1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/export.h>
4 #include <linux/nsproxy.h>
5 #include <linux/slab.h>
6 #include <linux/sched/signal.h>
7 #include <linux/user_namespace.h>
8 #include <linux/proc_ns.h>
9 #include <linux/highuid.h>
10 #include <linux/cred.h>
11 #include <linux/securebits.h>
12 #include <linux/keyctl.h>
13 #include <linux/key-type.h>
14 #include <keys/user-type.h>
15 #include <linux/seq_file.h>
17 #include <linux/uaccess.h>
18 #include <linux/ctype.h>
19 #include <linux/projid.h>
20 #include <linux/fs_struct.h>
21 #include <linux/bsearch.h>
22 #include <linux/sort.h>
24 static struct kmem_cache *user_ns_cachep __read_mostly;
25 static DEFINE_MUTEX(userns_state_mutex);
27 static bool new_idmap_permitted(const struct file *file,
28 struct user_namespace *ns, int cap_setid,
29 struct uid_gid_map *map);
30 static void free_user_ns(struct work_struct *work);
32 static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
34 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
37 static void dec_user_namespaces(struct ucounts *ucounts)
39 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
42 static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
44 /* Start with the same capabilities as init but useless for doing
45 * anything as the capabilities are bound to the new user namespace.
47 cred->securebits = SECUREBITS_DEFAULT;
48 cred->cap_inheritable = CAP_EMPTY_SET;
49 cred->cap_permitted = CAP_FULL_SET;
50 cred->cap_effective = CAP_FULL_SET;
51 cred->cap_ambient = CAP_EMPTY_SET;
52 cred->cap_bset = CAP_FULL_SET;
54 key_put(cred->request_key_auth);
55 cred->request_key_auth = NULL;
57 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
58 cred->user_ns = user_ns;
62 * Create a new user namespace, deriving the creator from the user in the
63 * passed credentials, and replacing that user with the new root user for the
66 * This is called by copy_creds(), which will finish setting the target task's
69 int create_user_ns(struct cred *new)
71 struct user_namespace *ns, *parent_ns = new->user_ns;
72 kuid_t owner = new->euid;
73 kgid_t group = new->egid;
74 struct ucounts *ucounts;
78 if (parent_ns->level > 32)
81 ucounts = inc_user_namespaces(parent_ns, owner);
86 * Verify that we can not violate the policy of which files
87 * may be accessed that is specified by the root directory,
88 * by verifying that the root directory is at the root of the
89 * mount namespace which allows all files to be accessed.
92 if (current_chrooted())
95 /* The creator needs a mapping in the parent user namespace
96 * or else we won't be able to reasonably tell userspace who
97 * created a user_namespace.
100 if (!kuid_has_mapping(parent_ns, owner) ||
101 !kgid_has_mapping(parent_ns, group))
105 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
109 ns->parent_could_setfcap = cap_raised(new->cap_effective, CAP_SETFCAP);
110 ret = ns_alloc_inum(&ns->ns);
113 ns->ns.ops = &userns_operations;
115 refcount_set(&ns->ns.count, 1);
116 /* Leave the new->user_ns reference with the new user namespace. */
117 ns->parent = parent_ns;
118 ns->level = parent_ns->level + 1;
121 INIT_WORK(&ns->work, free_user_ns);
122 for (i = 0; i < UCOUNT_COUNTS; i++) {
123 ns->ucount_max[i] = INT_MAX;
125 ns->ucounts = ucounts;
127 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
128 mutex_lock(&userns_state_mutex);
129 ns->flags = parent_ns->flags;
130 mutex_unlock(&userns_state_mutex);
133 INIT_LIST_HEAD(&ns->keyring_name_list);
134 init_rwsem(&ns->keyring_sem);
137 if (!setup_userns_sysctls(ns))
140 set_cred_user_ns(new, ns);
143 #ifdef CONFIG_PERSISTENT_KEYRINGS
144 key_put(ns->persistent_keyring_register);
146 ns_free_inum(&ns->ns);
148 kmem_cache_free(user_ns_cachep, ns);
150 dec_user_namespaces(ucounts);
155 int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
160 if (!(unshare_flags & CLONE_NEWUSER))
163 cred = prepare_creds();
