2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License as
4 * published by the Free Software Foundation, version 2 of the
8 #include <linux/export.h>
9 #include <linux/nsproxy.h>
10 #include <linux/slab.h>
11 #include <linux/sched/signal.h>
12 #include <linux/user_namespace.h>
13 #include <linux/proc_ns.h>
14 #include <linux/highuid.h>
15 #include <linux/cred.h>
16 #include <linux/securebits.h>
17 #include <linux/keyctl.h>
18 #include <linux/key-type.h>
19 #include <keys/user-type.h>
20 #include <linux/seq_file.h>
22 #include <linux/uaccess.h>
23 #include <linux/ctype.h>
24 #include <linux/projid.h>
25 #include <linux/fs_struct.h>
26 #include <linux/bsearch.h>
27 #include <linux/sort.h>
29 static struct kmem_cache *user_ns_cachep __read_mostly;
30 static DEFINE_MUTEX(userns_state_mutex);
32 static bool new_idmap_permitted(const struct file *file,
33 struct user_namespace *ns, int cap_setid,
34 struct uid_gid_map *map);
35 static void free_user_ns(struct work_struct *work);
37 static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
39 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
42 static void dec_user_namespaces(struct ucounts *ucounts)
44 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
47 static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
49 /* Start with the same capabilities as init but useless for doing
50 * anything as the capabilities are bound to the new user namespace.
52 cred->securebits = SECUREBITS_DEFAULT;
53 cred->cap_inheritable = CAP_EMPTY_SET;
54 cred->cap_permitted = CAP_FULL_SET;
55 cred->cap_effective = CAP_FULL_SET;
56 cred->cap_ambient = CAP_EMPTY_SET;
57 cred->cap_bset = CAP_FULL_SET;
59 key_put(cred->request_key_auth);
60 cred->request_key_auth = NULL;
62 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
63 cred->user_ns = user_ns;
67 * Create a new user namespace, deriving the creator from the user in the
68 * passed credentials, and replacing that user with the new root user for the
71 * This is called by copy_creds(), which will finish setting the target task's
74 int create_user_ns(struct cred *new)
76 struct user_namespace *ns, *parent_ns = new->user_ns;
77 kuid_t owner = new->euid;
78 kgid_t group = new->egid;
79 struct ucounts *ucounts;
83 if (parent_ns->level > 32)
86 ucounts = inc_user_namespaces(parent_ns, owner);
91 * Verify that we can not violate the policy of which files
92 * may be accessed that is specified by the root directory,
93 * by verifing that the root directory is at the root of the
94 * mount namespace which allows all files to be accessed.
97 if (current_chrooted())
100 /* The creator needs a mapping in the parent user namespace
101 * or else we won't be able to reasonably tell userspace who
102 * created a user_namespace.
105 if (!kuid_has_mapping(parent_ns, owner) ||
106 !kgid_has_mapping(parent_ns, group))
110 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
114 ret = ns_alloc_inum(&ns->ns);
117 ns->ns.ops = &userns_operations;
119 atomic_set(&ns->count, 1);
120 /* Leave the new->user_ns reference with the new user namespace. */
121 ns->parent = parent_ns;
122 ns->level = parent_ns->level + 1;
125 INIT_WORK(&ns->work, free_user_ns);
126 for (i = 0; i < UCOUNT_COUNTS; i++) {
127 ns->ucount_max[i] = INT_MAX;
129 ns->ucounts = ucounts;
131 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
132 mutex_lock(&userns_state_mutex);
133 ns->flags = parent_ns->flags;
134 mutex_unlock(&userns_state_mutex);
136 #ifdef CONFIG_PERSISTENT_KEYRINGS
137 init_rwsem(&ns->persistent_keyring_register_sem);
140 if (!setup_userns_sysctls(ns))
143 set_cred_user_ns(new, ns);
146 #ifdef CONFIG_PERSISTENT_KEYRINGS
147 key_put(ns->persistent_keyring_register);
149 ns_free_inum(&ns->ns);
151 kmem_cache_free(user_ns_cachep, ns);
153 dec_user_namespaces(ucounts);
158 int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
163 if (!(unshare_flags & CLONE_NEWUSER))
166 cred = prepare_creds();
