xref: /kernel/linux/linux-5.10/net/core/sock.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
4 *		operating system.  INET is implemented using the  BSD Socket
5 *		interface as the means of communication with the user level.
6 *
7 *		Generic socket support routines. Memory allocators, socket lock/release
8 *		handler for protocols to use and generic option handler.
9 *
10 * Authors:	Ross Biro
11 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *		Florian La Roche, <flla@stud.uni-sb.de>
13 *		Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 *		Alan Cox	: 	Numerous verify_area() problems
17 *		Alan Cox	:	Connecting on a connecting socket
18 *					now returns an error for tcp.
19 *		Alan Cox	:	sock->protocol is set correctly.
20 *					and is not sometimes left as 0.
21 *		Alan Cox	:	connect handles icmp errors on a
22 *					connect properly. Unfortunately there
23 *					is a restart syscall nasty there. I
24 *					can't match BSD without hacking the C
25 *					library. Ideas urgently sought!
26 *		Alan Cox	:	Disallow bind() to addresses that are
27 *					not ours - especially broadcast ones!!
28 *		Alan Cox	:	Socket 1024 _IS_ ok for users. (fencepost)
29 *		Alan Cox	:	sock_wfree/sock_rfree don't destroy sockets,
30 *					instead they leave that for the DESTROY timer.
31 *		Alan Cox	:	Clean up error flag in accept
32 *		Alan Cox	:	TCP ack handling is buggy, the DESTROY timer
33 *					was buggy. Put a remove_sock() in the handler
34 *					for memory when we hit 0. Also altered the timer
35 *					code. The ACK stuff can wait and needs major
36 *					TCP layer surgery.
37 *		Alan Cox	:	Fixed TCP ack bug, removed remove sock
38 *					and fixed timer/inet_bh race.
39 *		Alan Cox	:	Added zapped flag for TCP
40 *		Alan Cox	:	Move kfree_skb into skbuff.c and tidied up surplus code
41 *		Alan Cox	:	for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 *		Alan Cox	:	kfree_s calls now are kfree_skbmem so we can track skb resources
43 *		Alan Cox	:	Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 *		Alan Cox	:	Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 *		Rick Sladkey	:	Relaxed UDP rules for matching packets.
46 *		C.E.Hawkins	:	IFF_PROMISC/SIOCGHWADDR support
47 *	Pauline Middelink	:	identd support
48 *		Alan Cox	:	Fixed connect() taking signals I think.
49 *		Alan Cox	:	SO_LINGER supported
50 *		Alan Cox	:	Error reporting fixes
51 *		Anonymous	:	inet_create tidied up (sk->reuse setting)
52 *		Alan Cox	:	inet sockets don't set sk->type!
53 *		Alan Cox	:	Split socket option code
54 *		Alan Cox	:	Callbacks
55 *		Alan Cox	:	Nagle flag for Charles & Johannes stuff
56 *		Alex		:	Removed restriction on inet fioctl
57 *		Alan Cox	:	Splitting INET from NET core
58 *		Alan Cox	:	Fixed bogus SO_TYPE handling in getsockopt()
59 *		Adam Caldwell	:	Missing return in SO_DONTROUTE/SO_DEBUG code
60 *		Alan Cox	:	Split IP from generic code
61 *		Alan Cox	:	New kfree_skbmem()
62 *		Alan Cox	:	Make SO_DEBUG superuser only.
63 *		Alan Cox	:	Allow anyone to clear SO_DEBUG
64 *					(compatibility fix)
65 *		Alan Cox	:	Added optimistic memory grabbing for AF_UNIX throughput.
66 *		Alan Cox	:	Allocator for a socket is settable.
67 *		Alan Cox	:	SO_ERROR includes soft errors.
68 *		Alan Cox	:	Allow NULL arguments on some SO_ opts
69 *		Alan Cox	: 	Generic socket allocation to make hooks
70 *					easier (suggested by Craig Metz).
71 *		Michael Pall	:	SO_ERROR returns positive errno again
72 *              Steve Whitehouse:       Added default destructor to free
73 *                                      protocol private data.
74 *              Steve Whitehouse:       Added various other default routines
75 *                                      common to several socket families.
76 *              Chris Evans     :       Call suser() check last on F_SETOWN
77 *		Jay Schulist	:	Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 *		Andi Kleen	:	Add sock_kmalloc()/sock_kfree_s()
79 *		Andi Kleen	:	Fix write_space callback
80 *		Chris Evans	:	Security fixes - signedness again
81 *		Arnaldo C. Melo :       cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88#include <asm/unaligned.h>
89#include <linux/capability.h>
90#include <linux/errno.h>
91#include <linux/errqueue.h>
92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
100#include <linux/sched/mm.h>
101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
110#include <linux/init.h>
111#include <linux/highmem.h>
112#include <linux/user_namespace.h>
113#include <linux/static_key.h>
114#include <linux/memcontrol.h>
115#include <linux/prefetch.h>
116#include <linux/compat.h>
117
118#include <linux/uaccess.h>
119
120#include <linux/netdevice.h>
121#include <net/protocol.h>
122#include <linux/skbuff.h>
123#include <net/net_namespace.h>
124#include <net/request_sock.h>
125#include <net/sock.h>
126#include <linux/net_tstamp.h>
127#include <net/xfrm.h>
128#include <linux/ipsec.h>
129#include <net/cls_cgroup.h>
130#include <net/netprio_cgroup.h>
131#include <linux/sock_diag.h>
132
133#include <linux/filter.h>
134#include <net/sock_reuseport.h>
135#include <net/bpf_sk_storage.h>
136
137#include <trace/events/sock.h>
138
139#include <net/tcp.h>
140#include <net/busy_poll.h>
141
142static DEFINE_MUTEX(proto_list_mutex);
143static LIST_HEAD(proto_list);
144
145static void sock_inuse_add(struct net *net, int val);
146
147/**
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
152 *
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
156 */
157bool sk_ns_capable(const struct sock *sk,
158		   struct user_namespace *user_ns, int cap)
159{
160	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161		ns_capable(user_ns, cap);
162}
163EXPORT_SYMBOL(sk_ns_capable);
164
165/**
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
169 *
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
172 * namespaces.
173 */
174bool sk_capable(const struct sock *sk, int cap)
175{
176	return sk_ns_capable(sk, &init_user_ns, cap);
177}
178EXPORT_SYMBOL(sk_capable);
179
180/**
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
184 *
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
188 */
189bool sk_net_capable(const struct sock *sk, int cap)
190{
191	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
192}
193EXPORT_SYMBOL(sk_net_capable);
194
195/*
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
198 * userspace sockets.
199 */
200static struct lock_class_key af_family_keys[AF_MAX];
201static struct lock_class_key af_family_kern_keys[AF_MAX];
202static struct lock_class_key af_family_slock_keys[AF_MAX];
203static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
204
205/*
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
208 * locks is fast):
209 */
210
211#define _sock_locks(x)						  \
212  x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
213  x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
214  x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
215  x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
216  x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
217  x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
218  x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
219  x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
220  x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
221  x "27"       ,	x "28"          ,	x "AF_CAN"      , \
222  x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
223  x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
224  x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
225  x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
226  x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
227  x "AF_MAX"
228
229static const char *const af_family_key_strings[AF_MAX+1] = {
230	_sock_locks("sk_lock-")
231};
232static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233	_sock_locks("slock-")
234};
235static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236	_sock_locks("clock-")
237};
238
239static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240	_sock_locks("k-sk_lock-")
241};
242static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243	_sock_locks("k-slock-")
244};
245static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246	_sock_locks("k-clock-")
247};
248static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249	_sock_locks("rlock-")
250};
251static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
252	_sock_locks("wlock-")
253};
254static const char *const af_family_elock_key_strings[AF_MAX+1] = {
255	_sock_locks("elock-")
256};
257
258/*
259 * sk_callback_lock and sk queues locking rules are per-address-family,
260 * so split the lock classes by using a per-AF key:
261 */
262static struct lock_class_key af_callback_keys[AF_MAX];
263static struct lock_class_key af_rlock_keys[AF_MAX];
264static struct lock_class_key af_wlock_keys[AF_MAX];
265static struct lock_class_key af_elock_keys[AF_MAX];
266static struct lock_class_key af_kern_callback_keys[AF_MAX];
267
268/* Run time adjustable parameters. */
269__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270EXPORT_SYMBOL(sysctl_wmem_max);
271__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272EXPORT_SYMBOL(sysctl_rmem_max);
273__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
275
276/* Maximal space eaten by iovec or ancillary data plus some space */
277int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278EXPORT_SYMBOL(sysctl_optmem_max);
279
280int sysctl_tstamp_allow_data __read_mostly = 1;
281
282DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
283EXPORT_SYMBOL_GPL(memalloc_socks_key);
284
285/**
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
288 *
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
292 */
293void sk_set_memalloc(struct sock *sk)
294{
295	sock_set_flag(sk, SOCK_MEMALLOC);
296	sk->sk_allocation |= __GFP_MEMALLOC;
297	static_branch_inc(&memalloc_socks_key);
298}
299EXPORT_SYMBOL_GPL(sk_set_memalloc);
300
301void sk_clear_memalloc(struct sock *sk)
302{
303	sock_reset_flag(sk, SOCK_MEMALLOC);
304	sk->sk_allocation &= ~__GFP_MEMALLOC;
305	static_branch_dec(&memalloc_socks_key);
306
307	/*
308	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309	 * progress of swapping. SOCK_MEMALLOC may be cleared while
310	 * it has rmem allocations due to the last swapfile being deactivated
311	 * but there is a risk that the socket is unusable due to exceeding
312	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
313	 */
314	sk_mem_reclaim(sk);
315}
316EXPORT_SYMBOL_GPL(sk_clear_memalloc);
317
318int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
319{
320	int ret;
321	unsigned int noreclaim_flag;
322
323	/* these should have been dropped before queueing */
324	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
325
326	noreclaim_flag = memalloc_noreclaim_save();
327	ret = sk->sk_backlog_rcv(sk, skb);
328	memalloc_noreclaim_restore(noreclaim_flag);
329
330	return ret;
331}
332EXPORT_SYMBOL(__sk_backlog_rcv);
333
334static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
335{
336	struct __kernel_sock_timeval tv;
337
338	if (timeo == MAX_SCHEDULE_TIMEOUT) {
339		tv.tv_sec = 0;
340		tv.tv_usec = 0;
341	} else {
342		tv.tv_sec = timeo / HZ;
343		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
344	}
345
346	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348		*(struct old_timeval32 *)optval = tv32;
349		return sizeof(tv32);
350	}
351
352	if (old_timeval) {
353		struct __kernel_old_timeval old_tv;
354		old_tv.tv_sec = tv.tv_sec;
355		old_tv.tv_usec = tv.tv_usec;
356		*(struct __kernel_old_timeval *)optval = old_tv;
357		return sizeof(old_tv);
358	}
359
360	*(struct __kernel_sock_timeval *)optval = tv;
361	return sizeof(tv);
362}
363
364static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
365			    bool old_timeval)
366{
367	struct __kernel_sock_timeval tv;
368
369	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
370		struct old_timeval32 tv32;
371
372		if (optlen < sizeof(tv32))
373			return -EINVAL;
374
375		if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
376			return -EFAULT;
377		tv.tv_sec = tv32.tv_sec;
378		tv.tv_usec = tv32.tv_usec;
379	} else if (old_timeval) {
380		struct __kernel_old_timeval old_tv;
381
382		if (optlen < sizeof(old_tv))
383			return -EINVAL;
384		if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
385			return -EFAULT;
386		tv.tv_sec = old_tv.tv_sec;
387		tv.tv_usec = old_tv.tv_usec;
388	} else {
389		if (optlen < sizeof(tv))
390			return -EINVAL;
391		if (copy_from_sockptr(&tv, optval, sizeof(tv)))
392			return -EFAULT;
393	}
394	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
395		return -EDOM;
396
397	if (tv.tv_sec < 0) {
398		static int warned __read_mostly;
399
400		*timeo_p = 0;
401		if (warned < 10 && net_ratelimit()) {
402			warned++;
403			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404				__func__, current->comm, task_pid_nr(current));
405		}
406		return 0;
407	}
408	*timeo_p = MAX_SCHEDULE_TIMEOUT;
409	if (tv.tv_sec == 0 && tv.tv_usec == 0)
410		return 0;
411	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
412		*timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
413	return 0;
414}
415
416static bool sock_needs_netstamp(const struct sock *sk)
417{
418	switch (sk->sk_family) {
419	case AF_UNSPEC:
420	case AF_UNIX:
421		return false;
422	default:
423		return true;
424	}
425}
426
427static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
428{
429	if (sk->sk_flags & flags) {
430		sk->sk_flags &= ~flags;
431		if (sock_needs_netstamp(sk) &&
432		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
433			net_disable_timestamp();
434	}
435}
436
437
438int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
439{
440	unsigned long flags;
441	struct sk_buff_head *list = &sk->sk_receive_queue;
442
443	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
444		atomic_inc(&sk->sk_drops);
445		trace_sock_rcvqueue_full(sk, skb);
446		return -ENOMEM;
447	}
448
449	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
450		atomic_inc(&sk->sk_drops);
451		return -ENOBUFS;
452	}
453
454	skb->dev = NULL;
455	skb_set_owner_r(skb, sk);
456
457	/* we escape from rcu protected region, make sure we dont leak
458	 * a norefcounted dst
459	 */
460	skb_dst_force(skb);
461
462	spin_lock_irqsave(&list->lock, flags);
463	sock_skb_set_dropcount(sk, skb);
464	__skb_queue_tail(list, skb);
465	spin_unlock_irqrestore(&list->lock, flags);
466
467	if (!sock_flag(sk, SOCK_DEAD))
468		sk->sk_data_ready(sk);
469	return 0;
470}
471EXPORT_SYMBOL(__sock_queue_rcv_skb);
472
473int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
474{
475	int err;
476
477	err = sk_filter(sk, skb);
478	if (err)
479		return err;
480
481	return __sock_queue_rcv_skb(sk, skb);
482}
483EXPORT_SYMBOL(sock_queue_rcv_skb);
484
485int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
486		     const int nested, unsigned int trim_cap, bool refcounted)
487{
488	int rc = NET_RX_SUCCESS;
489
490	if (sk_filter_trim_cap(sk, skb, trim_cap))
491		goto discard_and_relse;
492
493	skb->dev = NULL;
494
495	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
496		atomic_inc(&sk->sk_drops);
497		goto discard_and_relse;
498	}
499	if (nested)
500		bh_lock_sock_nested(sk);
501	else
502		bh_lock_sock(sk);
503	if (!sock_owned_by_user(sk)) {
504		/*
505		 * trylock + unlock semantics:
506		 */
507		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
508
509		rc = sk_backlog_rcv(sk, skb);
510
511		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
512	} else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
513		bh_unlock_sock(sk);
514		atomic_inc(&sk->sk_drops);
515		goto discard_and_relse;
516	}
517
518	bh_unlock_sock(sk);
519out:
520	if (refcounted)
521		sock_put(sk);
522	return rc;
523discard_and_relse:
524	kfree_skb(skb);
525	goto out;
526}
527EXPORT_SYMBOL(__sk_receive_skb);
528
529struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
530{
531	struct dst_entry *dst = __sk_dst_get(sk);
532
533	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
534		sk_tx_queue_clear(sk);
535		WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
536		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
537		dst_release(dst);
538		return NULL;
539	}
540
541	return dst;
542}
543EXPORT_SYMBOL(__sk_dst_check);
544
545struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
546{
547	struct dst_entry *dst = sk_dst_get(sk);
548
549	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
550		sk_dst_reset(sk);
551		dst_release(dst);
552		return NULL;
553	}
554
555	return dst;
556}
557EXPORT_SYMBOL(sk_dst_check);
558
559static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
560{
561	int ret = -ENOPROTOOPT;
562#ifdef CONFIG_NETDEVICES
563	struct net *net = sock_net(sk);
564
565	/* Sorry... */
566	ret = -EPERM;
567	if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
568		goto out;
569
570	ret = -EINVAL;
571	if (ifindex < 0)
572		goto out;
573
574	sk->sk_bound_dev_if = ifindex;
575	if (sk->sk_prot->rehash)
576		sk->sk_prot->rehash(sk);
577	sk_dst_reset(sk);
578
579	ret = 0;
580
581out:
582#endif
583
584	return ret;
585}
586
587int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
588{
589	int ret;
590
591	if (lock_sk)
592		lock_sock(sk);
593	ret = sock_bindtoindex_locked(sk, ifindex);
594	if (lock_sk)
595		release_sock(sk);
596
597	return ret;
598}
599EXPORT_SYMBOL(sock_bindtoindex);
600
601static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
602{
603	int ret = -ENOPROTOOPT;
604#ifdef CONFIG_NETDEVICES
605	struct net *net = sock_net(sk);
606	char devname[IFNAMSIZ];
607	int index;
608
609	ret = -EINVAL;
610	if (optlen < 0)
611		goto out;
612
613	/* Bind this socket to a particular device like "eth0",
614	 * as specified in the passed interface name. If the
615	 * name is "" or the option length is zero the socket
616	 * is not bound.
