2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License as
4 * published by the Free Software Foundation, version 2 of the
8 #include <linux/export.h>
9 #include <linux/nsproxy.h>
10 #include <linux/slab.h>
11 #include <linux/user_namespace.h>
12 #include <linux/proc_fs.h>
13 #include <linux/highuid.h>
14 #include <linux/cred.h>
15 #include <linux/securebits.h>
16 #include <linux/keyctl.h>
17 #include <linux/key-type.h>
18 #include <keys/user-type.h>
19 #include <linux/seq_file.h>
21 #include <linux/uaccess.h>
22 #include <linux/ctype.h>
23 #include <linux/projid.h>
25 static struct kmem_cache
*user_ns_cachep __read_mostly
;
27 static bool new_idmap_permitted(struct user_namespace
*ns
, int cap_setid
,
28 struct uid_gid_map
*map
);
30 static void set_cred_user_ns(struct cred
*cred
, struct user_namespace
*user_ns
)
32 /* Start with the same capabilities as init but useless for doing
33 * anything as the capabilities are bound to the new user namespace.
35 cred
->securebits
= SECUREBITS_DEFAULT
;
36 cred
->cap_inheritable
= CAP_EMPTY_SET
;
37 cred
->cap_permitted
= CAP_FULL_SET
;
38 cred
->cap_effective
= CAP_FULL_SET
;
39 cred
->cap_bset
= CAP_FULL_SET
;
41 key_put(cred
->request_key_auth
);
42 cred
->request_key_auth
= NULL
;
44 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
45 cred
->user_ns
= user_ns
;
49 * Create a new user namespace, deriving the creator from the user in the
50 * passed credentials, and replacing that user with the new root user for the
53 * This is called by copy_creds(), which will finish setting the target task's
56 int create_user_ns(struct cred
*new)
58 struct user_namespace
*ns
, *parent_ns
= new->user_ns
;
59 kuid_t owner
= new->euid
;
60 kgid_t group
= new->egid
;
63 /* The creator needs a mapping in the parent user namespace
64 * or else we won't be able to reasonably tell userspace who
65 * created a user_namespace.
67 if (!kuid_has_mapping(parent_ns
, owner
) ||
68 !kgid_has_mapping(parent_ns
, group
))
71 ns
= kmem_cache_zalloc(user_ns_cachep
, GFP_KERNEL
);
75 ret
= proc_alloc_inum(&ns
->proc_inum
);
77 kmem_cache_free(user_ns_cachep
, ns
);
82 /* Leave the new->user_ns reference with the new user namespace. */
83 ns
->parent
= parent_ns
;
87 set_cred_user_ns(new, ns
);
92 int unshare_userns(unsigned long unshare_flags
, struct cred
**new_cred
)
96 if (!(unshare_flags
& CLONE_NEWUSER
))
99 cred
= prepare_creds();
104 return create_user_ns(cred
);
107 void free_user_ns(struct kref
*kref
)
109 struct user_namespace
*parent
, *ns
=
110 container_of(kref
, struct user_namespace
, kref
);
113 proc_free_inum(ns
->proc_inum
);
114 kmem_cache_free(user_ns_cachep
, ns
);
117 EXPORT_SYMBOL(free_user_ns
);
119 static u32
map_id_range_down(struct uid_gid_map
*map
, u32 id
, u32 count
)
121 unsigned idx
, extents
;
122 u32 first
, last
, id2
;
124 id2
= id
+ count
- 1;
126 /* Find the matching extent */
127 extents
= map
->nr_extents
;
128 smp_read_barrier_depends();
129 for (idx
= 0; idx
< extents
; idx
++) {
130 first
= map
->extent
[idx
].first
;
131 last
= first
+ map
->extent
[idx
].count
- 1;
132 if (id
>= first
&& id
<= last
&&
133 (id2
>= first
&& id2
<= last
))
136 /* Map the id or note failure */
138 id
= (id
- first
) + map
->extent
[idx
].lower_first
;
145 static u32
map_id_down(struct uid_gid_map
*map
, u32 id
)
147 unsigned idx
, extents
;
150 /* Find the matching extent */
151 extents
= map
->nr_extents
;
152 smp_read_barrier_depends();
153 for (idx
= 0; idx
< extents
; idx
++) {
154 first
= map
->extent
[idx
].first
;
155 last
= first
+ map
->extent
[idx
].count
- 1;
156 if (id
>= first
&& id
<= last
)
159 /* Map the id or note failure */
161 id
= (id
- first
) + map
->extent
[idx
].lower_first
;
168 static u32
map_id_up(struct uid_gid_map
*map
, u32 id
)
170 unsigned idx
, extents
;
173 /* Find the matching extent */
174 extents
= map
->nr_extents
;
175 smp_read_barrier_depends();
176 for (idx
= 0; idx
< extents
; idx
++) {
177 first
= map
->extent
[idx
].lower_first
;
178 last
= first
+ map
->extent
[idx
].count
- 1;
179 if (id
>= first
&& id
<= last
)
182 /* Map the id or note failure */
184 id
= (id
- first
) + map
->extent
[idx
].first
;
192 * make_kuid - Map a user-namespace uid pair into a kuid.