168 err = create_user_ns(cred);
178 static void free_user_ns(struct work_struct *work)
180 struct user_namespace *parent, *ns =
181 container_of(work, struct user_namespace, work);
184 struct ucounts *ucounts = ns->ucounts;
186 if (ns->gid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
187 kfree(ns->gid_map.forward);
188 kfree(ns->gid_map.reverse);
190 if (ns->uid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
191 kfree(ns->uid_map.forward);
192 kfree(ns->uid_map.reverse);
194 if (ns->projid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
195 kfree(ns->projid_map.forward);
196 kfree(ns->projid_map.reverse);
198 retire_userns_sysctls(ns);
199 #ifdef CONFIG_PERSISTENT_KEYRINGS
200 key_put(ns->persistent_keyring_register);
202 ns_free_inum(&ns->ns);
203 kmem_cache_free(user_ns_cachep, ns);
204 dec_user_namespaces(ucounts);
206 } while (atomic_dec_and_test(&parent->count));
209 void __put_user_ns(struct user_namespace *ns)
211 schedule_work(&ns->work);
213 EXPORT_SYMBOL(__put_user_ns);
216 * idmap_key struct holds the information necessary to find an idmapping in a
217 * sorted idmap array. It is passed to cmp_map_id() as first argument.
220 bool map_up; /* true -> id from kid; false -> kid from id */
221 u32 id; /* id to find */
222 u32 count; /* == 0 unless used with map_id_range_down() */
226 * cmp_map_id - Function to be passed to bsearch() to find the requested
227 * idmapping. Expects struct idmap_key to be passed via @k.
229 static int cmp_map_id(const void *k, const void *e)
231 u32 first, last, id2;
232 const struct idmap_key *key = k;
233 const struct uid_gid_extent *el = e;
235 id2 = key->id + key->count - 1;
237 /* handle map_id_{down,up}() */
239 first = el->lower_first;
243 last = first + el->count - 1;
245 if (key->id >= first && key->id <= last &&
246 (id2 >= first && id2 <= last))
249 if (key->id < first || id2 < first)
256 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
257 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
259 static struct uid_gid_extent *
260 map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
262 struct idmap_key key;
268 return bsearch(&key, map->forward, extents,
269 sizeof(struct uid_gid_extent), cmp_map_id);
273 * map_id_range_down_base - Find idmap via binary search in static extent array.
274 * Can only be called if number of mappings is equal or less than
275 * UID_GID_MAP_MAX_BASE_EXTENTS.
277 static struct uid_gid_extent *
278 map_id_range_down_base(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
281 u32 first, last, id2;
283 id2 = id + count - 1;
285 /* Find the matching extent */
286 for (idx = 0; idx < extents; idx++) {
287 first = map->extent[idx].first;
288 last = first + map->extent[idx].count - 1;
289 if (id >= first && id <= last &&
290 (id2 >= first && id2 <= last))
291 return &map->extent[idx];
296 static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
298 struct uid_gid_extent *extent;
299 unsigned extents = map->nr_extents;
302 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
303 extent = map_id_range_down_base(extents, map, id, count);
305 extent = map_id_range_down_max(extents, map, id, count);
307 /* Map the id or note failure */
309 id = (id - extent->first) + extent->lower_first;
316 static u32 map_id_down(struct uid_gid_map *map, u32 id)
318 return map_id_range_down(map, id, 1);
322 * map_id_up_base - Find idmap via binary search in static extent array.
323 * Can only be called if number of mappings is equal or less than
324 * UID_GID_MAP_MAX_BASE_EXTENTS.
326 static struct uid_gid_extent *
327 map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
332 /* Find the matching extent */
333 for (idx = 0; idx < extents; idx++) {
334 first = map->extent[idx].lower_first;
335 last = first + map->extent[idx].count - 1;
336 if (id >= first && id <= last)
337 return &map->extent[idx];
343 * map_id_up_max - Find idmap via binary search in ordered idmap array.
