1/*
   2 * NETLINK      Kernel-user communication protocol.
   3 *
   4 * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
   5 * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
   6 * 				Patrick McHardy <kaber@trash.net>
   7 *
   8 *		This program is free software; you can redistribute it and/or
   9 *		modify it under the terms of the GNU General Public License
  10 *		as published by the Free Software Foundation; either version
  11 *		2 of the License, or (at your option) any later version.
  12 *
  13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
  14 *                               added netlink_proto_exit
  15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
  16 * 				 use nlk_sk, as sk->protinfo is on a diet 8)
  17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
  18 * 				 - inc module use count of module that owns
  19 * 				   the kernel socket in case userspace opens
  20 * 				   socket of same protocol
  21 * 				 - remove all module support, since netlink is
  22 * 				   mandatory if CONFIG_NET=y these days
  23 */
  24
  25#include <linux/module.h>
  26
  27#include <linux/capability.h>
  28#include <linux/kernel.h>
  29#include <linux/init.h>
  30#include <linux/signal.h>
  31#include <linux/sched.h>
  32#include <linux/errno.h>
  33#include <linux/string.h>
  34#include <linux/stat.h>
  35#include <linux/socket.h>
  36#include <linux/un.h>
  37#include <linux/fcntl.h>
  38#include <linux/termios.h>
  39#include <linux/sockios.h>
  40#include <linux/net.h>
  41#include <linux/fs.h>
  42#include <linux/slab.h>
  43#include <linux/uaccess.h>
  44#include <linux/skbuff.h>
  45#include <linux/netdevice.h>
  46#include <linux/rtnetlink.h>
  47#include <linux/proc_fs.h>
  48#include <linux/seq_file.h>
  49#include <linux/notifier.h>
  50#include <linux/security.h>
  51#include <linux/jhash.h>
  52#include <linux/jiffies.h>
  53#include <linux/random.h>
  54#include <linux/bitops.h>
  55#include <linux/mm.h>
  56#include <linux/types.h>
  57#include <linux/audit.h>
  58#include <linux/mutex.h>
  59#include <linux/vmalloc.h>
  60#include <linux/if_arp.h>
  61#include <linux/rhashtable.h>
  62#include <asm/cacheflush.h>
  63#include <linux/hash.h>
  64#include <linux/genetlink.h>
  65
  66#include <net/net_namespace.h>
  67#include <net/sock.h>
  68#include <net/scm.h>
  69#include <net/netlink.h>
  70
  71#include "af_netlink.h"
  72
  73struct listeners {
  74	struct rcu_head		rcu;
  75	unsigned long		masks[0];
  76};
  77
  78/* state bits */
  79#define NETLINK_S_CONGESTED		0x0
  80
  81/* flags */
  82#define NETLINK_F_KERNEL_SOCKET		0x1
  83#define NETLINK_F_RECV_PKTINFO		0x2
  84#define NETLINK_F_BROADCAST_SEND_ERROR	0x4
  85#define NETLINK_F_RECV_NO_ENOBUFS	0x8
  86#define NETLINK_F_LISTEN_ALL_NSID	0x10
  87#define NETLINK_F_CAP_ACK		0x20
  88
  89static inline int netlink_is_kernel(struct sock *sk)
  90{
  91	return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
  92}
  93
  94struct netlink_table *nl_table __read_mostly;
  95EXPORT_SYMBOL_GPL(nl_table);
  96
  97static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
  98
  99static int netlink_dump(struct sock *sk);
 100static void netlink_skb_destructor(struct sk_buff *skb);
 101
 102/* nl_table locking explained:
 103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
 104 * and removal are protected with per bucket lock while using RCU list
 105 * modification primitives and may run in parallel to RCU protected lookups.
 106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
 107 * been acquired * either during or after the socket has been removed from
 108 * the list and after an RCU grace period.
 109 */
 110DEFINE_RWLOCK(nl_table_lock);
 111EXPORT_SYMBOL_GPL(nl_table_lock);
 112static atomic_t nl_table_users = ATOMIC_INIT(0);
 113
 114#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
 115
 116static BLOCKING_NOTIFIER_HEAD(netlink_chain);
 117
 118static DEFINE_SPINLOCK(netlink_tap_lock);
 119static struct list_head netlink_tap_all __read_mostly;
 120
 121static const struct rhashtable_params netlink_rhashtable_params;
 122
 123static inline u32 netlink_group_mask(u32 group)
 124{
 125	return group ? 1 << (group - 1) : 0;
 126}
 127
 128static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
 129					   gfp_t gfp_mask)
 130{
 131	unsigned int len = skb_end_offset(skb);
 132	struct sk_buff *new;
 133
 134	new = alloc_skb(len, gfp_mask);
 135	if (new == NULL)
 136		return NULL;
 137
 138	NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
 139	NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
 140	NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
 141
 142	memcpy(skb_put(new, len), skb->data, len);
 143	return new;
 144}
 145
 146int netlink_add_tap(struct netlink_tap *nt)
 147{
 148	if (unlikely(nt->dev->type != ARPHRD_NETLINK))
 149		return -EINVAL;
 150
 151	spin_lock(&netlink_tap_lock);
 152	list_add_rcu(&nt->list, &netlink_tap_all);
 153	spin_unlock(&netlink_tap_lock);
 154
 155	__module_get(nt->module);
 156
 157	return 0;
 158}
 159EXPORT_SYMBOL_GPL(netlink_add_tap);
 160
 161static int __netlink_remove_tap(struct netlink_tap *nt)
 162{
 163	bool found = false;
 164	struct netlink_tap *tmp;
 165
 166	spin_lock(&netlink_tap_lock);
 167
 168	list_for_each_entry(tmp, &netlink_tap_all, list) {
 169		if (nt == tmp) {
 170			list_del_rcu(&nt->list);
 171			found = true;
 172			goto out;
 173		}
 174	}
 175
 176	pr_warn("__netlink_remove_tap: %p not found\n", nt);
 177out:
 178	spin_unlock(&netlink_tap_lock);
 179
 180	if (found)
 181		module_put(nt->module);
 182
 183	return found ? 0 : -ENODEV;
 184}
 185
 186int netlink_remove_tap(struct netlink_tap *nt)
 187{
 188	int ret;
 189
 190	ret = __netlink_remove_tap(nt);
 191	synchronize_net();
 192
 193	return ret;
 194}
 195EXPORT_SYMBOL_GPL(netlink_remove_tap);
 196
 197static bool netlink_filter_tap(const struct sk_buff *skb)
 198{
 199	struct sock *sk = skb->sk;
 200
 201	/* We take the more conservative approach and
 202	 * whitelist socket protocols that may pass.
 203	 */
 204	switch (sk->sk_protocol) {
 205	case NETLINK_ROUTE:
 206	case NETLINK_USERSOCK:
 207	case NETLINK_SOCK_DIAG:
 208	case NETLINK_NFLOG:
 209	case NETLINK_XFRM:
 210	case NETLINK_FIB_LOOKUP:
 211	case NETLINK_NETFILTER:
 212	case NETLINK_GENERIC:
 213		return true;
 214	}
 215
 216	return false;
 217}
 218
 219static int __netlink_deliver_tap_skb(struct sk_buff *skb,
 220				     struct net_device *dev)
 221{
 222	struct sk_buff *nskb;
 223	struct sock *sk = skb->sk;
 224	int ret = -ENOMEM;
 225
 226	dev_hold(dev);
 227
 228	if (is_vmalloc_addr(skb->head))
 229		nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
 230	else
 231		nskb = skb_clone(skb, GFP_ATOMIC);
 232	if (nskb) {
 233		nskb->dev = dev;
 234		nskb->protocol = htons((u16) sk->sk_protocol);
 235		nskb->pkt_type = netlink_is_kernel(sk) ?
 236				 PACKET_KERNEL : PACKET_USER;
 237		skb_reset_network_header(nskb);
 238		ret = dev_queue_xmit(nskb);
 239		if (unlikely(ret > 0))
 240			ret = net_xmit_errno(ret);
 241	}
 242
 243	dev_put(dev);
 244	return ret;
 245}
 246
 247static void __netlink_deliver_tap(struct sk_buff *skb)
 248{
 249	int ret;
 250	struct netlink_tap *tmp;
 251
 252	if (!netlink_filter_tap(skb))
 253		return;
 254
 255	list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
 256		ret = __netlink_deliver_tap_skb(skb, tmp->dev);
 257		if (unlikely(ret))
 258			break;
 259	}
 260}
 261
 262static void netlink_deliver_tap(struct sk_buff *skb)
 263{
 264	rcu_read_lock();
 265
 266	if (unlikely(!list_empty(&netlink_tap_all)))
 267		__netlink_deliver_tap(skb);
 268
 269	rcu_read_unlock();
 270}
 271
 272static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
 273				       struct sk_buff *skb)
 274{
 275	if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
 276		netlink_deliver_tap(skb);
 277}
 278
 279static void netlink_overrun(struct sock *sk)
 280{
 281	struct netlink_sock *nlk = nlk_sk(sk);
 282
 283	if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
 284		if (!test_and_set_bit(NETLINK_S_CONGESTED,
 285				      &nlk_sk(sk)->state)) {
 286			sk->sk_err = ENOBUFS;
 287			sk->sk_error_report(sk);
 288		}
 289	}
 290	atomic_inc(&sk->sk_drops);
 291}
 292
 293static void netlink_rcv_wake(struct sock *sk)
 294{
 295	struct netlink_sock *nlk = nlk_sk(sk);
 296
 297	if (skb_queue_empty(&sk->sk_receive_queue))
 298		clear_bit(NETLINK_S_CONGESTED, &nlk->state);
 299	if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
 300		wake_up_interruptible(&nlk->wait);
 301}
 302
 303static void netlink_skb_destructor(struct sk_buff *skb)
 304{
 305	if (is_vmalloc_addr(skb->head)) {
 306		if (!skb->cloned ||
 307		    !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
 308			vfree(skb->head);
 309
 310		skb->head = NULL;
 311	}
 312	if (skb->sk != NULL)
 313		sock_rfree(skb);
 314}
 315
 316static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
 317{
 318	WARN_ON(skb->sk != NULL);
 319	skb->sk = sk;
 320	skb->destructor = netlink_skb_destructor;
 321	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
 322	sk_mem_charge(sk, skb->truesize);
 323}
 324
 325static void netlink_sock_destruct(struct sock *sk)
 326{
 327	struct netlink_sock *nlk = nlk_sk(sk);
 328
 329	if (nlk->cb_running) {
 330		if (nlk->cb.done)
 331			nlk->cb.done(&nlk->cb);
 
 332		module_put(nlk->cb.module);
 333		kfree_skb(nlk->cb.skb);
 334	}
 335
 336	skb_queue_purge(&sk->sk_receive_queue);
 337
 338	if (!sock_flag(sk, SOCK_DEAD)) {
 339		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
 340		return;
 341	}
 342
 343	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 344	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 345	WARN_ON(nlk_sk(sk)->groups);
 346}
 347
 348static void netlink_sock_destruct_work(struct work_struct *work)
 349{
 350	struct netlink_sock *nlk = container_of(work, struct netlink_sock,
 351						work);
 352
 353	sk_free(&nlk->sk);
 354}
 355
 356/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
 357 * SMP. Look, when several writers sleep and reader wakes them up, all but one
 358 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 359 * this, _but_ remember, it adds useless work on UP machines.
