kthread: rename probe_kthread_data() to kthread_probe_data()
[linux-2.6-block.git] / kernel / user_namespace.c
CommitLineData
acce292c
CLG
1/*
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
9984de1a 8#include <linux/export.h>
acce292c 9#include <linux/nsproxy.h>
1aeb272c 10#include <linux/slab.h>
acce292c 11#include <linux/user_namespace.h>
0bb80f24 12#include <linux/proc_ns.h>
5c1469de 13#include <linux/highuid.h>
18b6e041 14#include <linux/cred.h>
973c5914 15#include <linux/securebits.h>
22d917d8
EB
16#include <linux/keyctl.h>
17#include <linux/key-type.h>
18#include <keys/user-type.h>
19#include <linux/seq_file.h>
20#include <linux/fs.h>
21#include <linux/uaccess.h>
22#include <linux/ctype.h>
f76d207a 23#include <linux/projid.h>
e66eded8 24#include <linux/fs_struct.h>
acce292c 25
6164281a 26static struct kmem_cache *user_ns_cachep __read_mostly;
f0d62aec 27static DEFINE_MUTEX(userns_state_mutex);
6164281a 28
6708075f
EB
29static bool new_idmap_permitted(const struct file *file,
30 struct user_namespace *ns, int cap_setid,
22d917d8 31 struct uid_gid_map *map);
b032132c 32static void free_user_ns(struct work_struct *work);
22d917d8 33
25f9c081
EB
34static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
35{
36 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
37}
38
39static void dec_user_namespaces(struct ucounts *ucounts)
40{
41 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
42}
43
cde1975b
EB
44static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
45{
46 /* Start with the same capabilities as init but useless for doing
47 * anything as the capabilities are bound to the new user namespace.
48 */
49 cred->securebits = SECUREBITS_DEFAULT;
50 cred->cap_inheritable = CAP_EMPTY_SET;
51 cred->cap_permitted = CAP_FULL_SET;
52 cred->cap_effective = CAP_FULL_SET;
58319057 53 cred->cap_ambient = CAP_EMPTY_SET;
cde1975b
EB
54 cred->cap_bset = CAP_FULL_SET;
55#ifdef CONFIG_KEYS
56 key_put(cred->request_key_auth);
57 cred->request_key_auth = NULL;
58#endif
59 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
60 cred->user_ns = user_ns;
61}
62
77ec739d 63/*
18b6e041
SH
64 * Create a new user namespace, deriving the creator from the user in the
65 * passed credentials, and replacing that user with the new root user for the
66 * new namespace.
67 *
68 * This is called by copy_creds(), which will finish setting the target task's
69 * credentials.
77ec739d 70 */
18b6e041 71int create_user_ns(struct cred *new)
77ec739d 72{
0093ccb6 73 struct user_namespace *ns, *parent_ns = new->user_ns;
078de5f7
EB
74 kuid_t owner = new->euid;
75 kgid_t group = new->egid;
f6b2db1a 76 struct ucounts *ucounts;
25f9c081 77 int ret, i;
783291e6 78
df75e774 79 ret = -ENOSPC;
8742f229 80 if (parent_ns->level > 32)
b376c3e1
EB
81 goto fail;
82
f6b2db1a
EB
83 ucounts = inc_user_namespaces(parent_ns, owner);
84 if (!ucounts)
b376c3e1 85 goto fail;
8742f229 86
3151527e
EB
87 /*
88 * Verify that we can not violate the policy of which files
89 * may be accessed that is specified by the root directory,
90 * by verifing that the root directory is at the root of the
91 * mount namespace which allows all files to be accessed.
92 */
b376c3e1 93 ret = -EPERM;
3151527e 94 if (current_chrooted())
b376c3e1 95 goto fail_dec;
3151527e 96
783291e6
EB
97 /* The creator needs a mapping in the parent user namespace
98 * or else we won't be able to reasonably tell userspace who
99 * created a user_namespace.
