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 | |
52 | static inline struct nsproxy *create_nsproxy(void) |
53 | { |
54 | struct nsproxy *nsproxy; |
55 | |
56 | nsproxy = kmem_cache_alloc(cachep: nsproxy_cachep, GFP_KERNEL); |
57 | if (nsproxy) |
58 | refcount_set(r: &nsproxy->count, n: 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 | */ |
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(error: -ENOMEM); |
77 | |
78 | new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs); |
79 | if (IS_ERR(ptr: new_nsp->mnt_ns)) { |
80 | err = PTR_ERR(ptr: new_nsp->mnt_ns); |
81 | goto out_ns; |
82 | } |
83 | |
84 | new_nsp->uts_ns = copy_utsname(flags, user_ns, old_ns: tsk->nsproxy->uts_ns); |
85 | if (IS_ERR(ptr: new_nsp->uts_ns)) { |
86 | err = PTR_ERR(ptr: new_nsp->uts_ns); |
87 | goto out_uts; |
88 | } |
89 | |
90 | new_nsp->ipc_ns = copy_ipcs(flags, user_ns, ns: tsk->nsproxy->ipc_ns); |
91 | if (IS_ERR(ptr: new_nsp->ipc_ns)) { |
92 | err = PTR_ERR(ptr: new_nsp->ipc_ns); |
93 | goto out_ipc; |
94 | } |
95 | |
96 | new_nsp->pid_ns_for_children = |
97 | copy_pid_ns(flags, user_ns, ns: tsk->nsproxy->pid_ns_for_children); |
98 | if (IS_ERR(ptr: new_nsp->pid_ns_for_children)) { |
99 | err = PTR_ERR(ptr: new_nsp->pid_ns_for_children); |
100 | goto out_pid; |
101 | } |
102 | |
103 | new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns, |
104 | old_ns: tsk->nsproxy->cgroup_ns); |
105 | if (IS_ERR(ptr: new_nsp->cgroup_ns)) { |
106 | err = PTR_ERR(ptr: new_nsp->cgroup_ns); |
107 | goto out_cgroup; |
108 | } |
109 | |
110 | new_nsp->net_ns = copy_net_ns(flags, user_ns, old_net: tsk->nsproxy->net_ns); |
111 | if (IS_ERR(ptr: new_nsp->net_ns)) { |
112 | err = PTR_ERR(ptr: new_nsp->net_ns); |
113 | goto out_net; |
114 | } |
115 | |
116 | new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns, |
117 | old_ns: tsk->nsproxy->time_ns_for_children); |
118 | if (IS_ERR(ptr: new_nsp->time_ns_for_children)) { |
119 | err = PTR_ERR(ptr: new_nsp->time_ns_for_children); |
120 | goto out_time; |
121 | } |
122 | new_nsp->time_ns = get_time_ns(ns: tsk->nsproxy->time_ns); |
123 | |
124 | return new_nsp; |
125 | |
126 | out_time: |
127 | put_net(net: new_nsp->net_ns); |
128 | out_net: |
129 | put_cgroup_ns(ns: new_nsp->cgroup_ns); |
130 | out_cgroup: |
131 | if (new_nsp->pid_ns_for_children) |
132 | put_pid_ns(ns: new_nsp->pid_ns_for_children); |
133 | out_pid: |
134 | if (new_nsp->ipc_ns) |
135 | put_ipc_ns(ns: new_nsp->ipc_ns); |
136 | out_ipc: |
137 | if (new_nsp->uts_ns) |
138 | put_uts_ns(ns: new_nsp->uts_ns); |
139 | out_uts: |
140 | if (new_nsp->mnt_ns) |
141 | put_mnt_ns(ns: new_nsp->mnt_ns); |
142 | out_ns: |
143 | kmem_cache_free(s: nsproxy_cachep, objp: new_nsp); |
144 | return ERR_PTR(error: err); |
145 | } |
146 | |
147 | /* |
148 | * called from clone. This now handles copy for nsproxy and all |
149 | * namespaces therein. |
150 | */ |
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 ((flags & CLONE_VM) || |
161 | likely(old_ns->time_ns_for_children == old_ns->time_ns)) { |
162 | get_nsproxy(ns: old_ns); |
163 | return 0; |
164 | } |