 360 */
 361
 362void netlink_table_grab(void)
 363	__acquires(nl_table_lock)
 364{
 365	might_sleep();
 366
 367	write_lock_irq(&nl_table_lock);
 368
 369	if (atomic_read(&nl_table_users)) {
 370		DECLARE_WAITQUEUE(wait, current);
 371
 372		add_wait_queue_exclusive(&nl_table_wait, &wait);
 373		for (;;) {
 374			set_current_state(TASK_UNINTERRUPTIBLE);
 375			if (atomic_read(&nl_table_users) == 0)
 376				break;
 377			write_unlock_irq(&nl_table_lock);
 378			schedule();
 379			write_lock_irq(&nl_table_lock);
 380		}
 381
 382		__set_current_state(TASK_RUNNING);
 383		remove_wait_queue(&nl_table_wait, &wait);
 384	}
 385}
 386
 387void netlink_table_ungrab(void)
 388	__releases(nl_table_lock)
 389{
 390	write_unlock_irq(&nl_table_lock);
 391	wake_up(&nl_table_wait);
 392}
 393
 394static inline void
 395netlink_lock_table(void)
 396{
 397	/* read_lock() synchronizes us to netlink_table_grab */
 398
 399	read_lock(&nl_table_lock);
 400	atomic_inc(&nl_table_users);
 401	read_unlock(&nl_table_lock);
 402}
 403
 404static inline void
 405netlink_unlock_table(void)
 406{
 407	if (atomic_dec_and_test(&nl_table_users))
 408		wake_up(&nl_table_wait);
 409}
 410
 411struct netlink_compare_arg
 412{
 413	possible_net_t pnet;
 414	u32 portid;
 415};
 416
 417/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
 418#define netlink_compare_arg_len \
 419	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
 420
 421static inline int netlink_compare(struct rhashtable_compare_arg *arg,
 422				  const void *ptr)
 423{
 424	const struct netlink_compare_arg *x = arg->key;
 425	const struct netlink_sock *nlk = ptr;
 426
 427	return nlk->portid != x->portid ||
 428	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
 429}
 430
 431static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
 432				     struct net *net, u32 portid)
 433{
 434	memset(arg, 0, sizeof(*arg));
 435	write_pnet(&arg->pnet, net);
 436	arg->portid = portid;
 437}
 438
 439static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
 440				     struct net *net)
 441{
 442	struct netlink_compare_arg arg;
 443
 444	netlink_compare_arg_init(&arg, net, portid);
 445	return rhashtable_lookup_fast(&table->hash, &arg,
 446				      netlink_rhashtable_params);
 447}
 448
 449static int __netlink_insert(struct netlink_table *table, struct sock *sk)
 450{
 451	struct netlink_compare_arg arg;
 452
 453	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
 454	return rhashtable_lookup_insert_key(&table->hash, &arg,
 455					    &nlk_sk(sk)->node,
 456					    netlink_rhashtable_params);
 457}
 458
 459static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
 460{
 461	struct netlink_table *table = &nl_table[protocol];
 462	struct sock *sk;
 463
 464	rcu_read_lock();
 465	sk = __netlink_lookup(table, portid, net);
 466	if (sk)
 467		sock_hold(sk);
 468	rcu_read_unlock();
 469
 470	return sk;
 471}
 472
 473static const struct proto_ops netlink_ops;
 474
 475static void
 476netlink_update_listeners(struct sock *sk)
 477{
 478	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
 479	unsigned long mask;
 480	unsigned int i;
 481	struct listeners *listeners;
 482
 483	listeners = nl_deref_protected(tbl->listeners);
 484	if (!listeners)
 485		return;
 486
 487	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
 488		mask = 0;
 489		sk_for_each_bound(sk, &tbl->mc_list) {
 490			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
 491				mask |= nlk_sk(sk)->groups[i];
 492		}
 493		listeners->masks[i] = mask;
 494	}
 495	/* this function is only called with the netlink table "grabbed", which
 496	 * makes sure updates are visible before bind or setsockopt return. */
 497}
 498
 499static int netlink_insert(struct sock *sk, u32 portid)
 500{
 501	struct netlink_table *table = &nl_table[sk->sk_protocol];
 502	int err;
 503
 504	lock_sock(sk);
 505
 506	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
 507	if (nlk_sk(sk)->bound)
 508		goto err;
 509
 510	err = -ENOMEM;
 511	if (BITS_PER_LONG > 32 &&
 512	    unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
 513		goto err;
 514
 515	nlk_sk(sk)->portid = portid;
 516	sock_hold(sk);
 517
 518	err = __netlink_insert(table, sk);
 519	if (err) {
 520		/* In case the hashtable backend returns with -EBUSY
 521		 * from here, it must not escape to the caller.
 522		 */
 523		if (unlikely(err == -EBUSY))
 524			err = -EOVERFLOW;
 525		if (err == -EEXIST)
 526			err = -EADDRINUSE;
 527		sock_put(sk);
 528		goto err;
 529	}
 530
 531	/* We need to ensure that the socket is hashed and visible. */
 532	smp_wmb();
 533	nlk_sk(sk)->bound = portid;
 534
 535err:
 536	release_sock(sk);
 537	return err;
 538}
 539
 540static void netlink_remove(struct sock *sk)
 541{
 542	struct netlink_table *table;
 543
 544	table = &nl_table[sk->sk_protocol];
 545	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
 546				    netlink_rhashtable_params)) {
 547		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
 548		__sock_put(sk);
 549	}
 550
 551	netlink_table_grab();
 552	if (nlk_sk(sk)->subscriptions) {
 553		__sk_del_bind_node(sk);
 554		netlink_update_listeners(sk);
 555	}
 556	if (sk->sk_protocol == NETLINK_GENERIC)
 557		atomic_inc(&genl_sk_destructing_cnt);
 558	netlink_table_ungrab();
 559}
 560
 561static struct proto netlink_proto = {
 562	.name	  = "NETLINK",
 563	.owner	  = THIS_MODULE,
 564	.obj_size = sizeof(struct netlink_sock),
 565};
 566
 567static int __netlink_create(struct net *net, struct socket *sock,
 568			    struct mutex *cb_mutex, int protocol,
 569			    int kern)
 570{
 571	struct sock *sk;
 572	struct netlink_sock *nlk;
 573
 574	sock->ops = &netlink_ops;
 575
 576	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
 577	if (!sk)
 578		return -ENOMEM;
 579
 580	sock_init_data(sock, sk);
 581
 582	nlk = nlk_sk(sk);
 583	if (cb_mutex) {
 584		nlk->cb_mutex = cb_mutex;
 585	} else {
 586		nlk->cb_mutex = &nlk->cb_def_mutex;
 587		mutex_init(nlk->cb_mutex);
 588	}
 589	init_waitqueue_head(&nlk->wait);
 590
 591	sk->sk_destruct = netlink_sock_destruct;
 592	sk->sk_protocol = protocol;
 593	return 0;
 594}
 595
 596static int netlink_create(struct net *net, struct socket *sock, int protocol,
 597			  int kern)
 598{
 599	struct module *module = NULL;
 600	struct mutex *cb_mutex;
 601	struct netlink_sock *nlk;
 602	int (*bind)(struct net *net, int group);
 603	void (*unbind)(struct net *net, int group);
 604	int err = 0;
 605
 606	sock->state = SS_UNCONNECTED;
 607
 608	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 609		return -ESOCKTNOSUPPORT;
 610
 611	if (protocol < 0 || protocol >= MAX_LINKS)
 612		return -EPROTONOSUPPORT;
 613
 614	netlink_lock_table();
 615#ifdef CONFIG_MODULES
 616	if (!nl_table[protocol].registered) {
 617		netlink_unlock_table();
 618		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
 619		netlink_lock_table();
 620	}
 621#endif
 622	if (nl_table[protocol].registered &&
 623	    try_module_get(nl_table[protocol].module))
 624		module = nl_table[protocol].module;
 625	else
 626		err = -EPROTONOSUPPORT;
 627	cb_mutex = nl_table[protocol].cb_mutex;
 628	bind = nl_table[protocol].bind;
 629	unbind = nl_table[protocol].unbind;
 630	netlink_unlock_table();
 631
 632	if (err < 0)
 633		goto out;
 634
 635	err = __netlink_create(net, sock, cb_mutex, protocol, kern);
 636	if (err < 0)
 637		goto out_module;
 638
 639	local_bh_disable();
 640	sock_prot_inuse_add(net, &netlink_proto, 1);
 641	local_bh_enable();
 642
 643	nlk = nlk_sk(sock->sk);
 644	nlk->module = module;
 645	nlk->netlink_bind = bind;
 646	nlk->netlink_unbind = unbind;
 647out:
 648	return err;
 649
 650out_module:
 651	module_put(module);
 652	goto out;
 653}
 654
 655static void deferred_put_nlk_sk(struct rcu_head *head)
 656{
 657	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
 658	struct sock *sk = &nlk->sk;
 659
 660	if (!atomic_dec_and_test(&sk->sk_refcnt))
 661		return;
 662
 663	if (nlk->cb_running && nlk->cb.done) {
 664		INIT_WORK(&nlk->work, netlink_sock_destruct_work);
 665		schedule_work(&nlk->work);
 666		return;
 667	}
 668
 669	sk_free(sk);
 670}
 671
 672static int netlink_release(struct socket *sock)
 673{
 674	struct sock *sk = sock->sk;
 675	struct netlink_sock *nlk;
 676
 677	if (!sk)
 678		return 0;
 679
 680	netlink_remove(sk);
 681	sock_orphan(sk);
 682	nlk = nlk_sk(sk);
 683
 684	/*
 685	 * OK. Socket is unlinked, any packets that arrive now
 686	 * will be purged.