193 * @ns: User namespace that the uid is in
194 * @uid: User identifier
196 * Maps a user-namespace uid pair into a kernel internal kuid,
197 * and returns that kuid.
199 * When there is no mapping defined for the user-namespace uid
200 * pair INVALID_UID is returned. Callers are expected to test
201 * for and handle handle INVALID_UID being returned. INVALID_UID
202 * may be tested for using uid_valid().
204 kuid_t
make_kuid(struct user_namespace
*ns
, uid_t uid
)
206 /* Map the uid to a global kernel uid */
207 return KUIDT_INIT(map_id_down(&ns
->uid_map
, uid
));
209 EXPORT_SYMBOL(make_kuid
);
212 * from_kuid - Create a uid from a kuid user-namespace pair.
213 * @targ: The user namespace we want a uid in.
214 * @kuid: The kernel internal uid to start with.
216 * Map @kuid into the user-namespace specified by @targ and
217 * return the resulting uid.
219 * There is always a mapping into the initial user_namespace.
221 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
223 uid_t
from_kuid(struct user_namespace
*targ
, kuid_t kuid
)
225 /* Map the uid from a global kernel uid */
226 return map_id_up(&targ
->uid_map
, __kuid_val(kuid
));
228 EXPORT_SYMBOL(from_kuid
);
231 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
232 * @targ: The user namespace we want a uid in.
233 * @kuid: The kernel internal uid to start with.
235 * Map @kuid into the user-namespace specified by @targ and
236 * return the resulting uid.
238 * There is always a mapping into the initial user_namespace.
240 * Unlike from_kuid from_kuid_munged never fails and always
241 * returns a valid uid. This makes from_kuid_munged appropriate
242 * for use in syscalls like stat and getuid where failing the
243 * system call and failing to provide a valid uid are not an
246 * If @kuid has no mapping in @targ overflowuid is returned.
248 uid_t
from_kuid_munged(struct user_namespace
*targ
, kuid_t kuid
)
251 uid
= from_kuid(targ
, kuid
);
253 if (uid
== (uid_t
) -1)
257 EXPORT_SYMBOL(from_kuid_munged
);
260 * make_kgid - Map a user-namespace gid pair into a kgid.
261 * @ns: User namespace that the gid is in
262 * @uid: group identifier
264 * Maps a user-namespace gid pair into a kernel internal kgid,
265 * and returns that kgid.
267 * When there is no mapping defined for the user-namespace gid
268 * pair INVALID_GID is returned. Callers are expected to test
269 * for and handle INVALID_GID being returned. INVALID_GID may be
270 * tested for using gid_valid().
272 kgid_t
make_kgid(struct user_namespace
*ns
, gid_t gid
)
274 /* Map the gid to a global kernel gid */
275 return KGIDT_INIT(map_id_down(&ns
->gid_map
, gid
));
277 EXPORT_SYMBOL(make_kgid
);
280 * from_kgid - Create a gid from a kgid user-namespace pair.
281 * @targ: The user namespace we want a gid in.
282 * @kgid: The kernel internal gid to start with.
284 * Map @kgid into the user-namespace specified by @targ and
285 * return the resulting gid.
287 * There is always a mapping into the initial user_namespace.
289 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
291 gid_t
from_kgid(struct user_namespace
*targ
, kgid_t kgid
)
293 /* Map the gid from a global kernel gid */
294 return map_id_up(&targ
->gid_map
, __kgid_val(kgid
));
296 EXPORT_SYMBOL(from_kgid
);
299 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
300 * @targ: The user namespace we want a gid in.
301 * @kgid: The kernel internal gid to start with.
303 * Map @kgid into the user-namespace specified by @targ and
304 * return the resulting gid.
306 * There is always a mapping into the initial user_namespace.
308 * Unlike from_kgid from_kgid_munged never fails and always
309 * returns a valid gid. This makes from_kgid_munged appropriate
310 * for use in syscalls like stat and getgid where failing the
311 * system call and failing to provide a valid gid are not options.
