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
157 if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
158 CLONE_NEWPID | CLONE_NEWNET |
159 CLONE_NEWCGROUP | CLONE_NEWTIME)))) {
160 if (likely(old_ns->time_ns_for_children == old_ns->time_ns)) {
161 get_nsproxy(old_ns);
162 return 0;
163 }
164 } else if (!ns_capable(user_ns, CAP_SYS_ADMIN))
165 return -EPERM;
166
167 /*
168 * CLONE_NEWIPC must detach from the undolist: after switching
169 * to a new ipc namespace, the semaphore arrays from the old
170 * namespace are unreachable. In clone parlance, CLONE_SYSVSEM
171 * means share undolist with parent, so we must forbid using
172 * it along with CLONE_NEWIPC.
173 */
174 if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
175 (CLONE_NEWIPC | CLONE_SYSVSEM))
176 return -EINVAL;
177
178 new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
179 if (IS_ERR(new_ns))
180 return PTR_ERR(new_ns);
181
182 timens_on_fork(new_ns, tsk);
183
184 tsk->nsproxy = new_ns;
185 return 0;
186 }
187
free_nsproxy(struct nsproxy * ns)188 void free_nsproxy(struct nsproxy *ns)
189 {
190 if (ns->mnt_ns)
191 put_mnt_ns(ns->mnt_ns);
192 if (ns->uts_ns)
193 put_uts_ns(ns->uts_ns);
194 if (ns->ipc_ns)
195 put_ipc_ns(ns->ipc_ns);
196 if (ns->pid_ns_for_children)
197 put_pid_ns(ns->pid_ns_for_children);
198 if (ns->time_ns)
199 put_time_ns(ns->time_ns);
200 if (ns->time_ns_for_children)
201 put_time_ns(ns->time_ns_for_children);
202 put_cgroup_ns(ns->cgroup_ns);
203 put_net(ns->net_ns);
204 kmem_cache_free(nsproxy_cachep, ns);
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 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_task_namespaces(struct task_struct * p)252 void exit_task_namespaces(struct task_struct *p)
253 {
254 switch_task_namespaces(p, NULL);
255 }
256
check_setns_flags(unsigned long flags)257 static int check_setns_flags(unsigned long flags)
258 {
259 if (!flags || (flags & ~(CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
260 CLONE_NEWNET | CLONE_NEWTIME | CLONE_NEWUSER |
261 CLONE_NEWPID | CLONE_NEWCGROUP)))
262 return -EINVAL;
263
264 #ifndef CONFIG_USER_NS
265 if (flags & CLONE_NEWUSER)
266 return -EINVAL;
267 #endif
268 #ifndef CONFIG_PID_NS
269 if (flags & CLONE_NEWPID)
270 return -EINVAL;
271 #endif
272 #ifndef CONFIG_UTS_NS
273 if (flags & CLONE_NEWUTS)
274 return -EINVAL;
275 #endif
276 #ifndef CONFIG_IPC_NS
277 if (flags & CLONE_NEWIPC)
278 return -EINVAL;
279 #endif
280 #ifndef CONFIG_CGROUPS
281 if (flags & CLONE_NEWCGROUP)
282 return -EINVAL;
283 #endif
284 #ifndef CONFIG_NET_NS
285 if (flags & CLONE_NEWNET)
286 return -EINVAL;
287 #endif
288 #ifndef CONFIG_TIME_NS
289 if (flags & CLONE_NEWTIME)
290 return -EINVAL;
291 #endif
292
293 return 0;
294 }
295
put_nsset(struct nsset * nsset)296 static void put_nsset(struct nsset *nsset)
297 {
298 unsigned flags = nsset->flags;
299
300 if (flags & CLONE_NEWUSER)
301 put_cred(nsset_cred(nsset));
302 /*
303 * We only created a temporary copy if we attached to more than just
304 * the mount namespace.
