PageRenderTime 155ms CodeModel.GetById 24ms app.highlight 115ms RepoModel.GetById 1ms app.codeStats 0ms

/net/netlink/af_netlink.c

http://github.com/mirrors/linux
C | 2780 lines | 2127 code | 449 blank | 204 comment | 393 complexity | 517b2a5492a812d872ac4822651117e8 MD5 | raw file

Large files files are truncated, but you can click here to view the full file

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * NETLINK      Kernel-user communication protocol.
   4 *
   5 * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
   6 * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
   7 * 				Patrick McHardy <kaber@trash.net>
   8 *
   9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
  10 *                               added netlink_proto_exit
  11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
  12 * 				 use nlk_sk, as sk->protinfo is on a diet 8)
  13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
  14 * 				 - inc module use count of module that owns
  15 * 				   the kernel socket in case userspace opens
  16 * 				   socket of same protocol
  17 * 				 - remove all module support, since netlink is
  18 * 				   mandatory if CONFIG_NET=y these days
  19 */
  20
  21#include <linux/module.h>
  22
  23#include <linux/capability.h>
  24#include <linux/kernel.h>
  25#include <linux/init.h>
  26#include <linux/signal.h>
  27#include <linux/sched.h>
  28#include <linux/errno.h>
  29#include <linux/string.h>
  30#include <linux/stat.h>
  31#include <linux/socket.h>
  32#include <linux/un.h>
  33#include <linux/fcntl.h>
  34#include <linux/termios.h>
  35#include <linux/sockios.h>
  36#include <linux/net.h>
  37#include <linux/fs.h>
  38#include <linux/slab.h>
  39#include <linux/uaccess.h>
  40#include <linux/skbuff.h>
  41#include <linux/netdevice.h>
  42#include <linux/rtnetlink.h>
  43#include <linux/proc_fs.h>
  44#include <linux/seq_file.h>
  45#include <linux/notifier.h>
  46#include <linux/security.h>
  47#include <linux/jhash.h>
  48#include <linux/jiffies.h>
  49#include <linux/random.h>
  50#include <linux/bitops.h>
  51#include <linux/mm.h>
  52#include <linux/types.h>
  53#include <linux/audit.h>
  54#include <linux/mutex.h>
  55#include <linux/vmalloc.h>
  56#include <linux/if_arp.h>
  57#include <linux/rhashtable.h>
  58#include <asm/cacheflush.h>
  59#include <linux/hash.h>
  60#include <linux/genetlink.h>
  61#include <linux/net_namespace.h>
  62#include <linux/nospec.h>
  63
  64#include <net/net_namespace.h>
  65#include <net/netns/generic.h>
  66#include <net/sock.h>
  67#include <net/scm.h>
  68#include <net/netlink.h>
  69
  70#include "af_netlink.h"
  71
  72struct listeners {
  73	struct rcu_head		rcu;
  74	unsigned long		masks[];
  75};
  76
  77/* state bits */
  78#define NETLINK_S_CONGESTED		0x0
  79
  80static inline int netlink_is_kernel(struct sock *sk)
  81{
  82	return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
  83}
  84
  85struct netlink_table *nl_table __read_mostly;
  86EXPORT_SYMBOL_GPL(nl_table);
  87
  88static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
  89
  90static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
  91
  92static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
  93	"nlk_cb_mutex-ROUTE",
  94	"nlk_cb_mutex-1",
  95	"nlk_cb_mutex-USERSOCK",
  96	"nlk_cb_mutex-FIREWALL",
  97	"nlk_cb_mutex-SOCK_DIAG",
  98	"nlk_cb_mutex-NFLOG",
  99	"nlk_cb_mutex-XFRM",
 100	"nlk_cb_mutex-SELINUX",
 101	"nlk_cb_mutex-ISCSI",
 102	"nlk_cb_mutex-AUDIT",
 103	"nlk_cb_mutex-FIB_LOOKUP",
 104	"nlk_cb_mutex-CONNECTOR",
 105	"nlk_cb_mutex-NETFILTER",
 106	"nlk_cb_mutex-IP6_FW",
 107	"nlk_cb_mutex-DNRTMSG",
 108	"nlk_cb_mutex-KOBJECT_UEVENT",
 109	"nlk_cb_mutex-GENERIC",
 110	"nlk_cb_mutex-17",
 111	"nlk_cb_mutex-SCSITRANSPORT",
 112	"nlk_cb_mutex-ECRYPTFS",
 113	"nlk_cb_mutex-RDMA",
 114	"nlk_cb_mutex-CRYPTO",
 115	"nlk_cb_mutex-SMC",
 116	"nlk_cb_mutex-23",
 117	"nlk_cb_mutex-24",
 118	"nlk_cb_mutex-25",
 119	"nlk_cb_mutex-26",
 120	"nlk_cb_mutex-27",
 121	"nlk_cb_mutex-28",
 122	"nlk_cb_mutex-29",
 123	"nlk_cb_mutex-30",
 124	"nlk_cb_mutex-31",
 125	"nlk_cb_mutex-MAX_LINKS"
 126};
 127
 128static int netlink_dump(struct sock *sk);
 129
 130/* nl_table locking explained:
 131 * Lookup and traversal are protected with an RCU read-side lock. Insertion
 132 * and removal are protected with per bucket lock while using RCU list
 133 * modification primitives and may run in parallel to RCU protected lookups.
 134 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
 135 * been acquired * either during or after the socket has been removed from
 136 * the list and after an RCU grace period.
 137 */
 138DEFINE_RWLOCK(nl_table_lock);
 139EXPORT_SYMBOL_GPL(nl_table_lock);
 140static atomic_t nl_table_users = ATOMIC_INIT(0);
 141
 142#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
 143
 144static BLOCKING_NOTIFIER_HEAD(netlink_chain);
 145
 146
 147static const struct rhashtable_params netlink_rhashtable_params;
 148
 149static inline u32 netlink_group_mask(u32 group)
 150{
 151	return group ? 1 << (group - 1) : 0;
 152}
 153
 154static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
 155					   gfp_t gfp_mask)
 156{
 157	unsigned int len = skb_end_offset(skb);
 158	struct sk_buff *new;
 159
 160	new = alloc_skb(len, gfp_mask);
 161	if (new == NULL)
 162		return NULL;
 163
 164	NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
 165	NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
 166	NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
 167
 168	skb_put_data(new, skb->data, len);
 169	return new;
 170}
 171
 172static unsigned int netlink_tap_net_id;
 173
 174struct netlink_tap_net {
 175	struct list_head netlink_tap_all;
 176	struct mutex netlink_tap_lock;
 177};
 178
 179int netlink_add_tap(struct netlink_tap *nt)
 180{
 181	struct net *net = dev_net(nt->dev);
 182	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 183
 184	if (unlikely(nt->dev->type != ARPHRD_NETLINK))
 185		return -EINVAL;
 186
 187	mutex_lock(&nn->netlink_tap_lock);
 188	list_add_rcu(&nt->list, &nn->netlink_tap_all);
 189	mutex_unlock(&nn->netlink_tap_lock);
 190
 191	__module_get(nt->module);
 192
 193	return 0;
 194}
 195EXPORT_SYMBOL_GPL(netlink_add_tap);
 196
 197static int __netlink_remove_tap(struct netlink_tap *nt)
 198{
 199	struct net *net = dev_net(nt->dev);
 200	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 201	bool found = false;
 202	struct netlink_tap *tmp;
 203
 204	mutex_lock(&nn->netlink_tap_lock);
 205
 206	list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
 207		if (nt == tmp) {
 208			list_del_rcu(&nt->list);
 209			found = true;
 210			goto out;
 211		}
 212	}
 213
 214	pr_warn("__netlink_remove_tap: %p not found\n", nt);
 215out:
 216	mutex_unlock(&nn->netlink_tap_lock);
 217
 218	if (found)
 219		module_put(nt->module);
 220
 221	return found ? 0 : -ENODEV;
 222}
 223
 224int netlink_remove_tap(struct netlink_tap *nt)
 225{
 226	int ret;
 227
 228	ret = __netlink_remove_tap(nt);
 229	synchronize_net();
 230
 231	return ret;
 232}
 233EXPORT_SYMBOL_GPL(netlink_remove_tap);
 234
 235static __net_init int netlink_tap_init_net(struct net *net)
 236{
 237	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 238
 239	INIT_LIST_HEAD(&nn->netlink_tap_all);
 240	mutex_init(&nn->netlink_tap_lock);
 241	return 0;
 242}
 243
 244static struct pernet_operations netlink_tap_net_ops = {
 245	.init = netlink_tap_init_net,
 246	.id   = &netlink_tap_net_id,
 247	.size = sizeof(struct netlink_tap_net),
 248};
 249
 250static bool netlink_filter_tap(const struct sk_buff *skb)
 251{
 252	struct sock *sk = skb->sk;
 253
 254	/* We take the more conservative approach and
 255	 * whitelist socket protocols that may pass.