165 | } else if (!ns_capable(ns: user_ns, CAP_SYS_ADMIN)) |
166 | return -EPERM; |
167 | |
168 | /* |
169 | * CLONE_NEWIPC must detach from the undolist: after switching |
170 | * to a new ipc namespace, the semaphore arrays from the old |
171 | * namespace are unreachable. In clone parlance, CLONE_SYSVSEM |
172 | * means share undolist with parent, so we must forbid using |
173 | * it along with CLONE_NEWIPC. |
174 | */ |
175 | if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) == |
176 | (CLONE_NEWIPC | CLONE_SYSVSEM)) |
177 | return -EINVAL; |
178 | |
179 | new_ns = create_new_namespaces(flags, tsk, user_ns, new_fs: tsk->fs); |
180 | if (IS_ERR(ptr: new_ns)) |
181 | return PTR_ERR(ptr: new_ns); |
182 | |
183 | if ((flags & CLONE_VM) == 0) |
184 | timens_on_fork(nsproxy: new_ns, tsk); |
185 | |
186 | tsk->nsproxy = new_ns; |
187 | return 0; |
188 | } |
189 | |
190 | void free_nsproxy(struct nsproxy *ns) |
191 | { |
192 | if (ns->mnt_ns) |
193 | put_mnt_ns(ns: ns->mnt_ns); |
194 | if (ns->uts_ns) |
195 | put_uts_ns(ns: ns->uts_ns); |
196 | if (ns->ipc_ns) |
197 | put_ipc_ns(ns: ns->ipc_ns); |
198 | if (ns->pid_ns_for_children) |
199 | put_pid_ns(ns: ns->pid_ns_for_children); |
200 | if (ns->time_ns) |
201 | put_time_ns(ns: ns->time_ns); |
202 | if (ns->time_ns_for_children) |
203 | put_time_ns(ns: ns->time_ns_for_children); |
204 | put_cgroup_ns(ns: ns->cgroup_ns); |
205 | put_net(net: ns->net_ns); |
206 | kmem_cache_free(s: nsproxy_cachep, objp: ns); |
207 | } |
208 | |
209 | /* |
210 | * Called from unshare. Unshare all the namespaces part of nsproxy. |
211 | * On success, returns the new nsproxy. |
212 | */ |
213 | int unshare_nsproxy_namespaces(unsigned long unshare_flags, |
214 | struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs) |
215 | { |
216 | struct user_namespace *user_ns; |
217 | int err = 0; |
218 | |
219 | if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC | |
220 | CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP | |
221 | CLONE_NEWTIME))) |
222 | return 0; |
223 | |
224 | user_ns = new_cred ? new_cred->user_ns : current_user_ns(); |
225 | if (!ns_capable(ns: user_ns, CAP_SYS_ADMIN)) |
226 | return -EPERM; |
227 | |
228 | *new_nsp = create_new_namespaces(flags: unshare_flags, current, user_ns, |
229 | new_fs: new_fs ? new_fs : current->fs); |
230 | if (IS_ERR(ptr: *new_nsp)) { |
231 | err = PTR_ERR(ptr: *new_nsp); |
232 | goto out; |
233 | } |
234 | |
235 | out: |
236 | return err; |
237 | } |
238 | |
239 | void switch_task_namespaces(struct task_struct *p, struct nsproxy *new) |
240 | { |
241 | struct nsproxy *ns; |
242 | |
243 | might_sleep(); |
244 | |
245 | task_lock(p); |
246 | ns = p->nsproxy; |
247 | p->nsproxy = new; |
248 | task_unlock(p); |
249 | |
250 | if (ns) |
251 | put_nsproxy(ns); |
252 | } |
253 | |
254 | void exit_task_namespaces(struct task_struct *p) |
255 | { |
256 | switch_task_namespaces(p, NULL); |
257 | } |
258 | |
259 | int exec_task_namespaces(void) |
260 | { |
261 | struct task_struct *tsk = current; |
262 | struct nsproxy *new; |
263 | |
264 | if (tsk->nsproxy->time_ns_for_children == tsk->nsproxy->time_ns) |
265 | return 0; |
