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/ns/ns_common_types.h>
16 #include <linux/init_task.h>
17 #include <linux/mnt_namespace.h>
18 #include <linux/utsname.h>
19 #include <linux/pid_namespace.h>
20 #include <net/net_namespace.h>
21 #include <linux/ipc_namespace.h>
22 #include <linux/time_namespace.h>
23 #include <linux/fs_struct.h>
24 #include <linux/proc_fs.h>
25 #include <linux/proc_ns.h>
26 #include <linux/file.h>
27 #include <linux/syscalls.h>
28 #include <linux/cgroup.h>
29 #include <linux/perf_event.h>
30 #include <linux/nstree.h>
31
32 static struct kmem_cache *nsproxy_cachep;
33
34 struct nsproxy init_nsproxy = {
35 .count = REFCOUNT_INIT(1),
36 .uts_ns = &init_uts_ns,
37 #if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
38 .ipc_ns = &init_ipc_ns,
39 #endif
40 .mnt_ns = NULL,
41 .pid_ns_for_children = &init_pid_ns,
42 #ifdef CONFIG_NET
43 .net_ns = &init_net,
44 #endif
45 #ifdef CONFIG_CGROUPS
46 .cgroup_ns = &init_cgroup_ns,
47 #endif
48 #ifdef CONFIG_TIME_NS
49 .time_ns = &init_time_ns,
50 .time_ns_for_children = &init_time_ns,
51 #endif
52 };
53
create_nsproxy(void)54 static inline struct nsproxy *create_nsproxy(void)
55 {
56 struct nsproxy *nsproxy;
57
58 nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
59 if (nsproxy)
60 refcount_set(&nsproxy->count, 1);
61 return nsproxy;
62 }
63
nsproxy_free(struct nsproxy * ns)64 static inline void nsproxy_free(struct nsproxy *ns)
65 {
66 put_mnt_ns(ns->mnt_ns);
67 put_uts_ns(ns->uts_ns);
68 put_ipc_ns(ns->ipc_ns);
69 put_pid_ns(ns->pid_ns_for_children);
70 put_time_ns(ns->time_ns);
71 put_time_ns(ns->time_ns_for_children);
72 put_cgroup_ns(ns->cgroup_ns);
73 put_net(ns->net_ns);
74 kmem_cache_free(nsproxy_cachep, ns);
75 }
76
deactivate_nsproxy(struct nsproxy * ns)77 void deactivate_nsproxy(struct nsproxy *ns)
78 {
79 nsproxy_ns_active_put(ns);
80 nsproxy_free(ns);
81 }
82
83 /*
84 * Create new nsproxy and all of its the associated namespaces.
85 * Return the newly created nsproxy. Do not attach this to the task,
86 * leave it to the caller to do proper locking and attach it to task.
87 */
create_new_namespaces(u64 flags,struct task_struct * tsk,struct user_namespace * user_ns,struct fs_struct * new_fs)88 static struct nsproxy *create_new_namespaces(u64 flags,
89 struct task_struct *tsk, struct user_namespace *user_ns,
90 struct fs_struct *new_fs)
91 {
92 struct nsproxy *new_nsp;
93 int err;
94
95 new_nsp = create_nsproxy();
96 if (!new_nsp)
97 return ERR_PTR(-ENOMEM);
98
99 new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns,
100 user_ns, new_fs);
101 if (IS_ERR(new_nsp->mnt_ns)) {
102 err = PTR_ERR(new_nsp->mnt_ns);
103 goto out_ns;
104 }
105
106 new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
107 if (IS_ERR(new_nsp->uts_ns)) {
108 err = PTR_ERR(new_nsp->uts_ns);
109 goto out_uts;
110 }
111
112 new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
113 if (IS_ERR(new_nsp->ipc_ns)) {
114 err = PTR_ERR(new_nsp->ipc_ns);
115 goto out_ipc;
116 }
117
118 new_nsp->pid_ns_for_children =
119 copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
120 if (IS_ERR(new_nsp->pid_ns_for_children)) {
121 err = PTR_ERR(new_nsp->pid_ns_for_children);
122 goto out_pid;
123 }
124
125 new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns,
126 tsk->nsproxy->cgroup_ns);
127 if (IS_ERR(new_nsp->cgroup_ns)) {
128 err = PTR_ERR(new_nsp->cgroup_ns);
129 goto out_cgroup;
130 }
131
132 new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
133 if (IS_ERR(new_nsp->net_ns)) {
134 err = PTR_ERR(new_nsp->net_ns);
135 goto out_net;
136 }
137
138 new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns,
139 tsk->nsproxy->time_ns_for_children);
140 if (IS_ERR(new_nsp->time_ns_for_children)) {
141 err = PTR_ERR(new_nsp->time_ns_for_children);
142 goto out_time;
143 }
144 new_nsp->time_ns = get_time_ns(tsk->nsproxy->time_ns);
145
146 return new_nsp;
147
148 out_time:
149 put_net(new_nsp->net_ns);
150 out_net:
151 put_cgroup_ns(new_nsp->cgroup_ns);
152 out_cgroup:
153 put_pid_ns(new_nsp->pid_ns_for_children);
154 out_pid:
155 put_ipc_ns(new_nsp->ipc_ns);
156 out_ipc:
157 put_uts_ns(new_nsp->uts_ns);
158 out_uts:
159 put_mnt_ns(new_nsp->mnt_ns);
160 out_ns:
161 kmem_cache_free(nsproxy_cachep, new_nsp);
162 return ERR_PTR(err);
163 }
164
165 /*
166 * called from clone. This now handles copy for nsproxy and all
167 * namespaces therein.
