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/nstree.h>
30
31 static struct kmem_cache *nsproxy_cachep;
32
33 struct nsproxy init_nsproxy = {
34 .count = REFCOUNT_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 refcount_set(&nsproxy->count, 1);
60 return nsproxy;
61 }
62
nsproxy_free(struct nsproxy * ns)63 static inline void nsproxy_free(struct nsproxy *ns)
64 {
65 put_mnt_ns(ns->mnt_ns);
66 put_uts_ns(ns->uts_ns);
67 put_ipc_ns(ns->ipc_ns);
68 put_pid_ns(ns->pid_ns_for_children);
69 put_time_ns(ns->time_ns);
70 put_time_ns(ns->time_ns_for_children);
71 put_cgroup_ns(ns->cgroup_ns);
72 put_net(ns->net_ns);
73 kmem_cache_free(nsproxy_cachep, ns);
74 }
75
deactivate_nsproxy(struct nsproxy * ns)76 void deactivate_nsproxy(struct nsproxy *ns)
77 {
78 nsproxy_ns_active_put(ns);
79 nsproxy_free(ns);
80 }
81
82 /*
83 * Create new nsproxy and all of its the associated namespaces.
84 * Return the newly created nsproxy. Do not attach this to the task,
85 * leave it to the caller to do proper locking and attach it to task.
86 */
create_new_namespaces(u64 flags,struct task_struct * tsk,struct user_namespace * user_ns,struct fs_struct * new_fs)87 static struct nsproxy *create_new_namespaces(u64 flags,
88 struct task_struct *tsk, struct user_namespace *user_ns,
89 struct fs_struct *new_fs)
90 {
91 struct nsproxy *new_nsp;
92 int err;
93
94 new_nsp = create_nsproxy();
95 if (!new_nsp)
96 return ERR_PTR(-ENOMEM);
97
98 new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs);
99 if (IS_ERR(new_nsp->mnt_ns)) {
100 err = PTR_ERR(new_nsp->mnt_ns);
101 goto out_ns;
102 }
103
104 new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
105 if (IS_ERR(new_nsp->uts_ns)) {
106 err = PTR_ERR(new_nsp->uts_ns);
107 goto out_uts;
108 }
109
110 new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
111 if (IS_ERR(new_nsp->ipc_ns)) {
112 err = PTR_ERR(new_nsp->ipc_ns);
113 goto out_ipc;
114 }
115
116 new_nsp->pid_ns_for_children =
117 copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
118 if (IS_ERR(new_nsp->pid_ns_for_children)) {
119 err = PTR_ERR(new_nsp->pid_ns_for_children);
120 goto out_pid;
121 }
122
123 new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns,
124 tsk->nsproxy->cgroup_ns);
125 if (IS_ERR(new_nsp->cgroup_ns)) {
126 err = PTR_ERR(new_nsp->cgroup_ns);
127 goto out_cgroup;
128 }
129
130 new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
131 if (IS_ERR(new_nsp->net_ns)) {
132 err = PTR_ERR(new_nsp->net_ns);
133 goto out_net;
134 }
135
136 new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns,
137 tsk->nsproxy->time_ns_for_children);
138 if (IS_ERR(new_nsp->time_ns_for_children)) {
139 err = PTR_ERR(new_nsp->time_ns_for_children);
140 goto out_time;
141 }
142 new_nsp->time_ns = get_time_ns(tsk->nsproxy->time_ns);
143
144 return new_nsp;
145
146 out_time:
147 put_net(new_nsp->net_ns);
148 out_net:
149 put_cgroup_ns(new_nsp->cgroup_ns);
150 out_cgroup:
151 put_pid_ns(new_nsp->pid_ns_for_children);
152 out_pid:
153 put_ipc_ns(new_nsp->ipc_ns);
154 out_ipc:
155 put_uts_ns(new_nsp->uts_ns);
156 out_uts:
157 put_mnt_ns(new_nsp->mnt_ns);
158 out_ns:
159 kmem_cache_free(nsproxy_cachep, new_nsp);
160 return ERR_PTR(err);
161 }
162
163 /*
164 * called from clone. This now handles copy for nsproxy and all
165 * namespaces therein.
166 */
copy_namespaces(u64 flags,struct task_struct * tsk)167 int copy_namespaces(u64 flags, struct task_struct *tsk)
168 {
169 struct nsproxy *old_ns = tsk->nsproxy;
170 struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
171 struct nsproxy *new_ns;
172
173 if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
174 CLONE_NEWPID | CLONE_NEWNET |
175 CLONE_NEWCGROUP | CLONE_NEWTIME)))) {
176 if ((flags & CLONE_VM) ||
177 likely(old_ns->time_ns_for_children == old_ns->time_ns)) {
178 get_nsproxy(old_ns);
179 return 0;
180 }
181 } else if (!ns_capable(user_ns, CAP_SYS_ADMIN))
182 return -EPERM;
183
184 /*
185 * CLONE_NEWIPC must detach from the undolist: after switching
186 * to a new ipc namespace, the semaphore arrays from the old
187 * namespace are unreachable. In clone parlance, CLONE_SYSVSEM
188 * means share undolist with parent, so we must forbid using
189 * it along with CLONE_NEWIPC.