266 | |
267 | new = create_new_namespaces(flags: 0, tsk, current_user_ns(), new_fs: tsk->fs); |
268 | if (IS_ERR(ptr: new)) |
269 | return PTR_ERR(ptr: new); |
270 | |
271 | timens_on_fork(nsproxy: new, tsk); |
272 | switch_task_namespaces(p: tsk, new); |
273 | return 0; |
274 | } |
275 | |
276 | static int check_setns_flags(unsigned long flags) |
277 | { |
278 | if (!flags || (flags & ~(CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC | |
279 | CLONE_NEWNET | CLONE_NEWTIME | CLONE_NEWUSER | |
280 | CLONE_NEWPID | CLONE_NEWCGROUP))) |
281 | return -EINVAL; |
282 | |
283 | #ifndef CONFIG_USER_NS |
284 | if (flags & CLONE_NEWUSER) |
285 | return -EINVAL; |
286 | #endif |
287 | #ifndef CONFIG_PID_NS |
288 | if (flags & CLONE_NEWPID) |
289 | return -EINVAL; |
290 | #endif |
291 | #ifndef CONFIG_UTS_NS |
292 | if (flags & CLONE_NEWUTS) |
293 | return -EINVAL; |
294 | #endif |
295 | #ifndef CONFIG_IPC_NS |
296 | if (flags & CLONE_NEWIPC) |
297 | return -EINVAL; |
298 | #endif |
299 | #ifndef CONFIG_CGROUPS |
300 | if (flags & CLONE_NEWCGROUP) |
301 | return -EINVAL; |
302 | #endif |
303 | #ifndef CONFIG_NET_NS |
304 | if (flags & CLONE_NEWNET) |
305 | return -EINVAL; |
306 | #endif |
307 | #ifndef CONFIG_TIME_NS |
308 | if (flags & CLONE_NEWTIME) |
309 | return -EINVAL; |
310 | #endif |
311 | |
312 | return 0; |
313 | } |
314 | |
315 | static void put_nsset(struct nsset *nsset) |
316 | { |
317 | unsigned flags = nsset->flags; |
318 | |
319 | if (flags & CLONE_NEWUSER) |
320 | put_cred(cred: nsset_cred(set: nsset)); |
321 | /* |
322 | * We only created a temporary copy if we attached to more than just |
323 | * the mount namespace. |
324 | */ |
325 | if (nsset->fs && (flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) |
326 | free_fs_struct(nsset->fs); |
327 | if (nsset->nsproxy) |
328 | free_nsproxy(ns: nsset->nsproxy); |
329 | } |
330 | |
331 | static int prepare_nsset(unsigned flags, struct nsset *nsset) |
332 | { |
333 | struct task_struct *me = current; |
334 | |
335 | nsset->nsproxy = create_new_namespaces(flags: 0, tsk: me, current_user_ns(), new_fs: me->fs); |
336 | if (IS_ERR(ptr: nsset->nsproxy)) |
337 | return PTR_ERR(ptr: nsset->nsproxy); |
338 | |
339 | if (flags & CLONE_NEWUSER) |
340 | nsset->cred = prepare_creds(); |
341 | else |
342 | nsset->cred = current_cred(); |
343 | if (!nsset->cred) |
344 | goto out; |
345 | |
346 | /* Only create a temporary copy of fs_struct if we really need to. */ |
347 | if (flags == CLONE_NEWNS) { |
348 | nsset->fs = me->fs; |
349 | } else if (flags & CLONE_NEWNS) { |
350 | nsset->fs = copy_fs_struct(me->fs); |
351 | if (!nsset->fs) |
352 | goto out; |
353 | } |
354 | |
355 | nsset->flags = flags; |
356 | return 0; |
357 | |
358 | out: |
359 | put_nsset(nsset); |
360 | return -ENOMEM; |
361 | } |
362 | |
363 | static inline int validate_ns(struct nsset *nsset, struct ns_common *ns) |
364 | { |
365 | return ns->ops->install(nsset, ns); |
366 | } |
367 | |
368 | /* |
369 | * This is the inverse operation to unshare(). |
370 | * Ordering is equivalent to the standard ordering used everywhere else |
