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/sched/signal.h>
12 #include <linux/user_namespace.h>
13 #include <linux/proc_ns.h>
14 #include <linux/highuid.h>
15 #include <linux/cred.h>
16 #include <linux/securebits.h>
17 #include <linux/keyctl.h>
18 #include <linux/key-type.h>
19 #include <keys/user-type.h>
20 #include <linux/seq_file.h>
22 #include <linux/uaccess.h>
23 #include <linux/ctype.h>
24 #include <linux/projid.h>
25 #include <linux/fs_struct.h>
26 #include <linux/bsearch.h>
27 #include <linux/sort.h>
29 static struct kmem_cache
*user_ns_cachep __read_mostly
;
30 static DEFINE_MUTEX(userns_state_mutex
);
32 static bool new_idmap_permitted(const struct file
*file
,
33 struct user_namespace
*ns
, int cap_setid
,
34 struct uid_gid_map
*map
);
35 static void free_user_ns(struct work_struct
*work
);
37 static struct ucounts
*inc_user_namespaces(struct user_namespace
*ns
, kuid_t uid
)
39 return inc_ucount(ns
, uid
, UCOUNT_USER_NAMESPACES
);
42 static void dec_user_namespaces(struct ucounts
*ucounts
)
44 return dec_ucount(ucounts
, UCOUNT_USER_NAMESPACES
);
47 static void set_cred_user_ns(struct cred
*cred
, struct user_namespace
*user_ns
)
49 /* Start with the same capabilities as init but useless for doing
50 * anything as the capabilities are bound to the new user namespace.
52 cred
->securebits
= SECUREBITS_DEFAULT
;
53 cred
->cap_inheritable
= CAP_EMPTY_SET
;
54 cred
->cap_permitted
= CAP_FULL_SET
;
55 cred
->cap_effective
= CAP_FULL_SET
;
56 cred
->cap_ambient
= CAP_EMPTY_SET
;
57 cred
->cap_bset
= CAP_FULL_SET
;
59 key_put(cred
->request_key_auth
);
60 cred
->request_key_auth
= NULL
;
62 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
63 cred
->user_ns
= user_ns
;
67 * Create a new user namespace, deriving the creator from the user in the
68 * passed credentials, and replacing that user with the new root user for the
71 * This is called by copy_creds(), which will finish setting the target task's
74 int create_user_ns(struct cred
*new)
76 struct user_namespace
*ns
, *parent_ns
= new->user_ns
;
77 kuid_t owner
= new->euid
;
78 kgid_t group
= new->egid
;
79 struct ucounts
*ucounts
;
83 if (parent_ns
->level
> 32)
86 ucounts
= inc_user_namespaces(parent_ns
, owner
);
91 * Verify that we can not violate the policy of which files
92 * may be accessed that is specified by the root directory,
93 * by verifing that the root directory is at the root of the
94 * mount namespace which allows all files to be accessed.
97 if (current_chrooted())
100 /* The creator needs a mapping in the parent user namespace
101 * or else we won't be able to reasonably tell userspace who
102 * created a user_namespace.
105 if (!kuid_has_mapping(parent_ns
, owner
) ||
106 !kgid_has_mapping(parent_ns
, group
))
110 ns
= kmem_cache_zalloc(user_ns_cachep
, GFP_KERNEL
);
114 ret
= ns_alloc_inum(&ns
->ns
);
117 ns
->ns
.ops
= &userns_operations
;
119 atomic_set(&ns
->count
, 1);
120 /* Leave the new->user_ns reference with the new user namespace. */
121 ns
->parent
= parent_ns
;
122 ns
->level
= parent_ns
->level
+ 1;
125 INIT_WORK(&ns
->work
, free_user_ns
);
126 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
127 ns
->ucount_max
[i
] = INT_MAX
;
129 ns
->ucounts
= ucounts
;
131 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
132 mutex_lock(&userns_state_mutex
);
133 ns
->flags
= parent_ns
->flags
;
134 mutex_unlock(&userns_state_mutex
);
136 #ifdef CONFIG_PERSISTENT_KEYRINGS
137 init_rwsem(&ns
->persistent_keyring_register_sem
);
140 if (!setup_userns_sysctls(ns
))
143 set_cred_user_ns(new, ns
);
146 #ifdef CONFIG_PERSISTENT_KEYRINGS
147 key_put(ns
->persistent_keyring_register
);
149 ns_free_inum(&ns
->ns
);
151 kmem_cache_free(user_ns_cachep
, ns
);
153 dec_user_namespaces(ucounts
);
158 int unshare_userns(unsigned long unshare_flags
