Loading...
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
8#include <linux/export.h>
9#include <linux/nsproxy.h>
10#include <linux/slab.h>
11#include <linux/user_namespace.h>
12#include <linux/proc_ns.h>
13#include <linux/highuid.h>
14#include <linux/cred.h>
15#include <linux/securebits.h>
16#include <linux/keyctl.h>
17#include <linux/key-type.h>
18#include <keys/user-type.h>
19#include <linux/seq_file.h>
20#include <linux/fs.h>
21#include <linux/uaccess.h>
22#include <linux/ctype.h>
23#include <linux/projid.h>
24#include <linux/fs_struct.h>
25
26static struct kmem_cache *user_ns_cachep __read_mostly;
27static DEFINE_MUTEX(userns_state_mutex);
28
29static bool new_idmap_permitted(const struct file *file,
30 struct user_namespace *ns, int cap_setid,
31 struct uid_gid_map *map);
32static void free_user_ns(struct work_struct *work);
33
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
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;
53 cred->cap_ambient = CAP_EMPTY_SET;
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
63/*
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.
70 */
71int create_user_ns(struct cred *new)
72{
73 struct user_namespace *ns, *parent_ns = new->user_ns;
74 kuid_t owner = new->euid;
75 kgid_t group = new->egid;
76 struct ucounts *ucounts;
77 int ret, i;
78
79 ret = -ENOSPC;
80 if (parent_ns->level > 32)
81 goto fail;
82
83 ucounts = inc_user_namespaces(parent_ns, owner);
84 if (!ucounts)
85 goto fail;
86
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 */
93 ret = -EPERM;
94 if (current_chrooted())
95 goto fail_dec;
96
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 */
101 ret = -EPERM;
102 if (!kuid_has_mapping(parent_ns, owner) ||
103 !kgid_has_mapping(parent_ns, group))
104 goto fail_dec;
105
106 ret = -ENOMEM;
107 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
108 if (!ns)
109 goto fail_dec;
110
111 ret = ns_alloc_inum(&ns->ns);
112 if (ret)
113 goto fail_free;
114 ns->ns.ops = &userns_operations;
115
116 atomic_set(&ns->count, 1);
117 /* Leave the new->user_ns reference with the new user namespace. */
118 ns->parent = parent_ns;
119 ns->level = parent_ns->level + 1;
120 ns->owner = owner;
121 ns->group = group;
122 INIT_WORK(&ns->work, free_user_ns);
123 for (i = 0; i < UCOUNT_COUNTS; i++) {
124 ns->ucount_max[i] = INT_MAX;
125 }
126 ns->ucounts = ucounts;
127
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
133#ifdef CONFIG_PERSISTENT_KEYRINGS
134 init_rwsem(&ns->persistent_keyring_register_sem);
135#endif
136 ret = -ENOMEM;
137 if (!setup_userns_sysctls(ns))
138 goto fail_keyring;
139
140 set_cred_user_ns(new, ns);
141 return 0;
142fail_keyring:
143#ifdef CONFIG_PERSISTENT_KEYRINGS
144 key_put(ns->persistent_keyring_register);
145#endif
146 ns_free_inum(&ns->ns);
147fail_free:
148 kmem_cache_free(user_ns_cachep, ns);
149fail_dec:
150 dec_user_namespaces(ucounts);
151fail:
152 return ret;
153}
154
155int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
156{
157 struct cred *cred;
158 int err = -ENOMEM;
159
160 if (!(unshare_flags & CLONE_NEWUSER))
161 return 0;
162
163 cred = prepare_creds();
164 if (cred) {
165 err = create_user_ns(cred);
166 if (err)
167 put_cred(cred);
168 else
169 *new_cred = cred;
170 }
171
172 return err;
173}
174
175static void free_user_ns(struct work_struct *work)
176{
177 struct user_namespace *parent, *ns =
178 container_of(work, struct user_namespace, work);