344 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
346 static struct uid_gid_extent *
347 map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
349 struct idmap_key key;
355 return bsearch(&key, map->reverse, extents,
356 sizeof(struct uid_gid_extent), cmp_map_id);
359 static u32 map_id_up(struct uid_gid_map *map, u32 id)
361 struct uid_gid_extent *extent;
362 unsigned extents = map->nr_extents;
365 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
366 extent = map_id_up_base(extents, map, id);
368 extent = map_id_up_max(extents, map, id);
370 /* Map the id or note failure */
372 id = (id - extent->lower_first) + extent->first;
380 * make_kuid - Map a user-namespace uid pair into a kuid.
381 * @ns: User namespace that the uid is in
382 * @uid: User identifier
384 * Maps a user-namespace uid pair into a kernel internal kuid,
385 * and returns that kuid.
387 * When there is no mapping defined for the user-namespace uid
388 * pair INVALID_UID is returned. Callers are expected to test
389 * for and handle INVALID_UID being returned. INVALID_UID
390 * may be tested for using uid_valid().
392 kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
394 /* Map the uid to a global kernel uid */
395 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
397 EXPORT_SYMBOL(make_kuid);
400 * from_kuid - Create a uid from a kuid user-namespace pair.
401 * @targ: The user namespace we want a uid in.
402 * @kuid: The kernel internal uid to start with.
404 * Map @kuid into the user-namespace specified by @targ and
405 * return the resulting uid.
407 * There is always a mapping into the initial user_namespace.
409 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
411 uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
413 /* Map the uid from a global kernel uid */
414 return map_id_up(&targ->uid_map, __kuid_val(kuid));
416 EXPORT_SYMBOL(from_kuid);
419 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
420 * @targ: The user namespace we want a uid in.
421 * @kuid: The kernel internal uid to start with.
423 * Map @kuid into the user-namespace specified by @targ and
424 * return the resulting uid.
426 * There is always a mapping into the initial user_namespace.
428 * Unlike from_kuid from_kuid_munged never fails and always
429 * returns a valid uid. This makes from_kuid_munged appropriate
430 * for use in syscalls like stat and getuid where failing the
431 * system call and failing to provide a valid uid are not an
434 * If @kuid has no mapping in @targ overflowuid is returned.
436 uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
439 uid = from_kuid(targ, kuid);
441 if (uid == (uid_t) -1)
445 EXPORT_SYMBOL(from_kuid_munged);
448 * make_kgid - Map a user-namespace gid pair into a kgid.
449 * @ns: User namespace that the gid is in
450 * @gid: group identifier
452 * Maps a user-namespace gid pair into a kernel internal kgid,
453 * and returns that kgid.
455 * When there is no mapping defined for the user-namespace gid
456 * pair INVALID_GID is returned. Callers are expected to test
457 * for and handle INVALID_GID being returned. INVALID_GID may be
458 * tested for using gid_valid().
460 kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
462 /* Map the gid to a global kernel gid */
463 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
465 EXPORT_SYMBOL(make_kgid);
468 * from_kgid - Create a gid from a kgid user-namespace pair.
469 * @targ: The user namespace we want a gid in.
470 * @kgid: The kernel internal gid to start with.
472 * Map @kgid into the user-namespace specified by @targ and
473 * return the resulting gid.
475 * There is always a mapping into the initial user_namespace.
477 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
479 gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
481 /* Map the gid from a global kernel gid */
482 return map_id_up(&targ->gid_map, __kgid_val(kgid));
484 EXPORT_SYMBOL(from_kgid);
487 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
488 * @targ: The user namespace we want a gid in.
489 * @kgid: The kernel internal gid to start with.
491 * Map @kgid into the user-namespace specified by @targ and
492 * return the resulting gid.
494 * There is always a mapping into the initial user_namespace.
496 * Unlike from_kgid from_kgid_munged never fails and always
497 * returns a valid gid. This makes from_kgid_munged appropriate
498 * for use in syscalls like stat and getgid where failing the
499 * system call and failing to provide a valid gid are not options.
501 * If @kgid has no mapping in @targ overflowgid is returned.