165 err = create_user_ns(cred);
175 static void free_user_ns(struct work_struct *work)
177 struct user_namespace *parent, *ns =
178 container_of(work, struct user_namespace, work);
181 struct ucounts *ucounts = ns->ucounts;
183 if (ns->gid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
184 kfree(ns->gid_map.forward);
185 kfree(ns->gid_map.reverse);
187 if (ns->uid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
188 kfree(ns->uid_map.forward);
189 kfree(ns->uid_map.reverse);
191 if (ns->projid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
192 kfree(ns->projid_map.forward);
193 kfree(ns->projid_map.reverse);
195 retire_userns_sysctls(ns);
196 key_free_user_ns(ns);
197 ns_free_inum(&ns->ns);
198 kmem_cache_free(user_ns_cachep, ns);
199 dec_user_namespaces(ucounts);
201 } while (refcount_dec_and_test(&parent->ns.count));
204 void __put_user_ns(struct user_namespace *ns)
206 schedule_work(&ns->work);
208 EXPORT_SYMBOL(__put_user_ns);
211 * idmap_key struct holds the information necessary to find an idmapping in a
212 * sorted idmap array. It is passed to cmp_map_id() as first argument.
215 bool map_up; /* true -> id from kid; false -> kid from id */
216 u32 id; /* id to find */
217 u32 count; /* == 0 unless used with map_id_range_down() */
221 * cmp_map_id - Function to be passed to bsearch() to find the requested
222 * idmapping. Expects struct idmap_key to be passed via @k.
224 static int cmp_map_id(const void *k, const void *e)
226 u32 first, last, id2;
227 const struct idmap_key *key = k;
228 const struct uid_gid_extent *el = e;
230 id2 = key->id + key->count - 1;
232 /* handle map_id_{down,up}() */
234 first = el->lower_first;
238 last = first + el->count - 1;
240 if (key->id >= first && key->id <= last &&
241 (id2 >= first && id2 <= last))
244 if (key->id < first || id2 < first)
251 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
252 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
254 static struct uid_gid_extent *
255 map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
257 struct idmap_key key;
263 return bsearch(&key, map->forward, extents,
264 sizeof(struct uid_gid_extent), cmp_map_id);
268 * map_id_range_down_base - Find idmap via binary search in static extent array.
269 * Can only be called if number of mappings is equal or less than
270 * UID_GID_MAP_MAX_BASE_EXTENTS.
272 static struct uid_gid_extent *
273 map_id_range_down_base(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
276 u32 first, last, id2;
278 id2 = id + count - 1;
280 /* Find the matching extent */
281 for (idx = 0; idx < extents; idx++) {
282 first = map->extent[idx].first;
283 last = first + map->extent[idx].count - 1;
284 if (id >= first && id <= last &&
285 (id2 >= first && id2 <= last))
286 return &map->extent[idx];
291 static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
293 struct uid_gid_extent *extent;
294 unsigned extents = map->nr_extents;
297 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
298 extent = map_id_range_down_base(extents, map, id, count);
300 extent = map_id_range_down_max(extents, map, id, count);
302 /* Map the id or note failure */
304 id = (id - extent->first) + extent->lower_first;
311 static u32 map_id_down(struct uid_gid_map *map, u32 id)
313 return map_id_range_down(map, id, 1);
317 * map_id_up_base - Find idmap via binary search in static extent array.
318 * Can only be called if number of mappings is equal or less than
319 * UID_GID_MAP_MAX_BASE_EXTENTS.
321 static struct uid_gid_extent *
322 map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
327 /* Find the matching extent */
328 for (idx = 0; idx < extents; idx++) {
329 first = map->extent[idx].lower_first;
330 last = first + map->extent[idx].count - 1;
331 if (id >= first && id <= last)
332 return &map->extent[idx];
338 * map_id_up_max - Find idmap via binary search in ordered idmap array.