, struct cred
**new_cred
)
163 if (!(unshare_flags
& CLONE_NEWUSER
))
166 cred
= prepare_creds();
168 err
= create_user_ns(cred
);
178 static void free_user_ns(struct work_struct
*work
)
180 struct user_namespace
*parent
, *ns
=
181 container_of(work
, struct user_namespace
, work
);
184 struct ucounts
*ucounts
= ns
->ucounts
;
186 if (ns
->gid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
187 kfree(ns
->gid_map
.forward
);
188 kfree(ns
->gid_map
.reverse
);
190 if (ns
->uid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
191 kfree(ns
->uid_map
.forward
);
192 kfree(ns
->uid_map
.reverse
);
194 if (ns
->projid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
195 kfree(ns
->projid_map
.forward
);
196 kfree(ns
->projid_map
.reverse
);
198 retire_userns_sysctls(ns
);
199 #ifdef CONFIG_PERSISTENT_KEYRINGS
200 key_put(ns
->persistent_keyring_register
);
202 ns_free_inum(&ns
->ns
);
203 kmem_cache_free(user_ns_cachep
, ns
);
204 dec_user_namespaces(ucounts
);
206 } while (atomic_dec_and_test(&parent
->count
));
209 void __put_user_ns(struct user_namespace
*ns
)
211 schedule_work(&ns
->work
);
213 EXPORT_SYMBOL(__put_user_ns
);
216 * idmap_key struct holds the information necessary to find an idmapping in a
217 * sorted idmap array. It is passed to cmp_map_id() as first argument.
220 bool map_up
; /* true -> id from kid; false -> kid from id */
221 u32 id
; /* id to find */
222 u32 count
; /* == 0 unless used with map_id_range_down() */
226 * cmp_map_id - Function to be passed to bsearch() to find the requested
227 * idmapping. Expects struct idmap_key to be passed via @k.
229 static int cmp_map_id(const void *k
, const void *e
)
231 u32 first
, last
, id2
;
232 const struct idmap_key
*key
= k
;
233 const struct uid_gid_extent
*el
= e
;
235 id2
= key
->id
+ key
->count
- 1;
237 /* handle map_id_{down,up}() */
239 first
= el
->lower_first
;
243 last
= first
+ el
->count
- 1;
245 if (key
->id
>= first
&& key
->id
<= last
&&
246 (id2
>= first
&& id2
<= last
))
249 if (key
->id
< first
|| id2
< first
)
256 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
257 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
259 static struct uid_gid_extent
*
260 map_id_range_down_max(unsigned extents
, struct uid_gid_map
*map
, u32 id
, u32 count
)
262 struct idmap_key key
;
268 return bsearch(&key
, map
->forward
, extents
,
269 sizeof(struct uid_gid_extent
), cmp_map_id
);
273 * map_id_range_down_base - Find idmap via binary search in static extent array.
274 * Can only be called if number of mappings is equal or less than
275 * UID_GID_MAP_MAX_BASE_EXTENTS.
277 static struct uid_gid_extent
*
278 map_id_range_down_base(unsigned extents
, struct uid_gid_map
*map
, u32 id
, u32 count
)
281 u32 first
, last
, id2
;
283 id2
= id
+ count
- 1;
285 /* Find the matching extent */
286 for (idx
= 0; idx
< extents
; idx
++) {
287 first
= map
->extent
[idx
].first
;
288 last
= first
+ map
->extent
[idx
].count
- 1;
289 if (id
>= first
&& id
<= last
&&
290 (id2
>= first
&& id2
<= last
))
291 return &map
->extent
[idx
];
296 static u32
map_id_range_down(struct uid_gid_map
*map
, u32 id
, u32 count
)
298 struct uid_gid_extent
*extent
;
299 unsigned extents
= map
->nr_extents
;
302 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
303 extent
= map_id_range_down_base(extents
, map
, id
, count
);
305 extent
= map_id_range_down_max(extents
, map
, id
, count
);
307 /* Map the id or note failure */
309 id
= (id
- extent
->first
) + extent
->lower_first
;
316 static u32
map_id_down(struct uid_gid_map
*map
, u32 id
)
318 return map_id_range_down(map
, id
, 1);
322 * map_id_up_base - Find idmap via binary search in static extent array.
323 * Can only be called if number of mappings is equal or less than
324 * UID_GID_MAP_MAX_BASE_EXTENTS.