617	 */
618	if (optlen > IFNAMSIZ - 1)
619		optlen = IFNAMSIZ - 1;
620	memset(devname, 0, sizeof(devname));
621
622	ret = -EFAULT;
623	if (copy_from_sockptr(devname, optval, optlen))
624		goto out;
625
626	index = 0;
627	if (devname[0] != '\0') {
628		struct net_device *dev;
629
630		rcu_read_lock();
631		dev = dev_get_by_name_rcu(net, devname);
632		if (dev)
633			index = dev->ifindex;
634		rcu_read_unlock();
635		ret = -ENODEV;
636		if (!dev)
637			goto out;
638	}
639
640	return sock_bindtoindex(sk, index, true);
641out:
642#endif
643
644	return ret;
645}
646
647static int sock_getbindtodevice(struct sock *sk, char __user *optval,
648				int __user *optlen, int len)
649{
650	int ret = -ENOPROTOOPT;
651#ifdef CONFIG_NETDEVICES
652	struct net *net = sock_net(sk);
653	char devname[IFNAMSIZ];
654
655	if (sk->sk_bound_dev_if == 0) {
656		len = 0;
657		goto zero;
658	}
659
660	ret = -EINVAL;
661	if (len < IFNAMSIZ)
662		goto out;
663
664	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
665	if (ret)
666		goto out;
667
668	len = strlen(devname) + 1;
669
670	ret = -EFAULT;
671	if (copy_to_user(optval, devname, len))
672		goto out;
673
674zero:
675	ret = -EFAULT;
676	if (put_user(len, optlen))
677		goto out;
678
679	ret = 0;
680
681out:
682#endif
683
684	return ret;
685}
686
687bool sk_mc_loop(struct sock *sk)
688{
689	if (dev_recursion_level())
690		return false;
691	if (!sk)
692		return true;
693	/* IPV6_ADDRFORM can change sk->sk_family under us. */
694	switch (READ_ONCE(sk->sk_family)) {
695	case AF_INET:
696		return inet_sk(sk)->mc_loop;
697#if IS_ENABLED(CONFIG_IPV6)
698	case AF_INET6:
699		return inet6_sk(sk)->mc_loop;
700#endif
701	}
702	WARN_ON_ONCE(1);
703	return true;
704}
705EXPORT_SYMBOL(sk_mc_loop);
706
707void sock_set_reuseaddr(struct sock *sk)
708{
709	lock_sock(sk);
710	sk->sk_reuse = SK_CAN_REUSE;
711	release_sock(sk);
712}
713EXPORT_SYMBOL(sock_set_reuseaddr);
714
715void sock_set_reuseport(struct sock *sk)
716{
717	lock_sock(sk);
718	sk->sk_reuseport = true;
719	release_sock(sk);
720}
721EXPORT_SYMBOL(sock_set_reuseport);
722
723void sock_no_linger(struct sock *sk)
724{
725	lock_sock(sk);
726	sk->sk_lingertime = 0;
727	sock_set_flag(sk, SOCK_LINGER);
728	release_sock(sk);
729}
730EXPORT_SYMBOL(sock_no_linger);
731
732void sock_set_priority(struct sock *sk, u32 priority)
733{
734	lock_sock(sk);
735	sk->sk_priority = priority;
736	release_sock(sk);
737}
738EXPORT_SYMBOL(sock_set_priority);
739
740void sock_set_sndtimeo(struct sock *sk, s64 secs)
741{
742	lock_sock(sk);
743	if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
744		sk->sk_sndtimeo = secs * HZ;
745	else
746		sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
747	release_sock(sk);
748}
749EXPORT_SYMBOL(sock_set_sndtimeo);
750
751static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
752{
753	if (val)  {
754		sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
755		sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
756		sock_set_flag(sk, SOCK_RCVTSTAMP);
757		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
758	} else {
759		sock_reset_flag(sk, SOCK_RCVTSTAMP);
760		sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
761	}
762}
763
764void sock_enable_timestamps(struct sock *sk)
765{
766	lock_sock(sk);
767	__sock_set_timestamps(sk, true, false, true);
768	release_sock(sk);
769}
770EXPORT_SYMBOL(sock_enable_timestamps);
771
772void sock_set_keepalive(struct sock *sk)
773{
774	lock_sock(sk);
775	if (sk->sk_prot->keepalive)
776		sk->sk_prot->keepalive(sk, true);
777	sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
778	release_sock(sk);
779}
780EXPORT_SYMBOL(sock_set_keepalive);
781
782static void __sock_set_rcvbuf(struct sock *sk, int val)
783{
784	/* Ensure val * 2 fits into an int, to prevent max_t() from treating it
785	 * as a negative value.
786	 */
787	val = min_t(int, val, INT_MAX / 2);
788	sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
789
790	/* We double it on the way in to account for "struct sk_buff" etc.
791	 * overhead.   Applications assume that the SO_RCVBUF setting they make
792	 * will allow that much actual data to be received on that socket.
793	 *
794	 * Applications are unaware that "struct sk_buff" and other overheads
795	 * allocate from the receive buffer during socket buffer allocation.
796	 *
797	 * And after considering the possible alternatives, returning the value
798	 * we actually used in getsockopt is the most desirable behavior.
799	 */
800	WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
801}
802
803void sock_set_rcvbuf(struct sock *sk, int val)
804{
805	lock_sock(sk);
806	__sock_set_rcvbuf(sk, val);
807	release_sock(sk);
808}
809EXPORT_SYMBOL(sock_set_rcvbuf);
810
811static void __sock_set_mark(struct sock *sk, u32 val)
812{
813	if (val != sk->sk_mark) {
814		sk->sk_mark = val;
815		sk_dst_reset(sk);
816	}
817}
818
819void sock_set_mark(struct sock *sk, u32 val)
820{
821	lock_sock(sk);
822	__sock_set_mark(sk, val);
823	release_sock(sk);
824}
825EXPORT_SYMBOL(sock_set_mark);
826
827/*
828 *	This is meant for all protocols to use and covers goings on
829 *	at the socket level. Everything here is generic.
830 */
831
832int sock_setsockopt(struct socket *sock, int level, int optname,
833		    sockptr_t optval, unsigned int optlen)
834{
835	struct sock_txtime sk_txtime;
836	struct sock *sk = sock->sk;
837	int val;
838	int valbool;
839	struct linger ling;
840	int ret = 0;
841
842	/*
843	 *	Options without arguments
844	 */
845
846	if (optname == SO_BINDTODEVICE)
847		return sock_setbindtodevice(sk, optval, optlen);
848
849	if (optlen < sizeof(int))
850		return -EINVAL;
851
852	if (copy_from_sockptr(&val, optval, sizeof(val)))
853		return -EFAULT;
854
855	valbool = val ? 1 : 0;
856
857	lock_sock(sk);
858
859	switch (optname) {
860	case SO_DEBUG:
861		if (val && !capable(CAP_NET_ADMIN))
862			ret = -EACCES;
863		else
864			sock_valbool_flag(sk, SOCK_DBG, valbool);
865		break;
866	case SO_REUSEADDR:
867		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
868		break;
869	case SO_REUSEPORT:
870		sk->sk_reuseport = valbool;
871		break;
872	case SO_TYPE:
873	case SO_PROTOCOL:
874	case SO_DOMAIN:
875	case SO_ERROR:
876		ret = -ENOPROTOOPT;
877		break;
878	case SO_DONTROUTE:
879		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
880		sk_dst_reset(sk);
881		break;
882	case SO_BROADCAST:
883		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
884		break;
885	case SO_SNDBUF:
886		/* Don't error on this BSD doesn't and if you think
887		 * about it this is right. Otherwise apps have to
888		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
889		 * are treated in BSD as hints
890		 */
891		val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
892set_sndbuf:
893		/* Ensure val * 2 fits into an int, to prevent max_t()
894		 * from treating it as a negative value.
895		 */
896		val = min_t(int, val, INT_MAX / 2);
897		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
898		WRITE_ONCE(sk->sk_sndbuf,
899			   max_t(int, val * 2, SOCK_MIN_SNDBUF));
900		/* Wake up sending tasks if we upped the value. */
901		sk->sk_write_space(sk);
902		break;
903
904	case SO_SNDBUFFORCE:
905		if (!capable(CAP_NET_ADMIN)) {
906			ret = -EPERM;
907			break;
908		}
909
910		/* No negative values (to prevent underflow, as val will be
911		 * multiplied by 2).
912		 */
913		if (val < 0)
914			val = 0;
915		goto set_sndbuf;
916
917	case SO_RCVBUF:
918		/* Don't error on this BSD doesn't and if you think
919		 * about it this is right. Otherwise apps have to
920		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
921		 * are treated in BSD as hints
922		 */
923		__sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
924		break;
925
926	case SO_RCVBUFFORCE:
927		if (!capable(CAP_NET_ADMIN)) {
928			ret = -EPERM;
929			break;
930		}
931
932		/* No negative values (to prevent underflow, as val will be
933		 * multiplied by 2).
