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 verifing 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 ret = ns_alloc_inum(&ns->ns);
112 ns->ns.ops = &userns_operations;
114 atomic_set(&ns->count, 1);
115 /* Leave the new->user_ns reference with the new user namespace. */
116 ns->parent = parent_ns;
117 ns->level = parent_ns->level + 1;
120 INIT_WORK(&ns->work, free_user_ns);
121 for (i = 0; i < UCOUNT_COUNTS; i++) {
122 ns->ucount_max[i] = INT_MAX;
124 ns->ucounts = ucounts;
126 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
127 mutex_lock(&userns_state_mutex);
128 ns->flags = parent_ns->flags;
129 mutex_unlock(&userns_state_mutex);
132 INIT_LIST_HEAD(&ns->keyring_name_list);
133 init_rwsem(&ns->keyring_sem);
136 if (!setup_userns_sysctls(ns))
139 set_cred_user_ns(new, ns);
142 #ifdef CONFIG_PERSISTENT_KEYRINGS
143 key_put(ns->persistent_keyring_register);
145 ns_free_inum(&ns->ns);
147 kmem_cache_free(user_ns_cachep, ns);
149 dec_user_namespaces(ucounts);
154 int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
159 if (!(unshare_flags & CLONE_NEWUSER))
162 cred = prepare_creds();
164 err = create_user_ns(cred);
174 static void free_user_ns(struct work_struct *work)
176 struct user_namespace *parent, *ns =
177 container_of(work, struct user_namespace, work);
180 struct ucounts *ucounts = ns->ucounts;
182 if (ns->gid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
183 kfree(ns->gid_map.forward);
184 kfree(ns->gid_map.reverse);
186 if (ns->uid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
187 kfree(ns->uid_map.forward);
188 kfree(ns->uid_map.reverse);
190 if (ns->projid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
191 kfree(ns->projid_map.forward);
192 kfree(ns->projid_map.reverse);
194 retire_userns_sysctls(ns);
195 key_free_user_ns(ns);
196 ns_free_inum(&ns->ns);
197 kmem_cache_free(user_ns_cachep, ns);
198 dec_user_namespaces(ucounts);
200 } while (atomic_dec_and_test(&parent->count));
203 void __put_user_ns(struct user_namespace *ns)
205 schedule_work(&ns->work);
207 EXPORT_SYMBOL(__put_user_ns);
210 * idmap_key struct holds the information necessary to find an idmapping in a
211 * sorted idmap array. It is passed to cmp_map_id() as first argument.
214 bool map_up; /* true -> id from kid; false -> kid from id */
215 u32 id; /* id to find */
216 u32 count; /* == 0 unless used with map_id_range_down() */
220 * cmp_map_id - Function to be passed to bsearch() to find the requested
221 * idmapping. Expects struct idmap_key to be passed via @k.
223 static int cmp_map_id(const void *k, const void *e)
225 u32 first, last, id2;
226 const struct idmap_key *key = k;
227 const struct uid_gid_extent *el = e;
229 id2 = key->id + key->count - 1;
231 /* handle map_id_{down,up}() */
233 first = el->lower_first;
237 last = first + el->count - 1;
239 if (key->id >= first && key->id <= last &&
240 (id2 >= first && id2 <= last))
243 if (key->id < first || id2 < first)
250 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
251 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
253 static struct uid_gid_extent *
254 map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
256 struct idmap_key key;
262 return bsearch(&key, map->forward, extents,
263 sizeof(struct uid_gid_extent), cmp_map_id);
267 * map_id_range_down_base - Find idmap via binary search in static extent array.
268 * Can only be called if number of mappings is equal or less than
269 * UID_GID_MAP_MAX_BASE_EXTENTS.
