blob: 246d4d4ce5c70cde4b5c9ed7fd33f54c1efe5a3e [file] [log] [blame]
Cedric Le Goateracce2922007-07-15 23:40:59 -07001/*
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
5 * License.
6 */
7
Paul Gortmaker9984de12011-05-23 14:51:41 -04008#include <linux/export.h>
Cedric Le Goateracce2922007-07-15 23:40:59 -07009#include <linux/nsproxy.h>
Robert P. J. Day1aeb2722008-04-29 00:59:25 -070010#include <linux/slab.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010011#include <linux/sched/signal.h>
Cedric Le Goateracce2922007-07-15 23:40:59 -070012#include <linux/user_namespace.h>
David Howells0bb80f22013-04-12 01:50:06 +010013#include <linux/proc_ns.h>
Eric W. Biederman5c1469d2010-06-13 03:28:03 +000014#include <linux/highuid.h>
Serge Hallyn18b6e042008-10-15 16:38:45 -050015#include <linux/cred.h>
Eric W. Biederman973c5912011-11-17 01:59:07 -080016#include <linux/securebits.h>
Eric W. Biederman22d917d2011-11-17 00:11:58 -080017#include <linux/keyctl.h>
18#include <linux/key-type.h>
19#include <keys/user-type.h>
20#include <linux/seq_file.h>
21#include <linux/fs.h>
22#include <linux/uaccess.h>
23#include <linux/ctype.h>
Eric W. Biedermanf76d2072012-08-30 01:24:05 -070024#include <linux/projid.h>
Eric W. Biedermane66eded2013-03-13 11:51:49 -070025#include <linux/fs_struct.h>
Christian Brauner6397fac2017-10-25 00:04:41 +020026#include <linux/bsearch.h>
27#include <linux/sort.h>
Cedric Le Goateracce2922007-07-15 23:40:59 -070028
Pavel Emelyanov61642812011-01-12 17:00:46 -080029static struct kmem_cache *user_ns_cachep __read_mostly;
Eric W. Biedermanf0d62ae2014-12-09 14:03:14 -060030static DEFINE_MUTEX(userns_state_mutex);
Pavel Emelyanov61642812011-01-12 17:00:46 -080031
Eric W. Biederman67080752013-04-14 13:47:02 -070032static bool new_idmap_permitted(const struct file *file,
33 struct user_namespace *ns, int cap_setid,
Eric W. Biederman22d917d2011-11-17 00:11:58 -080034 struct uid_gid_map *map);
Eric W. Biedermanb0321322016-07-30 13:53:37 -050035static void free_user_ns(struct work_struct *work);
Eric W. Biederman22d917d2011-11-17 00:11:58 -080036
Eric W. Biederman25f9c082016-08-08 14:41:52 -050037static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
38{
39 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
40}
41
42static void dec_user_namespaces(struct ucounts *ucounts)
43{
44 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
45}
46
Eric W. Biedermancde19752012-07-26 06:24:06 -070047static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
48{
49 /* Start with the same capabilities as init but useless for doing
50 * anything as the capabilities are bound to the new user namespace.
51 */
52 cred->securebits = SECUREBITS_DEFAULT;
53 cred->cap_inheritable = CAP_EMPTY_SET;
54 cred->cap_permitted = CAP_FULL_SET;
55 cred->cap_effective = CAP_FULL_SET;
Andy Lutomirski58319052015-09-04 15:42:45 -070056 cred->cap_ambient = CAP_EMPTY_SET;
Eric W. Biedermancde19752012-07-26 06:24:06 -070057 cred->cap_bset = CAP_FULL_SET;
58#ifdef CONFIG_KEYS
59 key_put(cred->request_key_auth);
60 cred->request_key_auth = NULL;
61#endif
62 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
63 cred->user_ns = user_ns;
64}
65
Serge E. Hallyn77ec7392007-07-15 23:41:01 -070066/*
Serge Hallyn18b6e042008-10-15 16:38:45 -050067 * Create a new user namespace, deriving the creator from the user in the
68 * passed credentials, and replacing that user with the new root user for the
69 * new namespace.
70 *
71 * This is called by copy_creds(), which will finish setting the target task's
72 * credentials.
Serge E. Hallyn77ec7392007-07-15 23:41:01 -070073 */
Serge Hallyn18b6e042008-10-15 16:38:45 -050074int create_user_ns(struct cred *new)
Serge E. Hallyn77ec7392007-07-15 23:41:01 -070075{
Eric W. Biederman0093ccb2011-11-16 21:52:53 -080076 struct user_namespace *ns, *parent_ns = new->user_ns;
Eric W. Biederman078de5f2012-02-08 07:00:08 -080077 kuid_t owner = new->euid;
78 kgid_t group = new->egid;
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -050079 struct ucounts *ucounts;
Eric W. Biederman25f9c082016-08-08 14:41:52 -050080 int ret, i;
Eric W. Biederman783291e2011-11-17 01:32:59 -080081
Eric W. Biedermandf75e772016-09-22 13:08:36 -050082 ret = -ENOSPC;
Oleg Nesterov8742f222013-08-08 18:55:32 +020083 if (parent_ns->level > 32)
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -050084 goto fail;
85
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -050086 ucounts = inc_user_namespaces(parent_ns, owner);
87 if (!ucounts)
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -050088 goto fail;
Oleg Nesterov8742f222013-08-08 18:55:32 +020089
Eric W. Biederman31515272013-03-15 01:45:51 -070090 /*
91 * Verify that we can not violate the policy of which files
92 * may be accessed that is specified by the root directory,
93 * by verifing that the root directory is at the root of the
94 * mount namespace which allows all files to be accessed.
95 */
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -050096 ret = -EPERM;
Eric W. Biederman31515272013-03-15 01:45:51 -070097 if (current_chrooted())
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -050098 goto fail_dec;
Eric W. Biederman31515272013-03-15 01:45:51 -070099
Eric W. Biederman783291e2011-11-17 01:32:59 -0800100 /* The creator needs a mapping in the parent user namespace
101 * or else we won't be able to reasonably tell userspace who
102 * created a user_namespace.
