1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/export.h>
4 #include <linux/nsproxy.h>
5 #include <linux/slab.h>
6 #include <linux/sched/signal.h>
7 #include <linux/user_namespace.h>
8 #include <linux/proc_ns.h>
9 #include <linux/highuid.h>
10 #include <linux/cred.h>
11 #include <linux/securebits.h>
12 #include <linux/keyctl.h>
13 #include <linux/key-type.h>
14 #include <keys/user-type.h>
15 #include <linux/seq_file.h>
17 #include <linux/uaccess.h>
18 #include <linux/ctype.h>
19 #include <linux/projid.h>
20 #include <linux/fs_struct.h>
21 #include <linux/bsearch.h>
22 #include <linux/sort.h>
24 static struct kmem_cache
*user_ns_cachep __read_mostly
;
25 static DEFINE_MUTEX(userns_state_mutex
);
27 static bool new_idmap_permitted(const struct file
*file
,
28 struct user_namespace
*ns
, int cap_setid
,
29 struct uid_gid_map
*map
);
30 static void free_user_ns(struct work_struct
*work
);
32 static struct ucounts
*inc_user_namespaces(struct user_namespace
*ns
, kuid_t uid
)
34 return inc_ucount(ns
, uid
, UCOUNT_USER_NAMESPACES
);
37 static void dec_user_namespaces(struct ucounts
*ucounts
)
39 return dec_ucount(ucounts
, UCOUNT_USER_NAMESPACES
);
42 static void set_cred_user_ns(struct cred
*cred
, struct user_namespace
*user_ns
)
44 /* Start with the same capabilities as init but useless for doing
45 * anything as the capabilities are bound to the new user namespace.
47 cred
->securebits
= SECUREBITS_DEFAULT
;
48 cred
->cap_inheritable
= CAP_EMPTY_SET
;
49 cred
->cap_permitted
= CAP_FULL_SET
;
50 cred
->cap_effective
= CAP_FULL_SET
;
51 cred
->cap_ambient
= CAP_EMPTY_SET
;
52 cred
->cap_bset
= CAP_FULL_SET
;
54 key_put(cred
->request_key_auth
);
55 cred
->request_key_auth
= NULL
;
57 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
58 cred
->user_ns
= user_ns
;
62 * Create a new user namespace, deriving the creator from the user in the
63 * passed credentials, and replacing that user with the new root user for the
66 * This is called by copy_creds(), which will finish setting the target task's
69 int create_user_ns(struct cred
*new)
71 struct user_namespace
*ns
, *parent_ns
= new->user_ns
;
72 kuid_t owner
= new->euid
;
73 kgid_t group
= new->egid
;
74 struct ucounts
*ucounts
;
78 if (parent_ns
->level
> 32)
81 ucounts
= inc_user_namespaces(parent_ns
, owner
);
86 * Verify that we can not violate the policy of which files
87 * may be accessed that is specified by the root directory,
88 * by verifying that the root directory is at the root of the
89 * mount namespace which allows all files to be accessed.
92 if (current_chrooted())
95 /* The creator needs a mapping in the parent user namespace
96 * or else we won't be able to reasonably tell userspace who
97 * created a user_namespace.
100 if (!kuid_has_mapping(parent_ns
, owner
) ||
101 !kgid_has_mapping(parent_ns
, group
))
105 ns
= kmem_cache_zalloc(user_ns_cachep
, GFP_KERNEL
);
109 ns
->parent_could_setfcap
= cap_raised(new->cap_effective
, CAP_SETFCAP
);
110 ret
= ns_alloc_inum(&ns
->ns
);
113 ns
->ns
.ops
= &userns_operations
;
115 refcount_set(&ns
->ns
.count
, 1);
116 /* Leave the new->user_ns reference with the new user namespace. */
117 ns
->parent
= parent_ns
;
118 ns
->level
= parent_ns
->level
+ 1;
121 INIT_WORK(&ns
->work
, free_user_ns
);
122 for (i
= 0; i
< MAX_PER_NAMESPACE_UCOUNTS
; i
++) {
123 ns
->ucount_max
[i
] = INT_MAX
;
125 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
));
126 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_MSGQUEUE
, rlimit(RLIMIT_MSGQUEUE
));
127 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_SIGPENDING
, rlimit(RLIMIT_SIGPENDING
));
128 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_MEMLOCK
, rlimit(RLIMIT_MEMLOCK
));
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
);
137 INIT_LIST_HEAD(&ns
->keyring_name_list
);
138 init_rwsem(&ns
->keyring_sem
);
141 if (!setup_userns_sysctls(ns
))
144 set_cred_user_ns(new, ns
);
147 #ifdef CONFIG_PERSISTENT_KEYRINGS
148 key_put(ns
->persistent_keyring_register
);
150 ns_free_inum(&ns
->ns
);
152 kmem_cache_free(user_ns_cachep
, ns
);
154 dec_user_namespaces(ucounts
);
159 int unshare_userns(unsigned long unshare_flags
