Linux Audio

Check our new training course

Loading...
v6.8
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * NETLINK      Kernel-user communication protocol.
   4 *
   5 * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
   6 * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
   7 * 				Patrick McHardy <kaber@trash.net>
   8 *
   9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
  10 *                               added netlink_proto_exit
  11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
  12 * 				 use nlk_sk, as sk->protinfo is on a diet 8)
  13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
  14 * 				 - inc module use count of module that owns
  15 * 				   the kernel socket in case userspace opens
  16 * 				   socket of same protocol
  17 * 				 - remove all module support, since netlink is
  18 * 				   mandatory if CONFIG_NET=y these days
  19 */
  20
  21#include <linux/module.h>
  22
  23#include <linux/bpf.h>
  24#include <linux/capability.h>
  25#include <linux/kernel.h>
  26#include <linux/filter.h>
  27#include <linux/init.h>
  28#include <linux/signal.h>
  29#include <linux/sched.h>
  30#include <linux/errno.h>
  31#include <linux/string.h>
  32#include <linux/stat.h>
  33#include <linux/socket.h>
  34#include <linux/un.h>
  35#include <linux/fcntl.h>
  36#include <linux/termios.h>
  37#include <linux/sockios.h>
  38#include <linux/net.h>
  39#include <linux/fs.h>
  40#include <linux/slab.h>
  41#include <linux/uaccess.h>
  42#include <linux/skbuff.h>
  43#include <linux/netdevice.h>
  44#include <linux/rtnetlink.h>
  45#include <linux/proc_fs.h>
  46#include <linux/seq_file.h>
  47#include <linux/notifier.h>
  48#include <linux/security.h>
  49#include <linux/jhash.h>
  50#include <linux/jiffies.h>
  51#include <linux/random.h>
  52#include <linux/bitops.h>
  53#include <linux/mm.h>
  54#include <linux/types.h>
  55#include <linux/audit.h>
  56#include <linux/mutex.h>
  57#include <linux/vmalloc.h>
  58#include <linux/if_arp.h>
  59#include <linux/rhashtable.h>
  60#include <asm/cacheflush.h>
  61#include <linux/hash.h>
  62#include <linux/genetlink.h>
  63#include <linux/net_namespace.h>
  64#include <linux/nospec.h>
  65#include <linux/btf_ids.h>
  66
  67#include <net/net_namespace.h>
  68#include <net/netns/generic.h>
  69#include <net/sock.h>
  70#include <net/scm.h>
  71#include <net/netlink.h>
  72#define CREATE_TRACE_POINTS
  73#include <trace/events/netlink.h>
  74
  75#include "af_netlink.h"
  76
  77struct listeners {
  78	struct rcu_head		rcu;
  79	unsigned long		masks[];
  80};
  81
  82/* state bits */
  83#define NETLINK_S_CONGESTED		0x0
  84
  85static inline int netlink_is_kernel(struct sock *sk)
  86{
  87	return nlk_test_bit(KERNEL_SOCKET, sk);
  88}
  89
  90struct netlink_table *nl_table __read_mostly;
  91EXPORT_SYMBOL_GPL(nl_table);
  92
  93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
  94
  95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
  96
  97static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
  98	"nlk_cb_mutex-ROUTE",
  99	"nlk_cb_mutex-1",
 100	"nlk_cb_mutex-USERSOCK",
 101	"nlk_cb_mutex-FIREWALL",
 102	"nlk_cb_mutex-SOCK_DIAG",
 103	"nlk_cb_mutex-NFLOG",
 104	"nlk_cb_mutex-XFRM",
 105	"nlk_cb_mutex-SELINUX",
 106	"nlk_cb_mutex-ISCSI",
 107	"nlk_cb_mutex-AUDIT",
 108	"nlk_cb_mutex-FIB_LOOKUP",
 109	"nlk_cb_mutex-CONNECTOR",
 110	"nlk_cb_mutex-NETFILTER",
 111	"nlk_cb_mutex-IP6_FW",
 112	"nlk_cb_mutex-DNRTMSG",
 113	"nlk_cb_mutex-KOBJECT_UEVENT",
 114	"nlk_cb_mutex-GENERIC",
 115	"nlk_cb_mutex-17",
 116	"nlk_cb_mutex-SCSITRANSPORT",
 117	"nlk_cb_mutex-ECRYPTFS",
 118	"nlk_cb_mutex-RDMA",
 119	"nlk_cb_mutex-CRYPTO",
 120	"nlk_cb_mutex-SMC",
 121	"nlk_cb_mutex-23",
 122	"nlk_cb_mutex-24",
 123	"nlk_cb_mutex-25",
 124	"nlk_cb_mutex-26",
 125	"nlk_cb_mutex-27",
 126	"nlk_cb_mutex-28",
 127	"nlk_cb_mutex-29",
 128	"nlk_cb_mutex-30",
 129	"nlk_cb_mutex-31",
 130	"nlk_cb_mutex-MAX_LINKS"
 131};
 132
 133static int netlink_dump(struct sock *sk);
 134
 135/* nl_table locking explained:
 136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
 137 * and removal are protected with per bucket lock while using RCU list
 138 * modification primitives and may run in parallel to RCU protected lookups.
 139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
 140 * been acquired * either during or after the socket has been removed from
 141 * the list and after an RCU grace period.
 142 */
 143DEFINE_RWLOCK(nl_table_lock);
 144EXPORT_SYMBOL_GPL(nl_table_lock);
 145static atomic_t nl_table_users = ATOMIC_INIT(0);
 146
 147#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
 148
 149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
 150
 151
 152static const struct rhashtable_params netlink_rhashtable_params;
 153
 154void do_trace_netlink_extack(const char *msg)
 155{
 156	trace_netlink_extack(msg);
 157}
 158EXPORT_SYMBOL(do_trace_netlink_extack);
 159
 160static inline u32 netlink_group_mask(u32 group)
 161{
 162	if (group > 32)
 163		return 0;
 164	return group ? 1 << (group - 1) : 0;
 165}
 166
 167static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
 168					   gfp_t gfp_mask)
 169{
 170	unsigned int len = skb->len;
 171	struct sk_buff *new;
 172
 173	new = alloc_skb(len, gfp_mask);
 174	if (new == NULL)
 175		return NULL;
 176
 177	NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
 178	NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
 179	NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
 180
 181	skb_put_data(new, skb->data, len);
 182	return new;
 183}
 184
 185static unsigned int netlink_tap_net_id;
 186
 187struct netlink_tap_net {
 188	struct list_head netlink_tap_all;
 189	struct mutex netlink_tap_lock;
 190};
 191
 192int netlink_add_tap(struct netlink_tap *nt)
 193{
 194	struct net *net = dev_net(nt->dev);
 195	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 196
 197	if (unlikely(nt->dev->type != ARPHRD_NETLINK))
 198		return -EINVAL;
 199
 200	mutex_lock(&nn->netlink_tap_lock);
 201	list_add_rcu(&nt->list, &nn->netlink_tap_all);
 202	mutex_unlock(&nn->netlink_tap_lock);
 203
 204	__module_get(nt->module);
 205
 206	return 0;
 207}
 208EXPORT_SYMBOL_GPL(netlink_add_tap);
 209
 210static int __netlink_remove_tap(struct netlink_tap *nt)
 211{
 212	struct net *net = dev_net(nt->dev);
 213	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 214	bool found = false;
 215	struct netlink_tap *tmp;
 216
 217	mutex_lock(&nn->netlink_tap_lock);
 218
 219	list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
 220		if (nt == tmp) {
 221			list_del_rcu(&nt->list);
 222			found = true;
 223			goto out;
 224		}
 225	}
 226
 227	pr_warn("__netlink_remove_tap: %p not found\n", nt);
 228out:
 229	mutex_unlock(&nn->netlink_tap_lock);
 230
 231	if (found)
 232		module_put(nt->module);
 233
 234	return found ? 0 : -ENODEV;
 235}
 236
 237int netlink_remove_tap(struct netlink_tap *nt)
 238{
 239	int ret;
 240
 241	ret = __netlink_remove_tap(nt);
 242	synchronize_net();
 243
 244	return ret;
 245}
 246EXPORT_SYMBOL_GPL(netlink_remove_tap);
 247
 248static __net_init int netlink_tap_init_net(struct net *net)
 249{
 250	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 251
 252	INIT_LIST_HEAD(&nn->netlink_tap_all);
 253	mutex_init(&nn->netlink_tap_lock);
 254	return 0;
 255}
 256
 257static struct pernet_operations netlink_tap_net_ops = {
 258	.init = netlink_tap_init_net,
 259	.id   = &netlink_tap_net_id,
 260	.size = sizeof(struct netlink_tap_net),
 261};
 262
 263static bool netlink_filter_tap(const struct sk_buff *skb)
 264{
 265	struct sock *sk = skb->sk;
 266
 267	/* We take the more conservative approach and
 268	 * whitelist socket protocols that may pass.
 269	 */
 270	switch (sk->sk_protocol) {
 271	case NETLINK_ROUTE:
 272	case NETLINK_USERSOCK:
 273	case NETLINK_SOCK_DIAG:
 274	case NETLINK_NFLOG:
 275	case NETLINK_XFRM:
 276	case NETLINK_FIB_LOOKUP:
 277	case NETLINK_NETFILTER:
 278	case NETLINK_GENERIC:
 279		return true;
 280	}
 281
 282	return false;
 283}
 284
 285static int __netlink_deliver_tap_skb(struct sk_buff *skb,
 286				     struct net_device *dev)
 287{
 288	struct sk_buff *nskb;
 289	struct sock *sk = skb->sk;
 290	int ret = -ENOMEM;
 291
 292	if (!net_eq(dev_net(dev), sock_net(sk)))
 293		return 0;
 294
 295	dev_hold(dev);
 296
 297	if (is_vmalloc_addr(skb->head))
 298		nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
 299	else
 300		nskb = skb_clone(skb, GFP_ATOMIC);
 301	if (nskb) {
 302		nskb->dev = dev;
 303		nskb->protocol = htons((u16) sk->sk_protocol);
 304		nskb->pkt_type = netlink_is_kernel(sk) ?
 305				 PACKET_KERNEL : PACKET_USER;
 306		skb_reset_network_header(nskb);
 307		ret = dev_queue_xmit(nskb);
 308		if (unlikely(ret > 0))
 309			ret = net_xmit_errno(ret);
 310	}
 311
 312	dev_put(dev);
 313	return ret;
 314}
 315
 316static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
 317{
 318	int ret;
 319	struct netlink_tap *tmp;
 320
 321	if (!netlink_filter_tap(skb))
 322		return;
 323
 324	list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
 325		ret = __netlink_deliver_tap_skb(skb, tmp->dev);
 326		if (unlikely(ret))
 327			break;
 328	}
 329}
 330
 331static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
 332{
 333	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 334
 335	rcu_read_lock();
 336
 337	if (unlikely(!list_empty(&nn->netlink_tap_all)))
 338		__netlink_deliver_tap(skb, nn);
 339
 340	rcu_read_unlock();
 341}
 342
 343static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
 344				       struct sk_buff *skb)
 345{
 346	if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
 347		netlink_deliver_tap(sock_net(dst), skb);
 348}
 349
 350static void netlink_overrun(struct sock *sk)
 351{
 352	if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
 353		if (!test_and_set_bit(NETLINK_S_CONGESTED,
 354				      &nlk_sk(sk)->state)) {
 355			WRITE_ONCE(sk->sk_err, ENOBUFS);
 356			sk_error_report(sk);
 357		}
 358	}
 359	atomic_inc(&sk->sk_drops);
 360}
 361
 362static void netlink_rcv_wake(struct sock *sk)
 363{
 364	struct netlink_sock *nlk = nlk_sk(sk);
 365
 366	if (skb_queue_empty_lockless(&sk->sk_receive_queue))
 367		clear_bit(NETLINK_S_CONGESTED, &nlk->state);
 368	if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
 369		wake_up_interruptible(&nlk->wait);
 370}
 371
 372static void netlink_skb_destructor(struct sk_buff *skb)
 373{
 374	if (is_vmalloc_addr(skb->head)) {
 375		if (!skb->cloned ||
 376		    !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
 377			vfree_atomic(skb->head);
 378
 379		skb->head = NULL;
 380	}
 381	if (skb->sk != NULL)
 382		sock_rfree(skb);
 383}
 384
 385static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
 386{
 387	WARN_ON(skb->sk != NULL);
 388	skb->sk = sk;
 389	skb->destructor = netlink_skb_destructor;
 390	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
 391	sk_mem_charge(sk, skb->truesize);
 392}
 393
 394static void netlink_sock_destruct(struct sock *sk)
 395{
 396	struct netlink_sock *nlk = nlk_sk(sk);
 397
 398	if (nlk->cb_running) {
 399		if (nlk->cb.done)
 400			nlk->cb.done(&nlk->cb);
 401		module_put(nlk->cb.module);
 402		kfree_skb(nlk->cb.skb);
 403	}
 404
 405	skb_queue_purge(&sk->sk_receive_queue);
 406
 407	if (!sock_flag(sk, SOCK_DEAD)) {
 408		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
 409		return;
 410	}
 411
 412	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 413	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 414	WARN_ON(nlk_sk(sk)->groups);
 415}
 416
 417static void netlink_sock_destruct_work(struct work_struct *work)
 418{
 419	struct netlink_sock *nlk = container_of(work, struct netlink_sock,
 420						work);
 421
 422	sk_free(&nlk->sk);
 423}
 424
 425/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
 426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
 427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 428 * this, _but_ remember, it adds useless work on UP machines.
 429 */
 430
 431void netlink_table_grab(void)
 432	__acquires(nl_table_lock)
 433{
 434	might_sleep();
 435
 436	write_lock_irq(&nl_table_lock);
 437
 438	if (atomic_read(&nl_table_users)) {
 439		DECLARE_WAITQUEUE(wait, current);
 440
 441		add_wait_queue_exclusive(&nl_table_wait, &wait);
 442		for (;;) {
 443			set_current_state(TASK_UNINTERRUPTIBLE);
 444			if (atomic_read(&nl_table_users) == 0)
 445				break;
 446			write_unlock_irq(&nl_table_lock);
 447			schedule();
 448			write_lock_irq(&nl_table_lock);
 449		}
 450
 451		__set_current_state(TASK_RUNNING);
 452		remove_wait_queue(&nl_table_wait, &wait);
 453	}
 454}
 455
 456void netlink_table_ungrab(void)
 457	__releases(nl_table_lock)
 458{
 459	write_unlock_irq(&nl_table_lock);
 460	wake_up(&nl_table_wait);
 461}
 462
 463static inline void
 464netlink_lock_table(void)
 465{
 466	unsigned long flags;
 467
 468	/* read_lock() synchronizes us to netlink_table_grab */
 469
 470	read_lock_irqsave(&nl_table_lock, flags);
 471	atomic_inc(&nl_table_users);
 472	read_unlock_irqrestore(&nl_table_lock, flags);
 473}
 474
 475static inline void
 476netlink_unlock_table(void)
 477{
 478	if (atomic_dec_and_test(&nl_table_users))
 479		wake_up(&nl_table_wait);
 480}
 481
 482struct netlink_compare_arg
 483{
 484	possible_net_t pnet;
 485	u32 portid;
 486};
 487
 488/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
 489#define netlink_compare_arg_len \
 490	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
 491
 492static inline int netlink_compare(struct rhashtable_compare_arg *arg,
 493				  const void *ptr)
 494{
 495	const struct netlink_compare_arg *x = arg->key;
 496	const struct netlink_sock *nlk = ptr;
 497
 498	return nlk->portid != x->portid ||
 499	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
 500}
 501
 502static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
 503				     struct net *net, u32 portid)
 504{
 505	memset(arg, 0, sizeof(*arg));
 506	write_pnet(&arg->pnet, net);
 507	arg->portid = portid;
 508}
 509
 510static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
 511				     struct net *net)
 512{
 513	struct netlink_compare_arg arg;
 514
 515	netlink_compare_arg_init(&arg, net, portid);
 516	return rhashtable_lookup_fast(&table->hash, &arg,
 517				      netlink_rhashtable_params);
 518}
 519
 520static int __netlink_insert(struct netlink_table *table, struct sock *sk)
 521{
 522	struct netlink_compare_arg arg;
 523
 524	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
 525	return rhashtable_lookup_insert_key(&table->hash, &arg,
 526					    &nlk_sk(sk)->node,
 527					    netlink_rhashtable_params);
 528}
 529
 530static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
 531{
 532	struct netlink_table *table = &nl_table[protocol];
 533	struct sock *sk;
 534
 535	rcu_read_lock();
 536	sk = __netlink_lookup(table, portid, net);
 537	if (sk)
 538		sock_hold(sk);
 539	rcu_read_unlock();
 540
 541	return sk;
 542}
 543
 544static const struct proto_ops netlink_ops;
 545
 546static void
 547netlink_update_listeners(struct sock *sk)
 548{
 549	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
 550	unsigned long mask;
 551	unsigned int i;
 552	struct listeners *listeners;
 553
 554	listeners = nl_deref_protected(tbl->listeners);
 555	if (!listeners)
 556		return;
 557
 558	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
 559		mask = 0;
 560		sk_for_each_bound(sk, &tbl->mc_list) {
 561			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
 562				mask |= nlk_sk(sk)->groups[i];
 563		}
 564		listeners->masks[i] = mask;
 565	}
 566	/* this function is only called with the netlink table "grabbed", which
 567	 * makes sure updates are visible before bind or setsockopt return. */
 568}
 569
 570static int netlink_insert(struct sock *sk, u32 portid)
 571{
 572	struct netlink_table *table = &nl_table[sk->sk_protocol];
 573	int err;
 574
 575	lock_sock(sk);
 576
 577	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
 578	if (nlk_sk(sk)->bound)
 579		goto err;
 580
 581	/* portid can be read locklessly from netlink_getname(). */
 582	WRITE_ONCE(nlk_sk(sk)->portid, portid);
 583
 584	sock_hold(sk);
 585
 586	err = __netlink_insert(table, sk);
 587	if (err) {
 588		/* In case the hashtable backend returns with -EBUSY
 589		 * from here, it must not escape to the caller.
 590		 */
 591		if (unlikely(err == -EBUSY))
 592			err = -EOVERFLOW;
 593		if (err == -EEXIST)
 594			err = -EADDRINUSE;
 595		sock_put(sk);
 596		goto err;
 597	}
 598
 599	/* We need to ensure that the socket is hashed and visible. */
 600	smp_wmb();
 601	/* Paired with lockless reads from netlink_bind(),
 602	 * netlink_connect() and netlink_sendmsg().
 603	 */
 604	WRITE_ONCE(nlk_sk(sk)->bound, portid);
 605
 606err:
 607	release_sock(sk);
 608	return err;
 609}
 610
 611static void netlink_remove(struct sock *sk)
 612{
 613	struct netlink_table *table;
 614
 615	table = &nl_table[sk->sk_protocol];
 616	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
 617				    netlink_rhashtable_params)) {
 618		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 619		__sock_put(sk);
 620	}
 621
 622	netlink_table_grab();
 623	if (nlk_sk(sk)->subscriptions) {
 624		__sk_del_bind_node(sk);
 625		netlink_update_listeners(sk);
 626	}
 627	if (sk->sk_protocol == NETLINK_GENERIC)
 628		atomic_inc(&genl_sk_destructing_cnt);
 629	netlink_table_ungrab();
 630}
 631
 632static struct proto netlink_proto = {
 633	.name	  = "NETLINK",
 634	.owner	  = THIS_MODULE,
 635	.obj_size = sizeof(struct netlink_sock),
 636};
 637
 638static int __netlink_create(struct net *net, struct socket *sock,
 639			    struct mutex *cb_mutex, int protocol,
 640			    int kern)
 641{
 642	struct sock *sk;
 643	struct netlink_sock *nlk;
 644
 645	sock->ops = &netlink_ops;
 646
 647	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
 648	if (!sk)
 649		return -ENOMEM;
 650
 651	sock_init_data(sock, sk);
 652
 653	nlk = nlk_sk(sk);
 654	if (cb_mutex) {
 655		nlk->cb_mutex = cb_mutex;
 656	} else {
 657		nlk->cb_mutex = &nlk->cb_def_mutex;
 658		mutex_init(nlk->cb_mutex);
 659		lockdep_set_class_and_name(nlk->cb_mutex,
 660					   nlk_cb_mutex_keys + protocol,
 661					   nlk_cb_mutex_key_strings[protocol]);
 662	}
 663	init_waitqueue_head(&nlk->wait);
 664
 665	sk->sk_destruct = netlink_sock_destruct;
 666	sk->sk_protocol = protocol;
 667	return 0;
 668}
 669
 670static int netlink_create(struct net *net, struct socket *sock, int protocol,
 671			  int kern)
 672{
 673	struct module *module = NULL;
 674	struct mutex *cb_mutex;
 675	struct netlink_sock *nlk;
 676	int (*bind)(struct net *net, int group);
 677	void (*unbind)(struct net *net, int group);
 678	void (*release)(struct sock *sock, unsigned long *groups);
 679	int err = 0;
 680
 681	sock->state = SS_UNCONNECTED;
 682
 683	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 684		return -ESOCKTNOSUPPORT;
 685
 686	if (protocol < 0 || protocol >= MAX_LINKS)
 687		return -EPROTONOSUPPORT;
 688	protocol = array_index_nospec(protocol, MAX_LINKS);
 689
 690	netlink_lock_table();
 691#ifdef CONFIG_MODULES
 692	if (!nl_table[protocol].registered) {
 693		netlink_unlock_table();
 694		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
 695		netlink_lock_table();
 696	}
 697#endif
 698	if (nl_table[protocol].registered &&
 699	    try_module_get(nl_table[protocol].module))
 700		module = nl_table[protocol].module;
 701	else
 702		err = -EPROTONOSUPPORT;
 703	cb_mutex = nl_table[protocol].cb_mutex;
 704	bind = nl_table[protocol].bind;
 705	unbind = nl_table[protocol].unbind;
 706	release = nl_table[protocol].release;
 707	netlink_unlock_table();
 708
 709	if (err < 0)
 710		goto out;
 711
 712	err = __netlink_create(net, sock, cb_mutex, protocol, kern);
 713	if (err < 0)
 714		goto out_module;
 715
 716	sock_prot_inuse_add(net, &netlink_proto, 1);
 717
 718	nlk = nlk_sk(sock->sk);
 719	nlk->module = module;
 720	nlk->netlink_bind = bind;
 721	nlk->netlink_unbind = unbind;
 722	nlk->netlink_release = release;
 723out:
 724	return err;
 725
 726out_module:
 727	module_put(module);
 728	goto out;
 729}
 730
 731static void deferred_put_nlk_sk(struct rcu_head *head)
 732{
 733	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
 734	struct sock *sk = &nlk->sk;
 735
 736	kfree(nlk->groups);
 737	nlk->groups = NULL;
 738
 739	if (!refcount_dec_and_test(&sk->sk_refcnt))
 740		return;
 741
 742	if (nlk->cb_running && nlk->cb.done) {
 743		INIT_WORK(&nlk->work, netlink_sock_destruct_work);
 744		schedule_work(&nlk->work);
 745		return;
 746	}
 747
 748	sk_free(sk);
 749}
 750
 751static int netlink_release(struct socket *sock)
 752{
 753	struct sock *sk = sock->sk;
 754	struct netlink_sock *nlk;
 755
 756	if (!sk)
 757		return 0;
 758
 759	netlink_remove(sk);
 760	sock_orphan(sk);
 761	nlk = nlk_sk(sk);
 762
 763	/*
 764	 * OK. Socket is unlinked, any packets that arrive now
 765	 * will be purged.
 766	 */
 767	if (nlk->netlink_release)
 768		nlk->netlink_release(sk, nlk->groups);
 769
 770	/* must not acquire netlink_table_lock in any way again before unbind
 771	 * and notifying genetlink is done as otherwise it might deadlock
 772	 */
 773	if (nlk->netlink_unbind) {
 774		int i;
 775
 776		for (i = 0; i < nlk->ngroups; i++)
 777			if (test_bit(i, nlk->groups))
 778				nlk->netlink_unbind(sock_net(sk), i + 1);
 779	}
 780	if (sk->sk_protocol == NETLINK_GENERIC &&
 781	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
 782		wake_up(&genl_sk_destructing_waitq);
 783
 784	sock->sk = NULL;
 785	wake_up_interruptible_all(&nlk->wait);
 786
 787	skb_queue_purge(&sk->sk_write_queue);
 788
 789	if (nlk->portid && nlk->bound) {
 790		struct netlink_notify n = {
 791						.net = sock_net(sk),
 792						.protocol = sk->sk_protocol,
 793						.portid = nlk->portid,
 794					  };
 795		blocking_notifier_call_chain(&netlink_chain,
 796				NETLINK_URELEASE, &n);
 797	}
 798
 799	module_put(nlk->module);
 800
 801	if (netlink_is_kernel(sk)) {
 802		netlink_table_grab();
 803		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
 804		if (--nl_table[sk->sk_protocol].registered == 0) {
 805			struct listeners *old;
 806
 807			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
 808			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
 809			kfree_rcu(old, rcu);
 810			nl_table[sk->sk_protocol].module = NULL;
 811			nl_table[sk->sk_protocol].bind = NULL;
 812			nl_table[sk->sk_protocol].unbind = NULL;
 813			nl_table[sk->sk_protocol].flags = 0;
 814			nl_table[sk->sk_protocol].registered = 0;
 815		}
 816		netlink_table_ungrab();
 817	}
 818
 819	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
 820
 821	/* Because struct net might disappear soon, do not keep a pointer. */
 822	if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
 823		__netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
 824		/* Because of deferred_put_nlk_sk and use of work queue,
 825		 * it is possible  netns will be freed before this socket.
 826		 */
 827		sock_net_set(sk, &init_net);
 828		__netns_tracker_alloc(&init_net, &sk->ns_tracker,
 829				      false, GFP_KERNEL);
 830	}
 831	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
 832	return 0;
 833}
 834
 835static int netlink_autobind(struct socket *sock)
 836{
 837	struct sock *sk = sock->sk;
 838	struct net *net = sock_net(sk);
 839	struct netlink_table *table = &nl_table[sk->sk_protocol];
 840	s32 portid = task_tgid_vnr(current);
 841	int err;
 842	s32 rover = -4096;
 843	bool ok;
 844
 845retry:
 846	cond_resched();
 847	rcu_read_lock();
 848	ok = !__netlink_lookup(table, portid, net);
 849	rcu_read_unlock();
 850	if (!ok) {
 851		/* Bind collision, search negative portid values. */
 852		if (rover == -4096)
 853			/* rover will be in range [S32_MIN, -4097] */
 854			rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
 855		else if (rover >= -4096)
 856			rover = -4097;
 857		portid = rover--;
 858		goto retry;
 859	}
 860
 861	err = netlink_insert(sk, portid);
 862	if (err == -EADDRINUSE)
 863		goto retry;
 864
 865	/* If 2 threads race to autobind, that is fine.  */
 866	if (err == -EBUSY)
 867		err = 0;
 868
 869	return err;
 870}
 871
 872/**
 873 * __netlink_ns_capable - General netlink message capability test
 874 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
 875 * @user_ns: The user namespace of the capability to use
 876 * @cap: The capability to use
 877 *
 878 * Test to see if the opener of the socket we received the message
 879 * from had when the netlink socket was created and the sender of the
 880 * message has the capability @cap in the user namespace @user_ns.
 881 */
 882bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
 883			struct user_namespace *user_ns, int cap)
 884{
 885	return ((nsp->flags & NETLINK_SKB_DST) ||
 886		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
 887		ns_capable(user_ns, cap);
 888}
 889EXPORT_SYMBOL(__netlink_ns_capable);
 890
 891/**
 892 * netlink_ns_capable - General netlink message capability test
 893 * @skb: socket buffer holding a netlink command from userspace
 894 * @user_ns: The user namespace of the capability to use
 895 * @cap: The capability to use
 896 *
 897 * Test to see if the opener of the socket we received the message
 898 * from had when the netlink socket was created and the sender of the
 899 * message has the capability @cap in the user namespace @user_ns.
 900 */
 901bool netlink_ns_capable(const struct sk_buff *skb,
 902			struct user_namespace *user_ns, int cap)
 903{
 904	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
 905}
 906EXPORT_SYMBOL(netlink_ns_capable);
 907
 908/**
 909 * netlink_capable - Netlink global message capability test
 910 * @skb: socket buffer holding a netlink command from userspace
 911 * @cap: The capability to use
 912 *
 913 * Test to see if the opener of the socket we received the message
 914 * from had when the netlink socket was created and the sender of the
 915 * message has the capability @cap in all user namespaces.
 916 */
 917bool netlink_capable(const struct sk_buff *skb, int cap)
 918{
 919	return netlink_ns_capable(skb, &init_user_ns, cap);
 920}
 921EXPORT_SYMBOL(netlink_capable);
 922
 923/**
 924 * netlink_net_capable - Netlink network namespace message capability test
 925 * @skb: socket buffer holding a netlink command from userspace
 926 * @cap: The capability to use
 927 *
 928 * Test to see if the opener of the socket we received the message
 929 * from had when the netlink socket was created and the sender of the
 930 * message has the capability @cap over the network namespace of
 931 * the socket we received the message from.
 932 */
 933bool netlink_net_capable(const struct sk_buff *skb, int cap)
 934{
 935	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
 936}
 937EXPORT_SYMBOL(netlink_net_capable);
 938
 939static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
 940{
 941	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
 942		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
 943}
 944
 945static void
 946netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
 947{
 948	struct netlink_sock *nlk = nlk_sk(sk);
 949
 950	if (nlk->subscriptions && !subscriptions)
 951		__sk_del_bind_node(sk);
 952	else if (!nlk->subscriptions && subscriptions)
 953		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
 954	nlk->subscriptions = subscriptions;
 955}
 956
 957static int netlink_realloc_groups(struct sock *sk)
 958{
 959	struct netlink_sock *nlk = nlk_sk(sk);
 960	unsigned int groups;
 961	unsigned long *new_groups;
 962	int err = 0;
 963
 964	netlink_table_grab();
 965
 966	groups = nl_table[sk->sk_protocol].groups;
 967	if (!nl_table[sk->sk_protocol].registered) {
 968		err = -ENOENT;
 969		goto out_unlock;
 970	}
 971
 972	if (nlk->ngroups >= groups)
 973		goto out_unlock;
 974
 975	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
 976	if (new_groups == NULL) {
 977		err = -ENOMEM;
 978		goto out_unlock;
 979	}
 980	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
 981	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
 982
 983	nlk->groups = new_groups;
 984	nlk->ngroups = groups;
 985 out_unlock:
 986	netlink_table_ungrab();
 987	return err;
 988}
 989
 990static void netlink_undo_bind(int group, long unsigned int groups,
 991			      struct sock *sk)
 992{
 993	struct netlink_sock *nlk = nlk_sk(sk);
 994	int undo;
 995
 996	if (!nlk->netlink_unbind)
 997		return;
 998
 999	for (undo = 0; undo < group; undo++)
1000		if (test_bit(undo, &groups))
1001			nlk->netlink_unbind(sock_net(sk), undo + 1);
1002}
1003
1004static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1005			int addr_len)
1006{
1007	struct sock *sk = sock->sk;
1008	struct net *net = sock_net(sk);
1009	struct netlink_sock *nlk = nlk_sk(sk);
1010	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1011	int err = 0;
1012	unsigned long groups;
1013	bool bound;
1014
1015	if (addr_len < sizeof(struct sockaddr_nl))
1016		return -EINVAL;
1017
1018	if (nladdr->nl_family != AF_NETLINK)
1019		return -EINVAL;
1020	groups = nladdr->nl_groups;
1021
1022	/* Only superuser is allowed to listen multicasts */
1023	if (groups) {
1024		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1025			return -EPERM;
1026		err = netlink_realloc_groups(sk);
1027		if (err)
1028			return err;
1029	}
1030
1031	if (nlk->ngroups < BITS_PER_LONG)
1032		groups &= (1UL << nlk->ngroups) - 1;
1033
1034	/* Paired with WRITE_ONCE() in netlink_insert() */
1035	bound = READ_ONCE(nlk->bound);
1036	if (bound) {
1037		/* Ensure nlk->portid is up-to-date. */
1038		smp_rmb();
1039
1040		if (nladdr->nl_pid != nlk->portid)
1041			return -EINVAL;
1042	}
1043
1044	if (nlk->netlink_bind && groups) {
1045		int group;
1046
1047		/* nl_groups is a u32, so cap the maximum groups we can bind */
1048		for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1049			if (!test_bit(group, &groups))
1050				continue;
1051			err = nlk->netlink_bind(net, group + 1);
1052			if (!err)
1053				continue;
1054			netlink_undo_bind(group, groups, sk);
1055			return err;
1056		}
1057	}
1058
1059	/* No need for barriers here as we return to user-space without
1060	 * using any of the bound attributes.
1061	 */
1062	netlink_lock_table();
1063	if (!bound) {
1064		err = nladdr->nl_pid ?
1065			netlink_insert(sk, nladdr->nl_pid) :
1066			netlink_autobind(sock);
1067		if (err) {
1068			netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1069			goto unlock;
1070		}
1071	}
1072
1073	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1074		goto unlock;
1075	netlink_unlock_table();
1076
1077	netlink_table_grab();
1078	netlink_update_subscriptions(sk, nlk->subscriptions +
1079					 hweight32(groups) -
1080					 hweight32(nlk->groups[0]));
1081	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1082	netlink_update_listeners(sk);
1083	netlink_table_ungrab();
1084
1085	return 0;
1086
1087unlock:
1088	netlink_unlock_table();
1089	return err;
1090}
1091
1092static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1093			   int alen, int flags)
1094{
1095	int err = 0;
1096	struct sock *sk = sock->sk;
1097	struct netlink_sock *nlk = nlk_sk(sk);
1098	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1099
1100	if (alen < sizeof(addr->sa_family))
1101		return -EINVAL;
1102
1103	if (addr->sa_family == AF_UNSPEC) {
1104		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1105		WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1106		/* dst_portid and dst_group can be read locklessly */
1107		WRITE_ONCE(nlk->dst_portid, 0);
1108		WRITE_ONCE(nlk->dst_group, 0);
1109		return 0;
1110	}
1111	if (addr->sa_family != AF_NETLINK)
1112		return -EINVAL;
1113
1114	if (alen < sizeof(struct sockaddr_nl))
1115		return -EINVAL;
1116
1117	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1118	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1119		return -EPERM;
1120
1121	/* No need for barriers here as we return to user-space without
1122	 * using any of the bound attributes.
1123	 * Paired with WRITE_ONCE() in netlink_insert().
1124	 */
1125	if (!READ_ONCE(nlk->bound))
1126		err = netlink_autobind(sock);
1127
1128	if (err == 0) {
1129		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1130		WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1131		/* dst_portid and dst_group can be read locklessly */
1132		WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1133		WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1134	}
1135
1136	return err;
1137}
1138
1139static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1140			   int peer)
1141{
1142	struct sock *sk = sock->sk;
1143	struct netlink_sock *nlk = nlk_sk(sk);
1144	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1145
1146	nladdr->nl_family = AF_NETLINK;
1147	nladdr->nl_pad = 0;
1148
1149	if (peer) {
1150		/* Paired with WRITE_ONCE() in netlink_connect() */
1151		nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1152		nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1153	} else {
1154		/* Paired with WRITE_ONCE() in netlink_insert() */
1155		nladdr->nl_pid = READ_ONCE(nlk->portid);
1156		netlink_lock_table();
1157		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1158		netlink_unlock_table();
1159	}
1160	return sizeof(*nladdr);
1161}
1162
1163static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1164			 unsigned long arg)
1165{
1166	/* try to hand this ioctl down to the NIC drivers.
1167	 */
1168	return -ENOIOCTLCMD;
1169}
1170
1171static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1172{
1173	struct sock *sock;
1174	struct netlink_sock *nlk;
1175
1176	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1177	if (!sock)
1178		return ERR_PTR(-ECONNREFUSED);
1179
1180	/* Don't bother queuing skb if kernel socket has no input function */
1181	nlk = nlk_sk(sock);
1182	/* dst_portid and sk_state can be changed in netlink_connect() */
1183	if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1184	    READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1185		sock_put(sock);
1186		return ERR_PTR(-ECONNREFUSED);
1187	}
1188	return sock;
1189}
1190
1191struct sock *netlink_getsockbyfilp(struct file *filp)
1192{
1193	struct inode *inode = file_inode(filp);
1194	struct sock *sock;
1195
1196	if (!S_ISSOCK(inode->i_mode))
1197		return ERR_PTR(-ENOTSOCK);
1198
1199	sock = SOCKET_I(inode)->sk;
1200	if (sock->sk_family != AF_NETLINK)
1201		return ERR_PTR(-EINVAL);
1202
1203	sock_hold(sock);
1204	return sock;
1205}
1206
1207struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1208{
 
