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/user_namespace.h>
12 #include <linux/proc_ns.h>
13 #include <linux/highuid.h>
14 #include <linux/cred.h>
15 #include <linux/securebits.h>
16 #include <linux/keyctl.h>
17 #include <linux/key-type.h>
18 #include <keys/user-type.h>
19 #include <linux/seq_file.h>
21 #include <linux/uaccess.h>
22 #include <linux/ctype.h>
23 #include <linux/projid.h>
24 #include <linux/fs_struct.h>
26 static struct kmem_cache *user_ns_cachep __read_mostly;
27 static DEFINE_MUTEX(userns_state_mutex);
29 static bool new_idmap_permitted(const struct file *file,
30 struct user_namespace *ns, int cap_setid,
31 struct uid_gid_map *map);
33 static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
35 /* Start with the same capabilities as init but useless for doing
36 * anything as the capabilities are bound to the new user namespace.
38 cred->securebits = SECUREBITS_DEFAULT;
39 cred->cap_inheritable = CAP_EMPTY_SET;
40 cred->cap_permitted = CAP_FULL_SET;
41 cred->cap_effective = CAP_FULL_SET;
42 cred->cap_bset = CAP_FULL_SET;
44 key_put(cred->request_key_auth);
45 cred->request_key_auth = NULL;
47 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
48 cred->user_ns = user_ns;
52 * Create a new user namespace, deriving the creator from the user in the
53 * passed credentials, and replacing that user with the new root user for the
56 * This is called by copy_creds(), which will finish setting the target task's
59 int create_user_ns(struct cred *new)
61 struct user_namespace *ns, *parent_ns = new->user_ns;
62 kuid_t owner = new->euid;
63 kgid_t group = new->egid;
66 if (parent_ns->level > 32)
70 * Verify that we can not violate the policy of which files
71 * may be accessed that is specified by the root directory,
72 * by verifing that the root directory is at the root of the
73 * mount namespace which allows all files to be accessed.
75 if (current_chrooted())
78 /* The creator needs a mapping in the parent user namespace
79 * or else we won't be able to reasonably tell userspace who
80 * created a user_namespace.
82 if (!kuid_has_mapping(parent_ns, owner) ||
83 !kgid_has_mapping(parent_ns, group))
86 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
90 ret = proc_alloc_inum(&ns->proc_inum);
92 kmem_cache_free(user_ns_cachep, ns);
96 atomic_set(&ns->count, 1);
97 /* Leave the new->user_ns reference with the new user namespace. */
98 ns->parent = parent_ns;
99 ns->level = parent_ns->level + 1;
103 set_cred_user_ns(new, ns);
105 #ifdef CONFIG_PERSISTENT_KEYRINGS
106 init_rwsem(&ns->persistent_keyring_register_sem);
111 int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
116 if (!(unshare_flags & CLONE_NEWUSER))
119 cred = prepare_creds();
121 err = create_user_ns(cred);
131 void free_user_ns(struct user_namespace *ns)
133 struct user_namespace *parent;
137 #ifdef CONFIG_PERSISTENT_KEYRINGS
138 key_put(ns->persistent_keyring_register);
140 proc_free_inum(ns->proc_inum);
141 kmem_cache_free(user_ns_cachep, ns);
143 } while (atomic_dec_and_test(&parent->count));
145 EXPORT_SYMBOL(free_user_ns);
147 static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
149 unsigned idx, extents;
150 u32 first, last, id2;
152 id2 = id + count - 1;
154 /* Find the matching extent */
155 extents = map->nr_extents;
157 for (idx = 0; idx < extents; idx++) {
158 first = map->extent[idx].first;
159 last = first + map->extent[idx].count - 1;
160 if (id >= first && id <= last &&
161 (id2 >= first && id2 <= last))
164 /* Map the id or note failure */
166 id = (id - first) + map->extent[idx].lower_first;
173 static u32 map_id_down(struct uid_gid_map *map, u32 id)
175 unsigned idx, extents;
178 /* Find the matching extent */
179 extents = map->nr_extents;
181 for (idx = 0; idx < extents; idx++) {
182 first = map->extent[idx].first;
183 last = first + map->extent[idx].count - 1;
184 if (id >= first && id <= last)
187 /* Map the id or note failure */
189 id = (id - first) + map->extent[idx].lower_first;
196 static u32 map_id_up(struct uid_gid_map *map, u32 id)
198 unsigned idx, extents;
201 /* Find the matching extent */
202 extents = map->nr_extents;
204 for (idx = 0; idx < extents; idx++) {
205 first = map->extent[idx].lower_first;
206 last = first + map->extent[idx].count - 1;
207 if (id >= first && id <= last)
210 /* Map the id or note failure */
212 id = (id - first) + map->extent[idx].first;
220 * make_kuid - Map a user-namespace uid pair into a kuid.
