• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include <linux/hck/lite_hck_ced.h>
30 
31 static struct kmem_cache *nsproxy_cachep;
32 
33 struct nsproxy init_nsproxy = {
34 	.count			= ATOMIC_INIT(1),
35 	.uts_ns			= &init_uts_ns,
36 #if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
37 	.ipc_ns			= &init_ipc_ns,
38 #endif
39 	.mnt_ns			= NULL,
40 	.pid_ns_for_children	= &init_pid_ns,
41 #ifdef CONFIG_NET
42 	.net_ns			= &init_net,
43 #endif
44 #ifdef CONFIG_CGROUPS
45 	.cgroup_ns		= &init_cgroup_ns,
46 #endif
47 #ifdef CONFIG_TIME_NS
48 	.time_ns		= &init_time_ns,
49 	.time_ns_for_children	= &init_time_ns,
50 #endif
51 };
52 
create_nsproxy(void)53 static inline struct nsproxy *create_nsproxy(void)
54 {
55 	struct nsproxy *nsproxy;
56 
57 	nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
58 	if (nsproxy)
59 		atomic_set(&nsproxy->count, 1);
60 	return nsproxy;
61 }
62 
63 /*
64  * Create new nsproxy and all of its the associated namespaces.
65  * Return the newly created nsproxy.  Do not attach this to the task,
66  * leave it to the caller to do proper locking and attach it to task.
67  */
create_new_namespaces(unsigned long flags,struct task_struct * tsk,struct user_namespace * user_ns,struct fs_struct * new_fs)68 static struct nsproxy *create_new_namespaces(unsigned long flags,
69 	struct task_struct *tsk, struct user_namespace *user_ns,
70 	struct fs_struct *new_fs)
71 {
72 	struct nsproxy *new_nsp;
73 	int err;
74 
75 	new_nsp = create_nsproxy();
76 	if (!new_nsp)
77 		return ERR_PTR(-ENOMEM);
78 
79 	new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs);
80 	if (IS_ERR(new_nsp->mnt_ns)) {
81 		err = PTR_ERR(new_nsp->mnt_ns);
82 		goto out_ns;
83 	}
84 
85 	new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
86 	if (IS_ERR(new_nsp->uts_ns)) {
87 		err = PTR_ERR(new_nsp->uts_ns);
88 		goto out_uts;
89 	}
90 
91 	new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
92 	if (IS_ERR(new_nsp->ipc_ns)) {
93 		err = PTR_ERR(new_nsp->ipc_ns);
94 		goto out_ipc;
95 	}
96 
97 	new_nsp->pid_ns_for_children =
98 		copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
99 	if (IS_ERR(new_nsp->pid_ns_for_children)) {
100 		err = PTR_ERR(new_nsp->pid_ns_for_children);
101 		goto out_pid;
102 	}
103 
104 	new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns,
105 					    tsk->nsproxy->cgroup_ns);
106 	if (IS_ERR(new_nsp->cgroup_ns)) {
107 		err = PTR_ERR(new_nsp->cgroup_ns);
108 		goto out_cgroup;
109 	}
110 
111 	new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
112 	if (IS_ERR(new_nsp->net_ns)) {
113 		err = PTR_ERR(new_nsp->net_ns);
114 		goto out_net;
115 	}
116 
117 	new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns,
118 					tsk->nsproxy->time_ns_for_children);
119 	if (IS_ERR(new_nsp->time_ns_for_children)) {
120 		err = PTR_ERR(new_nsp->time_ns_for_children);
121 		goto out_time;
122 	}
123 	new_nsp->time_ns = get_time_ns(tsk->nsproxy->time_ns);
124 
125 	return new_nsp;
126 
127 out_time:
128 	put_net(new_nsp->net_ns);
129 out_net:
130 	put_cgroup_ns(new_nsp->cgroup_ns);
131 out_cgroup:
132 	if (new_nsp->pid_ns_for_children)
133 		put_pid_ns(new_nsp->pid_ns_for_children);
134 out_pid:
135 	if (new_nsp->ipc_ns)
136 		put_ipc_ns(new_nsp->ipc_ns);
137 out_ipc:
138 	if (new_nsp->uts_ns)
139 		put_uts_ns(new_nsp->uts_ns);
140 out_uts:
141 	if (new_nsp->mnt_ns)
142 		put_mnt_ns(new_nsp->mnt_ns);
143 out_ns:
144 	kmem_cache_free(nsproxy_cachep, new_nsp);
145 	return ERR_PTR(err);
146 }
147 
148 /*
149  * called from clone.  This now handles copy for nsproxy and all
150  * namespaces therein.