271 static struct uid_gid_extent *
272 map_id_range_down_base(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
275 u32 first, last, id2;
277 id2 = id + count - 1;
279 /* Find the matching extent */
280 for (idx = 0; idx < extents; idx++) {
281 first = map->extent[idx].first;
282 last = first + map->extent[idx].count - 1;
283 if (id >= first && id <= last &&
284 (id2 >= first && id2 <= last))
285 return &map->extent[idx];
290 static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
292 struct uid_gid_extent *extent;
293 unsigned extents = map->nr_extents;
296 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
297 extent = map_id_range_down_base(extents, map, id, count);
299 extent = map_id_range_down_max(extents, map, id, count);
301 /* Map the id or note failure */
303 id = (id - extent->first) + extent->lower_first;
310 static u32 map_id_down(struct uid_gid_map *map, u32 id)
312 return map_id_range_down(map, id, 1);
316 * map_id_up_base - Find idmap via binary search in static extent array.
317 * Can only be called if number of mappings is equal or less than
318 * UID_GID_MAP_MAX_BASE_EXTENTS.
320 static struct uid_gid_extent *
321 map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
326 /* Find the matching extent */
327 for (idx = 0; idx < extents; idx++) {
328 first = map->extent[idx].lower_first;
329 last = first + map->extent[idx].count - 1;
330 if (id >= first && id <= last)
331 return &map->extent[idx];
337 * map_id_up_max - Find idmap via binary search in ordered idmap array.
338 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
340 static struct uid_gid_extent *
341 map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
343 struct idmap_key key;
349 return bsearch(&key, map->reverse, extents,
350 sizeof(struct uid_gid_extent), cmp_map_id);
353 static u32 map_id_up(struct uid_gid_map *map, u32 id)
355 struct uid_gid_extent *extent;
356 unsigned extents = map->nr_extents;
359 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
360 extent = map_id_up_base(extents, map, id);
362 extent = map_id_up_max(extents, map, id);
364 /* Map the id or note failure */
366 id = (id - extent->lower_first) + extent->first;
374 * make_kuid - Map a user-namespace uid pair into a kuid.
375 * @ns: User namespace that the uid is in
376 * @uid: User identifier
378 * Maps a user-namespace uid pair into a kernel internal kuid,
379 * and returns that kuid.
381 * When there is no mapping defined for the user-namespace uid
382 * pair INVALID_UID is returned. Callers are expected to test
383 * for and handle INVALID_UID being returned. INVALID_UID
384 * may be tested for using uid_valid().
386 kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
388 /* Map the uid to a global kernel uid */
389 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
391 EXPORT_SYMBOL(make_kuid);
394 * from_kuid - Create a uid from a kuid user-namespace pair.
395 * @targ: The user namespace we want a uid in.
396 * @kuid: The kernel internal uid to start with.
398 * Map @kuid into the user-namespace specified by @targ and
399 * return the resulting uid.
401 * There is always a mapping into the initial user_namespace.
403 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
405 uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
407 /* Map the uid from a global kernel uid */
408 return map_id_up(&targ->uid_map, __kuid_val(kuid));
410 EXPORT_SYMBOL(from_kuid);
413 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
414 * @targ: The user namespace we want a uid in.
415 * @kuid: The kernel internal uid to start with.
417 * Map @kuid into the user-namespace specified by @targ and
418 * return the resulting uid.
420 * There is always a mapping into the initial user_namespace.
422 * Unlike from_kuid from_kuid_munged never fails and always
423 * returns a valid uid. This makes from_kuid_munged appropriate
424 * for use in syscalls like stat and getuid where failing the
425 * system call and failing to provide a valid uid are not an
428 * If @kuid has no mapping in @targ overflowuid is returned.
430 uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
433 uid = from_kuid(targ, kuid);
435 if (uid == (uid_t) -1)
439 EXPORT_SYMBOL(from_kuid_munged);
442 * make_kgid - Map a user-namespace gid pair into a kgid.
443 * @ns: User namespace that the gid is in
444 * @gid: group identifier
446 * Maps a user-namespace gid pair into a kernel internal kgid,
447 * and returns that kgid.