103 */
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500104 ret = -EPERM;
Eric W. Biederman783291e2011-11-17 01:32:59 -0800105 if (!kuid_has_mapping(parent_ns, owner) ||
106 !kgid_has_mapping(parent_ns, group))
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500107 goto fail_dec;
Serge E. Hallyn77ec7392007-07-15 23:41:01 -0700108
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500109 ret = -ENOMEM;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800110 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
Serge E. Hallyn77ec7392007-07-15 23:41:01 -0700111 if (!ns)
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500112 goto fail_dec;
Serge E. Hallyn77ec7392007-07-15 23:41:01 -0700113
Al Viro6344c432014-11-01 00:45:45 -0400114 ret = ns_alloc_inum(&ns->ns);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500115 if (ret)
116 goto fail_free;
Al Viro33c42942014-11-01 02:32:53 -0400117 ns->ns.ops = &userns_operations;
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700118
Eric W. Biedermanc61a2812012-12-28 18:58:39 -0800119 atomic_set(&ns->count, 1);
Eric W. Biedermancde19752012-07-26 06:24:06 -0700120 /* Leave the new->user_ns reference with the new user namespace. */
Eric W. Biedermanaeb3ae92011-11-16 21:59:43 -0800121 ns->parent = parent_ns;
Oleg Nesterov8742f222013-08-08 18:55:32 +0200122 ns->level = parent_ns->level + 1;
Eric W. Biederman783291e2011-11-17 01:32:59 -0800123 ns->owner = owner;
124 ns->group = group;
Eric W. Biedermanb0321322016-07-30 13:53:37 -0500125 INIT_WORK(&ns->work, free_user_ns);
Eric W. Biederman25f9c082016-08-08 14:41:52 -0500126 for (i = 0; i < UCOUNT_COUNTS; i++) {
127 ns->ucount_max[i] = INT_MAX;
128 }
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500129 ns->ucounts = ucounts;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800130
Eric W. Biederman9cc46512014-12-02 12:27:26 -0600131 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
132 mutex_lock(&userns_state_mutex);
133 ns->flags = parent_ns->flags;
134 mutex_unlock(&userns_state_mutex);
135
David Howellsf36f8c72013-09-24 10:35:19 +0100136#ifdef CONFIG_PERSISTENT_KEYRINGS
137 init_rwsem(&ns->persistent_keyring_register_sem);
138#endif
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500139 ret = -ENOMEM;
140 if (!setup_userns_sysctls(ns))
141 goto fail_keyring;
142
143 set_cred_user_ns(new, ns);
Serge Hallyn18b6e042008-10-15 16:38:45 -0500144 return 0;
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500145fail_keyring:
146#ifdef CONFIG_PERSISTENT_KEYRINGS
147 key_put(ns->persistent_keyring_register);
148#endif
149 ns_free_inum(&ns->ns);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500150fail_free:
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500151 kmem_cache_free(user_ns_cachep, ns);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500152fail_dec:
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500153 dec_user_namespaces(ucounts);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500154fail:
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500155 return ret;
Cedric Le Goateracce2922007-07-15 23:40:59 -0700156}
157
Eric W. Biedermanb2e0d9872012-07-26 05:15:35 -0700158int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
159{
160 struct cred *cred;
Oleg Nesterov61609682013-08-06 19:38:55 +0200161 int err = -ENOMEM;
Eric W. Biedermanb2e0d9872012-07-26 05:15:35 -0700162
163 if (!(unshare_flags & CLONE_NEWUSER))
164 return 0;
165
166 cred = prepare_creds();
Oleg Nesterov61609682013-08-06 19:38:55 +0200167 if (cred) {
168 err = create_user_ns(cred);
169 if (err)
170 put_cred(cred);
171 else
172 *new_cred = cred;
173 }
Eric W. Biedermanb2e0d9872012-07-26 05:15:35 -0700174
Oleg Nesterov61609682013-08-06 19:38:55 +0200175 return err;
Eric W. Biedermanb2e0d9872012-07-26 05:15:35 -0700176}
177
Eric W. Biedermanb0321322016-07-30 13:53:37 -0500178static void free_user_ns(struct work_struct *work)
David Howells51708362009-02-27 14:03:03 -0800179{
Eric W. Biedermanb0321322016-07-30 13:53:37 -0500180 struct user_namespace *parent, *ns =
181 container_of(work, struct user_namespace, work);
David Howells51708362009-02-27 14:03:03 -0800182
Eric W. Biedermanc61a2812012-12-28 18:58:39 -0800183 do {
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500184 struct ucounts *ucounts = ns->ucounts;
Eric W. Biedermanc61a2812012-12-28 18:58:39 -0800185 parent = ns->parent;
Christian Brauner6397fac2017-10-25 00:04:41 +0200186 if (ns->gid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
187 kfree(ns->gid_map.forward);
188 kfree(ns->gid_map.reverse);
189 }
190 if (ns->uid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
191 kfree(ns->uid_map.forward);
192 kfree(ns->uid_map.reverse);
193 }
194 if (ns->projid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
195 kfree(ns->projid_map.forward);
196 kfree(ns->projid_map.reverse);
197 }
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500198 retire_userns_sysctls(ns);
David Howellsf36f8c72013-09-24 10:35:19 +0100199#ifdef CONFIG_PERSISTENT_KEYRINGS
200 key_put(ns->persistent_keyring_register);
201#endif
Al Viro6344c432014-11-01 00:45:45 -0400202 ns_free_inum(&ns->ns);
Eric W. Biedermanc61a2812012-12-28 18:58:39 -0800203 kmem_cache_free(user_ns_cachep, ns);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500204 dec_user_namespaces(ucounts);
Eric W. Biedermanc61a2812012-12-28 18:58:39 -0800205 ns = parent;
206 } while (atomic_dec_and_test(&parent->count));
David Howells51708362009-02-27 14:03:03 -0800207}
Eric W. Biedermanb0321322016-07-30 13:53:37 -0500208
209void __put_user_ns(struct user_namespace *ns)
210{
211 schedule_work(&ns->work);
212}
213EXPORT_SYMBOL(__put_user_ns);
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000214
Christian Brauner6397fac2017-10-25 00:04:41 +0200215/**
216 * idmap_key struct holds the information necessary to find an idmapping in a
217 * sorted idmap array. It is passed to cmp_map_id() as first argument.
218 */
219struct idmap_key {
220 bool map_up; /* true -> id from kid; false -> kid from id */
221 u32 id; /* id to find */
222 u32 count; /* == 0 unless used with map_id_range_down() */
223};
224
225/**
226 * cmp_map_id - Function to be passed to bsearch() to find the requested
227 * idmapping. Expects struct idmap_key to be passed via @k.
228 */
229static int cmp_map_id(const void *k, const void *e)
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000230{
Christian Brauner6397fac2017-10-25 00:04:41 +0200231 u32 first, last, id2;
232 const struct idmap_key *key = k;
233 const struct uid_gid_extent *el = e;
234
Eric W. Biederman11a8b922017-10-31 15:54:32 -0500235 id2 = key->id + key->count - 1;
Christian Brauner6397fac2017-10-25 00:04:41 +0200236
237 /* handle map_id_{down,up}() */
238 if (key->map_up)
239 first = el->lower_first;
240 else
241 first = el->first;
242
243 last = first + el->count - 1;
244
245 if (key->id >= first && key->id <= last &&
246 (id2 >= first && id2 <= last))
247 return 0;
248
249 if (key->id < first || id2 < first)
250 return -1;
251
252 return 1;
253}
254
255/**
256 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
257 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
258 */
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500259static struct uid_gid_extent *
260map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
Christian Brauner6397fac2017-10-25 00:04:41 +0200261{
Christian Brauner6397fac2017-10-25 00:04:41 +0200262 struct idmap_key key;
263
264 key.map_up = false;
265 key.count = count;
266 key.id = id;
267
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500268 return bsearch(&key, map->forward, extents,
269 sizeof(struct uid_gid_extent), cmp_map_id);
270}
271
272/**
273 * map_id_range_down_base - Find idmap via binary search in static extent array.