179
180 do {
181 struct ucounts *ucounts = ns->ucounts;
182 parent = ns->parent;
183 retire_userns_sysctls(ns);
184#ifdef CONFIG_PERSISTENT_KEYRINGS
185 key_put(ns->persistent_keyring_register);
186#endif
187 ns_free_inum(&ns->ns);
188 kmem_cache_free(user_ns_cachep, ns);
189 dec_user_namespaces(ucounts);
190 ns = parent;
191 } while (atomic_dec_and_test(&parent->count));
192}
193
194void __put_user_ns(struct user_namespace *ns)
195{
196 schedule_work(&ns->work);
197}
198EXPORT_SYMBOL(__put_user_ns);
199
200static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
201{
202 unsigned idx, extents;
203 u32 first, last, id2;
204
205 id2 = id + count - 1;
206
207 /* Find the matching extent */
208 extents = map->nr_extents;
209 smp_rmb();
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;
233 smp_rmb();
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;
256 smp_rmb();
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;
262 }
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
282 * for and handle INVALID_UID being returned. INVALID_UID
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
343 * @gid: group identifier
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
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
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;
482
483 lower_ns = seq_user_ns(seq);
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;
495}
496
497static int gid_m_show(struct seq_file *seq, void *v)
498{
499 struct user_namespace *ns = seq->private;
500 struct uid_gid_extent *extent = v;
501 struct user_namespace *lower_ns;
502 gid_t lower;
503
504 lower_ns = seq_user_ns(seq);
505 if ((lower_ns == ns) && lower_ns->parent)
506 lower_ns = lower_ns->parent;
507
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
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
539static void *m_start(struct seq_file *seq, loff_t *ppos,
540 struct uid_gid_map *map)
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
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
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
583const struct seq_operations proc_uid_seq_operations = {
584 .start = uid_m_start,
585 .stop = m_stop,
586 .next = m_next,
587 .show = uid_m_show,
588};
589
590const struct seq_operations proc_gid_seq_operations = {
591 .start = gid_m_start,
592 .stop = m_stop,
593 .next = m_next,
594 .show = gid_m_show,
595};
596
597const struct seq_operations proc_projid_seq_operations = {
598 .start = projid_m_start,
599 .stop = m_stop,
600 .next = m_next,
601 .show = projid_m_show,
602};
603
604static bool mappings_overlap(struct uid_gid_map *new_map,
605 struct uid_gid_extent *extent)
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
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;
650 struct uid_gid_extent *extent = NULL;
651 char *kbuf = NULL, *pos, *next_line;
652 ssize_t ret = -EINVAL;
653
654 /*
655 * The userns_state_mutex serializes all writes to any given map.
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
670 * order and smp_rmb() is guaranteed that we don't have crazy
671 * architectures returning stale data.
672 */
673 mutex_lock(&userns_state_mutex);
674
675 ret = -EPERM;
676 /* Only allow one successful write to the map */
677 if (map->nr_extents != 0)
678 goto out;
679
680 /*
681 * Adjusting namespace settings requires capabilities on the target.