 687	 */
 688
 689	/* must not acquire netlink_table_lock in any way again before unbind
 690	 * and notifying genetlink is done as otherwise it might deadlock
 691	 */
 692	if (nlk->netlink_unbind) {
 693		int i;
 694
 695		for (i = 0; i < nlk->ngroups; i++)
 696			if (test_bit(i, nlk->groups))
 697				nlk->netlink_unbind(sock_net(sk), i + 1);
 698	}
 699	if (sk->sk_protocol == NETLINK_GENERIC &&
 700	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
 701		wake_up(&genl_sk_destructing_waitq);
 702
 703	sock->sk = NULL;
 704	wake_up_interruptible_all(&nlk->wait);
 705
 706	skb_queue_purge(&sk->sk_write_queue);
 707
 708	if (nlk->portid && nlk->bound) {
 709		struct netlink_notify n = {
 710						.net = sock_net(sk),
 711						.protocol = sk->sk_protocol,
 712						.portid = nlk->portid,
 713					  };
 714		blocking_notifier_call_chain(&netlink_chain,
 715				NETLINK_URELEASE, &n);
 716	}
 717
 718	module_put(nlk->module);
 719
 720	if (netlink_is_kernel(sk)) {
 721		netlink_table_grab();
 722		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
 723		if (--nl_table[sk->sk_protocol].registered == 0) {
 724			struct listeners *old;
 725
 726			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
 727			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
 728			kfree_rcu(old, rcu);
 729			nl_table[sk->sk_protocol].module = NULL;
 730			nl_table[sk->sk_protocol].bind = NULL;
 731			nl_table[sk->sk_protocol].unbind = NULL;
 732			nl_table[sk->sk_protocol].flags = 0;
 733			nl_table[sk->sk_protocol].registered = 0;
 734		}
 735		netlink_table_ungrab();
 736	}
 737
 738	kfree(nlk->groups);
 739	nlk->groups = NULL;
 740
 741	local_bh_disable();
 742	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
 743	local_bh_enable();
 744	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
 745	return 0;
 746}
 747
 748static int netlink_autobind(struct socket *sock)
 749{
 750	struct sock *sk = sock->sk;
 751	struct net *net = sock_net(sk);
 752	struct netlink_table *table = &nl_table[sk->sk_protocol];
 753	s32 portid = task_tgid_vnr(current);
 754	int err;
 755	s32 rover = -4096;
 756	bool ok;
 757
 758retry:
 759	cond_resched();
 760	rcu_read_lock();
 761	ok = !__netlink_lookup(table, portid, net);
 762	rcu_read_unlock();
 763	if (!ok) {
 764		/* Bind collision, search negative portid values. */
 765		if (rover == -4096)
 766			/* rover will be in range [S32_MIN, -4097] */
 767			rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
 768		else if (rover >= -4096)
 769			rover = -4097;
 770		portid = rover--;
 771		goto retry;
 772	}
 773
 774	err = netlink_insert(sk, portid);
 775	if (err == -EADDRINUSE)
 776		goto retry;
 777
 778	/* If 2 threads race to autobind, that is fine.  */
 779	if (err == -EBUSY)
 780		err = 0;
 781
 782	return err;
 783}
 784
 785/**
 786 * __netlink_ns_capable - General netlink message capability test
 787 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
 788 * @user_ns: The user namespace of the capability to use
 789 * @cap: The capability to use
 790 *
 791 * Test to see if the opener of the socket we received the message
 792 * from had when the netlink socket was created and the sender of the
 793 * message has has the capability @cap in the user namespace @user_ns.
 794 */
 795bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
 796			struct user_namespace *user_ns, int cap)
 797{
 798	return ((nsp->flags & NETLINK_SKB_DST) ||
 799		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
 800		ns_capable(user_ns, cap);
 801}
 802EXPORT_SYMBOL(__netlink_ns_capable);
 803
 804/**
 805 * netlink_ns_capable - General netlink message capability test
 806 * @skb: socket buffer holding a netlink command from userspace
 807 * @user_ns: The user namespace of the capability to use
 808 * @cap: The capability to use
 809 *
 810 * Test to see if the opener of the socket we received the message
 811 * from had when the netlink socket was created and the sender of the
 812 * message has has the capability @cap in the user namespace @user_ns.
 813 */
 814bool netlink_ns_capable(const struct sk_buff *skb,
 815			struct user_namespace *user_ns, int cap)
 816{
 817	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
 818}
 819EXPORT_SYMBOL(netlink_ns_capable);
 820
 821/**
 822 * netlink_capable - Netlink global message capability test
 823 * @skb: socket buffer holding a netlink command from userspace
 824 * @cap: The capability to use
 825 *
 826 * Test to see if the opener of the socket we received the message
 827 * from had when the netlink socket was created and the sender of the
 828 * message has has the capability @cap in all user namespaces.
 829 */
 830bool netlink_capable(const struct sk_buff *skb, int cap)
 831{
 832	return netlink_ns_capable(skb, &init_user_ns, cap);
 833}
 834EXPORT_SYMBOL(netlink_capable);
 835
 836/**
 837 * netlink_net_capable - Netlink network namespace message capability test
 838 * @skb: socket buffer holding a netlink command from userspace
 839 * @cap: The capability to use
 840 *
 841 * Test to see if the opener of the socket we received the message
 842 * from had when the netlink socket was created and the sender of the
 843 * message has has the capability @cap over the network namespace of
 844 * the socket we received the message from.
 845 */
 846bool netlink_net_capable(const struct sk_buff *skb, int cap)
 847{
 848	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
 849}
 850EXPORT_SYMBOL(netlink_net_capable);
 851
 852static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
 853{
 854	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
 855		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
 856}
 857
 858static void
 859netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
 860{
 861	struct netlink_sock *nlk = nlk_sk(sk);
 862
 863	if (nlk->subscriptions && !subscriptions)
 864		__sk_del_bind_node(sk);
 865	else if (!nlk->subscriptions && subscriptions)
 866		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
 867	nlk->subscriptions = subscriptions;
 868}
 869
 870static int netlink_realloc_groups(struct sock *sk)
 871{
 872	struct netlink_sock *nlk = nlk_sk(sk);
 873	unsigned int groups;
 874	unsigned long *new_groups;
 875	int err = 0;
 876
 877	netlink_table_grab();
 878
 879	groups = nl_table[sk->sk_protocol].groups;
 880	if (!nl_table[sk->sk_protocol].registered) {
 881		err = -ENOENT;
 882		goto out_unlock;
 883	}
 884
 885	if (nlk->ngroups >= groups)
 886		goto out_unlock;
 887
 888	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
 889	if (new_groups == NULL) {
 890		err = -ENOMEM;
 891		goto out_unlock;
 892	}
 893	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
 894	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
 895
 896	nlk->groups = new_groups;
 897	nlk->ngroups = groups;
 898 out_unlock:
 899	netlink_table_ungrab();
 900	return err;
 901}
 902
 903static void netlink_undo_bind(int group, long unsigned int groups,
 904			      struct sock *sk)
 905{
 906	struct netlink_sock *nlk = nlk_sk(sk);
 907	int undo;
 908
 909	if (!nlk->netlink_unbind)
 910		return;
 911
 912	for (undo = 0; undo < group; undo++)
 913		if (test_bit(undo, &groups))
 914			nlk->netlink_unbind(sock_net(sk), undo + 1);
 915}
 916
 917static int netlink_bind(struct socket *sock, struct sockaddr *addr,
 918			int addr_len)
 919{
 920	struct sock *sk = sock->sk;
 921	struct net *net = sock_net(sk);
 922	struct netlink_sock *nlk = nlk_sk(sk);
 923	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
 924	int err;
 925	long unsigned int groups = nladdr->nl_groups;
 926	bool bound;
 927
 928	if (addr_len < sizeof(struct sockaddr_nl))
 929		return -EINVAL;
 930
 931	if (nladdr->nl_family != AF_NETLINK)
 932		return -EINVAL;
 933
 934	/* Only superuser is allowed to listen multicasts */
 935	if (groups) {
 936		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
 937			return -EPERM;
 938		err = netlink_realloc_groups(sk);
 939		if (err)
 940			return err;
 941	}
 942
 943	bound = nlk->bound;
 944	if (bound) {
 945		/* Ensure nlk->portid is up-to-date. */
 946		smp_rmb();
 947
 948		if (nladdr->nl_pid != nlk->portid)
 949			return -EINVAL;
 950	}
 951
 952	if (nlk->netlink_bind && groups) {
 953		int group;
 954
 955		for (group = 0; group < nlk->ngroups; group++) {
 956			if (!test_bit(group, &groups))
 957				continue;
 958			err = nlk->netlink_bind(net, group + 1);
 959			if (!err)
 960				continue;
 961			netlink_undo_bind(group, groups, sk);
 962			return err;
 963		}
 964	}
 965
 966	/* No need for barriers here as we return to user-space without
 967	 * using any of the bound attributes.