221 * @ns: User namespace that the uid is in
222 * @uid: User identifier
224 * Maps a user-namespace uid pair into a kernel internal kuid,
225 * and returns that kuid.
227 * When there is no mapping defined for the user-namespace uid
228 * pair INVALID_UID is returned. Callers are expected to test
229 * for and handle INVALID_UID being returned. INVALID_UID
230 * may be tested for using uid_valid().
232 kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
234 /* Map the uid to a global kernel uid */
235 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
237 EXPORT_SYMBOL(make_kuid);
240 * from_kuid - Create a uid from a kuid user-namespace pair.
241 * @targ: The user namespace we want a uid in.
242 * @kuid: The kernel internal uid to start with.
244 * Map @kuid into the user-namespace specified by @targ and
245 * return the resulting uid.
247 * There is always a mapping into the initial user_namespace.
249 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
251 uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
253 /* Map the uid from a global kernel uid */
254 return map_id_up(&targ->uid_map, __kuid_val(kuid));
256 EXPORT_SYMBOL(from_kuid);
259 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
260 * @targ: The user namespace we want a uid in.
261 * @kuid: The kernel internal uid to start with.
263 * Map @kuid into the user-namespace specified by @targ and
264 * return the resulting uid.
266 * There is always a mapping into the initial user_namespace.
268 * Unlike from_kuid from_kuid_munged never fails and always
269 * returns a valid uid. This makes from_kuid_munged appropriate
270 * for use in syscalls like stat and getuid where failing the
271 * system call and failing to provide a valid uid are not an
274 * If @kuid has no mapping in @targ overflowuid is returned.
276 uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
279 uid = from_kuid(targ, kuid);
281 if (uid == (uid_t) -1)
285 EXPORT_SYMBOL(from_kuid_munged);
288 * make_kgid - Map a user-namespace gid pair into a kgid.
289 * @ns: User namespace that the gid is in
290 * @gid: group identifier
292 * Maps a user-namespace gid pair into a kernel internal kgid,
293 * and returns that kgid.
295 * When there is no mapping defined for the user-namespace gid
296 * pair INVALID_GID is returned. Callers are expected to test
297 * for and handle INVALID_GID being returned. INVALID_GID may be
298 * tested for using gid_valid().
300 kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
302 /* Map the gid to a global kernel gid */
303 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
305 EXPORT_SYMBOL(make_kgid);
308 * from_kgid - Create a gid from a kgid user-namespace pair.
309 * @targ: The user namespace we want a gid in.
310 * @kgid: The kernel internal gid to start with.
312 * Map @kgid into the user-namespace specified by @targ and
313 * return the resulting gid.
315 * There is always a mapping into the initial user_namespace.
317 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
319 gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
321 /* Map the gid from a global kernel gid */
322 return map_id_up(&targ->gid_map, __kgid_val(kgid));
324 EXPORT_SYMBOL(from_kgid);
327 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
328 * @targ: The user namespace we want a gid in.
329 * @kgid: The kernel internal gid to start with.
331 * Map @kgid into the user-namespace specified by @targ and
332 * return the resulting gid.
334 * There is always a mapping into the initial user_namespace.
336 * Unlike from_kgid from_kgid_munged never fails and always
337 * returns a valid gid. This makes from_kgid_munged appropriate
338 * for use in syscalls like stat and getgid where failing the
339 * system call and failing to provide a valid gid are not options.
341 * If @kgid has no mapping in @targ overflowgid is returned.
343 gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
346 gid = from_kgid(targ, kgid);
348 if (gid == (gid_t) -1)
352 EXPORT_SYMBOL(from_kgid_munged);
355 * make_kprojid - Map a user-namespace projid pair into a kprojid.
