Linux Audio

Check our new training course

Loading...
v4.6
 
   1#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   2
   3#include <linux/workqueue.h>
   4#include <linux/rtnetlink.h>
   5#include <linux/cache.h>
   6#include <linux/slab.h>
   7#include <linux/list.h>
   8#include <linux/delay.h>
   9#include <linux/sched.h>
  10#include <linux/idr.h>
  11#include <linux/rculist.h>
  12#include <linux/nsproxy.h>
  13#include <linux/fs.h>
  14#include <linux/proc_ns.h>
  15#include <linux/file.h>
  16#include <linux/export.h>
  17#include <linux/user_namespace.h>
  18#include <linux/net_namespace.h>
 
 
 
  19#include <net/sock.h>
  20#include <net/netlink.h>
  21#include <net/net_namespace.h>
  22#include <net/netns/generic.h>
  23
  24/*
  25 *	Our network namespace constructor/destructor lists
  26 */
  27
  28static LIST_HEAD(pernet_list);
  29static struct list_head *first_device = &pernet_list;
  30DEFINE_MUTEX(net_mutex);
  31
  32LIST_HEAD(net_namespace_list);
  33EXPORT_SYMBOL_GPL(net_namespace_list);
  34
 
 
 
 
 
 
 
 
  35struct net init_net = {
  36	.dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
 
 
 
 
  37};
  38EXPORT_SYMBOL(init_net);
  39
 
 
 
 
 
 
 
 
 
 
 
 
 
  40#define INITIAL_NET_GEN_PTRS	13 /* +1 for len +2 for rcu_head */
  41
  42static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  43
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  44static struct net_generic *net_alloc_generic(void)
  45{
  46	struct net_generic *ng;
  47	size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  48
  49	ng = kzalloc(generic_size, GFP_KERNEL);
  50	if (ng)
  51		ng->len = max_gen_ptrs;
  52
  53	return ng;
  54}
  55
  56static int net_assign_generic(struct net *net, int id, void *data)
  57{
  58	struct net_generic *ng, *old_ng;
  59
  60	BUG_ON(!mutex_is_locked(&net_mutex));
  61	BUG_ON(id == 0);
  62
  63	old_ng = rcu_dereference_protected(net->gen,
  64					   lockdep_is_held(&net_mutex));
  65	ng = old_ng;
  66	if (old_ng->len >= id)
  67		goto assign;
 
  68
  69	ng = net_alloc_generic();
  70	if (ng == NULL)
  71		return -ENOMEM;
  72
  73	/*
  74	 * Some synchronisation notes:
  75	 *
  76	 * The net_generic explores the net->gen array inside rcu
  77	 * read section. Besides once set the net->gen->ptr[x]
  78	 * pointer never changes (see rules in netns/generic.h).
  79	 *
  80	 * That said, we simply duplicate this array and schedule
  81	 * the old copy for kfree after a grace period.
  82	 */
  83
  84	memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
 
 
  85
  86	rcu_assign_pointer(net->gen, ng);
  87	kfree_rcu(old_ng, rcu);
  88assign:
  89	ng->ptr[id - 1] = data;
  90	return 0;
  91}
  92
  93static int ops_init(const struct pernet_operations *ops, struct net *net)
  94{
  95	int err = -ENOMEM;
  96	void *data = NULL;
  97
  98	if (ops->id && ops->size) {
  99		data = kzalloc(ops->size, GFP_KERNEL);
 100		if (!data)
 101			goto out;
 102
 103		err = net_assign_generic(net, *ops->id, data);
 104		if (err)
 105			goto cleanup;
 106	}
 107	err = 0;
 108	if (ops->init)
 109		err = ops->init(net);
 110	if (!err)
 111		return 0;
 112
 113cleanup:
 114	kfree(data);
 115
 116out:
 117	return err;
 118}
 119
 120static void ops_free(const struct pernet_operations *ops, struct net *net)
 121{
 122	if (ops->id && ops->size) {
 123		int id = *ops->id;
 124		kfree(net_generic(net, id));
 
 
 
 
 
 
 
 
 
 
 125	}
 126}
 127
 128static void ops_exit_list(const struct pernet_operations *ops,
 129			  struct list_head *net_exit_list)
 130{
 131	struct net *net;
 132	if (ops->exit) {
 133		list_for_each_entry(net, net_exit_list, exit_list)
 134			ops->exit(net);
 135	}
 136	if (ops->exit_batch)
 137		ops->exit_batch(net_exit_list);
 138}
 139
 140static void ops_free_list(const struct pernet_operations *ops,
 141			  struct list_head *net_exit_list)
 142{
 143	struct net *net;
 144	if (ops->size && ops->id) {
 145		list_for_each_entry(net, net_exit_list, exit_list)
 146			ops_free(ops, net);
 147	}
 148}
 149
 150/* should be called with nsid_lock held */
 151static int alloc_netid(struct net *net, struct net *peer, int reqid)
 152{
 153	int min = 0, max = 0;
 154
 155	if (reqid >= 0) {
 156		min = reqid;
 157		max = reqid + 1;
 158	}
 159
 160	return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
 161}
 162
 163/* This function is used by idr_for_each(). If net is equal to peer, the
 164 * function returns the id so that idr_for_each() stops. Because we cannot
 165 * returns the id 0 (idr_for_each() will not stop), we return the magic value
 166 * NET_ID_ZERO (-1) for it.
 167 */
 168#define NET_ID_ZERO -1
 169static int net_eq_idr(int id, void *net, void *peer)
 170{
 171	if (net_eq(net, peer))
 172		return id ? : NET_ID_ZERO;
 173	return 0;
 174}
 175
 176/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
 177 * is set to true, thus the caller knows that the new id must be notified via
 178 * rtnl.
 179 */
 180static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
 181{
 182	int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
 183	bool alloc_it = *alloc;
 184
 185	*alloc = false;
 186
 187	/* Magic value for id 0. */
 188	if (id == NET_ID_ZERO)
 189		return 0;
 190	if (id > 0)
 191		return id;
 192
 193	if (alloc_it) {
 194		id = alloc_netid(net, peer, -1);
 195		*alloc = true;
 196		return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
 197	}
 198
 199	return NETNSA_NSID_NOT_ASSIGNED;
 200}
 201
 202/* should be called with nsid_lock held */
 203static int __peernet2id(struct net *net, struct net *peer)
 204{
 205	bool no = false;
 206
 207	return __peernet2id_alloc(net, peer, &no);
 208}
 209
 210static void rtnl_net_notifyid(struct net *net, int cmd, int id);
 211/* This function returns the id of a peer netns. If no id is assigned, one will
 212 * be allocated and returned.
 213 */
 214int peernet2id_alloc(struct net *net, struct net *peer)
 215{
 216	unsigned long flags;
 217	bool alloc;
 218	int id;
 219
 220	spin_lock_irqsave(&net->nsid_lock, flags);
 221	alloc = atomic_read(&peer->count) == 0 ? false : true;
 222	id = __peernet2id_alloc(net, peer, &alloc);
 223	spin_unlock_irqrestore(&net->nsid_lock, flags);
 224	if (alloc && id >= 0)
 225		rtnl_net_notifyid(net, RTM_NEWNSID, id);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 226	return id;
 227}
 228EXPORT_SYMBOL(peernet2id_alloc);
 229
 230/* This function returns, if assigned, the id of a peer netns. */
 231int peernet2id(struct net *net, struct net *peer)
 232{
 233	unsigned long flags;
 234	int id;
 235
 236	spin_lock_irqsave(&net->nsid_lock, flags);
 237	id = __peernet2id(net, peer);
 238	spin_unlock_irqrestore(&net->nsid_lock, flags);
 
 239	return id;
 240}
 
 241
 242/* This function returns true is the peer netns has an id assigned into the
 243 * current netns.
 244 */
 245bool peernet_has_id(struct net *net, struct net *peer)
 246{
 247	return peernet2id(net, peer) >= 0;
 248}
 249
 250struct net *get_net_ns_by_id(struct net *net, int id)
 251{
 252	unsigned long flags;
 253	struct net *peer;
 254
 255	if (id < 0)
 256		return NULL;
 257
 258	rcu_read_lock();
 259	spin_lock_irqsave(&net->nsid_lock, flags);
 260	peer = idr_find(&net->netns_ids, id);
 261	if (peer)
 262		get_net(peer);
 263	spin_unlock_irqrestore(&net->nsid_lock, flags);
 264	rcu_read_unlock();
 265
 266	return peer;
 267}
 268
 269/*
 270 * setup_net runs the initializers for the network namespace object.
 271 */
 272static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 273{
 274	/* Must be called with net_mutex held */
 275	const struct pernet_operations *ops, *saved_ops;
 276	int error = 0;
 277	LIST_HEAD(net_exit_list);
 278
 279	atomic_set(&net->count, 1);
 280	atomic_set(&net->passive, 1);
 
 281	net->dev_base_seq = 1;
 282	net->user_ns = user_ns;
 283	idr_init(&net->netns_ids);
 284	spin_lock_init(&net->nsid_lock);
 
 285
 286	list_for_each_entry(ops, &pernet_list, list) {
 287		error = ops_init(ops, net);
 288		if (error < 0)
 289			goto out_undo;
 290	}
 
 
 
 291out:
 292	return error;
 293
 294out_undo:
 295	/* Walk through the list backwards calling the exit functions
 296	 * for the pernet modules whose init functions did not fail.
 297	 */
 298	list_add(&net->exit_list, &net_exit_list);
 299	saved_ops = ops;
 300	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 
 
