Loading...
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 <net/net_namespace.h>
19#include <net/netns/generic.h>
20
21/*
22 * Our network namespace constructor/destructor lists
23 */
24
25static LIST_HEAD(pernet_list);
26static struct list_head *first_device = &pernet_list;
27DEFINE_MUTEX(net_mutex);
28
29LIST_HEAD(net_namespace_list);
30EXPORT_SYMBOL_GPL(net_namespace_list);
31
32struct net init_net = {
33 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
34};
35EXPORT_SYMBOL(init_net);
36
37#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
38
39static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
40
41static struct net_generic *net_alloc_generic(void)
42{
43 struct net_generic *ng;
44 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
45
46 ng = kzalloc(generic_size, GFP_KERNEL);
47 if (ng)
48 ng->len = max_gen_ptrs;
49
50 return ng;
51}
52
53static int net_assign_generic(struct net *net, int id, void *data)
54{
55 struct net_generic *ng, *old_ng;
56
57 BUG_ON(!mutex_is_locked(&net_mutex));
58 BUG_ON(id == 0);
59
60 old_ng = rcu_dereference_protected(net->gen,
61 lockdep_is_held(&net_mutex));
62 ng = old_ng;
63 if (old_ng->len >= id)
64 goto assign;
65
66 ng = net_alloc_generic();
67 if (ng == NULL)
68 return -ENOMEM;
69
70 /*
71 * Some synchronisation notes:
72 *
73 * The net_generic explores the net->gen array inside rcu
74 * read section. Besides once set the net->gen->ptr[x]
75 * pointer never changes (see rules in netns/generic.h).
76 *
77 * That said, we simply duplicate this array and schedule
78 * the old copy for kfree after a grace period.
79 */
80
81 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
82
83 rcu_assign_pointer(net->gen, ng);
84 kfree_rcu(old_ng, rcu);
85assign:
86 ng->ptr[id - 1] = data;
87 return 0;
88}
89
90static int ops_init(const struct pernet_operations *ops, struct net *net)
91{
92 int err = -ENOMEM;
93 void *data = NULL;
94
95 if (ops->id && ops->size) {
96 data = kzalloc(ops->size, GFP_KERNEL);
97 if (!data)
98 goto out;
99
100 err = net_assign_generic(net, *ops->id, data);
101 if (err)
102 goto cleanup;
103 }
104 err = 0;
105 if (ops->init)
106 err = ops->init(net);
107 if (!err)
108 return 0;
109
110cleanup:
111 kfree(data);
112
113out:
114 return err;
115}
116
117static void ops_free(const struct pernet_operations *ops, struct net *net)
118{
119 if (ops->id && ops->size) {
120 int id = *ops->id;
121 kfree(net_generic(net, id));
122 }
123}
124
125static void ops_exit_list(const struct pernet_operations *ops,
126 struct list_head *net_exit_list)
127{
128 struct net *net;
129 if (ops->exit) {
130 list_for_each_entry(net, net_exit_list, exit_list)
131 ops->exit(net);
132 }
133 if (ops->exit_batch)
134 ops->exit_batch(net_exit_list);
135}
136
137static void ops_free_list(const struct pernet_operations *ops,
138 struct list_head *net_exit_list)
139{
140 struct net *net;
141 if (ops->size && ops->id) {
142 list_for_each_entry(net, net_exit_list, exit_list)
143 ops_free(ops, net);
144 }
145}
146
147/*
148 * setup_net runs the initializers for the network namespace object.
149 */
150static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
151{
152 /* Must be called with net_mutex held */
153 const struct pernet_operations *ops, *saved_ops;
154 int error = 0;
155 LIST_HEAD(net_exit_list);
156
157 atomic_set(&net->count, 1);
158 atomic_set(&net->passive, 1);
159 net->dev_base_seq = 1;
160 net->user_ns = user_ns;
161
162#ifdef NETNS_REFCNT_DEBUG
163 atomic_set(&net->use_count, 0);
164#endif
165
166 list_for_each_entry(ops, &pernet_list, list) {
167 error = ops_init(ops, net);
168 if (error < 0)
169 goto out_undo;
170 }
171out:
172 return error;
173
174out_undo:
175 /* Walk through the list backwards calling the exit functions
176 * for the pernet modules whose init functions did not fail.
