	 */
1959 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1960 		__sk_free(sk);
1961 }
1962 EXPORT_SYMBOL(sock_wfree);
1963 
1964 /* This variant of sock_wfree() is used by TCP,
1965  * since it sets SOCK_USE_WRITE_QUEUE.
1966  */
1967 void __sock_wfree(struct sk_buff *skb)
1968 {
1969 	struct sock *sk = skb->sk;
1970 
1971 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1972 		__sk_free(sk);
1973 }
1974 
1975 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1976 {
1977 	skb_orphan(skb);
1978 	skb->sk = sk;
1979 #ifdef CONFIG_INET
1980 	if (unlikely(!sk_fullsock(sk))) {
1981 		skb->destructor = sock_edemux;
1982 		sock_hold(sk);
1983 		return;
1984 	}
1985 #endif
1986 	skb->destructor = sock_wfree;
1987 	skb_set_hash_from_sk(skb, sk);
1988 	/*
1989 	 * We used to take a refcount on sk, but following operation
1990 	 * is enough to guarantee sk_free() wont free this sock until
1991 	 * all in-flight packets are completed
1992 	 */
1993 	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1994 }
1995 EXPORT_SYMBOL(skb_set_owner_w);
1996 
1997 /* This helper is used by netem, as it can hold packets in its
1998  * delay queue. We want to allow the owner socket to send more
1999  * packets, as if they were already TX completed by a typical driver.
2000  * But we also want to keep skb->sk set because some packet schedulers
2001  * rely on it (sch_fq for example).
2002  */
2003 void skb_orphan_partial(struct sk_buff *skb)
2004 {
2005 	if (skb_is_tcp_pure_ack(skb))
2006 		return;
2007 
2008 	if (skb->destructor == sock_wfree
2009 #ifdef CONFIG_INET
2010 	    || skb->destructor == tcp_wfree
2011 #endif
2012 		) {
2013 		struct sock *sk = skb->sk;
2014 
2015 		if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2016 			WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2017 			skb->destructor = sock_efree;
2018 		}
2019 	} else {
2020 		skb_orphan(skb);
2021 	}
2022 }
2023 EXPORT_SYMBOL(skb_orphan_partial);
2024 
2025 /*
2026  * Read buffer destructor automatically called from kfree_skb.
2027  */
2028 void sock_rfree(struct sk_buff *skb)
2029 {
2030 	struct sock *sk = skb->sk;
2031 	unsigned int len = skb->truesize;
2032 
2033 	atomic_sub(len, &sk->sk_rmem_alloc);
2034 	sk_mem_uncharge(sk, len);
2035 }
2036 EXPORT_SYMBOL(sock_rfree);
2037 
2038 /*
2039  * Buffer destructor for skbs that are not used directly in read or write
2040  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2041  */
2042 void sock_efree(struct sk_buff *skb)
2043 {
2044 	sock_put(skb->sk);
2045 }
2046 EXPORT_SYMBOL(sock_efree);
2047 
2048 kuid_t sock_i_uid(struct sock *sk)
2049 {
2050 	kuid_t uid;
2051 
2052 	read_lock_bh(&sk->sk_callback_lock);
2053 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2054 	read_unlock_bh(&sk->sk_callback_lock);
2055 	return uid;
2056 }
2057 EXPORT_SYMBOL(sock_i_uid);
2058 
2059 unsigned long sock_i_ino(struct sock *sk)
2060 {
2061 	unsigned long ino;
2062 
2063 	read_lock_bh(&sk->sk_callback_lock);
2064 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2065 	read_unlock_bh(&sk->sk_callback_lock);
2066 	return ino;
2067 }
2068 EXPORT_SYMBOL(sock_i_ino);
2069 
2070 /*
2071  * Allocate a skb from the socket's send buffer.
2072  */
2073 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2074 			     gfp_t priority)
2075 {
2076 	if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2077 		struct sk_buff *skb = alloc_skb(size, priority);
2078 		if (skb) {
2079 			skb_set_owner_w(skb, sk);
2080 			return skb;
2081 		}
2082 	}
2083 	return NULL;
2084 }
2085 EXPORT_SYMBOL(sock_wmalloc);
2086 
2087 static void sock_ofree(struct sk_buff *skb)
2088 {
2089 	struct sock *sk = skb->sk;
2090 
2091 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2092 }
2093 
2094 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2095 			     gfp_t priority)
2096 {
2097 	struct sk_buff *skb;
2098 
2099 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2100 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2101 	    sysctl_optmem_max)
2102 		return NULL;
2103 
2104 	skb = alloc_skb(size, priority);
2105 	if (!skb)
2106 		return NULL;
2107 
2108 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2109 	skb->sk = sk;
2110 	skb->destructor = sock_ofree;
2111 	return skb;
2112 }
2113 
2114 /*
2115  * Allocate a memory block from the socket's option memory buffer.
2116  */
2117 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2118 {
2119 	if ((unsigned int)size <= sysctl_optmem_max &&
2120 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2121 		void *mem;
2122 		/* First do the add, to avoid the race if kmalloc
2123 		 * might sleep.
2124 		 */
2125 		atomic_add(size, &sk->sk_omem_alloc);
2126 		mem = kmalloc(size, priority);
2127 		if (mem)
2128 			return mem;
2129 		atomic_sub(size, &sk->sk_omem_alloc);
2130 	}
2131 	return NULL;
2132 }
2133 EXPORT_SYMBOL(sock_kmalloc);
2134 
2135 /* Free an option memory block. Note, we actually want the inline
2136  * here as this allows gcc to detect the nullify and fold away the
2137  * condition entirely.
2138  */
2139 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2140 				  const bool nullify)
2141 {
2142 	if (WARN_ON_ONCE(!mem))
2143 		return;
2144 	if (nullify)
2145 		kzfree(mem);
2146 	else
2147 		kfree(mem);
2148 	atomic_sub(size, &sk->sk_omem_alloc);
2149 }
2150 
2151 void sock_kfree_s(struct sock *sk, void *mem, int size)
2152 {
2153 	__sock_kfree_s(sk, mem, size, false);
2154 }
2155 EXPORT_SYMBOL(sock_kfree_s);
2156 
2157 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2158 {
2159 	__sock_kfree_s(sk, mem, size, true);
2160 }
2161 EXPORT_SYMBOL(sock_kzfree_s);
2162 
2163 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2164    I think, these locks should be removed for datagram sockets.
