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 <asm/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 ATOMIC_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
 333		module_put(nlk->cb.module);
 334		kfree_skb(nlk->cb.skb);
 335	}
 336
 337	skb_queue_purge(&sk->sk_receive_queue);
 338
 339	if (!sock_flag(sk, SOCK_DEAD)) {
 340		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
 341		return;
 342	}
 343
 344	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 345	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 346	WARN_ON(nlk_sk(sk)->groups);
 347}
 348
 
 
 
 
 
 
 
 
 349/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
 350 * SMP. Look, when several writers sleep and reader wakes them up, all but one
 351 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 352 * this, _but_ remember, it adds useless work on UP machines.
 353 */
 354
 355void netlink_table_grab(void)
 356	__acquires(nl_table_lock)
 357{
 358	might_sleep();
 359
 360	write_lock_irq(&nl_table_lock);
 361
 362	if (atomic_read(&nl_table_users)) {
 363		DECLARE_WAITQUEUE(wait, current);
 364
 365		add_wait_queue_exclusive(&nl_table_wait, &wait);
 366		for (;;) {
 367			set_current_state(TASK_UNINTERRUPTIBLE);
 368			if (atomic_read(&nl_table_users) == 0)
 369				break;
 370			write_unlock_irq(&nl_table_lock);
 371			schedule();
 372			write_lock_irq(&nl_table_lock);
 373		}
 374
 375		__set_current_state(TASK_RUNNING);
 376		remove_wait_queue(&nl_table_wait, &wait);
 377	}
 378}
 379
 380void netlink_table_ungrab(void)
 381	__releases(nl_table_lock)
 382{
 383	write_unlock_irq(&nl_table_lock);
 384	wake_up(&nl_table_wait);
 385}
 386
 387static inline void
 388netlink_lock_table(void)
 389{
 390	/* read_lock() synchronizes us to netlink_table_grab */
 391
 392	read_lock(&nl_table_lock);
 393	atomic_inc(&nl_table_users);
 394	read_unlock(&nl_table_lock);
 395}
 396
 397static inline void
 398netlink_unlock_table(void)
 399{
 400	if (atomic_dec_and_test(&nl_table_users))
 401		wake_up(&nl_table_wait);
 402}
 403
 404struct netlink_compare_arg
 405{
 406	possible_net_t pnet;
 407	u32 portid;
 408};
 409
 410/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
 411#define netlink_compare_arg_len \
 412	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
 413
 414static inline int netlink_compare(struct rhashtable_compare_arg *arg,
 415				  const void *ptr)
 416{
 417	const struct netlink_compare_arg *x = arg->key;
 418	const struct netlink_sock *nlk = ptr;
 419
 420	return nlk->portid != x->portid ||
 421	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
 422}
 423
 424static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
 425				     struct net *net, u32 portid)
 426{
 427	memset(arg, 0, sizeof(*arg));
 428	write_pnet(&arg->pnet, net);
 429	arg->portid = portid;
 430}
 431
 432static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
 433				     struct net *net)
 434{
 435	struct netlink_compare_arg arg;
 436
 437	netlink_compare_arg_init(&arg, net, portid);
 438	return rhashtable_lookup_fast(&table->hash, &arg,
 439				      netlink_rhashtable_params);
 440}
 441
 442static int __netlink_insert(struct netlink_table *table, struct sock *sk)
 443{
 444	struct netlink_compare_arg arg;
 445
 446	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
 447	return rhashtable_lookup_insert_key(&table->hash, &arg,
 448					    &nlk_sk(sk)->node,
 449					    netlink_rhashtable_params);
 450}
 451
 452static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
 453{
 454	struct netlink_table *table = &nl_table[protocol];
 455	struct sock *sk;
 456
 457	rcu_read_lock();
 458	sk = __netlink_lookup(table, portid, net);
 459	if (sk)
 460		sock_hold(sk);
 461	rcu_read_unlock();
 462
 463	return sk;
 464}
 465
 466static const struct proto_ops netlink_ops;
 467
 468static void
 469netlink_update_listeners(struct sock *sk)
 470{
 471	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
 472	unsigned long mask;
 473	unsigned int i;
 474	struct listeners *listeners;
 475
 476	listeners = nl_deref_protected(tbl->listeners);
 477	if (!listeners)
 478		return;
 479
 480	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
 481		mask = 0;
 482		sk_for_each_bound(sk, &tbl->mc_list) {
 483			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
 484				mask |= nlk_sk(sk)->groups[i];
 485		}
 486		listeners->masks[i] = mask;
 487	}
 488	/* this function is only called with the netlink table "grabbed", which
 489	 * makes sure updates are visible before bind or setsockopt return. */
 490}
 491
 492static int netlink_insert(struct sock *sk, u32 portid)
 493{
 494	struct netlink_table *table = &nl_table[sk->sk_protocol];
 495	int err;
 496
 497	lock_sock(sk);
 498
 499	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
 500	if (nlk_sk(sk)->bound)
 501		goto err;
 502
 503	err = -ENOMEM;
 504	if (BITS_PER_LONG > 32 &&
 505	    unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
 506		goto err;
 507
 508	nlk_sk(sk)->portid = portid;
 509	sock_hold(sk);
 510
 511	err = __netlink_insert(table, sk);
 512	if (err) {
 513		/* In case the hashtable backend returns with -EBUSY
 514		 * from here, it must not escape to the caller.
 515		 */
 516		if (unlikely(err == -EBUSY))
 517			err = -EOVERFLOW;
 518		if (err == -EEXIST)
 519			err = -EADDRINUSE;
 520		sock_put(sk);
 521		goto err;
 522	}
 523
 524	/* We need to ensure that the socket is hashed and visible. */
 525	smp_wmb();
 526	nlk_sk(sk)->bound = portid;
 527
 528err:
 529	release_sock(sk);
 530	return err;
 531}
 532
 533static void netlink_remove(struct sock *sk)
 534{
 535	struct netlink_table *table;
 536
 537	table = &nl_table[sk->sk_protocol];
 538	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
 539				    netlink_rhashtable_params)) {
 540		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
 541		__sock_put(sk);
 542	}
 543
 544	netlink_table_grab();
 545	if (nlk_sk(sk)->subscriptions) {
 546		__sk_del_bind_node(sk);
 547		netlink_update_listeners(sk);
 548	}
 549	if (sk->sk_protocol == NETLINK_GENERIC)
 550		atomic_inc(&genl_sk_destructing_cnt);
 551	netlink_table_ungrab();
 552}
 553
 554static struct proto netlink_proto = {
 555	.name	  = "NETLINK",
 556	.owner	  = THIS_MODULE,
 557	.obj_size = sizeof(struct netlink_sock),
 558};
 559
 560static int __netlink_create(struct net *net, struct socket *sock,
 561			    struct mutex *cb_mutex, int protocol,
 562			    int kern)
 563{
 564	struct sock *sk;
 565	struct netlink_sock *nlk;
 566
 567	sock->ops = &netlink_ops;
 568
 569	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
 570	if (!sk)
 571		return -ENOMEM;
 572
 573	sock_init_data(sock, sk);
 574
 575	nlk = nlk_sk(sk);
 576	if (cb_mutex) {
 577		nlk->cb_mutex = cb_mutex;
 578	} else {
 579		nlk->cb_mutex = &nlk->cb_def_mutex;
 580		mutex_init(nlk->cb_mutex);
 
 
 
