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