 968	 */
 969	if (!bound) {
 970		err = nladdr->nl_pid ?
 971			netlink_insert(sk, nladdr->nl_pid) :
 972			netlink_autobind(sock);
 973		if (err) {
 974			netlink_undo_bind(nlk->ngroups, groups, sk);
 975			return err;
 976		}
 977	}
 978
 979	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
 980		return 0;
 981
 982	netlink_table_grab();
 983	netlink_update_subscriptions(sk, nlk->subscriptions +
 984					 hweight32(groups) -
 985					 hweight32(nlk->groups[0]));
 986	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
 987	netlink_update_listeners(sk);
 988	netlink_table_ungrab();
 989
 990	return 0;
 991}
 992
 993static int netlink_connect(struct socket *sock, struct sockaddr *addr,
 994			   int alen, int flags)
 995{
 996	int err = 0;
 997	struct sock *sk = sock->sk;
 998	struct netlink_sock *nlk = nlk_sk(sk);
 999	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1000
1001	if (alen < sizeof(addr->sa_family))
1002		return -EINVAL;
1003
1004	if (addr->sa_family == AF_UNSPEC) {
1005		sk->sk_state	= NETLINK_UNCONNECTED;
1006		nlk->dst_portid	= 0;
1007		nlk->dst_group  = 0;
1008		return 0;
1009	}
1010	if (addr->sa_family != AF_NETLINK)
1011		return -EINVAL;
1012
1013	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1014	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1015		return -EPERM;
1016
1017	/* No need for barriers here as we return to user-space without
1018	 * using any of the bound attributes.
1019	 */
1020	if (!nlk->bound)
1021		err = netlink_autobind(sock);
1022
1023	if (err == 0) {
1024		sk->sk_state	= NETLINK_CONNECTED;
1025		nlk->dst_portid = nladdr->nl_pid;
1026		nlk->dst_group  = ffs(nladdr->nl_groups);
1027	}
1028
1029	return err;
1030}
1031
1032static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1033			   int *addr_len, int peer)
1034{
1035	struct sock *sk = sock->sk;
1036	struct netlink_sock *nlk = nlk_sk(sk);
1037	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1038
1039	nladdr->nl_family = AF_NETLINK;
1040	nladdr->nl_pad = 0;
1041	*addr_len = sizeof(*nladdr);
1042
1043	if (peer) {
1044		nladdr->nl_pid = nlk->dst_portid;
1045		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1046	} else {
1047		nladdr->nl_pid = nlk->portid;
1048		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1049	}
1050	return 0;
1051}
1052
1053static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1054			 unsigned long arg)
1055{
1056	/* try to hand this ioctl down to the NIC drivers.
1057	 */
1058	return -ENOIOCTLCMD;
1059}
1060
1061static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1062{
1063	struct sock *sock;
1064	struct netlink_sock *nlk;
1065
1066	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1067	if (!sock)
1068		return ERR_PTR(-ECONNREFUSED);
1069
1070	/* Don't bother queuing skb if kernel socket has no input function */
1071	nlk = nlk_sk(sock);
1072	if (sock->sk_state == NETLINK_CONNECTED &&
1073	    nlk->dst_portid != nlk_sk(ssk)->portid) {
1074		sock_put(sock);
1075		return ERR_PTR(-ECONNREFUSED);
1076	}
1077	return sock;
1078}
1079
1080struct sock *netlink_getsockbyfilp(struct file *filp)
1081{
1082	struct inode *inode = file_inode(filp);
1083	struct sock *sock;
1084
1085	if (!S_ISSOCK(inode->i_mode))
1086		return ERR_PTR(-ENOTSOCK);
1087
1088	sock = SOCKET_I(inode)->sk;
1089	if (sock->sk_family != AF_NETLINK)
1090		return ERR_PTR(-EINVAL);
1091
1092	sock_hold(sock);
1093	return sock;
1094}
1095
1096static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1097					       int broadcast)
1098{
1099	struct sk_buff *skb;
1100	void *data;
1101
1102	if (size <= NLMSG_GOODSIZE || broadcast)
1103		return alloc_skb(size, GFP_KERNEL);
1104
1105	size = SKB_DATA_ALIGN(size) +
1106	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1107
1108	data = vmalloc(size);
1109	if (data == NULL)
1110		return NULL;
1111
1112	skb = __build_skb(data, size);
1113	if (skb == NULL)
1114		vfree(data);
1115	else
1116		skb->destructor = netlink_skb_destructor;
1117
1118	return skb;
1119}
1120
1121/*
1122 * Attach a skb to a netlink socket.
1123 * The caller must hold a reference to the destination socket. On error, the
1124 * reference is dropped. The skb is not send to the destination, just all
1125 * all error checks are performed and memory in the queue is reserved.
1126 * Return values:
1127 * < 0: error. skb freed, reference to sock dropped.
1128 * 0: continue
1129 * 1: repeat lookup - reference dropped while waiting for socket memory.
1130 */
1131int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1132		      long *timeo, struct sock *ssk)
1133{
1134	struct netlink_sock *nlk;
1135
1136	nlk = nlk_sk(sk);
1137
1138	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1139	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1140		DECLARE_WAITQUEUE(wait, current);
1141		if (!*timeo) {
1142			if (!ssk || netlink_is_kernel(ssk))
1143				netlink_overrun(sk);
1144			sock_put(sk);
1145			kfree_skb(skb);
1146			return -EAGAIN;
1147		}
1148
1149		__set_current_state(TASK_INTERRUPTIBLE);
1150		add_wait_queue(&nlk->wait, &wait);
1151
1152		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1153		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1154		    !sock_flag(sk, SOCK_DEAD))
1155			*timeo = schedule_timeout(*timeo);
1156
1157		__set_current_state(TASK_RUNNING);
1158		remove_wait_queue(&nlk->wait, &wait);
1159		sock_put(sk);
1160
1161		if (signal_pending(current)) {
1162			kfree_skb(skb);
1163			return sock_intr_errno(*timeo);
1164		}
1165		return 1;
1166	}
1167	netlink_skb_set_owner_r(skb, sk);
1168	return 0;
1169}
1170
1171static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1172{
1173	int len = skb->len;
1174
1175	netlink_deliver_tap(skb);
1176
1177	skb_queue_tail(&sk->sk_receive_queue, skb);
1178	sk->sk_data_ready(sk);
1179	return len;
1180}
1181
1182int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1183{
1184	int len = __netlink_sendskb(sk, skb);
1185
1186	sock_put(sk);
1187	return len;
1188}
1189
1190void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1191{
1192	kfree_skb(skb);
1193	sock_put(sk);
1194}
1195
1196static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1197{
1198	int delta;
1199
1200	WARN_ON(skb->sk != NULL);
1201	delta = skb->end - skb->tail;
1202	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1203		return skb;
1204
1205	if (skb_shared(skb)) {
1206		struct sk_buff *nskb = skb_clone(skb, allocation);
1207		if (!nskb)
1208			return skb;
1209		consume_skb(skb);
1210		skb = nskb;
1211	}
1212
1213	if (!pskb_expand_head(skb, 0, -delta, allocation))
1214		skb->truesize -= delta;
1215
1216	return skb;
1217}
1218
1219static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1220				  struct sock *ssk)
1221{
1222	int ret;
1223	struct netlink_sock *nlk = nlk_sk(sk);
1224
1225	ret = -ECONNREFUSED;
1226	if (nlk->netlink_rcv != NULL) {
1227		ret = skb->len;
1228		netlink_skb_set_owner_r(skb, sk);
1229		NETLINK_CB(skb).sk = ssk;
1230		netlink_deliver_tap_kernel(sk, ssk, skb);
1231		nlk->netlink_rcv(skb);
1232		consume_skb(skb);
1233	} else {
1234		kfree_skb(skb);
1235	}
1236	sock_put(sk);
1237	return ret;
1238}
1239
1240int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1241		    u32 portid, int nonblock)
1242{
1243	struct sock *sk;
1244	int err;
1245	long timeo;
1246
1247	skb = netlink_trim(skb, gfp_any());
1248
1249	timeo = sock_sndtimeo(ssk, nonblock);
1250retry:
1251	sk = netlink_getsockbyportid(ssk, portid);
1252	if (IS_ERR(sk)) {
1253		kfree_skb(skb);
1254		return PTR_ERR(sk);
1255	}
1256	if (netlink_is_kernel(sk))
1257		return netlink_unicast_kernel(sk, skb, ssk);
1258
1259	if (sk_filter(sk, skb)) {
1260		err = skb->len;
1261		kfree_skb(skb);
1262		sock_put(sk);
1263		return err;
1264	}
1265
1266	err = netlink_attachskb(sk, skb, &timeo, ssk);
1267	if (err == 1)
1268		goto retry;
1269	if (err)
1270		return err;
1271
1272	return netlink_sendskb(sk, skb);
1273}
1274EXPORT_SYMBOL(netlink_unicast);
1275
1276int netlink_has_listeners(struct sock *sk, unsigned int group)
1277{
1278	int res = 0;
1279	struct listeners *listeners;
1280
1281	BUG_ON(!netlink_is_kernel(sk));
1282
1283	rcu_read_lock();
1284	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1285
1286	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1287		res = test_bit(group - 1, listeners->masks);
1288
1289	rcu_read_unlock();
1290
1291	return res;
1292}
1293EXPORT_SYMBOL_GPL(netlink_has_listeners);
1294
1295static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1296{
1297	struct netlink_sock *nlk = nlk_sk(sk);
1298
1299	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1300	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1301		netlink_skb_set_owner_r(skb, sk);
1302		__netlink_sendskb(sk, skb);
1303		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1304	}
1305	return -1;
1306}
1307
1308struct netlink_broadcast_data {
1309	struct sock *exclude_sk;
1310	struct net *net;
1311	u32 portid;
1312	u32 group;
1313	int failure;
1314	int delivery_failure;
1315	int congested;
1316	int delivered;
1317	gfp_t allocation;
1318	struct sk_buff *skb, *skb2;
1319	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1320	void *tx_data;
1321};
1322
1323static void do_one_broadcast(struct sock *sk,
1324				    struct netlink_broadcast_data *p)
1325{
1326	struct netlink_sock *nlk = nlk_sk(sk);
1327	int val;
1328
1329	if (p->exclude_sk == sk)
1330		return;
1331
1332	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1333	    !test_bit(p->group - 1, nlk->groups))
1334		return;
1335
1336	if (!net_eq(sock_net(sk), p->net)) {
1337		if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1338			return;
1339
1340		if (!peernet_has_id(sock_net(sk), p->net))
1341			return;
1342
1343		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1344				     CAP_NET_BROADCAST))
1345			return;
1346	}
1347
1348	if (p->failure) {
1349		netlink_overrun(sk);
1350		return;
1351	}
1352
1353	sock_hold(sk);
1354	if (p->skb2 == NULL) {
1355		if (skb_shared(p->skb)) {
1356			p->skb2 = skb_clone(p->skb, p->allocation);
1357		} else {
1358			p->skb2 = skb_get(p->skb);
1359			/*
1360			 * skb ownership may have been set when
1361			 * delivered to a previous socket.