, struct cred
**new_cred
)
164 if (!(unshare_flags
& CLONE_NEWUSER
))
167 cred
= prepare_creds();
169 err
= create_user_ns(cred
);
179 static void free_user_ns(struct work_struct
*work
)
181 struct user_namespace
*parent
, *ns
=
182 container_of(work
, struct user_namespace
, work
);
185 struct ucounts
*ucounts
= ns
->ucounts
;
187 if (ns
->gid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
188 kfree(ns
->gid_map
.forward
);
189 kfree(ns
->gid_map
.reverse
);
191 if (ns
->uid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
192 kfree(ns
->uid_map
.forward
);
193 kfree(ns
->uid_map
.reverse
);
195 if (ns
->projid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
196 kfree(ns
->projid_map
.forward
);
197 kfree(ns
->projid_map
.reverse
);
199 retire_userns_sysctls(ns
);
200 key_free_user_ns(ns
);
201 ns_free_inum(&ns
->ns
);
202 kmem_cache_free(user_ns_cachep
, ns
);
203 dec_user_namespaces(ucounts
);
205 } while (refcount_dec_and_test(&parent
->ns
.count
));
208 void __put_user_ns(struct user_namespace
*ns
)
210 schedule_work(&ns
->work
);
212 EXPORT_SYMBOL(__put_user_ns
);
215 * idmap_key struct holds the information necessary to find an idmapping in a
216 * sorted idmap array. It is passed to cmp_map_id() as first argument.
219 bool map_up
; /* true -> id from kid; false -> kid from id */
220 u32 id
; /* id to find */
221 u32 count
; /* == 0 unless used with map_id_range_down() */
225 * cmp_map_id - Function to be passed to bsearch() to find the requested
226 * idmapping. Expects struct idmap_key to be passed via @k.
228 static int cmp_map_id(const void *k
, const void *e
)
230 u32 first
, last
, id2
;
231 const struct idmap_key
*key
= k
;
232 const struct uid_gid_extent
*el
= e
;
234 id2
= key
->id
+ key
->count
- 1;
236 /* handle map_id_{down,up}() */
238 first
= el
->lower_first
;
242 last
= first
+ el
->count
- 1;
244 if (key
->id
>= first
&& key
->id
<= last
&&
245 (id2
>= first
&& id2
<= last
))
248 if (key
->id
< first
|| id2
< first
)
255 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
256 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
258 static struct uid_gid_extent
*
259 map_id_range_down_max(unsigned extents
, struct uid_gid_map
*map
, u32 id
, u32 count
)
261 struct idmap_key key
;
267 return bsearch(&key
, map
->forward
, extents
,
268 sizeof(struct uid_gid_extent
), cmp_map_id
);
272 * map_id_range_down_base - Find idmap via binary search in static extent array.
273 * Can only be called if number of mappings is equal or less than
274 * UID_GID_MAP_MAX_BASE_EXTENTS.
276 static struct uid_gid_extent
*
277 map_id_range_down_base(unsigned extents
, struct uid_gid_map
*map
, u32 id
, u32 count
)
280 u32 first
, last
, id2
;
282 id2
= id
+ count
- 1;
284 /* Find the matching extent */
285 for (idx
= 0; idx
< extents
; idx
++) {
286 first
= map
->extent
[idx
].first
;
287 last
= first
+ map
->extent
[idx
].count
- 1;
288 if (id
>= first
&& id
<= last
&&
289 (id2
>= first
&& id2
<= last
))
290 return &map
->extent
[idx
];
295 static u32
map_id_range_down(struct uid_gid_map
*map
, u32 id
, u32 count
)
297 struct uid_gid_extent
*extent
;
298 unsigned extents
= map
->nr_extents
;
301 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
302 extent
= map_id_range_down_base(extents
, map
, id
, count
);
304 extent
= map_id_range_down_max(extents
, map
, id
, count
);
306 /* Map the id or note failure */
308 id
= (id
- extent
->first
) + extent
->lower_first
;
315 static u32
map_id_down(struct uid_gid_map
*map
, u32 id
)
317 return map_id_range_down(map
, id
, 1);
321 * map_id_up_base - Find idmap via binary search in static extent array.
322 * Can only be called if number of mappings is equal or less than
323 * UID_GID_MAP_MAX_BASE_EXTENTS.
325 static struct uid_gid_extent
*
326 map_id_up_base(unsigned extents
, struct uid_gid_map
*map
, u32 id
)
331 /* Find the matching extent */
332 for (idx
= 0; idx
< extents
; idx
++) {
333 first
= map
->extent
[idx
].lower_first
;
334 last
= first
+ map
->extent
[idx
].count
- 1;
335 if (id
>= first
&& id
<= last
)
336 return &map
->extent
[idx
];
342 * map_id_up_max - Find idmap via binary search in ordered idmap array.