339 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
341 static struct uid_gid_extent *
342 map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
344 struct idmap_key key;
350 return bsearch(&key, map->reverse, extents,
351 sizeof(struct uid_gid_extent), cmp_map_id);
354 static u32 map_id_up(struct uid_gid_map *map, u32 id)
356 struct uid_gid_extent *extent;
357 unsigned extents = map->nr_extents;
360 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
361 extent = map_id_up_base(extents, map, id);
363 extent = map_id_up_max(extents, map, id);
365 /* Map the id or note failure */
367 id = (id - extent->lower_first) + extent->first;
375 * make_kuid - Map a user-namespace uid pair into a kuid.
376 * @ns: User namespace that the uid is in
377 * @uid: User identifier
379 * Maps a user-namespace uid pair into a kernel internal kuid,
380 * and returns that kuid.
382 * When there is no mapping defined for the user-namespace uid
383 * pair INVALID_UID is returned. Callers are expected to test
384 * for and handle INVALID_UID being returned. INVALID_UID
385 * may be tested for using uid_valid().
387 kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
389 /* Map the uid to a global kernel uid */
390 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
392 EXPORT_SYMBOL(make_kuid);
395 * from_kuid - Create a uid from a kuid user-namespace pair.
396 * @targ: The user namespace we want a uid in.
397 * @kuid: The kernel internal uid to start with.
399 * Map @kuid into the user-namespace specified by @targ and
400 * return the resulting uid.
402 * There is always a mapping into the initial user_namespace.
404 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
406 uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
408 /* Map the uid from a global kernel uid */
409 return map_id_up(&targ->uid_map, __kuid_val(kuid));
411 EXPORT_SYMBOL(from_kuid);
414 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
415 * @targ: The user namespace we want a uid in.
416 * @kuid: The kernel internal uid to start with.
418 * Map @kuid into the user-namespace specified by @targ and
419 * return the resulting uid.
421 * There is always a mapping into the initial user_namespace.
423 * Unlike from_kuid from_kuid_munged never fails and always
424 * returns a valid uid. This makes from_kuid_munged appropriate
425 * for use in syscalls like stat and getuid where failing the
426 * system call and failing to provide a valid uid are not an
429 * If @kuid has no mapping in @targ overflowuid is returned.
431 uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
434 uid = from_kuid(targ, kuid);
436 if (uid == (uid_t) -1)
440 EXPORT_SYMBOL(from_kuid_munged);
443 * make_kgid - Map a user-namespace gid pair into a kgid.
444 * @ns: User namespace that the gid is in
445 * @gid: group identifier
447 * Maps a user-namespace gid pair into a kernel internal kgid,
448 * and returns that kgid.
450 * When there is no mapping defined for the user-namespace gid
451 * pair INVALID_GID is returned. Callers are expected to test
452 * for and handle INVALID_GID being returned. INVALID_GID may be
453 * tested for using gid_valid().
455 kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
457 /* Map the gid to a global kernel gid */
458 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
460 EXPORT_SYMBOL(make_kgid);
463 * from_kgid - Create a gid from a kgid user-namespace pair.
464 * @targ: The user namespace we want a gid in.
465 * @kgid: The kernel internal gid to start with.
467 * Map @kgid into the user-namespace specified by @targ and
468 * return the resulting gid.
470 * There is always a mapping into the initial user_namespace.
472 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
474 gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
476 /* Map the gid from a global kernel gid */
477 return map_id_up(&targ->gid_map, __kgid_val(kgid));
479 EXPORT_SYMBOL(from_kgid);
482 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
483 * @targ: The user namespace we want a gid in.
484 * @kgid: The kernel internal gid to start with.
486 * Map @kgid into the user-namespace specified by @targ and
487 * return the resulting gid.
489 * There is always a mapping into the initial user_namespace.
491 * Unlike from_kgid from_kgid_munged never fails and always
492 * returns a valid gid. This makes from_kgid_munged appropriate
493 * for use in syscalls like stat and getgid where failing the
494 * system call and failing to provide a valid gid are not options.