356 * @ns: User namespace that the projid is in
357 * @projid: Project identifier
359 * Maps a user-namespace uid pair into a kernel internal kuid,
360 * and returns that kuid.
362 * When there is no mapping defined for the user-namespace projid
363 * pair INVALID_PROJID is returned. Callers are expected to test
364 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
365 * may be tested for using projid_valid().
367 kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
369 /* Map the uid to a global kernel uid */
370 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
372 EXPORT_SYMBOL(make_kprojid);
375 * from_kprojid - Create a projid from a kprojid user-namespace pair.
376 * @targ: The user namespace we want a projid in.
377 * @kprojid: The kernel internal project identifier to start with.
379 * Map @kprojid into the user-namespace specified by @targ and
380 * return the resulting projid.
382 * There is always a mapping into the initial user_namespace.
384 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
386 projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
388 /* Map the uid from a global kernel uid */
389 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
391 EXPORT_SYMBOL(from_kprojid);
394 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
395 * @targ: The user namespace we want a projid in.
396 * @kprojid: The kernel internal projid to start with.
398 * Map @kprojid into the user-namespace specified by @targ and
399 * return the resulting projid.
401 * There is always a mapping into the initial user_namespace.
403 * Unlike from_kprojid from_kprojid_munged never fails and always
404 * returns a valid projid. This makes from_kprojid_munged
405 * appropriate for use in syscalls like stat and where
406 * failing the system call and failing to provide a valid projid are
409 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
411 projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
414 projid = from_kprojid(targ, kprojid);
416 if (projid == (projid_t) -1)
417 projid = OVERFLOW_PROJID;
420 EXPORT_SYMBOL(from_kprojid_munged);
423 static int uid_m_show(struct seq_file *seq, void *v)
425 struct user_namespace *ns = seq->private;
426 struct uid_gid_extent *extent = v;
427 struct user_namespace *lower_ns;
430 lower_ns = seq_user_ns(seq);
431 if ((lower_ns == ns) && lower_ns->parent)
432 lower_ns = lower_ns->parent;
434 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
436 seq_printf(seq, "%10u %10u %10u\n",
444 static int gid_m_show(struct seq_file *seq, void *v)
446 struct user_namespace *ns = seq->private;
447 struct uid_gid_extent *extent = v;
448 struct user_namespace *lower_ns;
451 lower_ns = seq_user_ns(seq);
452 if ((lower_ns == ns) && lower_ns->parent)
453 lower_ns = lower_ns->parent;
455 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
457 seq_printf(seq, "%10u %10u %10u\n",
465 static int projid_m_show(struct seq_file *seq, void *v)
467 struct user_namespace *ns = seq->private;
468 struct uid_gid_extent *extent = v;
469 struct user_namespace *lower_ns;
472 lower_ns = seq_user_ns(seq);
473 if ((lower_ns == ns) && lower_ns->parent)
474 lower_ns = lower_ns->parent;
476 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
478 seq_printf(seq, "%10u %10u %10u\n",
486 static void *m_start(struct seq_file *seq, loff_t *ppos,
487 struct uid_gid_map *map)
489 struct uid_gid_extent *extent = NULL;
492 if (pos < map->nr_extents)
493 extent = &map->extent[pos];
498 static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
500 struct user_namespace *ns = seq->private;
502 return m_start(seq, ppos, &ns->uid_map);
505 static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
507 struct user_namespace *ns = seq->private;
509 return m_start(seq, ppos, &ns->gid_map);
512 static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
514 struct user_namespace *ns = seq->private;
516 return m_start(seq, ppos, &ns->projid_map);
519 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
522 return seq->op->start(seq, pos);
525 static void m_stop(struct seq_file *seq, void *v)
530 const struct seq_operations proc_uid_seq_operations = {
531 .start = uid_m_start,
537 const struct seq_operations proc_gid_seq_operations = {
538 .start = gid_m_start,