326 static struct uid_gid_extent
*
327 map_id_up_base(unsigned extents
, struct uid_gid_map
*map
, u32 id
)
332 /* Find the matching extent */
333 for (idx
= 0; idx
< extents
; idx
++) {
334 first
= map
->extent
[idx
].lower_first
;
335 last
= first
+ map
->extent
[idx
].count
- 1;
336 if (id
>= first
&& id
<= last
)
337 return &map
->extent
[idx
];
343 * map_id_up_max - Find idmap via binary search in ordered idmap array.
344 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
346 static struct uid_gid_extent
*
347 map_id_up_max(unsigned extents
, struct uid_gid_map
*map
, u32 id
)
349 struct idmap_key key
;
355 return bsearch(&key
, map
->reverse
, extents
,
356 sizeof(struct uid_gid_extent
), cmp_map_id
);
359 static u32
map_id_up(struct uid_gid_map
*map
, u32 id
)
361 struct uid_gid_extent
*extent
;
362 unsigned extents
= map
->nr_extents
;
365 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
366 extent
= map_id_up_base(extents
, map
, id
);
368 extent
= map_id_up_max(extents
, map
, id
);
370 /* Map the id or note failure */
372 id
= (id
- extent
->lower_first
) + extent
->first
;
380 * make_kuid - Map a user-namespace uid pair into a kuid.
381 * @ns: User namespace that the uid is in
382 * @uid: User identifier
384 * Maps a user-namespace uid pair into a kernel internal kuid,
385 * and returns that kuid.
387 * When there is no mapping defined for the user-namespace uid
388 * pair INVALID_UID is returned. Callers are expected to test
389 * for and handle INVALID_UID being returned. INVALID_UID
390 * may be tested for using uid_valid().
392 kuid_t
make_kuid(struct user_namespace
*ns
, uid_t uid
)
394 /* Map the uid to a global kernel uid */
395 return KUIDT_INIT(map_id_down(&ns
->uid_map
, uid
));
397 EXPORT_SYMBOL(make_kuid
);
400 * from_kuid - Create a uid from a kuid user-namespace pair.
401 * @targ: The user namespace we want a uid in.
402 * @kuid: The kernel internal uid to start with.
404 * Map @kuid into the user-namespace specified by @targ and
405 * return the resulting uid.
407 * There is always a mapping into the initial user_namespace.
409 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
411 uid_t
from_kuid(struct user_namespace
*targ
, kuid_t kuid
)
413 /* Map the uid from a global kernel uid */
414 return map_id_up(&targ
->uid_map
, __kuid_val(kuid
));
416 EXPORT_SYMBOL(from_kuid
);
419 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
420 * @targ: The user namespace we want a uid in.
421 * @kuid: The kernel internal uid to start with.
423 * Map @kuid into the user-namespace specified by @targ and
424 * return the resulting uid.
426 * There is always a mapping into the initial user_namespace.
428 * Unlike from_kuid from_kuid_munged never fails and always
429 * returns a valid uid. This makes from_kuid_munged appropriate
430 * for use in syscalls like stat and getuid where failing the
431 * system call and failing to provide a valid uid are not an
434 * If @kuid has no mapping in @targ overflowuid is returned.
436 uid_t
from_kuid_munged(struct user_namespace
*targ
, kuid_t kuid
)
439 uid
= from_kuid(targ
, kuid
);
441 if (uid
== (uid_t
) -1)
445 EXPORT_SYMBOL(from_kuid_munged
);
448 * make_kgid - Map a user-namespace gid pair into a kgid.
449 * @ns: User namespace that the gid is in
450 * @gid: group identifier
452 * Maps a user-namespace gid pair into a kernel internal kgid,
453 * and returns that kgid.
455 * When there is no mapping defined for the user-namespace gid
456 * pair INVALID_GID is returned. Callers are expected to test
457 * for and handle INVALID_GID being returned. INVALID_GID may be
458 * tested for using gid_valid().
460 kgid_t
make_kgid(struct user_namespace
*ns
, gid_t gid
)
462 /* Map the gid to a global kernel gid */
463 return KGIDT_INIT(map_id_down(&ns
->gid_map
, gid
));
465 EXPORT_SYMBOL(make_kgid
);
468 * from_kgid - Create a gid from a kgid user-namespace pair.
469 * @targ: The user namespace we want a gid in.
470 * @kgid: The kernel internal gid to start with.
472 * Map @kgid into the user-namespace specified by @targ and
473 * return the resulting gid.
475 * There is always a mapping into the initial user_namespace.
477 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
479 gid_t
from_kgid(struct user_namespace
*targ
, kgid_t kgid
)
481 /* Map the gid from a global kernel gid */
482 return map_id_up(&targ
->gid_map
, __kgid_val(kgid
));
484 EXPORT_SYMBOL(from_kgid
);
487 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
488 * @targ: The user namespace we want a gid in.