496 * If @kgid has no mapping in @targ overflowgid is returned.
498 gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
501 gid = from_kgid(targ, kgid);
503 if (gid == (gid_t) -1)
507 EXPORT_SYMBOL(from_kgid_munged);
510 * make_kprojid - Map a user-namespace projid pair into a kprojid.
511 * @ns: User namespace that the projid is in
512 * @projid: Project identifier
514 * Maps a user-namespace uid pair into a kernel internal kuid,
515 * and returns that kuid.
517 * When there is no mapping defined for the user-namespace projid
518 * pair INVALID_PROJID is returned. Callers are expected to test
519 * for and handle INVALID_PROJID being returned. INVALID_PROJID
520 * may be tested for using projid_valid().
522 kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
524 /* Map the uid to a global kernel uid */
525 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
527 EXPORT_SYMBOL(make_kprojid);
530 * from_kprojid - Create a projid from a kprojid user-namespace pair.
531 * @targ: The user namespace we want a projid in.
532 * @kprojid: The kernel internal project identifier to start with.
534 * Map @kprojid into the user-namespace specified by @targ and
535 * return the resulting projid.
537 * There is always a mapping into the initial user_namespace.
539 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
541 projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
543 /* Map the uid from a global kernel uid */
544 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
546 EXPORT_SYMBOL(from_kprojid);
549 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
550 * @targ: The user namespace we want a projid in.
551 * @kprojid: The kernel internal projid to start with.
553 * Map @kprojid into the user-namespace specified by @targ and
554 * return the resulting projid.
556 * There is always a mapping into the initial user_namespace.
558 * Unlike from_kprojid from_kprojid_munged never fails and always
559 * returns a valid projid. This makes from_kprojid_munged
560 * appropriate for use in syscalls like stat and where
561 * failing the system call and failing to provide a valid projid are
564 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
566 projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
569 projid = from_kprojid(targ, kprojid);
571 if (projid == (projid_t) -1)
572 projid = OVERFLOW_PROJID;
575 EXPORT_SYMBOL(from_kprojid_munged);
578 static int uid_m_show(struct seq_file *seq, void *v)
580 struct user_namespace *ns = seq->private;
581 struct uid_gid_extent *extent = v;
582 struct user_namespace *lower_ns;
585 lower_ns = seq_user_ns(seq);
586 if ((lower_ns == ns) && lower_ns->parent)
587 lower_ns = lower_ns->parent;
589 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
591 seq_printf(seq, "%10u %10u %10u\n",
599 static int gid_m_show(struct seq_file *seq, void *v)
601 struct user_namespace *ns = seq->private;
602 struct uid_gid_extent *extent = v;
603 struct user_namespace *lower_ns;
606 lower_ns = seq_user_ns(seq);
607 if ((lower_ns == ns) && lower_ns->parent)
608 lower_ns = lower_ns->parent;
610 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
612 seq_printf(seq, "%10u %10u %10u\n",
620 static int projid_m_show(struct seq_file *seq, void *v)
622 struct user_namespace *ns = seq->private;
623 struct uid_gid_extent *extent = v;
624 struct user_namespace *lower_ns;
627 lower_ns = seq_user_ns(seq);
628 if ((lower_ns == ns) && lower_ns->parent)
629 lower_ns = lower_ns->parent;
631 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
633 seq_printf(seq, "%10u %10u %10u\n",
641 static void *m_start(struct seq_file *seq, loff_t *ppos,
642 struct uid_gid_map *map)
645 unsigned extents = map->nr_extents;
651 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
652 return &map->extent[pos];
654 return &map->forward[pos];
657 static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
659 struct user_namespace *ns = seq->private;
661 return m_start(seq, ppos, &ns->uid_map);
664 static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
666 struct user_namespace *ns = seq->private;
668 return m_start(seq, ppos, &ns->gid_map);
671 static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
