Linux Audio

Check our new training course

Linux kernel drivers training

Mar 31-Apr 9, 2025, special US time zones
Register
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
v3.5.6
  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/proc_fs.h>
 
 14#include <linux/file.h>
 15#include <linux/export.h>
 
 
 
 
 16#include <net/net_namespace.h>
 17#include <net/netns/generic.h>
 18
 19/*
 20 *	Our network namespace constructor/destructor lists
 21 */
 22
 23static LIST_HEAD(pernet_list);
 24static struct list_head *first_device = &pernet_list;
 25static DEFINE_MUTEX(net_mutex);
 26
 27LIST_HEAD(net_namespace_list);
 28EXPORT_SYMBOL_GPL(net_namespace_list);
 29
 30struct net init_net = {
 31	.dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
 32};
 33EXPORT_SYMBOL(init_net);
 34
 35#define INITIAL_NET_GEN_PTRS	13 /* +1 for len +2 for rcu_head */
 36
 37static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
 38
 39static struct net_generic *net_alloc_generic(void)
 40{
 41	struct net_generic *ng;
 42	size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
 43
 44	ng = kzalloc(generic_size, GFP_KERNEL);
 45	if (ng)
 46		ng->len = max_gen_ptrs;
 47
 48	return ng;
 49}
 50
 51static int net_assign_generic(struct net *net, int id, void *data)
 52{
 53	struct net_generic *ng, *old_ng;
 54
 55	BUG_ON(!mutex_is_locked(&net_mutex));
 56	BUG_ON(id == 0);
 57
 58	old_ng = rcu_dereference_protected(net->gen,
 59					   lockdep_is_held(&net_mutex));
 60	ng = old_ng;
 61	if (old_ng->len >= id)
 62		goto assign;
 63
 64	ng = net_alloc_generic();
 65	if (ng == NULL)
 66		return -ENOMEM;
 67
 68	/*
 69	 * Some synchronisation notes:
 70	 *
 71	 * The net_generic explores the net->gen array inside rcu
 72	 * read section. Besides once set the net->gen->ptr[x]
 73	 * pointer never changes (see rules in netns/generic.h).
 74	 *
 75	 * That said, we simply duplicate this array and schedule
 76	 * the old copy for kfree after a grace period.
 77	 */
 78
 79	memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
 80
 81	rcu_assign_pointer(net->gen, ng);
 82	kfree_rcu(old_ng, rcu);
 83assign:
 84	ng->ptr[id - 1] = data;
 85	return 0;
 86}
 87
 88static int ops_init(const struct pernet_operations *ops, struct net *net)
 89{
 90	int err = -ENOMEM;
 91	void *data = NULL;
 92
 93	if (ops->id && ops->size) {
 94		data = kzalloc(ops->size, GFP_KERNEL);
 95		if (!data)
 96			goto out;
 97
 98		err = net_assign_generic(net, *ops->id, data);
 99		if (err)
100			goto cleanup;
101	}
102	err = 0;
103	if (ops->init)
104		err = ops->init(net);
105	if (!err)
106		return 0;
107
108cleanup:
109	kfree(data);
110
111out:
112	return err;
113}
114
115static void ops_free(const struct pernet_operations *ops, struct net *net)
116{
117	if (ops->id && ops->size) {
118		int id = *ops->id;
119		kfree(net_generic(net, id));
120	}
121}
122
123static void ops_exit_list(const struct pernet_operations *ops,
124			  struct list_head *net_exit_list)
125{
126	struct net *net;
127	if (ops->exit) {
128		list_for_each_entry(net, net_exit_list, exit_list)
129			ops->exit(net);
130	}
131	if (ops->exit_batch)
132		ops->exit_batch(net_exit_list);
133}
134
135static void ops_free_list(const struct pernet_operations *ops,
136			  struct list_head *net_exit_list)
137{
138	struct net *net;
139	if (ops->size && ops->id) {
140		list_for_each_entry(net, net_exit_list, exit_list)
141			ops_free(ops, net);
142	}
143}
144
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
145/*
146 * setup_net runs the initializers for the network namespace object.
147 */
148static __net_init int setup_net(struct net *net)
149{
150	/* Must be called with net_mutex held */
151	const struct pernet_operations *ops, *saved_ops;
152	int error = 0;
153	LIST_HEAD(net_exit_list);
154
155	atomic_set(&net->count, 1);
156	atomic_set(&net->passive, 1);
157	net->dev_base_seq = 1;
158
159#ifdef NETNS_REFCNT_DEBUG
160	atomic_set(&net->use_count, 0);
161#endif
162
163	list_for_each_entry(ops, &pernet_list, list) {
164		error = ops_init(ops, net);
165		if (error < 0)
166			goto out_undo;
167	}
168out:
169	return error;
170
171out_undo:
172	/* Walk through the list backwards calling the exit functions
173	 * for the pernet modules whose init functions did not fail.
174	 */
175	list_add(&net->exit_list, &net_exit_list);
176	saved_ops = ops;
177	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
178		ops_exit_list(ops, &net_exit_list);
179
180	ops = saved_ops;
181	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
182		ops_free_list(ops, &net_exit_list);
183
184	rcu_barrier();
185	goto out;
186}
187
188
189#ifdef CONFIG_NET_NS
190static struct kmem_cache *net_cachep;
191static struct workqueue_struct *netns_wq;
192
193static struct net *net_alloc(void)
194{
195	struct net *net = NULL;
196	struct net_generic *ng;
197
198	ng = net_alloc_generic();
199	if (!ng)
200		goto out;
201
202	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
203	if (!net)
204		goto out_free;
205
206	rcu_assign_pointer(net->gen, ng);
207out:
208	return net;
209
210out_free:
211	kfree(ng);
212	goto out;
213}
214
215static void net_free(struct net *net)
216{
217#ifdef NETNS_REFCNT_DEBUG
218	if (unlikely(atomic_read(&net->use_count) != 0)) {
219		pr_emerg("network namespace not free! Usage: %d\n",
220			 atomic_read(&net->use_count));
221		return;
222	}
223#endif
224	kfree(net->gen);
225	kmem_cache_free(net_cachep, net);
226}
227
228void net_drop_ns(void *p)
229{
230	struct net *ns = p;
231	if (ns && atomic_dec_and_test(&ns->passive))
232		net_free(ns);
233}
234
235struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 
236{
237	struct net *net;
238	int rv;
239
240	if (!(flags & CLONE_NEWNET))
241		return get_net(old_net);
242
243	net = net_alloc();
244	if (!net)
245		return ERR_PTR(-ENOMEM);
 