343 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
345 static struct uid_gid_extent
*
346 map_id_up_max(unsigned extents
, struct uid_gid_map
*map
, u32 id
)
348 struct idmap_key key
;
354 return bsearch(&key
, map
->reverse
, extents
,
355 sizeof(struct uid_gid_extent
), cmp_map_id
);
358 static u32
map_id_up(struct uid_gid_map
*map
, u32 id
)
360 struct uid_gid_extent
*extent
;
361 unsigned extents
= map
->nr_extents
;
364 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
365 extent
= map_id_up_base(extents
, map
, id
);
367 extent
= map_id_up_max(extents
, map
, id
);
369 /* Map the id or note failure */
371 id
= (id
- extent
->lower_first
) + extent
->first
;
379 * make_kuid - Map a user-namespace uid pair into a kuid.
380 * @ns: User namespace that the uid is in
381 * @uid: User identifier
383 * Maps a user-namespace uid pair into a kernel internal kuid,
384 * and returns that kuid.
386 * When there is no mapping defined for the user-namespace uid
387 * pair INVALID_UID is returned. Callers are expected to test
388 * for and handle INVALID_UID being returned. INVALID_UID
389 * may be tested for using uid_valid().
391 kuid_t
make_kuid(struct user_namespace
*ns
, uid_t uid
)
393 /* Map the uid to a global kernel uid */
394 return KUIDT_INIT(map_id_down(&ns
->uid_map
, uid
));
396 EXPORT_SYMBOL(make_kuid
);
399 * from_kuid - Create a uid from a kuid user-namespace pair.
400 * @targ: The user namespace we want a uid in.
401 * @kuid: The kernel internal uid to start with.
403 * Map @kuid into the user-namespace specified by @targ and
404 * return the resulting uid.
406 * There is always a mapping into the initial user_namespace.
408 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
410 uid_t
from_kuid(struct user_namespace
*targ
, kuid_t kuid
)
412 /* Map the uid from a global kernel uid */
413 return map_id_up(&targ
->uid_map
, __kuid_val(kuid
));
415 EXPORT_SYMBOL(from_kuid
);
418 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
419 * @targ: The user namespace we want a uid in.
420 * @kuid: The kernel internal uid to start with.
422 * Map @kuid into the user-namespace specified by @targ and
423 * return the resulting uid.
425 * There is always a mapping into the initial user_namespace.
427 * Unlike from_kuid from_kuid_munged never fails and always
428 * returns a valid uid. This makes from_kuid_munged appropriate
429 * for use in syscalls like stat and getuid where failing the
430 * system call and failing to provide a valid uid are not an
433 * If @kuid has no mapping in @targ overflowuid is returned.
435 uid_t
from_kuid_munged(struct user_namespace
*targ
, kuid_t kuid
)
438 uid
= from_kuid(targ
, kuid
);
440 if (uid
== (uid_t
) -1)
444 EXPORT_SYMBOL(from_kuid_munged
);
447 * make_kgid - Map a user-namespace gid pair into a kgid.
448 * @ns: User namespace that the gid is in
449 * @gid: group identifier
451 * Maps a user-namespace gid pair into a kernel internal kgid,
452 * and returns that kgid.
454 * When there is no mapping defined for the user-namespace gid
455 * pair INVALID_GID is returned. Callers are expected to test
456 * for and handle INVALID_GID being returned. INVALID_GID may be
457 * tested for using gid_valid().
459 kgid_t
make_kgid(struct user_namespace
*ns
, gid_t gid
)
461 /* Map the gid to a global kernel gid */
462 return KGIDT_INIT(map_id_down(&ns
->gid_map
, gid
));
464 EXPORT_SYMBOL(make_kgid
);
467 * from_kgid - Create a gid from a kgid user-namespace pair.
468 * @targ: The user namespace we want a gid in.
469 * @kgid: The kernel internal gid to start with.
471 * Map @kgid into the user-namespace specified by @targ and
472 * return the resulting gid.
474 * There is always a mapping into the initial user_namespace.
476 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
478 gid_t
from_kgid(struct user_namespace
*targ
, kgid_t kgid
)
480 /* Map the gid from a global kernel gid */
481 return map_id_up(&targ
->gid_map
, __kgid_val(kgid
));
483 EXPORT_SYMBOL(from_kgid
);
486 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
487 * @targ: The user namespace we want a gid in.
488 * @kgid: The kernel internal gid to start with.
490 * Map @kgid into the user-namespace specified by @targ and
491 * return the resulting gid.
493 * There is always a mapping into the initial user_namespace.
495 * Unlike from_kgid from_kgid_munged never fails and always
496 * returns a valid gid. This makes from_kgid_munged appropriate
497 * for use in syscalls like stat and getgid where failing the
498 * system call and failing to provide a valid gid are not options.