934		 */
935		__sock_set_rcvbuf(sk, max(val, 0));
936		break;
937
938	case SO_KEEPALIVE:
939		if (sk->sk_prot->keepalive)
940			sk->sk_prot->keepalive(sk, valbool);
941		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
942		break;
943
944	case SO_OOBINLINE:
945		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
946		break;
947
948	case SO_NO_CHECK:
949		sk->sk_no_check_tx = valbool;
950		break;
951
952	case SO_PRIORITY:
953		if ((val >= 0 && val <= 6) ||
954		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
955			sk->sk_priority = val;
956		else
957			ret = -EPERM;
958		break;
959
960	case SO_LINGER:
961		if (optlen < sizeof(ling)) {
962			ret = -EINVAL;	/* 1003.1g */
963			break;
964		}
965		if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
966			ret = -EFAULT;
967			break;
968		}
969		if (!ling.l_onoff)
970			sock_reset_flag(sk, SOCK_LINGER);
971		else {
972#if (BITS_PER_LONG == 32)
973			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
974				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
975			else
976#endif
977				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
978			sock_set_flag(sk, SOCK_LINGER);
979		}
980		break;
981
982	case SO_BSDCOMPAT:
983		break;
984
985	case SO_PASSCRED:
986		if (valbool)
987			set_bit(SOCK_PASSCRED, &sock->flags);
988		else
989			clear_bit(SOCK_PASSCRED, &sock->flags);
990		break;
991
992	case SO_TIMESTAMP_OLD:
993		__sock_set_timestamps(sk, valbool, false, false);
994		break;
995	case SO_TIMESTAMP_NEW:
996		__sock_set_timestamps(sk, valbool, true, false);
997		break;
998	case SO_TIMESTAMPNS_OLD:
999		__sock_set_timestamps(sk, valbool, false, true);
1000		break;
1001	case SO_TIMESTAMPNS_NEW:
1002		__sock_set_timestamps(sk, valbool, true, true);
1003		break;
1004	case SO_TIMESTAMPING_NEW:
1005	case SO_TIMESTAMPING_OLD:
1006		if (val & ~SOF_TIMESTAMPING_MASK) {
1007			ret = -EINVAL;
1008			break;
1009		}
1010
1011		if (val & SOF_TIMESTAMPING_OPT_ID &&
1012		    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
1013			if (sk->sk_protocol == IPPROTO_TCP &&
1014			    sk->sk_type == SOCK_STREAM) {
1015				if ((1 << sk->sk_state) &
1016				    (TCPF_CLOSE | TCPF_LISTEN)) {
1017					ret = -EINVAL;
1018					break;
1019				}
1020				sk->sk_tskey = tcp_sk(sk)->snd_una;
1021			} else {
1022				sk->sk_tskey = 0;
1023			}
1024		}
1025
1026		if (val & SOF_TIMESTAMPING_OPT_STATS &&
1027		    !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1028			ret = -EINVAL;
1029			break;
1030		}
1031
1032		sk->sk_tsflags = val;
1033		sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
1034
1035		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1036			sock_enable_timestamp(sk,
1037					      SOCK_TIMESTAMPING_RX_SOFTWARE);
1038		else
1039			sock_disable_timestamp(sk,
1040					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1041		break;
1042
1043	case SO_RCVLOWAT:
1044		if (val < 0)
1045			val = INT_MAX;
1046		if (sock->ops->set_rcvlowat)
1047			ret = sock->ops->set_rcvlowat(sk, val);
1048		else
1049			WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1050		break;
1051
1052	case SO_RCVTIMEO_OLD:
1053	case SO_RCVTIMEO_NEW:
1054		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1055				       optlen, optname == SO_RCVTIMEO_OLD);
1056		break;
1057
1058	case SO_SNDTIMEO_OLD:
1059	case SO_SNDTIMEO_NEW:
1060		ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1061				       optlen, optname == SO_SNDTIMEO_OLD);
1062		break;
1063
1064	case SO_ATTACH_FILTER: {
1065		struct sock_fprog fprog;
1066
1067		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1068		if (!ret)
1069			ret = sk_attach_filter(&fprog, sk);
1070		break;
1071	}
1072	case SO_ATTACH_BPF:
1073		ret = -EINVAL;
1074		if (optlen == sizeof(u32)) {
1075			u32 ufd;
1076
1077			ret = -EFAULT;
1078			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1079				break;
1080
1081			ret = sk_attach_bpf(ufd, sk);
1082		}
1083		break;
1084
1085	case SO_ATTACH_REUSEPORT_CBPF: {
1086		struct sock_fprog fprog;
1087
1088		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1089		if (!ret)
1090			ret = sk_reuseport_attach_filter(&fprog, sk);
1091		break;
1092	}
1093	case SO_ATTACH_REUSEPORT_EBPF:
1094		ret = -EINVAL;
1095		if (optlen == sizeof(u32)) {
1096			u32 ufd;
1097
1098			ret = -EFAULT;
1099			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1100				break;
1101
1102			ret = sk_reuseport_attach_bpf(ufd, sk);
1103		}
1104		break;
1105
1106	case SO_DETACH_REUSEPORT_BPF:
1107		ret = reuseport_detach_prog(sk);
1108		break;
1109
1110	case SO_DETACH_FILTER:
1111		ret = sk_detach_filter(sk);
1112		break;
1113
1114	case SO_LOCK_FILTER:
1115		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1116			ret = -EPERM;
1117		else
1118			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1119		break;
1120
1121	case SO_PASSSEC:
1122		if (valbool)
1123			set_bit(SOCK_PASSSEC, &sock->flags);
1124		else
1125			clear_bit(SOCK_PASSSEC, &sock->flags);
1126		break;
1127	case SO_MARK:
1128		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1129			ret = -EPERM;
1130			break;
1131		}
1132
1133		__sock_set_mark(sk, val);
1134		break;
1135
1136	case SO_RXQ_OVFL:
1137		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1138		break;
1139
1140	case SO_WIFI_STATUS:
1141		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1142		break;
1143
1144	case SO_PEEK_OFF:
1145		if (sock->ops->set_peek_off)
1146			ret = sock->ops->set_peek_off(sk, val);
1147		else
1148			ret = -EOPNOTSUPP;
1149		break;
1150
1151	case SO_NOFCS:
1152		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1153		break;
1154
1155	case SO_SELECT_ERR_QUEUE:
1156		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1157		break;
1158
1159#ifdef CONFIG_NET_RX_BUSY_POLL
1160	case SO_BUSY_POLL:
1161		/* allow unprivileged users to decrease the value */
1162		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1163			ret = -EPERM;
1164		else {
1165			if (val < 0)
1166				ret = -EINVAL;
1167			else
1168				WRITE_ONCE(sk->sk_ll_usec, val);
1169		}
1170		break;
1171#endif
1172
1173	case SO_MAX_PACING_RATE:
1174		{
1175		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1176
1177		if (sizeof(ulval) != sizeof(val) &&
1178		    optlen >= sizeof(ulval) &&
1179		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1180			ret = -EFAULT;
1181			break;
1182		}
1183		if (ulval != ~0UL)
1184			cmpxchg(&sk->sk_pacing_status,
1185				SK_PACING_NONE,
1186				SK_PACING_NEEDED);
1187		/* Pairs with READ_ONCE() from sk_getsockopt() */
1188		WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1189		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1190		break;
1191		}
1192	case SO_INCOMING_CPU:
1193		WRITE_ONCE(sk->sk_incoming_cpu, val);
1194		break;
1195
1196	case SO_CNX_ADVICE:
1197		if (val == 1)
1198			dst_negative_advice(sk);
1199		break;
1200
1201	case SO_ZEROCOPY:
1202		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1203			if (!((sk->sk_type == SOCK_STREAM &&
1204			       sk->sk_protocol == IPPROTO_TCP) ||
1205			      (sk->sk_type == SOCK_DGRAM &&
1206			       sk->sk_protocol == IPPROTO_UDP)))
1207				ret = -ENOTSUPP;
1208		} else if (sk->sk_family != PF_RDS) {
1209			ret = -ENOTSUPP;
1210		}
1211		if (!ret) {
1212			if (val < 0 || val > 1)
1213				ret = -EINVAL;
1214			else
1215				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1216		}
1217		break;
1218
1219	case SO_TXTIME:
1220		if (optlen != sizeof(struct sock_txtime)) {
1221			ret = -EINVAL;
1222			break;
1223		} else if (copy_from_sockptr(&sk_txtime, optval,
1224			   sizeof(struct sock_txtime))) {
1225			ret = -EFAULT;
1226			break;
1227		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1228			ret = -EINVAL;
1229			break;
1230		}
1231		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1232		 * scheduler has enough safe guards.
1233		 */
1234		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1235		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1236			ret = -EPERM;
1237			break;
1238		}
1239		sock_valbool_flag(sk, SOCK_TXTIME, true);
1240		sk->sk_clockid = sk_txtime.clockid;
1241		sk->sk_txtime_deadline_mode =
1242			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1243		sk->sk_txtime_report_errors =
1244			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1245		break;
1246
1247	case SO_BINDTOIFINDEX:
1248		ret = sock_bindtoindex_locked(sk, val);
1249		break;
1250
1251	default:
1252		ret = -ENOPROTOOPT;
1253		break;
1254	}
1255	release_sock(sk);
1256	return ret;
1257}
1258EXPORT_SYMBOL(sock_setsockopt);
1259
1260static const struct cred *sk_get_peer_cred(struct sock *sk)
1261{
1262	const struct cred *cred;
1263
1264	spin_lock(&sk->sk_peer_lock);
1265	cred = get_cred(sk->sk_peer_cred);
1266	spin_unlock(&sk->sk_peer_lock);
1267
1268	return cred;
1269}
1270
1271static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1272			  struct ucred *ucred)
1273{
1274	ucred->pid = pid_vnr(pid);
1275	ucred->uid = ucred->gid = -1;
1276	if (cred) {
1277		struct user_namespace *current_ns = current_user_ns();
1278
1279		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1280		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1281	}
1282}
1283
1284static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1285{
1286	struct user_namespace *user_ns = current_user_ns();
1287	int i;
1288
1289	for (i = 0; i < src->ngroups; i++)
1290		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1291			return -EFAULT;
1292
1293	return 0;
1294}
1295
1296int sock_getsockopt(struct socket *sock, int level, int optname,
1297		    char __user *optval, int __user *optlen)
1298{
1299	struct sock *sk = sock->sk;
1300
1301	union {
1302		int val;
1303		u64 val64;
1304		unsigned long ulval;
1305		struct linger ling;
1306		struct old_timeval32 tm32;
1307		struct __kernel_old_timeval tm;
1308		struct  __kernel_sock_timeval stm;
1309		struct sock_txtime txtime;
1310	} v;
1311
1312	int lv = sizeof(int);
1313	int len;
1314
1315	if (get_user(len, optlen))
1316		return -EFAULT;
1317	if (len < 0)
1318		return -EINVAL;
1319
1320	memset(&v, 0, sizeof(v));
1321
1322	switch (optname) {
1323	case SO_DEBUG:
1324		v.val = sock_flag(sk, SOCK_DBG);
1325		break;
1326
1327	case SO_DONTROUTE:
1328		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1329		break;
1330
1331	case SO_BROADCAST:
1332		v.val = sock_flag(sk, SOCK_BROADCAST);
1333		break;
1334
1335	case SO_SNDBUF:
1336		v.val = READ_ONCE(sk->sk_sndbuf);
1337		break;
1338
1339	case SO_RCVBUF:
1340		v.val = READ_ONCE(sk->sk_rcvbuf);
1341		break;
1342
1343	case SO_REUSEADDR:
1344		v.val = sk->sk_reuse;
1345		break;
1346
1347	case SO_REUSEPORT:
1348		v.val = sk->sk_reuseport;
1349		break;
1350
1351	case SO_KEEPALIVE:
1352		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1353		break;
1354
1355	case SO_TYPE:
1356		v.val = sk->sk_type;
1357		break;
1358
1359	case SO_PROTOCOL:
1360		v.val = sk->sk_protocol;
1361		break;
1362
1363	case SO_DOMAIN:
1364		v.val = sk->sk_family;
1365		break;
1366
1367	case SO_ERROR:
1368		v.val = -sock_error(sk);
1369		if (v.val == 0)
1370			v.val = xchg(&sk->sk_err_soft, 0);
1371		break;
1372
1373	case SO_OOBINLINE:
1374		v.val = sock_flag(sk, SOCK_URGINLINE);
1375		break;
1376
1377	case SO_NO_CHECK:
1378		v.val = sk->sk_no_check_tx;
1379		break;
1380
1381	case SO_PRIORITY:
1382		v.val = sk->sk_priority;
1383		break;
1384
1385	case SO_LINGER:
1386		lv		= sizeof(v.ling);
1387		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1388		v.ling.l_linger	= sk->sk_lingertime / HZ;
1389		break;
1390
1391	case SO_BSDCOMPAT:
1392		break;
1393
1394	case SO_TIMESTAMP_OLD:
1395		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1396				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1397				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1398		break;
1399
1400	case SO_TIMESTAMPNS_OLD:
1401		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1402		break;
1403
1404	case SO_TIMESTAMP_NEW:
1405		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1406		break;
1407
1408	case SO_TIMESTAMPNS_NEW:
1409		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1410		break;
1411
1412	case SO_TIMESTAMPING_OLD:
1413		v.val = sk->sk_tsflags;
1414		break;
1415
1416	case SO_RCVTIMEO_OLD:
1417	case SO_RCVTIMEO_NEW:
1418		lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1419		break;
1420
1421	case SO_SNDTIMEO_OLD:
1422	case SO_SNDTIMEO_NEW:
1423		lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1424		break;
1425
1426	case SO_RCVLOWAT:
1427		v.val = READ_ONCE(sk->sk_rcvlowat);
1428		break;
1429
1430	case SO_SNDLOWAT:
1431		v.val = 1;
1432		break;
1433
1434	case SO_PASSCRED:
1435		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1436		break;
1437
1438	case SO_PEERCRED:
1439	{
1440		struct ucred peercred;
1441		if (len > sizeof(peercred))
1442			len = sizeof(peercred);
1443
1444		spin_lock(&sk->sk_peer_lock);
1445		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1446		spin_unlock(&sk->sk_peer_lock);
1447
1448		if (copy_to_user(optval, &peercred, len))
1449			return -EFAULT;
1450		goto lenout;
1451	}
1452
1453	case SO_PEERGROUPS:
1454	{
1455		const struct cred *cred;
1456		int ret, n;
1457
1458		cred = sk_get_peer_cred(sk);
1459		if (!cred)
1460			return -ENODATA;
1461
1462		n = cred->group_info->ngroups;
1463		if (len < n * sizeof(gid_t)) {
1464			len = n * sizeof(gid_t);
1465			put_cred(cred);