 
 
246	mutex_lock(&net_mutex);
247	rv = setup_net(net);
248	if (rv == 0) {
249		rtnl_lock();
250		list_add_tail_rcu(&net->list, &net_namespace_list);
251		rtnl_unlock();
252	}
253	mutex_unlock(&net_mutex);
254	if (rv < 0) {
 
255		net_drop_ns(net);
256		return ERR_PTR(rv);
257	}
258	return net;
259}
260
261static DEFINE_SPINLOCK(cleanup_list_lock);
262static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
263
264static void cleanup_net(struct work_struct *work)
265{
266	const struct pernet_operations *ops;
267	struct net *net, *tmp;
268	LIST_HEAD(net_kill_list);
269	LIST_HEAD(net_exit_list);
270
271	/* Atomically snapshot the list of namespaces to cleanup */
272	spin_lock_irq(&cleanup_list_lock);
273	list_replace_init(&cleanup_list, &net_kill_list);
274	spin_unlock_irq(&cleanup_list_lock);
275
276	mutex_lock(&net_mutex);
277
278	/* Don't let anyone else find us. */
279	rtnl_lock();
280	list_for_each_entry(net, &net_kill_list, cleanup_list) {
281		list_del_rcu(&net->list);
282		list_add_tail(&net->exit_list, &net_exit_list);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
283	}
284	rtnl_unlock();
285
286	/*
287	 * Another CPU might be rcu-iterating the list, wait for it.
288	 * This needs to be before calling the exit() notifiers, so
289	 * the rcu_barrier() below isn't sufficient alone.
290	 */
291	synchronize_rcu();
292
293	/* Run all of the network namespace exit methods */
294	list_for_each_entry_reverse(ops, &pernet_list, list)
295		ops_exit_list(ops, &net_exit_list);
296
297	/* Free the net generic variables */
298	list_for_each_entry_reverse(ops, &pernet_list, list)
299		ops_free_list(ops, &net_exit_list);
300
301	mutex_unlock(&net_mutex);
302
303	/* Ensure there are no outstanding rcu callbacks using this
304	 * network namespace.
305	 */
306	rcu_barrier();
307
308	/* Finally it is safe to free my network namespace structure */
309	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
310		list_del_init(&net->exit_list);
 
311		net_drop_ns(net);
312	}
313}
314static DECLARE_WORK(net_cleanup_work, cleanup_net);
315
316void __put_net(struct net *net)
317{
318	/* Cleanup the network namespace in process context */
319	unsigned long flags;
320
321	spin_lock_irqsave(&cleanup_list_lock, flags);
322	list_add(&net->cleanup_list, &cleanup_list);
323	spin_unlock_irqrestore(&cleanup_list_lock, flags);
324
325	queue_work(netns_wq, &net_cleanup_work);
326}
327EXPORT_SYMBOL_GPL(__put_net);
328
329struct net *get_net_ns_by_fd(int fd)
330{
331	struct proc_inode *ei;
332	struct file *file;
 
