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
create_nsproxy(void)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 */
create_new_namespaces(unsigned long flags,struct task_struct * tsk,struct user_namespace * user_ns,struct fs_struct * new_fs)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 */
copy_namespaces(unsigned long flags,struct task_struct * tsk)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
free_nsproxy(struct nsproxy * ns)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 */
unshare_nsproxy_namespaces(unsigned long unshare_flags,struct nsproxy ** new_nsp,struct cred * new_cred,struct fs_struct * new_fs)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
switch_task_namespaces(struct task_struct * p,struct nsproxy * new)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
exit_task_namespaces(struct task_struct * p)257 void exit_task_namespaces(struct task_struct *p)
258 {
259 switch_task_namespaces(p, NULL);
260 }
261
check_setns_flags(unsigned long flags)262 static int check_setns_flags(unsigned long flags)
263 {
264 if (!flags || (flags & ~(CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
265 CLONE_NEWNET | CLONE_NEWTIME | CLONE_NEWUSER |
266 CLONE_NEWPID | 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 #ifndef CONFIG_TIME_NS
294 if (flags & CLONE_NEWTIME)
295 return -EINVAL;
296 #endif
297
298 return 0;
299 }
300
put_nsset(struct nsset * nsset)301 static void put_nsset(struct nsset *nsset)
302 {
303 unsigned flags = nsset->flags;
304
305 if (flags & CLONE_NEWUSER)
306 put_cred(nsset_cred(nsset));
307 /*
308 * We only created a temporary copy if we attached to more than just
309 * the mount namespace.
310 */
311 if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS))
312 free_fs_struct(nsset->fs);
313 if (nsset->nsproxy)
314 free_nsproxy(nsset->nsproxy);
315 }
316
prepare_nsset(unsigned flags,struct nsset * nsset)317 static int prepare_nsset(unsigned flags, struct nsset *nsset)
318 {
319 struct task_struct *me = current;
320
321 nsset->nsproxy = create_new_namespaces(0, me, current_user_ns(), me->fs);
322 if (IS_ERR(nsset->nsproxy))
323 return PTR_ERR(nsset->nsproxy);
324
325 if (flags & CLONE_NEWUSER)
326 nsset->cred = prepare_creds();
327 else
328 nsset->cred = current_cred();
329 if (!nsset->cred)
330 goto out;
331
332 /* Only create a temporary copy of fs_struct if we really need to. */
333 if (flags == CLONE_NEWNS) {
334 nsset->fs = me->fs;
335 } else if (flags & CLONE_NEWNS) {
336 nsset->fs = copy_fs_struct(me->fs);
337 if (!nsset->fs)
338 goto out;
339 }
340
341 nsset->flags = flags;
342 return 0;
343
344 out:
345 put_nsset(nsset);
346 return -ENOMEM;
347 }
348
validate_ns(struct nsset * nsset,struct ns_common * ns)349 static inline int validate_ns(struct nsset *nsset, struct ns_common *ns)
350 {
351 return ns->ops->install(nsset, ns);
352 }
353
354 /*
355 * This is the inverse operation to unshare().
356 * Ordering is equivalent to the standard ordering used everywhere else
357 * during unshare and process creation. The switch to the new set of
358 * namespaces occurs at the point of no return after installation of
359 * all requested namespaces was successful in commit_nsset().
360 */
validate_nsset(struct nsset * nsset,struct pid * pid)361 static int validate_nsset(struct nsset *nsset, struct pid *pid)
362 {
363 int ret = 0;
364 unsigned flags = nsset->flags;
365 struct user_namespace *user_ns = NULL;
366 struct pid_namespace *pid_ns = NULL;
367 struct nsproxy *nsp;
368 struct task_struct *tsk;
369
370 /* Take a "snapshot" of the target task's namespaces. */
371 rcu_read_lock();
372 tsk = pid_task(pid, PIDTYPE_PID);
373 if (!tsk) {
374 rcu_read_unlock();
375 return -ESRCH;
376 }
377
378 if (!ptrace_may_access(tsk, PTRACE_MODE_READ_REALCREDS)) {
379 rcu_read_unlock();
380 return -EPERM;
381 }
382
383 task_lock(tsk);
384 nsp = tsk->nsproxy;
385 if (nsp)
386 get_nsproxy(nsp);
387 task_unlock(tsk);
388 if (!nsp) {
389 rcu_read_unlock();
390 return -ESRCH;
391 }
392
393 #ifdef CONFIG_PID_NS
394 if (flags & CLONE_NEWPID) {
395 pid_ns = task_active_pid_ns(tsk);
396 if (unlikely(!pid_ns)) {
397 rcu_read_unlock();
398 ret = -ESRCH;
399 goto out;
400 }
401 get_pid_ns(pid_ns);
402 }
403 #endif
404
405 #ifdef CONFIG_USER_NS
406 if (flags & CLONE_NEWUSER)
407 user_ns = get_user_ns(__task_cred(tsk)->user_ns);
408 #endif
409 rcu_read_unlock();
410
411 /*
412 * Install requested namespaces. The caller will have
413 * verified earlier that the requested namespaces are
414 * supported on this kernel. We don't report errors here
415 * if a namespace is requested that isn't supported.