1209	struct sk_buff *skb;
1210	void *data;
1211
1212	if (size <= NLMSG_GOODSIZE || broadcast)
1213		return alloc_skb(size, GFP_KERNEL);
1214
1215	size = SKB_DATA_ALIGN(size) +
1216	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1217
1218	data = vmalloc(size);
1219	if (data == NULL)
1220		return NULL;
1221
1222	skb = __build_skb(data, size);
1223	if (skb == NULL)
1224		vfree(data);
1225	else
1226		skb->destructor = netlink_skb_destructor;
1227
1228	return skb;
1229}
1230
1231/*
1232 * Attach a skb to a netlink socket.
1233 * The caller must hold a reference to the destination socket. On error, the
1234 * reference is dropped. The skb is not send to the destination, just all
1235 * all error checks are performed and memory in the queue is reserved.
1236 * Return values:
1237 * < 0: error. skb freed, reference to sock dropped.
1238 * 0: continue
1239 * 1: repeat lookup - reference dropped while waiting for socket memory.
1240 */
1241int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1242		      long *timeo, struct sock *ssk)
1243{
1244	struct netlink_sock *nlk;
1245
1246	nlk = nlk_sk(sk);
1247
1248	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1249	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1250		DECLARE_WAITQUEUE(wait, current);
1251		if (!*timeo) {
1252			if (!ssk || netlink_is_kernel(ssk))
1253				netlink_overrun(sk);
1254			sock_put(sk);
1255			kfree_skb(skb);
1256			return -EAGAIN;
1257		}
1258
1259		__set_current_state(TASK_INTERRUPTIBLE);
1260		add_wait_queue(&nlk->wait, &wait);
1261
1262		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1263		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1264		    !sock_flag(sk, SOCK_DEAD))
1265			*timeo = schedule_timeout(*timeo);
1266
1267		__set_current_state(TASK_RUNNING);
1268		remove_wait_queue(&nlk->wait, &wait);
1269		sock_put(sk);
1270
1271		if (signal_pending(current)) {
1272			kfree_skb(skb);
1273			return sock_intr_errno(*timeo);
1274		}
1275		return 1;
1276	}
1277	netlink_skb_set_owner_r(skb, sk);
1278	return 0;
1279}
1280
1281static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1282{
1283	int len = skb->len;
1284
1285	netlink_deliver_tap(sock_net(sk), skb);
1286
1287	skb_queue_tail(&sk->sk_receive_queue, skb);
1288	sk->sk_data_ready(sk);
1289	return len;
1290}
1291
1292int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1293{
1294	int len = __netlink_sendskb(sk, skb);
1295
1296	sock_put(sk);
1297	return len;
1298}
1299
1300void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1301{
1302	kfree_skb(skb);
1303	sock_put(sk);
1304}
1305
1306static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1307{
1308	int delta;
1309
1310	WARN_ON(skb->sk != NULL);
1311	delta = skb->end - skb->tail;
1312	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1313		return skb;
1314
1315	if (skb_shared(skb)) {
1316		struct sk_buff *nskb = skb_clone(skb, allocation);
1317		if (!nskb)
1318			return skb;
1319		consume_skb(skb);
1320		skb = nskb;
1321	}
1322
1323	pskb_expand_head(skb, 0, -delta,
1324			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1325			 __GFP_NOWARN | __GFP_NORETRY);
1326	return skb;
1327}
1328
1329static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1330				  struct sock *ssk)
1331{
1332	int ret;
1333	struct netlink_sock *nlk = nlk_sk(sk);
1334
1335	ret = -ECONNREFUSED;
1336	if (nlk->netlink_rcv != NULL) {
1337		ret = skb->len;
1338		netlink_skb_set_owner_r(skb, sk);
1339		NETLINK_CB(skb).sk = ssk;
1340		netlink_deliver_tap_kernel(sk, ssk, skb);
1341		nlk->netlink_rcv(skb);
1342		consume_skb(skb);
1343	} else {
1344		kfree_skb(skb);
1345	}
1346	sock_put(sk);
1347	return ret;
1348}
1349
1350int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1351		    u32 portid, int nonblock)
1352{
1353	struct sock *sk;
1354	int err;
1355	long timeo;
1356
1357	skb = netlink_trim(skb, gfp_any());
1358
1359	timeo = sock_sndtimeo(ssk, nonblock);
1360retry:
1361	sk = netlink_getsockbyportid(ssk, portid);
1362	if (IS_ERR(sk)) {
1363		kfree_skb(skb);
1364		return PTR_ERR(sk);
1365	}
1366	if (netlink_is_kernel(sk))
1367		return netlink_unicast_kernel(sk, skb, ssk);
1368
1369	if (sk_filter(sk, skb)) {
1370		err = skb->len;
1371		kfree_skb(skb);
1372		sock_put(sk);
1373		return err;
1374	}
1375
1376	err = netlink_attachskb(sk, skb, &timeo, ssk);
1377	if (err == 1)
1378		goto retry;
1379	if (err)
1380		return err;
1381
1382	return netlink_sendskb(sk, skb);
1383}
1384EXPORT_SYMBOL(netlink_unicast);
1385
1386int netlink_has_listeners(struct sock *sk, unsigned int group)
1387{
1388	int res = 0;
1389	struct listeners *listeners;
1390
1391	BUG_ON(!netlink_is_kernel(sk));
1392
1393	rcu_read_lock();
1394	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1395
1396	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1397		res = test_bit(group - 1, listeners->masks);
1398
1399	rcu_read_unlock();
1400
1401	return res;
1402}
1403EXPORT_SYMBOL_GPL(netlink_has_listeners);
1404
1405bool netlink_strict_get_check(struct sk_buff *skb)
1406{
1407	return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1408}
1409EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1410
1411static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1412{
1413	struct netlink_sock *nlk = nlk_sk(sk);
1414
1415	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1416	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1417		netlink_skb_set_owner_r(skb, sk);
1418		__netlink_sendskb(sk, skb);
1419		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1420	}
1421	return -1;
1422}
1423
1424struct netlink_broadcast_data {
1425	struct sock *exclude_sk;
1426	struct net *net;
1427	u32 portid;
1428	u32 group;
1429	int failure;
1430	int delivery_failure;
1431	int congested;
1432	int delivered;
1433	gfp_t allocation;
1434	struct sk_buff *skb, *skb2;
1435	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1436	void *tx_data;
1437};
1438
1439static void do_one_broadcast(struct sock *sk,
1440				    struct netlink_broadcast_data *p)
1441{
1442	struct netlink_sock *nlk = nlk_sk(sk);
1443	int val;
1444
1445	if (p->exclude_sk == sk)
1446		return;
1447
1448	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1449	    !test_bit(p->group - 1, nlk->groups))
1450		return;
1451
1452	if (!net_eq(sock_net(sk), p->net)) {
1453		if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1454			return;
1455
1456		if (!peernet_has_id(sock_net(sk), p->net))
1457			return;
1458
1459		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1460				     CAP_NET_BROADCAST))
1461			return;
1462	}
1463
1464	if (p->failure) {
1465		netlink_overrun(sk);
1466		return;
1467	}
1468
1469	sock_hold(sk);
1470	if (p->skb2 == NULL) {
1471		if (skb_shared(p->skb)) {
1472			p->skb2 = skb_clone(p->skb, p->allocation);
1473		} else {
1474			p->skb2 = skb_get(p->skb);
1475			/*
1476			 * skb ownership may have been set when
1477			 * delivered to a previous socket.
1478			 */
1479			skb_orphan(p->skb2);
1480		}
1481	}
1482	if (p->skb2 == NULL) {
1483		netlink_overrun(sk);
1484		/* Clone failed. Notify ALL listeners. */
1485		p->failure = 1;
1486		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1487			p->delivery_failure = 1;
1488		goto out;
1489	}
1490
1491	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1492		kfree_skb(p->skb2);
1493		p->skb2 = NULL;
1494		goto out;
1495	}
1496
1497	if (sk_filter(sk, p->skb2)) {
1498		kfree_skb(p->skb2);
1499		p->skb2 = NULL;
1500		goto out;
1501	}
1502	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1503	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1504		NETLINK_CB(p->skb2).nsid_is_set = true;
1505	val = netlink_broadcast_deliver(sk, p->skb2);
1506	if (val < 0) {
1507		netlink_overrun(sk);
1508		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1509			p->delivery_failure = 1;
1510	} else {
1511		p->congested |= val;
1512		p->delivered = 1;
1513		p->skb2 = NULL;
1514	}
1515out:
1516	sock_put(sk);
1517}
1518
1519int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1520			       u32 portid,
1521			       u32 group, gfp_t allocation,
1522			       netlink_filter_fn filter,
1523			       void *filter_data)
1524{
1525	struct net *net = sock_net(ssk);
1526	struct netlink_broadcast_data info;
1527	struct sock *sk;
1528
1529	skb = netlink_trim(skb, allocation);
1530
1531	info.exclude_sk = ssk;
1532	info.net = net;
1533	info.portid = portid;
1534	info.group = group;
1535	info.failure = 0;
1536	info.delivery_failure = 0;
1537	info.congested = 0;
1538	info.delivered = 0;
1539	info.allocation = allocation;
1540	info.skb = skb;
1541	info.skb2 = NULL;
1542	info.tx_filter = filter;
1543	info.tx_data = filter_data;
1544
1545	/* While we sleep in clone, do not allow to change socket list */
1546
1547	netlink_lock_table();
1548
1549	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1550		do_one_broadcast(sk, &info);
1551
1552	consume_skb(skb);
1553
1554	netlink_unlock_table();
1555
1556	if (info.delivery_failure) {
1557		kfree_skb(info.skb2);
1558		return -ENOBUFS;
1559	}
1560	consume_skb(info.skb2);
1561
1562	if (info.delivered) {
1563		if (info.congested && gfpflags_allow_blocking(allocation))
1564			yield();
1565		return 0;
1566	}
1567	return -ESRCH;
1568}
1569EXPORT_SYMBOL(netlink_broadcast_filtered);
1570
1571int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1572		      u32 group, gfp_t allocation)
1573{
1574	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1575					  NULL, NULL);
1576}
1577EXPORT_SYMBOL(netlink_broadcast);
1578
1579struct netlink_set_err_data {
1580	struct sock *exclude_sk;
1581	u32 portid;
1582	u32 group;
1583	int code;
1584};
1585
1586static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1587{
1588	struct netlink_sock *nlk = nlk_sk(sk);
1589	int ret = 0;
1590
1591	if (sk == p->exclude_sk)
1592		goto out;
1593
1594	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1595		goto out;
1596
1597	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1598	    !test_bit(p->group - 1, nlk->groups))
1599		goto out;
1600
1601	if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1602		ret = 1;
1603		goto out;
1604	}
1605
1606	WRITE_ONCE(sk->sk_err, p->code);
1607	sk_error_report(sk);
1608out:
1609	return ret;
1610}
1611
1612/**
1613 * netlink_set_err - report error to broadcast listeners
1614 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1615 * @portid: the PORTID of a process that we want to skip (if any)
1616 * @group: the broadcast group that will notice the error
1617 * @code: error code, must be negative (as usual in kernelspace)
1618 *
1619 * This function returns the number of broadcast listeners that have set the
1620 * NETLINK_NO_ENOBUFS socket option.
1621 */
1622int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1623{
1624	struct netlink_set_err_data info;
1625	unsigned long flags;
1626	struct sock *sk;
1627	int ret = 0;
1628
1629	info.exclude_sk = ssk;
1630	info.portid = portid;
1631	info.group = group;
1632	/* sk->sk_err wants a positive error value */
1633	info.code = -code;
1634
1635	read_lock_irqsave(&nl_table_lock, flags);
1636
1637	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1638		ret += do_one_set_err(sk, &info);
1639
1640	read_unlock_irqrestore(&nl_table_lock, flags);
1641	return ret;
1642}
1643EXPORT_SYMBOL(netlink_set_err);
1644
1645/* must be called with netlink table grabbed */
1646static void netlink_update_socket_mc(struct netlink_sock *nlk,
1647				     unsigned int group,
1648				     int is_new)
1649{
1650	int old, new = !!is_new, subscriptions;
1651
1652	old = test_bit(group - 1, nlk->groups);
1653	subscriptions = nlk->subscriptions - old + new;
1654	__assign_bit(group - 1, nlk->groups, new);
1655	netlink_update_subscriptions(&nlk->sk, subscriptions);
1656	netlink_update_listeners(&nlk->sk);
1657}
1658
1659static int netlink_setsockopt(struct socket *sock, int level, int optname,
1660			      sockptr_t optval, unsigned int optlen)
1661{
1662	struct sock *sk = sock->sk;
1663	struct netlink_sock *nlk = nlk_sk(sk);
1664	unsigned int val = 0;
1665	int nr = -1;
1666
1667	if (level != SOL_NETLINK)
1668		return -ENOPROTOOPT;
1669
1670	if (optlen >= sizeof(int) &&
1671	    copy_from_sockptr(&val, optval, sizeof(val)))
1672		return -EFAULT;
1673
1674	switch (optname) {
1675	case NETLINK_PKTINFO:
1676		nr = NETLINK_F_RECV_PKTINFO;
1677		break;
1678	case NETLINK_ADD_MEMBERSHIP:
1679	case NETLINK_DROP_MEMBERSHIP: {
1680		int err;
1681
1682		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1683			return -EPERM;
1684		err = netlink_realloc_groups(sk);
1685		if (err)
1686			return err;
1687		if (!val || val - 1 >= nlk->ngroups)
1688			return -EINVAL;
1689		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1690			err = nlk->netlink_bind(sock_net(sk), val);
1691			if (err)
1692				return err;
1693		}
1694		netlink_table_grab();
1695		netlink_update_socket_mc(nlk, val,
1696					 optname == NETLINK_ADD_MEMBERSHIP);
1697		netlink_table_ungrab();
1698		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1699			nlk->netlink_unbind(sock_net(sk), val);
1700
1701		break;
1702	}
1703	case NETLINK_BROADCAST_ERROR:
1704		nr = NETLINK_F_BROADCAST_SEND_ERROR;
1705		break;
1706	case NETLINK_NO_ENOBUFS:
1707		assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1708		if (val) {
1709			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1710			wake_up_interruptible(&nlk->wait);
1711		}
1712		break;
1713	case NETLINK_LISTEN_ALL_NSID:
1714		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1715			return -EPERM;
1716		nr = NETLINK_F_LISTEN_ALL_NSID;
1717		break;
1718	case NETLINK_CAP_ACK:
1719		nr = NETLINK_F_CAP_ACK;
1720		break;
1721	case NETLINK_EXT_ACK:
1722		nr = NETLINK_F_EXT_ACK;
1723		break;
1724	case NETLINK_GET_STRICT_CHK:
1725		nr = NETLINK_F_STRICT_CHK;
1726		break;
1727	default:
1728		return -ENOPROTOOPT;
1729	}
1730	if (nr >= 0)
1731		assign_bit(nr, &nlk->flags, val);
1732	return 0;
1733}
1734
1735static int netlink_getsockopt(struct socket *sock, int level, int optname,
1736			      char __user *optval, int __user *optlen)
1737{
1738	struct sock *sk = sock->sk;
1739	struct netlink_sock *nlk = nlk_sk(sk);
1740	unsigned int flag;
1741	int len, val;
1742
1743	if (level != SOL_NETLINK)
1744		return -ENOPROTOOPT;
1745
1746	if (get_user(len, optlen))
1747		return -EFAULT;
1748	if (len < 0)
1749		return -EINVAL;
1750
1751	switch (optname) {
1752	case NETLINK_PKTINFO:
1753		flag = NETLINK_F_RECV_PKTINFO;
1754		break;
1755	case NETLINK_BROADCAST_ERROR:
1756		flag = NETLINK_F_BROADCAST_SEND_ERROR;
1757		break;
1758	case NETLINK_NO_ENOBUFS:
1759		flag = NETLINK_F_RECV_NO_ENOBUFS;
1760		break;
1761	case NETLINK_LIST_MEMBERSHIPS: {
1762		int pos, idx, shift, err = 0;
1763
1764		netlink_lock_table();
1765		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1766			if (len - pos < sizeof(u32))
1767				break;
1768
1769			idx = pos / sizeof(unsigned long);
1770			shift = (pos % sizeof(unsigned long)) * 8;
1771			if (put_user((u32)(nlk->groups[idx] >> shift),
1772				     (u32 __user *)(optval + pos))) {
1773				err = -EFAULT;
1774				break;
1775			}
1776		}
1777		if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1778			err = -EFAULT;
1779		netlink_unlock_table();
1780		return err;
1781	}
 
