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
v6.13.7
   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#include <linux/cookie.h>
  23#include <linux/proc_fs.h>
  24
  25#include <net/sock.h>
  26#include <net/netlink.h>
  27#include <net/net_namespace.h>
  28#include <net/netns/generic.h>
  29
  30/*
  31 *	Our network namespace constructor/destructor lists
  32 */
  33
  34static LIST_HEAD(pernet_list);
  35static struct list_head *first_device = &pernet_list;
 
  36
  37LIST_HEAD(net_namespace_list);
  38EXPORT_SYMBOL_GPL(net_namespace_list);
  39
  40/* Protects net_namespace_list. Nests iside rtnl_lock() */
  41DECLARE_RWSEM(net_rwsem);
  42EXPORT_SYMBOL_GPL(net_rwsem);
  43
  44#ifdef CONFIG_KEYS
  45static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
  46#endif
  47
  48struct net init_net;
  49EXPORT_SYMBOL(init_net);
  50
  51static bool init_net_initialized;
  52/*
  53 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
  54 * init_net_initialized and first_device pointer.
  55 * This is internal net namespace object. Please, don't use it
  56 * outside.
  57 */
  58DECLARE_RWSEM(pernet_ops_rwsem);
  59
  60#define MIN_PERNET_OPS_ID	\
  61	((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
  62
  63#define INITIAL_NET_GEN_PTRS	13 /* +1 for len +2 for rcu_head */
  64
  65static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  66
  67DEFINE_COOKIE(net_cookie);
  68
  69static struct net_generic *net_alloc_generic(void)
  70{
  71	unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs);
  72	unsigned int generic_size;
  73	struct net_generic *ng;
  74
  75	generic_size = offsetof(struct net_generic, ptr[gen_ptrs]);
  76
  77	ng = kzalloc(generic_size, GFP_KERNEL);
  78	if (ng)
  79		ng->s.len = 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)
  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	struct net_generic *ng;
 124	int err = -ENOMEM;
 125	void *data = NULL;
 126
 127	if (ops->id) {
 128		data = kzalloc(ops->size, GFP_KERNEL);
 129		if (!data)
 130			goto out;
 131
 132		err = net_assign_generic(net, *ops->id, data);
 133		if (err)
 134			goto cleanup;
 135	}
 136	err = 0;
 137	if (ops->init)
 138		err = ops->init(net);
 139	if (!err)
 140		return 0;
 141
 142	if (ops->id) {
 143		ng = rcu_dereference_protected(net->gen,
 144					       lockdep_is_held(&pernet_ops_rwsem));
 145		ng->ptr[*ops->id] = NULL;
 146	}
 147
 148cleanup:
 149	kfree(data);
 150
 151out:
 152	return err;
 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			cond_resched();
 174		}
 175	}
 176	if (ops->exit_batch)
 177		ops->exit_batch(net_exit_list);
 178}
 179
 180static void ops_free_list(const struct pernet_operations *ops,
 181			  struct list_head *net_exit_list)
 182{
 183	struct net *net;
 184
 185	if (ops->id) {
 186		list_for_each_entry(net, net_exit_list, exit_list)
 187			kfree(net_generic(net, *ops->id));
 188	}
 189}
 190
 191/* should be called with nsid_lock held */
 192static int alloc_netid(struct net *net, struct net *peer, int reqid)
 193{
 194	int min = 0, max = 0;
 195
 196	if (reqid >= 0) {
 197		min = reqid;
 198		max = reqid + 1;
 199	}
 200
 201	return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
 202}
 203
 204/* This function is used by idr_for_each(). If net is equal to peer, the
 205 * function returns the id so that idr_for_each() stops. Because we cannot
 206 * returns the id 0 (idr_for_each() will not stop), we return the magic value
 207 * NET_ID_ZERO (-1) for it.
 208 */
 209#define NET_ID_ZERO -1
 210static int net_eq_idr(int id, void *net, void *peer)
 211{
 212	if (net_eq(net, peer))
 213		return id ? : NET_ID_ZERO;
 214	return 0;
 215}
 216
 217/* Must be called from RCU-critical section or with nsid_lock held */
 218static int __peernet2id(const struct net *net, struct net *peer)
 
 
 
 219{
 220	int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
 
 
 
 221
 222	/* Magic value for id 0. */
 223	if (id == NET_ID_ZERO)
 224		return 0;
 225	if (id > 0)
 226		return id;
 227
 
 
 
 
 
 
 228	return NETNSA_NSID_NOT_ASSIGNED;
 229}
 230
 231static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
 232			      struct nlmsghdr *nlh, gfp_t gfp);
 
 
 
 
 
 
 
 233/* This function returns the id of a peer netns. If no id is assigned, one will
 234 * be allocated and returned.
 235 */
 236int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
 237{
 
 
 238	int id;
 239
 240	if (refcount_read(&net->ns.count) == 0)
 241		return NETNSA_NSID_NOT_ASSIGNED;
 242
 243	spin_lock_bh(&net->nsid_lock);
 244	id = __peernet2id(net, peer);
 245	if (id >= 0) {
 246		spin_unlock_bh(&net->nsid_lock);
 247		return id;
 248	}
 249
 250	/* When peer is obtained from RCU lists, we may race with
 251	 * its cleanup. Check whether it's alive, and this guarantees
 252	 * we never hash a peer back to net->netns_ids, after it has
 253	 * just been idr_remove()'d from there in cleanup_net().
 254	 */
 255	if (!maybe_get_net(peer)) {
 256		spin_unlock_bh(&net->nsid_lock);
 257		return NETNSA_NSID_NOT_ASSIGNED;
 258	}
 259
 260	id = alloc_netid(net, peer, -1);
 261	spin_unlock_bh(&net->nsid_lock);
 262
 263	put_net(peer);
 264	if (id < 0)
 265		return NETNSA_NSID_NOT_ASSIGNED;
 266
 267	rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
 268
 269	return id;
 270}
 271EXPORT_SYMBOL_GPL(peernet2id_alloc);
 272
 273/* This function returns, if assigned, the id of a peer netns. */
 274int peernet2id(const struct net *net, struct net *peer)
 275{
 
