xref: /linux/kernel/nsproxy.c (revision 181d8e399f50c0683b12d40432bb6e1ca5c58d37) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  Copyright (C) 2006 IBM Corporation
4  *
5  *  Author: Serge Hallyn <serue@us.ibm.com>
6  *
7  *  Jun 2006 - namespaces support
8  *             OpenVZ, SWsoft Inc.
9  *             Pavel Emelianov <xemul@openvz.org>
10  */
11 
12 #include <linux/slab.h>
13 #include <linux/export.h>
14 #include <linux/nsproxy.h>
15 #include <linux/init_task.h>
16 #include <linux/mnt_namespace.h>
17 #include <linux/utsname.h>
18 #include <linux/pid_namespace.h>
19 #include <net/net_namespace.h>
20 #include <linux/ipc_namespace.h>
21 #include <linux/time_namespace.h>
22 #include <linux/fs_struct.h>
23 #include <linux/proc_fs.h>
24 #include <linux/proc_ns.h>
25 #include <linux/file.h>
26 #include <linux/syscalls.h>
27 #include <linux/cgroup.h>
28 #include <linux/perf_event.h>
29 
30 static struct kmem_cache *nsproxy_cachep;
31 
32 struct nsproxy init_nsproxy = {
33 	.count			= REFCOUNT_INIT(1),
34 	.uts_ns			= &init_uts_ns,
35 #if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
36 	.ipc_ns			= &init_ipc_ns,
37 #endif
38 	.mnt_ns			= NULL,
39 	.pid_ns_for_children	= &init_pid_ns,
40 #ifdef CONFIG_NET
41 	.net_ns			= &init_net,
42 #endif
43 #ifdef CONFIG_CGROUPS
44 	.cgroup_ns		= &init_cgroup_ns,
45 #endif
46 #ifdef CONFIG_TIME_NS
47 	.time_ns		= &init_time_ns,
48 	.time_ns_for_children	= &init_time_ns,
49 #endif
50 };
51 
52 static inline struct nsproxy *create_nsproxy(void)
53 {
54 	struct nsproxy *nsproxy;
55 
56 	nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
57 	if (nsproxy)
58 		refcount_set(&nsproxy->count, 1);
59 	return nsproxy;
60 }
61 
62 /*
63  * Create new nsproxy and all of its the associated namespaces.
64  * Return the newly created nsproxy.  Do not attach this to the task,
65  * leave it to the caller to do proper locking and attach it to task.
66  */
67 static struct nsproxy *create_new_namespaces(unsigned long flags,
68 	struct task_struct *tsk, struct user_namespace *user_ns,
69 	struct fs_struct *new_fs)
70 {
71 	struct nsproxy *new_nsp;
72 	int err;
73 
74 	new_nsp = create_nsproxy();
75 	if (!new_nsp)
76 		return ERR_PTR(-ENOMEM);
77 
78 	new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs);
79 	if (IS_ERR(new_nsp->mnt_ns)) {
80 		err = PTR_ERR(new_nsp->mnt_ns);
81 		goto out_ns;
82 	}
83 
84 	new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
85 	if (IS_ERR(new_nsp->uts_ns)) {
86 		err = PTR_ERR(new_nsp->uts_ns);
87 		goto out_uts;
88 	}
89 
90 	new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
91 	if (IS_ERR(new_nsp->ipc_ns)) {
92 		err = PTR_ERR(new_nsp->ipc_ns);
93 		goto out_ipc;
94 	}
95 
96 	new_nsp->pid_ns_for_children =
97 		copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
98 	if (IS_ERR(new_nsp->pid_ns_for_children)) {
99 		err = PTR_ERR(new_nsp->pid_ns_for_children);
100 		goto out_pid;
101 	}
102 
103 	new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns,