274 * Can only be called if number of mappings is equal or less than
275 * UID_GID_MAP_MAX_BASE_EXTENTS.
276 */
277static struct uid_gid_extent *
278map_id_range_down_base(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
279{
280 unsigned idx;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800281 u32 first, last, id2;
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000282
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800283 id2 = id + count - 1;
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000284
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800285 /* Find the matching extent */
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800286 for (idx = 0; idx < extents; idx++) {
287 first = map->extent[idx].first;
288 last = first + map->extent[idx].count - 1;
289 if (id >= first && id <= last &&
290 (id2 >= first && id2 <= last))
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500291 return &map->extent[idx];
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000292 }
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500293 return NULL;
294}
295
296static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
297{
298 struct uid_gid_extent *extent;
299 unsigned extents = map->nr_extents;
Christian Brauner6397fac2017-10-25 00:04:41 +0200300 smp_rmb();
301
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500302 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
303 extent = map_id_range_down_base(extents, map, id, count);
304 else
305 extent = map_id_range_down_max(extents, map, id, count);
306
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800307 /* Map the id or note failure */
Christian Brauner6397fac2017-10-25 00:04:41 +0200308 if (extent)
309 id = (id - extent->first) + extent->lower_first;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800310 else
311 id = (u32) -1;
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000312
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800313 return id;
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000314}
315
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800316static u32 map_id_down(struct uid_gid_map *map, u32 id)
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000317{
Eric W. Biedermanece66132017-10-31 16:53:09 -0500318 return map_id_range_down(map, id, 1);
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800319}
320
Christian Brauner6397fac2017-10-25 00:04:41 +0200321/**
322 * map_id_up_base - Find idmap via binary search in static extent array.
323 * Can only be called if number of mappings is equal or less than
324 * UID_GID_MAP_MAX_BASE_EXTENTS.
325 */
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500326static struct uid_gid_extent *
327map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800328{
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500329 unsigned idx;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800330 u32 first, last;
331
332 /* Find the matching extent */
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800333 for (idx = 0; idx < extents; idx++) {
334 first = map->extent[idx].lower_first;
335 last = first + map->extent[idx].count - 1;
336 if (id >= first && id <= last)
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500337 return &map->extent[idx];
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800338 }
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500339 return NULL;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800340}
341
342/**
Christian Brauner6397fac2017-10-25 00:04:41 +0200343 * map_id_up_max - Find idmap via binary search in ordered idmap array.
344 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
345 */
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500346static struct uid_gid_extent *
347map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
Christian Brauner6397fac2017-10-25 00:04:41 +0200348{
Christian Brauner6397fac2017-10-25 00:04:41 +0200349 struct idmap_key key;
350
351 key.map_up = true;
Eric W. Biederman11a8b922017-10-31 15:54:32 -0500352 key.count = 1;
Christian Brauner6397fac2017-10-25 00:04:41 +0200353 key.id = id;
354
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500355 return bsearch(&key, map->reverse, extents,
356 sizeof(struct uid_gid_extent), cmp_map_id);
357}
358
359static u32 map_id_up(struct uid_gid_map *map, u32 id)
360{
361 struct uid_gid_extent *extent;
362 unsigned extents = map->nr_extents;
Christian Brauner6397fac2017-10-25 00:04:41 +0200363 smp_rmb();
364
Eric W. Biederman3edf6522017-10-31 16:27:29 -0500365 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
366 extent = map_id_up_base(extents, map, id);
367 else
368 extent = map_id_up_max(extents, map, id);
369
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800370 /* Map the id or note failure */
Christian Brauner6397fac2017-10-25 00:04:41 +0200371 if (extent)
372 id = (id - extent->lower_first) + extent->first;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800373 else
374 id = (u32) -1;
375
376 return id;
377}
378
379/**
380 * make_kuid - Map a user-namespace uid pair into a kuid.
381 * @ns: User namespace that the uid is in
382 * @uid: User identifier
383 *
384 * Maps a user-namespace uid pair into a kernel internal kuid,
385 * and returns that kuid.
386 *
387 * When there is no mapping defined for the user-namespace uid
388 * pair INVALID_UID is returned. Callers are expected to test
Brian Campbellb080e042014-02-16 22:58:12 -0500389 * for and handle INVALID_UID being returned. INVALID_UID
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800390 * may be tested for using uid_valid().
391 */
392kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
393{
394 /* Map the uid to a global kernel uid */
395 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
396}
397EXPORT_SYMBOL(make_kuid);
398
399/**
400 * from_kuid - Create a uid from a kuid user-namespace pair.
401 * @targ: The user namespace we want a uid in.
402 * @kuid: The kernel internal uid to start with.
403 *
404 * Map @kuid into the user-namespace specified by @targ and
405 * return the resulting uid.
406 *
407 * There is always a mapping into the initial user_namespace.
408 *
409 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
410 */
411uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
412{
413 /* Map the uid from a global kernel uid */
414 return map_id_up(&targ->uid_map, __kuid_val(kuid));
415}
416EXPORT_SYMBOL(from_kuid);
417
418/**
419 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
420 * @targ: The user namespace we want a uid in.
421 * @kuid: The kernel internal uid to start with.
422 *
423 * Map @kuid into the user-namespace specified by @targ and
424 * return the resulting uid.
425 *
426 * There is always a mapping into the initial user_namespace.
427 *
428 * Unlike from_kuid from_kuid_munged never fails and always
429 * returns a valid uid. This makes from_kuid_munged appropriate
430 * for use in syscalls like stat and getuid where failing the
431 * system call and failing to provide a valid uid are not an
432 * options.
433 *
434 * If @kuid has no mapping in @targ overflowuid is returned.
435 */
436uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
437{
438 uid_t uid;
439 uid = from_kuid(targ, kuid);
440
441 if (uid == (uid_t) -1)
442 uid = overflowuid;
443 return uid;
444}
445EXPORT_SYMBOL(from_kuid_munged);
446
447/**
448 * make_kgid - Map a user-namespace gid pair into a kgid.
449 * @ns: User namespace that the gid is in
Fabian Frederick68a9a432014-06-06 14:37:21 -0700450 * @gid: group identifier
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800451 *
452 * Maps a user-namespace gid pair into a kernel internal kgid,
453 * and returns that kgid.
454 *
455 * When there is no mapping defined for the user-namespace gid
456 * pair INVALID_GID is returned. Callers are expected to test
457 * for and handle INVALID_GID being returned. INVALID_GID may be
458 * tested for using gid_valid().
