Linux Audio

Check our new training course

Loading...
v4.17
 
   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 <linux/sched/task.h>
 
  20
  21#include <net/sock.h>
  22#include <net/netlink.h>
  23#include <net/net_namespace.h>
  24#include <net/netns/generic.h>
  25
  26/*
  27 *	Our network namespace constructor/destructor lists
  28 */
  29
  30static LIST_HEAD(pernet_list);
  31static struct list_head *first_device = &pernet_list;
  32
  33LIST_HEAD(net_namespace_list);
  34EXPORT_SYMBOL_GPL(net_namespace_list);
  35
  36/* Protects net_namespace_list. Nests iside rtnl_lock() */
  37DECLARE_RWSEM(net_rwsem);
  38EXPORT_SYMBOL_GPL(net_rwsem);
  39
 
 
 
 
  40struct net init_net = {
  41	.count		= REFCOUNT_INIT(1),
  42	.dev_base_head	= LIST_HEAD_INIT(init_net.dev_base_head),
 
 
 
  43};
  44EXPORT_SYMBOL(init_net);
  45
  46static bool init_net_initialized;
  47/*
  48 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
  49 * init_net_initialized and first_device pointer.
  50 * This is internal net namespace object. Please, don't use it
  51 * outside.
  52 */
  53DECLARE_RWSEM(pernet_ops_rwsem);
  54EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
  55
  56#define MIN_PERNET_OPS_ID	\
  57	((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
  58
  59#define INITIAL_NET_GEN_PTRS	13 /* +1 for len +2 for rcu_head */
  60
  61static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  62
  63static struct net_generic *net_alloc_generic(void)
  64{
  65	struct net_generic *ng;
  66	unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  67
  68	ng = kzalloc(generic_size, GFP_KERNEL);
  69	if (ng)
  70		ng->s.len = max_gen_ptrs;
  71
  72	return ng;
  73}
  74
  75static int net_assign_generic(struct net *net, unsigned int id, void *data)
  76{
  77	struct net_generic *ng, *old_ng;
  78
  79	BUG_ON(id < MIN_PERNET_OPS_ID);
  80
  81	old_ng = rcu_dereference_protected(net->gen,
  82					   lockdep_is_held(&pernet_ops_rwsem));
  83	if (old_ng->s.len > id) {
  84		old_ng->ptr[id] = data;
  85		return 0;
  86	}
  87
  88	ng = net_alloc_generic();
  89	if (ng == NULL)
  90		return -ENOMEM;
  91
  92	/*
  93	 * Some synchronisation notes:
  94	 *
  95	 * The net_generic explores the net->gen array inside rcu
  96	 * read section. Besides once set the net->gen->ptr[x]
  97	 * pointer never changes (see rules in netns/generic.h).
  98	 *
  99	 * That said, we simply duplicate this array and schedule
 100	 * the old copy for kfree after a grace period.
 101	 */
 102
 103	memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
 104	       (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
 105	ng->ptr[id] = data;
 106
 107	rcu_assign_pointer(net->gen, ng);
 108	kfree_rcu(old_ng, s.rcu);
 109	return 0;
 110}
 111
 112static int ops_init(const struct pernet_operations *ops, struct net *net)
 113{
 114	int err = -ENOMEM;
 115	void *data = NULL;
 116
 117	if (ops->id && ops->size) {
 118		data = kzalloc(ops->size, GFP_KERNEL);
 119		if (!data)
 120			goto out;
 121
 122		err = net_assign_generic(net, *ops->id, data);
 123		if (err)
 124			goto cleanup;
 125	}
 126	err = 0;
 127	if (ops->init)
 128		err = ops->init(net);
 129	if (!err)
 130		return 0;
 131
 132cleanup:
 133	kfree(data);
 134
 135out:
 136	return err;
 137}
 138
 139static void ops_free(const struct pernet_operations *ops, struct net *net)
 140{
 141	if (ops->id && ops->size) {
 142		kfree(net_generic(net, *ops->id));
 143	}
 144}
 145
 
 
 
 
 
 
 
 
 
 
 
 146static void ops_exit_list(const struct pernet_operations *ops,
 147			  struct list_head *net_exit_list)
 148{
 149	struct net *net;
 150	if (ops->exit) {
 151		list_for_each_entry(net, net_exit_list, exit_list)
 152			ops->exit(net);
 153	}
 154	if (ops->exit_batch)
 155		ops->exit_batch(net_exit_list);
 156}
 157
 158static void ops_free_list(const struct pernet_operations *ops,
 159			  struct list_head *net_exit_list)
 160{
 161	struct net *net;
 162	if (ops->size && ops->id) {
 163		list_for_each_entry(net, net_exit_list, exit_list)
 164			ops_free(ops, net);
 165	}
 166}
 167
 168/* should be called with nsid_lock held */
 169static int alloc_netid(struct net *net, struct net *peer, int reqid)
 170{
 171	int min = 0, max = 0;
 172
 173	if (reqid >= 0) {
 174		min = reqid;
 175		max = reqid + 1;
 176	}
 177
 178	return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
 179}
 180
 181/* This function is used by idr_for_each(). If net is equal to peer, the
 182 * function returns the id so that idr_for_each() stops. Because we cannot
 183 * returns the id 0 (idr_for_each() will not stop), we return the magic value
 184 * NET_ID_ZERO (-1) for it.
 185 */
 186#define NET_ID_ZERO -1
 187static int net_eq_idr(int id, void *net, void *peer)
 188{
 189	if (net_eq(net, peer))
 190		return id ? : NET_ID_ZERO;
 191	return 0;
 192}
 193
 194/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
 195 * is set to true, thus the caller knows that the new id must be notified via
 196 * rtnl.
 197 */
 198static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
 199{
 200	int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
 201	bool alloc_it = *alloc;
 202
 203	*alloc = false;
 204
 205	/* Magic value for id 0. */
 206	if (id == NET_ID_ZERO)
 207		return 0;
 208	if (id > 0)
 209		return id;
 210
 211	if (alloc_it) {
 212		id = alloc_netid(net, peer, -1);
 213		*alloc = true;
 214		return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
 215	}
 216
 217	return NETNSA_NSID_NOT_ASSIGNED;
 218}
 219
 220/* should be called with nsid_lock held */
 221static int __peernet2id(struct net *net, struct net *peer)
 222{
 223	bool no = false;
 224
 225	return __peernet2id_alloc(net, peer, &no);
 226}
 227
 228static void rtnl_net_notifyid(struct net *net, int cmd, int id);
 
 229/* This function returns the id of a peer netns. If no id is assigned, one will
 230 * be allocated and returned.
 231 */
 232int peernet2id_alloc(struct net *net, struct net *peer)
 233{
 234	bool alloc = false, alive = false;
 235	int id;
 236
 237	if (refcount_read(&net->count) == 0)
 238		return NETNSA_NSID_NOT_ASSIGNED;
 239	spin_lock_bh(&net->nsid_lock);
 240	/*
 241	 * When peer is obtained from RCU lists, we may race with
 242	 * its cleanup. Check whether it's alive, and this guarantees
 243	 * we never hash a peer back to net->netns_ids, after it has
 244	 * just been idr_remove()'d from there in cleanup_net().
 245	 */
 246	if (maybe_get_net(peer))
 247		alive = alloc = true;
 248	id = __peernet2id_alloc(net, peer, &alloc);
 249	spin_unlock_bh(&net->nsid_lock);
 250	if (alloc && id >= 0)
 251		rtnl_net_notifyid(net, RTM_NEWNSID, id);
 252	if (alive)
 253		put_net(peer);
 254	return id;
 255}
 256EXPORT_SYMBOL_GPL(peernet2id_alloc);
 257
 258/* This function returns, if assigned, the id of a peer netns. */
 259int peernet2id(struct net *net, struct net *peer)
 260{
 261	int id;
 262
 263	spin_lock_bh(&net->nsid_lock);
 264	id = __peernet2id(net, peer);
 265	spin_unlock_bh(&net->nsid_lock);
 266	return id;
 267}
 268EXPORT_SYMBOL(peernet2id);
 269
 270/* This function returns true is the peer netns has an id assigned into the
 271 * current netns.
 272 */
 273bool peernet_has_id(struct net *net, struct net *peer)
 274{
 275	return peernet2id(net, peer) >= 0;
 276}
 277
 278struct net *get_net_ns_by_id(struct net *net, int id)
 279{
 280	struct net *peer;
 281
 282	if (id < 0)
 283		return NULL;
 284
 285	rcu_read_lock();
 286	peer = idr_find(&net->netns_ids, id);
 287	if (peer)
 288		peer = maybe_get_net(peer);
 289	rcu_read_unlock();
 290
 291	return peer;
 292}
 293
 294/*
 295 * setup_net runs the initializers for the network namespace object.
 296 */
 297static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 298{
 299	/* Must be called with pernet_ops_rwsem held */
 300	const struct pernet_operations *ops, *saved_ops;
 301	int error = 0;
 302	LIST_HEAD(net_exit_list);
 303
 304	refcount_set(&net->count, 1);
 305	refcount_set(&net->passive, 1);
 
 306	net->dev_base_seq = 1;
 307	net->user_ns = user_ns;
 308	idr_init(&net->netns_ids);
 309	spin_lock_init(&net->nsid_lock);
 310	mutex_init(&net->ipv4.ra_mutex);
 311
 312	list_for_each_entry(ops, &pernet_list, list) {
 313		error = ops_init(ops, net);
 314		if (error < 0)
 315			goto out_undo;
 316	}
 317	down_write(&net_rwsem);
 318	list_add_tail_rcu(&net->list, &net_namespace_list);
 319	up_write(&net_rwsem);
 320out:
 321	return error;
 322
 323out_undo:
 324	/* Walk through the list backwards calling the exit functions
 325	 * for the pernet modules whose init functions did not fail.
 326	 */
 327	list_add(&net->exit_list, &net_exit_list);
 328	saved_ops = ops;
 329	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 
 