 256	 */
 257	switch (sk->sk_protocol) {
 258	case NETLINK_ROUTE:
 259	case NETLINK_USERSOCK:
 260	case NETLINK_SOCK_DIAG:
 261	case NETLINK_NFLOG:
 262	case NETLINK_XFRM:
 263	case NETLINK_FIB_LOOKUP:
 264	case NETLINK_NETFILTER:
 265	case NETLINK_GENERIC:
 266		return true;
 267	}
 268
 269	return false;
 270}
 271
 272static int __netlink_deliver_tap_skb(struct sk_buff *skb,
 273				     struct net_device *dev)
 274{
 275	struct sk_buff *nskb;
 276	struct sock *sk = skb->sk;
 277	int ret = -ENOMEM;
 278
 279	if (!net_eq(dev_net(dev), sock_net(sk)))
 280		return 0;
 281
 282	dev_hold(dev);
 283
 284	if (is_vmalloc_addr(skb->head))
 285		nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
 286	else
 287		nskb = skb_clone(skb, GFP_ATOMIC);
 288	if (nskb) {
 289		nskb->dev = dev;
 290		nskb->protocol = htons((u16) sk->sk_protocol);
 291		nskb->pkt_type = netlink_is_kernel(sk) ?
 292				 PACKET_KERNEL : PACKET_USER;
 293		skb_reset_network_header(nskb);
 294		ret = dev_queue_xmit(nskb);
 295		if (unlikely(ret > 0))
 296			ret = net_xmit_errno(ret);
 297	}
 298
 299	dev_put(dev);
 300	return ret;
 301}
 302
 303static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
 304{
 305	int ret;
 306	struct netlink_tap *tmp;
 307
 308	if (!netlink_filter_tap(skb))
 309		return;
 310
 311	list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
 312		ret = __netlink_deliver_tap_skb(skb, tmp->dev);
 313		if (unlikely(ret))
 314			break;
 315	}
 316}
 317
 318static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
 319{
 320	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 321
 322	rcu_read_lock();
 323
 324	if (unlikely(!list_empty(&nn->netlink_tap_all)))
 325		__netlink_deliver_tap(skb, nn);
 326
 327	rcu_read_unlock();
 328}
 329
 330static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
 331				       struct sk_buff *skb)
 332{
 333	if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
 334		netlink_deliver_tap(sock_net(dst), skb);
 335}
 336
 337static void netlink_overrun(struct sock *sk)
 338{
 339	struct netlink_sock *nlk = nlk_sk(sk);
 340
 341	if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
 342		if (!test_and_set_bit(NETLINK_S_CONGESTED,
 343				      &nlk_sk(sk)->state)) {
 344			sk->sk_err = ENOBUFS;
 345			sk->sk_error_report(sk);
 346		}
 347	}
 348	atomic_inc(&sk->sk_drops);
 349}
 350
 351static void netlink_rcv_wake(struct sock *sk)
 352{
 353	struct netlink_sock *nlk = nlk_sk(sk);
 354
 355	if (skb_queue_empty(&sk->sk_receive_queue))
 356		clear_bit(NETLINK_S_CONGESTED, &nlk->state);
 357	if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
 358		wake_up_interruptible(&nlk->wait);
 359}
 360
 361static void netlink_skb_destructor(struct sk_buff *skb)
 362{
 363	if (is_vmalloc_addr(skb->head)) {
 364		if (!skb->cloned ||
 365		    !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
 366			vfree(skb->head);
 367
 368		skb->head = NULL;
 369	}
 370	if (skb->sk != NULL)
 371		sock_rfree(skb);
 372}
 373
 374static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
 375{
 376	WARN_ON(skb->sk != NULL);
 377	skb->sk = sk;
 378	skb->destructor = netlink_skb_destructor;
 379	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
 380	sk_mem_charge(sk, skb->truesize);
 381}
 382
 383static void netlink_sock_destruct(struct sock *sk)
 384{
 385	struct netlink_sock *nlk = nlk_sk(sk);
 386
 387	if (nlk->cb_running) {
 388		if (nlk->cb.done)
 389			nlk->cb.done(&nlk->cb);
 390		module_put(nlk->cb.module);
 391		kfree_skb(nlk->cb.skb);
 392	}
 393
 394	skb_queue_purge(&sk->sk_receive_queue);
 395
 396	if (!sock_flag(sk, SOCK_DEAD)) {
 397		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
 398		return;
 399	}
 400
 401	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 402	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 403	WARN_ON(nlk_sk(sk)->groups);
 404}
 405
 406static void netlink_sock_destruct_work(struct work_struct *work)
 407{
 408	struct netlink_sock *nlk = container_of(work, struct netlink_sock,
 409						work);
 410
 411	sk_free(&nlk->sk);
 412}
 413
 414/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
 415 * SMP. Look, when several writers sleep and reader wakes them up, all but one
 416 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 417 * this, _but_ remember, it adds useless work on UP machines.