 
 
 
 
 301		ops_exit_list(ops, &net_exit_list);
 302
 303	ops = saved_ops;
 304	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 305		ops_free_list(ops, &net_exit_list);
 306
 307	rcu_barrier();
 308	goto out;
 309}
 310
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 311
 312#ifdef CONFIG_NET_NS
 313static struct kmem_cache *net_cachep;
 
 
 
 
 
 
 
 
 
 
 314static struct workqueue_struct *netns_wq;
 315
 316static struct net *net_alloc(void)
 317{
 318	struct net *net = NULL;
 319	struct net_generic *ng;
 320
 321	ng = net_alloc_generic();
 322	if (!ng)
 323		goto out;
 324
 325	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 326	if (!net)
 327		goto out_free;
 328
 
 
 
 
 
 
 
 329	rcu_assign_pointer(net->gen, ng);
 330out:
 331	return net;
 332
 
 
 
 
 
 333out_free:
 334	kfree(ng);
 335	goto out;
 336}
 337
 338static void net_free(struct net *net)
 339{
 340	kfree(rcu_access_pointer(net->gen));
 341	kmem_cache_free(net_cachep, net);
 342}
 343
 344void net_drop_ns(void *p)
 345{
 346	struct net *ns = p;
 347	if (ns && atomic_dec_and_test(&ns->passive))
 348		net_free(ns);
 349}
 350
 351struct net *copy_net_ns(unsigned long flags,
 352			struct user_namespace *user_ns, struct net *old_net)
 353{
 
 354	struct net *net;
 355	int rv;
 356
 357	if (!(flags & CLONE_NEWNET))
 358		return get_net(old_net);
 359
 360	net = net_alloc();
 361	if (!net)
 362		return ERR_PTR(-ENOMEM);
 363
 
 
 
 
 
 
 
 364	get_user_ns(user_ns);
 365
 366	mutex_lock(&net_mutex);
 
 
 
 367	rv = setup_net(net, user_ns);
 368	if (rv == 0) {
 369		rtnl_lock();
 370		list_add_tail_rcu(&net->list, &net_namespace_list);
 371		rtnl_unlock();
 372	}
 373	mutex_unlock(&net_mutex);
 374	if (rv < 0) {
 
 
 375		put_user_ns(user_ns);
 376		net_drop_ns(net);
 
 
 377		return ERR_PTR(rv);
 378	}
 379	return net;
 380}
 381
 382static DEFINE_SPINLOCK(cleanup_list_lock);
 383static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 384
 385static void cleanup_net(struct work_struct *work)
 386{
 387	const struct pernet_operations *ops;
 388	struct net *net, *tmp;
 389	struct list_head net_kill_list;
 390	LIST_HEAD(net_exit_list);
 391
 392	/* Atomically snapshot the list of namespaces to cleanup */
 393	spin_lock_irq(&cleanup_list_lock);
 394	list_replace_init(&cleanup_list, &net_kill_list);
 395	spin_unlock_irq(&cleanup_list_lock);
 396
 397	mutex_lock(&net_mutex);
 398
 399	/* Don't let anyone else find us. */
 400	rtnl_lock();
 401	list_for_each_entry(net, &net_kill_list, cleanup_list) {
 402		list_del_rcu(&net->list);
 403		list_add_tail(&net->exit_list, &net_exit_list);
 404		for_each_net(tmp) {
 405			int id;
 406
 407			spin_lock_irq(&tmp->nsid_lock);
 408			id = __peernet2id(tmp, net);
 409			if (id >= 0)
 410				idr_remove(&tmp->netns_ids, id);
 411			spin_unlock_irq(&tmp->nsid_lock);
 412			if (id >= 0)
 413				rtnl_net_notifyid(tmp, RTM_DELNSID, id);
 414		}
 415		spin_lock_irq(&net->nsid_lock);
 416		idr_destroy(&net->netns_ids);
 417		spin_unlock_irq(&net->nsid_lock);
 418
 
 
 
 419	}
 420	rtnl_unlock();
 
 
 
 421
 422	/*
 423	 * Another CPU might be rcu-iterating the list, wait for it.
 424	 * This needs to be before calling the exit() notifiers, so
 425	 * the rcu_barrier() below isn't sufficient alone.
 
 426	 */
 427	synchronize_rcu();
 428
 429	/* Run all of the network namespace exit methods */
 430	list_for_each_entry_reverse(ops, &pernet_list, list)
 431		ops_exit_list(ops, &net_exit_list);
 432
 433	/* Free the net generic variables */
 434	list_for_each_entry_reverse(ops, &pernet_list, list)
 435		ops_free_list(ops, &net_exit_list);
 436
 437	mutex_unlock(&net_mutex);
 438
 439	/* Ensure there are no outstanding rcu callbacks using this
 440	 * network namespace.
 441	 */
 442	rcu_barrier();
 443
 444	/* Finally it is safe to free my network namespace structure */
 445	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 446		list_del_init(&net->exit_list);
 
 
 447		put_user_ns(net->user_ns);
 448		net_drop_ns(net);
 449	}
 450}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 451static DECLARE_WORK(net_cleanup_work, cleanup_net);
 452
 453void __put_net(struct net *net)
 454{
 455	/* Cleanup the network namespace in process context */
 456	unsigned long flags;
 457
 458	spin_lock_irqsave(&cleanup_list_lock, flags);
 459	list_add(&net->cleanup_list, &cleanup_list);
 460	spin_unlock_irqrestore(&cleanup_list_lock, flags);
 461
 462	queue_work(netns_wq, &net_cleanup_work);
 463}
 464EXPORT_SYMBOL_GPL(__put_net);
 465
 466struct net *get_net_ns_by_fd(int fd)
 467{
 468	struct file *file;
 469	struct ns_common *ns;
 470	struct net *net;
 471
 472	file = proc_ns_fget(fd);
 473	if (IS_ERR(file))
 474		return ERR_CAST(file);
 475
 476	ns = get_proc_ns(file_inode(file));
 477	if (ns->ops == &netns_operations)
 478		net = get_net(container_of(ns, struct net, ns));
 479	else
 480		net = ERR_PTR(-EINVAL);
 481
 482	fput(file);
 483	return net;
 484}
 485
 486#else
 487struct net *get_net_ns_by_fd(int fd)
 488{
 489	return ERR_PTR(-EINVAL);
 490}
 491#endif
 492EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
 493
 494struct net *get_net_ns_by_pid(pid_t pid)
 495{
 496	struct task_struct *tsk;
 497	struct net *net;
 498
 499	/* Lookup the network namespace */
 500	net = ERR_PTR(-ESRCH);
 501	rcu_read_lock();
 502	tsk = find_task_by_vpid(pid);
 503	if (tsk) {
 504		struct nsproxy *nsproxy;
 505		task_lock(tsk);
 506		nsproxy = tsk->nsproxy;
 507		if (nsproxy)
 508			net = get_net(nsproxy->net_ns);
 509		task_unlock(tsk);
 510	}
 511	rcu_read_unlock();
 512	return net;
 513}
 514EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 515
 516static __net_init int net_ns_net_init(struct net *net)
 517{
 518#ifdef CONFIG_NET_NS
 519	net->ns.ops = &netns_operations;
 520#endif
 521	return ns_alloc_inum(&net->ns);
 522}
 523
 524static __net_exit void net_ns_net_exit(struct net *net)
 525{
 526	ns_free_inum(&net->ns);
 527}
 528
 529static struct pernet_operations __net_initdata net_ns_ops = {
 530	.init = net_ns_net_init,
 531	.exit = net_ns_net_exit,
 532};
 533
 534static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
 535	[NETNSA_NONE]		= { .type = NLA_UNSPEC },
 536	[NETNSA_NSID]		= { .type = NLA_S32 },
 537	[NETNSA_PID]		= { .type = NLA_U32 },
 538	[NETNSA_FD]		= { .type = NLA_U32 },
 
 539};
 540
 541static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
 
 542{
 543	struct net *net = sock_net(skb->sk);
 544	struct nlattr *tb[NETNSA_MAX + 1];
 545	unsigned long flags;
 546	struct net *peer;
 547	int nsid, err;
 548
 549	err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
 550			  rtnl_net_policy);
 551	if (err < 0)
 552		return err;
 553	if (!tb[NETNSA_NSID])
 
 554		return -EINVAL;
 
 555	nsid = nla_get_s32(tb[NETNSA_NSID]);
 556
 557	if (tb[NETNSA_PID])
 558		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 559	else if (tb[NETNSA_FD])
 
 560		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 561	else
 
 
 562		return -EINVAL;
 563	if (IS_ERR(peer))
 
 
 
 564		return PTR_ERR(peer);
 
 565
 566	spin_lock_irqsave(&net->nsid_lock, flags);
 567	if (__peernet2id(net, peer) >= 0) {
 568		spin_unlock_irqrestore(&net->nsid_lock, flags);
 569		err = -EEXIST;
 
 
 
 570		goto out;
 571	}
 572
 573	err = alloc_netid(net, peer, nsid);
 574	spin_unlock_irqrestore(&net->nsid_lock, flags);
 575	if (err >= 0) {
 576		rtnl_net_notifyid(net, RTM_NEWNSID, err);
 
 577		err = 0;
 
 
 