 
 
 
 
 330		ops_exit_list(ops, &net_exit_list);
 331
 332	ops = saved_ops;
 333	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 334		ops_free_list(ops, &net_exit_list);
 335
 336	rcu_barrier();
 337	goto out;
 338}
 339
 340static int __net_init net_defaults_init_net(struct net *net)
 341{
 342	net->core.sysctl_somaxconn = SOMAXCONN;
 343	return 0;
 344}
 345
 346static struct pernet_operations net_defaults_ops = {
 347	.init = net_defaults_init_net,
 348};
 349
 350static __init int net_defaults_init(void)
 351{
 352	if (register_pernet_subsys(&net_defaults_ops))
 353		panic("Cannot initialize net default settings");
 354
 355	return 0;
 356}
 357
 358core_initcall(net_defaults_init);
 359
 360#ifdef CONFIG_NET_NS
 361static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
 362{
 363	return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
 364}
 365
 366static void dec_net_namespaces(struct ucounts *ucounts)
 367{
 368	dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
 369}
 370
 371static struct kmem_cache *net_cachep __ro_after_init;
 372static struct workqueue_struct *netns_wq;
 373
 374static struct net *net_alloc(void)
 375{
 376	struct net *net = NULL;
 377	struct net_generic *ng;
 378
 379	ng = net_alloc_generic();
 380	if (!ng)
 381		goto out;
 382
 383	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 384	if (!net)
 385		goto out_free;
 386
 
 
 
 
 
 
 
 387	rcu_assign_pointer(net->gen, ng);
 388out:
 389	return net;
 390
 
 
 
 
 
 391out_free:
 392	kfree(ng);
 393	goto out;
 394}
 395
 396static void net_free(struct net *net)
 397{
 398	kfree(rcu_access_pointer(net->gen));
 399	kmem_cache_free(net_cachep, net);
 400}
 401
 402void net_drop_ns(void *p)
 403{
 404	struct net *ns = p;
 405	if (ns && refcount_dec_and_test(&ns->passive))
 406		net_free(ns);
 407}
 408
 409struct net *copy_net_ns(unsigned long flags,
 410			struct user_namespace *user_ns, struct net *old_net)
 411{
 412	struct ucounts *ucounts;
 413	struct net *net;
 414	int rv;
 415
 416	if (!(flags & CLONE_NEWNET))
 417		return get_net(old_net);
 418
 419	ucounts = inc_net_namespaces(user_ns);
 420	if (!ucounts)
 421		return ERR_PTR(-ENOSPC);
 422
 423	net = net_alloc();
 424	if (!net) {
 425		rv = -ENOMEM;
 426		goto dec_ucounts;
 427	}
 428	refcount_set(&net->passive, 1);
 429	net->ucounts = ucounts;
 430	get_user_ns(user_ns);
 431
 432	rv = down_read_killable(&pernet_ops_rwsem);
 433	if (rv < 0)
 434		goto put_userns;
 435
 436	rv = setup_net(net, user_ns);
 437
 438	up_read(&pernet_ops_rwsem);
 439
 440	if (rv < 0) {
 441put_userns:
 
 442		put_user_ns(user_ns);
 443		net_drop_ns(net);
 444dec_ucounts:
 445		dec_net_namespaces(ucounts);
 446		return ERR_PTR(rv);
 447	}
 448	return net;
 449}
 450
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 451static void unhash_nsid(struct net *net, struct net *last)
 452{
 453	struct net *tmp;
 454	/* This function is only called from cleanup_net() work,
 455	 * and this work is the only process, that may delete
 456	 * a net from net_namespace_list. So, when the below
 457	 * is executing, the list may only grow. Thus, we do not
 458	 * use for_each_net_rcu() or net_rwsem.
 459	 */
 460	for_each_net(tmp) {
 461		int id;
 462
 463		spin_lock_bh(&tmp->nsid_lock);
 464		id = __peernet2id(tmp, net);
 465		if (id >= 0)
 466			idr_remove(&tmp->netns_ids, id);
 467		spin_unlock_bh(&tmp->nsid_lock);
 468		if (id >= 0)
 469			rtnl_net_notifyid(tmp, RTM_DELNSID, id);
 
 470		if (tmp == last)
 471			break;
 472	}
 473	spin_lock_bh(&net->nsid_lock);
 474	idr_destroy(&net->netns_ids);
 475	spin_unlock_bh(&net->nsid_lock);
 476}
 477
 478static LLIST_HEAD(cleanup_list);
 479
 480static void cleanup_net(struct work_struct *work)
 481{
 482	const struct pernet_operations *ops;
 483	struct net *net, *tmp, *last;
 484	struct llist_node *net_kill_list;
 485	LIST_HEAD(net_exit_list);
 486
 487	/* Atomically snapshot the list of namespaces to cleanup */
 488	net_kill_list = llist_del_all(&cleanup_list);
 489
 490	down_read(&pernet_ops_rwsem);
 491
 492	/* Don't let anyone else find us. */
 493	down_write(&net_rwsem);
 494	llist_for_each_entry(net, net_kill_list, cleanup_list)
 495		list_del_rcu(&net->list);
 496	/* Cache last net. After we unlock rtnl, no one new net
 497	 * added to net_namespace_list can assign nsid pointer
 498	 * to a net from net_kill_list (see peernet2id_alloc()).
 499	 * So, we skip them in unhash_nsid().
 500	 *
 501	 * Note, that unhash_nsid() does not delete nsid links
 502	 * between net_kill_list's nets, as they've already
 503	 * deleted from net_namespace_list. But, this would be
 504	 * useless anyway, as netns_ids are destroyed there.
 505	 */
 506	last = list_last_entry(&net_namespace_list, struct net, list);
 507	up_write(&net_rwsem);
 508
 509	llist_for_each_entry(net, net_kill_list, cleanup_list) {
 510		unhash_nsid(net, last);
 511		list_add_tail(&net->exit_list, &net_exit_list);
 512	}
 513
 
 
 
 
 514	/*
 515	 * Another CPU might be rcu-iterating the list, wait for it.
 516	 * This needs to be before calling the exit() notifiers, so
 517	 * the rcu_barrier() below isn't sufficient alone.
 
 518	 */
 519	synchronize_rcu();
 520
 521	/* Run all of the network namespace exit methods */
 522	list_for_each_entry_reverse(ops, &pernet_list, list)
 523		ops_exit_list(ops, &net_exit_list);
 524
 525	/* Free the net generic variables */
 526	list_for_each_entry_reverse(ops, &pernet_list, list)
 527		ops_free_list(ops, &net_exit_list);
 528
 529	up_read(&pernet_ops_rwsem);
 530
 531	/* Ensure there are no outstanding rcu callbacks using this
 532	 * network namespace.
 533	 */
 534	rcu_barrier();
 535
 536	/* Finally it is safe to free my network namespace structure */
 537	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 538		list_del_init(&net->exit_list);
 539		dec_net_namespaces(net->ucounts);
 
 540		put_user_ns(net->user_ns);
 541		net_drop_ns(net);
 542	}
 543}
 544
 545/**
 546 * net_ns_barrier - wait until concurrent net_cleanup_work is done
 547 *
 548 * cleanup_net runs from work queue and will first remove namespaces
 549 * from the global list, then run net exit functions.
 550 *
 551 * Call this in module exit path to make sure that all netns
 552 * ->exit ops have been invoked before the function is removed.
 553 */
 554void net_ns_barrier(void)
 555{
 556	down_write(&pernet_ops_rwsem);
 557	up_write(&pernet_ops_rwsem);
 558}
 559EXPORT_SYMBOL(net_ns_barrier);
 560
 561static DECLARE_WORK(net_cleanup_work, cleanup_net);
 562
 563void __put_net(struct net *net)
 564{
 565	/* Cleanup the network namespace in process context */
 566	if (llist_add(&net->cleanup_list, &cleanup_list))
 567		queue_work(netns_wq, &net_cleanup_work);
 568}
 569EXPORT_SYMBOL_GPL(__put_net);
 570
 571struct net *get_net_ns_by_fd(int fd)
 572{
 573	struct file *file;
 574	struct ns_common *ns;
 575	struct net *net;
 576
 577	file = proc_ns_fget(fd);
 578	if (IS_ERR(file))
 579		return ERR_CAST(file);
 580
 581	ns = get_proc_ns(file_inode(file));
 582	if (ns->ops == &netns_operations)
 583		net = get_net(container_of(ns, struct net, ns));
 584	else
 585		net = ERR_PTR(-EINVAL);
 586
 587	fput(file);
 588	return net;
 589}
 590
 591#else
 592struct net *get_net_ns_by_fd(int fd)
 593{
 594	return ERR_PTR(-EINVAL);
 595}
 596#endif
 597EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
 598
 599struct net *get_net_ns_by_pid(pid_t pid)
 600{
 601	struct task_struct *tsk;
 602	struct net *net;
 603
 604	/* Lookup the network namespace */
 605	net = ERR_PTR(-ESRCH);
 606	rcu_read_lock();
 607	tsk = find_task_by_vpid(pid);
 608	if (tsk) {
 609		struct nsproxy *nsproxy;
 610		task_lock(tsk);
 611		nsproxy = tsk->nsproxy;
 612		if (nsproxy)
 613			net = get_net(nsproxy->net_ns);
 614		task_unlock(tsk);
 615	}
 616	rcu_read_unlock();
 617	return net;
 618}
 619EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 620
 621static __net_init int net_ns_net_init(struct net *net)
 622{
 623#ifdef CONFIG_NET_NS
 624	net->ns.ops = &netns_operations;
 625#endif
 626	return ns_alloc_inum(&net->ns);
 627}
 628
 629static __net_exit void net_ns_net_exit(struct net *net)
 630{
 631	ns_free_inum(&net->ns);
 632}
 633
 634static struct pernet_operations __net_initdata net_ns_ops = {
 635	.init = net_ns_net_init,
 636	.exit = net_ns_net_exit,
 637};
 638
 639static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
 640	[NETNSA_NONE]		= { .type = NLA_UNSPEC },
 641	[NETNSA_NSID]		= { .type = NLA_S32 },
 642	[NETNSA_PID]		= { .type = NLA_U32 },
 643	[NETNSA_FD]		= { .type = NLA_U32 },
 