100 */
b376c3e1 101 ret = -EPERM;
783291e6
EB
102 if (!kuid_has_mapping(parent_ns, owner) ||
103 !kgid_has_mapping(parent_ns, group))
b376c3e1 104 goto fail_dec;
77ec739d 105
b376c3e1 106 ret = -ENOMEM;
22d917d8 107 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
77ec739d 108 if (!ns)
b376c3e1 109 goto fail_dec;
77ec739d 110
6344c433 111 ret = ns_alloc_inum(&ns->ns);
b376c3e1
EB
112 if (ret)
113 goto fail_free;
33c42940 114 ns->ns.ops = &userns_operations;
98f842e6 115
c61a2810 116 atomic_set(&ns->count, 1);
cde1975b 117 /* Leave the new->user_ns reference with the new user namespace. */
aeb3ae9d 118 ns->parent = parent_ns;
8742f229 119 ns->level = parent_ns->level + 1;
783291e6
EB
120 ns->owner = owner;
121 ns->group = group;
b032132c 122 INIT_WORK(&ns->work, free_user_ns);
25f9c081
EB
123 for (i = 0; i < UCOUNT_COUNTS; i++) {
124 ns->ucount_max[i] = INT_MAX;
125 }
f6b2db1a 126 ns->ucounts = ucounts;
22d917d8 127
9cc46516
EB
128 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
129 mutex_lock(&userns_state_mutex);
130 ns->flags = parent_ns->flags;
131 mutex_unlock(&userns_state_mutex);
132
f36f8c75
DH
133#ifdef CONFIG_PERSISTENT_KEYRINGS
134 init_rwsem(&ns->persistent_keyring_register_sem);
135#endif
dbec2846
EB
136 ret = -ENOMEM;
137 if (!setup_userns_sysctls(ns))
138 goto fail_keyring;
139
140 set_cred_user_ns(new, ns);
18b6e041 141 return 0;
dbec2846
EB
142fail_keyring:
143#ifdef CONFIG_PERSISTENT_KEYRINGS
144 key_put(ns->persistent_keyring_register);
145#endif
146 ns_free_inum(&ns->ns);
b376c3e1 147fail_free:
dbec2846 148 kmem_cache_free(user_ns_cachep, ns);
b376c3e1 149fail_dec:
f6b2db1a 150 dec_user_namespaces(ucounts);
b376c3e1 151fail:
dbec2846 152 return ret;
acce292c
CLG
153}
154
b2e0d987
EB
155int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
156{
157 struct cred *cred;
6160968c 158 int err = -ENOMEM;
b2e0d987
EB
159
160 if (!(unshare_flags & CLONE_NEWUSER))
161 return 0;
162
163 cred = prepare_creds();
6160968c
ON
164 if (cred) {
165 err = create_user_ns(cred);
166 if (err)
167 put_cred(cred);
168 else
169 *new_cred = cred;
170 }
b2e0d987 171
6160968c 172 return err;
b2e0d987
EB
173}
174
b032132c 175static void free_user_ns(struct work_struct *work)
acce292c 176{
b032132c
EB
177 struct user_namespace *parent, *ns =
178 container_of(work, struct user_namespace, work);
783291e6 179
c61a2810 180 do {
f6b2db1a 181 struct ucounts *ucounts = ns->ucounts;
c61a2810 182 parent = ns->parent;
dbec2846 183 retire_userns_sysctls(ns);
f36f8c75
DH
184#ifdef CONFIG_PERSISTENT_KEYRINGS
185 key_put(ns->persistent_keyring_register);
186#endif
6344c433 187 ns_free_inum(&ns->ns);
c61a2810 188 kmem_cache_free(user_ns_cachep, ns);
f6b2db1a 189 dec_user_namespaces(ucounts);
c61a2810
EB
190 ns = parent;
191 } while (atomic_dec_and_test(&parent->count));
acce292c 192}
b032132c
EB
193
194void __put_user_ns(struct user_namespace *ns)
195{
196 schedule_work(&ns->work);
197}
198EXPORT_SYMBOL(__put_user_ns);
5c1469de 199
22d917d8 200static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
5c1469de 201{
22d917d8
EB
202 unsigned idx, extents;
203 u32 first, last, id2;
5c1469de 204
22d917d8 205 id2 = id + count - 1;
5c1469de 206
22d917d8
EB
207 /* Find the matching extent */
208 extents = map->nr_extents;
e79323bd 209 smp_rmb();
22d917d8
EB
210 for (idx = 0; idx < extents; idx++) {
211 first = map->extent[idx].first;
212 last = first + map->extent[idx].count - 1;
213 if (id >= first && id <= last &&
214 (id2 >= first && id2 <= last))
215 break;
216 }
217 /* Map the id or note failure */
218 if (idx < extents)
219 id = (id - first) + map->extent[idx].lower_first;
220 else
221 id = (u32) -1;
222
223 return id;
224}
225
226static u32 map_id_down(struct uid_gid_map *map, u32 id)
227{
228 unsigned idx, extents;
229 u32 first, last;
230
231 /* Find the matching extent */
232 extents = map->nr_extents;
e79323bd 233 smp_rmb();
22d917d8
EB
234 for (idx = 0; idx < extents; idx++) {
235 first = map->extent[idx].first;
236 last = first + map->extent[idx].count - 1;
237 if (id >= first && id <= last)
238 break;
239 }
240 /* Map the id or note failure */
241 if (idx < extents)
242 id = (id - first) + map->extent[idx].lower_first;
243 else
244 id = (u32) -1;
245
246 return id;
247}
248
249static u32 map_id_up(struct uid_gid_map *map, u32 id)
250{
251 unsigned idx, extents;
252 u32 first, last;
253
254 /* Find the matching extent */
255 extents = map->nr_extents;
e79323bd 256 smp_rmb();
22d917d8
EB
257 for (idx = 0; idx < extents; idx++) {
258 first = map->extent[idx].lower_first;
259 last = first + map->extent[idx].count - 1;
260 if (id >= first && id <= last)
261 break;
5c1469de 262 }
22d917d8
EB
263 /* Map the id or note failure */
264 if (idx < extents)
265 id = (id - first) + map->extent[idx].first;
266 else
267 id = (u32) -1;
268
269 return id;
270}
271
272/**
273 * make_kuid - Map a user-namespace uid pair into a kuid.
