1/*
2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
3 *		operating system.  INET is implemented using the  BSD Socket
4 *		interface as the means of communication with the user level.
5 *
6 *		Generic socket support routines. Memory allocators, socket lock/release
7 *		handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors:	Ross Biro
11 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *		Florian La Roche, <flla@stud.uni-sb.de>
13 *		Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 *		Alan Cox	: 	Numerous verify_area() problems
17 *		Alan Cox	:	Connecting on a connecting socket
18 *					now returns an error for tcp.
19 *		Alan Cox	:	sock->protocol is set correctly.
20 *					and is not sometimes left as 0.
21 *		Alan Cox	:	connect handles icmp errors on a
22 *					connect properly. Unfortunately there
23 *					is a restart syscall nasty there. I
24 *					can't match BSD without hacking the C
25 *					library. Ideas urgently sought!
26 *		Alan Cox	:	Disallow bind() to addresses that are
27 *					not ours - especially broadcast ones!!
28 *		Alan Cox	:	Socket 1024 _IS_ ok for users. (fencepost)
29 *		Alan Cox	:	sock_wfree/sock_rfree don't destroy sockets,
30 *					instead they leave that for the DESTROY timer.
31 *		Alan Cox	:	Clean up error flag in accept
32 *		Alan Cox	:	TCP ack handling is buggy, the DESTROY timer
33 *					was buggy. Put a remove_sock() in the handler
34 *					for memory when we hit 0. Also altered the timer
35 *					code. The ACK stuff can wait and needs major
36 *					TCP layer surgery.
37 *		Alan Cox	:	Fixed TCP ack bug, removed remove sock
38 *					and fixed timer/inet_bh race.
39 *		Alan Cox	:	Added zapped flag for TCP
40 *		Alan Cox	:	Move kfree_skb into skbuff.c and tidied up surplus code
41 *		Alan Cox	:	for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 *		Alan Cox	:	kfree_s calls now are kfree_skbmem so we can track skb resources
43 *		Alan Cox	:	Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 *		Alan Cox	:	Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 *		Rick Sladkey	:	Relaxed UDP rules for matching packets.
46 *		C.E.Hawkins	:	IFF_PROMISC/SIOCGHWADDR support
47 *	Pauline Middelink	:	identd support
48 *		Alan Cox	:	Fixed connect() taking signals I think.
49 *		Alan Cox	:	SO_LINGER supported
50 *		Alan Cox	:	Error reporting fixes
51 *		Anonymous	:	inet_create tidied up (sk->reuse setting)
52 *		Alan Cox	:	inet sockets don't set sk->type!
53 *		Alan Cox	:	Split socket option code
54 *		Alan Cox	:	Callbacks
55 *		Alan Cox	:	Nagle flag for Charles & Johannes stuff
56 *		Alex		:	Removed restriction on inet fioctl
57 *		Alan Cox	:	Splitting INET from NET core
58 *		Alan Cox	:	Fixed bogus SO_TYPE handling in getsockopt()
59 *		Adam Caldwell	:	Missing return in SO_DONTROUTE/SO_DEBUG code
60 *		Alan Cox	:	Split IP from generic code
61 *		Alan Cox	:	New kfree_skbmem()
62 *		Alan Cox	:	Make SO_DEBUG superuser only.
63 *		Alan Cox	:	Allow anyone to clear SO_DEBUG
64 *					(compatibility fix)
65 *		Alan Cox	:	Added optimistic memory grabbing for AF_UNIX throughput.
66 *		Alan Cox	:	Allocator for a socket is settable.
67 *		Alan Cox	:	SO_ERROR includes soft errors.
68 *		Alan Cox	:	Allow NULL arguments on some SO_ opts
69 *		Alan Cox	: 	Generic socket allocation to make hooks
70 *					easier (suggested by Craig Metz).
71 *		Michael Pall	:	SO_ERROR returns positive errno again
72 *              Steve Whitehouse:       Added default destructor to free
73 *                                      protocol private data.
74 *              Steve Whitehouse:       Added various other default routines
75 *                                      common to several socket families.
76 *              Chris Evans     :       Call suser() check last on F_SETOWN
77 *		Jay Schulist	:	Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 *		Andi Kleen	:	Add sock_kmalloc()/sock_kfree_s()
79 *		Andi Kleen	:	Fix write_space callback
80 *		Chris Evans	:	Security fixes - signedness again
81 *		Arnaldo C. Melo :       cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 *		This program is free software; you can redistribute it and/or
87 *		modify it under the terms of the GNU General Public License
88 *		as published by the Free Software Foundation; either version
89 *		2 of the License, or (at your option) any later version.
90 */
91
92#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94#include <linux/capability.h>
95#include <linux/errno.h>
96#include <linux/errqueue.h>
97#include <linux/types.h>
98#include <linux/socket.h>
99#include <linux/in.h>
100#include <linux/kernel.h>
101#include <linux/module.h>
102#include <linux/proc_fs.h>
103#include <linux/seq_file.h>
104#include <linux/sched.h>
105#include <linux/timer.h>
106#include <linux/string.h>
107#include <linux/sockios.h>
108#include <linux/net.h>
109#include <linux/mm.h>
110#include <linux/slab.h>
111#include <linux/interrupt.h>
112#include <linux/poll.h>
113#include <linux/tcp.h>
114#include <linux/init.h>
115#include <linux/highmem.h>
116#include <linux/user_namespace.h>
117#include <linux/static_key.h>
118#include <linux/memcontrol.h>
119#include <linux/prefetch.h>
120
121#include <asm/uaccess.h>
122
123#include <linux/netdevice.h>
124#include <net/protocol.h>
125#include <linux/skbuff.h>
126#include <net/net_namespace.h>
127#include <net/request_sock.h>
128#include <net/sock.h>
129#include <linux/net_tstamp.h>
130#include <net/xfrm.h>
131#include <linux/ipsec.h>
132#include <net/cls_cgroup.h>
133#include <net/netprio_cgroup.h>
134
135#include <linux/filter.h>
136
137#include <trace/events/sock.h>
138
139#ifdef CONFIG_INET
140#include <net/tcp.h>
141#endif
142
143#include <net/busy_poll.h>
144
145static DEFINE_MUTEX(proto_list_mutex);
146static LIST_HEAD(proto_list);
147
148/**
149 * sk_ns_capable - General socket capability test
150 * @sk: Socket to use a capability on or through
151 * @user_ns: The user namespace of the capability to use
152 * @cap: The capability to use
153 *
154 * Test to see if the opener of the socket had when the socket was
155 * created and the current process has the capability @cap in the user
156 * namespace @user_ns.
157 */
158bool sk_ns_capable(const struct sock *sk,
159		   struct user_namespace *user_ns, int cap)
160{
161	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162		ns_capable(user_ns, cap);
163}
164EXPORT_SYMBOL(sk_ns_capable);
165
166/**
167 * sk_capable - Socket global capability test
168 * @sk: Socket to use a capability on or through
169 * @cap: The global capability to use
170 *
171 * Test to see if the opener of the socket had when the socket was
172 * created and the current process has the capability @cap in all user
173 * namespaces.
174 */
175bool sk_capable(const struct sock *sk, int cap)
176{
177	return sk_ns_capable(sk, &init_user_ns, cap);
178}
179EXPORT_SYMBOL(sk_capable);
180
181/**
182 * sk_net_capable - Network namespace socket capability test
183 * @sk: Socket to use a capability on or through
184 * @cap: The capability to use
185 *
186 * Test to see if the opener of the socket had when the socket was created
187 * and the current process has the capability @cap over the network namespace
188 * the socket is a member of.
189 */
190bool sk_net_capable(const struct sock *sk, int cap)
191{
192	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193}
194EXPORT_SYMBOL(sk_net_capable);
195
196
197#ifdef CONFIG_MEMCG_KMEM
198int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
199{
200	struct proto *proto;
201	int ret = 0;
202
203	mutex_lock(&proto_list_mutex);
204	list_for_each_entry(proto, &proto_list, node) {
205		if (proto->init_cgroup) {
206			ret = proto->init_cgroup(memcg, ss);
207			if (ret)
208				goto out;
209		}
210	}
211
212	mutex_unlock(&proto_list_mutex);
213	return ret;
214out:
215	list_for_each_entry_continue_reverse(proto, &proto_list, node)
216		if (proto->destroy_cgroup)
217			proto->destroy_cgroup(memcg);
218	mutex_unlock(&proto_list_mutex);
219	return ret;
220}
221
222void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
223{
224	struct proto *proto;
225
226	mutex_lock(&proto_list_mutex);
227	list_for_each_entry_reverse(proto, &proto_list, node)
228		if (proto->destroy_cgroup)
229			proto->destroy_cgroup(memcg);
230	mutex_unlock(&proto_list_mutex);
231}
232#endif
233
234/*
235 * Each address family might have different locking rules, so we have
236 * one slock key per address family:
237 */
238static struct lock_class_key af_family_keys[AF_MAX];
239static struct lock_class_key af_family_slock_keys[AF_MAX];
240
241#if defined(CONFIG_MEMCG_KMEM)
242struct static_key memcg_socket_limit_enabled;
243EXPORT_SYMBOL(memcg_socket_limit_enabled);
244#endif
245
246/*
247 * Make lock validator output more readable. (we pre-construct these
248 * strings build-time, so that runtime initialization of socket
249 * locks is fast):
250 */
251static const char *const af_family_key_strings[AF_MAX+1] = {
252  "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
253  "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
254  "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
255  "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
256  "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
257  "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
258  "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
259  "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
260  "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
261  "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
262  "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
263  "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
264  "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
265  "sk_lock-AF_NFC"   , "sk_lock-AF_VSOCK"    , "sk_lock-AF_MAX"
266};
267static const char *const af_family_slock_key_strings[AF_MAX+1] = {
268  "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
269  "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
270  "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
271  "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
272  "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
273  "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
274  "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
275  "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
276  "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
277  "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
278  "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
279  "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
280  "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
281  "slock-AF_NFC"   , "slock-AF_VSOCK"    ,"slock-AF_MAX"
282};
283static const char *const af_family_clock_key_strings[AF_MAX+1] = {
284  "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
285  "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
286  "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
287  "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
288  "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
289  "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
290  "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
291  "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
292  "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
293  "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
294  "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
295  "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
296  "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
297  "clock-AF_NFC"   , "clock-AF_VSOCK"    , "clock-AF_MAX"
298};
299
300/*
301 * sk_callback_lock locking rules are per-address-family,
302 * so split the lock classes by using a per-AF key:
303 */
304static struct lock_class_key af_callback_keys[AF_MAX];
305
306/* Take into consideration the size of the struct sk_buff overhead in the
307 * determination of these values, since that is non-constant across
308 * platforms.  This makes socket queueing behavior and performance
309 * not depend upon such differences.