104 					    tsk->nsproxy->cgroup_ns);
105 	if (IS_ERR(new_nsp->cgroup_ns)) {
106 		err = PTR_ERR(new_nsp->cgroup_ns);
107 		goto out_cgroup;
108 	}
109 
110 	new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
111 	if (IS_ERR(new_nsp->net_ns)) {
112 		err = PTR_ERR(new_nsp->net_ns);
113 		goto out_net;
114 	}
115 
116 	new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns,
117 					tsk->nsproxy->time_ns_for_children);
118 	if (IS_ERR(new_nsp->time_ns_for_children)) {
119 		err = PTR_ERR(new_nsp->time_ns_for_children);
120 		goto out_time;
121 	}
122 	new_nsp->time_ns = get_time_ns(tsk->nsproxy->time_ns);
123 
124 	return new_nsp;
125 
126 out_time:
127 	put_net(new_nsp->net_ns);
128 out_net:
129 	put_cgroup_ns(new_nsp->cgroup_ns);
130 out_cgroup:
131 	put_pid_ns(new_nsp->pid_ns_for_children);
132 out_pid:
133 	put_ipc_ns(new_nsp->ipc_ns);
134 out_ipc:
135 	put_uts_ns(new_nsp->uts_ns);
136 out_uts:
137 	put_mnt_ns(new_nsp->mnt_ns);
138 out_ns:
139 	kmem_cache_free(nsproxy_cachep, new_nsp);
140 	return ERR_PTR(err);
141 }
142 
143 /*
144  * called from clone.  This now handles copy for nsproxy and all
145  * namespaces therein.
146  */
147 int copy_namespaces(unsigned long flags, struct task_struct *tsk)
148 {
149 	struct nsproxy *old_ns = tsk->nsproxy;
150 	struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
151 	struct nsproxy *new_ns;
152 
153 	if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
154 			      CLONE_NEWPID | CLONE_NEWNET |
155 			      CLONE_NEWCGROUP | CLONE_NEWTIME)))) {
156 		if ((flags & CLONE_VM) ||
157 		    likely(old_ns->time_ns_for_children == old_ns->time_ns)) {
158 			get_nsproxy(old_ns);
159 			return 0;
160 		}
161 	} else if (!ns_capable(user_ns, CAP_SYS_ADMIN))
162 		return -EPERM;
163 
164 	/*
165 	 * CLONE_NEWIPC must detach from the undolist: after switching
166 	 * to a new ipc namespace, the semaphore arrays from the old
167 	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
168 	 * means share undolist with parent, so we must forbid using
169 	 * it along with CLONE_NEWIPC.
170 	 */
171 	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
172 		(CLONE_NEWIPC | CLONE_SYSVSEM))
173 		return -EINVAL;
174 
175 	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
176 	if (IS_ERR(new_ns))
177 		return  PTR_ERR(new_ns);
178 
179 	if ((flags & CLONE_VM) == 0)
180 		timens_on_fork(new_ns, tsk);
181 
182 	tsk->nsproxy = new_ns;
183 	return 0;
184 }
185 
186 void free_nsproxy(struct nsproxy *ns)
187 {
188 	put_mnt_ns(ns->mnt_ns);
189 	put_uts_ns(ns->uts_ns);
190 	put_ipc_ns(ns->ipc_ns);
191 	put_pid_ns(ns->pid_ns_for_children);
192 	put_time_ns(ns->time_ns);
193 	put_time_ns(ns->time_ns_for_children);
194 	put_cgroup_ns(ns->cgroup_ns);
195 	put_net(ns->net_ns);
196 	kmem_cache_free(nsproxy_cachep, ns);
197 }
198 
199 /*
200  * Called from unshare. Unshare all the namespaces part of nsproxy.
201  * On success, returns the new nsproxy.