 276	int id;
 277
 278	rcu_read_lock();
 279	id = __peernet2id(net, peer);
 280	rcu_read_unlock();
 281
 282	return id;
 283}
 284EXPORT_SYMBOL(peernet2id);
 285
 286/* This function returns true is the peer netns has an id assigned into the
 287 * current netns.
 288 */
 289bool peernet_has_id(const struct net *net, struct net *peer)
 290{
 291	return peernet2id(net, peer) >= 0;
 292}
 293
 294struct net *get_net_ns_by_id(const struct net *net, int id)
 295{
 
 296	struct net *peer;
 297
 298	if (id < 0)
 299		return NULL;
 300
 301	rcu_read_lock();
 
 302	peer = idr_find(&net->netns_ids, id);
 303	if (peer)
 304		peer = maybe_get_net(peer);
 
 305	rcu_read_unlock();
 306
 307	return peer;
 308}
 309EXPORT_SYMBOL_GPL(get_net_ns_by_id);
 310
 311static __net_init void preinit_net_sysctl(struct net *net)
 312{
 313	net->core.sysctl_somaxconn = SOMAXCONN;
 314	/* Limits per socket sk_omem_alloc usage.
 315	 * TCP zerocopy regular usage needs 128 KB.
 316	 */
 317	net->core.sysctl_optmem_max = 128 * 1024;
 318	net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
 319	net->core.sysctl_tstamp_allow_data = 1;
 320}
 321
 322/* init code that must occur even if setup_net() is not called. */
 323static __net_init void preinit_net(struct net *net, struct user_namespace *user_ns)
 324{
 325	refcount_set(&net->passive, 1);
 326	refcount_set(&net->ns.count, 1);
 327	ref_tracker_dir_init(&net->refcnt_tracker, 128, "net refcnt");
 328	ref_tracker_dir_init(&net->notrefcnt_tracker, 128, "net notrefcnt");
 329
 330	get_random_bytes(&net->hash_mix, sizeof(u32));
 331	net->dev_base_seq = 1;
 332	net->user_ns = user_ns;
 333
 334	idr_init(&net->netns_ids);
 335	spin_lock_init(&net->nsid_lock);
 336	mutex_init(&net->ipv4.ra_mutex);
 337
 338#ifdef CONFIG_DEBUG_NET_SMALL_RTNL
 339	mutex_init(&net->rtnl_mutex);
 340	lock_set_cmp_fn(&net->rtnl_mutex, rtnl_net_lock_cmp_fn, NULL);
 341#endif
 342
 343	preinit_net_sysctl(net);
 344}
 345
 346/*
 347 * setup_net runs the initializers for the network namespace object.
 348 */
 349static __net_init int setup_net(struct net *net)
 350{
 351	/* Must be called with pernet_ops_rwsem held */
 352	const struct pernet_operations *ops, *saved_ops;
 
 353	LIST_HEAD(net_exit_list);
 354	LIST_HEAD(dev_kill_list);
 355	int error = 0;
 356
 357	preempt_disable();
 358	net->net_cookie = gen_cookie_next(&net_cookie);
 359	preempt_enable();
 
 
 
 360
 361	list_for_each_entry(ops, &pernet_list, list) {
 362		error = ops_init(ops, net);
 363		if (error < 0)
 364			goto out_undo;
 365	}
 366	down_write(&net_rwsem);
 367	list_add_tail_rcu(&net->list, &net_namespace_list);
 368	up_write(&net_rwsem);
 369out:
 370	return error;
 371
 372out_undo:
 373	/* Walk through the list backwards calling the exit functions
 374	 * for the pernet modules whose init functions did not fail.
 375	 */
 376	list_add(&net->exit_list, &net_exit_list);
 377	saved_ops = ops;
 378	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 379		ops_pre_exit_list(ops, &net_exit_list);
 380
 381	synchronize_rcu();
 382
 383	ops = saved_ops;
 384	rtnl_lock();
 385	list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
 386		if (ops->exit_batch_rtnl)
 387			ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
 388	}
 389	unregister_netdevice_many(&dev_kill_list);
 390	rtnl_unlock();
 391
 392	ops = saved_ops;
 393	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 394		ops_exit_list(ops, &net_exit_list);
 395
 396	ops = saved_ops;
 397	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 398		ops_free_list(ops, &net_exit_list);
 399
 400	rcu_barrier();
 401	goto out;
 402}
 403
 