489 * @kgid: The kernel internal gid to start with.
491 * Map @kgid into the user-namespace specified by @targ and
492 * return the resulting gid.
494 * There is always a mapping into the initial user_namespace.
496 * Unlike from_kgid from_kgid_munged never fails and always
497 * returns a valid gid. This makes from_kgid_munged appropriate
498 * for use in syscalls like stat and getgid where failing the
499 * system call and failing to provide a valid gid are not options.
501 * If @kgid has no mapping in @targ overflowgid is returned.
503 gid_t
from_kgid_munged(struct user_namespace
*targ
, kgid_t kgid
)
506 gid
= from_kgid(targ
, kgid
);
508 if (gid
== (gid_t
) -1)
512 EXPORT_SYMBOL(from_kgid_munged
);
515 * make_kprojid - Map a user-namespace projid pair into a kprojid.
516 * @ns: User namespace that the projid is in
517 * @projid: Project identifier
519 * Maps a user-namespace uid pair into a kernel internal kuid,
520 * and returns that kuid.
522 * When there is no mapping defined for the user-namespace projid
523 * pair INVALID_PROJID is returned. Callers are expected to test
524 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
525 * may be tested for using projid_valid().
527 kprojid_t
make_kprojid(struct user_namespace
*ns
, projid_t projid
)
529 /* Map the uid to a global kernel uid */
530 return KPROJIDT_INIT(map_id_down(&ns
->projid_map
, projid
));
532 EXPORT_SYMBOL(make_kprojid
);
535 * from_kprojid - Create a projid from a kprojid user-namespace pair.
536 * @targ: The user namespace we want a projid in.
537 * @kprojid: The kernel internal project identifier to start with.
539 * Map @kprojid into the user-namespace specified by @targ and
540 * return the resulting projid.
542 * There is always a mapping into the initial user_namespace.
544 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
546 projid_t
from_kprojid(struct user_namespace
*targ
, kprojid_t kprojid
)
548 /* Map the uid from a global kernel uid */
549 return map_id_up(&targ
->projid_map
, __kprojid_val(kprojid
));
551 EXPORT_SYMBOL(from_kprojid
);
554 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
555 * @targ: The user namespace we want a projid in.
556 * @kprojid: The kernel internal projid to start with.
558 * Map @kprojid into the user-namespace specified by @targ and
559 * return the resulting projid.
561 * There is always a mapping into the initial user_namespace.
563 * Unlike from_kprojid from_kprojid_munged never fails and always
564 * returns a valid projid. This makes from_kprojid_munged
565 * appropriate for use in syscalls like stat and where
566 * failing the system call and failing to provide a valid projid are
569 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
571 projid_t
from_kprojid_munged(struct user_namespace
*targ
, kprojid_t kprojid
)
574 projid
= from_kprojid(targ
, kprojid
);
576 if (projid
== (projid_t
) -1)
577 projid
= OVERFLOW_PROJID
;
580 EXPORT_SYMBOL(from_kprojid_munged
);
583 static int uid_m_show(struct seq_file
*seq
, void *v
)
585 struct user_namespace
*ns
= seq
->private;
586 struct uid_gid_extent
*extent
= v
;
587 struct user_namespace
*lower_ns
;
590 lower_ns
= seq_user_ns(seq
);
591 if ((lower_ns
== ns
) && lower_ns
->parent
)
592 lower_ns
= lower_ns
->parent
;
594 lower
= from_kuid(lower_ns
, KUIDT_INIT(extent
->lower_first
));
596 seq_printf(seq
, "%10u %10u %10u\n",
604 static int gid_m_show(struct seq_file
*seq
, void *v
)
606 struct user_namespace
*ns
= seq
->private;
607 struct uid_gid_extent
*extent
= v
;
608 struct user_namespace
*lower_ns
;
611 lower_ns
= seq_user_ns(seq
);
612 if ((lower_ns
== ns
) && lower_ns
->parent
)
613 lower_ns
= lower_ns
->parent
;
615 lower
= from_kgid(lower_ns
, KGIDT_INIT(extent
->lower_first
));
617 seq_printf(seq
, "%10u %10u %10u\n",
625 static int projid_m_show(struct seq_file
*seq
, void *v
)
627 struct user_namespace
*ns
= seq
->private;
628 struct uid_gid_extent
*extent
= v
;
629 struct user_namespace
*lower_ns
;
632 lower_ns
= seq_user_ns(seq
);
633 if ((lower_ns
== ns
) && lower_ns
->parent
)
