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_array(UID_GID_MAP_MAX_EXTENTS,
768 sizeof(struct uid_gid_extent),
769 GFP_KERNEL);
770 if (!forward)
771 return -ENOMEM;
772
773 /* Copy over memory. Only set up memory for the forward pointer.
774 * Defer the memory setup for the reverse pointer.
775 */
776 memcpy(forward, map->extent,
777 map->nr_extents * sizeof(map->extent[0]));
778
779 map->forward = forward;
780 map->reverse = NULL;
781 }
782
783 if (map->nr_extents < UID_GID_MAP_MAX_BASE_EXTENTS)
784 dest = &map->extent[map->nr_extents];
785 else
786 dest = &map->forward[map->nr_extents];
787
788 *dest = *extent;
789 map->nr_extents++;
790 return 0;
791}
792
793/* cmp function to sort() forward mappings */
794static int cmp_extents_forward(const void *a, const void *b)
795{
796 const struct uid_gid_extent *e1 = a;
797 const struct uid_gid_extent *e2 = b;
798
799 if (e1->first < e2->first)
800 return -1;
801
802 if (e1->first > e2->first)
803 return 1;
804
805 return 0;
806}
807
808/* cmp function to sort() reverse mappings */
809static int cmp_extents_reverse(const void *a, const void *b)
810{
811 const struct uid_gid_extent *e1 = a;
812 const struct uid_gid_extent *e2 = b;
813
814 if (e1->lower_first < e2->lower_first)
815 return -1;
816
817 if (e1->lower_first > e2->lower_first)
818 return 1;
819
820 return 0;
821}
822
823/**
824 * sort_idmaps - Sorts an array of idmap entries.
825 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
826 */
827static int sort_idmaps(struct uid_gid_map *map)
828{
829 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
830 return 0;
831
832 /* Sort forward array. */
833 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
834 cmp_extents_forward, NULL);
835
836 /* Only copy the memory from forward we actually need. */
837 map->reverse = kmemdup(map->forward,
838 map->nr_extents * sizeof(struct uid_gid_extent),
839 GFP_KERNEL);
840 if (!map->reverse)
841 return -ENOMEM;
842
843 /* Sort reverse array. */
844 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
845 cmp_extents_reverse, NULL);
846
847 return 0;
848}
849
850static ssize_t map_write(struct file *file, const char __user *buf,
851 size_t count, loff_t *ppos,
852 int cap_setid,
853 struct uid_gid_map *map,
854 struct uid_gid_map *parent_map)
855{
856 struct seq_file *seq = file->private_data;
857 struct user_namespace *ns = seq->private;
858 struct uid_gid_map new_map;
859 unsigned idx;
860 struct uid_gid_extent extent;
861 char *kbuf = NULL, *pos, *next_line;
862 ssize_t ret;
863
864 /* Only allow < page size writes at the beginning of the file */
865 if ((*ppos != 0) || (count >= PAGE_SIZE))
866 return -EINVAL;
867
868 /* Slurp in the user data */
869 kbuf = memdup_user_nul(buf, count);
870 if (IS_ERR(kbuf))
871 return PTR_ERR(kbuf);
872
873 /*
874 * The userns_state_mutex serializes all writes to any given map.
875 *
876 * Any map is only ever written once.
877 *
878 * An id map fits within 1 cache line on most architectures.
879 *
880 * On read nothing needs to be done unless you are on an
881 * architecture with a crazy cache coherency model like alpha.
882 *
883 * There is a one time data dependency between reading the
884 * count of the extents and the values of the extents. The
885 * desired behavior is to see the values of the extents that
886 * were written before the count of the extents.
887 *
888 * To achieve this smp_wmb() is used on guarantee the write
889 * order and smp_rmb() is guaranteed that we don't have crazy
890 * architectures returning stale data.
891 */
892 mutex_lock(&userns_state_mutex);
893
894 memset(&new_map, 0, sizeof(struct uid_gid_map));
895
896 ret = -EPERM;
897 /* Only allow one successful write to the map */
898 if (map->nr_extents != 0)
899 goto out;
900
901 /*
902 * Adjusting namespace settings requires capabilities on the target.
903 */
904 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
905 goto out;
906
907 /* Parse the user data */
908 ret = -EINVAL;
909 pos = kbuf;
910 for (; pos; pos = next_line) {
911
912 /* Find the end of line and ensure I don't look past it */
913 next_line = strchr(pos, '\n');
914 if (next_line) {
915 *next_line = '\0';
916 next_line++;
917 if (*next_line == '\0')
918 next_line = NULL;
919 }
920
921 pos = skip_spaces(pos);
922 extent.first = simple_strtoul(pos, &pos, 10);
923 if (!isspace(*pos))
924 goto out;
925
926 pos = skip_spaces(pos);
927 extent.lower_first = simple_strtoul(pos, &pos, 10);
928 if (!isspace(*pos))
929 goto out;
930
931 pos = skip_spaces(pos);
932 extent.count = simple_strtoul(pos, &pos, 10);
933 if (*pos && !isspace(*pos))
934 goto out;
935
936 /* Verify there is not trailing junk on the line */
937 pos = skip_spaces(pos);
938 if (*pos != '\0')
939 goto out;
940
941 /* Verify we have been given valid starting values */
942 if ((extent.first == (u32) -1) ||
943 (extent.lower_first == (u32) -1))
944 goto out;
945
946 /* Verify count is not zero and does not cause the
947 * extent to wrap
948 */
949 if ((extent.first + extent.count) <= extent.first)
950 goto out;
951 if ((extent.lower_first + extent.count) <=
952 extent.lower_first)
953 goto out;
954
955 /* Do the ranges in extent overlap any previous extents? */
956 if (mappings_overlap(&new_map, &extent))
957 goto out;
958
959 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
960 (next_line != NULL))
961 goto out;
962
963 ret = insert_extent(&new_map, &extent);
964 if (ret < 0)
965 goto out;
966 ret = -EINVAL;
967 }
968 /* Be very certaint the new map actually exists */
969 if (new_map.nr_extents == 0)
970 goto out;
971
972 ret = -EPERM;
973 /* Validate the user is allowed to use user id's mapped to. */
974 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
975 goto out;
976
977 ret = -EPERM;
978 /* Map the lower ids from the parent user namespace to the
979 * kernel global id space.