682 */
683 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
684 goto out;
685
686 /* Only allow < page size writes at the beginning of the file */
687 ret = -EINVAL;
688 if ((*ppos != 0) || (count >= PAGE_SIZE))
689 goto out;
690
691 /* Slurp in the user data */
692 kbuf = memdup_user_nul(buf, count);
693 if (IS_ERR(kbuf)) {
694 ret = PTR_ERR(kbuf);
695 kbuf = NULL;
696 goto out;
697 }
698
699 /* Parse the user data */
700 ret = -EINVAL;
701 pos = kbuf;
702 new_map.nr_extents = 0;
703 for (; pos; pos = next_line) {
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;
713 }
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) ||
737 (extent->lower_first == (u32) -1))
738 goto out;
739
740 /* Verify count is not zero and does not cause the
741 * extent to wrap
742 */
743 if ((extent->first + extent->count) <= extent->first)
744 goto out;
745 if ((extent->lower_first + extent->count) <=
746 extent->lower_first)
747 goto out;
748
749 /* Do the ranges in extent overlap any previous extents? */
750 if (mappings_overlap(&new_map, extent))
751 goto out;
752
753 new_map.nr_extents++;
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;
759 }
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. */
766 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
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:
798 mutex_unlock(&userns_state_mutex);
799 kfree(kbuf);
800 return ret;
801}
802
803ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
804 size_t size, loff_t *ppos)
805{
806 struct seq_file *seq = file->private_data;
807 struct user_namespace *ns = seq->private;
808 struct user_namespace *seq_ns = seq_user_ns(seq);
809
810 if (!ns->parent)
811 return -EPERM;
812
813 if ((seq_ns != ns) && (seq_ns != ns->parent))
814 return -EPERM;
815
816 return map_write(file, buf, size, ppos, CAP_SETUID,
817 &ns->uid_map, &ns->parent->uid_map);
818}
819
820ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
821 size_t size, loff_t *ppos)
822{
823 struct seq_file *seq = file->private_data;
824 struct user_namespace *ns = seq->private;
825 struct user_namespace *seq_ns = seq_user_ns(seq);
826
827 if (!ns->parent)
828 return -EPERM;
829
830 if ((seq_ns != ns) && (seq_ns != ns->parent))
831 return -EPERM;
832
833 return map_write(file, buf, size, ppos, CAP_SETGID,
834 &ns->gid_map, &ns->parent->gid_map);
835}
836
837ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
838 size_t size, loff_t *ppos)
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
855static bool new_idmap_permitted(const struct file *file,
856 struct user_namespace *ns, int cap_setid,
857 struct uid_gid_map *new_map)
858{
859 const struct cred *cred = file->f_cred;
860 /* Don't allow mappings that would allow anything that wouldn't
861 * be allowed without the establishment of unprivileged mappings.
862 */
863 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
864 uid_eq(ns->owner, cred->euid)) {
865 u32 id = new_map->extent[0].lower_first;
866 if (cap_setid == CAP_SETUID) {
867 kuid_t uid = make_kuid(ns->parent, id);
868 if (uid_eq(uid, cred->euid))
869 return true;
870 } else if (cap_setid == CAP_SETGID) {
871 kgid_t gid = make_kgid(ns->parent, id);
872 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
873 gid_eq(gid, cred->egid))
874 return true;
875 }
876 }
877
878 /* Allow anyone to set a mapping that doesn't require privilege */
879 if (!cap_valid(cap_setid))
880 return true;
881
882 /* Allow the specified ids if we have the appropriate capability
883 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
884 * And the opener of the id file also had the approprpiate capability.
885 */
886 if (ns_capable(ns->parent, cap_setid) &&
887 file_ns_capable(file, ns->parent, cap_setid))
888 return true;
889
890 return false;
891}
892
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
971bool userns_may_setgroups(const struct user_namespace *ns)
972{
973 bool allowed;
974
975 mutex_lock(&userns_state_mutex);
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;
980 /* Is setgroups allowed? */
981 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
982 mutex_unlock(&userns_state_mutex);
983
984 return allowed;
985}
986
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
1001static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1002{
1003 return container_of(ns, struct user_namespace, ns);
1004}
1005
1006static struct ns_common *userns_get(struct task_struct *task)
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
1014 return user_ns ? &user_ns->ns : NULL;
1015}
1016
1017static void userns_put(struct ns_common *ns)
1018{
1019 put_user_ns(to_user_ns(ns));