503 gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
506 gid = from_kgid(targ, kgid);
508 if (gid == (gid_t) -1)
512 EXPORT_SYMBOL(from_kgid_munged);
515 * make_kprojid - Map a user-namespace projid pair into a kprojid.
516 * @ns: User namespace that the projid is in
517 * @projid: Project identifier
519 * Maps a user-namespace uid pair into a kernel internal kuid,
520 * and returns that kuid.
522 * When there is no mapping defined for the user-namespace projid
523 * pair INVALID_PROJID is returned. Callers are expected to test
524 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
525 * may be tested for using projid_valid().
527 kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
529 /* Map the uid to a global kernel uid */
530 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
532 EXPORT_SYMBOL(make_kprojid);
535 * from_kprojid - Create a projid from a kprojid user-namespace pair.
536 * @targ: The user namespace we want a projid in.
537 * @kprojid: The kernel internal project identifier to start with.
539 * Map @kprojid into the user-namespace specified by @targ and
540 * return the resulting projid.
542 * There is always a mapping into the initial user_namespace.
544 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
546 projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
548 /* Map the uid from a global kernel uid */
549 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
551 EXPORT_SYMBOL(from_kprojid);
554 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
555 * @targ: The user namespace we want a projid in.
556 * @kprojid: The kernel internal projid to start with.
558 * Map @kprojid into the user-namespace specified by @targ and
559 * return the resulting projid.
561 * There is always a mapping into the initial user_namespace.
563 * Unlike from_kprojid from_kprojid_munged never fails and always
564 * returns a valid projid. This makes from_kprojid_munged
565 * appropriate for use in syscalls like stat and where
566 * failing the system call and failing to provide a valid projid are
569 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
571 projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
574 projid = from_kprojid(targ, kprojid);
576 if (projid == (projid_t) -1)
577 projid = OVERFLOW_PROJID;
580 EXPORT_SYMBOL(from_kprojid_munged);
583 static int uid_m_show(struct seq_file *seq, void *v)
585 struct user_namespace *ns = seq->private;
586 struct uid_gid_extent *extent = v;
587 struct user_namespace *lower_ns;
590 lower_ns = seq_user_ns(seq);
591 if ((lower_ns == ns) && lower_ns->parent)
592 lower_ns = lower_ns->parent;
594 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
596 seq_printf(seq, "%10u %10u %10u\n",
604 static int gid_m_show(struct seq_file *seq, void *v)
606 struct user_namespace *ns = seq->private;
607 struct uid_gid_extent *extent = v;
608 struct user_namespace *lower_ns;
611 lower_ns = seq_user_ns(seq);
612 if ((lower_ns == ns) && lower_ns->parent)
613 lower_ns = lower_ns->parent;
615 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
617 seq_printf(seq, "%10u %10u %10u\n",
625 static int projid_m_show(struct seq_file *seq, void *v)
627 struct user_namespace *ns = seq->private;
628 struct uid_gid_extent *extent = v;
629 struct user_namespace *lower_ns;
632 lower_ns = seq_user_ns(seq);
633 if ((lower_ns == ns) && lower_ns->parent)
634 lower_ns = lower_ns->parent;
636 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
638 seq_printf(seq, "%10u %10u %10u\n",
646 static void *m_start(struct seq_file *seq, loff_t *ppos,
647 struct uid_gid_map *map)
650 unsigned extents = map->nr_extents;
656 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
657 return &map->extent[pos];
659 return &map->forward[pos];
662 static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
664 struct user_namespace *ns = seq->private;
666 return m_start(seq, ppos, &ns->uid_map);
669 static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
671 struct user_namespace *ns = seq->private;
673 return m_start(seq, ppos, &ns->gid_map);
676 static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
678 struct user_namespace *ns = seq->private;
680 return m_start(seq, ppos, &ns->projid_map);
683 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
686 return seq->op->start(seq, pos);
689 static void m_stop(struct seq_file *seq, void *v)
694 const struct seq_operations proc_uid_seq_operations = {
695 .start = uid_m_start,
701 const struct seq_operations proc_gid_seq_operations = {
702 .start = gid_m_start,
708 const struct seq_operations proc_projid_seq_operations = {
709 .start = projid_m_start,
712 .show = projid_m_show,