274 * @ns: User namespace that the uid is in
275 * @uid: User identifier
276 *
277 * Maps a user-namespace uid pair into a kernel internal kuid,
278 * and returns that kuid.
279 *
280 * When there is no mapping defined for the user-namespace uid
281 * pair INVALID_UID is returned. Callers are expected to test
b080e047 282 * for and handle INVALID_UID being returned. INVALID_UID
22d917d8
EB
283 * may be tested for using uid_valid().
284 */
285kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
286{
287 /* Map the uid to a global kernel uid */
288 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
289}
290EXPORT_SYMBOL(make_kuid);
291
292/**
293 * from_kuid - Create a uid from a kuid user-namespace pair.
294 * @targ: The user namespace we want a uid in.
295 * @kuid: The kernel internal uid to start with.
296 *
297 * Map @kuid into the user-namespace specified by @targ and
298 * return the resulting uid.
299 *
300 * There is always a mapping into the initial user_namespace.
301 *
302 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
303 */
304uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
305{
306 /* Map the uid from a global kernel uid */
307 return map_id_up(&targ->uid_map, __kuid_val(kuid));
308}
309EXPORT_SYMBOL(from_kuid);
310
311/**
312 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
313 * @targ: The user namespace we want a uid in.
314 * @kuid: The kernel internal uid to start with.
315 *
316 * Map @kuid into the user-namespace specified by @targ and
317 * return the resulting uid.
318 *
319 * There is always a mapping into the initial user_namespace.
320 *
321 * Unlike from_kuid from_kuid_munged never fails and always
322 * returns a valid uid. This makes from_kuid_munged appropriate
323 * for use in syscalls like stat and getuid where failing the
324 * system call and failing to provide a valid uid are not an
325 * options.
326 *
327 * If @kuid has no mapping in @targ overflowuid is returned.
328 */
329uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
330{
331 uid_t uid;
332 uid = from_kuid(targ, kuid);
333
334 if (uid == (uid_t) -1)
335 uid = overflowuid;
336 return uid;
337}
338EXPORT_SYMBOL(from_kuid_munged);
339
340/**
341 * make_kgid - Map a user-namespace gid pair into a kgid.
342 * @ns: User namespace that the gid is in
68a9a435 343 * @gid: group identifier
22d917d8
EB
344 *
345 * Maps a user-namespace gid pair into a kernel internal kgid,
346 * and returns that kgid.
347 *
348 * When there is no mapping defined for the user-namespace gid
349 * pair INVALID_GID is returned. Callers are expected to test
350 * for and handle INVALID_GID being returned. INVALID_GID may be
351 * tested for using gid_valid().
352 */
353kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
354{
355 /* Map the gid to a global kernel gid */
356 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
357}
358EXPORT_SYMBOL(make_kgid);
359
360/**
361 * from_kgid - Create a gid from a kgid user-namespace pair.
362 * @targ: The user namespace we want a gid in.
363 * @kgid: The kernel internal gid to start with.
364 *
365 * Map @kgid into the user-namespace specified by @targ and
366 * return the resulting gid.
367 *
368 * There is always a mapping into the initial user_namespace.
369 *
370 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
371 */
372gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
373{
374 /* Map the gid from a global kernel gid */
375 return map_id_up(&targ->gid_map, __kgid_val(kgid));
376}
377EXPORT_SYMBOL(from_kgid);
378
379/**
380 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
381 * @targ: The user namespace we want a gid in.
382 * @kgid: The kernel internal gid to start with.
383 *
384 * Map @kgid into the user-namespace specified by @targ and
385 * return the resulting gid.
386 *
387 * There is always a mapping into the initial user_namespace.