2165  */
2166 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2167 {
2168 	DEFINE_WAIT(wait);
2169 
2170 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2171 	for (;;) {
2172 		if (!timeo)
2173 			break;
2174 		if (signal_pending(current))
2175 			break;
2176 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2177 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2178 		if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2179 			break;
2180 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2181 			break;
2182 		if (sk->sk_err)
2183 			break;
2184 		timeo = schedule_timeout(timeo);
2185 	}
2186 	finish_wait(sk_sleep(sk), &wait);
2187 	return timeo;
2188 }
2189 
2190 
2191 /*
2192  *	Generic send/receive buffer handlers
2193  */
2194 
2195 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2196 				     unsigned long data_len, int noblock,
2197 				     int *errcode, int max_page_order)
2198 {
2199 	struct sk_buff *skb;
2200 	long timeo;
2201 	int err;
2202 
2203 	timeo = sock_sndtimeo(sk, noblock);
2204 	for (;;) {
2205 		err = sock_error(sk);
2206 		if (err != 0)
2207 			goto failure;
2208 
2209 		err = -EPIPE;
2210 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2211 			goto failure;
2212 
2213 		if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2214 			break;
2215 
2216 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2217 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2218 		err = -EAGAIN;
2219 		if (!timeo)
2220 			goto failure;
2221 		if (signal_pending(current))
2222 			goto interrupted;
2223 		timeo = sock_wait_for_wmem(sk, timeo);
2224 	}
2225 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2226 				   errcode, sk->sk_allocation);
2227 	if (skb)
2228 		skb_set_owner_w(skb, sk);
2229 	return skb;
2230 
2231 interrupted:
2232 	err = sock_intr_errno(timeo);
2233 failure:
2234 	*errcode = err;
2235 	return NULL;
2236 }
2237 EXPORT_SYMBOL(sock_alloc_send_pskb);
2238 
2239 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2240 				    int noblock, int *errcode)
2241 {
2242 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2243 }
2244 EXPORT_SYMBOL(sock_alloc_send_skb);
2245 
2246 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2247 		     struct sockcm_cookie *sockc)
2248 {
2249 	u32 tsflags;
2250 
2251 	switch (cmsg->cmsg_type) {
2252 	case SO_MARK:
2253 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2254 			return -EPERM;
2255 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2256 			return -EINVAL;
2257 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2258 		break;
2259 	case SO_TIMESTAMPING_OLD:
2260 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2261 			return -EINVAL;
2262 
2263 		tsflags = *(u32 *)CMSG_DATA(cmsg);
2264 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2265 			return -EINVAL;
2266 
2267 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2268 		sockc->tsflags |= tsflags;
2269 		break;
2270 	case SCM_TXTIME:
2271 		if (!sock_flag(sk, SOCK_TXTIME))
2272 			return -EINVAL;
2273 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2274 			return -EINVAL;
2275 		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2276 		break;
2277 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2278 	case SCM_RIGHTS:
2279 	case SCM_CREDENTIALS:
2280 		break;
2281 	default:
2282 		return -EINVAL;
2283 	}
2284 	return 0;
2285 }
2286 EXPORT_SYMBOL(__sock_cmsg_send);
2287 
2288 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2289 		   struct sockcm_cookie *sockc)
2290 {
2291 	struct cmsghdr *cmsg;
2292 	int ret;
2293 
2294 	for_each_cmsghdr(cmsg, msg) {
2295 		if (!CMSG_OK(msg, cmsg))
2296 			return -EINVAL;
2297 		if (cmsg->cmsg_level != SOL_SOCKET)
2298 			continue;
2299 		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2300 		if (ret)
2301 			return ret;
2302 	}
2303 	return 0;
2304 }
2305 EXPORT_SYMBOL(sock_cmsg_send);
2306 
2307 static void sk_enter_memory_pressure(struct sock *sk)
2308 {
2309 	if (!sk->sk_prot->enter_memory_pressure)
2310 		return;
2311 
2312 	sk->sk_prot->enter_memory_pressure(sk);
2313 }
2314 
2315 static void sk_leave_memory_pressure(struct sock *sk)
2316 {
2317 	if (sk->sk_prot->leave_memory_pressure) {
2318 		sk->sk_prot->leave_memory_pressure(sk);
2319 	} else {
2320 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2321 
2322 		if (memory_pressure && *memory_pressure)
2323 			*memory_pressure = 0;
2324 	}
2325 }
2326 
2327 /* On 32bit arches, an skb frag is limited to 2^15 */
2328 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
2329 
2330 /**
2331  * skb_page_frag_refill - check that a page_frag contains enough room
2332  * @sz: minimum size of the fragment we want to get
2333  * @pfrag: pointer to page_frag
2334  * @gfp: priority for memory allocation
2335  *
2336  * Note: While this allocator tries to use high order pages, there is
2337  * no guarantee that allocations succeed. Therefore, @sz MUST be
2338  * less or equal than PAGE_SIZE.