 644};
 645
 646static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
 647			  struct netlink_ext_ack *extack)
 648{
 649	struct net *net = sock_net(skb->sk);
 650	struct nlattr *tb[NETNSA_MAX + 1];
 651	struct nlattr *nla;
 652	struct net *peer;
 653	int nsid, err;
 654
 655	err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
 656			  rtnl_net_policy, extack);
 657	if (err < 0)
 658		return err;
 659	if (!tb[NETNSA_NSID]) {
 660		NL_SET_ERR_MSG(extack, "nsid is missing");
 661		return -EINVAL;
 662	}
 663	nsid = nla_get_s32(tb[NETNSA_NSID]);
 664
 665	if (tb[NETNSA_PID]) {
 666		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 667		nla = tb[NETNSA_PID];
 668	} else if (tb[NETNSA_FD]) {
 669		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 670		nla = tb[NETNSA_FD];
 671	} else {
 672		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 673		return -EINVAL;
 674	}
 675	if (IS_ERR(peer)) {
 676		NL_SET_BAD_ATTR(extack, nla);
 677		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 678		return PTR_ERR(peer);
 679	}
 680
 681	spin_lock_bh(&net->nsid_lock);
 682	if (__peernet2id(net, peer) >= 0) {
 683		spin_unlock_bh(&net->nsid_lock);
 684		err = -EEXIST;
 685		NL_SET_BAD_ATTR(extack, nla);
 686		NL_SET_ERR_MSG(extack,
 687			       "Peer netns already has a nsid assigned");
 688		goto out;
 689	}
 690
 691	err = alloc_netid(net, peer, nsid);
 692	spin_unlock_bh(&net->nsid_lock);
 693	if (err >= 0) {
 694		rtnl_net_notifyid(net, RTM_NEWNSID, err);
 
 695		err = 0;
 696	} else if (err == -ENOSPC && nsid >= 0) {
 697		err = -EEXIST;
 698		NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
 699		NL_SET_ERR_MSG(extack, "The specified nsid is already used");
 700	}
 701out:
 702	put_net(peer);
 703	return err;
 704}
 705
 706static int rtnl_net_get_size(void)
 707{
 708	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
 709	       + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
 
 710	       ;
 711}
 712
 713static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
 714			 int cmd, struct net *net, int nsid)
 
 
 
 
 
 
 
 
 
 715{
 716	struct nlmsghdr *nlh;
 717	struct rtgenmsg *rth;
 718
 719	nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
 
 720	if (!nlh)
 721		return -EMSGSIZE;
 722
 723	rth = nlmsg_data(nlh);
 724	rth->rtgen_family = AF_UNSPEC;
 725
 726	if (nla_put_s32(skb, NETNSA_NSID, nsid))
 
 
 
 
 727		goto nla_put_failure;
 728
 729	nlmsg_end(skb, nlh);
 730	return 0;
 731
 732nla_put_failure:
 733	nlmsg_cancel(skb, nlh);
 734	return -EMSGSIZE;
 735}
 736
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 737static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
 738			  struct netlink_ext_ack *extack)
 739{
 740	struct net *net = sock_net(skb->sk);
 741	struct nlattr *tb[NETNSA_MAX + 1];
 
 
 
 
 
 
 742	struct nlattr *nla;
 743	struct sk_buff *msg;
 744	struct net *peer;
 745	int err, id;
 746
 747	err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
 748			  rtnl_net_policy, extack);
 749	if (err < 0)
 750		return err;
 751	if (tb[NETNSA_PID]) {
 752		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 753		nla = tb[NETNSA_PID];
 754	} else if (tb[NETNSA_FD]) {
 755		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 756		nla = tb[NETNSA_FD];
 
 
 
 
 
 757	} else {
 758		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 759		return -EINVAL;
 760	}
 761
 762	if (IS_ERR(peer)) {
 763		NL_SET_BAD_ATTR(extack, nla);
 764		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 765		return PTR_ERR(peer);
 766	}
 767
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 768	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 769	if (!msg) {
 770		err = -ENOMEM;
 771		goto out;
 772	}
 773
 774	id = peernet2id(net, peer);
 775	err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
 776			    RTM_NEWNSID, net, id);
 777	if (err < 0)
 778		goto err_out;
 779
 780	err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
 781	goto out;
 782
 783err_out:
 784	nlmsg_free(msg);
 785out:
 
 
 786	put_net(peer);
 787	return err;
 788}
 789
 790struct rtnl_net_dump_cb {
 791	struct net *net;
 
 792	struct sk_buff *skb;
 793	struct netlink_callback *cb;
 794	int idx;
 795	int s_idx;
 796};
 797
 798static int rtnl_net_dumpid_one(int id, void *peer, void *data)
 799{
 800	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
 801	int ret;
 802
 803	if (net_cb->idx < net_cb->s_idx)
 804		goto cont;
 805
 806	ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
 807			    net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
 808			    RTM_NEWNSID, net_cb->net, id);
 
 809	if (ret < 0)
 810		return ret;
 811
 812cont:
 813	net_cb->idx++;
 814	return 0;
 815}
 816
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 817static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
 818{
 819	struct net *net = sock_net(skb->sk);
 820	struct rtnl_net_dump_cb net_cb = {
 821		.net = net,
 822		.skb = skb,
 823		.cb = cb,
 
 
 
 
 
 824		.idx = 0,
 825		.s_idx = cb->args[0],
 826	};
 
 827
 828	spin_lock_bh(&net->nsid_lock);
 829	idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
 830	spin_unlock_bh(&net->nsid_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 831
 832	cb->args[0] = net_cb.idx;
 833	return skb->len;
 
 
 
 834}
 835
 836static void rtnl_net_notifyid(struct net *net, int cmd, int id)
 837{
 
 
 
 
 
 
 
 838	struct sk_buff *msg;
 839	int err = -ENOMEM;
 840
 841	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 842	if (!msg)
 843		goto out;
 844
 845	err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
 846	if (err < 0)
 847		goto err_out;
 848
 849	rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
 850	return;
 851
 852err_out:
 853	nlmsg_free(msg);
 854out:
 855	rtnl_set_sk_err(net, RTNLGRP_NSID, err);
 856}
 857
 858static int __init net_ns_init(void)
 859{
 860	struct net_generic *ng;
 861
 862#ifdef CONFIG_NET_NS
 863	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 864					SMP_CACHE_BYTES,
 865					SLAB_PANIC|SLAB_ACCOUNT, NULL);
 866
 867	/* Create workqueue for cleanup */
 868	netns_wq = create_singlethread_workqueue("netns");
 869	if (!netns_wq)
 870		panic("Could not create netns workq");
 871#endif
 872
 873	ng = net_alloc_generic();
 874	if (!ng)
 875		panic("Could not allocate generic netns");
 876
 877	rcu_assign_pointer(init_net.gen, ng);
 878
 879	down_write(&pernet_ops_rwsem);
 880	if (setup_net(&init_net, &init_user_ns))
 881		panic("Could not setup the initial network namespace");
 882
 883	init_net_initialized = true;
 884	up_write(&pernet_ops_rwsem);
 885
 886	register_pernet_subsys(&net_ns_ops);
 
 887
 888	rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
 889		      RTNL_FLAG_DOIT_UNLOCKED);
 890	rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
 891		      RTNL_FLAG_DOIT_UNLOCKED);
 892
 893	return 0;
 894}
 895
 896pure_initcall(net_ns_init);
 897
 898#ifdef CONFIG_NET_NS
 899static int __register_pernet_operations(struct list_head *list,
 900					struct pernet_operations *ops)
 901{
 902	struct net *net;
 903	int error;
 904	LIST_HEAD(net_exit_list);
 905
 906	list_add_tail(&ops->list, list);
 907	if (ops->init || (ops->id && ops->size)) {
 908		/* We held write locked pernet_ops_rwsem, and parallel
 909		 * setup_net() and cleanup_net() are not possible.
 910		 */
 911		for_each_net(net) {
 912			error = ops_init(ops, net);
 913			if (error)
 914				goto out_undo;
 915			list_add_tail(&net->exit_list, &net_exit_list);
 916		}
 917	}
 918	return 0;
 919
 920out_undo:
 921	/* If I have an error cleanup all namespaces I initialized */
 922	list_del(&ops->list);
 
 
 923	ops_exit_list(ops, &net_exit_list);
 924	ops_free_list(ops, &net_exit_list);
 925	return error;
 926}
 927
 928static void __unregister_pernet_operations(struct pernet_operations *ops)
 929{
 930	struct net *net;
 931	LIST_HEAD(net_exit_list);
 932
 933	list_del(&ops->list);
 934	/* See comment in __register_pernet_operations() */
 935	for_each_net(net)
 936		list_add_tail(&net->exit_list, &net_exit_list);
 