459 */
460kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
461{
462 /* Map the gid to a global kernel gid */
463 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
464}
465EXPORT_SYMBOL(make_kgid);
466
467/**
468 * from_kgid - Create a gid from a kgid user-namespace pair.
469 * @targ: The user namespace we want a gid in.
470 * @kgid: The kernel internal gid to start with.
471 *
472 * Map @kgid into the user-namespace specified by @targ and
473 * return the resulting gid.
474 *
475 * There is always a mapping into the initial user_namespace.
476 *
477 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
478 */
479gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
480{
481 /* Map the gid from a global kernel gid */
482 return map_id_up(&targ->gid_map, __kgid_val(kgid));
483}
484EXPORT_SYMBOL(from_kgid);
485
486/**
487 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
488 * @targ: The user namespace we want a gid in.
489 * @kgid: The kernel internal gid to start with.
490 *
491 * Map @kgid into the user-namespace specified by @targ and
492 * return the resulting gid.
493 *
494 * There is always a mapping into the initial user_namespace.
495 *
496 * Unlike from_kgid from_kgid_munged never fails and always
497 * returns a valid gid. This makes from_kgid_munged appropriate
498 * for use in syscalls like stat and getgid where failing the
499 * system call and failing to provide a valid gid are not options.
500 *
501 * If @kgid has no mapping in @targ overflowgid is returned.
502 */
503gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
504{
505 gid_t gid;
506 gid = from_kgid(targ, kgid);
507
508 if (gid == (gid_t) -1)
509 gid = overflowgid;
510 return gid;
511}
512EXPORT_SYMBOL(from_kgid_munged);
513
Eric W. Biedermanf76d2072012-08-30 01:24:05 -0700514/**
515 * make_kprojid - Map a user-namespace projid pair into a kprojid.
516 * @ns: User namespace that the projid is in
517 * @projid: Project identifier
518 *
519 * Maps a user-namespace uid pair into a kernel internal kuid,
520 * and returns that kuid.
521 *
522 * When there is no mapping defined for the user-namespace projid
523 * pair INVALID_PROJID is returned. Callers are expected to test
524 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
525 * may be tested for using projid_valid().
526 */
527kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
528{
529 /* Map the uid to a global kernel uid */
530 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
531}
532EXPORT_SYMBOL(make_kprojid);
533
534/**
535 * from_kprojid - Create a projid from a kprojid user-namespace pair.
536 * @targ: The user namespace we want a projid in.
537 * @kprojid: The kernel internal project identifier to start with.
538 *
539 * Map @kprojid into the user-namespace specified by @targ and
540 * return the resulting projid.
541 *
542 * There is always a mapping into the initial user_namespace.
543 *
544 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
545 */
546projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
547{
548 /* Map the uid from a global kernel uid */
549 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
550}
551EXPORT_SYMBOL(from_kprojid);
552
553/**
554 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
555 * @targ: The user namespace we want a projid in.
556 * @kprojid: The kernel internal projid to start with.
557 *
558 * Map @kprojid into the user-namespace specified by @targ and
559 * return the resulting projid.
560 *
561 * There is always a mapping into the initial user_namespace.
562 *
563 * Unlike from_kprojid from_kprojid_munged never fails and always
564 * returns a valid projid. This makes from_kprojid_munged
565 * appropriate for use in syscalls like stat and where
566 * failing the system call and failing to provide a valid projid are
567 * not an options.
568 *
569 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
570 */
571projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
572{
573 projid_t projid;
574 projid = from_kprojid(targ, kprojid);
575
576 if (projid == (projid_t) -1)
577 projid = OVERFLOW_PROJID;
578 return projid;
579}
580EXPORT_SYMBOL(from_kprojid_munged);
581
582
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800583static int uid_m_show(struct seq_file *seq, void *v)
584{
585 struct user_namespace *ns = seq->private;
586 struct uid_gid_extent *extent = v;
587 struct user_namespace *lower_ns;
588 uid_t lower;
589
Eric W. Biedermanc450f372012-08-14 21:25:13 -0700590 lower_ns = seq_user_ns(seq);
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800591 if ((lower_ns == ns) && lower_ns->parent)
592 lower_ns = lower_ns->parent;
593
594 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
595
596 seq_printf(seq, "%10u %10u %10u\n",
597 extent->first,
598 lower,
599 extent->count);
600
601 return 0;
602}
603
604static int gid_m_show(struct seq_file *seq, void *v)
605{
606 struct user_namespace *ns = seq->private;
607 struct uid_gid_extent *extent = v;
608 struct user_namespace *lower_ns;
609 gid_t lower;
610
Eric W. Biedermanc450f372012-08-14 21:25:13 -0700611 lower_ns = seq_user_ns(seq);
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800612 if ((lower_ns == ns) && lower_ns->parent)
613 lower_ns = lower_ns->parent;
614
615 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
616
617 seq_printf(seq, "%10u %10u %10u\n",
618 extent->first,
619 lower,
620 extent->count);
621
622 return 0;
623}
624
Eric W. Biedermanf76d2072012-08-30 01:24:05 -0700625static int projid_m_show(struct seq_file *seq, void *v)
626{
627 struct user_namespace *ns = seq->private;
628 struct uid_gid_extent *extent = v;
629 struct user_namespace *lower_ns;
630 projid_t lower;
631
632 lower_ns = seq_user_ns(seq);
633 if ((lower_ns == ns) && lower_ns->parent)
634 lower_ns = lower_ns->parent;
635
636 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
637
638 seq_printf(seq, "%10u %10u %10u\n",
639 extent->first,
640 lower,
641 extent->count);
642
643 return 0;
644}
645
Fabian Frederick68a9a432014-06-06 14:37:21 -0700646static void *m_start(struct seq_file *seq, loff_t *ppos,
647 struct uid_gid_map *map)
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800648{
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800649 loff_t pos = *ppos;
Eric W. Biedermand5e7b3c2017-10-31 17:09:34 -0500650 unsigned extents = map->nr_extents;
651 smp_rmb();
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800652
Eric W. Biedermand5e7b3c2017-10-31 17:09:34 -0500653 if (pos >= extents)
Christian Brauner6397fac2017-10-25 00:04:41 +0200654 return NULL;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800655
Eric W. Biedermand5e7b3c2017-10-31 17:09:34 -0500656 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
Christian Brauner6397fac2017-10-25 00:04:41 +0200657 return &map->extent[pos];
658
659 return &map->forward[pos];
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800660}