544 const struct seq_operations proc_projid_seq_operations = {
545 .start = projid_m_start,
548 .show = projid_m_show,
551 static bool mappings_overlap(struct uid_gid_map *new_map,
552 struct uid_gid_extent *extent)
554 u32 upper_first, lower_first, upper_last, lower_last;
557 upper_first = extent->first;
558 lower_first = extent->lower_first;
559 upper_last = upper_first + extent->count - 1;
560 lower_last = lower_first + extent->count - 1;
562 for (idx = 0; idx < new_map->nr_extents; idx++) {
563 u32 prev_upper_first, prev_lower_first;
564 u32 prev_upper_last, prev_lower_last;
565 struct uid_gid_extent *prev;
567 prev = &new_map->extent[idx];
569 prev_upper_first = prev->first;
570 prev_lower_first = prev->lower_first;
571 prev_upper_last = prev_upper_first + prev->count - 1;
572 prev_lower_last = prev_lower_first + prev->count - 1;
574 /* Does the upper range intersect a previous extent? */
575 if ((prev_upper_first <= upper_last) &&
576 (prev_upper_last >= upper_first))
579 /* Does the lower range intersect a previous extent? */
580 if ((prev_lower_first <= lower_last) &&
581 (prev_lower_last >= lower_first))
587 static ssize_t map_write(struct file *file, const char __user *buf,
588 size_t count, loff_t *ppos,
590 struct uid_gid_map *map,
591 struct uid_gid_map *parent_map)
593 struct seq_file *seq = file->private_data;
594 struct user_namespace *ns = seq->private;
595 struct uid_gid_map new_map;
597 struct uid_gid_extent *extent = NULL;
598 unsigned long page = 0;
599 char *kbuf, *pos, *next_line;
600 ssize_t ret = -EINVAL;
603 * The userns_state_mutex serializes all writes to any given map.
605 * Any map is only ever written once.
607 * An id map fits within 1 cache line on most architectures.
609 * On read nothing needs to be done unless you are on an
610 * architecture with a crazy cache coherency model like alpha.
612 * There is a one time data dependency between reading the
613 * count of the extents and the values of the extents. The
614 * desired behavior is to see the values of the extents that
615 * were written before the count of the extents.
617 * To achieve this smp_wmb() is used on guarantee the write
618 * order and smp_rmb() is guaranteed that we don't have crazy
619 * architectures returning stale data.
621 mutex_lock(&userns_state_mutex);
624 /* Only allow one successful write to the map */
625 if (map->nr_extents != 0)
629 * Adjusting namespace settings requires capabilities on the target.
631 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
636 page = __get_free_page(GFP_TEMPORARY);
637 kbuf = (char *) page;
641 /* Only allow <= page size writes at the beginning of the file */
643 if ((*ppos != 0) || (count >= PAGE_SIZE))
646 /* Slurp in the user data */
648 if (copy_from_user(kbuf, buf, count))
652 /* Parse the user data */
655 new_map.nr_extents = 0;
656 for (; pos; pos = next_line) {
657 extent = &new_map.extent[new_map.nr_extents];
659 /* Find the end of line and ensure I don't look past it */
660 next_line = strchr(pos, '\n');
664 if (*next_line == '\0')
668 pos = skip_spaces(pos);
669 extent->first = simple_strtoul(pos, &pos, 10);
673 pos = skip_spaces(pos);
674 extent->lower_first = simple_strtoul(pos, &pos, 10);
678 pos = skip_spaces(pos);
679 extent->count = simple_strtoul(pos, &pos, 10);
680 if (*pos && !isspace(*pos))
683 /* Verify there is not trailing junk on the line */
684 pos = skip_spaces(pos);
688 /* Verify we have been given valid starting values */
689 if ((extent->first == (u32) -1) ||
690 (extent->lower_first == (u32) -1))
693 /* Verify count is not zero and does not cause the
696 if ((extent->first + extent->count) <= extent->first)
698 if ((extent->lower_first + extent->count) <=
702 /* Do the ranges in extent overlap any previous extents? */
703 if (mappings_overlap(&new_map, extent))
706 new_map.nr_extents++;
708 /* Fail if the file contains too many extents */
709 if ((new_map.nr_extents == UID_GID_MAP_MAX_EXTENTS) &&
713 /* Be very certaint the new map actually exists */
714 if (new_map.nr_extents == 0)
718 /* Validate the user is allowed to use user id's mapped to. */
719 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
722 /* Map the lower ids from the parent user namespace to the
723 * kernel global id space.