1020}
1021
1022static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1023{
1024 struct user_namespace *user_ns = to_user_ns(ns);
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
1033 /* Tasks that share a thread group must share a user namespace */
1034 if (!thread_group_empty(current))
1035 return -EINVAL;
1036
1037 if (current->fs->users != 1)
1038 return -EINVAL;
1039
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
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
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,
1082 .owner = userns_owner,
1083 .get_parent = ns_get_owner,
1084};
1085
1086static __init int user_namespaces_init(void)
1087{
1088 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1089 return 0;
1090}
1091subsys_initcall(user_namespaces_init);
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
8#include <linux/export.h>
9#include <linux/nsproxy.h>
10#include <linux/slab.h>
11#include <linux/sched/signal.h>
12#include <linux/user_namespace.h>
13#include <linux/proc_ns.h>
14#include <linux/highuid.h>
15#include <linux/cred.h>
16#include <linux/securebits.h>
17#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>
24#include <linux/projid.h>
25#include <linux/fs_struct.h>
26#include <linux/bsearch.h>
27#include <linux/sort.h>
28
29static struct kmem_cache *user_ns_cachep __read_mostly;
30static DEFINE_MUTEX(userns_state_mutex);
31
32static bool new_idmap_permitted(const struct file *file,
33 struct user_namespace *ns, int cap_setid,
34 struct uid_gid_map *map);
35static void free_user_ns(struct work_struct *work);
36
37static 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
47static 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;
56 cred->cap_ambient = CAP_EMPTY_SET;
57 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
66/*
67 * 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.
73 */
74int create_user_ns(struct cred *new)
75{
76 struct user_namespace *ns, *parent_ns = new->user_ns;
77 kuid_t owner = new->euid;
78 kgid_t group = new->egid;
79 struct ucounts *ucounts;
80 int ret, i;
81
82 ret = -ENOSPC;
83 if (parent_ns->level > 32)
84 goto fail;
85
86 ucounts = inc_user_namespaces(parent_ns, owner);
87 if (!ucounts)
88 goto fail;
89
90 /*
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 */
96 ret = -EPERM;
97 if (current_chrooted())
98 goto fail_dec;
99
100 /* 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 */
104 ret = -EPERM;
105 if (!kuid_has_mapping(parent_ns, owner) ||
106 !kgid_has_mapping(parent_ns, group))
107 goto fail_dec;
108
109 ret = -ENOMEM;
110 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
111 if (!ns)
112 goto fail_dec;
113
114 ret = ns_alloc_inum(&ns->ns);
115 if (ret)
116 goto fail_free;
117 ns->ns.ops = &userns_operations;
118
119 atomic_set(&ns->count, 1);
120 /* Leave the new->user_ns reference with the new user namespace. */
121 ns->parent = parent_ns;
122 ns->level = parent_ns->level + 1;
123 ns->owner = owner;
124 ns->group = group;
125 INIT_WORK(&ns->work, free_user_ns);
126 for (i = 0; i < UCOUNT_COUNTS; i++) {
127 ns->ucount_max[i] = INT_MAX;
128 }
129 ns->ucounts = ucounts;
130
131 /* 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
136#ifdef CONFIG_PERSISTENT_KEYRINGS
137 init_rwsem(&ns->persistent_keyring_register_sem);
138#endif
139 ret = -ENOMEM;
140 if (!setup_userns_sysctls(ns))
141 goto fail_keyring;
142
143 set_cred_user_ns(new, ns);
144 return 0;
145fail_keyring:
146#ifdef CONFIG_PERSISTENT_KEYRINGS
147 key_put(ns->persistent_keyring_register);
148#endif
149 ns_free_inum(&ns->ns);
150fail_free:
151 kmem_cache_free(user_ns_cachep, ns);
152fail_dec:
153 dec_user_namespaces(ucounts);
154fail:
155 return ret;
156}
157
158int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
159{
160 struct cred *cred;
161 int err = -ENOMEM;
162
163 if (!(unshare_flags & CLONE_NEWUSER))
164 return 0;
165
166 cred = prepare_creds();
167 if (cred) {
168 err = create_user_ns(cred);
169 if (err)
170 put_cred(cred);
171 else
172 *new_cred = cred;
173 }
174
175 return err;
176}
177
178static void free_user_ns(struct work_struct *work)
179{
180 struct user_namespace *parent, *ns =
181 container_of(work, struct user_namespace, work);
182
183 do {
184 struct ucounts *ucounts = ns->ucounts;
185 parent = ns->parent;
186 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 }
198 retire_userns_sysctls(ns);
199#ifdef CONFIG_PERSISTENT_KEYRINGS
200 key_put(ns->persistent_keyring_register);
201#endif
202 ns_free_inum(&ns->ns);
203 kmem_cache_free(user_ns_cachep, ns);
204 dec_user_namespaces(ucounts);
205 ns = parent;
