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;
61 static unsigned long enforced_nproc_rlimit(void)
63 unsigned long limit = RLIM_INFINITY;
65 /* Is RLIMIT_NPROC currently enforced? */
66 if (!uid_eq(current_uid(), GLOBAL_ROOT_UID) ||
67 (current_user_ns() != &init_user_ns))
68 limit = rlimit(RLIMIT_NPROC);
74 * Create a new user namespace, deriving the creator from the user in the
75 * passed credentials, and replacing that user with the new root user for the
78 * This is called by copy_creds(), which will finish setting the target task's
81 int create_user_ns(struct cred *new)
83 struct user_namespace *ns, *parent_ns = new->user_ns;
84 kuid_t owner = new->euid;
85 kgid_t group = new->egid;
86 struct ucounts *ucounts;
90 if (parent_ns->level > 32)
93 ucounts = inc_user_namespaces(parent_ns, owner);
98 * Verify that we can not violate the policy of which files
99 * may be accessed that is specified by the root directory,
100 * by verifying that the root directory is at the root of the
101 * mount namespace which allows all files to be accessed.
104 if (current_chrooted())
107 /* The creator needs a mapping in the parent user namespace
108 * or else we won't be able to reasonably tell userspace who
109 * created a user_namespace.
112 if (!kuid_has_mapping(parent_ns, owner) ||
113 !kgid_has_mapping(parent_ns, group))
117 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
121 ns->parent_could_setfcap = cap_raised(new->cap_effective, CAP_SETFCAP);
122 ret = ns_alloc_inum(&ns->ns);
125 ns->ns.ops = &userns_operations;
127 refcount_set(&ns->ns.count, 1);
128 /* Leave the new->user_ns reference with the new user namespace. */
129 ns->parent = parent_ns;
130 ns->level = parent_ns->level + 1;
133 INIT_WORK(&ns->work, free_user_ns);
134 for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++) {
135 ns->ucount_max[i] = INT_MAX;
137 set_rlimit_ucount_max(ns, UCOUNT_RLIMIT_NPROC, enforced_nproc_rlimit());
138 set_rlimit_ucount_max(ns, UCOUNT_RLIMIT_MSGQUEUE, rlimit(RLIMIT_MSGQUEUE));
139 set_rlimit_ucount_max(ns, UCOUNT_RLIMIT_SIGPENDING, rlimit(RLIMIT_SIGPENDING));
140 set_rlimit_ucount_max(ns, UCOUNT_RLIMIT_MEMLOCK, rlimit(RLIMIT_MEMLOCK));
141 ns->ucounts = ucounts;
143 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
144 mutex_lock(&userns_state_mutex);
145 ns->flags = parent_ns->flags;
146 mutex_unlock(&userns_state_mutex);
149 INIT_LIST_HEAD(&ns->keyring_name_list);
150 init_rwsem(&ns->keyring_sem);
153 if (!setup_userns_sysctls(ns))
156 set_cred_user_ns(new, ns);
159 #ifdef CONFIG_PERSISTENT_KEYRINGS
160 key_put(ns->persistent_keyring_register);
162 ns_free_inum(&ns->ns);
164 kmem_cache_free(user_ns_cachep, ns);
166 dec_user_namespaces(ucounts);
171 int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
176 if (!(unshare_flags & CLONE_NEWUSER))
179 cred = prepare_creds();
181 err = create_user_ns(cred);
191 static void free_user_ns(struct work_struct *work)
193 struct user_namespace *parent, *ns =
194 container_of(work, struct user_namespace, work);
197 struct ucounts *ucounts = ns->ucounts;
199 if (ns->gid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
200 kfree(ns->gid_map.forward);
201 kfree(ns->gid_map.reverse);
203 if (ns->uid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
204 kfree(ns->uid_map.forward);
205 kfree(ns->uid_map.reverse);
207 if (ns->projid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
208 kfree(ns->projid_map.forward);
209 kfree(ns->projid_map.reverse);
211 retire_userns_sysctls(ns);
212 key_free_user_ns(ns);
213 ns_free_inum(&ns->ns);
214 kmem_cache_free(user_ns_cachep, ns);
215 dec_user_namespaces(ucounts);
217 } while (refcount_dec_and_test(&parent->ns.count));
220 void __put_user_ns(struct user_namespace *ns)
222 schedule_work(&ns->work);
224 EXPORT_SYMBOL(__put_user_ns);
227 * idmap_key struct holds the information necessary to find an idmapping in a
228 * sorted idmap array. It is passed to cmp_map_id() as first argument.