2339  */
2340 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2341 {
2342 	if (pfrag->page) {
2343 		if (page_ref_count(pfrag->page) == 1) {
2344 			pfrag->offset = 0;
2345 			return true;
2346 		}
2347 		if (pfrag->offset + sz <= pfrag->size)
2348 			return true;
2349 		put_page(pfrag->page);
2350 	}
2351 
2352 	pfrag->offset = 0;
2353 	if (SKB_FRAG_PAGE_ORDER) {
2354 		/* Avoid direct reclaim but allow kswapd to wake */
2355 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2356 					  __GFP_COMP | __GFP_NOWARN |
2357 					  __GFP_NORETRY,
2358 					  SKB_FRAG_PAGE_ORDER);
2359 		if (likely(pfrag->page)) {
2360 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2361 			return true;
2362 		}
2363 	}
2364 	pfrag->page = alloc_page(gfp);
2365 	if (likely(pfrag->page)) {
2366 		pfrag->size = PAGE_SIZE;
2367 		return true;
2368 	}
2369 	return false;
2370 }
2371 EXPORT_SYMBOL(skb_page_frag_refill);
2372 
2373 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2374 {
2375 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2376 		return true;
2377 
2378 	sk_enter_memory_pressure(sk);
2379 	sk_stream_moderate_sndbuf(sk);
2380 	return false;
2381 }
2382 EXPORT_SYMBOL(sk_page_frag_refill);
2383 
2384 static void __lock_sock(struct sock *sk)
2385 	__releases(&sk->sk_lock.slock)
2386 	__acquires(&sk->sk_lock.slock)
2387 {
2388 	DEFINE_WAIT(wait);
2389 
2390 	for (;;) {
2391 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2392 					TASK_UNINTERRUPTIBLE);
2393 		spin_unlock_bh(&sk->sk_lock.slock);
2394 		schedule();
2395 		spin_lock_bh(&sk->sk_lock.slock);
2396 		if (!sock_owned_by_user(sk))
2397 			break;
2398 	}
2399 	finish_wait(&sk->sk_lock.wq, &wait);
2400 }
2401 
2402 void __release_sock(struct sock *sk)
2403 	__releases(&sk->sk_lock.slock)
2404 	__acquires(&sk->sk_lock.slock)
2405 {
2406 	struct sk_buff *skb, *next;
2407 
2408 	while ((skb = sk->sk_backlog.head) != NULL) {
2409 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2410 
2411 		spin_unlock_bh(&sk->sk_lock.slock);
2412 
2413 		do {
2414 			next = skb->next;
2415 			prefetch(next);
2416 			WARN_ON_ONCE(skb_dst_is_noref(skb));
2417 			skb_mark_not_on_list(skb);
2418 			sk_backlog_rcv(sk, skb);
2419 
2420 			cond_resched();
2421 
2422 			skb = next;
2423 		} while (skb != NULL);
2424 
2425 		spin_lock_bh(&sk->sk_lock.slock);
2426 	}
2427 
2428 	/*
2429 	 * Doing the zeroing here guarantee we can not loop forever
2430 	 * while a wild producer attempts to flood us.
2431 	 */
2432 	sk->sk_backlog.len = 0;
2433 }
2434 
2435 void __sk_flush_backlog(struct sock *sk)
2436 {
2437 	spin_lock_bh(&sk->sk_lock.slock);
2438 	__release_sock(sk);
2439 	spin_unlock_bh(&sk->sk_lock.slock);
2440 }
2441 
2442 /**
2443  * sk_wait_data - wait for data to arrive at sk_receive_queue
2444  * @sk:    sock to wait on
2445  * @timeo: for how long
2446  * @skb:   last skb seen on sk_receive_queue
2447  *
2448  * Now socket state including sk->sk_err is changed only under lock,
2449  * hence we may omit checks after joining wait queue.
2450  * We check receive queue before schedule() only as optimization;
2451  * it is very likely that release_sock() added new data.
2452  */
2453 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2454 {
2455 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2456 	int rc;
2457 
2458 	add_wait_queue(sk_sleep(sk), &wait);
2459 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2460 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2461 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2462 	remove_wait_queue(sk_sleep(sk), &wait);
2463 	return rc;
2464 }
2465 EXPORT_SYMBOL(sk_wait_data);
2466 
2467 /**
2468  *	__sk_mem_raise_allocated - increase memory_allocated
2469  *	@sk: socket
2470  *	@size: memory size to allocate
2471  *	@amt: pages to allocate
2472  *	@kind: allocation type
2473  *
2474  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2475  */
2476 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2477 {
2478 	struct proto *prot = sk->sk_prot;
2479 	long allocated = sk_memory_allocated_add(sk, amt);
2480 	bool charged = true;
2481 
2482 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2483 	    !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2484 		goto suppress_allocation;
2485 
2486 	/* Under limit. */
2487 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2488 		sk_leave_memory_pressure(sk);
2489 		return 1;
2490 	}
2491 
2492 	/* Under pressure. */
2493 	if (allocated > sk_prot_mem_limits(sk, 1))
2494 		sk_enter_memory_pressure(sk);
2495 
2496 	/* Over hard limit. */
2497 	if (allocated > sk_prot_mem_limits(sk, 2))
2498 		goto suppress_allocation;
2499 
2500 	/* guarantee minimum buffer size under pressure */
2501 	if (kind == SK_MEM_RECV) {
2502 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2503 			return 1;
2504 
2505 	} else { /* SK_MEM_SEND */
2506 		int wmem0 = sk_get_wmem0(sk, prot);
2507 
2508 		if (sk->sk_type == SOCK_STREAM) {
2509 			if (sk->sk_wmem_queued < wmem0)
2510 				return 1;
2511 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2512 				return 1;
2513 		}
2514 	}
2515 
2516 	if (sk_has_memory_pressure(sk)) {
2517 		u64 alloc;
2518 
2519 		if (!sk_under_memory_pressure(sk))
2520 			return 1;
2521 		alloc = sk_sockets_allocated_read_positive(sk);
2522 		if (sk_prot_mem_limits(sk, 2) > alloc *
2523 		    sk_mem_pages(sk->sk_wmem_queued +
2524 				 atomic_read(&sk->sk_rmem_alloc) +
2525 				 sk->sk_forward_alloc))
2526 			return 1;
2527 	}
2528 
2529 suppress_allocation:
2530 
2531 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2532 		sk_stream_moderate_sndbuf(sk);
2533 
2534 		/* Fail only if socket is _under_ its sndbuf.
2535 		 * In this case we cannot block, so that we have to fail.
2536 		 */
2537 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2538 			return 1;
2539 	}
2540 
2541 	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2542 		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2543 
2544 	sk_memory_allocated_sub(sk, amt);
2545 
2546 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2547 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2548 
2549 	return 0;
2550 }
2551 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2552 
2553 /**
2554  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2555  *	@sk: socket
2556  *	@size: memory size to allocate
2557  *	@kind: allocation type
2558  *
2559  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2560  *	rmem allocation. This function assumes that protocols which have
2561  *	memory_pressure use sk_wmem_queued as write buffer accounting.