1466			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1467		}
1468		len = n * sizeof(gid_t);
1469
1470		ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1471		put_cred(cred);
1472		if (ret)
1473			return ret;
1474		goto lenout;
1475	}
1476
1477	case SO_PEERNAME:
1478	{
1479		char address[128];
1480
1481		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1482		if (lv < 0)
1483			return -ENOTCONN;
1484		if (lv < len)
1485			return -EINVAL;
1486		if (copy_to_user(optval, address, len))
1487			return -EFAULT;
1488		goto lenout;
1489	}
1490
1491	/* Dubious BSD thing... Probably nobody even uses it, but
1492	 * the UNIX standard wants it for whatever reason... -DaveM
1493	 */
1494	case SO_ACCEPTCONN:
1495		v.val = sk->sk_state == TCP_LISTEN;
1496		break;
1497
1498	case SO_PASSSEC:
1499		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1500		break;
1501
1502	case SO_PEERSEC:
1503		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1504
1505	case SO_MARK:
1506		v.val = sk->sk_mark;
1507		break;
1508
1509	case SO_RXQ_OVFL:
1510		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1511		break;
1512
1513	case SO_WIFI_STATUS:
1514		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1515		break;
1516
1517	case SO_PEEK_OFF:
1518		if (!sock->ops->set_peek_off)
1519			return -EOPNOTSUPP;
1520
1521		v.val = READ_ONCE(sk->sk_peek_off);
1522		break;
1523	case SO_NOFCS:
1524		v.val = sock_flag(sk, SOCK_NOFCS);
1525		break;
1526
1527	case SO_BINDTODEVICE:
1528		return sock_getbindtodevice(sk, optval, optlen, len);
1529
1530	case SO_GET_FILTER:
1531		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1532		if (len < 0)
1533			return len;
1534
1535		goto lenout;
1536
1537	case SO_LOCK_FILTER:
1538		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1539		break;
1540
1541	case SO_BPF_EXTENSIONS:
1542		v.val = bpf_tell_extensions();
1543		break;
1544
1545	case SO_SELECT_ERR_QUEUE:
1546		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1547		break;
1548
1549#ifdef CONFIG_NET_RX_BUSY_POLL
1550	case SO_BUSY_POLL:
1551		v.val = READ_ONCE(sk->sk_ll_usec);
1552		break;
1553#endif
1554
1555	case SO_MAX_PACING_RATE:
1556		/* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1557		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1558			lv = sizeof(v.ulval);
1559			v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1560		} else {
1561			/* 32bit version */
1562			v.val = min_t(unsigned long, ~0U,
1563				      READ_ONCE(sk->sk_max_pacing_rate));
1564		}
1565		break;
1566
1567	case SO_INCOMING_CPU:
1568		v.val = READ_ONCE(sk->sk_incoming_cpu);
1569		break;
1570
1571	case SO_MEMINFO:
1572	{
1573		u32 meminfo[SK_MEMINFO_VARS];
1574
1575		sk_get_meminfo(sk, meminfo);
1576
1577		len = min_t(unsigned int, len, sizeof(meminfo));
1578		if (copy_to_user(optval, &meminfo, len))
1579			return -EFAULT;
1580
1581		goto lenout;
1582	}
1583
1584#ifdef CONFIG_NET_RX_BUSY_POLL
1585	case SO_INCOMING_NAPI_ID:
1586		v.val = READ_ONCE(sk->sk_napi_id);
1587
1588		/* aggregate non-NAPI IDs down to 0 */
1589		if (v.val < MIN_NAPI_ID)
1590			v.val = 0;
1591
1592		break;
1593#endif
1594
1595	case SO_COOKIE:
1596		lv = sizeof(u64);
1597		if (len < lv)
1598			return -EINVAL;
1599		v.val64 = sock_gen_cookie(sk);
1600		break;
1601
1602	case SO_ZEROCOPY:
1603		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1604		break;
1605
1606	case SO_TXTIME:
1607		lv = sizeof(v.txtime);
1608		v.txtime.clockid = sk->sk_clockid;
1609		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1610				  SOF_TXTIME_DEADLINE_MODE : 0;
1611		v.txtime.flags |= sk->sk_txtime_report_errors ?
1612				  SOF_TXTIME_REPORT_ERRORS : 0;
1613		break;
1614
1615	case SO_BINDTOIFINDEX:
1616		v.val = sk->sk_bound_dev_if;
1617		break;
1618
1619	default:
1620		/* We implement the SO_SNDLOWAT etc to not be settable
1621		 * (1003.1g 7).
1622		 */
1623		return -ENOPROTOOPT;
1624	}
1625
1626	if (len > lv)
1627		len = lv;
1628	if (copy_to_user(optval, &v, len))
1629		return -EFAULT;
1630lenout:
1631	if (put_user(len, optlen))
1632		return -EFAULT;
1633	return 0;
1634}
1635
1636/*
1637 * Initialize an sk_lock.
1638 *
1639 * (We also register the sk_lock with the lock validator.)
1640 */
1641static inline void sock_lock_init(struct sock *sk)
1642{
1643	if (sk->sk_kern_sock)
1644		sock_lock_init_class_and_name(
1645			sk,
1646			af_family_kern_slock_key_strings[sk->sk_family],
1647			af_family_kern_slock_keys + sk->sk_family,
1648			af_family_kern_key_strings[sk->sk_family],
1649			af_family_kern_keys + sk->sk_family);
1650	else
1651		sock_lock_init_class_and_name(
1652			sk,
1653			af_family_slock_key_strings[sk->sk_family],
1654			af_family_slock_keys + sk->sk_family,
1655			af_family_key_strings[sk->sk_family],
1656			af_family_keys + sk->sk_family);
1657}
1658
1659/*
1660 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1661 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1662 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1663 */
1664static void sock_copy(struct sock *nsk, const struct sock *osk)
1665{
1666	const struct proto *prot = READ_ONCE(osk->sk_prot);
1667#ifdef CONFIG_SECURITY_NETWORK
1668	void *sptr = nsk->sk_security;
1669#endif
1670	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1671
1672	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1673	       prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1674
1675#ifdef CONFIG_SECURITY_NETWORK
1676	nsk->sk_security = sptr;
1677	security_sk_clone(osk, nsk);
1678#endif
1679}
1680
1681static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1682		int family)
1683{
1684	struct sock *sk;
1685	struct kmem_cache *slab;
1686
1687	slab = prot->slab;
1688	if (slab != NULL) {
1689		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1690		if (!sk)
1691			return sk;
1692		if (want_init_on_alloc(priority))
1693			sk_prot_clear_nulls(sk, prot->obj_size);
1694	} else
1695		sk = kmalloc(prot->obj_size, priority);
1696
1697	if (sk != NULL) {
1698		if (security_sk_alloc(sk, family, priority))
1699			goto out_free;
1700
1701		if (!try_module_get(prot->owner))
1702			goto out_free_sec;
1703		sk_tx_queue_clear(sk);
1704	}
1705
1706	return sk;
1707
1708out_free_sec:
1709	security_sk_free(sk);
1710out_free:
1711	if (slab != NULL)
1712		kmem_cache_free(slab, sk);
1713	else
1714		kfree(sk);
1715	return NULL;
1716}
1717
1718static void sk_prot_free(struct proto *prot, struct sock *sk)
1719{
1720	struct kmem_cache *slab;
1721	struct module *owner;
1722
1723	owner = prot->owner;
1724	slab = prot->slab;
1725
1726	cgroup_sk_free(&sk->sk_cgrp_data);
1727	mem_cgroup_sk_free(sk);
1728	security_sk_free(sk);
1729	if (slab != NULL)
1730		kmem_cache_free(slab, sk);
1731	else
1732		kfree(sk);
1733	module_put(owner);
1734}
1735
1736/**
1737 *	sk_alloc - All socket objects are allocated here
1738 *	@net: the applicable net namespace
1739 *	@family: protocol family
1740 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1741 *	@prot: struct proto associated with this new sock instance
1742 *	@kern: is this to be a kernel socket?
1743 */
1744struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1745		      struct proto *prot, int kern)
1746{
1747	struct sock *sk;
1748
1749	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1750	if (sk) {
1751		sk->sk_family = family;
1752		/*
1753		 * See comment in struct sock definition to understand
1754		 * why we need sk_prot_creator -acme
1755		 */
1756		sk->sk_prot = sk->sk_prot_creator = prot;
1757		sk->sk_kern_sock = kern;
1758		sock_lock_init(sk);
1759		sk->sk_net_refcnt = kern ? 0 : 1;
1760		if (likely(sk->sk_net_refcnt)) {
1761			get_net(net);
1762			sock_inuse_add(net, 1);
1763		}
1764
1765		sock_net_set(sk, net);
1766		refcount_set(&sk->sk_wmem_alloc, 1);
1767
1768		mem_cgroup_sk_alloc(sk);
1769		cgroup_sk_alloc(&sk->sk_cgrp_data);
1770		sock_update_classid(&sk->sk_cgrp_data);
1771		sock_update_netprioidx(&sk->sk_cgrp_data);
1772		sk_tx_queue_clear(sk);
1773	}
1774
1775	return sk;
1776}
1777EXPORT_SYMBOL(sk_alloc);
1778
1779/* Sockets having SOCK_RCU_FREE will call this function after one RCU
1780 * grace period. This is the case for UDP sockets and TCP listeners.
1781 */
1782static void __sk_destruct(struct rcu_head *head)
1783{
1784	struct sock *sk = container_of(head, struct sock, sk_rcu);
1785	struct sk_filter *filter;
1786
1787	if (sk->sk_destruct)
1788		sk->sk_destruct(sk);
1789
1790	filter = rcu_dereference_check(sk->sk_filter,
1791				       refcount_read(&sk->sk_wmem_alloc) == 0);
1792	if (filter) {
1793		sk_filter_uncharge(sk, filter);
1794		RCU_INIT_POINTER(sk->sk_filter, NULL);
1795	}
1796
1797	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1798
1799#ifdef CONFIG_BPF_SYSCALL
1800	bpf_sk_storage_free(sk);
1801#endif
1802
1803	if (atomic_read(&sk->sk_omem_alloc))
1804		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1805			 __func__, atomic_read(&sk->sk_omem_alloc));
1806
1807	if (sk->sk_frag.page) {
1808		put_page(sk->sk_frag.page);
1809		sk->sk_frag.page = NULL;
1810	}
1811
1812	/* We do not need to acquire sk->sk_peer_lock, we are the last user. */
1813	put_cred(sk->sk_peer_cred);
1814	put_pid(sk->sk_peer_pid);
1815
1816	if (likely(sk->sk_net_refcnt))
1817		put_net(sock_net(sk));
1818	sk_prot_free(sk->sk_prot_creator, sk);
1819}
1820
1821void sk_destruct(struct sock *sk)
1822{
1823	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1824
1825	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1826		reuseport_detach_sock(sk);
1827		use_call_rcu = true;
1828	}
1829
1830	if (use_call_rcu)
1831		call_rcu(&sk->sk_rcu, __sk_destruct);
1832	else
1833		__sk_destruct(&sk->sk_rcu);
1834}
1835
1836static void __sk_free(struct sock *sk)
1837{
1838	if (likely(sk->sk_net_refcnt))
1839		sock_inuse_add(sock_net(sk), -1);
1840
1841	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1842		sock_diag_broadcast_destroy(sk);
1843	else
1844		sk_destruct(sk);
1845}
1846
1847void sk_free(struct sock *sk)
1848{
1849	/*
1850	 * We subtract one from sk_wmem_alloc and can know if
1851	 * some packets are still in some tx queue.
1852	 * If not null, sock_wfree() will call __sk_free(sk) later
1853	 */
1854	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1855		__sk_free(sk);
1856}
1857EXPORT_SYMBOL(sk_free);
1858
1859static void sk_init_common(struct sock *sk)
1860{
1861	skb_queue_head_init(&sk->sk_receive_queue);
1862	skb_queue_head_init(&sk->sk_write_queue);
1863	skb_queue_head_init(&sk->sk_error_queue);
1864
1865	rwlock_init(&sk->sk_callback_lock);
1866	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1867			af_rlock_keys + sk->sk_family,
1868			af_family_rlock_key_strings[sk->sk_family]);
1869	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1870			af_wlock_keys + sk->sk_family,
1871			af_family_wlock_key_strings[sk->sk_family]);
1872	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1873			af_elock_keys + sk->sk_family,
1874			af_family_elock_key_strings[sk->sk_family]);
1875	lockdep_set_class_and_name(&sk->sk_callback_lock,
1876			af_callback_keys + sk->sk_family,
1877			af_family_clock_key_strings[sk->sk_family]);
1878}
1879
1880/**
1881 *	sk_clone_lock - clone a socket, and lock its clone
1882 *	@sk: the socket to clone
1883 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1884 *
1885 *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1886 */
1887struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1888{
1889	struct proto *prot = READ_ONCE(sk->sk_prot);
1890	struct sk_filter *filter;
1891	bool is_charged = true;
1892	struct sock *newsk;
1893
1894	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1895	if (!newsk)
1896		goto out;
1897
1898	sock_copy(newsk, sk);
1899
1900	newsk->sk_prot_creator = prot;
1901
1902	/* SANITY */
1903	if (likely(newsk->sk_net_refcnt)) {
1904		get_net(sock_net(newsk));
1905		sock_inuse_add(sock_net(newsk), 1);
1906	}
1907	sk_node_init(&newsk->sk_node);
1908	sock_lock_init(newsk);
1909	bh_lock_sock(newsk);
1910	newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1911	newsk->sk_backlog.len = 0;
1912
1913	atomic_set(&newsk->sk_rmem_alloc, 0);
1914
1915	/* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
1916	refcount_set(&newsk->sk_wmem_alloc, 1);
1917
1918	atomic_set(&newsk->sk_omem_alloc, 0);
1919	sk_init_common(newsk);
1920
1921	newsk->sk_dst_cache	= NULL;
1922	newsk->sk_dst_pending_confirm = 0;
1923	newsk->sk_wmem_queued	= 0;
1924	newsk->sk_forward_alloc = 0;
1925	atomic_set(&newsk->sk_drops, 0);
1926	newsk->sk_send_head	= NULL;
1927	newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1928	atomic_set(&newsk->sk_zckey, 0);
1929
1930	sock_reset_flag(newsk, SOCK_DONE);
1931
1932	/* sk->sk_memcg will be populated at accept() time */
1933	newsk->sk_memcg = NULL;
1934
1935	cgroup_sk_clone(&newsk->sk_cgrp_data);
1936
1937	rcu_read_lock();
1938	filter = rcu_dereference(sk->sk_filter);
1939	if (filter != NULL)
1940		/* though it's an empty new sock, the charging may fail
1941		 * if sysctl_optmem_max was changed between creation of
1942		 * original socket and cloning
1943		 */
1944		is_charged = sk_filter_charge(newsk, filter);
1945	RCU_INIT_POINTER(newsk->sk_filter, filter);
1946	rcu_read_unlock();
1947
1948	if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1949		/* We need to make sure that we don't uncharge the new
1950		 * socket if we couldn't charge it in the first place
1951		 * as otherwise we uncharge the parent's filter.