305 */
306 if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS))
307 free_fs_struct(nsset->fs);
308 if (nsset->nsproxy)
309 free_nsproxy(nsset->nsproxy);
310 }
311
prepare_nsset(unsigned flags,struct nsset * nsset)312 static int prepare_nsset(unsigned flags, struct nsset *nsset)
313 {
314 struct task_struct *me = current;
315
316 nsset->nsproxy = create_new_namespaces(0, me, current_user_ns(), me->fs);
317 if (IS_ERR(nsset->nsproxy))
318 return PTR_ERR(nsset->nsproxy);
319
320 if (flags & CLONE_NEWUSER)
321 nsset->cred = prepare_creds();
322 else
323 nsset->cred = current_cred();
324 if (!nsset->cred)
325 goto out;
326
327 /* Only create a temporary copy of fs_struct if we really need to. */
328 if (flags == CLONE_NEWNS) {
329 nsset->fs = me->fs;
330 } else if (flags & CLONE_NEWNS) {
331 nsset->fs = copy_fs_struct(me->fs);
332 if (!nsset->fs)
333 goto out;
334 }
335
336 nsset->flags = flags;
337 return 0;
338
339 out:
340 put_nsset(nsset);
341 return -ENOMEM;
342 }
343
validate_ns(struct nsset * nsset,struct ns_common * ns)344 static inline int validate_ns(struct nsset *nsset, struct ns_common *ns)
345 {
346 return ns->ops->install(nsset, ns);
347 }
348
349 /*
350 * This is the inverse operation to unshare().
351 * Ordering is equivalent to the standard ordering used everywhere else
352 * during unshare and process creation. The switch to the new set of
353 * namespaces occurs at the point of no return after installation of
354 * all requested namespaces was successful in commit_nsset().
355 */
validate_nsset(struct nsset * nsset,struct pid * pid)356 static int validate_nsset(struct nsset *nsset, struct pid *pid)
357 {
358 int ret = 0;
359 unsigned flags = nsset->flags;
360 struct user_namespace *user_ns = NULL;
361 struct pid_namespace *pid_ns = NULL;
362 struct nsproxy *nsp;
363 struct task_struct *tsk;
364
365 /* Take a "snapshot" of the target task's namespaces. */
366 rcu_read_lock();
367 tsk = pid_task(pid, PIDTYPE_PID);
368 if (!tsk) {
369 rcu_read_unlock();
370 return -ESRCH;
371 }
372
373 if (!ptrace_may_access(tsk, PTRACE_MODE_READ_REALCREDS)) {
374 rcu_read_unlock();
375 return -EPERM;
376 }
377
378 task_lock(tsk);
379 nsp = tsk->nsproxy;
380 if (nsp)
381 get_nsproxy(nsp);
382 task_unlock(tsk);
383 if (!nsp) {
384 rcu_read_unlock();
385 return -ESRCH;
386 }
387
388 #ifdef CONFIG_PID_NS
389 if (flags & CLONE_NEWPID) {
390 pid_ns = task_active_pid_ns(tsk);
391 if (unlikely(!pid_ns)) {
392 rcu_read_unlock();
393 ret = -ESRCH;
394 goto out;
395 }
396 get_pid_ns(pid_ns);
397 }
398 #endif
399
400 #ifdef CONFIG_USER_NS
401 if (flags & CLONE_NEWUSER)
402 user_ns = get_user_ns(__task_cred(tsk)->user_ns);
403 #endif
404 rcu_read_unlock();
405
406 /*
407 * Install requested namespaces. The caller will have
408 * verified earlier that the requested namespaces are
409 * supported on this kernel. We don't report errors here
410 * if a namespace is requested that isn't supported.