2562  */
2563 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2564 {
2565 	int ret, amt = sk_mem_pages(size);
2566 
2567 	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2568 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2569 	if (!ret)
2570 		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2571 	return ret;
2572 }
2573 EXPORT_SYMBOL(__sk_mem_schedule);
2574 
2575 /**
2576  *	__sk_mem_reduce_allocated - reclaim memory_allocated
2577  *	@sk: socket
2578  *	@amount: number of quanta
2579  *
2580  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2581  */
2582 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2583 {
2584 	sk_memory_allocated_sub(sk, amount);
2585 
2586 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2587 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2588 
2589 	if (sk_under_memory_pressure(sk) &&
2590 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2591 		sk_leave_memory_pressure(sk);
2592 }
2593 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2594 
2595 /**
2596  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2597  *	@sk: socket
2598  *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2599  */
2600 void __sk_mem_reclaim(struct sock *sk, int amount)
2601 {
2602 	amount >>= SK_MEM_QUANTUM_SHIFT;
2603 	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2604 	__sk_mem_reduce_allocated(sk, amount);
2605 }
2606 EXPORT_SYMBOL(__sk_mem_reclaim);
2607 
2608 int sk_set_peek_off(struct sock *sk, int val)
2609 {
2610 	sk->sk_peek_off = val;
2611 	return 0;
2612 }
2613 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2614 
2615 /*
2616  * Set of default routines for initialising struct proto_ops when
2617  * the protocol does not support a particular function. In certain
2618  * cases where it makes no sense for a protocol to have a "do nothing"
2619  * function, some default processing is provided.
2620  */
2621 
2622 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2623 {
2624 	return -EOPNOTSUPP;
2625 }
2626 EXPORT_SYMBOL(sock_no_bind);
2627 
2628 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2629 		    int len, int flags)
2630 {
2631 	return -EOPNOTSUPP;
2632 }
2633 EXPORT_SYMBOL(sock_no_connect);
2634 
2635 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2636 {
2637 	return -EOPNOTSUPP;
2638 }
2639 EXPORT_SYMBOL(sock_no_socketpair);
2640 
2641 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2642 		   bool kern)
2643 {
2644 	return -EOPNOTSUPP;
2645 }
2646 EXPORT_SYMBOL(sock_no_accept);
2647 
2648 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2649 		    int peer)
2650 {
2651 	return -EOPNOTSUPP;
2652 }
2653 EXPORT_SYMBOL(sock_no_getname);
2654 
2655 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2656 {
2657 	return -EOPNOTSUPP;
2658 }
2659 EXPORT_SYMBOL(sock_no_ioctl);
2660 
2661 int sock_no_listen(struct socket *sock, int backlog)
2662 {
2663 	return -EOPNOTSUPP;
2664 }
2665 EXPORT_SYMBOL(sock_no_listen);
2666 
2667 int sock_no_shutdown(struct socket *sock, int how)
2668 {
2669 	return -EOPNOTSUPP;
2670 }
2671 EXPORT_SYMBOL(sock_no_shutdown);
2672 
2673 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2674 		    char __user *optval, unsigned int optlen)
2675 {
2676 	return -EOPNOTSUPP;
2677 }
2678 EXPORT_SYMBOL(sock_no_setsockopt);
2679 
2680 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2681 		    char __user *optval, int __user *optlen)
2682 {
2683 	return -EOPNOTSUPP;
2684 }
2685 EXPORT_SYMBOL(sock_no_getsockopt);
2686 
2687 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2688 {
2689 	return -EOPNOTSUPP;
2690 }
2691 EXPORT_SYMBOL(sock_no_sendmsg);
2692 
2693 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2694 {
2695 	return -EOPNOTSUPP;
2696 }
2697 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2698 
2699 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2700 		    int flags)
2701 {
2702 	return -EOPNOTSUPP;
2703 }
2704 EXPORT_SYMBOL(sock_no_recvmsg);
2705 
2706 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2707 {
2708 	/* Mirror missing mmap method error code */
2709 	return -ENODEV;
2710 }
2711 EXPORT_SYMBOL(sock_no_mmap);
2712 
2713 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2714 {
2715 	ssize_t res;
2716 	struct msghdr msg = {.msg_flags = flags};
2717 	struct kvec iov;
2718 	char *kaddr = kmap(page);
2719 	iov.iov_base = kaddr + offset;
2720 	iov.iov_len = size;
2721 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2722 	kunmap(page);
2723 	return res;
2724 }
2725 EXPORT_SYMBOL(sock_no_sendpage);
2726 
2727 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2728 				int offset, size_t size, int flags)
2729 {
2730 	ssize_t res;
2731 	struct msghdr msg = {.msg_flags = flags};
2732 	struct kvec iov;
2733 	char *kaddr = kmap(page);
2734 
2735 	iov.iov_base = kaddr + offset;
2736 	iov.iov_len = size;
2737 	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2738 	kunmap(page);
2739 	return res;
2740 }
2741 EXPORT_SYMBOL(sock_no_sendpage_locked);
2742 
2743 /*
2744  *	Default Socket Callbacks
2745  */
2746 
2747 static void sock_def_wakeup(struct sock *sk)
2748 {
2749 	struct socket_wq *wq;
2750 
2751 	rcu_read_lock();
2752 	wq = rcu_dereference(sk->sk_wq);
2753 	if (skwq_has_sleeper(wq))
2754 		wake_up_interruptible_all(&wq->wait);
2755 	rcu_read_unlock();