 404#ifdef CONFIG_NET_NS
 405static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
 406{
 407	return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
 408}
 409
 410static void dec_net_namespaces(struct ucounts *ucounts)
 411{
 412	dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
 413}
 414
 415static struct kmem_cache *net_cachep __ro_after_init;
 416static struct workqueue_struct *netns_wq;
 417
 418static struct net *net_alloc(void)
 419{
 420	struct net *net = NULL;
 421	struct net_generic *ng;
 422
 423	ng = net_alloc_generic();
 424	if (!ng)
 425		goto out;
 426
 427	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 428	if (!net)
 429		goto out_free;
 430
 431#ifdef CONFIG_KEYS
 432	net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
 433	if (!net->key_domain)
 434		goto out_free_2;
 435	refcount_set(&net->key_domain->usage, 1);
 436#endif
 437
 438	rcu_assign_pointer(net->gen, ng);
 439out:
 440	return net;
 441
 442#ifdef CONFIG_KEYS
 443out_free_2:
 444	kmem_cache_free(net_cachep, net);
 445	net = NULL;
 446#endif
 447out_free:
 448	kfree(ng);
 449	goto out;
 450}
 451
 452static LLIST_HEAD(defer_free_list);
 453
 454static void net_complete_free(void)
 455{
 456	struct llist_node *kill_list;
 457	struct net *net, *next;
 458
 459	/* Get the list of namespaces to free from last round. */
 460	kill_list = llist_del_all(&defer_free_list);
 461
 462	llist_for_each_entry_safe(net, next, kill_list, defer_free_list)
 463		kmem_cache_free(net_cachep, net);
 464
 465}
 466
 467void net_passive_dec(struct net *net)
 468{
 469	if (refcount_dec_and_test(&net->passive)) {
 470		kfree(rcu_access_pointer(net->gen));
 471
 472		/* There should not be any trackers left there. */
 473		ref_tracker_dir_exit(&net->notrefcnt_tracker);
 474
 475		/* Wait for an extra rcu_barrier() before final free. */
 476		llist_add(&net->defer_free_list, &defer_free_list);
 477	}
 478}
 479
 480void net_drop_ns(void *p)
 481{
 482	struct net *net = (struct net *)p;
 483
 484	if (net)
 485		net_passive_dec(net);
 486}
 487
 488struct net *copy_net_ns(unsigned long flags,
 489			struct user_namespace *user_ns, struct net *old_net)
 490{
 491	struct ucounts *ucounts;
 492	struct net *net;
 493	int rv;
 494
 495	if (!(flags & CLONE_NEWNET))
 496		return get_net(old_net);
 497
 498	ucounts = inc_net_namespaces(user_ns);
 499	if (!ucounts)
 500		return ERR_PTR(-ENOSPC);
 501
 502	net = net_alloc();
 503	if (!net) {
 504		rv = -ENOMEM;
 505		goto dec_ucounts;
 506	}
 507
 508	preinit_net(net, user_ns);
 509	net->ucounts = ucounts;
 510	get_user_ns(user_ns);
 511
 512	rv = down_read_killable(&pernet_ops_rwsem);
 513	if (rv < 0)
 514		goto put_userns;
 515
 516	rv = setup_net(net);
 517
 518	up_read(&pernet_ops_rwsem);
 519
 520	if (rv < 0) {
 521put_userns:
 522#ifdef CONFIG_KEYS
 523		key_remove_domain(net->key_domain);
 524#endif
 525		put_user_ns(user_ns);
 526		net_passive_dec(net);
 527dec_ucounts:
 528		dec_net_namespaces(ucounts);
 529		return ERR_PTR(rv);
 530	}
 531	return net;
 532}
 533
 534/**
 535 * net_ns_get_ownership - get sysfs ownership data for @net
 536 * @net: network namespace in question (can be NULL)
 537 * @uid: kernel user ID for sysfs objects
 538 * @gid: kernel group ID for sysfs objects
 539 *
 540 * Returns the uid/gid pair of root in the user namespace associated with the
 541 * given network namespace.
 542 */
 543void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
 544{
 545	if (net) {
 546		kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
 547		kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
 548
 549		if (uid_valid(ns_root_uid))
 550			*uid = ns_root_uid;
 551
 552		if (gid_valid(ns_root_gid))
 553			*gid = ns_root_gid;
 554	} else {
 555		*uid = GLOBAL_ROOT_UID;
 556		*gid = GLOBAL_ROOT_GID;
 557	}
 558}
 559EXPORT_SYMBOL_GPL(net_ns_get_ownership);
 560
 561static void unhash_nsid(struct net *net, struct net *last)
 562{
 563	struct net *tmp;
 564	/* This function is only called from cleanup_net() work,
 565	 * and this work is the only process, that may delete
 566	 * a net from net_namespace_list. So, when the below
 567	 * is executing, the list may only grow. Thus, we do not
 568	 * use for_each_net_rcu() or net_rwsem.
 569	 */
 570	for_each_net(tmp) {
 571		int id;
 572
 573		spin_lock_bh(&tmp->nsid_lock);
 574		id = __peernet2id(tmp, net);
 575		if (id >= 0)
 576			idr_remove(&tmp->netns_ids, id);
 577		spin_unlock_bh(&tmp->nsid_lock);
 578		if (id >= 0)
 579			rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
 580					  GFP_KERNEL);
 581		if (tmp == last)
 582			break;
 583	}
 584	spin_lock_bh(&net->nsid_lock);
 585	idr_destroy(&net->netns_ids);
 586	spin_unlock_bh(&net->nsid_lock);
 587}
 588
 589static LLIST_HEAD(cleanup_list);
 590
 591static void cleanup_net(struct work_struct *work)
 592{
 593	const struct pernet_operations *ops;
 594	struct net *net, *tmp, *last;
 595	struct llist_node *net_kill_list;
 596	LIST_HEAD(net_exit_list);
 597	LIST_HEAD(dev_kill_list);
 598
 599	/* Atomically snapshot the list of namespaces to cleanup */
 600	net_kill_list = llist_del_all(&cleanup_list);
 
 
 601
 602	down_read(&pernet_ops_rwsem);
 603
 604	/* Don't let anyone else find us. */
 605	down_write(&net_rwsem);
 606	llist_for_each_entry(net, net_kill_list, cleanup_list)
 607		list_del_rcu(&net->list);
 608	/* Cache last net. After we unlock rtnl, no one new net
 609	 * added to net_namespace_list can assign nsid pointer
 610	 * to a net from net_kill_list (see peernet2id_alloc()).
 611	 * So, we skip them in unhash_nsid().
 612	 *
 613	 * Note, that unhash_nsid() does not delete nsid links
 614	 * between net_kill_list's nets, as they've already
 615	 * deleted from net_namespace_list. But, this would be
 616	 * useless anyway, as netns_ids are destroyed there.
 617	 */
 618	last = list_last_entry(&net_namespace_list, struct net, list);
 619	up_write(&net_rwsem);
 
