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>
25 * sysctl determining whether unprivileged users may unshare a new
26 * userns. Allowed by default
28 int unprivileged_userns_clone
= 1;
30 static struct kmem_cache
*user_ns_cachep __read_mostly
;
31 static DEFINE_MUTEX(userns_state_mutex
);
33 static bool new_idmap_permitted(const struct file
*file
,
34 struct user_namespace
*ns
, int cap_setid
,
35 struct uid_gid_map
*map
);
36 static void free_user_ns(struct work_struct
*work
);
38 static struct ucounts
*inc_user_namespaces(struct user_namespace
*ns
, kuid_t uid
)
40 return inc_ucount(ns
, uid
, UCOUNT_USER_NAMESPACES
);
43 static void dec_user_namespaces(struct ucounts
*ucounts
)
45 return dec_ucount(ucounts
, UCOUNT_USER_NAMESPACES
);
48 static void set_cred_user_ns(struct cred
*cred
, struct user_namespace
*user_ns
)
50 /* Start with the same capabilities as init but useless for doing
51 * anything as the capabilities are bound to the new user namespace.
53 cred
->securebits
= SECUREBITS_DEFAULT
;
54 cred
->cap_inheritable
= CAP_EMPTY_SET
;
55 cred
->cap_permitted
= CAP_FULL_SET
;
56 cred
->cap_effective
= CAP_FULL_SET
;
57 cred
->cap_ambient
= CAP_EMPTY_SET
;
58 cred
->cap_bset
= CAP_FULL_SET
;
60 key_put(cred
->request_key_auth
);
61 cred
->request_key_auth
= NULL
;
63 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
64 cred
->user_ns
= user_ns
;
68 * Create a new user namespace, deriving the creator from the user in the
69 * passed credentials, and replacing that user with the new root user for the
72 * This is called by copy_creds(), which will finish setting the target task's
75 int create_user_ns(struct cred
*new)
77 struct user_namespace
*ns
, *parent_ns
= new->user_ns
;
78 kuid_t owner
= new->euid
;
79 kgid_t group
= new->egid
;
80 struct ucounts
*ucounts
;
84 if (parent_ns
->level
> 32)
87 ucounts
= inc_user_namespaces(parent_ns
, owner
);
92 * Verify that we can not violate the policy of which files
93 * may be accessed that is specified by the root directory,
94 * by verifying that the root directory is at the root of the
95 * mount namespace which allows all files to be accessed.
98 if (current_chrooted())
101 /* The creator needs a mapping in the parent user namespace
102 * or else we won't be able to reasonably tell userspace who
103 * created a user_namespace.
106 if (!kuid_has_mapping(parent_ns
, owner
) ||
107 !kgid_has_mapping(parent_ns
, group
))
111 ns
= kmem_cache_zalloc(user_ns_cachep
, GFP_KERNEL
);
115 ns
->parent_could_setfcap
= cap_raised(new->cap_effective
, CAP_SETFCAP
);
116 ret
= ns_alloc_inum(&ns
->ns
);
119 ns
->ns
.ops
= &userns_operations
;
121 refcount_set(&ns
->ns
.count
, 1);
122 /* Leave the new->user_ns reference with the new user namespace. */
123 ns
->parent
= parent_ns
;
124 ns
->level
= parent_ns
->level
+ 1;
127 INIT_WORK(&ns
->work
, free_user_ns
);
128 for (i
= 0; i
< MAX_PER_NAMESPACE_UCOUNTS
; i
++) {
129 ns
->ucount_max
[i
] = INT_MAX
;
131 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
));
132 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_MSGQUEUE
, rlimit(RLIMIT_MSGQUEUE
));
133 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_SIGPENDING
, rlimit(RLIMIT_SIGPENDING
));
134 set_rlimit_ucount_max(ns
, UCOUNT_RLIMIT_MEMLOCK
, rlimit(RLIMIT_MEMLOCK
));
135 ns
->ucounts
= ucounts
;
137 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
138 mutex_lock(&userns_state_mutex
);
139 ns
->flags
= parent_ns
->flags
;
140 mutex_unlock(&userns_state_mutex
);
143 INIT_LIST_HEAD(&ns
->keyring_name_list
);
144 init_rwsem(&ns
->keyring_sem
);
147 if (!setup_userns_sysctls(ns
))
150 set_cred_user_ns(new, ns
);
153 #ifdef CONFIG_PERSISTENT_KEYRINGS
154 key_put(ns
->persistent_keyring_register
);
156 ns_free_inum(&ns
->ns
);
158 kmem_cache_free(user_ns_cachep
, ns
);
160 dec_user_namespaces(ucounts
);
165 int unshare_userns(unsigned long unshare_flags
, struct cred
**new_cred
)
170 if (!(unshare_flags
& CLONE_NEWUSER
))
173 cred
= prepare_creds();
175 err
= create_user_ns(cred
);
185 static void free_user_ns(struct work_struct
*work
)
187 struct user_namespace
*parent
, *ns
=
188 container_of(work
, struct user_namespace
, work
);
191 struct ucounts
*ucounts
= ns
->ucounts
;
193 if (ns
->gid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
194 kfree(ns
->gid_map
.forward
);
195 kfree(ns
->gid_map
.reverse
);
197 if (ns
->uid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
198 kfree(ns
->uid_map
.forward
);
199 kfree(ns
->uid_map
.reverse
);
201 if (ns
->projid_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
202 kfree(ns
->projid_map
.forward
);
203 kfree(ns
->projid_map
.reverse
);
205 retire_userns_sysctls(ns
);
206 key_free_user_ns(ns
);
207 ns_free_inum(&ns
->ns
);
208 kmem_cache_free(user_ns_cachep
, ns
);
209 dec_user_namespaces(ucounts
);
211 } while (refcount_dec_and_test(&parent
->ns
.count
));
214 void __put_user_ns(struct user_namespace
*ns
)
216 schedule_work(&ns
->work
);
218 EXPORT_SYMBOL(__put_user_ns
);
221 * idmap_key struct holds the information necessary to find an idmapping in a
222 * sorted idmap array. It is passed to cmp_map_id() as first argument.