1362			 */
1363			skb_orphan(p->skb2);
1364		}
1365	}
1366	if (p->skb2 == NULL) {
1367		netlink_overrun(sk);
1368		/* Clone failed. Notify ALL listeners. */
1369		p->failure = 1;
1370		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1371			p->delivery_failure = 1;
1372		goto out;
1373	}
1374	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1375		kfree_skb(p->skb2);
1376		p->skb2 = NULL;
1377		goto out;
1378	}
1379	if (sk_filter(sk, p->skb2)) {
1380		kfree_skb(p->skb2);
1381		p->skb2 = NULL;
1382		goto out;
1383	}
1384	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1385	NETLINK_CB(p->skb2).nsid_is_set = true;
1386	val = netlink_broadcast_deliver(sk, p->skb2);
1387	if (val < 0) {
1388		netlink_overrun(sk);
1389		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1390			p->delivery_failure = 1;
1391	} else {
1392		p->congested |= val;
1393		p->delivered = 1;
1394		p->skb2 = NULL;
1395	}
1396out:
1397	sock_put(sk);
1398}
1399
1400int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1401	u32 group, gfp_t allocation,
1402	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1403	void *filter_data)
1404{
1405	struct net *net = sock_net(ssk);
1406	struct netlink_broadcast_data info;
1407	struct sock *sk;
1408
1409	skb = netlink_trim(skb, allocation);
1410
1411	info.exclude_sk = ssk;
1412	info.net = net;
1413	info.portid = portid;
1414	info.group = group;
1415	info.failure = 0;
1416	info.delivery_failure = 0;
1417	info.congested = 0;
1418	info.delivered = 0;
1419	info.allocation = allocation;
1420	info.skb = skb;
1421	info.skb2 = NULL;
1422	info.tx_filter = filter;
1423	info.tx_data = filter_data;
1424
1425	/* While we sleep in clone, do not allow to change socket list */
1426
1427	netlink_lock_table();
1428
1429	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1430		do_one_broadcast(sk, &info);
1431
1432	consume_skb(skb);
1433
1434	netlink_unlock_table();
1435
1436	if (info.delivery_failure) {
1437		kfree_skb(info.skb2);
1438		return -ENOBUFS;
1439	}
1440	consume_skb(info.skb2);
1441
1442	if (info.delivered) {
1443		if (info.congested && gfpflags_allow_blocking(allocation))
1444			yield();
1445		return 0;
1446	}
1447	return -ESRCH;
1448}
1449EXPORT_SYMBOL(netlink_broadcast_filtered);
1450
1451int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1452		      u32 group, gfp_t allocation)
1453{
1454	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1455		NULL, NULL);
1456}
1457EXPORT_SYMBOL(netlink_broadcast);
1458
1459struct netlink_set_err_data {
1460	struct sock *exclude_sk;
1461	u32 portid;
1462	u32 group;
1463	int code;
1464};
1465
1466static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1467{
1468	struct netlink_sock *nlk = nlk_sk(sk);
1469	int ret = 0;
1470
1471	if (sk == p->exclude_sk)
1472		goto out;
1473
1474	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1475		goto out;
1476
1477	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1478	    !test_bit(p->group - 1, nlk->groups))
1479		goto out;
1480
1481	if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1482		ret = 1;
1483		goto out;
1484	}
1485
1486	sk->sk_err = p->code;
1487	sk->sk_error_report(sk);
1488out:
1489	return ret;
1490}
1491
1492/**
1493 * netlink_set_err - report error to broadcast listeners
1494 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1495 * @portid: the PORTID of a process that we want to skip (if any)
1496 * @group: the broadcast group that will notice the error
1497 * @code: error code, must be negative (as usual in kernelspace)
1498 *
1499 * This function returns the number of broadcast listeners that have set the
1500 * NETLINK_NO_ENOBUFS socket option.
1501 */
1502int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1503{
1504	struct netlink_set_err_data info;
1505	struct sock *sk;
1506	int ret = 0;
1507
1508	info.exclude_sk = ssk;
1509	info.portid = portid;
1510	info.group = group;
1511	/* sk->sk_err wants a positive error value */
1512	info.code = -code;
1513
1514	read_lock(&nl_table_lock);
1515
1516	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1517		ret += do_one_set_err(sk, &info);
1518
1519	read_unlock(&nl_table_lock);
1520	return ret;
1521}
1522EXPORT_SYMBOL(netlink_set_err);
1523
1524/* must be called with netlink table grabbed */
1525static void netlink_update_socket_mc(struct netlink_sock *nlk,
1526				     unsigned int group,
1527				     int is_new)
1528{
1529	int old, new = !!is_new, subscriptions;
1530
1531	old = test_bit(group - 1, nlk->groups);
1532	subscriptions = nlk->subscriptions - old + new;
1533	if (new)
1534		__set_bit(group - 1, nlk->groups);
1535	else
1536		__clear_bit(group - 1, nlk->groups);
1537	netlink_update_subscriptions(&nlk->sk, subscriptions);
1538	netlink_update_listeners(&nlk->sk);
1539}
1540
1541static int netlink_setsockopt(struct socket *sock, int level, int optname,
1542			      char __user *optval, unsigned int optlen)
1543{
1544	struct sock *sk = sock->sk;
1545	struct netlink_sock *nlk = nlk_sk(sk);
1546	unsigned int val = 0;
1547	int err;
1548
1549	if (level != SOL_NETLINK)
1550		return -ENOPROTOOPT;
1551
1552	if (optlen >= sizeof(int) &&
1553	    get_user(val, (unsigned int __user *)optval))
1554		return -EFAULT;
1555
1556	switch (optname) {
1557	case NETLINK_PKTINFO:
1558		if (val)
1559			nlk->flags |= NETLINK_F_RECV_PKTINFO;
1560		else
1561			nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1562		err = 0;
1563		break;
1564	case NETLINK_ADD_MEMBERSHIP:
1565	case NETLINK_DROP_MEMBERSHIP: {
1566		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1567			return -EPERM;
1568		err = netlink_realloc_groups(sk);
1569		if (err)
1570			return err;
1571		if (!val || val - 1 >= nlk->ngroups)
1572			return -EINVAL;
1573		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1574			err = nlk->netlink_bind(sock_net(sk), val);
1575			if (err)
1576				return err;
1577		}
1578		netlink_table_grab();
1579		netlink_update_socket_mc(nlk, val,
1580					 optname == NETLINK_ADD_MEMBERSHIP);
1581		netlink_table_ungrab();
1582		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1583			nlk->netlink_unbind(sock_net(sk), val);
1584
1585		err = 0;
1586		break;
1587	}
1588	case NETLINK_BROADCAST_ERROR:
1589		if (val)
1590			nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1591		else
1592			nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1593		err = 0;
1594		break;
1595	case NETLINK_NO_ENOBUFS:
1596		if (val) {
1597			nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1598			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1599			wake_up_interruptible(&nlk->wait);
1600		} else {
1601			nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1602		}
1603		err = 0;
1604		break;
1605	case NETLINK_LISTEN_ALL_NSID:
1606		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1607			return -EPERM;
1608
1609		if (val)
1610			nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1611		else
1612			nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1613		err = 0;
1614		break;
1615	case NETLINK_CAP_ACK:
1616		if (val)
1617			nlk->flags |= NETLINK_F_CAP_ACK;
1618		else
1619			nlk->flags &= ~NETLINK_F_CAP_ACK;
1620		err = 0;
1621		break;
1622	default:
1623		err = -ENOPROTOOPT;
1624	}
1625	return err;
1626}
1627
1628static int netlink_getsockopt(struct socket *sock, int level, int optname,
1629			      char __user *optval, int __user *optlen)
1630{
1631	struct sock *sk = sock->sk;
1632	struct netlink_sock *nlk = nlk_sk(sk);