661
662static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
663{
664 struct user_namespace *ns = seq->private;
665
666 return m_start(seq, ppos, &ns->uid_map);
667}
668
669static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
670{
671 struct user_namespace *ns = seq->private;
672
673 return m_start(seq, ppos, &ns->gid_map);
674}
675
Eric W. Biedermanf76d2072012-08-30 01:24:05 -0700676static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
677{
678 struct user_namespace *ns = seq->private;
679
680 return m_start(seq, ppos, &ns->projid_map);
681}
682
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800683static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
684{
685 (*pos)++;
686 return seq->op->start(seq, pos);
687}
688
689static void m_stop(struct seq_file *seq, void *v)
690{
691 return;
692}
693
Fabian Frederickccf94f12014-08-08 14:21:22 -0700694const struct seq_operations proc_uid_seq_operations = {
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800695 .start = uid_m_start,
696 .stop = m_stop,
697 .next = m_next,
698 .show = uid_m_show,
699};
700
Fabian Frederickccf94f12014-08-08 14:21:22 -0700701const struct seq_operations proc_gid_seq_operations = {
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800702 .start = gid_m_start,
703 .stop = m_stop,
704 .next = m_next,
705 .show = gid_m_show,
706};
707
Fabian Frederickccf94f12014-08-08 14:21:22 -0700708const struct seq_operations proc_projid_seq_operations = {
Eric W. Biedermanf76d2072012-08-30 01:24:05 -0700709 .start = projid_m_start,
710 .stop = m_stop,
711 .next = m_next,
712 .show = projid_m_show,
713};
714
Fabian Frederick68a9a432014-06-06 14:37:21 -0700715static bool mappings_overlap(struct uid_gid_map *new_map,
716 struct uid_gid_extent *extent)
Eric W. Biederman0bd14b42012-12-27 22:27:29 -0800717{
718 u32 upper_first, lower_first, upper_last, lower_last;
719 unsigned idx;
720
721 upper_first = extent->first;
722 lower_first = extent->lower_first;
723 upper_last = upper_first + extent->count - 1;
724 lower_last = lower_first + extent->count - 1;
725
726 for (idx = 0; idx < new_map->nr_extents; idx++) {
727 u32 prev_upper_first, prev_lower_first;
728 u32 prev_upper_last, prev_lower_last;
729 struct uid_gid_extent *prev;
730
Christian Brauner6397fac2017-10-25 00:04:41 +0200731 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
732 prev = &new_map->extent[idx];
733 else
734 prev = &new_map->forward[idx];
Eric W. Biederman0bd14b42012-12-27 22:27:29 -0800735
736 prev_upper_first = prev->first;
737 prev_lower_first = prev->lower_first;
738 prev_upper_last = prev_upper_first + prev->count - 1;
739 prev_lower_last = prev_lower_first + prev->count - 1;
740
741 /* Does the upper range intersect a previous extent? */
742 if ((prev_upper_first <= upper_last) &&
743 (prev_upper_last >= upper_first))
744 return true;
745
746 /* Does the lower range intersect a previous extent? */
747 if ((prev_lower_first <= lower_last) &&
748 (prev_lower_last >= lower_first))
749 return true;
750 }
751 return false;
752}
753
Christian Brauner6397fac2017-10-25 00:04:41 +0200754/**
755 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
756 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
757 * UID_GID_MAP_MAX_BASE_EXTENTS.
758 */
759static int insert_extent(struct uid_gid_map *map, struct uid_gid_extent *extent)
760{
Eric W. Biederman3fda0e72017-10-31 17:15:30 -0500761 struct uid_gid_extent *dest;
Christian Brauner6397fac2017-10-25 00:04:41 +0200762
763 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
764 struct uid_gid_extent *forward;
765
766 /* Allocate memory for 340 mappings. */
767 forward = kmalloc(sizeof(struct uid_gid_extent) *
768 UID_GID_MAP_MAX_EXTENTS, GFP_KERNEL);
769 if (!forward)
770 return -ENOMEM;
771
772 /* Copy over memory. Only set up memory for the forward pointer.
773 * Defer the memory setup for the reverse pointer.
774 */
775 memcpy(forward, map->extent,
776 map->nr_extents * sizeof(map->extent[0]));
777
778 map->forward = forward;
779 map->reverse = NULL;
780 }
781
Eric W. Biederman3fda0e72017-10-31 17:15:30 -0500782 if (map->nr_extents < UID_GID_MAP_MAX_BASE_EXTENTS)
783 dest = &map->extent[map->nr_extents];
784 else
785 dest = &map->forward[map->nr_extents];
786
787 *dest = *extent;
788 map->nr_extents++;
Christian Brauner6397fac2017-10-25 00:04:41 +0200789 return 0;
790}
791
792/* cmp function to sort() forward mappings */
793static int cmp_extents_forward(const void *a, const void *b)
794{
795 const struct uid_gid_extent *e1 = a;
796 const struct uid_gid_extent *e2 = b;
797
798 if (e1->first < e2->first)
799 return -1;
800
801 if (e1->first > e2->first)
802 return 1;
803
804 return 0;
805}
806
807/* cmp function to sort() reverse mappings */
808static int cmp_extents_reverse(const void *a, const void *b)
809{
810 const struct uid_gid_extent *e1 = a;
811 const struct uid_gid_extent *e2 = b;
812
813 if (e1->lower_first < e2->lower_first)
814 return -1;
815
816 if (e1->lower_first > e2->lower_first)
817 return 1;
818
819 return 0;
820}
821
822/**
823 * sort_idmaps - Sorts an array of idmap entries.
824 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
825 */
826static int sort_idmaps(struct uid_gid_map *map)
827{
828 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
829 return 0;
830
831 /* Sort forward array. */
832 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
833 cmp_extents_forward, NULL);
834
835 /* Only copy the memory from forward we actually need. */
836 map->reverse = kmemdup(map->forward,
837 map->nr_extents * sizeof(struct uid_gid_extent),
838 GFP_KERNEL);
839 if (!map->reverse)
840 return -ENOMEM;
841
842 /* Sort reverse array. */
843 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
844 cmp_extents_reverse, NULL);
845
846 return 0;
847}
848
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800849static ssize_t map_write(struct file *file, const char __user *buf,
850 size_t count, loff_t *ppos,
851 int cap_setid,
852 struct uid_gid_map *map,
853 struct uid_gid_map *parent_map)
854{
855 struct seq_file *seq = file->private_data;
856 struct user_namespace *ns = seq->private;
857 struct uid_gid_map new_map;
858 unsigned idx;
Christian Brauner6397fac2017-10-25 00:04:41 +0200859 struct uid_gid_extent extent;
Al Viro70f6cbb2015-12-24 00:13:10 -0500860 char *kbuf = NULL, *pos, *next_line;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800861 ssize_t ret = -EINVAL;
862
863 /*
Eric W. Biedermanf0d62ae2014-12-09 14:03:14 -0600864 * The userns_state_mutex serializes all writes to any given map.
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800865 *
866 * Any map is only ever written once.
867 *
868 * An id map fits within 1 cache line on most architectures.
869 *
870 * On read nothing needs to be done unless you are on an
871 * architecture with a crazy cache coherency model like alpha.