634 lower_ns
= lower_ns
->parent
;
636 lower
= from_kprojid(lower_ns
, KPROJIDT_INIT(extent
->lower_first
));
638 seq_printf(seq
, "%10u %10u %10u\n",
646 static void *m_start(struct seq_file
*seq
, loff_t
*ppos
,
647 struct uid_gid_map
*map
)
650 unsigned extents
= map
->nr_extents
;
656 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
657 return &map
->extent
[pos
];
659 return &map
->forward
[pos
];
662 static void *uid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
664 struct user_namespace
*ns
= seq
->private;
666 return m_start(seq
, ppos
, &ns
->uid_map
);
669 static void *gid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
671 struct user_namespace
*ns
= seq
->private;
673 return m_start(seq
, ppos
, &ns
->gid_map
);
676 static void *projid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
678 struct user_namespace
*ns
= seq
->private;
680 return m_start(seq
, ppos
, &ns
->projid_map
);
683 static void *m_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
686 return seq
->op
->start(seq
, pos
);
689 static void m_stop(struct seq_file
*seq
, void *v
)
694 const struct seq_operations proc_uid_seq_operations
= {
695 .start
= uid_m_start
,
701 const struct seq_operations proc_gid_seq_operations
= {
702 .start
= gid_m_start
,
708 const struct seq_operations proc_projid_seq_operations
= {
709 .start
= projid_m_start
,
712 .show
= projid_m_show
,
715 static bool mappings_overlap(struct uid_gid_map
*new_map
,
716 struct uid_gid_extent
*extent
)
718 u32 upper_first
, lower_first
, upper_last
, lower_last
;
721 upper_first
= extent
->first
;
722 lower_first
= extent
->lower_first
;
723 upper_last
= upper_first
+ extent
->count
- 1;
724 lower_last
= lower_first
+ extent
->count
- 1;
726 for (idx
= 0; idx
< new_map
->nr_extents
; idx
++) {
727 u32 prev_upper_first
, prev_lower_first
;
728 u32 prev_upper_last
, prev_lower_last
;
729 struct uid_gid_extent
*prev
;
731 if (new_map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
732 prev
= &new_map
->extent
[idx
];
734 prev
= &new_map
->forward
[idx
];
736 prev_upper_first
= prev
->first
;
737 prev_lower_first
= prev
->lower_first
;
738 prev_upper_last
= prev_upper_first
+ prev
->count
- 1;
739 prev_lower_last
= prev_lower_first
+ prev
->count
- 1;
741 /* Does the upper range intersect a previous extent? */
742 if ((prev_upper_first
<= upper_last
) &&
743 (prev_upper_last
>= upper_first
))
746 /* Does the lower range intersect a previous extent? */
747 if ((prev_lower_first
<= lower_last
) &&
748 (prev_lower_last
>= lower_first
))
755 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
756 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
757 * UID_GID_MAP_MAX_BASE_EXTENTS.
759 static int insert_extent(struct uid_gid_map
*map
, struct uid_gid_extent
*extent
)
761 struct uid_gid_extent
*dest
;
763 if (map
->nr_extents
== UID_GID_MAP_MAX_BASE_EXTENTS
) {
764 struct uid_gid_extent
*forward
;
766 /* Allocate memory for 340 mappings. */
767 forward
= kmalloc_array(UID_GID_MAP_MAX_EXTENTS
,
768 sizeof(struct uid_gid_extent
),
773 /* Copy over memory. Only set up memory for the forward pointer.
774 * Defer the memory setup for the reverse pointer.
776 memcpy(forward
, map
->extent
,
777 map
->nr_extents
* sizeof(map
->extent
[0]));
779 map
->forward
= forward
;
783 if (map
->nr_extents
< UID_GID_MAP_MAX_BASE_EXTENTS
)
784 dest
= &map
->extent
[map
->nr_extents
];
786 dest
= &map
->forward
[map
->nr_extents
];
793 /* cmp function to sort() forward mappings */
794 static int cmp_extents_forward(const void *a
, const void *b
)
796 const struct uid_gid_extent
*e1
= a
;
797 const struct uid_gid_extent
*e2
= b
;
799 if (e1
->first
< e2
->first
)
802 if (e1
->first
> e2
->first
)
808 /* cmp function to sort() reverse mappings */
809 static int cmp_extents_reverse(const void *a
, const void *b
)
811 const struct uid_gid_extent
*e1
= a
;
812 const struct uid_gid_extent
*e2
= b
;
814 if (e1
->lower_first
< e2
->lower_first
)
817 if (e1
->lower_first
> e2
->lower_first
)
824 * sort_idmaps - Sorts an array of idmap entries.