 
 
 620
 621	llist_for_each_entry(net, net_kill_list, cleanup_list) {
 622		unhash_nsid(net, last);
 623		list_add_tail(&net->exit_list, &net_exit_list);
 624	}
 625
 626	/* Run all of the network namespace pre_exit methods */
 627	list_for_each_entry_reverse(ops, &pernet_list, list)
 628		ops_pre_exit_list(ops, &net_exit_list);
 629
 630	/*
 631	 * Another CPU might be rcu-iterating the list, wait for it.
 632	 * This needs to be before calling the exit() notifiers, so
 633	 * the rcu_barrier() below isn't sufficient alone.
 634	 * Also the pre_exit() and exit() methods need this barrier.
 635	 */
 636	synchronize_rcu_expedited();
 637
 638	rtnl_lock();
 639	list_for_each_entry_reverse(ops, &pernet_list, list) {
 640		if (ops->exit_batch_rtnl)
 641			ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
 642	}
 643	unregister_netdevice_many(&dev_kill_list);
 644	rtnl_unlock();
 645
 646	/* Run all of the network namespace exit methods */
 647	list_for_each_entry_reverse(ops, &pernet_list, list)
 648		ops_exit_list(ops, &net_exit_list);
 649
 650	/* Free the net generic variables */
 651	list_for_each_entry_reverse(ops, &pernet_list, list)
 652		ops_free_list(ops, &net_exit_list);
 653
 654	up_read(&pernet_ops_rwsem);
 655
 656	/* Ensure there are no outstanding rcu callbacks using this
 657	 * network namespace.
 658	 */
 659	rcu_barrier();
 660
 661	net_complete_free();
 662
 663	/* Finally it is safe to free my network namespace structure */
 664	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 665		list_del_init(&net->exit_list);
 666		dec_net_namespaces(net->ucounts);
 667#ifdef CONFIG_KEYS
 668		key_remove_domain(net->key_domain);
 669#endif
 670		put_user_ns(net->user_ns);
 671		net_passive_dec(net);
 672	}
 673}
 674
 675/**
 676 * net_ns_barrier - wait until concurrent net_cleanup_work is done
 677 *
 678 * cleanup_net runs from work queue and will first remove namespaces
 679 * from the global list, then run net exit functions.
 680 *
 681 * Call this in module exit path to make sure that all netns
 682 * ->exit ops have been invoked before the function is removed.
 683 */
 684void net_ns_barrier(void)
 685{
 686	down_write(&pernet_ops_rwsem);
 687	up_write(&pernet_ops_rwsem);
 688}
 689EXPORT_SYMBOL(net_ns_barrier);
 690
 691static DECLARE_WORK(net_cleanup_work, cleanup_net);
 692
 693void __put_net(struct net *net)
 694{
 695	ref_tracker_dir_exit(&net->refcnt_tracker);
 696	/* Cleanup the network namespace in process context */
 697	if (llist_add(&net->cleanup_list, &cleanup_list))
 698		queue_work(netns_wq, &net_cleanup_work);
 
 
 
 
 
 699}
 700EXPORT_SYMBOL_GPL(__put_net);
 701
 702/**
 703 * get_net_ns - increment the refcount of the network namespace
 704 * @ns: common namespace (net)
 705 *
 706 * Returns the net's common namespace or ERR_PTR() if ref is zero.
 707 */
 708struct ns_common *get_net_ns(struct ns_common *ns)
 709{
 
 
 710	struct net *net;
 711
 712	net = maybe_get_net(container_of(ns, struct net, ns));
 713	if (net)
 714		return &net->ns;
 715	return ERR_PTR(-EINVAL);
 
 
 
 
 
 
 
 
 716}
 717EXPORT_SYMBOL_GPL(get_net_ns);
 718
 
 719struct net *get_net_ns_by_fd(int fd)
 720{
 721	CLASS(fd, f)(fd);
 722
 723	if (fd_empty(f))
 724		return ERR_PTR(-EBADF);
 725
 726	if (proc_ns_file(fd_file(f))) {
 727		struct ns_common *ns = get_proc_ns(file_inode(fd_file(f)));
 728		if (ns->ops == &netns_operations)
 729			return get_net(container_of(ns, struct net, ns));
 730	}
 731
 732	return ERR_PTR(-EINVAL);
 733}
 