313 * If @kgid has no mapping in @targ overflowgid is returned.
315 gid_t
from_kgid_munged(struct user_namespace
*targ
, kgid_t kgid
)
318 gid
= from_kgid(targ
, kgid
);
320 if (gid
== (gid_t
) -1)
324 EXPORT_SYMBOL(from_kgid_munged
);
327 * make_kprojid - Map a user-namespace projid pair into a kprojid.
328 * @ns: User namespace that the projid is in
329 * @projid: Project identifier
331 * Maps a user-namespace uid pair into a kernel internal kuid,
332 * and returns that kuid.
334 * When there is no mapping defined for the user-namespace projid
335 * pair INVALID_PROJID is returned. Callers are expected to test
336 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
337 * may be tested for using projid_valid().
339 kprojid_t
make_kprojid(struct user_namespace
*ns
, projid_t projid
)
341 /* Map the uid to a global kernel uid */
342 return KPROJIDT_INIT(map_id_down(&ns
->projid_map
, projid
));
344 EXPORT_SYMBOL(make_kprojid
);
347 * from_kprojid - Create a projid from a kprojid user-namespace pair.
348 * @targ: The user namespace we want a projid in.
349 * @kprojid: The kernel internal project identifier to start with.
351 * Map @kprojid into the user-namespace specified by @targ and
352 * return the resulting projid.
354 * There is always a mapping into the initial user_namespace.
356 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
358 projid_t
from_kprojid(struct user_namespace
*targ
, kprojid_t kprojid
)
360 /* Map the uid from a global kernel uid */
361 return map_id_up(&targ
->projid_map
, __kprojid_val(kprojid
));
363 EXPORT_SYMBOL(from_kprojid
);
366 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
367 * @targ: The user namespace we want a projid in.
368 * @kprojid: The kernel internal projid to start with.
370 * Map @kprojid into the user-namespace specified by @targ and
371 * return the resulting projid.
373 * There is always a mapping into the initial user_namespace.
375 * Unlike from_kprojid from_kprojid_munged never fails and always
376 * returns a valid projid. This makes from_kprojid_munged
377 * appropriate for use in syscalls like stat and where
378 * failing the system call and failing to provide a valid projid are
381 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
383 projid_t
from_kprojid_munged(struct user_namespace
*targ
, kprojid_t kprojid
)
386 projid
= from_kprojid(targ
, kprojid
);
388 if (projid
== (projid_t
) -1)
389 projid
= OVERFLOW_PROJID
;
392 EXPORT_SYMBOL(from_kprojid_munged
);
395 static int uid_m_show(struct seq_file
*seq
, void *v
)
397 struct user_namespace
*ns
= seq
->private;
398 struct uid_gid_extent
*extent
= v
;
399 struct user_namespace
*lower_ns
;
402 lower_ns
= seq_user_ns(seq
);
403 if ((lower_ns
== ns
) && lower_ns
->parent
)
404 lower_ns
= lower_ns
->parent
;
406 lower
= from_kuid(lower_ns
, KUIDT_INIT(extent
->lower_first
));
408 seq_printf(seq
, "%10u %10u %10u\n",
416 static int gid_m_show(struct seq_file
*seq
, void *v
)
418 struct user_namespace
*ns
= seq
->private;
419 struct uid_gid_extent
*extent
= v
;
420 struct user_namespace
*lower_ns
;
423 lower_ns
= seq_user_ns(seq
);
424 if ((lower_ns
== ns
) && lower_ns
->parent
)
425 lower_ns
= lower_ns
->parent
;
427 lower
= from_kgid(lower_ns
, KGIDT_INIT(extent
->lower_first
));
429 seq_printf(seq
, "%10u %10u %10u\n",
437 static int projid_m_show(struct seq_file
*seq
, void *v
)
439 struct user_namespace
*ns
= seq
->private;
440 struct uid_gid_extent
*extent
= v
;
441 struct user_namespace
*lower_ns
;
444 lower_ns
= seq_user_ns(seq
);
445 if ((lower_ns
== ns
) && lower_ns
->parent
)
446 lower_ns
= lower_ns
->parent
;
448 lower
= from_kprojid(lower_ns
, KPROJIDT_INIT(extent
->lower_first
));
450 seq_printf(seq
, "%10u %10u %10u\n",
458 static void *m_start(struct seq_file
*seq
, loff_t
*ppos
, struct uid_gid_map
*map
)
460 struct uid_gid_extent
*extent
= NULL
;
463 if (pos
< map
->nr_extents
)
464 extent
= &map
->extent
[pos
];
469 static void *uid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
471 struct user_namespace
*ns
= seq
->private;
473 return m_start(seq
, ppos
, &ns
->uid_map
);
476 static void *gid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