673 struct user_namespace *ns = seq->private;
675 return m_start(seq, ppos, &ns->projid_map);
678 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
681 return seq->op->start(seq, pos);
684 static void m_stop(struct seq_file *seq, void *v)
689 const struct seq_operations proc_uid_seq_operations = {
690 .start = uid_m_start,
696 const struct seq_operations proc_gid_seq_operations = {
697 .start = gid_m_start,
703 const struct seq_operations proc_projid_seq_operations = {
704 .start = projid_m_start,
707 .show = projid_m_show,
710 static bool mappings_overlap(struct uid_gid_map *new_map,
711 struct uid_gid_extent *extent)
713 u32 upper_first, lower_first, upper_last, lower_last;
716 upper_first = extent->first;
717 lower_first = extent->lower_first;
718 upper_last = upper_first + extent->count - 1;
719 lower_last = lower_first + extent->count - 1;
721 for (idx = 0; idx < new_map->nr_extents; idx++) {
722 u32 prev_upper_first, prev_lower_first;
723 u32 prev_upper_last, prev_lower_last;
724 struct uid_gid_extent *prev;
726 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
727 prev = &new_map->extent[idx];
729 prev = &new_map->forward[idx];
731 prev_upper_first = prev->first;
732 prev_lower_first = prev->lower_first;
733 prev_upper_last = prev_upper_first + prev->count - 1;
734 prev_lower_last = prev_lower_first + prev->count - 1;
736 /* Does the upper range intersect a previous extent? */
737 if ((prev_upper_first <= upper_last) &&
738 (prev_upper_last >= upper_first))
741 /* Does the lower range intersect a previous extent? */
742 if ((prev_lower_first <= lower_last) &&
743 (prev_lower_last >= lower_first))
750 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
751 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
752 * UID_GID_MAP_MAX_BASE_EXTENTS.
754 static int insert_extent(struct uid_gid_map *map, struct uid_gid_extent *extent)
756 struct uid_gid_extent *dest;
758 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
759 struct uid_gid_extent *forward;
761 /* Allocate memory for 340 mappings. */
762 forward = kmalloc_array(UID_GID_MAP_MAX_EXTENTS,
763 sizeof(struct uid_gid_extent),
768 /* Copy over memory. Only set up memory for the forward pointer.
769 * Defer the memory setup for the reverse pointer.
771 memcpy(forward, map->extent,
772 map->nr_extents * sizeof(map->extent[0]));
774 map->forward = forward;
778 if (map->nr_extents < UID_GID_MAP_MAX_BASE_EXTENTS)
779 dest = &map->extent[map->nr_extents];
781 dest = &map->forward[map->nr_extents];
788 /* cmp function to sort() forward mappings */
789 static int cmp_extents_forward(const void *a, const void *b)
791 const struct uid_gid_extent *e1 = a;
792 const struct uid_gid_extent *e2 = b;
794 if (e1->first < e2->first)
797 if (e1->first > e2->first)
803 /* cmp function to sort() reverse mappings */
804 static int cmp_extents_reverse(const void *a, const void *b)
806 const struct uid_gid_extent *e1 = a;
807 const struct uid_gid_extent *e2 = b;
809 if (e1->lower_first < e2->lower_first)
812 if (e1->lower_first > e2->lower_first)
819 * sort_idmaps - Sorts an array of idmap entries.
820 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
822 static int sort_idmaps(struct uid_gid_map *map)
824 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
827 /* Sort forward array. */
828 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
829 cmp_extents_forward, NULL);
831 /* Only copy the memory from forward we actually need. */
832 map->reverse = kmemdup(map->forward,
833 map->nr_extents * sizeof(struct uid_gid_extent),
838 /* Sort reverse array. */
839 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
840 cmp_extents_reverse, NULL);
846 * verify_root_map() - check the uid 0 mapping
847 * @file: idmapping file
848 * @map_ns: user namespace of the target process
849 * @new_map: requested idmap
851 * If a process requests mapping parent uid 0 into the new ns, verify that the
852 * process writing the map had the CAP_SETFCAP capability as the target process
853 * will be able to write fscaps that are valid in ancestor user namespaces.
855 * Return: true if the mapping is allowed, false if not.
