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