225 bool map_up
; /* true -> id from kid; false -> kid from id */
226 u32 id
; /* id to find */
227 u32 count
; /* == 0 unless used with map_id_range_down() */
231 * cmp_map_id - Function to be passed to bsearch() to find the requested
232 * idmapping. Expects struct idmap_key to be passed via @k.
234 static int cmp_map_id(const void *k
, const void *e
)
236 u32 first
, last
, id2
;
237 const struct idmap_key
*key
= k
;
238 const struct uid_gid_extent
*el
= e
;
240 id2
= key
->id
+ key
->count
- 1;
242 /* handle map_id_{down,up}() */
244 first
= el
->lower_first
;
248 last
= first
+ el
->count
- 1;
250 if (key
->id
>= first
&& key
->id
<= last
&&
251 (id2
>= first
&& id2
<= last
))
254 if (key
->id
< first
|| id2
< first
)
261 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
262 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
264 static struct uid_gid_extent
*
265 map_id_range_down_max(unsigned extents
, struct uid_gid_map
*map
, u32 id
, u32 count
)
267 struct idmap_key key
;
273 return bsearch(&key
, map
->forward
, extents
,
274 sizeof(struct uid_gid_extent
), cmp_map_id
);
278 * map_id_range_down_base - Find idmap via binary search in static extent array.
279 * Can only be called if number of mappings is equal or less than
280 * UID_GID_MAP_MAX_BASE_EXTENTS.
282 static struct uid_gid_extent
*
283 map_id_range_down_base(unsigned extents
, struct uid_gid_map
*map
, u32 id
, u32 count
)
286 u32 first
, last
, id2
;
288 id2
= id
+ count
- 1;
290 /* Find the matching extent */
291 for (idx
= 0; idx
< extents
; idx
++) {
292 first
= map
->extent
[idx
].first
;
293 last
= first
+ map
->extent
[idx
].count
- 1;
294 if (id
>= first
&& id
<= last
&&
295 (id2
>= first
&& id2
<= last
))
296 return &map
->extent
[idx
];
301 static u32
map_id_range_down(struct uid_gid_map
*map
, u32 id
, u32 count
)
303 struct uid_gid_extent
*extent
;
304 unsigned extents
= map
->nr_extents
;
307 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
308 extent
= map_id_range_down_base(extents
, map
, id
, count
);
310 extent
= map_id_range_down_max(extents
, map
, id
, count
);
312 /* Map the id or note failure */
314 id
= (id
- extent
->first
) + extent
->lower_first
;
321 static u32
map_id_down(struct uid_gid_map
*map
, u32 id
)
323 return map_id_range_down(map
, id
, 1);
327 * map_id_up_base - Find idmap via binary search in static extent array.
328 * Can only be called if number of mappings is equal or less than
329 * UID_GID_MAP_MAX_BASE_EXTENTS.
331 static struct uid_gid_extent
*
332 map_id_up_base(unsigned extents
, struct uid_gid_map
*map
, u32 id
)
337 /* Find the matching extent */
338 for (idx
= 0; idx
< extents
; idx
++) {
339 first
= map
->extent
[idx
].lower_first
;
340 last
= first
+ map
->extent
[idx
].count
- 1;
341 if (id
>= first
&& id
<= last
)
342 return &map
->extent
[idx
];
348 * map_id_up_max - Find idmap via binary search in ordered idmap array.