857 static bool verify_root_map(const struct file *file,
858 struct user_namespace *map_ns,
859 struct uid_gid_map *new_map)
862 const struct user_namespace *file_ns = file->f_cred->user_ns;
863 struct uid_gid_extent *extent0 = NULL;
865 for (idx = 0; idx < new_map->nr_extents; idx++) {
866 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
867 extent0 = &new_map->extent[idx];
869 extent0 = &new_map->forward[idx];
870 if (extent0->lower_first == 0)
879 if (map_ns == file_ns) {
880 /* The process unshared its ns and is writing to its own
881 * /proc/self/uid_map. User already has full capabilites in
882 * the new namespace. Verify that the parent had CAP_SETFCAP
885 if (!file_ns->parent_could_setfcap)
888 /* Process p1 is writing to uid_map of p2, who is in a child
889 * user namespace to p1's. Verify that the opener of the map
890 * file has CAP_SETFCAP against the parent of the new map
892 if (!file_ns_capable(file, map_ns->parent, CAP_SETFCAP))
899 static ssize_t map_write(struct file *file, const char __user *buf,
900 size_t count, loff_t *ppos,
902 struct uid_gid_map *map,
903 struct uid_gid_map *parent_map)
905 struct seq_file *seq = file->private_data;
906 struct user_namespace *map_ns = seq->private;
907 struct uid_gid_map new_map;
909 struct uid_gid_extent extent;
910 char *kbuf = NULL, *pos, *next_line;
913 /* Only allow < page size writes at the beginning of the file */
914 if ((*ppos != 0) || (count >= PAGE_SIZE))
917 /* Slurp in the user data */
918 kbuf = memdup_user_nul(buf, count);
920 return PTR_ERR(kbuf);
923 * The userns_state_mutex serializes all writes to any given map.
925 * Any map is only ever written once.
927 * An id map fits within 1 cache line on most architectures.
929 * On read nothing needs to be done unless you are on an
930 * architecture with a crazy cache coherency model like alpha.
932 * There is a one time data dependency between reading the
933 * count of the extents and the values of the extents. The
934 * desired behavior is to see the values of the extents that
935 * were written before the count of the extents.
937 * To achieve this smp_wmb() is used on guarantee the write
938 * order and smp_rmb() is guaranteed that we don't have crazy
939 * architectures returning stale data.
941 mutex_lock(&userns_state_mutex);
943 memset(&new_map, 0, sizeof(struct uid_gid_map));
946 /* Only allow one successful write to the map */
947 if (map->nr_extents != 0)
951 * Adjusting namespace settings requires capabilities on the target.
953 if (cap_valid(cap_setid) && !file_ns_capable(file, map_ns, CAP_SYS_ADMIN))
956 /* Parse the user data */
959 for (; pos; pos = next_line) {
961 /* Find the end of line and ensure I don't look past it */
962 next_line = strchr(pos, '\n');
966 if (*next_line == '\0')
970 pos = skip_spaces(pos);
971 extent.first = simple_strtoul(pos, &pos, 10);
975 pos = skip_spaces(pos);
976 extent.lower_first = simple_strtoul(pos, &pos, 10);
980 pos = skip_spaces(pos);
981 extent.count = simple_strtoul(pos, &pos, 10);
982 if (*pos && !isspace(*pos))
985 /* Verify there is not trailing junk on the line */
986 pos = skip_spaces(pos);
990 /* Verify we have been given valid starting values */
991 if ((extent.first == (u32) -1) ||
992 (extent.lower_first == (u32) -1))
995 /* Verify count is not zero and does not cause the
998 if ((extent.first + extent.count) <= extent.first)
1000 if ((extent.lower_first + extent.count) <=
1004 /* Do the ranges in extent overlap any previous extents? */
1005 if (mappings_overlap(&new_map, &extent))
1008 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
1009 (next_line != NULL))
1012 ret = insert_extent(&new_map, &extent);
1017 /* Be very certain the new map actually exists */
1018 if (new_map.nr_extents == 0)
1022 /* Validate the user is allowed to use user id's mapped to. */
1023 if (!new_idmap_permitted(file, map_ns, cap_setid, &new_map))
1027 /* Map the lower ids from the parent user namespace to the
1028 * kernel global id space.