500 * If @kgid has no mapping in @targ overflowgid is returned.
502 gid_t
from_kgid_munged(struct user_namespace
*targ
, kgid_t kgid
)
505 gid
= from_kgid(targ
, kgid
);
507 if (gid
== (gid_t
) -1)
511 EXPORT_SYMBOL(from_kgid_munged
);
514 * make_kprojid - Map a user-namespace projid pair into a kprojid.
515 * @ns: User namespace that the projid is in
516 * @projid: Project identifier
518 * Maps a user-namespace uid pair into a kernel internal kuid,
519 * and returns that kuid.
521 * When there is no mapping defined for the user-namespace projid
522 * pair INVALID_PROJID is returned. Callers are expected to test
523 * for and handle INVALID_PROJID being returned. INVALID_PROJID
524 * may be tested for using projid_valid().
526 kprojid_t
make_kprojid(struct user_namespace
*ns
, projid_t projid
)
528 /* Map the uid to a global kernel uid */
529 return KPROJIDT_INIT(map_id_down(&ns
->projid_map
, projid
));
531 EXPORT_SYMBOL(make_kprojid
);
534 * from_kprojid - Create a projid from a kprojid user-namespace pair.
535 * @targ: The user namespace we want a projid in.
536 * @kprojid: The kernel internal project identifier to start with.
538 * Map @kprojid into the user-namespace specified by @targ and
539 * return the resulting projid.
541 * There is always a mapping into the initial user_namespace.
543 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
545 projid_t
from_kprojid(struct user_namespace
*targ
, kprojid_t kprojid
)
547 /* Map the uid from a global kernel uid */
548 return map_id_up(&targ
->projid_map
, __kprojid_val(kprojid
));
550 EXPORT_SYMBOL(from_kprojid
);
553 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
554 * @targ: The user namespace we want a projid in.
555 * @kprojid: The kernel internal projid to start with.
557 * Map @kprojid into the user-namespace specified by @targ and
558 * return the resulting projid.
560 * There is always a mapping into the initial user_namespace.
562 * Unlike from_kprojid from_kprojid_munged never fails and always
563 * returns a valid projid. This makes from_kprojid_munged
564 * appropriate for use in syscalls like stat and where
565 * failing the system call and failing to provide a valid projid are
568 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
570 projid_t
from_kprojid_munged(struct user_namespace
*targ
, kprojid_t kprojid
)
573 projid
= from_kprojid(targ
, kprojid
);
575 if (projid
== (projid_t
) -1)
576 projid
= OVERFLOW_PROJID
;
579 EXPORT_SYMBOL(from_kprojid_munged
);
582 static int uid_m_show(struct seq_file
*seq
, void *v
)
584 struct user_namespace
*ns
= seq
->private;
585 struct uid_gid_extent
*extent
= v
;
586 struct user_namespace
*lower_ns
;
589 lower_ns
= seq_user_ns(seq
);
590 if ((lower_ns
== ns
) && lower_ns
->parent
)
591 lower_ns
= lower_ns
->parent
;
593 lower
= from_kuid(lower_ns
, KUIDT_INIT(extent
->lower_first
));
595 seq_printf(seq
, "%10u %10u %10u\n",
603 static int gid_m_show(struct seq_file
*seq
, void *v
)
605 struct user_namespace
*ns
= seq
->private;
606 struct uid_gid_extent
*extent
= v
;
607 struct user_namespace
*lower_ns
;
610 lower_ns
= seq_user_ns(seq
);
611 if ((lower_ns
== ns
) && lower_ns
->parent
)
612 lower_ns
= lower_ns
->parent
;
614 lower
= from_kgid(lower_ns
, KGIDT_INIT(extent
->lower_first
));
616 seq_printf(seq
, "%10u %10u %10u\n",
624 static int projid_m_show(struct seq_file
*seq
, void *v
)
626 struct user_namespace
*ns
= seq
->private;
627 struct uid_gid_extent
*extent
= v
;
628 struct user_namespace
*lower_ns
;
631 lower_ns
= seq_user_ns(seq
);
632 if ((lower_ns
== ns
) && lower_ns
->parent
)
633 lower_ns
= lower_ns
->parent
;
635 lower
= from_kprojid(lower_ns
, KPROJIDT_INIT(extent
->lower_first
));
637 seq_printf(seq
, "%10u %10u %10u\n",
645 static void *m_start(struct seq_file
*seq
, loff_t
*ppos
,
646 struct uid_gid_map
*map
)
649 unsigned extents
= map
->nr_extents
;
655 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
656 return &map
->extent
[pos
];
658 return &map
->forward
[pos
];
661 static void *uid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
663 struct user_namespace
*ns
= seq
->private;
665 return m_start(seq
, ppos
, &ns
->uid_map
);
668 static void *gid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
670 struct user_namespace
*ns
= seq
->private;
672 return m_start(seq
, ppos
, &ns
->gid_map
);