872 *
873 * There is a one time data dependency between reading the
874 * count of the extents and the values of the extents. The
875 * desired behavior is to see the values of the extents that
876 * were written before the count of the extents.
877 *
878 * To achieve this smp_wmb() is used on guarantee the write
Mikulas Patockae79323b2014-04-14 16:58:55 -0400879 * order and smp_rmb() is guaranteed that we don't have crazy
880 * architectures returning stale data.
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000881 */
Eric W. Biedermanf0d62ae2014-12-09 14:03:14 -0600882 mutex_lock(&userns_state_mutex);
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800883
Christian Brauner6397fac2017-10-25 00:04:41 +0200884 memset(&new_map, 0, sizeof(struct uid_gid_map));
885
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800886 ret = -EPERM;
887 /* Only allow one successful write to the map */
888 if (map->nr_extents != 0)
889 goto out;
890
Andy Lutomirski41c21e32013-04-14 11:44:04 -0700891 /*
892 * Adjusting namespace settings requires capabilities on the target.
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800893 */
Andy Lutomirski41c21e32013-04-14 11:44:04 -0700894 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800895 goto out;
896
Eric W. Biederman36476be2014-12-05 20:03:28 -0600897 /* Only allow < page size writes at the beginning of the file */
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800898 ret = -EINVAL;
899 if ((*ppos != 0) || (count >= PAGE_SIZE))
900 goto out;
901
902 /* Slurp in the user data */
Al Viro70f6cbb2015-12-24 00:13:10 -0500903 kbuf = memdup_user_nul(buf, count);
904 if (IS_ERR(kbuf)) {
905 ret = PTR_ERR(kbuf);
906 kbuf = NULL;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800907 goto out;
Al Viro70f6cbb2015-12-24 00:13:10 -0500908 }
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800909
910 /* Parse the user data */
911 ret = -EINVAL;
912 pos = kbuf;
Fabian Frederick68a9a432014-06-06 14:37:21 -0700913 for (; pos; pos = next_line) {
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800914
915 /* Find the end of line and ensure I don't look past it */
916 next_line = strchr(pos, '\n');
917 if (next_line) {
918 *next_line = '\0';
919 next_line++;
920 if (*next_line == '\0')
921 next_line = NULL;
Eric W. Biederman5c1469d2010-06-13 03:28:03 +0000922 }
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800923
924 pos = skip_spaces(pos);
Christian Brauner6397fac2017-10-25 00:04:41 +0200925 extent.first = simple_strtoul(pos, &pos, 10);
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800926 if (!isspace(*pos))
927 goto out;
928
929 pos = skip_spaces(pos);
Christian Brauner6397fac2017-10-25 00:04:41 +0200930 extent.lower_first = simple_strtoul(pos, &pos, 10);
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800931 if (!isspace(*pos))
932 goto out;
933
934 pos = skip_spaces(pos);
Christian Brauner6397fac2017-10-25 00:04:41 +0200935 extent.count = simple_strtoul(pos, &pos, 10);
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800936 if (*pos && !isspace(*pos))
937 goto out;
938
939 /* Verify there is not trailing junk on the line */
940 pos = skip_spaces(pos);
941 if (*pos != '\0')
942 goto out;
943
944 /* Verify we have been given valid starting values */
Christian Brauner6397fac2017-10-25 00:04:41 +0200945 if ((extent.first == (u32) -1) ||
946 (extent.lower_first == (u32) -1))
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800947 goto out;
948
Fabian Frederick68a9a432014-06-06 14:37:21 -0700949 /* Verify count is not zero and does not cause the
950 * extent to wrap
951 */
Christian Brauner6397fac2017-10-25 00:04:41 +0200952 if ((extent.first + extent.count) <= extent.first)
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800953 goto out;
Christian Brauner6397fac2017-10-25 00:04:41 +0200954 if ((extent.lower_first + extent.count) <=
955 extent.lower_first)
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800956 goto out;
957
Eric W. Biederman0bd14b42012-12-27 22:27:29 -0800958 /* Do the ranges in extent overlap any previous extents? */
Christian Brauner6397fac2017-10-25 00:04:41 +0200959 if (mappings_overlap(&new_map, &extent))
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800960 goto out;
961
Christian Brauner6397fac2017-10-25 00:04:41 +0200962 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800963 (next_line != NULL))
964 goto out;
Christian Brauner6397fac2017-10-25 00:04:41 +0200965
966 ret = insert_extent(&new_map, &extent);
967 if (ret < 0)
968 goto out;
969 ret = -EINVAL;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800970 }
971 /* Be very certaint the new map actually exists */
972 if (new_map.nr_extents == 0)
973 goto out;
974
975 ret = -EPERM;
976 /* Validate the user is allowed to use user id's mapped to. */
Eric W. Biederman67080752013-04-14 13:47:02 -0700977 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800978 goto out;
979
Christian Brauner6397fac2017-10-25 00:04:41 +0200980 ret = sort_idmaps(&new_map);
981 if (ret < 0)
982 goto out;
983
984 ret = -EPERM;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800985 /* Map the lower ids from the parent user namespace to the
986 * kernel global id space.
987 */
988 for (idx = 0; idx < new_map.nr_extents; idx++) {
Christian Brauner6397fac2017-10-25 00:04:41 +0200989 struct uid_gid_extent *e;
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800990 u32 lower_first;
Christian Brauner6397fac2017-10-25 00:04:41 +0200991
992 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
993 e = &new_map.extent[idx];
994 else
995 e = &new_map.forward[idx];
Eric W. Biederman22d917d2011-11-17 00:11:58 -0800996
997 lower_first = map_id_range_down(parent_map,
Christian Brauner6397fac2017-10-25 00:04:41 +0200998 e->lower_first,
999 e->count);
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001000
1001 /* Fail if we can not map the specified extent to
1002 * the kernel global id space.