449 * When there is no mapping defined for the user-namespace gid
450 * pair INVALID_GID is returned. Callers are expected to test
451 * for and handle INVALID_GID being returned. INVALID_GID may be
452 * tested for using gid_valid().
454 kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
456 /* Map the gid to a global kernel gid */
457 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
459 EXPORT_SYMBOL(make_kgid);
462 * from_kgid - Create a gid from a kgid user-namespace pair.
463 * @targ: The user namespace we want a gid in.
464 * @kgid: The kernel internal gid to start with.
466 * Map @kgid into the user-namespace specified by @targ and
467 * return the resulting gid.
469 * There is always a mapping into the initial user_namespace.
471 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
473 gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
475 /* Map the gid from a global kernel gid */
476 return map_id_up(&targ->gid_map, __kgid_val(kgid));
478 EXPORT_SYMBOL(from_kgid);
481 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
482 * @targ: The user namespace we want a gid in.
483 * @kgid: The kernel internal gid to start with.
485 * Map @kgid into the user-namespace specified by @targ and
486 * return the resulting gid.
488 * There is always a mapping into the initial user_namespace.
490 * Unlike from_kgid from_kgid_munged never fails and always
491 * returns a valid gid. This makes from_kgid_munged appropriate
492 * for use in syscalls like stat and getgid where failing the
493 * system call and failing to provide a valid gid are not options.
495 * If @kgid has no mapping in @targ overflowgid is returned.
497 gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
500 gid = from_kgid(targ, kgid);
502 if (gid == (gid_t) -1)
506 EXPORT_SYMBOL(from_kgid_munged);
509 * make_kprojid - Map a user-namespace projid pair into a kprojid.
510 * @ns: User namespace that the projid is in
511 * @projid: Project identifier
513 * Maps a user-namespace uid pair into a kernel internal kuid,
514 * and returns that kuid.
516 * When there is no mapping defined for the user-namespace projid
517 * pair INVALID_PROJID is returned. Callers are expected to test
518 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
519 * may be tested for using projid_valid().
521 kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
523 /* Map the uid to a global kernel uid */
524 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
526 EXPORT_SYMBOL(make_kprojid);
529 * from_kprojid - Create a projid from a kprojid user-namespace pair.
530 * @targ: The user namespace we want a projid in.
531 * @kprojid: The kernel internal project identifier to start with.
533 * Map @kprojid into the user-namespace specified by @targ and
534 * return the resulting projid.
536 * There is always a mapping into the initial user_namespace.
538 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
540 projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
542 /* Map the uid from a global kernel uid */
543 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
545 EXPORT_SYMBOL(from_kprojid);
548 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
549 * @targ: The user namespace we want a projid in.
550 * @kprojid: The kernel internal projid to start with.
552 * Map @kprojid into the user-namespace specified by @targ and
553 * return the resulting projid.
555 * There is always a mapping into the initial user_namespace.