310 */
311#define _SK_MEM_PACKETS		256
312#define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
313#define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
314#define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315
316/* Run time adjustable parameters. */
317__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
318EXPORT_SYMBOL(sysctl_wmem_max);
319__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
320EXPORT_SYMBOL(sysctl_rmem_max);
321__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
322__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
323
324/* Maximal space eaten by iovec or ancillary data plus some space */
325int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
326EXPORT_SYMBOL(sysctl_optmem_max);
327
328int sysctl_tstamp_allow_data __read_mostly = 1;
329
330struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
331EXPORT_SYMBOL_GPL(memalloc_socks);
332
333/**
334 * sk_set_memalloc - sets %SOCK_MEMALLOC
335 * @sk: socket to set it on
336 *
337 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
338 * It's the responsibility of the admin to adjust min_free_kbytes
339 * to meet the requirements
340 */
341void sk_set_memalloc(struct sock *sk)
342{
343	sock_set_flag(sk, SOCK_MEMALLOC);
344	sk->sk_allocation |= __GFP_MEMALLOC;
345	static_key_slow_inc(&memalloc_socks);
346}
347EXPORT_SYMBOL_GPL(sk_set_memalloc);
348
349void sk_clear_memalloc(struct sock *sk)
350{
351	sock_reset_flag(sk, SOCK_MEMALLOC);
352	sk->sk_allocation &= ~__GFP_MEMALLOC;
353	static_key_slow_dec(&memalloc_socks);
354
355	/*
356	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
357	 * progress of swapping. SOCK_MEMALLOC may be cleared while
358	 * it has rmem allocations due to the last swapfile being deactivated
359	 * but there is a risk that the socket is unusable due to exceeding
360	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
361	 */
362	sk_mem_reclaim(sk);
363}
364EXPORT_SYMBOL_GPL(sk_clear_memalloc);
365
366int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
367{
368	int ret;
369	unsigned long pflags = current->flags;
370
371	/* these should have been dropped before queueing */
372	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
373
374	current->flags |= PF_MEMALLOC;
375	ret = sk->sk_backlog_rcv(sk, skb);
376	tsk_restore_flags(current, pflags, PF_MEMALLOC);
377
378	return ret;
379}
380EXPORT_SYMBOL(__sk_backlog_rcv);
381
382static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
383{
384	struct timeval tv;
385
386	if (optlen < sizeof(tv))
387		return -EINVAL;
388	if (copy_from_user(&tv, optval, sizeof(tv)))
389		return -EFAULT;
390	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
391		return -EDOM;
392
393	if (tv.tv_sec < 0) {
394		static int warned __read_mostly;
395
396		*timeo_p = 0;
397		if (warned < 10 && net_ratelimit()) {
398			warned++;
399			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
400				__func__, current->comm, task_pid_nr(current));
401		}
402		return 0;
403	}
404	*timeo_p = MAX_SCHEDULE_TIMEOUT;
405	if (tv.tv_sec == 0 && tv.tv_usec == 0)
406		return 0;
407	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
408		*timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
409	return 0;
410}
411
412static void sock_warn_obsolete_bsdism(const char *name)
413{
414	static int warned;
415	static char warncomm[TASK_COMM_LEN];
416	if (strcmp(warncomm, current->comm) && warned < 5) {
417		strcpy(warncomm,  current->comm);
418		pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
419			warncomm, name);
420		warned++;
421	}
422}
423
424static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
425{
426	if (sk->sk_flags & flags) {
427		sk->sk_flags &= ~flags;
428		if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
429			net_disable_timestamp();
430	}
431}
432
433
434int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
435{
436	int err;
437	unsigned long flags;
438	struct sk_buff_head *list = &sk->sk_receive_queue;
439
440	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
441		atomic_inc(&sk->sk_drops);
442		trace_sock_rcvqueue_full(sk, skb);
443		return -ENOMEM;
444	}
445
446	err = sk_filter(sk, skb);
447	if (err)
448		return err;
449
450	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
451		atomic_inc(&sk->sk_drops);
452		return -ENOBUFS;
453	}
454
455	skb->dev = NULL;
456	skb_set_owner_r(skb, sk);
457
458	/* we escape from rcu protected region, make sure we dont leak
459	 * a norefcounted dst
460	 */
461	skb_dst_force(skb);
462
463	spin_lock_irqsave(&list->lock, flags);
464	sock_skb_set_dropcount(sk, skb);
465	__skb_queue_tail(list, skb);
466	spin_unlock_irqrestore(&list->lock, flags);
467
468	if (!sock_flag(sk, SOCK_DEAD))
469		sk->sk_data_ready(sk);
470	return 0;
471}
472EXPORT_SYMBOL(sock_queue_rcv_skb);
473
474int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
475{
476	int rc = NET_RX_SUCCESS;
477
478	if (sk_filter(sk, skb))
479		goto discard_and_relse;
480
481	skb->dev = NULL;
482
483	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
484		atomic_inc(&sk->sk_drops);
485		goto discard_and_relse;
486	}
487	if (nested)
488		bh_lock_sock_nested(sk);
489	else
490		bh_lock_sock(sk);
491	if (!sock_owned_by_user(sk)) {
492		/*
493		 * trylock + unlock semantics:
494		 */
495		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
496
497		rc = sk_backlog_rcv(sk, skb);
498
499		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
500	} else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
501		bh_unlock_sock(sk);
502		atomic_inc(&sk->sk_drops);
503		goto discard_and_relse;
504	}
505
506	bh_unlock_sock(sk);
507out:
508	sock_put(sk);
509	return rc;
510discard_and_relse:
511	kfree_skb(skb);
512	goto out;
513}
514EXPORT_SYMBOL(sk_receive_skb);
515
516struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
517{
518	struct dst_entry *dst = __sk_dst_get(sk);
519
520	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
521		sk_tx_queue_clear(sk);
522		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
523		dst_release(dst);
524		return NULL;
525	}
526
527	return dst;
528}
529EXPORT_SYMBOL(__sk_dst_check);
530
531struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
532{
533	struct dst_entry *dst = sk_dst_get(sk);
534
535	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
536		sk_dst_reset(sk);
537		dst_release(dst);
538		return NULL;
539	}
540
541	return dst;
542}
543EXPORT_SYMBOL(sk_dst_check);
544
545static int sock_setbindtodevice(struct sock *sk, char __user *optval,
546				int optlen)
547{
548	int ret = -ENOPROTOOPT;
549#ifdef CONFIG_NETDEVICES
550	struct net *net = sock_net(sk);
551	char devname[IFNAMSIZ];
552	int index;
553
554	/* Sorry... */
555	ret = -EPERM;
556	if (!ns_capable(net->user_ns, CAP_NET_RAW))
557		goto out;
558
559	ret = -EINVAL;
560	if (optlen < 0)
561		goto out;
562
563	/* Bind this socket to a particular device like "eth0",
564	 * as specified in the passed interface name. If the
565	 * name is "" or the option length is zero the socket
566	 * is not bound.
567	 */
568	if (optlen > IFNAMSIZ - 1)
569		optlen = IFNAMSIZ - 1;
570	memset(devname, 0, sizeof(devname));
571
572	ret = -EFAULT;
573	if (copy_from_user(devname, optval, optlen))
574		goto out;
575
576	index = 0;
577	if (devname[0] != '\0') {
578		struct net_device *dev;
579
580		rcu_read_lock();
581		dev = dev_get_by_name_rcu(net, devname);
582		if (dev)
583			index = dev->ifindex;
584		rcu_read_unlock();
585		ret = -ENODEV;
586		if (!dev)
587			goto out;
588	}
589
590	lock_sock(sk);
591	sk->sk_bound_dev_if = index;
592	sk_dst_reset(sk);
593	release_sock(sk);
594
595	ret = 0;
596
597out:
598#endif
599
600	return ret;
601}
602
603static int sock_getbindtodevice(struct sock *sk, char __user *optval,
604				int __user *optlen, int len)
605{
606	int ret = -ENOPROTOOPT;
607#ifdef CONFIG_NETDEVICES
608	struct net *net = sock_net(sk);
609	char devname[IFNAMSIZ];
610
611	if (sk->sk_bound_dev_if == 0) {
612		len = 0;
613		goto zero;
614	}
615
616	ret = -EINVAL;
617	if (len < IFNAMSIZ)
618		goto out;
619
620	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
621	if (ret)
622		goto out;
623
624	len = strlen(devname) + 1;
625
626	ret = -EFAULT;
627	if (copy_to_user(optval, devname, len))
628		goto out;
629
630zero:
631	ret = -EFAULT;
632	if (put_user(len, optlen))
633		goto out;
634
635	ret = 0;
636
637out:
638#endif
639
640	return ret;
641}
642
643static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
644{
645	if (valbool)
646		sock_set_flag(sk, bit);
647	else
648		sock_reset_flag(sk, bit);
649}
650
651bool sk_mc_loop(struct sock *sk)
652{
653	if (dev_recursion_level())
654		return false;
655	if (!sk)
656		return true;
657	switch (sk->sk_family) {
658	case AF_INET:
659		return inet_sk(sk)->mc_loop;
660#if IS_ENABLED(CONFIG_IPV6)
661	case AF_INET6:
662		return inet6_sk(sk)->mc_loop;
663#endif
664	}
665	WARN_ON(1);
666	return true;
667}
668EXPORT_SYMBOL(sk_mc_loop);
669
670/*
671 *	This is meant for all protocols to use and covers goings on
672 *	at the socket level. Everything here is generic.
673 */
674
675int sock_setsockopt(struct socket *sock, int level, int optname,
676		    char __user *optval, unsigned int optlen)
677{
678	struct sock *sk = sock->sk;
679	int val;
680	int valbool;
681	struct linger ling;
682	int ret = 0;
683
684	/*
685	 *	Options without arguments
686	 */
687
688	if (optname == SO_BINDTODEVICE)
689		return sock_setbindtodevice(sk, optval, optlen);
690
691	if (optlen < sizeof(int))
692		return -EINVAL;
693
694	if (get_user(val, (int __user *)optval))
695		return -EFAULT;
696
697	valbool = val ? 1 : 0;
698
699	lock_sock(sk);
700
701	switch (optname) {
702	case SO_DEBUG:
703		if (val && !capable(CAP_NET_ADMIN))
704			ret = -EACCES;
705		else
706			sock_valbool_flag(sk, SOCK_DBG, valbool);
707		break;
708	case SO_REUSEADDR:
709		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
710		break;
711	case SO_REUSEPORT:
712		sk->sk_reuseport = valbool;
713		break;
714	case SO_TYPE:
715	case SO_PROTOCOL:
716	case SO_DOMAIN:
717	case SO_ERROR:
718		ret = -ENOPROTOOPT;
719		break;
720	case SO_DONTROUTE:
721		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
722		break;
723	case SO_BROADCAST:
724		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
725		break;
726	case SO_SNDBUF:
727		/* Don't error on this BSD doesn't and if you think
728		 * about it this is right. Otherwise apps have to
729		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
730		 * are treated in BSD as hints
731		 */
732		val = min_t(u32, val, sysctl_wmem_max);
733set_sndbuf:
734		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
735		sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
736		/* Wake up sending tasks if we upped the value. */
737		sk->sk_write_space(sk);
738		break;
739
740	case SO_SNDBUFFORCE:
741		if (!capable(CAP_NET_ADMIN)) {
742			ret = -EPERM;
743			break;
744		}
745		goto set_sndbuf;
746
747	case SO_RCVBUF:
748		/* Don't error on this BSD doesn't and if you think
749		 * about it this is right. Otherwise apps have to
750		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
751		 * are treated in BSD as hints
752		 */
753		val = min_t(u32, val, sysctl_rmem_max);
754set_rcvbuf:
755		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
756		/*
757		 * We double it on the way in to account for
758		 * "struct sk_buff" etc. overhead.   Applications
759		 * assume that the SO_RCVBUF setting they make will
760		 * allow that much actual data to be received on that
761		 * socket.
