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;
28 static bool new_idmap_permitted(const struct file *file,
29 struct user_namespace *ns, int cap_setid,
30 struct uid_gid_map *map);
32 static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
34 /* Start with the same capabilities as init but useless for doing
35 * anything as the capabilities are bound to the new user namespace.
37 cred->securebits = SECUREBITS_DEFAULT;
38 cred->cap_inheritable = CAP_EMPTY_SET;
39 cred->cap_permitted = CAP_FULL_SET;
40 cred->cap_effective = CAP_FULL_SET;
41 cred->cap_bset = CAP_FULL_SET;
43 key_put(cred->request_key_auth);
44 cred->request_key_auth = NULL;
46 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
47 cred->user_ns = user_ns;
51 * Create a new user namespace, deriving the creator from the user in the
52 * passed credentials, and replacing that user with the new root user for the
55 * This is called by copy_creds(), which will finish setting the target task's
58 int create_user_ns(struct cred *new)
60 struct user_namespace *ns, *parent_ns = new->user_ns;
61 kuid_t owner = new->euid;
62 kgid_t group = new->egid;
65 if (parent_ns->level > 32)
69 * Verify that we can not violate the policy of which files
70 * may be accessed that is specified by the root directory,
71 * by verifing that the root directory is at the root of the
72 * mount namespace which allows all files to be accessed.
74 if (current_chrooted())
77 /* The creator needs a mapping in the parent user namespace
78 * or else we won't be able to reasonably tell userspace who
79 * created a user_namespace.
81 if (!kuid_has_mapping(parent_ns, owner) ||
82 !kgid_has_mapping(parent_ns, group))
85 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
89 ret = proc_alloc_inum(&ns->proc_inum);
91 kmem_cache_free(user_ns_cachep, ns);
95 atomic_set(&ns->count, 1);
96 /* Leave the new->user_ns reference with the new user namespace. */
97 ns->parent = parent_ns;
98 ns->level = parent_ns->level + 1;
102 set_cred_user_ns(new, ns);
104 #ifdef CONFIG_PERSISTENT_KEYRINGS
105 init_rwsem(&ns->persistent_keyring_register_sem);
110 int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
115 if (!(unshare_flags & CLONE_NEWUSER))
118 cred = prepare_creds();
120 err = create_user_ns(cred);
130 void free_user_ns(struct user_namespace *ns)
132 struct user_namespace *parent;
136 #ifdef CONFIG_PERSISTENT_KEYRINGS
137 key_put(ns->persistent_keyring_register);
139 proc_free_inum(ns->proc_inum);
140 kmem_cache_free(user_ns_cachep, ns);
142 } while (atomic_dec_and_test(&parent->count));
144 EXPORT_SYMBOL(free_user_ns);
146 static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
148 unsigned idx, extents;
149 u32 first, last, id2;
151 id2 = id + count - 1;
153 /* Find the matching extent */
154 extents = map->nr_extents;
156 for (idx = 0; idx < extents; idx++) {
157 first = map->extent[idx].first;
158 last = first + map->extent[idx].count - 1;
159 if (id >= first && id <= last &&
160 (id2 >= first && id2 <= last))
163 /* Map the id or note failure */
165 id = (id - first) + map->extent[idx].lower_first;
172 static u32 map_id_down(struct uid_gid_map *map, u32 id)
174 unsigned idx, extents;
177 /* Find the matching extent */
178 extents = map->nr_extents;
180 for (idx = 0; idx < extents; idx++) {
181 first = map->extent[idx].first;
182 last = first + map->extent[idx].count - 1;
183 if (id >= first && id <= last)
186 /* Map the id or note failure */
188 id = (id - first) + map->extent[idx].lower_first;
195 static u32 map_id_up(struct uid_gid_map *map, u32 id)
197 unsigned idx, extents;
200 /* Find the matching extent */
201 extents = map->nr_extents;
203 for (idx = 0; idx < extents; idx++) {
204 first = map->extent[idx].lower_first;
205 last = first + map->extent[idx].count - 1;
206 if (id >= first && id <= last)
209 /* Map the id or note failure */
211 id = (id - first) + map->extent[idx].first;
219 * make_kuid - Map a user-namespace uid pair into a kuid.
220 * @ns: User namespace that the uid is in
221 * @uid: User identifier
223 * Maps a user-namespace uid pair into a kernel internal kuid,
224 * and returns that kuid.
226 * When there is no mapping defined for the user-namespace uid
227 * pair INVALID_UID is returned. Callers are expected to test
228 * for and handle INVALID_UID being returned. INVALID_UID
229 * may be tested for using uid_valid().