 418 */
 419
 420void netlink_table_grab(void)
 421	__acquires(nl_table_lock)
 422{
 423	might_sleep();
 424
 425	write_lock_irq(&nl_table_lock);
 426
 427	if (atomic_read(&nl_table_users)) {
 428		DECLARE_WAITQUEUE(wait, current);
 429
 430		add_wait_queue_exclusive(&nl_table_wait, &wait);
 431		for (;;) {
 432			set_current_state(TASK_UNINTERRUPTIBLE);
 433			if (atomic_read(&nl_table_users) == 0)
 434				break;
 435			write_unlock_irq(&nl_table_lock);
 436			schedule();
 437			write_lock_irq(&nl_table_lock);
 438		}
 439
 440		__set_current_state(TASK_RUNNING);
 441		remove_wait_queue(&nl_table_wait, &wait);
 442	}
 443}
 444
 445void netlink_table_ungrab(void)
 446	__releases(nl_table_lock)
 447{
 448	write_unlock_irq(&nl_table_lock);
 449	wake_up(&nl_table_wait);
 450}
 451
 452static inline void
 453netlink_lock_table(void)
 454{
 455	/* read_lock() synchronizes us to netlink_table_grab */
 456
 457	read_lock(&nl_table_lock);
 458	atomic_inc(&nl_table_users);
 459	read_unlock(&nl_table_lock);
 460}
 461
 462static inline void
 463netlink_unlock_table(void)
 464{
 465	if (atomic_dec_and_test(&nl_table_users))
 466		wake_up(&nl_table_wait);
 467}
 468
 469struct netlink_compare_arg
 470{
 471	possible_net_t pnet;
 472	u32 portid;
 473};
 474
 475/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
 476#define netlink_compare_arg_len \
 477	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
 478
 479static inline int netlink_compare(struct rhashtable_compare_arg *arg,
 480				  const void *ptr)
 481{
 482	const struct netlink_compare_arg *x = arg->key;
 483	const struct netlink_sock *nlk = ptr;
 484
 485	return nlk->portid != x->portid ||
 486	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
 487}
 488
 489static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
 490				     struct net *net, u32 portid)
 491{
 492	memset(arg, 0, sizeof(*arg));
 493	write_pnet(&arg->pnet, net);
 494	arg->portid = portid;
 495}
 496
 497static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
 498				     struct net *net)
 499{
 500	struct netlink_compare_arg arg;
 501
 502	netlink_compare_arg_init(&arg, net, portid);
 503	return rhashtable_lookup_fast(&table->hash, &arg,
 504				      netlink_rhashtable_params);
 505}
 506
 507static int __netlink_insert(struct netlink_table *table, struct sock *sk)
 508{
 509	struct netlink_compare_arg arg;
 510
 511	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
 512	return rhashtable_lookup_insert_key(&table->hash, &arg,
 513					    &nlk_sk(sk)->node,
 514					    netlink_rhashtable_params);
 515}
 516
 517static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
 518{
 519	struct netlink_table *table = &nl_table[protocol];
 520	struct sock *sk;
 521
 522	rcu_read_lock();
 523	sk = __netlink_lookup(table, portid, net);
 524	if (sk)
 525		sock_hold(sk);
 526	rcu_read_unlock();
 527
 528	return sk;
 529}
 530
 531static const struct proto_ops netlink_ops;
 532
 533static void
 534netlink_update_listeners(struct sock *sk)
 535{
 536	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
 537	unsigned long mask;
 538	unsigned int i;
 539	struct listeners *listeners;
 540
 541	listeners = nl_deref_protected(tbl->listeners);
 542	if (!listeners)
 543		return;
 544
 545	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
 546		mask = 0;
 547		sk_for_each_bound(sk, &tbl->mc_list) {
 548			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
 549				mask |= nlk_sk(sk)->groups[i];
 550		}
 551		listeners->masks[i] = mask;
 552	}
 553	/* this function is only called with the netlink table "grabbed", which
 554	 * makes sure updates are visible before bind or setsockopt return. */
 555}
 556
 557static int netlink_insert(struct sock *sk, u32 portid)
 558{
 559	struct netlink_table *table = &nl_table[sk->sk_protocol];
 560	int err;
 561
 562	lock_sock(sk);
 563
 564	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
 565	if (nlk_sk(sk)->bound)
 566		goto err;
 567
 568	nlk_sk(sk)->portid = portid;
 569	sock_hold(sk);
 570
 571	err = __netlink_insert(table, sk);
 572	if (err) {
 573		/* In case the hashtable backend returns with -EBUSY
 574		 * from here, it must not escape to the caller.
 575		 */
 576		if (unlikely(err == -EBUSY))
 577			err = -EOVERFLOW;
 578		if (err == -EEXIST)
 579			err = -EADDRINUSE;
 580		sock_put(sk);
 581		goto err;
 582	}
 583
 584	/* We need to ensure that the socket is hashed and visible. */
 585	smp_wmb();
 586	nlk_sk(sk)->bound = portid;
 587
 588err:
 589	release_sock(sk);
 590	return err;
 591}
 592
 593static void netlink_remove(struct sock *sk)
 594{
 595	struct netlink_table *table;
 596
 597	table = &nl_table[sk->sk_protocol];
 598	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
 599				    netlink_rhashtable_params)) {
 600		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 601		__sock_put(sk);
 602	}
 603
 604	netlink_table_grab();
 605	if (nlk_sk(sk)->subscriptions) {
 606		__sk_del_bind_node(sk);
 607		netlink_update_listeners(sk);
 608	}
 609	if (sk->sk_protocol == NETLINK_GENERIC)
 610		atomic_inc(&genl_sk_destructing_cnt);
 611	netlink_table_ungrab();
 612}
 613
 614static struct proto netlink_proto = {
 615	.name	  = "NETLINK",
 616	.owner	  = THIS_MODULE,
 617	.obj_size = sizeof(struct netlink_sock),
 618};
 619
 620static int __netlink_create(struct net *net, struct socket *sock,
 621			    struct mutex *cb_mutex, int protocol,
 622			    int kern)
 623{
 624	struct sock *sk;
 625	struct netlink_sock *nlk;
 626
 627	sock->ops = &netlink_ops;
 628
 629	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
 630	if (!sk)
 631		return -ENOMEM;
 632
 633	sock_init_data(sock, sk);
 634
 635	nlk = nlk_sk(sk);
 636	if (cb_mutex) {
 637		nlk->cb_mutex = cb_mutex;
 638	} else {
 639		nlk->cb_mutex = &nlk->cb_def_mutex;
 640		mutex_init(nlk->cb_mutex);
 641		lockdep_set_class_and_name(nlk->cb_mutex,
 642					   nlk_cb_mutex_keys + protocol,
 643					   nlk_cb_mutex_key_strings[protocol]);
 644	}
 645	init_waitqueue_head(&nlk->wait);
 646
 647	sk->sk_destruct = netlink_sock_destruct;
 648	sk->sk_protocol = protocol;
 649	return 0;
 650}
 651
 652static int netlink_create(struct net *net, struct socket *sock, int protocol,
 653			  int kern)
 654{
 655	struct module *module = NULL;
 656	struct mutex *cb_mutex;
 657	struct netlink_sock *nlk;
 658	int (*bind)(struct net *net, int group);
 659	void (*unbind)(struct net *net, int group);
 660	int err = 0;
 661
 662	sock->state = SS_UNCONNECTED;
 663
 664	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 665		return -ESOCKTNOSUPPORT;
 666
 667	if (protocol < 0 || protocol >= MAX_LINKS)
 668		return -EPROTONOSUPPORT;
 669	protocol = array_index_nospec(protocol, MAX_LINKS);
 670
 671	netlink_lock_table();
 672#ifdef CONFIG_MODULES
 673	if (!nl_table[protocol].registered) {
 674		netlink_unlock_table();
 675		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
 676		netlink_lock_table();
 677	}
 678#endif
 679	if (nl_table[protocol].registered &&
 680	    try_module_get(nl_table[protocol].module))
 681		module = nl_table[protocol].module;
 682	else
 683		err = -EPROTONOSUPPORT;
 684	cb_mutex = nl_table[protocol].cb_mutex;
 685	bind = nl_table[protocol].bind;
 686	unbind = nl_table[protocol].unbind;
 687	netlink_unlock_table();
 688
 689	if (err < 0)
 690		goto out;
 691
 692	err = __netlink_create(net, sock, cb_mutex, protocol, kern);
 693	if (err < 0)
 694		goto out_module;
 695
 696	local_bh_disable();
 697	sock_prot_inuse_add(net, &netlink_proto, 1);
 698	local_bh_enable();
 699
 700	nlk = nlk_sk(sock->sk);
 701	nlk->module = module;
 702	nlk->netlink_bind = bind;
 703	nlk->netlink_unbind = unbind;
 704out:
 705	return err;
 706
 707out_module:
 708	module_put(module);
 709	goto out;
 710}
 711
 712static void deferred_put_nlk_sk(struct rcu_head *head)
 713{
 714	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
 715	struct sock *sk = &nlk->sk;
 716
 717	kfree(nlk->groups);
 718	nlk->groups = NULL;
 719
 720	if (!refcount_dec_and_test(&sk->sk_refcnt))
 721		return;
 722
 723	if (nlk->cb_running && nlk->cb.done) {
 724		INIT_WORK(&nlk->work, netlink_sock_destruct_work);
 725		schedule_work(&nlk->work);
 726		return;
 727	}
 728
 729	sk_free(sk);
 730}
 731
 732static int netlink_release(struct socket *sock)
 733{
 734	struct sock *sk = sock->sk;
 735	struct netlink_sock *nlk;
 736
 737	if (!sk)
 738		return 0;
 739
 740	netlink_remove(sk);
 741	sock_orphan(sk);
 742	nlk = nlk_sk(sk);
 743
 744	/*
 745	 * OK. Socket is unlinked, any packets that arrive now
 746	 * will be purged.