557 * Unlike from_kprojid from_kprojid_munged never fails and always
558 * returns a valid projid. This makes from_kprojid_munged
559 * appropriate for use in syscalls like stat and where
560 * failing the system call and failing to provide a valid projid are
563 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
565 projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
568 projid = from_kprojid(targ, kprojid);
570 if (projid == (projid_t) -1)
571 projid = OVERFLOW_PROJID;
574 EXPORT_SYMBOL(from_kprojid_munged);
577 static int uid_m_show(struct seq_file *seq, void *v)
579 struct user_namespace *ns = seq->private;
580 struct uid_gid_extent *extent = v;
581 struct user_namespace *lower_ns;
584 lower_ns = seq_user_ns(seq);
585 if ((lower_ns == ns) && lower_ns->parent)
586 lower_ns = lower_ns->parent;
588 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
590 seq_printf(seq, "%10u %10u %10u\n",
598 static int gid_m_show(struct seq_file *seq, void *v)
600 struct user_namespace *ns = seq->private;
601 struct uid_gid_extent *extent = v;
602 struct user_namespace *lower_ns;
605 lower_ns = seq_user_ns(seq);
606 if ((lower_ns == ns) && lower_ns->parent)
607 lower_ns = lower_ns->parent;
609 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
611 seq_printf(seq, "%10u %10u %10u\n",
619 static int projid_m_show(struct seq_file *seq, void *v)
621 struct user_namespace *ns = seq->private;
622 struct uid_gid_extent *extent = v;
623 struct user_namespace *lower_ns;
626 lower_ns = seq_user_ns(seq);
627 if ((lower_ns == ns) && lower_ns->parent)
628 lower_ns = lower_ns->parent;
630 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
632 seq_printf(seq, "%10u %10u %10u\n",
640 static void *m_start(struct seq_file *seq, loff_t *ppos,
641 struct uid_gid_map *map)
644 unsigned extents = map->nr_extents;
650 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
651 return &map->extent[pos];
653 return &map->forward[pos];
656 static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
658 struct user_namespace *ns = seq->private;
660 return m_start(seq, ppos, &ns->uid_map);
663 static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
665 struct user_namespace *ns = seq->private;
667 return m_start(seq, ppos, &ns->gid_map);
670 static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
672 struct user_namespace *ns = seq->private;
674 return m_start(seq, ppos, &ns->projid_map);
677 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
680 return seq->op->start(seq, pos);
683 static void m_stop(struct seq_file *seq, void *v)
688 const struct seq_operations proc_uid_seq_operations = {
689 .start = uid_m_start,
695 const struct seq_operations proc_gid_seq_operations = {
696 .start = gid_m_start,
702 const struct seq_operations proc_projid_seq_operations = {
703 .start = projid_m_start,
706 .show = projid_m_show,
709 static bool mappings_overlap(struct uid_gid_map *new_map,
710 struct uid_gid_extent *extent)
712 u32 upper_first, lower_first, upper_last, lower_last;
715 upper_first = extent->first;
716 lower_first = extent->lower_first;
717 upper_last = upper_first + extent->count - 1;
718 lower_last = lower_first + extent->count - 1;
720 for (idx = 0; idx < new_map->nr_extents; idx++) {
721 u32 prev_upper_first, prev_lower_first;
722 u32 prev_upper_last, prev_lower_last;
723 struct uid_gid_extent *prev;
725 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
726 prev = &new_map->extent[idx];
728 prev = &new_map->forward[idx];
730 prev_upper_first = prev->first;
731 prev_lower_first = prev->lower_first;
732 prev_upper_last = prev_upper_first + prev->count - 1;
733 prev_lower_last = prev_lower_first + prev->count - 1;
735 /* Does the upper range intersect a previous extent? */
736 if ((prev_upper_first <= upper_last) &&
737 (prev_upper_last >= upper_first))
740 /* Does the lower range intersect a previous extent? */
741 if ((prev_lower_first <= lower_last) &&
742 (prev_lower_last >= lower_first))
749 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
750 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
751 * UID_GID_MAP_MAX_BASE_EXTENTS.
753 static int insert_extent(struct uid_gid_map *map, struct uid_gid_extent *extent)
755 struct uid_gid_extent *dest;
757 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
758 struct uid_gid_extent *forward;
760 /* Allocate memory for 340 mappings. */
761 forward = kmalloc_array(UID_GID_MAP_MAX_EXTENTS,
762 sizeof(struct uid_gid_extent),
767 /* Copy over memory. Only set up memory for the forward pointer.
768 * Defer the memory setup for the reverse pointer.