231 kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
233 /* Map the uid to a global kernel uid */
234 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
236 EXPORT_SYMBOL(make_kuid);
239 * from_kuid - Create a uid from a kuid user-namespace pair.
240 * @targ: The user namespace we want a uid in.
241 * @kuid: The kernel internal uid to start with.
243 * Map @kuid into the user-namespace specified by @targ and
244 * return the resulting uid.
246 * There is always a mapping into the initial user_namespace.
248 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
250 uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
252 /* Map the uid from a global kernel uid */
253 return map_id_up(&targ->uid_map, __kuid_val(kuid));
255 EXPORT_SYMBOL(from_kuid);
258 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
259 * @targ: The user namespace we want a uid in.
260 * @kuid: The kernel internal uid to start with.
262 * Map @kuid into the user-namespace specified by @targ and
263 * return the resulting uid.
265 * There is always a mapping into the initial user_namespace.
267 * Unlike from_kuid from_kuid_munged never fails and always
268 * returns a valid uid. This makes from_kuid_munged appropriate
269 * for use in syscalls like stat and getuid where failing the
270 * system call and failing to provide a valid uid are not an
273 * If @kuid has no mapping in @targ overflowuid is returned.
275 uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
278 uid = from_kuid(targ, kuid);
280 if (uid == (uid_t) -1)
284 EXPORT_SYMBOL(from_kuid_munged);
287 * make_kgid - Map a user-namespace gid pair into a kgid.
288 * @ns: User namespace that the gid is in
289 * @gid: group identifier
291 * Maps a user-namespace gid pair into a kernel internal kgid,
292 * and returns that kgid.
294 * When there is no mapping defined for the user-namespace gid
295 * pair INVALID_GID is returned. Callers are expected to test
296 * for and handle INVALID_GID being returned. INVALID_GID may be
297 * tested for using gid_valid().
299 kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
301 /* Map the gid to a global kernel gid */
302 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
304 EXPORT_SYMBOL(make_kgid);
307 * from_kgid - Create a gid from a kgid user-namespace pair.
308 * @targ: The user namespace we want a gid in.
309 * @kgid: The kernel internal gid to start with.
311 * Map @kgid into the user-namespace specified by @targ and
312 * return the resulting gid.
314 * There is always a mapping into the initial user_namespace.
316 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
318 gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
320 /* Map the gid from a global kernel gid */
321 return map_id_up(&targ->gid_map, __kgid_val(kgid));
323 EXPORT_SYMBOL(from_kgid);
326 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
327 * @targ: The user namespace we want a gid in.
328 * @kgid: The kernel internal gid to start with.
330 * Map @kgid into the user-namespace specified by @targ and
331 * return the resulting gid.
333 * There is always a mapping into the initial user_namespace.
335 * Unlike from_kgid from_kgid_munged never fails and always
336 * returns a valid gid. This makes from_kgid_munged appropriate
337 * for use in syscalls like stat and getgid where failing the
338 * system call and failing to provide a valid gid are not options.
340 * If @kgid has no mapping in @targ overflowgid is returned.
342 gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
345 gid = from_kgid(targ, kgid);
347 if (gid == (gid_t) -1)
351 EXPORT_SYMBOL(from_kgid_munged);
354 * make_kprojid - Map a user-namespace projid pair into a kprojid.
355 * @ns: User namespace that the projid is in
356 * @projid: Project identifier
358 * Maps a user-namespace uid pair into a kernel internal kuid,
359 * and returns that kuid.
361 * When there is no mapping defined for the user-namespace projid
362 * pair INVALID_PROJID is returned. Callers are expected to test
363 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
364 * may be tested for using projid_valid().
366 kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
368 /* Map the uid to a global kernel uid */
369 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
371 EXPORT_SYMBOL(make_kprojid);
374 * from_kprojid - Create a projid from a kprojid user-namespace pair.
375 * @targ: The user namespace we want a projid in.
376 * @kprojid: The kernel internal project identifier to start with.
378 * Map @kprojid into the user-namespace specified by @targ and
379 * return the resulting projid.
381 * There is always a mapping into the initial user_namespace.
383 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
385 projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
387 /* Map the uid from a global kernel uid */
388 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
390 EXPORT_SYMBOL(from_kprojid);
393 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
394 * @targ: The user namespace we want a projid in.
395 * @kprojid: The kernel internal projid to start with.
397 * Map @kprojid into the user-namespace specified by @targ and
398 * return the resulting projid.
400 * There is always a mapping into the initial user_namespace.