762		 *
763		 * Applications are unaware that "struct sk_buff" and
764		 * other overheads allocate from the receive buffer
765		 * during socket buffer allocation.
766		 *
767		 * And after considering the possible alternatives,
768		 * returning the value we actually used in getsockopt
769		 * is the most desirable behavior.
770		 */
771		sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
772		break;
773
774	case SO_RCVBUFFORCE:
775		if (!capable(CAP_NET_ADMIN)) {
776			ret = -EPERM;
777			break;
778		}
779		goto set_rcvbuf;
780
781	case SO_KEEPALIVE:
782#ifdef CONFIG_INET
783		if (sk->sk_protocol == IPPROTO_TCP &&
784		    sk->sk_type == SOCK_STREAM)
785			tcp_set_keepalive(sk, valbool);
786#endif
787		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
788		break;
789
790	case SO_OOBINLINE:
791		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
792		break;
793
794	case SO_NO_CHECK:
795		sk->sk_no_check_tx = valbool;
796		break;
797
798	case SO_PRIORITY:
799		if ((val >= 0 && val <= 6) ||
800		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
801			sk->sk_priority = val;
802		else
803			ret = -EPERM;
804		break;
805
806	case SO_LINGER:
807		if (optlen < sizeof(ling)) {
808			ret = -EINVAL;	/* 1003.1g */
809			break;
810		}
811		if (copy_from_user(&ling, optval, sizeof(ling))) {
812			ret = -EFAULT;
813			break;
814		}
815		if (!ling.l_onoff)
816			sock_reset_flag(sk, SOCK_LINGER);
817		else {
818#if (BITS_PER_LONG == 32)
819			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
820				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
821			else
822#endif
823				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
824			sock_set_flag(sk, SOCK_LINGER);
825		}
826		break;
827
828	case SO_BSDCOMPAT:
829		sock_warn_obsolete_bsdism("setsockopt");
830		break;
831
832	case SO_PASSCRED:
833		if (valbool)
834			set_bit(SOCK_PASSCRED, &sock->flags);
835		else
836			clear_bit(SOCK_PASSCRED, &sock->flags);
837		break;
838
839	case SO_TIMESTAMP:
840	case SO_TIMESTAMPNS:
841		if (valbool)  {
842			if (optname == SO_TIMESTAMP)
843				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
844			else
845				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
846			sock_set_flag(sk, SOCK_RCVTSTAMP);
847			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
848		} else {
849			sock_reset_flag(sk, SOCK_RCVTSTAMP);
850			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
851		}
852		break;
853
854	case SO_TIMESTAMPING:
855		if (val & ~SOF_TIMESTAMPING_MASK) {
856			ret = -EINVAL;
857			break;
858		}
859
860		if (val & SOF_TIMESTAMPING_OPT_ID &&
861		    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
862			if (sk->sk_protocol == IPPROTO_TCP &&
863			    sk->sk_type == SOCK_STREAM) {
864				if (sk->sk_state != TCP_ESTABLISHED) {
865					ret = -EINVAL;
866					break;
867				}
868				sk->sk_tskey = tcp_sk(sk)->snd_una;
869			} else {
870				sk->sk_tskey = 0;
871			}
872		}
873		sk->sk_tsflags = val;
874		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
875			sock_enable_timestamp(sk,
876					      SOCK_TIMESTAMPING_RX_SOFTWARE);
877		else
878			sock_disable_timestamp(sk,
879					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
880		break;
881
882	case SO_RCVLOWAT:
883		if (val < 0)
884			val = INT_MAX;
885		sk->sk_rcvlowat = val ? : 1;
886		break;
887
888	case SO_RCVTIMEO:
889		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
890		break;
891
892	case SO_SNDTIMEO:
893		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
894		break;
895
896	case SO_ATTACH_FILTER:
897		ret = -EINVAL;
898		if (optlen == sizeof(struct sock_fprog)) {
899			struct sock_fprog fprog;
900
901			ret = -EFAULT;
902			if (copy_from_user(&fprog, optval, sizeof(fprog)))
903				break;
904
905			ret = sk_attach_filter(&fprog, sk);
906		}
907		break;
908
909	case SO_ATTACH_BPF:
910		ret = -EINVAL;
911		if (optlen == sizeof(u32)) {
912			u32 ufd;
913
914			ret = -EFAULT;
915			if (copy_from_user(&ufd, optval, sizeof(ufd)))
916				break;
917
918			ret = sk_attach_bpf(ufd, sk);
919		}
920		break;
921
922	case SO_DETACH_FILTER:
923		ret = sk_detach_filter(sk);
924		break;
925
926	case SO_LOCK_FILTER:
927		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
928			ret = -EPERM;
929		else
930			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
931		break;
932
933	case SO_PASSSEC:
934		if (valbool)
935			set_bit(SOCK_PASSSEC, &sock->flags);
936		else
937			clear_bit(SOCK_PASSSEC, &sock->flags);
938		break;
939	case SO_MARK:
940		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
941			ret = -EPERM;
942		else
943			sk->sk_mark = val;
944		break;
945
946	case SO_RXQ_OVFL:
947		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
948		break;
949
950	case SO_WIFI_STATUS:
951		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
952		break;
953
954	case SO_PEEK_OFF:
955		if (sock->ops->set_peek_off)
956			ret = sock->ops->set_peek_off(sk, val);
957		else
958			ret = -EOPNOTSUPP;
959		break;
960
961	case SO_NOFCS:
962		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
963		break;
964
965	case SO_SELECT_ERR_QUEUE:
966		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
967		break;
968
969#ifdef CONFIG_NET_RX_BUSY_POLL
970	case SO_BUSY_POLL:
971		/* allow unprivileged users to decrease the value */
972		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
973			ret = -EPERM;
974		else {
975			if (val < 0)
976				ret = -EINVAL;
977			else
978				sk->sk_ll_usec = val;
979		}
980		break;
981#endif
982
983	case SO_MAX_PACING_RATE:
984		sk->sk_max_pacing_rate = val;
985		sk->sk_pacing_rate = min(sk->sk_pacing_rate,
986					 sk->sk_max_pacing_rate);
987		break;
988
989	default:
990		ret = -ENOPROTOOPT;
991		break;
992	}
993	release_sock(sk);
994	return ret;
995}
996EXPORT_SYMBOL(sock_setsockopt);
997
998
999static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1000			  struct ucred *ucred)
1001{
1002	ucred->pid = pid_vnr(pid);
1003	ucred->uid = ucred->gid = -1;
1004	if (cred) {
1005		struct user_namespace *current_ns = current_user_ns();
1006
1007		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1008		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1009	}
1010}
1011
1012int sock_getsockopt(struct socket *sock, int level, int optname,
1013		    char __user *optval, int __user *optlen)
1014{
1015	struct sock *sk = sock->sk;
1016
1017	union {
1018		int val;
1019		struct linger ling;
1020		struct timeval tm;
1021	} v;
1022
1023	int lv = sizeof(int);
1024	int len;
1025
1026	if (get_user(len, optlen))
1027		return -EFAULT;
1028	if (len < 0)
1029		return -EINVAL;
1030
1031	memset(&v, 0, sizeof(v));
1032
1033	switch (optname) {
1034	case SO_DEBUG:
1035		v.val = sock_flag(sk, SOCK_DBG);
1036		break;
1037
1038	case SO_DONTROUTE:
1039		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1040		break;
1041
1042	case SO_BROADCAST:
1043		v.val = sock_flag(sk, SOCK_BROADCAST);
1044		break;
1045
1046	case SO_SNDBUF:
1047		v.val = sk->sk_sndbuf;
1048		break;
1049
1050	case SO_RCVBUF:
1051		v.val = sk->sk_rcvbuf;
1052		break;
1053
1054	case SO_REUSEADDR:
1055		v.val = sk->sk_reuse;
1056		break;
1057
1058	case SO_REUSEPORT:
1059		v.val = sk->sk_reuseport;
1060		break;
1061
1062	case SO_KEEPALIVE:
1063		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1064		break;
1065
1066	case SO_TYPE:
1067		v.val = sk->sk_type;
1068		break;
1069
1070	case SO_PROTOCOL:
1071		v.val = sk->sk_protocol;
1072		break;
1073
1074	case SO_DOMAIN:
1075		v.val = sk->sk_family;
1076		break;
1077
1078	case SO_ERROR:
1079		v.val = -sock_error(sk);
1080		if (v.val == 0)
1081			v.val = xchg(&sk->sk_err_soft, 0);
1082		break;
1083
1084	case SO_OOBINLINE:
1085		v.val = sock_flag(sk, SOCK_URGINLINE);
1086		break;
1087
1088	case SO_NO_CHECK:
1089		v.val = sk->sk_no_check_tx;
1090		break;
1091
1092	case SO_PRIORITY:
1093		v.val = sk->sk_priority;
1094		break;
1095
1096	case SO_LINGER:
1097		lv		= sizeof(v.ling);
1098		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1099		v.ling.l_linger	= sk->sk_lingertime / HZ;
1100		break;
1101
1102	case SO_BSDCOMPAT:
1103		sock_warn_obsolete_bsdism("getsockopt");
1104		break;
1105
1106	case SO_TIMESTAMP:
1107		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1108				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1109		break;
1110
1111	case SO_TIMESTAMPNS:
1112		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1113		break;
1114
1115	case SO_TIMESTAMPING:
1116		v.val = sk->sk_tsflags;
1117		break;
1118
1119	case SO_RCVTIMEO:
1120		lv = sizeof(struct timeval);
1121		if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1122			v.tm.tv_sec = 0;
1123			v.tm.tv_usec = 0;
1124		} else {
1125			v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1126			v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1127		}
1128		break;
1129
1130	case SO_SNDTIMEO:
1131		lv = sizeof(struct timeval);
1132		if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1133			v.tm.tv_sec = 0;
1134			v.tm.tv_usec = 0;
1135		} else {
1136			v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1137			v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1138		}
1139		break;
1140
1141	case SO_RCVLOWAT:
1142		v.val = sk->sk_rcvlowat;
1143		break;
1144
1145	case SO_SNDLOWAT:
1146		v.val = 1;
1147		break;
1148
1149	case SO_PASSCRED:
1150		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1151		break;
1152
1153	case SO_PEERCRED:
1154	{
1155		struct ucred peercred;
1156		if (len > sizeof(peercred))
1157			len = sizeof(peercred);
1158		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1159		if (copy_to_user(optval, &peercred, len))
1160			return -EFAULT;
1161		goto lenout;
1162	}
1163
1164	case SO_PEERNAME:
1165	{
1166		char address[128];
1167
1168		if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1169			return -ENOTCONN;
1170		if (lv < len)
1171			return -EINVAL;
1172		if (copy_to_user(optval, address, len))
1173			return -EFAULT;
1174		goto lenout;
1175	}
1176
1177	/* Dubious BSD thing... Probably nobody even uses it, but
1178	 * the UNIX standard wants it for whatever reason... -DaveM
1179	 */
1180	case SO_ACCEPTCONN:
1181		v.val = sk->sk_state == TCP_LISTEN;
1182		break;
1183
1184	case SO_PASSSEC:
1185		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1186		break;
1187
1188	case SO_PEERSEC:
1189		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1190
1191	case SO_MARK:
1192		v.val = sk->sk_mark;
1193		break;
1194
1195	case SO_RXQ_OVFL:
1196		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1197		break;
1198
1199	case SO_WIFI_STATUS:
1200		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1201		break;
1202
1203	case SO_PEEK_OFF:
1204		if (!sock->ops->set_peek_off)
1205			return -EOPNOTSUPP;
1206
1207		v.val = sk->sk_peek_off;
1208		break;
1209	case SO_NOFCS:
1210		v.val = sock_flag(sk, SOCK_NOFCS);
1211		break;
1212
1213	case SO_BINDTODEVICE:
1214		return sock_getbindtodevice(sk, optval, optlen, len);
1215
1216	case SO_GET_FILTER:
1217		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1218		if (len < 0)
1219			return len;
1220
1221		goto lenout;
1222
1223	case SO_LOCK_FILTER:
1224		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1225		break;
1226
1227	case SO_BPF_EXTENSIONS:
1228		v.val = bpf_tell_extensions();
1229		break;
1230
1231	case SO_SELECT_ERR_QUEUE:
1232		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1233		break;
1234
1235#ifdef CONFIG_NET_RX_BUSY_POLL
1236	case SO_BUSY_POLL:
1237		v.val = sk->sk_ll_usec;
1238		break;
1239#endif
1240
1241	case SO_MAX_PACING_RATE:
1242		v.val = sk->sk_max_pacing_rate;
1243		break;
1244
1245	case SO_INCOMING_CPU:
1246		v.val = sk->sk_incoming_cpu;
1247		break;
1248
1249	default:
1250		/* We implement the SO_SNDLOWAT etc to not be settable
1251		 * (1003.1g 7).