168 */
copy_namespaces(u64 flags,struct task_struct * tsk)169 int copy_namespaces(u64 flags, struct task_struct *tsk)
170 {
171 struct nsproxy *old_ns = tsk->nsproxy;
172 struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
173 struct nsproxy *new_ns;
174
175 if (likely(!(flags & (CLONE_NS_ALL & ~CLONE_NEWUSER)))) {
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 u64 flags = unshare_flags;
216 int err = 0;
217
218 if (!(flags & (CLONE_NS_ALL & ~CLONE_NEWUSER)))
219 return 0;
220
221 user_ns = new_cred ? new_cred->user_ns : current_user_ns();
222 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
223 return -EPERM;
224
225 /*
226 * Convert the 32-bit UNSHARE_EMPTY_MNTNS (which aliases
227 * CLONE_PARENT_SETTID) to the unique 64-bit CLONE_EMPTY_MNTNS.
228 */
229 if (flags & UNSHARE_EMPTY_MNTNS) {
230 flags &= ~(u64)UNSHARE_EMPTY_MNTNS;
231 flags |= CLONE_EMPTY_MNTNS;
232 }
233
234 *new_nsp = create_new_namespaces(flags, current, user_ns,
235 new_fs ? new_fs : current->fs);
236 if (IS_ERR(*new_nsp)) {
237 err = PTR_ERR(*new_nsp);
238 goto out;
239 }
240
241 out:
242 return err;
243 }
244
switch_task_namespaces(struct task_struct * p,struct nsproxy * new)245 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
246 {
247 struct nsproxy *ns;
248
249 might_sleep();
250
251 if (new)
252 nsproxy_ns_active_get(new);
253
254 task_lock(p);
255 ns = p->nsproxy;
256 p->nsproxy = new;
257 task_unlock(p);
258
259 if (ns)
260 put_nsproxy(ns);
261 }
262
exit_nsproxy_namespaces(struct task_struct * p)263 void exit_nsproxy_namespaces(struct task_struct *p)
264 {
265 switch_task_namespaces(p, NULL);
266 }
267
switch_cred_namespaces(const struct cred * old,const struct cred * new)268 void switch_cred_namespaces(const struct cred *old, const struct cred *new)
269 {
270 ns_ref_active_get(new->user_ns);
271 ns_ref_active_put(old->user_ns);
272 }
273
get_cred_namespaces(struct task_struct * tsk)274 void get_cred_namespaces(struct task_struct *tsk)
275 {
276 ns_ref_active_get(tsk->real_cred->user_ns);
277 }
278
exit_cred_namespaces(struct task_struct * tsk)279 void exit_cred_namespaces(struct task_struct *tsk)
280 {
281 ns_ref_active_put(tsk->real_cred->user_ns);
282 }
283
exec_task_namespaces(void)284 int exec_task_namespaces(void)
285 {
286 struct task_struct *tsk = current;
287 struct nsproxy *new;
288
289 if (tsk->nsproxy->time_ns_for_children == tsk->nsproxy->time_ns)
290 return 0;
291
292 new = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
293 if (IS_ERR(new))
294 return PTR_ERR(new);
295
296 timens_on_fork(new, tsk);
297 switch_task_namespaces(tsk, new);
298 return 0;
299 }
300
check_setns_flags(unsigned long flags)301 static int check_setns_flags(unsigned long flags)
302 {
303 if (!flags || (flags & ~CLONE_NS_ALL))
304 return -EINVAL;
305
306 #ifndef CONFIG_USER_NS
307 if (flags & CLONE_NEWUSER)
308 return -EINVAL;
309 #endif
310 #ifndef CONFIG_PID_NS
311 if (flags & CLONE_NEWPID)
312 return -EINVAL;
313 #endif
314 #ifndef CONFIG_UTS_NS
315 if (flags & CLONE_NEWUTS)
316 return -EINVAL;
317 #endif
318 #ifndef CONFIG_IPC_NS
319 if (flags & CLONE_NEWIPC)
320 return -EINVAL;
321 #endif
322 #ifndef CONFIG_CGROUPS
323 if (flags & CLONE_NEWCGROUP)
324 return -EINVAL;
325 #endif
326 #ifndef CONFIG_NET_NS
327 if (flags & CLONE_NEWNET)
328 return -EINVAL;
329 #endif
330 #ifndef CONFIG_TIME_NS
331 if (flags & CLONE_NEWTIME)
332 return -EINVAL;
333 #endif
334
335 return 0;
336 }
337
put_nsset(struct nsset * nsset)338 static void put_nsset(struct nsset *nsset)
339 {
340 unsigned flags = nsset->flags;
341
342 if (flags & CLONE_NEWUSER)
343 put_cred(nsset_cred(nsset));
344 /*
345 * We only created a temporary copy if we attached to more than just
346 * the mount namespace.