 
 
1782	case NETLINK_CAP_ACK:
1783		flag = NETLINK_F_CAP_ACK;
1784		break;
1785	case NETLINK_EXT_ACK:
1786		flag = NETLINK_F_EXT_ACK;
1787		break;
1788	case NETLINK_GET_STRICT_CHK:
1789		flag = NETLINK_F_STRICT_CHK;
1790		break;
1791	default:
1792		return -ENOPROTOOPT;
1793	}
1794
1795	if (len < sizeof(int))
1796		return -EINVAL;
1797
1798	len = sizeof(int);
1799	val = test_bit(flag, &nlk->flags);
1800
1801	if (put_user(len, optlen) ||
1802	    copy_to_user(optval, &val, len))
1803		return -EFAULT;
1804
1805	return 0;
1806}
1807
1808static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1809{
1810	struct nl_pktinfo info;
1811
1812	info.group = NETLINK_CB(skb).dst_group;
1813	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1814}
1815
1816static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1817					 struct sk_buff *skb)
1818{
1819	if (!NETLINK_CB(skb).nsid_is_set)
1820		return;
1821
1822	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1823		 &NETLINK_CB(skb).nsid);
1824}
1825
1826static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1827{
1828	struct sock *sk = sock->sk;
1829	struct netlink_sock *nlk = nlk_sk(sk);
1830	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1831	u32 dst_portid;
1832	u32 dst_group;
1833	struct sk_buff *skb;
1834	int err;
1835	struct scm_cookie scm;
1836	u32 netlink_skb_flags = 0;
1837
1838	if (msg->msg_flags & MSG_OOB)
1839		return -EOPNOTSUPP;
1840
1841	if (len == 0) {
1842		pr_warn_once("Zero length message leads to an empty skb\n");
1843		return -ENODATA;
1844	}
1845
1846	err = scm_send(sock, msg, &scm, true);
1847	if (err < 0)
1848		return err;
1849
1850	if (msg->msg_namelen) {
1851		err = -EINVAL;
1852		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1853			goto out;
1854		if (addr->nl_family != AF_NETLINK)
1855			goto out;
1856		dst_portid = addr->nl_pid;
1857		dst_group = ffs(addr->nl_groups);
1858		err =  -EPERM;
1859		if ((dst_group || dst_portid) &&
1860		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1861			goto out;
1862		netlink_skb_flags |= NETLINK_SKB_DST;
1863	} else {
1864		/* Paired with WRITE_ONCE() in netlink_connect() */
1865		dst_portid = READ_ONCE(nlk->dst_portid);
1866		dst_group = READ_ONCE(nlk->dst_group);
1867	}
1868
1869	/* Paired with WRITE_ONCE() in netlink_insert() */
1870	if (!READ_ONCE(nlk->bound)) {
1871		err = netlink_autobind(sock);
1872		if (err)
1873			goto out;
1874	} else {
1875		/* Ensure nlk is hashed and visible. */
1876		smp_rmb();
1877	}
1878
1879	err = -EMSGSIZE;
1880	if (len > sk->sk_sndbuf - 32)
1881		goto out;
1882	err = -ENOBUFS;
1883	skb = netlink_alloc_large_skb(len, dst_group);
1884	if (skb == NULL)
1885		goto out;
1886
1887	NETLINK_CB(skb).portid	= nlk->portid;
1888	NETLINK_CB(skb).dst_group = dst_group;
1889	NETLINK_CB(skb).creds	= scm.creds;
1890	NETLINK_CB(skb).flags	= netlink_skb_flags;
1891
1892	err = -EFAULT;
1893	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1894		kfree_skb(skb);
1895		goto out;
1896	}
1897
1898	err = security_netlink_send(sk, skb);
1899	if (err) {
1900		kfree_skb(skb);
1901		goto out;
1902	}
1903
1904	if (dst_group) {
1905		refcount_inc(&skb->users);
1906		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1907	}
1908	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1909
1910out:
1911	scm_destroy(&scm);
1912	return err;
1913}
1914
1915static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1916			   int flags)
1917{
1918	struct scm_cookie scm;
1919	struct sock *sk = sock->sk;
1920	struct netlink_sock *nlk = nlk_sk(sk);
1921	size_t copied, max_recvmsg_len;
1922	struct sk_buff *skb, *data_skb;
1923	int err, ret;
1924
1925	if (flags & MSG_OOB)
1926		return -EOPNOTSUPP;
1927
1928	copied = 0;
1929
1930	skb = skb_recv_datagram(sk, flags, &err);
1931	if (skb == NULL)
1932		goto out;
1933
1934	data_skb = skb;
1935
1936#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1937	if (unlikely(skb_shinfo(skb)->frag_list)) {
1938		/*
1939		 * If this skb has a frag_list, then here that means that we
1940		 * will have to use the frag_list skb's data for compat tasks
1941		 * and the regular skb's data for normal (non-compat) tasks.
1942		 *
1943		 * If we need to send the compat skb, assign it to the
1944		 * 'data_skb' variable so that it will be used below for data
1945		 * copying. We keep 'skb' for everything else, including
1946		 * freeing both later.
1947		 */
1948		if (flags & MSG_CMSG_COMPAT)
1949			data_skb = skb_shinfo(skb)->frag_list;
1950	}
1951#endif
1952
1953	/* Record the max length of recvmsg() calls for future allocations */
1954	max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1955	max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1956				SKB_WITH_OVERHEAD(32768));
1957	WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1958
1959	copied = data_skb->len;
1960	if (len < copied) {
1961		msg->msg_flags |= MSG_TRUNC;
1962		copied = len;
1963	}
1964
1965	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1966
1967	if (msg->msg_name) {
1968		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1969		addr->nl_family = AF_NETLINK;
1970		addr->nl_pad    = 0;
1971		addr->nl_pid	= NETLINK_CB(skb).portid;
1972		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1973		msg->msg_namelen = sizeof(*addr);
1974	}
1975
1976	if (nlk_test_bit(RECV_PKTINFO, sk))
1977		netlink_cmsg_recv_pktinfo(msg, skb);
1978	if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1979		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1980
1981	memset(&scm, 0, sizeof(scm));
1982	scm.creds = *NETLINK_CREDS(skb);
1983	if (flags & MSG_TRUNC)
1984		copied = data_skb->len;
1985
1986	skb_free_datagram(sk, skb);
1987
1988	if (READ_ONCE(nlk->cb_running) &&
1989	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1990		ret = netlink_dump(sk);
1991		if (ret) {
1992			WRITE_ONCE(sk->sk_err, -ret);
1993			sk_error_report(sk);
1994		}
1995	}
1996
1997	scm_recv(sock, msg, &scm, flags);
1998out:
1999	netlink_rcv_wake(sk);
2000	return err ? : copied;
2001}
2002
2003static void netlink_data_ready(struct sock *sk)
2004{
2005	BUG();
2006}
2007
2008/*
2009 *	We export these functions to other modules. They provide a
2010 *	complete set of kernel non-blocking support for message
2011 *	queueing.
2012 */
2013
2014struct sock *
2015__netlink_kernel_create(struct net *net, int unit, struct module *module,
2016			struct netlink_kernel_cfg *cfg)
2017{
2018	struct socket *sock;
2019	struct sock *sk;
2020	struct netlink_sock *nlk;
2021	struct listeners *listeners = NULL;
2022	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2023	unsigned int groups;
2024
2025	BUG_ON(!nl_table);
2026
2027	if (unit < 0 || unit >= MAX_LINKS)
2028		return NULL;
2029
2030	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2031		return NULL;
2032
2033	if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2034		goto out_sock_release_nosk;
2035
2036	sk = sock->sk;
2037
2038	if (!cfg || cfg->groups < 32)
2039		groups = 32;
2040	else
2041		groups = cfg->groups;
2042
2043	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2044	if (!listeners)
2045		goto out_sock_release;
2046
2047	sk->sk_data_ready = netlink_data_ready;
2048	if (cfg && cfg->input)
2049		nlk_sk(sk)->netlink_rcv = cfg->input;
2050
2051	if (netlink_insert(sk, 0))
2052		goto out_sock_release;
2053
2054	nlk = nlk_sk(sk);
2055	set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2056
2057	netlink_table_grab();
2058	if (!nl_table[unit].registered) {
2059		nl_table[unit].groups = groups;
2060		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2061		nl_table[unit].cb_mutex = cb_mutex;
2062		nl_table[unit].module = module;
2063		if (cfg) {
2064			nl_table[unit].bind = cfg->bind;
2065			nl_table[unit].unbind = cfg->unbind;
2066			nl_table[unit].release = cfg->release;
2067			nl_table[unit].flags = cfg->flags;
2068		}
2069		nl_table[unit].registered = 1;
2070	} else {
2071		kfree(listeners);
2072		nl_table[unit].registered++;
2073	}
2074	netlink_table_ungrab();
2075	return sk;
2076
2077out_sock_release:
2078	kfree(listeners);
2079	netlink_kernel_release(sk);
2080	return NULL;
2081
2082out_sock_release_nosk:
2083	sock_release(sock);
2084	return NULL;
2085}
2086EXPORT_SYMBOL(__netlink_kernel_create);
2087
2088void
2089netlink_kernel_release(struct sock *sk)
2090{
2091	if (sk == NULL || sk->sk_socket == NULL)
2092		return;
2093
2094	sock_release(sk->sk_socket);
2095}
2096EXPORT_SYMBOL(netlink_kernel_release);
2097
2098int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2099{
2100	struct listeners *new, *old;
2101	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2102
2103	if (groups < 32)
2104		groups = 32;
2105
2106	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2107		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2108		if (!new)
2109			return -ENOMEM;
2110		old = nl_deref_protected(tbl->listeners);
2111		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2112		rcu_assign_pointer(tbl->listeners, new);
2113
2114		kfree_rcu(old, rcu);
2115	}
2116	tbl->groups = groups;
2117
2118	return 0;
2119}
2120
2121/**
2122 * netlink_change_ngroups - change number of multicast groups
2123 *
2124 * This changes the number of multicast groups that are available
2125 * on a certain netlink family. Note that it is not possible to
2126 * change the number of groups to below 32. Also note that it does
2127 * not implicitly call netlink_clear_multicast_users() when the
2128 * number of groups is reduced.
2129 *
2130 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2131 * @groups: The new number of groups.
2132 */
2133int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2134{
2135	int err;
2136
2137	netlink_table_grab();
2138	err = __netlink_change_ngroups(sk, groups);
2139	netlink_table_ungrab();
2140
2141	return err;
2142}
2143
2144void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2145{
2146	struct sock *sk;
2147	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2148
2149	sk_for_each_bound(sk, &tbl->mc_list)
2150		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2151}
2152
2153struct nlmsghdr *
2154__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2155{
2156	struct nlmsghdr *nlh;
2157	int size = nlmsg_msg_size(len);
2158
2159	nlh = skb_put(skb, NLMSG_ALIGN(size));
2160	nlh->nlmsg_type = type;
2161	nlh->nlmsg_len = size;
2162	nlh->nlmsg_flags = flags;
2163	nlh->nlmsg_pid = portid;
2164	nlh->nlmsg_seq = seq;
2165	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2166		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2167	return nlh;
2168}
2169EXPORT_SYMBOL(__nlmsg_put);
2170
2171/*
2172 * It looks a bit ugly.
2173 * It would be better to create kernel thread.
2174 */
2175
2176static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2177			     struct netlink_callback *cb,
2178			     struct netlink_ext_ack *extack)
2179{
2180	struct nlmsghdr *nlh;
2181
2182	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2183			       NLM_F_MULTI | cb->answer_flags);
2184	if (WARN_ON(!nlh))
2185		return -ENOBUFS;
2186
2187	nl_dump_check_consistent(cb, nlh);
2188	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2189
2190	if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2191		nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2192		if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2193			nlmsg_end(skb, nlh);
2194	}
2195
2196	return 0;
2197}
2198
2199static int netlink_dump(struct sock *sk)
2200{
2201	struct netlink_sock *nlk = nlk_sk(sk);
2202	struct netlink_ext_ack extack = {};
2203	struct netlink_callback *cb;
2204	struct sk_buff *skb = NULL;
2205	size_t max_recvmsg_len;
2206	struct module *module;
2207	int err = -ENOBUFS;
2208	int alloc_min_size;
2209	int alloc_size;
2210
2211	mutex_lock(nlk->cb_mutex);
 