1030 for (idx = 0; idx < new_map.nr_extents; idx++) {
1031 struct uid_gid_extent *e;
1034 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
1035 e = &new_map.extent[idx];
1037 e = &new_map.forward[idx];
1039 lower_first = map_id_range_down(parent_map,
1043 /* Fail if we can not map the specified extent to
1044 * the kernel global id space.
1046 if (lower_first == (u32) -1)
1049 e->lower_first = lower_first;
1053 * If we want to use binary search for lookup, this clones the extent
1054 * array and sorts both copies.
1056 ret = sort_idmaps(&new_map);
1060 /* Install the map */
1061 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1062 memcpy(map->extent, new_map.extent,
1063 new_map.nr_extents * sizeof(new_map.extent[0]));
1065 map->forward = new_map.forward;
1066 map->reverse = new_map.reverse;
1069 map->nr_extents = new_map.nr_extents;
1074 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1075 kfree(new_map.forward);
1076 kfree(new_map.reverse);
1077 map->forward = NULL;
1078 map->reverse = NULL;
1079 map->nr_extents = 0;
1082 mutex_unlock(&userns_state_mutex);
1087 ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1088 size_t size, loff_t *ppos)
1090 struct seq_file *seq = file->private_data;
1091 struct user_namespace *ns = seq->private;
1092 struct user_namespace *seq_ns = seq_user_ns(seq);
1097 if ((seq_ns != ns) && (seq_ns != ns->parent))
1100 return map_write(file, buf, size, ppos, CAP_SETUID,
1101 &ns->uid_map, &ns->parent->uid_map);
1104 ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1105 size_t size, loff_t *ppos)
1107 struct seq_file *seq = file->private_data;
1108 struct user_namespace *ns = seq->private;
1109 struct user_namespace *seq_ns = seq_user_ns(seq);
1114 if ((seq_ns != ns) && (seq_ns != ns->parent))
1117 return map_write(file, buf, size, ppos, CAP_SETGID,
1118 &ns->gid_map, &ns->parent->gid_map);
1121 ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1122 size_t size, loff_t *ppos)
1124 struct seq_file *seq = file->private_data;
1125 struct user_namespace *ns = seq->private;
1126 struct user_namespace *seq_ns = seq_user_ns(seq);
1131 if ((seq_ns != ns) && (seq_ns != ns->parent))
1134 /* Anyone can set any valid project id no capability needed */
1135 return map_write(file, buf, size, ppos, -1,
1136 &ns->projid_map, &ns->parent->projid_map);
1139 static bool new_idmap_permitted(const struct file *file,
1140 struct user_namespace *ns, int cap_setid,
1141 struct uid_gid_map *new_map)
1143 const struct cred *cred = file->f_cred;
1145 if (cap_setid == CAP_SETUID && !verify_root_map(file, ns, new_map))
1148 /* Don't allow mappings that would allow anything that wouldn't
1149 * be allowed without the establishment of unprivileged mappings.
1151 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1152 uid_eq(ns->owner, cred->euid)) {
1153 u32 id = new_map->extent[0].lower_first;
1154 if (cap_setid == CAP_SETUID) {
1155 kuid_t uid = make_kuid(ns->parent, id);
1156 if (uid_eq(uid, cred->euid))
1158 } else if (cap_setid == CAP_SETGID) {
1159 kgid_t gid = make_kgid(ns->parent, id);
1160 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1161 gid_eq(gid, cred->egid))
1166 /* Allow anyone to set a mapping that doesn't require privilege */
1167 if (!cap_valid(cap_setid))
1170 /* Allow the specified ids if we have the appropriate capability
1171 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1172 * And the opener of the id file also has the appropriate capability.
