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/highuid.h>
13 #include <linux/cred.h>
14 #include <linux/securebits.h>
15 #include <linux/keyctl.h>
16 #include <linux/key-type.h>
17 #include <keys/user-type.h>
18 #include <linux/seq_file.h>
20 #include <linux/uaccess.h>
21 #include <linux/ctype.h>
23 static struct kmem_cache *user_ns_cachep __read_mostly;
25 static bool new_idmap_permitted(struct user_namespace *ns, int cap_setid,
26 struct uid_gid_map *map);
29 * Create a new user namespace, deriving the creator from the user in the
30 * passed credentials, and replacing that user with the new root user for the
33 * This is called by copy_creds(), which will finish setting the target task's
36 int create_user_ns(struct cred *new)
38 struct user_namespace *ns, *parent_ns = new->user_ns;
39 kuid_t owner = new->euid;
40 kgid_t group = new->egid;
42 /* The creator needs a mapping in the parent user namespace
43 * or else we won't be able to reasonably tell userspace who
44 * created a user_namespace.
46 if (!kuid_has_mapping(parent_ns, owner) ||
47 !kgid_has_mapping(parent_ns, group))
50 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
55 ns->parent = parent_ns;
59 /* Start with the same capabilities as init but useless for doing
60 * anything as the capabilities are bound to the new user namespace.
62 new->securebits = SECUREBITS_DEFAULT;
63 new->cap_inheritable = CAP_EMPTY_SET;
64 new->cap_permitted = CAP_FULL_SET;
65 new->cap_effective = CAP_FULL_SET;
66 new->cap_bset = CAP_FULL_SET;
68 key_put(new->request_key_auth);
69 new->request_key_auth = NULL;
71 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
73 /* Leave the new->user_ns reference with the new user namespace. */
74 /* Leave the reference to our user_ns with the new cred. */
80 void free_user_ns(struct kref *kref)
82 struct user_namespace *parent, *ns =
83 container_of(kref, struct user_namespace, kref);
86 kmem_cache_free(user_ns_cachep, ns);
89 EXPORT_SYMBOL(free_user_ns);
91 static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
93 unsigned idx, extents;
98 /* Find the matching extent */
99 extents = map->nr_extents;
100 smp_read_barrier_depends();
101 for (idx = 0; idx < extents; idx++) {
102 first = map->extent[idx].first;
103 last = first + map->extent[idx].count - 1;
104 if (id >= first && id <= last &&
105 (id2 >= first && id2 <= last))
108 /* Map the id or note failure */
110 id = (id - first) + map->extent[idx].lower_first;
117 static u32 map_id_down(struct uid_gid_map *map, u32 id)
119 unsigned idx, extents;
122 /* Find the matching extent */
123 extents = map->nr_extents;
124 smp_read_barrier_depends();
125 for (idx = 0; idx < extents; idx++) {
126 first = map->extent[idx].first;
127 last = first + map->extent[idx].count - 1;
128 if (id >= first && id <= last)
131 /* Map the id or note failure */
133 id = (id - first) + map->extent[idx].lower_first;
140 static u32 map_id_up(struct uid_gid_map *map, u32 id)
142 unsigned idx, extents;
145 /* Find the matching extent */
146 extents = map->nr_extents;
147 smp_read_barrier_depends();
148 for (idx = 0; idx < extents; idx++) {
149 first = map->extent[idx].lower_first;
150 last = first + map->extent[idx].count - 1;
151 if (id >= first && id <= last)
154 /* Map the id or note failure */
156 id = (id - first) + map->extent[idx].first;
164 * make_kuid - Map a user-namespace uid pair into a kuid.
165 * @ns: User namespace that the uid is in
166 * @uid: User identifier
168 * Maps a user-namespace uid pair into a kernel internal kuid,
169 * and returns that kuid.
171 * When there is no mapping defined for the user-namespace uid
172 * pair INVALID_UID is returned. Callers are expected to test
173 * for and handle handle INVALID_UID being returned. INVALID_UID
174 * may be tested for using uid_valid().
176 kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
178 /* Map the uid to a global kernel uid */
179 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
181 EXPORT_SYMBOL(make_kuid);
184 * from_kuid - Create a uid from a kuid user-namespace pair.
185 * @targ: The user namespace we want a uid in.
186 * @kuid: The kernel internal uid to start with.
188 * Map @kuid into the user-namespace specified by @targ and
189 * return the resulting uid.
191 * There is always a mapping into the initial user_namespace.
193 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
195 uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
197 /* Map the uid from a global kernel uid */
198 return map_id_up(&targ->uid_map, __kuid_val(kuid));
200 EXPORT_SYMBOL(from_kuid);
203 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
204 * @targ: The user namespace we want a uid in.