2212	if (!nlk->cb_running) {
2213		err = -EINVAL;
2214		goto errout_skb;
2215	}
2216
2217	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2218		goto errout_skb;
2219
2220	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2221	 * required, but it makes sense to _attempt_ a 16K bytes allocation
2222	 * to reduce number of system calls on dump operations, if user
2223	 * ever provided a big enough buffer.
2224	 */
2225	cb = &nlk->cb;
2226	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2227
2228	max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2229	if (alloc_min_size < max_recvmsg_len) {
2230		alloc_size = max_recvmsg_len;
2231		skb = alloc_skb(alloc_size,
2232				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2233				__GFP_NOWARN | __GFP_NORETRY);
2234	}
2235	if (!skb) {
2236		alloc_size = alloc_min_size;
2237		skb = alloc_skb(alloc_size, GFP_KERNEL);
2238	}
2239	if (!skb)
2240		goto errout_skb;
2241
2242	/* Trim skb to allocated size. User is expected to provide buffer as
2243	 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2244	 * netlink_recvmsg())). dump will pack as many smaller messages as
2245	 * could fit within the allocated skb. skb is typically allocated
2246	 * with larger space than required (could be as much as near 2x the
2247	 * requested size with align to next power of 2 approach). Allowing
2248	 * dump to use the excess space makes it difficult for a user to have a
2249	 * reasonable static buffer based on the expected largest dump of a
2250	 * single netdev. The outcome is MSG_TRUNC error.
2251	 */
2252	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2253
2254	/* Make sure malicious BPF programs can not read unitialized memory
2255	 * from skb->head -> skb->data
2256	 */
2257	skb_reset_network_header(skb);
2258	skb_reset_mac_header(skb);
2259
2260	netlink_skb_set_owner_r(skb, sk);
2261
2262	if (nlk->dump_done_errno > 0) {
 
 
2263		cb->extack = &extack;
 
 
 
 
 
2264		nlk->dump_done_errno = cb->dump(skb, cb);
 
 
 
 
 
 
 
 
 
 
 
 
2265		cb->extack = NULL;
2266	}
2267
2268	if (nlk->dump_done_errno > 0 ||
2269	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2270		mutex_unlock(nlk->cb_mutex);
2271
2272		if (sk_filter(sk, skb))
2273			kfree_skb(skb);
2274		else
2275			__netlink_sendskb(sk, skb);
2276		return 0;
2277	}
2278
2279	if (netlink_dump_done(nlk, skb, cb, &extack))
2280		goto errout_skb;
2281
2282#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2283	/* frag_list skb's data is used for compat tasks
2284	 * and the regular skb's data for normal (non-compat) tasks.
2285	 * See netlink_recvmsg().
2286	 */
2287	if (unlikely(skb_shinfo(skb)->frag_list)) {
2288		if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2289			goto errout_skb;
2290	}
2291#endif
2292
2293	if (sk_filter(sk, skb))
2294		kfree_skb(skb);
2295	else
2296		__netlink_sendskb(sk, skb);
2297
2298	if (cb->done)
2299		cb->done(cb);
2300
2301	WRITE_ONCE(nlk->cb_running, false);
2302	module = cb->module;
2303	skb = cb->skb;
2304	mutex_unlock(nlk->cb_mutex);
2305	module_put(module);
2306	consume_skb(skb);
2307	return 0;
2308
2309errout_skb:
2310	mutex_unlock(nlk->cb_mutex);
2311	kfree_skb(skb);
2312	return err;
2313}
2314
2315int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2316			 const struct nlmsghdr *nlh,
2317			 struct netlink_dump_control *control)
2318{
2319	struct netlink_callback *cb;
2320	struct netlink_sock *nlk;
2321	struct sock *sk;
2322	int ret;
2323
2324	refcount_inc(&skb->users);
2325
2326	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2327	if (sk == NULL) {
2328		ret = -ECONNREFUSED;
2329		goto error_free;
2330	}
2331
2332	nlk = nlk_sk(sk);
2333	mutex_lock(nlk->cb_mutex);
2334	/* A dump is in progress... */
2335	if (nlk->cb_running) {
2336		ret = -EBUSY;
2337		goto error_unlock;
2338	}
2339	/* add reference of module which cb->dump belongs to */
2340	if (!try_module_get(control->module)) {
2341		ret = -EPROTONOSUPPORT;
2342		goto error_unlock;
2343	}
2344
2345	cb = &nlk->cb;
2346	memset(cb, 0, sizeof(*cb));
2347	cb->dump = control->dump;
2348	cb->done = control->done;
2349	cb->nlh = nlh;
2350	cb->data = control->data;
2351	cb->module = control->module;
2352	cb->min_dump_alloc = control->min_dump_alloc;
 