1252		 */
1253		return -ENOPROTOOPT;
1254	}
1255
1256	if (len > lv)
1257		len = lv;
1258	if (copy_to_user(optval, &v, len))
1259		return -EFAULT;
1260lenout:
1261	if (put_user(len, optlen))
1262		return -EFAULT;
1263	return 0;
1264}
1265
1266/*
1267 * Initialize an sk_lock.
1268 *
1269 * (We also register the sk_lock with the lock validator.)
1270 */
1271static inline void sock_lock_init(struct sock *sk)
1272{
1273	sock_lock_init_class_and_name(sk,
1274			af_family_slock_key_strings[sk->sk_family],
1275			af_family_slock_keys + sk->sk_family,
1276			af_family_key_strings[sk->sk_family],
1277			af_family_keys + sk->sk_family);
1278}
1279
1280/*
1281 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1282 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1283 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1284 */
1285static void sock_copy(struct sock *nsk, const struct sock *osk)
1286{
1287#ifdef CONFIG_SECURITY_NETWORK
1288	void *sptr = nsk->sk_security;
1289#endif
1290	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1291
1292	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1293	       osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1294
1295#ifdef CONFIG_SECURITY_NETWORK
1296	nsk->sk_security = sptr;
1297	security_sk_clone(osk, nsk);
1298#endif
1299}
1300
1301void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1302{
1303	unsigned long nulls1, nulls2;
1304
1305	nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1306	nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1307	if (nulls1 > nulls2)
1308		swap(nulls1, nulls2);
1309
1310	if (nulls1 != 0)
1311		memset((char *)sk, 0, nulls1);
1312	memset((char *)sk + nulls1 + sizeof(void *), 0,
1313	       nulls2 - nulls1 - sizeof(void *));
1314	memset((char *)sk + nulls2 + sizeof(void *), 0,
1315	       size - nulls2 - sizeof(void *));
1316}
1317EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1318
1319static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1320		int family)
1321{
1322	struct sock *sk;
1323	struct kmem_cache *slab;
1324
1325	slab = prot->slab;
1326	if (slab != NULL) {
1327		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1328		if (!sk)
1329			return sk;
1330		if (priority & __GFP_ZERO) {
1331			if (prot->clear_sk)
1332				prot->clear_sk(sk, prot->obj_size);
1333			else
1334				sk_prot_clear_nulls(sk, prot->obj_size);
1335		}
1336	} else
1337		sk = kmalloc(prot->obj_size, priority);
1338
1339	if (sk != NULL) {
1340		kmemcheck_annotate_bitfield(sk, flags);
1341
1342		if (security_sk_alloc(sk, family, priority))
1343			goto out_free;
1344
1345		if (!try_module_get(prot->owner))
1346			goto out_free_sec;
1347		sk_tx_queue_clear(sk);
1348	}
1349
1350	return sk;
1351
1352out_free_sec:
1353	security_sk_free(sk);
1354out_free:
1355	if (slab != NULL)
1356		kmem_cache_free(slab, sk);
1357	else
1358		kfree(sk);
1359	return NULL;
1360}
1361
1362static void sk_prot_free(struct proto *prot, struct sock *sk)
1363{
1364	struct kmem_cache *slab;
1365	struct module *owner;
1366
1367	owner = prot->owner;
1368	slab = prot->slab;
1369
1370	security_sk_free(sk);
1371	if (slab != NULL)
1372		kmem_cache_free(slab, sk);
1373	else
1374		kfree(sk);
1375	module_put(owner);
1376}
1377
1378#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1379void sock_update_netprioidx(struct sock *sk)
1380{
1381	if (in_interrupt())
1382		return;
1383
1384	sk->sk_cgrp_prioidx = task_netprioidx(current);
1385}
1386EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1387#endif
1388
1389/**
1390 *	sk_alloc - All socket objects are allocated here
1391 *	@net: the applicable net namespace
1392 *	@family: protocol family
1393 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1394 *	@prot: struct proto associated with this new sock instance
1395 */
1396struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1397		      struct proto *prot)
1398{
1399	struct sock *sk;
1400
1401	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1402	if (sk) {
1403		sk->sk_family = family;
1404		/*
1405		 * See comment in struct sock definition to understand
1406		 * why we need sk_prot_creator -acme
1407		 */
1408		sk->sk_prot = sk->sk_prot_creator = prot;
1409		sock_lock_init(sk);
1410		sock_net_set(sk, get_net(net));
1411		atomic_set(&sk->sk_wmem_alloc, 1);
1412
1413		sock_update_classid(sk);
1414		sock_update_netprioidx(sk);
1415	}
1416
1417	return sk;
1418}
1419EXPORT_SYMBOL(sk_alloc);
1420
1421static void __sk_free(struct sock *sk)
1422{
1423	struct sk_filter *filter;
1424
1425	if (sk->sk_destruct)
1426		sk->sk_destruct(sk);
1427
1428	filter = rcu_dereference_check(sk->sk_filter,
1429				       atomic_read(&sk->sk_wmem_alloc) == 0);
1430	if (filter) {
1431		sk_filter_uncharge(sk, filter);
1432		RCU_INIT_POINTER(sk->sk_filter, NULL);
1433	}
1434
1435	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1436
1437	if (atomic_read(&sk->sk_omem_alloc))
1438		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1439			 __func__, atomic_read(&sk->sk_omem_alloc));
1440
1441	if (sk->sk_peer_cred)
1442		put_cred(sk->sk_peer_cred);
1443	put_pid(sk->sk_peer_pid);
1444	put_net(sock_net(sk));
1445	sk_prot_free(sk->sk_prot_creator, sk);
1446}
1447
1448void sk_free(struct sock *sk)
1449{
1450	/*
1451	 * We subtract one from sk_wmem_alloc and can know if
1452	 * some packets are still in some tx queue.
1453	 * If not null, sock_wfree() will call __sk_free(sk) later
1454	 */
1455	if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1456		__sk_free(sk);
1457}
1458EXPORT_SYMBOL(sk_free);
1459
1460/*
1461 * Last sock_put should drop reference to sk->sk_net. It has already
1462 * been dropped in sk_change_net. Taking reference to stopping namespace
1463 * is not an option.
1464 * Take reference to a socket to remove it from hash _alive_ and after that
1465 * destroy it in the context of init_net.
1466 */
1467void sk_release_kernel(struct sock *sk)
1468{
1469	if (sk == NULL || sk->sk_socket == NULL)
1470		return;
1471
1472	sock_hold(sk);
1473	sock_release(sk->sk_socket);
1474	sock_net_set(sk, get_net(&init_net));
1475	sock_put(sk);
1476}
1477EXPORT_SYMBOL(sk_release_kernel);
1478
1479static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1480{
1481	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1482		sock_update_memcg(newsk);
1483}
1484
1485/**
1486 *	sk_clone_lock - clone a socket, and lock its clone
1487 *	@sk: the socket to clone
1488 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1489 *
1490 *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1491 */
1492struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1493{
1494	struct sock *newsk;
1495	bool is_charged = true;
1496
1497	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1498	if (newsk != NULL) {
1499		struct sk_filter *filter;
1500
1501		sock_copy(newsk, sk);
1502
1503		/* SANITY */
1504		get_net(sock_net(newsk));
1505		sk_node_init(&newsk->sk_node);
1506		sock_lock_init(newsk);
1507		bh_lock_sock(newsk);
1508		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1509		newsk->sk_backlog.len = 0;
1510
1511		atomic_set(&newsk->sk_rmem_alloc, 0);
1512		/*
1513		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1514		 */
1515		atomic_set(&newsk->sk_wmem_alloc, 1);
1516		atomic_set(&newsk->sk_omem_alloc, 0);
1517		skb_queue_head_init(&newsk->sk_receive_queue);
1518		skb_queue_head_init(&newsk->sk_write_queue);
1519
1520		spin_lock_init(&newsk->sk_dst_lock);
1521		rwlock_init(&newsk->sk_callback_lock);
1522		lockdep_set_class_and_name(&newsk->sk_callback_lock,
1523				af_callback_keys + newsk->sk_family,
1524				af_family_clock_key_strings[newsk->sk_family]);
1525
1526		newsk->sk_dst_cache	= NULL;
1527		newsk->sk_wmem_queued	= 0;
1528		newsk->sk_forward_alloc = 0;
1529		newsk->sk_send_head	= NULL;
1530		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1531
1532		sock_reset_flag(newsk, SOCK_DONE);
1533		skb_queue_head_init(&newsk->sk_error_queue);
1534
1535		filter = rcu_dereference_protected(newsk->sk_filter, 1);
1536		if (filter != NULL)
1537			/* though it's an empty new sock, the charging may fail
1538			 * if sysctl_optmem_max was changed between creation of
1539			 * original socket and cloning
1540			 */
1541			is_charged = sk_filter_charge(newsk, filter);
1542
1543		if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1544			/* It is still raw copy of parent, so invalidate
1545			 * destructor and make plain sk_free() */
1546			newsk->sk_destruct = NULL;
1547			bh_unlock_sock(newsk);
1548			sk_free(newsk);
1549			newsk = NULL;
1550			goto out;
1551		}
1552
1553		newsk->sk_err	   = 0;
1554		newsk->sk_priority = 0;
1555		newsk->sk_incoming_cpu = raw_smp_processor_id();
1556		atomic64_set(&newsk->sk_cookie, 0);
1557		/*
1558		 * Before updating sk_refcnt, we must commit prior changes to memory
1559		 * (Documentation/RCU/rculist_nulls.txt for details)
1560		 */
1561		smp_wmb();
1562		atomic_set(&newsk->sk_refcnt, 2);
1563
1564		/*
1565		 * Increment the counter in the same struct proto as the master
1566		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1567		 * is the same as sk->sk_prot->socks, as this field was copied
1568		 * with memcpy).