1003 */
1004 if (lower_first == (u32) -1)
1005 goto out;
1006
Christian Brauner6397fac2017-10-25 00:04:41 +02001007 e->lower_first = lower_first;
Eric W. Biederman5c1469d2010-06-13 03:28:03 +00001008 }
1009
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001010 /* Install the map */
Christian Brauner6397fac2017-10-25 00:04:41 +02001011 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1012 memcpy(map->extent, new_map.extent,
1013 new_map.nr_extents * sizeof(new_map.extent[0]));
1014 } else {
1015 map->forward = new_map.forward;
1016 map->reverse = new_map.reverse;
1017 }
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001018 smp_wmb();
1019 map->nr_extents = new_map.nr_extents;
1020
1021 *ppos = count;
1022 ret = count;
1023out:
Christian Brauner6397fac2017-10-25 00:04:41 +02001024 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1025 kfree(new_map.forward);
1026 kfree(new_map.reverse);
1027 map->forward = NULL;
1028 map->reverse = NULL;
1029 map->nr_extents = 0;
1030 }
1031
Eric W. Biedermanf0d62ae2014-12-09 14:03:14 -06001032 mutex_unlock(&userns_state_mutex);
Al Viro70f6cbb2015-12-24 00:13:10 -05001033 kfree(kbuf);
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001034 return ret;
1035}
1036
Fabian Frederick68a9a432014-06-06 14:37:21 -07001037ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1038 size_t size, loff_t *ppos)
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001039{
1040 struct seq_file *seq = file->private_data;
1041 struct user_namespace *ns = seq->private;
Eric W. Biedermanc450f372012-08-14 21:25:13 -07001042 struct user_namespace *seq_ns = seq_user_ns(seq);
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001043
1044 if (!ns->parent)
1045 return -EPERM;
1046
Eric W. Biedermanc450f372012-08-14 21:25:13 -07001047 if ((seq_ns != ns) && (seq_ns != ns->parent))
1048 return -EPERM;
1049
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001050 return map_write(file, buf, size, ppos, CAP_SETUID,
1051 &ns->uid_map, &ns->parent->uid_map);
1052}
1053
Fabian Frederick68a9a432014-06-06 14:37:21 -07001054ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1055 size_t size, loff_t *ppos)
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001056{
1057 struct seq_file *seq = file->private_data;
1058 struct user_namespace *ns = seq->private;
Eric W. Biedermanc450f372012-08-14 21:25:13 -07001059 struct user_namespace *seq_ns = seq_user_ns(seq);
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001060
1061 if (!ns->parent)
1062 return -EPERM;
1063
Eric W. Biedermanc450f372012-08-14 21:25:13 -07001064 if ((seq_ns != ns) && (seq_ns != ns->parent))
1065 return -EPERM;
1066
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001067 return map_write(file, buf, size, ppos, CAP_SETGID,
1068 &ns->gid_map, &ns->parent->gid_map);
1069}
1070
Fabian Frederick68a9a432014-06-06 14:37:21 -07001071ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1072 size_t size, loff_t *ppos)
Eric W. Biedermanf76d2072012-08-30 01:24:05 -07001073{
1074 struct seq_file *seq = file->private_data;
1075 struct user_namespace *ns = seq->private;
1076 struct user_namespace *seq_ns = seq_user_ns(seq);
1077
1078 if (!ns->parent)
1079 return -EPERM;
1080
1081 if ((seq_ns != ns) && (seq_ns != ns->parent))
1082 return -EPERM;
1083
1084 /* Anyone can set any valid project id no capability needed */
1085 return map_write(file, buf, size, ppos, -1,
1086 &ns->projid_map, &ns->parent->projid_map);
1087}
1088
Fabian Frederick68a9a432014-06-06 14:37:21 -07001089static bool new_idmap_permitted(const struct file *file,
Eric W. Biederman67080752013-04-14 13:47:02 -07001090 struct user_namespace *ns, int cap_setid,
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001091 struct uid_gid_map *new_map)
1092{
Eric W. Biedermanf95d7912014-11-26 23:22:14 -06001093 const struct cred *cred = file->f_cred;
Eric W. Biederman0542f172014-12-05 17:51:47 -06001094 /* Don't allow mappings that would allow anything that wouldn't
1095 * be allowed without the establishment of unprivileged mappings.
1096 */
Eric W. Biedermanf95d7912014-11-26 23:22:14 -06001097 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1098 uid_eq(ns->owner, cred->euid)) {
Eric W. Biederman37657da2012-07-27 06:21:27 -07001099 u32 id = new_map->extent[0].lower_first;
1100 if (cap_setid == CAP_SETUID) {
1101 kuid_t uid = make_kuid(ns->parent, id);
Eric W. Biedermanf95d7912014-11-26 23:22:14 -06001102 if (uid_eq(uid, cred->euid))
Eric W. Biederman37657da2012-07-27 06:21:27 -07001103 return true;
Fabian Frederick68a9a432014-06-06 14:37:21 -07001104 } else if (cap_setid == CAP_SETGID) {
Eric W. Biederman37657da2012-07-27 06:21:27 -07001105 kgid_t gid = make_kgid(ns->parent, id);
Eric W. Biederman66d2f332014-12-05 19:36:04 -06001106 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1107 gid_eq(gid, cred->egid))
Eric W. Biederman37657da2012-07-27 06:21:27 -07001108 return true;
1109 }
1110 }
1111
Eric W. Biedermanf76d2072012-08-30 01:24:05 -07001112 /* Allow anyone to set a mapping that doesn't require privilege */
1113 if (!cap_valid(cap_setid))
1114 return true;
1115
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001116 /* Allow the specified ids if we have the appropriate capability
1117 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
Eric W. Biederman67080752013-04-14 13:47:02 -07001118 * And the opener of the id file also had the approprpiate capability.
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001119 */
Eric W. Biederman67080752013-04-14 13:47:02 -07001120 if (ns_capable(ns->parent, cap_setid) &&
1121 file_ns_capable(file, ns->parent, cap_setid))
Eric W. Biederman22d917d2011-11-17 00:11:58 -08001122 return true;
1123
1124 return false;
Eric W. Biederman5c1469d2010-06-13 03:28:03 +00001125}
Pavel Emelyanov61642812011-01-12 17:00:46 -08001126
Eric W. Biederman9cc46512014-12-02 12:27:26 -06001127int proc_setgroups_show(struct seq_file *seq, void *v)
1128{
1129 struct user_namespace *ns = seq->private;
Mark Rutland6aa7de02017-10-23 14:07:29 -07001130 unsigned long userns_flags = READ_ONCE(ns->flags);
Eric W. Biederman9cc46512014-12-02 12:27:26 -06001131
1132 seq_printf(seq, "%s\n",
1133 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1134 "allow" : "deny");
1135 return 0;
1136}
1137
1138ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1139 size_t count, loff_t *ppos)
1140{
1141 struct seq_file *seq = file->private_data;
1142 struct user_namespace *ns = seq->private;
1143 char kbuf[8], *pos;
1144 bool setgroups_allowed;
1145 ssize_t ret;
1146
1147 /* Only allow a very narrow range of strings to be written */
1148 ret = -EINVAL;
1149 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1150 goto out;
1151
1152 /* What was written? */
1153 ret = -EFAULT;
1154 if (copy_from_user(kbuf, buf, count))
1155 goto out;
1156 kbuf[count] = '\0';
1157 pos = kbuf;
1158
1159 /* What is being requested? */
1160 ret = -EINVAL;
1161 if (strncmp(pos, "allow", 5) == 0) {
1162 pos += 5;
1163 setgroups_allowed = true;
1164 }
1165 else if (strncmp(pos, "deny", 4) == 0) {
1166 pos += 4;
1167 setgroups_allowed = false;
1168 }
1169 else
1170 goto out;
1171
1172 /* Verify there is not trailing junk on the line */
1173 pos = skip_spaces(pos);
1174 if (*pos != '\0')
1175 goto out;
1176
1177 ret = -EPERM;
1178 mutex_lock(&userns_state_mutex);
1179 if (setgroups_allowed) {
1180 /* Enabling setgroups after setgroups has been disabled
1181 * is not allowed.