206 } while (atomic_dec_and_test(&parent->count));
207}
208
209void __put_user_ns(struct user_namespace *ns)
210{
211 schedule_work(&ns->work);
212}
213EXPORT_SYMBOL(__put_user_ns);
214
215/**
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)
230{
231 u32 first, last, id2;
232 const struct idmap_key *key = k;
233 const struct uid_gid_extent *el = e;
234
235 id2 = key->id + key->count - 1;
236
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 */
259static struct uid_gid_extent *
260map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
261{
262 struct idmap_key key;
263
264 key.map_up = false;
265 key.count = count;
266 key.id = id;
267
268 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;
281 u32 first, last, id2;
282
283 id2 = id + count - 1;
284
285 /* Find the matching extent */
286 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))
291 return &map->extent[idx];
292 }
293 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;
300 smp_rmb();
301
302 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
307 /* Map the id or note failure */
308 if (extent)
309 id = (id - extent->first) + extent->lower_first;
310 else
311 id = (u32) -1;
312
313 return id;
314}
315
316static u32 map_id_down(struct uid_gid_map *map, u32 id)
317{
318 return map_id_range_down(map, id, 1);
319}
320
321/**
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 */
326static struct uid_gid_extent *
327map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
328{
329 unsigned idx;
330 u32 first, last;
331
332 /* Find the matching extent */
333 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)
337 return &map->extent[idx];
338 }
339 return NULL;
340}
341
342/**
343 * 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 */
346static struct uid_gid_extent *
347map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
348{
349 struct idmap_key key;
350
351 key.map_up = true;
352 key.count = 1;
353 key.id = id;
354
355 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;
363 smp_rmb();
364
365 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
370 /* Map the id or note failure */
371 if (extent)
372 id = (id - extent->lower_first) + extent->first;
373 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
389 * for and handle INVALID_UID being returned. INVALID_UID
390 * 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
450 * @gid: group identifier
451 *
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
514/**
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
583static 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
590 lower_ns = seq_user_ns(seq);
591 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
611 lower_ns = seq_user_ns(seq);
612 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
625static 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
646static void *m_start(struct seq_file *seq, loff_t *ppos,
647 struct uid_gid_map *map)
648{
649 loff_t pos = *ppos;
650 unsigned extents = map->nr_extents;
651 smp_rmb();
652
653 if (pos >= extents)
654 return NULL;
655
656 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
657 return &map->extent[pos];
658
659 return &map->forward[pos];
660}
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
676static 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
683static 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
694const struct seq_operations proc_uid_seq_operations = {
695 .start = uid_m_start,
696 .stop = m_stop,
697 .next = m_next,
698 .show = uid_m_show,
699};
700
701const struct seq_operations proc_gid_seq_operations = {
702 .start = gid_m_start,
703 .stop = m_stop,
704 .next = m_next,
705 .show = gid_m_show,
706};
707
708const struct seq_operations proc_projid_seq_operations = {
709 .start = projid_m_start,
710 .stop = m_stop,
711 .next = m_next,
712 .show = projid_m_show,
713};
714
715static bool mappings_overlap(struct uid_gid_map *new_map,
716 struct uid_gid_extent *extent)
717{
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
731 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];
735
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
754/**
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{
761 struct uid_gid_extent *dest;
762
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
782 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++;
789 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
849static 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;
859 struct uid_gid_extent extent;
860 char *kbuf = NULL, *pos, *next_line;
861 ssize_t ret = -EINVAL;
862
863 /*
864 * The userns_state_mutex serializes all writes to any given map.