2756 }
2757 
2758 static void sock_def_error_report(struct sock *sk)
2759 {
2760 	struct socket_wq *wq;
2761 
2762 	rcu_read_lock();
2763 	wq = rcu_dereference(sk->sk_wq);
2764 	if (skwq_has_sleeper(wq))
2765 		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2766 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2767 	rcu_read_unlock();
2768 }
2769 
2770 static void sock_def_readable(struct sock *sk)
2771 {
2772 	struct socket_wq *wq;
2773 
2774 	rcu_read_lock();
2775 	wq = rcu_dereference(sk->sk_wq);
2776 	if (skwq_has_sleeper(wq))
2777 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2778 						EPOLLRDNORM | EPOLLRDBAND);
2779 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2780 	rcu_read_unlock();
2781 }
2782 
2783 static void sock_def_write_space(struct sock *sk)
2784 {
2785 	struct socket_wq *wq;
2786 
2787 	rcu_read_lock();
2788 
2789 	/* Do not wake up a writer until he can make "significant"
2790 	 * progress.  --DaveM
2791 	 */
2792 	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2793 		wq = rcu_dereference(sk->sk_wq);
2794 		if (skwq_has_sleeper(wq))
2795 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2796 						EPOLLWRNORM | EPOLLWRBAND);
2797 
2798 		/* Should agree with poll, otherwise some programs break */
2799 		if (sock_writeable(sk))
2800 			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2801 	}
2802 
2803 	rcu_read_unlock();
2804 }
2805 
2806 static void sock_def_destruct(struct sock *sk)
2807 {
2808 }
2809 
2810 void sk_send_sigurg(struct sock *sk)
2811 {
2812 	if (sk->sk_socket && sk->sk_socket->file)
2813 		if (send_sigurg(&sk->sk_socket->file->f_owner))
2814 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2815 }
2816 EXPORT_SYMBOL(sk_send_sigurg);
2817 
2818 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2819 		    unsigned long expires)
2820 {
2821 	if (!mod_timer(timer, expires))
2822 		sock_hold(sk);
2823 }
2824 EXPORT_SYMBOL(sk_reset_timer);
2825 
2826 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2827 {
2828 	if (del_timer(timer))
2829 		__sock_put(sk);
2830 }
2831 EXPORT_SYMBOL(sk_stop_timer);
2832 
2833 void sock_init_data(struct socket *sock, struct sock *sk)
2834 {
2835 	sk_init_common(sk);
2836 	sk->sk_send_head	=	NULL;
2837 
2838 	timer_setup(&sk->sk_timer, NULL, 0);
2839 
2840 	sk->sk_allocation	=	GFP_KERNEL;
2841 	sk->sk_rcvbuf		=	sysctl_rmem_default;
2842 	sk->sk_sndbuf		=	sysctl_wmem_default;
2843 	sk->sk_state		=	TCP_CLOSE;
2844 	sk_set_socket(sk, sock);
2845 
2846 	sock_set_flag(sk, SOCK_ZAPPED);
2847 
2848 	if (sock) {
2849 		sk->sk_type	=	sock->type;
2850 		RCU_INIT_POINTER(sk->sk_wq, sock->wq);
2851 		sock->sk	=	sk;
2852 		sk->sk_uid	=	SOCK_INODE(sock)->i_uid;
2853 	} else {
2854 		RCU_INIT_POINTER(sk->sk_wq, NULL);
2855 		sk->sk_uid	=	make_kuid(sock_net(sk)->user_ns, 0);
2856 	}
2857 
2858 	rwlock_init(&sk->sk_callback_lock);
2859 	if (sk->sk_kern_sock)
2860 		lockdep_set_class_and_name(
2861 			&sk->sk_callback_lock,
2862 			af_kern_callback_keys + sk->sk_family,
2863 			af_family_kern_clock_key_strings[sk->sk_family]);
2864 	else
2865 		lockdep_set_class_and_name(
2866 			&sk->sk_callback_lock,
2867 			af_callback_keys + sk->sk_family,
2868 			af_family_clock_key_strings[sk->sk_family]);
2869 
2870 	sk->sk_state_change	=	sock_def_wakeup;
2871 	sk->sk_data_ready	=	sock_def_readable;
2872 	sk->sk_write_space	=	sock_def_write_space;
2873 	sk->sk_error_report	=	sock_def_error_report;
2874 	sk->sk_destruct		=	sock_def_destruct;
2875 
2876 	sk->sk_frag.page	=	NULL;
2877 	sk->sk_frag.offset	=	0;
2878 	sk->sk_peek_off		=	-1;
2879 
2880 	sk->sk_peer_pid 	=	NULL;
2881 	sk->sk_peer_cred	=	NULL;
2882 	sk->sk_write_pending	=	0;
2883 	sk->sk_rcvlowat		=	1;
2884 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
2885 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
2886 
2887 	sk->sk_stamp = SK_DEFAULT_STAMP;
2888 #if BITS_PER_LONG==32
2889 	seqlock_init(&sk->sk_stamp_seq);
2890 #endif
2891 	atomic_set(&sk->sk_zckey, 0);
2892 
2893 #ifdef CONFIG_NET_RX_BUSY_POLL
2894 	sk->sk_napi_id		=	0;
2895 	sk->sk_ll_usec		=	sysctl_net_busy_read;
2896 #endif
2897 
2898 	sk->sk_max_pacing_rate = ~0UL;
2899 	sk->sk_pacing_rate = ~0UL;
2900 	sk->sk_pacing_shift = 10;
2901 	sk->sk_incoming_cpu = -1;
2902 
2903 	sk_rx_queue_clear(sk);
2904 	/*
2905 	 * Before updating sk_refcnt, we must commit prior changes to memory
2906 	 * (Documentation/RCU/rculist_nulls.txt for details)
2907 	 */
2908 	smp_wmb();
2909 	refcount_set(&sk->sk_refcnt, 1);
2910 	atomic_set(&sk->sk_drops, 0);
2911 }
2912 EXPORT_SYMBOL(sock_init_data);
2913 
2914 void lock_sock_nested(struct sock *sk, int subclass)
2915 {
2916 	might_sleep();
2917 	spin_lock_bh(&sk->sk_lock.slock);
2918 	if (sk->sk_lock.owned)
2919 		__lock_sock(sk);
2920 	sk->sk_lock.owned = 1;
2921 	spin_unlock(&sk->sk_lock.slock);
2922 	/*
2923 	 * The sk_lock has mutex_lock() semantics here:
2924 	 */
2925 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2926 	local_bh_enable();
2927 }
2928 EXPORT_SYMBOL(lock_sock_nested);
2929 
2930 void release_sock(struct sock *sk)
2931 {
2932 	spin_lock_bh(&sk->sk_lock.slock);
2933 	if (sk->sk_backlog.tail)
2934 		__release_sock(sk);
2935 
2936 	/* Warning : release_cb() might need to release sk ownership,
2937 	 * ie call sock_release_ownership(sk) before us.