1633	int len, val, err;
1634
1635	if (level != SOL_NETLINK)
1636		return -ENOPROTOOPT;
1637
1638	if (get_user(len, optlen))
1639		return -EFAULT;
1640	if (len < 0)
1641		return -EINVAL;
1642
1643	switch (optname) {
1644	case NETLINK_PKTINFO:
1645		if (len < sizeof(int))
1646			return -EINVAL;
1647		len = sizeof(int);
1648		val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1649		if (put_user(len, optlen) ||
1650		    put_user(val, optval))
1651			return -EFAULT;
1652		err = 0;
1653		break;
1654	case NETLINK_BROADCAST_ERROR:
1655		if (len < sizeof(int))
1656			return -EINVAL;
1657		len = sizeof(int);
1658		val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1659		if (put_user(len, optlen) ||
1660		    put_user(val, optval))
1661			return -EFAULT;
1662		err = 0;
1663		break;
1664	case NETLINK_NO_ENOBUFS:
1665		if (len < sizeof(int))
1666			return -EINVAL;
1667		len = sizeof(int);
1668		val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1669		if (put_user(len, optlen) ||
1670		    put_user(val, optval))
1671			return -EFAULT;
1672		err = 0;
1673		break;
1674	case NETLINK_LIST_MEMBERSHIPS: {
1675		int pos, idx, shift;
1676
1677		err = 0;
1678		netlink_lock_table();
1679		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1680			if (len - pos < sizeof(u32))
1681				break;
1682
1683			idx = pos / sizeof(unsigned long);
1684			shift = (pos % sizeof(unsigned long)) * 8;
1685			if (put_user((u32)(nlk->groups[idx] >> shift),
1686				     (u32 __user *)(optval + pos))) {
1687				err = -EFAULT;
1688				break;
1689			}
1690		}
1691		if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1692			err = -EFAULT;
1693		netlink_unlock_table();
1694		break;
1695	}
1696	case NETLINK_CAP_ACK:
1697		if (len < sizeof(int))
1698			return -EINVAL;
1699		len = sizeof(int);
1700		val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1701		if (put_user(len, optlen) ||
1702		    put_user(val, optval))
1703			return -EFAULT;
1704		err = 0;
1705		break;
1706	default:
1707		err = -ENOPROTOOPT;
1708	}
1709	return err;
1710}
1711
1712static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1713{
1714	struct nl_pktinfo info;
1715
1716	info.group = NETLINK_CB(skb).dst_group;
1717	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1718}
1719
1720static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1721					 struct sk_buff *skb)
1722{
1723	if (!NETLINK_CB(skb).nsid_is_set)
1724		return;
1725
1726	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1727		 &NETLINK_CB(skb).nsid);
1728}
1729
1730static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1731{
1732	struct sock *sk = sock->sk;
1733	struct netlink_sock *nlk = nlk_sk(sk);
1734	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1735	u32 dst_portid;
1736	u32 dst_group;
1737	struct sk_buff *skb;
1738	int err;
1739	struct scm_cookie scm;
1740	u32 netlink_skb_flags = 0;
1741
1742	if (msg->msg_flags&MSG_OOB)
1743		return -EOPNOTSUPP;
1744
1745	err = scm_send(sock, msg, &scm, true);
1746	if (err < 0)
1747		return err;
1748
1749	if (msg->msg_namelen) {
1750		err = -EINVAL;
1751		if (addr->nl_family != AF_NETLINK)
1752			goto out;
1753		dst_portid = addr->nl_pid;
1754		dst_group = ffs(addr->nl_groups);
1755		err =  -EPERM;
1756		if ((dst_group || dst_portid) &&
1757		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1758			goto out;
1759		netlink_skb_flags |= NETLINK_SKB_DST;
1760	} else {
1761		dst_portid = nlk->dst_portid;
1762		dst_group = nlk->dst_group;
1763	}
1764
1765	if (!nlk->bound) {
1766		err = netlink_autobind(sock);
1767		if (err)
1768			goto out;
1769	} else {
1770		/* Ensure nlk is hashed and visible. */
1771		smp_rmb();
1772	}
1773
1774	err = -EMSGSIZE;
1775	if (len > sk->sk_sndbuf - 32)
1776		goto out;
1777	err = -ENOBUFS;
1778	skb = netlink_alloc_large_skb(len, dst_group);
1779	if (skb == NULL)
1780		goto out;
1781
1782	NETLINK_CB(skb).portid	= nlk->portid;
1783	NETLINK_CB(skb).dst_group = dst_group;
1784	NETLINK_CB(skb).creds	= scm.creds;
1785	NETLINK_CB(skb).flags	= netlink_skb_flags;
1786
1787	err = -EFAULT;
1788	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1789		kfree_skb(skb);
1790		goto out;
1791	}
1792
1793	err = security_netlink_send(sk, skb);
1794	if (err) {
1795		kfree_skb(skb);
1796		goto out;
1797	}
1798
1799	if (dst_group) {
1800		atomic_inc(&skb->users);
1801		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1802	}
1803	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1804
1805out:
1806	scm_destroy(&scm);
1807	return err;
1808}
1809
1810static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1811			   int flags)
1812{
1813	struct scm_cookie scm;
1814	struct sock *sk = sock->sk;
1815	struct netlink_sock *nlk = nlk_sk(sk);
1816	int noblock = flags&MSG_DONTWAIT;
1817	size_t copied;
1818	struct sk_buff *skb, *data_skb;
1819	int err, ret;
1820
1821	if (flags&MSG_OOB)
1822		return -EOPNOTSUPP;
1823
1824	copied = 0;
1825
1826	skb = skb_recv_datagram(sk, flags, noblock, &err);
1827	if (skb == NULL)
1828		goto out;
1829
1830	data_skb = skb;
1831
1832#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1833	if (unlikely(skb_shinfo(skb)->frag_list)) {
1834		/*
1835		 * If this skb has a frag_list, then here that means that we
1836		 * will have to use the frag_list skb's data for compat tasks
1837		 * and the regular skb's data for normal (non-compat) tasks.
1838		 *
1839		 * If we need to send the compat skb, assign it to the
1840		 * 'data_skb' variable so that it will be used below for data
1841		 * copying. We keep 'skb' for everything else, including
1842		 * freeing both later.
1843		 */
1844		if (flags & MSG_CMSG_COMPAT)
1845			data_skb = skb_shinfo(skb)->frag_list;
1846	}
1847#endif
1848
1849	/* Record the max length of recvmsg() calls for future allocations */
1850	nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1851	nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1852				     SKB_WITH_OVERHEAD(32768));
1853
1854	copied = data_skb->len;
1855	if (len < copied) {
1856		msg->msg_flags |= MSG_TRUNC;
1857		copied = len;
1858	}
1859
1860	skb_reset_transport_header(data_skb);
1861	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1862
1863	if (msg->msg_name) {
1864		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1865		addr->nl_family = AF_NETLINK;
1866		addr->nl_pad    = 0;
1867		addr->nl_pid	= NETLINK_CB(skb).portid;
1868		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1869		msg->msg_namelen = sizeof(*addr);
1870	}
1871
1872	if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1873		netlink_cmsg_recv_pktinfo(msg, skb);
1874	if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1875		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1876
1877	memset(&scm, 0, sizeof(scm));
1878	scm.creds = *NETLINK_CREDS(skb);
1879	if (flags & MSG_TRUNC)
1880		copied = data_skb->len;
1881
1882	skb_free_datagram(sk, skb);
1883
1884	if (nlk->cb_running &&
1885	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1886		ret = netlink_dump(sk);
1887		if (ret) {
1888			sk->sk_err = -ret;
1889			sk->sk_error_report(sk);
1890		}
1891	}
1892
1893	scm_recv(sock, msg, &scm, flags);
1894out:
1895	netlink_rcv_wake(sk);
1896	return err ? : copied;
1897}
1898
1899static void netlink_data_ready(struct sock *sk)
1900{
1901	BUG();
1902}
1903
1904/*
1905 *	We export these functions to other modules. They provide a
1906 *	complete set of kernel non-blocking support for message
1907 *	queueing.