770 memcpy(forward, map->extent,
771 map->nr_extents * sizeof(map->extent[0]));
773 map->forward = forward;
777 if (map->nr_extents < UID_GID_MAP_MAX_BASE_EXTENTS)
778 dest = &map->extent[map->nr_extents];
780 dest = &map->forward[map->nr_extents];
787 /* cmp function to sort() forward mappings */
788 static int cmp_extents_forward(const void *a, const void *b)
790 const struct uid_gid_extent *e1 = a;
791 const struct uid_gid_extent *e2 = b;
793 if (e1->first < e2->first)
796 if (e1->first > e2->first)
802 /* cmp function to sort() reverse mappings */
803 static int cmp_extents_reverse(const void *a, const void *b)
805 const struct uid_gid_extent *e1 = a;
806 const struct uid_gid_extent *e2 = b;
808 if (e1->lower_first < e2->lower_first)
811 if (e1->lower_first > e2->lower_first)
818 * sort_idmaps - Sorts an array of idmap entries.
819 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
821 static int sort_idmaps(struct uid_gid_map *map)
823 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
826 /* Sort forward array. */
827 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
828 cmp_extents_forward, NULL);
830 /* Only copy the memory from forward we actually need. */
831 map->reverse = kmemdup(map->forward,
832 map->nr_extents * sizeof(struct uid_gid_extent),
837 /* Sort reverse array. */
838 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
839 cmp_extents_reverse, NULL);
844 static ssize_t map_write(struct file *file, const char __user *buf,
845 size_t count, loff_t *ppos,
847 struct uid_gid_map *map,
848 struct uid_gid_map *parent_map)
850 struct seq_file *seq = file->private_data;
851 struct user_namespace *ns = seq->private;
852 struct uid_gid_map new_map;
854 struct uid_gid_extent extent;
855 char *kbuf = NULL, *pos, *next_line;
858 /* Only allow < page size writes at the beginning of the file */
859 if ((*ppos != 0) || (count >= PAGE_SIZE))
862 /* Slurp in the user data */
863 kbuf = memdup_user_nul(buf, count);
865 return PTR_ERR(kbuf);
868 * The userns_state_mutex serializes all writes to any given map.
870 * Any map is only ever written once.
872 * An id map fits within 1 cache line on most architectures.
874 * On read nothing needs to be done unless you are on an
875 * architecture with a crazy cache coherency model like alpha.
877 * There is a one time data dependency between reading the
878 * count of the extents and the values of the extents. The
879 * desired behavior is to see the values of the extents that
880 * were written before the count of the extents.
882 * To achieve this smp_wmb() is used on guarantee the write
883 * order and smp_rmb() is guaranteed that we don't have crazy
884 * architectures returning stale data.
886 mutex_lock(&userns_state_mutex);
888 memset(&new_map, 0, sizeof(struct uid_gid_map));
891 /* Only allow one successful write to the map */
892 if (map->nr_extents != 0)
896 * Adjusting namespace settings requires capabilities on the target.
898 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
901 /* Parse the user data */
904 for (; pos; pos = next_line) {
906 /* Find the end of line and ensure I don't look past it */
907 next_line = strchr(pos, '\n');
911 if (*next_line == '\0')
915 pos = skip_spaces(pos);
916 extent.first = simple_strtoul(pos, &pos, 10);
920 pos = skip_spaces(pos);
921 extent.lower_first = simple_strtoul(pos, &pos, 10);
925 pos = skip_spaces(pos);
926 extent.count = simple_strtoul(pos, &pos, 10);
927 if (*pos && !isspace(*pos))
930 /* Verify there is not trailing junk on the line */
931 pos = skip_spaces(pos);
935 /* Verify we have been given valid starting values */
936 if ((extent.first == (u32) -1) ||
937 (extent.lower_first == (u32) -1))
940 /* Verify count is not zero and does not cause the
943 if ((extent.first + extent.count) <= extent.first)
945 if ((extent.lower_first + extent.count) <=
949 /* Do the ranges in extent overlap any previous extents? */
950 if (mappings_overlap(&new_map, &extent))
953 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
957 ret = insert_extent(&new_map, &extent);
962 /* Be very certaint the new map actually exists */
963 if (new_map.nr_extents == 0)
967 /* Validate the user is allowed to use user id's mapped to. */
968 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
972 /* Map the lower ids from the parent user namespace to the
973 * kernel global id space.