231 bool map_up; /* true -> id from kid; false -> kid from id */
232 u32 id; /* id to find */
233 u32 count; /* == 0 unless used with map_id_range_down() */
237 * cmp_map_id - Function to be passed to bsearch() to find the requested
238 * idmapping. Expects struct idmap_key to be passed via @k.
240 static int cmp_map_id(const void *k, const void *e)
242 u32 first, last, id2;
243 const struct idmap_key *key = k;
244 const struct uid_gid_extent *el = e;
246 id2 = key->id + key->count - 1;
248 /* handle map_id_{down,up}() */
250 first = el->lower_first;
254 last = first + el->count - 1;
256 if (key->id >= first && key->id <= last &&
257 (id2 >= first && id2 <= last))
260 if (key->id < first || id2 < first)
267 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
268 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
270 static struct uid_gid_extent *
271 map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
273 struct idmap_key key;
279 return bsearch(&key, map->forward, extents,
280 sizeof(struct uid_gid_extent), cmp_map_id);
284 * map_id_range_down_base - Find idmap via binary search in static extent array.
285 * Can only be called if number of mappings is equal or less than
286 * UID_GID_MAP_MAX_BASE_EXTENTS.
288 static struct uid_gid_extent *
289 map_id_range_down_base(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
292 u32 first, last, id2;
294 id2 = id + count - 1;
296 /* Find the matching extent */
297 for (idx = 0; idx < extents; idx++) {
298 first = map->extent[idx].first;
299 last = first + map->extent[idx].count - 1;
300 if (id >= first && id <= last &&
301 (id2 >= first && id2 <= last))
302 return &map->extent[idx];
307 static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
309 struct uid_gid_extent *extent;
310 unsigned extents = map->nr_extents;
313 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
314 extent = map_id_range_down_base(extents, map, id, count);
316 extent = map_id_range_down_max(extents, map, id, count);
318 /* Map the id or note failure */
320 id = (id - extent->first) + extent->lower_first;
327 static u32 map_id_down(struct uid_gid_map *map, u32 id)
329 return map_id_range_down(map, id, 1);
333 * map_id_up_base - Find idmap via binary search in static extent array.
334 * Can only be called if number of mappings is equal or less than
335 * UID_GID_MAP_MAX_BASE_EXTENTS.
337 static struct uid_gid_extent *
338 map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
343 /* Find the matching extent */
344 for (idx = 0; idx < extents; idx++) {
345 first = map->extent[idx].lower_first;
346 last = first + map->extent[idx].count - 1;
347 if (id >= first && id <= last)
348 return &map->extent[idx];
354 * map_id_up_max - Find idmap via binary search in ordered idmap array.
355 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
357 static struct uid_gid_extent *
358 map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
360 struct idmap_key key;
366 return bsearch(&key, map->reverse, extents,
367 sizeof(struct uid_gid_extent), cmp_map_id);
370 static u32 map_id_up(struct uid_gid_map *map, u32 id)
372 struct uid_gid_extent *extent;
373 unsigned extents = map->nr_extents;
376 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
377 extent = map_id_up_base(extents, map, id);
379 extent = map_id_up_max(extents, map, id);
381 /* Map the id or note failure */
383 id = (id - extent->lower_first) + extent->first;
391 * make_kuid - Map a user-namespace uid pair into a kuid.