1569		 *
1570		 * This _changes_ the previous behaviour, where
1571		 * tcp_create_openreq_child always was incrementing the
1572		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1573		 * to be taken into account in all callers. -acme
1574		 */
1575		sk_refcnt_debug_inc(newsk);
1576		sk_set_socket(newsk, NULL);
1577		newsk->sk_wq = NULL;
1578
1579		sk_update_clone(sk, newsk);
1580
1581		if (newsk->sk_prot->sockets_allocated)
1582			sk_sockets_allocated_inc(newsk);
1583
1584		if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1585			net_enable_timestamp();
1586	}
1587out:
1588	return newsk;
1589}
1590EXPORT_SYMBOL_GPL(sk_clone_lock);
1591
1592void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1593{
1594	__sk_dst_set(sk, dst);
1595	sk->sk_route_caps = dst->dev->features;
1596	if (sk->sk_route_caps & NETIF_F_GSO)
1597		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1598	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1599	if (sk_can_gso(sk)) {
1600		if (dst->header_len) {
1601			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1602		} else {
1603			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1604			sk->sk_gso_max_size = dst->dev->gso_max_size;
1605			sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1606		}
1607	}
1608}
1609EXPORT_SYMBOL_GPL(sk_setup_caps);
1610
1611/*
1612 *	Simple resource managers for sockets.
1613 */
1614
1615
1616/*
1617 * Write buffer destructor automatically called from kfree_skb.
1618 */
1619void sock_wfree(struct sk_buff *skb)
1620{
1621	struct sock *sk = skb->sk;
1622	unsigned int len = skb->truesize;
1623
1624	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1625		/*
1626		 * Keep a reference on sk_wmem_alloc, this will be released
1627		 * after sk_write_space() call
1628		 */
1629		atomic_sub(len - 1, &sk->sk_wmem_alloc);
1630		sk->sk_write_space(sk);
1631		len = 1;
1632	}
1633	/*
1634	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1635	 * could not do because of in-flight packets
1636	 */
1637	if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1638		__sk_free(sk);
1639}
1640EXPORT_SYMBOL(sock_wfree);
1641
1642void skb_orphan_partial(struct sk_buff *skb)
1643{
1644	/* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1645	 * so we do not completely orphan skb, but transfert all
1646	 * accounted bytes but one, to avoid unexpected reorders.
1647	 */
1648	if (skb->destructor == sock_wfree
1649#ifdef CONFIG_INET
1650	    || skb->destructor == tcp_wfree
1651#endif
1652		) {
1653		atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1654		skb->truesize = 1;
1655	} else {
1656		skb_orphan(skb);
1657	}
1658}
1659EXPORT_SYMBOL(skb_orphan_partial);
1660
1661/*
1662 * Read buffer destructor automatically called from kfree_skb.
1663 */
1664void sock_rfree(struct sk_buff *skb)
1665{
1666	struct sock *sk = skb->sk;
1667	unsigned int len = skb->truesize;
1668
1669	atomic_sub(len, &sk->sk_rmem_alloc);
1670	sk_mem_uncharge(sk, len);
1671}
1672EXPORT_SYMBOL(sock_rfree);
1673
1674/*
1675 * Buffer destructor for skbs that are not used directly in read or write
1676 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1677 */
1678void sock_efree(struct sk_buff *skb)
1679{
1680	sock_put(skb->sk);
1681}
1682EXPORT_SYMBOL(sock_efree);
1683
1684kuid_t sock_i_uid(struct sock *sk)
1685{
1686	kuid_t uid;
1687
1688	read_lock_bh(&sk->sk_callback_lock);
1689	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1690	read_unlock_bh(&sk->sk_callback_lock);
1691	return uid;
1692}
1693EXPORT_SYMBOL(sock_i_uid);
1694
1695unsigned long sock_i_ino(struct sock *sk)
1696{
1697	unsigned long ino;
1698
1699	read_lock_bh(&sk->sk_callback_lock);
1700	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1701	read_unlock_bh(&sk->sk_callback_lock);
1702	return ino;
1703}
1704EXPORT_SYMBOL(sock_i_ino);
1705
1706/*
1707 * Allocate a skb from the socket's send buffer.
1708 */
1709struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1710			     gfp_t priority)
1711{
1712	if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1713		struct sk_buff *skb = alloc_skb(size, priority);
1714		if (skb) {
1715			skb_set_owner_w(skb, sk);
1716			return skb;
1717		}
1718	}
1719	return NULL;
1720}
1721EXPORT_SYMBOL(sock_wmalloc);
1722
1723/*
1724 * Allocate a memory block from the socket's option memory buffer.
1725 */
1726void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1727{
1728	if ((unsigned int)size <= sysctl_optmem_max &&
1729	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1730		void *mem;
1731		/* First do the add, to avoid the race if kmalloc
1732		 * might sleep.
1733		 */
1734		atomic_add(size, &sk->sk_omem_alloc);
1735		mem = kmalloc(size, priority);
1736		if (mem)
1737			return mem;
1738		atomic_sub(size, &sk->sk_omem_alloc);
1739	}
1740	return NULL;
1741}
1742EXPORT_SYMBOL(sock_kmalloc);
1743
1744/* Free an option memory block. Note, we actually want the inline
1745 * here as this allows gcc to detect the nullify and fold away the
1746 * condition entirely.
1747 */
1748static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1749				  const bool nullify)
1750{
1751	if (WARN_ON_ONCE(!mem))
1752		return;
1753	if (nullify)
1754		kzfree(mem);
1755	else
1756		kfree(mem);
1757	atomic_sub(size, &sk->sk_omem_alloc);
1758}
1759
1760void sock_kfree_s(struct sock *sk, void *mem, int size)
1761{
1762	__sock_kfree_s(sk, mem, size, false);
1763}
1764EXPORT_SYMBOL(sock_kfree_s);
1765
1766void sock_kzfree_s(struct sock *sk, void *mem, int size)
1767{
1768	__sock_kfree_s(sk, mem, size, true);
1769}
1770EXPORT_SYMBOL(sock_kzfree_s);
1771
1772/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1773   I think, these locks should be removed for datagram sockets.
1774 */
1775static long sock_wait_for_wmem(struct sock *sk, long timeo)
1776{
1777	DEFINE_WAIT(wait);
1778
1779	clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1780	for (;;) {
1781		if (!timeo)
1782			break;
1783		if (signal_pending(current))
1784			break;
1785		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1786		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1787		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1788			break;
1789		if (sk->sk_shutdown & SEND_SHUTDOWN)
1790			break;
1791		if (sk->sk_err)
1792			break;
1793		timeo = schedule_timeout(timeo);
1794	}
1795	finish_wait(sk_sleep(sk), &wait);
1796	return timeo;
1797}
1798
1799
1800/*
1801 *	Generic send/receive buffer handlers
1802 */
1803
1804struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1805				     unsigned long data_len, int noblock,
1806				     int *errcode, int max_page_order)
1807{
1808	struct sk_buff *skb;
1809	long timeo;
1810	int err;
1811
1812	timeo = sock_sndtimeo(sk, noblock);
1813	for (;;) {
1814		err = sock_error(sk);
1815		if (err != 0)
1816			goto failure;
1817
1818		err = -EPIPE;
1819		if (sk->sk_shutdown & SEND_SHUTDOWN)
1820			goto failure;
1821
1822		if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1823			break;
1824
1825		set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1826		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1827		err = -EAGAIN;
1828		if (!timeo)
1829			goto failure;
1830		if (signal_pending(current))
1831			goto interrupted;
1832		timeo = sock_wait_for_wmem(sk, timeo);
1833	}
1834	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1835				   errcode, sk->sk_allocation);
1836	if (skb)
1837		skb_set_owner_w(skb, sk);
1838	return skb;
1839
1840interrupted:
1841	err = sock_intr_errno(timeo);
1842failure:
1843	*errcode = err;
1844	return NULL;
1845}
1846EXPORT_SYMBOL(sock_alloc_send_pskb);
1847
1848struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1849				    int noblock, int *errcode)
1850{
1851	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1852}
1853EXPORT_SYMBOL(sock_alloc_send_skb);
1854
1855/* On 32bit arches, an skb frag is limited to 2^15 */
1856#define SKB_FRAG_PAGE_ORDER	get_order(32768)
1857
1858/**
1859 * skb_page_frag_refill - check that a page_frag contains enough room
1860 * @sz: minimum size of the fragment we want to get
1861 * @pfrag: pointer to page_frag
1862 * @gfp: priority for memory allocation
1863 *
1864 * Note: While this allocator tries to use high order pages, there is
1865 * no guarantee that allocations succeed. Therefore, @sz MUST be
1866 * less or equal than PAGE_SIZE.