 
 578	}
 579out:
 580	put_net(peer);
 581	return err;
 582}
 583
 584static int rtnl_net_get_size(void)
 585{
 586	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
 587	       + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
 
 588	       ;
 589}
 590
 591static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
 592			 int cmd, struct net *net, int nsid)
 
 
 
 
 
 
 
 
 
 593{
 594	struct nlmsghdr *nlh;
 595	struct rtgenmsg *rth;
 596
 597	nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
 
 598	if (!nlh)
 599		return -EMSGSIZE;
 600
 601	rth = nlmsg_data(nlh);
 602	rth->rtgen_family = AF_UNSPEC;
 603
 604	if (nla_put_s32(skb, NETNSA_NSID, nsid))
 
 
 
 
 605		goto nla_put_failure;
 606
 607	nlmsg_end(skb, nlh);
 608	return 0;
 609
 610nla_put_failure:
 611	nlmsg_cancel(skb, nlh);
 612	return -EMSGSIZE;
 613}
 614
 615static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 616{
 617	struct net *net = sock_net(skb->sk);
 618	struct nlattr *tb[NETNSA_MAX + 1];
 
 
 
 
 
 
 
 619	struct sk_buff *msg;
 620	struct net *peer;
 621	int err, id;
 622
 623	err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
 624			  rtnl_net_policy);
 625	if (err < 0)
 626		return err;
 627	if (tb[NETNSA_PID])
 628		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 629	else if (tb[NETNSA_FD])
 
 630		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 631	else
 
 
 
 
 
 
 
 632		return -EINVAL;
 
 633
 634	if (IS_ERR(peer))
 
 
 635		return PTR_ERR(peer);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 636
 637	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 638	if (!msg) {
 639		err = -ENOMEM;
 640		goto out;
 641	}
 642
 643	id = peernet2id(net, peer);
 644	err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
 645			    RTM_NEWNSID, net, id);
 646	if (err < 0)
 647		goto err_out;
 648
 649	err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
 650	goto out;
 651
 652err_out:
 653	nlmsg_free(msg);
 654out:
 
 
 655	put_net(peer);
 656	return err;
 657}
 658
 659struct rtnl_net_dump_cb {
 660	struct net *net;
 
 661	struct sk_buff *skb;
 662	struct netlink_callback *cb;
 663	int idx;
 664	int s_idx;
 665};
 666
 
 667static int rtnl_net_dumpid_one(int id, void *peer, void *data)
 668{
 669	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
 670	int ret;
 671
 672	if (net_cb->idx < net_cb->s_idx)
 673		goto cont;
 674
 675	ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
 676			    net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
 677			    RTM_NEWNSID, net_cb->net, id);
 
 678	if (ret < 0)
 679		return ret;
 680
 681cont:
 682	net_cb->idx++;
 683	return 0;
 684}
 685
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 686static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
 687{
 688	struct net *net = sock_net(skb->sk);
 689	struct rtnl_net_dump_cb net_cb = {
 690		.net = net,
 691		.skb = skb,
 692		.cb = cb,
 
 
 
 
 
 693		.idx = 0,
 694		.s_idx = cb->args[0],
 695	};
 696	unsigned long flags;
 697
 698	spin_lock_irqsave(&net->nsid_lock, flags);
 699	idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
 700	spin_unlock_irqrestore(&net->nsid_lock, flags);
 
 
 
 
 
 
 701
 702	cb->args[0] = net_cb.idx;
 703	return skb->len;
 
 
 
 704}
 705
 706static void rtnl_net_notifyid(struct net *net, int cmd, int id)
 707{
 
 
 
 
 
 
 
 708	struct sk_buff *msg;
 709	int err = -ENOMEM;
 710
 711	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 712	if (!msg)
 713		goto out;
 714
 715	err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
 716	if (err < 0)
 717		goto err_out;
 718
 719	rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
 720	return;
 721
 722err_out:
 723	nlmsg_free(msg);
 724out:
 725	rtnl_set_sk_err(net, RTNLGRP_NSID, err);
 726}
 727
 728static int __init net_ns_init(void)
 729{
 730	struct net_generic *ng;
 731
 732#ifdef CONFIG_NET_NS
 733	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 734					SMP_CACHE_BYTES,
 735					SLAB_PANIC, NULL);
 736
 737	/* Create workqueue for cleanup */
 738	netns_wq = create_singlethread_workqueue("netns");
 739	if (!netns_wq)
 740		panic("Could not create netns workq");
 741#endif
 742
 743	ng = net_alloc_generic();
 744	if (!ng)
 745		panic("Could not allocate generic netns");
 746
 747	rcu_assign_pointer(init_net.gen, ng);
 
 748
 749	mutex_lock(&net_mutex);
 750	if (setup_net(&init_net, &init_user_ns))
 751		panic("Could not setup the initial network namespace");
 752
 753	rtnl_lock();
 754	list_add_tail_rcu(&init_net.list, &net_namespace_list);
 755	rtnl_unlock();
 756
 757	mutex_unlock(&net_mutex);
 758
 759	register_pernet_subsys(&net_ns_ops);
 
 760
 761	rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
 
 762	rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
 763		      NULL);
 764
 765	return 0;
 766}
 767
 768pure_initcall(net_ns_init);
 769
 770#ifdef CONFIG_NET_NS
 771static int __register_pernet_operations(struct list_head *list,
 772					struct pernet_operations *ops)
 773{
 774	struct net *net;
 775	int error;
 776	LIST_HEAD(net_exit_list);
 777
 778	list_add_tail(&ops->list, list);
 779	if (ops->init || (ops->id && ops->size)) {
 
 
 
 780		for_each_net(net) {
 781			error = ops_init(ops, net);
 782			if (error)
 783				goto out_undo;
 784			list_add_tail(&net->exit_list, &net_exit_list);
 785		}
 786	}
 787	return 0;
 788
 789out_undo:
 790	/* If I have an error cleanup all namespaces I initialized */
 791	list_del(&ops->list);
 
 
 792	ops_exit_list(ops, &net_exit_list);
 793	ops_free_list(ops, &net_exit_list);
 794	return error;
 795}
 796
 797static void __unregister_pernet_operations(struct pernet_operations *ops)
 798{
 799	struct net *net;
 800	LIST_HEAD(net_exit_list);
 801
 802	list_del(&ops->list);
 
 803	for_each_net(net)
 804		list_add_tail(&net->exit_list, &net_exit_list);
 
 
 805	ops_exit_list(ops, &net_exit_list);
 806	ops_free_list(ops, &net_exit_list);
 807}
 808
 809#else
 810
 811static int __register_pernet_operations(struct list_head *list,
 812					struct pernet_operations *ops)
 813{
 
 
 
 
 
 814	return ops_init(ops, &init_net);
 815}
 816
 817static void __unregister_pernet_operations(struct pernet_operations *ops)
 818{
 819	LIST_HEAD(net_exit_list);
 820	list_add(&init_net.exit_list, &net_exit_list);
 821	ops_exit_list(ops, &net_exit_list);
 822	ops_free_list(ops, &net_exit_list);
 
 
 
 
 