202  */
203 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
204 	struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
205 {
206 	struct user_namespace *user_ns;
207 	int err = 0;
208 
209 	if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
210 			       CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP |
211 			       CLONE_NEWTIME)))
212 		return 0;
213 
214 	user_ns = new_cred ? new_cred->user_ns : current_user_ns();
215 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
216 		return -EPERM;
217 
218 	*new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
219 					 new_fs ? new_fs : current->fs);
220 	if (IS_ERR(*new_nsp)) {
221 		err = PTR_ERR(*new_nsp);
222 		goto out;
223 	}
224 
225 out:
226 	return err;
227 }
228 
229 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
230 {
231 	struct nsproxy *ns;
232 
233 	might_sleep();
234 
235 	task_lock(p);
236 	ns = p->nsproxy;
237 	p->nsproxy = new;
238 	task_unlock(p);
239 
240 	if (ns)
241 		put_nsproxy(ns);
242 }
243 
244 void exit_task_namespaces(struct task_struct *p)
245 {
246 	switch_task_namespaces(p, NULL);
247 }
248 
249 int exec_task_namespaces(void)
250 {
251 	struct task_struct *tsk = current;
252 	struct nsproxy *new;
253 
254 	if (tsk->nsproxy->time_ns_for_children == tsk->nsproxy->time_ns)
255 		return 0;
256 
257 	new = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
258 	if (IS_ERR(new))
259 		return PTR_ERR(new);
260 
261 	timens_on_fork(new, tsk);
262 	switch_task_namespaces(tsk, new);
263 	return 0;
264 }
265 
266 static int check_setns_flags(unsigned long flags)
267 {
268 	if (!flags || (flags & ~(CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
269 				 CLONE_NEWNET | CLONE_NEWTIME | CLONE_NEWUSER |
270 				 CLONE_NEWPID | CLONE_NEWCGROUP)))
271 		return -EINVAL;
272 
273 #ifndef CONFIG_USER_NS
274 	if (flags & CLONE_NEWUSER)
275 		return -EINVAL;
276 #endif
277 #ifndef CONFIG_PID_NS
278 	if (flags & CLONE_NEWPID)
279 		return -EINVAL;
280 #endif
281 #ifndef CONFIG_UTS_NS
282 	if (flags & CLONE_NEWUTS)
283 		return -EINVAL;
284 #endif
285 #ifndef CONFIG_IPC_NS
286 	if (flags & CLONE_NEWIPC)
287 		return -EINVAL;
288 #endif
289 #ifndef CONFIG_CGROUPS
290 	if (flags & CLONE_NEWCGROUP)
291 		return -EINVAL;
292 #endif
293 #ifndef CONFIG_NET_NS
294 	if (flags & CLONE_NEWNET)
295 		return -EINVAL;
296 #endif
297 #ifndef CONFIG_TIME_NS
298 	if (flags & CLONE_NEWTIME)
299 		return -EINVAL;
300 #endif
301 
302 	return 0;
303 }
304 
305 static void put_nsset(struct nsset *nsset)
306 {
307 	unsigned flags = nsset->flags;
308 
309 	if (flags & CLONE_NEWUSER)
310 		put_cred(nsset_cred(nsset));
311 	/*
312 	 * We only created a temporary copy if we attached to more than just
313 	 * the mount namespace.
314 	 */
315 	if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS))
316 		free_fs_struct(nsset->fs);
317 	if (nsset->nsproxy)
318 		free_nsproxy(nsset->nsproxy);
319 }
320 
321 static int prepare_nsset(unsigned flags, struct nsset *nsset)
322 {
323 	struct task_struct *me = current;
324 
325 	nsset->nsproxy = create_new_namespaces(0, me, current_user_ns(), me->fs);
326 	if (IS_ERR(nsset->nsproxy))
327 		return PTR_ERR(nsset->nsproxy);
328 
329 	if (flags & CLONE_NEWUSER)
330 		nsset->cred = prepare_creds();
331 	else
332 		nsset->cred = current_cred();
333 	if (!nsset->cred)
334 		goto out;
335 
336 	/* Only create a temporary copy of fs_struct if we really need to. */
337 	if (flags == CLONE_NEWNS) {
338 		nsset->fs = me->fs;
339 	} else if (flags & CLONE_NEWNS) {
340 		nsset->fs = copy_fs_struct(me->fs);
341 		if (!nsset->fs)
342 			goto out;
343 	}
344 
345 	nsset->flags = flags;
346 	return 0;
347 
348 out:
349 	put_nsset(nsset);
350 	return -ENOMEM;
351 }
352 
353 static inline int validate_ns(struct nsset *nsset, struct ns_common *ns)
354 {
355 	return ns->ops->install(nsset, ns);
356 }
357 
358 /*
359  * This is the inverse operation to unshare().