715 static bool mappings_overlap(struct uid_gid_map *new_map,
716 struct uid_gid_extent *extent)
718 u32 upper_first, lower_first, upper_last, lower_last;
721 upper_first = extent->first;
722 lower_first = extent->lower_first;
723 upper_last = upper_first + extent->count - 1;
724 lower_last = lower_first + extent->count - 1;
726 for (idx = 0; idx < new_map->nr_extents; idx++) {
727 u32 prev_upper_first, prev_lower_first;
728 u32 prev_upper_last, prev_lower_last;
729 struct uid_gid_extent *prev;
731 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
732 prev = &new_map->extent[idx];
734 prev = &new_map->forward[idx];
736 prev_upper_first = prev->first;
737 prev_lower_first = prev->lower_first;
738 prev_upper_last = prev_upper_first + prev->count - 1;
739 prev_lower_last = prev_lower_first + prev->count - 1;
741 /* Does the upper range intersect a previous extent? */
742 if ((prev_upper_first <= upper_last) &&
743 (prev_upper_last >= upper_first))
746 /* Does the lower range intersect a previous extent? */
747 if ((prev_lower_first <= lower_last) &&
748 (prev_lower_last >= lower_first))
755 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
756 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
757 * UID_GID_MAP_MAX_BASE_EXTENTS.
759 static int insert_extent(struct uid_gid_map *map, struct uid_gid_extent *extent)
761 struct uid_gid_extent *dest;
763 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
764 struct uid_gid_extent *forward;
766 /* Allocate memory for 340 mappings. */
767 forward = kmalloc(sizeof(struct uid_gid_extent) *
768 UID_GID_MAP_MAX_EXTENTS, GFP_KERNEL);
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);
849 static ssize_t map_write(struct file *file, const char __user *buf,
850 size_t count, loff_t *ppos,
852 struct uid_gid_map *map,
853 struct uid_gid_map *parent_map)
855 struct seq_file *seq = file->private_data;
856 struct user_namespace *ns = seq->private;
857 struct uid_gid_map new_map;
859 struct uid_gid_extent extent;
860 char *kbuf = NULL, *pos, *next_line;
861 ssize_t ret = -EINVAL;
864 * The userns_state_mutex serializes all writes to any given map.
866 * Any map is only ever written once.
868 * An id map fits within 1 cache line on most architectures.
870 * On read nothing needs to be done unless you are on an
871 * architecture with a crazy cache coherency model like alpha.
873 * There is a one time data dependency between reading the
874 * count of the extents and the values of the extents. The
875 * desired behavior is to see the values of the extents that
876 * were written before the count of the extents.
878 * To achieve this smp_wmb() is used on guarantee the write
879 * order and smp_rmb() is guaranteed that we don't have crazy
880 * architectures returning stale data.
882 mutex_lock(&userns_state_mutex);
884 memset(&new_map, 0, sizeof(struct uid_gid_map));
887 /* Only allow one successful write to the map */
888 if (map->nr_extents != 0)
892 * Adjusting namespace settings requires capabilities on the target.
894 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
897 /* Only allow < page size writes at the beginning of the file */
899 if ((*ppos != 0) || (count >= PAGE_SIZE))
902 /* Slurp in the user data */
903 kbuf = memdup_user_nul(buf, count);
910 /* Parse the user data */
913 for (; pos; pos = next_line) {
915 /* Find the end of line and ensure I don't look past it */
916 next_line = strchr(pos, '\n');
920 if (*next_line == '\0')
924 pos = skip_spaces(pos);
925 extent.first = simple_strtoul(pos, &pos, 10);
929 pos = skip_spaces(pos);
930 extent.lower_first = simple_strtoul(pos, &pos, 10);
934 pos = skip_spaces(pos);
935 extent.count = simple_strtoul(pos, &pos, 10);
936 if (*pos && !isspace(*pos))
939 /* Verify there is not trailing junk on the line */
940 pos = skip_spaces(pos);
944 /* Verify we have been given valid starting values */
945 if ((extent.first == (u32) -1) ||
946 (extent.lower_first == (u32) -1))
949 /* Verify count is not zero and does not cause the
952 if ((extent.first + extent.count) <= extent.first)
954 if ((extent.lower_first + extent.count) <=
958 /* Do the ranges in extent overlap any previous extents? */
959 if (mappings_overlap(&new_map, &extent))
962 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
966 ret = insert_extent(&new_map, &extent);
971 /* Be very certaint the new map actually exists */
972 if (new_map.nr_extents == 0)
976 /* Validate the user is allowed to use user id's mapped to. */
977 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
980 ret = sort_idmaps(&new_map);
985 /* Map the lower ids from the parent user namespace to the
986 * kernel global id space.