2938 	 */
2939 	if (sk->sk_prot->release_cb)
2940 		sk->sk_prot->release_cb(sk);
2941 
2942 	sock_release_ownership(sk);
2943 	if (waitqueue_active(&sk->sk_lock.wq))
2944 		wake_up(&sk->sk_lock.wq);
2945 	spin_unlock_bh(&sk->sk_lock.slock);
2946 }
2947 EXPORT_SYMBOL(release_sock);
2948 
2949 /**
2950  * lock_sock_fast - fast version of lock_sock
2951  * @sk: socket
2952  *
2953  * This version should be used for very small section, where process wont block
2954  * return false if fast path is taken:
2955  *
2956  *   sk_lock.slock locked, owned = 0, BH disabled
2957  *
2958  * return true if slow path is taken:
2959  *
2960  *   sk_lock.slock unlocked, owned = 1, BH enabled
2961  */
2962 bool lock_sock_fast(struct sock *sk)
2963 {
2964 	might_sleep();
2965 	spin_lock_bh(&sk->sk_lock.slock);
2966 
2967 	if (!sk->sk_lock.owned)
2968 		/*
2969 		 * Note : We must disable BH
2970 		 */
2971 		return false;
2972 
2973 	__lock_sock(sk);
2974 	sk->sk_lock.owned = 1;
2975 	spin_unlock(&sk->sk_lock.slock);
2976 	/*
2977 	 * The sk_lock has mutex_lock() semantics here:
2978 	 */
2979 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2980 	local_bh_enable();
2981 	return true;
2982 }
2983 EXPORT_SYMBOL(lock_sock_fast);
2984 
2985 int sock_gettstamp(struct socket *sock, void __user *userstamp,
2986 		   bool timeval, bool time32)
2987 {
2988 	struct sock *sk = sock->sk;
2989 	struct timespec64 ts;
2990 
2991 	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2992 	ts = ktime_to_timespec64(sock_read_timestamp(sk));
2993 	if (ts.tv_sec == -1)
2994 		return -ENOENT;
2995 	if (ts.tv_sec == 0) {
2996 		ktime_t kt = ktime_get_real();
2997 		sock_write_timestamp(sk, kt);;
2998 		ts = ktime_to_timespec64(kt);
2999 	}
3000 
3001 	if (timeval)
3002 		ts.tv_nsec /= 1000;
3003 
3004 #ifdef CONFIG_COMPAT_32BIT_TIME
3005 	if (time32)
3006 		return put_old_timespec32(&ts, userstamp);
3007 #endif
3008 #ifdef CONFIG_SPARC64
3009 	/* beware of padding in sparc64 timeval */
3010 	if (timeval && !in_compat_syscall()) {
3011 		struct __kernel_old_timeval __user tv = {
3012 			.tv_sec = ts.tv_sec,
3013 			.tv_usec = ts.tv_nsec,
3014 		};
3015 		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3016 			return -EFAULT;
3017 		return 0;
3018 	}
3019 #endif
3020 	return put_timespec64(&ts, userstamp);
3021 }
3022 EXPORT_SYMBOL(sock_gettstamp);
3023 
3024 void sock_enable_timestamp(struct sock *sk, int flag)
3025 {
3026 	if (!sock_flag(sk, flag)) {
3027 		unsigned long previous_flags = sk->sk_flags;
3028 
3029 		sock_set_flag(sk, flag);
3030 		/*
3031 		 * we just set one of the two flags which require net
3032 		 * time stamping, but time stamping might have been on
3033 		 * already because of the other one
3034 		 */
3035 		if (sock_needs_netstamp(sk) &&
3036 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3037 			net_enable_timestamp();
3038 	}
3039 }
3040 
3041 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3042 		       int level, int type)
3043 {
3044 	struct sock_exterr_skb *serr;
3045 	struct sk_buff *skb;
3046 	int copied, err;
3047 
3048 	err = -EAGAIN;
3049 	skb = sock_dequeue_err_skb(sk);
3050 	if (skb == NULL)
3051 		goto out;
3052 
3053 	copied = skb->len;
3054 	if (copied > len) {
3055 		msg->msg_flags |= MSG_TRUNC;
3056 		copied = len;
3057 	}
3058 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3059 	if (err)
3060 		goto out_free_skb;
3061 
3062 	sock_recv_timestamp(msg, sk, skb);
3063 
3064 	serr = SKB_EXT_ERR(skb);
3065 	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3066 
3067 	msg->msg_flags |= MSG_ERRQUEUE;
3068 	err = copied;
3069 
3070 out_free_skb:
3071 	kfree_skb(skb);
3072 out:
3073 	return err;
3074 }
3075 EXPORT_SYMBOL(sock_recv_errqueue);
3076 
3077 /*
3078  *	Get a socket option on an socket.
3079  *
3080  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3081  *	asynchronous errors should be reported by getsockopt. We assume
3082  *	this means if you specify SO_ERROR (otherwise whats the point of it).
3083  */
3084 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3085 			   char __user *optval, int __user *optlen)
3086 {
3087 	struct sock *sk = sock->sk;
3088 
3089 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3090 }
3091 EXPORT_SYMBOL(sock_common_getsockopt);
3092 
3093 #ifdef CONFIG_COMPAT
3094 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3095 				  char __user *optval, int __user *optlen)
3096 {
3097 	struct sock *sk = sock->sk;
3098 
3099 	if (sk->sk_prot->compat_getsockopt != NULL)
3100 		return sk->sk_prot->compat_getsockopt(sk, level, optname,
3101 						      optval, optlen);
3102 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3103 }
3104 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3105 #endif
3106 
3107 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3108 			int flags)
3109 {
3110 	struct sock *sk = sock->sk;
3111 	int addr_len = 0;
3112 	int err;
3113 
3114 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3115 				   flags & ~MSG_DONTWAIT, &addr_len);
3116 	if (err >= 0)
3117 		msg->msg_namelen = addr_len;
3118 	return err;
3119 }
3120 EXPORT_SYMBOL(sock_common_recvmsg);
3121 
3122 /*
3123  *	Set socket options on an inet socket.
3124  */
3125 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3126 			   char __user *optval, unsigned int optlen)
3127 {
3128 	struct sock *sk = sock->sk;
3129 
3130 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3131 }
3132 EXPORT_SYMBOL(sock_common_setsockopt);
3133 
3134 #ifdef CONFIG_COMPAT
3135 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3136 				  char __user *optval, unsigned int optlen)
3137 {
3138 	struct sock *sk = sock->sk;
3139 
3140 	if (sk->sk_prot->compat_setsockopt != NULL)
3141 		return sk->sk_prot->compat_setsockopt(sk, level, optname,
3142 						      optval, optlen);
3143 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3144 }
3145 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3146 #endif
3147 
3148 void sk_common_release(struct sock *sk)
3149 {
3150 	if (sk->sk_prot->destroy)
3151 		sk->sk_prot->destroy(sk);
3152 
3153 	/*
3154 	 * Observation: when sock_common_release is called, processes have
3155 	 * no access to socket. But net still has.
3156 	 * Step one, detach it from networking:
3157 	 *
3158 	 * A. Remove from hash tables.
3159 	 */
3160 
3161 	sk->sk_prot->unhash(sk);
3162 
3163 	/*
3164 	 * In this point socket cannot receive new packets, but it is possible
3165 	 * that some packets are in flight because some CPU runs receiver and
3166 	 * did hash table lookup before we unhashed socket. They will achieve
3167 	 * receive queue and will be purged by socket destructor.