1174 if (ns_capable(ns->parent, cap_setid) &&
1175 file_ns_capable(file, ns->parent, cap_setid))
1181 int proc_setgroups_show(struct seq_file *seq, void *v)
1183 struct user_namespace *ns = seq->private;
1184 unsigned long userns_flags = READ_ONCE(ns->flags);
1186 seq_printf(seq, "%s\n",
1187 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1192 ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1193 size_t count, loff_t *ppos)
1195 struct seq_file *seq = file->private_data;
1196 struct user_namespace *ns = seq->private;
1198 bool setgroups_allowed;
1201 /* Only allow a very narrow range of strings to be written */
1203 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1206 /* What was written? */
1208 if (copy_from_user(kbuf, buf, count))
1213 /* What is being requested? */
1215 if (strncmp(pos, "allow", 5) == 0) {
1217 setgroups_allowed = true;
1219 else if (strncmp(pos, "deny", 4) == 0) {
1221 setgroups_allowed = false;
1226 /* Verify there is not trailing junk on the line */
1227 pos = skip_spaces(pos);
1232 mutex_lock(&userns_state_mutex);
1233 if (setgroups_allowed) {
1234 /* Enabling setgroups after setgroups has been disabled
1237 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1240 /* Permanently disabling setgroups after setgroups has
1241 * been enabled by writing the gid_map is not allowed.
1243 if (ns->gid_map.nr_extents != 0)
1245 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1247 mutex_unlock(&userns_state_mutex);
1249 /* Report a successful write */
1255 mutex_unlock(&userns_state_mutex);
1259 bool userns_may_setgroups(const struct user_namespace *ns)
1263 mutex_lock(&userns_state_mutex);
1264 /* It is not safe to use setgroups until a gid mapping in
1265 * the user namespace has been established.
1267 allowed = ns->gid_map.nr_extents != 0;
1268 /* Is setgroups allowed? */
1269 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1270 mutex_unlock(&userns_state_mutex);
1276 * Returns true if @child is the same namespace or a descendant of
1279 bool in_userns(const struct user_namespace *ancestor,
1280 const struct user_namespace *child)
1282 const struct user_namespace *ns;
1283 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1285 return (ns == ancestor);
1288 bool current_in_userns(const struct user_namespace *target_ns)
1290 return in_userns(target_ns, current_user_ns());
1292 EXPORT_SYMBOL(current_in_userns);
1294 static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1296 return container_of(ns, struct user_namespace, ns);
1299 static struct ns_common *userns_get(struct task_struct *task)
1301 struct user_namespace *user_ns;
1304 user_ns = get_user_ns(__task_cred(task)->user_ns);
1307 return user_ns ? &user_ns->ns : NULL;
1310 static void userns_put(struct ns_common *ns)
1312 put_user_ns(to_user_ns(ns));
1315 static int userns_install(struct nsset *nsset, struct ns_common *ns)
1317 struct user_namespace *user_ns = to_user_ns(ns);
1320 /* Don't allow gaining capabilities by reentering
1321 * the same user namespace.
1323 if (user_ns == current_user_ns())
1326 /* Tasks that share a thread group must share a user namespace */
1327 if (!thread_group_empty(current))
1330 if (current->fs->users != 1)
1333 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1336 cred = nsset_cred(nsset);
1340 put_user_ns(cred->user_ns);
1341 set_cred_user_ns(cred, get_user_ns(user_ns));
1346 struct ns_common *ns_get_owner(struct ns_common *ns)
1348 struct user_namespace *my_user_ns = current_user_ns();
1349 struct user_namespace *owner, *p;
1351 /* See if the owner is in the current user namespace */
1352 owner = p = ns->ops->owner(ns);
1355 return ERR_PTR(-EPERM);
1356 if (p == my_user_ns)
1361 return &get_user_ns(owner)->ns;
1364 static struct user_namespace *userns_owner(struct ns_common *ns)
1366 return to_user_ns(ns)->parent;
1369 const struct proc_ns_operations userns_operations = {
1371 .type = CLONE_NEWUSER,
1374 .install = userns_install,
1375 .owner = userns_owner,
1376 .get_parent = ns_get_owner,
1379 static __init int user_namespaces_init(void)
1381 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1384 subsys_initcall(user_namespaces_init);