 
 937	ops_exit_list(ops, &net_exit_list);
 938	ops_free_list(ops, &net_exit_list);
 939}
 940
 941#else
 942
 943static int __register_pernet_operations(struct list_head *list,
 944					struct pernet_operations *ops)
 945{
 946	if (!init_net_initialized) {
 947		list_add_tail(&ops->list, list);
 948		return 0;
 949	}
 950
 951	return ops_init(ops, &init_net);
 952}
 953
 954static void __unregister_pernet_operations(struct pernet_operations *ops)
 955{
 956	if (!init_net_initialized) {
 957		list_del(&ops->list);
 958	} else {
 959		LIST_HEAD(net_exit_list);
 960		list_add(&init_net.exit_list, &net_exit_list);
 
 
 961		ops_exit_list(ops, &net_exit_list);
 962		ops_free_list(ops, &net_exit_list);
 963	}
 964}
 965
 966#endif /* CONFIG_NET_NS */
 967
 968static DEFINE_IDA(net_generic_ids);
 969
 970static int register_pernet_operations(struct list_head *list,
 971				      struct pernet_operations *ops)
 972{
 973	int error;
 974
 975	if (ops->id) {
 976again:
 977		error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id);
 978		if (error < 0) {
 979			if (error == -EAGAIN) {
 980				ida_pre_get(&net_generic_ids, GFP_KERNEL);
 981				goto again;
 982			}
 983			return error;
 984		}
 985		max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
 986	}
 987	error = __register_pernet_operations(list, ops);
 988	if (error) {
 989		rcu_barrier();
 990		if (ops->id)
 991			ida_remove(&net_generic_ids, *ops->id);
 992	}
 993
 994	return error;
 995}
 996
 997static void unregister_pernet_operations(struct pernet_operations *ops)
 998{
 999	__unregister_pernet_operations(ops);
1000	rcu_barrier();
1001	if (ops->id)
1002		ida_remove(&net_generic_ids, *ops->id);
1003}
1004
1005/**
1006 *      register_pernet_subsys - register a network namespace subsystem
1007 *	@ops:  pernet operations structure for the subsystem
1008 *
1009 *	Register a subsystem which has init and exit functions
1010 *	that are called when network namespaces are created and
1011 *	destroyed respectively.
1012 *
1013 *	When registered all network namespace init functions are
1014 *	called for every existing network namespace.  Allowing kernel
1015 *	modules to have a race free view of the set of network namespaces.
1016 *
1017 *	When a new network namespace is created all of the init
1018 *	methods are called in the order in which they were registered.
1019 *
1020 *	When a network namespace is destroyed all of the exit methods
1021 *	are called in the reverse of the order with which they were
1022 *	registered.
1023 */
1024int register_pernet_subsys(struct pernet_operations *ops)
1025{
1026	int error;
1027	down_write(&pernet_ops_rwsem);
1028	error =  register_pernet_operations(first_device, ops);
1029	up_write(&pernet_ops_rwsem);
1030	return error;
1031}
1032EXPORT_SYMBOL_GPL(register_pernet_subsys);
1033
1034/**
1035 *      unregister_pernet_subsys - unregister a network namespace subsystem
1036 *	@ops: pernet operations structure to manipulate
1037 *
1038 *	Remove the pernet operations structure from the list to be
1039 *	used when network namespaces are created or destroyed.  In
1040 *	addition run the exit method for all existing network
1041 *	namespaces.
1042 */
1043void unregister_pernet_subsys(struct pernet_operations *ops)
1044{
1045	down_write(&pernet_ops_rwsem);
1046	unregister_pernet_operations(ops);
1047	up_write(&pernet_ops_rwsem);
1048}
1049EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1050
1051/**
1052 *      register_pernet_device - register a network namespace device
1053 *	@ops:  pernet operations structure for the subsystem
1054 *
1055 *	Register a device which has init and exit functions
1056 *	that are called when network namespaces are created and
1057 *	destroyed respectively.
1058 *
1059 *	When registered all network namespace init functions are
1060 *	called for every existing network namespace.  Allowing kernel
1061 *	modules to have a race free view of the set of network namespaces.
1062 *
1063 *	When a new network namespace is created all of the init
1064 *	methods are called in the order in which they were registered.
1065 *
1066 *	When a network namespace is destroyed all of the exit methods
1067 *	are called in the reverse of the order with which they were
1068 *	registered.
1069 */
1070int register_pernet_device(struct pernet_operations *ops)
1071{
1072	int error;
1073	down_write(&pernet_ops_rwsem);
1074	error = register_pernet_operations(&pernet_list, ops);
1075	if (!error && (first_device == &pernet_list))
1076		first_device = &ops->list;
1077	up_write(&pernet_ops_rwsem);
1078	return error;
1079}
1080EXPORT_SYMBOL_GPL(register_pernet_device);
1081
1082/**
1083 *      unregister_pernet_device - unregister a network namespace netdevice
1084 *	@ops: pernet operations structure to manipulate
1085 *
1086 *	Remove the pernet operations structure from the list to be
1087 *	used when network namespaces are created or destroyed.  In
1088 *	addition run the exit method for all existing network
1089 *	namespaces.
1090 */
1091void unregister_pernet_device(struct pernet_operations *ops)
1092{
1093	down_write(&pernet_ops_rwsem);
1094	if (&ops->list == first_device)
1095		first_device = first_device->next;
1096	unregister_pernet_operations(ops);
1097	up_write(&pernet_ops_rwsem);
1098}
1099EXPORT_SYMBOL_GPL(unregister_pernet_device);
1100
1101#ifdef CONFIG_NET_NS
1102static struct ns_common *netns_get(struct task_struct *task)
1103{
1104	struct net *net = NULL;
1105	struct nsproxy *nsproxy;
1106
1107	task_lock(task);
1108	nsproxy = task->nsproxy;
1109	if (nsproxy)
1110		net = get_net(nsproxy->net_ns);
1111	task_unlock(task);
1112
1113	return net ? &net->ns : NULL;
1114}
1115
1116static inline struct net *to_net_ns(struct ns_common *ns)
1117{
1118	return container_of(ns, struct net, ns);
1119}
1120
1121static void netns_put(struct ns_common *ns)
1122{
1123	put_net(to_net_ns(ns));
1124}
1125
1126static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1127{
1128	struct net *net = to_net_ns(ns);
1129
1130	if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1131	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1132		return -EPERM;
1133
1134	put_net(nsproxy->net_ns);
1135	nsproxy->net_ns = get_net(net);
1136	return 0;
1137}
1138
1139static struct user_namespace *netns_owner(struct ns_common *ns)
1140{
1141	return to_net_ns(ns)->user_ns;
1142}
1143
1144const struct proc_ns_operations netns_operations = {
1145	.name		= "net",
1146	.type		= CLONE_NEWNET,
1147	.get		= netns_get,
1148	.put		= netns_put,
1149	.install	= netns_install,
1150	.owner		= netns_owner,
1151};
1152#endif
v5.4
   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 struct net_generic *net_alloc_generic(void)
  73{
  74	struct net_generic *ng;
  75	unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  76
  77	ng = kzalloc(generic_size, GFP_KERNEL);
  78	if (ng)
  79		ng->s.len = max_gen_ptrs;
  80
  81	return ng;
  82}
  83
  84static int net_assign_generic(struct net *net, unsigned int id, void *data)
  85{
  86	struct net_generic *ng, *old_ng;
  87
  88	BUG_ON(id < MIN_PERNET_OPS_ID);
  89
  90	old_ng = rcu_dereference_protected(net->gen,
  91					   lockdep_is_held(&pernet_ops_rwsem));
  92	if (old_ng->s.len > id) {
  93		old_ng->ptr[id] = data;
  94		return 0;
  95	}
  96
  97	ng = net_alloc_generic();
  98	if (ng == NULL)
  99		return -ENOMEM;
 100
 101	/*
 102	 * Some synchronisation notes:
 103	 *
 104	 * The net_generic explores the net->gen array inside rcu
 105	 * read section. Besides once set the net->gen->ptr[x]
 106	 * pointer never changes (see rules in netns/generic.h).
 107	 *
 108	 * That said, we simply duplicate this array and schedule
 109	 * the old copy for kfree after a grace period.
 110	 */
 111
 112	memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
 113	       (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
 114	ng->ptr[id] = data;
 115
 116	rcu_assign_pointer(net->gen, ng);
 117	kfree_rcu(old_ng, s.rcu);
 118	return 0;
 119}
 120
 121static int ops_init(const struct pernet_operations *ops, struct net *net)
 122{
 123	int err = -ENOMEM;
 124	void *data = NULL;
 125
 126	if (ops->id && ops->size) {
 127		data = kzalloc(ops->size, GFP_KERNEL);
 128		if (!data)
 129			goto out;
 130
 131		err = net_assign_generic(net, *ops->id, data);
 132		if (err)
 133			goto cleanup;
 134	}
 135	err = 0;
 136	if (ops->init)
 137		err = ops->init(net);
 138	if (!err)
 139		return 0;
 140
 141cleanup:
 142	kfree(data);
 143
 144out:
 145	return err;
 146}
 147
 148static void ops_free(const struct pernet_operations *ops, struct net *net)
 149{
 150	if (ops->id && ops->size) {
 151		kfree(net_generic(net, *ops->id));
 152	}
 153}
 154
 155static void ops_pre_exit_list(const struct pernet_operations *ops,
 156			      struct list_head *net_exit_list)
 157{
 158	struct net *net;
 159
 160	if (ops->pre_exit) {
 161		list_for_each_entry(net, net_exit_list, exit_list)
 162			ops->pre_exit(net);
 163	}
 164}
 165
 166static void ops_exit_list(const struct pernet_operations *ops,
 167			  struct list_head *net_exit_list)
 168{
 169	struct net *net;
 170	if (ops->exit) {
 171		list_for_each_entry(net, net_exit_list, exit_list)
 172			ops->exit(net);
 173	}
 174	if (ops->exit_batch)
 175		ops->exit_batch(net_exit_list);
 176}
 177
 178static void ops_free_list(const struct pernet_operations *ops,
 179			  struct list_head *net_exit_list)
 180{
 181	struct net *net;
 182	if (ops->size && ops->id) {
 183		list_for_each_entry(net, net_exit_list, exit_list)
 184			ops_free(ops, net);
 185	}
 186}
 187
 188/* should be called with nsid_lock held */
