1 #include <linux/workqueue.h>
2 #include <linux/rtnetlink.h>
3 #include <linux/cache.h>
4 #include <linux/slab.h>
5 #include <linux/list.h>
6 #include <linux/delay.h>
7 #include <linux/sched.h>
9 #include <net/net_namespace.h>
10 #include <net/netns/generic.h>
13 * Our network namespace constructor/destructor lists
16 static LIST_HEAD(pernet_list);
17 static struct list_head *first_device = &pernet_list;
18 static DEFINE_MUTEX(net_mutex);
20 LIST_HEAD(net_namespace_list);
21 EXPORT_SYMBOL_GPL(net_namespace_list);
24 EXPORT_SYMBOL(init_net);
26 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
29 * setup_net runs the initializers for the network namespace object.
31 static __net_init int setup_net(struct net *net)
33 /* Must be called with net_mutex held */
34 struct pernet_operations *ops;
36 struct net_generic *ng;
38 atomic_set(&net->count, 1);
39 #ifdef NETNS_REFCNT_DEBUG
40 atomic_set(&net->use_count, 0);
44 ng = kzalloc(sizeof(struct net_generic) +
45 INITIAL_NET_GEN_PTRS * sizeof(void *), GFP_KERNEL);
49 ng->len = INITIAL_NET_GEN_PTRS;
50 INIT_RCU_HEAD(&ng->rcu);
51 rcu_assign_pointer(net->gen, ng);
54 list_for_each_entry(ops, &pernet_list, list) {
56 error = ops->init(net);
65 /* Walk through the list backwards calling the exit functions
66 * for the pernet modules whose init functions did not fail.
68 list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
79 static struct kmem_cache *net_cachep;
80 static struct workqueue_struct *netns_wq;
82 static struct net *net_alloc(void)
84 return kmem_cache_zalloc(net_cachep, GFP_KERNEL);
87 static void net_free(struct net *net)
92 #ifdef NETNS_REFCNT_DEBUG
93 if (unlikely(atomic_read(&net->use_count) != 0)) {
94 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
95 atomic_read(&net->use_count));
100 kmem_cache_free(net_cachep, net);
103 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
105 struct net *new_net = NULL;
110 if (!(flags & CLONE_NEWNET))
114 new_net = net_alloc();
118 mutex_lock(&net_mutex);
119 err = setup_net(new_net);
124 list_add_tail(&new_net->list, &net_namespace_list);
129 mutex_unlock(&net_mutex);
134 new_net = ERR_PTR(err);
139 static void cleanup_net(struct work_struct *work)
141 struct pernet_operations *ops;
144 /* Be very certain incoming network packets will not find us */
147 net = container_of(work, struct net, work);
149 mutex_lock(&net_mutex);
151 /* Don't let anyone else find us. */
153 list_del(&net->list);
156 /* Run all of the network namespace exit methods */
157 list_for_each_entry_reverse(ops, &pernet_list, list) {
162 mutex_unlock(&net_mutex);
164 /* Ensure there are no outstanding rcu callbacks using this
169 /* Finally it is safe to free my network namespace structure */
173 void __put_net(struct net *net)
175 /* Cleanup the network namespace in process context */
176 INIT_WORK(&net->work, cleanup_net);
177 queue_work(netns_wq, &net->work);
179 EXPORT_SYMBOL_GPL(__put_net);
182 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
184 if (flags & CLONE_NEWNET)
185 return ERR_PTR(-EINVAL);
190 static int __init net_ns_init(void)
194 printk(KERN_INFO "net_namespace: %zd bytes\n", sizeof(struct net));
196 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
200 /* Create workqueue for cleanup */
201 netns_wq = create_singlethread_workqueue("netns");
203 panic("Could not create netns workq");
206 mutex_lock(&net_mutex);
207 err = setup_net(&init_net);
210 list_add_tail(&init_net.list, &net_namespace_list);
213 mutex_unlock(&net_mutex);
215 panic("Could not setup the initial network namespace");
220 pure_initcall(net_ns_init);
223 static int register_pernet_operations(struct list_head *list,
224 struct pernet_operations *ops)
226 struct net *net, *undo_net;
229 list_add_tail(&ops->list, list);
232 error = ops->init(net);
240 /* If I have an error cleanup all namespaces I initialized */
241 list_del(&ops->list);
243 for_each_net(undo_net) {
253 static void unregister_pernet_operations(struct pernet_operations *ops)
257 list_del(&ops->list);
265 static int register_pernet_operations(struct list_head *list,
266 struct pernet_operations *ops)
268 if (ops->init == NULL)
270 return ops->init(&init_net);
273 static void unregister_pernet_operations(struct pernet_operations *ops)
276 ops->exit(&init_net);
280 static DEFINE_IDA(net_generic_ids);
283 * register_pernet_subsys - register a network namespace subsystem
284 * @ops: pernet operations structure for the subsystem
286 * Register a subsystem which has init and exit functions
287 * that are called when network namespaces are created and
288 * destroyed respectively.