1182 */
1183 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1184 goto out_unlock;
1185 } else {
1186 /* Permanently disabling setgroups after setgroups has
1187 * been enabled by writing the gid_map is not allowed.
1188 */
1189 if (ns->gid_map.nr_extents != 0)
1190 goto out_unlock;
1191 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1192 }
1193 mutex_unlock(&userns_state_mutex);
1194
1195 /* Report a successful write */
1196 *ppos = count;
1197 ret = count;
1198out:
1199 return ret;
1200out_unlock:
1201 mutex_unlock(&userns_state_mutex);
1202 goto out;
1203}
1204
Eric W. Biederman273d2c62014-12-05 18:01:11 -06001205bool userns_may_setgroups(const struct user_namespace *ns)
1206{
1207 bool allowed;
1208
Eric W. Biedermanf0d62ae2014-12-09 14:03:14 -06001209 mutex_lock(&userns_state_mutex);
Eric W. Biederman273d2c62014-12-05 18:01:11 -06001210 /* It is not safe to use setgroups until a gid mapping in
1211 * the user namespace has been established.
1212 */
1213 allowed = ns->gid_map.nr_extents != 0;
Eric W. Biederman9cc46512014-12-02 12:27:26 -06001214 /* Is setgroups allowed? */
1215 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
Eric W. Biedermanf0d62ae2014-12-09 14:03:14 -06001216 mutex_unlock(&userns_state_mutex);
Eric W. Biederman273d2c62014-12-05 18:01:11 -06001217
1218 return allowed;
1219}
1220
Seth Forsheed07b8462015-09-23 15:16:04 -05001221/*
Eric W. Biedermana2b42622017-04-29 14:12:15 -05001222 * Returns true if @child is the same namespace or a descendant of
1223 * @ancestor.
Seth Forsheed07b8462015-09-23 15:16:04 -05001224 */
Eric W. Biedermana2b42622017-04-29 14:12:15 -05001225bool in_userns(const struct user_namespace *ancestor,
1226 const struct user_namespace *child)
1227{
1228 const struct user_namespace *ns;
1229 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1230 ;
1231 return (ns == ancestor);
1232}
1233
Seth Forsheed07b8462015-09-23 15:16:04 -05001234bool current_in_userns(const struct user_namespace *target_ns)
1235{
Eric W. Biedermana2b42622017-04-29 14:12:15 -05001236 return in_userns(target_ns, current_user_ns());
Seth Forsheed07b8462015-09-23 15:16:04 -05001237}
1238
Al Viro3c041182014-11-01 00:25:30 -04001239static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1240{
1241 return container_of(ns, struct user_namespace, ns);
1242}
1243
Al Viro64964522014-11-01 00:37:32 -04001244static struct ns_common *userns_get(struct task_struct *task)
Eric W. Biedermancde19752012-07-26 06:24:06 -07001245{
1246 struct user_namespace *user_ns;
1247
1248 rcu_read_lock();
1249 user_ns = get_user_ns(__task_cred(task)->user_ns);
1250 rcu_read_unlock();
1251
Al Viro3c041182014-11-01 00:25:30 -04001252 return user_ns ? &user_ns->ns : NULL;
Eric W. Biedermancde19752012-07-26 06:24:06 -07001253}
1254
Al Viro64964522014-11-01 00:37:32 -04001255static void userns_put(struct ns_common *ns)
Eric W. Biedermancde19752012-07-26 06:24:06 -07001256{
Al Viro3c041182014-11-01 00:25:30 -04001257 put_user_ns(to_user_ns(ns));
Eric W. Biedermancde19752012-07-26 06:24:06 -07001258}
1259
Al Viro64964522014-11-01 00:37:32 -04001260static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biedermancde19752012-07-26 06:24:06 -07001261{
Al Viro3c041182014-11-01 00:25:30 -04001262 struct user_namespace *user_ns = to_user_ns(ns);
Eric W. Biedermancde19752012-07-26 06:24:06 -07001263 struct cred *cred;
1264
1265 /* Don't allow gaining capabilities by reentering
1266 * the same user namespace.
1267 */
1268 if (user_ns == current_user_ns())
1269 return -EINVAL;
1270
Eric W. Biedermanfaf00da2015-08-10 18:25:44 -05001271 /* Tasks that share a thread group must share a user namespace */
1272 if (!thread_group_empty(current))
Eric W. Biedermancde19752012-07-26 06:24:06 -07001273 return -EINVAL;
1274
Eric W. Biedermane66eded2013-03-13 11:51:49 -07001275 if (current->fs->users != 1)
1276 return -EINVAL;
1277
Eric W. Biedermancde19752012-07-26 06:24:06 -07001278 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1279 return -EPERM;
1280
1281 cred = prepare_creds();
1282 if (!cred)
1283 return -ENOMEM;
1284
1285 put_user_ns(cred->user_ns);
1286 set_cred_user_ns(cred, get_user_ns(user_ns));
1287
1288 return commit_creds(cred);
1289}
1290
Andrey Vaginbcac25a2016-09-06 00:47:13 -07001291struct ns_common *ns_get_owner(struct ns_common *ns)
1292{
1293 struct user_namespace *my_user_ns = current_user_ns();
1294 struct user_namespace *owner, *p;
1295
1296 /* See if the owner is in the current user namespace */
1297 owner = p = ns->ops->owner(ns);
1298 for (;;) {
1299 if (!p)
1300 return ERR_PTR(-EPERM);
1301 if (p == my_user_ns)
1302 break;
1303 p = p->parent;
1304 }
1305
1306 return &get_user_ns(owner)->ns;
1307}
1308
1309static struct user_namespace *userns_owner(struct ns_common *ns)
1310{
1311 return to_user_ns(ns)->parent;
1312}
1313
Eric W. Biedermancde19752012-07-26 06:24:06 -07001314const struct proc_ns_operations userns_operations = {
1315 .name = "user",
1316 .type = CLONE_NEWUSER,
1317 .get = userns_get,
1318 .put = userns_put,
1319 .install = userns_install,
Andrey Vaginbcac25a2016-09-06 00:47:13 -07001320 .owner = userns_owner,
Andrey Vagina7306ed2016-09-06 00:47:15 -07001321 .get_parent = ns_get_owner,
Eric W. Biedermancde19752012-07-26 06:24:06 -07001322};
1323
Pavel Emelyanov61642812011-01-12 17:00:46 -08001324static __init int user_namespaces_init(void)
1325{
1326 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1327 return 0;
1328}
Paul Gortmakerc96d6662014-04-03 14:48:35 -07001329subsys_initcall(user_namespaces_init);