1952		 */
1953		if (!is_charged)
1954			RCU_INIT_POINTER(newsk->sk_filter, NULL);
1955		sk_free_unlock_clone(newsk);
1956		newsk = NULL;
1957		goto out;
1958	}
1959	RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1960
1961	if (bpf_sk_storage_clone(sk, newsk)) {
1962		sk_free_unlock_clone(newsk);
1963		newsk = NULL;
1964		goto out;
1965	}
1966
1967	/* Clear sk_user_data if parent had the pointer tagged
1968	 * as not suitable for copying when cloning.
1969	 */
1970	if (sk_user_data_is_nocopy(newsk))
1971		newsk->sk_user_data = NULL;
1972
1973	newsk->sk_err	   = 0;
1974	newsk->sk_err_soft = 0;
1975	newsk->sk_priority = 0;
1976	newsk->sk_incoming_cpu = raw_smp_processor_id();
1977
1978	/* Before updating sk_refcnt, we must commit prior changes to memory
1979	 * (Documentation/RCU/rculist_nulls.rst for details)
1980	 */
1981	smp_wmb();
1982	refcount_set(&newsk->sk_refcnt, 2);
1983
1984	/* Increment the counter in the same struct proto as the master
1985	 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1986	 * is the same as sk->sk_prot->socks, as this field was copied
1987	 * with memcpy).
1988	 *
1989	 * This _changes_ the previous behaviour, where
1990	 * tcp_create_openreq_child always was incrementing the
1991	 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1992	 * to be taken into account in all callers. -acme
1993	 */
1994	sk_refcnt_debug_inc(newsk);
1995	sk_set_socket(newsk, NULL);
1996	sk_tx_queue_clear(newsk);
1997	RCU_INIT_POINTER(newsk->sk_wq, NULL);
1998
1999	if (newsk->sk_prot->sockets_allocated)
2000		sk_sockets_allocated_inc(newsk);
2001
2002	if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2003		net_enable_timestamp();
2004out:
2005	return newsk;
2006}
2007EXPORT_SYMBOL_GPL(sk_clone_lock);
2008
2009void sk_free_unlock_clone(struct sock *sk)
2010{
2011	/* It is still raw copy of parent, so invalidate
2012	 * destructor and make plain sk_free() */
2013	sk->sk_destruct = NULL;
2014	bh_unlock_sock(sk);
2015	sk_free(sk);
2016}
2017EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2018
2019void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2020{
2021	u32 max_segs = 1;
2022
2023	sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2024	if (sk->sk_route_caps & NETIF_F_GSO)
2025		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2026	sk->sk_route_caps &= ~sk->sk_route_nocaps;
2027	if (sk_can_gso(sk)) {
2028		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2029			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2030		} else {
2031			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2032			sk->sk_gso_max_size = dst->dev->gso_max_size;
2033			max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2034		}
2035	}
2036	sk->sk_gso_max_segs = max_segs;
2037	sk_dst_set(sk, dst);
2038}
2039EXPORT_SYMBOL_GPL(sk_setup_caps);
2040
2041/*
2042 *	Simple resource managers for sockets.
2043 */
2044
2045
2046/*
2047 * Write buffer destructor automatically called from kfree_skb.
2048 */
2049void sock_wfree(struct sk_buff *skb)
2050{
2051	struct sock *sk = skb->sk;
2052	unsigned int len = skb->truesize;
2053
2054	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2055		/*
2056		 * Keep a reference on sk_wmem_alloc, this will be released
2057		 * after sk_write_space() call
2058		 */
2059		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2060		sk->sk_write_space(sk);
2061		len = 1;
2062	}
2063	/*
2064	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2065	 * could not do because of in-flight packets
2066	 */
2067	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2068		__sk_free(sk);
2069}
2070EXPORT_SYMBOL(sock_wfree);
2071
2072/* This variant of sock_wfree() is used by TCP,
2073 * since it sets SOCK_USE_WRITE_QUEUE.
2074 */
2075void __sock_wfree(struct sk_buff *skb)
2076{
2077	struct sock *sk = skb->sk;
2078
2079	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2080		__sk_free(sk);
2081}
2082
2083void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2084{
2085	skb_orphan(skb);
2086	skb->sk = sk;
2087#ifdef CONFIG_INET
2088	if (unlikely(!sk_fullsock(sk))) {
2089		skb->destructor = sock_edemux;
2090		sock_hold(sk);
2091		return;
2092	}
2093#endif
2094	skb->destructor = sock_wfree;
2095	skb_set_hash_from_sk(skb, sk);
2096	/*
2097	 * We used to take a refcount on sk, but following operation
2098	 * is enough to guarantee sk_free() wont free this sock until
2099	 * all in-flight packets are completed
2100	 */
2101	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2102}
2103EXPORT_SYMBOL(skb_set_owner_w);
2104
2105static bool can_skb_orphan_partial(const struct sk_buff *skb)
2106{
2107#ifdef CONFIG_TLS_DEVICE
2108	/* Drivers depend on in-order delivery for crypto offload,
2109	 * partial orphan breaks out-of-order-OK logic.
2110	 */
2111	if (skb->decrypted)
2112		return false;
2113#endif
2114	return (skb->destructor == sock_wfree ||
2115		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2116}
2117
2118/* This helper is used by netem, as it can hold packets in its
2119 * delay queue. We want to allow the owner socket to send more
2120 * packets, as if they were already TX completed by a typical driver.
2121 * But we also want to keep skb->sk set because some packet schedulers
2122 * rely on it (sch_fq for example).
2123 */
2124void skb_orphan_partial(struct sk_buff *skb)
2125{
2126	if (skb_is_tcp_pure_ack(skb))
2127		return;
2128
2129	if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2130		return;
2131
2132	skb_orphan(skb);
2133}
2134EXPORT_SYMBOL(skb_orphan_partial);
2135
2136/*
2137 * Read buffer destructor automatically called from kfree_skb.
2138 */
2139void sock_rfree(struct sk_buff *skb)
2140{
2141	struct sock *sk = skb->sk;
2142	unsigned int len = skb->truesize;
2143
2144	atomic_sub(len, &sk->sk_rmem_alloc);
2145	sk_mem_uncharge(sk, len);
2146}
2147EXPORT_SYMBOL(sock_rfree);
2148
2149/*
2150 * Buffer destructor for skbs that are not used directly in read or write
2151 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2152 */
2153void sock_efree(struct sk_buff *skb)
2154{
2155	sock_put(skb->sk);
2156}
2157EXPORT_SYMBOL(sock_efree);
2158
2159/* Buffer destructor for prefetch/receive path where reference count may
2160 * not be held, e.g. for listen sockets.
2161 */
2162#ifdef CONFIG_INET
2163void sock_pfree(struct sk_buff *skb)
2164{
2165	if (sk_is_refcounted(skb->sk))
2166		sock_gen_put(skb->sk);
2167}
2168EXPORT_SYMBOL(sock_pfree);
2169#endif /* CONFIG_INET */
2170
2171kuid_t sock_i_uid(struct sock *sk)
2172{
2173	kuid_t uid;
2174
2175	read_lock_bh(&sk->sk_callback_lock);
2176	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2177	read_unlock_bh(&sk->sk_callback_lock);
2178	return uid;
2179}
2180EXPORT_SYMBOL(sock_i_uid);
2181
2182unsigned long __sock_i_ino(struct sock *sk)
2183{
2184	unsigned long ino;
2185
2186	read_lock(&sk->sk_callback_lock);
2187	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2188	read_unlock(&sk->sk_callback_lock);
2189	return ino;
2190}
2191EXPORT_SYMBOL(__sock_i_ino);
2192
2193unsigned long sock_i_ino(struct sock *sk)
2194{
2195	unsigned long ino;
2196
2197	local_bh_disable();
2198	ino = __sock_i_ino(sk);
2199	local_bh_enable();
2200	return ino;
2201}
2202EXPORT_SYMBOL(sock_i_ino);
2203
2204/*
2205 * Allocate a skb from the socket's send buffer.
2206 */
2207struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2208			     gfp_t priority)
2209{
2210	if (force ||
2211	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2212		struct sk_buff *skb = alloc_skb(size, priority);
2213
2214		if (skb) {
2215			skb_set_owner_w(skb, sk);
2216			return skb;
2217		}
2218	}
2219	return NULL;
2220}
2221EXPORT_SYMBOL(sock_wmalloc);
2222
2223static void sock_ofree(struct sk_buff *skb)
2224{
2225	struct sock *sk = skb->sk;
2226
2227	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2228}
2229
2230struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2231			     gfp_t priority)
2232{
2233	struct sk_buff *skb;
2234
2235	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2236	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2237	    READ_ONCE(sysctl_optmem_max))
2238		return NULL;
2239
2240	skb = alloc_skb(size, priority);
2241	if (!skb)
2242		return NULL;
2243
2244	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2245	skb->sk = sk;
2246	skb->destructor = sock_ofree;
2247	return skb;
2248}
2249
2250/*
2251 * Allocate a memory block from the socket's option memory buffer.
2252 */
2253void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2254{
2255	int optmem_max = READ_ONCE(sysctl_optmem_max);
2256
2257	if ((unsigned int)size <= optmem_max &&
2258	    atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2259		void *mem;
2260		/* First do the add, to avoid the race if kmalloc
2261		 * might sleep.
2262		 */
2263		atomic_add(size, &sk->sk_omem_alloc);
2264		mem = kmalloc(size, priority);
2265		if (mem)
2266			return mem;
2267		atomic_sub(size, &sk->sk_omem_alloc);
2268	}
2269	return NULL;
2270}
2271EXPORT_SYMBOL(sock_kmalloc);
2272
2273/* Free an option memory block. Note, we actually want the inline
2274 * here as this allows gcc to detect the nullify and fold away the
2275 * condition entirely.
2276 */
2277static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2278				  const bool nullify)
2279{
2280	if (WARN_ON_ONCE(!mem))
2281		return;
2282	if (nullify)
2283		kfree_sensitive(mem);
2284	else
2285		kfree(mem);
2286	atomic_sub(size, &sk->sk_omem_alloc);
2287}
2288
2289void sock_kfree_s(struct sock *sk, void *mem, int size)
2290{
2291	__sock_kfree_s(sk, mem, size, false);
2292}
2293EXPORT_SYMBOL(sock_kfree_s);
2294
2295void sock_kzfree_s(struct sock *sk, void *mem, int size)
2296{
2297	__sock_kfree_s(sk, mem, size, true);
2298}
2299EXPORT_SYMBOL(sock_kzfree_s);
2300
2301/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2302   I think, these locks should be removed for datagram sockets.