360  * Ordering is equivalent to the standard ordering used everywhere else
361  * during unshare and process creation. The switch to the new set of
362  * namespaces occurs at the point of no return after installation of
363  * all requested namespaces was successful in commit_nsset().
364  */
365 static int validate_nsset(struct nsset *nsset, struct pid *pid)
366 {
367 	int ret = 0;
368 	unsigned flags = nsset->flags;
369 	struct user_namespace *user_ns = NULL;
370 	struct pid_namespace *pid_ns = NULL;
371 	struct nsproxy *nsp;
372 	struct task_struct *tsk;
373 
374 	/* Take a "snapshot" of the target task's namespaces. */
375 	rcu_read_lock();
376 	tsk = pid_task(pid, PIDTYPE_PID);
377 	if (!tsk) {
378 		rcu_read_unlock();
379 		return -ESRCH;
380 	}
381 
382 	if (!ptrace_may_access(tsk, PTRACE_MODE_READ_REALCREDS)) {
383 		rcu_read_unlock();
384 		return -EPERM;
385 	}
386 
387 	task_lock(tsk);
388 	nsp = tsk->nsproxy;
389 	if (nsp)
390 		get_nsproxy(nsp);
391 	task_unlock(tsk);
392 	if (!nsp) {
393 		rcu_read_unlock();
394 		return -ESRCH;
395 	}
396 
397 #ifdef CONFIG_PID_NS
398 	if (flags & CLONE_NEWPID) {
399 		pid_ns = task_active_pid_ns(tsk);
400 		if (unlikely(!pid_ns)) {
401 			rcu_read_unlock();
402 			ret = -ESRCH;
403 			goto out;
404 		}
405 		get_pid_ns(pid_ns);
406 	}
407 #endif
408 
409 #ifdef CONFIG_USER_NS
410 	if (flags & CLONE_NEWUSER)
411 		user_ns = get_user_ns(__task_cred(tsk)->user_ns);
412 #endif
413 	rcu_read_unlock();
414 
415 	/*
416 	 * Install requested namespaces. The caller will have
417 	 * verified earlier that the requested namespaces are
418 	 * supported on this kernel. We don't report errors here
419 	 * if a namespace is requested that isn't supported.
420 	 */
421 #ifdef CONFIG_USER_NS
422 	if (flags & CLONE_NEWUSER) {
423 		ret = validate_ns(nsset, &user_ns->ns);
424 		if (ret)
425 			goto out;
426 	}
427 #endif
428 
429 	if (flags & CLONE_NEWNS) {
430 		ret = validate_ns(nsset, from_mnt_ns(nsp->mnt_ns));
431 		if (ret)
432 			goto out;
433 	}
434 
435 #ifdef CONFIG_UTS_NS
436 	if (flags & CLONE_NEWUTS) {
437 		ret = validate_ns(nsset, &nsp->uts_ns->ns);
438 		if (ret)
439 			goto out;
440 	}
441 #endif
442 
443 #ifdef CONFIG_IPC_NS
444 	if (flags & CLONE_NEWIPC) {
445 		ret = validate_ns(nsset, &nsp->ipc_ns->ns);
446 		if (ret)
447 			goto out;
448 	}
449 #endif
450 
451 #ifdef CONFIG_PID_NS
452 	if (flags & CLONE_NEWPID) {
453 		ret = validate_ns(nsset, &pid_ns->ns);
454 		if (ret)
455 			goto out;
456 	}
457 #endif
458 
459 #ifdef CONFIG_CGROUPS
460 	if (flags & CLONE_NEWCGROUP) {
461 		ret = validate_ns(nsset, &nsp->cgroup_ns->ns);
462 		if (ret)
463 			goto out;
464 	}
465 #endif
466 
467 #ifdef CONFIG_NET_NS
468 	if (flags & CLONE_NEWNET) {
469 		ret = validate_ns(nsset, &nsp->net_ns->ns);
470 		if (ret)
471 			goto out;
472 	}
473 #endif
474 
475 #ifdef CONFIG_TIME_NS
476 	if (flags & CLONE_NEWTIME) {
477 		ret = validate_ns(nsset, &nsp->time_ns->ns);
478 		if (ret)
479 			goto out;
480 	}
481 #endif
482 
483 out:
484 	if (pid_ns)
485 		put_pid_ns(pid_ns);
486 	if (nsp)
487 		put_nsproxy(nsp);
488 	put_user_ns(user_ns);
489 
490 	return ret;
491 }
492 
493 /*
494  * This is the point of no return. There are just a few namespaces
495  * that do some actual work here and it's sufficiently minimal that
496  * a separate ns_common operation seems unnecessary for now.