388 *
389 * Unlike from_kgid from_kgid_munged never fails and always
390 * returns a valid gid. This makes from_kgid_munged appropriate
391 * for use in syscalls like stat and getgid where failing the
392 * system call and failing to provide a valid gid are not options.
393 *
394 * If @kgid has no mapping in @targ overflowgid is returned.
395 */
396gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
397{
398 gid_t gid;
399 gid = from_kgid(targ, kgid);
400
401 if (gid == (gid_t) -1)
402 gid = overflowgid;
403 return gid;
404}
405EXPORT_SYMBOL(from_kgid_munged);
406
f76d207a
EB
407/**
408 * make_kprojid - Map a user-namespace projid pair into a kprojid.
409 * @ns: User namespace that the projid is in
410 * @projid: Project identifier
411 *
412 * Maps a user-namespace uid pair into a kernel internal kuid,
413 * and returns that kuid.
414 *
415 * When there is no mapping defined for the user-namespace projid
416 * pair INVALID_PROJID is returned. Callers are expected to test
417 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
418 * may be tested for using projid_valid().
419 */
420kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
421{
422 /* Map the uid to a global kernel uid */
423 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
424}
425EXPORT_SYMBOL(make_kprojid);
426
427/**
428 * from_kprojid - Create a projid from a kprojid user-namespace pair.
429 * @targ: The user namespace we want a projid in.
430 * @kprojid: The kernel internal project identifier to start with.
431 *
432 * Map @kprojid into the user-namespace specified by @targ and
433 * return the resulting projid.
434 *
435 * There is always a mapping into the initial user_namespace.
436 *
437 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
438 */
439projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
440{
441 /* Map the uid from a global kernel uid */
442 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
443}
444EXPORT_SYMBOL(from_kprojid);
445
446/**
447 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
448 * @targ: The user namespace we want a projid in.
449 * @kprojid: The kernel internal projid to start with.
450 *
451 * Map @kprojid into the user-namespace specified by @targ and
452 * return the resulting projid.
453 *
454 * There is always a mapping into the initial user_namespace.
455 *
456 * Unlike from_kprojid from_kprojid_munged never fails and always
457 * returns a valid projid. This makes from_kprojid_munged
458 * appropriate for use in syscalls like stat and where
459 * failing the system call and failing to provide a valid projid are
460 * not an options.
461 *
462 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
463 */
464projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
465{
466 projid_t projid;
467 projid = from_kprojid(targ, kprojid);
468
469 if (projid == (projid_t) -1)
470 projid = OVERFLOW_PROJID;
471 return projid;
472}
473EXPORT_SYMBOL(from_kprojid_munged);
474
475
22d917d8
EB
476static int uid_m_show(struct seq_file *seq, void *v)
477{
478 struct user_namespace *ns = seq->private;
479 struct uid_gid_extent *extent = v;
480 struct user_namespace *lower_ns;
481 uid_t lower;
5c1469de 482
c450f371 483 lower_ns = seq_user_ns(seq);
22d917d8
EB
484 if ((lower_ns == ns) && lower_ns->parent)
485 lower_ns = lower_ns->parent;
486
487 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
488
489 seq_printf(seq, "%10u %10u %10u\n",
490 extent->first,
491 lower,
492 extent->count);
493
494 return 0;
5c1469de
EB
495}
496
22d917d8 497static int gid_m_show(struct seq_file *seq, void *v)
5c1469de 498{
22d917d8
EB
499 struct user_namespace *ns = seq->private;
500 struct uid_gid_extent *extent = v;
501 struct user_namespace *lower_ns;
502 gid_t lower;
5c1469de 503
c450f371 504 lower_ns = seq_user_ns(seq);
22d917d8
EB
505 if ((lower_ns == ns) && lower_ns->parent)
506 lower_ns = lower_ns->parent;
5c1469de 507
22d917d8
EB
508 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
509
510 seq_printf(seq, "%10u %10u %10u\n",
511 extent->first,
512 lower,
513 extent->count);
514
515 return 0;
516}
517
f76d207a
EB
518static int projid_m_show(struct seq_file *seq, void *v)
519{
520 struct user_namespace *ns = seq->private;
521 struct uid_gid_extent *extent = v;
522 struct user_namespace *lower_ns;
523 projid_t lower;
524
525 lower_ns = seq_user_ns(seq);
526 if ((lower_ns == ns) && lower_ns->parent)
527 lower_ns = lower_ns->parent;
528