1908 */
1909
1910struct sock *
1911__netlink_kernel_create(struct net *net, int unit, struct module *module,
1912			struct netlink_kernel_cfg *cfg)
1913{
1914	struct socket *sock;
1915	struct sock *sk;
1916	struct netlink_sock *nlk;
1917	struct listeners *listeners = NULL;
1918	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1919	unsigned int groups;
1920
1921	BUG_ON(!nl_table);
1922
1923	if (unit < 0 || unit >= MAX_LINKS)
1924		return NULL;
1925
1926	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1927		return NULL;
1928
1929	if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1930		goto out_sock_release_nosk;
1931
1932	sk = sock->sk;
1933
1934	if (!cfg || cfg->groups < 32)
1935		groups = 32;
1936	else
1937		groups = cfg->groups;
1938
1939	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1940	if (!listeners)
1941		goto out_sock_release;
1942
1943	sk->sk_data_ready = netlink_data_ready;
1944	if (cfg && cfg->input)
1945		nlk_sk(sk)->netlink_rcv = cfg->input;
1946
1947	if (netlink_insert(sk, 0))
1948		goto out_sock_release;
1949
1950	nlk = nlk_sk(sk);
1951	nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1952
1953	netlink_table_grab();
1954	if (!nl_table[unit].registered) {
1955		nl_table[unit].groups = groups;
1956		rcu_assign_pointer(nl_table[unit].listeners, listeners);
1957		nl_table[unit].cb_mutex = cb_mutex;
1958		nl_table[unit].module = module;
1959		if (cfg) {
1960			nl_table[unit].bind = cfg->bind;
1961			nl_table[unit].unbind = cfg->unbind;
1962			nl_table[unit].flags = cfg->flags;
1963			if (cfg->compare)
1964				nl_table[unit].compare = cfg->compare;
1965		}
1966		nl_table[unit].registered = 1;
1967	} else {
1968		kfree(listeners);
1969		nl_table[unit].registered++;
1970	}
1971	netlink_table_ungrab();
1972	return sk;
1973
1974out_sock_release:
1975	kfree(listeners);
1976	netlink_kernel_release(sk);
1977	return NULL;
1978
1979out_sock_release_nosk:
1980	sock_release(sock);
1981	return NULL;
1982}
1983EXPORT_SYMBOL(__netlink_kernel_create);
1984
1985void
1986netlink_kernel_release(struct sock *sk)
1987{
1988	if (sk == NULL || sk->sk_socket == NULL)
1989		return;
1990
1991	sock_release(sk->sk_socket);
1992}
1993EXPORT_SYMBOL(netlink_kernel_release);
1994
1995int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1996{
1997	struct listeners *new, *old;
1998	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1999
2000	if (groups < 32)
2001		groups = 32;
2002
2003	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2004		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2005		if (!new)
2006			return -ENOMEM;
2007		old = nl_deref_protected(tbl->listeners);
2008		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2009		rcu_assign_pointer(tbl->listeners, new);
2010
2011		kfree_rcu(old, rcu);
2012	}
2013	tbl->groups = groups;
2014
2015	return 0;
2016}
2017
2018/**
2019 * netlink_change_ngroups - change number of multicast groups
2020 *
2021 * This changes the number of multicast groups that are available
2022 * on a certain netlink family. Note that it is not possible to
2023 * change the number of groups to below 32. Also note that it does
2024 * not implicitly call netlink_clear_multicast_users() when the
2025 * number of groups is reduced.
2026 *
2027 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2028 * @groups: The new number of groups.
2029 */
2030int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2031{
2032	int err;
2033
2034	netlink_table_grab();
2035	err = __netlink_change_ngroups(sk, groups);
2036	netlink_table_ungrab();
2037
2038	return err;
2039}
2040
2041void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2042{
2043	struct sock *sk;
2044	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2045
2046	sk_for_each_bound(sk, &tbl->mc_list)
2047		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2048}
2049
2050struct nlmsghdr *
2051__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2052{
2053	struct nlmsghdr *nlh;
2054	int size = nlmsg_msg_size(len);
2055
2056	nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2057	nlh->nlmsg_type = type;
2058	nlh->nlmsg_len = size;
2059	nlh->nlmsg_flags = flags;
2060	nlh->nlmsg_pid = portid;
2061	nlh->nlmsg_seq = seq;
2062	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2063		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2064	return nlh;
2065}
2066EXPORT_SYMBOL(__nlmsg_put);
2067
2068/*
2069 * It looks a bit ugly.
2070 * It would be better to create kernel thread.
2071 */
2072
2073static int netlink_dump(struct sock *sk)
2074{
2075	struct netlink_sock *nlk = nlk_sk(sk);
2076	struct netlink_callback *cb;
2077	struct sk_buff *skb = NULL;
2078	struct nlmsghdr *nlh;
2079	struct module *module;
2080	int len, err = -ENOBUFS;
2081	int alloc_min_size;
2082	int alloc_size;
2083
2084	mutex_lock(nlk->cb_mutex);
2085	if (!nlk->cb_running) {
2086		err = -EINVAL;
2087		goto errout_skb;
2088	}
2089
2090	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2091		goto errout_skb;
2092
2093	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2094	 * required, but it makes sense to _attempt_ a 16K bytes allocation
2095	 * to reduce number of system calls on dump operations, if user
2096	 * ever provided a big enough buffer.
2097	 */
2098	cb = &nlk->cb;
2099	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2100
2101	if (alloc_min_size < nlk->max_recvmsg_len) {
2102		alloc_size = nlk->max_recvmsg_len;
2103		skb = alloc_skb(alloc_size,
2104				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2105				__GFP_NOWARN | __GFP_NORETRY);
2106	}
2107	if (!skb) {
2108		alloc_size = alloc_min_size;
2109		skb = alloc_skb(alloc_size, GFP_KERNEL);
2110	}
2111	if (!skb)
2112		goto errout_skb;
2113
2114	/* Trim skb to allocated size. User is expected to provide buffer as
2115	 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2116	 * netlink_recvmsg())). dump will pack as many smaller messages as
2117	 * could fit within the allocated skb. skb is typically allocated
2118	 * with larger space than required (could be as much as near 2x the
2119	 * requested size with align to next power of 2 approach). Allowing
2120	 * dump to use the excess space makes it difficult for a user to have a
2121	 * reasonable static buffer based on the expected largest dump of a
2122	 * single netdev. The outcome is MSG_TRUNC error.
2123	 */
2124	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2125	netlink_skb_set_owner_r(skb, sk);
2126
2127	len = cb->dump(skb, cb);
2128
2129	if (len > 0) {
2130		mutex_unlock(nlk->cb_mutex);
2131
2132		if (sk_filter(sk, skb))
2133			kfree_skb(skb);
2134		else
2135			__netlink_sendskb(sk, skb);
2136		return 0;
2137	}
2138
2139	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2140	if (!nlh)
2141		goto errout_skb;
2142
2143	nl_dump_check_consistent(cb, nlh);
2144
2145	memcpy(nlmsg_data(nlh), &len, sizeof(len));
2146
2147	if (sk_filter(sk, skb))
2148		kfree_skb(skb);
2149	else
2150		__netlink_sendskb(sk, skb);
2151
2152	if (cb->done)
2153		cb->done(cb);
2154
2155	nlk->cb_running = false;
2156	module = cb->module;
2157	skb = cb->skb;
2158	mutex_unlock(nlk->cb_mutex);
2159	module_put(module);
2160	consume_skb(skb);
2161	return 0;
2162
2163errout_skb:
2164	mutex_unlock(nlk->cb_mutex);
2165	kfree_skb(skb);
2166	return err;
2167}
2168
2169int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2170			 const struct nlmsghdr *nlh,
2171			 struct netlink_dump_control *control)
2172{
2173	struct netlink_callback *cb;
2174	struct sock *sk;
2175	struct netlink_sock *nlk;
2176	int ret;
2177
2178	atomic_inc(&skb->users);
2179
2180	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2181	if (sk == NULL) {
2182		ret = -ECONNREFUSED;
2183		goto error_free;
2184	}
2185
2186	nlk = nlk_sk(sk);
2187	mutex_lock(nlk->cb_mutex);
2188	/* A dump is in progress... */
2189	if (nlk->cb_running) {
2190		ret = -EBUSY;
2191		goto error_unlock;
2192	}
2193	/* add reference of module which cb->dump belongs to */
2194	if (!try_module_get(control->module)) {
2195		ret = -EPROTONOSUPPORT;
2196		goto error_unlock;
2197	}
2198
2199	cb = &nlk->cb;
2200	memset(cb, 0, sizeof(*cb));
2201	cb->start = control->start;
2202	cb->dump = control->dump;
2203	cb->done = control->done;
2204	cb->nlh = nlh;
2205	cb->data = control->data;
2206	cb->module = control->module;
2207	cb->min_dump_alloc = control->min_dump_alloc;
2208	cb->skb = skb;
2209
2210	nlk->cb_running = true;
2211
2212	mutex_unlock(nlk->cb_mutex);
2213
2214	if (cb->start)
2215		cb->start(cb);
2216
2217	ret = netlink_dump(sk);
2218	sock_put(sk);
2219
2220	if (ret)
2221		return ret;
2222
2223	/* We successfully started a dump, by returning -EINTR we
2224	 * signal not to send ACK even if it was requested.
2225	 */
2226	return -EINTR;
2227
2228error_unlock:
2229	sock_put(sk);
2230	mutex_unlock(nlk->cb_mutex);
2231error_free:
2232	kfree_skb(skb);
2233	return ret;
2234}
2235EXPORT_SYMBOL(__netlink_dump_start);
2236
2237void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2238{
2239	struct sk_buff *skb;
2240	struct nlmsghdr *rep;
2241	struct nlmsgerr *errmsg;
2242	size_t payload = sizeof(*errmsg);
2243	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2244
2245	/* Error messages get the original request appened, unless the user
2246	 * requests to cap the error message.