 189static int alloc_netid(struct net *net, struct net *peer, int reqid)
 190{
 191	int min = 0, max = 0;
 192
 193	if (reqid >= 0) {
 194		min = reqid;
 195		max = reqid + 1;
 196	}
 197
 198	return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
 199}
 200
 201/* This function is used by idr_for_each(). If net is equal to peer, the
 202 * function returns the id so that idr_for_each() stops. Because we cannot
 203 * returns the id 0 (idr_for_each() will not stop), we return the magic value
 204 * NET_ID_ZERO (-1) for it.
 205 */
 206#define NET_ID_ZERO -1
 207static int net_eq_idr(int id, void *net, void *peer)
 208{
 209	if (net_eq(net, peer))
 210		return id ? : NET_ID_ZERO;
 211	return 0;
 212}
 213
 214/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
 215 * is set to true, thus the caller knows that the new id must be notified via
 216 * rtnl.
 217 */
 218static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
 219{
 220	int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
 221	bool alloc_it = *alloc;
 222
 223	*alloc = false;
 224
 225	/* Magic value for id 0. */
 226	if (id == NET_ID_ZERO)
 227		return 0;
 228	if (id > 0)
 229		return id;
 230
 231	if (alloc_it) {
 232		id = alloc_netid(net, peer, -1);
 233		*alloc = true;
 234		return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
 235	}
 236
 237	return NETNSA_NSID_NOT_ASSIGNED;
 238}
 239
 240/* should be called with nsid_lock held */
 241static int __peernet2id(struct net *net, struct net *peer)
 242{
 243	bool no = false;
 244
 245	return __peernet2id_alloc(net, peer, &no);
 246}
 247
 248static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
 249			      struct nlmsghdr *nlh, gfp_t gfp);
 250/* This function returns the id of a peer netns. If no id is assigned, one will
 251 * be allocated and returned.
 252 */
 253int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
 254{
 255	bool alloc = false, alive = false;
 256	int id;
 257
 258	if (refcount_read(&net->count) == 0)
 259		return NETNSA_NSID_NOT_ASSIGNED;
 260	spin_lock_bh(&net->nsid_lock);
 261	/*
 262	 * When peer is obtained from RCU lists, we may race with
 263	 * its cleanup. Check whether it's alive, and this guarantees
 264	 * we never hash a peer back to net->netns_ids, after it has
 265	 * just been idr_remove()'d from there in cleanup_net().
 266	 */
 267	if (maybe_get_net(peer))
 268		alive = alloc = true;
 269	id = __peernet2id_alloc(net, peer, &alloc);
 270	spin_unlock_bh(&net->nsid_lock);
 271	if (alloc && id >= 0)
 272		rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
 273	if (alive)
 274		put_net(peer);
 275	return id;
 276}
 277EXPORT_SYMBOL_GPL(peernet2id_alloc);
 278
 279/* This function returns, if assigned, the id of a peer netns. */
 280int peernet2id(struct net *net, struct net *peer)
 281{
 282	int id;
 283
 284	spin_lock_bh(&net->nsid_lock);
 285	id = __peernet2id(net, peer);
 286	spin_unlock_bh(&net->nsid_lock);
 287	return id;
 288}
 289EXPORT_SYMBOL(peernet2id);
 290
 291/* This function returns true is the peer netns has an id assigned into the
 292 * current netns.
 293 */
 294bool peernet_has_id(struct net *net, struct net *peer)
 295{
 296	return peernet2id(net, peer) >= 0;
 297}
 298
 299struct net *get_net_ns_by_id(struct net *net, int id)
 300{
 301	struct net *peer;
 302
 303	if (id < 0)
 304		return NULL;
 305
 306	rcu_read_lock();
 307	peer = idr_find(&net->netns_ids, id);
 308	if (peer)
 309		peer = maybe_get_net(peer);
 310	rcu_read_unlock();
 311
 312	return peer;
 313}
 314
 315/*
 316 * setup_net runs the initializers for the network namespace object.
 317 */
 318static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 319{
 320	/* Must be called with pernet_ops_rwsem held */
 321	const struct pernet_operations *ops, *saved_ops;
 322	int error = 0;
 323	LIST_HEAD(net_exit_list);
 324
 325	refcount_set(&net->count, 1);
 326	refcount_set(&net->passive, 1);
 327	get_random_bytes(&net->hash_mix, sizeof(u32));
 328	net->dev_base_seq = 1;
 329	net->user_ns = user_ns;
 330	idr_init(&net->netns_ids);
 331	spin_lock_init(&net->nsid_lock);
 332	mutex_init(&net->ipv4.ra_mutex);
 333
 334	list_for_each_entry(ops, &pernet_list, list) {
 335		error = ops_init(ops, net);
 336		if (error < 0)
 337			goto out_undo;
 338	}
 339	down_write(&net_rwsem);
 340	list_add_tail_rcu(&net->list, &net_namespace_list);
 341	up_write(&net_rwsem);
 342out:
 343	return error;
 344
 345out_undo:
 346	/* Walk through the list backwards calling the exit functions
 347	 * for the pernet modules whose init functions did not fail.
 348	 */
 349	list_add(&net->exit_list, &net_exit_list);
 350	saved_ops = ops;
 351	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 352		ops_pre_exit_list(ops, &net_exit_list);
 353
 354	synchronize_rcu();
 355
 356	ops = saved_ops;
 357	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 358		ops_exit_list(ops, &net_exit_list);
 359
 360	ops = saved_ops;
 361	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 362		ops_free_list(ops, &net_exit_list);
 363
 364	rcu_barrier();
 365	goto out;
 366}
 367
 368static int __net_init net_defaults_init_net(struct net *net)
 369{
 370	net->core.sysctl_somaxconn = SOMAXCONN;
 371	return 0;
 372}
 373
 374static struct pernet_operations net_defaults_ops = {
 375	.init = net_defaults_init_net,
 376};
 377
 378static __init int net_defaults_init(void)
 379{
 380	if (register_pernet_subsys(&net_defaults_ops))
 381		panic("Cannot initialize net default settings");
 382
 383	return 0;
 384}
 385
 386core_initcall(net_defaults_init);
 387
 388#ifdef CONFIG_NET_NS
 389static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
 390{
 391	return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
 392}
 393
 394static void dec_net_namespaces(struct ucounts *ucounts)
 395{
 396	dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
 397}
 398
 399static struct kmem_cache *net_cachep __ro_after_init;
 400static struct workqueue_struct *netns_wq;
 401
 402static struct net *net_alloc(void)
 403{
 404	struct net *net = NULL;
 405	struct net_generic *ng;
 406
 407	ng = net_alloc_generic();
 408	if (!ng)
 409		goto out;
 410
 411	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 412	if (!net)
 413		goto out_free;
 414
 415#ifdef CONFIG_KEYS
 416	net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
 417	if (!net->key_domain)
 418		goto out_free_2;
 419	refcount_set(&net->key_domain->usage, 1);
 420#endif
 421
 422	rcu_assign_pointer(net->gen, ng);
 423out:
 424	return net;
 425
 426#ifdef CONFIG_KEYS
 427out_free_2:
 428	kmem_cache_free(net_cachep, net);
 429	net = NULL;
 430#endif
 431out_free:
 432	kfree(ng);
 433	goto out;
 434}
 435
 436static void net_free(struct net *net)
 437{
 438	kfree(rcu_access_pointer(net->gen));
 439	kmem_cache_free(net_cachep, net);
 440}
 441
 442void net_drop_ns(void *p)
 443{
 444	struct net *ns = p;
 445	if (ns && refcount_dec_and_test(&ns->passive))
 446		net_free(ns);
 447}
 448
 449struct net *copy_net_ns(unsigned long flags,
 450			struct user_namespace *user_ns, struct net *old_net)
 451{
 452	struct ucounts *ucounts;
 453	struct net *net;
 454	int rv;
 455
 456	if (!(flags & CLONE_NEWNET))
 457		return get_net(old_net);
 458
 459	ucounts = inc_net_namespaces(user_ns);
 460	if (!ucounts)
 461		return ERR_PTR(-ENOSPC);
 462
 463	net = net_alloc();
 464	if (!net) {
 465		rv = -ENOMEM;
 466		goto dec_ucounts;
 467	}
 468	refcount_set(&net->passive, 1);
 469	net->ucounts = ucounts;
 470	get_user_ns(user_ns);
 471
 472	rv = down_read_killable(&pernet_ops_rwsem);
 473	if (rv < 0)
 474		goto put_userns;
 475
 476	rv = setup_net(net, user_ns);
 477
 478	up_read(&pernet_ops_rwsem);
 479
 480	if (rv < 0) {
 481put_userns:
 482		key_remove_domain(net->key_domain);
 483		put_user_ns(user_ns);
 484		net_drop_ns(net);
 485dec_ucounts:
 486		dec_net_namespaces(ucounts);
 487		return ERR_PTR(rv);
 488	}
 489	return net;
 490}
 491
 492/**
 493 * net_ns_get_ownership - get sysfs ownership data for @net
 494 * @net: network namespace in question (can be NULL)
 495 * @uid: kernel user ID for sysfs objects
 496 * @gid: kernel group ID for sysfs objects
 497 *
 498 * Returns the uid/gid pair of root in the user namespace associated with the
 499 * given network namespace.
 500 */
 501void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
 502{
 503	if (net) {
 504		kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
 505		kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
 506
 507		if (uid_valid(ns_root_uid))
 508			*uid = ns_root_uid;
 509
 510		if (gid_valid(ns_root_gid))
 511			*gid = ns_root_gid;
 512	} else {
 513		*uid = GLOBAL_ROOT_UID;
 514		*gid = GLOBAL_ROOT_GID;
 515	}
 516}
 517EXPORT_SYMBOL_GPL(net_ns_get_ownership);
 518
 519static void unhash_nsid(struct net *net, struct net *last)
 520{
 521	struct net *tmp;
 522	/* This function is only called from cleanup_net() work,
 523	 * and this work is the only process, that may delete
 524	 * a net from net_namespace_list. So, when the below
 525	 * is executing, the list may only grow. Thus, we do not
 526	 * use for_each_net_rcu() or net_rwsem.
 527	 */
 528	for_each_net(tmp) {
 529		int id;
 530
 531		spin_lock_bh(&tmp->nsid_lock);
 532		id = __peernet2id(tmp, net);
 533		if (id >= 0)
 534			idr_remove(&tmp->netns_ids, id);
 535		spin_unlock_bh(&tmp->nsid_lock);
 536		if (id >= 0)
 537			rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
 538					  GFP_KERNEL);
 539		if (tmp == last)
 540			break;
 541	}
 542	spin_lock_bh(&net->nsid_lock);
 543	idr_destroy(&net->netns_ids);