 
 823}
 824
 825#endif /* CONFIG_NET_NS */
 826
 827static DEFINE_IDA(net_generic_ids);
 828
 829static int register_pernet_operations(struct list_head *list,
 830				      struct pernet_operations *ops)
 831{
 832	int error;
 833
 834	if (ops->id) {
 835again:
 836		error = ida_get_new_above(&net_generic_ids, 1, ops->id);
 837		if (error < 0) {
 838			if (error == -EAGAIN) {
 839				ida_pre_get(&net_generic_ids, GFP_KERNEL);
 840				goto again;
 841			}
 842			return error;
 843		}
 844		max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
 845	}
 846	error = __register_pernet_operations(list, ops);
 847	if (error) {
 848		rcu_barrier();
 849		if (ops->id)
 850			ida_remove(&net_generic_ids, *ops->id);
 851	}
 852
 853	return error;
 854}
 855
 856static void unregister_pernet_operations(struct pernet_operations *ops)
 857{
 858	
 859	__unregister_pernet_operations(ops);
 860	rcu_barrier();
 861	if (ops->id)
 862		ida_remove(&net_generic_ids, *ops->id);
 863}
 864
 865/**
 866 *      register_pernet_subsys - register a network namespace subsystem
 867 *	@ops:  pernet operations structure for the subsystem
 868 *
 869 *	Register a subsystem which has init and exit functions
 870 *	that are called when network namespaces are created and
 871 *	destroyed respectively.
 872 *
 873 *	When registered all network namespace init functions are
 874 *	called for every existing network namespace.  Allowing kernel
 875 *	modules to have a race free view of the set of network namespaces.
 876 *
 877 *	When a new network namespace is created all of the init
 878 *	methods are called in the order in which they were registered.
 879 *
 880 *	When a network namespace is destroyed all of the exit methods
 881 *	are called in the reverse of the order with which they were
 882 *	registered.
 883 */
 884int register_pernet_subsys(struct pernet_operations *ops)
 885{
 886	int error;
 887	mutex_lock(&net_mutex);
 888	error =  register_pernet_operations(first_device, ops);
 889	mutex_unlock(&net_mutex);
 890	return error;
 891}
 892EXPORT_SYMBOL_GPL(register_pernet_subsys);
 893
 894/**
 895 *      unregister_pernet_subsys - unregister a network namespace subsystem
 896 *	@ops: pernet operations structure to manipulate
 897 *
 898 *	Remove the pernet operations structure from the list to be
 899 *	used when network namespaces are created or destroyed.  In
 900 *	addition run the exit method for all existing network
 901 *	namespaces.
 902 */
 903void unregister_pernet_subsys(struct pernet_operations *ops)
 904{
 905	mutex_lock(&net_mutex);
 906	unregister_pernet_operations(ops);
 907	mutex_unlock(&net_mutex);
 908}
 909EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 910
 911/**
 912 *      register_pernet_device - register a network namespace device
 913 *	@ops:  pernet operations structure for the subsystem
 914 *
 915 *	Register a device which has init and exit functions
 916 *	that are called when network namespaces are created and
 917 *	destroyed respectively.
 918 *
 919 *	When registered all network namespace init functions are
 920 *	called for every existing network namespace.  Allowing kernel
 921 *	modules to have a race free view of the set of network namespaces.
 922 *
 923 *	When a new network namespace is created all of the init
 924 *	methods are called in the order in which they were registered.
 925 *
 926 *	When a network namespace is destroyed all of the exit methods
 927 *	are called in the reverse of the order with which they were
 928 *	registered.
 929 */
 930int register_pernet_device(struct pernet_operations *ops)
 931{
 932	int error;
 933	mutex_lock(&net_mutex);
 934	error = register_pernet_operations(&pernet_list, ops);
 935	if (!error && (first_device == &pernet_list))
 936		first_device = &ops->list;
 937	mutex_unlock(&net_mutex);
 938	return error;
 939}
 940EXPORT_SYMBOL_GPL(register_pernet_device);
 941
 942/**
 943 *      unregister_pernet_device - unregister a network namespace netdevice
 944 *	@ops: pernet operations structure to manipulate
 945 *
 946 *	Remove the pernet operations structure from the list to be
 947 *	used when network namespaces are created or destroyed.  In
 948 *	addition run the exit method for all existing network
 949 *	namespaces.
 950 */
 951void unregister_pernet_device(struct pernet_operations *ops)
 952{
 953	mutex_lock(&net_mutex);
 954	if (&ops->list == first_device)
 955		first_device = first_device->next;
 956	unregister_pernet_operations(ops);
 957	mutex_unlock(&net_mutex);
 958}
 959EXPORT_SYMBOL_GPL(unregister_pernet_device);
 960
 961#ifdef CONFIG_NET_NS
 962static struct ns_common *netns_get(struct task_struct *task)
 963{
 964	struct net *net = NULL;
 965	struct nsproxy *nsproxy;
 966
 967	task_lock(task);
 968	nsproxy = task->nsproxy;
 969	if (nsproxy)
 970		net = get_net(nsproxy->net_ns);
 971	task_unlock(task);
 972
 973	return net ? &net->ns : NULL;
 974}
 975
 976static inline struct net *to_net_ns(struct ns_common *ns)
 977{
 978	return container_of(ns, struct net, ns);
 979}
 980
 981static void netns_put(struct ns_common *ns)
 982{
 983	put_net(to_net_ns(ns));
 984}
 985
 986static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
 987{
 
 988	struct net *net = to_net_ns(ns);
 989
 990	if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
 991	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 992		return -EPERM;
 993
 994	put_net(nsproxy->net_ns);
 995	nsproxy->net_ns = get_net(net);
 996	return 0;
 997}
 998
 
 
 
 
 
 999const struct proc_ns_operations netns_operations = {
1000	.name		= "net",
1001	.type		= CLONE_NEWNET,
1002	.get		= netns_get,
1003	.put		= netns_put,
1004	.install	= netns_install,
 
1005};
1006#endif
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   3
   4#include <linux/workqueue.h>
   5#include <linux/rtnetlink.h>
   6#include <linux/cache.h>
   7#include <linux/slab.h>
   8#include <linux/list.h>
   9#include <linux/delay.h>
  10#include <linux/sched.h>
  11#include <linux/idr.h>
  12#include <linux/rculist.h>
  13#include <linux/nsproxy.h>
  14#include <linux/fs.h>
  15#include <linux/proc_ns.h>
  16#include <linux/file.h>
  17#include <linux/export.h>
  18#include <linux/user_namespace.h>
  19#include <linux/net_namespace.h>
  20#include <linux/sched/task.h>
  21#include <linux/uidgid.h>
  22
  23#include <net/sock.h>
  24#include <net/netlink.h>
  25#include <net/net_namespace.h>
  26#include <net/netns/generic.h>
  27
  28/*
  29 *	Our network namespace constructor/destructor lists
  30 */
  31
  32static LIST_HEAD(pernet_list);
  33static struct list_head *first_device = &pernet_list;
 
  34
  35LIST_HEAD(net_namespace_list);
  36EXPORT_SYMBOL_GPL(net_namespace_list);
  37
  38/* Protects net_namespace_list. Nests iside rtnl_lock() */
  39DECLARE_RWSEM(net_rwsem);
  40EXPORT_SYMBOL_GPL(net_rwsem);
  41
  42#ifdef CONFIG_KEYS
  43static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
  44#endif
  45
  46struct net init_net = {
  47	.count		= REFCOUNT_INIT(1),
  48	.dev_base_head	= LIST_HEAD_INIT(init_net.dev_base_head),
  49#ifdef CONFIG_KEYS
  50	.key_domain	= &init_net_key_domain,
  51#endif
  52};
  53EXPORT_SYMBOL(init_net);
  54
  55static bool init_net_initialized;
  56/*
  57 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
  58 * init_net_initialized and first_device pointer.
  59 * This is internal net namespace object. Please, don't use it
  60 * outside.
  61 */
  62DECLARE_RWSEM(pernet_ops_rwsem);
  63EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
  64
  65#define MIN_PERNET_OPS_ID	\
  66	((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
  67
  68#define INITIAL_NET_GEN_PTRS	13 /* +1 for len +2 for rcu_head */
  69
  70static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  71
  72static atomic64_t cookie_gen;
  73
  74u64 net_gen_cookie(struct net *net)
  75{
  76	while (1) {
  77		u64 res = atomic64_read(&net->net_cookie);
  78
  79		if (res)
  80			return res;
  81		res = atomic64_inc_return(&cookie_gen);
  82		atomic64_cmpxchg(&net->net_cookie, 0, res);
  83	}
  84}
  85
  86static struct net_generic *net_alloc_generic(void)
  87{
  88	struct net_generic *ng;
  89	unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  90
  91	ng = kzalloc(generic_size, GFP_KERNEL);
  92	if (ng)
  93		ng->s.len = max_gen_ptrs;
  94
  95	return ng;
  96}
  97
  98static int net_assign_generic(struct net *net, unsigned int id, void *data)
  99{
 100	struct net_generic *ng, *old_ng;
 101
 102	BUG_ON(id < MIN_PERNET_OPS_ID);
 
 103
 104	old_ng = rcu_dereference_protected(net->gen,
 105					   lockdep_is_held(&pernet_ops_rwsem));
 106	if (old_ng->s.len > id) {
 107		old_ng->ptr[id] = data;
 108		return 0;
 109	}
 110
 111	ng = net_alloc_generic();
 112	if (ng == NULL)
 113		return -ENOMEM;
 114
 115	/*
 116	 * Some synchronisation notes:
 117	 *
 118	 * The net_generic explores the net->gen array inside rcu
 119	 * read section. Besides once set the net->gen->ptr[x]
 120	 * pointer never changes (see rules in netns/generic.h).
 121	 *
 122	 * That said, we simply duplicate this array and schedule
 123	 * the old copy for kfree after a grace period.
 124	 */
 125
 126	memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
 127	       (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
 128	ng->ptr[id] = data;
 129
 130	rcu_assign_pointer(net->gen, ng);
 131	kfree_rcu(old_ng, s.rcu);
 
 
 132	return 0;
 133}
 134
 135static int ops_init(const struct pernet_operations *ops, struct net *net)
 136{
 137	int err = -ENOMEM;
 138	void *data = NULL;
 139
 140	if (ops->id && ops->size) {
 141		data = kzalloc(ops->size, GFP_KERNEL);
 142		if (!data)
 143			goto out;
 144
 145		err = net_assign_generic(net, *ops->id, data);
 146		if (err)
 147			goto cleanup;
 148	}
 149	err = 0;
 150	if (ops->init)
 151		err = ops->init(net);
 152	if (!err)
 153		return 0;
 154
 155cleanup:
 156	kfree(data);
 157
 158out:
 159	return err;
 160}
 161
 162static void ops_free(const struct pernet_operations *ops, struct net *net)
 163{
 164	if (ops->id && ops->size) {
 165		kfree(net_generic(net, *ops->id));
 166	}
 167}
 168
 169static void ops_pre_exit_list(const struct pernet_operations *ops,
 170			      struct list_head *net_exit_list)
 171{
 172	struct net *net;
 173
 174	if (ops->pre_exit) {
 175		list_for_each_entry(net, net_exit_list, exit_list)
 176			ops->pre_exit(net);
 177	}
 178}
 179
 180static void ops_exit_list(const struct pernet_operations *ops,
 181			  struct list_head *net_exit_list)
 182{
 183	struct net *net;
 184	if (ops->exit) {
 185		list_for_each_entry(net, net_exit_list, exit_list)
 186			ops->exit(net);
 187	}
 188	if (ops->exit_batch)
 189		ops->exit_batch(net_exit_list);
 190}
 191
 192static void ops_free_list(const struct pernet_operations *ops,
 193			  struct list_head *net_exit_list)
 194{
 195	struct net *net;
 196	if (ops->size && ops->id) {
 197		list_for_each_entry(net, net_exit_list, exit_list)
 198			ops_free(ops, net);
 199	}
 200}
 201
 202/* should be called with nsid_lock held */
 203static int alloc_netid(struct net *net, struct net *peer, int reqid)
 204{
 205	int min = 0, max = 0;
 206
 207	if (reqid >= 0) {
 208		min = reqid;
 209		max = reqid + 1;
 210	}
 211
 212	return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
 213}
 214
 215/* This function is used by idr_for_each(). If net is equal to peer, the
 216 * function returns the id so that idr_for_each() stops. Because we cannot
 217 * returns the id 0 (idr_for_each() will not stop), we return the magic value
 218 * NET_ID_ZERO (-1) for it.
 219 */
 220#define NET_ID_ZERO -1
 221static int net_eq_idr(int id, void *net, void *peer)
 222{
 223	if (net_eq(net, peer))
 224		return id ? : NET_ID_ZERO;
 225	return 0;
 226}
 227
 228/* Must be called from RCU-critical section or with nsid_lock held */
 229static int __peernet2id(const struct net *net, struct net *peer)
 