190 */
191 if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
192 (CLONE_NEWIPC | CLONE_SYSVSEM))
193 return -EINVAL;
194
195 new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
196 if (IS_ERR(new_ns))
197 return PTR_ERR(new_ns);
198
199 if ((flags & CLONE_VM) == 0)
200 timens_on_fork(new_ns, tsk);
201
202 nsproxy_ns_active_get(new_ns);
203 tsk->nsproxy = new_ns;
204 return 0;
205 }
206
207 /*
208 * Called from unshare. Unshare all the namespaces part of nsproxy.
209 * On success, returns the new nsproxy.
210 */
unshare_nsproxy_namespaces(unsigned long unshare_flags,struct nsproxy ** new_nsp,struct cred * new_cred,struct fs_struct * new_fs)211 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
212 struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
213 {
214 struct user_namespace *user_ns;
215 int err = 0;
216
217 if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
218 CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP |
219 CLONE_NEWTIME)))
220 return 0;
221
222 user_ns = new_cred ? new_cred->user_ns : current_user_ns();
223 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
224 return -EPERM;
225
226 *new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
227 new_fs ? new_fs : current->fs);
228 if (IS_ERR(*new_nsp)) {
229 err = PTR_ERR(*new_nsp);
230 goto out;
231 }
232
233 out:
234 return err;
235 }
236
switch_task_namespaces(struct task_struct * p,struct nsproxy * new)237 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
238 {
239 struct nsproxy *ns;
240
241 might_sleep();
242
243 if (new)
244 nsproxy_ns_active_get(new);
245
246 task_lock(p);
247 ns = p->nsproxy;
248 p->nsproxy = new;
249 task_unlock(p);
250
251 if (ns)
252 put_nsproxy(ns);
253 }
254
exit_nsproxy_namespaces(struct task_struct * p)255 void exit_nsproxy_namespaces(struct task_struct *p)
256 {
257 switch_task_namespaces(p, NULL);
258 }
259
switch_cred_namespaces(const struct cred * old,const struct cred * new)260 void switch_cred_namespaces(const struct cred *old, const struct cred *new)
261 {
262 ns_ref_active_get(new->user_ns);
263 ns_ref_active_put(old->user_ns);
264 }
265
get_cred_namespaces(struct task_struct * tsk)266 void get_cred_namespaces(struct task_struct *tsk)
267 {
268 ns_ref_active_get(tsk->real_cred->user_ns);
269 }
270
exit_cred_namespaces(struct task_struct * tsk)271 void exit_cred_namespaces(struct task_struct *tsk)
272 {
273 ns_ref_active_put(tsk->real_cred->user_ns);
274 }
275
exec_task_namespaces(void)276 int exec_task_namespaces(void)
277 {
278 struct task_struct *tsk = current;
279 struct nsproxy *new;
280
281 if (tsk->nsproxy->time_ns_for_children == tsk->nsproxy->time_ns)
282 return 0;
283
284 new = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
285 if (IS_ERR(new))
286 return PTR_ERR(new);
287
288 timens_on_fork(new, tsk);
289 switch_task_namespaces(tsk, new);
290 return 0;
291 }
292
check_setns_flags(unsigned long flags)293 static int check_setns_flags(unsigned long flags)
294 {
295 if (!flags || (flags & ~(CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
296 CLONE_NEWNET | CLONE_NEWTIME | CLONE_NEWUSER |
297 CLONE_NEWPID | CLONE_NEWCGROUP)))
298 return -EINVAL;
299
300 #ifndef CONFIG_USER_NS
301 if (flags & CLONE_NEWUSER)
302 return -EINVAL;
303 #endif
304 #ifndef CONFIG_PID_NS
305 if (flags & CLONE_NEWPID)
306 return -EINVAL;
307 #endif
308 #ifndef CONFIG_UTS_NS
309 if (flags & CLONE_NEWUTS)
310 return -EINVAL;
311 #endif
312 #ifndef CONFIG_IPC_NS
313 if (flags & CLONE_NEWIPC)
314 return -EINVAL;
315 #endif
316 #ifndef CONFIG_CGROUPS
317 if (flags & CLONE_NEWCGROUP)
318 return -EINVAL;
319 #endif
320 #ifndef CONFIG_NET_NS
321 if (flags & CLONE_NEWNET)
322 return -EINVAL;
323 #endif
324 #ifndef CONFIG_TIME_NS
325 if (flags & CLONE_NEWTIME)
326 return -EINVAL;
327 #endif
328
329 return 0;
330 }
331
put_nsset(struct nsset * nsset)332 static void put_nsset(struct nsset *nsset)
333 {
334 unsigned flags = nsset->flags;
335
336 if (flags & CLONE_NEWUSER)
337 put_cred(nsset_cred(nsset));
338 /*
339 * We only created a temporary copy if we attached to more than just
340 * the mount namespace.