402 * Unlike from_kprojid from_kprojid_munged never fails and always
403 * returns a valid projid. This makes from_kprojid_munged
404 * appropriate for use in syscalls like stat and where
405 * failing the system call and failing to provide a valid projid are
408 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
410 projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
413 projid = from_kprojid(targ, kprojid);
415 if (projid == (projid_t) -1)
416 projid = OVERFLOW_PROJID;
419 EXPORT_SYMBOL(from_kprojid_munged);
422 static int uid_m_show(struct seq_file *seq, void *v)
424 struct user_namespace *ns = seq->private;
425 struct uid_gid_extent *extent = v;
426 struct user_namespace *lower_ns;
429 lower_ns = seq_user_ns(seq);
430 if ((lower_ns == ns) && lower_ns->parent)
431 lower_ns = lower_ns->parent;
433 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
435 seq_printf(seq, "%10u %10u %10u\n",
443 static int gid_m_show(struct seq_file *seq, void *v)
445 struct user_namespace *ns = seq->private;
446 struct uid_gid_extent *extent = v;
447 struct user_namespace *lower_ns;
450 lower_ns = seq_user_ns(seq);
451 if ((lower_ns == ns) && lower_ns->parent)
452 lower_ns = lower_ns->parent;
454 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
456 seq_printf(seq, "%10u %10u %10u\n",
464 static int projid_m_show(struct seq_file *seq, void *v)
466 struct user_namespace *ns = seq->private;
467 struct uid_gid_extent *extent = v;
468 struct user_namespace *lower_ns;
471 lower_ns = seq_user_ns(seq);
472 if ((lower_ns == ns) && lower_ns->parent)
473 lower_ns = lower_ns->parent;
475 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
477 seq_printf(seq, "%10u %10u %10u\n",
485 static void *m_start(struct seq_file *seq, loff_t *ppos,
486 struct uid_gid_map *map)
488 struct uid_gid_extent *extent = NULL;
491 if (pos < map->nr_extents)
492 extent = &map->extent[pos];
497 static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
499 struct user_namespace *ns = seq->private;
501 return m_start(seq, ppos, &ns->uid_map);
504 static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
506 struct user_namespace *ns = seq->private;
508 return m_start(seq, ppos, &ns->gid_map);
511 static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
513 struct user_namespace *ns = seq->private;
515 return m_start(seq, ppos, &ns->projid_map);
518 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
521 return seq->op->start(seq, pos);
524 static void m_stop(struct seq_file *seq, void *v)
529 const struct seq_operations proc_uid_seq_operations = {
530 .start = uid_m_start,
536 const struct seq_operations proc_gid_seq_operations = {
537 .start = gid_m_start,
543 const struct seq_operations proc_projid_seq_operations = {
544 .start = projid_m_start,
547 .show = projid_m_show,
550 static bool mappings_overlap(struct uid_gid_map *new_map,
551 struct uid_gid_extent *extent)
553 u32 upper_first, lower_first, upper_last, lower_last;
556 upper_first = extent->first;
557 lower_first = extent->lower_first;
558 upper_last = upper_first + extent->count - 1;
559 lower_last = lower_first + extent->count - 1;
561 for (idx = 0; idx < new_map->nr_extents; idx++) {
562 u32 prev_upper_first, prev_lower_first;
563 u32 prev_upper_last, prev_lower_last;
564 struct uid_gid_extent *prev;
566 prev = &new_map->extent[idx];
568 prev_upper_first = prev->first;
569 prev_lower_first = prev->lower_first;
570 prev_upper_last = prev_upper_first + prev->count - 1;
571 prev_lower_last = prev_lower_first + prev->count - 1;
573 /* Does the upper range intersect a previous extent? */
574 if ((prev_upper_first <= upper_last) &&
575 (prev_upper_last >= upper_first))
578 /* Does the lower range intersect a previous extent? */
579 if ((prev_lower_first <= lower_last) &&
580 (prev_lower_last >= lower_first))
587 static DEFINE_MUTEX(id_map_mutex);
589 static ssize_t map_write(struct file *file, const char __user *buf,
590 size_t count, loff_t *ppos,
592 struct uid_gid_map *map,
593 struct uid_gid_map *parent_map)
595 struct seq_file *seq = file->private_data;
596 struct user_namespace *ns = seq->private;
597 struct uid_gid_map new_map;
599 struct uid_gid_extent *extent = NULL;
600 unsigned long page = 0;
601 char *kbuf, *pos, *next_line;
602 ssize_t ret = -EINVAL;
605 * The id_map_mutex serializes all writes to any given map.
607 * Any map is only ever written once.
609 * An id map fits within 1 cache line on most architectures.
611 * On read nothing needs to be done unless you are on an
612 * architecture with a crazy cache coherency model like alpha.