975 for (idx = 0; idx < new_map.nr_extents; idx++) {
976 struct uid_gid_extent *e;
979 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
980 e = &new_map.extent[idx];
982 e = &new_map.forward[idx];
984 lower_first = map_id_range_down(parent_map,
988 /* Fail if we can not map the specified extent to
989 * the kernel global id space.
991 if (lower_first == (u32) -1)
994 e->lower_first = lower_first;
998 * If we want to use binary search for lookup, this clones the extent
999 * array and sorts both copies.
1001 ret = sort_idmaps(&new_map);
1005 /* Install the map */
1006 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1007 memcpy(map->extent, new_map.extent,
1008 new_map.nr_extents * sizeof(new_map.extent[0]));
1010 map->forward = new_map.forward;
1011 map->reverse = new_map.reverse;
1014 map->nr_extents = new_map.nr_extents;
1019 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1020 kfree(new_map.forward);
1021 kfree(new_map.reverse);
1022 map->forward = NULL;
1023 map->reverse = NULL;
1024 map->nr_extents = 0;
1027 mutex_unlock(&userns_state_mutex);
1032 ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1033 size_t size, loff_t *ppos)
1035 struct seq_file *seq = file->private_data;
1036 struct user_namespace *ns = seq->private;
1037 struct user_namespace *seq_ns = seq_user_ns(seq);
1042 if ((seq_ns != ns) && (seq_ns != ns->parent))
1045 return map_write(file, buf, size, ppos, CAP_SETUID,
1046 &ns->uid_map, &ns->parent->uid_map);
1049 ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1050 size_t size, loff_t *ppos)
1052 struct seq_file *seq = file->private_data;
1053 struct user_namespace *ns = seq->private;
1054 struct user_namespace *seq_ns = seq_user_ns(seq);
1059 if ((seq_ns != ns) && (seq_ns != ns->parent))
1062 return map_write(file, buf, size, ppos, CAP_SETGID,
1063 &ns->gid_map, &ns->parent->gid_map);
1066 ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1067 size_t size, loff_t *ppos)
1069 struct seq_file *seq = file->private_data;
1070 struct user_namespace *ns = seq->private;
1071 struct user_namespace *seq_ns = seq_user_ns(seq);
1076 if ((seq_ns != ns) && (seq_ns != ns->parent))
1079 /* Anyone can set any valid project id no capability needed */
1080 return map_write(file, buf, size, ppos, -1,
1081 &ns->projid_map, &ns->parent->projid_map);
1084 static bool new_idmap_permitted(const struct file *file,
1085 struct user_namespace *ns, int cap_setid,
1086 struct uid_gid_map *new_map)
1088 const struct cred *cred = file->f_cred;
1089 /* Don't allow mappings that would allow anything that wouldn't
1090 * be allowed without the establishment of unprivileged mappings.
1092 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1093 uid_eq(ns->owner, cred->euid)) {
1094 u32 id = new_map->extent[0].lower_first;
1095 if (cap_setid == CAP_SETUID) {
1096 kuid_t uid = make_kuid(ns->parent, id);
1097 if (uid_eq(uid, cred->euid))
1099 } else if (cap_setid == CAP_SETGID) {
1100 kgid_t gid = make_kgid(ns->parent, id);
1101 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1102 gid_eq(gid, cred->egid))
1107 /* Allow anyone to set a mapping that doesn't require privilege */
1108 if (!cap_valid(cap_setid))
1111 /* Allow the specified ids if we have the appropriate capability
1112 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1113 * And the opener of the id file also had the approprpiate capability.