347 */
348 if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS))
349 free_fs_struct(nsset->fs);
350 if (nsset->nsproxy)
351 nsproxy_free(nsset->nsproxy);
352 }
353
prepare_nsset(unsigned flags,struct nsset * nsset)354 static int prepare_nsset(unsigned flags, struct nsset *nsset)
355 {
356 struct task_struct *me = current;
357
358 nsset->nsproxy = create_new_namespaces(0, me, current_user_ns(), me->fs);
359 if (IS_ERR(nsset->nsproxy))
360 return PTR_ERR(nsset->nsproxy);
361
362 if (flags & CLONE_NEWUSER)
363 nsset->cred = prepare_creds();
364 else
365 nsset->cred = current_cred();
366 if (!nsset->cred)
367 goto out;
368
369 /* Only create a temporary copy of fs_struct if we really need to. */
370 if (flags == CLONE_NEWNS) {
371 nsset->fs = me->fs;
372 } else if (flags & CLONE_NEWNS) {
373 nsset->fs = copy_fs_struct(me->fs);
374 if (!nsset->fs)
375 goto out;
376 }
377
378 nsset->flags = flags;
379 return 0;
380
381 out:
382 put_nsset(nsset);
383 return -ENOMEM;
384 }
385
validate_ns(struct nsset * nsset,struct ns_common * ns)386 static inline int validate_ns(struct nsset *nsset, struct ns_common *ns)
387 {
388 return ns->ops->install(nsset, ns);
389 }
390
391 /*
392 * This is the inverse operation to unshare().
393 * Ordering is equivalent to the standard ordering used everywhere else
394 * during unshare and process creation. The switch to the new set of
395 * namespaces occurs at the point of no return after installation of
396 * all requested namespaces was successful in commit_nsset().
397 */
validate_nsset(struct nsset * nsset,struct pid * pid)398 static int validate_nsset(struct nsset *nsset, struct pid *pid)
399 {
400 int ret = 0;
401 unsigned flags = nsset->flags;
402 struct user_namespace *user_ns = NULL;
403 struct pid_namespace *pid_ns = NULL;
404 struct nsproxy *nsp;
405 struct task_struct *tsk;
406
407 /* Take a "snapshot" of the target task's namespaces. */
408 rcu_read_lock();
409 tsk = pid_task(pid, PIDTYPE_PID);
410 if (!tsk) {
411 rcu_read_unlock();
412 return -ESRCH;
413 }
414
415 if (!ptrace_may_access(tsk, PTRACE_MODE_READ_REALCREDS)) {
416 rcu_read_unlock();
417 return -EPERM;
418 }
419
420 task_lock(tsk);
421 nsp = tsk->nsproxy;
422 if (nsp)
423 get_nsproxy(nsp);
424 task_unlock(tsk);
425 if (!nsp) {
426 rcu_read_unlock();
427 return -ESRCH;
428 }
429
430 #ifdef CONFIG_PID_NS
431 if (flags & CLONE_NEWPID) {
432 pid_ns = task_active_pid_ns(tsk);
433 if (unlikely(!pid_ns)) {
434 rcu_read_unlock();
435 ret = -ESRCH;
436 goto out;
437 }
438 get_pid_ns(pid_ns);
439 }
440 #endif
441
442 #ifdef CONFIG_USER_NS
443 if (flags & CLONE_NEWUSER)
444 user_ns = get_user_ns(__task_cred(tsk)->user_ns);
445 #endif
446 rcu_read_unlock();
447
448 /*
449 * Install requested namespaces. The caller will have
450 * verified earlier that the requested namespaces are
451 * supported on this kernel. We don't report errors here
452 * if a namespace is requested that isn't supported.