 581	}
 582	init_waitqueue_head(&nlk->wait);
 583
 584	sk->sk_destruct = netlink_sock_destruct;
 585	sk->sk_protocol = protocol;
 586	return 0;
 587}
 588
 589static int netlink_create(struct net *net, struct socket *sock, int protocol,
 590			  int kern)
 591{
 592	struct module *module = NULL;
 593	struct mutex *cb_mutex;
 594	struct netlink_sock *nlk;
 595	int (*bind)(struct net *net, int group);
 596	void (*unbind)(struct net *net, int group);
 597	int err = 0;
 598
 599	sock->state = SS_UNCONNECTED;
 600
 601	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 602		return -ESOCKTNOSUPPORT;
 603
 604	if (protocol < 0 || protocol >= MAX_LINKS)
 605		return -EPROTONOSUPPORT;
 606
 607	netlink_lock_table();
 608#ifdef CONFIG_MODULES
 609	if (!nl_table[protocol].registered) {
 610		netlink_unlock_table();
 611		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
 612		netlink_lock_table();
 613	}
 614#endif
 615	if (nl_table[protocol].registered &&
 616	    try_module_get(nl_table[protocol].module))
 617		module = nl_table[protocol].module;
 618	else
 619		err = -EPROTONOSUPPORT;
 620	cb_mutex = nl_table[protocol].cb_mutex;
 621	bind = nl_table[protocol].bind;
 622	unbind = nl_table[protocol].unbind;
 623	netlink_unlock_table();
 624
 625	if (err < 0)
 626		goto out;
 627
 628	err = __netlink_create(net, sock, cb_mutex, protocol, kern);
 629	if (err < 0)
 630		goto out_module;
 631
 632	local_bh_disable();
 633	sock_prot_inuse_add(net, &netlink_proto, 1);
 634	local_bh_enable();
 635
 636	nlk = nlk_sk(sock->sk);
 637	nlk->module = module;
 638	nlk->netlink_bind = bind;
 639	nlk->netlink_unbind = unbind;
 640out:
 641	return err;
 642
 643out_module:
 644	module_put(module);
 645	goto out;
 646}
 647
 648static void deferred_put_nlk_sk(struct rcu_head *head)
 649{
 650	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
 
 
 
 
 
 
 
 
 
 
 
 
 
 651
 652	sock_put(&nlk->sk);
 653}
 654
 655static int netlink_release(struct socket *sock)
 656{
 657	struct sock *sk = sock->sk;
 658	struct netlink_sock *nlk;
 659
 660	if (!sk)
 661		return 0;
 662
 663	netlink_remove(sk);
 664	sock_orphan(sk);
 665	nlk = nlk_sk(sk);
 666
 667	/*
 668	 * OK. Socket is unlinked, any packets that arrive now
 669	 * will be purged.
 670	 */
 671
 672	/* must not acquire netlink_table_lock in any way again before unbind
 673	 * and notifying genetlink is done as otherwise it might deadlock
 674	 */
 675	if (nlk->netlink_unbind) {
 676		int i;
 677
 678		for (i = 0; i < nlk->ngroups; i++)
 679			if (test_bit(i, nlk->groups))
 680				nlk->netlink_unbind(sock_net(sk), i + 1);
 681	}
 682	if (sk->sk_protocol == NETLINK_GENERIC &&
 683	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
 684		wake_up(&genl_sk_destructing_waitq);
 685
 686	sock->sk = NULL;
 687	wake_up_interruptible_all(&nlk->wait);
 688
 689	skb_queue_purge(&sk->sk_write_queue);
 690
 691	if (nlk->portid && nlk->bound) {
 692		struct netlink_notify n = {
 693						.net = sock_net(sk),
 694						.protocol = sk->sk_protocol,
 695						.portid = nlk->portid,
 696					  };
 697		atomic_notifier_call_chain(&netlink_chain,
 698				NETLINK_URELEASE, &n);
 699	}
 700
 701	module_put(nlk->module);
 702
 703	if (netlink_is_kernel(sk)) {
 704		netlink_table_grab();
 705		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
 706		if (--nl_table[sk->sk_protocol].registered == 0) {
 707			struct listeners *old;
 708
 709			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
 710			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
 711			kfree_rcu(old, rcu);
 712			nl_table[sk->sk_protocol].module = NULL;
 713			nl_table[sk->sk_protocol].bind = NULL;
 714			nl_table[sk->sk_protocol].unbind = NULL;
 715			nl_table[sk->sk_protocol].flags = 0;
 716			nl_table[sk->sk_protocol].registered = 0;
 717		}
 718		netlink_table_ungrab();
 719	}
 720
 721	kfree(nlk->groups);
 722	nlk->groups = NULL;
 723
 724	local_bh_disable();
 725	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
 726	local_bh_enable();
 727	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
 728	return 0;
 729}
 730
 731static int netlink_autobind(struct socket *sock)
 732{
 733	struct sock *sk = sock->sk;
 734	struct net *net = sock_net(sk);
 735	struct netlink_table *table = &nl_table[sk->sk_protocol];
 736	s32 portid = task_tgid_vnr(current);
 737	int err;
 738	s32 rover = -4096;
 739	bool ok;
 740
 741retry:
 742	cond_resched();
 743	rcu_read_lock();
 744	ok = !__netlink_lookup(table, portid, net);
 745	rcu_read_unlock();
 746	if (!ok) {
 747		/* Bind collision, search negative portid values. */
 748		if (rover == -4096)
 749			/* rover will be in range [S32_MIN, -4097] */
 750			rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
 751		else if (rover >= -4096)
 752			rover = -4097;
 753		portid = rover--;
 754		goto retry;
 755	}
 756
 757	err = netlink_insert(sk, portid);
 758	if (err == -EADDRINUSE)
 759		goto retry;
 760
 761	/* If 2 threads race to autobind, that is fine.  */
 762	if (err == -EBUSY)
 763		err = 0;
 764
 765	return err;
 766}
 767
 768/**
 769 * __netlink_ns_capable - General netlink message capability test
 770 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
 771 * @user_ns: The user namespace of the capability to use
 772 * @cap: The capability to use
 773 *
 774 * Test to see if the opener of the socket we received the message
 775 * from had when the netlink socket was created and the sender of the
 776 * message has has the capability @cap in the user namespace @user_ns.
 777 */
 778bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
 779			struct user_namespace *user_ns, int cap)
 780{
 781	return ((nsp->flags & NETLINK_SKB_DST) ||
 782		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
 783		ns_capable(user_ns, cap);
 784}
 785EXPORT_SYMBOL(__netlink_ns_capable);
 786
 787/**
 788 * netlink_ns_capable - General netlink message capability test
 789 * @skb: socket buffer holding a netlink command from userspace
 790 * @user_ns: The user namespace of the capability to use
 791 * @cap: The capability to use
 792 *
 793 * Test to see if the opener of the socket we received the message
 794 * from had when the netlink socket was created and the sender of the
 795 * message has has the capability @cap in the user namespace @user_ns.
 796 */
 797bool netlink_ns_capable(const struct sk_buff *skb,
 798			struct user_namespace *user_ns, int cap)
 799{
 800	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
 801}
 802EXPORT_SYMBOL(netlink_ns_capable);
 803
 804/**
 805 * netlink_capable - Netlink global message capability test
 806 * @skb: socket buffer holding a netlink command from userspace
 807 * @cap: The capability to use
 808 *
 809 * Test to see if the opener of the socket we received the message
 810 * from had when the netlink socket was created and the sender of the
 811 * message has has the capability @cap in all user namespaces.
 812 */
 813bool netlink_capable(const struct sk_buff *skb, int cap)
 814{
 815	return netlink_ns_capable(skb, &init_user_ns, cap);
 816}
 817EXPORT_SYMBOL(netlink_capable);
 818
 819/**
 820 * netlink_net_capable - Netlink network namespace message capability test
 821 * @skb: socket buffer holding a netlink command from userspace
 822 * @cap: The capability to use
 823 *
 824 * Test to see if the opener of the socket we received the message
 825 * from had when the netlink socket was created and the sender of the
 826 * message has has the capability @cap over the network namespace of
 827 * the socket we received the message from.
 828 */
 829bool netlink_net_capable(const struct sk_buff *skb, int cap)
 830{
 831	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
 832}
 833EXPORT_SYMBOL(netlink_net_capable);
 834
 835static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
 836{
 837	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
 838		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
 839}
 840
 841static void
 842netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
 843{
 844	struct netlink_sock *nlk = nlk_sk(sk);
 845
 846	if (nlk->subscriptions && !subscriptions)
 847		__sk_del_bind_node(sk);
 848	else if (!nlk->subscriptions && subscriptions)
 849		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
 850	nlk->subscriptions = subscriptions;
 851}
 852
 853static int netlink_realloc_groups(struct sock *sk)
 854{
 855	struct netlink_sock *nlk = nlk_sk(sk);
 856	unsigned int groups;
 857	unsigned long *new_groups;
 858	int err = 0;
 859
 860	netlink_table_grab();
 861
 862	groups = nl_table[sk->sk_protocol].groups;
 863	if (!nl_table[sk->sk_protocol].registered) {
 864		err = -ENOENT;
 865		goto out_unlock;
 866	}
 867
 868	if (nlk->ngroups >= groups)
 869		goto out_unlock;
 870
 871	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
 872	if (new_groups == NULL) {
 873		err = -ENOMEM;
 874		goto out_unlock;
 875	}
 876	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
 877	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
 878
 879	nlk->groups = new_groups;
 880	nlk->ngroups = groups;
 881 out_unlock:
 882	netlink_table_ungrab();
 883	return err;
 884}
 885
 886static void netlink_undo_bind(int group, long unsigned int groups,
 887			      struct sock *sk)
 888{
 889	struct netlink_sock *nlk = nlk_sk(sk);
 890	int undo;
 891
 892	if (!nlk->netlink_unbind)
 893		return;
 894
 895	for (undo = 0; undo < group; undo++)
 896		if (test_bit(undo, &groups))
 897			nlk->netlink_unbind(sock_net(sk), undo + 1);
 898}
 899
 900static int netlink_bind(struct socket *sock, struct sockaddr *addr,
 901			int addr_len)
 902{
 903	struct sock *sk = sock->sk;
 904	struct net *net = sock_net(sk);
 905	struct netlink_sock *nlk = nlk_sk(sk);
 906	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
 907	int err;
 908	long unsigned int groups = nladdr->nl_groups;
 909	bool bound;
 910
 911	if (addr_len < sizeof(struct sockaddr_nl))
 912		return -EINVAL;
 913
 914	if (nladdr->nl_family != AF_NETLINK)
 915		return -EINVAL;
 916
 917	/* Only superuser is allowed to listen multicasts */
 918	if (groups) {
 919		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
 920			return -EPERM;
 921		err = netlink_realloc_groups(sk);
 922		if (err)
 923			return err;
 924	}
 925
 926	bound = nlk->bound;
 927	if (bound) {
 928		/* Ensure nlk->portid is up-to-date. */
 929		smp_rmb();
 930
 931		if (nladdr->nl_pid != nlk->portid)
 932			return -EINVAL;
 933	}
 934
 