290 * When registered all network namespace init functions are
291 * called for every existing network namespace. Allowing kernel
292 * modules to have a race free view of the set of network namespaces.
294 * When a new network namespace is created all of the init
295 * methods are called in the order in which they were registered.
297 * When a network namespace is destroyed all of the exit methods
298 * are called in the reverse of the order with which they were
301 int register_pernet_subsys(struct pernet_operations *ops)
304 mutex_lock(&net_mutex);
305 error = register_pernet_operations(first_device, ops);
306 mutex_unlock(&net_mutex);
309 EXPORT_SYMBOL_GPL(register_pernet_subsys);
312 * unregister_pernet_subsys - unregister a network namespace subsystem
313 * @ops: pernet operations structure to manipulate
315 * Remove the pernet operations structure from the list to be
316 * used when network namespaces are created or destroyed. In
317 * addition run the exit method for all existing network
320 void unregister_pernet_subsys(struct pernet_operations *module)
322 mutex_lock(&net_mutex);
323 unregister_pernet_operations(module);
324 mutex_unlock(&net_mutex);
326 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
329 * register_pernet_device - register a network namespace device
330 * @ops: pernet operations structure for the subsystem
332 * Register a device which has init and exit functions
333 * that are called when network namespaces are created and
334 * destroyed respectively.
336 * When registered all network namespace init functions are
337 * called for every existing network namespace. Allowing kernel
338 * modules to have a race free view of the set of network namespaces.
340 * When a new network namespace is created all of the init
341 * methods are called in the order in which they were registered.
343 * When a network namespace is destroyed all of the exit methods
344 * are called in the reverse of the order with which they were
347 int register_pernet_device(struct pernet_operations *ops)
350 mutex_lock(&net_mutex);
351 error = register_pernet_operations(&pernet_list, ops);
352 if (!error && (first_device == &pernet_list))
353 first_device = &ops->list;
354 mutex_unlock(&net_mutex);
357 EXPORT_SYMBOL_GPL(register_pernet_device);
359 int register_pernet_gen_device(int *id, struct pernet_operations *ops)
362 mutex_lock(&net_mutex);
364 error = ida_get_new_above(&net_generic_ids, 1, id);
366 if (error == -EAGAIN) {
367 ida_pre_get(&net_generic_ids, GFP_KERNEL);
372 error = register_pernet_operations(&pernet_list, ops);
374 ida_remove(&net_generic_ids, *id);
375 else if (first_device == &pernet_list)
376 first_device = &ops->list;
378 mutex_unlock(&net_mutex);
381 EXPORT_SYMBOL_GPL(register_pernet_gen_device);
384 * unregister_pernet_device - unregister a network namespace netdevice
385 * @ops: pernet operations structure to manipulate
387 * Remove the pernet operations structure from the list to be
388 * used when network namespaces are created or destroyed. In
389 * addition run the exit method for all existing network
392 void unregister_pernet_device(struct pernet_operations *ops)
394 mutex_lock(&net_mutex);
395 if (&ops->list == first_device)
396 first_device = first_device->next;
397 unregister_pernet_operations(ops);
398 mutex_unlock(&net_mutex);
400 EXPORT_SYMBOL_GPL(unregister_pernet_device);
402 void unregister_pernet_gen_device(int id, struct pernet_operations *ops)
404 mutex_lock(&net_mutex);
405 if (&ops->list == first_device)
406 first_device = first_device->next;
407 unregister_pernet_operations(ops);
408 ida_remove(&net_generic_ids, id);
409 mutex_unlock(&net_mutex);
411 EXPORT_SYMBOL_GPL(unregister_pernet_gen_device);
413 static void net_generic_release(struct rcu_head *rcu)
415 struct net_generic *ng;
417 ng = container_of(rcu, struct net_generic, rcu);
421 int net_assign_generic(struct net *net, int id, void *data)
423 struct net_generic *ng, *old_ng;
425 BUG_ON(!mutex_is_locked(&net_mutex));
428 ng = old_ng = net->gen;
429 if (old_ng->len >= id)
432 ng = kzalloc(sizeof(struct net_generic) +
433 id * sizeof(void *), GFP_KERNEL);
438 * Some synchronisation notes:
440 * The net_generic explores the net->gen array inside rcu
441 * read section. Besides once set the net->gen->ptr[x]
442 * pointer never changes (see rules in netns/generic.h).
444 * That said, we simply duplicate this array and schedule
445 * the old copy for kfree after a grace period.
449 INIT_RCU_HEAD(&ng->rcu);
450 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len);
452 rcu_assign_pointer(net->gen, ng);
453 call_rcu(&old_ng->rcu, net_generic_release);
455 ng->ptr[id - 1] = data;
458 EXPORT_SYMBOL_GPL(net_assign_generic);