1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
14 #include <linux/proc_ns.h>
15 #include <linux/file.h>
16 #include <linux/export.h>
17 #include <linux/user_namespace.h>
18 #include <linux/net_namespace.h>
19 #include <linux/rtnetlink.h>
21 #include <net/netlink.h>
22 #include <net/net_namespace.h>
23 #include <net/netns/generic.h>
26 * Our network namespace constructor/destructor lists
29 static LIST_HEAD(pernet_list
);
30 static struct list_head
*first_device
= &pernet_list
;
31 DEFINE_MUTEX(net_mutex
);
33 LIST_HEAD(net_namespace_list
);
34 EXPORT_SYMBOL_GPL(net_namespace_list
);
36 struct net init_net
= {
37 .dev_base_head
= LIST_HEAD_INIT(init_net
.dev_base_head
),
39 EXPORT_SYMBOL(init_net
);
41 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
43 static unsigned int max_gen_ptrs
= INITIAL_NET_GEN_PTRS
;
45 static struct net_generic
*net_alloc_generic(void)
47 struct net_generic
*ng
;
48 size_t generic_size
= offsetof(struct net_generic
, ptr
[max_gen_ptrs
]);
50 ng
= kzalloc(generic_size
, GFP_KERNEL
);
52 ng
->len
= max_gen_ptrs
;
57 static int net_assign_generic(struct net
*net
, int id
, void *data
)
59 struct net_generic
*ng
, *old_ng
;
61 BUG_ON(!mutex_is_locked(&net_mutex
));
64 old_ng
= rcu_dereference_protected(net
->gen
,
65 lockdep_is_held(&net_mutex
));
67 if (old_ng
->len
>= id
)
70 ng
= net_alloc_generic();
75 * Some synchronisation notes:
77 * The net_generic explores the net->gen array inside rcu
78 * read section. Besides once set the net->gen->ptr[x]
79 * pointer never changes (see rules in netns/generic.h).
81 * That said, we simply duplicate this array and schedule
82 * the old copy for kfree after a grace period.
85 memcpy(&ng
->ptr
, &old_ng
->ptr
, old_ng
->len
* sizeof(void*));
87 rcu_assign_pointer(net
->gen
, ng
);
88 kfree_rcu(old_ng
, rcu
);
90 ng
->ptr
[id
- 1] = data
;
94 static int ops_init(const struct pernet_operations
*ops
, struct net
*net
)
99 if (ops
->id
&& ops
->size
) {
100 data
= kzalloc(ops
->size
, GFP_KERNEL
);
104 err
= net_assign_generic(net
, *ops
->id
, data
);
110 err
= ops
->init(net
);
121 static void ops_free(const struct pernet_operations
*ops
, struct net
*net
)
123 if (ops
->id
&& ops
->size
) {
125 kfree(net_generic(net
, id
));
129 static void ops_exit_list(const struct pernet_operations
*ops
,
130 struct list_head
*net_exit_list
)
134 list_for_each_entry(net
, net_exit_list
, exit_list
)
138 ops
->exit_batch(net_exit_list
);
141 static void ops_free_list(const struct pernet_operations
*ops
,
142 struct list_head
*net_exit_list
)
145 if (ops
->size
&& ops
->id
) {
146 list_for_each_entry(net
, net_exit_list
, exit_list
)
151 static int alloc_netid(struct net
*net
, struct net
*peer
, int reqid
)
153 int min
= 0, max
= 0;
162 return idr_alloc(&net
->netns_ids
, peer
, min
, max
, GFP_KERNEL
);
165 /* This function is used by idr_for_each(). If net is equal to peer, the
166 * function returns the id so that idr_for_each() stops. Because we cannot
167 * returns the id 0 (idr_for_each() will not stop), we return the magic value
168 * NET_ID_ZERO (-1) for it.
170 #define NET_ID_ZERO -1
171 static int net_eq_idr(int id
, void *net
, void *peer
)
173 if (net_eq(net
, peer
))
174 return id
? : NET_ID_ZERO
;
178 static int __peernet2id(struct net
*net
, struct net
*peer
, bool alloc
)
180 int id
= idr_for_each(&net
->netns_ids
, net_eq_idr
, peer
);
184 /* Magic value for id 0. */
185 if (id
== NET_ID_ZERO
)
191 return alloc_netid(net
, peer
, -1);
196 /* This function returns the id of a peer netns. If no id is assigned, one will
197 * be allocated and returned.