1867 */
1868bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1869{
1870	if (pfrag->page) {
1871		if (atomic_read(&pfrag->page->_count) == 1) {
1872			pfrag->offset = 0;
1873			return true;
1874		}
1875		if (pfrag->offset + sz <= pfrag->size)
1876			return true;
1877		put_page(pfrag->page);
1878	}
1879
1880	pfrag->offset = 0;
1881	if (SKB_FRAG_PAGE_ORDER) {
1882		pfrag->page = alloc_pages((gfp & ~__GFP_WAIT) | __GFP_COMP |
1883					  __GFP_NOWARN | __GFP_NORETRY,
1884					  SKB_FRAG_PAGE_ORDER);
1885		if (likely(pfrag->page)) {
1886			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1887			return true;
1888		}
1889	}
1890	pfrag->page = alloc_page(gfp);
1891	if (likely(pfrag->page)) {
1892		pfrag->size = PAGE_SIZE;
1893		return true;
1894	}
1895	return false;
1896}
1897EXPORT_SYMBOL(skb_page_frag_refill);
1898
1899bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1900{
1901	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1902		return true;
1903
1904	sk_enter_memory_pressure(sk);
1905	sk_stream_moderate_sndbuf(sk);
1906	return false;
1907}
1908EXPORT_SYMBOL(sk_page_frag_refill);
1909
1910static void __lock_sock(struct sock *sk)
1911	__releases(&sk->sk_lock.slock)
1912	__acquires(&sk->sk_lock.slock)
1913{
1914	DEFINE_WAIT(wait);
1915
1916	for (;;) {
1917		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1918					TASK_UNINTERRUPTIBLE);
1919		spin_unlock_bh(&sk->sk_lock.slock);
1920		schedule();
1921		spin_lock_bh(&sk->sk_lock.slock);
1922		if (!sock_owned_by_user(sk))
1923			break;
1924	}
1925	finish_wait(&sk->sk_lock.wq, &wait);
1926}
1927
1928static void __release_sock(struct sock *sk)
1929	__releases(&sk->sk_lock.slock)
1930	__acquires(&sk->sk_lock.slock)
1931{
1932	struct sk_buff *skb = sk->sk_backlog.head;
1933
1934	do {
1935		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1936		bh_unlock_sock(sk);
1937
1938		do {
1939			struct sk_buff *next = skb->next;
1940
1941			prefetch(next);
1942			WARN_ON_ONCE(skb_dst_is_noref(skb));
1943			skb->next = NULL;
1944			sk_backlog_rcv(sk, skb);
1945
1946			/*
1947			 * We are in process context here with softirqs
1948			 * disabled, use cond_resched_softirq() to preempt.
1949			 * This is safe to do because we've taken the backlog
1950			 * queue private:
1951			 */
1952			cond_resched_softirq();
1953
1954			skb = next;
1955		} while (skb != NULL);
1956
1957		bh_lock_sock(sk);
1958	} while ((skb = sk->sk_backlog.head) != NULL);
1959
1960	/*
1961	 * Doing the zeroing here guarantee we can not loop forever
1962	 * while a wild producer attempts to flood us.
1963	 */
1964	sk->sk_backlog.len = 0;
1965}
1966
1967/**
1968 * sk_wait_data - wait for data to arrive at sk_receive_queue
1969 * @sk:    sock to wait on
1970 * @timeo: for how long
1971 *
1972 * Now socket state including sk->sk_err is changed only under lock,
1973 * hence we may omit checks after joining wait queue.
1974 * We check receive queue before schedule() only as optimization;
1975 * it is very likely that release_sock() added new data.
1976 */
1977int sk_wait_data(struct sock *sk, long *timeo)
1978{
1979	int rc;
1980	DEFINE_WAIT(wait);
1981
1982	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1983	set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1984	rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1985	clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1986	finish_wait(sk_sleep(sk), &wait);
1987	return rc;
1988}
1989EXPORT_SYMBOL(sk_wait_data);
1990
1991/**
1992 *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1993 *	@sk: socket
1994 *	@size: memory size to allocate
1995 *	@kind: allocation type
1996 *
1997 *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1998 *	rmem allocation. This function assumes that protocols which have
1999 *	memory_pressure use sk_wmem_queued as write buffer accounting.
2000 */
2001int __sk_mem_schedule(struct sock *sk, int size, int kind)
2002{
2003	struct proto *prot = sk->sk_prot;
2004	int amt = sk_mem_pages(size);
2005	long allocated;
2006	int parent_status = UNDER_LIMIT;
2007
2008	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2009
2010	allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2011
2012	/* Under limit. */
2013	if (parent_status == UNDER_LIMIT &&
2014			allocated <= sk_prot_mem_limits(sk, 0)) {
2015		sk_leave_memory_pressure(sk);
2016		return 1;
2017	}
2018
2019	/* Under pressure. (we or our parents) */
2020	if ((parent_status > SOFT_LIMIT) ||
2021			allocated > sk_prot_mem_limits(sk, 1))
2022		sk_enter_memory_pressure(sk);
2023
2024	/* Over hard limit (we or our parents) */
2025	if ((parent_status == OVER_LIMIT) ||
2026			(allocated > sk_prot_mem_limits(sk, 2)))
2027		goto suppress_allocation;
2028
2029	/* guarantee minimum buffer size under pressure */
2030	if (kind == SK_MEM_RECV) {
2031		if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2032			return 1;
2033
2034	} else { /* SK_MEM_SEND */
2035		if (sk->sk_type == SOCK_STREAM) {
2036			if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2037				return 1;
2038		} else if (atomic_read(&sk->sk_wmem_alloc) <
2039			   prot->sysctl_wmem[0])
2040				return 1;
2041	}
2042
2043	if (sk_has_memory_pressure(sk)) {
2044		int alloc;
2045
2046		if (!sk_under_memory_pressure(sk))
2047			return 1;
2048		alloc = sk_sockets_allocated_read_positive(sk);
2049		if (sk_prot_mem_limits(sk, 2) > alloc *
2050		    sk_mem_pages(sk->sk_wmem_queued +
2051				 atomic_read(&sk->sk_rmem_alloc) +
2052				 sk->sk_forward_alloc))
2053			return 1;
2054	}
2055
2056suppress_allocation:
2057
2058	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2059		sk_stream_moderate_sndbuf(sk);
2060
2061		/* Fail only if socket is _under_ its sndbuf.
2062		 * In this case we cannot block, so that we have to fail.
2063		 */
2064		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2065			return 1;
2066	}
2067
2068	trace_sock_exceed_buf_limit(sk, prot, allocated);
2069
2070	/* Alas. Undo changes. */
2071	sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2072
2073	sk_memory_allocated_sub(sk, amt);
2074
2075	return 0;
2076}
2077EXPORT_SYMBOL(__sk_mem_schedule);
2078
2079/**
2080 *	__sk_reclaim - reclaim memory_allocated
2081 *	@sk: socket
2082 */
2083void __sk_mem_reclaim(struct sock *sk)
2084{
2085	sk_memory_allocated_sub(sk,
2086				sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2087	sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2088
2089	if (sk_under_memory_pressure(sk) &&
2090	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2091		sk_leave_memory_pressure(sk);
2092}
2093EXPORT_SYMBOL(__sk_mem_reclaim);
2094
2095
2096/*
2097 * Set of default routines for initialising struct proto_ops when
2098 * the protocol does not support a particular function. In certain
2099 * cases where it makes no sense for a protocol to have a "do nothing"
2100 * function, some default processing is provided.
2101 */
2102
2103int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2104{
2105	return -EOPNOTSUPP;
2106}
2107EXPORT_SYMBOL(sock_no_bind);
2108
2109int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2110		    int len, int flags)
2111{
2112	return -EOPNOTSUPP;
2113}
2114EXPORT_SYMBOL(sock_no_connect);
2115
2116int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2117{
2118	return -EOPNOTSUPP;
2119}
2120EXPORT_SYMBOL(sock_no_socketpair);
2121
2122int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2123{
2124	return -EOPNOTSUPP;
2125}
2126EXPORT_SYMBOL(sock_no_accept);
2127
2128int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2129		    int *len, int peer)
2130{
2131	return -EOPNOTSUPP;
2132}
2133EXPORT_SYMBOL(sock_no_getname);
2134
2135unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2136{
2137	return 0;
2138}
2139EXPORT_SYMBOL(sock_no_poll);
2140
2141int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2142{
2143	return -EOPNOTSUPP;
2144}
2145EXPORT_SYMBOL(sock_no_ioctl);
2146
2147int sock_no_listen(struct socket *sock, int backlog)
2148{
2149	return -EOPNOTSUPP;
2150}
2151EXPORT_SYMBOL(sock_no_listen);
2152
2153int sock_no_shutdown(struct socket *sock, int how)
2154{
2155	return -EOPNOTSUPP;
2156}
2157EXPORT_SYMBOL(sock_no_shutdown);
2158
2159int sock_no_setsockopt(struct socket *sock, int level, int optname,
2160		    char __user *optval, unsigned int optlen)
2161{
2162	return -EOPNOTSUPP;
2163}
2164EXPORT_SYMBOL(sock_no_setsockopt);
2165
2166int sock_no_getsockopt(struct socket *sock, int level, int optname,
2167		    char __user *optval, int __user *optlen)
2168{
2169	return -EOPNOTSUPP;
2170}
2171EXPORT_SYMBOL(sock_no_getsockopt);
2172
2173int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2174{
2175	return -EOPNOTSUPP;
2176}
2177EXPORT_SYMBOL(sock_no_sendmsg);
2178
2179int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2180		    int flags)
2181{
2182	return -EOPNOTSUPP;
2183}
2184EXPORT_SYMBOL(sock_no_recvmsg);
2185
2186int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2187{
2188	/* Mirror missing mmap method error code */
2189	return -ENODEV;
2190}
2191EXPORT_SYMBOL(sock_no_mmap);
2192
2193ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2194{
2195	ssize_t res;
2196	struct msghdr msg = {.msg_flags = flags};
2197	struct kvec iov;
2198	char *kaddr = kmap(page);
2199	iov.iov_base = kaddr + offset;
2200	iov.iov_len = size;
2201	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2202	kunmap(page);
2203	return res;
2204}
2205EXPORT_SYMBOL(sock_no_sendpage);
2206
2207/*
2208 *	Default Socket Callbacks
2209 */
2210
2211static void sock_def_wakeup(struct sock *sk)
2212{
2213	struct socket_wq *wq;
2214
2215	rcu_read_lock();
2216	wq = rcu_dereference(sk->sk_wq);
2217	if (wq_has_sleeper(wq))
2218		wake_up_interruptible_all(&wq->wait);
2219	rcu_read_unlock();
2220}
2221
2222static void sock_def_error_report(struct sock *sk)
2223{
2224	struct socket_wq *wq;
2225
2226	rcu_read_lock();
2227	wq = rcu_dereference(sk->sk_wq);
2228	if (wq_has_sleeper(wq))
2229		wake_up_interruptible_poll(&wq->wait, POLLERR);
2230	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2231	rcu_read_unlock();
2232}
2233
2234static void sock_def_readable(struct sock *sk)
2235{
2236	struct socket_wq *wq;
2237
2238	rcu_read_lock();
2239	wq = rcu_dereference(sk->sk_wq);
2240	if (wq_has_sleeper(wq))
2241		wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2242						POLLRDNORM | POLLRDBAND);