 935	if (nlk->netlink_bind && groups) {
 936		int group;
 937
 938		for (group = 0; group < nlk->ngroups; group++) {
 939			if (!test_bit(group, &groups))
 940				continue;
 941			err = nlk->netlink_bind(net, group + 1);
 942			if (!err)
 943				continue;
 944			netlink_undo_bind(group, groups, sk);
 945			return err;
 946		}
 947	}
 948
 949	/* No need for barriers here as we return to user-space without
 950	 * using any of the bound attributes.
 951	 */
 952	if (!bound) {
 953		err = nladdr->nl_pid ?
 954			netlink_insert(sk, nladdr->nl_pid) :
 955			netlink_autobind(sock);
 956		if (err) {
 957			netlink_undo_bind(nlk->ngroups, groups, sk);
 958			return err;
 959		}
 960	}
 961
 962	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
 963		return 0;
 
 964
 965	netlink_table_grab();
 966	netlink_update_subscriptions(sk, nlk->subscriptions +
 967					 hweight32(groups) -
 968					 hweight32(nlk->groups[0]));
 969	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
 970	netlink_update_listeners(sk);
 971	netlink_table_ungrab();
 972
 973	return 0;
 
 
 
 
 974}
 975
 976static int netlink_connect(struct socket *sock, struct sockaddr *addr,
 977			   int alen, int flags)
 978{
 979	int err = 0;
 980	struct sock *sk = sock->sk;
 981	struct netlink_sock *nlk = nlk_sk(sk);
 982	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
 983
 984	if (alen < sizeof(addr->sa_family))
 985		return -EINVAL;
 986
 987	if (addr->sa_family == AF_UNSPEC) {
 988		sk->sk_state	= NETLINK_UNCONNECTED;
 989		nlk->dst_portid	= 0;
 990		nlk->dst_group  = 0;
 991		return 0;
 992	}
 993	if (addr->sa_family != AF_NETLINK)
 994		return -EINVAL;
 995
 
 
 
 996	if ((nladdr->nl_groups || nladdr->nl_pid) &&
 997	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
 998		return -EPERM;
 999
1000	/* No need for barriers here as we return to user-space without
1001	 * using any of the bound attributes.
1002	 */
1003	if (!nlk->bound)
1004		err = netlink_autobind(sock);
1005
1006	if (err == 0) {
1007		sk->sk_state	= NETLINK_CONNECTED;
1008		nlk->dst_portid = nladdr->nl_pid;
1009		nlk->dst_group  = ffs(nladdr->nl_groups);
1010	}
1011
1012	return err;
1013}
1014
1015static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1016			   int *addr_len, int peer)
1017{
1018	struct sock *sk = sock->sk;
1019	struct netlink_sock *nlk = nlk_sk(sk);
1020	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1021
1022	nladdr->nl_family = AF_NETLINK;
1023	nladdr->nl_pad = 0;
1024	*addr_len = sizeof(*nladdr);
1025
1026	if (peer) {
1027		nladdr->nl_pid = nlk->dst_portid;
1028		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1029	} else {
1030		nladdr->nl_pid = nlk->portid;
 