865 *
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
879 * order and smp_rmb() is guaranteed that we don't have crazy
880 * architectures returning stale data.
881 */
882 mutex_lock(&userns_state_mutex);
883
884 memset(&new_map, 0, sizeof(struct uid_gid_map));
885
886 ret = -EPERM;
887 /* Only allow one successful write to the map */
888 if (map->nr_extents != 0)
889 goto out;
890
891 /*
892 * Adjusting namespace settings requires capabilities on the target.
893 */
894 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
895 goto out;
896
897 /* Only allow < page size writes at the beginning of the file */
898 ret = -EINVAL;
899 if ((*ppos != 0) || (count >= PAGE_SIZE))
900 goto out;
901
902 /* Slurp in the user data */
903 kbuf = memdup_user_nul(buf, count);
904 if (IS_ERR(kbuf)) {
905 ret = PTR_ERR(kbuf);
906 kbuf = NULL;
907 goto out;
908 }
909
910 /* Parse the user data */
911 ret = -EINVAL;
912 pos = kbuf;
913 for (; pos; pos = next_line) {
914
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;
922 }
923
924 pos = skip_spaces(pos);
925 extent.first = simple_strtoul(pos, &pos, 10);
926 if (!isspace(*pos))
927 goto out;
928
929 pos = skip_spaces(pos);
930 extent.lower_first = simple_strtoul(pos, &pos, 10);
931 if (!isspace(*pos))
932 goto out;
933
934 pos = skip_spaces(pos);
935 extent.count = simple_strtoul(pos, &pos, 10);
936 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 */
945 if ((extent.first == (u32) -1) ||
946 (extent.lower_first == (u32) -1))
947 goto out;
948
949 /* Verify count is not zero and does not cause the
950 * extent to wrap
951 */
952 if ((extent.first + extent.count) <= extent.first)
953 goto out;
954 if ((extent.lower_first + extent.count) <=
955 extent.lower_first)
956 goto out;
957
958 /* Do the ranges in extent overlap any previous extents? */
959 if (mappings_overlap(&new_map, &extent))
960 goto out;
961
962 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
963 (next_line != NULL))
964 goto out;
965
966 ret = insert_extent(&new_map, &extent);
967 if (ret < 0)
968 goto out;
969 ret = -EINVAL;
970 }
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. */
977 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
978 goto out;
979
980 ret = sort_idmaps(&new_map);
981 if (ret < 0)
982 goto out;
983
984 ret = -EPERM;
985 /* 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++) {
989 struct uid_gid_extent *e;
990 u32 lower_first;
991
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];
996
997 lower_first = map_id_range_down(parent_map,
998 e->lower_first,
999 e->count);
1000
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
1007 e->lower_first = lower_first;
1008 }
1009
1010 /* Install the map */
1011 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 }
1018 smp_wmb();
1019 map->nr_extents = new_map.nr_extents;
1020
1021 *ppos = count;
1022 ret = count;
1023out:
1024 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
1032 mutex_unlock(&userns_state_mutex);
1033 kfree(kbuf);
1034 return ret;
1035}
1036
1037ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1038 size_t size, loff_t *ppos)
1039{
1040 struct seq_file *seq = file->private_data;
1041 struct user_namespace *ns = seq->private;
1042 struct user_namespace *seq_ns = seq_user_ns(seq);
1043
1044 if (!ns->parent)
1045 return -EPERM;
1046
1047 if ((seq_ns != ns) && (seq_ns != ns->parent))
1048 return -EPERM;
1049
1050 return map_write(file, buf, size, ppos, CAP_SETUID,
1051 &ns->uid_map, &ns->parent->uid_map);
1052}
1053
1054ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1055 size_t size, loff_t *ppos)
1056{
1057 struct seq_file *seq = file->private_data;
1058 struct user_namespace *ns = seq->private;
1059 struct user_namespace *seq_ns = seq_user_ns(seq);
1060
1061 if (!ns->parent)
1062 return -EPERM;
1063
1064 if ((seq_ns != ns) && (seq_ns != ns->parent))
1065 return -EPERM;
1066
1067 return map_write(file, buf, size, ppos, CAP_SETGID,
1068 &ns->gid_map, &ns->parent->gid_map);
1069}
1070
1071ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1072 size_t size, loff_t *ppos)
1073{
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
1089static bool new_idmap_permitted(const struct file *file,
1090 struct user_namespace *ns, int cap_setid,
1091 struct uid_gid_map *new_map)
1092{
1093 const struct cred *cred = file->f_cred;
1094 /* Don't allow mappings that would allow anything that wouldn't
1095 * be allowed without the establishment of unprivileged mappings.