 734EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
 735#endif
 736
 737struct net *get_net_ns_by_pid(pid_t pid)
 738{
 739	struct task_struct *tsk;
 740	struct net *net;
 741
 742	/* Lookup the network namespace */
 743	net = ERR_PTR(-ESRCH);
 744	rcu_read_lock();
 745	tsk = find_task_by_vpid(pid);
 746	if (tsk) {
 747		struct nsproxy *nsproxy;
 748		task_lock(tsk);
 749		nsproxy = tsk->nsproxy;
 750		if (nsproxy)
 751			net = get_net(nsproxy->net_ns);
 752		task_unlock(tsk);
 753	}
 754	rcu_read_unlock();
 755	return net;
 756}
 757EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 758
 759static __net_init int net_ns_net_init(struct net *net)
 760{
 761#ifdef CONFIG_NET_NS
 762	net->ns.ops = &netns_operations;
 763#endif
 764	return ns_alloc_inum(&net->ns);
 765}
 766
 767static __net_exit void net_ns_net_exit(struct net *net)
 768{
 769	ns_free_inum(&net->ns);
 770}
 771
 772static struct pernet_operations __net_initdata net_ns_ops = {
 773	.init = net_ns_net_init,
 774	.exit = net_ns_net_exit,
 775};
 776
 777static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
 778	[NETNSA_NONE]		= { .type = NLA_UNSPEC },
 779	[NETNSA_NSID]		= { .type = NLA_S32 },
 780	[NETNSA_PID]		= { .type = NLA_U32 },
 781	[NETNSA_FD]		= { .type = NLA_U32 },
 782	[NETNSA_TARGET_NSID]	= { .type = NLA_S32 },
 783};
 784
 785static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
 786			  struct netlink_ext_ack *extack)
 787{
 788	struct net *net = sock_net(skb->sk);
 789	struct nlattr *tb[NETNSA_MAX + 1];
 790	struct nlattr *nla;
 791	struct net *peer;
 792	int nsid, err;
 793
 794	err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
 795				     NETNSA_MAX, rtnl_net_policy, extack);
 796	if (err < 0)
 797		return err;
 798	if (!tb[NETNSA_NSID]) {
 799		NL_SET_ERR_MSG(extack, "nsid is missing");
 800		return -EINVAL;
 801	}
 802	nsid = nla_get_s32(tb[NETNSA_NSID]);
 803
 804	if (tb[NETNSA_PID]) {
 805		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 806		nla = tb[NETNSA_PID];
 807	} else if (tb[NETNSA_FD]) {
 808		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 809		nla = tb[NETNSA_FD];
 810	} else {
 811		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 812		return -EINVAL;
 813	}
 814	if (IS_ERR(peer)) {
 815		NL_SET_BAD_ATTR(extack, nla);
 816		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 817		return PTR_ERR(peer);
 818	}
 819
 820	spin_lock_bh(&net->nsid_lock);
 821	if (__peernet2id(net, peer) >= 0) {
 822		spin_unlock_bh(&net->nsid_lock);
 823		err = -EEXIST;
 824		NL_SET_BAD_ATTR(extack, nla);
 825		NL_SET_ERR_MSG(extack,
 826			       "Peer netns already has a nsid assigned");
 827		goto out;
 828	}
 829
 830	err = alloc_netid(net, peer, nsid);
 831	spin_unlock_bh(&net->nsid_lock);
 832	if (err >= 0) {
 833		rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
 834				  nlh, GFP_KERNEL);
 835		err = 0;
 836	} else if (err == -ENOSPC && nsid >= 0) {
 837		err = -EEXIST;
 838		NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
 839		NL_SET_ERR_MSG(extack, "The specified nsid is already used");
 840	}
 841out:
 842	put_net(peer);
 843	return err;
 844}
 845
 846static int rtnl_net_get_size(void)
 847{
 848	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
 849	       + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
 850	       + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
 851	       ;
 852}
 853
 854struct net_fill_args {
 855	u32 portid;
 856	u32 seq;
 857	int flags;
 858	int cmd;
 859	int nsid;
 860	bool add_ref;
 861	int ref_nsid;
 862};
 863
 864static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
 865{
 866	struct nlmsghdr *nlh;
 867	struct rtgenmsg *rth;
 868
 869	nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
 870			args->flags);
 871	if (!nlh)
 872		return -EMSGSIZE;
 873
 874	rth = nlmsg_data(nlh);
 875	rth->rtgen_family = AF_UNSPEC;
 876
 877	if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
 878		goto nla_put_failure;
 879
 880	if (args->add_ref &&
 881	    nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
 882		goto nla_put_failure;
 883
 884	nlmsg_end(skb, nlh);
 885	return 0;
 886
 887nla_put_failure:
 888	nlmsg_cancel(skb, nlh);
 889	return -EMSGSIZE;
 890}
 891
 892static int rtnl_net_valid_getid_req(struct sk_buff *skb,
 893				    const struct nlmsghdr *nlh,
 894				    struct nlattr **tb,
 895				    struct netlink_ext_ack *extack)
 896{
 897	int i, err;
 898
 899	if (!netlink_strict_get_check(skb))
 900		return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
 901					      tb, NETNSA_MAX, rtnl_net_policy,
 902					      extack);
 903
 904	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
 905					    NETNSA_MAX, rtnl_net_policy,
 906					    extack);
 907	if (err)
 908		return err;
 909
 910	for (i = 0; i <= NETNSA_MAX; i++) {
 911		if (!tb[i])
 912			continue;
 913
 914		switch (i) {
 915		case NETNSA_PID:
 916		case NETNSA_FD:
 917		case NETNSA_NSID:
 918		case NETNSA_TARGET_NSID:
 919			break;
 920		default:
 921			NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
 922			return -EINVAL;
 923		}
 924	}
 925
 926	return 0;
 927}
 928
 929static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
 930			  struct netlink_ext_ack *extack)
 931{
 932	struct net *net = sock_net(skb->sk);
 933	struct nlattr *tb[NETNSA_MAX + 1];
 934	struct net_fill_args fillargs = {
 935		.portid = NETLINK_CB(skb).portid,
 936		.seq = nlh->nlmsg_seq,
 937		.cmd = RTM_NEWNSID,
 938	};
 939	struct net *peer, *target = net;
 940	struct nlattr *nla;
 941	struct sk_buff *msg;
 942	int err;
 