2303 */
2304static long sock_wait_for_wmem(struct sock *sk, long timeo)
2305{
2306	DEFINE_WAIT(wait);
2307
2308	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2309	for (;;) {
2310		if (!timeo)
2311			break;
2312		if (signal_pending(current))
2313			break;
2314		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2315		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2316		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2317			break;
2318		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2319			break;
2320		if (READ_ONCE(sk->sk_err))
2321			break;
2322		timeo = schedule_timeout(timeo);
2323	}
2324	finish_wait(sk_sleep(sk), &wait);
2325	return timeo;
2326}
2327
2328
2329/*
2330 *	Generic send/receive buffer handlers
2331 */
2332
2333struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2334				     unsigned long data_len, int noblock,
2335				     int *errcode, int max_page_order)
2336{
2337	struct sk_buff *skb;
2338	long timeo;
2339	int err;
2340
2341	timeo = sock_sndtimeo(sk, noblock);
2342	for (;;) {
2343		err = sock_error(sk);
2344		if (err != 0)
2345			goto failure;
2346
2347		err = -EPIPE;
2348		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2349			goto failure;
2350
2351		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2352			break;
2353
2354		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2355		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2356		err = -EAGAIN;
2357		if (!timeo)
2358			goto failure;
2359		if (signal_pending(current))
2360			goto interrupted;
2361		timeo = sock_wait_for_wmem(sk, timeo);
2362	}
2363	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2364				   errcode, sk->sk_allocation);
2365	if (skb)
2366		skb_set_owner_w(skb, sk);
2367	return skb;
2368
2369interrupted:
2370	err = sock_intr_errno(timeo);
2371failure:
2372	*errcode = err;
2373	return NULL;
2374}
2375EXPORT_SYMBOL(sock_alloc_send_pskb);
2376
2377struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2378				    int noblock, int *errcode)
2379{
2380	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2381}
2382EXPORT_SYMBOL(sock_alloc_send_skb);
2383
2384int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2385		     struct sockcm_cookie *sockc)
2386{
2387	u32 tsflags;
2388
2389	switch (cmsg->cmsg_type) {
2390	case SO_MARK:
2391		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2392			return -EPERM;
2393		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2394			return -EINVAL;
2395		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2396		break;
2397	case SO_TIMESTAMPING_OLD:
2398	case SO_TIMESTAMPING_NEW:
2399		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2400			return -EINVAL;
2401
2402		tsflags = *(u32 *)CMSG_DATA(cmsg);
2403		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2404			return -EINVAL;
2405
2406		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2407		sockc->tsflags |= tsflags;
2408		break;
2409	case SCM_TXTIME:
2410		if (!sock_flag(sk, SOCK_TXTIME))
2411			return -EINVAL;
2412		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2413			return -EINVAL;
2414		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2415		break;
2416	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2417	case SCM_RIGHTS:
2418	case SCM_CREDENTIALS:
2419		break;
2420	default:
2421		return -EINVAL;
2422	}
2423	return 0;
2424}
2425EXPORT_SYMBOL(__sock_cmsg_send);
2426
2427int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2428		   struct sockcm_cookie *sockc)
2429{
2430	struct cmsghdr *cmsg;
2431	int ret;
2432
2433	for_each_cmsghdr(cmsg, msg) {
2434		if (!CMSG_OK(msg, cmsg))
2435			return -EINVAL;
2436		if (cmsg->cmsg_level != SOL_SOCKET)
2437			continue;
2438		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2439		if (ret)
2440			return ret;
2441	}
2442	return 0;
2443}
2444EXPORT_SYMBOL(sock_cmsg_send);
2445
2446static void sk_enter_memory_pressure(struct sock *sk)
2447{
2448	if (!sk->sk_prot->enter_memory_pressure)
2449		return;
2450
2451	sk->sk_prot->enter_memory_pressure(sk);
2452}
2453
2454static void sk_leave_memory_pressure(struct sock *sk)
2455{
2456	if (sk->sk_prot->leave_memory_pressure) {
2457		sk->sk_prot->leave_memory_pressure(sk);
2458	} else {
2459		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2460
2461		if (memory_pressure && READ_ONCE(*memory_pressure))
2462			WRITE_ONCE(*memory_pressure, 0);
2463	}
2464}
2465
2466DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2467
2468/**
2469 * skb_page_frag_refill - check that a page_frag contains enough room
2470 * @sz: minimum size of the fragment we want to get
2471 * @pfrag: pointer to page_frag
2472 * @gfp: priority for memory allocation
2473 *
2474 * Note: While this allocator tries to use high order pages, there is
2475 * no guarantee that allocations succeed. Therefore, @sz MUST be
2476 * less or equal than PAGE_SIZE.
2477 */
2478bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2479{
2480	if (pfrag->page) {
2481		if (page_ref_count(pfrag->page) == 1) {
2482			pfrag->offset = 0;
2483			return true;
2484		}
2485		if (pfrag->offset + sz <= pfrag->size)
2486			return true;
2487		put_page(pfrag->page);
2488	}
2489
2490	pfrag->offset = 0;
2491	if (SKB_FRAG_PAGE_ORDER &&
2492	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2493		/* Avoid direct reclaim but allow kswapd to wake */
2494		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2495					  __GFP_COMP | __GFP_NOWARN |
2496					  __GFP_NORETRY,
2497					  SKB_FRAG_PAGE_ORDER);
2498		if (likely(pfrag->page)) {
2499			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2500			return true;
2501		}
2502	}
2503	pfrag->page = alloc_page(gfp);
2504	if (likely(pfrag->page)) {
2505		pfrag->size = PAGE_SIZE;
2506		return true;
2507	}
2508	return false;
2509}
2510EXPORT_SYMBOL(skb_page_frag_refill);
2511
2512bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2513{
2514	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2515		return true;
2516
2517	sk_enter_memory_pressure(sk);
2518	sk_stream_moderate_sndbuf(sk);
2519	return false;
2520}
2521EXPORT_SYMBOL(sk_page_frag_refill);
2522
2523static void __lock_sock(struct sock *sk)
2524	__releases(&sk->sk_lock.slock)
2525	__acquires(&sk->sk_lock.slock)
2526{
2527	DEFINE_WAIT(wait);
2528
2529	for (;;) {
2530		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2531					TASK_UNINTERRUPTIBLE);
2532		spin_unlock_bh(&sk->sk_lock.slock);
2533		schedule();
2534		spin_lock_bh(&sk->sk_lock.slock);
2535		if (!sock_owned_by_user(sk))
2536			break;
2537	}
2538	finish_wait(&sk->sk_lock.wq, &wait);
2539}
2540
2541void __release_sock(struct sock *sk)
2542	__releases(&sk->sk_lock.slock)
2543	__acquires(&sk->sk_lock.slock)
2544{
2545	struct sk_buff *skb, *next;
2546
2547	while ((skb = sk->sk_backlog.head) != NULL) {
2548		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2549
2550		spin_unlock_bh(&sk->sk_lock.slock);
2551
2552		do {
2553			next = skb->next;
2554			prefetch(next);
2555			WARN_ON_ONCE(skb_dst_is_noref(skb));
2556			skb_mark_not_on_list(skb);
2557			sk_backlog_rcv(sk, skb);
2558
2559			cond_resched();
2560
2561			skb = next;
2562		} while (skb != NULL);
2563
2564		spin_lock_bh(&sk->sk_lock.slock);
2565	}
2566
2567	/*
2568	 * Doing the zeroing here guarantee we can not loop forever
2569	 * while a wild producer attempts to flood us.
2570	 */
2571	sk->sk_backlog.len = 0;
2572}
2573
2574void __sk_flush_backlog(struct sock *sk)
2575{
2576	spin_lock_bh(&sk->sk_lock.slock);
2577	__release_sock(sk);
2578	spin_unlock_bh(&sk->sk_lock.slock);
2579}
2580
2581/**
2582 * sk_wait_data - wait for data to arrive at sk_receive_queue
2583 * @sk:    sock to wait on
2584 * @timeo: for how long
2585 * @skb:   last skb seen on sk_receive_queue
2586 *
2587 * Now socket state including sk->sk_err is changed only under lock,
2588 * hence we may omit checks after joining wait queue.
2589 * We check receive queue before schedule() only as optimization;
2590 * it is very likely that release_sock() added new data.
2591 */
2592int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2593{
2594	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2595	int rc;
2596
2597	add_wait_queue(sk_sleep(sk), &wait);
2598	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2599	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2600	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2601	remove_wait_queue(sk_sleep(sk), &wait);
2602	return rc;
2603}
2604EXPORT_SYMBOL(sk_wait_data);
2605
2606/**
2607 *	__sk_mem_raise_allocated - increase memory_allocated
2608 *	@sk: socket
2609 *	@size: memory size to allocate
2610 *	@amt: pages to allocate
2611 *	@kind: allocation type
2612 *
2613 *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2614 */
2615int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2616{
2617	struct proto *prot = sk->sk_prot;
2618	long allocated = sk_memory_allocated_add(sk, amt);
2619	bool charged = true;
2620
2621	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2622	    !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2623		goto suppress_allocation;
2624
2625	/* Under limit. */
2626	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2627		sk_leave_memory_pressure(sk);
2628		return 1;
2629	}
2630
2631	/* Under pressure. */
2632	if (allocated > sk_prot_mem_limits(sk, 1))
2633		sk_enter_memory_pressure(sk);
2634
2635	/* Over hard limit. */
2636	if (allocated > sk_prot_mem_limits(sk, 2))
2637		goto suppress_allocation;
2638
2639	/* guarantee minimum buffer size under pressure */
2640	if (kind == SK_MEM_RECV) {
2641		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2642			return 1;
2643
2644	} else { /* SK_MEM_SEND */
2645		int wmem0 = sk_get_wmem0(sk, prot);
2646
2647		if (sk->sk_type == SOCK_STREAM) {
2648			if (sk->sk_wmem_queued < wmem0)
2649				return 1;
2650		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2651				return 1;
2652		}
2653	}
2654
2655	if (sk_has_memory_pressure(sk)) {
2656		u64 alloc;
2657
2658		if (!sk_under_memory_pressure(sk))
2659			return 1;
2660		alloc = sk_sockets_allocated_read_positive(sk);
2661		if (sk_prot_mem_limits(sk, 2) > alloc *
2662		    sk_mem_pages(sk->sk_wmem_queued +
2663				 atomic_read(&sk->sk_rmem_alloc) +
2664				 sk->sk_forward_alloc))
2665			return 1;
2666	}
2667
2668suppress_allocation:
2669
2670	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2671		sk_stream_moderate_sndbuf(sk);
2672
2673		/* Fail only if socket is _under_ its sndbuf.
2674		 * In this case we cannot block, so that we have to fail.
2675		 */
2676		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2677			return 1;
2678	}
2679
2680	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2681		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2682
2683	sk_memory_allocated_sub(sk, amt);
2684
2685	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2686		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2687
2688	return 0;
2689}
2690EXPORT_SYMBOL(__sk_mem_raise_allocated);
2691
2692/**
2693 *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2694 *	@sk: socket
2695 *	@size: memory size to allocate
2696 *	@kind: allocation type
2697 *
2698 *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2699 *	rmem allocation. This function assumes that protocols which have
2700 *	memory_pressure use sk_wmem_queued as write buffer accounting.
2701 */
2702int __sk_mem_schedule(struct sock *sk, int size, int kind)
2703{
2704	int ret, amt = sk_mem_pages(size);
2705
2706	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2707	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2708	if (!ret)
2709		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2710	return ret;
2711}
2712EXPORT_SYMBOL(__sk_mem_schedule);
2713
2714/**
2715 *	__sk_mem_reduce_allocated - reclaim memory_allocated
2716 *	@sk: socket
2717 *	@amount: number of quanta
2718 *
2719 *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2720 */
2721void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2722{
2723	sk_memory_allocated_sub(sk, amount);
2724
2725	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2726		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2727
2728	if (sk_under_global_memory_pressure(sk) &&
2729	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2730		sk_leave_memory_pressure(sk);
2731}
2732EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2733
2734/**
2735 *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2736 *	@sk: socket
2737 *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2738 */
2739void __sk_mem_reclaim(struct sock *sk, int amount)
2740{
2741	amount >>= SK_MEM_QUANTUM_SHIFT;
2742	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2743	__sk_mem_reduce_allocated(sk, amount);
2744}
2745EXPORT_SYMBOL(__sk_mem_reclaim);
2746
2747int sk_set_peek_off(struct sock *sk, int val)
2748{
2749	WRITE_ONCE(sk->sk_peek_off, val);
2750	return 0;
2751}
2752EXPORT_SYMBOL_GPL(sk_set_peek_off);
2753
2754/*
2755 * Set of default routines for initialising struct proto_ops when
2756 * the protocol does not support a particular function. In certain
2757 * cases where it makes no sense for a protocol to have a "do nothing"
2758 * function, some default processing is provided.
2759 */
2760
2761int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2762{
2763	return -EOPNOTSUPP;
2764}
2765EXPORT_SYMBOL(sock_no_bind);
2766
2767int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2768		    int len, int flags)
2769{
2770	return -EOPNOTSUPP;
2771}
2772EXPORT_SYMBOL(sock_no_connect);
2773
2774int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2775{
2776	return -EOPNOTSUPP;
2777}
2778EXPORT_SYMBOL(sock_no_socketpair);
2779
2780int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2781		   bool kern)
2782{
2783	return -EOPNOTSUPP;
2784}
2785EXPORT_SYMBOL(sock_no_accept);
2786
2787int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2788		    int peer)
2789{
2790	return -EOPNOTSUPP;
2791}
2792EXPORT_SYMBOL(sock_no_getname);
2793
2794int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2795{
2796	return -EOPNOTSUPP;
2797}
2798EXPORT_SYMBOL(sock_no_ioctl);
2799
2800int sock_no_listen(struct socket *sock, int backlog)
2801{
2802	return -EOPNOTSUPP;
2803}
2804EXPORT_SYMBOL(sock_no_listen);
2805
2806int sock_no_shutdown(struct socket *sock, int how)
2807{
2808	return -EOPNOTSUPP;
2809}
2810EXPORT_SYMBOL(sock_no_shutdown);
2811
2812int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2813{
2814	return -EOPNOTSUPP;
2815}
2816EXPORT_SYMBOL(sock_no_sendmsg);
2817
2818int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2819{
2820	return -EOPNOTSUPP;
2821}
2822EXPORT_SYMBOL(sock_no_sendmsg_locked);
2823
2824int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2825		    int flags)
2826{
2827	return -EOPNOTSUPP;
2828}
2829EXPORT_SYMBOL(sock_no_recvmsg);
2830
2831int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2832{
2833	/* Mirror missing mmap method error code */
2834	return -ENODEV;
2835}
2836EXPORT_SYMBOL(sock_no_mmap);
2837
2838/*
2839 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2840 * various sock-based usage counts.
2841 */
2842void __receive_sock(struct file *file)
2843{
2844	struct socket *sock;
2845	int error;
2846
2847	/*
2848	 * The resulting value of "error" is ignored here since we only
2849	 * need to take action when the file is a socket and testing
2850	 * "sock" for NULL is sufficient.