392 * @ns: User namespace that the uid is in
393 * @uid: User identifier
395 * Maps a user-namespace uid pair into a kernel internal kuid,
396 * and returns that kuid.
398 * When there is no mapping defined for the user-namespace uid
399 * pair INVALID_UID is returned. Callers are expected to test
400 * for and handle INVALID_UID being returned. INVALID_UID
401 * may be tested for using uid_valid().
403 kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
405 /* Map the uid to a global kernel uid */
406 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
408 EXPORT_SYMBOL(make_kuid);
411 * from_kuid - Create a uid from a kuid user-namespace pair.
412 * @targ: The user namespace we want a uid in.
413 * @kuid: The kernel internal uid to start with.
415 * Map @kuid into the user-namespace specified by @targ and
416 * return the resulting uid.
418 * There is always a mapping into the initial user_namespace.
420 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
422 uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
424 /* Map the uid from a global kernel uid */
425 return map_id_up(&targ->uid_map, __kuid_val(kuid));
427 EXPORT_SYMBOL(from_kuid);
430 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
431 * @targ: The user namespace we want a uid in.
432 * @kuid: The kernel internal uid to start with.
434 * Map @kuid into the user-namespace specified by @targ and
435 * return the resulting uid.
437 * There is always a mapping into the initial user_namespace.
439 * Unlike from_kuid from_kuid_munged never fails and always
440 * returns a valid uid. This makes from_kuid_munged appropriate
441 * for use in syscalls like stat and getuid where failing the
442 * system call and failing to provide a valid uid are not an
445 * If @kuid has no mapping in @targ overflowuid is returned.
447 uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
450 uid = from_kuid(targ, kuid);
452 if (uid == (uid_t) -1)
456 EXPORT_SYMBOL(from_kuid_munged);
459 * make_kgid - Map a user-namespace gid pair into a kgid.
460 * @ns: User namespace that the gid is in
461 * @gid: group identifier
463 * Maps a user-namespace gid pair into a kernel internal kgid,
464 * and returns that kgid.
466 * When there is no mapping defined for the user-namespace gid
467 * pair INVALID_GID is returned. Callers are expected to test
468 * for and handle INVALID_GID being returned. INVALID_GID may be
469 * tested for using gid_valid().
471 kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
473 /* Map the gid to a global kernel gid */
474 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
476 EXPORT_SYMBOL(make_kgid);
479 * from_kgid - Create a gid from a kgid user-namespace pair.
480 * @targ: The user namespace we want a gid in.
481 * @kgid: The kernel internal gid to start with.
483 * Map @kgid into the user-namespace specified by @targ and
484 * return the resulting gid.
486 * There is always a mapping into the initial user_namespace.
488 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
490 gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
492 /* Map the gid from a global kernel gid */
493 return map_id_up(&targ->gid_map, __kgid_val(kgid));
495 EXPORT_SYMBOL(from_kgid);
498 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
499 * @targ: The user namespace we want a gid in.
500 * @kgid: The kernel internal gid to start with.
502 * Map @kgid into the user-namespace specified by @targ and
503 * return the resulting gid.
505 * There is always a mapping into the initial user_namespace.
507 * Unlike from_kgid from_kgid_munged never fails and always
508 * returns a valid gid. This makes from_kgid_munged appropriate
509 * for use in syscalls like stat and getgid where failing the
510 * system call and failing to provide a valid gid are not options.
512 * If @kgid has no mapping in @targ overflowgid is returned.
514 gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
517 gid = from_kgid(targ, kgid);
519 if (gid == (gid_t) -1)
523 EXPORT_SYMBOL(from_kgid_munged);
526 * make_kprojid - Map a user-namespace projid pair into a kprojid.
527 * @ns: User namespace that the projid is in
528 * @projid: Project identifier
530 * Maps a user-namespace uid pair into a kernel internal kuid,
531 * and returns that kuid.