 544	spin_unlock_bh(&net->nsid_lock);
 545}
 546
 547static LLIST_HEAD(cleanup_list);
 548
 549static void cleanup_net(struct work_struct *work)
 550{
 551	const struct pernet_operations *ops;
 552	struct net *net, *tmp, *last;
 553	struct llist_node *net_kill_list;
 554	LIST_HEAD(net_exit_list);
 555
 556	/* Atomically snapshot the list of namespaces to cleanup */
 557	net_kill_list = llist_del_all(&cleanup_list);
 558
 559	down_read(&pernet_ops_rwsem);
 560
 561	/* Don't let anyone else find us. */
 562	down_write(&net_rwsem);
 563	llist_for_each_entry(net, net_kill_list, cleanup_list)
 564		list_del_rcu(&net->list);
 565	/* Cache last net. After we unlock rtnl, no one new net
 566	 * added to net_namespace_list can assign nsid pointer
 567	 * to a net from net_kill_list (see peernet2id_alloc()).
 568	 * So, we skip them in unhash_nsid().
 569	 *
 570	 * Note, that unhash_nsid() does not delete nsid links
 571	 * between net_kill_list's nets, as they've already
 572	 * deleted from net_namespace_list. But, this would be
 573	 * useless anyway, as netns_ids are destroyed there.
 574	 */
 575	last = list_last_entry(&net_namespace_list, struct net, list);
 576	up_write(&net_rwsem);
 577
 578	llist_for_each_entry(net, net_kill_list, cleanup_list) {
 579		unhash_nsid(net, last);
 580		list_add_tail(&net->exit_list, &net_exit_list);
 581	}
 582
 583	/* Run all of the network namespace pre_exit methods */
 584	list_for_each_entry_reverse(ops, &pernet_list, list)
 585		ops_pre_exit_list(ops, &net_exit_list);
 586
 587	/*
 588	 * Another CPU might be rcu-iterating the list, wait for it.
 589	 * This needs to be before calling the exit() notifiers, so
 590	 * the rcu_barrier() below isn't sufficient alone.
 591	 * Also the pre_exit() and exit() methods need this barrier.
 592	 */
 593	synchronize_rcu();
 594
 595	/* Run all of the network namespace exit methods */
 596	list_for_each_entry_reverse(ops, &pernet_list, list)
 597		ops_exit_list(ops, &net_exit_list);
 598
 599	/* Free the net generic variables */
 600	list_for_each_entry_reverse(ops, &pernet_list, list)
 601		ops_free_list(ops, &net_exit_list);
 602
 603	up_read(&pernet_ops_rwsem);
 604
 605	/* Ensure there are no outstanding rcu callbacks using this
 606	 * network namespace.
 607	 */
 608	rcu_barrier();
 609
 610	/* Finally it is safe to free my network namespace structure */
 611	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 612		list_del_init(&net->exit_list);
 613		dec_net_namespaces(net->ucounts);
 614		key_remove_domain(net->key_domain);
 615		put_user_ns(net->user_ns);
 616		net_drop_ns(net);
 617	}
 618}
 619
 620/**
 621 * net_ns_barrier - wait until concurrent net_cleanup_work is done
 622 *
 623 * cleanup_net runs from work queue and will first remove namespaces
 624 * from the global list, then run net exit functions.
 625 *
 626 * Call this in module exit path to make sure that all netns
 627 * ->exit ops have been invoked before the function is removed.
 628 */
 629void net_ns_barrier(void)
 630{
 631	down_write(&pernet_ops_rwsem);
 632	up_write(&pernet_ops_rwsem);
 633}
 634EXPORT_SYMBOL(net_ns_barrier);
 635
 636static DECLARE_WORK(net_cleanup_work, cleanup_net);
 637
 638void __put_net(struct net *net)
 639{
 640	/* Cleanup the network namespace in process context */
 641	if (llist_add(&net->cleanup_list, &cleanup_list))
 642		queue_work(netns_wq, &net_cleanup_work);
 643}
 644EXPORT_SYMBOL_GPL(__put_net);
 645
 646struct net *get_net_ns_by_fd(int fd)
 647{
 648	struct file *file;
 649	struct ns_common *ns;
 650	struct net *net;
 651
 652	file = proc_ns_fget(fd);
 653	if (IS_ERR(file))
 654		return ERR_CAST(file);
 655
 656	ns = get_proc_ns(file_inode(file));
 657	if (ns->ops == &netns_operations)
 658		net = get_net(container_of(ns, struct net, ns));
 659	else
 660		net = ERR_PTR(-EINVAL);
 661
 662	fput(file);
 663	return net;
 664}
 665
 666#else
 667struct net *get_net_ns_by_fd(int fd)
 668{
 669	return ERR_PTR(-EINVAL);
 670}
 671#endif
 672EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
 673
 674struct net *get_net_ns_by_pid(pid_t pid)
 675{
 676	struct task_struct *tsk;
 677	struct net *net;
 678
 679	/* Lookup the network namespace */
 680	net = ERR_PTR(-ESRCH);
 681	rcu_read_lock();
 682	tsk = find_task_by_vpid(pid);
 683	if (tsk) {
 684		struct nsproxy *nsproxy;
 685		task_lock(tsk);
 686		nsproxy = tsk->nsproxy;
 687		if (nsproxy)
 688			net = get_net(nsproxy->net_ns);
 689		task_unlock(tsk);
 690	}
 691	rcu_read_unlock();
 692	return net;
 693}
 694EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 695
 696static __net_init int net_ns_net_init(struct net *net)
 697{
 698#ifdef CONFIG_NET_NS
 699	net->ns.ops = &netns_operations;
 700#endif
 701	return ns_alloc_inum(&net->ns);
 702}
 703
 704static __net_exit void net_ns_net_exit(struct net *net)
 705{
 706	ns_free_inum(&net->ns);
 707}
 708
 709static struct pernet_operations __net_initdata net_ns_ops = {
 710	.init = net_ns_net_init,
 711	.exit = net_ns_net_exit,
 712};
 713
 714static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
 715	[NETNSA_NONE]		= { .type = NLA_UNSPEC },
 716	[NETNSA_NSID]		= { .type = NLA_S32 },
 717	[NETNSA_PID]		= { .type = NLA_U32 },
 718	[NETNSA_FD]		= { .type = NLA_U32 },
 719	[NETNSA_TARGET_NSID]	= { .type = NLA_S32 },
 720};
 721
 722static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
 723			  struct netlink_ext_ack *extack)
 724{
 725	struct net *net = sock_net(skb->sk);
 726	struct nlattr *tb[NETNSA_MAX + 1];
 727	struct nlattr *nla;
 728	struct net *peer;
 729	int nsid, err;
 730
 731	err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
 732				     NETNSA_MAX, rtnl_net_policy, extack);
 733	if (err < 0)
 734		return err;
 735	if (!tb[NETNSA_NSID]) {
 736		NL_SET_ERR_MSG(extack, "nsid is missing");
 737		return -EINVAL;
 738	}
 739	nsid = nla_get_s32(tb[NETNSA_NSID]);
 740
 741	if (tb[NETNSA_PID]) {
 742		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 743		nla = tb[NETNSA_PID];
 744	} else if (tb[NETNSA_FD]) {
 745		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 746		nla = tb[NETNSA_FD];
 747	} else {
 748		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 749		return -EINVAL;
 750	}
 751	if (IS_ERR(peer)) {
 752		NL_SET_BAD_ATTR(extack, nla);
 753		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 754		return PTR_ERR(peer);
 755	}
 756
 757	spin_lock_bh(&net->nsid_lock);
 758	if (__peernet2id(net, peer) >= 0) {
 759		spin_unlock_bh(&net->nsid_lock);
 760		err = -EEXIST;
 761		NL_SET_BAD_ATTR(extack, nla);
 762		NL_SET_ERR_MSG(extack,
 763			       "Peer netns already has a nsid assigned");
 764		goto out;
 765	}
 766
 767	err = alloc_netid(net, peer, nsid);
 768	spin_unlock_bh(&net->nsid_lock);
 769	if (err >= 0) {
 770		rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
 771				  nlh, GFP_KERNEL);
 772		err = 0;
 773	} else if (err == -ENOSPC && nsid >= 0) {
 774		err = -EEXIST;
 775		NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
 776		NL_SET_ERR_MSG(extack, "The specified nsid is already used");
 777	}
 778out:
 779	put_net(peer);
 780	return err;
 781}
 782
 783static int rtnl_net_get_size(void)
 784{
 785	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
 786	       + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
 787	       + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
 788	       ;
 789}
 790
 791struct net_fill_args {
 792	u32 portid;
 793	u32 seq;
 794	int flags;
 795	int cmd;
 796	int nsid;
 797	bool add_ref;
 798	int ref_nsid;
 799};
 800
 801static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
 802{
 803	struct nlmsghdr *nlh;
 804	struct rtgenmsg *rth;
 805
 806	nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
 807			args->flags);
 808	if (!nlh)
 809		return -EMSGSIZE;
 810
 811	rth = nlmsg_data(nlh);
 812	rth->rtgen_family = AF_UNSPEC;
 813
 814	if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
 815		goto nla_put_failure;
 816
 817	if (args->add_ref &&
 818	    nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
 819		goto nla_put_failure;
 820
 821	nlmsg_end(skb, nlh);
 822	return 0;
 823
 824nla_put_failure:
 825	nlmsg_cancel(skb, nlh);
 826	return -EMSGSIZE;
 827}
 828
 829static int rtnl_net_valid_getid_req(struct sk_buff *skb,
 830				    const struct nlmsghdr *nlh,
 831				    struct nlattr **tb,
 832				    struct netlink_ext_ack *extack)
 833{
 834	int i, err;
 835
 836	if (!netlink_strict_get_check(skb))
 837		return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
 838					      tb, NETNSA_MAX, rtnl_net_policy,
 839					      extack);
 840
 841	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
 842					    NETNSA_MAX, rtnl_net_policy,
 843					    extack);
 844	if (err)
 845		return err;
 846
 847	for (i = 0; i <= NETNSA_MAX; i++) {
 848		if (!tb[i])
 849			continue;
 850
 851		switch (i) {
 852		case NETNSA_PID:
 853		case NETNSA_FD:
 854		case NETNSA_NSID:
 855		case NETNSA_TARGET_NSID:
 856			break;
 857		default:
 858			NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
 859			return -EINVAL;
 860		}
 861	}
 862
 863	return 0;
 864}
 865
 866static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
 867			  struct netlink_ext_ack *extack)
 868{
 869	struct net *net = sock_net(skb->sk);
 870	struct nlattr *tb[NETNSA_MAX + 1];
 871	struct net_fill_args fillargs = {
 872		.portid = NETLINK_CB(skb).portid,
 873		.seq = nlh->nlmsg_seq,
 874		.cmd = RTM_NEWNSID,
 875	};
 876	struct net *peer, *target = net;
 877	struct nlattr *nla;
 878	struct sk_buff *msg;
 879	int err;
 