177 */
178 list_add(&net->exit_list, &net_exit_list);
179 saved_ops = ops;
180 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
181 ops_exit_list(ops, &net_exit_list);
182
183 ops = saved_ops;
184 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
185 ops_free_list(ops, &net_exit_list);
186
187 rcu_barrier();
188 goto out;
189}
190
191
192#ifdef CONFIG_NET_NS
193static struct kmem_cache *net_cachep;
194static struct workqueue_struct *netns_wq;
195
196static struct net *net_alloc(void)
197{
198 struct net *net = NULL;
199 struct net_generic *ng;
200
201 ng = net_alloc_generic();
202 if (!ng)
203 goto out;
204
205 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
206 if (!net)
207 goto out_free;
208
209 rcu_assign_pointer(net->gen, ng);
210out:
211 return net;
212
213out_free:
214 kfree(ng);
215 goto out;
216}
217
218static void net_free(struct net *net)
219{
220#ifdef NETNS_REFCNT_DEBUG
221 if (unlikely(atomic_read(&net->use_count) != 0)) {
222 pr_emerg("network namespace not free! Usage: %d\n",
223 atomic_read(&net->use_count));
224 return;
225 }
226#endif
227 kfree(net->gen);
228 kmem_cache_free(net_cachep, net);
229}
230
231void net_drop_ns(void *p)
232{
233 struct net *ns = p;
234 if (ns && atomic_dec_and_test(&ns->passive))
235 net_free(ns);
236}
237
238struct net *copy_net_ns(unsigned long flags,
239 struct user_namespace *user_ns, struct net *old_net)
240{
241 struct net *net;
242 int rv;
243
244 if (!(flags & CLONE_NEWNET))
245 return get_net(old_net);
246
247 net = net_alloc();
248 if (!net)
249 return ERR_PTR(-ENOMEM);
250
251 get_user_ns(user_ns);
252
253 mutex_lock(&net_mutex);
254 rv = setup_net(net, user_ns);
255 if (rv == 0) {
256 rtnl_lock();
257 list_add_tail_rcu(&net->list, &net_namespace_list);
258 rtnl_unlock();
259 }
260 mutex_unlock(&net_mutex);
261 if (rv < 0) {
262 put_user_ns(user_ns);
263 net_drop_ns(net);
264 return ERR_PTR(rv);
265 }
266 return net;
267}
268
269static DEFINE_SPINLOCK(cleanup_list_lock);
270static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
271
272static void cleanup_net(struct work_struct *work)
273{
274 const struct pernet_operations *ops;
275 struct net *net, *tmp;
276 LIST_HEAD(net_kill_list);
277 LIST_HEAD(net_exit_list);
278
279 /* Atomically snapshot the list of namespaces to cleanup */
280 spin_lock_irq(&cleanup_list_lock);
281 list_replace_init(&cleanup_list, &net_kill_list);
282 spin_unlock_irq(&cleanup_list_lock);
283
284 mutex_lock(&net_mutex);
285
286 /* Don't let anyone else find us. */
287 rtnl_lock();
288 list_for_each_entry(net, &net_kill_list, cleanup_list) {
289 list_del_rcu(&net->list);
290 list_add_tail(&net->exit_list, &net_exit_list);
291 }
292 rtnl_unlock();
293
294 /*
295 * Another CPU might be rcu-iterating the list, wait for it.
296 * This needs to be before calling the exit() notifiers, so
297 * the rcu_barrier() below isn't sufficient alone.
298 */
299 synchronize_rcu();
300
301 /* Run all of the network namespace exit methods */
302 list_for_each_entry_reverse(ops, &pernet_list, list)
303 ops_exit_list(ops, &net_exit_list);
304
305 /* Free the net generic variables */
306 list_for_each_entry_reverse(ops, &pernet_list, list)
307 ops_free_list(ops, &net_exit_list);
308
309 mutex_unlock(&net_mutex);
310
311 /* Ensure there are no outstanding rcu callbacks using this
312 * network namespace.