533 * When there is no mapping defined for the user-namespace projid
534 * pair INVALID_PROJID is returned. Callers are expected to test
535 * for and handle INVALID_PROJID being returned. INVALID_PROJID
536 * may be tested for using projid_valid().
538 kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
540 /* Map the uid to a global kernel uid */
541 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
543 EXPORT_SYMBOL(make_kprojid);
546 * from_kprojid - Create a projid from a kprojid user-namespace pair.
547 * @targ: The user namespace we want a projid in.
548 * @kprojid: The kernel internal project identifier to start with.
550 * Map @kprojid into the user-namespace specified by @targ and
551 * return the resulting projid.
553 * There is always a mapping into the initial user_namespace.
555 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
557 projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
559 /* Map the uid from a global kernel uid */
560 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
562 EXPORT_SYMBOL(from_kprojid);
565 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
566 * @targ: The user namespace we want a projid in.
567 * @kprojid: The kernel internal projid to start with.
569 * Map @kprojid into the user-namespace specified by @targ and
570 * return the resulting projid.
572 * There is always a mapping into the initial user_namespace.
574 * Unlike from_kprojid from_kprojid_munged never fails and always
575 * returns a valid projid. This makes from_kprojid_munged
576 * appropriate for use in syscalls like stat and where
577 * failing the system call and failing to provide a valid projid are
580 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
582 projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
585 projid = from_kprojid(targ, kprojid);
587 if (projid == (projid_t) -1)
588 projid = OVERFLOW_PROJID;
591 EXPORT_SYMBOL(from_kprojid_munged);
594 static int uid_m_show(struct seq_file *seq, void *v)
596 struct user_namespace *ns = seq->private;
597 struct uid_gid_extent *extent = v;
598 struct user_namespace *lower_ns;
601 lower_ns = seq_user_ns(seq);
602 if ((lower_ns == ns) && lower_ns->parent)
603 lower_ns = lower_ns->parent;
605 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
607 seq_printf(seq, "%10u %10u %10u\n",
615 static int gid_m_show(struct seq_file *seq, void *v)
617 struct user_namespace *ns = seq->private;
618 struct uid_gid_extent *extent = v;
619 struct user_namespace *lower_ns;
622 lower_ns = seq_user_ns(seq);
623 if ((lower_ns == ns) && lower_ns->parent)
624 lower_ns = lower_ns->parent;
626 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
628 seq_printf(seq, "%10u %10u %10u\n",
636 static int projid_m_show(struct seq_file *seq, void *v)
638 struct user_namespace *ns = seq->private;
639 struct uid_gid_extent *extent = v;
640 struct user_namespace *lower_ns;
643 lower_ns = seq_user_ns(seq);
644 if ((lower_ns == ns) && lower_ns->parent)
645 lower_ns = lower_ns->parent;
647 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
649 seq_printf(seq, "%10u %10u %10u\n",
657 static void *m_start(struct seq_file *seq, loff_t *ppos,
658 struct uid_gid_map *map)
661 unsigned extents = map->nr_extents;
667 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
668 return &map->extent[pos];
670 return &map->forward[pos];
673 static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
675 struct user_namespace *ns = seq->private;
677 return m_start(seq, ppos, &ns->uid_map);
680 static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
682 struct user_namespace *ns = seq->private;
684 return m_start(seq, ppos, &ns->gid_map);
687 static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
689 struct user_namespace *ns = seq->private;
691 return m_start(seq, ppos, &ns->projid_map);
694 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
697 return seq->op->start(seq, pos);
700 static void m_stop(struct seq_file *seq, void *v)
705 const struct seq_operations proc_uid_seq_operations = {
706 .start = uid_m_start,
712 const struct seq_operations proc_gid_seq_operations = {
713 .start = gid_m_start,
719 const struct seq_operations proc_projid_seq_operations = {
720 .start = projid_m_start,
723 .show = projid_m_show,
726 static bool mappings_overlap(struct uid_gid_map *new_map,
727 struct uid_gid_extent *extent)
729 u32 upper_first, lower_first, upper_last, lower_last;
732 upper_first = extent->first;
733 lower_first = extent->lower_first;
734 upper_last = upper_first + extent->count - 1;
735 lower_last = lower_first + extent->count - 1;
737 for (idx = 0; idx < new_map->nr_extents; idx++) {
738 u32 prev_upper_first, prev_lower_first;
739 u32 prev_upper_last, prev_lower_last;
740 struct uid_gid_extent *prev;
742 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
743 prev = &new_map->extent[idx];
745 prev = &new_map->forward[idx];
747 prev_upper_first = prev->first;
748 prev_lower_first = prev->lower_first;
749 prev_upper_last = prev_upper_first + prev->count - 1;
750 prev_lower_last = prev_lower_first + prev->count - 1;
752 /* Does the upper range intersect a previous extent? */
753 if ((prev_upper_first <= upper_last) &&
754 (prev_upper_last >= upper_first))
757 /* Does the lower range intersect a previous extent? */
758 if ((prev_lower_first <= lower_last) &&
759 (prev_lower_last >= lower_first))
766 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
767 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
768 * UID_GID_MAP_MAX_BASE_EXTENTS.
