xref: /openbmc/linux/kernel/nsproxy.c (revision 11c416e3) 
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			= ATOMIC_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 		atomic_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 	if (new_nsp->pid_ns_for_children)
132 		put_pid_ns(new_nsp->pid_ns_for_children);
133 out_pid:
134 	if (new_nsp->ipc_ns)
135 		put_ipc_ns(new_nsp->ipc_ns);
136 out_ipc:
137 	if (new_nsp->uts_ns)
138 		put_uts_ns(new_nsp->uts_ns);
139 out_uts:
140 	if (new_nsp->mnt_ns)
141 		put_mnt_ns(new_nsp->mnt_ns);
142 out_ns:
143 	kmem_cache_free(nsproxy_cachep, new_nsp);
144 	return ERR_PTR(err);
145 }
146 
147 /*
148  * called from clone.  This now handles copy for nsproxy and all
149  * namespaces therein.
150  */
151 int copy_namespaces(unsigned long flags, struct task_struct *tsk)
152 {
153 	struct nsproxy *old_ns = tsk->nsproxy;
154 	struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
155 	struct nsproxy *new_ns;
156 	int ret;
157 
158 	if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
159 			      CLONE_NEWPID | CLONE_NEWNET |
160 			      CLONE_NEWCGROUP | CLONE_NEWTIME)))) {
161 		if (likely(old_ns->time_ns_for_children == old_ns->time_ns)) {
162 			get_nsproxy(old_ns);
163 			return 0;
164 		}
165 	} else if (!ns_capable(user_ns, CAP_SYS_ADMIN))
166 		return -EPERM;
167 
168 	/*
169 	 * CLONE_NEWIPC must detach from the undolist: after switching
170 	 * to a new ipc namespace, the semaphore arrays from the old
171 	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
172 	 * means share undolist with parent, so we must forbid using
173 	 * it along with CLONE_NEWIPC.
174 	 */
175 	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
176 		(CLONE_NEWIPC | CLONE_SYSVSEM))
177 		return -EINVAL;
178 
179 	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
180 	if (IS_ERR(new_ns))
181 		return  PTR_ERR(new_ns);
182 
183 	ret = timens_on_fork(new_ns, tsk);
184 	if (ret) {
185 		free_nsproxy(new_ns);
186 		return ret;
187 	}
188 
189 	tsk->nsproxy = new_ns;
190 	return 0;
191 }
192 
193 void free_nsproxy(struct nsproxy *ns)
194 {
195 	if (ns->mnt_ns)
196 		put_mnt_ns(ns->mnt_ns);
197 	if (ns->uts_ns)
198 		put_uts_ns(ns->uts_ns);
199 	if (ns->ipc_ns)
200 		put_ipc_ns(ns->ipc_ns);
201 	if (ns->pid_ns_for_children)
202 		put_pid_ns(ns->pid_ns_for_children);
203 	if (ns->time_ns)
204 		put_time_ns(ns->time_ns);
205 	if (ns->time_ns_for_children)
206 		put_time_ns(ns->time_ns_for_children);
207 	put_cgroup_ns(ns->cgroup_ns);
208 	put_net(ns->net_ns);
209 	kmem_cache_free(nsproxy_cachep, ns);
210 }
211 
212 /*
213  * Called from unshare. Unshare all the namespaces part of nsproxy.
214  * On success, returns the new nsproxy.
215  */
216 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
217 	struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
218 {
219 	struct user_namespace *user_ns;
220 	int err = 0;
221 
222 	if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
223 			       CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP |
224 			       CLONE_NEWTIME)))
225 		return 0;
226 
227 	user_ns = new_cred ? new_cred->user_ns : current_user_ns();
228 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
229 		return -EPERM;
230 
231 	*new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
232 					 new_fs ? new_fs : current->fs);
233 	if (IS_ERR(*new_nsp)) {
234 		err = PTR_ERR(*new_nsp);
235 		goto out;
236 	}
237 
238 out:
239 	return err;
240 }
241 
242 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
243 {
244 	struct nsproxy *ns;
245 
246 	might_sleep();
247 
248 	task_lock(p);
249 	ns = p->nsproxy;
250 	p->nsproxy = new;
251 	task_unlock(p);
252 
253 	if (ns && atomic_dec_and_test(&ns->count))
254 		free_nsproxy(ns);
255 }
256 
257 void exit_task_namespaces(struct task_struct *p)
258 {
259 	switch_task_namespaces(p, NULL);
260 }
261 
262 static int check_setns_flags(unsigned long flags)
263 {
264 	if (!flags || (flags & ~(CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
265 				 CLONE_NEWNET | CLONE_NEWUSER | CLONE_NEWPID |
266 				 CLONE_NEWCGROUP)))
267 		return -EINVAL;
268 
269 #ifndef CONFIG_USER_NS
270 	if (flags & CLONE_NEWUSER)
271 		return -EINVAL;
272 #endif
273 #ifndef CONFIG_PID_NS
274 	if (flags & CLONE_NEWPID)
275 		return -EINVAL;
276 #endif
277 #ifndef CONFIG_UTS_NS
278 	if (flags & CLONE_NEWUTS)
279 		return -EINVAL;
280 #endif
281 #ifndef CONFIG_IPC_NS
282 	if (flags & CLONE_NEWIPC)
283 		return -EINVAL;
284 #endif
285 #ifndef CONFIG_CGROUPS
286 	if (flags & CLONE_NEWCGROUP)
287 		return -EINVAL;
288 #endif
289 #ifndef CONFIG_NET_NS
290 	if (flags & CLONE_NEWNET)
291 		return -EINVAL;
292 #endif
293 
294 	return 0;
295 }
296 
297 static void put_nsset(struct nsset *nsset)
298 {
299 	unsigned flags = nsset->flags;
300 
301 	if (flags & CLONE_NEWUSER)
302 		put_cred(nsset_cred(nsset));
303 	/*
304 	 * We only created a temporary copy if we attached to more than just
305 	 * the mount namespace.