1115 if (ns_capable(ns->parent, cap_setid) &&
1116 file_ns_capable(file, ns->parent, cap_setid))
1122 int proc_setgroups_show(struct seq_file *seq, void *v)
1124 struct user_namespace *ns = seq->private;
1125 unsigned long userns_flags = READ_ONCE(ns->flags);
1127 seq_printf(seq, "%s\n",
1128 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1133 ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1134 size_t count, loff_t *ppos)
1136 struct seq_file *seq = file->private_data;
1137 struct user_namespace *ns = seq->private;
1139 bool setgroups_allowed;
1142 /* Only allow a very narrow range of strings to be written */
1144 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1147 /* What was written? */
1149 if (copy_from_user(kbuf, buf, count))
1154 /* What is being requested? */
1156 if (strncmp(pos, "allow", 5) == 0) {
1158 setgroups_allowed = true;
1160 else if (strncmp(pos, "deny", 4) == 0) {
1162 setgroups_allowed = false;
1167 /* Verify there is not trailing junk on the line */
1168 pos = skip_spaces(pos);
1173 mutex_lock(&userns_state_mutex);
1174 if (setgroups_allowed) {
1175 /* Enabling setgroups after setgroups has been disabled
1178 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1181 /* Permanently disabling setgroups after setgroups has
1182 * been enabled by writing the gid_map is not allowed.
1184 if (ns->gid_map.nr_extents != 0)
1186 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1188 mutex_unlock(&userns_state_mutex);
1190 /* Report a successful write */
1196 mutex_unlock(&userns_state_mutex);
1200 bool userns_may_setgroups(const struct user_namespace *ns)
1204 mutex_lock(&userns_state_mutex);
1205 /* It is not safe to use setgroups until a gid mapping in
1206 * the user namespace has been established.
1208 allowed = ns->gid_map.nr_extents != 0;
1209 /* Is setgroups allowed? */
1210 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1211 mutex_unlock(&userns_state_mutex);
1217 * Returns true if @child is the same namespace or a descendant of
1220 bool in_userns(const struct user_namespace *ancestor,
1221 const struct user_namespace *child)
1223 const struct user_namespace *ns;
1224 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1226 return (ns == ancestor);
1229 bool current_in_userns(const struct user_namespace *target_ns)
1231 return in_userns(target_ns, current_user_ns());
1233 EXPORT_SYMBOL(current_in_userns);
1235 static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1237 return container_of(ns, struct user_namespace, ns);
1240 static struct ns_common *userns_get(struct task_struct *task)
1242 struct user_namespace *user_ns;
1245 user_ns = get_user_ns(__task_cred(task)->user_ns);
1248 return user_ns ? &user_ns->ns : NULL;
1251 static void userns_put(struct ns_common *ns)
1253 put_user_ns(to_user_ns(ns));
1256 static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1258 struct user_namespace *user_ns = to_user_ns(ns);
1261 /* Don't allow gaining capabilities by reentering
1262 * the same user namespace.
1264 if (user_ns == current_user_ns())
1267 /* Tasks that share a thread group must share a user namespace */
1268 if (!thread_group_empty(current))
1271 if (current->fs->users != 1)
1274 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1277 cred = prepare_creds();
1281 put_user_ns(cred->user_ns);
1282 set_cred_user_ns(cred, get_user_ns(user_ns));
1284 return commit_creds(cred);
1287 struct ns_common *ns_get_owner(struct ns_common *ns)
1289 struct user_namespace *my_user_ns = current_user_ns();
1290 struct user_namespace *owner, *p;
1292 /* See if the owner is in the current user namespace */
1293 owner = p = ns->ops->owner(ns);
1296 return ERR_PTR(-EPERM);
1297 if (p == my_user_ns)
1302 return &get_user_ns(owner)->ns;
1305 static struct user_namespace *userns_owner(struct ns_common *ns)
1307 return to_user_ns(ns)->parent;
1310 const struct proc_ns_operations userns_operations = {
1312 .type = CLONE_NEWUSER,
1315 .install = userns_install,
1316 .owner = userns_owner,
1317 .get_parent = ns_get_owner,
1320 static __init int user_namespaces_init(void)
1322 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1325 subsys_initcall(user_namespaces_init);