988 for (idx = 0; idx < new_map.nr_extents; idx++) {
989 struct uid_gid_extent *e;
992 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
993 e = &new_map.extent[idx];
995 e = &new_map.forward[idx];
997 lower_first = map_id_range_down(parent_map,
1001 /* Fail if we can not map the specified extent to
1002 * the kernel global id space.
1004 if (lower_first == (u32) -1)
1007 e->lower_first = lower_first;
1010 /* Install the map */
1011 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1012 memcpy(map->extent, new_map.extent,
1013 new_map.nr_extents * sizeof(new_map.extent[0]));
1015 map->forward = new_map.forward;
1016 map->reverse = new_map.reverse;
1019 map->nr_extents = new_map.nr_extents;
1024 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1025 kfree(new_map.forward);
1026 kfree(new_map.reverse);
1027 map->forward = NULL;
1028 map->reverse = NULL;
1029 map->nr_extents = 0;
1032 mutex_unlock(&userns_state_mutex);
1037 ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1038 size_t size, loff_t *ppos)
1040 struct seq_file *seq = file->private_data;
1041 struct user_namespace *ns = seq->private;
1042 struct user_namespace *seq_ns = seq_user_ns(seq);
1047 if ((seq_ns != ns) && (seq_ns != ns->parent))
1050 return map_write(file, buf, size, ppos, CAP_SETUID,
1051 &ns->uid_map, &ns->parent->uid_map);
1054 ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1055 size_t size, loff_t *ppos)
1057 struct seq_file *seq = file->private_data;
1058 struct user_namespace *ns = seq->private;
1059 struct user_namespace *seq_ns = seq_user_ns(seq);
1064 if ((seq_ns != ns) && (seq_ns != ns->parent))
1067 return map_write(file, buf, size, ppos, CAP_SETGID,
1068 &ns->gid_map, &ns->parent->gid_map);
1071 ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1072 size_t size, loff_t *ppos)
1074 struct seq_file *seq = file->private_data;
1075 struct user_namespace *ns = seq->private;
1076 struct user_namespace *seq_ns = seq_user_ns(seq);
1081 if ((seq_ns != ns) && (seq_ns != ns->parent))
1084 /* Anyone can set any valid project id no capability needed */
1085 return map_write(file, buf, size, ppos, -1,
1086 &ns->projid_map, &ns->parent->projid_map);
1089 static bool new_idmap_permitted(const struct file *file,
1090 struct user_namespace *ns, int cap_setid,
1091 struct uid_gid_map *new_map)
1093 const struct cred *cred = file->f_cred;
1094 /* Don't allow mappings that would allow anything that wouldn't
1095 * be allowed without the establishment of unprivileged mappings.
1097 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1098 uid_eq(ns->owner, cred->euid)) {
1099 u32 id = new_map->extent[0].lower_first;
1100 if (cap_setid == CAP_SETUID) {
1101 kuid_t uid = make_kuid(ns->parent, id);
1102 if (uid_eq(uid, cred->euid))
1104 } else if (cap_setid == CAP_SETGID) {
1105 kgid_t gid = make_kgid(ns->parent, id);
1106 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1107 gid_eq(gid, cred->egid))
1112 /* Allow anyone to set a mapping that doesn't require privilege */
1113 if (!cap_valid(cap_setid))
1116 /* Allow the specified ids if we have the appropriate capability
1117 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1118 * And the opener of the id file also had the approprpiate capability.