825 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
827 static int sort_idmaps(struct uid_gid_map
*map
)
829 if (map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
832 /* Sort forward array. */
833 sort(map
->forward
, map
->nr_extents
, sizeof(struct uid_gid_extent
),
834 cmp_extents_forward
, NULL
);
836 /* Only copy the memory from forward we actually need. */
837 map
->reverse
= kmemdup(map
->forward
,
838 map
->nr_extents
* sizeof(struct uid_gid_extent
),
843 /* Sort reverse array. */
844 sort(map
->reverse
, map
->nr_extents
, sizeof(struct uid_gid_extent
),
845 cmp_extents_reverse
, NULL
);
850 static ssize_t
map_write(struct file
*file
, const char __user
*buf
,
851 size_t count
, loff_t
*ppos
,
853 struct uid_gid_map
*map
,
854 struct uid_gid_map
*parent_map
)
856 struct seq_file
*seq
= file
->private_data
;
857 struct user_namespace
*ns
= seq
->private;
858 struct uid_gid_map new_map
;
860 struct uid_gid_extent extent
;
861 char *kbuf
= NULL
, *pos
, *next_line
;
864 /* Only allow < page size writes at the beginning of the file */
865 if ((*ppos
!= 0) || (count
>= PAGE_SIZE
))
868 /* Slurp in the user data */
869 kbuf
= memdup_user_nul(buf
, count
);
871 return PTR_ERR(kbuf
);
874 * The userns_state_mutex serializes all writes to any given map.
876 * Any map is only ever written once.
878 * An id map fits within 1 cache line on most architectures.
880 * On read nothing needs to be done unless you are on an
881 * architecture with a crazy cache coherency model like alpha.
883 * There is a one time data dependency between reading the
884 * count of the extents and the values of the extents. The
885 * desired behavior is to see the values of the extents that
886 * were written before the count of the extents.
888 * To achieve this smp_wmb() is used on guarantee the write
889 * order and smp_rmb() is guaranteed that we don't have crazy
890 * architectures returning stale data.
892 mutex_lock(&userns_state_mutex
);
894 memset(&new_map
, 0, sizeof(struct uid_gid_map
));
897 /* Only allow one successful write to the map */
898 if (map
->nr_extents
!= 0)
902 * Adjusting namespace settings requires capabilities on the target.
904 if (cap_valid(cap_setid
) && !file_ns_capable(file
, ns
, CAP_SYS_ADMIN
))
907 /* Parse the user data */
910 for (; pos
; pos
= next_line
) {
912 /* Find the end of line and ensure I don't look past it */
913 next_line
= strchr(pos
, '\n');
917 if (*next_line
== '\0')
921 pos
= skip_spaces(pos
);
922 extent
.first
= simple_strtoul(pos
, &pos
, 10);
926 pos
= skip_spaces(pos
);
927 extent
.lower_first
= simple_strtoul(pos
, &pos
, 10);
931 pos
= skip_spaces(pos
);
932 extent
.count
= simple_strtoul(pos
, &pos
, 10);
933 if (*pos
&& !isspace(*pos
))
936 /* Verify there is not trailing junk on the line */
937 pos
= skip_spaces(pos
);
941 /* Verify we have been given valid starting values */
942 if ((extent
.first
== (u32
) -1) ||
943 (extent
.lower_first
== (u32
) -1))
946 /* Verify count is not zero and does not cause the
949 if ((extent
.first
+ extent
.count
) <= extent
.first
)
951 if ((extent
.lower_first
+ extent
.count
) <=
955 /* Do the ranges in extent overlap any previous extents? */
956 if (mappings_overlap(&new_map
, &extent
))
959 if ((new_map
.nr_extents
+ 1) == UID_GID_MAP_MAX_EXTENTS
&&
963 ret
= insert_extent(&new_map
, &extent
);
968 /* Be very certaint the new map actually exists */
969 if (new_map
.nr_extents
== 0)
973 /* Validate the user is allowed to use user id's mapped to. */
974 if (!new_idmap_permitted(file
, ns
, cap_setid
, &new_map
))
978 /* Map the lower ids from the parent user namespace to the
979 * kernel global id space.
981 for (idx
= 0; idx
< new_map
.nr_extents
; idx
++) {
982 struct uid_gid_extent
*e
;
985 if (new_map
.nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
986 e
= &new_map
.extent
[idx
];
988 e
= &new_map
.forward
[idx
];
990 lower_first
= map_id_range_down(parent_map
,
994 /* Fail if we can not map the specified extent to
995 * the kernel global id space.