478 struct user_namespace
*ns
= seq
->private;
480 return m_start(seq
, ppos
, &ns
->gid_map
);
483 static void *projid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
485 struct user_namespace
*ns
= seq
->private;
487 return m_start(seq
, ppos
, &ns
->projid_map
);
490 static void *m_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
493 return seq
->op
->start(seq
, pos
);
496 static void m_stop(struct seq_file
*seq
, void *v
)
501 struct seq_operations proc_uid_seq_operations
= {
502 .start
= uid_m_start
,
508 struct seq_operations proc_gid_seq_operations
= {
509 .start
= gid_m_start
,
515 struct seq_operations proc_projid_seq_operations
= {
516 .start
= projid_m_start
,
519 .show
= projid_m_show
,
522 static DEFINE_MUTEX(id_map_mutex
);
524 static ssize_t
map_write(struct file
*file
, const char __user
*buf
,
525 size_t count
, loff_t
*ppos
,
527 struct uid_gid_map
*map
,
528 struct uid_gid_map
*parent_map
)
530 struct seq_file
*seq
= file
->private_data
;
531 struct user_namespace
*ns
= seq
->private;
532 struct uid_gid_map new_map
;
534 struct uid_gid_extent
*extent
, *last
= NULL
;
535 unsigned long page
= 0;
536 char *kbuf
, *pos
, *next_line
;
537 ssize_t ret
= -EINVAL
;
540 * The id_map_mutex serializes all writes to any given map.
542 * Any map is only ever written once.
544 * An id map fits within 1 cache line on most architectures.
546 * On read nothing needs to be done unless you are on an
547 * architecture with a crazy cache coherency model like alpha.
549 * There is a one time data dependency between reading the
550 * count of the extents and the values of the extents. The
551 * desired behavior is to see the values of the extents that
552 * were written before the count of the extents.
554 * To achieve this smp_wmb() is used on guarantee the write
555 * order and smp_read_barrier_depends() is guaranteed that we
556 * don't have crazy architectures returning stale data.
559 mutex_lock(&id_map_mutex
);
562 /* Only allow one successful write to the map */
563 if (map
->nr_extents
!= 0)
566 /* Require the appropriate privilege CAP_SETUID or CAP_SETGID
567 * over the user namespace in order to set the id mapping.
569 if (cap_valid(cap_setid
) && !ns_capable(ns
, cap_setid
))
574 page
= __get_free_page(GFP_TEMPORARY
);
575 kbuf
= (char *) page
;
579 /* Only allow <= page size writes at the beginning of the file */
581 if ((*ppos
!= 0) || (count
>= PAGE_SIZE
))
584 /* Slurp in the user data */
586 if (copy_from_user(kbuf
, buf
, count
))
590 /* Parse the user data */
593 new_map
.nr_extents
= 0;
594 for (;pos
; pos
= next_line
) {
595 extent
= &new_map
.extent
[new_map
.nr_extents
];
597 /* Find the end of line and ensure I don't look past it */
598 next_line
= strchr(pos
, '\n');
602 if (*next_line
== '\0')
606 pos
= skip_spaces(pos
);
607 extent
->first
= simple_strtoul(pos
, &pos
, 10);
611 pos
= skip_spaces(pos
);
612 extent
->lower_first
= simple_strtoul(pos
, &pos
, 10);
616 pos
= skip_spaces(pos
);
617 extent
->count
= simple_strtoul(pos
, &pos
, 10);
618 if (*pos
&& !isspace(*pos
))
621 /* Verify there is not trailing junk on the line */
622 pos
= skip_spaces(pos
);
626 /* Verify we have been given valid starting values */
627 if ((extent
->first
== (u32
) -1) ||
628 (extent
->lower_first
== (u32
) -1 ))
631 /* Verify count is not zero and does not cause the extent to wrap */
632 if ((extent
->first
+ extent
->count
) <= extent
->first
)
634 if ((extent
->lower_first
+ extent
->count
) <= extent
->lower_first
)
637 /* For now only accept extents that are strictly in order */
639 (((last
->first
+ last
->count
) > extent
->first
) ||
640 ((last
->lower_first
+ last
->count
) > extent
->lower_first
)))
643 new_map
.nr_extents
++;
646 /* Fail if the file contains too many extents */
647 if ((new_map
.nr_extents
== UID_GID_MAP_MAX_EXTENTS
) &&
651 /* Be very certaint the new map actually exists */
652 if (new_map
.nr_extents
== 0)
656 /* Validate the user is allowed to use user id's mapped to. */
657 if (!new_idmap_permitted(ns
, cap_setid
, &new_map
))
660 /* Map the lower ids from the parent user namespace to the
661 * kernel global id space.