529 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
530
531 seq_printf(seq, "%10u %10u %10u\n",
532 extent->first,
533 lower,
534 extent->count);
535
536 return 0;
537}
538
68a9a435
FF
539static void *m_start(struct seq_file *seq, loff_t *ppos,
540 struct uid_gid_map *map)
22d917d8
EB
541{
542 struct uid_gid_extent *extent = NULL;
543 loff_t pos = *ppos;
544
545 if (pos < map->nr_extents)
546 extent = &map->extent[pos];
547
548 return extent;
549}
550
551static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
552{
553 struct user_namespace *ns = seq->private;
554
555 return m_start(seq, ppos, &ns->uid_map);
556}
557
558static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
559{
560 struct user_namespace *ns = seq->private;
561
562 return m_start(seq, ppos, &ns->gid_map);
563}
564
f76d207a
EB
565static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
566{
567 struct user_namespace *ns = seq->private;
568
569 return m_start(seq, ppos, &ns->projid_map);
570}
571
22d917d8
EB
572static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
573{
574 (*pos)++;
575 return seq->op->start(seq, pos);
576}
577
578static void m_stop(struct seq_file *seq, void *v)
579{
580 return;
581}
582
ccf94f1b 583const struct seq_operations proc_uid_seq_operations = {
22d917d8
EB
584 .start = uid_m_start,
585 .stop = m_stop,
586 .next = m_next,
587 .show = uid_m_show,
588};
589
ccf94f1b 590const struct seq_operations proc_gid_seq_operations = {
22d917d8
EB
591 .start = gid_m_start,
592 .stop = m_stop,
593 .next = m_next,
594 .show = gid_m_show,
595};
596
ccf94f1b 597const struct seq_operations proc_projid_seq_operations = {
f76d207a
EB
598 .start = projid_m_start,
599 .stop = m_stop,
600 .next = m_next,
601 .show = projid_m_show,
602};
603
68a9a435
FF
604static bool mappings_overlap(struct uid_gid_map *new_map,
605 struct uid_gid_extent *extent)
0bd14b4f
EB
606{
607 u32 upper_first, lower_first, upper_last, lower_last;
608 unsigned idx;
609
610 upper_first = extent->first;
611 lower_first = extent->lower_first;
612 upper_last = upper_first + extent->count - 1;
613 lower_last = lower_first + extent->count - 1;
614
615 for (idx = 0; idx < new_map->nr_extents; idx++) {
616 u32 prev_upper_first, prev_lower_first;
617 u32 prev_upper_last, prev_lower_last;
618 struct uid_gid_extent *prev;
619
620 prev = &new_map->extent[idx];
621
622 prev_upper_first = prev->first;
623 prev_lower_first = prev->lower_first;
624 prev_upper_last = prev_upper_first + prev->count - 1;
625 prev_lower_last = prev_lower_first + prev->count - 1;
626
627 /* Does the upper range intersect a previous extent? */
628 if ((prev_upper_first <= upper_last) &&
629 (prev_upper_last >= upper_first))
630 return true;
631
632 /* Does the lower range intersect a previous extent? */
633 if ((prev_lower_first <= lower_last) &&
634 (prev_lower_last >= lower_first))
635 return true;
636 }
637 return false;
638}
639
22d917d8
EB
640static ssize_t map_write(struct file *file, const char __user *buf,
641 size_t count, loff_t *ppos,
642 int cap_setid,
643 struct uid_gid_map *map,
644 struct uid_gid_map *parent_map)
645{
646 struct seq_file *seq = file->private_data;
647 struct user_namespace *ns = seq->private;
648 struct uid_gid_map new_map;
649 unsigned idx;
0bd14b4f 650 struct uid_gid_extent *extent = NULL;
70f6cbb6 651 char *kbuf = NULL, *pos, *next_line;
22d917d8
EB
652 ssize_t ret = -EINVAL;
653
654 /*
f0d62aec 655 * The userns_state_mutex serializes all writes to any given map.
22d917d8
EB
656 *
657 * Any map is only ever written once.
658 *
659 * An id map fits within 1 cache line on most architectures.
660 *
661 * On read nothing needs to be done unless you are on an
662 * architecture with a crazy cache coherency model like alpha.
663 *
664 * There is a one time data dependency between reading the
665 * count of the extents and the values of the extents. The
666 * desired behavior is to see the values of the extents that
667 * were written before the count of the extents.
668 *
669 * To achieve this smp_wmb() is used on guarantee the write
e79323bd
MP
670 * order and smp_rmb() is guaranteed that we don't have crazy
671 * architectures returning stale data.
22d917d8 672 */
f0d62aec 673 mutex_lock(&userns_state_mutex);
22d917d8
EB
674
675 ret = -EPERM;
676 /* Only allow one successful write to the map */
677 if (map->nr_extents != 0)
678 goto out;
679
41c21e35
AL
680 /*
681 * Adjusting namespace settings requires capabilities on the target.