 
 
 230{
 231	int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
 
 
 
 232
 233	/* Magic value for id 0. */
 234	if (id == NET_ID_ZERO)
 235		return 0;
 236	if (id > 0)
 237		return id;
 238
 
 
 
 
 
 
 239	return NETNSA_NSID_NOT_ASSIGNED;
 240}
 241
 242static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
 243			      struct nlmsghdr *nlh, gfp_t gfp);
 
 
 
 
 
 
 
 244/* This function returns the id of a peer netns. If no id is assigned, one will
 245 * be allocated and returned.
 246 */
 247int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
 248{
 
 
 249	int id;
 250
 251	if (refcount_read(&net->count) == 0)
 252		return NETNSA_NSID_NOT_ASSIGNED;
 253
 254	spin_lock_bh(&net->nsid_lock);
 255	id = __peernet2id(net, peer);
 256	if (id >= 0) {
 257		spin_unlock_bh(&net->nsid_lock);
 258		return id;
 259	}
 260
 261	/* When peer is obtained from RCU lists, we may race with
 262	 * its cleanup. Check whether it's alive, and this guarantees
 263	 * we never hash a peer back to net->netns_ids, after it has
 264	 * just been idr_remove()'d from there in cleanup_net().
 265	 */
 266	if (!maybe_get_net(peer)) {
 267		spin_unlock_bh(&net->nsid_lock);
 268		return NETNSA_NSID_NOT_ASSIGNED;
 269	}
 270
 271	id = alloc_netid(net, peer, -1);
 272	spin_unlock_bh(&net->nsid_lock);
 273
 274	put_net(peer);
 275	if (id < 0)
 276		return NETNSA_NSID_NOT_ASSIGNED;
 277
 278	rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
 279
 280	return id;
 281}
 282EXPORT_SYMBOL_GPL(peernet2id_alloc);
 283
 284/* This function returns, if assigned, the id of a peer netns. */
 285int peernet2id(const struct net *net, struct net *peer)
 286{
 
 287	int id;
 288
 289	rcu_read_lock();
 290	id = __peernet2id(net, peer);
 291	rcu_read_unlock();
 292
 293	return id;
 294}
 295EXPORT_SYMBOL(peernet2id);
 296
 297/* This function returns true is the peer netns has an id assigned into the
 298 * current netns.
 299 */
 300bool peernet_has_id(const struct net *net, struct net *peer)
 301{
 302	return peernet2id(net, peer) >= 0;
 303}
 304
 305struct net *get_net_ns_by_id(const struct net *net, int id)
 306{
 
 307	struct net *peer;
 308
 309	if (id < 0)
 310		return NULL;
 311
 312	rcu_read_lock();
 
 313	peer = idr_find(&net->netns_ids, id);
 314	if (peer)
 315		peer = maybe_get_net(peer);
 
 316	rcu_read_unlock();
 317
 318	return peer;
 319}
 320
 321/*
 322 * setup_net runs the initializers for the network namespace object.
 323 */
 324static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 325{
 326	/* Must be called with pernet_ops_rwsem held */
 327	const struct pernet_operations *ops, *saved_ops;
 328	int error = 0;
 329	LIST_HEAD(net_exit_list);
 330
 331	refcount_set(&net->count, 1);
 332	refcount_set(&net->passive, 1);
 333	get_random_bytes(&net->hash_mix, sizeof(u32));
 334	net->dev_base_seq = 1;
 335	net->user_ns = user_ns;
 336	idr_init(&net->netns_ids);
 337	spin_lock_init(&net->nsid_lock);
 338	mutex_init(&net->ipv4.ra_mutex);
 339
 340	list_for_each_entry(ops, &pernet_list, list) {
 341		error = ops_init(ops, net);
 342		if (error < 0)
 343			goto out_undo;
 344	}
 345	down_write(&net_rwsem);
 346	list_add_tail_rcu(&net->list, &net_namespace_list);
 347	up_write(&net_rwsem);
 348out:
 349	return error;
 350
 351out_undo:
 352	/* Walk through the list backwards calling the exit functions
 353	 * for the pernet modules whose init functions did not fail.
 354	 */
 355	list_add(&net->exit_list, &net_exit_list);
 356	saved_ops = ops;
 357	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 358		ops_pre_exit_list(ops, &net_exit_list);
 359
 360	synchronize_rcu();
 361
 362	ops = saved_ops;
 363	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 364		ops_exit_list(ops, &net_exit_list);
 365
 366	ops = saved_ops;
 367	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 368		ops_free_list(ops, &net_exit_list);
 369
 370	rcu_barrier();
 371	goto out;
 372}
 373
 374static int __net_init net_defaults_init_net(struct net *net)
 375{
 376	net->core.sysctl_somaxconn = SOMAXCONN;
 377	return 0;
 378}
 379
 380static struct pernet_operations net_defaults_ops = {
 381	.init = net_defaults_init_net,
 382};
 383
 384static __init int net_defaults_init(void)
 385{
 386	if (register_pernet_subsys(&net_defaults_ops))
 387		panic("Cannot initialize net default settings");
 388
 389	return 0;
 390}
 391
 392core_initcall(net_defaults_init);
 393
 394#ifdef CONFIG_NET_NS
 395static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
 396{
 397	return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
 398}
 399
 400static void dec_net_namespaces(struct ucounts *ucounts)
 401{
 402	dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
 403}
 404
 405static struct kmem_cache *net_cachep __ro_after_init;
 406static struct workqueue_struct *netns_wq;
 407
 408static struct net *net_alloc(void)
 409{
 410	struct net *net = NULL;
 411	struct net_generic *ng;
 412
 413	ng = net_alloc_generic();
 414	if (!ng)
 415		goto out;
 416
 417	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 418	if (!net)
 419		goto out_free;
 420
 421#ifdef CONFIG_KEYS
 422	net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
 423	if (!net->key_domain)
 424		goto out_free_2;
 425	refcount_set(&net->key_domain->usage, 1);
 426#endif
 427
 428	rcu_assign_pointer(net->gen, ng);
 429out:
 430	return net;
 431
 432#ifdef CONFIG_KEYS
 433out_free_2:
 434	kmem_cache_free(net_cachep, net);
 435	net = NULL;
 436#endif
 437out_free:
 438	kfree(ng);
 439	goto out;
 440}
 441
 442static void net_free(struct net *net)
 443{
 444	kfree(rcu_access_pointer(net->gen));
 445	kmem_cache_free(net_cachep, net);
 446}
 447
 448void net_drop_ns(void *p)
 449{
 450	struct net *ns = p;
 451	if (ns && refcount_dec_and_test(&ns->passive))
 452		net_free(ns);
 453}
 454
 455struct net *copy_net_ns(unsigned long flags,
 456			struct user_namespace *user_ns, struct net *old_net)
 457{
 458	struct ucounts *ucounts;
 459	struct net *net;
 460	int rv;
 461
 462	if (!(flags & CLONE_NEWNET))
 463		return get_net(old_net);
 464
 465	ucounts = inc_net_namespaces(user_ns);
 466	if (!ucounts)
 467		return ERR_PTR(-ENOSPC);
 468
 469	net = net_alloc();
 470	if (!net) {
 471		rv = -ENOMEM;
 472		goto dec_ucounts;
 473	}
 474	refcount_set(&net->passive, 1);
 475	net->ucounts = ucounts;
 476	get_user_ns(user_ns);
 477
 478	rv = down_read_killable(&pernet_ops_rwsem);
 479	if (rv < 0)
 480		goto put_userns;
 481
 482	rv = setup_net(net, user_ns);
 483
 484	up_read(&pernet_ops_rwsem);
 485
 
 
 