349 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
351 static struct uid_gid_extent
*
352 map_id_up_max(unsigned extents
, struct uid_gid_map
*map
, u32 id
)
354 struct idmap_key key
;
360 return bsearch(&key
, map
->reverse
, extents
,
361 sizeof(struct uid_gid_extent
), cmp_map_id
);
364 static u32
map_id_up(struct uid_gid_map
*map
, u32 id
)
366 struct uid_gid_extent
*extent
;
367 unsigned extents
= map
->nr_extents
;
370 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
371 extent
= map_id_up_base(extents
, map
, id
);
373 extent
= map_id_up_max(extents
, map
, id
);
375 /* Map the id or note failure */
377 id
= (id
- extent
->lower_first
) + extent
->first
;
385 * make_kuid - Map a user-namespace uid pair into a kuid.
386 * @ns: User namespace that the uid is in
387 * @uid: User identifier
389 * Maps a user-namespace uid pair into a kernel internal kuid,
390 * and returns that kuid.
392 * When there is no mapping defined for the user-namespace uid
393 * pair INVALID_UID is returned. Callers are expected to test
394 * for and handle INVALID_UID being returned. INVALID_UID
395 * may be tested for using uid_valid().
397 kuid_t
make_kuid(struct user_namespace
*ns
, uid_t uid
)
399 /* Map the uid to a global kernel uid */
400 return KUIDT_INIT(map_id_down(&ns
->uid_map
, uid
));
402 EXPORT_SYMBOL(make_kuid
);
405 * from_kuid - Create a uid from a kuid user-namespace pair.
406 * @targ: The user namespace we want a uid in.
407 * @kuid: The kernel internal uid to start with.
409 * Map @kuid into the user-namespace specified by @targ and
410 * return the resulting uid.
412 * There is always a mapping into the initial user_namespace.
414 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
416 uid_t
from_kuid(struct user_namespace
*targ
, kuid_t kuid
)
418 /* Map the uid from a global kernel uid */
419 return map_id_up(&targ
->uid_map
, __kuid_val(kuid
));
421 EXPORT_SYMBOL(from_kuid
);
424 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
425 * @targ: The user namespace we want a uid in.
426 * @kuid: The kernel internal uid to start with.
428 * Map @kuid into the user-namespace specified by @targ and
429 * return the resulting uid.
431 * There is always a mapping into the initial user_namespace.
433 * Unlike from_kuid from_kuid_munged never fails and always
434 * returns a valid uid. This makes from_kuid_munged appropriate
435 * for use in syscalls like stat and getuid where failing the
436 * system call and failing to provide a valid uid are not an
439 * If @kuid has no mapping in @targ overflowuid is returned.
441 uid_t
from_kuid_munged(struct user_namespace
*targ
, kuid_t kuid
)
444 uid
= from_kuid(targ
, kuid
);
446 if (uid
== (uid_t
) -1)
450 EXPORT_SYMBOL(from_kuid_munged
);
453 * make_kgid - Map a user-namespace gid pair into a kgid.
454 * @ns: User namespace that the gid is in
455 * @gid: group identifier
457 * Maps a user-namespace gid pair into a kernel internal kgid,
458 * and returns that kgid.
460 * When there is no mapping defined for the user-namespace gid
461 * pair INVALID_GID is returned. Callers are expected to test
462 * for and handle INVALID_GID being returned. INVALID_GID may be
463 * tested for using gid_valid().
465 kgid_t
make_kgid(struct user_namespace
*ns
, gid_t gid
)
467 /* Map the gid to a global kernel gid */
468 return KGIDT_INIT(map_id_down(&ns
->gid_map
, gid
));
470 EXPORT_SYMBOL(make_kgid
);
473 * from_kgid - Create a gid from a kgid user-namespace pair.
474 * @targ: The user namespace we want a gid in.
475 * @kgid: The kernel internal gid to start with.
477 * Map @kgid into the user-namespace specified by @targ and
478 * return the resulting gid.
480 * There is always a mapping into the initial user_namespace.
482 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
484 gid_t
from_kgid(struct user_namespace
*targ
, kgid_t kgid
)
486 /* Map the gid from a global kernel gid */
487 return map_id_up(&targ
->gid_map
, __kgid_val(kgid
));
489 EXPORT_SYMBOL(from_kgid
);
492 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
493 * @targ: The user namespace we want a gid in.
494 * @kgid: The kernel internal gid to start with.
496 * Map @kgid into the user-namespace specified by @targ and
497 * return the resulting gid.
499 * There is always a mapping into the initial user_namespace.
501 * Unlike from_kgid from_kgid_munged never fails and always
502 * returns a valid gid. This makes from_kgid_munged appropriate
503 * for use in syscalls like stat and getgid where failing the
504 * system call and failing to provide a valid gid are not options.
506 * If @kgid has no mapping in @targ overflowgid is returned.
508 gid_t
from_kgid_munged(struct user_namespace
*targ
, kgid_t kgid
)
511 gid
= from_kgid(targ
, kgid
);
513 if (gid
== (gid_t
) -1)
517 EXPORT_SYMBOL(from_kgid_munged
);
520 * make_kprojid - Map a user-namespace projid pair into a kprojid.