675 static void *projid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
677 struct user_namespace
*ns
= seq
->private;
679 return m_start(seq
, ppos
, &ns
->projid_map
);
682 static void *m_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
685 return seq
->op
->start(seq
, pos
);
688 static void m_stop(struct seq_file
*seq
, void *v
)
693 const struct seq_operations proc_uid_seq_operations
= {
694 .start
= uid_m_start
,
700 const struct seq_operations proc_gid_seq_operations
= {
701 .start
= gid_m_start
,
707 const struct seq_operations proc_projid_seq_operations
= {
708 .start
= projid_m_start
,
711 .show
= projid_m_show
,
714 static bool mappings_overlap(struct uid_gid_map
*new_map
,
715 struct uid_gid_extent
*extent
)
717 u32 upper_first
, lower_first
, upper_last
, lower_last
;
720 upper_first
= extent
->first
;
721 lower_first
= extent
->lower_first
;
722 upper_last
= upper_first
+ extent
->count
- 1;
723 lower_last
= lower_first
+ extent
->count
- 1;
725 for (idx
= 0; idx
< new_map
->nr_extents
; idx
++) {
726 u32 prev_upper_first
, prev_lower_first
;
727 u32 prev_upper_last
, prev_lower_last
;
728 struct uid_gid_extent
*prev
;
730 if (new_map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
731 prev
= &new_map
->extent
[idx
];
733 prev
= &new_map
->forward
[idx
];
735 prev_upper_first
= prev
->first
;
736 prev_lower_first
= prev
->lower_first
;
737 prev_upper_last
= prev_upper_first
+ prev
->count
- 1;
738 prev_lower_last
= prev_lower_first
+ prev
->count
- 1;
740 /* Does the upper range intersect a previous extent? */
741 if ((prev_upper_first
<= upper_last
) &&
742 (prev_upper_last
>= upper_first
))
745 /* Does the lower range intersect a previous extent? */
746 if ((prev_lower_first
<= lower_last
) &&
747 (prev_lower_last
>= lower_first
))
754 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
755 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
756 * UID_GID_MAP_MAX_BASE_EXTENTS.
758 static int insert_extent(struct uid_gid_map
*map
, struct uid_gid_extent
*extent
)
760 struct uid_gid_extent
*dest
;
762 if (map
->nr_extents
== UID_GID_MAP_MAX_BASE_EXTENTS
) {
763 struct uid_gid_extent
*forward
;
765 /* Allocate memory for 340 mappings. */
766 forward
= kmalloc_array(UID_GID_MAP_MAX_EXTENTS
,
767 sizeof(struct uid_gid_extent
),
772 /* Copy over memory. Only set up memory for the forward pointer.
773 * Defer the memory setup for the reverse pointer.
775 memcpy(forward
, map
->extent
,
776 map
->nr_extents
* sizeof(map
->extent
[0]));
778 map
->forward
= forward
;
782 if (map
->nr_extents
< UID_GID_MAP_MAX_BASE_EXTENTS
)
783 dest
= &map
->extent
[map
->nr_extents
];
785 dest
= &map
->forward
[map
->nr_extents
];
792 /* cmp function to sort() forward mappings */
793 static int cmp_extents_forward(const void *a
, const void *b
)
795 const struct uid_gid_extent
*e1
= a
;
796 const struct uid_gid_extent
*e2
= b
;
798 if (e1
->first
< e2
->first
)
801 if (e1
->first
> e2
->first
)
807 /* cmp function to sort() reverse mappings */
808 static int cmp_extents_reverse(const void *a
, const void *b
)
810 const struct uid_gid_extent
*e1
= a
;
811 const struct uid_gid_extent
*e2
= b
;
813 if (e1
->lower_first
< e2
->lower_first
)
816 if (e1
->lower_first
> e2
->lower_first
)
823 * sort_idmaps - Sorts an array of idmap entries.
824 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
826 static int sort_idmaps(struct uid_gid_map
*map
)
828 if (map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
831 /* Sort forward array. */
832 sort(map
->forward
, map
->nr_extents
, sizeof(struct uid_gid_extent
),
833 cmp_extents_forward
, NULL
);
835 /* Only copy the memory from forward we actually need. */
836 map
->reverse
= kmemdup(map
->forward
,
837 map
->nr_extents
* sizeof(struct uid_gid_extent
),
842 /* Sort reverse array. */
843 sort(map
->reverse
, map
->nr_extents
, sizeof(struct uid_gid_extent
),
844 cmp_extents_reverse
, NULL
);
850 * verify_root_map() - check the uid 0 mapping
851 * @file: idmapping file
852 * @map_ns: user namespace of the target process
853 * @new_map: requested idmap
855 * If a process requests mapping parent uid 0 into the new ns, verify that the
856 * process writing the map had the CAP_SETFCAP capability as the target process
857 * will be able to write fscaps that are valid in ancestor user namespaces.