614 * There is a one time data dependency between reading the
615 * count of the extents and the values of the extents. The
616 * desired behavior is to see the values of the extents that
617 * were written before the count of the extents.
619 * To achieve this smp_wmb() is used on guarantee the write
620 * order and smp_rmb() is guaranteed that we don't have crazy
621 * architectures returning stale data.
623 mutex_lock(&id_map_mutex);
626 /* Only allow one successful write to the map */
627 if (map->nr_extents != 0)
631 * Adjusting namespace settings requires capabilities on the target.
633 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
638 page = __get_free_page(GFP_TEMPORARY);
639 kbuf = (char *) page;
643 /* Only allow <= page size writes at the beginning of the file */
645 if ((*ppos != 0) || (count >= PAGE_SIZE))
648 /* Slurp in the user data */
650 if (copy_from_user(kbuf, buf, count))
654 /* Parse the user data */
657 new_map.nr_extents = 0;
658 for (; pos; pos = next_line) {
659 extent = &new_map.extent[new_map.nr_extents];
661 /* Find the end of line and ensure I don't look past it */
662 next_line = strchr(pos, '\n');
666 if (*next_line == '\0')
670 pos = skip_spaces(pos);
671 extent->first = simple_strtoul(pos, &pos, 10);
675 pos = skip_spaces(pos);
676 extent->lower_first = simple_strtoul(pos, &pos, 10);
680 pos = skip_spaces(pos);
681 extent->count = simple_strtoul(pos, &pos, 10);
682 if (*pos && !isspace(*pos))
685 /* Verify there is not trailing junk on the line */
686 pos = skip_spaces(pos);
690 /* Verify we have been given valid starting values */
691 if ((extent->first == (u32) -1) ||
692 (extent->lower_first == (u32) -1))
695 /* Verify count is not zero and does not cause the
698 if ((extent->first + extent->count) <= extent->first)
700 if ((extent->lower_first + extent->count) <=
704 /* Do the ranges in extent overlap any previous extents? */
705 if (mappings_overlap(&new_map, extent))
708 new_map.nr_extents++;
710 /* Fail if the file contains too many extents */
711 if ((new_map.nr_extents == UID_GID_MAP_MAX_EXTENTS) &&
715 /* Be very certaint the new map actually exists */
716 if (new_map.nr_extents == 0)
720 /* Validate the user is allowed to use user id's mapped to. */
721 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
724 /* Map the lower ids from the parent user namespace to the
725 * kernel global id space.
727 for (idx = 0; idx < new_map.nr_extents; idx++) {
729 extent = &new_map.extent[idx];
731 lower_first = map_id_range_down(parent_map,
735 /* Fail if we can not map the specified extent to
736 * the kernel global id space.
738 if (lower_first == (u32) -1)
741 extent->lower_first = lower_first;
744 /* Install the map */
745 memcpy(map->extent, new_map.extent,
746 new_map.nr_extents*sizeof(new_map.extent[0]));
748 map->nr_extents = new_map.nr_extents;
753 mutex_unlock(&id_map_mutex);
759 ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
760 size_t size, loff_t *ppos)
762 struct seq_file *seq = file->private_data;
763 struct user_namespace *ns = seq->private;
764 struct user_namespace *seq_ns = seq_user_ns(seq);
769 if ((seq_ns != ns) && (seq_ns != ns->parent))
772 return map_write(file, buf, size, ppos, CAP_SETUID,
773 &ns->uid_map, &ns->parent->uid_map);
776 ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
777 size_t size, loff_t *ppos)
779 struct seq_file *seq = file->private_data;
780 struct user_namespace *ns = seq->private;
781 struct user_namespace *seq_ns = seq_user_ns(seq);
786 if ((seq_ns != ns) && (seq_ns != ns->parent))
789 return map_write(file, buf, size, ppos, CAP_SETGID,
790 &ns->gid_map, &ns->parent->gid_map);
793 ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
794 size_t size, loff_t *ppos)
796 struct seq_file *seq = file->private_data;
797 struct user_namespace *ns = seq->private;
798 struct user_namespace *seq_ns = seq_user_ns(seq);
803 if ((seq_ns != ns) && (seq_ns != ns->parent))
806 /* Anyone can set any valid project id no capability needed */
807 return map_write(file, buf, size, ppos, -1,
808 &ns->projid_map, &ns->parent->projid_map);
811 static bool new_idmap_permitted(const struct file *file,
812 struct user_namespace *ns, int cap_setid,
813 struct uid_gid_map *new_map)
815 /* Don't allow mappings that would allow anything that wouldn't
816 * be allowed without the establishment of unprivileged mappings.
818 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1)) {
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, file->f_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(&id_map_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(&id_map_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);