521 * @ns: User namespace that the projid is in
522 * @projid: Project identifier
524 * Maps a user-namespace uid pair into a kernel internal kuid,
525 * and returns that kuid.
527 * When there is no mapping defined for the user-namespace projid
528 * pair INVALID_PROJID is returned. Callers are expected to test
529 * for and handle INVALID_PROJID being returned. INVALID_PROJID
530 * may be tested for using projid_valid().
532 kprojid_t
make_kprojid(struct user_namespace
*ns
, projid_t projid
)
534 /* Map the uid to a global kernel uid */
535 return KPROJIDT_INIT(map_id_down(&ns
->projid_map
, projid
));
537 EXPORT_SYMBOL(make_kprojid
);
540 * from_kprojid - Create a projid from a kprojid user-namespace pair.
541 * @targ: The user namespace we want a projid in.
542 * @kprojid: The kernel internal project identifier to start with.
544 * Map @kprojid into the user-namespace specified by @targ and
545 * return the resulting projid.
547 * There is always a mapping into the initial user_namespace.
549 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
551 projid_t
from_kprojid(struct user_namespace
*targ
, kprojid_t kprojid
)
553 /* Map the uid from a global kernel uid */
554 return map_id_up(&targ
->projid_map
, __kprojid_val(kprojid
));
556 EXPORT_SYMBOL(from_kprojid
);
559 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
560 * @targ: The user namespace we want a projid in.
561 * @kprojid: The kernel internal projid to start with.
563 * Map @kprojid into the user-namespace specified by @targ and
564 * return the resulting projid.
566 * There is always a mapping into the initial user_namespace.
568 * Unlike from_kprojid from_kprojid_munged never fails and always
569 * returns a valid projid. This makes from_kprojid_munged
570 * appropriate for use in syscalls like stat and where
571 * failing the system call and failing to provide a valid projid are
574 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
576 projid_t
from_kprojid_munged(struct user_namespace
*targ
, kprojid_t kprojid
)
579 projid
= from_kprojid(targ
, kprojid
);
581 if (projid
== (projid_t
) -1)
582 projid
= OVERFLOW_PROJID
;
585 EXPORT_SYMBOL(from_kprojid_munged
);
588 static int uid_m_show(struct seq_file
*seq
, void *v
)
590 struct user_namespace
*ns
= seq
->private;
591 struct uid_gid_extent
*extent
= v
;
592 struct user_namespace
*lower_ns
;
595 lower_ns
= seq_user_ns(seq
);
596 if ((lower_ns
== ns
) && lower_ns
->parent
)
597 lower_ns
= lower_ns
->parent
;
599 lower
= from_kuid(lower_ns
, KUIDT_INIT(extent
->lower_first
));
601 seq_printf(seq
, "%10u %10u %10u\n",
609 static int gid_m_show(struct seq_file
*seq
, void *v
)
611 struct user_namespace
*ns
= seq
->private;
612 struct uid_gid_extent
*extent
= v
;
613 struct user_namespace
*lower_ns
;
616 lower_ns
= seq_user_ns(seq
);
617 if ((lower_ns
== ns
) && lower_ns
->parent
)
618 lower_ns
= lower_ns
->parent
;
620 lower
= from_kgid(lower_ns
, KGIDT_INIT(extent
->lower_first
));
622 seq_printf(seq
, "%10u %10u %10u\n",
630 static int projid_m_show(struct seq_file
*seq
, void *v
)
632 struct user_namespace
*ns
= seq
->private;
633 struct uid_gid_extent
*extent
= v
;
634 struct user_namespace
*lower_ns
;
637 lower_ns
= seq_user_ns(seq
);
638 if ((lower_ns
== ns
) && lower_ns
->parent
)
639 lower_ns
= lower_ns
->parent
;
641 lower
= from_kprojid(lower_ns
, KPROJIDT_INIT(extent
->lower_first
));
643 seq_printf(seq
, "%10u %10u %10u\n",
651 static void *m_start(struct seq_file
*seq
, loff_t
*ppos
,
652 struct uid_gid_map
*map
)
655 unsigned extents
= map
->nr_extents
;
661 if (extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
662 return &map
->extent
[pos
];
664 return &map
->forward
[pos
];
667 static void *uid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
669 struct user_namespace
*ns
= seq
->private;
671 return m_start(seq
, ppos
, &ns
->uid_map
);
674 static void *gid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
676 struct user_namespace
*ns
= seq
->private;
678 return m_start(seq
, ppos
, &ns
->gid_map
);
681 static void *projid_m_start(struct seq_file
*seq
, loff_t
*ppos
)
683 struct user_namespace
*ns
= seq
->private;
685 return m_start(seq
, ppos
, &ns
->projid_map
);