 943
 944	err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
 
 945	if (err < 0)
 946		return err;
 947	if (tb[NETNSA_PID]) {
 948		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 949		nla = tb[NETNSA_PID];
 950	} else if (tb[NETNSA_FD]) {
 951		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 952		nla = tb[NETNSA_FD];
 953	} else if (tb[NETNSA_NSID]) {
 954		peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
 955		if (!peer)
 956			peer = ERR_PTR(-ENOENT);
 957		nla = tb[NETNSA_NSID];
 958	} else {
 959		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 960		return -EINVAL;
 961	}
 962
 963	if (IS_ERR(peer)) {
 964		NL_SET_BAD_ATTR(extack, nla);
 965		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 966		return PTR_ERR(peer);
 967	}
 968
 969	if (tb[NETNSA_TARGET_NSID]) {
 970		int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
 971
 972		target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
 973		if (IS_ERR(target)) {
 974			NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
 975			NL_SET_ERR_MSG(extack,
 976				       "Target netns reference is invalid");
 977			err = PTR_ERR(target);
 978			goto out;
 979		}
 980		fillargs.add_ref = true;
 981		fillargs.ref_nsid = peernet2id(net, peer);
 982	}
 983
 984	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 985	if (!msg) {
 986		err = -ENOMEM;
 987		goto out;
 988	}
 989
 990	fillargs.nsid = peernet2id(target, peer);
 991	err = rtnl_net_fill(msg, &fillargs);
 
 992	if (err < 0)
 993		goto err_out;
 994
 995	err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
 996	goto out;
 997
 998err_out:
 999	nlmsg_free(msg);
1000out:
1001	if (fillargs.add_ref)
1002		put_net(target);
1003	put_net(peer);
1004	return err;
1005}
1006
1007struct rtnl_net_dump_cb {
1008	struct net *tgt_net;
1009	struct net *ref_net;
1010	struct sk_buff *skb;
1011	struct net_fill_args fillargs;
1012	int idx;
1013	int s_idx;
1014};
1015
1016/* Runs in RCU-critical section. */
1017static int rtnl_net_dumpid_one(int id, void *peer, void *data)
1018{
1019	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
1020	int ret;
1021
1022	if (net_cb->idx < net_cb->s_idx)
1023		goto cont;
1024
1025	net_cb->fillargs.nsid = id;
1026	if (net_cb->fillargs.add_ref)
1027		net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
1028	ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1029	if (ret < 0)
1030		return ret;
1031
1032cont:
1033	net_cb->idx++;
1034	return 0;
1035}
1036
1037static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1038				   struct rtnl_net_dump_cb *net_cb,
1039				   struct netlink_callback *cb)
1040{
1041	struct netlink_ext_ack *extack = cb->extack;
1042	struct nlattr *tb[NETNSA_MAX + 1];
1043	int err, i;
1044
1045	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1046					    NETNSA_MAX, rtnl_net_policy,
1047					    extack);
1048	if (err < 0)
1049		return err;
1050
1051	for (i = 0; i <= NETNSA_MAX; i++) {
1052		if (!tb[i])
1053			continue;
1054
1055		if (i == NETNSA_TARGET_NSID) {
1056			struct net *net;
1057
1058			net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1059			if (IS_ERR(net)) {
1060				NL_SET_BAD_ATTR(extack, tb[i]);
1061				NL_SET_ERR_MSG(extack,
1062					       "Invalid target network namespace id");
1063				return PTR_ERR(net);
1064			}
1065			net_cb->fillargs.add_ref = true;
1066			net_cb->ref_net = net_cb->tgt_net;
1067			net_cb->tgt_net = net;
1068		} else {
1069			NL_SET_BAD_ATTR(extack, tb[i]);
1070			NL_SET_ERR_MSG(extack,
1071				       "Unsupported attribute in dump request");
1072			return -EINVAL;
1073		}
1074	}
1075
1076	return 0;
1077}
1078
1079static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1080{
 
1081	struct rtnl_net_dump_cb net_cb = {
1082		.tgt_net = sock_net(skb->sk),
1083		.skb = skb,
1084		.fillargs = {
1085			.portid = NETLINK_CB(cb->skb).portid,
1086			.seq = cb->nlh->nlmsg_seq,
1087			.flags = NLM_F_MULTI,
1088			.cmd = RTM_NEWNSID,
1089		},
1090		.idx = 0,
1091		.s_idx = cb->args[0],
1092	};
1093	int err = 0;
1094
1095	if (cb->strict_check) {
1096		err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1097		if (err < 0)
1098			goto end;
1099	}
1100
1101	rcu_read_lock();
1102	idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1103	rcu_read_unlock();
1104
1105	cb->args[0] = net_cb.idx;
1106end:
1107	if (net_cb.fillargs.add_ref)
1108		put_net(net_cb.tgt_net);
1109	return err;
1110}
1111
1112static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1113			      struct nlmsghdr *nlh, gfp_t gfp)
1114{
1115	struct net_fill_args fillargs = {
1116		.portid = portid,
1117		.seq = nlh ? nlh->nlmsg_seq : 0,
1118		.cmd = cmd,
1119		.nsid = id,
1120	};
1121	struct sk_buff *msg;
1122	int err = -ENOMEM;
1123
1124	msg = nlmsg_new(rtnl_net_get_size(), gfp);
1125	if (!msg)
1126		goto out;
1127
1128	err = rtnl_net_fill(msg, &fillargs);
1129	if (err < 0)
1130		goto err_out;
1131
1132	rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1133	return;
1134
1135err_out:
1136	nlmsg_free(msg);
1137out:
1138	rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1139}
1140
1141#ifdef CONFIG_NET_NS
1142static void __init netns_ipv4_struct_check(void)
1143{
1144	/* TX readonly hotpath cache lines */
1145	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1146				      sysctl_tcp_early_retrans);
1147	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1148				      sysctl_tcp_tso_win_divisor);
1149	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1150				      sysctl_tcp_tso_rtt_log);
1151	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1152				      sysctl_tcp_autocorking);
1153	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1154				      sysctl_tcp_min_snd_mss);
1155	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1156				      sysctl_tcp_notsent_lowat);
1157	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1158				      sysctl_tcp_limit_output_bytes);
1159	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1160				      sysctl_tcp_min_rtt_wlen);
1161	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1162				      sysctl_tcp_wmem);
1163	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1164				      sysctl_ip_fwd_use_pmtu);
1165	CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33);
1166
1167	/* TXRX readonly hotpath cache lines */
1168	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx,
1169				      sysctl_tcp_moderate_rcvbuf);
1170	CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1);
1171
1172	/* RX readonly hotpath cache line */
1173	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1174				      sysctl_ip_early_demux);
1175	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1176				      sysctl_tcp_early_demux);
1177	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1178				      sysctl_tcp_l3mdev_accept);
1179	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1180				      sysctl_tcp_reordering);
1181	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1182				      sysctl_tcp_rmem);
1183	CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 22);
1184}
1185#endif
1186
1187static const struct rtnl_msg_handler net_ns_rtnl_msg_handlers[] __initconst = {
1188	{.msgtype = RTM_NEWNSID, .doit = rtnl_net_newid,
1189	 .flags = RTNL_FLAG_DOIT_UNLOCKED},
1190	{.msgtype = RTM_GETNSID, .doit = rtnl_net_getid,
1191	 .dumpit = rtnl_net_dumpid,
1192	 .flags = RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED},
1193};
1194
1195void __init net_ns_init(void)
1196{
1197	struct net_generic *ng;
1198
1199#ifdef CONFIG_NET_NS
1200	netns_ipv4_struct_check();
1201	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1202					SMP_CACHE_BYTES,
1203					SLAB_PANIC|SLAB_ACCOUNT, NULL);
1204
1205	/* Create workqueue for cleanup */
1206	netns_wq = create_singlethread_workqueue("netns");
1207	if (!netns_wq)
1208		panic("Could not create netns workq");
1209#endif
1210
1211	ng = net_alloc_generic();
1212	if (!ng)
1213		panic("Could not allocate generic netns");
1214
1215	rcu_assign_pointer(init_net.gen, ng);
1216
1217#ifdef CONFIG_KEYS
1218	init_net.key_domain = &init_net_key_domain;
1219#endif
1220	preinit_net(&init_net, &init_user_ns);
1221
1222	down_write(&pernet_ops_rwsem);
1223	if (setup_net(&init_net))
1224		panic("Could not setup the initial network namespace");
1225
1226	init_net_initialized = true;
1227	up_write(&pernet_ops_rwsem);
 