 747	 */
 748
 749	/* must not acquire netlink_table_lock in any way again before unbind
 750	 * and notifying genetlink is done as otherwise it might deadlock
 751	 */
 752	if (nlk->netlink_unbind) {
 753		int i;
 754
 755		for (i = 0; i < nlk->ngroups; i++)
 756			if (test_bit(i, nlk->groups))
 757				nlk->netlink_unbind(sock_net(sk), i + 1);
 758	}
 759	if (sk->sk_protocol == NETLINK_GENERIC &&
 760	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
 761		wake_up(&genl_sk_destructing_waitq);
 762
 763	sock->sk = NULL;
 764	wake_up_interruptible_all(&nlk->wait);
 765
 766	skb_queue_purge(&sk->sk_write_queue);
 767
 768	if (nlk->portid && nlk->bound) {
 769		struct netlink_notify n = {
 770						.net = sock_net(sk),
 771						.protocol = sk->sk_protocol,
 772						.portid = nlk->portid,
 773					  };
 774		blocking_notifier_call_chain(&netlink_chain,
 775				NETLINK_URELEASE, &n);
 776	}
 777
 778	module_put(nlk->module);
 779
 780	if (netlink_is_kernel(sk)) {
 781		netlink_table_grab();
 782		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
 783		if (--nl_table[sk->sk_protocol].registered == 0) {
 784			struct listeners *old;
 785
 786			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
 787			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
 788			kfree_rcu(old, rcu);
 789			nl_table[sk->sk_protocol].module = NULL;
 790			nl_table[sk->sk_protocol].bind = NULL;
 791			nl_table[sk->sk_protocol].unbind = NULL;
 792			nl_table[sk->sk_protocol].flags = 0;
 793			nl_table[sk->sk_protocol].registered = 0;
 794		}
 795		netlink_table_ungrab();
 796	}
 797
 798	local_bh_disable();
 799	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
 800	local_bh_enable();
 801	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
 802	return 0;
 803}
 804
 805static int netlink_autobind(struct socket *sock)
 806{
 807	struct sock *sk = sock->sk;
 808	struct net *net = sock_net(sk);
 809	struct netlink_table *table = &nl_table[sk->sk_protocol];
 810	s32 portid = task_tgid_vnr(current);
 811	int err;
 812	s32 rover = -4096;
 813	bool ok;
 814
 815retry:
 816	cond_resched();
 817	rcu_read_lock();
 818	ok = !__netlink_lookup(table, portid, net);
 819	rcu_read_unlock();
 820	if (!ok) {
 821		/* Bind collision, search negative portid values. */
 822		if (rover == -4096)
 823			/* rover will be in range [S32_MIN, -4097] */
 824			rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
 825		else if (rover >= -4096)
 826			rover = -4097;
 827		portid = rover--;
 828		goto retry;
 829	}
 830
 831	err = netlink_insert(sk, portid);
 832	if (err == -EADDRINUSE)
 833		goto retry;
 834
 835	/* If 2 threads race to autobind, that is fine.  */
 836	if (err == -EBUSY)
 837		err = 0;
 838
 839	return err;
 840}
 841
 842/**
 843 * __netlink_ns_capable - General netlink message capability test
 844 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
 845 * @user_ns: The user namespace of the capability to use
 846 * @cap: The capability to use
 847 *
 848 * Test to see if the opener of the socket we received the message
 849 * from had when the netlink socket was created and the sender of the
 850 * message has has the capability @cap in the user namespace @user_ns.
 851 */
 852bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
 853			struct user_namespace *user_ns, int cap)
 854{
 855	return ((nsp->flags & NETLINK_SKB_DST) ||
 856		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
 857		ns_capable(user_ns, cap);
 858}
 859EXPORT_SYMBOL(__netlink_ns_capable);
 860
 861/**
 862 * netlink_ns_capable - General netlink message capability test
 863 * @skb: socket buffer holding a netlink command from userspace
 864 * @user_ns: The user namespace of the capability to use
 865 * @cap: The capability to use
 866 *
 867 * Test to see if the opener of the socket we received the message
 868 * from had when the netlink socket was created and the sender of the
 869 * message has has the capability @cap in the user namespace @user_ns.
 870 */
 871bool netlink_ns_capable(const struct sk_buff *skb,
 872			struct user_namespace *user_ns, int cap)
 873{
 874	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
 875}
 876EXPORT_SYMBOL(netlink_ns_capable);
 877
 878/**
 879 * netlink_capable - Netlink global message capability test
 880 * @skb: socket buffer holding a netlink command from userspace
 881 * @cap: The capability to use
 882 *
 883 * Test to see if the opener of the socket we received the message
 884 * from had when the netlink socket was created and the sender of the
 885 * message has has the capability @cap in all user namespaces.
 886 */
 887bool netlink_capable(const struct sk_buff *skb, int cap)
 888{
 889	return netlink_ns_capable(skb, &init_user_ns, cap);
 890}
 891EXPORT_SYMBOL(netlink_capable);
 892
 893/**
 894 * netlink_net_capable - Netlink network namespace message capability test
 895 * @skb: socket buffer holding a netlink command from userspace
 896 * @cap: The capability to use
 897 *
 898 * Test to see if the opener of the socket we received the message
 899 * from had when the netlink socket was created and the sender of the
 900 * message has has the capability @cap over the network namespace of
 901 * the socket we received the message from.
 902 */
 903bool netlink_net_capable(const struct sk_buff *skb, int cap)
 904{
 905	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
 906}
 907EXPORT_SYMBOL(netlink_net_capable);
 908
 909static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
 910{
 911	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
 912		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
 913}
 914
 915static void
 916netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
 917{
 918	struct netlink_sock *nlk = nlk_sk(sk);
 919
 920	if (nlk->subscriptions && !subscriptions)
 921		__sk_del_bind_node(sk);
 922	else if (!nlk->subscriptions && subscriptions)
 923		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
 924	nlk->subscriptions = subscriptions;
 925}
 926
 927static int netlink_realloc_groups(struct sock *sk)
 928{
 929	struct netlink_sock *nlk = nlk_sk(sk);
 930	unsigned int groups;
 931	unsigned long *new_groups;
 932	int err = 0;
 933
 934	netlink_table_grab();
 935
 936	groups = nl_table[sk->sk_protocol].groups;
 937	if (!nl_table[sk->sk_protocol].registered) {
 938		err = -ENOENT;
 939		goto out_unlock;
 940	}
 941
 942	if (nlk->ngroups >= groups)
 943		goto out_unlock;
 944
 945	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
 946	if (new_groups == NULL) {
 947		err = -ENOMEM;
 948		goto out_unlock;
 949	}
 950	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
 951	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
 952
 953	nlk->groups = new_groups;
 954	nlk->ngroups = groups;
 955 out_unlock:
 956	netlink_table_ungrab();
 957	return err;
 958}
 959
 960static void netlink_undo_bind(int group, long unsigned int groups,
 961			      struct sock *sk)
 962{
 963	struct netlink_sock *nlk = nlk_sk(sk);
 964	int undo;
 965
 966	if (!nlk->netlink_unbind)
 967		return;
 968
 969	for (undo = 0; undo < group; undo++)
 970		if (test_bit(undo, &groups))
 971			nlk->netlink_unbind(sock_net(sk), undo + 1);
 972}
 973
 974static int netlink_bind(struct socket *sock, struct sockaddr *addr,
 975			int addr_len)
 976{
 977	struct sock *sk = sock->sk;
 978	struct net *net = sock_net(sk);
 979	struct netlink_sock *nlk = nlk_sk(sk);
 980	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
 981	int err = 0;
 982	unsigned long groups;
 983	bool bound;
 984
 985	if (addr_len < sizeof(struct sockaddr_nl))
 986		return -EINVAL;
 987
 988	if (nladdr->nl_family != AF_NETLINK)
 989		return -EINVAL;
 990	groups = nladdr->nl_groups;
 991
 992	/* Only superuser is allowed to listen multicasts */
 993	if (groups) {
 994		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
 995			return -EPERM;
 996		err = netlink_realloc_groups(sk);
 997		if (err)
 998			return err;
 999	}
1000
1001	if (nlk->ngroups < BITS_PER_LONG)
1002		groups &= (1UL << nlk->ngroups) - 1;
1003
1004	bound = nlk->bound;
1005	if (bound) {
1006		/* Ensure nlk->portid is up-to-date. */
1007		smp_rmb();
1008
1009		if (nladdr->nl_pid != nlk->portid)
1010			return -EINVAL;
1011	}
1012
1013	netlink_lock_table();
1014	if (nlk->netlink_bind && groups) {
1015		int group;
1016
1017		/* nl_groups is a u32, so cap the maximum groups we can bind */
1018		for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1019			if (!test_bit(group, &groups))
1020				continue;
1021			err = nlk->netlink_bind(net, group + 1);
1022			if (!err)
1023				continue;
1024			netlink_undo_bind(group, groups, sk);
1025			goto unlock;
1026		}
1027	}
1028
1029	/* No need for barriers here as we return to user-space without
1030	 * using any of the bound attributes.