341 */
342 if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS))
343 free_fs_struct(nsset->fs);
344 if (nsset->nsproxy)
345 nsproxy_free(nsset->nsproxy);
346 }
347
prepare_nsset(unsigned flags,struct nsset * nsset)348 static int prepare_nsset(unsigned flags, struct nsset *nsset)
349 {
350 struct task_struct *me = current;
351
352 nsset->nsproxy = create_new_namespaces(0, me, current_user_ns(), me->fs);
353 if (IS_ERR(nsset->nsproxy))
354 return PTR_ERR(nsset->nsproxy);
355
356 if (flags & CLONE_NEWUSER)
357 nsset->cred = prepare_creds();
358 else
359 nsset->cred = current_cred();
360 if (!nsset->cred)
361 goto out;
362
363 /* Only create a temporary copy of fs_struct if we really need to. */
364 if (flags == CLONE_NEWNS) {
365 nsset->fs = me->fs;
366 } else if (flags & CLONE_NEWNS) {
367 nsset->fs = copy_fs_struct(me->fs);
368 if (!nsset->fs)
369 goto out;
370 }
371
372 nsset->flags = flags;
373 return 0;
374
375 out:
376 put_nsset(nsset);
377 return -ENOMEM;
378 }
379
validate_ns(struct nsset * nsset,struct ns_common * ns)380 static inline int validate_ns(struct nsset *nsset, struct ns_common *ns)
381 {
382 return ns->ops->install(nsset, ns);
383 }
384
385 /*
386 * This is the inverse operation to unshare().
387 * Ordering is equivalent to the standard ordering used everywhere else
388 * during unshare and process creation. The switch to the new set of
389 * namespaces occurs at the point of no return after installation of
390 * all requested namespaces was successful in commit_nsset().
391 */
validate_nsset(struct nsset * nsset,struct pid * pid)392 static int validate_nsset(struct nsset *nsset, struct pid *pid)
393 {
394 int ret = 0;
395 unsigned flags = nsset->flags;
396 struct user_namespace *user_ns = NULL;
397 struct pid_namespace *pid_ns = NULL;
398 struct nsproxy *nsp;
399 struct task_struct *tsk;
400
401 /* Take a "snapshot" of the target task's namespaces. */
402 rcu_read_lock();
403 tsk = pid_task(pid, PIDTYPE_PID);
404 if (!tsk) {
405 rcu_read_unlock();
406 return -ESRCH;
407 }
408
409 if (!ptrace_may_access(tsk, PTRACE_MODE_READ_REALCREDS)) {
410 rcu_read_unlock();
411 return -EPERM;
412 }
413
414 task_lock(tsk);
415 nsp = tsk->nsproxy;
416 if (nsp)
417 get_nsproxy(nsp);
418 task_unlock(tsk);
419 if (!nsp) {
420 rcu_read_unlock();
421 return -ESRCH;
422 }
423
424 #ifdef CONFIG_PID_NS
425 if (flags & CLONE_NEWPID) {
426 pid_ns = task_active_pid_ns(tsk);
427 if (unlikely(!pid_ns)) {
428 rcu_read_unlock();
429 ret = -ESRCH;
430 goto out;
431 }
432 get_pid_ns(pid_ns);
433 }
434 #endif
435
436 #ifdef CONFIG_USER_NS
437 if (flags & CLONE_NEWUSER)
438 user_ns = get_user_ns(__task_cred(tsk)->user_ns);
439 #endif
440 rcu_read_unlock();
441
442 /*
443 * Install requested namespaces. The caller will have
444 * verified earlier that the requested namespaces are
445 * supported on this kernel. We don't report errors here
446 * if a namespace is requested that isn't supported.