770 static int insert_extent(struct uid_gid_map *map, struct uid_gid_extent *extent)
772 struct uid_gid_extent *dest;
774 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
775 struct uid_gid_extent *forward;
777 /* Allocate memory for 340 mappings. */
778 forward = kmalloc_array(UID_GID_MAP_MAX_EXTENTS,
779 sizeof(struct uid_gid_extent),
784 /* Copy over memory. Only set up memory for the forward pointer.
785 * Defer the memory setup for the reverse pointer.
787 memcpy(forward, map->extent,
788 map->nr_extents * sizeof(map->extent[0]));
790 map->forward = forward;
794 if (map->nr_extents < UID_GID_MAP_MAX_BASE_EXTENTS)
795 dest = &map->extent[map->nr_extents];
797 dest = &map->forward[map->nr_extents];
804 /* cmp function to sort() forward mappings */
805 static int cmp_extents_forward(const void *a, const void *b)
807 const struct uid_gid_extent *e1 = a;
808 const struct uid_gid_extent *e2 = b;
810 if (e1->first < e2->first)
813 if (e1->first > e2->first)
819 /* cmp function to sort() reverse mappings */
820 static int cmp_extents_reverse(const void *a, const void *b)
822 const struct uid_gid_extent *e1 = a;
823 const struct uid_gid_extent *e2 = b;
825 if (e1->lower_first < e2->lower_first)
828 if (e1->lower_first > e2->lower_first)
835 * sort_idmaps - Sorts an array of idmap entries.
836 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
838 static int sort_idmaps(struct uid_gid_map *map)
840 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
843 /* Sort forward array. */
844 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
845 cmp_extents_forward, NULL);
847 /* Only copy the memory from forward we actually need. */
848 map->reverse = kmemdup(map->forward,
849 map->nr_extents * sizeof(struct uid_gid_extent),
854 /* Sort reverse array. */
855 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
856 cmp_extents_reverse, NULL);
862 * verify_root_map() - check the uid 0 mapping
863 * @file: idmapping file
864 * @map_ns: user namespace of the target process
865 * @new_map: requested idmap
867 * If a process requests mapping parent uid 0 into the new ns, verify that the
868 * process writing the map had the CAP_SETFCAP capability as the target process
869 * will be able to write fscaps that are valid in ancestor user namespaces.
871 * Return: true if the mapping is allowed, false if not.