3168 	 *
3169 	 * Also we still have packets pending on receive queue and probably,
3170 	 * our own packets waiting in device queues. sock_destroy will drain
3171 	 * receive queue, but transmitted packets will delay socket destruction
3172 	 * until the last reference will be released.
3173 	 */
3174 
3175 	sock_orphan(sk);
3176 
3177 	xfrm_sk_free_policy(sk);
3178 
3179 	sk_refcnt_debug_release(sk);
3180 
3181 	sock_put(sk);
3182 }
3183 EXPORT_SYMBOL(sk_common_release);
3184 
3185 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3186 {
3187 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3188 
3189 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3190 	mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3191 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3192 	mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3193 	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3194 	mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3195 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3196 	mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3197 	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3198 }
3199 
3200 #ifdef CONFIG_PROC_FS
3201 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
3202 struct prot_inuse {
3203 	int val[PROTO_INUSE_NR];
3204 };
3205 
3206 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3207 
3208 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3209 {
3210 	__this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3211 }
3212 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3213 
3214 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3215 {
3216 	int cpu, idx = prot->inuse_idx;
3217 	int res = 0;
3218 
3219 	for_each_possible_cpu(cpu)
3220 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3221 
3222 	return res >= 0 ? res : 0;
3223 }
3224 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3225 
3226 static void sock_inuse_add(struct net *net, int val)
3227 {
3228 	this_cpu_add(*net->core.sock_inuse, val);
3229 }
3230 
3231 int sock_inuse_get(struct net *net)
3232 {
3233 	int cpu, res = 0;
3234 
3235 	for_each_possible_cpu(cpu)
3236 		res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3237 
3238 	return res;
3239 }
3240 
3241 EXPORT_SYMBOL_GPL(sock_inuse_get);
3242 
3243 static int __net_init sock_inuse_init_net(struct net *net)
3244 {
3245 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3246 	if (net->core.prot_inuse == NULL)
3247 		return -ENOMEM;
3248 
3249 	net->core.sock_inuse = alloc_percpu(int);
3250 	if (net->core.sock_inuse == NULL)
3251 		goto out;
3252 
3253 	return 0;
3254 
3255 out:
3256 	free_percpu(net->core.prot_inuse);
3257 	return -ENOMEM;
3258 }
3259 
3260 static void __net_exit sock_inuse_exit_net(struct net *net)
3261 {
3262 	free_percpu(net->core.prot_inuse);
3263 	free_percpu(net->core.sock_inuse);
3264 }
3265 
3266 static struct pernet_operations net_inuse_ops = {
3267 	.init = sock_inuse_init_net,
3268 	.exit = sock_inuse_exit_net,
3269 };
3270 
3271 static __init int net_inuse_init(void)
3272 {
3273 	if (register_pernet_subsys(&net_inuse_ops))
3274 		panic("Cannot initialize net inuse counters");
3275 
3276 	return 0;
3277 }
3278 
3279 core_initcall(net_inuse_init);
3280 
3281 static void assign_proto_idx(struct proto *prot)
3282 {
3283 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3284 
3285 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3286 		pr_err("PROTO_INUSE_NR exhausted\n");
3287 		return;
3288 	}
3289 
3290 	set_bit(prot->inuse_idx, proto_inuse_idx);
3291 }
3292 
3293 static void release_proto_idx(struct proto *prot)
3294 {
3295 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3296 		clear_bit(prot->inuse_idx, proto_inuse_idx);
3297 }
3298 #else
3299 static inline void assign_proto_idx(struct proto *prot)
3300 {
3301 }
3302 
3303 static inline void release_proto_idx(struct proto *prot)
3304 {
3305 }
3306 
3307 static void sock_inuse_add(struct net *net, int val)
3308 {
3309 }
3310 #endif
3311 
3312 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3313 {
3314 	if (!rsk_prot)
3315 		return;
3316 	kfree(rsk_prot->slab_name);
3317 	rsk_prot->slab_name = NULL;
3318 	kmem_cache_destroy(rsk_prot->slab);
3319 	rsk_prot->slab = NULL;
3320 }
3321 
3322 static int req_prot_init(const struct proto *prot)
3323 {
3324 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3325 
3326 	if (!rsk_prot)
3327 		return 0;
3328 
3329 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3330 					prot->name);
3331 	if (!rsk_prot->slab_name)
3332 		return -ENOMEM;
3333 
3334 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3335 					   rsk_prot->obj_size, 0,
3336 					   SLAB_ACCOUNT | prot->slab_flags,
3337 					   NULL);
3338 
3339 	if (!rsk_prot->slab) {
3340 		pr_crit("%s: Can't create request sock SLAB cache!\n",
3341 			prot->name);
3342 		return -ENOMEM;
3343 	}
3344 	return 0;
3345 }
3346 
3347 int proto_register(struct proto *prot, int alloc_slab)
3348 {
3349 	if (alloc_slab) {
3350 		prot->slab = kmem_cache_create_usercopy(prot->name,
3351 					prot->obj_size, 0,
3352 					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3353 					prot->slab_flags,
3354 					prot->useroffset, prot->usersize,
3355 					NULL);
3356 
3357 		if (prot->slab == NULL) {
3358 			pr_crit("%s: Can't create sock SLAB cache!\n",
3359 				prot->name);
3360 			goto out;
3361 		}
3362 
3363 		if (req_prot_init(prot))
3364 			goto out_free_request_sock_slab;
3365 
3366 		if (prot->twsk_prot != NULL) {
3367 			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3368 
3369 			if (prot->twsk_prot->twsk_slab_name == NULL)
3370 				goto out_free_request_sock_slab;
3371 
3372 			prot->twsk_prot->twsk_slab =
3373 				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3374 						  prot->twsk_prot->twsk_obj_size,
3375 						  0,
3376 						  SLAB_ACCOUNT |
3377 						  prot->slab_flags,
3378 						  NULL);
3379 			if (prot->twsk_prot->twsk_slab == NULL)