5c1469de 682 */
41c21e35 683 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
22d917d8
EB
684 goto out;
685
36476bea 686 /* Only allow < page size writes at the beginning of the file */
22d917d8
EB
687 ret = -EINVAL;
688 if ((*ppos != 0) || (count >= PAGE_SIZE))
689 goto out;
690
691 /* Slurp in the user data */
70f6cbb6
AV
692 kbuf = memdup_user_nul(buf, count);
693 if (IS_ERR(kbuf)) {
694 ret = PTR_ERR(kbuf);
695 kbuf = NULL;
22d917d8 696 goto out;
70f6cbb6 697 }
22d917d8
EB
698
699 /* Parse the user data */
700 ret = -EINVAL;
701 pos = kbuf;
702 new_map.nr_extents = 0;
68a9a435 703 for (; pos; pos = next_line) {
22d917d8
EB
704 extent = &new_map.extent[new_map.nr_extents];
705
706 /* Find the end of line and ensure I don't look past it */
707 next_line = strchr(pos, '\n');
708 if (next_line) {
709 *next_line = '\0';
710 next_line++;
711 if (*next_line == '\0')
712 next_line = NULL;
5c1469de 713 }
22d917d8
EB
714
715 pos = skip_spaces(pos);
716 extent->first = simple_strtoul(pos, &pos, 10);
717 if (!isspace(*pos))
718 goto out;
719
720 pos = skip_spaces(pos);
721 extent->lower_first = simple_strtoul(pos, &pos, 10);
722 if (!isspace(*pos))
723 goto out;
724
725 pos = skip_spaces(pos);
726 extent->count = simple_strtoul(pos, &pos, 10);
727 if (*pos && !isspace(*pos))
728 goto out;
729
730 /* Verify there is not trailing junk on the line */
731 pos = skip_spaces(pos);
732 if (*pos != '\0')
733 goto out;
734
735 /* Verify we have been given valid starting values */
736 if ((extent->first == (u32) -1) ||
68a9a435 737 (extent->lower_first == (u32) -1))
22d917d8
EB
738 goto out;
739
68a9a435
FF
740 /* Verify count is not zero and does not cause the
741 * extent to wrap
742 */
22d917d8
EB
743 if ((extent->first + extent->count) <= extent->first)
744 goto out;
68a9a435
FF
745 if ((extent->lower_first + extent->count) <=
746 extent->lower_first)
22d917d8
EB
747 goto out;
748
0bd14b4f
EB
749 /* Do the ranges in extent overlap any previous extents? */
750 if (mappings_overlap(&new_map, extent))
22d917d8
EB
751 goto out;
752
753 new_map.nr_extents++;
22d917d8
EB
754
755 /* Fail if the file contains too many extents */
756 if ((new_map.nr_extents == UID_GID_MAP_MAX_EXTENTS) &&
757 (next_line != NULL))
758 goto out;
5c1469de 759 }
22d917d8
EB
760 /* Be very certaint the new map actually exists */
761 if (new_map.nr_extents == 0)
762 goto out;
763
764 ret = -EPERM;
765 /* Validate the user is allowed to use user id's mapped to. */
6708075f 766 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
22d917d8
EB
767 goto out;
768
769 /* Map the lower ids from the parent user namespace to the
770 * kernel global id space.
771 */
772 for (idx = 0; idx < new_map.nr_extents; idx++) {
773 u32 lower_first;
774 extent = &new_map.extent[idx];
775
776 lower_first = map_id_range_down(parent_map,
777 extent->lower_first,
778 extent->count);
779
780 /* Fail if we can not map the specified extent to
781 * the kernel global id space.