1031	 */
1032	if (!bound) {
1033		err = nladdr->nl_pid ?
1034			netlink_insert(sk, nladdr->nl_pid) :
1035			netlink_autobind(sock);
1036		if (err) {
1037			netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1038			goto unlock;
1039		}
1040	}
1041
1042	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1043		goto unlock;
1044	netlink_unlock_table();
1045
1046	netlink_table_grab();
1047	netlink_update_subscriptions(sk, nlk->subscriptions +
1048					 hweight32(groups) -
1049					 hweight32(nlk->groups[0]));
1050	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1051	netlink_update_listeners(sk);
1052	netlink_table_ungrab();
1053
1054	return 0;
1055
1056unlock:
1057	netlink_unlock_table();
1058	return err;
1059}
1060
1061static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1062			   int alen, int flags)
1063{
1064	int err = 0;
1065	struct sock *sk = sock->sk;
1066	struct netlink_sock *nlk = nlk_sk(sk);
1067	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1068
1069	if (alen < sizeof(addr->sa_family))
1070		return -EINVAL;
1071
1072	if (addr->sa_family == AF_UNSPEC) {
1073		sk->sk_state	= NETLINK_UNCONNECTED;
1074		nlk->dst_portid	= 0;
1075		nlk->dst_group  = 0;
1076		return 0;
1077	}
1078	if (addr->sa_family != AF_NETLINK)
1079		return -EINVAL;
1080
1081	if (alen < sizeof(struct sockaddr_nl))
1082		return -EINVAL;
1083
1084	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1085	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1086		return -EPERM;
1087
1088	/* No need for barriers here as we return to user-space without
1089	 * using any of the bound attributes.
1090	 */
1091	if (!nlk->bound)
1092		err = netlink_autobind(sock);
1093
1094	if (err == 0) {
1095		sk->sk_state	= NETLINK_CONNECTED;
1096		nlk->dst_portid = nladdr->nl_pid;
1097		nlk->dst_group  = ffs(nladdr->nl_groups);
1098	}
1099
1100	return err;
1101}
1102
1103static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1104			   int peer)
1105{
1106	struct sock *sk = sock->sk;
1107	struct netlink_sock *nlk = nlk_sk(sk);
1108	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1109
1110	nladdr->nl_family = AF_NETLINK;
1111	nladdr->nl_pad = 0;
1112
1113	if (peer) {
1114		nladdr->nl_pid = nlk->dst_portid;
1115		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1116	} else {
1117		nladdr->nl_pid = nlk->portid;
1118		netlink_lock_table();
1119		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1120		netlink_unlock_table();
1121	}
1122	return sizeof(*nladdr);
1123}
1124
1125static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1126			 unsigned long arg)
1127{
1128	/* try to hand this ioctl down to the NIC drivers.
1129	 */
1130	return -ENOIOCTLCMD;
1131}
1132
1133static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1134{
1135	struct sock *sock;
1136	struct netlink_sock *nlk;
1137
1138	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1139	if (!sock)
1140		return ERR_PTR(-ECONNREFUSED);
1141
1142	/* Don't bother queuing skb if kernel socket has no input function */
1143	nlk = nlk_sk(sock);
1144	if (sock->sk_state == NETLINK_CONNECTED &&
1145	    nlk->dst_portid != nlk_sk(ssk)->portid) {
1146		sock_put(sock);
1147		return ERR_PTR(-ECONNREFUSED);
1148	}
1149	return sock;
1150}
1151
1152struct sock *netlink_getsockbyfilp(struct file *filp)
1153{
1154	struct inode *inode = file_inode(filp);
1155	struct sock *sock;
1156
1157	if (!S_ISSOCK(inode->i_mode))
1158		return ERR_PTR(-ENOTSOCK);
1159
1160	sock = SOCKET_I(inode)->sk;
1161	if (sock->sk_family != AF_NETLINK)
1162		return ERR_PTR(-EINVAL);
1163
1164	sock_hold(sock);
1165	return sock;
1166}
1167
1168static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1169					       int broadcast)
1170{
1171	struct sk_buff *skb;
1172	void *data;
1173
1174	if (size <= NLMSG_GOODSIZE || broadcast)
1175		return alloc_skb(size, GFP_KERNEL);
1176
1177	size = SKB_DATA_ALIGN(size) +
1178	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1179
1180	data = vmalloc(size);
1181	if (data == NULL)
1182		return NULL;
1183
1184	skb = __build_skb(data, size);
1185	if (skb == NULL)
1186		vfree(data);
1187	else
1188		skb->destructor = netlink_skb_destructor;
1189
1190	return skb;
1191}
1192
1193/*
1194 * Attach a skb to a netlink socket.
1195 * The caller must hold a reference to the destination socket. On error, the
1196 * reference is dropped. The skb is not send to the destination, just all
1197 * all error checks are performed and memory in the queue is reserved.
1198 * Return values:
1199 * < 0: error. skb freed, reference to sock dropped.
1200 * 0: continue
1201 * 1: repeat lookup - reference dropped while waiting for socket memory.