371 | * during unshare and process creation. The switch to the new set of |
372 | * namespaces occurs at the point of no return after installation of |
373 | * all requested namespaces was successful in commit_nsset(). |
374 | */ |
375 | static int validate_nsset(struct nsset *nsset, struct pid *pid) |
376 | { |
377 | int ret = 0; |
378 | unsigned flags = nsset->flags; |
379 | struct user_namespace *user_ns = NULL; |
380 | struct pid_namespace *pid_ns = NULL; |
381 | struct nsproxy *nsp; |
382 | struct task_struct *tsk; |
383 | |
384 | /* Take a "snapshot" of the target task's namespaces. */ |
385 | rcu_read_lock(); |
386 | tsk = pid_task(pid, PIDTYPE_PID); |
387 | if (!tsk) { |
388 | rcu_read_unlock(); |
389 | return -ESRCH; |
390 | } |
391 | |
392 | if (!ptrace_may_access(task: tsk, PTRACE_MODE_READ_REALCREDS)) { |
393 | rcu_read_unlock(); |
394 | return -EPERM; |
395 | } |
396 | |
397 | task_lock(p: tsk); |
398 | nsp = tsk->nsproxy; |
399 | if (nsp) |
400 | get_nsproxy(ns: nsp); |
401 | task_unlock(p: tsk); |
402 | if (!nsp) { |
403 | rcu_read_unlock(); |
404 | return -ESRCH; |
405 | } |
406 | |
407 | #ifdef CONFIG_PID_NS |
408 | if (flags & CLONE_NEWPID) { |
409 | pid_ns = task_active_pid_ns(tsk); |
410 | if (unlikely(!pid_ns)) { |
411 | rcu_read_unlock(); |
412 | ret = -ESRCH; |
413 | goto out; |
414 | } |
415 | get_pid_ns(ns: pid_ns); |
416 | } |
417 | #endif |
418 | |
419 | #ifdef CONFIG_USER_NS |
420 | if (flags & CLONE_NEWUSER) |
421 | user_ns = get_user_ns(__task_cred(tsk)->user_ns); |
422 | #endif |
423 | rcu_read_unlock(); |
424 | |
425 | /* |
426 | * Install requested namespaces. The caller will have |
427 | * verified earlier that the requested namespaces are |
428 | * supported on this kernel. We don't report errors here |
429 | * if a namespace is requested that isn't supported. |
430 | */ |
431 | #ifdef CONFIG_USER_NS |
432 | if (flags & CLONE_NEWUSER) { |
433 | ret = validate_ns(nsset, ns: &user_ns->ns); |
434 | if (ret) |
435 | goto out; |
436 | } |
437 | #endif |
438 | |
439 | if (flags & CLONE_NEWNS) { |
440 | ret = validate_ns(nsset, ns: from_mnt_ns(nsp->mnt_ns)); |
441 | if (ret) |
442 | goto out; |
443 | } |
444 | |
445 | #ifdef CONFIG_UTS_NS |
446 | if (flags & CLONE_NEWUTS) { |
447 | ret = validate_ns(nsset, ns: &nsp->uts_ns->ns); |
448 | if (ret) |
449 | goto out; |
450 | } |
451 | #endif |
452 | |
453 | #ifdef CONFIG_IPC_NS |
454 | if (flags & CLONE_NEWIPC) { |
455 | ret = validate_ns(nsset, ns: &nsp->ipc_ns->ns); |
456 | if (ret) |
457 | goto out; |
458 | } |
459 | #endif |
460 | |
461 | #ifdef CONFIG_PID_NS |
462 | if (flags & CLONE_NEWPID) { |
463 | ret = validate_ns(nsset, ns: &pid_ns->ns); |
464 | if (ret) |
465 | goto out; |
466 | } |
467 | #endif |
468 | |
469 | #ifdef CONFIG_CGROUPS |
470 | if (flags & CLONE_NEWCGROUP) { |
471 | ret = validate_ns(nsset, ns: &nsp->cgroup_ns->ns); |
472 | if (ret) |
473 | goto out; |
474 | } |
475 | #endif |
476 | |
477 | #ifdef CONFIG_NET_NS |
478 | if (flags & CLONE_NEWNET) { |
479 | ret = validate_ns(nsset, ns: &nsp->net_ns->ns); |
480 | if (ret) |
481 | goto out; |
482 | } |
483 | #endif |
484 | |
485 | #ifdef CONFIG_TIME_NS |