859 * Return: true if the mapping is allowed, false if not.
861 static bool verify_root_map(const struct file
*file
,
862 struct user_namespace
*map_ns
,
863 struct uid_gid_map
*new_map
)
866 const struct user_namespace
*file_ns
= file
->f_cred
->user_ns
;
867 struct uid_gid_extent
*extent0
= NULL
;
869 for (idx
= 0; idx
< new_map
->nr_extents
; idx
++) {
870 if (new_map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
871 extent0
= &new_map
->extent
[idx
];
873 extent0
= &new_map
->forward
[idx
];
874 if (extent0
->lower_first
== 0)
883 if (map_ns
== file_ns
) {
884 /* The process unshared its ns and is writing to its own
885 * /proc/self/uid_map. User already has full capabilites in
886 * the new namespace. Verify that the parent had CAP_SETFCAP
889 if (!file_ns
->parent_could_setfcap
)
892 /* Process p1 is writing to uid_map of p2, who is in a child
893 * user namespace to p1's. Verify that the opener of the map
894 * file has CAP_SETFCAP against the parent of the new map
896 if (!file_ns_capable(file
, map_ns
->parent
, CAP_SETFCAP
))
903 static ssize_t
map_write(struct file
*file
, const char __user
*buf
,
904 size_t count
, loff_t
*ppos
,
906 struct uid_gid_map
*map
,
907 struct uid_gid_map
*parent_map
)
909 struct seq_file
*seq
= file
->private_data
;
910 struct user_namespace
*map_ns
= seq
->private;
911 struct uid_gid_map new_map
;
913 struct uid_gid_extent extent
;
914 char *kbuf
= NULL
, *pos
, *next_line
;
917 /* Only allow < page size writes at the beginning of the file */
918 if ((*ppos
!= 0) || (count
>= PAGE_SIZE
))
921 /* Slurp in the user data */
922 kbuf
= memdup_user_nul(buf
, count
);
924 return PTR_ERR(kbuf
);
927 * The userns_state_mutex serializes all writes to any given map.
929 * Any map is only ever written once.
931 * An id map fits within 1 cache line on most architectures.
933 * On read nothing needs to be done unless you are on an
934 * architecture with a crazy cache coherency model like alpha.
936 * There is a one time data dependency between reading the
937 * count of the extents and the values of the extents. The
938 * desired behavior is to see the values of the extents that
939 * were written before the count of the extents.
941 * To achieve this smp_wmb() is used on guarantee the write
942 * order and smp_rmb() is guaranteed that we don't have crazy
943 * architectures returning stale data.
945 mutex_lock(&userns_state_mutex
);
947 memset(&new_map
, 0, sizeof(struct uid_gid_map
));
950 /* Only allow one successful write to the map */
951 if (map
->nr_extents
!= 0)
955 * Adjusting namespace settings requires capabilities on the target.
957 if (cap_valid(cap_setid
) && !file_ns_capable(file
, map_ns
, CAP_SYS_ADMIN
))
960 /* Parse the user data */
963 for (; pos
; pos
= next_line
) {
965 /* Find the end of line and ensure I don't look past it */
966 next_line
= strchr(pos
, '\n');
970 if (*next_line
== '\0')
974 pos
= skip_spaces(pos
);
975 extent
.first
= simple_strtoul(pos
, &pos
, 10);
979 pos
= skip_spaces(pos
);
980 extent
.lower_first
= simple_strtoul(pos
, &pos
, 10);
984 pos
= skip_spaces(pos
);
985 extent
.count
= simple_strtoul(pos
, &pos
, 10);
986 if (*pos
&& !isspace(*pos
))
989 /* Verify there is not trailing junk on the line */
990 pos
= skip_spaces(pos
);
994 /* Verify we have been given valid starting values */
995 if ((extent
.first
== (u32
) -1) ||
996 (extent
.lower_first
== (u32
) -1))
999 /* Verify count is not zero and does not cause the
1002 if ((extent
.first
+ extent
.count
) <= extent
.first
)
1004 if ((extent
.lower_first
+ extent
.count
) <=
1008 /* Do the ranges in extent overlap any previous extents? */
1009 if (mappings_overlap(&new_map
, &extent
))
1012 if ((new_map
.nr_extents
+ 1) == UID_GID_MAP_MAX_EXTENTS
&&
1013 (next_line
!= NULL
))
1016 ret
= insert_extent(&new_map
, &extent
);
1021 /* Be very certain the new map actually exists */
1022 if (new_map
.nr_extents
== 0)
1026 /* Validate the user is allowed to use user id's mapped to. */
1027 if (!new_idmap_permitted(file
, map_ns
, cap_setid
, &new_map
))
1031 /* Map the lower ids from the parent user namespace to the
1032 * kernel global id space.