2247	 */
2248	if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2249		payload += nlmsg_len(nlh);
2250
2251	skb = nlmsg_new(payload, GFP_KERNEL);
2252	if (!skb) {
2253		struct sock *sk;
2254
2255		sk = netlink_lookup(sock_net(in_skb->sk),
2256				    in_skb->sk->sk_protocol,
2257				    NETLINK_CB(in_skb).portid);
2258		if (sk) {
2259			sk->sk_err = ENOBUFS;
2260			sk->sk_error_report(sk);
2261			sock_put(sk);
2262		}
2263		return;
2264	}
2265
2266	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2267			  NLMSG_ERROR, payload, 0);
2268	errmsg = nlmsg_data(rep);
2269	errmsg->error = err;
2270	memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2271	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2272}
2273EXPORT_SYMBOL(netlink_ack);
2274
2275int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2276						     struct nlmsghdr *))
2277{
2278	struct nlmsghdr *nlh;
2279	int err;
2280
2281	while (skb->len >= nlmsg_total_size(0)) {
2282		int msglen;
2283
2284		nlh = nlmsg_hdr(skb);
2285		err = 0;
2286
2287		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2288			return 0;
2289
2290		/* Only requests are handled by the kernel */
2291		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2292			goto ack;
2293
2294		/* Skip control messages */
2295		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2296			goto ack;
2297
2298		err = cb(skb, nlh);
2299		if (err == -EINTR)
2300			goto skip;
2301
2302ack:
2303		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2304			netlink_ack(skb, nlh, err);
2305
2306skip:
2307		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2308		if (msglen > skb->len)
2309			msglen = skb->len;
2310		skb_pull(skb, msglen);
2311	}
2312
2313	return 0;
2314}
2315EXPORT_SYMBOL(netlink_rcv_skb);
2316
2317/**
2318 * nlmsg_notify - send a notification netlink message
2319 * @sk: netlink socket to use
2320 * @skb: notification message
2321 * @portid: destination netlink portid for reports or 0
2322 * @group: destination multicast group or 0
2323 * @report: 1 to report back, 0 to disable
2324 * @flags: allocation flags
2325 */
2326int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2327		 unsigned int group, int report, gfp_t flags)
2328{
2329	int err = 0;
2330
2331	if (group) {
2332		int exclude_portid = 0;
2333
2334		if (report) {
2335			atomic_inc(&skb->users);
2336			exclude_portid = portid;
2337		}
2338
2339		/* errors reported via destination sk->sk_err, but propagate
2340		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2341		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2342	}
2343
2344	if (report) {
2345		int err2;
2346
2347		err2 = nlmsg_unicast(sk, skb, portid);
2348		if (!err || err == -ESRCH)
2349			err = err2;
2350	}
2351
2352	return err;
2353}
2354EXPORT_SYMBOL(nlmsg_notify);
2355
2356#ifdef CONFIG_PROC_FS
2357struct nl_seq_iter {
2358	struct seq_net_private p;
2359	struct rhashtable_iter hti;
2360	int link;
2361};
2362
2363static int netlink_walk_start(struct nl_seq_iter *iter)
2364{
2365	int err;
2366
2367	err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2368				   GFP_KERNEL);
2369	if (err) {
2370		iter->link = MAX_LINKS;
2371		return err;
2372	}
2373
2374	err = rhashtable_walk_start(&iter->hti);
2375	return err == -EAGAIN ? 0 : err;
2376}
2377
2378static void netlink_walk_stop(struct nl_seq_iter *iter)
2379{
2380	rhashtable_walk_stop(&iter->hti);
2381	rhashtable_walk_exit(&iter->hti);
2382}
2383
2384static void *__netlink_seq_next(struct seq_file *seq)
2385{
2386	struct nl_seq_iter *iter = seq->private;
2387	struct netlink_sock *nlk;
2388
2389	do {
2390		for (;;) {
2391			int err;
2392
2393			nlk = rhashtable_walk_next(&iter->hti);
2394
2395			if (IS_ERR(nlk)) {
2396				if (PTR_ERR(nlk) == -EAGAIN)
2397					continue;
2398
2399				return nlk;
2400			}
2401
2402			if (nlk)
2403				break;
2404
2405			netlink_walk_stop(iter);
2406			if (++iter->link >= MAX_LINKS)
2407				return NULL;
2408
2409			err = netlink_walk_start(iter);
2410			if (err)
2411				return ERR_PTR(err);
2412		}
2413	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2414
2415	return nlk;
2416}
2417
2418static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2419{
2420	struct nl_seq_iter *iter = seq->private;
2421	void *obj = SEQ_START_TOKEN;
2422	loff_t pos;
2423	int err;
2424
2425	iter->link = 0;
2426
2427	err = netlink_walk_start(iter);
2428	if (err)
2429		return ERR_PTR(err);
2430
2431	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2432		obj = __netlink_seq_next(seq);
2433
2434	return obj;
2435}
2436
2437static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2438{
2439	++*pos;
2440	return __netlink_seq_next(seq);
2441}
2442
2443static void netlink_seq_stop(struct seq_file *seq, void *v)
2444{
2445	struct nl_seq_iter *iter = seq->private;
2446
2447	if (iter->link >= MAX_LINKS)
2448		return;
2449
2450	netlink_walk_stop(iter);
2451}
2452
2453
2454static int netlink_seq_show(struct seq_file *seq, void *v)
2455{
2456	if (v == SEQ_START_TOKEN) {
2457		seq_puts(seq,
2458			 "sk       Eth Pid    Groups   "
2459			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
2460	} else {
2461		struct sock *s = v;
2462		struct netlink_sock *nlk = nlk_sk(s);
2463
2464		seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2465			   s,
2466			   s->sk_protocol,
2467			   nlk->portid,
2468			   nlk->groups ? (u32)nlk->groups[0] : 0,
2469			   sk_rmem_alloc_get(s),
2470			   sk_wmem_alloc_get(s),
2471			   nlk->cb_running,
2472			   atomic_read(&s->sk_refcnt),
2473			   atomic_read(&s->sk_drops),
2474			   sock_i_ino(s)
2475			);
2476
2477	}
2478	return 0;
2479}
2480
2481static const struct seq_operations netlink_seq_ops = {
2482	.start  = netlink_seq_start,
2483	.next   = netlink_seq_next,
2484	.stop   = netlink_seq_stop,
2485	.show   = netlink_seq_show,
2486};
2487
2488
2489static int netlink_seq_open(struct inode *inode, struct file *file)
2490{
2491	return seq_open_net(inode, file, &netlink_seq_ops,
2492				sizeof(struct nl_seq_iter));
2493}
2494
2495static const struct file_operations netlink_seq_fops = {
2496	.owner		= THIS_MODULE,
2497	.open		= netlink_seq_open,
2498	.read		= seq_read,
2499	.llseek		= seq_lseek,
2500	.release	= seq_release_net,
2501};
2502
2503#endif
2504
2505int netlink_register_notifier(struct notifier_block *nb)
2506{
2507	return blocking_notifier_chain_register(&netlink_chain, nb);
2508}
2509EXPORT_SYMBOL(netlink_register_notifier);
2510
2511int netlink_unregister_notifier(struct notifier_block *nb)
2512{
2513	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2514}
2515EXPORT_SYMBOL(netlink_unregister_notifier);
2516
2517static const struct proto_ops netlink_ops = {
2518	.family =	PF_NETLINK,
2519	.owner =	THIS_MODULE,
2520	.release =	netlink_release,
2521	.bind =		netlink_bind,
2522	.connect =	netlink_connect,
2523	.socketpair =	sock_no_socketpair,
2524	.accept =	sock_no_accept,
2525	.getname =	netlink_getname,
2526	.poll =		datagram_poll,
2527	.ioctl =	netlink_ioctl,
2528	.listen =	sock_no_listen,
2529	.shutdown =	sock_no_shutdown,
2530	.setsockopt =	netlink_setsockopt,
2531	.getsockopt =	netlink_getsockopt,
2532	.sendmsg =	netlink_sendmsg,
2533	.recvmsg =	netlink_recvmsg,
2534	.mmap =		sock_no_mmap,
2535	.sendpage =	sock_no_sendpage,
2536};
2537
2538static const struct net_proto_family netlink_family_ops = {
2539	.family = PF_NETLINK,
2540	.create = netlink_create,
2541	.owner	= THIS_MODULE,	/* for consistency 8) */
2542};
2543
2544static int __net_init netlink_net_init(struct net *net)
2545{
2546#ifdef CONFIG_PROC_FS
2547	if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2548		return -ENOMEM;
2549#endif
2550	return 0;
2551}
2552
2553static void __net_exit netlink_net_exit(struct net *net)
2554{
2555#ifdef CONFIG_PROC_FS
2556	remove_proc_entry("netlink", net->proc_net);
2557#endif
2558}
2559
2560static void __init netlink_add_usersock_entry(void)
2561{
2562	struct listeners *listeners;
2563	int groups = 32;
2564
2565	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2566	if (!listeners)
2567		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2568
2569	netlink_table_grab();
2570
2571	nl_table[NETLINK_USERSOCK].groups = groups;
2572	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2573	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2574	nl_table[NETLINK_USERSOCK].registered = 1;
2575	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2576
2577	netlink_table_ungrab();
2578}
2579
2580static struct pernet_operations __net_initdata netlink_net_ops = {
2581	.init = netlink_net_init,
2582	.exit = netlink_net_exit,
2583};
2584
2585static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2586{
2587	const struct netlink_sock *nlk = data;
2588	struct netlink_compare_arg arg;
2589
2590	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2591	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2592}
2593
2594static const struct rhashtable_params netlink_rhashtable_params = {
2595	.head_offset = offsetof(struct netlink_sock, node),
2596	.key_len = netlink_compare_arg_len,
2597	.obj_hashfn = netlink_hash,
2598	.obj_cmpfn = netlink_compare,
2599	.automatic_shrinking = true,
2600};
2601
2602static int __init netlink_proto_init(void)
2603{
2604	int i;
2605	int err = proto_register(&netlink_proto, 0);
2606
2607	if (err != 0)
2608		goto out;
2609
2610	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2611
2612	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2613	if (!nl_table)
2614		goto panic;
2615
2616	for (i = 0; i < MAX_LINKS; i++) {
2617		if (rhashtable_init(&nl_table[i].hash,
2618				    &netlink_rhashtable_params) < 0) {
2619			while (--i > 0)
2620				rhashtable_destroy(&nl_table[i].hash);
2621			kfree(nl_table);
2622			goto panic;
2623		}
2624	}
2625
2626	INIT_LIST_HEAD(&netlink_tap_all);
2627
2628	netlink_add_usersock_entry();
2629
2630	sock_register(&netlink_family_ops);
2631	register_pernet_subsys(&netlink_net_ops);
2632	/* The netlink device handler may be needed early. */
2633	rtnetlink_init();
2634out:
2635	return err;
2636panic:
2637	panic("netlink_init: Cannot allocate nl_table\n");
2638}
2639
2640core_initcall(netlink_proto_init);