2353	cb->skb = skb;
2354
2355	cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2356
2357	if (control->start) {
2358		cb->extack = control->extack;
2359		ret = control->start(cb);
2360		cb->extack = NULL;
2361		if (ret)
2362			goto error_put;
2363	}
2364
2365	WRITE_ONCE(nlk->cb_running, true);
2366	nlk->dump_done_errno = INT_MAX;
2367
2368	mutex_unlock(nlk->cb_mutex);
2369
2370	ret = netlink_dump(sk);
2371
2372	sock_put(sk);
2373
2374	if (ret)
2375		return ret;
2376
2377	/* We successfully started a dump, by returning -EINTR we
2378	 * signal not to send ACK even if it was requested.
2379	 */
2380	return -EINTR;
2381
2382error_put:
2383	module_put(control->module);
2384error_unlock:
2385	sock_put(sk);
2386	mutex_unlock(nlk->cb_mutex);
2387error_free:
2388	kfree_skb(skb);
2389	return ret;
2390}
2391EXPORT_SYMBOL(__netlink_dump_start);
2392
2393static size_t
2394netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2395		    const struct netlink_ext_ack *extack)
2396{
2397	size_t tlvlen;
2398
2399	if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2400		return 0;
2401
2402	tlvlen = 0;
2403	if (extack->_msg)
2404		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2405	if (extack->cookie_len)
2406		tlvlen += nla_total_size(extack->cookie_len);
2407
2408	/* Following attributes are only reported as error (not warning) */
2409	if (!err)
2410		return tlvlen;
2411
2412	if (extack->bad_attr)
2413		tlvlen += nla_total_size(sizeof(u32));
2414	if (extack->policy)
2415		tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2416	if (extack->miss_type)
2417		tlvlen += nla_total_size(sizeof(u32));
2418	if (extack->miss_nest)
2419		tlvlen += nla_total_size(sizeof(u32));
2420
2421	return tlvlen;
2422}
2423
2424static void
2425netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2426		     struct nlmsghdr *nlh, int err,
2427		     const struct netlink_ext_ack *extack)
2428{
2429	if (extack->_msg)
2430		WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2431	if (extack->cookie_len)
2432		WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2433				extack->cookie_len, extack->cookie));
2434
2435	if (!err)
2436		return;
2437
2438	if (extack->bad_attr &&
2439	    !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2440		     (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2441		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2442				    (u8 *)extack->bad_attr - (u8 *)nlh));
2443	if (extack->policy)
2444		netlink_policy_dump_write_attr(skb, extack->policy,
2445					       NLMSGERR_ATTR_POLICY);
2446	if (extack->miss_type)
2447		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2448				    extack->miss_type));
2449	if (extack->miss_nest &&
2450	    !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2451		     (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2452		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2453				    (u8 *)extack->miss_nest - (u8 *)nlh));
2454}
2455
2456void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2457		 const struct netlink_ext_ack *extack)
2458{
2459	struct sk_buff *skb;
2460	struct nlmsghdr *rep;
2461	struct nlmsgerr *errmsg;
2462	size_t payload = sizeof(*errmsg);
2463	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2464	unsigned int flags = 0;
2465	size_t tlvlen;
2466
2467	/* Error messages get the original request appened, unless the user
2468	 * requests to cap the error message, and get extra error data if
2469	 * requested.
2470	 */
2471	if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2472		payload += nlmsg_len(nlh);
2473	else
2474		flags |= NLM_F_CAPPED;
2475
2476	tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2477	if (tlvlen)
2478		flags |= NLM_F_ACK_TLVS;
2479
2480	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2481	if (!skb)
2482		goto err_skb;
2483
2484	rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2485			NLMSG_ERROR, sizeof(*errmsg), flags);
2486	if (!rep)
2487		goto err_bad_put;
2488	errmsg = nlmsg_data(rep);
2489	errmsg->error = err;
2490	errmsg->msg = *nlh;
2491
2492	if (!(flags & NLM_F_CAPPED)) {
2493		if (!nlmsg_append(skb, nlmsg_len(nlh)))
2494			goto err_bad_put;
2495
2496		memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2497		       nlmsg_len(nlh));
2498	}
2499
2500	if (tlvlen)
2501		netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2502
2503	nlmsg_end(skb, rep);
2504
2505	nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2506
2507	return;
2508
2509err_bad_put:
2510	nlmsg_free(skb);
2511err_skb:
2512	WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2513	sk_error_report(NETLINK_CB(in_skb).sk);
2514}
2515EXPORT_SYMBOL(netlink_ack);
2516
2517int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2518						   struct nlmsghdr *,
2519						   struct netlink_ext_ack *))
2520{
2521	struct netlink_ext_ack extack;
2522	struct nlmsghdr *nlh;
2523	int err;
2524
2525	while (skb->len >= nlmsg_total_size(0)) {
2526		int msglen;
2527
2528		memset(&extack, 0, sizeof(extack));
2529		nlh = nlmsg_hdr(skb);
2530		err = 0;
2531
2532		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2533			return 0;
2534
2535		/* Only requests are handled by the kernel */
2536		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2537			goto ack;
2538
2539		/* Skip control messages */
2540		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2541			goto ack;
2542
2543		err = cb(skb, nlh, &extack);
2544		if (err == -EINTR)
2545			goto skip;
2546
2547ack:
2548		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2549			netlink_ack(skb, nlh, err, &extack);
2550
2551skip:
2552		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2553		if (msglen > skb->len)
2554			msglen = skb->len;
2555		skb_pull(skb, msglen);
2556	}
2557
2558	return 0;
2559}
2560EXPORT_SYMBOL(netlink_rcv_skb);
2561
2562/**
2563 * nlmsg_notify - send a notification netlink message
2564 * @sk: netlink socket to use
2565 * @skb: notification message
2566 * @portid: destination netlink portid for reports or 0
2567 * @group: destination multicast group or 0
2568 * @report: 1 to report back, 0 to disable
2569 * @flags: allocation flags
2570 */
2571int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2572		 unsigned int group, int report, gfp_t flags)
2573{
2574	int err = 0;
2575
2576	if (group) {
2577		int exclude_portid = 0;
2578
2579		if (report) {
2580			refcount_inc(&skb->users);
2581			exclude_portid = portid;
2582		}
2583
2584		/* errors reported via destination sk->sk_err, but propagate
2585		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2586		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2587		if (err == -ESRCH)
2588			err = 0;
2589	}
2590
2591	if (report) {
2592		int err2;
2593
2594		err2 = nlmsg_unicast(sk, skb, portid);
2595		if (!err)
2596			err = err2;
2597	}
2598
2599	return err;
2600}
2601EXPORT_SYMBOL(nlmsg_notify);
2602
2603#ifdef CONFIG_PROC_FS
2604struct nl_seq_iter {
2605	struct seq_net_private p;
2606	struct rhashtable_iter hti;
2607	int link;
2608};
2609
2610static void netlink_walk_start(struct nl_seq_iter *iter)
2611{
2612	rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2613	rhashtable_walk_start(&iter->hti);
2614}
2615
2616static void netlink_walk_stop(struct nl_seq_iter *iter)
2617{
2618	rhashtable_walk_stop(&iter->hti);
2619	rhashtable_walk_exit(&iter->hti);
2620}
2621
2622static void *__netlink_seq_next(struct seq_file *seq)
2623{
2624	struct nl_seq_iter *iter = seq->private;
2625	struct netlink_sock *nlk;
2626
2627	do {
2628		for (;;) {
2629			nlk = rhashtable_walk_next(&iter->hti);
2630
2631			if (IS_ERR(nlk)) {
2632				if (PTR_ERR(nlk) == -EAGAIN)
2633					continue;
2634
2635				return nlk;
2636			}
2637
2638			if (nlk)
2639				break;
2640
2641			netlink_walk_stop(iter);
2642			if (++iter->link >= MAX_LINKS)
2643				return NULL;
2644
2645			netlink_walk_start(iter);
2646		}
2647	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2648
2649	return nlk;
2650}
2651
2652static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2653	__acquires(RCU)
2654{
2655	struct nl_seq_iter *iter = seq->private;
2656	void *obj = SEQ_START_TOKEN;
2657	loff_t pos;
2658
2659	iter->link = 0;
2660
2661	netlink_walk_start(iter);
2662
2663	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2664		obj = __netlink_seq_next(seq);
2665
2666	return obj;
2667}
2668
2669static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2670{
2671	++*pos;
2672	return __netlink_seq_next(seq);
2673}
2674
2675static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2676{
2677	struct nl_seq_iter *iter = seq->private;
2678
2679	if (iter->link >= MAX_LINKS)
2680		return;
2681
2682	netlink_walk_stop(iter);
2683}
2684
2685
2686static int netlink_native_seq_show(struct seq_file *seq, void *v)
2687{
2688	if (v == SEQ_START_TOKEN) {
2689		seq_puts(seq,
2690			 "sk               Eth Pid        Groups   "
2691			 "Rmem     Wmem     Dump  Locks    Drops    Inode\n");
2692	} else {
2693		struct sock *s = v;
2694		struct netlink_sock *nlk = nlk_sk(s);
2695
2696		seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2697			   s,
2698			   s->sk_protocol,
2699			   nlk->portid,
2700			   nlk->groups ? (u32)nlk->groups[0] : 0,
2701			   sk_rmem_alloc_get(s),
2702			   sk_wmem_alloc_get(s),
2703			   READ_ONCE(nlk->cb_running),
2704			   refcount_read(&s->sk_refcnt),
2705			   atomic_read(&s->sk_drops),
2706			   sock_i_ino(s)
2707			);
2708
2709	}
2710	return 0;
2711}
2712
2713#ifdef CONFIG_BPF_SYSCALL
2714struct bpf_iter__netlink {
2715	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2716	__bpf_md_ptr(struct netlink_sock *, sk);
2717};
2718
2719DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2720
2721static int netlink_prog_seq_show(struct bpf_prog *prog,
2722				  struct bpf_iter_meta *meta,
2723				  void *v)
2724{
2725	struct bpf_iter__netlink ctx;
2726
2727	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2728	ctx.meta = meta;
2729	ctx.sk = nlk_sk((struct sock *)v);
2730	return bpf_iter_run_prog(prog, &ctx);
2731}
2732
2733static int netlink_seq_show(struct seq_file *seq, void *v)
2734{
2735	struct bpf_iter_meta meta;
2736	struct bpf_prog *prog;
2737
2738	meta.seq = seq;
2739	prog = bpf_iter_get_info(&meta, false);
2740	if (!prog)
2741		return netlink_native_seq_show(seq, v);
2742
2743	if (v != SEQ_START_TOKEN)
2744		return netlink_prog_seq_show(prog, &meta, v);
2745
2746	return 0;
2747}
2748
2749static void netlink_seq_stop(struct seq_file *seq, void *v)
2750{
2751	struct bpf_iter_meta meta;
2752	struct bpf_prog *prog;
2753
2754	if (!v) {
2755		meta.seq = seq;
2756		prog = bpf_iter_get_info(&meta, true);
2757		if (prog)
2758			(void)netlink_prog_seq_show(prog, &meta, v);
2759	}
2760
2761	netlink_native_seq_stop(seq, v);
2762}
2763#else
2764static int netlink_seq_show(struct seq_file *seq, void *v)
2765{
2766	return netlink_native_seq_show(seq, v);
2767}
2768
2769static void netlink_seq_stop(struct seq_file *seq, void *v)
2770{
2771	netlink_native_seq_stop(seq, v);
2772}
2773#endif
2774
2775static const struct seq_operations netlink_seq_ops = {
2776	.start  = netlink_seq_start,
2777	.next   = netlink_seq_next,
2778	.stop   = netlink_seq_stop,
2779	.show   = netlink_seq_show,
2780};
2781#endif
2782
2783int netlink_register_notifier(struct notifier_block *nb)
2784{
2785	return blocking_notifier_chain_register(&netlink_chain, nb);
2786}
2787EXPORT_SYMBOL(netlink_register_notifier);
2788
2789int netlink_unregister_notifier(struct notifier_block *nb)
2790{
2791	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2792}
2793EXPORT_SYMBOL(netlink_unregister_notifier);
2794
2795static const struct proto_ops netlink_ops = {
2796	.family =	PF_NETLINK,
2797	.owner =	THIS_MODULE,
2798	.release =	netlink_release,
2799	.bind =		netlink_bind,
2800	.connect =	netlink_connect,
2801	.socketpair =	sock_no_socketpair,
2802	.accept =	sock_no_accept,
2803	.getname =	netlink_getname,
2804	.poll =		datagram_poll,
2805	.ioctl =	netlink_ioctl,
2806	.listen =	sock_no_listen,
2807	.shutdown =	sock_no_shutdown,
2808	.setsockopt =	netlink_setsockopt,
2809	.getsockopt =	netlink_getsockopt,
2810	.sendmsg =	netlink_sendmsg,
2811	.recvmsg =	netlink_recvmsg,
2812	.mmap =		sock_no_mmap,
2813};
2814
2815static const struct net_proto_family netlink_family_ops = {
2816	.family = PF_NETLINK,
2817	.create = netlink_create,
2818	.owner	= THIS_MODULE,	/* for consistency 8) */
2819};
2820
2821static int __net_init netlink_net_init(struct net *net)
2822{
2823#ifdef CONFIG_PROC_FS
2824	if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2825			sizeof(struct nl_seq_iter)))
2826		return -ENOMEM;
2827#endif
2828	return 0;
2829}
2830
2831static void __net_exit netlink_net_exit(struct net *net)
2832{
2833#ifdef CONFIG_PROC_FS
2834	remove_proc_entry("netlink", net->proc_net);
2835#endif
2836}
2837
2838static void __init netlink_add_usersock_entry(void)
2839{
2840	struct listeners *listeners;
2841	int groups = 32;
2842
2843	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2844	if (!listeners)
2845		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2846
2847	netlink_table_grab();
2848
2849	nl_table[NETLINK_USERSOCK].groups = groups;
2850	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2851	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2852	nl_table[NETLINK_USERSOCK].registered = 1;
2853	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2854
2855	netlink_table_ungrab();
2856}
2857
2858static struct pernet_operations __net_initdata netlink_net_ops = {
2859	.init = netlink_net_init,
2860	.exit = netlink_net_exit,
2861};
2862
2863static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2864{
2865	const struct netlink_sock *nlk = data;
2866	struct netlink_compare_arg arg;
2867
2868	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2869	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2870}
2871
2872static const struct rhashtable_params netlink_rhashtable_params = {
2873	.head_offset = offsetof(struct netlink_sock, node),
2874	.key_len = netlink_compare_arg_len,
2875	.obj_hashfn = netlink_hash,
2876	.obj_cmpfn = netlink_compare,
2877	.automatic_shrinking = true,
2878};
2879
2880#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2881BTF_ID_LIST(btf_netlink_sock_id)
2882BTF_ID(struct, netlink_sock)
2883
2884static const struct bpf_iter_seq_info netlink_seq_info = {
2885	.seq_ops		= &netlink_seq_ops,
2886	.init_seq_private	= bpf_iter_init_seq_net,
2887	.fini_seq_private	= bpf_iter_fini_seq_net,
2888	.seq_priv_size		= sizeof(struct nl_seq_iter),
2889};
2890
2891static struct bpf_iter_reg netlink_reg_info = {
2892	.target			= "netlink",
2893	.ctx_arg_info_size	= 1,
2894	.ctx_arg_info		= {
2895		{ offsetof(struct bpf_iter__netlink, sk),
2896		  PTR_TO_BTF_ID_OR_NULL },
2897	},
2898	.seq_info		= &netlink_seq_info,
2899};
2900
2901static int __init bpf_iter_register(void)
2902{
2903	netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2904	return bpf_iter_reg_target(&netlink_reg_info);
2905}
2906#endif
2907
2908static int __init netlink_proto_init(void)
2909{
2910	int i;
2911	int err = proto_register(&netlink_proto, 0);
2912
2913	if (err != 0)
2914		goto out;
2915
2916#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2917	err = bpf_iter_register();
2918	if (err)
2919		goto out;
2920#endif
2921
2922	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2923
2924	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2925	if (!nl_table)
2926		goto panic;
2927
2928	for (i = 0; i < MAX_LINKS; i++) {
2929		if (rhashtable_init(&nl_table[i].hash,
2930				    &netlink_rhashtable_params) < 0) {
2931			while (--i > 0)
2932				rhashtable_destroy(&nl_table[i].hash);
2933			kfree(nl_table);
2934			goto panic;
2935		}
2936	}
2937
2938	netlink_add_usersock_entry();
2939
2940	sock_register(&netlink_family_ops);
2941	register_pernet_subsys(&netlink_net_ops);
2942	register_pernet_subsys(&netlink_tap_net_ops);
2943	/* The netlink device handler may be needed early. */
2944	rtnetlink_init();
2945out:
2946	return err;
2947panic:
2948	panic("netlink_init: Cannot allocate nl_table\n");
2949}
2950
2951core_initcall(netlink_proto_init);
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * NETLINK      Kernel-user communication protocol.
   4 *
   5 * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
   6 * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
   7 * 				Patrick McHardy <kaber@trash.net>
   8 *
   9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
  10 *                               added netlink_proto_exit
  11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
  12 * 				 use nlk_sk, as sk->protinfo is on a diet 8)
  13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
  14 * 				 - inc module use count of module that owns
  15 * 				   the kernel socket in case userspace opens
  16 * 				   socket of same protocol
  17 * 				 - remove all module support, since netlink is
  18 * 				   mandatory if CONFIG_NET=y these days
  19 */
  20
  21#include <linux/module.h>
  22
  23#include <linux/bpf.h>
  24#include <linux/capability.h>
  25#include <linux/kernel.h>
  26#include <linux/filter.h>
  27#include <linux/init.h>
  28#include <linux/signal.h>
  29#include <linux/sched.h>
  30#include <linux/errno.h>
  31#include <linux/string.h>
  32#include <linux/stat.h>
  33#include <linux/socket.h>
  34#include <linux/un.h>
  35#include <linux/fcntl.h>
  36#include <linux/termios.h>
  37#include <linux/sockios.h>
  38#include <linux/net.h>
  39#include <linux/fs.h>
  40#include <linux/slab.h>
  41#include <linux/uaccess.h>
  42#include <linux/skbuff.h>
  43#include <linux/netdevice.h>
  44#include <linux/rtnetlink.h>
  45#include <linux/proc_fs.h>
  46#include <linux/seq_file.h>
  47#include <linux/notifier.h>
  48#include <linux/security.h>
  49#include <linux/jhash.h>
  50#include <linux/jiffies.h>
  51#include <linux/random.h>
  52#include <linux/bitops.h>
  53#include <linux/mm.h>
  54#include <linux/types.h>
  55#include <linux/audit.h>
  56#include <linux/mutex.h>
  57#include <linux/vmalloc.h>
  58#include <linux/if_arp.h>
  59#include <linux/rhashtable.h>
  60#include <asm/cacheflush.h>
  61#include <linux/hash.h>
  62#include <linux/genetlink.h>
  63#include <linux/net_namespace.h>
  64#include <linux/nospec.h>
  65#include <linux/btf_ids.h>
  66
  67#include <net/net_namespace.h>
  68#include <net/netns/generic.h>
  69#include <net/sock.h>
  70#include <net/scm.h>
  71#include <net/netlink.h>
  72#define CREATE_TRACE_POINTS
  73#include <trace/events/netlink.h>
  74
  75#include "af_netlink.h"
  76
  77struct listeners {
  78	struct rcu_head		rcu;
  79	unsigned long		masks[];
  80};
  81
  82/* state bits */
  83#define NETLINK_S_CONGESTED		0x0
  84
  85static inline int netlink_is_kernel(struct sock *sk)
  86{
  87	return nlk_test_bit(KERNEL_SOCKET, sk);
  88}
  89
  90struct netlink_table *nl_table __read_mostly;
  91EXPORT_SYMBOL_GPL(nl_table);
  92
  93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
  94
  95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
  96
  97static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
  98	"nlk_cb_mutex-ROUTE",
  99	"nlk_cb_mutex-1",
 100	"nlk_cb_mutex-USERSOCK",
 101	"nlk_cb_mutex-FIREWALL",
 102	"nlk_cb_mutex-SOCK_DIAG",
 103	"nlk_cb_mutex-NFLOG",
 104	"nlk_cb_mutex-XFRM",
 105	"nlk_cb_mutex-SELINUX",
 106	"nlk_cb_mutex-ISCSI",
 107	"nlk_cb_mutex-AUDIT",
 108	"nlk_cb_mutex-FIB_LOOKUP",
 109	"nlk_cb_mutex-CONNECTOR",
 110	"nlk_cb_mutex-NETFILTER",
 111	"nlk_cb_mutex-IP6_FW",
 112	"nlk_cb_mutex-DNRTMSG",
 113	"nlk_cb_mutex-KOBJECT_UEVENT",
 114	"nlk_cb_mutex-GENERIC",
 115	"nlk_cb_mutex-17",
 116	"nlk_cb_mutex-SCSITRANSPORT",
 117	"nlk_cb_mutex-ECRYPTFS",
 118	"nlk_cb_mutex-RDMA",
 119	"nlk_cb_mutex-CRYPTO",
 120	"nlk_cb_mutex-SMC",
 121	"nlk_cb_mutex-23",
 122	"nlk_cb_mutex-24",
 123	"nlk_cb_mutex-25",
 124	"nlk_cb_mutex-26",
 125	"nlk_cb_mutex-27",
 126	"nlk_cb_mutex-28",
 127	"nlk_cb_mutex-29",
 128	"nlk_cb_mutex-30",
 129	"nlk_cb_mutex-31",
 130	"nlk_cb_mutex-MAX_LINKS"
 131};
 132
 133static int netlink_dump(struct sock *sk, bool lock_taken);
 134
 135/* nl_table locking explained:
 136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
 137 * and removal are protected with per bucket lock while using RCU list
 138 * modification primitives and may run in parallel to RCU protected lookups.
 139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
 140 * been acquired * either during or after the socket has been removed from
 141 * the list and after an RCU grace period.
 142 */
 143DEFINE_RWLOCK(nl_table_lock);
 144EXPORT_SYMBOL_GPL(nl_table_lock);
 145static atomic_t nl_table_users = ATOMIC_INIT(0);
 146
 147#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
 148
 149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
 150
 151
 152static const struct rhashtable_params netlink_rhashtable_params;
 153
 154void do_trace_netlink_extack(const char *msg)
 155{
 156	trace_netlink_extack(msg);
 157}
 158EXPORT_SYMBOL(do_trace_netlink_extack);
 159
 160static inline u32 netlink_group_mask(u32 group)
 161{
 162	if (group > 32)
 163		return 0;
 164	return group ? 1 << (group - 1) : 0;
 165}
 166
 167static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
 168					   gfp_t gfp_mask)
 169{
 170	unsigned int len = skb->len;
 171	struct sk_buff *new;
 172
 173	new = alloc_skb(len, gfp_mask);
 174	if (new == NULL)
 175		return NULL;
 176
 177	NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
 178	NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
 179	NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
 180
 181	skb_put_data(new, skb->data, len);
 182	return new;
 183}
 184
 185static unsigned int netlink_tap_net_id;
 186
 187struct netlink_tap_net {
 188	struct list_head netlink_tap_all;
 189	struct mutex netlink_tap_lock;
 190};
 191
 192int netlink_add_tap(struct netlink_tap *nt)
 193{
 194	struct net *net = dev_net(nt->dev);
 195	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 196
 197	if (unlikely(nt->dev->type != ARPHRD_NETLINK))
 198		return -EINVAL;
 199
 200	mutex_lock(&nn->netlink_tap_lock);
 201	list_add_rcu(&nt->list, &nn->netlink_tap_all);
 202	mutex_unlock(&nn->netlink_tap_lock);
 203
 204	__module_get(nt->module);
 205
 206	return 0;
 207}
 208EXPORT_SYMBOL_GPL(netlink_add_tap);
 209
 210static int __netlink_remove_tap(struct netlink_tap *nt)
 211{
 212	struct net *net = dev_net(nt->dev);
 213	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 214	bool found = false;
 215	struct netlink_tap *tmp;
 216
 217	mutex_lock(&nn->netlink_tap_lock);
 218
 219	list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
 220		if (nt == tmp) {
 221			list_del_rcu(&nt->list);
 222			found = true;
 223			goto out;
 224		}
 225	}
 226
 227	pr_warn("__netlink_remove_tap: %p not found\n", nt);
 228out:
 229	mutex_unlock(&nn->netlink_tap_lock);
 230
 231	if (found)
 232		module_put(nt->module);
 233
 234	return found ? 0 : -ENODEV;
 235}
 236
 237int netlink_remove_tap(struct netlink_tap *nt)
 238{
 239	int ret;
 240
 241	ret = __netlink_remove_tap(nt);
 242	synchronize_net();
 243
 244	return ret;
 245}
 246EXPORT_SYMBOL_GPL(netlink_remove_tap);
 247
 248static __net_init int netlink_tap_init_net(struct net *net)
 249{
 250	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 251
 252	INIT_LIST_HEAD(&nn->netlink_tap_all);
 253	mutex_init(&nn->netlink_tap_lock);
 254	return 0;
 255}
 256
 257static struct pernet_operations netlink_tap_net_ops = {
 258	.init = netlink_tap_init_net,
 259	.id   = &netlink_tap_net_id,
 260	.size = sizeof(struct netlink_tap_net),
 261};
 262
 263static bool netlink_filter_tap(const struct sk_buff *skb)
 264{
 265	struct sock *sk = skb->sk;
 266
 267	/* We take the more conservative approach and
 268	 * whitelist socket protocols that may pass.
 269	 */
 270	switch (sk->sk_protocol) {
 271	case NETLINK_ROUTE:
 272	case NETLINK_USERSOCK:
 273	case NETLINK_SOCK_DIAG:
 274	case NETLINK_NFLOG:
 275	case NETLINK_XFRM:
 276	case NETLINK_FIB_LOOKUP:
 277	case NETLINK_NETFILTER:
 278	case NETLINK_GENERIC:
 279		return true;
 280	}
 281
 282	return false;
 283}
 284
 285static int __netlink_deliver_tap_skb(struct sk_buff *skb,
 286				     struct net_device *dev)
 287{
 288	struct sk_buff *nskb;
 289	struct sock *sk = skb->sk;
 290	int ret = -ENOMEM;
 291
 292	if (!net_eq(dev_net(dev), sock_net(sk)))
 293		return 0;
 294
 295	dev_hold(dev);
 296
 297	if (is_vmalloc_addr(skb->head))
 298		nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
 299	else
 300		nskb = skb_clone(skb, GFP_ATOMIC);
 301	if (nskb) {
 302		nskb->dev = dev;
 303		nskb->protocol = htons((u16) sk->sk_protocol);
 304		nskb->pkt_type = netlink_is_kernel(sk) ?
 305				 PACKET_KERNEL : PACKET_USER;
 306		skb_reset_network_header(nskb);
 307		ret = dev_queue_xmit(nskb);
 308		if (unlikely(ret > 0))
 309			ret = net_xmit_errno(ret);
 310	}
 311
 312	dev_put(dev);
 313	return ret;
 314}
 315
 316static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
 317{
 318	int ret;
 319	struct netlink_tap *tmp;
 320
 321	if (!netlink_filter_tap(skb))
 322		return;
 323
 324	list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
 325		ret = __netlink_deliver_tap_skb(skb, tmp->dev);
 326		if (unlikely(ret))
 327			break;
 328	}
 329}
 330
 331static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
 332{
 333	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
 334
 335	rcu_read_lock();
 336
 337	if (unlikely(!list_empty(&nn->netlink_tap_all)))
 338		__netlink_deliver_tap(skb, nn);
 339
 340	rcu_read_unlock();
 341}
 342
 343static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
 344				       struct sk_buff *skb)
 345{
 346	if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
 347		netlink_deliver_tap(sock_net(dst), skb);
 348}
 349
 350static void netlink_overrun(struct sock *sk)
 351{
 352	if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
 353		if (!test_and_set_bit(NETLINK_S_CONGESTED,
 354				      &nlk_sk(sk)->state)) {
 355			WRITE_ONCE(sk->sk_err, ENOBUFS);
 356			sk_error_report(sk);
 357		}
 358	}
 359	atomic_inc(&sk->sk_drops);
 360}
 361
 362static void netlink_rcv_wake(struct sock *sk)
 363{
 364	struct netlink_sock *nlk = nlk_sk(sk);
 365
 366	if (skb_queue_empty_lockless(&sk->sk_receive_queue))
 367		clear_bit(NETLINK_S_CONGESTED, &nlk->state);
 368	if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
 369		wake_up_interruptible(&nlk->wait);
 370}
 371
 372static void netlink_skb_destructor(struct sk_buff *skb)
 373{
 374	if (is_vmalloc_addr(skb->head)) {
 375		if (!skb->cloned ||
 376		    !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
 377			vfree_atomic(skb->head);
 378
 379		skb->head = NULL;
 380	}
 381	if (skb->sk != NULL)
 382		sock_rfree(skb);
 383}
 384
 385static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
 386{
 387	WARN_ON(skb->sk != NULL);
 388	skb->sk = sk;
 389	skb->destructor = netlink_skb_destructor;
 390	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
 391	sk_mem_charge(sk, skb->truesize);
 392}
 393
 394static void netlink_sock_destruct(struct sock *sk)
 395{
 396	struct netlink_sock *nlk = nlk_sk(sk);
 397
 398	if (nlk->cb_running) {
 399		if (nlk->cb.done)
 400			nlk->cb.done(&nlk->cb);
 401		module_put(nlk->cb.module);
 402		kfree_skb(nlk->cb.skb);
 403	}
 404
 405	skb_queue_purge(&sk->sk_receive_queue);
 406
 407	if (!sock_flag(sk, SOCK_DEAD)) {
 408		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
 409		return;
 410	}
 411
 412	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 413	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 414	WARN_ON(nlk_sk(sk)->groups);
 415}
 416
 417static void netlink_sock_destruct_work(struct work_struct *work)
 418{
 419	struct netlink_sock *nlk = container_of(work, struct netlink_sock,
 420						work);
 421
 422	sk_free(&nlk->sk);
 423}
 424
 425/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
 426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
 427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 428 * this, _but_ remember, it adds useless work on UP machines.
 429 */
 430
 431void netlink_table_grab(void)
 432	__acquires(nl_table_lock)
 433{
 434	might_sleep();
 435
 436	write_lock_irq(&nl_table_lock);
 437
 438	if (atomic_read(&nl_table_users)) {
 439		DECLARE_WAITQUEUE(wait, current);
 440
 441		add_wait_queue_exclusive(&nl_table_wait, &wait);
 442		for (;;) {
 443			set_current_state(TASK_UNINTERRUPTIBLE);
 444			if (atomic_read(&nl_table_users) == 0)
 445				break;
 446			write_unlock_irq(&nl_table_lock);
 447			schedule();
 448			write_lock_irq(&nl_table_lock);
 449		}
 450
 451		__set_current_state(TASK_RUNNING);
 452		remove_wait_queue(&nl_table_wait, &wait);
 453	}
 454}
 455
 456void netlink_table_ungrab(void)
 457	__releases(nl_table_lock)
 458{
 459	write_unlock_irq(&nl_table_lock);
 460	wake_up(&nl_table_wait);
 461}
 462
 463static inline void
 464netlink_lock_table(void)
 465{
 466	unsigned long flags;
 467
 468	/* read_lock() synchronizes us to netlink_table_grab */
 469
 470	read_lock_irqsave(&nl_table_lock, flags);
 471	atomic_inc(&nl_table_users);
 472	read_unlock_irqrestore(&nl_table_lock, flags);
 473}
 474
 475static inline void
 476netlink_unlock_table(void)
 477{
 478	if (atomic_dec_and_test(&nl_table_users))
 479		wake_up(&nl_table_wait);
 480}
 481
 482struct netlink_compare_arg
 483{
 484	possible_net_t pnet;
 485	u32 portid;
 486};
 487
 488/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
 489#define netlink_compare_arg_len \
 490	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
 491
 492static inline int netlink_compare(struct rhashtable_compare_arg *arg,
 493				  const void *ptr)
 494{
 495	const struct netlink_compare_arg *x = arg->key;
 496	const struct netlink_sock *nlk = ptr;
 497
 498	return nlk->portid != x->portid ||
 499	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
 500}
 501
 502static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
 503				     struct net *net, u32 portid)
 504{
 505	memset(arg, 0, sizeof(*arg));
 506	write_pnet(&arg->pnet, net);
 507	arg->portid = portid;
 508}
 509
 510static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
 511				     struct net *net)
 512{
 513	struct netlink_compare_arg arg;
 514
 515	netlink_compare_arg_init(&arg, net, portid);
 516	return rhashtable_lookup_fast(&table->hash, &arg,
 517				      netlink_rhashtable_params);
 518}
 519
 520static int __netlink_insert(struct netlink_table *table, struct sock *sk)
 521{
 522	struct netlink_compare_arg arg;
 523
 524	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
 525	return rhashtable_lookup_insert_key(&table->hash, &arg,
 526					    &nlk_sk(sk)->node,
 527					    netlink_rhashtable_params);
 528}
 529
 530static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
 531{
 532	struct netlink_table *table = &nl_table[protocol];
 533	struct sock *sk;
 534
 535	rcu_read_lock();
 536	sk = __netlink_lookup(table, portid, net);
 537	if (sk)
 538		sock_hold(sk);
 539	rcu_read_unlock();
 540
 541	return sk;
 542}
 543
 544static const struct proto_ops netlink_ops;
 545
 546static void
 547netlink_update_listeners(struct sock *sk)
 548{
 549	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
 550	unsigned long mask;
 551	unsigned int i;
 552	struct listeners *listeners;
 553
 554	listeners = nl_deref_protected(tbl->listeners);
 555	if (!listeners)
 556		return;
 557
 558	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
 559		mask = 0;
 560		sk_for_each_bound(sk, &tbl->mc_list) {
 561			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
 562				mask |= nlk_sk(sk)->groups[i];
 563		}
 564		listeners->masks[i] = mask;
 565	}
 566	/* this function is only called with the netlink table "grabbed", which
 567	 * makes sure updates are visible before bind or setsockopt return. */
 568}
 569
 570static int netlink_insert(struct sock *sk, u32 portid)
 571{
 572	struct netlink_table *table = &nl_table[sk->sk_protocol];
 573	int err;
 574
 575	lock_sock(sk);
 576
 577	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
 578	if (nlk_sk(sk)->bound)
 579		goto err;
 580
 581	/* portid can be read locklessly from netlink_getname(). */
 582	WRITE_ONCE(nlk_sk(sk)->portid, portid);
 583
 584	sock_hold(sk);
 585
 586	err = __netlink_insert(table, sk);
 587	if (err) {
 588		/* In case the hashtable backend returns with -EBUSY
 589		 * from here, it must not escape to the caller.
 590		 */
 591		if (unlikely(err == -EBUSY))
 592			err = -EOVERFLOW;
 593		if (err == -EEXIST)
 594			err = -EADDRINUSE;
 595		sock_put(sk);
 596		goto err;
 597	}
 598
 599	/* We need to ensure that the socket is hashed and visible. */
 600	smp_wmb();
 601	/* Paired with lockless reads from netlink_bind(),
 602	 * netlink_connect() and netlink_sendmsg().
 603	 */
 604	WRITE_ONCE(nlk_sk(sk)->bound, portid);
 605
 606err:
 607	release_sock(sk);
 608	return err;
 609}
 610
 611static void netlink_remove(struct sock *sk)
 612{
 613	struct netlink_table *table;
 614
 615	table = &nl_table[sk->sk_protocol];
 616	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
 617				    netlink_rhashtable_params)) {
 618		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 619		__sock_put(sk);
 620	}
 621
 622	netlink_table_grab();
 623	if (nlk_sk(sk)->subscriptions) {
 624		__sk_del_bind_node(sk);
 625		netlink_update_listeners(sk);
 626	}
 627	if (sk->sk_protocol == NETLINK_GENERIC)
 628		atomic_inc(&genl_sk_destructing_cnt);
 629	netlink_table_ungrab();
 630}
 631
 632static struct proto netlink_proto = {
 633	.name	  = "NETLINK",
 634	.owner	  = THIS_MODULE,
 635	.obj_size = sizeof(struct netlink_sock),
 636};
 637
 638static int __netlink_create(struct net *net, struct socket *sock,
 639			    struct mutex *dump_cb_mutex, int protocol,
 640			    int kern)
 641{
 642	struct sock *sk;
 643	struct netlink_sock *nlk;
 644
 645	sock->ops = &netlink_ops;
 646
 647	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
 648	if (!sk)
 649		return -ENOMEM;
 650
 651	sock_init_data(sock, sk);
 652
 653	nlk = nlk_sk(sk);
 654	mutex_init(&nlk->nl_cb_mutex);
 655	lockdep_set_class_and_name(&nlk->nl_cb_mutex,
 
 
 