1034 for (idx
= 0; idx
< new_map
.nr_extents
; idx
++) {
1035 struct uid_gid_extent
*e
;
1038 if (new_map
.nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
1039 e
= &new_map
.extent
[idx
];
1041 e
= &new_map
.forward
[idx
];
1043 lower_first
= map_id_range_down(parent_map
,
1047 /* Fail if we can not map the specified extent to
1048 * the kernel global id space.
1050 if (lower_first
== (u32
) -1)
1053 e
->lower_first
= lower_first
;
1057 * If we want to use binary search for lookup, this clones the extent
1058 * array and sorts both copies.
1060 ret
= sort_idmaps(&new_map
);
1064 /* Install the map */
1065 if (new_map
.nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
) {
1066 memcpy(map
->extent
, new_map
.extent
,
1067 new_map
.nr_extents
* sizeof(new_map
.extent
[0]));
1069 map
->forward
= new_map
.forward
;
1070 map
->reverse
= new_map
.reverse
;
1073 map
->nr_extents
= new_map
.nr_extents
;
1078 if (ret
< 0 && new_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
1079 kfree(new_map
.forward
);
1080 kfree(new_map
.reverse
);
1081 map
->forward
= NULL
;
1082 map
->reverse
= NULL
;
1083 map
->nr_extents
= 0;
1086 mutex_unlock(&userns_state_mutex
);
1091 ssize_t
proc_uid_map_write(struct file
*file
, const char __user
*buf
,
1092 size_t size
, loff_t
*ppos
)
1094 struct seq_file
*seq
= file
->private_data
;
1095 struct user_namespace
*ns
= seq
->private;
1096 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1101 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1104 return map_write(file
, buf
, size
, ppos
, CAP_SETUID
,
1105 &ns
->uid_map
, &ns
->parent
->uid_map
);
1108 ssize_t
proc_gid_map_write(struct file
*file
, const char __user
*buf
,
1109 size_t size
, loff_t
*ppos
)
1111 struct seq_file
*seq
= file
->private_data
;
1112 struct user_namespace
*ns
= seq
->private;
1113 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1118 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1121 return map_write(file
, buf
, size
, ppos
, CAP_SETGID
,
1122 &ns
->gid_map
, &ns
->parent
->gid_map
);
1125 ssize_t
proc_projid_map_write(struct file
*file
, const char __user
*buf
,
1126 size_t size
, loff_t
*ppos
)
1128 struct seq_file
*seq
= file
->private_data
;
1129 struct user_namespace
*ns
= seq
->private;
1130 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1135 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1138 /* Anyone can set any valid project id no capability needed */
1139 return map_write(file
, buf
, size
, ppos
, -1,
1140 &ns
->projid_map
, &ns
->parent
->projid_map
);
1143 static bool new_idmap_permitted(const struct file
*file
,
1144 struct user_namespace
*ns
, int cap_setid
,
1145 struct uid_gid_map
*new_map
)
1147 const struct cred
*cred
= file
->f_cred
;
1149 if (cap_setid
== CAP_SETUID
&& !verify_root_map(file
, ns
, new_map
))
1152 /* Don't allow mappings that would allow anything that wouldn't
1153 * be allowed without the establishment of unprivileged mappings.
1155 if ((new_map
->nr_extents
== 1) && (new_map
->extent
[0].count
== 1) &&
1156 uid_eq(ns
->owner
, cred
->euid
)) {
1157 u32 id
= new_map
->extent
[0].lower_first
;
1158 if (cap_setid
== CAP_SETUID
) {
1159 kuid_t uid
= make_kuid(ns
->parent
, id
);
1160 if (uid_eq(uid
, cred
->euid
))
1162 } else if (cap_setid
== CAP_SETGID
) {
1163 kgid_t gid
= make_kgid(ns
->parent
, id
);
1164 if (!(ns
->flags
& USERNS_SETGROUPS_ALLOWED
) &&
1165 gid_eq(gid
, cred
->egid
))
1170 /* Allow anyone to set a mapping that doesn't require privilege */
1171 if (!cap_valid(cap_setid
))
1174 /* Allow the specified ids if we have the appropriate capability
1175 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1176 * And the opener of the id file also has the appropriate capability.
1178 if (ns_capable(ns
->parent
, cap_setid
) &&
1179 file_ns_capable(file
, ns
->parent
, cap_setid
))
1185 int proc_setgroups_show(struct seq_file
*seq
, void *v
)
1187 struct user_namespace
*ns
= seq
->private;
1188 unsigned long userns_flags
= READ_ONCE(ns
->flags
);
1190 seq_printf(seq
, "%s\n",
1191 (userns_flags
& USERNS_SETGROUPS_ALLOWED
) ?