663 for (idx
= 0; idx
< new_map
.nr_extents
; idx
++) {
665 extent
= &new_map
.extent
[idx
];
667 lower_first
= map_id_range_down(parent_map
,
671 /* Fail if we can not map the specified extent to
672 * the kernel global id space.
674 if (lower_first
== (u32
) -1)
677 extent
->lower_first
= lower_first
;
680 /* Install the map */
681 memcpy(map
->extent
, new_map
.extent
,
682 new_map
.nr_extents
*sizeof(new_map
.extent
[0]));
684 map
->nr_extents
= new_map
.nr_extents
;
689 mutex_unlock(&id_map_mutex
);
695 ssize_t
proc_uid_map_write(struct file
*file
, const char __user
*buf
, size_t size
, loff_t
*ppos
)
697 struct seq_file
*seq
= file
->private_data
;
698 struct user_namespace
*ns
= seq
->private;
699 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
704 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
707 return map_write(file
, buf
, size
, ppos
, CAP_SETUID
,
708 &ns
->uid_map
, &ns
->parent
->uid_map
);
711 ssize_t
proc_gid_map_write(struct file
*file
, const char __user
*buf
, size_t size
, loff_t
*ppos
)
713 struct seq_file
*seq
= file
->private_data
;
714 struct user_namespace
*ns
= seq
->private;
715 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
720 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
723 return map_write(file
, buf
, size
, ppos
, CAP_SETGID
,
724 &ns
->gid_map
, &ns
->parent
->gid_map
);
727 ssize_t
proc_projid_map_write(struct file
*file
, const char __user
*buf
, size_t size
, loff_t
*ppos
)
729 struct seq_file
*seq
= file
->private_data
;
730 struct user_namespace
*ns
= seq
->private;
731 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
736 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
739 /* Anyone can set any valid project id no capability needed */
740 return map_write(file
, buf
, size
, ppos
, -1,
741 &ns
->projid_map
, &ns
->parent
->projid_map
);
744 static bool new_idmap_permitted(struct user_namespace
*ns
, int cap_setid
,
745 struct uid_gid_map
*new_map
)
747 /* Allow mapping to your own filesystem ids */
748 if ((new_map
->nr_extents
== 1) && (new_map
->extent
[0].count
== 1)) {
749 u32 id
= new_map
->extent
[0].lower_first
;
750 if (cap_setid
== CAP_SETUID
) {
751 kuid_t uid
= make_kuid(ns
->parent
, id
);
752 if (uid_eq(uid
, current_fsuid()))
755 else if (cap_setid
== CAP_SETGID
) {
756 kgid_t gid
= make_kgid(ns
->parent
, id
);
757 if (gid_eq(gid
, current_fsgid()))
762 /* Allow anyone to set a mapping that doesn't require privilege */
763 if (!cap_valid(cap_setid
))
766 /* Allow the specified ids if we have the appropriate capability
767 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
769 if (ns_capable(ns
->parent
, cap_setid
))
775 static void *userns_get(struct task_struct
*task
)
777 struct user_namespace
*user_ns
;
780 user_ns
= get_user_ns(__task_cred(task
)->user_ns
);
786 static void userns_put(void *ns
)
791 static int userns_install(struct nsproxy
*nsproxy
, void *ns
)
793 struct user_namespace
*user_ns
= ns
;
796 /* Don't allow gaining capabilities by reentering
797 * the same user namespace.
799 if (user_ns
== current_user_ns())
802 /* Threaded processes may not enter a different user namespace */
803 if (atomic_read(¤t
->mm
->mm_users
) > 1)
806 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
809 cred
= prepare_creds();
813 put_user_ns(cred
->user_ns
);
814 set_cred_user_ns(cred
, get_user_ns(user_ns
));
816 return commit_creds(cred
);
819 static unsigned int userns_inum(void *ns
)
821 struct user_namespace
*user_ns
= ns
;
822 return user_ns
->proc_inum
;
825 const struct proc_ns_operations userns_operations
= {
827 .type
= CLONE_NEWUSER
,
830 .install
= userns_install
,
834 static __init
int user_namespaces_init(void)
836 user_ns_cachep
= KMEM_CACHE(user_namespace
, SLAB_PANIC
);
839 module_init(user_namespaces_init
);