411 */
412 #ifdef CONFIG_USER_NS
413 if (flags & CLONE_NEWUSER) {
414 ret = validate_ns(nsset, &user_ns->ns);
415 if (ret)
416 goto out;
417 }
418 #endif
419
420 if (flags & CLONE_NEWNS) {
421 ret = validate_ns(nsset, from_mnt_ns(nsp->mnt_ns));
422 if (ret)
423 goto out;
424 }
425
426 #ifdef CONFIG_UTS_NS
427 if (flags & CLONE_NEWUTS) {
428 ret = validate_ns(nsset, &nsp->uts_ns->ns);
429 if (ret)
430 goto out;
431 }
432 #endif
433
434 #ifdef CONFIG_IPC_NS
435 if (flags & CLONE_NEWIPC) {
436 ret = validate_ns(nsset, &nsp->ipc_ns->ns);
437 if (ret)
438 goto out;
439 }
440 #endif
441
442 #ifdef CONFIG_PID_NS
443 if (flags & CLONE_NEWPID) {
444 ret = validate_ns(nsset, &pid_ns->ns);
445 if (ret)
446 goto out;
447 }
448 #endif
449
450 #ifdef CONFIG_CGROUPS
451 if (flags & CLONE_NEWCGROUP) {
452 ret = validate_ns(nsset, &nsp->cgroup_ns->ns);
453 if (ret)
454 goto out;
455 }
456 #endif
457
458 #ifdef CONFIG_NET_NS
459 if (flags & CLONE_NEWNET) {
460 ret = validate_ns(nsset, &nsp->net_ns->ns);
461 if (ret)
462 goto out;
463 }
464 #endif
465
466 #ifdef CONFIG_TIME_NS
467 if (flags & CLONE_NEWTIME) {
468 ret = validate_ns(nsset, &nsp->time_ns->ns);
469 if (ret)
470 goto out;
471 }
472 #endif
473
474 out:
475 if (pid_ns)
476 put_pid_ns(pid_ns);
477 if (nsp)
478 put_nsproxy(nsp);
479 put_user_ns(user_ns);
480
481 return ret;
482 }
483
484 /*
485 * This is the point of no return. There are just a few namespaces
486 * that do some actual work here and it's sufficiently minimal that
487 * a separate ns_common operation seems unnecessary for now.
488 * Unshare is doing the same thing. If we'll end up needing to do
489 * more in a given namespace or a helper here is ultimately not
490 * exported anymore a simple commit handler for each namespace
491 * should be added to ns_common.
492 */
commit_nsset(struct nsset * nsset)493 static void commit_nsset(struct nsset *nsset)
494 {
495 unsigned flags = nsset->flags;
496 struct task_struct *me = current;
497
498 #ifdef CONFIG_USER_NS
499 if (flags & CLONE_NEWUSER) {
500 /* transfer ownership */
501 commit_creds(nsset_cred(nsset));
502 nsset->cred = NULL;
503 }
504 #endif
505
506 /* We only need to commit if we have used a temporary fs_struct. */
507 if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) {
508 set_fs_root(me->fs, &nsset->fs->root);
509 set_fs_pwd(me->fs, &nsset->fs->pwd);
510 }
511
512 #ifdef CONFIG_IPC_NS
513 if (flags & CLONE_NEWIPC)
514 exit_sem(me);
515 #endif
516
517 #ifdef CONFIG_TIME_NS
518 if (flags & CLONE_NEWTIME)
519 timens_commit(me, nsset->nsproxy->time_ns);
520 #endif
521
522 /* transfer ownership */
523 switch_task_namespaces(me, nsset->nsproxy);
524 nsset->nsproxy = NULL;
525 }
526
SYSCALL_DEFINE2(setns,int,fd,int,flags)527 SYSCALL_DEFINE2(setns, int, fd, int, flags)
528 {
529 struct file *file;
530 struct ns_common *ns = NULL;
531 struct nsset nsset = {};
532 int err = 0;
533
534 file = fget(fd);
535 if (!file)
536 return -EBADF;
537
538 if (proc_ns_file(file)) {
539 ns = get_proc_ns(file_inode(file));
540 if (flags && (ns->ops->type != flags))
541 err = -EINVAL;
542 flags = ns->ops->type;
543 } else if (!IS_ERR(pidfd_pid(file))) {
544 err = check_setns_flags(flags);
545 } else {
546 err = -EINVAL;
547 }
548 if (err)
549 goto out;
550
551 err = prepare_nsset(flags, &nsset);
552 if (err)
553 goto out;
554
555 if (proc_ns_file(file))
556 err = validate_ns(&nsset, ns);
557 else
558 err = validate_nsset(&nsset, file->private_data);
559 if (!err) {
560 commit_nsset(&nsset);
561 perf_event_namespaces(current);
562 }
563 put_nsset(&nsset);
564 out:
565 fput(file);
566 return err;
567 }
568
nsproxy_cache_init(void)569 int __init nsproxy_cache_init(void)
570 {
571 nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC|SLAB_ACCOUNT);
572 return 0;
573 }
574