873 static bool verify_root_map(const struct file *file,
874 struct user_namespace *map_ns,
875 struct uid_gid_map *new_map)
878 const struct user_namespace *file_ns = file->f_cred->user_ns;
879 struct uid_gid_extent *extent0 = NULL;
881 for (idx = 0; idx < new_map->nr_extents; idx++) {
882 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
883 extent0 = &new_map->extent[idx];
885 extent0 = &new_map->forward[idx];
886 if (extent0->lower_first == 0)
895 if (map_ns == file_ns) {
896 /* The process unshared its ns and is writing to its own
897 * /proc/self/uid_map. User already has full capabilites in
898 * the new namespace. Verify that the parent had CAP_SETFCAP
901 if (!file_ns->parent_could_setfcap)
904 /* Process p1 is writing to uid_map of p2, who is in a child
905 * user namespace to p1's. Verify that the opener of the map
906 * file has CAP_SETFCAP against the parent of the new map
908 if (!file_ns_capable(file, map_ns->parent, CAP_SETFCAP))
915 static ssize_t map_write(struct file *file, const char __user *buf,
916 size_t count, loff_t *ppos,
918 struct uid_gid_map *map,
919 struct uid_gid_map *parent_map)
921 struct seq_file *seq = file->private_data;
922 struct user_namespace *map_ns = seq->private;
923 struct uid_gid_map new_map;
925 struct uid_gid_extent extent;
926 char *kbuf = NULL, *pos, *next_line;
929 /* Only allow < page size writes at the beginning of the file */
930 if ((*ppos != 0) || (count >= PAGE_SIZE))
933 /* Slurp in the user data */
934 kbuf = memdup_user_nul(buf, count);
936 return PTR_ERR(kbuf);
939 * The userns_state_mutex serializes all writes to any given map.
941 * Any map is only ever written once.
943 * An id map fits within 1 cache line on most architectures.
945 * On read nothing needs to be done unless you are on an
946 * architecture with a crazy cache coherency model like alpha.
948 * There is a one time data dependency between reading the
949 * count of the extents and the values of the extents. The
950 * desired behavior is to see the values of the extents that
951 * were written before the count of the extents.
953 * To achieve this smp_wmb() is used on guarantee the write
954 * order and smp_rmb() is guaranteed that we don't have crazy
955 * architectures returning stale data.
957 mutex_lock(&userns_state_mutex);
959 memset(&new_map, 0, sizeof(struct uid_gid_map));
962 /* Only allow one successful write to the map */
963 if (map->nr_extents != 0)
967 * Adjusting namespace settings requires capabilities on the target.
969 if (cap_valid(cap_setid) && !file_ns_capable(file, map_ns, CAP_SYS_ADMIN))
972 /* Parse the user data */
975 for (; pos; pos = next_line) {
977 /* Find the end of line and ensure I don't look past it */
978 next_line = strchr(pos, '\n');
982 if (*next_line == '\0')
986 pos = skip_spaces(pos);
987 extent.first = simple_strtoul(pos, &pos, 10);
991 pos = skip_spaces(pos);
992 extent.lower_first = simple_strtoul(pos, &pos, 10);
996 pos = skip_spaces(pos);
997 extent.count = simple_strtoul(pos, &pos, 10);
998 if (*pos && !isspace(*pos))
1001 /* Verify there is not trailing junk on the line */
1002 pos = skip_spaces(pos);
1006 /* Verify we have been given valid starting values */
1007 if ((extent.first == (u32) -1) ||
1008 (extent.lower_first == (u32) -1))
1011 /* Verify count is not zero and does not cause the
1014 if ((extent.first + extent.count) <= extent.first)
1016 if ((extent.lower_first + extent.count) <=
1020 /* Do the ranges in extent overlap any previous extents? */
1021 if (mappings_overlap(&new_map, &extent))
1024 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
1025 (next_line != NULL))
1028 ret = insert_extent(&new_map, &extent);
1033 /* Be very certain the new map actually exists */
1034 if (new_map.nr_extents == 0)
1038 /* Validate the user is allowed to use user id's mapped to. */
1039 if (!new_idmap_permitted(file, map_ns, cap_setid, &new_map))
1043 /* Map the lower ids from the parent user namespace to the
1044 * kernel global id space.