782 */
783 if (lower_first == (u32) -1)
784 goto out;
785
786 extent->lower_first = lower_first;
787 }
788
789 /* Install the map */
790 memcpy(map->extent, new_map.extent,
791 new_map.nr_extents*sizeof(new_map.extent[0]));
792 smp_wmb();
793 map->nr_extents = new_map.nr_extents;
794
795 *ppos = count;
796 ret = count;
797out:
f0d62aec 798 mutex_unlock(&userns_state_mutex);
70f6cbb6 799 kfree(kbuf);
22d917d8
EB
800 return ret;
801}
802
68a9a435
FF
803ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
804 size_t size, loff_t *ppos)
22d917d8
EB
805{
806 struct seq_file *seq = file->private_data;
807 struct user_namespace *ns = seq->private;
c450f371 808 struct user_namespace *seq_ns = seq_user_ns(seq);
22d917d8
EB
809
810 if (!ns->parent)
811 return -EPERM;
812
c450f371
EB
813 if ((seq_ns != ns) && (seq_ns != ns->parent))
814 return -EPERM;
815
22d917d8
EB
816 return map_write(file, buf, size, ppos, CAP_SETUID,
817 &ns->uid_map, &ns->parent->uid_map);
818}
819
68a9a435
FF
820ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
821 size_t size, loff_t *ppos)
22d917d8
EB
822{
823 struct seq_file *seq = file->private_data;
824 struct user_namespace *ns = seq->private;
c450f371 825 struct user_namespace *seq_ns = seq_user_ns(seq);
22d917d8
EB
826
827 if (!ns->parent)
828 return -EPERM;
829
c450f371
EB
830 if ((seq_ns != ns) && (seq_ns != ns->parent))
831 return -EPERM;
832
22d917d8
EB
833 return map_write(file, buf, size, ppos, CAP_SETGID,
834 &ns->gid_map, &ns->parent->gid_map);
835}
836
68a9a435
FF
837ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
838 size_t size, loff_t *ppos)
f76d207a
EB
839{
840 struct seq_file *seq = file->private_data;
841 struct user_namespace *ns = seq->private;
842 struct user_namespace *seq_ns = seq_user_ns(seq);
843
844 if (!ns->parent)
845 return -EPERM;
846
847 if ((seq_ns != ns) && (seq_ns != ns->parent))
848 return -EPERM;
849
850 /* Anyone can set any valid project id no capability needed */
851 return map_write(file, buf, size, ppos, -1,
852 &ns->projid_map, &ns->parent->projid_map);
853}
854
68a9a435 855static bool new_idmap_permitted(const struct file *file,
6708075f 856 struct user_namespace *ns, int cap_setid,
22d917d8
EB
857 struct uid_gid_map *new_map)
858{
f95d7918 859 const struct cred *cred = file->f_cred;
0542f17b
EB
860 /* Don't allow mappings that would allow anything that wouldn't
861 * be allowed without the establishment of unprivileged mappings.
862 */
f95d7918
EB
863 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
864 uid_eq(ns->owner, cred->euid)) {
37657da3
EB
865 u32 id = new_map->extent[0].lower_first;
866 if (cap_setid == CAP_SETUID) {
867 kuid_t uid = make_kuid(ns->parent, id);
f95d7918 868 if (uid_eq(uid, cred->euid))
37657da3 869 return true;
68a9a435 870 } else if (cap_setid == CAP_SETGID) {
37657da3 871 kgid_t gid = make_kgid(ns->parent, id);
66d2f338
EB
872 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
873 gid_eq(gid, cred->egid))
37657da3
EB
874 return true;
875 }
876 }
877
f76d207a
EB
878 /* Allow anyone to set a mapping that doesn't require privilege */
879 if (!cap_valid(cap_setid))
880 return true;
881
22d917d8
EB
882 /* Allow the specified ids if we have the appropriate capability
883 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
6708075f 884 * And the opener of the id file also had the approprpiate capability.
22d917d8 885 */
6708075f
EB
886 if (ns_capable(ns->parent, cap_setid) &&
887 file_ns_capable(file, ns->parent, cap_setid))
22d917d8 888 return true;
5c1469de 889
22d917d8 890 return false;
5c1469de 891}
6164281a 892
9cc46516
EB
893int proc_setgroups_show(struct seq_file *seq, void *v)
894{
895 struct user_namespace *ns = seq->private;
896 unsigned long userns_flags = ACCESS_ONCE(ns->flags);
897
898 seq_printf(seq, "%s\n",
899 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
900 "allow" : "deny");
901 return 0;
902}
903
904ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
905 size_t count, loff_t *ppos)
906{
907 struct seq_file *seq = file->private_data;
908 struct user_namespace *ns = seq->private;
909 char kbuf[8], *pos;
910 bool setgroups_allowed;
911 ssize_t ret;
912
913 /* Only allow a very narrow range of strings to be written */
914 ret = -EINVAL;
915 if ((*ppos != 0) || (count >= sizeof(kbuf)))
916 goto out;
917
918 /* What was written? */
919 ret = -EFAULT;
920 if (copy_from_user(kbuf, buf, count))
921 goto out;
922 kbuf[count] = '\0';
923 pos = kbuf;
924
925 /* What is being requested? */
926 ret = -EINVAL;
927 if (strncmp(pos, "allow", 5) == 0) {
928 pos += 5;
929 setgroups_allowed = true;
930 }
931 else if (strncmp(pos, "deny", 4) == 0) {
932 pos += 4;
933 setgroups_allowed = false;
934 }
935 else
936 goto out;
937
938 /* Verify there is not trailing junk on the line */
939 pos = skip_spaces(pos);
940 if (*pos != '\0')
941 goto out;
942
943 ret = -EPERM;
944 mutex_lock(&userns_state_mutex);
945 if (setgroups_allowed) {
946 /* Enabling setgroups after setgroups has been disabled
947 * is not allowed.