333	struct net *net;
334
335	file = proc_ns_fget(fd);
336	if (IS_ERR(file))
337		return ERR_CAST(file);
338
339	ei = PROC_I(file->f_dentry->d_inode);
340	if (ei->ns_ops == &netns_operations)
341		net = get_net(ei->ns);
342	else
343		net = ERR_PTR(-EINVAL);
344
345	fput(file);
346	return net;
347}
348
349#else
350struct net *copy_net_ns(unsigned long flags, struct net *old_net)
351{
352	if (flags & CLONE_NEWNET)
353		return ERR_PTR(-EINVAL);
354	return old_net;
355}
356
357struct net *get_net_ns_by_fd(int fd)
358{
359	return ERR_PTR(-EINVAL);
360}
361#endif
 
362
363struct net *get_net_ns_by_pid(pid_t pid)
364{
365	struct task_struct *tsk;
366	struct net *net;
367
368	/* Lookup the network namespace */
369	net = ERR_PTR(-ESRCH);
370	rcu_read_lock();
371	tsk = find_task_by_vpid(pid);
372	if (tsk) {
373		struct nsproxy *nsproxy;
374		nsproxy = task_nsproxy(tsk);
 
375		if (nsproxy)
376			net = get_net(nsproxy->net_ns);
 
377	}
378	rcu_read_unlock();
379	return net;
380}
381EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
382
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
383static int __init net_ns_init(void)
384{
385	struct net_generic *ng;
386
387#ifdef CONFIG_NET_NS
388	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
389					SMP_CACHE_BYTES,
390					SLAB_PANIC, NULL);
391
392	/* Create workqueue for cleanup */
393	netns_wq = create_singlethread_workqueue("netns");
394	if (!netns_wq)
395		panic("Could not create netns workq");
396#endif
397
398	ng = net_alloc_generic();
399	if (!ng)
400		panic("Could not allocate generic netns");
401
402	rcu_assign_pointer(init_net.gen, ng);
403
404	mutex_lock(&net_mutex);
405	if (setup_net(&init_net))
406		panic("Could not setup the initial network namespace");
407
408	rtnl_lock();
409	list_add_tail_rcu(&init_net.list, &net_namespace_list);
410	rtnl_unlock();
411
412	mutex_unlock(&net_mutex);
413
 
 
 
 
 