1228
1229	if (register_pernet_subsys(&net_ns_ops))
1230		panic("Could not register network namespace subsystems");
1231
1232	rtnl_register_many(net_ns_rtnl_msg_handlers);
1233}
1234
1235static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1236{
1237	ops_pre_exit_list(ops, net_exit_list);
1238	synchronize_rcu();
1239
1240	if (ops->exit_batch_rtnl) {
1241		LIST_HEAD(dev_kill_list);
1242
1243		rtnl_lock();
1244		ops->exit_batch_rtnl(net_exit_list, &dev_kill_list);
1245		unregister_netdevice_many(&dev_kill_list);
1246		rtnl_unlock();
1247	}
1248	ops_exit_list(ops, net_exit_list);
1249
1250	ops_free_list(ops, net_exit_list);
1251}
1252
1253#ifdef CONFIG_NET_NS
1254static int __register_pernet_operations(struct list_head *list,
1255					struct pernet_operations *ops)
1256{
1257	struct net *net;
1258	int error;
1259	LIST_HEAD(net_exit_list);
1260
1261	list_add_tail(&ops->list, list);
1262	if (ops->init || ops->id) {
1263		/* We held write locked pernet_ops_rwsem, and parallel
1264		 * setup_net() and cleanup_net() are not possible.
1265		 */
1266		for_each_net(net) {
1267			error = ops_init(ops, net);
1268			if (error)
1269				goto out_undo;
1270			list_add_tail(&net->exit_list, &net_exit_list);
1271		}
1272	}
1273	return 0;
1274
1275out_undo:
1276	/* If I have an error cleanup all namespaces I initialized */
1277	list_del(&ops->list);
1278	free_exit_list(ops, &net_exit_list);
 
1279	return error;
1280}
1281
1282static void __unregister_pernet_operations(struct pernet_operations *ops)
1283{
1284	struct net *net;
1285	LIST_HEAD(net_exit_list);
1286
1287	list_del(&ops->list);
1288	/* See comment in __register_pernet_operations() */
1289	for_each_net(net)
1290		list_add_tail(&net->exit_list, &net_exit_list);
1291
1292	free_exit_list(ops, &net_exit_list);
1293}
1294
1295#else
1296
1297static int __register_pernet_operations(struct list_head *list,
1298					struct pernet_operations *ops)
1299{
1300	if (!init_net_initialized) {
1301		list_add_tail(&ops->list, list);
1302		return 0;
1303	}
1304
1305	return ops_init(ops, &init_net);
1306}
1307
1308static void __unregister_pernet_operations(struct pernet_operations *ops)
1309{
1310	if (!init_net_initialized) {
1311		list_del(&ops->list);
1312	} else {
1313		LIST_HEAD(net_exit_list);
1314		list_add(&init_net.exit_list, &net_exit_list);
1315		free_exit_list(ops, &net_exit_list);
1316	}
1317}
1318
1319#endif /* CONFIG_NET_NS */
1320
1321static DEFINE_IDA(net_generic_ids);
1322
1323static int register_pernet_operations(struct list_head *list,
1324				      struct pernet_operations *ops)
1325{
1326	int error;
1327
1328	if (WARN_ON(!!ops->id ^ !!ops->size))
1329		return -EINVAL;
1330
1331	if (ops->id) {
1332		error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1333				GFP_KERNEL);
1334		if (error < 0)
 
 
 