416 */
417 #ifdef CONFIG_USER_NS
418 if (flags & CLONE_NEWUSER) {
419 ret = validate_ns(nsset, &user_ns->ns);
420 if (ret)
421 goto out;
422 }
423 #endif
424
425 if (flags & CLONE_NEWNS) {
426 ret = validate_ns(nsset, from_mnt_ns(nsp->mnt_ns));
427 if (ret)
428 goto out;
429 }
430
431 #ifdef CONFIG_UTS_NS
432 if (flags & CLONE_NEWUTS) {
433 ret = validate_ns(nsset, &nsp->uts_ns->ns);
434 if (ret)
435 goto out;
436 }
437 #endif
438
439 #ifdef CONFIG_IPC_NS
440 if (flags & CLONE_NEWIPC) {
441 ret = validate_ns(nsset, &nsp->ipc_ns->ns);
442 if (ret)
443 goto out;
444 }
445 #endif
446
447 #ifdef CONFIG_PID_NS
448 if (flags & CLONE_NEWPID) {
449 ret = validate_ns(nsset, &pid_ns->ns);
450 if (ret)
451 goto out;
452 }
453 #endif
454
455 #ifdef CONFIG_CGROUPS
456 if (flags & CLONE_NEWCGROUP) {
457 ret = validate_ns(nsset, &nsp->cgroup_ns->ns);
458 if (ret)
459 goto out;
460 }
461 #endif
462
463 #ifdef CONFIG_NET_NS
464 if (flags & CLONE_NEWNET) {
465 ret = validate_ns(nsset, &nsp->net_ns->ns);
466 if (ret)
467 goto out;
468 }
469 #endif
470
471 #ifdef CONFIG_TIME_NS
472 if (flags & CLONE_NEWTIME) {
473 ret = validate_ns(nsset, &nsp->time_ns->ns);
474 if (ret)
475 goto out;
476 }
477 #endif
478
479 out:
480 if (pid_ns)
481 put_pid_ns(pid_ns);
482 if (nsp)
483 put_nsproxy(nsp);
484 put_user_ns(user_ns);
485
486 return ret;
487 }
488
489 /*
490 * This is the point of no return. There are just a few namespaces
491 * that do some actual work here and it's sufficiently minimal that
492 * a separate ns_common operation seems unnecessary for now.
493 * Unshare is doing the same thing. If we'll end up needing to do
494 * more in a given namespace or a helper here is ultimately not
495 * exported anymore a simple commit handler for each namespace
496 * should be added to ns_common.
497 */
commit_nsset(struct nsset * nsset)498 static void commit_nsset(struct nsset *nsset)
499 {
500 unsigned flags = nsset->flags;
501 struct task_struct *me = current;
502
503 #ifdef CONFIG_USER_NS
504 if (flags & CLONE_NEWUSER) {
505 /* transfer ownership */
506 commit_creds(nsset_cred(nsset));
507 nsset->cred = NULL;
508 }
509 #endif
510
511 /* We only need to commit if we have used a temporary fs_struct. */
512 if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) {
513 set_fs_root(me->fs, &nsset->fs->root);
514 set_fs_pwd(me->fs, &nsset->fs->pwd);
515 }
516
517 #ifdef CONFIG_IPC_NS
518 if (flags & CLONE_NEWIPC)
519 exit_sem(me);
520 #endif
521
522 #ifdef CONFIG_TIME_NS
523 if (flags & CLONE_NEWTIME)
524 timens_commit(me, nsset->nsproxy->time_ns);
525 #endif
526
527 /* transfer ownership */
528 switch_task_namespaces(me, nsset->nsproxy);
529 nsset->nsproxy = NULL;
530 }
531
SYSCALL_DEFINE2(setns,int,fd,int,flags)532 SYSCALL_DEFINE2(setns, int, fd, int, flags)
533 {
534 struct file *file;
535 struct ns_common *ns = NULL;
536 struct nsset nsset = {};
537 int err = 0;
538
539 file = fget(fd);
540 if (!file)
541 return -EBADF;
542
543 if (proc_ns_file(file)) {
544 ns = get_proc_ns(file_inode(file));
545 if (flags && (ns->ops->type != flags))
546 err = -EINVAL;
547 flags = ns->ops->type;
548 } else if (!IS_ERR(pidfd_pid(file))) {
549 err = check_setns_flags(flags);
550 } else {
551 err = -EINVAL;
552 }
553 if (err)
554 goto out;
555
556 err = prepare_nsset(flags, &nsset);
557 if (err)
558 goto out;
559
560 if (proc_ns_file(file))
561 err = validate_ns(&nsset, ns);
562 else
563 err = validate_nsset(&nsset, file->private_data);
564 if (!err) {
565 commit_nsset(&nsset);
566 perf_event_namespaces(current);
567 }
568 put_nsset(&nsset);
569 out:
570 fput(file);
571 return err;
572 }
573
nsproxy_cache_init(void)574 int __init nsproxy_cache_init(void)
575 {
576 nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC|SLAB_ACCOUNT);
577 return 0;
578 }
579