2851	 */
2852	sock = sock_from_file(file, &error);
2853	if (sock) {
2854		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2855		sock_update_classid(&sock->sk->sk_cgrp_data);
2856	}
2857}
2858
2859ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2860{
2861	ssize_t res;
2862	struct msghdr msg = {.msg_flags = flags};
2863	struct kvec iov;
2864	char *kaddr = kmap(page);
2865	iov.iov_base = kaddr + offset;
2866	iov.iov_len = size;
2867	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2868	kunmap(page);
2869	return res;
2870}
2871EXPORT_SYMBOL(sock_no_sendpage);
2872
2873ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2874				int offset, size_t size, int flags)
2875{
2876	ssize_t res;
2877	struct msghdr msg = {.msg_flags = flags};
2878	struct kvec iov;
2879	char *kaddr = kmap(page);
2880
2881	iov.iov_base = kaddr + offset;
2882	iov.iov_len = size;
2883	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2884	kunmap(page);
2885	return res;
2886}
2887EXPORT_SYMBOL(sock_no_sendpage_locked);
2888
2889/*
2890 *	Default Socket Callbacks
2891 */
2892
2893static void sock_def_wakeup(struct sock *sk)
2894{
2895	struct socket_wq *wq;
2896
2897	rcu_read_lock();
2898	wq = rcu_dereference(sk->sk_wq);
2899	if (skwq_has_sleeper(wq))
2900		wake_up_interruptible_all(&wq->wait);
2901	rcu_read_unlock();
2902}
2903
2904static void sock_def_error_report(struct sock *sk)
2905{
2906	struct socket_wq *wq;
2907
2908	rcu_read_lock();
2909	wq = rcu_dereference(sk->sk_wq);
2910	if (skwq_has_sleeper(wq))
2911		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2912	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2913	rcu_read_unlock();
2914}
2915
2916void sock_def_readable(struct sock *sk)
2917{
2918	struct socket_wq *wq;
2919
2920	rcu_read_lock();
2921	wq = rcu_dereference(sk->sk_wq);
2922	if (skwq_has_sleeper(wq))
2923		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2924						EPOLLRDNORM | EPOLLRDBAND);
2925	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2926	rcu_read_unlock();
2927}
2928
2929static void sock_def_write_space(struct sock *sk)
2930{
2931	struct socket_wq *wq;
2932
2933	rcu_read_lock();
2934
2935	/* Do not wake up a writer until he can make "significant"
2936	 * progress.  --DaveM
2937	 */
2938	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2939		wq = rcu_dereference(sk->sk_wq);
2940		if (skwq_has_sleeper(wq))
2941			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2942						EPOLLWRNORM | EPOLLWRBAND);
2943
2944		/* Should agree with poll, otherwise some programs break */
2945		if (sock_writeable(sk))
2946			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2947	}
2948
2949	rcu_read_unlock();
2950}
2951
2952static void sock_def_destruct(struct sock *sk)
2953{
2954}
2955
2956void sk_send_sigurg(struct sock *sk)
2957{
2958	if (sk->sk_socket && sk->sk_socket->file)
2959		if (send_sigurg(&sk->sk_socket->file->f_owner))
2960			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2961}
2962EXPORT_SYMBOL(sk_send_sigurg);
2963
2964void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2965		    unsigned long expires)
2966{
2967	if (!mod_timer(timer, expires))
2968		sock_hold(sk);
2969}
2970EXPORT_SYMBOL(sk_reset_timer);
2971
2972void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2973{
2974	if (del_timer(timer))
2975		__sock_put(sk);
2976}
2977EXPORT_SYMBOL(sk_stop_timer);
2978
2979void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
2980{
2981	if (del_timer_sync(timer))
2982		__sock_put(sk);
2983}
2984EXPORT_SYMBOL(sk_stop_timer_sync);
2985
2986void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
2987{
2988	sk_init_common(sk);
2989	sk->sk_send_head	=	NULL;
2990
2991	timer_setup(&sk->sk_timer, NULL, 0);
2992
2993	sk->sk_allocation	=	GFP_KERNEL;
2994	sk->sk_rcvbuf		=	READ_ONCE(sysctl_rmem_default);
2995	sk->sk_sndbuf		=	READ_ONCE(sysctl_wmem_default);
2996	sk->sk_state		=	TCP_CLOSE;
2997	sk_set_socket(sk, sock);
2998
2999	sock_set_flag(sk, SOCK_ZAPPED);
3000
3001	if (sock) {
3002		sk->sk_type	=	sock->type;
3003		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3004		sock->sk	=	sk;
3005	} else {
3006		RCU_INIT_POINTER(sk->sk_wq, NULL);
3007	}
3008	sk->sk_uid	=	uid;
3009
3010	rwlock_init(&sk->sk_callback_lock);
3011	if (sk->sk_kern_sock)
3012		lockdep_set_class_and_name(
3013			&sk->sk_callback_lock,
3014			af_kern_callback_keys + sk->sk_family,
3015			af_family_kern_clock_key_strings[sk->sk_family]);
3016	else
3017		lockdep_set_class_and_name(
3018			&sk->sk_callback_lock,
3019			af_callback_keys + sk->sk_family,
3020			af_family_clock_key_strings[sk->sk_family]);
3021
3022	sk->sk_state_change	=	sock_def_wakeup;
3023	sk->sk_data_ready	=	sock_def_readable;
3024	sk->sk_write_space	=	sock_def_write_space;
3025	sk->sk_error_report	=	sock_def_error_report;
3026	sk->sk_destruct		=	sock_def_destruct;
3027
3028	sk->sk_frag.page	=	NULL;
3029	sk->sk_frag.offset	=	0;
3030	sk->sk_peek_off		=	-1;
3031
3032	sk->sk_peer_pid 	=	NULL;
3033	sk->sk_peer_cred	=	NULL;
3034	spin_lock_init(&sk->sk_peer_lock);
3035
3036	sk->sk_write_pending	=	0;
3037	sk->sk_rcvlowat		=	1;
3038	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3039	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3040
3041	sk->sk_stamp = SK_DEFAULT_STAMP;
3042#if BITS_PER_LONG==32
3043	seqlock_init(&sk->sk_stamp_seq);
3044#endif
3045	atomic_set(&sk->sk_zckey, 0);
3046
3047#ifdef CONFIG_NET_RX_BUSY_POLL
3048	sk->sk_napi_id		=	0;
3049	sk->sk_ll_usec		=	READ_ONCE(sysctl_net_busy_read);
3050#endif
3051
3052	sk->sk_max_pacing_rate = ~0UL;
3053	sk->sk_pacing_rate = ~0UL;
3054	WRITE_ONCE(sk->sk_pacing_shift, 10);
3055	sk->sk_incoming_cpu = -1;
3056
3057	sk_rx_queue_clear(sk);
3058	/*
3059	 * Before updating sk_refcnt, we must commit prior changes to memory
3060	 * (Documentation/RCU/rculist_nulls.rst for details)
3061	 */
3062	smp_wmb();
3063	refcount_set(&sk->sk_refcnt, 1);
3064	atomic_set(&sk->sk_drops, 0);
3065}
3066EXPORT_SYMBOL(sock_init_data_uid);
3067
3068void sock_init_data(struct socket *sock, struct sock *sk)
3069{
3070	kuid_t uid = sock ?
3071		SOCK_INODE(sock)->i_uid :
3072		make_kuid(sock_net(sk)->user_ns, 0);
3073
3074	sock_init_data_uid(sock, sk, uid);
3075}
3076EXPORT_SYMBOL(sock_init_data);
3077
3078void lock_sock_nested(struct sock *sk, int subclass)
3079{
3080	might_sleep();
3081	spin_lock_bh(&sk->sk_lock.slock);
3082	if (sk->sk_lock.owned)
3083		__lock_sock(sk);
3084	sk->sk_lock.owned = 1;
3085	spin_unlock(&sk->sk_lock.slock);
3086	/*
3087	 * The sk_lock has mutex_lock() semantics here:
3088	 */
3089	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3090	local_bh_enable();
3091}
3092EXPORT_SYMBOL(lock_sock_nested);
3093
3094void release_sock(struct sock *sk)
3095{
3096	spin_lock_bh(&sk->sk_lock.slock);
3097	if (sk->sk_backlog.tail)
3098		__release_sock(sk);
3099
3100	/* Warning : release_cb() might need to release sk ownership,
3101	 * ie call sock_release_ownership(sk) before us.
3102	 */
3103	if (sk->sk_prot->release_cb)
3104		sk->sk_prot->release_cb(sk);
3105
3106	sock_release_ownership(sk);
3107	if (waitqueue_active(&sk->sk_lock.wq))
3108		wake_up(&sk->sk_lock.wq);
3109	spin_unlock_bh(&sk->sk_lock.slock);
3110}
3111EXPORT_SYMBOL(release_sock);
3112
3113/**
3114 * lock_sock_fast - fast version of lock_sock
3115 * @sk: socket
3116 *
3117 * This version should be used for very small section, where process wont block
3118 * return false if fast path is taken:
3119 *
3120 *   sk_lock.slock locked, owned = 0, BH disabled
3121 *
3122 * return true if slow path is taken:
3123 *
3124 *   sk_lock.slock unlocked, owned = 1, BH enabled
3125 */
3126bool lock_sock_fast(struct sock *sk)
3127{
3128	might_sleep();
3129	spin_lock_bh(&sk->sk_lock.slock);
3130
3131	if (!sk->sk_lock.owned)
3132		/*
3133		 * Note : We must disable BH
3134		 */
3135		return false;
3136
3137	__lock_sock(sk);
3138	sk->sk_lock.owned = 1;
3139	spin_unlock(&sk->sk_lock.slock);
3140	/*
3141	 * The sk_lock has mutex_lock() semantics here:
3142	 */
3143	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3144	local_bh_enable();
3145	return true;
3146}
3147EXPORT_SYMBOL(lock_sock_fast);
3148
3149int sock_gettstamp(struct socket *sock, void __user *userstamp,
3150		   bool timeval, bool time32)
3151{
3152	struct sock *sk = sock->sk;
3153	struct timespec64 ts;
3154
3155	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3156	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3157	if (ts.tv_sec == -1)
3158		return -ENOENT;
3159	if (ts.tv_sec == 0) {
3160		ktime_t kt = ktime_get_real();
3161		sock_write_timestamp(sk, kt);
3162		ts = ktime_to_timespec64(kt);
3163	}
3164
3165	if (timeval)
3166		ts.tv_nsec /= 1000;
3167
3168#ifdef CONFIG_COMPAT_32BIT_TIME
3169	if (time32)
3170		return put_old_timespec32(&ts, userstamp);
3171#endif
3172#ifdef CONFIG_SPARC64
3173	/* beware of padding in sparc64 timeval */
3174	if (timeval && !in_compat_syscall()) {
3175		struct __kernel_old_timeval __user tv = {
3176			.tv_sec = ts.tv_sec,
3177			.tv_usec = ts.tv_nsec,
3178		};
3179		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3180			return -EFAULT;
3181		return 0;
3182	}
3183#endif
3184	return put_timespec64(&ts, userstamp);
3185}
3186EXPORT_SYMBOL(sock_gettstamp);
3187
3188void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3189{
3190	if (!sock_flag(sk, flag)) {
3191		unsigned long previous_flags = sk->sk_flags;
3192
3193		sock_set_flag(sk, flag);
3194		/*
3195		 * we just set one of the two flags which require net
3196		 * time stamping, but time stamping might have been on
3197		 * already because of the other one
3198		 */
3199		if (sock_needs_netstamp(sk) &&
3200		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3201			net_enable_timestamp();
3202	}
3203}
3204
3205int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3206		       int level, int type)
3207{
3208	struct sock_exterr_skb *serr;
3209	struct sk_buff *skb;
3210	int copied, err;
3211
3212	err = -EAGAIN;
3213	skb = sock_dequeue_err_skb(sk);
3214	if (skb == NULL)
3215		goto out;
3216
3217	copied = skb->len;
3218	if (copied > len) {
3219		msg->msg_flags |= MSG_TRUNC;
3220		copied = len;
3221	}
3222	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3223	if (err)
3224		goto out_free_skb;
3225
3226	sock_recv_timestamp(msg, sk, skb);
3227
3228	serr = SKB_EXT_ERR(skb);
3229	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3230
3231	msg->msg_flags |= MSG_ERRQUEUE;
3232	err = copied;
3233
3234out_free_skb:
3235	kfree_skb(skb);
3236out:
3237	return err;
3238}
3239EXPORT_SYMBOL(sock_recv_errqueue);
3240
3241/*
3242 *	Get a socket option on an socket.
3243 *
3244 *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3245 *	asynchronous errors should be reported by getsockopt. We assume
3246 *	this means if you specify SO_ERROR (otherwise whats the point of it).
3247 */
3248int sock_common_getsockopt(struct socket *sock, int level, int optname,
3249			   char __user *optval, int __user *optlen)
3250{
3251	struct sock *sk = sock->sk;
3252
3253	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
3254	return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3255}
3256EXPORT_SYMBOL(sock_common_getsockopt);
3257
3258int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3259			int flags)
3260{
3261	struct sock *sk = sock->sk;
3262	int addr_len = 0;
3263	int err;
3264
3265	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3266				   flags & ~MSG_DONTWAIT, &addr_len);
3267	if (err >= 0)
3268		msg->msg_namelen = addr_len;
3269	return err;
3270}
3271EXPORT_SYMBOL(sock_common_recvmsg);
3272
3273/*
3274 *	Set socket options on an inet socket.
