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