199 int peernet2id(struct net
*net
, struct net
*peer
)
201 int id
= __peernet2id(net
, peer
, true);
203 return id
>= 0 ? id
: NETNSA_NSID_NOT_ASSIGNED
;
205 EXPORT_SYMBOL(peernet2id
);
207 struct net
*get_net_ns_by_id(struct net
*net
, int id
)
215 peer
= idr_find(&net
->netns_ids
, id
);
224 * setup_net runs the initializers for the network namespace object.
226 static __net_init
int setup_net(struct net
*net
, struct user_namespace
*user_ns
)
228 /* Must be called with net_mutex held */
229 const struct pernet_operations
*ops
, *saved_ops
;
231 LIST_HEAD(net_exit_list
);
233 atomic_set(&net
->count
, 1);
234 atomic_set(&net
->passive
, 1);
235 net
->dev_base_seq
= 1;
236 net
->user_ns
= user_ns
;
237 idr_init(&net
->netns_ids
);
239 #ifdef NETNS_REFCNT_DEBUG
240 atomic_set(&net
->use_count
, 0);
243 list_for_each_entry(ops
, &pernet_list
, list
) {
244 error
= ops_init(ops
, net
);
252 /* Walk through the list backwards calling the exit functions
253 * for the pernet modules whose init functions did not fail.
255 list_add(&net
->exit_list
, &net_exit_list
);
257 list_for_each_entry_continue_reverse(ops
, &pernet_list
, list
)
258 ops_exit_list(ops
, &net_exit_list
);
261 list_for_each_entry_continue_reverse(ops
, &pernet_list
, list
)
262 ops_free_list(ops
, &net_exit_list
);
270 static struct kmem_cache
*net_cachep
;
271 static struct workqueue_struct
*netns_wq
;
273 static struct net
*net_alloc(void)
275 struct net
*net
= NULL
;
276 struct net_generic
*ng
;
278 ng
= net_alloc_generic();
282 net
= kmem_cache_zalloc(net_cachep
, GFP_KERNEL
);
286 rcu_assign_pointer(net
->gen
, ng
);
295 static void net_free(struct net
*net
)
297 #ifdef NETNS_REFCNT_DEBUG
298 if (unlikely(atomic_read(&net
->use_count
) != 0)) {
299 pr_emerg("network namespace not free! Usage: %d\n",
300 atomic_read(&net
->use_count
));
304 kfree(rcu_access_pointer(net
->gen
));
305 kmem_cache_free(net_cachep
, net
);
308 void net_drop_ns(void *p
)
311 if (ns
&& atomic_dec_and_test(&ns
->passive
))
315 struct net
*copy_net_ns(unsigned long flags
,
316 struct user_namespace
*user_ns
, struct net
*old_net
)
321 if (!(flags
& CLONE_NEWNET
))
322 return get_net(old_net
);
326 return ERR_PTR(-ENOMEM
);
328 get_user_ns(user_ns
);
330 mutex_lock(&net_mutex
);
331 rv
= setup_net(net
, user_ns
);
334 list_add_tail_rcu(&net
->list
, &net_namespace_list
);
337 mutex_unlock(&net_mutex
);
339 put_user_ns(user_ns
);
346 static DEFINE_SPINLOCK(cleanup_list_lock
);
347 static LIST_HEAD(cleanup_list
); /* Must hold cleanup_list_lock to touch */
349 static void cleanup_net(struct work_struct
*work
)
351 const struct pernet_operations
*ops
;
352 struct net
*net
, *tmp
;
353 struct list_head net_kill_list
;
354 LIST_HEAD(net_exit_list
);
356 /* Atomically snapshot the list of namespaces to cleanup */
357 spin_lock_irq(&cleanup_list_lock
);
358 list_replace_init(&cleanup_list
, &net_kill_list
);
359 spin_unlock_irq(&cleanup_list_lock
);
361 mutex_lock(&net_mutex
);
363 /* Don't let anyone else find us. */
365 list_for_each_entry(net
, &net_kill_list
, cleanup_list
) {
366 list_del_rcu(&net
->list
);
367 list_add_tail(&net
->exit_list
, &net_exit_list
);
369 int id
= __peernet2id(tmp
, net
, false);
372 idr_remove(&tmp
->netns_ids
, id
);
374 idr_destroy(&net
->netns_ids
);
380 * Another CPU might be rcu-iterating the list, wait for it.
381 * This needs to be before calling the exit() notifiers, so
382 * the rcu_barrier() below isn't sufficient alone.