1031		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
 
1032	}
1033	return 0;
1034}
1035
1036static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1037			 unsigned long arg)
1038{
1039	/* try to hand this ioctl down to the NIC drivers.
1040	 */
1041	return -ENOIOCTLCMD;
1042}
1043
1044static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1045{
1046	struct sock *sock;
1047	struct netlink_sock *nlk;
1048
1049	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1050	if (!sock)
1051		return ERR_PTR(-ECONNREFUSED);
1052
1053	/* Don't bother queuing skb if kernel socket has no input function */
1054	nlk = nlk_sk(sock);
1055	if (sock->sk_state == NETLINK_CONNECTED &&
1056	    nlk->dst_portid != nlk_sk(ssk)->portid) {
1057		sock_put(sock);
1058		return ERR_PTR(-ECONNREFUSED);
1059	}
1060	return sock;
1061}
1062
1063struct sock *netlink_getsockbyfilp(struct file *filp)
1064{
1065	struct inode *inode = file_inode(filp);
1066	struct sock *sock;
1067
1068	if (!S_ISSOCK(inode->i_mode))
1069		return ERR_PTR(-ENOTSOCK);
1070
1071	sock = SOCKET_I(inode)->sk;
1072	if (sock->sk_family != AF_NETLINK)
1073		return ERR_PTR(-EINVAL);
1074
1075	sock_hold(sock);
1076	return sock;
1077}
1078
1079static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1080					       int broadcast)
1081{
1082	struct sk_buff *skb;
1083	void *data;
1084
1085	if (size <= NLMSG_GOODSIZE || broadcast)
1086		return alloc_skb(size, GFP_KERNEL);
1087
1088	size = SKB_DATA_ALIGN(size) +
1089	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1090
1091	data = vmalloc(size);
1092	if (data == NULL)
1093		return NULL;
1094
1095	skb = __build_skb(data, size);
1096	if (skb == NULL)
1097		vfree(data);
1098	else
1099		skb->destructor = netlink_skb_destructor;
1100
1101	return skb;
1102}
1103
1104/*
1105 * Attach a skb to a netlink socket.
1106 * The caller must hold a reference to the destination socket. On error, the
1107 * reference is dropped. The skb is not send to the destination, just all
1108 * all error checks are performed and memory in the queue is reserved.
1109 * Return values:
1110 * < 0: error. skb freed, reference to sock dropped.
1111 * 0: continue
1112 * 1: repeat lookup - reference dropped while waiting for socket memory.
1113 */
1114int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1115		      long *timeo, struct sock *ssk)
1116{
1117	struct netlink_sock *nlk;
1118
1119	nlk = nlk_sk(sk);
1120
1121	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1122	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1123		DECLARE_WAITQUEUE(wait, current);
1124		if (!*timeo) {
1125			if (!ssk || netlink_is_kernel(ssk))
1126				netlink_overrun(sk);
1127			sock_put(sk);
1128			kfree_skb(skb);
1129			return -EAGAIN;
1130		}
1131
1132		__set_current_state(TASK_INTERRUPTIBLE);
1133		add_wait_queue(&nlk->wait, &wait);
1134
1135		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1136		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1137		    !sock_flag(sk, SOCK_DEAD))
1138			*timeo = schedule_timeout(*timeo);
1139
1140		__set_current_state(TASK_RUNNING);
1141		remove_wait_queue(&nlk->wait, &wait);
1142		sock_put(sk);
1143
1144		if (signal_pending(current)) {
1145			kfree_skb(skb);
1146			return sock_intr_errno(*timeo);
1147		}
1148		return 1;
1149	}
1150	netlink_skb_set_owner_r(skb, sk);
1151	return 0;
1152}
1153
1154static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1155{
1156	int len = skb->len;
1157
1158	netlink_deliver_tap(skb);
1159
1160	skb_queue_tail(&sk->sk_receive_queue, skb);
1161	sk->sk_data_ready(sk);
1162	return len;
1163}
1164
1165int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1166{
1167	int len = __netlink_sendskb(sk, skb);
1168
1169	sock_put(sk);
1170	return len;
1171}
1172
1173void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1174{
1175	kfree_skb(skb);
1176	sock_put(sk);
1177}
1178
1179static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1180{
1181	int delta;
1182
1183	WARN_ON(skb->sk != NULL);
1184	delta = skb->end - skb->tail;
1185	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1186		return skb;
1187
1188	if (skb_shared(skb)) {
1189		struct sk_buff *nskb = skb_clone(skb, allocation);
1190		if (!nskb)
1191			return skb;
1192		consume_skb(skb);
1193		skb = nskb;
1194	}
1195
1196	if (!pskb_expand_head(skb, 0, -delta, allocation))
1197		skb->truesize -= delta;
1198
1199	return skb;
1200}
1201
1202static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1203				  struct sock *ssk)
1204{
1205	int ret;
1206	struct netlink_sock *nlk = nlk_sk(sk);
1207
1208	ret = -ECONNREFUSED;
1209	if (nlk->netlink_rcv != NULL) {
1210		ret = skb->len;
1211		netlink_skb_set_owner_r(skb, sk);
1212		NETLINK_CB(skb).sk = ssk;
1213		netlink_deliver_tap_kernel(sk, ssk, skb);
1214		nlk->netlink_rcv(skb);
1215		consume_skb(skb);
1216	} else {
1217		kfree_skb(skb);
1218	}
1219	sock_put(sk);
1220	return ret;
1221}
1222
1223int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1224		    u32 portid, int nonblock)
1225{
1226	struct sock *sk;
1227	int err;
1228	long timeo;
1229
1230	skb = netlink_trim(skb, gfp_any());
1231
1232	timeo = sock_sndtimeo(ssk, nonblock);
1233retry:
1234	sk = netlink_getsockbyportid(ssk, portid);
1235	if (IS_ERR(sk)) {
1236		kfree_skb(skb);
1237		return PTR_ERR(sk);
1238	}
1239	if (netlink_is_kernel(sk))
1240		return netlink_unicast_kernel(sk, skb, ssk);
1241
1242	if (sk_filter(sk, skb)) {
1243		err = skb->len;
1244		kfree_skb(skb);
1245		sock_put(sk);
1246		return err;
1247	}
1248
1249	err = netlink_attachskb(sk, skb, &timeo, ssk);
1250	if (err == 1)
1251		goto retry;
1252	if (err)
1253		return err;
1254
1255	return netlink_sendskb(sk, skb);
1256}
1257EXPORT_SYMBOL(netlink_unicast);
1258
1259int netlink_has_listeners(struct sock *sk, unsigned int group)
1260{
1261	int res = 0;
1262	struct listeners *listeners;
1263
1264	BUG_ON(!netlink_is_kernel(sk));
1265
1266	rcu_read_lock();
1267	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1268
1269	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1270		res = test_bit(group - 1, listeners->masks);
1271
1272	rcu_read_unlock();
1273
1274	return res;
1275}
1276EXPORT_SYMBOL_GPL(netlink_has_listeners);
1277
1278static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1279{
1280	struct netlink_sock *nlk = nlk_sk(sk);
1281
1282	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1283	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1284		netlink_skb_set_owner_r(skb, sk);
1285		__netlink_sendskb(sk, skb);
1286		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1287	}
1288	return -1;
1289}
1290
1291struct netlink_broadcast_data {
1292	struct sock *exclude_sk;
1293	struct net *net;
1294	u32 portid;
1295	u32 group;
1296	int failure;
1297	int delivery_failure;
1298	int congested;
1299	int delivered;
1300	gfp_t allocation;
1301	struct sk_buff *skb, *skb2;
1302	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1303	void *tx_data;
1304};
1305
1306static void do_one_broadcast(struct sock *sk,
1307				    struct netlink_broadcast_data *p)
1308{
1309	struct netlink_sock *nlk = nlk_sk(sk);
1310	int val;
1311
1312	if (p->exclude_sk == sk)
1313		return;
1314
1315	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1316	    !test_bit(p->group - 1, nlk->groups))
1317		return;
1318
1319	if (!net_eq(sock_net(sk), p->net)) {
1320		if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1321			return;
1322
1323		if (!peernet_has_id(sock_net(sk), p->net))
1324			return;
1325
1326		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1327				     CAP_NET_BROADCAST))
1328			return;
1329	}
1330
1331	if (p->failure) {
1332		netlink_overrun(sk);
1333		return;
1334	}
1335
1336	sock_hold(sk);
1337	if (p->skb2 == NULL) {
1338		if (skb_shared(p->skb)) {
1339			p->skb2 = skb_clone(p->skb, p->allocation);
1340		} else {
1341			p->skb2 = skb_get(p->skb);
1342			/*
1343			 * skb ownership may have been set when
1344			 * delivered to a previous socket.
1345			 */
1346			skb_orphan(p->skb2);
1347		}
1348	}
1349	if (p->skb2 == NULL) {
1350		netlink_overrun(sk);
1351		/* Clone failed. Notify ALL listeners. */
1352		p->failure = 1;
1353		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1354			p->delivery_failure = 1;
1355		goto out;
1356	}
1357	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1358		kfree_skb(p->skb2);
1359		p->skb2 = NULL;
1360		goto out;
1361	}
1362	if (sk_filter(sk, p->skb2)) {
1363		kfree_skb(p->skb2);
1364		p->skb2 = NULL;
1365		goto out;
1366	}
1367	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1368	NETLINK_CB(p->skb2).nsid_is_set = true;
 