 
 656					   nlk_cb_mutex_keys + protocol,
 657					   nlk_cb_mutex_key_strings[protocol]);
 658	nlk->dump_cb_mutex = dump_cb_mutex;
 659	init_waitqueue_head(&nlk->wait);
 660
 661	sk->sk_destruct = netlink_sock_destruct;
 662	sk->sk_protocol = protocol;
 663	return 0;
 664}
 665
 666static int netlink_create(struct net *net, struct socket *sock, int protocol,
 667			  int kern)
 668{
 669	struct module *module = NULL;
 670	struct mutex *cb_mutex;
 671	struct netlink_sock *nlk;
 672	int (*bind)(struct net *net, int group);
 673	void (*unbind)(struct net *net, int group);
 674	void (*release)(struct sock *sock, unsigned long *groups);
 675	int err = 0;
 676
 677	sock->state = SS_UNCONNECTED;
 678
 679	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 680		return -ESOCKTNOSUPPORT;
 681
 682	if (protocol < 0 || protocol >= MAX_LINKS)
 683		return -EPROTONOSUPPORT;
 684	protocol = array_index_nospec(protocol, MAX_LINKS);
 685
 686	netlink_lock_table();
 687#ifdef CONFIG_MODULES
 688	if (!nl_table[protocol].registered) {
 689		netlink_unlock_table();
 690		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
 691		netlink_lock_table();
 692	}
 693#endif
 694	if (nl_table[protocol].registered &&
 695	    try_module_get(nl_table[protocol].module))
 696		module = nl_table[protocol].module;
 697	else
 698		err = -EPROTONOSUPPORT;
 699	cb_mutex = nl_table[protocol].cb_mutex;
 700	bind = nl_table[protocol].bind;
 701	unbind = nl_table[protocol].unbind;
 702	release = nl_table[protocol].release;
 703	netlink_unlock_table();
 704
 705	if (err < 0)
 706		goto out;
 707
 708	err = __netlink_create(net, sock, cb_mutex, protocol, kern);
 709	if (err < 0)
 710		goto out_module;
 711
 712	sock_prot_inuse_add(net, &netlink_proto, 1);
 713
 714	nlk = nlk_sk(sock->sk);
 715	nlk->module = module;
 716	nlk->netlink_bind = bind;
 717	nlk->netlink_unbind = unbind;
 718	nlk->netlink_release = release;
 719out:
 720	return err;
 721
 722out_module:
 723	module_put(module);
 724	goto out;
 725}
 726
 727static void deferred_put_nlk_sk(struct rcu_head *head)
 728{
 729	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
 730	struct sock *sk = &nlk->sk;
 731
 732	kfree(nlk->groups);
 733	nlk->groups = NULL;
 734
 735	if (!refcount_dec_and_test(&sk->sk_refcnt))
 736		return;
 737
 738	if (nlk->cb_running && nlk->cb.done) {
 739		INIT_WORK(&nlk->work, netlink_sock_destruct_work);
 740		schedule_work(&nlk->work);
 741		return;
 742	}
 743
 744	sk_free(sk);
 745}
 746
 747static int netlink_release(struct socket *sock)
 748{
 749	struct sock *sk = sock->sk;
 750	struct netlink_sock *nlk;
 751
 752	if (!sk)
 753		return 0;
 754
 755	netlink_remove(sk);
 756	sock_orphan(sk);
 757	nlk = nlk_sk(sk);
 758
 759	/*
 760	 * OK. Socket is unlinked, any packets that arrive now
 761	 * will be purged.
 762	 */
 763	if (nlk->netlink_release)
 764		nlk->netlink_release(sk, nlk->groups);
 765
 766	/* must not acquire netlink_table_lock in any way again before unbind
 767	 * and notifying genetlink is done as otherwise it might deadlock
 768	 */
 769	if (nlk->netlink_unbind) {
 770		int i;
 771
 772		for (i = 0; i < nlk->ngroups; i++)
 773			if (test_bit(i, nlk->groups))
 774				nlk->netlink_unbind(sock_net(sk), i + 1);
 775	}
 776	if (sk->sk_protocol == NETLINK_GENERIC &&
 777	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
 778		wake_up(&genl_sk_destructing_waitq);
 779
 780	sock->sk = NULL;
 781	wake_up_interruptible_all(&nlk->wait);
 782
 783	skb_queue_purge(&sk->sk_write_queue);
 784
 785	if (nlk->portid && nlk->bound) {
 786		struct netlink_notify n = {
 787						.net = sock_net(sk),
 788						.protocol = sk->sk_protocol,
 789						.portid = nlk->portid,
 790					  };
 791		blocking_notifier_call_chain(&netlink_chain,
 792				NETLINK_URELEASE, &n);
 793	}
 794
 795	module_put(nlk->module);
 796
 797	if (netlink_is_kernel(sk)) {
 798		netlink_table_grab();
 799		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
 800		if (--nl_table[sk->sk_protocol].registered == 0) {
 801			struct listeners *old;
 802
 803			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
 804			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
 805			kfree_rcu(old, rcu);
 806			nl_table[sk->sk_protocol].module = NULL;
 807			nl_table[sk->sk_protocol].bind = NULL;
 808			nl_table[sk->sk_protocol].unbind = NULL;
 809			nl_table[sk->sk_protocol].flags = 0;
 810			nl_table[sk->sk_protocol].registered = 0;
 811		}
 812		netlink_table_ungrab();
 813	}
 814
 815	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
 816
 817	/* Because struct net might disappear soon, do not keep a pointer. */
 818	if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
 819		__netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
 820		/* Because of deferred_put_nlk_sk and use of work queue,
 821		 * it is possible  netns will be freed before this socket.
 822		 */
 823		sock_net_set(sk, &init_net);
 824		__netns_tracker_alloc(&init_net, &sk->ns_tracker,
 825				      false, GFP_KERNEL);
 826	}
 827	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
 828	return 0;
 829}
 830
 831static int netlink_autobind(struct socket *sock)
 832{
 833	struct sock *sk = sock->sk;
 834	struct net *net = sock_net(sk);
 835	struct netlink_table *table = &nl_table[sk->sk_protocol];
 836	s32 portid = task_tgid_vnr(current);
 837	int err;
 838	s32 rover = -4096;
 839	bool ok;
 840
 841retry:
 842	cond_resched();
 843	rcu_read_lock();
 844	ok = !__netlink_lookup(table, portid, net);
 845	rcu_read_unlock();
 846	if (!ok) {
 847		/* Bind collision, search negative portid values. */
 848		if (rover == -4096)
 849			/* rover will be in range [S32_MIN, -4097] */
 850			rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
 851		else if (rover >= -4096)
 852			rover = -4097;
 853		portid = rover--;
 854		goto retry;
 855	}
 856
 857	err = netlink_insert(sk, portid);
 858	if (err == -EADDRINUSE)
 859		goto retry;
 860
 861	/* If 2 threads race to autobind, that is fine.  */
 862	if (err == -EBUSY)
 863		err = 0;
 864
 865	return err;
 866}
 867
 868/**
 869 * __netlink_ns_capable - General netlink message capability test
 870 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
 871 * @user_ns: The user namespace of the capability to use
 872 * @cap: The capability to use
 873 *
 874 * Test to see if the opener of the socket we received the message
 875 * from had when the netlink socket was created and the sender of the
 876 * message has the capability @cap in the user namespace @user_ns.
 877 */
 878bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
 879			struct user_namespace *user_ns, int cap)
 880{
 881	return ((nsp->flags & NETLINK_SKB_DST) ||
 882		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
 883		ns_capable(user_ns, cap);
 884}
 885EXPORT_SYMBOL(__netlink_ns_capable);
 886
 887/**
 888 * netlink_ns_capable - General netlink message capability test
 889 * @skb: socket buffer holding a netlink command from userspace
 890 * @user_ns: The user namespace of the capability to use
 891 * @cap: The capability to use
 892 *
 893 * Test to see if the opener of the socket we received the message
 894 * from had when the netlink socket was created and the sender of the
 895 * message has the capability @cap in the user namespace @user_ns.
 896 */
 897bool netlink_ns_capable(const struct sk_buff *skb,
 898			struct user_namespace *user_ns, int cap)
 899{
 900	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
 901}
 902EXPORT_SYMBOL(netlink_ns_capable);
 903
 904/**
 905 * netlink_capable - Netlink global message capability test
 906 * @skb: socket buffer holding a netlink command from userspace
 907 * @cap: The capability to use
 908 *
 909 * Test to see if the opener of the socket we received the message
 910 * from had when the netlink socket was created and the sender of the
 911 * message has the capability @cap in all user namespaces.
 912 */
 913bool netlink_capable(const struct sk_buff *skb, int cap)
 914{
 915	return netlink_ns_capable(skb, &init_user_ns, cap);
 916}
 917EXPORT_SYMBOL(netlink_capable);
 918
 919/**
 920 * netlink_net_capable - Netlink network namespace message capability test
 921 * @skb: socket buffer holding a netlink command from userspace
 922 * @cap: The capability to use
 923 *
 924 * Test to see if the opener of the socket we received the message
 925 * from had when the netlink socket was created and the sender of the
 926 * message has the capability @cap over the network namespace of
 927 * the socket we received the message from.
 928 */
 929bool netlink_net_capable(const struct sk_buff *skb, int cap)
 930{
 931	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
 932}
 933EXPORT_SYMBOL(netlink_net_capable);
 934
 935static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
 936{
 937	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
 938		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
 939}
 940
 941static void
 942netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
 943{
 944	struct netlink_sock *nlk = nlk_sk(sk);
 945
 946	if (nlk->subscriptions && !subscriptions)
 947		__sk_del_bind_node(sk);
 948	else if (!nlk->subscriptions && subscriptions)
 949		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
 950	nlk->subscriptions = subscriptions;
 951}
 952
 953static int netlink_realloc_groups(struct sock *sk)
 954{
 955	struct netlink_sock *nlk = nlk_sk(sk);
 956	unsigned int groups;
 957	unsigned long *new_groups;
 958	int err = 0;
 959
 960	netlink_table_grab();
 961
 962	groups = nl_table[sk->sk_protocol].groups;
 963	if (!nl_table[sk->sk_protocol].registered) {
 964		err = -ENOENT;
 965		goto out_unlock;
 966	}
 967
 968	if (nlk->ngroups >= groups)
 969		goto out_unlock;
 970
 971	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
 972	if (new_groups == NULL) {
 973		err = -ENOMEM;
 974		goto out_unlock;
 975	}
 976	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
 977	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
 978
 979	nlk->groups = new_groups;
 980	nlk->ngroups = groups;
 981 out_unlock:
 982	netlink_table_ungrab();
 983	return err;
 984}
 985
 986static void netlink_undo_bind(int group, long unsigned int groups,
 987			      struct sock *sk)
 988{
 989	struct netlink_sock *nlk = nlk_sk(sk);
 990	int undo;
 991
 992	if (!nlk->netlink_unbind)
 993		return;
 994
 995	for (undo = 0; undo < group; undo++)
 996		if (test_bit(undo, &groups))
 997			nlk->netlink_unbind(sock_net(sk), undo + 1);
 998}
 999
1000static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1001			int addr_len)
1002{
1003	struct sock *sk = sock->sk;
1004	struct net *net = sock_net(sk);
1005	struct netlink_sock *nlk = nlk_sk(sk);
1006	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1007	int err = 0;
1008	unsigned long groups;
1009	bool bound;
1010
1011	if (addr_len < sizeof(struct sockaddr_nl))
1012		return -EINVAL;
1013
1014	if (nladdr->nl_family != AF_NETLINK)
1015		return -EINVAL;
1016	groups = nladdr->nl_groups;
1017
1018	/* Only superuser is allowed to listen multicasts */
1019	if (groups) {
1020		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1021			return -EPERM;
1022		err = netlink_realloc_groups(sk);
1023		if (err)
1024			return err;
1025	}
1026
1027	if (nlk->ngroups < BITS_PER_LONG)
1028		groups &= (1UL << nlk->ngroups) - 1;
1029
1030	/* Paired with WRITE_ONCE() in netlink_insert() */
1031	bound = READ_ONCE(nlk->bound);
1032	if (bound) {
1033		/* Ensure nlk->portid is up-to-date. */
1034		smp_rmb();
1035
1036		if (nladdr->nl_pid != nlk->portid)
1037			return -EINVAL;
1038	}
1039
1040	if (nlk->netlink_bind && groups) {
1041		int group;
1042
1043		/* nl_groups is a u32, so cap the maximum groups we can bind */
1044		for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1045			if (!test_bit(group, &groups))
1046				continue;
1047			err = nlk->netlink_bind(net, group + 1);
1048			if (!err)
1049				continue;
1050			netlink_undo_bind(group, groups, sk);
1051			return err;
1052		}
1053	}
1054
1055	/* No need for barriers here as we return to user-space without
1056	 * using any of the bound attributes.
1057	 */
1058	netlink_lock_table();
1059	if (!bound) {
1060		err = nladdr->nl_pid ?
1061			netlink_insert(sk, nladdr->nl_pid) :
1062			netlink_autobind(sock);
1063		if (err) {
1064			netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1065			goto unlock;
1066		}
1067	}
1068
1069	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1070		goto unlock;
1071	netlink_unlock_table();
1072
1073	netlink_table_grab();
1074	netlink_update_subscriptions(sk, nlk->subscriptions +
1075					 hweight32(groups) -
1076					 hweight32(nlk->groups[0]));
1077	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1078	netlink_update_listeners(sk);
1079	netlink_table_ungrab();
1080
1081	return 0;
1082
1083unlock:
1084	netlink_unlock_table();
1085	return err;
1086}
1087
1088static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1089			   int alen, int flags)
1090{
1091	int err = 0;
1092	struct sock *sk = sock->sk;
1093	struct netlink_sock *nlk = nlk_sk(sk);
1094	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1095
1096	if (alen < sizeof(addr->sa_family))
1097		return -EINVAL;
1098
1099	if (addr->sa_family == AF_UNSPEC) {
1100		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1101		WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1102		/* dst_portid and dst_group can be read locklessly */
1103		WRITE_ONCE(nlk->dst_portid, 0);
1104		WRITE_ONCE(nlk->dst_group, 0);
1105		return 0;
1106	}
1107	if (addr->sa_family != AF_NETLINK)
1108		return -EINVAL;
1109
1110	if (alen < sizeof(struct sockaddr_nl))
1111		return -EINVAL;
1112
1113	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1114	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1115		return -EPERM;
1116
1117	/* No need for barriers here as we return to user-space without
1118	 * using any of the bound attributes.
1119	 * Paired with WRITE_ONCE() in netlink_insert().
1120	 */
1121	if (!READ_ONCE(nlk->bound))
1122		err = netlink_autobind(sock);
1123
1124	if (err == 0) {
1125		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1126		WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1127		/* dst_portid and dst_group can be read locklessly */
1128		WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1129		WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1130	}
1131
1132	return err;
1133}
1134
1135static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1136			   int peer)
1137{
1138	struct sock *sk = sock->sk;
1139	struct netlink_sock *nlk = nlk_sk(sk);
1140	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1141
1142	nladdr->nl_family = AF_NETLINK;
1143	nladdr->nl_pad = 0;
1144
1145	if (peer) {
1146		/* Paired with WRITE_ONCE() in netlink_connect() */
1147		nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1148		nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1149	} else {
1150		/* Paired with WRITE_ONCE() in netlink_insert() */
1151		nladdr->nl_pid = READ_ONCE(nlk->portid);
1152		netlink_lock_table();
1153		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1154		netlink_unlock_table();
1155	}
1156	return sizeof(*nladdr);
1157}
1158
1159static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1160			 unsigned long arg)
1161{
1162	/* try to hand this ioctl down to the NIC drivers.
1163	 */
1164	return -ENOIOCTLCMD;
1165}
1166
1167static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1168{
1169	struct sock *sock;
1170	struct netlink_sock *nlk;
1171
1172	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1173	if (!sock)
1174		return ERR_PTR(-ECONNREFUSED);
1175
1176	/* Don't bother queuing skb if kernel socket has no input function */
1177	nlk = nlk_sk(sock);
1178	/* dst_portid and sk_state can be changed in netlink_connect() */
1179	if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1180	    READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1181		sock_put(sock);
1182		return ERR_PTR(-ECONNREFUSED);
1183	}
1184	return sock;
1185}
1186
1187struct sock *netlink_getsockbyfilp(struct file *filp)
1188{
1189	struct inode *inode = file_inode(filp);
1190	struct sock *sock;
1191
1192	if (!S_ISSOCK(inode->i_mode))
1193		return ERR_PTR(-ENOTSOCK);
1194
1195	sock = SOCKET_I(inode)->sk;
1196	if (sock->sk_family != AF_NETLINK)
1197		return ERR_PTR(-EINVAL);
1198
1199	sock_hold(sock);
1200	return sock;
1201}
1202
1203struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1204{
1205	size_t head_size = SKB_HEAD_ALIGN(size);
1206	struct sk_buff *skb;
1207	void *data;
1208
1209	if (head_size <= PAGE_SIZE || broadcast)
1210		return alloc_skb(size, GFP_KERNEL);
1211
1212	data = kvmalloc(head_size, GFP_KERNEL);
1213	if (!data)
 
 
 