688 static void *m_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
691 return seq
->op
->start(seq
, pos
);
694 static void m_stop(struct seq_file
*seq
, void *v
)
699 const struct seq_operations proc_uid_seq_operations
= {
700 .start
= uid_m_start
,
706 const struct seq_operations proc_gid_seq_operations
= {
707 .start
= gid_m_start
,
713 const struct seq_operations proc_projid_seq_operations
= {
714 .start
= projid_m_start
,
717 .show
= projid_m_show
,
720 static bool mappings_overlap(struct uid_gid_map
*new_map
,
721 struct uid_gid_extent
*extent
)
723 u32 upper_first
, lower_first
, upper_last
, lower_last
;
726 upper_first
= extent
->first
;
727 lower_first
= extent
->lower_first
;
728 upper_last
= upper_first
+ extent
->count
- 1;
729 lower_last
= lower_first
+ extent
->count
- 1;
731 for (idx
= 0; idx
< new_map
->nr_extents
; idx
++) {
732 u32 prev_upper_first
, prev_lower_first
;
733 u32 prev_upper_last
, prev_lower_last
;
734 struct uid_gid_extent
*prev
;
736 if (new_map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
737 prev
= &new_map
->extent
[idx
];
739 prev
= &new_map
->forward
[idx
];
741 prev_upper_first
= prev
->first
;
742 prev_lower_first
= prev
->lower_first
;
743 prev_upper_last
= prev_upper_first
+ prev
->count
- 1;
744 prev_lower_last
= prev_lower_first
+ prev
->count
- 1;
746 /* Does the upper range intersect a previous extent? */
747 if ((prev_upper_first
<= upper_last
) &&
748 (prev_upper_last
>= upper_first
))
751 /* Does the lower range intersect a previous extent? */
752 if ((prev_lower_first
<= lower_last
) &&
753 (prev_lower_last
>= lower_first
))
760 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
761 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
762 * UID_GID_MAP_MAX_BASE_EXTENTS.
764 static int insert_extent(struct uid_gid_map
*map
, struct uid_gid_extent
*extent
)
766 struct uid_gid_extent
*dest
;
768 if (map
->nr_extents
== UID_GID_MAP_MAX_BASE_EXTENTS
) {
769 struct uid_gid_extent
*forward
;
771 /* Allocate memory for 340 mappings. */
772 forward
= kmalloc_array(UID_GID_MAP_MAX_EXTENTS
,
773 sizeof(struct uid_gid_extent
),
778 /* Copy over memory. Only set up memory for the forward pointer.
779 * Defer the memory setup for the reverse pointer.
781 memcpy(forward
, map
->extent
,
782 map
->nr_extents
* sizeof(map
->extent
[0]));
784 map
->forward
= forward
;
788 if (map
->nr_extents
< UID_GID_MAP_MAX_BASE_EXTENTS
)
789 dest
= &map
->extent
[map
->nr_extents
];
791 dest
= &map
->forward
[map
->nr_extents
];
798 /* cmp function to sort() forward mappings */
799 static int cmp_extents_forward(const void *a
, const void *b
)
801 const struct uid_gid_extent
*e1
= a
;
802 const struct uid_gid_extent
*e2
= b
;
804 if (e1
->first
< e2
->first
)
807 if (e1
->first
> e2
->first
)
813 /* cmp function to sort() reverse mappings */
814 static int cmp_extents_reverse(const void *a
, const void *b
)
816 const struct uid_gid_extent
*e1
= a
;
817 const struct uid_gid_extent
*e2
= b
;
819 if (e1
->lower_first
< e2
->lower_first
)
822 if (e1
->lower_first
> e2
->lower_first
)
829 * sort_idmaps - Sorts an array of idmap entries.
830 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
832 static int sort_idmaps(struct uid_gid_map
*map
)
834 if (map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
837 /* Sort forward array. */
838 sort(map
->forward
, map
->nr_extents
, sizeof(struct uid_gid_extent
),
839 cmp_extents_forward
, NULL
);
841 /* Only copy the memory from forward we actually need. */
842 map
->reverse
= kmemdup(map
->forward
,
843 map
->nr_extents
* sizeof(struct uid_gid_extent
),
848 /* Sort reverse array. */
849 sort(map
->reverse
, map
->nr_extents
, sizeof(struct uid_gid_extent
),
850 cmp_extents_reverse
, NULL
);
856 * verify_root_map() - check the uid 0 mapping
857 * @file: idmapping file
858 * @map_ns: user namespace of the target process
859 * @new_map: requested idmap
861 * If a process requests mapping parent uid 0 into the new ns, verify that the
862 * process writing the map had the CAP_SETFCAP capability as the target process
863 * will be able to write fscaps that are valid in ancestor user namespaces.
865 * Return: true if the mapping is allowed, false if not.