1369	val = netlink_broadcast_deliver(sk, p->skb2);
1370	if (val < 0) {
1371		netlink_overrun(sk);
1372		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1373			p->delivery_failure = 1;
1374	} else {
1375		p->congested |= val;
1376		p->delivered = 1;
1377		p->skb2 = NULL;
1378	}
1379out:
1380	sock_put(sk);
1381}
1382
1383int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1384	u32 group, gfp_t allocation,
1385	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1386	void *filter_data)
1387{
1388	struct net *net = sock_net(ssk);
1389	struct netlink_broadcast_data info;
1390	struct sock *sk;
1391
1392	skb = netlink_trim(skb, allocation);
1393
1394	info.exclude_sk = ssk;
1395	info.net = net;
1396	info.portid = portid;
1397	info.group = group;
1398	info.failure = 0;
1399	info.delivery_failure = 0;
1400	info.congested = 0;
1401	info.delivered = 0;
1402	info.allocation = allocation;
1403	info.skb = skb;
1404	info.skb2 = NULL;
1405	info.tx_filter = filter;
1406	info.tx_data = filter_data;
1407
1408	/* While we sleep in clone, do not allow to change socket list */
1409
1410	netlink_lock_table();
1411
1412	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1413		do_one_broadcast(sk, &info);
1414
1415	consume_skb(skb);
1416
1417	netlink_unlock_table();
1418
1419	if (info.delivery_failure) {
1420		kfree_skb(info.skb2);
1421		return -ENOBUFS;
1422	}
1423	consume_skb(info.skb2);
1424
1425	if (info.delivered) {
1426		if (info.congested && gfpflags_allow_blocking(allocation))
1427			yield();
1428		return 0;
1429	}
1430	return -ESRCH;
1431}
1432EXPORT_SYMBOL(netlink_broadcast_filtered);
1433
1434int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1435		      u32 group, gfp_t allocation)
1436{
1437	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1438		NULL, NULL);
1439}
1440EXPORT_SYMBOL(netlink_broadcast);
1441
1442struct netlink_set_err_data {
1443	struct sock *exclude_sk;
1444	u32 portid;
1445	u32 group;
1446	int code;
1447};
1448
1449static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1450{
1451	struct netlink_sock *nlk = nlk_sk(sk);
1452	int ret = 0;
1453
1454	if (sk == p->exclude_sk)
1455		goto out;
1456
1457	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1458		goto out;
1459
1460	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1461	    !test_bit(p->group - 1, nlk->groups))
1462		goto out;
1463
1464	if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1465		ret = 1;
1466		goto out;
1467	}
1468
1469	sk->sk_err = p->code;
1470	sk->sk_error_report(sk);
1471out:
1472	return ret;
1473}
1474
1475/**
1476 * netlink_set_err - report error to broadcast listeners
1477 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1478 * @portid: the PORTID of a process that we want to skip (if any)
1479 * @group: the broadcast group that will notice the error
1480 * @code: error code, must be negative (as usual in kernelspace)
1481 *
1482 * This function returns the number of broadcast listeners that have set the
1483 * NETLINK_NO_ENOBUFS socket option.
1484 */
1485int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1486{
1487	struct netlink_set_err_data info;
1488	struct sock *sk;
1489	int ret = 0;
1490
1491	info.exclude_sk = ssk;
1492	info.portid = portid;
1493	info.group = group;
1494	/* sk->sk_err wants a positive error value */
1495	info.code = -code;
1496
1497	read_lock(&nl_table_lock);
1498
1499	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1500		ret += do_one_set_err(sk, &info);
1501
1502	read_unlock(&nl_table_lock);
1503	return ret;
1504}
1505EXPORT_SYMBOL(netlink_set_err);
1506
1507/* must be called with netlink table grabbed */
1508static void netlink_update_socket_mc(struct netlink_sock *nlk,
1509				     unsigned int group,
1510				     int is_new)
1511{
1512	int old, new = !!is_new, subscriptions;
1513
1514	old = test_bit(group - 1, nlk->groups);
1515	subscriptions = nlk->subscriptions - old + new;
1516	if (new)
1517		__set_bit(group - 1, nlk->groups);
1518	else
1519		__clear_bit(group - 1, nlk->groups);
1520	netlink_update_subscriptions(&nlk->sk, subscriptions);
1521	netlink_update_listeners(&nlk->sk);
1522}
1523
1524static int netlink_setsockopt(struct socket *sock, int level, int optname,
1525			      char __user *optval, unsigned int optlen)
1526{
1527	struct sock *sk = sock->sk;
1528	struct netlink_sock *nlk = nlk_sk(sk);
1529	unsigned int val = 0;
1530	int err;
1531
1532	if (level != SOL_NETLINK)
1533		return -ENOPROTOOPT;
1534
1535	if (optlen >= sizeof(int) &&
1536	    get_user(val, (unsigned int __user *)optval))
1537		return -EFAULT;
1538
1539	switch (optname) {
1540	case NETLINK_PKTINFO:
1541		if (val)
1542			nlk->flags |= NETLINK_F_RECV_PKTINFO;
1543		else
1544			nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1545		err = 0;
1546		break;
1547	case NETLINK_ADD_MEMBERSHIP:
1548	case NETLINK_DROP_MEMBERSHIP: {
1549		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1550			return -EPERM;
1551		err = netlink_realloc_groups(sk);
1552		if (err)
1553			return err;
1554		if (!val || val - 1 >= nlk->ngroups)
1555			return -EINVAL;
1556		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1557			err = nlk->netlink_bind(sock_net(sk), val);
1558			if (err)
1559				return err;
1560		}
1561		netlink_table_grab();
1562		netlink_update_socket_mc(nlk, val,
1563					 optname == NETLINK_ADD_MEMBERSHIP);
1564		netlink_table_ungrab();
1565		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1566			nlk->netlink_unbind(sock_net(sk), val);
1567
1568		err = 0;
1569		break;
1570	}
1571	case NETLINK_BROADCAST_ERROR:
1572		if (val)
1573			nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1574		else
1575			nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1576		err = 0;
1577		break;
1578	case NETLINK_NO_ENOBUFS:
1579		if (val) {
1580			nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1581			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1582			wake_up_interruptible(&nlk->wait);
1583		} else {
1584			nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1585		}
1586		err = 0;
1587		break;
1588	case NETLINK_LISTEN_ALL_NSID:
1589		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1590			return -EPERM;
1591
1592		if (val)
1593			nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1594		else
1595			nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1596		err = 0;
1597		break;
1598	case NETLINK_CAP_ACK:
1599		if (val)
1600			nlk->flags |= NETLINK_F_CAP_ACK;
1601		else
1602			nlk->flags &= ~NETLINK_F_CAP_ACK;
1603		err = 0;
1604		break;
 
 
 
 
 
 
 
1605	default:
1606		err = -ENOPROTOOPT;
1607	}
1608	return err;
1609}
1610
1611static int netlink_getsockopt(struct socket *sock, int level, int optname,
1612			      char __user *optval, int __user *optlen)
1613{
1614	struct sock *sk = sock->sk;
1615	struct netlink_sock *nlk = nlk_sk(sk);
1616	int len, val, err;
1617
1618	if (level != SOL_NETLINK)
1619		return -ENOPROTOOPT;
1620
1621	if (get_user(len, optlen))
1622		return -EFAULT;
1623	if (len < 0)
1624		return -EINVAL;
1625
1626	switch (optname) {
1627	case NETLINK_PKTINFO:
1628		if (len < sizeof(int))
1629			return -EINVAL;
1630		len = sizeof(int);
1631		val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1632		if (put_user(len, optlen) ||
1633		    put_user(val, optval))
1634			return -EFAULT;
1635		err = 0;
1636		break;
1637	case NETLINK_BROADCAST_ERROR:
1638		if (len < sizeof(int))
1639			return -EINVAL;
1640		len = sizeof(int);
1641		val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1642		if (put_user(len, optlen) ||
1643		    put_user(val, optval))
1644			return -EFAULT;
1645		err = 0;
1646		break;
1647	case NETLINK_NO_ENOBUFS:
1648		if (len < sizeof(int))
1649			return -EINVAL;
1650		len = sizeof(int);
1651		val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1652		if (put_user(len, optlen) ||
1653		    put_user(val, optval))
1654			return -EFAULT;
1655		err = 0;
1656		break;
1657	case NETLINK_LIST_MEMBERSHIPS: {
1658		int pos, idx, shift;
1659
1660		err = 0;
1661		netlink_lock_table();
1662		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1663			if (len - pos < sizeof(u32))
1664				break;
1665
1666			idx = pos / sizeof(unsigned long);
1667			shift = (pos % sizeof(unsigned long)) * 8;
1668			if (put_user((u32)(nlk->groups[idx] >> shift),
1669				     (u32 __user *)(optval + pos))) {
1670				err = -EFAULT;
1671				break;
1672			}
1673		}
1674		if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1675			err = -EFAULT;
1676		netlink_unlock_table();
1677		break;
1678	}
1679	case NETLINK_CAP_ACK:
1680		if (len < sizeof(int))
1681			return -EINVAL;
1682		len = sizeof(int);
1683		val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1684		if (put_user(len, optlen) ||
1685		    put_user(val, optval))
1686			return -EFAULT;
1687		err = 0;
1688		break;
 
 
 
 
 
 
 
 
 
1689	default:
1690		err = -ENOPROTOOPT;
1691	}
1692	return err;
1693}
1694
1695static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1696{
1697	struct nl_pktinfo info;
1698
1699	info.group = NETLINK_CB(skb).dst_group;
1700	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1701}
1702
1703static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1704					 struct sk_buff *skb)
1705{
1706	if (!NETLINK_CB(skb).nsid_is_set)
1707		return;
1708
1709	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1710		 &NETLINK_CB(skb).nsid);
1711}
1712
1713static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1714{
1715	struct sock *sk = sock->sk;
1716	struct netlink_sock *nlk = nlk_sk(sk);
1717	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1718	u32 dst_portid;
1719	u32 dst_group;
1720	struct sk_buff *skb;
1721	int err;
1722	struct scm_cookie scm;
1723	u32 netlink_skb_flags = 0;
1724
1725	if (msg->msg_flags&MSG_OOB)
1726		return -EOPNOTSUPP;
1727
1728	err = scm_send(sock, msg, &scm, true);
1729	if (err < 0)
1730		return err;
1731
1732	if (msg->msg_namelen) {
1733		err = -EINVAL;
 