2243	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2244	rcu_read_unlock();
2245}
2246
2247static void sock_def_write_space(struct sock *sk)
2248{
2249	struct socket_wq *wq;
2250
2251	rcu_read_lock();
2252
2253	/* Do not wake up a writer until he can make "significant"
2254	 * progress.  --DaveM
2255	 */
2256	if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2257		wq = rcu_dereference(sk->sk_wq);
2258		if (wq_has_sleeper(wq))
2259			wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2260						POLLWRNORM | POLLWRBAND);
2261
2262		/* Should agree with poll, otherwise some programs break */
2263		if (sock_writeable(sk))
2264			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2265	}
2266
2267	rcu_read_unlock();
2268}
2269
2270static void sock_def_destruct(struct sock *sk)
2271{
2272	kfree(sk->sk_protinfo);
2273}
2274
2275void sk_send_sigurg(struct sock *sk)
2276{
2277	if (sk->sk_socket && sk->sk_socket->file)
2278		if (send_sigurg(&sk->sk_socket->file->f_owner))
2279			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2280}
2281EXPORT_SYMBOL(sk_send_sigurg);
2282
2283void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2284		    unsigned long expires)
2285{
2286	if (!mod_timer(timer, expires))
2287		sock_hold(sk);
2288}
2289EXPORT_SYMBOL(sk_reset_timer);
2290
2291void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2292{
2293	if (del_timer(timer))
2294		__sock_put(sk);
2295}
2296EXPORT_SYMBOL(sk_stop_timer);
2297
2298void sock_init_data(struct socket *sock, struct sock *sk)
2299{
2300	skb_queue_head_init(&sk->sk_receive_queue);
2301	skb_queue_head_init(&sk->sk_write_queue);
2302	skb_queue_head_init(&sk->sk_error_queue);
2303
2304	sk->sk_send_head	=	NULL;
2305
2306	init_timer(&sk->sk_timer);
2307
2308	sk->sk_allocation	=	GFP_KERNEL;
2309	sk->sk_rcvbuf		=	sysctl_rmem_default;
2310	sk->sk_sndbuf		=	sysctl_wmem_default;
2311	sk->sk_state		=	TCP_CLOSE;
2312	sk_set_socket(sk, sock);
2313
2314	sock_set_flag(sk, SOCK_ZAPPED);
2315
2316	if (sock) {
2317		sk->sk_type	=	sock->type;
2318		sk->sk_wq	=	sock->wq;
2319		sock->sk	=	sk;
2320	} else
2321		sk->sk_wq	=	NULL;
2322
2323	spin_lock_init(&sk->sk_dst_lock);
2324	rwlock_init(&sk->sk_callback_lock);
2325	lockdep_set_class_and_name(&sk->sk_callback_lock,
2326			af_callback_keys + sk->sk_family,
2327			af_family_clock_key_strings[sk->sk_family]);
2328
2329	sk->sk_state_change	=	sock_def_wakeup;
2330	sk->sk_data_ready	=	sock_def_readable;
2331	sk->sk_write_space	=	sock_def_write_space;
2332	sk->sk_error_report	=	sock_def_error_report;
2333	sk->sk_destruct		=	sock_def_destruct;
2334
2335	sk->sk_frag.page	=	NULL;
2336	sk->sk_frag.offset	=	0;
2337	sk->sk_peek_off		=	-1;
2338
2339	sk->sk_peer_pid 	=	NULL;
2340	sk->sk_peer_cred	=	NULL;
2341	sk->sk_write_pending	=	0;
2342	sk->sk_rcvlowat		=	1;
2343	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
2344	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
2345
2346	sk->sk_stamp = ktime_set(-1L, 0);
2347
2348#ifdef CONFIG_NET_RX_BUSY_POLL
2349	sk->sk_napi_id		=	0;
2350	sk->sk_ll_usec		=	sysctl_net_busy_read;
2351#endif
2352
2353	sk->sk_max_pacing_rate = ~0U;
2354	sk->sk_pacing_rate = ~0U;
2355	/*
2356	 * Before updating sk_refcnt, we must commit prior changes to memory
2357	 * (Documentation/RCU/rculist_nulls.txt for details)
2358	 */
2359	smp_wmb();
2360	atomic_set(&sk->sk_refcnt, 1);
2361	atomic_set(&sk->sk_drops, 0);
2362}
2363EXPORT_SYMBOL(sock_init_data);
2364
2365void lock_sock_nested(struct sock *sk, int subclass)
2366{
2367	might_sleep();
2368	spin_lock_bh(&sk->sk_lock.slock);
2369	if (sk->sk_lock.owned)
2370		__lock_sock(sk);
2371	sk->sk_lock.owned = 1;
2372	spin_unlock(&sk->sk_lock.slock);
2373	/*
2374	 * The sk_lock has mutex_lock() semantics here:
2375	 */
2376	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2377	local_bh_enable();
2378}
2379EXPORT_SYMBOL(lock_sock_nested);
2380
2381void release_sock(struct sock *sk)
2382{
2383	/*
2384	 * The sk_lock has mutex_unlock() semantics:
2385	 */
2386	mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2387
2388	spin_lock_bh(&sk->sk_lock.slock);
2389	if (sk->sk_backlog.tail)
2390		__release_sock(sk);
2391
2392	/* Warning : release_cb() might need to release sk ownership,
2393	 * ie call sock_release_ownership(sk) before us.
2394	 */
2395	if (sk->sk_prot->release_cb)
2396		sk->sk_prot->release_cb(sk);
2397
2398	sock_release_ownership(sk);
2399	if (waitqueue_active(&sk->sk_lock.wq))
2400		wake_up(&sk->sk_lock.wq);
2401	spin_unlock_bh(&sk->sk_lock.slock);
2402}
2403EXPORT_SYMBOL(release_sock);
2404
2405/**
2406 * lock_sock_fast - fast version of lock_sock
2407 * @sk: socket
2408 *
2409 * This version should be used for very small section, where process wont block
2410 * return false if fast path is taken
2411 *   sk_lock.slock locked, owned = 0, BH disabled
2412 * return true if slow path is taken
2413 *   sk_lock.slock unlocked, owned = 1, BH enabled
2414 */
2415bool lock_sock_fast(struct sock *sk)
2416{
2417	might_sleep();
2418	spin_lock_bh(&sk->sk_lock.slock);
2419
2420	if (!sk->sk_lock.owned)
2421		/*
2422		 * Note : We must disable BH
2423		 */
2424		return false;
2425
2426	__lock_sock(sk);
2427	sk->sk_lock.owned = 1;
2428	spin_unlock(&sk->sk_lock.slock);
2429	/*
2430	 * The sk_lock has mutex_lock() semantics here:
2431	 */
2432	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2433	local_bh_enable();
2434	return true;
2435}
2436EXPORT_SYMBOL(lock_sock_fast);
2437
2438int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2439{
2440	struct timeval tv;
2441	if (!sock_flag(sk, SOCK_TIMESTAMP))
2442		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2443	tv = ktime_to_timeval(sk->sk_stamp);
2444	if (tv.tv_sec == -1)
2445		return -ENOENT;
2446	if (tv.tv_sec == 0) {
2447		sk->sk_stamp = ktime_get_real();
2448		tv = ktime_to_timeval(sk->sk_stamp);
2449	}
2450	return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2451}
2452EXPORT_SYMBOL(sock_get_timestamp);
2453
2454int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2455{
2456	struct timespec ts;
2457	if (!sock_flag(sk, SOCK_TIMESTAMP))
2458		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2459	ts = ktime_to_timespec(sk->sk_stamp);
2460	if (ts.tv_sec == -1)
2461		return -ENOENT;
2462	if (ts.tv_sec == 0) {
2463		sk->sk_stamp = ktime_get_real();
2464		ts = ktime_to_timespec(sk->sk_stamp);
2465	}
2466	return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2467}
2468EXPORT_SYMBOL(sock_get_timestampns);
2469
2470void sock_enable_timestamp(struct sock *sk, int flag)
2471{
2472	if (!sock_flag(sk, flag)) {
2473		unsigned long previous_flags = sk->sk_flags;
2474
2475		sock_set_flag(sk, flag);
2476		/*
2477		 * we just set one of the two flags which require net
2478		 * time stamping, but time stamping might have been on
2479		 * already because of the other one
2480		 */
2481		if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2482			net_enable_timestamp();
2483	}
2484}
2485
2486int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2487		       int level, int type)
2488{
2489	struct sock_exterr_skb *serr;
2490	struct sk_buff *skb;
2491	int copied, err;
2492
2493	err = -EAGAIN;
2494	skb = sock_dequeue_err_skb(sk);
2495	if (skb == NULL)
2496		goto out;
2497
2498	copied = skb->len;
2499	if (copied > len) {
2500		msg->msg_flags |= MSG_TRUNC;
2501		copied = len;
2502	}
2503	err = skb_copy_datagram_msg(skb, 0, msg, copied);
2504	if (err)
2505		goto out_free_skb;
2506
2507	sock_recv_timestamp(msg, sk, skb);
2508
2509	serr = SKB_EXT_ERR(skb);
2510	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2511
2512	msg->msg_flags |= MSG_ERRQUEUE;
2513	err = copied;
2514
2515out_free_skb:
2516	kfree_skb(skb);
2517out:
2518	return err;
2519}
2520EXPORT_SYMBOL(sock_recv_errqueue);
2521
2522/*
2523 *	Get a socket option on an socket.
2524 *
2525 *	FIX: POSIX 1003.1g is very ambiguous here. It states that
2526 *	asynchronous errors should be reported by getsockopt. We assume
2527 *	this means if you specify SO_ERROR (otherwise whats the point of it).
2528 */
2529int sock_common_getsockopt(struct socket *sock, int level, int optname,
2530			   char __user *optval, int __user *optlen)
2531{
2532	struct sock *sk = sock->sk;
2533
2534	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2535}
2536EXPORT_SYMBOL(sock_common_getsockopt);
2537
2538#ifdef CONFIG_COMPAT
2539int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2540				  char __user *optval, int __user *optlen)
2541{
2542	struct sock *sk = sock->sk;
2543
2544	if (sk->sk_prot->compat_getsockopt != NULL)
2545		return sk->sk_prot->compat_getsockopt(sk, level, optname,
2546						      optval, optlen);
2547	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2548}
2549EXPORT_SYMBOL(compat_sock_common_getsockopt);
2550#endif
2551
2552int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2553			int flags)
2554{
2555	struct sock *sk = sock->sk;
2556	int addr_len = 0;
2557	int err;
2558
2559	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2560				   flags & ~MSG_DONTWAIT, &addr_len);
2561	if (err >= 0)
2562		msg->msg_namelen = addr_len;
2563	return err;
2564}
2565EXPORT_SYMBOL(sock_common_recvmsg);
2566
2567/*
2568 *	Set socket options on an inet socket.
2569 */
2570int sock_common_setsockopt(struct socket *sock, int level, int optname,
2571			   char __user *optval, unsigned int optlen)
2572{
2573	struct sock *sk = sock->sk;
2574
2575	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2576}
2577EXPORT_SYMBOL(sock_common_setsockopt);
2578
2579#ifdef CONFIG_COMPAT
2580int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2581				  char __user *optval, unsigned int optlen)
2582{
2583	struct sock *sk = sock->sk;
2584
2585	if (sk->sk_prot->compat_setsockopt != NULL)
2586		return sk->sk_prot->compat_setsockopt(sk, level, optname,
2587						      optval, optlen);
2588	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2589}
2590EXPORT_SYMBOL(compat_sock_common_setsockopt);
2591#endif
2592
2593void sk_common_release(struct sock *sk)
2594{
2595	if (sk->sk_prot->destroy)
2596		sk->sk_prot->destroy(sk);
2597
2598	/*
2599	 * Observation: when sock_common_release is called, processes have
2600	 * no access to socket. But net still has.
2601	 * Step one, detach it from networking:
2602	 *
2603	 * A. Remove from hash tables.