1196 ssize_t
proc_setgroups_write(struct file
*file
, const char __user
*buf
,
1197 size_t count
, loff_t
*ppos
)
1199 struct seq_file
*seq
= file
->private_data
;
1200 struct user_namespace
*ns
= seq
->private;
1202 bool setgroups_allowed
;
1205 /* Only allow a very narrow range of strings to be written */
1207 if ((*ppos
!= 0) || (count
>= sizeof(kbuf
)))
1210 /* What was written? */
1212 if (copy_from_user(kbuf
, buf
, count
))
1217 /* What is being requested? */
1219 if (strncmp(pos
, "allow", 5) == 0) {
1221 setgroups_allowed
= true;
1223 else if (strncmp(pos
, "deny", 4) == 0) {
1225 setgroups_allowed
= false;
1230 /* Verify there is not trailing junk on the line */
1231 pos
= skip_spaces(pos
);
1236 mutex_lock(&userns_state_mutex
);
1237 if (setgroups_allowed
) {
1238 /* Enabling setgroups after setgroups has been disabled
1241 if (!(ns
->flags
& USERNS_SETGROUPS_ALLOWED
))
1244 /* Permanently disabling setgroups after setgroups has
1245 * been enabled by writing the gid_map is not allowed.
1247 if (ns
->gid_map
.nr_extents
!= 0)
1249 ns
->flags
&= ~USERNS_SETGROUPS_ALLOWED
;
1251 mutex_unlock(&userns_state_mutex
);
1253 /* Report a successful write */
1259 mutex_unlock(&userns_state_mutex
);
1263 bool userns_may_setgroups(const struct user_namespace
*ns
)
1267 mutex_lock(&userns_state_mutex
);
1268 /* It is not safe to use setgroups until a gid mapping in
1269 * the user namespace has been established.
1271 allowed
= ns
->gid_map
.nr_extents
!= 0;
1272 /* Is setgroups allowed? */
1273 allowed
= allowed
&& (ns
->flags
& USERNS_SETGROUPS_ALLOWED
);
1274 mutex_unlock(&userns_state_mutex
);
1280 * Returns true if @child is the same namespace or a descendant of
1283 bool in_userns(const struct user_namespace
*ancestor
,
1284 const struct user_namespace
*child
)
1286 const struct user_namespace
*ns
;
1287 for (ns
= child
; ns
->level
> ancestor
->level
; ns
= ns
->parent
)
1289 return (ns
== ancestor
);
1292 bool current_in_userns(const struct user_namespace
*target_ns
)
1294 return in_userns(target_ns
, current_user_ns());
1296 EXPORT_SYMBOL(current_in_userns
);
1298 static inline struct user_namespace
*to_user_ns(struct ns_common
*ns
)
1300 return container_of(ns
, struct user_namespace
, ns
);
1303 static struct ns_common
*userns_get(struct task_struct
*task
)
1305 struct user_namespace
*user_ns
;
1308 user_ns
= get_user_ns(__task_cred(task
)->user_ns
);
1311 return user_ns
? &user_ns
->ns
: NULL
;
1314 static void userns_put(struct ns_common
*ns
)
1316 put_user_ns(to_user_ns(ns
));
1319 static int userns_install(struct nsset
*nsset
, struct ns_common
*ns
)
1321 struct user_namespace
*user_ns
= to_user_ns(ns
);
1324 /* Don't allow gaining capabilities by reentering
1325 * the same user namespace.
1327 if (user_ns
== current_user_ns())
1330 /* Tasks that share a thread group must share a user namespace */
1331 if (!thread_group_empty(current
))
1334 if (current
->fs
->users
!= 1)
1337 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
1340 cred
= nsset_cred(nsset
);
1344 put_user_ns(cred
->user_ns
);
1345 set_cred_user_ns(cred
, get_user_ns(user_ns
));
1347 if (set_cred_ucounts(cred
) < 0)
1353 struct ns_common
*ns_get_owner(struct ns_common
*ns
)
1355 struct user_namespace
*my_user_ns
= current_user_ns();
1356 struct user_namespace
*owner
, *p
;
1358 /* See if the owner is in the current user namespace */
1359 owner
= p
= ns
->ops
->owner(ns
);
1362 return ERR_PTR(-EPERM
);
1363 if (p
== my_user_ns
)
1368 return &get_user_ns(owner
)->ns
;
1371 static struct user_namespace
*userns_owner(struct ns_common
*ns
)
1373 return to_user_ns(ns
)->parent
;
1376 const struct proc_ns_operations userns_operations
= {
1378 .type
= CLONE_NEWUSER
,
1381 .install
= userns_install
,
1382 .owner
= userns_owner
,
1383 .get_parent
= ns_get_owner
,
1386 static __init
int user_namespaces_init(void)
1388 user_ns_cachep
= KMEM_CACHE(user_namespace
, SLAB_PANIC
| SLAB_ACCOUNT
);
1391 subsys_initcall(user_namespaces_init
);