 
1734		if (addr->nl_family != AF_NETLINK)
1735			goto out;
1736		dst_portid = addr->nl_pid;
1737		dst_group = ffs(addr->nl_groups);
1738		err =  -EPERM;
1739		if ((dst_group || dst_portid) &&
1740		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1741			goto out;
1742		netlink_skb_flags |= NETLINK_SKB_DST;
1743	} else {
1744		dst_portid = nlk->dst_portid;
1745		dst_group = nlk->dst_group;
1746	}
1747
1748	if (!nlk->bound) {
1749		err = netlink_autobind(sock);
1750		if (err)
1751			goto out;
1752	} else {
1753		/* Ensure nlk is hashed and visible. */
1754		smp_rmb();
1755	}
1756
1757	err = -EMSGSIZE;
1758	if (len > sk->sk_sndbuf - 32)
1759		goto out;
1760	err = -ENOBUFS;
1761	skb = netlink_alloc_large_skb(len, dst_group);
1762	if (skb == NULL)
1763		goto out;
1764
1765	NETLINK_CB(skb).portid	= nlk->portid;
1766	NETLINK_CB(skb).dst_group = dst_group;
1767	NETLINK_CB(skb).creds	= scm.creds;
1768	NETLINK_CB(skb).flags	= netlink_skb_flags;
1769
1770	err = -EFAULT;
1771	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1772		kfree_skb(skb);
1773		goto out;
1774	}
1775
1776	err = security_netlink_send(sk, skb);
1777	if (err) {
1778		kfree_skb(skb);
1779		goto out;
1780	}
1781
1782	if (dst_group) {
1783		atomic_inc(&skb->users);
1784		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1785	}
1786	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1787
1788out:
1789	scm_destroy(&scm);
1790	return err;
1791}
1792
1793static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1794			   int flags)
1795{
1796	struct scm_cookie scm;
1797	struct sock *sk = sock->sk;
1798	struct netlink_sock *nlk = nlk_sk(sk);
1799	int noblock = flags&MSG_DONTWAIT;
1800	size_t copied;
1801	struct sk_buff *skb, *data_skb;
1802	int err, ret;
1803
1804	if (flags&MSG_OOB)
1805		return -EOPNOTSUPP;
1806
1807	copied = 0;
1808
1809	skb = skb_recv_datagram(sk, flags, noblock, &err);
1810	if (skb == NULL)
1811		goto out;
1812
1813	data_skb = skb;
1814
1815#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1816	if (unlikely(skb_shinfo(skb)->frag_list)) {
1817		/*
1818		 * If this skb has a frag_list, then here that means that we
1819		 * will have to use the frag_list skb's data for compat tasks
1820		 * and the regular skb's data for normal (non-compat) tasks.
1821		 *
1822		 * If we need to send the compat skb, assign it to the
1823		 * 'data_skb' variable so that it will be used below for data
1824		 * copying. We keep 'skb' for everything else, including
1825		 * freeing both later.
1826		 */
1827		if (flags & MSG_CMSG_COMPAT)
1828			data_skb = skb_shinfo(skb)->frag_list;
1829	}
1830#endif
1831
1832	/* Record the max length of recvmsg() calls for future allocations */
1833	nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1834	nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1835				     16384);
1836
1837	copied = data_skb->len;
1838	if (len < copied) {
1839		msg->msg_flags |= MSG_TRUNC;
1840		copied = len;
1841	}
1842
1843	skb_reset_transport_header(data_skb);
1844	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1845
1846	if (msg->msg_name) {
1847		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1848		addr->nl_family = AF_NETLINK;
1849		addr->nl_pad    = 0;
1850		addr->nl_pid	= NETLINK_CB(skb).portid;
1851		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1852		msg->msg_namelen = sizeof(*addr);
1853	}
1854
1855	if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1856		netlink_cmsg_recv_pktinfo(msg, skb);
1857	if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1858		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1859
1860	memset(&scm, 0, sizeof(scm));
1861	scm.creds = *NETLINK_CREDS(skb);
1862	if (flags & MSG_TRUNC)
1863		copied = data_skb->len;
1864
1865	skb_free_datagram(sk, skb);
1866
1867	if (nlk->cb_running &&
1868	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1869		ret = netlink_dump(sk);
1870		if (ret) {
1871			sk->sk_err = -ret;
1872			sk->sk_error_report(sk);
1873		}
1874	}
1875
1876	scm_recv(sock, msg, &scm, flags);
1877out:
1878	netlink_rcv_wake(sk);
1879	return err ? : copied;
1880}
1881
1882static void netlink_data_ready(struct sock *sk)
1883{
1884	BUG();
1885}
1886
1887/*
1888 *	We export these functions to other modules. They provide a
1889 *	complete set of kernel non-blocking support for message
1890 *	queueing.
1891 */
1892
1893struct sock *
1894__netlink_kernel_create(struct net *net, int unit, struct module *module,
1895			struct netlink_kernel_cfg *cfg)
1896{
1897	struct socket *sock;
1898	struct sock *sk;
1899	struct netlink_sock *nlk;
1900	struct listeners *listeners = NULL;
1901	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1902	unsigned int groups;
1903
1904	BUG_ON(!nl_table);
1905
1906	if (unit < 0 || unit >= MAX_LINKS)
1907		return NULL;
1908
1909	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1910		return NULL;
1911
1912	if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1913		goto out_sock_release_nosk;
1914
1915	sk = sock->sk;
1916
1917	if (!cfg || cfg->groups < 32)
1918		groups = 32;
1919	else
1920		groups = cfg->groups;
1921
1922	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1923	if (!listeners)
1924		goto out_sock_release;
1925
1926	sk->sk_data_ready = netlink_data_ready;
1927	if (cfg && cfg->input)
1928		nlk_sk(sk)->netlink_rcv = cfg->input;
1929
1930	if (netlink_insert(sk, 0))
1931		goto out_sock_release;
1932
1933	nlk = nlk_sk(sk);
1934	nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1935
1936	netlink_table_grab();
1937	if (!nl_table[unit].registered) {
1938		nl_table[unit].groups = groups;
1939		rcu_assign_pointer(nl_table[unit].listeners, listeners);
1940		nl_table[unit].cb_mutex = cb_mutex;
1941		nl_table[unit].module = module;
1942		if (cfg) {
1943			nl_table[unit].bind = cfg->bind;
1944			nl_table[unit].unbind = cfg->unbind;
1945			nl_table[unit].flags = cfg->flags;
1946			if (cfg->compare)
1947				nl_table[unit].compare = cfg->compare;
1948		}
1949		nl_table[unit].registered = 1;
1950	} else {
1951		kfree(listeners);
1952		nl_table[unit].registered++;
1953	}
1954	netlink_table_ungrab();
1955	return sk;
1956
1957out_sock_release:
1958	kfree(listeners);
1959	netlink_kernel_release(sk);
1960	return NULL;
1961
1962out_sock_release_nosk:
1963	sock_release(sock);
1964	return NULL;
1965}
1966EXPORT_SYMBOL(__netlink_kernel_create);
1967
1968void
1969netlink_kernel_release(struct sock *sk)
1970{
1971	if (sk == NULL || sk->sk_socket == NULL)
1972		return;
1973
1974	sock_release(sk->sk_socket);
1975}
1976EXPORT_SYMBOL(netlink_kernel_release);
1977
1978int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1979{
1980	struct listeners *new, *old;
1981	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1982
1983	if (groups < 32)
1984		groups = 32;
1985
1986	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1987		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1988		if (!new)
1989			return -ENOMEM;
1990		old = nl_deref_protected(tbl->listeners);
1991		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1992		rcu_assign_pointer(tbl->listeners, new);
1993
1994		kfree_rcu(old, rcu);
1995	}
1996	tbl->groups = groups;
1997
1998	return 0;
1999}
2000
2001/**
2002 * netlink_change_ngroups - change number of multicast groups
2003 *
2004 * This changes the number of multicast groups that are available
2005 * on a certain netlink family. Note that it is not possible to
2006 * change the number of groups to below 32. Also note that it does
2007 * not implicitly call netlink_clear_multicast_users() when the
2008 * number of groups is reduced.
2009 *
2010 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2011 * @groups: The new number of groups.
2012 */
2013int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2014{
2015	int err;
2016
2017	netlink_table_grab();
2018	err = __netlink_change_ngroups(sk, groups);
2019	netlink_table_ungrab();
2020
2021	return err;
2022}
2023
2024void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2025{
2026	struct sock *sk;
2027	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2028
2029	sk_for_each_bound(sk, &tbl->mc_list)
2030		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2031}
2032
2033struct nlmsghdr *
2034__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2035{
2036	struct nlmsghdr *nlh;
2037	int size = nlmsg_msg_size(len);
2038
2039	nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2040	nlh->nlmsg_type = type;
2041	nlh->nlmsg_len = size;
2042	nlh->nlmsg_flags = flags;
2043	nlh->nlmsg_pid = portid;
2044	nlh->nlmsg_seq = seq;
2045	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2046		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2047	return nlh;
2048}
2049EXPORT_SYMBOL(__nlmsg_put);
2050
2051/*
2052 * It looks a bit ugly.
2053 * It would be better to create kernel thread.
2054 */
2055
2056static int netlink_dump(struct sock *sk)
2057{
2058	struct netlink_sock *nlk = nlk_sk(sk);
2059	struct netlink_callback *cb;
2060	struct sk_buff *skb = NULL;
2061	struct nlmsghdr *nlh;
2062	int len, err = -ENOBUFS;
 