1202 */
1203int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1204		      long *timeo, struct sock *ssk)
1205{
1206	struct netlink_sock *nlk;
1207
1208	nlk = nlk_sk(sk);
1209
1210	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1211	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1212		DECLARE_WAITQUEUE(wait, current);
1213		if (!*timeo) {
1214			if (!ssk || netlink_is_kernel(ssk))
1215				netlink_overrun(sk);
1216			sock_put(sk);
1217			kfree_skb(skb);
1218			return -EAGAIN;
1219		}
1220
1221		__set_current_state(TASK_INTERRUPTIBLE);
1222		add_wait_queue(&nlk->wait, &wait);
1223
1224		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1225		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1226		    !sock_flag(sk, SOCK_DEAD))
1227			*timeo = schedule_timeout(*timeo);
1228
1229		__set_current_state(TASK_RUNNING);
1230		remove_wait_queue(&nlk->wait, &wait);
1231		sock_put(sk);
1232
1233		if (signal_pending(current)) {
1234			kfree_skb(skb);
1235			return sock_intr_errno(*timeo);
1236		}
1237		return 1;
1238	}
1239	netlink_skb_set_owner_r(skb, sk);
1240	return 0;
1241}
1242
1243static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1244{
1245	int len = skb->len;
1246
1247	netlink_deliver_tap(sock_net(sk), skb);
1248
1249	skb_queue_tail(&sk->sk_receive_queue, skb);
1250	sk->sk_data_ready(sk);
1251	return len;
1252}
1253
1254int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1255{
1256	int len = __netlink_sendskb(sk, skb);
1257
1258	sock_put(sk);
1259	return len;
1260}
1261
1262void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1263{
1264	kfree_skb(skb);
1265	sock_put(sk);
1266}
1267
1268static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1269{
1270	int delta;
1271
1272	WARN_ON(skb->sk != NULL);
1273	delta = skb->end - skb->tail;
1274	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1275		return skb;
1276
1277	if (skb_shared(skb)) {
1278		struct sk_buff *nskb = skb_clone(skb, allocation);
1279		if (!nskb)
1280			return skb;
1281		consume_skb(skb);
1282		skb = nskb;
1283	}
1284
1285	pskb_expand_head(skb, 0, -delta,
1286			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1287			 __GFP_NOWARN | __GFP_NORETRY);
1288	return skb;
1289}
1290
1291static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1292				  struct sock *ssk)
1293{
1294	int ret;
1295	struct netlink_sock *nlk = nlk_sk(sk);
1296
1297	ret = -ECONNREFUSED;
1298	if (nlk->netlink_rcv != NULL) {
1299		ret = skb->len;
1300		netlink_skb_set_owner_r(skb, sk);
1301		NETLINK_CB(skb).sk = ssk;
1302		netlink_deliver_tap_kernel(sk, ssk, skb);
1303		nlk->netlink_rcv(skb);
1304		consume_skb(skb);
1305	} else {
1306		kfree_skb(skb);
1307	}
1308	sock_put(sk);
1309	return ret;
1310}
1311
1312int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1313		    u32 portid, int nonblock)
1314{
1315	struct sock *sk;
1316	int err;
1317	long timeo;
1318
1319	skb = netlink_trim(skb, gfp_any());
1320
1321	timeo = sock_sndtimeo(ssk, nonblock);
1322retry:
1323	sk = netlink_getsockbyportid(ssk, portid);
1324	if (IS_ERR(sk)) {
1325		kfree_skb(skb);
1326		return PTR_ERR(sk);
1327	}
1328	if (netlink_is_kernel(sk))
1329		return netlink_unicast_kernel(sk, skb, ssk);
1330
1331	if (sk_filter(sk, skb)) {
1332		err = skb->len;
1333		kfree_skb(skb);
1334		sock_put(sk);
1335		return err;
1336	}
1337
1338	err = netlink_attachskb(sk, skb, &timeo, ssk);
1339	if (err == 1)
1340		goto retry;
1341	if (err)
1342		return err;
1343
1344	return netlink_sendskb(sk, skb);
1345}
1346EXPORT_SYMBOL(netlink_unicast);
1347
1348int netlink_has_listeners(struct sock *sk, unsigned int group)
1349{
1350	int res = 0;
1351	struct listeners *listeners;
1352
1353	BUG_ON(!netlink_is_kernel(sk));
1354
1355	rcu_read_lock();
1356	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1357
1358	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1359		res = test_bit(group - 1, listeners->masks);
1360
1361	rcu_read_unlock();
1362
1363	return res;
1364}
1365EXPORT_SYMBOL_GPL(netlink_has_listeners);
1366
1367bool netlink_strict_get_check(struct sk_buff *skb)
1368{
1369	const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1370
1371	return nlk->flags & NETLINK_F_STRICT_CHK;
1372}
1373EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1374
1375static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1376{
1377	struct netlink_sock *nlk = nlk_sk(sk);
1378
1379	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1380	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1381		netlink_skb_set_owner_r(skb, sk);
1382		__netlink_sendskb(sk, skb);
1383		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1384	}
1385	return -1;
1386}
1387
1388struct netlink_broadcast_data {
1389	struct sock *exclude_sk;
1390	struct net *net;
1391	u32 portid;
1392	u32 group;
1393	int failure;
1394	int delivery_failure;
1395	int congested;
1396	int delivered;
1397	gfp_t allocation;
1398	struct sk_buff *skb, *skb2;
1399	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1400	void *tx_data;
1401};
1402
1403static void do_one_broadcast(struct sock *sk,
1404				    struct netlink_broadcast_data *p)
1405{
1406	struct netlink_sock *nlk = nlk_sk(sk);
1407	int val;
1408
1409	if (p->exclude_sk == sk)
1410		return;
1411
1412	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1413	    !test_bit(p->group - 1, nlk->groups))
1414		return;
1415
1416	if (!net_eq(sock_net(sk), p->net)) {
1417		if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1418			return;
1419
1420		if (!peernet_has_id(sock_net(sk), p->net))
1421			return;
1422
1423		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1424				     CAP_NET_BROADCAST))
1425			return;
1426	}
1427
1428	if (p->failure) {
1429		netlink_overrun(sk);
1430		return;
1431	}
1432
1433	sock_hold(sk);
1434	if (p->skb2 == NULL) {
1435		if (skb_shared(p->skb)) {
1436			p->skb2 = skb_clone(p->skb, p->allocation);
1437		} else {
1438			p->skb2 = skb_get(p->skb);
1439			/*
1440			 * skb ownership may have been set when
1441			 * delivered to a previous socket.
1442			 */
1443			skb_orphan(p->skb2);
1444		}
1445	}
1446	if (p->skb2 == NULL) {
1447		netlink_overrun(sk);
1448		/* Clone failed. Notify ALL listeners. */
1449		p->failure = 1;
1450		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1451			p->delivery_failure = 1;
1452		goto out;
1453	}
1454	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1455		kfree_skb(p->skb2);
1456		p->skb2 = NULL;
1457		goto out;
1458	}
1459	if (sk_filter(sk, p->skb2)) {
1460		kfree_skb(p->skb2);
1461		p->skb2 = NULL;
1462		goto out;
1463	}
1464	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1465	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1466		NETLINK_CB(p->skb2).nsid_is_set = true;
1467	val = netlink_broadcast_deliver(sk, p->skb2);
1468	if (val < 0) {
1469		netlink_overrun(sk);
1470		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1471			p->delivery_failure = 1;
1472	} else {
1473		p->congested |= val;
1474		p->delivered = 1;
1475		p->skb2 = NULL;
1476	}
1477out:
1478	sock_put(sk);
1479}
1480
1481int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1482	u32 group, gfp_t allocation,
1483	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1484	void *filter_data)
1485{
1486	struct net *net = sock_net(ssk);
1487	struct netlink_broadcast_data info;
1488	struct sock *sk;
1489
1490	skb = netlink_trim(skb, allocation);
1491
1492	info.exclude_sk = ssk;
1493	info.net = net;
1494	info.portid = portid;
1495	info.group = group;
1496	info.failure = 0;
1497	info.delivery_failure = 0;
1498	info.congested = 0;
1499	info.delivered = 0;
1500	info.allocation = allocation;
1501	info.skb = skb;
1502	info.skb2 = NULL;
1503	info.tx_filter = filter;
1504	info.tx_data = filter_data;
1505
1506	/* While we sleep in clone, do not allow to change socket list */
1507
1508	netlink_lock_table();
1509
1510	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1511		do_one_broadcast(sk, &info);
1512
1513	consume_skb(skb);
1514
1515	netlink_unlock_table();
1516
1517	if (info.delivery_failure) {
1518		kfree_skb(info.skb2);
1519		return -ENOBUFS;
1520	}
1521	consume_skb(info.skb2);
1522
1523	if (info.delivered) {
1524		if (info.congested && gfpflags_allow_blocking(allocation))
1525			yield();
1526		return 0;
1527	}
1528	return -ESRCH;
1529}
1530EXPORT_SYMBOL(netlink_broadcast_filtered);
1531
1532int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1533		      u32 group, gfp_t allocation)
1534{
1535	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1536		NULL, NULL);
1537}
1538EXPORT_SYMBOL(netlink_broadcast);
1539
1540struct netlink_set_err_data {
1541	struct sock *exclude_sk;
1542	u32 portid;
1543	u32 group;
1544	int code;
1545};
1546
1547static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1548{
1549	struct netlink_sock *nlk = nlk_sk(sk);
1550	int ret = 0;
1551
1552	if (sk == p->exclude_sk)
1553		goto out;
1554
1555	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1556		goto out;
1557
1558	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1559	    !test_bit(p->group - 1, nlk->groups))
1560		goto out;
1561
1562	if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1563		ret = 1;
1564		goto out;
1565	}
1566
1567	sk->sk_err = p->code;
1568	sk->sk_error_report(sk);
1569out:
1570	return ret;
1571}
1572
1573/**
1574 * netlink_set_err - report error to broadcast listeners
1575 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1576 * @portid: the PORTID of a process that we want to skip (if any)
1577 * @group: the broadcast group that will notice the error
1578 * @code: error code, must be negative (as usual in kernelspace)
1579 *
1580 * This function returns the number of broadcast listeners that have set the
1581 * NETLINK_NO_ENOBUFS socket option.