306 	 */
307 	if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS))
308 		free_fs_struct(nsset->fs);
309 	if (nsset->nsproxy)
310 		free_nsproxy(nsset->nsproxy);
311 }
312 
313 static int prepare_nsset(unsigned flags, struct nsset *nsset)
314 {
315 	struct task_struct *me = current;
316 
317 	nsset->nsproxy = create_new_namespaces(0, me, current_user_ns(), me->fs);
318 	if (IS_ERR(nsset->nsproxy))
319 		return PTR_ERR(nsset->nsproxy);
320 
321 	if (flags & CLONE_NEWUSER)
322 		nsset->cred = prepare_creds();
323 	else
324 		nsset->cred = current_cred();
325 	if (!nsset->cred)
326 		goto out;
327 
328 	/* Only create a temporary copy of fs_struct if we really need to. */
329 	if (flags == CLONE_NEWNS) {
330 		nsset->fs = me->fs;
331 	} else if (flags & CLONE_NEWNS) {
332 		nsset->fs = copy_fs_struct(me->fs);
333 		if (!nsset->fs)
334 			goto out;
335 	}
336 
337 	nsset->flags = flags;
338 	return 0;
339 
340 out:
341 	put_nsset(nsset);
342 	return -ENOMEM;
343 }
344 
345 static inline int validate_ns(struct nsset *nsset, struct ns_common *ns)
346 {
347 	return ns->ops->install(nsset, ns);
348 }
349 
350 /*
351  * This is the inverse operation to unshare().
352  * Ordering is equivalent to the standard ordering used everywhere else
353  * during unshare and process creation. The switch to the new set of
354  * namespaces occurs at the point of no return after installation of
355  * all requested namespaces was successful in commit_nsset().
356  */
357 static int validate_nsset(struct nsset *nsset, struct pid *pid)
358 {
359 	int ret = 0;
360 	unsigned flags = nsset->flags;
361 	struct user_namespace *user_ns = NULL;
362 	struct pid_namespace *pid_ns = NULL;
363 	struct nsproxy *nsp;
364 	struct task_struct *tsk;
365 
366 	/* Take a "snapshot" of the target task's namespaces. */
367 	rcu_read_lock();
368 	tsk = pid_task(pid, PIDTYPE_PID);
369 	if (!tsk) {
370 		rcu_read_unlock();
371 		return -ESRCH;
372 	}
373 
374 	if (!ptrace_may_access(tsk, PTRACE_MODE_READ_REALCREDS)) {
375 		rcu_read_unlock();
376 		return -EPERM;
377 	}
378 
379 	task_lock(tsk);
380 	nsp = tsk->nsproxy;
381 	if (nsp)
382 		get_nsproxy(nsp);
383 	task_unlock(tsk);
384 	if (!nsp) {
385 		rcu_read_unlock();
386 		return -ESRCH;
387 	}
388 
389 #ifdef CONFIG_PID_NS
390 	if (flags & CLONE_NEWPID) {
391 		pid_ns = task_active_pid_ns(tsk);
392 		if (unlikely(!pid_ns)) {
393 			rcu_read_unlock();
394 			ret = -ESRCH;
395 			goto out;
396 		}
397 		get_pid_ns(pid_ns);
398 	}
399 #endif
400 
401 #ifdef CONFIG_USER_NS
402 	if (flags & CLONE_NEWUSER)
403 		user_ns = get_user_ns(__task_cred(tsk)->user_ns);
404 #endif
405 	rcu_read_unlock();
406 
407 	/*
408 	 * Install requested namespaces. The caller will have
409 	 * verified earlier that the requested namespaces are
410 	 * supported on this kernel. We don't report errors here
411 	 * if a namespace is requested that isn't supported.