205 * @kuid: The kernel internal uid to start with.
207 * Map @kuid into the user-namespace specified by @targ and
208 * return the resulting uid.
210 * There is always a mapping into the initial user_namespace.
212 * Unlike from_kuid from_kuid_munged never fails and always
213 * returns a valid uid. This makes from_kuid_munged appropriate
214 * for use in syscalls like stat and getuid where failing the
215 * system call and failing to provide a valid uid are not an
218 * If @kuid has no mapping in @targ overflowuid is returned.
220 uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
223 uid = from_kuid(targ, kuid);
225 if (uid == (uid_t) -1)
229 EXPORT_SYMBOL(from_kuid_munged);
232 * make_kgid - Map a user-namespace gid pair into a kgid.
233 * @ns: User namespace that the gid is in
234 * @uid: group identifier
236 * Maps a user-namespace gid pair into a kernel internal kgid,
237 * and returns that kgid.
239 * When there is no mapping defined for the user-namespace gid
240 * pair INVALID_GID is returned. Callers are expected to test
241 * for and handle INVALID_GID being returned. INVALID_GID may be
242 * tested for using gid_valid().
244 kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
246 /* Map the gid to a global kernel gid */
247 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
249 EXPORT_SYMBOL(make_kgid);
252 * from_kgid - Create a gid from a kgid user-namespace pair.
253 * @targ: The user namespace we want a gid in.
254 * @kgid: The kernel internal gid to start with.
256 * Map @kgid into the user-namespace specified by @targ and
257 * return the resulting gid.
259 * There is always a mapping into the initial user_namespace.
261 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
263 gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
265 /* Map the gid from a global kernel gid */
266 return map_id_up(&targ->gid_map, __kgid_val(kgid));
268 EXPORT_SYMBOL(from_kgid);
271 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
272 * @targ: The user namespace we want a gid in.
273 * @kgid: The kernel internal gid to start with.
275 * Map @kgid into the user-namespace specified by @targ and
276 * return the resulting gid.
278 * There is always a mapping into the initial user_namespace.
280 * Unlike from_kgid from_kgid_munged never fails and always
281 * returns a valid gid. This makes from_kgid_munged appropriate
282 * for use in syscalls like stat and getgid where failing the
283 * system call and failing to provide a valid gid are not options.
285 * If @kgid has no mapping in @targ overflowgid is returned.
287 gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
290 gid = from_kgid(targ, kgid);
292 if (gid == (gid_t) -1)
296 EXPORT_SYMBOL(from_kgid_munged);
298 static int uid_m_show(struct seq_file *seq, void *v)
300 struct user_namespace *ns = seq->private;
301 struct uid_gid_extent *extent = v;
302 struct user_namespace *lower_ns;
305 lower_ns = current_user_ns();
306 if ((lower_ns == ns) && lower_ns->parent)
307 lower_ns = lower_ns->parent;
309 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
311 seq_printf(seq, "%10u %10u %10u\n",
319 static int gid_m_show(struct seq_file *seq, void *v)
321 struct user_namespace *ns = seq->private;
322 struct uid_gid_extent *extent = v;
323 struct user_namespace *lower_ns;
326 lower_ns = current_user_ns();
327 if ((lower_ns == ns) && lower_ns->parent)
328 lower_ns = lower_ns->parent;
330 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
332 seq_printf(seq, "%10u %10u %10u\n",
340 static void *m_start(struct seq_file *seq, loff_t *ppos, struct uid_gid_map *map)
342 struct uid_gid_extent *extent = NULL;
345 if (pos < map->nr_extents)
346 extent = &map->extent[pos];
351 static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
353 struct user_namespace *ns = seq->private;
355 return m_start(seq, ppos, &ns->uid_map);
358 static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
360 struct user_namespace *ns = seq->private;
362 return m_start(seq, ppos, &ns->gid_map);
365 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
368 return seq->op->start(seq, pos);
371 static void m_stop(struct seq_file *seq, void *v)
376 struct seq_operations proc_uid_seq_operations = {
377 .start = uid_m_start,
383 struct seq_operations proc_gid_seq_operations = {
384 .start = gid_m_start,
390 static DEFINE_MUTEX(id_map_mutex);
392 static ssize_t map_write(struct file *file, const char __user *buf,
393 size_t count, loff_t *ppos,
395 struct uid_gid_map *map,
396 struct uid_gid_map *parent_map)
398 struct seq_file *seq = file->private_data;
399 struct user_namespace *ns = seq->private;
400 struct uid_gid_map new_map;
402 struct uid_gid_extent *extent, *last = NULL;
403 unsigned long page = 0;
404 char *kbuf, *pos, *next_line;
405 ssize_t ret = -EINVAL;
408 * The id_map_mutex serializes all writes to any given map.