151  */
copy_namespaces(unsigned long flags,struct task_struct * tsk)152 int copy_namespaces(unsigned long flags, struct task_struct *tsk)
153 {
154 	struct nsproxy *old_ns = tsk->nsproxy;
155 	struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
156 	struct nsproxy *new_ns;
157 	int ret;
158 
159 	if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
160 			      CLONE_NEWPID | CLONE_NEWNET |
161 			      CLONE_NEWCGROUP | CLONE_NEWTIME)))) {
162 		if (likely(old_ns->time_ns_for_children == old_ns->time_ns)) {
163 			get_nsproxy(old_ns);
164 			return 0;
165 		}
166 	} else if (!ns_capable(user_ns, CAP_SYS_ADMIN))
167 		return -EPERM;
168 
169 	/*
170 	 * CLONE_NEWIPC must detach from the undolist: after switching
171 	 * to a new ipc namespace, the semaphore arrays from the old
172 	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
173 	 * means share undolist with parent, so we must forbid using
174 	 * it along with CLONE_NEWIPC.
175 	 */
176 	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
177 		(CLONE_NEWIPC | CLONE_SYSVSEM))
178 		return -EINVAL;
179 
180 	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
181 	if (IS_ERR(new_ns))
182 		return  PTR_ERR(new_ns);
183 
184 	ret = timens_on_fork(new_ns, tsk);
185 	if (ret) {
186 		free_nsproxy(new_ns);
187 		return ret;
188 	}
189 
190 	tsk->nsproxy = new_ns;
191 	return 0;
192 }
193 
free_nsproxy(struct nsproxy * ns)194 void free_nsproxy(struct nsproxy *ns)
195 {
196 	if (ns->mnt_ns)
197 		put_mnt_ns(ns->mnt_ns);
198 	if (ns->uts_ns)
199 		put_uts_ns(ns->uts_ns);
200 	if (ns->ipc_ns)
201 		put_ipc_ns(ns->ipc_ns);
202 	if (ns->pid_ns_for_children)
203 		put_pid_ns(ns->pid_ns_for_children);
204 	if (ns->time_ns)
205 		put_time_ns(ns->time_ns);
206 	if (ns->time_ns_for_children)
207 		put_time_ns(ns->time_ns_for_children);
208 	put_cgroup_ns(ns->cgroup_ns);
209 	put_net(ns->net_ns);
210 	kmem_cache_free(nsproxy_cachep, ns);
211 }
212 
213 /*
214  * Called from unshare. Unshare all the namespaces part of nsproxy.
215  * On success, returns the new nsproxy.