386 /* Run all of the network namespace exit methods */
387 list_for_each_entry_reverse(ops
, &pernet_list
, list
)
388 ops_exit_list(ops
, &net_exit_list
);
390 /* Free the net generic variables */
391 list_for_each_entry_reverse(ops
, &pernet_list
, list
)
392 ops_free_list(ops
, &net_exit_list
);
394 mutex_unlock(&net_mutex
);
396 /* Ensure there are no outstanding rcu callbacks using this
401 /* Finally it is safe to free my network namespace structure */
402 list_for_each_entry_safe(net
, tmp
, &net_exit_list
, exit_list
) {
403 list_del_init(&net
->exit_list
);
404 put_user_ns(net
->user_ns
);
408 static DECLARE_WORK(net_cleanup_work
, cleanup_net
);
410 void __put_net(struct net
*net
)
412 /* Cleanup the network namespace in process context */
415 spin_lock_irqsave(&cleanup_list_lock
, flags
);
416 list_add(&net
->cleanup_list
, &cleanup_list
);
417 spin_unlock_irqrestore(&cleanup_list_lock
, flags
);
419 queue_work(netns_wq
, &net_cleanup_work
);
421 EXPORT_SYMBOL_GPL(__put_net
);
423 struct net
*get_net_ns_by_fd(int fd
)
426 struct ns_common
*ns
;
429 file
= proc_ns_fget(fd
);
431 return ERR_CAST(file
);
433 ns
= get_proc_ns(file_inode(file
));
434 if (ns
->ops
== &netns_operations
)
435 net
= get_net(container_of(ns
, struct net
, ns
));
437 net
= ERR_PTR(-EINVAL
);
444 struct net
*get_net_ns_by_fd(int fd
)
446 return ERR_PTR(-EINVAL
);
449 EXPORT_SYMBOL_GPL(get_net_ns_by_fd
);
451 struct net
*get_net_ns_by_pid(pid_t pid
)
453 struct task_struct
*tsk
;
456 /* Lookup the network namespace */
457 net
= ERR_PTR(-ESRCH
);
459 tsk
= find_task_by_vpid(pid
);
461 struct nsproxy
*nsproxy
;
463 nsproxy
= tsk
->nsproxy
;
465 net
= get_net(nsproxy
->net_ns
);
471 EXPORT_SYMBOL_GPL(get_net_ns_by_pid
);
473 static __net_init
int net_ns_net_init(struct net
*net
)
476 net
->ns
.ops
= &netns_operations
;
478 return ns_alloc_inum(&net
->ns
);
481 static __net_exit
void net_ns_net_exit(struct net
*net
)
483 ns_free_inum(&net
->ns
);
486 static struct pernet_operations __net_initdata net_ns_ops
= {
487 .init
= net_ns_net_init
,
488 .exit
= net_ns_net_exit
,
491 static struct nla_policy rtnl_net_policy
[NETNSA_MAX
+ 1] = {
492 [NETNSA_NONE
] = { .type
= NLA_UNSPEC
},
493 [NETNSA_NSID
] = { .type
= NLA_S32
},
494 [NETNSA_PID
] = { .type
= NLA_U32
},
495 [NETNSA_FD
] = { .type
= NLA_U32
},
498 static int rtnl_net_newid(struct sk_buff
*skb
, struct nlmsghdr
*nlh
)
500 struct net
*net
= sock_net(skb
->sk
);
501 struct nlattr
*tb
[NETNSA_MAX
+ 1];
505 err
= nlmsg_parse(nlh
, sizeof(struct rtgenmsg
), tb
, NETNSA_MAX
,
509 if (!tb
[NETNSA_NSID
])
511 nsid
= nla_get_s32(tb
[NETNSA_NSID
]);
514 peer
= get_net_ns_by_pid(nla_get_u32(tb
[NETNSA_PID
]));
515 else if (tb
[NETNSA_FD
])
516 peer
= get_net_ns_by_fd(nla_get_u32(tb
[NETNSA_FD
]));
520 return PTR_ERR(peer
);
522 if (__peernet2id(net
, peer
, false) >= 0) {
527 err
= alloc_netid(net
, peer
, nsid
);
535 static int rtnl_net_get_size(void)
537 return NLMSG_ALIGN(sizeof(struct rtgenmsg
))
538 + nla_total_size(sizeof(s32
)) /* NETNSA_NSID */
542 static int rtnl_net_fill(struct sk_buff
*skb
, u32 portid
, u32 seq
, int flags
,
543 int cmd
, struct net
*net
, struct net
*peer
)
545 struct nlmsghdr
*nlh
;
546 struct rtgenmsg
*rth
;
551 nlh
= nlmsg_put(skb
, portid
, seq
, cmd
, sizeof(*rth
), flags
);
555 rth
= nlmsg_data(nlh
);
556 rth
->rtgen_family
= AF_UNSPEC
;
558 id
= __peernet2id(net
, peer
, false);
560 id
= NETNSA_NSID_NOT_ASSIGNED
;
561 if (nla_put_s32(skb
, NETNSA_NSID
, id
))
562 goto nla_put_failure
;
568 nlmsg_cancel(skb
, nlh
);
572 static int rtnl_net_getid(struct sk_buff