1096 */
1097 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1098 uid_eq(ns->owner, cred->euid)) {
1099 u32 id = new_map->extent[0].lower_first;
1100 if (cap_setid == CAP_SETUID) {
1101 kuid_t uid = make_kuid(ns->parent, id);
1102 if (uid_eq(uid, cred->euid))
1103 return true;
1104 } else if (cap_setid == CAP_SETGID) {
1105 kgid_t gid = make_kgid(ns->parent, id);
1106 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1107 gid_eq(gid, cred->egid))
1108 return true;
1109 }
1110 }
1111
1112 /* Allow anyone to set a mapping that doesn't require privilege */
1113 if (!cap_valid(cap_setid))
1114 return true;
1115
1116 /* Allow the specified ids if we have the appropriate capability
1117 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1118 * And the opener of the id file also had the approprpiate capability.
1119 */
1120 if (ns_capable(ns->parent, cap_setid) &&
1121 file_ns_capable(file, ns->parent, cap_setid))
1122 return true;
1123
1124 return false;
1125}
1126
1127int proc_setgroups_show(struct seq_file *seq, void *v)
1128{
1129 struct user_namespace *ns = seq->private;
1130 unsigned long userns_flags = READ_ONCE(ns->flags);
1131
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
1205bool userns_may_setgroups(const struct user_namespace *ns)
1206{
1207 bool allowed;
1208
1209 mutex_lock(&userns_state_mutex);
1210 /* 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;
1214 /* Is setgroups allowed? */
1215 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1216 mutex_unlock(&userns_state_mutex);
1217
1218 return allowed;
1219}
1220
1221/*
1222 * Returns true if @child is the same namespace or a descendant of
1223 * @ancestor.
1224 */
1225bool 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
1234bool current_in_userns(const struct user_namespace *target_ns)
1235{
1236 return in_userns(target_ns, current_user_ns());
1237}
1238
1239static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1240{
1241 return container_of(ns, struct user_namespace, ns);
1242}
1243
1244static struct ns_common *userns_get(struct task_struct *task)
1245{
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
1252 return user_ns ? &user_ns->ns : NULL;
1253}
1254
1255static void userns_put(struct ns_common *ns)
1256{
1257 put_user_ns(to_user_ns(ns));
1258}
1259
1260static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1261{
1262 struct user_namespace *user_ns = to_user_ns(ns);
1263 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
1271 /* Tasks that share a thread group must share a user namespace */
1272 if (!thread_group_empty(current))
1273 return -EINVAL;
1274
1275 if (current->fs->users != 1)
1276 return -EINVAL;
1277
1278 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
1291struct 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
1314const 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,
1320 .owner = userns_owner,
1321 .get_parent = ns_get_owner,
1322};
1323
1324static __init int user_namespaces_init(void)
1325{
1326 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1327 return 0;
1328}
1329subsys_initcall(user_namespaces_init);