867 static bool verify_root_map(const struct file
*file
,
868 struct user_namespace
*map_ns
,
869 struct uid_gid_map
*new_map
)
872 const struct user_namespace
*file_ns
= file
->f_cred
->user_ns
;
873 struct uid_gid_extent
*extent0
= NULL
;
875 for (idx
= 0; idx
< new_map
->nr_extents
; idx
++) {
876 if (new_map
->nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
877 extent0
= &new_map
->extent
[idx
];
879 extent0
= &new_map
->forward
[idx
];
880 if (extent0
->lower_first
== 0)
889 if (map_ns
== file_ns
) {
890 /* The process unshared its ns and is writing to its own
891 * /proc/self/uid_map. User already has full capabilites in
892 * the new namespace. Verify that the parent had CAP_SETFCAP
895 if (!file_ns
->parent_could_setfcap
)
898 /* Process p1 is writing to uid_map of p2, who is in a child
899 * user namespace to p1's. Verify that the opener of the map
900 * file has CAP_SETFCAP against the parent of the new map
902 if (!file_ns_capable(file
, map_ns
->parent
, CAP_SETFCAP
))
909 static ssize_t
map_write(struct file
*file
, const char __user
*buf
,
910 size_t count
, loff_t
*ppos
,
912 struct uid_gid_map
*map
,
913 struct uid_gid_map
*parent_map
)
915 struct seq_file
*seq
= file
->private_data
;
916 struct user_namespace
*map_ns
= seq
->private;
917 struct uid_gid_map new_map
;
919 struct uid_gid_extent extent
;
920 char *kbuf
= NULL
, *pos
, *next_line
;
923 /* Only allow < page size writes at the beginning of the file */
924 if ((*ppos
!= 0) || (count
>= PAGE_SIZE
))
927 /* Slurp in the user data */
928 kbuf
= memdup_user_nul(buf
, count
);
930 return PTR_ERR(kbuf
);
933 * The userns_state_mutex serializes all writes to any given map.
935 * Any map is only ever written once.
937 * An id map fits within 1 cache line on most architectures.
939 * On read nothing needs to be done unless you are on an
940 * architecture with a crazy cache coherency model like alpha.
942 * There is a one time data dependency between reading the
943 * count of the extents and the values of the extents. The
944 * desired behavior is to see the values of the extents that
945 * were written before the count of the extents.
947 * To achieve this smp_wmb() is used on guarantee the write
948 * order and smp_rmb() is guaranteed that we don't have crazy
949 * architectures returning stale data.
951 mutex_lock(&userns_state_mutex
);
953 memset(&new_map
, 0, sizeof(struct uid_gid_map
));
956 /* Only allow one successful write to the map */
957 if (map
->nr_extents
!= 0)
961 * Adjusting namespace settings requires capabilities on the target.
963 if (cap_valid(cap_setid
) && !file_ns_capable(file
, map_ns
, CAP_SYS_ADMIN
))
966 /* Parse the user data */
969 for (; pos
; pos
= next_line
) {
971 /* Find the end of line and ensure I don't look past it */
972 next_line
= strchr(pos
, '\n');
976 if (*next_line
== '\0')
980 pos
= skip_spaces(pos
);
981 extent
.first
= simple_strtoul(pos
, &pos
, 10);
985 pos
= skip_spaces(pos
);
986 extent
.lower_first
= simple_strtoul(pos
, &pos
, 10);
990 pos
= skip_spaces(pos
);
991 extent
.count
= simple_strtoul(pos
, &pos
, 10);
992 if (*pos
&& !isspace(*pos
))
995 /* Verify there is not trailing junk on the line */
996 pos
= skip_spaces(pos
);
1000 /* Verify we have been given valid starting values */
1001 if ((extent
.first
== (u32
) -1) ||
1002 (extent
.lower_first
== (u32
) -1))
1005 /* Verify count is not zero and does not cause the
1008 if ((extent
.first
+ extent
.count
) <= extent
.first
)
1010 if ((extent
.lower_first
+ extent
.count
) <=
1014 /* Do the ranges in extent overlap any previous extents? */
1015 if (mappings_overlap(&new_map
, &extent
))
1018 if ((new_map
.nr_extents
+ 1) == UID_GID_MAP_MAX_EXTENTS
&&
1019 (next_line
!= NULL
))
1022 ret
= insert_extent(&new_map
, &extent
);
1027 /* Be very certain the new map actually exists */
1028 if (new_map
.nr_extents
== 0)
1032 /* Validate the user is allowed to use user id's mapped to. */
1033 if (!new_idmap_permitted(file
, map_ns
, cap_setid
, &new_map
))
1037 /* Map the lower ids from the parent user namespace to the
1038 * kernel global id space.
1040 for (idx
= 0; idx
< new_map
.nr_extents
; idx
++) {
1041 struct uid_gid_extent
*e
;
1044 if (new_map
.nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
)
1045 e
= &new_map
.extent
[idx
];
1047 e
= &new_map
.forward
[idx
];
1049 lower_first
= map_id_range_down(parent_map
,
1053 /* Fail if we can not map the specified extent to
1054 * the kernel global id space.