216  */
unshare_nsproxy_namespaces(unsigned long unshare_flags,struct nsproxy ** new_nsp,struct cred * new_cred,struct fs_struct * new_fs)217 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
218 	struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
219 {
220 	struct user_namespace *user_ns;
221 	int err = 0;
222 
223 	if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
224 			       CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP |
225 			       CLONE_NEWTIME)))
226 		return 0;
227 
228 	user_ns = new_cred ? new_cred->user_ns : current_user_ns();
229 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
230 		return -EPERM;
231 
232 	*new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
233 					 new_fs ? new_fs : current->fs);
234 	if (IS_ERR(*new_nsp)) {
235 		err = PTR_ERR(*new_nsp);
236 		goto out;
237 	}
238 
239 out:
240 	return err;
241 }
242 
switch_task_namespaces(struct task_struct * p,struct nsproxy * new)243 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
244 {
245 	struct nsproxy *ns;
246 	int ret = 0;
247 	CALL_HCK_LITE_HOOK(ced_switch_task_namespaces_lhck, new);
248 	CALL_HCK_LITE_HOOK(ced_switch_task_namespaces_permission_lhck, new, &ret);
249 	if (ret)
250 		return;
251 
252 	might_sleep();
253 
254 	task_lock(p);
255 	ns = p->nsproxy;
256 	p->nsproxy = new;
257 	task_unlock(p);
258 
259 	if (ns && atomic_dec_and_test(&ns->count))
260 		free_nsproxy(ns);
261 }
262 
exit_task_namespaces(struct task_struct * p)263 void exit_task_namespaces(struct task_struct *p)
264 {
265 	switch_task_namespaces(p, NULL);
266 }
267 
check_setns_flags(unsigned long flags)268 static int check_setns_flags(unsigned long flags)
269 {
270 	if (!flags || (flags & ~(CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
271 				 CLONE_NEWNET | CLONE_NEWTIME | CLONE_NEWUSER |
272 				 CLONE_NEWPID | CLONE_NEWCGROUP)))
273 		return -EINVAL;
274 
275 #ifndef CONFIG_USER_NS
276 	if (flags & CLONE_NEWUSER)
277 		return -EINVAL;
278 #endif
279 #ifndef CONFIG_PID_NS
280 	if (flags & CLONE_NEWPID)
281 		return -EINVAL;
282 #endif
283 #ifndef CONFIG_UTS_NS
284 	if (flags & CLONE_NEWUTS)
285 		return -EINVAL;
286 #endif
287 #ifndef CONFIG_IPC_NS
288 	if (flags & CLONE_NEWIPC)
289 		return -EINVAL;
290 #endif
291 #ifndef CONFIG_CGROUPS
292 	if (flags & CLONE_NEWCGROUP)
293 		return -EINVAL;
294 #endif
295 #ifndef CONFIG_NET_NS
296 	if (flags & CLONE_NEWNET)
297 		return -EINVAL;
298 #endif
299 #ifndef CONFIG_TIME_NS
300 	if (flags & CLONE_NEWTIME)
301 		return -EINVAL;
302 #endif
303 
304 	return 0;
305 }
306 
put_nsset(struct nsset * nsset)307 static void put_nsset(struct nsset *nsset)
308 {
309 	unsigned flags = nsset->flags;
310 
311 	if (flags & CLONE_NEWUSER)
312 		put_cred(nsset_cred(nsset));
313 	/*
314 	 * We only created a temporary copy if we attached to more than just
315 	 * the mount namespace.
316 	 */
317 	if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS))
318 		free_fs_struct(nsset->fs);
319 	if (nsset->nsproxy)
320 		free_nsproxy(nsset->nsproxy);
321 }
322 
prepare_nsset(unsigned flags,struct nsset * nsset)323 static int prepare_nsset(unsigned flags, struct nsset *nsset)
324 {
325 	struct task_struct *me = current;
326 
327 	nsset->nsproxy = create_new_namespaces(0, me, current_user_ns(), me->fs);
328 	if (IS_ERR(nsset->nsproxy))
329 		return PTR_ERR(nsset->nsproxy);
330 
331 	if (flags & CLONE_NEWUSER)
332 		nsset->cred = prepare_creds();
333 	else
334 		nsset->cred = current_cred();
335 	if (!nsset->cred)
336 		goto out;
337 
338 	/* Only create a temporary copy of fs_struct if we really need to. */
339 	if (flags == CLONE_NEWNS) {
340 		nsset->fs = me->fs;
341 	} else if (flags & CLONE_NEWNS) {
342 		nsset->fs = copy_fs_struct(me->fs);
343 		if (!nsset->fs)
344 			goto out;
345 	}
346 
347 	nsset->flags = flags;
348 	return 0;
349 
350 out:
351 	put_nsset(nsset);
352 	return -ENOMEM;
353 }
354 
validate_ns(struct nsset * nsset,struct ns_common * ns)355 static inline int validate_ns(struct nsset *nsset, struct ns_common *ns)
356 {
357 	return ns->ops->install(nsset, ns);
358 }
359 
360 /*
361  * This is the inverse operation to unshare().