997 if (lower_first
== (u32
) -1)
1000 e
->lower_first
= lower_first
;
1004 * If we want to use binary search for lookup, this clones the extent
1005 * array and sorts both copies.
1007 ret
= sort_idmaps(&new_map
);
1011 /* Install the map */
1012 if (new_map
.nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
) {
1013 memcpy(map
->extent
, new_map
.extent
,
1014 new_map
.nr_extents
* sizeof(new_map
.extent
[0]));
1016 map
->forward
= new_map
.forward
;
1017 map
->reverse
= new_map
.reverse
;
1020 map
->nr_extents
= new_map
.nr_extents
;
1025 if (ret
< 0 && new_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
1026 kfree(new_map
.forward
);
1027 kfree(new_map
.reverse
);
1028 map
->forward
= NULL
;
1029 map
->reverse
= NULL
;
1030 map
->nr_extents
= 0;
1033 mutex_unlock(&userns_state_mutex
);
1038 ssize_t
proc_uid_map_write(struct file
*file
, const char __user
*buf
,
1039 size_t size
, loff_t
*ppos
)
1041 struct seq_file
*seq
= file
->private_data
;
1042 struct user_namespace
*ns
= seq
->private;
1043 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1048 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1051 return map_write(file
, buf
, size
, ppos
, CAP_SETUID
,
1052 &ns
->uid_map
, &ns
->parent
->uid_map
);
1055 ssize_t
proc_gid_map_write(struct file
*file
, const char __user
*buf
,
1056 size_t size
, loff_t
*ppos
)
1058 struct seq_file
*seq
= file
->private_data
;
1059 struct user_namespace
*ns
= seq
->private;
1060 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1065 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1068 return map_write(file
, buf
, size
, ppos
, CAP_SETGID
,
1069 &ns
->gid_map
, &ns
->parent
->gid_map
);
1072 ssize_t
proc_projid_map_write(struct file
*file
, const char __user
*buf
,
1073 size_t size
, loff_t
*ppos
)
1075 struct seq_file
*seq
= file
->private_data
;
1076 struct user_namespace
*ns
= seq
->private;
1077 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1082 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1085 /* Anyone can set any valid project id no capability needed */
1086 return map_write(file
, buf
, size
, ppos
, -1,
1087 &ns
->projid_map
, &ns
->parent
->projid_map
);
1090 static bool new_idmap_permitted(const struct file
*file
,
1091 struct user_namespace
*ns
, int cap_setid
,
1092 struct uid_gid_map
*new_map
)
1094 const struct cred
*cred
= file
->f_cred
;
1095 /* Don't allow mappings that would allow anything that wouldn't
1096 * be allowed without the establishment of unprivileged mappings.
1098 if ((new_map
->nr_extents
== 1) && (new_map
->extent
[0].count
== 1) &&
1099 uid_eq(ns
->owner
, cred
->euid
)) {
1100 u32 id
= new_map
->extent
[0].lower_first
;
1101 if (cap_setid
== CAP_SETUID
) {
1102 kuid_t uid
= make_kuid(ns
->parent
, id
);
1103 if (uid_eq(uid
, cred
->euid
))
1105 } else if (cap_setid
== CAP_SETGID
) {
1106 kgid_t gid
= make_kgid(ns
->parent
, id
);
1107 if (!(ns
->flags
& USERNS_SETGROUPS_ALLOWED
) &&
1108 gid_eq(gid
, cred
->egid
))
1113 /* Allow anyone to set a mapping that doesn't require privilege */
1114 if (!cap_valid(cap_setid
))
1117 /* Allow the specified ids if we have the appropriate capability
1118 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1119 * And the opener of the id file also had the approprpiate capability.
1121 if (ns_capable(ns
->parent
, cap_setid
) &&
1122 file_ns_capable(file
, ns
->parent
, cap_setid
))
1128 int proc_setgroups_show(struct seq_file
*seq
, void *v
)
1130 struct user_namespace
*ns
= seq
->private;
1131 unsigned long userns_flags
= READ_ONCE(ns
->flags
);
1133 seq_printf(seq
, "%s\n",
1134 (userns_flags
& USERNS_SETGROUPS_ALLOWED
) ?