1046 for (idx = 0; idx < new_map.nr_extents; idx++) {
1047 struct uid_gid_extent *e;
1050 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
1051 e = &new_map.extent[idx];
1053 e = &new_map.forward[idx];
1055 lower_first = map_id_range_down(parent_map,
1059 /* Fail if we can not map the specified extent to
1060 * the kernel global id space.
1062 if (lower_first == (u32) -1)
1065 e->lower_first = lower_first;
1069 * If we want to use binary search for lookup, this clones the extent
1070 * array and sorts both copies.
1072 ret = sort_idmaps(&new_map);
1076 /* Install the map */
1077 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1078 memcpy(map->extent, new_map.extent,
1079 new_map.nr_extents * sizeof(new_map.extent[0]));
1081 map->forward = new_map.forward;
1082 map->reverse = new_map.reverse;
1085 map->nr_extents = new_map.nr_extents;
1090 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1091 kfree(new_map.forward);
1092 kfree(new_map.reverse);
1093 map->forward = NULL;
1094 map->reverse = NULL;
1095 map->nr_extents = 0;
1098 mutex_unlock(&userns_state_mutex);
1103 ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1104 size_t size, loff_t *ppos)
1106 struct seq_file *seq = file->private_data;
1107 struct user_namespace *ns = seq->private;
1108 struct user_namespace *seq_ns = seq_user_ns(seq);
1113 if ((seq_ns != ns) && (seq_ns != ns->parent))
1116 return map_write(file, buf, size, ppos, CAP_SETUID,
1117 &ns->uid_map, &ns->parent->uid_map);
1120 ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1121 size_t size, loff_t *ppos)
1123 struct seq_file *seq = file->private_data;
1124 struct user_namespace *ns = seq->private;
1125 struct user_namespace *seq_ns = seq_user_ns(seq);
1130 if ((seq_ns != ns) && (seq_ns != ns->parent))
1133 return map_write(file, buf, size, ppos, CAP_SETGID,
1134 &ns->gid_map, &ns->parent->gid_map);
1137 ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1138 size_t size, loff_t *ppos)
1140 struct seq_file *seq = file->private_data;
1141 struct user_namespace *ns = seq->private;
1142 struct user_namespace *seq_ns = seq_user_ns(seq);
1147 if ((seq_ns != ns) && (seq_ns != ns->parent))
1150 /* Anyone can set any valid project id no capability needed */
1151 return map_write(file, buf, size, ppos, -1,
1152 &ns->projid_map, &ns->parent->projid_map);
1155 static bool new_idmap_permitted(const struct file *file,
1156 struct user_namespace *ns, int cap_setid,
1157 struct uid_gid_map *new_map)
1159 const struct cred *cred = file->f_cred;
1161 if (cap_setid == CAP_SETUID && !verify_root_map(file, ns, new_map))
1164 /* Don't allow mappings that would allow anything that wouldn't
1165 * be allowed without the establishment of unprivileged mappings.
1167 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1168 uid_eq(ns->owner, cred->euid)) {
1169 u32 id = new_map->extent[0].lower_first;
1170 if (cap_setid == CAP_SETUID) {
1171 kuid_t uid = make_kuid(ns->parent, id);
1172 if (uid_eq(uid, cred->euid))
1174 } else if (cap_setid == CAP_SETGID) {
1175 kgid_t gid = make_kgid(ns->parent, id);
1176 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1177 gid_eq(gid, cred->egid))
1182 /* Allow anyone to set a mapping that doesn't require privilege */
1183 if (!cap_valid(cap_setid))
1186 /* Allow the specified ids if we have the appropriate capability
1187 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1188 * And the opener of the id file also has the appropriate capability.