 
414	return 0;
415}
416
417pure_initcall(net_ns_init);
418
419#ifdef CONFIG_NET_NS
420static int __register_pernet_operations(struct list_head *list,
421					struct pernet_operations *ops)
422{
423	struct net *net;
424	int error;
425	LIST_HEAD(net_exit_list);
426
427	list_add_tail(&ops->list, list);
428	if (ops->init || (ops->id && ops->size)) {
429		for_each_net(net) {
430			error = ops_init(ops, net);
431			if (error)
432				goto out_undo;
433			list_add_tail(&net->exit_list, &net_exit_list);
434		}
435	}
436	return 0;
437
438out_undo:
439	/* If I have an error cleanup all namespaces I initialized */
440	list_del(&ops->list);
441	ops_exit_list(ops, &net_exit_list);
442	ops_free_list(ops, &net_exit_list);
443	return error;
444}
445
446static void __unregister_pernet_operations(struct pernet_operations *ops)
447{
448	struct net *net;
449	LIST_HEAD(net_exit_list);
450
451	list_del(&ops->list);
452	for_each_net(net)
453		list_add_tail(&net->exit_list, &net_exit_list);
454	ops_exit_list(ops, &net_exit_list);
455	ops_free_list(ops, &net_exit_list);
456}
457
458#else
459
460static int __register_pernet_operations(struct list_head *list,
461					struct pernet_operations *ops)
462{
463	return ops_init(ops, &init_net);
464}
465
466static void __unregister_pernet_operations(struct pernet_operations *ops)
467{
468	LIST_HEAD(net_exit_list);
469	list_add(&init_net.exit_list, &net_exit_list);
470	ops_exit_list(ops, &net_exit_list);
471	ops_free_list(ops, &net_exit_list);
472}
473
474#endif /* CONFIG_NET_NS */
475
476static DEFINE_IDA(net_generic_ids);
477
478static int register_pernet_operations(struct list_head *list,
479				      struct pernet_operations *ops)
480{
481	int error;
482
483	if (ops->id) {
484again:
485		error = ida_get_new_above(&net_generic_ids, 1, ops->id);
486		if (error < 0) {
487			if (error == -EAGAIN) {
488				ida_pre_get(&net_generic_ids, GFP_KERNEL);
489				goto again;
490			}
491			return error;
492		}
493		max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
494	}
495	error = __register_pernet_operations(list, ops);
496	if (error) {
497		rcu_barrier();
498		if (ops->id)
499			ida_remove(&net_generic_ids, *ops->id);
500	}
501
502	return error;
503}
504
505static void unregister_pernet_operations(struct pernet_operations *ops)
506{
507	
508	__unregister_pernet_operations(ops);
509	rcu_barrier();
510	if (ops->id)
511		ida_remove(&net_generic_ids, *ops->id);
512}
513
514/**
515 *      register_pernet_subsys - register a network namespace subsystem
516 *	@ops:  pernet operations structure for the subsystem
517 *
518 *	Register a subsystem which has init and exit functions
519 *	that are called when network namespaces are created and
520 *	destroyed respectively.
521 *
522 *	When registered all network namespace init functions are
523 *	called for every existing network namespace.  Allowing kernel
524 *	modules to have a race free view of the set of network namespaces.
525 *
526 *	When a new network namespace is created all of the init
527 *	methods are called in the order in which they were registered.
528 *
529 *	When a network namespace is destroyed all of the exit methods
530 *	are called in the reverse of the order with which they were
531 *	registered.
532 */
533int register_pernet_subsys(struct pernet_operations *ops)
534{
535	int error;
536	mutex_lock(&net_mutex);
537	error =  register_pernet_operations(first_device, ops);
538	mutex_unlock(&net_mutex);
539	return error;
540}
541EXPORT_SYMBOL_GPL(register_pernet_subsys);
542
543/**
544 *      unregister_pernet_subsys - unregister a network namespace subsystem
545 *	@ops: pernet operations structure to manipulate
546 *
547 *	Remove the pernet operations structure from the list to be
548 *	used when network namespaces are created or destroyed.  In
549 *	addition run the exit method for all existing network
550 *	namespaces.
551 */
552void unregister_pernet_subsys(struct pernet_operations *ops)
553{
554	mutex_lock(&net_mutex);
555	unregister_pernet_operations(ops);
556	mutex_unlock(&net_mutex);
557}
558EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
559
560/**
561 *      register_pernet_device - register a network namespace device
562 *	@ops:  pernet operations structure for the subsystem
563 *
564 *	Register a device which has init and exit functions
565 *	that are called when network namespaces are created and
566 *	destroyed respectively.
567 *
568 *	When registered all network namespace init functions are
569 *	called for every existing network namespace.  Allowing kernel
570 *	modules to have a race free view of the set of network namespaces.
571 *
572 *	When a new network namespace is created all of the init
573 *	methods are called in the order in which they were registered.
574 *
575 *	When a network namespace is destroyed all of the exit methods
576 *	are called in the reverse of the order with which they were
577 *	registered.
578 */
579int register_pernet_device(struct pernet_operations *ops)
580{
581	int error;
582	mutex_lock(&net_mutex);
583	error = register_pernet_operations(&pernet_list, ops);
584	if (!error && (first_device == &pernet_list))
585		first_device = &ops->list;
586	mutex_unlock(&net_mutex);
587	return error;
588}
589EXPORT_SYMBOL_GPL(register_pernet_device);
590
591/**
592 *      unregister_pernet_device - unregister a network namespace netdevice
593 *	@ops: pernet operations structure to manipulate
594 *
595 *	Remove the pernet operations structure from the list to be
596 *	used when network namespaces are created or destroyed.  In
597 *	addition run the exit method for all existing network
598 *	namespaces.
599 */
600void unregister_pernet_device(struct pernet_operations *ops)
601{
602	mutex_lock(&net_mutex);
603	if (&ops->list == first_device)
604		first_device = first_device->next;
605	unregister_pernet_operations(ops);
606	mutex_unlock(&net_mutex);
607}
608EXPORT_SYMBOL_GPL(unregister_pernet_device);
609
610#ifdef CONFIG_NET_NS
611static void *netns_get(struct task_struct *task)
612{
613	struct net *net = NULL;
614	struct nsproxy *nsproxy;
615
616	rcu_read_lock();
617	nsproxy = task_nsproxy(task);
618	if (nsproxy)
619		net = get_net(nsproxy->net_ns);
620	rcu_read_unlock();
 
 
 
621
622	return net;
 
 
623}
624
625static void netns_put(void *ns)
626{
627	put_net(ns);
628}
629
630static int netns_install(struct nsproxy *nsproxy, void *ns)
631{
 
 
 
 
 
 
632	put_net(nsproxy->net_ns);
633	nsproxy->net_ns = get_net(ns);
634	return 0;
635}
636
637const struct proc_ns_operations netns_operations = {
638	.name		= "net",
639	.type		= CLONE_NEWNET,
640	.get		= netns_get,
641	.put		= netns_put,
642	.install	= netns_install,
643};
644#endif