410 * Any map is only ever written once.
412 * An id map fits within 1 cache line on most architectures.
414 * On read nothing needs to be done unless you are on an
415 * architecture with a crazy cache coherency model like alpha.
417 * There is a one time data dependency between reading the
418 * count of the extents and the values of the extents. The
419 * desired behavior is to see the values of the extents that
420 * were written before the count of the extents.
422 * To achieve this smp_wmb() is used on guarantee the write
423 * order and smp_read_barrier_depends() is guaranteed that we
424 * don't have crazy architectures returning stale data.
427 mutex_lock(&id_map_mutex);
430 /* Only allow one successful write to the map */
431 if (map->nr_extents != 0)
434 /* Require the appropriate privilege CAP_SETUID or CAP_SETGID
435 * over the user namespace in order to set the id mapping.
437 if (!ns_capable(ns, cap_setid))
442 page = __get_free_page(GFP_TEMPORARY);
443 kbuf = (char *) page;
447 /* Only allow <= page size writes at the beginning of the file */
449 if ((*ppos != 0) || (count >= PAGE_SIZE))
452 /* Slurp in the user data */
454 if (copy_from_user(kbuf, buf, count))
458 /* Parse the user data */
461 new_map.nr_extents = 0;
462 for (;pos; pos = next_line) {
463 extent = &new_map.extent[new_map.nr_extents];
465 /* Find the end of line and ensure I don't look past it */
466 next_line = strchr(pos, '\n');
470 if (*next_line == '\0')
474 pos = skip_spaces(pos);
475 extent->first = simple_strtoul(pos, &pos, 10);
479 pos = skip_spaces(pos);
480 extent->lower_first = simple_strtoul(pos, &pos, 10);
484 pos = skip_spaces(pos);
485 extent->count = simple_strtoul(pos, &pos, 10);
486 if (*pos && !isspace(*pos))
489 /* Verify there is not trailing junk on the line */
490 pos = skip_spaces(pos);
494 /* Verify we have been given valid starting values */
495 if ((extent->first == (u32) -1) ||
496 (extent->lower_first == (u32) -1 ))
499 /* Verify count is not zero and does not cause the extent to wrap */
500 if ((extent->first + extent->count) <= extent->first)
502 if ((extent->lower_first + extent->count) <= extent->lower_first)
505 /* For now only accept extents that are strictly in order */
507 (((last->first + last->count) > extent->first) ||
508 ((last->lower_first + last->count) > extent->lower_first)))
511 new_map.nr_extents++;
514 /* Fail if the file contains too many extents */
515 if ((new_map.nr_extents == UID_GID_MAP_MAX_EXTENTS) &&
519 /* Be very certaint the new map actually exists */
520 if (new_map.nr_extents == 0)
524 /* Validate the user is allowed to use user id's mapped to. */
525 if (!new_idmap_permitted(ns, cap_setid, &new_map))
528 /* Map the lower ids from the parent user namespace to the
529 * kernel global id space.
531 for (idx = 0; idx < new_map.nr_extents; idx++) {
533 extent = &new_map.extent[idx];
535 lower_first = map_id_range_down(parent_map,
539 /* Fail if we can not map the specified extent to
540 * the kernel global id space.
542 if (lower_first == (u32) -1)
545 extent->lower_first = lower_first;
548 /* Install the map */
549 memcpy(map->extent, new_map.extent,
550 new_map.nr_extents*sizeof(new_map.extent[0]));
552 map->nr_extents = new_map.nr_extents;
557 mutex_unlock(&id_map_mutex);
563 ssize_t proc_uid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos)
565 struct seq_file *seq = file->private_data;
566 struct user_namespace *ns = seq->private;
571 return map_write(file, buf, size, ppos, CAP_SETUID,
572 &ns->uid_map, &ns->parent->uid_map);
575 ssize_t proc_gid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos)
577 struct seq_file *seq = file->private_data;
578 struct user_namespace *ns = seq->private;
583 return map_write(file, buf, size, ppos, CAP_SETGID,
584 &ns->gid_map, &ns->parent->gid_map);
587 static bool new_idmap_permitted(struct user_namespace *ns, int cap_setid,
588 struct uid_gid_map *new_map)
590 /* Allow the specified ids if we have the appropriate capability
591 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
593 if (ns_capable(ns->parent, cap_setid))
599 static __init int user_namespaces_init(void)
601 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
604 module_init(user_namespaces_init);