1190 if (ns_capable(ns->parent, cap_setid) &&
1191 file_ns_capable(file, ns->parent, cap_setid))
1197 int proc_setgroups_show(struct seq_file *seq, void *v)
1199 struct user_namespace *ns = seq->private;
1200 unsigned long userns_flags = READ_ONCE(ns->flags);
1202 seq_printf(seq, "%s\n",
1203 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1208 ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1209 size_t count, loff_t *ppos)
1211 struct seq_file *seq = file->private_data;
1212 struct user_namespace *ns = seq->private;
1214 bool setgroups_allowed;
1217 /* Only allow a very narrow range of strings to be written */
1219 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1222 /* What was written? */
1224 if (copy_from_user(kbuf, buf, count))
1229 /* What is being requested? */
1231 if (strncmp(pos, "allow", 5) == 0) {
1233 setgroups_allowed = true;
1235 else if (strncmp(pos, "deny", 4) == 0) {
1237 setgroups_allowed = false;
1242 /* Verify there is not trailing junk on the line */
1243 pos = skip_spaces(pos);
1248 mutex_lock(&userns_state_mutex);
1249 if (setgroups_allowed) {
1250 /* Enabling setgroups after setgroups has been disabled
1253 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1256 /* Permanently disabling setgroups after setgroups has
1257 * been enabled by writing the gid_map is not allowed.
1259 if (ns->gid_map.nr_extents != 0)
1261 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1263 mutex_unlock(&userns_state_mutex);
1265 /* Report a successful write */
1271 mutex_unlock(&userns_state_mutex);
1275 bool userns_may_setgroups(const struct user_namespace *ns)
1279 mutex_lock(&userns_state_mutex);
1280 /* It is not safe to use setgroups until a gid mapping in
1281 * the user namespace has been established.
1283 allowed = ns->gid_map.nr_extents != 0;
1284 /* Is setgroups allowed? */
1285 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1286 mutex_unlock(&userns_state_mutex);
1292 * Returns true if @child is the same namespace or a descendant of
1295 bool in_userns(const struct user_namespace *ancestor,
1296 const struct user_namespace *child)
1298 const struct user_namespace *ns;
1299 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1301 return (ns == ancestor);
1304 bool current_in_userns(const struct user_namespace *target_ns)
1306 return in_userns(target_ns, current_user_ns());
1308 EXPORT_SYMBOL(current_in_userns);
1310 static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1312 return container_of(ns, struct user_namespace, ns);
1315 static struct ns_common *userns_get(struct task_struct *task)
1317 struct user_namespace *user_ns;
1320 user_ns = get_user_ns(__task_cred(task)->user_ns);
1323 return user_ns ? &user_ns->ns : NULL;
1326 static void userns_put(struct ns_common *ns)
1328 put_user_ns(to_user_ns(ns));
1331 static int userns_install(struct nsset *nsset, struct ns_common *ns)
1333 struct user_namespace *user_ns = to_user_ns(ns);
1336 /* Don't allow gaining capabilities by reentering
1337 * the same user namespace.
1339 if (user_ns == current_user_ns())
1342 /* Tasks that share a thread group must share a user namespace */
1343 if (!thread_group_empty(current))
1346 if (current->fs->users != 1)
1349 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1352 cred = nsset_cred(nsset);
1356 put_user_ns(cred->user_ns);
1357 set_cred_user_ns(cred, get_user_ns(user_ns));
1359 if (set_cred_ucounts(cred) < 0)
1365 struct ns_common *ns_get_owner(struct ns_common *ns)
1367 struct user_namespace *my_user_ns = current_user_ns();
1368 struct user_namespace *owner, *p;
1370 /* See if the owner is in the current user namespace */
1371 owner = p = ns->ops->owner(ns);
1374 return ERR_PTR(-EPERM);
1375 if (p == my_user_ns)
1380 return &get_user_ns(owner)->ns;
1383 static struct user_namespace *userns_owner(struct ns_common *ns)
1385 return to_user_ns(ns)->parent;
1388 const struct proc_ns_operations userns_operations = {
1390 .type = CLONE_NEWUSER,
1393 .install = userns_install,
1394 .owner = userns_owner,
1395 .get_parent = ns_get_owner,
1398 static __init int user_namespaces_init(void)
1400 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC | SLAB_ACCOUNT);
1403 subsys_initcall(user_namespaces_init);