3275 */
3276int sock_common_setsockopt(struct socket *sock, int level, int optname,
3277			   sockptr_t optval, unsigned int optlen)
3278{
3279	struct sock *sk = sock->sk;
3280
3281	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
3282	return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3283}
3284EXPORT_SYMBOL(sock_common_setsockopt);
3285
3286void sk_common_release(struct sock *sk)
3287{
3288	if (sk->sk_prot->destroy)
3289		sk->sk_prot->destroy(sk);
3290
3291	/*
3292	 * Observation: when sk_common_release is called, processes have
3293	 * no access to socket. But net still has.
3294	 * Step one, detach it from networking:
3295	 *
3296	 * A. Remove from hash tables.
3297	 */
3298
3299	sk->sk_prot->unhash(sk);
3300
3301	/*
3302	 * In this point socket cannot receive new packets, but it is possible
3303	 * that some packets are in flight because some CPU runs receiver and
3304	 * did hash table lookup before we unhashed socket. They will achieve
3305	 * receive queue and will be purged by socket destructor.
3306	 *
3307	 * Also we still have packets pending on receive queue and probably,
3308	 * our own packets waiting in device queues. sock_destroy will drain
3309	 * receive queue, but transmitted packets will delay socket destruction
3310	 * until the last reference will be released.
3311	 */
3312
3313	sock_orphan(sk);
3314
3315	xfrm_sk_free_policy(sk);
3316
3317	sk_refcnt_debug_release(sk);
3318
3319	sock_put(sk);
3320}
3321EXPORT_SYMBOL(sk_common_release);
3322
3323void sk_get_meminfo(const struct sock *sk, u32 *mem)
3324{
3325	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3326
3327	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3328	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3329	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3330	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3331	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3332	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3333	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3334	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3335	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3336}
3337
3338#ifdef CONFIG_PROC_FS
3339#define PROTO_INUSE_NR	64	/* should be enough for the first time */
3340struct prot_inuse {
3341	int val[PROTO_INUSE_NR];
3342};
3343
3344static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3345
3346void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3347{
3348	__this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3349}
3350EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3351
3352int sock_prot_inuse_get(struct net *net, struct proto *prot)
3353{
3354	int cpu, idx = prot->inuse_idx;
3355	int res = 0;
3356
3357	for_each_possible_cpu(cpu)
3358		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3359
3360	return res >= 0 ? res : 0;
3361}
3362EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3363
3364static void sock_inuse_add(struct net *net, int val)
3365{
3366	this_cpu_add(*net->core.sock_inuse, val);
3367}
3368
3369int sock_inuse_get(struct net *net)
3370{
3371	int cpu, res = 0;
3372
3373	for_each_possible_cpu(cpu)
3374		res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3375
3376	return res;
3377}
3378
3379EXPORT_SYMBOL_GPL(sock_inuse_get);
3380
3381static int __net_init sock_inuse_init_net(struct net *net)
3382{
3383	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3384	if (net->core.prot_inuse == NULL)
3385		return -ENOMEM;
3386
3387	net->core.sock_inuse = alloc_percpu(int);
3388	if (net->core.sock_inuse == NULL)
3389		goto out;
3390
3391	return 0;
3392
3393out:
3394	free_percpu(net->core.prot_inuse);
3395	return -ENOMEM;
3396}
3397
3398static void __net_exit sock_inuse_exit_net(struct net *net)
3399{
3400	free_percpu(net->core.prot_inuse);
3401	free_percpu(net->core.sock_inuse);
3402}
3403
3404static struct pernet_operations net_inuse_ops = {
3405	.init = sock_inuse_init_net,
3406	.exit = sock_inuse_exit_net,
3407};
3408
3409static __init int net_inuse_init(void)
3410{
3411	if (register_pernet_subsys(&net_inuse_ops))
3412		panic("Cannot initialize net inuse counters");
3413
3414	return 0;
3415}
3416
3417core_initcall(net_inuse_init);
3418
3419static int assign_proto_idx(struct proto *prot)
3420{
3421	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3422
3423	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3424		pr_err("PROTO_INUSE_NR exhausted\n");
3425		return -ENOSPC;
3426	}
3427
3428	set_bit(prot->inuse_idx, proto_inuse_idx);
3429	return 0;
3430}
3431
3432static void release_proto_idx(struct proto *prot)
3433{
3434	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3435		clear_bit(prot->inuse_idx, proto_inuse_idx);
3436}
3437#else
3438static inline int assign_proto_idx(struct proto *prot)
3439{
3440	return 0;
3441}
3442
3443static inline void release_proto_idx(struct proto *prot)
3444{
3445}
3446
3447static void sock_inuse_add(struct net *net, int val)
3448{
3449}
3450#endif
3451
3452static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3453{
3454	if (!twsk_prot)
3455		return;
3456	kfree(twsk_prot->twsk_slab_name);
3457	twsk_prot->twsk_slab_name = NULL;
3458	kmem_cache_destroy(twsk_prot->twsk_slab);
3459	twsk_prot->twsk_slab = NULL;
3460}
3461
3462static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3463{
3464	if (!rsk_prot)
3465		return;
3466	kfree(rsk_prot->slab_name);
3467	rsk_prot->slab_name = NULL;
3468	kmem_cache_destroy(rsk_prot->slab);
3469	rsk_prot->slab = NULL;
3470}
3471
3472static int req_prot_init(const struct proto *prot)
3473{
3474	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3475
3476	if (!rsk_prot)
3477		return 0;
3478
3479	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3480					prot->name);
3481	if (!rsk_prot->slab_name)
3482		return -ENOMEM;
3483
3484	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3485					   rsk_prot->obj_size, 0,
3486					   SLAB_ACCOUNT | prot->slab_flags,
3487					   NULL);
3488
3489	if (!rsk_prot->slab) {
3490		pr_crit("%s: Can't create request sock SLAB cache!\n",
3491			prot->name);
3492		return -ENOMEM;
3493	}
3494	return 0;
3495}
3496
3497int proto_register(struct proto *prot, int alloc_slab)
3498{
3499	int ret = -ENOBUFS;
3500
3501	if (alloc_slab) {
3502		prot->slab = kmem_cache_create_usercopy(prot->name,
3503					prot->obj_size, 0,
3504					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3505					prot->slab_flags,
3506					prot->useroffset, prot->usersize,
3507					NULL);
3508
3509		if (prot->slab == NULL) {
3510			pr_crit("%s: Can't create sock SLAB cache!\n",
3511				prot->name);
3512			goto out;
3513		}
3514
3515		if (req_prot_init(prot))
3516			goto out_free_request_sock_slab;
3517
3518		if (prot->twsk_prot != NULL) {
3519			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3520
3521			if (prot->twsk_prot->twsk_slab_name == NULL)
3522				goto out_free_request_sock_slab;
3523
3524			prot->twsk_prot->twsk_slab =
3525				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3526						  prot->twsk_prot->twsk_obj_size,
3527						  0,
3528						  SLAB_ACCOUNT |
3529						  prot->slab_flags,
3530						  NULL);
3531			if (prot->twsk_prot->twsk_slab == NULL)
3532				goto out_free_timewait_sock_slab;
3533		}
3534	}
3535
3536	mutex_lock(&proto_list_mutex);
3537	ret = assign_proto_idx(prot);
3538	if (ret) {
3539		mutex_unlock(&proto_list_mutex);
3540		goto out_free_timewait_sock_slab;
3541	}
3542	list_add(&prot->node, &proto_list);
3543	mutex_unlock(&proto_list_mutex);
3544	return ret;
3545
3546out_free_timewait_sock_slab:
3547	if (alloc_slab && prot->twsk_prot)
3548		tw_prot_cleanup(prot->twsk_prot);
3549out_free_request_sock_slab:
3550	if (alloc_slab) {
3551		req_prot_cleanup(prot->rsk_prot);
3552
3553		kmem_cache_destroy(prot->slab);
3554		prot->slab = NULL;
3555	}
3556out:
3557	return ret;
3558}
3559EXPORT_SYMBOL(proto_register);
3560
3561void proto_unregister(struct proto *prot)
3562{
3563	mutex_lock(&proto_list_mutex);
3564	release_proto_idx(prot);
3565	list_del(&prot->node);
3566	mutex_unlock(&proto_list_mutex);
3567
3568	kmem_cache_destroy(prot->slab);
3569	prot->slab = NULL;
3570
3571	req_prot_cleanup(prot->rsk_prot);
3572	tw_prot_cleanup(prot->twsk_prot);
3573}
3574EXPORT_SYMBOL(proto_unregister);
3575
3576int sock_load_diag_module(int family, int protocol)
3577{
3578	if (!protocol) {
3579		if (!sock_is_registered(family))
3580			return -ENOENT;
3581
3582		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3583				      NETLINK_SOCK_DIAG, family);
3584	}
3585
3586#ifdef CONFIG_INET
3587	if (family == AF_INET &&
3588	    protocol != IPPROTO_RAW &&
3589	    protocol < MAX_INET_PROTOS &&
3590	    !rcu_access_pointer(inet_protos[protocol]))
3591		return -ENOENT;
3592#endif
3593
3594	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3595			      NETLINK_SOCK_DIAG, family, protocol);
3596}
3597EXPORT_SYMBOL(sock_load_diag_module);
3598
3599#ifdef CONFIG_PROC_FS
3600static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3601	__acquires(proto_list_mutex)
3602{
3603	mutex_lock(&proto_list_mutex);
3604	return seq_list_start_head(&proto_list, *pos);
3605}
3606
3607static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3608{
3609	return seq_list_next(v, &proto_list, pos);
3610}
3611
3612static void proto_seq_stop(struct seq_file *seq, void *v)
3613	__releases(proto_list_mutex)
3614{
3615	mutex_unlock(&proto_list_mutex);
3616}
3617
3618static char proto_method_implemented(const void *method)
3619{
3620	return method == NULL ? 'n' : 'y';
3621}
3622static long sock_prot_memory_allocated(struct proto *proto)
3623{
3624	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3625}
3626
3627static const char *sock_prot_memory_pressure(struct proto *proto)
3628{
3629	return proto->memory_pressure != NULL ?
3630	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3631}
3632
3633static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3634{
3635
3636	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3637			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3638		   proto->name,
3639		   proto->obj_size,
3640		   sock_prot_inuse_get(seq_file_net(seq), proto),
3641		   sock_prot_memory_allocated(proto),
3642		   sock_prot_memory_pressure(proto),
3643		   proto->max_header,
3644		   proto->slab == NULL ? "no" : "yes",
3645		   module_name(proto->owner),
3646		   proto_method_implemented(proto->close),
3647		   proto_method_implemented(proto->connect),
3648		   proto_method_implemented(proto->disconnect),
3649		   proto_method_implemented(proto->accept),
3650		   proto_method_implemented(proto->ioctl),
3651		   proto_method_implemented(proto->init),
3652		   proto_method_implemented(proto->destroy),
3653		   proto_method_implemented(proto->shutdown),
3654		   proto_method_implemented(proto->setsockopt),
3655		   proto_method_implemented(proto->getsockopt),
3656		   proto_method_implemented(proto->sendmsg),
3657		   proto_method_implemented(proto->recvmsg),
3658		   proto_method_implemented(proto->sendpage),
3659		   proto_method_implemented(proto->bind),
3660		   proto_method_implemented(proto->backlog_rcv),
3661		   proto_method_implemented(proto->hash),
3662		   proto_method_implemented(proto->unhash),
3663		   proto_method_implemented(proto->get_port),
3664		   proto_method_implemented(proto->enter_memory_pressure));
3665}
3666
3667static int proto_seq_show(struct seq_file *seq, void *v)
3668{
3669	if (v == &proto_list)
3670		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3671			   "protocol",
3672			   "size",
3673			   "sockets",
3674			   "memory",
3675			   "press",
3676			   "maxhdr",
3677			   "slab",
3678			   "module",
3679			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3680	else
3681		proto_seq_printf(seq, list_entry(v, struct proto, node));
3682	return 0;
3683}
3684
3685static const struct seq_operations proto_seq_ops = {
3686	.start  = proto_seq_start,
3687	.next   = proto_seq_next,
3688	.stop   = proto_seq_stop,
3689	.show   = proto_seq_show,
3690};
3691
3692static __net_init int proto_init_net(struct net *net)
3693{
3694	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3695			sizeof(struct seq_net_private)))
3696		return -ENOMEM;
3697
3698	return 0;
3699}
3700
3701static __net_exit void proto_exit_net(struct net *net)
3702{
3703	remove_proc_entry("protocols", net->proc_net);
3704}
3705
3706
3707static __net_initdata struct pernet_operations proto_net_ops = {
3708	.init = proto_init_net,
3709	.exit = proto_exit_net,
3710};
3711
3712static int __init proto_init(void)
3713{
3714	return register_pernet_subsys(&proto_net_ops);
3715}
3716
3717subsys_initcall(proto_init);
3718
3719#endif /* PROC_FS */
3720
3721#ifdef CONFIG_NET_RX_BUSY_POLL
3722bool sk_busy_loop_end(void *p, unsigned long start_time)
3723{
3724	struct sock *sk = p;
3725
3726	return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3727	       sk_busy_loop_timeout(sk, start_time);
3728}
3729EXPORT_SYMBOL(sk_busy_loop_end);
3730#endif /* CONFIG_NET_RX_BUSY_POLL */
3731
3732int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3733{
3734	if (!sk->sk_prot->bind_add)
3735		return -EOPNOTSUPP;
3736	return sk->sk_prot->bind_add(sk, addr, addr_len);
3737}
3738EXPORT_SYMBOL(sock_bind_add);
3739