1214		return NULL;
1215
1216	skb = __build_skb(data, head_size);
1217	if (!skb)
1218		kvfree(data);
1219	else if (is_vmalloc_addr(data))
1220		skb->destructor = netlink_skb_destructor;
1221
1222	return skb;
1223}
1224
1225/*
1226 * Attach a skb to a netlink socket.
1227 * The caller must hold a reference to the destination socket. On error, the
1228 * reference is dropped. The skb is not send to the destination, just all
1229 * all error checks are performed and memory in the queue is reserved.
1230 * Return values:
1231 * < 0: error. skb freed, reference to sock dropped.
1232 * 0: continue
1233 * 1: repeat lookup - reference dropped while waiting for socket memory.
1234 */
1235int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1236		      long *timeo, struct sock *ssk)
1237{
1238	struct netlink_sock *nlk;
1239
1240	nlk = nlk_sk(sk);
1241
1242	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1243	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1244		DECLARE_WAITQUEUE(wait, current);
1245		if (!*timeo) {
1246			if (!ssk || netlink_is_kernel(ssk))
1247				netlink_overrun(sk);
1248			sock_put(sk);
1249			kfree_skb(skb);
1250			return -EAGAIN;
1251		}
1252
1253		__set_current_state(TASK_INTERRUPTIBLE);
1254		add_wait_queue(&nlk->wait, &wait);
1255
1256		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1257		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1258		    !sock_flag(sk, SOCK_DEAD))
1259			*timeo = schedule_timeout(*timeo);
1260
1261		__set_current_state(TASK_RUNNING);
1262		remove_wait_queue(&nlk->wait, &wait);
1263		sock_put(sk);
1264
1265		if (signal_pending(current)) {
1266			kfree_skb(skb);
1267			return sock_intr_errno(*timeo);
1268		}
1269		return 1;
1270	}
1271	netlink_skb_set_owner_r(skb, sk);
1272	return 0;
1273}
1274
1275static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1276{
1277	int len = skb->len;
1278
1279	netlink_deliver_tap(sock_net(sk), skb);
1280
1281	skb_queue_tail(&sk->sk_receive_queue, skb);
1282	sk->sk_data_ready(sk);
1283	return len;
1284}
1285
1286int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1287{
1288	int len = __netlink_sendskb(sk, skb);
1289
1290	sock_put(sk);
1291	return len;
1292}
1293
1294void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1295{
1296	kfree_skb(skb);
1297	sock_put(sk);
1298}
1299
1300static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1301{
1302	int delta;
1303
1304	WARN_ON(skb->sk != NULL);
1305	delta = skb->end - skb->tail;
1306	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1307		return skb;
1308
1309	if (skb_shared(skb)) {
1310		struct sk_buff *nskb = skb_clone(skb, allocation);
1311		if (!nskb)
1312			return skb;
1313		consume_skb(skb);
1314		skb = nskb;
1315	}
1316
1317	pskb_expand_head(skb, 0, -delta,
1318			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1319			 __GFP_NOWARN | __GFP_NORETRY);
1320	return skb;
1321}
1322
1323static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1324				  struct sock *ssk)
1325{
1326	int ret;
1327	struct netlink_sock *nlk = nlk_sk(sk);
1328
1329	ret = -ECONNREFUSED;
1330	if (nlk->netlink_rcv != NULL) {
1331		ret = skb->len;
1332		netlink_skb_set_owner_r(skb, sk);
1333		NETLINK_CB(skb).sk = ssk;
1334		netlink_deliver_tap_kernel(sk, ssk, skb);
1335		nlk->netlink_rcv(skb);
1336		consume_skb(skb);
1337	} else {
1338		kfree_skb(skb);
1339	}
1340	sock_put(sk);
1341	return ret;
1342}
1343
1344int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1345		    u32 portid, int nonblock)
1346{
1347	struct sock *sk;
1348	int err;
1349	long timeo;
1350
1351	skb = netlink_trim(skb, gfp_any());
1352
1353	timeo = sock_sndtimeo(ssk, nonblock);
1354retry:
1355	sk = netlink_getsockbyportid(ssk, portid);
1356	if (IS_ERR(sk)) {
1357		kfree_skb(skb);
1358		return PTR_ERR(sk);
1359	}
1360	if (netlink_is_kernel(sk))
1361		return netlink_unicast_kernel(sk, skb, ssk);
1362
1363	if (sk_filter(sk, skb)) {
1364		err = skb->len;
1365		kfree_skb(skb);
1366		sock_put(sk);
1367		return err;
1368	}
1369
1370	err = netlink_attachskb(sk, skb, &timeo, ssk);
1371	if (err == 1)
1372		goto retry;
1373	if (err)
1374		return err;
1375
1376	return netlink_sendskb(sk, skb);
1377}
1378EXPORT_SYMBOL(netlink_unicast);
1379
1380int netlink_has_listeners(struct sock *sk, unsigned int group)
1381{
1382	int res = 0;
1383	struct listeners *listeners;
1384
1385	BUG_ON(!netlink_is_kernel(sk));
1386
1387	rcu_read_lock();
1388	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1389
1390	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1391		res = test_bit(group - 1, listeners->masks);
1392
1393	rcu_read_unlock();
1394
1395	return res;
1396}
1397EXPORT_SYMBOL_GPL(netlink_has_listeners);
1398
1399bool netlink_strict_get_check(struct sk_buff *skb)
1400{
1401	return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1402}
1403EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1404
1405static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1406{
1407	struct netlink_sock *nlk = nlk_sk(sk);
1408
1409	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1410	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1411		netlink_skb_set_owner_r(skb, sk);
1412		__netlink_sendskb(sk, skb);
1413		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1414	}
1415	return -1;
1416}
1417
1418struct netlink_broadcast_data {
1419	struct sock *exclude_sk;
1420	struct net *net;
1421	u32 portid;
1422	u32 group;
1423	int failure;
1424	int delivery_failure;
1425	int congested;
1426	int delivered;
1427	gfp_t allocation;
1428	struct sk_buff *skb, *skb2;
1429	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1430	void *tx_data;
1431};
1432
1433static void do_one_broadcast(struct sock *sk,
1434				    struct netlink_broadcast_data *p)
1435{
1436	struct netlink_sock *nlk = nlk_sk(sk);
1437	int val;
1438
1439	if (p->exclude_sk == sk)
1440		return;
1441
1442	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1443	    !test_bit(p->group - 1, nlk->groups))
1444		return;
1445
1446	if (!net_eq(sock_net(sk), p->net)) {
1447		if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1448			return;
1449
1450		if (!peernet_has_id(sock_net(sk), p->net))
1451			return;
1452
1453		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1454				     CAP_NET_BROADCAST))
1455			return;
1456	}
1457
1458	if (p->failure) {
1459		netlink_overrun(sk);
1460		return;
1461	}
1462
1463	sock_hold(sk);
1464	if (p->skb2 == NULL) {
1465		if (skb_shared(p->skb)) {
1466			p->skb2 = skb_clone(p->skb, p->allocation);
1467		} else {
1468			p->skb2 = skb_get(p->skb);
1469			/*
1470			 * skb ownership may have been set when
1471			 * delivered to a previous socket.
1472			 */
1473			skb_orphan(p->skb2);
1474		}
1475	}
1476	if (p->skb2 == NULL) {
1477		netlink_overrun(sk);
1478		/* Clone failed. Notify ALL listeners. */
1479		p->failure = 1;
1480		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1481			p->delivery_failure = 1;
1482		goto out;
1483	}
1484
1485	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1486		kfree_skb(p->skb2);
1487		p->skb2 = NULL;
1488		goto out;
1489	}
1490
1491	if (sk_filter(sk, p->skb2)) {
1492		kfree_skb(p->skb2);
1493		p->skb2 = NULL;
1494		goto out;
1495	}
1496	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1497	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1498		NETLINK_CB(p->skb2).nsid_is_set = true;
1499	val = netlink_broadcast_deliver(sk, p->skb2);
1500	if (val < 0) {
1501		netlink_overrun(sk);
1502		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1503			p->delivery_failure = 1;
1504	} else {
1505		p->congested |= val;
1506		p->delivered = 1;
1507		p->skb2 = NULL;
1508	}
1509out:
1510	sock_put(sk);
1511}
1512
1513int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1514			       u32 portid,
1515			       u32 group, gfp_t allocation,
1516			       netlink_filter_fn filter,
1517			       void *filter_data)
1518{
1519	struct net *net = sock_net(ssk);
1520	struct netlink_broadcast_data info;
1521	struct sock *sk;
1522
1523	skb = netlink_trim(skb, allocation);
1524
1525	info.exclude_sk = ssk;
1526	info.net = net;
1527	info.portid = portid;
1528	info.group = group;
1529	info.failure = 0;
1530	info.delivery_failure = 0;
1531	info.congested = 0;
1532	info.delivered = 0;
1533	info.allocation = allocation;
1534	info.skb = skb;
1535	info.skb2 = NULL;
1536	info.tx_filter = filter;
1537	info.tx_data = filter_data;
1538
1539	/* While we sleep in clone, do not allow to change socket list */
1540
1541	netlink_lock_table();
1542
1543	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1544		do_one_broadcast(sk, &info);
1545
1546	consume_skb(skb);
1547
1548	netlink_unlock_table();
1549
1550	if (info.delivery_failure) {
1551		kfree_skb(info.skb2);
1552		return -ENOBUFS;
1553	}
1554	consume_skb(info.skb2);
1555
1556	if (info.delivered) {
1557		if (info.congested && gfpflags_allow_blocking(allocation))
1558			yield();
1559		return 0;
1560	}
1561	return -ESRCH;
1562}
1563EXPORT_SYMBOL(netlink_broadcast_filtered);
1564
1565int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1566		      u32 group, gfp_t allocation)
1567{
1568	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1569					  NULL, NULL);
1570}
1571EXPORT_SYMBOL(netlink_broadcast);
1572
1573struct netlink_set_err_data {
1574	struct sock *exclude_sk;
1575	u32 portid;
1576	u32 group;
1577	int code;
1578};
1579
1580static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1581{
1582	struct netlink_sock *nlk = nlk_sk(sk);
1583	int ret = 0;
1584
1585	if (sk == p->exclude_sk)
1586		goto out;
1587
1588	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1589		goto out;
1590
1591	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1592	    !test_bit(p->group - 1, nlk->groups))
1593		goto out;
1594
1595	if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1596		ret = 1;
1597		goto out;
1598	}
1599
1600	WRITE_ONCE(sk->sk_err, p->code);
1601	sk_error_report(sk);
1602out:
1603	return ret;
1604}
1605
1606/**
1607 * netlink_set_err - report error to broadcast listeners
1608 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1609 * @portid: the PORTID of a process that we want to skip (if any)
1610 * @group: the broadcast group that will notice the error
1611 * @code: error code, must be negative (as usual in kernelspace)
1612 *
1613 * This function returns the number of broadcast listeners that have set the
1614 * NETLINK_NO_ENOBUFS socket option.
1615 */
1616int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1617{
1618	struct netlink_set_err_data info;
1619	unsigned long flags;
1620	struct sock *sk;
1621	int ret = 0;
1622
1623	info.exclude_sk = ssk;
1624	info.portid = portid;
1625	info.group = group;
1626	/* sk->sk_err wants a positive error value */
1627	info.code = -code;
1628
1629	read_lock_irqsave(&nl_table_lock, flags);
1630
1631	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1632		ret += do_one_set_err(sk, &info);
1633
1634	read_unlock_irqrestore(&nl_table_lock, flags);
1635	return ret;
1636}
1637EXPORT_SYMBOL(netlink_set_err);
1638
1639/* must be called with netlink table grabbed */
1640static void netlink_update_socket_mc(struct netlink_sock *nlk,
1641				     unsigned int group,
1642				     int is_new)
1643{
1644	int old, new = !!is_new, subscriptions;
1645
1646	old = test_bit(group - 1, nlk->groups);
1647	subscriptions = nlk->subscriptions - old + new;
1648	__assign_bit(group - 1, nlk->groups, new);
1649	netlink_update_subscriptions(&nlk->sk, subscriptions);
1650	netlink_update_listeners(&nlk->sk);
1651}
1652
1653static int netlink_setsockopt(struct socket *sock, int level, int optname,
1654			      sockptr_t optval, unsigned int optlen)
1655{
1656	struct sock *sk = sock->sk;
1657	struct netlink_sock *nlk = nlk_sk(sk);
1658	unsigned int val = 0;
1659	int nr = -1;
1660
1661	if (level != SOL_NETLINK)
1662		return -ENOPROTOOPT;
1663
1664	if (optlen >= sizeof(int) &&
1665	    copy_from_sockptr(&val, optval, sizeof(val)))
1666		return -EFAULT;
1667
1668	switch (optname) {
1669	case NETLINK_PKTINFO:
1670		nr = NETLINK_F_RECV_PKTINFO;
1671		break;
1672	case NETLINK_ADD_MEMBERSHIP:
1673	case NETLINK_DROP_MEMBERSHIP: {
1674		int err;
1675
1676		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1677			return -EPERM;
1678		err = netlink_realloc_groups(sk);
1679		if (err)
1680			return err;
1681		if (!val || val - 1 >= nlk->ngroups)
1682			return -EINVAL;
1683		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1684			err = nlk->netlink_bind(sock_net(sk), val);
1685			if (err)
1686				return err;
1687		}
1688		netlink_table_grab();
1689		netlink_update_socket_mc(nlk, val,
1690					 optname == NETLINK_ADD_MEMBERSHIP);
1691		netlink_table_ungrab();
1692		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1693			nlk->netlink_unbind(sock_net(sk), val);
1694
1695		break;
1696	}
1697	case NETLINK_BROADCAST_ERROR:
1698		nr = NETLINK_F_BROADCAST_SEND_ERROR;
1699		break;
1700	case NETLINK_NO_ENOBUFS:
1701		assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1702		if (val) {
1703			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1704			wake_up_interruptible(&nlk->wait);
1705		}
1706		break;
1707	case NETLINK_LISTEN_ALL_NSID:
1708		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1709			return -EPERM;
1710		nr = NETLINK_F_LISTEN_ALL_NSID;
1711		break;
1712	case NETLINK_CAP_ACK:
1713		nr = NETLINK_F_CAP_ACK;
1714		break;
1715	case NETLINK_EXT_ACK:
1716		nr = NETLINK_F_EXT_ACK;
1717		break;
1718	case NETLINK_GET_STRICT_CHK:
1719		nr = NETLINK_F_STRICT_CHK;
1720		break;
1721	default:
1722		return -ENOPROTOOPT;
1723	}
1724	if (nr >= 0)
1725		assign_bit(nr, &nlk->flags, val);
1726	return 0;
1727}
1728
1729static int netlink_getsockopt(struct socket *sock, int level, int optname,
1730			      char __user *optval, int __user *optlen)
1731{
1732	struct sock *sk = sock->sk;
1733	struct netlink_sock *nlk = nlk_sk(sk);
1734	unsigned int flag;
1735	int len, val;
1736
1737	if (level != SOL_NETLINK)
1738		return -ENOPROTOOPT;
1739
1740	if (get_user(len, optlen))
1741		return -EFAULT;
1742	if (len < 0)
1743		return -EINVAL;
1744
1745	switch (optname) {
1746	case NETLINK_PKTINFO:
1747		flag = NETLINK_F_RECV_PKTINFO;
1748		break;
1749	case NETLINK_BROADCAST_ERROR:
1750		flag = NETLINK_F_BROADCAST_SEND_ERROR;
1751		break;
1752	case NETLINK_NO_ENOBUFS:
1753		flag = NETLINK_F_RECV_NO_ENOBUFS;
1754		break;
1755	case NETLINK_LIST_MEMBERSHIPS: {
1756		int pos, idx, shift, err = 0;
1757
1758		netlink_lock_table();
1759		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1760			if (len - pos < sizeof(u32))
1761				break;
1762
1763			idx = pos / sizeof(unsigned long);
1764			shift = (pos % sizeof(unsigned long)) * 8;
1765			if (put_user((u32)(nlk->groups[idx] >> shift),
1766				     (u32 __user *)(optval + pos))) {
1767				err = -EFAULT;
1768				break;
1769			}
1770		}
1771		if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1772			err = -EFAULT;
1773		netlink_unlock_table();
1774		return err;
1775	}
1776	case NETLINK_LISTEN_ALL_NSID:
1777		flag = NETLINK_F_LISTEN_ALL_NSID;
1778		break;
1779	case NETLINK_CAP_ACK:
1780		flag = NETLINK_F_CAP_ACK;
1781		break;
1782	case NETLINK_EXT_ACK:
1783		flag = NETLINK_F_EXT_ACK;
1784		break;
1785	case NETLINK_GET_STRICT_CHK:
1786		flag = NETLINK_F_STRICT_CHK;
1787		break;
1788	default:
1789		return -ENOPROTOOPT;
1790	}
1791
1792	if (len < sizeof(int))
1793		return -EINVAL;
1794
1795	len = sizeof(int);
1796	val = test_bit(flag, &nlk->flags);
1797
1798	if (put_user(len, optlen) ||
1799	    copy_to_user(optval, &val, len))
1800		return -EFAULT;
1801
1802	return 0;
1803}
1804
1805static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1806{
1807	struct nl_pktinfo info;
1808
1809	info.group = NETLINK_CB(skb).dst_group;
1810	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1811}
1812
1813static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1814					 struct sk_buff *skb)
1815{
1816	if (!NETLINK_CB(skb).nsid_is_set)
1817		return;
1818
1819	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1820		 &NETLINK_CB(skb).nsid);
1821}
1822
1823static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1824{
1825	struct sock *sk = sock->sk;
1826	struct netlink_sock *nlk = nlk_sk(sk);
1827	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1828	u32 dst_portid;
1829	u32 dst_group;
1830	struct sk_buff *skb;
1831	int err;
1832	struct scm_cookie scm;
1833	u32 netlink_skb_flags = 0;
1834
1835	if (msg->msg_flags & MSG_OOB)
1836		return -EOPNOTSUPP;
1837
1838	if (len == 0) {
1839		pr_warn_once("Zero length message leads to an empty skb\n");
1840		return -ENODATA;
1841	}
1842
1843	err = scm_send(sock, msg, &scm, true);
1844	if (err < 0)
1845		return err;
1846
1847	if (msg->msg_namelen) {
1848		err = -EINVAL;
1849		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1850			goto out;
1851		if (addr->nl_family != AF_NETLINK)
1852			goto out;
1853		dst_portid = addr->nl_pid;
1854		dst_group = ffs(addr->nl_groups);
1855		err =  -EPERM;
1856		if ((dst_group || dst_portid) &&
1857		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1858			goto out;
1859		netlink_skb_flags |= NETLINK_SKB_DST;
1860	} else {
1861		/* Paired with WRITE_ONCE() in netlink_connect() */
1862		dst_portid = READ_ONCE(nlk->dst_portid);
1863		dst_group = READ_ONCE(nlk->dst_group);
1864	}
1865
1866	/* Paired with WRITE_ONCE() in netlink_insert() */
1867	if (!READ_ONCE(nlk->bound)) {
1868		err = netlink_autobind(sock);
1869		if (err)
1870			goto out;
1871	} else {
1872		/* Ensure nlk is hashed and visible. */
1873		smp_rmb();
1874	}
1875
1876	err = -EMSGSIZE;
1877	if (len > sk->sk_sndbuf - 32)
1878		goto out;
1879	err = -ENOBUFS;
1880	skb = netlink_alloc_large_skb(len, dst_group);
1881	if (skb == NULL)
1882		goto out;
1883
1884	NETLINK_CB(skb).portid	= nlk->portid;
1885	NETLINK_CB(skb).dst_group = dst_group;
1886	NETLINK_CB(skb).creds	= scm.creds;
1887	NETLINK_CB(skb).flags	= netlink_skb_flags;
1888
1889	err = -EFAULT;
1890	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1891		kfree_skb(skb);
1892		goto out;
1893	}
1894
1895	err = security_netlink_send(sk, skb);
1896	if (err) {
1897		kfree_skb(skb);
1898		goto out;
1899	}
1900
1901	if (dst_group) {
1902		refcount_inc(&skb->users);
1903		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1904	}
1905	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1906
1907out:
1908	scm_destroy(&scm);
1909	return err;
1910}
1911
1912static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1913			   int flags)
1914{
1915	struct scm_cookie scm;
1916	struct sock *sk = sock->sk;
1917	struct netlink_sock *nlk = nlk_sk(sk);
1918	size_t copied, max_recvmsg_len;
1919	struct sk_buff *skb, *data_skb;
1920	int err, ret;
1921
1922	if (flags & MSG_OOB)
1923		return -EOPNOTSUPP;
1924
1925	copied = 0;
1926
1927	skb = skb_recv_datagram(sk, flags, &err);
1928	if (skb == NULL)
1929		goto out;
1930
1931	data_skb = skb;
1932
1933#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1934	if (unlikely(skb_shinfo(skb)->frag_list)) {
1935		/*
1936		 * If this skb has a frag_list, then here that means that we
1937		 * will have to use the frag_list skb's data for compat tasks
1938		 * and the regular skb's data for normal (non-compat) tasks.
1939		 *
1940		 * If we need to send the compat skb, assign it to the
1941		 * 'data_skb' variable so that it will be used below for data
1942		 * copying. We keep 'skb' for everything else, including
1943		 * freeing both later.
1944		 */
1945		if (flags & MSG_CMSG_COMPAT)
1946			data_skb = skb_shinfo(skb)->frag_list;
1947	}
1948#endif
1949
1950	/* Record the max length of recvmsg() calls for future allocations */
1951	max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1952	max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1953				SKB_WITH_OVERHEAD(32768));
1954	WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1955
1956	copied = data_skb->len;
1957	if (len < copied) {
1958		msg->msg_flags |= MSG_TRUNC;
1959		copied = len;
1960	}
1961
1962	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1963
1964	if (msg->msg_name) {
1965		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1966		addr->nl_family = AF_NETLINK;
1967		addr->nl_pad    = 0;
1968		addr->nl_pid	= NETLINK_CB(skb).portid;
1969		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1970		msg->msg_namelen = sizeof(*addr);
1971	}
1972
1973	if (nlk_test_bit(RECV_PKTINFO, sk))
1974		netlink_cmsg_recv_pktinfo(msg, skb);
1975	if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1976		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1977
1978	memset(&scm, 0, sizeof(scm));
1979	scm.creds = *NETLINK_CREDS(skb);
1980	if (flags & MSG_TRUNC)
1981		copied = data_skb->len;
1982
1983	skb_free_datagram(sk, skb);
1984
1985	if (READ_ONCE(nlk->cb_running) &&
1986	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1987		ret = netlink_dump(sk, false);
1988		if (ret) {
1989			WRITE_ONCE(sk->sk_err, -ret);
1990			sk_error_report(sk);
1991		}
1992	}
1993
1994	scm_recv(sock, msg, &scm, flags);
1995out:
1996	netlink_rcv_wake(sk);
1997	return err ? : copied;
1998}
1999
2000static void netlink_data_ready(struct sock *sk)
2001{
2002	BUG();
2003}
2004
2005/*
2006 *	We export these functions to other modules. They provide a
2007 *	complete set of kernel non-blocking support for message
2008 *	queueing.
2009 */
2010
2011struct sock *
2012__netlink_kernel_create(struct net *net, int unit, struct module *module,
2013			struct netlink_kernel_cfg *cfg)
2014{
2015	struct socket *sock;
2016	struct sock *sk;
2017	struct netlink_sock *nlk;
2018	struct listeners *listeners = NULL;
2019	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2020	unsigned int groups;
2021
2022	BUG_ON(!nl_table);
2023
2024	if (unit < 0 || unit >= MAX_LINKS)
2025		return NULL;
2026
2027	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2028		return NULL;
2029
2030	if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2031		goto out_sock_release_nosk;
2032
2033	sk = sock->sk;
2034
2035	if (!cfg || cfg->groups < 32)
2036		groups = 32;
2037	else
2038		groups = cfg->groups;
2039
2040	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2041	if (!listeners)
2042		goto out_sock_release;
2043
2044	sk->sk_data_ready = netlink_data_ready;
2045	if (cfg && cfg->input)
2046		nlk_sk(sk)->netlink_rcv = cfg->input;
2047
2048	if (netlink_insert(sk, 0))
2049		goto out_sock_release;
2050
2051	nlk = nlk_sk(sk);
2052	set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2053
2054	netlink_table_grab();
2055	if (!nl_table[unit].registered) {
2056		nl_table[unit].groups = groups;
2057		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2058		nl_table[unit].cb_mutex = cb_mutex;
2059		nl_table[unit].module = module;
2060		if (cfg) {
2061			nl_table[unit].bind = cfg->bind;
2062			nl_table[unit].unbind = cfg->unbind;
2063			nl_table[unit].release = cfg->release;
2064			nl_table[unit].flags = cfg->flags;
2065		}
2066		nl_table[unit].registered = 1;
2067	} else {
2068		kfree(listeners);
2069		nl_table[unit].registered++;
2070	}
2071	netlink_table_ungrab();
2072	return sk;
2073
2074out_sock_release:
2075	kfree(listeners);
2076	netlink_kernel_release(sk);
2077	return NULL;
2078
2079out_sock_release_nosk:
2080	sock_release(sock);
2081	return NULL;
2082}
2083EXPORT_SYMBOL(__netlink_kernel_create);
2084
2085void
2086netlink_kernel_release(struct sock *sk)
2087{
2088	if (sk == NULL || sk->sk_socket == NULL)
2089		return;
2090
2091	sock_release(sk->sk_socket);
2092}
2093EXPORT_SYMBOL(netlink_kernel_release);
2094
2095int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2096{
2097	struct listeners *new, *old;
2098	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2099
2100	if (groups < 32)
2101		groups = 32;
2102
2103	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2104		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2105		if (!new)
2106			return -ENOMEM;
2107		old = nl_deref_protected(tbl->listeners);
2108		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2109		rcu_assign_pointer(tbl->listeners, new);
2110
2111		kfree_rcu(old, rcu);
2112	}
2113	tbl->groups = groups;
2114
2115	return 0;
2116}
2117
2118/**
2119 * netlink_change_ngroups - change number of multicast groups
2120 *
2121 * This changes the number of multicast groups that are available
2122 * on a certain netlink family. Note that it is not possible to
2123 * change the number of groups to below 32. Also note that it does
2124 * not implicitly call netlink_clear_multicast_users() when the
2125 * number of groups is reduced.
2126 *
2127 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2128 * @groups: The new number of groups.
2129 */
2130int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2131{
2132	int err;
2133
2134	netlink_table_grab();
2135	err = __netlink_change_ngroups(sk, groups);
2136	netlink_table_ungrab();
2137
2138	return err;
2139}
2140
2141void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2142{
2143	struct sock *sk;
2144	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2145
2146	sk_for_each_bound(sk, &tbl->mc_list)
2147		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2148}
2149
2150struct nlmsghdr *
2151__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2152{
2153	struct nlmsghdr *nlh;
2154	int size = nlmsg_msg_size(len);
2155
2156	nlh = skb_put(skb, NLMSG_ALIGN(size));
2157	nlh->nlmsg_type = type;
2158	nlh->nlmsg_len = size;
2159	nlh->nlmsg_flags = flags;
2160	nlh->nlmsg_pid = portid;
2161	nlh->nlmsg_seq = seq;
2162	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2163		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2164	return nlh;
2165}
2166EXPORT_SYMBOL(__nlmsg_put);
2167
2168/*
2169 * It looks a bit ugly.
2170 * It would be better to create kernel thread.
2171 */
2172
2173static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2174			     struct netlink_callback *cb,
2175			     struct netlink_ext_ack *extack)
2176{
2177	struct nlmsghdr *nlh;
2178
2179	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2180			       NLM_F_MULTI | cb->answer_flags);
2181	if (WARN_ON(!nlh))
2182		return -ENOBUFS;
2183
2184	nl_dump_check_consistent(cb, nlh);
2185	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2186
2187	if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2188		nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2189		if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2190			nlmsg_end(skb, nlh);
2191	}
2192
2193	return 0;
2194}
2195
2196static int netlink_dump(struct sock *sk, bool lock_taken)
2197{
2198	struct netlink_sock *nlk = nlk_sk(sk);
2199	struct netlink_ext_ack extack = {};
2200	struct netlink_callback *cb;
2201	struct sk_buff *skb = NULL;
2202	size_t max_recvmsg_len;
2203	struct module *module;
2204	int err = -ENOBUFS;
2205	int alloc_min_size;
2206	int alloc_size;
2207
2208	if (!lock_taken)
2209		mutex_lock(&nlk->nl_cb_mutex);
2210	if (!nlk->cb_running) {
2211		err = -EINVAL;
2212		goto errout_skb;
2213	}
2214
2215	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2216		goto errout_skb;
2217
2218	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2219	 * required, but it makes sense to _attempt_ a 16K bytes allocation
2220	 * to reduce number of system calls on dump operations, if user
2221	 * ever provided a big enough buffer.
2222	 */
2223	cb = &nlk->cb;
2224	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2225
2226	max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2227	if (alloc_min_size < max_recvmsg_len) {
2228		alloc_size = max_recvmsg_len;
2229		skb = alloc_skb(alloc_size,
2230				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2231				__GFP_NOWARN | __GFP_NORETRY);
2232	}
2233	if (!skb) {
2234		alloc_size = alloc_min_size;
2235		skb = alloc_skb(alloc_size, GFP_KERNEL);
2236	}
2237	if (!skb)
2238		goto errout_skb;
2239
2240	/* Trim skb to allocated size. User is expected to provide buffer as
2241	 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2242	 * netlink_recvmsg())). dump will pack as many smaller messages as
2243	 * could fit within the allocated skb. skb is typically allocated
2244	 * with larger space than required (could be as much as near 2x the
2245	 * requested size with align to next power of 2 approach). Allowing
2246	 * dump to use the excess space makes it difficult for a user to have a
2247	 * reasonable static buffer based on the expected largest dump of a
2248	 * single netdev. The outcome is MSG_TRUNC error.
2249	 */
2250	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2251
2252	/* Make sure malicious BPF programs can not read unitialized memory
2253	 * from skb->head -> skb->data
2254	 */
2255	skb_reset_network_header(skb);
2256	skb_reset_mac_header(skb);
2257
2258	netlink_skb_set_owner_r(skb, sk);
2259
2260	if (nlk->dump_done_errno > 0) {
2261		struct mutex *extra_mutex = nlk->dump_cb_mutex;
2262
2263		cb->extack = &extack;
2264
2265		if (cb->flags & RTNL_FLAG_DUMP_UNLOCKED)
2266			extra_mutex = NULL;
2267		if (extra_mutex)
2268			mutex_lock(extra_mutex);
2269		nlk->dump_done_errno = cb->dump(skb, cb);
2270		if (extra_mutex)
2271			mutex_unlock(extra_mutex);
2272
2273		/* EMSGSIZE plus something already in the skb means
2274		 * that there's more to dump but current skb has filled up.
2275		 * If the callback really wants to return EMSGSIZE to user space
2276		 * it needs to do so again, on the next cb->dump() call,
2277		 * without putting data in the skb.
2278		 */
2279		if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2280			nlk->dump_done_errno = skb->len;
2281
2282		cb->extack = NULL;
2283	}
2284
2285	if (nlk->dump_done_errno > 0 ||
2286	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2287		mutex_unlock(&nlk->nl_cb_mutex);
2288
2289		if (sk_filter(sk, skb))
2290			kfree_skb(skb);
2291		else
2292			__netlink_sendskb(sk, skb);
2293		return 0;
2294	}
2295
2296	if (netlink_dump_done(nlk, skb, cb, &extack))
2297		goto errout_skb;
2298
2299#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2300	/* frag_list skb's data is used for compat tasks
2301	 * and the regular skb's data for normal (non-compat) tasks.
2302	 * See netlink_recvmsg().
2303	 */
2304	if (unlikely(skb_shinfo(skb)->frag_list)) {
2305		if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2306			goto errout_skb;
2307	}
2308#endif
2309
2310	if (sk_filter(sk, skb))
2311		kfree_skb(skb);
2312	else
2313		__netlink_sendskb(sk, skb);
2314
2315	if (cb->done)
2316		cb->done(cb);
2317
2318	WRITE_ONCE(nlk->cb_running, false);
2319	module = cb->module;
2320	skb = cb->skb;
2321	mutex_unlock(&nlk->nl_cb_mutex);
2322	module_put(module);
2323	consume_skb(skb);
2324	return 0;
2325
2326errout_skb:
2327	mutex_unlock(&nlk->nl_cb_mutex);
2328	kfree_skb(skb);
2329	return err;
2330}
2331
2332int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2333			 const struct nlmsghdr *nlh,
2334			 struct netlink_dump_control *control)
2335{
2336	struct netlink_callback *cb;
2337	struct netlink_sock *nlk;
2338	struct sock *sk;
2339	int ret;
2340
2341	refcount_inc(&skb->users);
2342
2343	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2344	if (sk == NULL) {
2345		ret = -ECONNREFUSED;
2346		goto error_free;
2347	}
2348
2349	nlk = nlk_sk(sk);
2350	mutex_lock(&nlk->nl_cb_mutex);
2351	/* A dump is in progress... */
2352	if (nlk->cb_running) {
2353		ret = -EBUSY;
2354		goto error_unlock;
2355	}
2356	/* add reference of module which cb->dump belongs to */
2357	if (!try_module_get(control->module)) {
2358		ret = -EPROTONOSUPPORT;
2359		goto error_unlock;
2360	}
2361
2362	cb = &nlk->cb;
2363	memset(cb, 0, sizeof(*cb));
2364	cb->dump = control->dump;
2365	cb->done = control->done;
2366	cb->nlh = nlh;
2367	cb->data = control->data;
2368	cb->module = control->module;
2369	cb->min_dump_alloc = control->min_dump_alloc;
2370	cb->flags = control->flags;
2371	cb->skb = skb;
2372
2373	cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2374
2375	if (control->start) {
2376		cb->extack = control->extack;
2377		ret = control->start(cb);
2378		cb->extack = NULL;
2379		if (ret)
2380			goto error_put;
2381	}
2382
2383	WRITE_ONCE(nlk->cb_running, true);
2384	nlk->dump_done_errno = INT_MAX;
2385
2386	ret = netlink_dump(sk, true);
 