313 */
314 rcu_barrier();
315
316 /* Finally it is safe to free my network namespace structure */
317 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
318 list_del_init(&net->exit_list);
319 put_user_ns(net->user_ns);
320 net_drop_ns(net);
321 }
322}
323static DECLARE_WORK(net_cleanup_work, cleanup_net);
324
325void __put_net(struct net *net)
326{
327 /* Cleanup the network namespace in process context */
328 unsigned long flags;
329
330 spin_lock_irqsave(&cleanup_list_lock, flags);
331 list_add(&net->cleanup_list, &cleanup_list);
332 spin_unlock_irqrestore(&cleanup_list_lock, flags);
333
334 queue_work(netns_wq, &net_cleanup_work);
335}
336EXPORT_SYMBOL_GPL(__put_net);
337
338struct net *get_net_ns_by_fd(int fd)
339{
340 struct proc_ns *ei;
341 struct file *file;
342 struct net *net;
343
344 file = proc_ns_fget(fd);
345 if (IS_ERR(file))
346 return ERR_CAST(file);
347
348 ei = get_proc_ns(file_inode(file));
349 if (ei->ns_ops == &netns_operations)
350 net = get_net(ei->ns);
351 else
352 net = ERR_PTR(-EINVAL);
353
354 fput(file);
355 return net;
356}
357
358#else
359struct net *get_net_ns_by_fd(int fd)
360{
361 return ERR_PTR(-EINVAL);
362}
363#endif
364
365struct net *get_net_ns_by_pid(pid_t pid)
366{
367 struct task_struct *tsk;
368 struct net *net;
369
370 /* Lookup the network namespace */
371 net = ERR_PTR(-ESRCH);
372 rcu_read_lock();
373 tsk = find_task_by_vpid(pid);
374 if (tsk) {
375 struct nsproxy *nsproxy;
376 nsproxy = task_nsproxy(tsk);
377 if (nsproxy)
378 net = get_net(nsproxy->net_ns);
379 }
380 rcu_read_unlock();
381 return net;
382}
383EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
384
385static __net_init int net_ns_net_init(struct net *net)
386{
387 return proc_alloc_inum(&net->proc_inum);
388}
389
390static __net_exit void net_ns_net_exit(struct net *net)
391{
392 proc_free_inum(net->proc_inum);
393}
394
395static struct pernet_operations __net_initdata net_ns_ops = {
396 .init = net_ns_net_init,
397 .exit = net_ns_net_exit,
398};
399
400static int __init net_ns_init(void)
401{
402 struct net_generic *ng;
403
404#ifdef CONFIG_NET_NS
405 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
406 SMP_CACHE_BYTES,
407 SLAB_PANIC, NULL);
408
409 /* Create workqueue for cleanup */
410 netns_wq = create_singlethread_workqueue("netns");
411 if (!netns_wq)
412 panic("Could not create netns workq");
413#endif
414
415 ng = net_alloc_generic();
416 if (!ng)
417 panic("Could not allocate generic netns");
418
419 rcu_assign_pointer(init_net.gen, ng);
420
421 mutex_lock(&net_mutex);
422 if (setup_net(&init_net, &init_user_ns))
423 panic("Could not setup the initial network namespace");
424
425 rtnl_lock();
426 list_add_tail_rcu(&init_net.list, &net_namespace_list);
427 rtnl_unlock();
428
429 mutex_unlock(&net_mutex);
430
431 register_pernet_subsys(&net_ns_ops);
432
433 return 0;
434}
435
436pure_initcall(net_ns_init);
437
438#ifdef CONFIG_NET_NS
439static int __register_pernet_operations(struct list_head *list,
440 struct pernet_operations *ops)
441{
442 struct net *net;
443 int error;
444 LIST_HEAD(net_exit_list);
445
446 list_add_tail(&ops->list, list);
447 if (ops->init || (ops->id && ops->size)) {
448 for_each_net(net) {
449 error = ops_init(ops, net);
450 if (error)
451 goto out_undo;
452 list_add_tail(&net->exit_list, &net_exit_list);
453 }
454 }
455 return 0;
456
457out_undo:
458 /* If I have an error cleanup all namespaces I initialized */
459 list_del(&ops->list);
460 ops_exit_list(ops, &net_exit_list);
461 ops_free_list(ops, &net_exit_list);
462 return error;
463}
464
465static void __unregister_pernet_operations(struct pernet_operations *ops)
466{
467 struct net *net;
468 LIST_HEAD(net_exit_list);
469
470 list_del(&ops->list);
471 for_each_net(net)
472 list_add_tail(&net->exit_list, &net_exit_list);
473 ops_exit_list(ops, &net_exit_list);
474 ops_free_list(ops, &net_exit_list);
475}
476
477#else
478
479static int __register_pernet_operations(struct list_head *list,
480 struct pernet_operations *ops)
481{
482 return ops_init(ops, &init_net);
483}
484
485static void __unregister_pernet_operations(struct pernet_operations *ops)
486{
487 LIST_HEAD(net_exit_list);
488 list_add(&init_net.exit_list, &net_exit_list);
489 ops_exit_list(ops, &net_exit_list);
490 ops_free_list(ops, &net_exit_list);
491}
492
493#endif /* CONFIG_NET_NS */
494
495static DEFINE_IDA(net_generic_ids);
496
497static int register_pernet_operations(struct list_head *list,
498 struct pernet_operations *ops)
499{
500 int error;
501
502 if (ops->id) {
503again:
504 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
505 if (error < 0) {
506 if (error == -EAGAIN) {
507 ida_pre_get(&net_generic_ids, GFP_KERNEL);
508 goto again;
509 }
510 return error;
511 }
512 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
513 }
514 error = __register_pernet_operations(list, ops);
515 if (error) {
516 rcu_barrier();
517 if (ops->id)
518 ida_remove(&net_generic_ids, *ops->id);
519 }
520
521 return error;
522}
523
524static void unregister_pernet_operations(struct pernet_operations *ops)
525{
526
527 __unregister_pernet_operations(ops);
528 rcu_barrier();
529 if (ops->id)
530 ida_remove(&net_generic_ids, *ops->id);
531}
532
533/**
534 * register_pernet_subsys - register a network namespace subsystem
535 * @ops: pernet operations structure for the subsystem
536 *
537 * Register a subsystem which has init and exit functions
538 * that are called when network namespaces are created and
539 * destroyed respectively.