497  * Unshare is doing the same thing. If we'll end up needing to do
498  * more in a given namespace or a helper here is ultimately not
499  * exported anymore a simple commit handler for each namespace
500  * should be added to ns_common.
501  */
502 static void commit_nsset(struct nsset *nsset)
503 {
504 	unsigned flags = nsset->flags;
505 	struct task_struct *me = current;
506 
507 #ifdef CONFIG_USER_NS
508 	if (flags & CLONE_NEWUSER) {
509 		/* transfer ownership */
510 		commit_creds(nsset_cred(nsset));
511 		nsset->cred = NULL;
512 	}
513 #endif
514 
515 	/* We only need to commit if we have used a temporary fs_struct. */
516 	if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) {
517 		set_fs_root(me->fs, &nsset->fs->root);
518 		set_fs_pwd(me->fs, &nsset->fs->pwd);
519 	}
520 
521 #ifdef CONFIG_IPC_NS
522 	if (flags & CLONE_NEWIPC)
523 		exit_sem(me);
524 #endif
525 
526 #ifdef CONFIG_TIME_NS
527 	if (flags & CLONE_NEWTIME)
528 		timens_commit(me, nsset->nsproxy->time_ns);
529 #endif
530 
531 	/* transfer ownership */
532 	switch_task_namespaces(me, nsset->nsproxy);
533 	nsset->nsproxy = NULL;
534 }
535 
536 SYSCALL_DEFINE2(setns, int, fd, int, flags)
537 {
538 	CLASS(fd, f)(fd);
539 	struct ns_common *ns = NULL;
540 	struct nsset nsset = {};
541 	int err = 0;
542 
543 	if (fd_empty(f))
544 		return -EBADF;
545 
546 	if (proc_ns_file(fd_file(f))) {
547 		ns = get_proc_ns(file_inode(fd_file(f)));
548 		if (flags && (ns->ops->type != flags))
549 			err = -EINVAL;
550 		flags = ns->ops->type;
551 	} else if (!IS_ERR(pidfd_pid(fd_file(f)))) {
552 		err = check_setns_flags(flags);
553 	} else {
554 		err = -EINVAL;
555 	}
556 	if (err)
557 		goto out;
558 
559 	err = prepare_nsset(flags, &nsset);
560 	if (err)
561 		goto out;
562 
563 	if (proc_ns_file(fd_file(f)))
564 		err = validate_ns(&nsset, ns);
565 	else
566 		err = validate_nsset(&nsset, pidfd_pid(fd_file(f)));
567 	if (!err) {
568 		commit_nsset(&nsset);
569 		perf_event_namespaces(current);
570 	}
571 	put_nsset(&nsset);
572 out:
573 	return err;
574 }
575 
576 int __init nsproxy_cache_init(void)
577 {
578 	nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC|SLAB_ACCOUNT);
579 	return 0;
580 }
581