2063	int alloc_min_size;
2064	int alloc_size;
2065
2066	mutex_lock(nlk->cb_mutex);
2067	if (!nlk->cb_running) {
2068		err = -EINVAL;
2069		goto errout_skb;
2070	}
2071
2072	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2073		goto errout_skb;
2074
2075	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2076	 * required, but it makes sense to _attempt_ a 16K bytes allocation
2077	 * to reduce number of system calls on dump operations, if user
2078	 * ever provided a big enough buffer.
2079	 */
2080	cb = &nlk->cb;
2081	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2082
2083	if (alloc_min_size < nlk->max_recvmsg_len) {
2084		alloc_size = nlk->max_recvmsg_len;
2085		skb = alloc_skb(alloc_size, GFP_KERNEL |
2086					    __GFP_NOWARN | __GFP_NORETRY);
 
2087	}
2088	if (!skb) {
2089		alloc_size = alloc_min_size;
2090		skb = alloc_skb(alloc_size, GFP_KERNEL);
2091	}
2092	if (!skb)
2093		goto errout_skb;
2094
2095	/* Trim skb to allocated size. User is expected to provide buffer as
2096	 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2097	 * netlink_recvmsg())). dump will pack as many smaller messages as
2098	 * could fit within the allocated skb. skb is typically allocated
2099	 * with larger space than required (could be as much as near 2x the
2100	 * requested size with align to next power of 2 approach). Allowing
2101	 * dump to use the excess space makes it difficult for a user to have a
2102	 * reasonable static buffer based on the expected largest dump of a
2103	 * single netdev. The outcome is MSG_TRUNC error.
2104	 */
2105	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2106	netlink_skb_set_owner_r(skb, sk);
2107
2108	len = cb->dump(skb, cb);
 
2109
2110	if (len > 0) {
 
2111		mutex_unlock(nlk->cb_mutex);
2112
2113		if (sk_filter(sk, skb))
2114			kfree_skb(skb);
2115		else
2116			__netlink_sendskb(sk, skb);
2117		return 0;
2118	}
2119
2120	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2121	if (!nlh)
 
2122		goto errout_skb;
2123
2124	nl_dump_check_consistent(cb, nlh);
2125
2126	memcpy(nlmsg_data(nlh), &len, sizeof(len));
 
2127
2128	if (sk_filter(sk, skb))
2129		kfree_skb(skb);
2130	else
2131		__netlink_sendskb(sk, skb);
2132
2133	if (cb->done)
2134		cb->done(cb);
2135
2136	nlk->cb_running = false;
 
 
2137	mutex_unlock(nlk->cb_mutex);
2138	module_put(cb->module);
2139	consume_skb(cb->skb);
2140	return 0;
2141
2142errout_skb:
2143	mutex_unlock(nlk->cb_mutex);
2144	kfree_skb(skb);
2145	return err;
2146}
2147
2148int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2149			 const struct nlmsghdr *nlh,
2150			 struct netlink_dump_control *control)
2151{
2152	struct netlink_callback *cb;
2153	struct sock *sk;
2154	struct netlink_sock *nlk;
2155	int ret;
2156
2157	atomic_inc(&skb->users);
2158
2159	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2160	if (sk == NULL) {
2161		ret = -ECONNREFUSED;
2162		goto error_free;
2163	}
2164
2165	nlk = nlk_sk(sk);
2166	mutex_lock(nlk->cb_mutex);
2167	/* A dump is in progress... */
2168	if (nlk->cb_running) {
2169		ret = -EBUSY;
2170		goto error_unlock;
2171	}
2172	/* add reference of module which cb->dump belongs to */
2173	if (!try_module_get(control->module)) {
2174		ret = -EPROTONOSUPPORT;
2175		goto error_unlock;
2176	}
2177
2178	cb = &nlk->cb;
2179	memset(cb, 0, sizeof(*cb));
2180	cb->start = control->start;
2181	cb->dump = control->dump;
2182	cb->done = control->done;
2183	cb->nlh = nlh;
2184	cb->data = control->data;
2185	cb->module = control->module;
2186	cb->min_dump_alloc = control->min_dump_alloc;
2187	cb->skb = skb;
2188
 
 
 
 
 
 
2189	nlk->cb_running = true;
 
2190
2191	mutex_unlock(nlk->cb_mutex);
2192
2193	if (cb->start)
2194		cb->start(cb);
2195
2196	ret = netlink_dump(sk);
 
2197	sock_put(sk);
2198
2199	if (ret)
2200		return ret;
2201
2202	/* We successfully started a dump, by returning -EINTR we
2203	 * signal not to send ACK even if it was requested.
2204	 */
2205	return -EINTR;
2206
 
 
2207error_unlock:
2208	sock_put(sk);
2209	mutex_unlock(nlk->cb_mutex);
2210error_free:
2211	kfree_skb(skb);
2212	return ret;
2213}
2214EXPORT_SYMBOL(__netlink_dump_start);
2215
2216void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
 
2217{
2218	struct sk_buff *skb;
2219	struct nlmsghdr *rep;
2220	struct nlmsgerr *errmsg;
2221	size_t payload = sizeof(*errmsg);
 
2222	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
 
 
2223
2224	/* Error messages get the original request appened, unless the user
2225	 * requests to cap the error message.
 
2226	 */
2227	if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2228		payload += nlmsg_len(nlh);
2229
2230	skb = nlmsg_new(payload, GFP_KERNEL);
2231	if (!skb) {
2232		struct sock *sk;
 
 
 
 
 
 
2233
2234		sk = netlink_lookup(sock_net(in_skb->sk),
2235				    in_skb->sk->sk_protocol,
2236				    NETLINK_CB(in_skb).portid);
2237		if (sk) {
2238			sk->sk_err = ENOBUFS;
2239			sk->sk_error_report(sk);
2240			sock_put(sk);
2241		}
 
 
 
2242		return;
2243	}
2244
2245	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2246			  NLMSG_ERROR, payload, 0);
2247	errmsg = nlmsg_data(rep);
2248	errmsg->error = err;
2249	memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2250	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2251}
2252EXPORT_SYMBOL(netlink_ack);
2253
2254int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2255						     struct nlmsghdr *))
 
2256{
 
2257	struct nlmsghdr *nlh;
2258	int err;
2259
2260	while (skb->len >= nlmsg_total_size(0)) {
2261		int msglen;
2262
 
2263		nlh = nlmsg_hdr(skb);
2264		err = 0;
2265
2266		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2267			return 0;
2268
2269		/* Only requests are handled by the kernel */
2270		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2271			goto ack;
2272
2273		/* Skip control messages */
2274		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2275			goto ack;
2276
2277		err = cb(skb, nlh);
2278		if (err == -EINTR)
2279			goto skip;
2280
2281ack:
2282		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2283			netlink_ack(skb, nlh, err);
2284
2285skip:
2286		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2287		if (msglen > skb->len)
2288			msglen = skb->len;
2289		skb_pull(skb, msglen);
2290	}
2291
2292	return 0;
2293}
2294EXPORT_SYMBOL(netlink_rcv_skb);
2295
2296/**
2297 * nlmsg_notify - send a notification netlink message
2298 * @sk: netlink socket to use
2299 * @skb: notification message
2300 * @portid: destination netlink portid for reports or 0
2301 * @group: destination multicast group or 0
2302 * @report: 1 to report back, 0 to disable
2303 * @flags: allocation flags
2304 */
2305int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2306		 unsigned int group, int report, gfp_t flags)
2307{
2308	int err = 0;
2309
2310	if (group) {
2311		int exclude_portid = 0;
2312
2313		if (report) {
2314			atomic_inc(&skb->users);
2315			exclude_portid = portid;
2316		}
2317
2318		/* errors reported via destination sk->sk_err, but propagate
2319		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2320		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2321	}
2322
2323	if (report) {
2324		int err2;
2325
2326		err2 = nlmsg_unicast(sk, skb, portid);
2327		if (!err || err == -ESRCH)
2328			err = err2;
2329	}
2330
2331	return err;
2332}
2333EXPORT_SYMBOL(nlmsg_notify);
2334
2335#ifdef CONFIG_PROC_FS
2336struct nl_seq_iter {
2337	struct seq_net_private p;
2338	struct rhashtable_iter hti;
2339	int link;
2340};
2341
2342static int netlink_walk_start(struct nl_seq_iter *iter)
2343{
2344	int err;
2345
2346	err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
 