1056 if (lower_first
== (u32
) -1)
1059 e
->lower_first
= lower_first
;
1063 * If we want to use binary search for lookup, this clones the extent
1064 * array and sorts both copies.
1066 ret
= sort_idmaps(&new_map
);
1070 /* Install the map */
1071 if (new_map
.nr_extents
<= UID_GID_MAP_MAX_BASE_EXTENTS
) {
1072 memcpy(map
->extent
, new_map
.extent
,
1073 new_map
.nr_extents
* sizeof(new_map
.extent
[0]));
1075 map
->forward
= new_map
.forward
;
1076 map
->reverse
= new_map
.reverse
;
1079 map
->nr_extents
= new_map
.nr_extents
;
1084 if (ret
< 0 && new_map
.nr_extents
> UID_GID_MAP_MAX_BASE_EXTENTS
) {
1085 kfree(new_map
.forward
);
1086 kfree(new_map
.reverse
);
1087 map
->forward
= NULL
;
1088 map
->reverse
= NULL
;
1089 map
->nr_extents
= 0;
1092 mutex_unlock(&userns_state_mutex
);
1097 ssize_t
proc_uid_map_write(struct file
*file
, const char __user
*buf
,
1098 size_t size
, loff_t
*ppos
)
1100 struct seq_file
*seq
= file
->private_data
;
1101 struct user_namespace
*ns
= seq
->private;
1102 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1107 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1110 return map_write(file
, buf
, size
, ppos
, CAP_SETUID
,
1111 &ns
->uid_map
, &ns
->parent
->uid_map
);
1114 ssize_t
proc_gid_map_write(struct file
*file
, const char __user
*buf
,
1115 size_t size
, loff_t
*ppos
)
1117 struct seq_file
*seq
= file
->private_data
;
1118 struct user_namespace
*ns
= seq
->private;
1119 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1124 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1127 return map_write(file
, buf
, size
, ppos
, CAP_SETGID
,
1128 &ns
->gid_map
, &ns
->parent
->gid_map
);
1131 ssize_t
proc_projid_map_write(struct file
*file
, const char __user
*buf
,
1132 size_t size
, loff_t
*ppos
)
1134 struct seq_file
*seq
= file
->private_data
;
1135 struct user_namespace
*ns
= seq
->private;
1136 struct user_namespace
*seq_ns
= seq_user_ns(seq
);
1141 if ((seq_ns
!= ns
) && (seq_ns
!= ns
->parent
))
1144 /* Anyone can set any valid project id no capability needed */
1145 return map_write(file
, buf
, size
, ppos
, -1,
1146 &ns
->projid_map
, &ns
->parent
->projid_map
);
1149 static bool new_idmap_permitted(const struct file
*file
,
1150 struct user_namespace
*ns
, int cap_setid
,
1151 struct uid_gid_map
*new_map
)
1153 const struct cred
*cred
= file
->f_cred
;
1155 if (cap_setid
== CAP_SETUID
&& !verify_root_map(file
, ns
, new_map
))
1158 /* Don't allow mappings that would allow anything that wouldn't
1159 * be allowed without the establishment of unprivileged mappings.
1161 if ((new_map
->nr_extents
== 1) && (new_map
->extent
[0].count
== 1) &&
1162 uid_eq(ns
->owner
, cred
->euid
)) {
1163 u32 id
= new_map
->extent
[0].lower_first
;
1164 if (cap_setid
== CAP_SETUID
) {
1165 kuid_t uid
= make_kuid(ns
->parent
, id
);
1166 if (uid_eq(uid
, cred
->euid
))
1168 } else if (cap_setid
== CAP_SETGID
) {
1169 kgid_t gid
= make_kgid(ns
->parent
, id
);
1170 if (!(ns
->flags
& USERNS_SETGROUPS_ALLOWED
) &&
1171 gid_eq(gid
, cred
->egid
))
1176 /* Allow anyone to set a mapping that doesn't require privilege */
1177 if (!cap_valid(cap_setid
))
1180 /* Allow the specified ids if we have the appropriate capability
1181 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1182 * And the opener of the id file also has the appropriate capability.
1184 if (ns_capable(ns
->parent
, cap_setid
) &&
1185 file_ns_capable(file
, ns
->parent
, cap_setid
))
1191 int proc_setgroups_show(struct seq_file
*seq
, void *v
)
1193 struct user_namespace
*ns
= seq
->private;
1194 unsigned long userns_flags
= READ_ONCE(ns
->flags
);
1196 seq_printf(seq
, "%s\n",
1197 (userns_flags
& USERNS_SETGROUPS_ALLOWED
) ?