412 	 */
413 #ifdef CONFIG_USER_NS
414 	if (flags & CLONE_NEWUSER) {
415 		ret = validate_ns(nsset, &user_ns->ns);
416 		if (ret)
417 			goto out;
418 	}
419 #endif
420 
421 	if (flags & CLONE_NEWNS) {
422 		ret = validate_ns(nsset, from_mnt_ns(nsp->mnt_ns));
423 		if (ret)
424 			goto out;
425 	}
426 
427 #ifdef CONFIG_UTS_NS
428 	if (flags & CLONE_NEWUTS) {
429 		ret = validate_ns(nsset, &nsp->uts_ns->ns);
430 		if (ret)
431 			goto out;
432 	}
433 #endif
434 
435 #ifdef CONFIG_IPC_NS
436 	if (flags & CLONE_NEWIPC) {
437 		ret = validate_ns(nsset, &nsp->ipc_ns->ns);
438 		if (ret)
439 			goto out;
440 	}
441 #endif
442 
443 #ifdef CONFIG_PID_NS
444 	if (flags & CLONE_NEWPID) {
445 		ret = validate_ns(nsset, &pid_ns->ns);
446 		if (ret)
447 			goto out;
448 	}
449 #endif
450 
451 #ifdef CONFIG_CGROUPS
452 	if (flags & CLONE_NEWCGROUP) {
453 		ret = validate_ns(nsset, &nsp->cgroup_ns->ns);
454 		if (ret)
455 			goto out;
456 	}
457 #endif
458 
459 #ifdef CONFIG_NET_NS
460 	if (flags & CLONE_NEWNET) {
461 		ret = validate_ns(nsset, &nsp->net_ns->ns);
462 		if (ret)
463 			goto out;
464 	}
465 #endif
466 
467 out:
468 	if (pid_ns)
469 		put_pid_ns(pid_ns);
470 	if (nsp)
471 		put_nsproxy(nsp);
472 	put_user_ns(user_ns);
473 
474 	return ret;
475 }
476 
477 /*
478  * This is the point of no return. There are just a few namespaces
479  * that do some actual work here and it's sufficiently minimal that
480  * a separate ns_common operation seems unnecessary for now.
481  * Unshare is doing the same thing. If we'll end up needing to do
482  * more in a given namespace or a helper here is ultimately not
483  * exported anymore a simple commit handler for each namespace
484  * should be added to ns_common.
485  */
486 static void commit_nsset(struct nsset *nsset)
487 {
488 	unsigned flags = nsset->flags;
489 	struct task_struct *me = current;
490 
491 #ifdef CONFIG_USER_NS
492 	if (flags & CLONE_NEWUSER) {
493 		/* transfer ownership */
494 		commit_creds(nsset_cred(nsset));
495 		nsset->cred = NULL;
496 	}
497 #endif
498 
499 	/* We only need to commit if we have used a temporary fs_struct. */
500 	if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) {
501 		set_fs_root(me->fs, &nsset->fs->root);
502 		set_fs_pwd(me->fs, &nsset->fs->pwd);
503 	}
504 
505 #ifdef CONFIG_IPC_NS
506 	if (flags & CLONE_NEWIPC)
507 		exit_sem(me);
508 #endif
509 
510 	/* transfer ownership */
511 	switch_task_namespaces(me, nsset->nsproxy);
512 	nsset->nsproxy = NULL;
513 }
514 
515 SYSCALL_DEFINE2(setns, int, fd, int, flags)
516 {
517 	struct file *file;
518 	struct ns_common *ns = NULL;
519 	struct nsset nsset = {};
520 	int err = 0;
521 
522 	file = fget(fd);
523 	if (!file)
524 		return -EBADF;
525 
526 	if (proc_ns_file(file)) {
527 		ns = get_proc_ns(file_inode(file));
528 		if (flags && (ns->ops->type != flags))
529 			err = -EINVAL;
530 		flags = ns->ops->type;
531 	} else if (!IS_ERR(pidfd_pid(file))) {
532 		err = check_setns_flags(flags);
533 	} else {
534 		err = -EINVAL;
535 	}
536 	if (err)
537 		goto out;
538 
539 	err = prepare_nsset(flags, &nsset);
540 	if (err)
541 		goto out;
542 
543 	if (proc_ns_file(file))
544 		err = validate_ns(&nsset, ns);
545 	else
546 		err = validate_nsset(&nsset, file->private_data);
547 	if (!err) {
548 		commit_nsset(&nsset);
549 		perf_event_namespaces(current);
550 	}
551 	put_nsset(&nsset);
552 out:
553 	fput(file);
554 	return err;
555 }
556 
557 int __init nsproxy_cache_init(void)
558 {
559 	nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
560 	return 0;
561 }
562