 486	if (rv < 0) {
 487put_userns:
 488		key_remove_domain(net->key_domain);
 489		put_user_ns(user_ns);
 490		net_drop_ns(net);
 491dec_ucounts:
 492		dec_net_namespaces(ucounts);
 493		return ERR_PTR(rv);
 494	}
 495	return net;
 496}
 497
 498/**
 499 * net_ns_get_ownership - get sysfs ownership data for @net
 500 * @net: network namespace in question (can be NULL)
 501 * @uid: kernel user ID for sysfs objects
 502 * @gid: kernel group ID for sysfs objects
 503 *
 504 * Returns the uid/gid pair of root in the user namespace associated with the
 505 * given network namespace.
 506 */
 507void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
 508{
 509	if (net) {
 510		kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
 511		kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
 512
 513		if (uid_valid(ns_root_uid))
 514			*uid = ns_root_uid;
 515
 516		if (gid_valid(ns_root_gid))
 517			*gid = ns_root_gid;
 518	} else {
 519		*uid = GLOBAL_ROOT_UID;
 520		*gid = GLOBAL_ROOT_GID;
 521	}
 522}
 523EXPORT_SYMBOL_GPL(net_ns_get_ownership);
 524
 525static void unhash_nsid(struct net *net, struct net *last)
 526{
 527	struct net *tmp;
 528	/* This function is only called from cleanup_net() work,
 529	 * and this work is the only process, that may delete
 530	 * a net from net_namespace_list. So, when the below
 531	 * is executing, the list may only grow. Thus, we do not
 532	 * use for_each_net_rcu() or net_rwsem.
 533	 */
 534	for_each_net(tmp) {
 535		int id;
 536
 537		spin_lock_bh(&tmp->nsid_lock);
 538		id = __peernet2id(tmp, net);
 539		if (id >= 0)
 540			idr_remove(&tmp->netns_ids, id);
 541		spin_unlock_bh(&tmp->nsid_lock);
 542		if (id >= 0)
 543			rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
 544					  GFP_KERNEL);
 545		if (tmp == last)
 546			break;
 547	}
 548	spin_lock_bh(&net->nsid_lock);
 549	idr_destroy(&net->netns_ids);
 550	spin_unlock_bh(&net->nsid_lock);
 551}
 552
 553static LLIST_HEAD(cleanup_list);
 554
 555static void cleanup_net(struct work_struct *work)
 556{
 557	const struct pernet_operations *ops;
 558	struct net *net, *tmp, *last;
 559	struct llist_node *net_kill_list;
 560	LIST_HEAD(net_exit_list);
 561
 562	/* Atomically snapshot the list of namespaces to cleanup */
 563	net_kill_list = llist_del_all(&cleanup_list);
 
 
 564
 565	down_read(&pernet_ops_rwsem);
 566
 567	/* Don't let anyone else find us. */
 568	down_write(&net_rwsem);
 569	llist_for_each_entry(net, net_kill_list, cleanup_list)
 570		list_del_rcu(&net->list);
 571	/* Cache last net. After we unlock rtnl, no one new net
 572	 * added to net_namespace_list can assign nsid pointer
 573	 * to a net from net_kill_list (see peernet2id_alloc()).
 574	 * So, we skip them in unhash_nsid().
 575	 *
 576	 * Note, that unhash_nsid() does not delete nsid links
 577	 * between net_kill_list's nets, as they've already
 578	 * deleted from net_namespace_list. But, this would be
 579	 * useless anyway, as netns_ids are destroyed there.
 580	 */
 581	last = list_last_entry(&net_namespace_list, struct net, list);
 582	up_write(&net_rwsem);
 
 
 
 583
 584	llist_for_each_entry(net, net_kill_list, cleanup_list) {
 585		unhash_nsid(net, last);
 586		list_add_tail(&net->exit_list, &net_exit_list);
 587	}
 588
 589	/* Run all of the network namespace pre_exit methods */
 590	list_for_each_entry_reverse(ops, &pernet_list, list)
 591		ops_pre_exit_list(ops, &net_exit_list);
 592
 593	/*
 594	 * Another CPU might be rcu-iterating the list, wait for it.
 595	 * This needs to be before calling the exit() notifiers, so
 596	 * the rcu_barrier() below isn't sufficient alone.
 597	 * Also the pre_exit() and exit() methods need this barrier.
 598	 */
 599	synchronize_rcu();
 600
 601	/* Run all of the network namespace exit methods */
 602	list_for_each_entry_reverse(ops, &pernet_list, list)
 603		ops_exit_list(ops, &net_exit_list);
 604
 605	/* Free the net generic variables */
 606	list_for_each_entry_reverse(ops, &pernet_list, list)
 607		ops_free_list(ops, &net_exit_list);
 608
 609	up_read(&pernet_ops_rwsem);
 610
 611	/* Ensure there are no outstanding rcu callbacks using this
 612	 * network namespace.
 613	 */
 614	rcu_barrier();
 615
 616	/* Finally it is safe to free my network namespace structure */
 617	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 618		list_del_init(&net->exit_list);
 619		dec_net_namespaces(net->ucounts);
 620		key_remove_domain(net->key_domain);
 621		put_user_ns(net->user_ns);
 622		net_drop_ns(net);
 623	}
 624}
 625
 626/**
 627 * net_ns_barrier - wait until concurrent net_cleanup_work is done
 628 *
 629 * cleanup_net runs from work queue and will first remove namespaces
 630 * from the global list, then run net exit functions.
 631 *
 632 * Call this in module exit path to make sure that all netns
 633 * ->exit ops have been invoked before the function is removed.
 634 */
 635void net_ns_barrier(void)
 636{
 637	down_write(&pernet_ops_rwsem);
 638	up_write(&pernet_ops_rwsem);
 639}
 640EXPORT_SYMBOL(net_ns_barrier);
 641
 642static DECLARE_WORK(net_cleanup_work, cleanup_net);
 643
 644void __put_net(struct net *net)
 645{
 646	/* Cleanup the network namespace in process context */
 647	if (llist_add(&net->cleanup_list, &cleanup_list))
 648		queue_work(netns_wq, &net_cleanup_work);
 
 
 
 
 
 649}
 650EXPORT_SYMBOL_GPL(__put_net);
 651
 652struct net *get_net_ns_by_fd(int fd)
 653{
 654	struct file *file;
 655	struct ns_common *ns;
 656	struct net *net;
 657
 658	file = proc_ns_fget(fd);
 659	if (IS_ERR(file))
 660		return ERR_CAST(file);
 661
 662	ns = get_proc_ns(file_inode(file));
 663	if (ns->ops == &netns_operations)
 664		net = get_net(container_of(ns, struct net, ns));
 665	else
 666		net = ERR_PTR(-EINVAL);
 667
 668	fput(file);
 669	return net;
 670}
 671
 672#else
 673struct net *get_net_ns_by_fd(int fd)
 674{
 675	return ERR_PTR(-EINVAL);
 676}
 677#endif
 678EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
 679
 680struct net *get_net_ns_by_pid(pid_t pid)
 681{
 682	struct task_struct *tsk;
 683	struct net *net;
 684
 685	/* Lookup the network namespace */
 686	net = ERR_PTR(-ESRCH);
 687	rcu_read_lock();
 688	tsk = find_task_by_vpid(pid);
 689	if (tsk) {
 690		struct nsproxy *nsproxy;
 691		task_lock(tsk);
 692		nsproxy = tsk->nsproxy;
 693		if (nsproxy)
 694			net = get_net(nsproxy->net_ns);
 695		task_unlock(tsk);
 696	}
 697	rcu_read_unlock();
 698	return net;
 699}
 700EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 701
 702static __net_init int net_ns_net_init(struct net *net)
 703{
 704#ifdef CONFIG_NET_NS
 705	net->ns.ops = &netns_operations;
 706#endif
 707	return ns_alloc_inum(&net->ns);
 708}
 709
 710static __net_exit void net_ns_net_exit(struct net *net)
 711{
 712	ns_free_inum(&net->ns);
 713}
 714
 715static struct pernet_operations __net_initdata net_ns_ops = {
 716	.init = net_ns_net_init,
 717	.exit = net_ns_net_exit,
 718};
 719
 720static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
 721	[NETNSA_NONE]		= { .type = NLA_UNSPEC },
 722	[NETNSA_NSID]		= { .type = NLA_S32 },
 723	[NETNSA_PID]		= { .type = NLA_U32 },
 724	[NETNSA_FD]		= { .type = NLA_U32 },
 725	[NETNSA_TARGET_NSID]	= { .type = NLA_S32 },
 726};
 727
 728static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
 729			  struct netlink_ext_ack *extack)
 730{
 731	struct net *net = sock_net(skb->sk);
 732	struct nlattr *tb[NETNSA_MAX + 1];
 733	struct nlattr *nla;
 734	struct net *peer;
 735	int nsid, err;
 736
 737	err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
 738				     NETNSA_MAX, rtnl_net_policy, extack);
 739	if (err < 0)
 740		return err;
 741	if (!tb[NETNSA_NSID]) {
 742		NL_SET_ERR_MSG(extack, "nsid is missing");
 743		return -EINVAL;
 744	}
 745	nsid = nla_get_s32(tb[NETNSA_NSID]);
 746
 747	if (tb[NETNSA_PID]) {
 748		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 749		nla = tb[NETNSA_PID];
 750	} else if (tb[NETNSA_FD]) {
 751		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 752		nla = tb[NETNSA_FD];
 753	} else {
 754		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 755		return -EINVAL;
 756	}
 757	if (IS_ERR(peer)) {
 758		NL_SET_BAD_ATTR(extack, nla);
 759		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 760		return PTR_ERR(peer);
 761	}
 762
 763	spin_lock_bh(&net->nsid_lock);
 764	if (__peernet2id(net, peer) >= 0) {
 765		spin_unlock_bh(&net->nsid_lock);
 766		err = -EEXIST;
 767		NL_SET_BAD_ATTR(extack, nla);
 768		NL_SET_ERR_MSG(extack,
 769			       "Peer netns already has a nsid assigned");
 770		goto out;
 771	}
 772
 773	err = alloc_netid(net, peer, nsid);
 774	spin_unlock_bh(&net->nsid_lock);
 775	if (err >= 0) {
 776		rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
 777				  nlh, GFP_KERNEL);
 778		err = 0;
 779	} else if (err == -ENOSPC && nsid >= 0) {
 780		err = -EEXIST;
 781		NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
 782		NL_SET_ERR_MSG(extack, "The specified nsid is already used");
 783	}
 784out:
 785	put_net(peer);
 786	return err;
 787}
 788
 789static int rtnl_net_get_size(void)
 790{
 791	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
 792	       + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
 793	       + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
 794	       ;
 795}
 796
 797struct net_fill_args {
 798	u32 portid;
 799	u32 seq;
 800	int flags;
 801	int cmd;
 802	int nsid;
 803	bool add_ref;
 804	int ref_nsid;
 805};
 806
 807static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
 808{
 809	struct nlmsghdr *nlh;
 810	struct rtgenmsg *rth;
 811
 812	nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
 813			args->flags);
 814	if (!nlh)
 815		return -EMSGSIZE;
 816
 817	rth = nlmsg_data(nlh);
 818	rth->rtgen_family = AF_UNSPEC;
 819
 820	if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
 821		goto nla_put_failure;
 822
 823	if (args->add_ref &&
 824	    nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
 825		goto nla_put_failure;
 826
 827	nlmsg_end(skb, nlh);
 828	return 0;
 829
 830nla_put_failure:
 831	nlmsg_cancel(skb, nlh);
 832	return -EMSGSIZE;
 833}
 834
 835static int rtnl_net_valid_getid_req(struct sk_buff *skb,
 836				    const struct nlmsghdr *nlh,
 837				    struct nlattr **tb,
 838				    struct netlink_ext_ack *extack)
 839{
 840	int i, err;
 841
 842	if (!netlink_strict_get_check(skb))
 843		return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
 844					      tb, NETNSA_MAX, rtnl_net_policy,
 845					      extack);
 846
 847	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
 848					    NETNSA_MAX, rtnl_net_policy,
 849					    extack);
 850	if (err)
 851		return err;
 852
 853	for (i = 0; i <= NETNSA_MAX; i++) {
 854		if (!tb[i])
 855			continue;
 856
 857		switch (i) {
 858		case NETNSA_PID:
 859		case NETNSA_FD:
 860		case NETNSA_NSID:
 861		case NETNSA_TARGET_NSID:
 862			break;
 863		default:
 864			NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
 865			return -EINVAL;
 866		}
 867	}
 868
 869	return 0;
 870}
 871
 872static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
 873			  struct netlink_ext_ack *extack)
 874{
 875	struct net *net = sock_net(skb->sk);
 876	struct nlattr *tb[NETNSA_MAX + 1];
 877	struct net_fill_args fillargs = {
 878		.portid = NETLINK_CB(skb).portid,
 879		.seq = nlh->nlmsg_seq,
 880		.cmd = RTM_NEWNSID,
 881	};
 882	struct net *peer, *target = net;
 883	struct nlattr *nla;
 884	struct sk_buff *msg;
 885	int err;
 