 
2387
2388	sock_put(sk);
2389
2390	if (ret)
2391		return ret;
2392
2393	/* We successfully started a dump, by returning -EINTR we
2394	 * signal not to send ACK even if it was requested.
2395	 */
2396	return -EINTR;
2397
2398error_put:
2399	module_put(control->module);
2400error_unlock:
2401	sock_put(sk);
2402	mutex_unlock(&nlk->nl_cb_mutex);
2403error_free:
2404	kfree_skb(skb);
2405	return ret;
2406}
2407EXPORT_SYMBOL(__netlink_dump_start);
2408
2409static size_t
2410netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2411		    const struct netlink_ext_ack *extack)
2412{
2413	size_t tlvlen;
2414
2415	if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2416		return 0;
2417
2418	tlvlen = 0;
2419	if (extack->_msg)
2420		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2421	if (extack->cookie_len)
2422		tlvlen += nla_total_size(extack->cookie_len);
2423
2424	/* Following attributes are only reported as error (not warning) */
2425	if (!err)
2426		return tlvlen;
2427
2428	if (extack->bad_attr)
2429		tlvlen += nla_total_size(sizeof(u32));
2430	if (extack->policy)
2431		tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2432	if (extack->miss_type)
2433		tlvlen += nla_total_size(sizeof(u32));
2434	if (extack->miss_nest)
2435		tlvlen += nla_total_size(sizeof(u32));
2436
2437	return tlvlen;
2438}
2439
2440static void
2441netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2442		     struct nlmsghdr *nlh, int err,
2443		     const struct netlink_ext_ack *extack)
2444{
2445	if (extack->_msg)
2446		WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2447	if (extack->cookie_len)
2448		WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2449				extack->cookie_len, extack->cookie));
2450
2451	if (!err)
2452		return;
2453
2454	if (extack->bad_attr &&
2455	    !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2456		     (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2457		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2458				    (u8 *)extack->bad_attr - (u8 *)nlh));
2459	if (extack->policy)
2460		netlink_policy_dump_write_attr(skb, extack->policy,
2461					       NLMSGERR_ATTR_POLICY);
2462	if (extack->miss_type)
2463		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2464				    extack->miss_type));
2465	if (extack->miss_nest &&
2466	    !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2467		     (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2468		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2469				    (u8 *)extack->miss_nest - (u8 *)nlh));
2470}
2471
2472void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2473		 const struct netlink_ext_ack *extack)
2474{
2475	struct sk_buff *skb;
2476	struct nlmsghdr *rep;
2477	struct nlmsgerr *errmsg;
2478	size_t payload = sizeof(*errmsg);
2479	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2480	unsigned int flags = 0;
2481	size_t tlvlen;
2482
2483	/* Error messages get the original request appened, unless the user
2484	 * requests to cap the error message, and get extra error data if
2485	 * requested.
2486	 */
2487	if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2488		payload += nlmsg_len(nlh);
2489	else
2490		flags |= NLM_F_CAPPED;
2491
2492	tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2493	if (tlvlen)
2494		flags |= NLM_F_ACK_TLVS;
2495
2496	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2497	if (!skb)
2498		goto err_skb;
2499
2500	rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2501			NLMSG_ERROR, sizeof(*errmsg), flags);
2502	if (!rep)
2503		goto err_bad_put;
2504	errmsg = nlmsg_data(rep);
2505	errmsg->error = err;
2506	errmsg->msg = *nlh;
2507
2508	if (!(flags & NLM_F_CAPPED)) {
2509		if (!nlmsg_append(skb, nlmsg_len(nlh)))
2510			goto err_bad_put;
2511
2512		memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2513		       nlmsg_len(nlh));
2514	}
2515
2516	if (tlvlen)
2517		netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2518
2519	nlmsg_end(skb, rep);
2520
2521	nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2522
2523	return;
2524
2525err_bad_put:
2526	nlmsg_free(skb);
2527err_skb:
2528	WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2529	sk_error_report(NETLINK_CB(in_skb).sk);
2530}
2531EXPORT_SYMBOL(netlink_ack);
2532
2533int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2534						   struct nlmsghdr *,
2535						   struct netlink_ext_ack *))
2536{
2537	struct netlink_ext_ack extack;
2538	struct nlmsghdr *nlh;
2539	int err;
2540
2541	while (skb->len >= nlmsg_total_size(0)) {
2542		int msglen;
2543
2544		memset(&extack, 0, sizeof(extack));
2545		nlh = nlmsg_hdr(skb);
2546		err = 0;
2547
2548		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2549			return 0;
2550
2551		/* Only requests are handled by the kernel */
2552		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2553			goto ack;
2554
2555		/* Skip control messages */
2556		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2557			goto ack;
2558
2559		err = cb(skb, nlh, &extack);
2560		if (err == -EINTR)
2561			goto skip;
2562
2563ack:
2564		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2565			netlink_ack(skb, nlh, err, &extack);
2566
2567skip:
2568		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2569		if (msglen > skb->len)
2570			msglen = skb->len;
2571		skb_pull(skb, msglen);
2572	}
2573
2574	return 0;
2575}
2576EXPORT_SYMBOL(netlink_rcv_skb);
2577
2578/**
2579 * nlmsg_notify - send a notification netlink message
2580 * @sk: netlink socket to use
2581 * @skb: notification message
2582 * @portid: destination netlink portid for reports or 0
2583 * @group: destination multicast group or 0
2584 * @report: 1 to report back, 0 to disable
2585 * @flags: allocation flags
2586 */
2587int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2588		 unsigned int group, int report, gfp_t flags)
2589{
2590	int err = 0;
2591
2592	if (group) {
2593		int exclude_portid = 0;
2594
2595		if (report) {
2596			refcount_inc(&skb->users);
2597			exclude_portid = portid;
2598		}
2599
2600		/* errors reported via destination sk->sk_err, but propagate
2601		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2602		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2603		if (err == -ESRCH)
2604			err = 0;
2605	}
2606
2607	if (report) {
2608		int err2;
2609
2610		err2 = nlmsg_unicast(sk, skb, portid);
2611		if (!err)
2612			err = err2;
2613	}
2614
2615	return err;
2616}
2617EXPORT_SYMBOL(nlmsg_notify);
2618
2619#ifdef CONFIG_PROC_FS
2620struct nl_seq_iter {
2621	struct seq_net_private p;
2622	struct rhashtable_iter hti;
2623	int link;
2624};
2625
2626static void netlink_walk_start(struct nl_seq_iter *iter)
2627{
2628	rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2629	rhashtable_walk_start(&iter->hti);
2630}
2631
2632static void netlink_walk_stop(struct nl_seq_iter *iter)
2633{
2634	rhashtable_walk_stop(&iter->hti);
2635	rhashtable_walk_exit(&iter->hti);
2636}
2637
2638static void *__netlink_seq_next(struct seq_file *seq)
2639{
2640	struct nl_seq_iter *iter = seq->private;
2641	struct netlink_sock *nlk;
2642
2643	do {
2644		for (;;) {
2645			nlk = rhashtable_walk_next(&iter->hti);
2646
2647			if (IS_ERR(nlk)) {
2648				if (PTR_ERR(nlk) == -EAGAIN)
2649					continue;
2650
2651				return nlk;
2652			}
2653
2654			if (nlk)
2655				break;
2656
2657			netlink_walk_stop(iter);
2658			if (++iter->link >= MAX_LINKS)
2659				return NULL;
2660
2661			netlink_walk_start(iter);
2662		}
2663	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2664
2665	return nlk;
2666}
2667
2668static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2669	__acquires(RCU)
2670{
2671	struct nl_seq_iter *iter = seq->private;
2672	void *obj = SEQ_START_TOKEN;
2673	loff_t pos;
2674
2675	iter->link = 0;
2676
2677	netlink_walk_start(iter);
2678
2679	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2680		obj = __netlink_seq_next(seq);
2681
2682	return obj;
2683}
2684
2685static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2686{
2687	++*pos;
2688	return __netlink_seq_next(seq);
2689}
2690
2691static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2692{
2693	struct nl_seq_iter *iter = seq->private;
2694
2695	if (iter->link >= MAX_LINKS)
2696		return;
2697
2698	netlink_walk_stop(iter);
2699}
2700
2701
2702static int netlink_native_seq_show(struct seq_file *seq, void *v)
2703{
2704	if (v == SEQ_START_TOKEN) {
2705		seq_puts(seq,
2706			 "sk               Eth Pid        Groups   "
2707			 "Rmem     Wmem     Dump  Locks    Drops    Inode\n");
2708	} else {
2709		struct sock *s = v;
2710		struct netlink_sock *nlk = nlk_sk(s);
2711
2712		seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2713			   s,
2714			   s->sk_protocol,
2715			   nlk->portid,
2716			   nlk->groups ? (u32)nlk->groups[0] : 0,
2717			   sk_rmem_alloc_get(s),
2718			   sk_wmem_alloc_get(s),
2719			   READ_ONCE(nlk->cb_running),
2720			   refcount_read(&s->sk_refcnt),
2721			   atomic_read(&s->sk_drops),
2722			   sock_i_ino(s)
2723			);
2724
2725	}
2726	return 0;
2727}
2728
2729#ifdef CONFIG_BPF_SYSCALL
2730struct bpf_iter__netlink {
2731	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2732	__bpf_md_ptr(struct netlink_sock *, sk);
2733};
2734
2735DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2736
2737static int netlink_prog_seq_show(struct bpf_prog *prog,
2738				  struct bpf_iter_meta *meta,
2739				  void *v)
2740{
2741	struct bpf_iter__netlink ctx;
2742
2743	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2744	ctx.meta = meta;
2745	ctx.sk = nlk_sk((struct sock *)v);
2746	return bpf_iter_run_prog(prog, &ctx);
2747}
2748
2749static int netlink_seq_show(struct seq_file *seq, void *v)
2750{
2751	struct bpf_iter_meta meta;
2752	struct bpf_prog *prog;
2753
2754	meta.seq = seq;
2755	prog = bpf_iter_get_info(&meta, false);
2756	if (!prog)
2757		return netlink_native_seq_show(seq, v);
2758
2759	if (v != SEQ_START_TOKEN)
2760		return netlink_prog_seq_show(prog, &meta, v);
2761
2762	return 0;
2763}
2764
2765static void netlink_seq_stop(struct seq_file *seq, void *v)
2766{
2767	struct bpf_iter_meta meta;
2768	struct bpf_prog *prog;
2769
2770	if (!v) {
2771		meta.seq = seq;
2772		prog = bpf_iter_get_info(&meta, true);
2773		if (prog)
2774			(void)netlink_prog_seq_show(prog, &meta, v);
2775	}
2776
2777	netlink_native_seq_stop(seq, v);
2778}
2779#else
2780static int netlink_seq_show(struct seq_file *seq, void *v)
2781{
2782	return netlink_native_seq_show(seq, v);
2783}
2784
2785static void netlink_seq_stop(struct seq_file *seq, void *v)
2786{
2787	netlink_native_seq_stop(seq, v);
2788}
2789#endif
2790
2791static const struct seq_operations netlink_seq_ops = {
2792	.start  = netlink_seq_start,
2793	.next   = netlink_seq_next,
2794	.stop   = netlink_seq_stop,
2795	.show   = netlink_seq_show,
2796};
2797#endif
2798
2799int netlink_register_notifier(struct notifier_block *nb)
2800{
2801	return blocking_notifier_chain_register(&netlink_chain, nb);
2802}
2803EXPORT_SYMBOL(netlink_register_notifier);
2804
2805int netlink_unregister_notifier(struct notifier_block *nb)
2806{
2807	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2808}
2809EXPORT_SYMBOL(netlink_unregister_notifier);
2810
2811static const struct proto_ops netlink_ops = {
2812	.family =	PF_NETLINK,
2813	.owner =	THIS_MODULE,
2814	.release =	netlink_release,
2815	.bind =		netlink_bind,
2816	.connect =	netlink_connect,
2817	.socketpair =	sock_no_socketpair,
2818	.accept =	sock_no_accept,
2819	.getname =	netlink_getname,
2820	.poll =		datagram_poll,
2821	.ioctl =	netlink_ioctl,
2822	.listen =	sock_no_listen,
2823	.shutdown =	sock_no_shutdown,
2824	.setsockopt =	netlink_setsockopt,
2825	.getsockopt =	netlink_getsockopt,
2826	.sendmsg =	netlink_sendmsg,
2827	.recvmsg =	netlink_recvmsg,
2828	.mmap =		sock_no_mmap,
2829};
2830
2831static const struct net_proto_family netlink_family_ops = {
2832	.family = PF_NETLINK,
2833	.create = netlink_create,
2834	.owner	= THIS_MODULE,	/* for consistency 8) */
2835};
2836
2837static int __net_init netlink_net_init(struct net *net)
2838{
2839#ifdef CONFIG_PROC_FS
2840	if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2841			sizeof(struct nl_seq_iter)))
2842		return -ENOMEM;
2843#endif
2844	return 0;
2845}
2846
2847static void __net_exit netlink_net_exit(struct net *net)
2848{
2849#ifdef CONFIG_PROC_FS
2850	remove_proc_entry("netlink", net->proc_net);
2851#endif
2852}
2853
2854static void __init netlink_add_usersock_entry(void)
2855{
2856	struct listeners *listeners;
2857	int groups = 32;
2858
2859	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2860	if (!listeners)
2861		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2862
2863	netlink_table_grab();
2864
2865	nl_table[NETLINK_USERSOCK].groups = groups;
2866	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2867	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2868	nl_table[NETLINK_USERSOCK].registered = 1;
2869	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2870
2871	netlink_table_ungrab();
2872}
2873
2874static struct pernet_operations __net_initdata netlink_net_ops = {
2875	.init = netlink_net_init,
2876	.exit = netlink_net_exit,
2877};
2878
2879static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2880{
2881	const struct netlink_sock *nlk = data;
2882	struct netlink_compare_arg arg;
2883
2884	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2885	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2886}
2887
2888static const struct rhashtable_params netlink_rhashtable_params = {
2889	.head_offset = offsetof(struct netlink_sock, node),
2890	.key_len = netlink_compare_arg_len,
2891	.obj_hashfn = netlink_hash,
2892	.obj_cmpfn = netlink_compare,
2893	.automatic_shrinking = true,
2894};
2895
2896#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2897BTF_ID_LIST(btf_netlink_sock_id)
2898BTF_ID(struct, netlink_sock)
2899
2900static const struct bpf_iter_seq_info netlink_seq_info = {
2901	.seq_ops		= &netlink_seq_ops,
2902	.init_seq_private	= bpf_iter_init_seq_net,
2903	.fini_seq_private	= bpf_iter_fini_seq_net,
2904	.seq_priv_size		= sizeof(struct nl_seq_iter),
2905};
2906
2907static struct bpf_iter_reg netlink_reg_info = {
2908	.target			= "netlink",
2909	.ctx_arg_info_size	= 1,
2910	.ctx_arg_info		= {
2911		{ offsetof(struct bpf_iter__netlink, sk),
2912		  PTR_TO_BTF_ID_OR_NULL },
2913	},
2914	.seq_info		= &netlink_seq_info,
2915};
2916
2917static int __init bpf_iter_register(void)
2918{
2919	netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2920	return bpf_iter_reg_target(&netlink_reg_info);
2921}
2922#endif
2923
2924static int __init netlink_proto_init(void)
2925{
2926	int i;
2927	int err = proto_register(&netlink_proto, 0);
2928
2929	if (err != 0)
2930		goto out;
2931
2932#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2933	err = bpf_iter_register();
2934	if (err)
2935		goto out;
2936#endif
2937
2938	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2939
2940	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2941	if (!nl_table)
2942		goto panic;
2943
2944	for (i = 0; i < MAX_LINKS; i++) {
2945		if (rhashtable_init(&nl_table[i].hash,
2946				    &netlink_rhashtable_params) < 0) {
2947			while (--i > 0)
2948				rhashtable_destroy(&nl_table[i].hash);
2949			kfree(nl_table);
2950			goto panic;
2951		}
2952	}
2953
2954	netlink_add_usersock_entry();
2955
2956	sock_register(&netlink_family_ops);
2957	register_pernet_subsys(&netlink_net_ops);
2958	register_pernet_subsys(&netlink_tap_net_ops);
2959	/* The netlink device handler may be needed early. */
2960	rtnetlink_init();
2961out:
2962	return err;
2963panic:
2964	panic("netlink_init: Cannot allocate nl_table\n");
2965}
2966
2967core_initcall(netlink_proto_init);