540 *
541 * When registered all network namespace init functions are
542 * called for every existing network namespace. Allowing kernel
543 * modules to have a race free view of the set of network namespaces.
544 *
545 * When a new network namespace is created all of the init
546 * methods are called in the order in which they were registered.
547 *
548 * When a network namespace is destroyed all of the exit methods
549 * are called in the reverse of the order with which they were
550 * registered.
551 */
552int register_pernet_subsys(struct pernet_operations *ops)
553{
554 int error;
555 mutex_lock(&net_mutex);
556 error = register_pernet_operations(first_device, ops);
557 mutex_unlock(&net_mutex);
558 return error;
559}
560EXPORT_SYMBOL_GPL(register_pernet_subsys);
561
562/**
563 * unregister_pernet_subsys - unregister a network namespace subsystem
564 * @ops: pernet operations structure to manipulate
565 *
566 * Remove the pernet operations structure from the list to be
567 * used when network namespaces are created or destroyed. In
568 * addition run the exit method for all existing network
569 * namespaces.
570 */
571void unregister_pernet_subsys(struct pernet_operations *ops)
572{
573 mutex_lock(&net_mutex);
574 unregister_pernet_operations(ops);
575 mutex_unlock(&net_mutex);
576}
577EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
578
579/**
580 * register_pernet_device - register a network namespace device
581 * @ops: pernet operations structure for the subsystem
582 *
583 * Register a device which has init and exit functions
584 * that are called when network namespaces are created and
585 * destroyed respectively.
586 *
587 * When registered all network namespace init functions are
588 * called for every existing network namespace. Allowing kernel
589 * modules to have a race free view of the set of network namespaces.
590 *
591 * When a new network namespace is created all of the init
592 * methods are called in the order in which they were registered.
593 *
594 * When a network namespace is destroyed all of the exit methods
595 * are called in the reverse of the order with which they were
596 * registered.
597 */
598int register_pernet_device(struct pernet_operations *ops)
599{
600 int error;
601 mutex_lock(&net_mutex);
602 error = register_pernet_operations(&pernet_list, ops);
603 if (!error && (first_device == &pernet_list))
604 first_device = &ops->list;
605 mutex_unlock(&net_mutex);
606 return error;
607}
608EXPORT_SYMBOL_GPL(register_pernet_device);
609
610/**
611 * unregister_pernet_device - unregister a network namespace netdevice
612 * @ops: pernet operations structure to manipulate
613 *
614 * Remove the pernet operations structure from the list to be
615 * used when network namespaces are created or destroyed. In
616 * addition run the exit method for all existing network
617 * namespaces.
618 */
619void unregister_pernet_device(struct pernet_operations *ops)
620{
621 mutex_lock(&net_mutex);
622 if (&ops->list == first_device)
623 first_device = first_device->next;
624 unregister_pernet_operations(ops);
625 mutex_unlock(&net_mutex);
626}
627EXPORT_SYMBOL_GPL(unregister_pernet_device);
628
629#ifdef CONFIG_NET_NS
630static void *netns_get(struct task_struct *task)
631{
632 struct net *net = NULL;
633 struct nsproxy *nsproxy;
634
635 rcu_read_lock();
636 nsproxy = task_nsproxy(task);
637 if (nsproxy)
638 net = get_net(nsproxy->net_ns);
639 rcu_read_unlock();
640
641 return net;
642}
643
644static void netns_put(void *ns)
645{
646 put_net(ns);
647}
648
649static int netns_install(struct nsproxy *nsproxy, void *ns)
650{
651 struct net *net = ns;
652
653 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
654 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
655 return -EPERM;
656
657 put_net(nsproxy->net_ns);
658 nsproxy->net_ns = get_net(net);
659 return 0;
660}
661
662static unsigned int netns_inum(void *ns)
663{
664 struct net *net = ns;
665 return net->proc_inum;
666}
667
668const struct proc_ns_operations netns_operations = {
669 .name = "net",
670 .type = CLONE_NEWNET,
671 .get = netns_get,
672 .put = netns_put,
673 .install = netns_install,
674 .inum = netns_inum,
675};
676#endif
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