725 for (idx = 0; idx < new_map.nr_extents; idx++) {
727 extent = &new_map.extent[idx];
729 lower_first = map_id_range_down(parent_map,
733 /* Fail if we can not map the specified extent to
734 * the kernel global id space.
736 if (lower_first == (u32) -1)
739 extent->lower_first = lower_first;
742 /* Install the map */
743 memcpy(map->extent, new_map.extent,
744 new_map.nr_extents*sizeof(new_map.extent[0]));
746 map->nr_extents = new_map.nr_extents;
751 mutex_unlock(&userns_state_mutex);
757 ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
758 size_t size, loff_t *ppos)
760 struct seq_file *seq = file->private_data;
761 struct user_namespace *ns = seq->private;
762 struct user_namespace *seq_ns = seq_user_ns(seq);
767 if ((seq_ns != ns) && (seq_ns != ns->parent))
770 return map_write(file, buf, size, ppos, CAP_SETUID,
771 &ns->uid_map, &ns->parent->uid_map);
774 ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
775 size_t size, loff_t *ppos)
777 struct seq_file *seq = file->private_data;
778 struct user_namespace *ns = seq->private;
779 struct user_namespace *seq_ns = seq_user_ns(seq);
784 if ((seq_ns != ns) && (seq_ns != ns->parent))
787 return map_write(file, buf, size, ppos, CAP_SETGID,
788 &ns->gid_map, &ns->parent->gid_map);
791 ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
792 size_t size, loff_t *ppos)
794 struct seq_file *seq = file->private_data;
795 struct user_namespace *ns = seq->private;
796 struct user_namespace *seq_ns = seq_user_ns(seq);
801 if ((seq_ns != ns) && (seq_ns != ns->parent))
804 /* Anyone can set any valid project id no capability needed */
805 return map_write(file, buf, size, ppos, -1,
806 &ns->projid_map, &ns->parent->projid_map);
809 static bool new_idmap_permitted(const struct file *file,
810 struct user_namespace *ns, int cap_setid,
811 struct uid_gid_map *new_map)
813 const struct cred *cred = file->f_cred;
814 /* Don't allow mappings that would allow anything that wouldn't
815 * be allowed without the establishment of unprivileged mappings.
817 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
818 uid_eq(ns->owner, cred->euid)) {
819 u32 id = new_map->extent[0].lower_first;
820 if (cap_setid == CAP_SETUID) {
821 kuid_t uid = make_kuid(ns->parent, id);
822 if (uid_eq(uid, cred->euid))
827 /* Allow anyone to set a mapping that doesn't require privilege */
828 if (!cap_valid(cap_setid))
831 /* Allow the specified ids if we have the appropriate capability
832 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
833 * And the opener of the id file also had the approprpiate capability.
835 if (ns_capable(ns->parent, cap_setid) &&
836 file_ns_capable(file, ns->parent, cap_setid))
842 bool userns_may_setgroups(const struct user_namespace *ns)
846 mutex_lock(&userns_state_mutex);
847 /* It is not safe to use setgroups until a gid mapping in
848 * the user namespace has been established.
850 allowed = ns->gid_map.nr_extents != 0;
851 mutex_unlock(&userns_state_mutex);
856 static void *userns_get(struct task_struct *task)
858 struct user_namespace *user_ns;
861 user_ns = get_user_ns(__task_cred(task)->user_ns);
867 static void userns_put(void *ns)
872 static int userns_install(struct nsproxy *nsproxy, void *ns)
874 struct user_namespace *user_ns = ns;
877 /* Don't allow gaining capabilities by reentering
878 * the same user namespace.
880 if (user_ns == current_user_ns())
883 /* Threaded processes may not enter a different user namespace */
884 if (atomic_read(¤t->mm->mm_users) > 1)
887 if (current->fs->users != 1)
890 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
893 cred = prepare_creds();
897 put_user_ns(cred->user_ns);
898 set_cred_user_ns(cred, get_user_ns(user_ns));
900 return commit_creds(cred);
903 static unsigned int userns_inum(void *ns)
905 struct user_namespace *user_ns = ns;
906 return user_ns->proc_inum;
909 const struct proc_ns_operations userns_operations = {
911 .type = CLONE_NEWUSER,
914 .install = userns_install,
918 static __init int user_namespaces_init(void)
920 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
923 subsys_initcall(user_namespaces_init);