*skb
, struct nlmsghdr
*nlh
)
574 struct net
*net
= sock_net(skb
->sk
);
575 struct nlattr
*tb
[NETNSA_MAX
+ 1];
580 err
= nlmsg_parse(nlh
, sizeof(struct rtgenmsg
), tb
, NETNSA_MAX
,
585 peer
= get_net_ns_by_pid(nla_get_u32(tb
[NETNSA_PID
]));
586 else if (tb
[NETNSA_FD
])
587 peer
= get_net_ns_by_fd(nla_get_u32(tb
[NETNSA_FD
]));
592 return PTR_ERR(peer
);
594 msg
= nlmsg_new(rtnl_net_get_size(), GFP_KERNEL
);
600 err
= rtnl_net_fill(msg
, NETLINK_CB(skb
).portid
, nlh
->nlmsg_seq
, 0,
601 RTM_GETNSID
, net
, peer
);
605 err
= rtnl_unicast(msg
, net
, NETLINK_CB(skb
).portid
);
615 static int __init
net_ns_init(void)
617 struct net_generic
*ng
;
620 net_cachep
= kmem_cache_create("net_namespace", sizeof(struct net
),
624 /* Create workqueue for cleanup */
625 netns_wq
= create_singlethread_workqueue("netns");
627 panic("Could not create netns workq");
630 ng
= net_alloc_generic();
632 panic("Could not allocate generic netns");
634 rcu_assign_pointer(init_net
.gen
, ng
);
636 mutex_lock(&net_mutex
);
637 if (setup_net(&init_net
, &init_user_ns
))
638 panic("Could not setup the initial network namespace");
641 list_add_tail_rcu(&init_net
.list
, &net_namespace_list
);
644 mutex_unlock(&net_mutex
);
646 register_pernet_subsys(&net_ns_ops
);
648 rtnl_register(PF_UNSPEC
, RTM_NEWNSID
, rtnl_net_newid
, NULL
, NULL
);
649 rtnl_register(PF_UNSPEC
, RTM_GETNSID
, rtnl_net_getid
, NULL
, NULL
);
654 pure_initcall(net_ns_init
);
657 static int __register_pernet_operations(struct list_head
*list
,
658 struct pernet_operations
*ops
)
662 LIST_HEAD(net_exit_list
);
664 list_add_tail(&ops
->list
, list
);
665 if (ops
->init
|| (ops
->id
&& ops
->size
)) {
667 error
= ops_init(ops
, net
);
670 list_add_tail(&net
->exit_list
, &net_exit_list
);
676 /* If I have an error cleanup all namespaces I initialized */
677 list_del(&ops
->list
);
678 ops_exit_list(ops
, &net_exit_list
);
679 ops_free_list(ops
, &net_exit_list
);
683 static void __unregister_pernet_operations(struct pernet_operations
*ops
)
686 LIST_HEAD(net_exit_list
);
688 list_del(&ops
->list
);
690 list_add_tail(&net
->exit_list
, &net_exit_list
);
691 ops_exit_list(ops
, &net_exit_list
);
692 ops_free_list(ops
, &net_exit_list
);
697 static int __register_pernet_operations(struct list_head
*list
,
698 struct pernet_operations
*ops
)
700 return ops_init(ops
, &init_net
);
703 static void __unregister_pernet_operations(struct pernet_operations
*ops
)
705 LIST_HEAD(net_exit_list
);
706 list_add(&init_net
.exit_list
, &net_exit_list
);
707 ops_exit_list(ops
, &net_exit_list
);
708 ops_free_list(ops
, &net_exit_list
);
711 #endif /* CONFIG_NET_NS */
713 static DEFINE_IDA(net_generic_ids
);
715 static int register_pernet_operations(struct list_head
*list
,
716 struct pernet_operations
*ops
)
722 error
= ida_get_new_above(&net_generic_ids
, 1, ops
->id
);
724 if (error
== -EAGAIN
) {
725 ida_pre_get(&net_generic_ids
, GFP_KERNEL
);
730 max_gen_ptrs
= max_t(unsigned int, max_gen_ptrs
, *ops
->id
);
732 error
= __register_pernet_operations(list
, ops
);
736 ida_remove(&net_generic_ids
, *ops
->id
);
742 static void unregister_pernet_operations(struct pernet_operations
*ops
)
745 __unregister_pernet_operations(ops
);
748 ida_remove(&net_generic_ids
, *ops
->id
);
752 * register_pernet_subsys - register a network namespace subsystem
753 * @ops: pernet operations structure for the subsystem
755 * Register a subsystem which has init and exit functions
756 * that are called when network namespaces are created and
757 * destroyed respectively.