1582 */
1583int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1584{
1585	struct netlink_set_err_data info;
1586	struct sock *sk;
1587	int ret = 0;
1588
1589	info.exclude_sk = ssk;
1590	info.portid = portid;
1591	info.group = group;
1592	/* sk->sk_err wants a positive error value */
1593	info.code = -code;
1594
1595	read_lock(&nl_table_lock);
1596
1597	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1598		ret += do_one_set_err(sk, &info);
1599
1600	read_unlock(&nl_table_lock);
1601	return ret;
1602}
1603EXPORT_SYMBOL(netlink_set_err);
1604
1605/* must be called with netlink table grabbed */
1606static void netlink_update_socket_mc(struct netlink_sock *nlk,
1607				     unsigned int group,
1608				     int is_new)
1609{
1610	int old, new = !!is_new, subscriptions;
1611
1612	old = test_bit(group - 1, nlk->groups);
1613	subscriptions = nlk->subscriptions - old + new;
1614	if (new)
1615		__set_bit(group - 1, nlk->groups);
1616	else
1617		__clear_bit(group - 1, nlk->groups);
1618	netlink_update_subscriptions(&nlk->sk, subscriptions);
1619	netlink_update_listeners(&nlk->sk);
1620}
1621
1622static int netlink_setsockopt(struct socket *sock, int level, int optname,
1623			      char __user *optval, unsigned int optlen)
1624{
1625	struct sock *sk = sock->sk;
1626	struct netlink_sock *nlk = nlk_sk(sk);
1627	unsigned int val = 0;
1628	int err;
1629
1630	if (level != SOL_NETLINK)
1631		return -ENOPROTOOPT;
1632
1633	if (optlen >= sizeof(int) &&
1634	    get_user(val, (unsigned int __user *)optval))
1635		return -EFAULT;
1636
1637	switch (optname) {
1638	case NETLINK_PKTINFO:
1639		if (val)
1640			nlk->flags |= NETLINK_F_RECV_PKTINFO;
1641		else
1642			nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1643		err = 0;
1644		break;
1645	case NETLINK_ADD_MEMBERSHIP:
1646	case NETLINK_DROP_MEMBERSHIP: {
1647		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1648			return -EPERM;
1649		err = netlink_realloc_groups(sk);
1650		if (err)
1651			return err;
1652		if (!val || val - 1 >= nlk->ngroups)
1653			return -EINVAL;
1654		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1655			err = nlk->netlink_bind(sock_net(sk), val);
1656			if (err)
1657				return err;
1658		}
1659		netlink_table_grab();
1660		netlink_update_socket_mc(nlk, val,
1661					 optname == NETLINK_ADD_MEMBERSHIP);
1662		netlink_table_ungrab();
1663		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1664			nlk->netlink_unbind(sock_net(sk), val);
1665
1666		err = 0;
1667		break;
1668	}
1669	case NETLINK_BROADCAST_ERROR:
1670		if (val)
1671			nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1672		else
1673			nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1674		err = 0;
1675		break;
1676	case NETLINK_NO_ENOBUFS:
1677		if (val) {
1678			nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1679			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1680			wake_up_interruptible(&nlk->wait);
1681		} else {
1682			nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1683		}
1684		err = 0;
1685		break;
1686	case NETLINK_LISTEN_ALL_NSID:
1687		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1688			return -EPERM;
1689
1690		if (val)
1691			nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1692		else
1693			nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1694		err = 0;
1695		break;
1696	case NETLINK_CAP_ACK:
1697		if (val)
1698			nlk->flags |= NETLINK_F_CAP_ACK;
1699		else
1700			nlk->flags &= ~NETLINK_F_CAP_ACK;
1701		err = 0;
1702		break;
1703	case NETLINK_EXT_ACK:
1704		if (val)
1705			nlk->flags |= NETLINK_F_EXT_ACK;
1706		else
1707			nlk->flags &= ~NETLINK_F_EXT_ACK;
1708		err = 0;
1709		break;
1710	case NETLINK_GET_STRICT_CHK:
1711		if (val)
1712			nlk->flags |= NETLINK_F_STRICT_CHK;
1713		else
1714			nlk->flags &= ~NETLINK_F_STRICT_CHK;
1715		err = 0;
1716		break;
1717	default:
1718		err = -ENOPROTOOPT;
1719	}
1720	return err;
1721}
1722
1723static int netlink_getsockopt(struct socket *sock, int level, int optname,
1724			      char __user *optval, int __user *optlen)
1725{
1726	struct sock *sk = sock->sk;
1727	struct netlink_sock *nlk = nlk_sk(sk);
1728	int len, val, err;
1729
1730	if (level != SOL_NETLINK)
1731		return -ENOPROTOOPT;
1732
1733	if (get_user(len, optlen))
1734		return -EFAULT;
1735	if (len < 0)
1736		return -EINVAL;
1737
1738	switch (optname) {
1739	case NETLINK_PKTINFO:
1740		if (len < sizeof(int))
1741			return -EINVAL;
1742		len = sizeof(int);
1743		val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1744		if (put_user(len, optlen) ||
1745		    put_user(val, optval))
1746			return -EFAULT;
1747		err = 0;
1748		break;
1749	case NETLINK_BROADCAST_ERROR:
1750		if (len < sizeof(int))
1751			return -EINVAL;
1752		len = sizeof(int);
1753		val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1754		if (put_user(len, optlen) ||
1755		    put_user(val, optval))
1756			return -EFAULT;
1757		err = 0;
1758		break;
1759	case NETLINK_NO_ENOBUFS:
1760		if (len < sizeof(int))
1761			return -EINVAL;
1762		len = sizeof(int);
1763		val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1764		if (put_user(len, optlen) ||
1765		    put_user(val, optval))
1766			return -EFAULT;
1767		err = 0;
1768		break;
1769	case NETLINK_LIST_MEMBERSHIPS: {
1770		int pos, idx, shift;
1771
1772		err = 0;
1773		netlink_lock_table();
1774		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1775			if (len - pos < sizeof(u32))
1776				break;
1777
1778			idx = pos / sizeof(unsigned long);
1779			shift = (pos % sizeof(unsigned long)) * 8;
1780			if (put_user((u32)(nlk->groups[idx] >> shift),
1781				     (u32 __user *)(optval + pos))) {
1782				err = -EFAULT;
1783				break;
1784			}
1785		}
1786		if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1787			err = -EFAULT;
1788		netlink_unlock_table();
1789		break;
1790	}
1791	case NETLINK_CAP_ACK:
1792		if (len < sizeof(int))
1793			return -EINVAL;
1794		len = sizeof(int);
1795		val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1796		if (put_user(len, optlen) ||
1797		    put_user(val, optval))
1798			return -EFAULT;
1799		err = 0;
1800		break;
1801	case NETLINK_EXT_ACK:
1802		if (len < sizeof(int))
1803			return -EINVAL;
1804		len = sizeof(int);
1805		val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1806		if (put_user(len, optlen) || put_user(val, optval))
1807			return -EFAULT;
1808		err = 0;
1809		break;
1810	case NETLINK_GET_STRICT_CHK:
1811		if (len < sizeof(int))
1812			return -EINVAL;
1813		len = sizeof(int);
1814		val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1815		if (put_user(len, optlen) || put_user(val, optval))
1816			return -EFAULT;
1817		err = 0;
1818		break;
1819	default:
1820		err = -ENOPROTOOPT;
1821	}
1822	return err;
1823}
1824
1825static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1826{
1827	struct nl_pktinfo info;
1828
1829	info.group = NETLINK_CB(skb).dst_group;
1830	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1831}
1832
1833static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1834					 struct sk_buff *skb)
1835{
1836	if (!NETLINK_CB(skb).nsid_is_set)
1837		return;