948 */
949 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
950 goto out_unlock;
951 } else {
952 /* Permanently disabling setgroups after setgroups has
953 * been enabled by writing the gid_map is not allowed.
954 */
955 if (ns->gid_map.nr_extents != 0)
956 goto out_unlock;
957 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
958 }
959 mutex_unlock(&userns_state_mutex);
960
961 /* Report a successful write */
962 *ppos = count;
963 ret = count;
964out:
965 return ret;
966out_unlock:
967 mutex_unlock(&userns_state_mutex);
968 goto out;
969}
970
273d2c67
EB
971bool userns_may_setgroups(const struct user_namespace *ns)
972{
973 bool allowed;
974
f0d62aec 975 mutex_lock(&userns_state_mutex);
273d2c67
EB
976 /* It is not safe to use setgroups until a gid mapping in
977 * the user namespace has been established.
978 */
979 allowed = ns->gid_map.nr_extents != 0;
9cc46516
EB
980 /* Is setgroups allowed? */
981 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
f0d62aec 982 mutex_unlock(&userns_state_mutex);
273d2c67
EB
983
984 return allowed;
985}
986
d07b846f
SF
987/*
988 * Returns true if @ns is the same namespace as or a descendant of
989 * @target_ns.
990 */
991bool current_in_userns(const struct user_namespace *target_ns)
992{
993 struct user_namespace *ns;
994 for (ns = current_user_ns(); ns; ns = ns->parent) {
995 if (ns == target_ns)
996 return true;
997 }
998 return false;
999}
1000
3c041184
AV
1001static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1002{
1003 return container_of(ns, struct user_namespace, ns);
1004}
1005
64964528 1006static struct ns_common *userns_get(struct task_struct *task)
cde1975b
EB
1007{
1008 struct user_namespace *user_ns;
1009
1010 rcu_read_lock();
1011 user_ns = get_user_ns(__task_cred(task)->user_ns);
1012 rcu_read_unlock();
1013
3c041184 1014 return user_ns ? &user_ns->ns : NULL;
cde1975b
EB
1015}
1016
64964528 1017static void userns_put(struct ns_common *ns)
cde1975b 1018{
3c041184 1019 put_user_ns(to_user_ns(ns));
cde1975b
EB
1020}
1021
64964528 1022static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
cde1975b 1023{
3c041184 1024 struct user_namespace *user_ns = to_user_ns(ns);
cde1975b
EB
1025 struct cred *cred;
1026
1027 /* Don't allow gaining capabilities by reentering
1028 * the same user namespace.
1029 */
1030 if (user_ns == current_user_ns())
1031 return -EINVAL;
1032
faf00da5
EB
1033 /* Tasks that share a thread group must share a user namespace */
1034 if (!thread_group_empty(current))
cde1975b
EB
1035 return -EINVAL;
1036
e66eded8
EB
1037 if (current->fs->users != 1)
1038 return -EINVAL;
1039
cde1975b
EB
1040 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1041 return -EPERM;
1042
1043 cred = prepare_creds();
1044 if (!cred)
1045 return -ENOMEM;
1046
1047 put_user_ns(cred->user_ns);
1048 set_cred_user_ns(cred, get_user_ns(user_ns));
1049
1050 return commit_creds(cred);
1051}
1052
bcac25a5
AV
1053struct ns_common *ns_get_owner(struct ns_common *ns)
1054{
1055 struct user_namespace *my_user_ns = current_user_ns();
1056 struct user_namespace *owner, *p;
1057
1058 /* See if the owner is in the current user namespace */
1059 owner = p = ns->ops->owner(ns);
1060 for (;;) {
1061 if (!p)
1062 return ERR_PTR(-EPERM);
1063 if (p == my_user_ns)
1064 break;
1065 p = p->parent;
1066 }
1067
1068 return &get_user_ns(owner)->ns;
1069}
1070
1071static struct user_namespace *userns_owner(struct ns_common *ns)
1072{
1073 return to_user_ns(ns)->parent;
1074}
1075
cde1975b
EB
1076const struct proc_ns_operations userns_operations = {
1077 .name = "user",
1078 .type = CLONE_NEWUSER,
1079 .get = userns_get,
1080 .put = userns_put,
1081 .install = userns_install,
bcac25a5 1082 .owner = userns_owner,
a7306ed8 1083 .get_parent = ns_get_owner,
cde1975b
EB
1084};
1085
6164281a
PE
1086static __init int user_namespaces_init(void)
1087{
1088 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1089 return 0;
1090}
c96d6660 1091subsys_initcall(user_namespaces_init);