 886
 887	err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
 
 888	if (err < 0)
 889		return err;
 890	if (tb[NETNSA_PID]) {
 891		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 892		nla = tb[NETNSA_PID];
 893	} else if (tb[NETNSA_FD]) {
 894		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 895		nla = tb[NETNSA_FD];
 896	} else if (tb[NETNSA_NSID]) {
 897		peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
 898		if (!peer)
 899			peer = ERR_PTR(-ENOENT);
 900		nla = tb[NETNSA_NSID];
 901	} else {
 902		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 903		return -EINVAL;
 904	}
 905
 906	if (IS_ERR(peer)) {
 907		NL_SET_BAD_ATTR(extack, nla);
 908		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 909		return PTR_ERR(peer);
 910	}
 911
 912	if (tb[NETNSA_TARGET_NSID]) {
 913		int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
 914
 915		target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
 916		if (IS_ERR(target)) {
 917			NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
 918			NL_SET_ERR_MSG(extack,
 919				       "Target netns reference is invalid");
 920			err = PTR_ERR(target);
 921			goto out;
 922		}
 923		fillargs.add_ref = true;
 924		fillargs.ref_nsid = peernet2id(net, peer);
 925	}
 926
 927	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 928	if (!msg) {
 929		err = -ENOMEM;
 930		goto out;
 931	}
 932
 933	fillargs.nsid = peernet2id(target, peer);
 934	err = rtnl_net_fill(msg, &fillargs);
 
 935	if (err < 0)
 936		goto err_out;
 937
 938	err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
 939	goto out;
 940
 941err_out:
 942	nlmsg_free(msg);
 943out:
 944	if (fillargs.add_ref)
 945		put_net(target);
 946	put_net(peer);
 947	return err;
 948}
 949
 950struct rtnl_net_dump_cb {
 951	struct net *tgt_net;
 952	struct net *ref_net;
 953	struct sk_buff *skb;
 954	struct net_fill_args fillargs;
 955	int idx;
 956	int s_idx;
 957};
 958
 959/* Runs in RCU-critical section. */
 960static int rtnl_net_dumpid_one(int id, void *peer, void *data)
 961{
 962	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
 963	int ret;
 964
 965	if (net_cb->idx < net_cb->s_idx)
 966		goto cont;
 967
 968	net_cb->fillargs.nsid = id;
 969	if (net_cb->fillargs.add_ref)
 970		net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
 971	ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
 972	if (ret < 0)
 973		return ret;
 974
 975cont:
 976	net_cb->idx++;
 977	return 0;
 978}
 979
 980static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
 981				   struct rtnl_net_dump_cb *net_cb,
 982				   struct netlink_callback *cb)
 983{
 984	struct netlink_ext_ack *extack = cb->extack;
 985	struct nlattr *tb[NETNSA_MAX + 1];
 986	int err, i;
 987
 988	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
 989					    NETNSA_MAX, rtnl_net_policy,
 990					    extack);
 991	if (err < 0)
 992		return err;
 993
 994	for (i = 0; i <= NETNSA_MAX; i++) {
 995		if (!tb[i])
 996			continue;
 997
 998		if (i == NETNSA_TARGET_NSID) {
 999			struct net *net;
1000
1001			net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1002			if (IS_ERR(net)) {
1003				NL_SET_BAD_ATTR(extack, tb[i]);
1004				NL_SET_ERR_MSG(extack,
1005					       "Invalid target network namespace id");
1006				return PTR_ERR(net);
1007			}
1008			net_cb->fillargs.add_ref = true;
1009			net_cb->ref_net = net_cb->tgt_net;
1010			net_cb->tgt_net = net;
1011		} else {
1012			NL_SET_BAD_ATTR(extack, tb[i]);
1013			NL_SET_ERR_MSG(extack,
1014				       "Unsupported attribute in dump request");
1015			return -EINVAL;
1016		}
1017	}
1018
1019	return 0;
1020}
1021
1022static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1023{
 
1024	struct rtnl_net_dump_cb net_cb = {
1025		.tgt_net = sock_net(skb->sk),
1026		.skb = skb,
1027		.fillargs = {
1028			.portid = NETLINK_CB(cb->skb).portid,
1029			.seq = cb->nlh->nlmsg_seq,
1030			.flags = NLM_F_MULTI,
1031			.cmd = RTM_NEWNSID,
1032		},
1033		.idx = 0,
1034		.s_idx = cb->args[0],
1035	};
1036	int err = 0;
1037
1038	if (cb->strict_check) {
1039		err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1040		if (err < 0)
1041			goto end;
1042	}
1043
1044	rcu_read_lock();
1045	idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1046	rcu_read_unlock();
1047
1048	cb->args[0] = net_cb.idx;
1049end:
1050	if (net_cb.fillargs.add_ref)
1051		put_net(net_cb.tgt_net);
1052	return err < 0 ? err : skb->len;
1053}
1054
1055static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1056			      struct nlmsghdr *nlh, gfp_t gfp)
1057{
1058	struct net_fill_args fillargs = {
1059		.portid = portid,
1060		.seq = nlh ? nlh->nlmsg_seq : 0,
1061		.cmd = cmd,
1062		.nsid = id,
1063	};
1064	struct sk_buff *msg;
1065	int err = -ENOMEM;
1066
1067	msg = nlmsg_new(rtnl_net_get_size(), gfp);
1068	if (!msg)
1069		goto out;
1070
1071	err = rtnl_net_fill(msg, &fillargs);
1072	if (err < 0)
1073		goto err_out;
1074
1075	rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1076	return;
1077
1078err_out:
1079	nlmsg_free(msg);
1080out:
1081	rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1082}
1083
1084static int __init net_ns_init(void)
1085{
1086	struct net_generic *ng;
1087
1088#ifdef CONFIG_NET_NS
1089	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1090					SMP_CACHE_BYTES,
1091					SLAB_PANIC|SLAB_ACCOUNT, NULL);
1092
1093	/* Create workqueue for cleanup */
1094	netns_wq = create_singlethread_workqueue("netns");
1095	if (!netns_wq)
1096		panic("Could not create netns workq");
1097#endif
1098
1099	ng = net_alloc_generic();
1100	if (!ng)
1101		panic("Could not allocate generic netns");
1102
1103	rcu_assign_pointer(init_net.gen, ng);
1104	net_gen_cookie(&init_net);
1105
1106	down_write(&pernet_ops_rwsem);
1107	if (setup_net(&init_net, &init_user_ns))
1108		panic("Could not setup the initial network namespace");
1109
1110	init_net_initialized = true;
1111	up_write(&pernet_ops_rwsem);
 