1838
1839	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1840		 &NETLINK_CB(skb).nsid);
1841}
1842
1843static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1844{
1845	struct sock *sk = sock->sk;
1846	struct netlink_sock *nlk = nlk_sk(sk);
1847	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1848	u32 dst_portid;
1849	u32 dst_group;
1850	struct sk_buff *skb;
1851	int err;
1852	struct scm_cookie scm;
1853	u32 netlink_skb_flags = 0;
1854
1855	if (msg->msg_flags&MSG_OOB)
1856		return -EOPNOTSUPP;
1857
1858	err = scm_send(sock, msg, &scm, true);
1859	if (err < 0)
1860		return err;
1861
1862	if (msg->msg_namelen) {
1863		err = -EINVAL;
1864		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1865			goto out;
1866		if (addr->nl_family != AF_NETLINK)
1867			goto out;
1868		dst_portid = addr->nl_pid;
1869		dst_group = ffs(addr->nl_groups);
1870		err =  -EPERM;
1871		if ((dst_group || dst_portid) &&
1872		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1873			goto out;
1874		netlink_skb_flags |= NETLINK_SKB_DST;
1875	} else {
1876		dst_portid = nlk->dst_portid;
1877		dst_group = nlk->dst_group;
1878	}
1879
1880	if (!nlk->bound) {
1881		err = netlink_autobind(sock);
1882		if (err)
1883			goto out;
1884	} else {
1885		/* Ensure nlk is hashed and visible. */
1886		smp_rmb();
1887	}
1888
1889	err = -EMSGSIZE;
1890	if (len > sk->sk_sndbuf - 32)
1891		goto out;
1892	err = -ENOBUFS;
1893	skb = netlink_alloc_large_skb(len, dst_group);
1894	if (skb == NULL)
1895		goto out;
1896
1897	NETLINK_CB(skb).portid	= nlk->portid;
1898	NETLINK_CB(skb).dst_group = dst_group;
1899	NETLINK_CB(skb).creds	= scm.creds;
1900	NETLINK_CB(skb).flags	= netlink_skb_flags;
1901
1902	err = -EFAULT;
1903	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1904		kfree_skb(skb);
1905		goto out;
1906	}
1907
1908	err = security_netlink_send(sk, skb);
1909	if (err) {
1910		kfree_skb(skb);
1911		goto out;
1912	}
1913
1914	if (dst_group) {
1915		refcount_inc(&skb->users);
1916		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1917	}
1918	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1919
1920out:
1921	scm_destroy(&scm);
1922	return err;
1923}
1924
1925static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1926			   int flags)
1927{
1928	struct scm_cookie scm;
1929	struct sock *sk = sock->sk;
1930	struct netlink_sock *nlk = nlk_sk(sk);
1931	int noblock = flags&MSG_DONTWAIT;
1932	size_t copied;
1933	struct sk_buff *skb, *data_skb;
1934	int err, ret;
1935
1936	if (flags&MSG_OOB)
1937		return -EOPNOTSUPP;
1938
1939	copied = 0;
1940
1941	skb = skb_recv_datagram(sk, flags, noblock, &err);
1942	if (skb == NULL)
1943		goto out;
1944
1945	data_skb = skb;
1946
1947#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1948	if (unlikely(skb_shinfo(skb)->frag_list)) {
1949		/*
1950		 * If this skb has a frag_list, then here that means that we
1951		 * will have to use the frag_list skb's data for compat tasks
1952		 * and the regular skb's data for normal (non-compat) tasks.
1953		 *
1954		 * If we need to send the compat skb, assign it to the
1955		 * 'data_skb' variable so that it will be used below for data
1956		 * copying. We keep 'skb' for everything else, including
1957		 * freeing both later.
1958		 */
1959		if (flags & MSG_CMSG_COMPAT)
1960			data_skb = skb_shinfo(skb)->frag_list;
1961	}
1962#endif
1963
1964	/* Record the max length of recvmsg() calls for future allocations */
1965	nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1966	nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1967				     SKB_WITH_OVERHEAD(32768));
1968
1969	copied = data_skb->len;
1970	if (len < copied) {
1971		msg->msg_flags |= MSG_TRUNC;
1972		copied = len;
1973	}
1974
1975	skb_reset_transport_header(data_skb);
1976	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1977
1978	if (msg->msg_name) {
1979		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1980		addr->nl_family = AF_NETLINK;
1981		addr->nl_pad    = 0;
1982		addr->nl_pid	= NETLINK_CB(skb).portid;
1983		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1984		msg->msg_namelen = sizeof(*addr);
1985	}
1986
1987	if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1988		netlink_cmsg_recv_pktinfo(msg, skb);
1989	if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1990		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1991
1992	memset(&scm, 0, sizeof(scm));
1993	scm.creds = *NETLINK_CREDS(skb);
1994	if (flags & MSG_TRUNC)
1995		copied = data_skb->len;
1996
1997	skb_free_datagram(sk, skb);
1998
1999	if (nlk->cb_running &&
2000	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2001		ret = netlink_dump(sk);
2002		if (ret) {
2003			sk->sk_err = -ret;
2004			sk->sk_error_report(sk);
2005		}
2006	}
2007
2008	scm_recv(sock, msg, &scm, flags);
2009out:
2010	netlink_rcv_wake(sk);
2011	return err ? : copied;
2012}
2013
2014static void netlink_data_ready(struct sock *sk)
2015{
2016	BUG();
2017}
2018
2019/*
2020 *	We export these functions to other modules. They provide a
2021 *	complete set of kernel non-blocking support for message
2022 *	queueing.
2023 */
2024
2025struct sock *
2026__netlink_kernel_create(struct net *net, int unit, struct module *module,
2027			struct netlink_kernel_cfg *cfg)
2028{
2029	struct socket *sock;
2030	struct sock *sk;
2031	struct netlink_sock *nlk;
2032	struct listeners *listeners = NULL;
2033	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2034	unsigned int groups;
2035
2036	BUG_ON(!nl_table);
2037
2038	if (unit < 0 || unit >= MAX_LINKS)
2039		return NULL;
2040
2041	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2042		return NULL;
2043
2044	if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2045		goto out_sock_release_nosk;
2046
2047	sk = sock->sk;
2048
2049	if (!cfg || cfg->groups < 32)
2050		groups = 32;
2051	else
2052		groups = cfg->groups;
2053
2054	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2055	if (!listeners)
2056		goto out_sock_release;
2057
2058	sk->sk_data_ready = netlink_data_ready;
2059	if (cfg && cfg->input)
2060		nlk_sk(sk)->netlink_rcv = cfg->input;
2061
2062	if (netlink_insert(sk, 0))
2063		goto out_sock_release;
2064
2065	nlk = nlk_sk(sk);
2066	nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2067
2068	netlink_table_grab();
2069	if (!nl_table[unit].registered) {
2070		nl_table[unit].groups = groups;
2071		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2072		nl_table[unit].cb_mutex = cb_mutex;
2073		nl_table[unit].module = module;
2074		if (cfg) {
2075			nl_table[unit].bind = cfg->bind;
2076			nl_table[unit].unbind = cfg->unbind;
2077			nl_table[unit].flags = cfg->flags;
2078			if (cfg->compare)
2079				nl_table[unit].compare = cfg->compare;
2080		}
2081		nl_table[unit].registered = 1;
2082	} else {
2083		kfree(listeners);
2084		nl_table[unit].registered++;
2085	}
2086	netlink_table_ungrab();
2087	return sk;
2088
2089out_sock_release:
2090	kfree(listeners);
2091	netlink_kernel_release(sk);
2092	return NULL;
2093
2094out_sock_release_nosk:
2095	sock_release(…

Large files files are truncated, but you can click here to view the full file