453 */
454 #ifdef CONFIG_USER_NS
455 if (flags & CLONE_NEWUSER) {
456 ret = validate_ns(nsset, &user_ns->ns);
457 if (ret)
458 goto out;
459 }
460 #endif
461
462 if (flags & CLONE_NEWNS) {
463 ret = validate_ns(nsset, from_mnt_ns(nsp->mnt_ns));
464 if (ret)
465 goto out;
466 }
467
468 #ifdef CONFIG_UTS_NS
469 if (flags & CLONE_NEWUTS) {
470 ret = validate_ns(nsset, &nsp->uts_ns->ns);
471 if (ret)
472 goto out;
473 }
474 #endif
475
476 #ifdef CONFIG_IPC_NS
477 if (flags & CLONE_NEWIPC) {
478 ret = validate_ns(nsset, &nsp->ipc_ns->ns);
479 if (ret)
480 goto out;
481 }
482 #endif
483
484 #ifdef CONFIG_PID_NS
485 if (flags & CLONE_NEWPID) {
486 ret = validate_ns(nsset, &pid_ns->ns);
487 if (ret)
488 goto out;
489 }
490 #endif
491
492 #ifdef CONFIG_CGROUPS
493 if (flags & CLONE_NEWCGROUP) {
494 ret = validate_ns(nsset, &nsp->cgroup_ns->ns);
495 if (ret)
496 goto out;
497 }
498 #endif
499
500 #ifdef CONFIG_NET_NS
501 if (flags & CLONE_NEWNET) {
502 ret = validate_ns(nsset, &nsp->net_ns->ns);
503 if (ret)
504 goto out;
505 }
506 #endif
507
508 #ifdef CONFIG_TIME_NS
509 if (flags & CLONE_NEWTIME) {
510 ret = validate_ns(nsset, &nsp->time_ns->ns);
511 if (ret)
512 goto out;
513 }
514 #endif
515
516 out:
517 if (pid_ns)
518 put_pid_ns(pid_ns);
519 if (nsp)
520 put_nsproxy(nsp);
521 put_user_ns(user_ns);
522
523 return ret;
524 }
525
526 /*
527 * This is the point of no return. There are just a few namespaces
528 * that do some actual work here and it's sufficiently minimal that
529 * a separate ns_common operation seems unnecessary for now.
530 * Unshare is doing the same thing. If we'll end up needing to do
531 * more in a given namespace or a helper here is ultimately not
532 * exported anymore a simple commit handler for each namespace
533 * should be added to ns_common.
534 */
commit_nsset(struct nsset * nsset)535 static void commit_nsset(struct nsset *nsset)
536 {
537 unsigned flags = nsset->flags;
538 struct task_struct *me = current;
539
540 #ifdef CONFIG_USER_NS
541 if (flags & CLONE_NEWUSER) {
542 /* transfer ownership */
543 commit_creds(nsset_cred(nsset));
544 nsset->cred = NULL;
545 }
546 #endif
547
548 /* We only need to commit if we have used a temporary fs_struct. */
549 if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) {
550 set_fs_root(me->fs, &nsset->fs->root);
551 set_fs_pwd(me->fs, &nsset->fs->pwd);
552 }
553
554 #ifdef CONFIG_IPC_NS
555 if (flags & CLONE_NEWIPC)
556 exit_sem(me);
557 #endif
558
559 #ifdef CONFIG_TIME_NS
560 if (flags & CLONE_NEWTIME)
561 timens_commit(me, nsset->nsproxy->time_ns);
562 #endif
563
564 /* transfer ownership */
565 switch_task_namespaces(me, nsset->nsproxy);
566 nsset->nsproxy = NULL;
567 }
568
SYSCALL_DEFINE2(setns,int,fd,int,flags)569 SYSCALL_DEFINE2(setns, int, fd, int, flags)
570 {
571 CLASS(fd, f)(fd);
572 struct ns_common *ns = NULL;
573 struct nsset nsset = {};
574 int err = 0;
575
576 if (fd_empty(f))
577 return -EBADF;
578
579 if (proc_ns_file(fd_file(f))) {
580 ns = get_proc_ns(file_inode(fd_file(f)));
581 if (flags && (ns->ns_type != flags))
582 err = -EINVAL;
583 flags = ns->ns_type;
584 } else if (!IS_ERR(pidfd_pid(fd_file(f)))) {
585 err = check_setns_flags(flags);
586 } else {
587 err = -EINVAL;
588 }
589 if (err)
590 goto out;
591
592 err = prepare_nsset(flags, &nsset);
593 if (err)
594 goto out;
595
596 if (proc_ns_file(fd_file(f)))
597 err = validate_ns(&nsset, ns);
598 else
599 err = validate_nsset(&nsset, pidfd_pid(fd_file(f)));
600 if (!err) {
601 commit_nsset(&nsset);
602 perf_event_namespaces(current);
603 }
604 put_nsset(&nsset);
605 out:
606 return err;
607 }
608
nsproxy_cache_init(void)609 int __init nsproxy_cache_init(void)
610 {
611 nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC|SLAB_ACCOUNT);
612 return 0;
613 }
614