2604	 */
2605
2606	sk->sk_prot->unhash(sk);
2607
2608	/*
2609	 * In this point socket cannot receive new packets, but it is possible
2610	 * that some packets are in flight because some CPU runs receiver and
2611	 * did hash table lookup before we unhashed socket. They will achieve
2612	 * receive queue and will be purged by socket destructor.
2613	 *
2614	 * Also we still have packets pending on receive queue and probably,
2615	 * our own packets waiting in device queues. sock_destroy will drain
2616	 * receive queue, but transmitted packets will delay socket destruction
2617	 * until the last reference will be released.
2618	 */
2619
2620	sock_orphan(sk);
2621
2622	xfrm_sk_free_policy(sk);
2623
2624	sk_refcnt_debug_release(sk);
2625
2626	if (sk->sk_frag.page) {
2627		put_page(sk->sk_frag.page);
2628		sk->sk_frag.page = NULL;
2629	}
2630
2631	sock_put(sk);
2632}
2633EXPORT_SYMBOL(sk_common_release);
2634
2635#ifdef CONFIG_PROC_FS
2636#define PROTO_INUSE_NR	64	/* should be enough for the first time */
2637struct prot_inuse {
2638	int val[PROTO_INUSE_NR];
2639};
2640
2641static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2642
2643#ifdef CONFIG_NET_NS
2644void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2645{
2646	__this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2647}
2648EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2649
2650int sock_prot_inuse_get(struct net *net, struct proto *prot)
2651{
2652	int cpu, idx = prot->inuse_idx;
2653	int res = 0;
2654
2655	for_each_possible_cpu(cpu)
2656		res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2657
2658	return res >= 0 ? res : 0;
2659}
2660EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2661
2662static int __net_init sock_inuse_init_net(struct net *net)
2663{
2664	net->core.inuse = alloc_percpu(struct prot_inuse);
2665	return net->core.inuse ? 0 : -ENOMEM;
2666}
2667
2668static void __net_exit sock_inuse_exit_net(struct net *net)
2669{
2670	free_percpu(net->core.inuse);
2671}
2672
2673static struct pernet_operations net_inuse_ops = {
2674	.init = sock_inuse_init_net,
2675	.exit = sock_inuse_exit_net,
2676};
2677
2678static __init int net_inuse_init(void)
2679{
2680	if (register_pernet_subsys(&net_inuse_ops))
2681		panic("Cannot initialize net inuse counters");
2682
2683	return 0;
2684}
2685
2686core_initcall(net_inuse_init);
2687#else
2688static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2689
2690void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2691{
2692	__this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2693}
2694EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2695
2696int sock_prot_inuse_get(struct net *net, struct proto *prot)
2697{
2698	int cpu, idx = prot->inuse_idx;
2699	int res = 0;
2700
2701	for_each_possible_cpu(cpu)
2702		res += per_cpu(prot_inuse, cpu).val[idx];
2703
2704	return res >= 0 ? res : 0;
2705}
2706EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2707#endif
2708
2709static void assign_proto_idx(struct proto *prot)
2710{
2711	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2712
2713	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2714		pr_err("PROTO_INUSE_NR exhausted\n");
2715		return;
2716	}
2717
2718	set_bit(prot->inuse_idx, proto_inuse_idx);
2719}
2720
2721static void release_proto_idx(struct proto *prot)
2722{
2723	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2724		clear_bit(prot->inuse_idx, proto_inuse_idx);
2725}
2726#else
2727static inline void assign_proto_idx(struct proto *prot)
2728{
2729}
2730
2731static inline void release_proto_idx(struct proto *prot)
2732{
2733}
2734#endif
2735
2736static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2737{
2738	if (!rsk_prot)
2739		return;
2740	kfree(rsk_prot->slab_name);
2741	rsk_prot->slab_name = NULL;
2742	if (rsk_prot->slab) {
2743		kmem_cache_destroy(rsk_prot->slab);
2744		rsk_prot->slab = NULL;
2745	}
2746}
2747
2748static int req_prot_init(const struct proto *prot)
2749{
2750	struct request_sock_ops *rsk_prot = prot->rsk_prot;
2751
2752	if (!rsk_prot)
2753		return 0;
2754
2755	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2756					prot->name);
2757	if (!rsk_prot->slab_name)
2758		return -ENOMEM;
2759
2760	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2761					   rsk_prot->obj_size, 0,
2762					   0, NULL);
2763
2764	if (!rsk_prot->slab) {
2765		pr_crit("%s: Can't create request sock SLAB cache!\n",
2766			prot->name);
2767		return -ENOMEM;
2768	}
2769	return 0;
2770}
2771
2772int proto_register(struct proto *prot, int alloc_slab)
2773{
2774	if (alloc_slab) {
2775		prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2776					SLAB_HWCACHE_ALIGN | prot->slab_flags,
2777					NULL);
2778
2779		if (prot->slab == NULL) {
2780			pr_crit("%s: Can't create sock SLAB cache!\n",
2781				prot->name);
2782			goto out;
2783		}
2784
2785		if (req_prot_init(prot))
2786			goto out_free_request_sock_slab;
2787
2788		if (prot->twsk_prot != NULL) {
2789			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2790
2791			if (prot->twsk_prot->twsk_slab_name == NULL)
2792				goto out_free_request_sock_slab;
2793
2794			prot->twsk_prot->twsk_slab =
2795				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2796						  prot->twsk_prot->twsk_obj_size,
2797						  0,
2798						  prot->slab_flags,
2799						  NULL);
2800			if (prot->twsk_prot->twsk_slab == NULL)
2801				goto out_free_timewait_sock_slab_name;
2802		}
2803	}
2804
2805	mutex_lock(&proto_list_mutex);
2806	list_add(&prot->node, &proto_list);
2807	assign_proto_idx(prot);
2808	mutex_unlock(&proto_list_mutex);
2809	return 0;
2810
2811out_free_timewait_sock_slab_name:
2812	kfree(prot->twsk_prot->twsk_slab_name);
2813out_free_request_sock_slab:
2814	req_prot_cleanup(prot->rsk_prot);
2815
2816	kmem_cache_destroy(prot->slab);
2817	prot->slab = NULL;
2818out:
2819	return -ENOBUFS;
2820}
2821EXPORT_SYMBOL(proto_register);
2822
2823void proto_unregister(struct proto *prot)
2824{
2825	mutex_lock(&proto_list_mutex);
2826	release_proto_idx(prot);
2827	list_del(&prot->node);
2828	mutex_unlock(&proto_list_mutex);
2829
2830	if (prot->slab != NULL) {
2831		kmem_cache_destroy(prot->slab);
2832		prot->slab = NULL;
2833	}
2834
2835	req_prot_cleanup(prot->rsk_prot);
2836
2837	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2838		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2839		kfree(prot->twsk_prot->twsk_slab_name);
2840		prot->twsk_prot->twsk_slab = NULL;
2841	}
2842}
2843EXPORT_SYMBOL(proto_unregister);
2844
2845#ifdef CONFIG_PROC_FS
2846static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2847	__acquires(proto_list_mutex)
2848{
2849	mutex_lock(&proto_list_mutex);
2850	return seq_list_start_head(&proto_list, *pos);
2851}
2852
2853static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2854{
2855	return seq_list_next(v, &proto_list, pos);
2856}
2857
2858static void proto_seq_stop(struct seq_file *seq, void *v)
2859	__releases(proto_list_mutex)
2860{
2861	mutex_unlock(&proto_list_mutex);
2862}
2863
2864static char proto_method_implemented(const void *method)
2865{
2866	return method == NULL ? 'n' : 'y';
2867}
2868static long sock_prot_memory_allocated(struct proto *proto)
2869{
2870	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2871}
2872
2873static char *sock_prot_memory_pressure(struct proto *proto)
2874{
2875	return proto->memory_pressure != NULL ?
2876	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2877}
2878
2879static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2880{
2881
2882	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2883			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2884		   proto->name,
2885		   proto->obj_size,
2886		   sock_prot_inuse_get(seq_file_net(seq), proto),
2887		   sock_prot_memory_allocated(proto),
2888		   sock_prot_memory_pressure(proto),
2889		   proto->max_header,
2890		   proto->slab == NULL ? "no" : "yes",
2891		   module_name(proto->owner),
2892		   proto_method_implemented(proto->close),
2893		   proto_method_implemented(proto->connect),
2894		   proto_method_implemented(proto->disconnect),
2895		   proto_method_implemented(proto->accept),
2896		   proto_method_implemented(proto->ioctl),
2897		   proto_method_implemented(proto->init),
2898		   proto_method_implemented(proto->destroy),
2899		   proto_method_implemented(proto->shutdown),
2900		   proto_method_implemented(proto->setsockopt),
2901		   proto_method_implemented(proto->getsockopt),
2902		   proto_method_implemented(proto->sendmsg),
2903		   proto_method_implemented(proto->recvmsg),
2904		   proto_method_implemented(proto->sendpage),
2905		   proto_method_implemented(proto->bind),
2906		   proto_method_implemented(proto->backlog_rcv),
2907		   proto_method_implemented(proto->hash),
2908		   proto_method_implemented(proto->unhash),
2909		   proto_method_implemented(proto->get_port),
2910		   proto_method_implemented(proto->enter_memory_pressure));
2911}
2912
2913static int proto_seq_show(struct seq_file *seq, void *v)
2914{
2915	if (v == &proto_list)
2916		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2917			   "protocol",
2918			   "size",
2919			   "sockets",
2920			   "memory",
2921			   "press",
2922			   "maxhdr",
2923			   "slab",
2924			   "module",
2925			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2926	else
2927		proto_seq_printf(seq, list_entry(v, struct proto, node));
2928	return 0;
2929}
2930
2931static const struct seq_operations proto_seq_ops = {
2932	.start  = proto_seq_start,
2933	.next   = proto_seq_next,
2934	.stop   = proto_seq_stop,
2935	.show   = proto_seq_show,
2936};
2937
2938static int proto_seq_open(struct inode *inode, struct file *file)
2939{
2940	return seq_open_net(inode, file, &proto_seq_ops,
2941			    sizeof(struct seq_net_private));
2942}
2943
2944static const struct file_operations proto_seq_fops = {
2945	.owner		= THIS_MODULE,
2946	.open		= proto_seq_open,
2947	.read		= seq_read,
2948	.llseek		= seq_lseek,
2949	.release	= seq_release_net,
2950};
2951
2952static __net_init int proto_init_net(struct net *net)
2953{
2954	if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2955		return -ENOMEM;
2956
2957	return 0;
2958}
2959
2960static __net_exit void proto_exit_net(struct net *net)
2961{
2962	remove_proc_entry("protocols", net->proc_net);
2963}
2964
2965
2966static __net_initdata struct pernet_operations proto_net_ops = {
2967	.init = proto_init_net,
2968	.exit = proto_exit_net,
2969};
2970
2971static int __init proto_init(void)
2972{
2973	return register_pernet_subsys(&proto_net_ops);
2974}
2975
2976subsys_initcall(proto_init);
2977
2978#endif /* PROC_FS */
2979