486 | if (flags & CLONE_NEWTIME) { |
487 | ret = validate_ns(nsset, ns: &nsp->time_ns->ns); |
488 | if (ret) |
489 | goto out; |
490 | } |
491 | #endif |
492 | |
493 | out: |
494 | if (pid_ns) |
495 | put_pid_ns(ns: pid_ns); |
496 | if (nsp) |
497 | put_nsproxy(ns: nsp); |
498 | put_user_ns(ns: user_ns); |
499 | |
500 | return ret; |
501 | } |
502 | |
503 | /* |
504 | * This is the point of no return. There are just a few namespaces |
505 | * that do some actual work here and it's sufficiently minimal that |
506 | * a separate ns_common operation seems unnecessary for now. |
507 | * Unshare is doing the same thing. If we'll end up needing to do |
508 | * more in a given namespace or a helper here is ultimately not |
509 | * exported anymore a simple commit handler for each namespace |
510 | * should be added to ns_common. |
511 | */ |
512 | static void commit_nsset(struct nsset *nsset) |
513 | { |
514 | unsigned flags = nsset->flags; |
515 | struct task_struct *me = current; |
516 | |
517 | #ifdef CONFIG_USER_NS |
518 | if (flags & CLONE_NEWUSER) { |
519 | /* transfer ownership */ |
520 | commit_creds(nsset_cred(set: nsset)); |
521 | nsset->cred = NULL; |
522 | } |
523 | #endif |
524 | |
525 | /* We only need to commit if we have used a temporary fs_struct. */ |
526 | if ((flags & CLONE_NEWNS) && (flags & ~CLONE_NEWNS)) { |
527 | set_fs_root(me->fs, &nsset->fs->root); |
528 | set_fs_pwd(me->fs, &nsset->fs->pwd); |
529 | } |
530 | |
531 | #ifdef CONFIG_IPC_NS |
532 | if (flags & CLONE_NEWIPC) |
533 | exit_sem(tsk: me); |
534 | #endif |
535 | |
536 | #ifdef CONFIG_TIME_NS |
537 | if (flags & CLONE_NEWTIME) |
538 | timens_commit(tsk: me, ns: nsset->nsproxy->time_ns); |
539 | #endif |
540 | |
541 | /* transfer ownership */ |
542 | switch_task_namespaces(p: me, new: nsset->nsproxy); |
543 | nsset->nsproxy = NULL; |
544 | } |
545 | |
546 | SYSCALL_DEFINE2(setns, int, fd, int, flags) |
547 | { |
548 | struct fd f = fdget(fd); |
549 | struct ns_common *ns = NULL; |
550 | struct nsset nsset = {}; |
551 | int err = 0; |
552 | |
553 | if (!f.file) |
554 | return -EBADF; |
555 | |
556 | if (proc_ns_file(file: f.file)) { |
557 | ns = get_proc_ns(file_inode(f.file)); |
558 | if (flags && (ns->ops->type != flags)) |
559 | err = -EINVAL; |
560 | flags = ns->ops->type; |
561 | } else if (!IS_ERR(ptr: pidfd_pid(file: f.file))) { |
562 | err = check_setns_flags(flags); |
563 | } else { |
564 | err = -EINVAL; |
565 | } |
566 | if (err) |
567 | goto out; |
568 | |
569 | err = prepare_nsset(flags, nsset: &nsset); |
570 | if (err) |
571 | goto out; |
572 | |
573 | if (proc_ns_file(file: f.file)) |
574 | err = validate_ns(nsset: &nsset, ns); |
575 | else |
576 | err = validate_nsset(nsset: &nsset, pid: f.file->private_data); |
577 | if (!err) { |
578 | commit_nsset(nsset: &nsset); |
579 | perf_event_namespaces(current); |
580 | } |
581 | put_nsset(nsset: &nsset); |
582 | out: |
583 | fdput(fd: f); |
584 | return err; |
585 | } |
586 | |
587 | int __init nsproxy_cache_init(void) |
588 | { |
589 | nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC|SLAB_ACCOUNT); |
590 | return 0; |
591 | } |
592 | |