362  * Ordering is equivalent to the standard ordering used everywhere else
363  * during unshare and process creation. The switch to the new set of
364  * namespaces occurs at the point of no return after installation of
365  * all requested namespaces was successful in commit_nsset().
366  */
validate_nsset(struct nsset * nsset,struct pid * pid)367 static int validate_nsset(struct nsset *nsset, struct pid *pid)
368 {
369 	int ret = 0;
370 	unsigned flags = nsset->flags;
371 	struct user_namespace *user_ns = NULL;
372 	struct pid_namespace *pid_ns = NULL;
373 	struct nsproxy *nsp;
374 	struct task_struct *tsk;
375 
376 	/* Take a "snapshot" of the target task's namespaces. */
377 	rcu_read_lock();
378 	tsk = pid_task(pid, PIDTYPE_PID);
379 	if (!tsk) {
380 		rcu_read_unlock();
381 		return -ESRCH;
382 	}
383 
384 	if (!ptrace_may_access(tsk, PTRACE_MODE_READ_REALCREDS)) {
385 		rcu_read_unlock();
386 		return -EPERM;
387 	}
388 
389 	task_lock(tsk);
390 	nsp = tsk->nsproxy;
391 	if (nsp)
392 		get_nsproxy(nsp);
393 	task_unlock(tsk);
394 	if (!nsp) {
395 		rcu_read_unlock();
396 		return -ESRCH;
397 	}
398 
399 #ifdef CONFIG_PID_NS
400 	if (flags & CLONE_NEWPID) {
401 		pid_ns = task_active_pid_ns(tsk);
402 		if (unlikely(!pid_ns)) {
403 			rcu_read_unlock();
404 			ret = -ESRCH;
405 			goto out;
406 		}
407 		get_pid_ns(pid_ns);
408 	}
409 #endif
410 
411 #ifdef CONFIG_USER_NS
412 	if (flags & CLONE_NEWUSER)
413 		user_ns = get_user_ns(__task_cred(tsk)->user_ns);
414 #endif
415 	rcu_read_unlock();
416 
417 	/*
418 	 * Install requested namespaces. The caller will have
419 	 * verified earlier that the requested namespaces are
420 	 * supported on this kernel. We don't report errors here
421 	 * if a namespace is requested that isn't supported.
422 	 */
423 #ifdef CONFIG_USER_NS
424 	if (flags & CLONE_NEWUSER) {
425 		ret = validate_ns(nsset, &user_ns->ns);
426 		if (ret)
427 			goto out;
428 	}
429 #endif
430 
431 	if (flags & CLONE_NEWNS) {
432 		ret = validate_ns(nsset, from_mnt_ns(nsp->mnt_ns));
433 		if (ret)
434 			goto out;
435 	}
436 
437 #ifdef CONFIG_UTS_NS
438 	if (flags & CLONE_NEWUTS) {
439 		ret = validate_ns(nsset, &nsp->uts_ns->ns);
440 		if (ret)
441 			goto out;
442 	}
443 #endif
444 
445 #ifdef CONFIG_IPC_NS
446 	if (flags & CLONE_NEWIPC) {
447 		ret = validate_ns(nsset, &nsp->ipc_ns->ns);
448 		if (ret)
449 			goto out;
450 	}
451 #endif
452 
453 #ifdef CONFIG_PID_NS
454 	if (flags & CLONE_NEWPID) {
455 		ret = validate_ns(nsset, &pid_ns->ns);
456 		if (ret)
457 			goto out;
458 	}
459 #endif
460 
461 #ifdef CONFIG_CGROUPS
462 	if (flags & CLONE_NEWCGROUP) {
463 		ret = validate_ns(nsset, &nsp->cgroup_ns->ns);
464 		if (ret)
465 			goto out;
466 	}
467 #endif
468 
469 #ifdef CONFIG_NET_NS
470 	if (flags & CLONE_NEWNET) {
471 		ret = validate_ns(nsset, &nsp->net_ns->ns);
472 		if (ret)
473 			goto out;
474 	}
475 #endif
476 
477 #ifdef CONFIG_TIME_NS
478 	if (flags & CLONE_NEWTIME) {
479 		ret = validate_ns(nsset, &nsp->time_ns->ns);
480 		if (ret)
481 			goto out;
482 	}
483 #endif
484 
485 out:
486 	if (pid_ns)
487 		put_pid_ns(pid_ns);
488 	if (nsp)
489 		put_nsproxy(nsp);
490 	put_user_ns(user_ns);
491 
492 	return ret;
493 }
494 
495 /*
496  * This is the point of no return. There are just a few namespaces
497  * that do some actual work here and it's sufficiently minimal that
498  * a separate ns_common operation seems unnecessary for now.