2347	if (err) {
2348		iter->link = MAX_LINKS;
2349		return err;
2350	}
2351
2352	err = rhashtable_walk_start(&iter->hti);
2353	return err == -EAGAIN ? 0 : err;
 
2354}
2355
2356static void netlink_walk_stop(struct nl_seq_iter *iter)
2357{
2358	rhashtable_walk_stop(&iter->hti);
2359	rhashtable_walk_exit(&iter->hti);
2360}
2361
2362static void *__netlink_seq_next(struct seq_file *seq)
2363{
2364	struct nl_seq_iter *iter = seq->private;
2365	struct netlink_sock *nlk;
2366
2367	do {
2368		for (;;) {
2369			int err;
2370
2371			nlk = rhashtable_walk_next(&iter->hti);
2372
2373			if (IS_ERR(nlk)) {
2374				if (PTR_ERR(nlk) == -EAGAIN)
2375					continue;
2376
2377				return nlk;
2378			}
2379
2380			if (nlk)
2381				break;
2382
2383			netlink_walk_stop(iter);
2384			if (++iter->link >= MAX_LINKS)
2385				return NULL;
2386
2387			err = netlink_walk_start(iter);
2388			if (err)
2389				return ERR_PTR(err);
2390		}
2391	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2392
2393	return nlk;
2394}
2395
2396static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2397{
2398	struct nl_seq_iter *iter = seq->private;
2399	void *obj = SEQ_START_TOKEN;
2400	loff_t pos;
2401	int err;
2402
2403	iter->link = 0;
2404
2405	err = netlink_walk_start(iter);
2406	if (err)
2407		return ERR_PTR(err);
2408
2409	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2410		obj = __netlink_seq_next(seq);
2411
2412	return obj;
2413}
2414
2415static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2416{
2417	++*pos;
2418	return __netlink_seq_next(seq);
2419}
2420
2421static void netlink_seq_stop(struct seq_file *seq, void *v)
2422{
2423	struct nl_seq_iter *iter = seq->private;
2424
2425	if (iter->link >= MAX_LINKS)
2426		return;
2427
2428	netlink_walk_stop(iter);
2429}
2430
2431
2432static int netlink_seq_show(struct seq_file *seq, void *v)
2433{
2434	if (v == SEQ_START_TOKEN) {
2435		seq_puts(seq,
2436			 "sk       Eth Pid    Groups   "
2437			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
2438	} else {
2439		struct sock *s = v;
2440		struct netlink_sock *nlk = nlk_sk(s);
2441
2442		seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2443			   s,
2444			   s->sk_protocol,
2445			   nlk->portid,
2446			   nlk->groups ? (u32)nlk->groups[0] : 0,
2447			   sk_rmem_alloc_get(s),
2448			   sk_wmem_alloc_get(s),
2449			   nlk->cb_running,
2450			   atomic_read(&s->sk_refcnt),
2451			   atomic_read(&s->sk_drops),
2452			   sock_i_ino(s)
2453			);
2454
2455	}
2456	return 0;
2457}
2458
2459static const struct seq_operations netlink_seq_ops = {
2460	.start  = netlink_seq_start,
2461	.next   = netlink_seq_next,
2462	.stop   = netlink_seq_stop,
2463	.show   = netlink_seq_show,
2464};
2465
2466
2467static int netlink_seq_open(struct inode *inode, struct file *file)
2468{
2469	return seq_open_net(inode, file, &netlink_seq_ops,
2470				sizeof(struct nl_seq_iter));
2471}
2472
2473static const struct file_operations netlink_seq_fops = {
2474	.owner		= THIS_MODULE,
2475	.open		= netlink_seq_open,
2476	.read		= seq_read,
2477	.llseek		= seq_lseek,
2478	.release	= seq_release_net,
2479};
2480
2481#endif
2482
2483int netlink_register_notifier(struct notifier_block *nb)
2484{
2485	return atomic_notifier_chain_register(&netlink_chain, nb);
2486}
2487EXPORT_SYMBOL(netlink_register_notifier);
2488
2489int netlink_unregister_notifier(struct notifier_block *nb)
2490{
2491	return atomic_notifier_chain_unregister(&netlink_chain, nb);
2492}
2493EXPORT_SYMBOL(netlink_unregister_notifier);
2494
2495static const struct proto_ops netlink_ops = {
2496	.family =	PF_NETLINK,
2497	.owner =	THIS_MODULE,
2498	.release =	netlink_release,
2499	.bind =		netlink_bind,
2500	.connect =	netlink_connect,
2501	.socketpair =	sock_no_socketpair,
2502	.accept =	sock_no_accept,
2503	.getname =	netlink_getname,
2504	.poll =		datagram_poll,
2505	.ioctl =	netlink_ioctl,
2506	.listen =	sock_no_listen,
2507	.shutdown =	sock_no_shutdown,
2508	.setsockopt =	netlink_setsockopt,
2509	.getsockopt =	netlink_getsockopt,
2510	.sendmsg =	netlink_sendmsg,
2511	.recvmsg =	netlink_recvmsg,
2512	.mmap =		sock_no_mmap,
2513	.sendpage =	sock_no_sendpage,
2514};
2515
2516static const struct net_proto_family netlink_family_ops = {
2517	.family = PF_NETLINK,
2518	.create = netlink_create,
2519	.owner	= THIS_MODULE,	/* for consistency 8) */
2520};
2521
2522static int __net_init netlink_net_init(struct net *net)
2523{
2524#ifdef CONFIG_PROC_FS
2525	if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2526		return -ENOMEM;
2527#endif
2528	return 0;
2529}
2530
2531static void __net_exit netlink_net_exit(struct net *net)
2532{
2533#ifdef CONFIG_PROC_FS
2534	remove_proc_entry("netlink", net->proc_net);
2535#endif
2536}
2537
2538static void __init netlink_add_usersock_entry(void)
2539{
2540	struct listeners *listeners;
2541	int groups = 32;
2542
2543	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2544	if (!listeners)
2545		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2546
2547	netlink_table_grab();
2548
2549	nl_table[NETLINK_USERSOCK].groups = groups;
2550	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2551	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2552	nl_table[NETLINK_USERSOCK].registered = 1;
2553	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2554
2555	netlink_table_ungrab();
2556}
2557
2558static struct pernet_operations __net_initdata netlink_net_ops = {
2559	.init = netlink_net_init,
2560	.exit = netlink_net_exit,
2561};
2562
2563static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2564{
2565	const struct netlink_sock *nlk = data;
2566	struct netlink_compare_arg arg;
2567
2568	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2569	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2570}
2571
2572static const struct rhashtable_params netlink_rhashtable_params = {
2573	.head_offset = offsetof(struct netlink_sock, node),
2574	.key_len = netlink_compare_arg_len,
2575	.obj_hashfn = netlink_hash,
2576	.obj_cmpfn = netlink_compare,
2577	.automatic_shrinking = true,
2578};
2579
2580static int __init netlink_proto_init(void)
2581{
2582	int i;
2583	int err = proto_register(&netlink_proto, 0);
2584
2585	if (err != 0)
2586		goto out;
2587
2588	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2589
2590	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2591	if (!nl_table)
2592		goto panic;
2593
2594	for (i = 0; i < MAX_LINKS; i++) {
2595		if (rhashtable_init(&nl_table[i].hash,
2596				    &netlink_rhashtable_params) < 0) {
2597			while (--i > 0)
2598				rhashtable_destroy(&nl_table[i].hash);
2599			kfree(nl_table);
2600			goto panic;
2601		}
2602	}
2603
2604	INIT_LIST_HEAD(&netlink_tap_all);
2605
2606	netlink_add_usersock_entry();
2607
2608	sock_register(&netlink_family_ops);
2609	register_pernet_subsys(&netlink_net_ops);
 
2610	/* The netlink device handler may be needed early. */
2611	rtnetlink_init();
2612out:
2613	return err;
2614panic:
2615	panic("netlink_init: Cannot allocate nl_table\n");
2616}
2617
2618core_initcall(netlink_proto_init);