3380 				goto out_free_timewait_sock_slab_name;
3381 		}
3382 	}
3383 
3384 	mutex_lock(&proto_list_mutex);
3385 	list_add(&prot->node, &proto_list);
3386 	assign_proto_idx(prot);
3387 	mutex_unlock(&proto_list_mutex);
3388 	return 0;
3389 
3390 out_free_timewait_sock_slab_name:
3391 	kfree(prot->twsk_prot->twsk_slab_name);
3392 out_free_request_sock_slab:
3393 	req_prot_cleanup(prot->rsk_prot);
3394 
3395 	kmem_cache_destroy(prot->slab);
3396 	prot->slab = NULL;
3397 out:
3398 	return -ENOBUFS;
3399 }
3400 EXPORT_SYMBOL(proto_register);
3401 
3402 void proto_unregister(struct proto *prot)
3403 {
3404 	mutex_lock(&proto_list_mutex);
3405 	release_proto_idx(prot);
3406 	list_del(&prot->node);
3407 	mutex_unlock(&proto_list_mutex);
3408 
3409 	kmem_cache_destroy(prot->slab);
3410 	prot->slab = NULL;
3411 
3412 	req_prot_cleanup(prot->rsk_prot);
3413 
3414 	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3415 		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3416 		kfree(prot->twsk_prot->twsk_slab_name);
3417 		prot->twsk_prot->twsk_slab = NULL;
3418 	}
3419 }
3420 EXPORT_SYMBOL(proto_unregister);
3421 
3422 int sock_load_diag_module(int family, int protocol)
3423 {
3424 	if (!protocol) {
3425 		if (!sock_is_registered(family))
3426 			return -ENOENT;
3427 
3428 		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3429 				      NETLINK_SOCK_DIAG, family);
3430 	}
3431 
3432 #ifdef CONFIG_INET
3433 	if (family == AF_INET &&
3434 	    protocol != IPPROTO_RAW &&
3435 	    !rcu_access_pointer(inet_protos[protocol]))
3436 		return -ENOENT;
3437 #endif
3438 
3439 	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3440 			      NETLINK_SOCK_DIAG, family, protocol);
3441 }
3442 EXPORT_SYMBOL(sock_load_diag_module);
3443 
3444 #ifdef CONFIG_PROC_FS
3445 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3446 	__acquires(proto_list_mutex)
3447 {
3448 	mutex_lock(&proto_list_mutex);
3449 	return seq_list_start_head(&proto_list, *pos);
3450 }
3451 
3452 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3453 {
3454 	return seq_list_next(v, &proto_list, pos);
3455 }
3456 
3457 static void proto_seq_stop(struct seq_file *seq, void *v)
3458 	__releases(proto_list_mutex)
3459 {
3460 	mutex_unlock(&proto_list_mutex);
3461 }
3462 
3463 static char proto_method_implemented(const void *method)
3464 {
3465 	return method == NULL ? 'n' : 'y';
3466 }
3467 static long sock_prot_memory_allocated(struct proto *proto)
3468 {
3469 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3470 }
3471 
3472 static char *sock_prot_memory_pressure(struct proto *proto)
3473 {
3474 	return proto->memory_pressure != NULL ?
3475 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3476 }
3477 
3478 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3479 {
3480 
3481 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3482 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3483 		   proto->name,
3484 		   proto->obj_size,
3485 		   sock_prot_inuse_get(seq_file_net(seq), proto),
3486 		   sock_prot_memory_allocated(proto),
3487 		   sock_prot_memory_pressure(proto),
3488 		   proto->max_header,
3489 		   proto->slab == NULL ? "no" : "yes",
3490 		   module_name(proto->owner),
3491 		   proto_method_implemented(proto->close),
3492 		   proto_method_implemented(proto->connect),
3493 		   proto_method_implemented(proto->disconnect),
3494 		   proto_method_implemented(proto->accept),
3495 		   proto_method_implemented(proto->ioctl),
3496 		   proto_method_implemented(proto->init),
3497 		   proto_method_implemented(proto->destroy),
3498 		   proto_method_implemented(proto->shutdown),
3499 		   proto_method_implemented(proto->setsockopt),
3500 		   proto_method_implemented(proto->getsockopt),
3501 		   proto_method_implemented(proto->sendmsg),
3502 		   proto_method_implemented(proto->recvmsg),
3503 		   proto_method_implemented(proto->sendpage),
3504 		   proto_method_implemented(proto->bind),
3505 		   proto_method_implemented(proto->backlog_rcv),
3506 		   proto_method_implemented(proto->hash),
3507 		   proto_method_implemented(proto->unhash),
3508 		   proto_method_implemented(proto->get_port),
3509 		   proto_method_implemented(proto->enter_memory_pressure));
3510 }
3511 
3512 static int proto_seq_show(struct seq_file *seq, void *v)
3513 {
3514 	if (v == &proto_list)
3515 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3516 			   "protocol",
3517 			   "size",
3518 			   "sockets",
3519 			   "memory",
3520 			   "press",
3521 			   "maxhdr",
3522 			   "slab",
3523 			   "module",
3524 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3525 	else
3526 		proto_seq_printf(seq, list_entry(v, struct proto, node));
3527 	return 0;
3528 }
3529 
3530 static const struct seq_operations proto_seq_ops = {
3531 	.start  = proto_seq_start,
3532 	.next   = proto_seq_next,
3533 	.stop   = proto_seq_stop,
3534 	.show   = proto_seq_show,
3535 };
3536 
3537 static __net_init int proto_init_net(struct net *net)
3538 {
3539 	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3540 			sizeof(struct seq_net_private)))
3541 		return -ENOMEM;
3542 
3543 	return 0;
3544 }
3545 
3546 static __net_exit void proto_exit_net(struct net *net)
3547 {
3548 	remove_proc_entry("protocols", net->proc_net);
3549 }
3550 
3551 
3552 static __net_initdata struct pernet_operations proto_net_ops = {
3553 	.init = proto_init_net,
3554 	.exit = proto_exit_net,
3555 };
3556 
3557 static int __init proto_init(void)
3558 {
3559 	return register_pernet_subsys(&proto_net_ops);
3560 }
3561 
3562 subsys_initcall(proto_init);
3563 
3564 #endif /* PROC_FS */
3565 
3566 #ifdef CONFIG_NET_RX_BUSY_POLL
3567 bool sk_busy_loop_end(void *p, unsigned long start_time)
3568 {
3569 	struct sock *sk = p;
3570 
3571 	return !skb_queue_empty(&sk->sk_receive_queue) ||
3572 	       sk_busy_loop_timeout(sk, start_time);
3573 }
3574 EXPORT_SYMBOL(sk_busy_loop_end);
3575 #endif /* CONFIG_NET_RX_BUSY_POLL */
3576