1120 if (ns_capable(ns->parent, cap_setid) &&
1121 file_ns_capable(file, ns->parent, cap_setid))
1127 int proc_setgroups_show(struct seq_file *seq, void *v)
1129 struct user_namespace *ns = seq->private;
1130 unsigned long userns_flags = READ_ONCE(ns->flags);
1132 seq_printf(seq, "%s\n",
1133 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1138 ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1139 size_t count, loff_t *ppos)
1141 struct seq_file *seq = file->private_data;
1142 struct user_namespace *ns = seq->private;
1144 bool setgroups_allowed;
1147 /* Only allow a very narrow range of strings to be written */
1149 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1152 /* What was written? */
1154 if (copy_from_user(kbuf, buf, count))
1159 /* What is being requested? */
1161 if (strncmp(pos, "allow", 5) == 0) {
1163 setgroups_allowed = true;
1165 else if (strncmp(pos, "deny", 4) == 0) {
1167 setgroups_allowed = false;
1172 /* Verify there is not trailing junk on the line */
1173 pos = skip_spaces(pos);
1178 mutex_lock(&userns_state_mutex);
1179 if (setgroups_allowed) {
1180 /* Enabling setgroups after setgroups has been disabled
1183 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1186 /* Permanently disabling setgroups after setgroups has
1187 * been enabled by writing the gid_map is not allowed.
1189 if (ns->gid_map.nr_extents != 0)
1191 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1193 mutex_unlock(&userns_state_mutex);
1195 /* Report a successful write */
1201 mutex_unlock(&userns_state_mutex);
1205 bool userns_may_setgroups(const struct user_namespace *ns)
1209 mutex_lock(&userns_state_mutex);
1210 /* It is not safe to use setgroups until a gid mapping in
1211 * the user namespace has been established.
1213 allowed = ns->gid_map.nr_extents != 0;
1214 /* Is setgroups allowed? */
1215 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1216 mutex_unlock(&userns_state_mutex);
1222 * Returns true if @child is the same namespace or a descendant of
1225 bool in_userns(const struct user_namespace *ancestor,
1226 const struct user_namespace *child)
1228 const struct user_namespace *ns;
1229 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1231 return (ns == ancestor);
1234 bool current_in_userns(const struct user_namespace *target_ns)
1236 return in_userns(target_ns, current_user_ns());
1239 static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1241 return container_of(ns, struct user_namespace, ns);
1244 static struct ns_common *userns_get(struct task_struct *task)
1246 struct user_namespace *user_ns;
1249 user_ns = get_user_ns(__task_cred(task)->user_ns);
1252 return user_ns ? &user_ns->ns : NULL;
1255 static void userns_put(struct ns_common *ns)
1257 put_user_ns(to_user_ns(ns));
1260 static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1262 struct user_namespace *user_ns = to_user_ns(ns);
1265 /* Don't allow gaining capabilities by reentering
1266 * the same user namespace.
1268 if (user_ns == current_user_ns())
1271 /* Tasks that share a thread group must share a user namespace */
1272 if (!thread_group_empty(current))
1275 if (current->fs->users != 1)
1278 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1281 cred = prepare_creds();
1285 put_user_ns(cred->user_ns);
1286 set_cred_user_ns(cred, get_user_ns(user_ns));
1288 return commit_creds(cred);
1291 struct ns_common *ns_get_owner(struct ns_common *ns)
1293 struct user_namespace *my_user_ns = current_user_ns();
1294 struct user_namespace *owner, *p;
1296 /* See if the owner is in the current user namespace */
1297 owner = p = ns->ops->owner(ns);
1300 return ERR_PTR(-EPERM);
1301 if (p == my_user_ns)
1306 return &get_user_ns(owner)->ns;
1309 static struct user_namespace *userns_owner(struct ns_common *ns)
1311 return to_user_ns(ns)->parent;
1314 const struct proc_ns_operations userns_operations = {
1316 .type = CLONE_NEWUSER,
1319 .install = userns_install,
1320 .owner = userns_owner,
1321 .get_parent = ns_get_owner,
1324 static __init int user_namespaces_init(void)
1326 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1329 subsys_initcall(user_namespaces_init);