 
 
1112
1113	if (register_pernet_subsys(&net_ns_ops))
1114		panic("Could not register network namespace subsystems");
1115
1116	rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1117		      RTNL_FLAG_DOIT_UNLOCKED);
1118	rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1119		      RTNL_FLAG_DOIT_UNLOCKED);
1120
1121	return 0;
1122}
1123
1124pure_initcall(net_ns_init);
1125
1126#ifdef CONFIG_NET_NS
1127static int __register_pernet_operations(struct list_head *list,
1128					struct pernet_operations *ops)
1129{
1130	struct net *net;
1131	int error;
1132	LIST_HEAD(net_exit_list);
1133
1134	list_add_tail(&ops->list, list);
1135	if (ops->init || (ops->id && ops->size)) {
1136		/* We held write locked pernet_ops_rwsem, and parallel
1137		 * setup_net() and cleanup_net() are not possible.
1138		 */
1139		for_each_net(net) {
1140			error = ops_init(ops, net);
1141			if (error)
1142				goto out_undo;
1143			list_add_tail(&net->exit_list, &net_exit_list);
1144		}
1145	}
1146	return 0;
1147
1148out_undo:
1149	/* If I have an error cleanup all namespaces I initialized */
1150	list_del(&ops->list);
1151	ops_pre_exit_list(ops, &net_exit_list);
1152	synchronize_rcu();
1153	ops_exit_list(ops, &net_exit_list);
1154	ops_free_list(ops, &net_exit_list);
1155	return error;
1156}
1157
1158static void __unregister_pernet_operations(struct pernet_operations *ops)
1159{
1160	struct net *net;
1161	LIST_HEAD(net_exit_list);
1162
1163	list_del(&ops->list);
1164	/* See comment in __register_pernet_operations() */
1165	for_each_net(net)
1166		list_add_tail(&net->exit_list, &net_exit_list);
1167	ops_pre_exit_list(ops, &net_exit_list);
1168	synchronize_rcu();
1169	ops_exit_list(ops, &net_exit_list);
1170	ops_free_list(ops, &net_exit_list);
1171}
1172
1173#else
1174
1175static int __register_pernet_operations(struct list_head *list,
1176					struct pernet_operations *ops)
1177{
1178	if (!init_net_initialized) {
1179		list_add_tail(&ops->list, list);
1180		return 0;
1181	}
1182
1183	return ops_init(ops, &init_net);
1184}
1185
1186static void __unregister_pernet_operations(struct pernet_operations *ops)
1187{
1188	if (!init_net_initialized) {
1189		list_del(&ops->list);
1190	} else {
1191		LIST_HEAD(net_exit_list);
1192		list_add(&init_net.exit_list, &net_exit_list);
1193		ops_pre_exit_list(ops, &net_exit_list);
1194		synchronize_rcu();
1195		ops_exit_list(ops, &net_exit_list);
1196		ops_free_list(ops, &net_exit_list);
1197	}
1198}
1199
1200#endif /* CONFIG_NET_NS */
1201
1202static DEFINE_IDA(net_generic_ids);
1203
1204static int register_pernet_operations(struct list_head *list,
1205				      struct pernet_operations *ops)
1206{
1207	int error;
1208
1209	if (ops->id) {
1210		error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1211				GFP_KERNEL);
1212		if (error < 0)
 
 
 
 
1213			return error;
1214		*ops->id = error;
1215		max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1216	}
1217	error = __register_pernet_operations(list, ops);
1218	if (error) {
1219		rcu_barrier();
1220		if (ops->id)
1221			ida_free(&net_generic_ids, *ops->id);
1222	}
1223
1224	return error;
1225}
1226
1227static void unregister_pernet_operations(struct pernet_operations *ops)
1228{
 
1229	__unregister_pernet_operations(ops);
1230	rcu_barrier();
1231	if (ops->id)
1232		ida_free(&net_generic_ids, *ops->id);
1233}
1234
1235/**
1236 *      register_pernet_subsys - register a network namespace subsystem
1237 *	@ops:  pernet operations structure for the subsystem
1238 *
1239 *	Register a subsystem which has init and exit functions
1240 *	that are called when network namespaces are created and
1241 *	destroyed respectively.
1242 *
1243 *	When registered all network namespace init functions are
1244 *	called for every existing network namespace.  Allowing kernel
1245 *	modules to have a race free view of the set of network namespaces.
1246 *
1247 *	When a new network namespace is created all of the init
1248 *	methods are called in the order in which they were registered.
1249 *
1250 *	When a network namespace is destroyed all of the exit methods
1251 *	are called in the reverse of the order with which they were
1252 *	registered.
1253 */
1254int register_pernet_subsys(struct pernet_operations *ops)
1255{
1256	int error;
1257	down_write(&pernet_ops_rwsem);
1258	error =  register_pernet_operations(first_device, ops);
1259	up_write(&pernet_ops_rwsem);
1260	return error;
1261}
1262EXPORT_SYMBOL_GPL(register_pernet_subsys);
1263
1264/**
1265 *      unregister_pernet_subsys - unregister a network namespace subsystem
1266 *	@ops: pernet operations structure to manipulate
1267 *
1268 *	Remove the pernet operations structure from the list to be
1269 *	used when network namespaces are created or destroyed.  In
1270 *	addition run the exit method for all existing network
1271 *	namespaces.
1272 */
1273void unregister_pernet_subsys(struct pernet_operations *ops)
1274{
1275	down_write(&pernet_ops_rwsem);
1276	unregister_pernet_operations(ops);
1277	up_write(&pernet_ops_rwsem);
1278}
1279EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1280
1281/**
1282 *      register_pernet_device - register a network namespace device
1283 *	@ops:  pernet operations structure for the subsystem
1284 *
1285 *	Register a device which has init and exit functions
1286 *	that are called when network namespaces are created and
1287 *	destroyed respectively.
1288 *
1289 *	When registered all network namespace init functions are
1290 *	called for every existing network namespace.  Allowing kernel
1291 *	modules to have a race free view of the set of network namespaces.
1292 *
1293 *	When a new network namespace is created all of the init
1294 *	methods are called in the order in which they were registered.
1295 *
1296 *	When a network namespace is destroyed all of the exit methods
1297 *	are called in the reverse of the order with which they were
1298 *	registered.
1299 */
1300int register_pernet_device(struct pernet_operations *ops)
1301{
1302	int error;
1303	down_write(&pernet_ops_rwsem);
1304	error = register_pernet_operations(&pernet_list, ops);
1305	if (!error && (first_device == &pernet_list))
1306		first_device = &ops->list;
1307	up_write(&pernet_ops_rwsem);
1308	return error;
1309}
1310EXPORT_SYMBOL_GPL(register_pernet_device);
1311
1312/**
1313 *      unregister_pernet_device - unregister a network namespace netdevice
1314 *	@ops: pernet operations structure to manipulate
1315 *
1316 *	Remove the pernet operations structure from the list to be
1317 *	used when network namespaces are created or destroyed.  In
1318 *	addition run the exit method for all existing network
1319 *	namespaces.
1320 */
1321void unregister_pernet_device(struct pernet_operations *ops)
1322{
1323	down_write(&pernet_ops_rwsem);
1324	if (&ops->list == first_device)
1325		first_device = first_device->next;
1326	unregister_pernet_operations(ops);
1327	up_write(&pernet_ops_rwsem);
1328}
1329EXPORT_SYMBOL_GPL(unregister_pernet_device);
1330
1331#ifdef CONFIG_NET_NS
1332static struct ns_common *netns_get(struct task_struct *task)
1333{
1334	struct net *net = NULL;
1335	struct nsproxy *nsproxy;
1336
1337	task_lock(task);
1338	nsproxy = task->nsproxy;
1339	if (nsproxy)
1340		net = get_net(nsproxy->net_ns);
1341	task_unlock(task);
1342
1343	return net ? &net->ns : NULL;
1344}
1345
1346static inline struct net *to_net_ns(struct ns_common *ns)
1347{
1348	return container_of(ns, struct net, ns);
1349}
1350
1351static void netns_put(struct ns_common *ns)
1352{
1353	put_net(to_net_ns(ns));
1354}
1355
1356static int netns_install(struct nsset *nsset, struct ns_common *ns)
1357{
1358	struct nsproxy *nsproxy = nsset->nsproxy;
1359	struct net *net = to_net_ns(ns);
1360
1361	if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1362	    !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1363		return -EPERM;
1364
1365	put_net(nsproxy->net_ns);
1366	nsproxy->net_ns = get_net(net);
1367	return 0;
1368}
1369
1370static struct user_namespace *netns_owner(struct ns_common *ns)
1371{
1372	return to_net_ns(ns)->user_ns;
1373}
1374
1375const struct proc_ns_operations netns_operations = {
1376	.name		= "net",
1377	.type		= CLONE_NEWNET,
1378	.get		= netns_get,
1379	.put		= netns_put,
1380	.install	= netns_install,
1381	.owner		= netns_owner,
1382};
1383#endif