447 */
448 #ifdef CONFIG_USER_NS
449 if (flags & CLONE_NEWUSER) {
450 ret = validate_ns(nsset, &user_ns->ns);
451 if (ret)
452 goto out;
453 }
454 #endif
455
456 if (flags & CLONE_NEWNS) {
457 ret = validate_ns(nsset, from_mnt_ns(nsp->mnt_ns));
458 if (ret)
459 goto out;
460 }
461
462 #ifdef CONFIG_UTS_NS
463 if (flags & CLONE_NEWUTS) {
464 ret = validate_ns(nsset, &nsp->uts_ns->ns);
465 if (ret)
466 goto out;
467 }
468 #endif
469
470 #ifdef CONFIG_IPC_NS
471 if (flags & CLONE_NEWIPC) {
472 ret = validate_ns(nsset, &nsp->ipc_ns->ns);
473 if (ret)
474 goto out;
475 }
476 #endif
477
478 #ifdef CONFIG_PID_NS
479 if (flags & CLONE_NEWPID) {
480 ret = validate_ns(nsset, &pid_ns->ns);
481 if (ret)
482 goto out;
483 }
484 #endif
485
486 #ifdef CONFIG_CGROUPS
487 if (flags & CLONE_NEWCGROUP) {
488 ret = validate_ns(nsset, &nsp->cgroup_ns->ns);
489 if (ret)
490 goto out;
491 }
492 #endif
493
494 #ifdef CONFIG_NET_NS
495 if (flags & CLONE_NEWNET) {
496 ret = validate_ns(nsset, &nsp->net_ns->ns);
497 if (ret)
498 goto out;
499 }
500 #endif
501
502 #ifdef CONFIG_TIME_NS
503 if (flags & CLONE_NEWTIME) {
504 ret = validate_ns(nsset, &nsp->time_ns->ns);
505 if (ret)
506 goto out;
507 }
508 #endif
509
510 out:
511 if (pid_ns)
512 put_pid_ns(pid_ns);
513 if (nsp)
514 put_nsproxy(nsp);
515 put_user_ns(user_ns);
516
517 return ret;
518 }
519
520 /*
521 * This is the point of no return. There are just a few namespaces
522 * that do some actual work here and it's sufficiently minimal that
523 * a separate ns_common operation seems unnecessary for now.
524 * Unshare is doing the same thing. If we'll end up needing to do
525 * more in a given namespace or a helper here is ultimately not
526 * exported anymore a simple commit handler for each namespace
527 * should be added to ns_common.
528 */
commit_nsset(struct nsset * nsset)529 static void commit_nsset(struct nsset *nsset)
530 {
531 unsigned flags = nsset->flags;
532 struct task_struct *me = current;
533
534 #ifdef CONFIG_USER_NS
535 if (flags & CLONE_NEWUSER) {
536 /* transfer ownership */
537 commit_creds(nsset_cred(nsset));
538 nsset->cred = NULL;
539 }
540 #endif
541
542 /* We only need to commit if we have used a temporary fs_struct. */
543 if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) {
544 set_fs_root(me->fs, &nsset->fs->root);
545 set_fs_pwd(me->fs, &nsset->fs->pwd);
546 }
547
548 #ifdef CONFIG_IPC_NS
549 if (flags & CLONE_NEWIPC)
550 exit_sem(me);
551 #endif
552
553 #ifdef CONFIG_TIME_NS
554 if (flags & CLONE_NEWTIME)
555 timens_commit(me, nsset->nsproxy->time_ns);
556 #endif
557
558 /* transfer ownership */
559 switch_task_namespaces(me, nsset->nsproxy);
560 nsset->nsproxy = NULL;
561 }
562
SYSCALL_DEFINE2(setns,int,fd,int,flags)563 SYSCALL_DEFINE2(setns, int, fd, int, flags)
564 {
565 CLASS(fd, f)(fd);
566 struct ns_common *ns = NULL;
567 struct nsset nsset = {};
568 int err = 0;
569
570 if (fd_empty(f))
571 return -EBADF;
572
573 if (proc_ns_file(fd_file(f))) {
574 ns = get_proc_ns(file_inode(fd_file(f)));
575 if (flags && (ns->ns_type != flags))
576 err = -EINVAL;
577 flags = ns->ns_type;
578 } else if (!IS_ERR(pidfd_pid(fd_file(f)))) {
579 err = check_setns_flags(flags);
580 } else {
581 err = -EINVAL;
582 }
583 if (err)
584 goto out;
585
586 err = prepare_nsset(flags, &nsset);
587 if (err)
588 goto out;
589
590 if (proc_ns_file(fd_file(f)))
591 err = validate_ns(&nsset, ns);
592 else
593 err = validate_nsset(&nsset, pidfd_pid(fd_file(f)));
594 if (!err) {
595 commit_nsset(&nsset);
596 perf_event_namespaces(current);
597 }
598 put_nsset(&nsset);
599 out:
600 return err;
601 }
602
nsproxy_cache_init(void)603 int __init nsproxy_cache_init(void)
604 {
605 nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC|SLAB_ACCOUNT);
606 return 0;
607 }
608