759 * When registered all network namespace init functions are
760 * called for every existing network namespace. Allowing kernel
761 * modules to have a race free view of the set of network namespaces.
763 * When a new network namespace is created all of the init
764 * methods are called in the order in which they were registered.
766 * When a network namespace is destroyed all of the exit methods
767 * are called in the reverse of the order with which they were
770 int register_pernet_subsys(struct pernet_operations
*ops
)
773 mutex_lock(&net_mutex
);
774 error
= register_pernet_operations(first_device
, ops
);
775 mutex_unlock(&net_mutex
);
778 EXPORT_SYMBOL_GPL(register_pernet_subsys
);
781 * unregister_pernet_subsys - unregister a network namespace subsystem
782 * @ops: pernet operations structure to manipulate
784 * Remove the pernet operations structure from the list to be
785 * used when network namespaces are created or destroyed. In
786 * addition run the exit method for all existing network
789 void unregister_pernet_subsys(struct pernet_operations
*ops
)
791 mutex_lock(&net_mutex
);
792 unregister_pernet_operations(ops
);
793 mutex_unlock(&net_mutex
);
795 EXPORT_SYMBOL_GPL(unregister_pernet_subsys
);
798 * register_pernet_device - register a network namespace device
799 * @ops: pernet operations structure for the subsystem
801 * Register a device which has init and exit functions
802 * that are called when network namespaces are created and
803 * destroyed respectively.
805 * When registered all network namespace init functions are
806 * called for every existing network namespace. Allowing kernel
807 * modules to have a race free view of the set of network namespaces.
809 * When a new network namespace is created all of the init
810 * methods are called in the order in which they were registered.
812 * When a network namespace is destroyed all of the exit methods
813 * are called in the reverse of the order with which they were
816 int register_pernet_device(struct pernet_operations
*ops
)
819 mutex_lock(&net_mutex
);
820 error
= register_pernet_operations(&pernet_list
, ops
);
821 if (!error
&& (first_device
== &pernet_list
))
822 first_device
= &ops
->list
;
823 mutex_unlock(&net_mutex
);
826 EXPORT_SYMBOL_GPL(register_pernet_device
);
829 * unregister_pernet_device - unregister a network namespace netdevice
830 * @ops: pernet operations structure to manipulate
832 * Remove the pernet operations structure from the list to be
833 * used when network namespaces are created or destroyed. In
834 * addition run the exit method for all existing network
837 void unregister_pernet_device(struct pernet_operations
*ops
)
839 mutex_lock(&net_mutex
);
840 if (&ops
->list
== first_device
)
841 first_device
= first_device
->next
;
842 unregister_pernet_operations(ops
);
843 mutex_unlock(&net_mutex
);
845 EXPORT_SYMBOL_GPL(unregister_pernet_device
);
848 static struct ns_common
*netns_get(struct task_struct
*task
)
850 struct net
*net
= NULL
;
851 struct nsproxy
*nsproxy
;
854 nsproxy
= task
->nsproxy
;
856 net
= get_net(nsproxy
->net_ns
);
859 return net
? &net
->ns
: NULL
;
862 static inline struct net
*to_net_ns(struct ns_common
*ns
)
864 return container_of(ns
, struct net
, ns
);
867 static void netns_put(struct ns_common
*ns
)
869 put_net(to_net_ns(ns
));
872 static int netns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
874 struct net
*net
= to_net_ns(ns
);
876 if (!ns_capable(net
->user_ns
, CAP_SYS_ADMIN
) ||
877 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
880 put_net(nsproxy
->net_ns
);
881 nsproxy
->net_ns
= get_net(net
);
885 const struct proc_ns_operations netns_operations
= {
887 .type
= CLONE_NEWNET
,
890 .install
= netns_install
,