1202 ssize_t
proc_setgroups_write(struct file
*file
, const char __user
*buf
,
1203 size_t count
, loff_t
*ppos
)
1205 struct seq_file
*seq
= file
->private_data
;
1206 struct user_namespace
*ns
= seq
->private;
1208 bool setgroups_allowed
;
1211 /* Only allow a very narrow range of strings to be written */
1213 if ((*ppos
!= 0) || (count
>= sizeof(kbuf
)))
1216 /* What was written? */
1218 if (copy_from_user(kbuf
, buf
, count
))
1223 /* What is being requested? */
1225 if (strncmp(pos
, "allow", 5) == 0) {
1227 setgroups_allowed
= true;
1229 else if (strncmp(pos
, "deny", 4) == 0) {
1231 setgroups_allowed
= false;
1236 /* Verify there is not trailing junk on the line */
1237 pos
= skip_spaces(pos
);
1242 mutex_lock(&userns_state_mutex
);
1243 if (setgroups_allowed
) {
1244 /* Enabling setgroups after setgroups has been disabled
1247 if (!(ns
->flags
& USERNS_SETGROUPS_ALLOWED
))
1250 /* Permanently disabling setgroups after setgroups has
1251 * been enabled by writing the gid_map is not allowed.
1253 if (ns
->gid_map
.nr_extents
!= 0)
1255 ns
->flags
&= ~USERNS_SETGROUPS_ALLOWED
;
1257 mutex_unlock(&userns_state_mutex
);
1259 /* Report a successful write */
1265 mutex_unlock(&userns_state_mutex
);
1269 bool userns_may_setgroups(const struct user_namespace
*ns
)
1273 mutex_lock(&userns_state_mutex
);
1274 /* It is not safe to use setgroups until a gid mapping in
1275 * the user namespace has been established.
1277 allowed
= ns
->gid_map
.nr_extents
!= 0;
1278 /* Is setgroups allowed? */
1279 allowed
= allowed
&& (ns
->flags
& USERNS_SETGROUPS_ALLOWED
);
1280 mutex_unlock(&userns_state_mutex
);
1286 * Returns true if @child is the same namespace or a descendant of
1289 bool in_userns(const struct user_namespace
*ancestor
,
1290 const struct user_namespace
*child
)
1292 const struct user_namespace
*ns
;
1293 for (ns
= child
; ns
->level
> ancestor
->level
; ns
= ns
->parent
)
1295 return (ns
== ancestor
);
1298 bool current_in_userns(const struct user_namespace
*target_ns
)
1300 return in_userns(target_ns
, current_user_ns());
1302 EXPORT_SYMBOL(current_in_userns
);
1304 static inline struct user_namespace
*to_user_ns(struct ns_common
*ns
)
1306 return container_of(ns
, struct user_namespace
, ns
);
1309 static struct ns_common
*userns_get(struct task_struct
*task
)
1311 struct user_namespace
*user_ns
;
1314 user_ns
= get_user_ns(__task_cred(task
)->user_ns
);
1317 return user_ns
? &user_ns
->ns
: NULL
;
1320 static void userns_put(struct ns_common
*ns
)
1322 put_user_ns(to_user_ns(ns
));
1325 static int userns_install(struct nsset
*nsset
, struct ns_common
*ns
)
1327 struct user_namespace
*user_ns
= to_user_ns(ns
);
1330 /* Don't allow gaining capabilities by reentering
1331 * the same user namespace.
1333 if (user_ns
== current_user_ns())
1336 /* Tasks that share a thread group must share a user namespace */
1337 if (!thread_group_empty(current
))
1340 if (current
->fs
->users
!= 1)
1343 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
1346 cred
= nsset_cred(nsset
);
1350 put_user_ns(cred
->user_ns
);
1351 set_cred_user_ns(cred
, get_user_ns(user_ns
));
1353 if (set_cred_ucounts(cred
) < 0)
1359 struct ns_common
*ns_get_owner(struct ns_common
*ns
)
1361 struct user_namespace
*my_user_ns
= current_user_ns();
1362 struct user_namespace
*owner
, *p
;
1364 /* See if the owner is in the current user namespace */
1365 owner
= p
= ns
->ops
->owner(ns
);
1368 return ERR_PTR(-EPERM
);
1369 if (p
== my_user_ns
)
1374 return &get_user_ns(owner
)->ns
;
1377 static struct user_namespace
*userns_owner(struct ns_common
*ns
)
1379 return to_user_ns(ns
)->parent
;
1382 const struct proc_ns_operations userns_operations
= {
1384 .type
= CLONE_NEWUSER
,
1387 .install
= userns_install
,
1388 .owner
= userns_owner
,
1389 .get_parent
= ns_get_owner
,
1392 static __init
int user_namespaces_init(void)
1394 user_ns_cachep
= KMEM_CACHE(user_namespace
, SLAB_PANIC
| SLAB_ACCOUNT
);
1397 subsys_initcall(user_namespaces_init
);