 880
 881	err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
 
 882	if (err < 0)
 883		return err;
 884	if (tb[NETNSA_PID]) {
 885		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 886		nla = tb[NETNSA_PID];
 887	} else if (tb[NETNSA_FD]) {
 888		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 889		nla = tb[NETNSA_FD];
 890	} else if (tb[NETNSA_NSID]) {
 891		peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
 892		if (!peer)
 893			peer = ERR_PTR(-ENOENT);
 894		nla = tb[NETNSA_NSID];
 895	} else {
 896		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 897		return -EINVAL;
 898	}
 899
 900	if (IS_ERR(peer)) {
 901		NL_SET_BAD_ATTR(extack, nla);
 902		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 903		return PTR_ERR(peer);
 904	}
 905
 906	if (tb[NETNSA_TARGET_NSID]) {
 907		int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
 908
 909		target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
 910		if (IS_ERR(target)) {
 911			NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
 912			NL_SET_ERR_MSG(extack,
 913				       "Target netns reference is invalid");
 914			err = PTR_ERR(target);
 915			goto out;
 916		}
 917		fillargs.add_ref = true;
 918		fillargs.ref_nsid = peernet2id(net, peer);
 919	}
 920
 921	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 922	if (!msg) {
 923		err = -ENOMEM;
 924		goto out;
 925	}
 926
 927	fillargs.nsid = peernet2id(target, peer);
 928	err = rtnl_net_fill(msg, &fillargs);
 
 929	if (err < 0)
 930		goto err_out;
 931
 932	err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
 933	goto out;
 934
 935err_out:
 936	nlmsg_free(msg);
 937out:
 938	if (fillargs.add_ref)
 939		put_net(target);
 940	put_net(peer);
 941	return err;
 942}
 943
 944struct rtnl_net_dump_cb {
 945	struct net *tgt_net;
 946	struct net *ref_net;
 947	struct sk_buff *skb;
 948	struct net_fill_args fillargs;
 949	int idx;
 950	int s_idx;
 951};
 952
 953static int rtnl_net_dumpid_one(int id, void *peer, void *data)
 954{
 955	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
 956	int ret;
 957
 958	if (net_cb->idx < net_cb->s_idx)
 959		goto cont;
 960
 961	net_cb->fillargs.nsid = id;
 962	if (net_cb->fillargs.add_ref)
 963		net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
 964	ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
 965	if (ret < 0)
 966		return ret;
 967
 968cont:
 969	net_cb->idx++;
 970	return 0;
 971}
 972
 973static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
 974				   struct rtnl_net_dump_cb *net_cb,
 975				   struct netlink_callback *cb)
 976{
 977	struct netlink_ext_ack *extack = cb->extack;
 978	struct nlattr *tb[NETNSA_MAX + 1];
 979	int err, i;
 980
 981	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
 982					    NETNSA_MAX, rtnl_net_policy,
 983					    extack);
 984	if (err < 0)
 985		return err;
 986
 987	for (i = 0; i <= NETNSA_MAX; i++) {
 988		if (!tb[i])
 989			continue;
 990
 991		if (i == NETNSA_TARGET_NSID) {
 992			struct net *net;
 993
 994			net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
 995			if (IS_ERR(net)) {
 996				NL_SET_BAD_ATTR(extack, tb[i]);
 997				NL_SET_ERR_MSG(extack,
 998					       "Invalid target network namespace id");
 999				return PTR_ERR(net);
1000			}
1001			net_cb->fillargs.add_ref = true;
1002			net_cb->ref_net = net_cb->tgt_net;
1003			net_cb->tgt_net = net;
1004		} else {
1005			NL_SET_BAD_ATTR(extack, tb[i]);
1006			NL_SET_ERR_MSG(extack,
1007				       "Unsupported attribute in dump request");
1008			return -EINVAL;
1009		}
1010	}
1011
1012	return 0;
1013}
1014
1015static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1016{
 