 
1335			return error;
1336		*ops->id = error;
1337		/* This does not require READ_ONCE as writers already hold
1338		 * pernet_ops_rwsem. But WRITE_ONCE is needed to protect
1339		 * net_alloc_generic.
1340		 */
1341		WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1));
1342	}
1343	error = __register_pernet_operations(list, ops);
1344	if (error) {
1345		rcu_barrier();
1346		if (ops->id)
1347			ida_free(&net_generic_ids, *ops->id);
1348	}
1349
1350	return error;
1351}
1352
1353static void unregister_pernet_operations(struct pernet_operations *ops)
1354{
 
1355	__unregister_pernet_operations(ops);
1356	rcu_barrier();
1357	if (ops->id)
1358		ida_free(&net_generic_ids, *ops->id);
1359}
1360
1361/**
1362 *      register_pernet_subsys - register a network namespace subsystem
1363 *	@ops:  pernet operations structure for the subsystem
1364 *
1365 *	Register a subsystem which has init and exit functions
1366 *	that are called when network namespaces are created and
1367 *	destroyed respectively.
1368 *
1369 *	When registered all network namespace init functions are
1370 *	called for every existing network namespace.  Allowing kernel
1371 *	modules to have a race free view of the set of network namespaces.
1372 *
1373 *	When a new network namespace is created all of the init
1374 *	methods are called in the order in which they were registered.
1375 *
1376 *	When a network namespace is destroyed all of the exit methods
1377 *	are called in the reverse of the order with which they were
1378 *	registered.
1379 */
1380int register_pernet_subsys(struct pernet_operations *ops)
1381{
1382	int error;
1383	down_write(&pernet_ops_rwsem);
1384	error =  register_pernet_operations(first_device, ops);
1385	up_write(&pernet_ops_rwsem);
1386	return error;
1387}
1388EXPORT_SYMBOL_GPL(register_pernet_subsys);
1389
1390/**
1391 *      unregister_pernet_subsys - unregister a network namespace subsystem
1392 *	@ops: pernet operations structure to manipulate
1393 *
1394 *	Remove the pernet operations structure from the list to be
1395 *	used when network namespaces are created or destroyed.  In
1396 *	addition run the exit method for all existing network
1397 *	namespaces.
1398 */
1399void unregister_pernet_subsys(struct pernet_operations *ops)
1400{
1401	down_write(&pernet_ops_rwsem);
1402	unregister_pernet_operations(ops);
1403	up_write(&pernet_ops_rwsem);
1404}
1405EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1406
1407/**
1408 *      register_pernet_device - register a network namespace device
1409 *	@ops:  pernet operations structure for the subsystem
1410 *
1411 *	Register a device which has init and exit functions
1412 *	that are called when network namespaces are created and
1413 *	destroyed respectively.
1414 *
1415 *	When registered all network namespace init functions are
1416 *	called for every existing network namespace.  Allowing kernel
1417 *	modules to have a race free view of the set of network namespaces.
1418 *
1419 *	When a new network namespace is created all of the init
1420 *	methods are called in the order in which they were registered.
1421 *
1422 *	When a network namespace is destroyed all of the exit methods
1423 *	are called in the reverse of the order with which they were
1424 *	registered.
1425 */
1426int register_pernet_device(struct pernet_operations *ops)
1427{
1428	int error;
1429	down_write(&pernet_ops_rwsem);
1430	error = register_pernet_operations(&pernet_list, ops);
1431	if (!error && (first_device == &pernet_list))
1432		first_device = &ops->list;
1433	up_write(&pernet_ops_rwsem);
1434	return error;
1435}
1436EXPORT_SYMBOL_GPL(register_pernet_device);
1437
1438/**
1439 *      unregister_pernet_device - unregister a network namespace netdevice
1440 *	@ops: pernet operations structure to manipulate
1441 *
1442 *	Remove the pernet operations structure from the list to be
1443 *	used when network namespaces are created or destroyed.  In
1444 *	addition run the exit method for all existing network
1445 *	namespaces.
1446 */
1447void unregister_pernet_device(struct pernet_operations *ops)
1448{
1449	down_write(&pernet_ops_rwsem);
1450	if (&ops->list == first_device)
1451		first_device = first_device->next;
1452	unregister_pernet_operations(ops);
1453	up_write(&pernet_ops_rwsem);
1454}
1455EXPORT_SYMBOL_GPL(unregister_pernet_device);
1456
1457#ifdef CONFIG_NET_NS
1458static struct ns_common *netns_get(struct task_struct *task)
1459{
1460	struct net *net = NULL;
1461	struct nsproxy *nsproxy;
1462
1463	task_lock(task);
1464	nsproxy = task->nsproxy;
1465	if (nsproxy)
1466		net = get_net(nsproxy->net_ns);
1467	task_unlock(task);
1468
1469	return net ? &net->ns : NULL;
1470}
1471
1472static inline struct net *to_net_ns(struct ns_common *ns)
1473{
1474	return container_of(ns, struct net, ns);
1475}
1476
1477static void netns_put(struct ns_common *ns)
1478{
1479	put_net(to_net_ns(ns));
1480}
1481
1482static int netns_install(struct nsset *nsset, struct ns_common *ns)
1483{
1484	struct nsproxy *nsproxy = nsset->nsproxy;
1485	struct net *net = to_net_ns(ns);
1486
1487	if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1488	    !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1489		return -EPERM;
1490
1491	put_net(nsproxy->net_ns);
1492	nsproxy->net_ns = get_net(net);
1493	return 0;
1494}
1495
1496static struct user_namespace *netns_owner(struct ns_common *ns)
1497{
1498	return to_net_ns(ns)->user_ns;
1499}
1500
1501const struct proc_ns_operations netns_operations = {
1502	.name		= "net",
1503	.type		= CLONE_NEWNET,
1504	.get		= netns_get,
1505	.put		= netns_put,
1506	.install	= netns_install,
1507	.owner		= netns_owner,
1508};
1509#endif