1139 ssize_t
proc_setgroups_write(struct file
*file
, const char __user
*buf
,
1140 size_t count
, loff_t
*ppos
)
1142 struct seq_file
*seq
= file
->private_data
;
1143 struct user_namespace
*ns
= seq
->private;
1145 bool setgroups_allowed
;
1148 /* Only allow a very narrow range of strings to be written */
1150 if ((*ppos
!= 0) || (count
>= sizeof(kbuf
)))
1153 /* What was written? */
1155 if (copy_from_user(kbuf
, buf
, count
))
1160 /* What is being requested? */
1162 if (strncmp(pos
, "allow", 5) == 0) {
1164 setgroups_allowed
= true;
1166 else if (strncmp(pos
, "deny", 4) == 0) {
1168 setgroups_allowed
= false;
1173 /* Verify there is not trailing junk on the line */
1174 pos
= skip_spaces(pos
);
1179 mutex_lock(&userns_state_mutex
);
1180 if (setgroups_allowed
) {
1181 /* Enabling setgroups after setgroups has been disabled
1184 if (!(ns
->flags
& USERNS_SETGROUPS_ALLOWED
))
1187 /* Permanently disabling setgroups after setgroups has
1188 * been enabled by writing the gid_map is not allowed.
1190 if (ns
->gid_map
.nr_extents
!= 0)
1192 ns
->flags
&= ~USERNS_SETGROUPS_ALLOWED
;
1194 mutex_unlock(&userns_state_mutex
);
1196 /* Report a successful write */
1202 mutex_unlock(&userns_state_mutex
);
1206 bool userns_may_setgroups(const struct user_namespace
*ns
)
1210 mutex_lock(&userns_state_mutex
);
1211 /* It is not safe to use setgroups until a gid mapping in
1212 * the user namespace has been established.
1214 allowed
= ns
->gid_map
.nr_extents
!= 0;
1215 /* Is setgroups allowed? */
1216 allowed
= allowed
&& (ns
->flags
& USERNS_SETGROUPS_ALLOWED
);
1217 mutex_unlock(&userns_state_mutex
);
1223 * Returns true if @child is the same namespace or a descendant of
1226 bool in_userns(const struct user_namespace
*ancestor
,
1227 const struct user_namespace
*child
)
1229 const struct user_namespace
*ns
;
1230 for (ns
= child
; ns
->level
> ancestor
->level
; ns
= ns
->parent
)
1232 return (ns
== ancestor
);
1235 bool current_in_userns(const struct user_namespace
*target_ns
)
1237 return in_userns(target_ns
, current_user_ns());
1239 EXPORT_SYMBOL(current_in_userns
);
1241 static inline struct user_namespace
*to_user_ns(struct ns_common
*ns
)
1243 return container_of(ns
, struct user_namespace
, ns
);
1246 static struct ns_common
*userns_get(struct task_struct
*task
)
1248 struct user_namespace
*user_ns
;
1251 user_ns
= get_user_ns(__task_cred(task
)->user_ns
);
1254 return user_ns
? &user_ns
->ns
: NULL
;
1257 static void userns_put(struct ns_common
*ns
)
1259 put_user_ns(to_user_ns(ns
));
1262 static int userns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
1264 struct user_namespace
*user_ns
= to_user_ns(ns
);
1267 /* Don't allow gaining capabilities by reentering
1268 * the same user namespace.
1270 if (user_ns
== current_user_ns())
1273 /* Tasks that share a thread group must share a user namespace */
1274 if (!thread_group_empty(current
))
1277 if (current
->fs
->users
!= 1)
1280 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
1283 cred
= prepare_creds();
1287 put_user_ns(cred
->user_ns
);
1288 set_cred_user_ns(cred
, get_user_ns(user_ns
));
1290 return commit_creds(cred
);
1293 struct ns_common
*ns_get_owner(struct ns_common
*ns
)
1295 struct user_namespace
*my_user_ns
= current_user_ns();
1296 struct user_namespace
*owner
, *p
;
1298 /* See if the owner is in the current user namespace */
1299 owner
= p
= ns
->ops
->owner(ns
);
1302 return ERR_PTR(-EPERM
);
1303 if (p
== my_user_ns
)
1308 return &get_user_ns(owner
)->ns
;
1311 static struct user_namespace
*userns_owner(struct ns_common
*ns
)
1313 return to_user_ns(ns
)->parent
;
1316 const struct proc_ns_operations userns_operations
= {
1318 .type
= CLONE_NEWUSER
,
1321 .install
= userns_install
,
1322 .owner
= userns_owner
,
1323 .get_parent
= ns_get_owner
,
1326 static __init
int user_namespaces_init(void)
1328 user_ns_cachep
= KMEM_CACHE(user_namespace
, SLAB_PANIC
);
1331 subsys_initcall(user_namespaces_init
);