1017	struct rtnl_net_dump_cb net_cb = {
1018		.tgt_net = sock_net(skb->sk),
1019		.skb = skb,
1020		.fillargs = {
1021			.portid = NETLINK_CB(cb->skb).portid,
1022			.seq = cb->nlh->nlmsg_seq,
1023			.flags = NLM_F_MULTI,
1024			.cmd = RTM_NEWNSID,
1025		},
1026		.idx = 0,
1027		.s_idx = cb->args[0],
1028	};
1029	int err = 0;
1030
1031	if (cb->strict_check) {
1032		err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1033		if (err < 0)
1034			goto end;
1035	}
1036
1037	spin_lock_bh(&net_cb.tgt_net->nsid_lock);
1038	if (net_cb.fillargs.add_ref &&
1039	    !net_eq(net_cb.ref_net, net_cb.tgt_net) &&
1040	    !spin_trylock_bh(&net_cb.ref_net->nsid_lock)) {
1041		spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1042		err = -EAGAIN;
1043		goto end;
1044	}
1045	idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1046	if (net_cb.fillargs.add_ref &&
1047	    !net_eq(net_cb.ref_net, net_cb.tgt_net))
1048		spin_unlock_bh(&net_cb.ref_net->nsid_lock);
1049	spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1050
1051	cb->args[0] = net_cb.idx;
1052end:
1053	if (net_cb.fillargs.add_ref)
1054		put_net(net_cb.tgt_net);
1055	return err < 0 ? err : skb->len;
1056}
1057
1058static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1059			      struct nlmsghdr *nlh, gfp_t gfp)
1060{
1061	struct net_fill_args fillargs = {
1062		.portid = portid,
1063		.seq = nlh ? nlh->nlmsg_seq : 0,
1064		.cmd = cmd,
1065		.nsid = id,
1066	};
1067	struct sk_buff *msg;
1068	int err = -ENOMEM;
1069
1070	msg = nlmsg_new(rtnl_net_get_size(), gfp);
1071	if (!msg)
1072		goto out;
1073
1074	err = rtnl_net_fill(msg, &fillargs);
1075	if (err < 0)
1076		goto err_out;
1077
1078	rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1079	return;
1080
1081err_out:
1082	nlmsg_free(msg);
1083out:
1084	rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1085}
1086
1087static int __init net_ns_init(void)
1088{
1089	struct net_generic *ng;
1090
1091#ifdef CONFIG_NET_NS
1092	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1093					SMP_CACHE_BYTES,
1094					SLAB_PANIC|SLAB_ACCOUNT, NULL);
1095
1096	/* Create workqueue for cleanup */
1097	netns_wq = create_singlethread_workqueue("netns");
1098	if (!netns_wq)
1099		panic("Could not create netns workq");
1100#endif
1101
1102	ng = net_alloc_generic();
1103	if (!ng)
1104		panic("Could not allocate generic netns");
1105
1106	rcu_assign_pointer(init_net.gen, ng);
1107
1108	down_write(&pernet_ops_rwsem);
1109	if (setup_net(&init_net, &init_user_ns))
1110		panic("Could not setup the initial network namespace");
1111
1112	init_net_initialized = true;
1113	up_write(&pernet_ops_rwsem);
1114
1115	if (register_pernet_subsys(&net_ns_ops))
1116		panic("Could not register network namespace subsystems");
1117
1118	rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1119		      RTNL_FLAG_DOIT_UNLOCKED);
1120	rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1121		      RTNL_FLAG_DOIT_UNLOCKED);
1122
1123	return 0;
1124}
1125
1126pure_initcall(net_ns_init);
1127
1128#ifdef CONFIG_NET_NS
1129static int __register_pernet_operations(struct list_head *list,
1130					struct pernet_operations *ops)
1131{
1132	struct net *net;
1133	int error;
1134	LIST_HEAD(net_exit_list);
1135
1136	list_add_tail(&ops->list, list);
1137	if (ops->init || (ops->id && ops->size)) {
1138		/* We held write locked pernet_ops_rwsem, and parallel
1139		 * setup_net() and cleanup_net() are not possible.
1140		 */
1141		for_each_net(net) {
1142			error = ops_init(ops, net);
1143			if (error)
1144				goto out_undo;
1145			list_add_tail(&net->exit_list, &net_exit_list);
1146		}
1147	}
1148	return 0;
1149
1150out_undo:
1151	/* If I have an error cleanup all namespaces I initialized */
1152	list_del(&ops->list);
1153	ops_pre_exit_list(ops, &net_exit_list);
1154	synchronize_rcu();
1155	ops_exit_list(ops, &net_exit_list);
1156	ops_free_list(ops, &net_exit_list);
1157	return error;
1158}
1159
1160static void __unregister_pernet_operations(struct pernet_operations *ops)
1161{
1162	struct net *net;
1163	LIST_HEAD(net_exit_list);
1164
1165	list_del(&ops->list);
1166	/* See comment in __register_pernet_operations() */
1167	for_each_net(net)
1168		list_add_tail(&net->exit_list, &net_exit_list);
1169	ops_pre_exit_list(ops, &net_exit_list);
1170	synchronize_rcu();
1171	ops_exit_list(ops, &net_exit_list);
1172	ops_free_list(ops, &net_exit_list);
1173}
1174
1175#else
1176
1177static int __register_pernet_operations(struct list_head *list,
1178					struct pernet_operations *ops)
1179{
1180	if (!init_net_initialized) {
1181		list_add_tail(&ops->list, list);
1182		return 0;
1183	}
1184
1185	return ops_init(ops, &init_net);
1186}
1187
1188static void __unregister_pernet_operations(struct pernet_operations *ops)
1189{
1190	if (!init_net_initialized) {
1191		list_del(&ops->list);
1192	} else {
1193		LIST_HEAD(net_exit_list);
1194		list_add(&init_net.exit_list, &net_exit_list);
1195		ops_pre_exit_list(ops, &net_exit_list);
1196		synchronize_rcu();
1197		ops_exit_list(ops, &net_exit_list);
1198		ops_free_list(ops, &net_exit_list);
1199	}
1200}
1201
1202#endif /* CONFIG_NET_NS */
1203
1204static DEFINE_IDA(net_generic_ids);
1205
1206static int register_pernet_operations(struct list_head *list,
1207				      struct pernet_operations *ops)
1208{
1209	int error;
1210
1211	if (ops->id) {
1212		error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1213				GFP_KERNEL);
1214		if (error < 0)
 
 
 
 
1215			return error;
1216		*ops->id = error;
1217		max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1218	}
1219	error = __register_pernet_operations(list, ops);
1220	if (error) {
1221		rcu_barrier();
1222		if (ops->id)
1223			ida_free(&net_generic_ids, *ops->id);
1224	}
1225
1226	return error;
1227}
1228
1229static void unregister_pernet_operations(struct pernet_operations *ops)
1230{
1231	__unregister_pernet_operations(ops);
1232	rcu_barrier();
1233	if (ops->id)
1234		ida_free(&net_generic_ids, *ops->id);
1235}
1236
1237/**
1238 *      register_pernet_subsys - register a network namespace subsystem
1239 *	@ops:  pernet operations structure for the subsystem
1240 *
1241 *	Register a subsystem which has init and exit functions
1242 *	that are called when network namespaces are created and
1243 *	destroyed respectively.
1244 *
1245 *	When registered all network namespace init functions are
1246 *	called for every existing network namespace.  Allowing kernel
1247 *	modules to have a race free view of the set of network namespaces.
1248 *
1249 *	When a new network namespace is created all of the init
1250 *	methods are called in the order in which they were registered.
1251 *
1252 *	When a network namespace is destroyed all of the exit methods
1253 *	are called in the reverse of the order with which they were
1254 *	registered.
1255 */
1256int register_pernet_subsys(struct pernet_operations *ops)
1257{
1258	int error;
1259	down_write(&pernet_ops_rwsem);
1260	error =  register_pernet_operations(first_device, ops);
1261	up_write(&pernet_ops_rwsem);
1262	return error;
1263}
1264EXPORT_SYMBOL_GPL(register_pernet_subsys);
1265
1266/**
1267 *      unregister_pernet_subsys - unregister a network namespace subsystem
1268 *	@ops: pernet operations structure to manipulate
1269 *
1270 *	Remove the pernet operations structure from the list to be
1271 *	used when network namespaces are created or destroyed.  In
1272 *	addition run the exit method for all existing network
1273 *	namespaces.
1274 */
1275void unregister_pernet_subsys(struct pernet_operations *ops)
1276{
1277	down_write(&pernet_ops_rwsem);
1278	unregister_pernet_operations(ops);
1279	up_write(&pernet_ops_rwsem);
1280}
1281EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1282
1283/**
1284 *      register_pernet_device - register a network namespace device
1285 *	@ops:  pernet operations structure for the subsystem
1286 *
1287 *	Register a device which has init and exit functions
1288 *	that are called when network namespaces are created and
1289 *	destroyed respectively.
1290 *
1291 *	When registered all network namespace init functions are
1292 *	called for every existing network namespace.  Allowing kernel
1293 *	modules to have a race free view of the set of network namespaces.
1294 *
1295 *	When a new network namespace is created all of the init
1296 *	methods are called in the order in which they were registered.
1297 *
1298 *	When a network namespace is destroyed all of the exit methods
1299 *	are called in the reverse of the order with which they were
1300 *	registered.
1301 */
1302int register_pernet_device(struct pernet_operations *ops)
1303{
1304	int error;
1305	down_write(&pernet_ops_rwsem);
1306	error = register_pernet_operations(&pernet_list, ops);
1307	if (!error && (first_device == &pernet_list))
1308		first_device = &ops->list;
1309	up_write(&pernet_ops_rwsem);
1310	return error;
1311}
1312EXPORT_SYMBOL_GPL(register_pernet_device);
1313
1314/**
1315 *      unregister_pernet_device - unregister a network namespace netdevice
1316 *	@ops: pernet operations structure to manipulate
1317 *
1318 *	Remove the pernet operations structure from the list to be
1319 *	used when network namespaces are created or destroyed.  In
1320 *	addition run the exit method for all existing network
1321 *	namespaces.
1322 */
1323void unregister_pernet_device(struct pernet_operations *ops)
1324{
1325	down_write(&pernet_ops_rwsem);
1326	if (&ops->list == first_device)
1327		first_device = first_device->next;
1328	unregister_pernet_operations(ops);
1329	up_write(&pernet_ops_rwsem);
1330}
1331EXPORT_SYMBOL_GPL(unregister_pernet_device);
1332
1333#ifdef CONFIG_NET_NS
1334static struct ns_common *netns_get(struct task_struct *task)
1335{
1336	struct net *net = NULL;
1337	struct nsproxy *nsproxy;
1338
1339	task_lock(task);
1340	nsproxy = task->nsproxy;
1341	if (nsproxy)
1342		net = get_net(nsproxy->net_ns);
1343	task_unlock(task);
1344
1345	return net ? &net->ns : NULL;
1346}
1347
1348static inline struct net *to_net_ns(struct ns_common *ns)
1349{
1350	return container_of(ns, struct net, ns);
1351}
1352
1353static void netns_put(struct ns_common *ns)
1354{
1355	put_net(to_net_ns(ns));
1356}
1357
1358static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1359{
1360	struct net *net = to_net_ns(ns);
1361
1362	if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1363	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1364		return -EPERM;
1365
1366	put_net(nsproxy->net_ns);
1367	nsproxy->net_ns = get_net(net);
1368	return 0;
1369}
1370
1371static struct user_namespace *netns_owner(struct ns_common *ns)
1372{
1373	return to_net_ns(ns)->user_ns;
1374}
1375
1376const struct proc_ns_operations netns_operations = {
1377	.name		= "net",
1378	.type		= CLONE_NEWNET,
1379	.get		= netns_get,
1380	.put		= netns_put,
1381	.install	= netns_install,
1382	.owner		= netns_owner,
1383};
1384#endif