499  * Unshare is doing the same thing. If we'll end up needing to do
500  * more in a given namespace or a helper here is ultimately not
501  * exported anymore a simple commit handler for each namespace
502  * should be added to ns_common.
503  */
commit_nsset(struct nsset * nsset)504 static void commit_nsset(struct nsset *nsset)
505 {
506 	unsigned flags = nsset->flags;
507 	struct task_struct *me = current;
508 
509 #ifdef CONFIG_USER_NS
510 	if (flags & CLONE_NEWUSER) {
511 		/* transfer ownership */
512 		commit_creds(nsset_cred(nsset));
513 		nsset->cred = NULL;
514 	}
515 #endif
516 
517 	/* We only need to commit if we have used a temporary fs_struct. */
518 	if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) {
519 		set_fs_root(me->fs, &nsset->fs->root);
520 		set_fs_pwd(me->fs, &nsset->fs->pwd);
521 	}
522 
523 #ifdef CONFIG_IPC_NS
524 	if (flags & CLONE_NEWIPC)
525 		exit_sem(me);
526 #endif
527 
528 #ifdef CONFIG_TIME_NS
529 	if (flags & CLONE_NEWTIME)
530 		timens_commit(me, nsset->nsproxy->time_ns);
531 #endif
532 
533 	/* transfer ownership */
534 	switch_task_namespaces(me, nsset->nsproxy);
535 	nsset->nsproxy = NULL;
536 }
537 
SYSCALL_DEFINE2(setns,int,fd,int,flags)538 SYSCALL_DEFINE2(setns, int, fd, int, flags)
539 {
540 	struct file *file;
541 	struct ns_common *ns = NULL;
542 	struct nsset nsset = {};
543 	int err = 0;
544 
545 	file = fget(fd);
546 	if (!file)
547 		return -EBADF;
548 
549 	if (proc_ns_file(file)) {
550 		ns = get_proc_ns(file_inode(file));
551 		if (flags && (ns->ops->type != flags))
552 			err = -EINVAL;
553 		flags = ns->ops->type;
554 	} else if (!IS_ERR(pidfd_pid(file))) {
555 		err = check_setns_flags(flags);
556 	} else {
557 		err = -EINVAL;
558 	}
559 	if (err)
560 		goto out;
561 
562 	err = prepare_nsset(flags, &nsset);
563 	if (err)
564 		goto out;
565 
566 	if (proc_ns_file(file))
567 		err = validate_ns(&nsset, ns);
568 	else
569 		err = validate_nsset(&nsset, file->private_data);
570 	if (!err) {
571 		commit_nsset(&nsset);
572 		perf_event_namespaces(current);
573 	}
574 	put_nsset(&nsset);
575 out:
576 	fput(file);
577 	return err;
578 }
579 
nsproxy_cache_init(void)580 int __init nsproxy_cache_init(void)
581 {
582 	nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC|SLAB_ACCOUNT);
583 	return 0;
584 }
585