980 */
981 for (idx = 0; idx < new_map.nr_extents; idx++) {
982 struct uid_gid_extent *e;
983 u32 lower_first;
984
985 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
986 e = &new_map.extent[idx];
987 else
988 e = &new_map.forward[idx];
989
990 lower_first = map_id_range_down(parent_map,
991 e->lower_first,
992 e->count);
993
994 /* Fail if we can not map the specified extent to
995 * the kernel global id space.
996 */
997 if (lower_first == (u32) -1)
998 goto out;
999
1000 e->lower_first = lower_first;
1001 }
1002
1003 /*
1004 * If we want to use binary search for lookup, this clones the extent
1005 * array and sorts both copies.
1006 */
1007 ret = sort_idmaps(&new_map);
1008 if (ret < 0)
1009 goto out;
1010
1011 /* Install the map */
1012 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1013 memcpy(map->extent, new_map.extent,
1014 new_map.nr_extents * sizeof(new_map.extent[0]));
1015 } else {
1016 map->forward = new_map.forward;
1017 map->reverse = new_map.reverse;
1018 }
1019 smp_wmb();
1020 map->nr_extents = new_map.nr_extents;
1021
1022 *ppos = count;
1023 ret = count;
1024out:
1025 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1026 kfree(new_map.forward);
1027 kfree(new_map.reverse);
1028 map->forward = NULL;
1029 map->reverse = NULL;
1030 map->nr_extents = 0;
1031 }
1032
1033 mutex_unlock(&userns_state_mutex);
1034 kfree(kbuf);
1035 return ret;
1036}
1037
1038ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1039 size_t size, loff_t *ppos)
1040{
1041 struct seq_file *seq = file->private_data;
1042 struct user_namespace *ns = seq->private;
1043 struct user_namespace *seq_ns = seq_user_ns(seq);
1044
1045 if (!ns->parent)
1046 return -EPERM;
1047
1048 if ((seq_ns != ns) && (seq_ns != ns->parent))
1049 return -EPERM;
1050
1051 return map_write(file, buf, size, ppos, CAP_SETUID,
1052 &ns->uid_map, &ns->parent->uid_map);
1053}
1054
1055ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1056 size_t size, loff_t *ppos)
1057{
1058 struct seq_file *seq = file->private_data;
1059 struct user_namespace *ns = seq->private;
1060 struct user_namespace *seq_ns = seq_user_ns(seq);
1061
1062 if (!ns->parent)
1063 return -EPERM;
1064
1065 if ((seq_ns != ns) && (seq_ns != ns->parent))
1066 return -EPERM;
1067
1068 return map_write(file, buf, size, ppos, CAP_SETGID,
1069 &ns->gid_map, &ns->parent->gid_map);
1070}
1071
1072ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1073 size_t size, loff_t *ppos)
1074{
1075 struct seq_file *seq = file->private_data;
1076 struct user_namespace *ns = seq->private;
1077 struct user_namespace *seq_ns = seq_user_ns(seq);
1078
1079 if (!ns->parent)
1080 return -EPERM;
1081
1082 if ((seq_ns != ns) && (seq_ns != ns->parent))
1083 return -EPERM;
1084
1085 /* Anyone can set any valid project id no capability needed */
1086 return map_write(file, buf, size, ppos, -1,
1087 &ns->projid_map, &ns->parent->projid_map);
1088}
1089
1090static bool new_idmap_permitted(const struct file *file,
1091 struct user_namespace *ns, int cap_setid,
1092 struct uid_gid_map *new_map)
1093{
1094 const struct cred *cred = file->f_cred;
1095 /* Don't allow mappings that would allow anything that wouldn't
1096 * be allowed without the establishment of unprivileged mappings.
1097 */
1098 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1099 uid_eq(ns->owner, cred->euid)) {
1100 u32 id = new_map->extent[0].lower_first;
1101 if (cap_setid == CAP_SETUID) {
1102 kuid_t uid = make_kuid(ns->parent, id);
1103 if (uid_eq(uid, cred->euid))
1104 return true;
1105 } else if (cap_setid == CAP_SETGID) {
1106 kgid_t gid = make_kgid(ns->parent, id);
1107 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1108 gid_eq(gid, cred->egid))
1109 return true;
1110 }
1111 }
1112
1113 /* Allow anyone to set a mapping that doesn't require privilege */
1114 if (!cap_valid(cap_setid))
1115 return true;
1116
1117 /* Allow the specified ids if we have the appropriate capability
1118 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1119 * And the opener of the id file also had the approprpiate capability.
1120 */
1121 if (ns_capable(ns->parent, cap_setid) &&
1122 file_ns_capable(file, ns->parent, cap_setid))
1123 return true;
1124
1125 return false;
1126}
1127
1128int proc_setgroups_show(struct seq_file *seq, void *v)
1129{
1130 struct user_namespace *ns = seq->private;
1131 unsigned long userns_flags = READ_ONCE(ns->flags);
1132
1133 seq_printf(seq, "%s\n",
1134 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1135 "allow" : "deny");
1136 return 0;
1137}
1138
1139ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1140 size_t count, loff_t *ppos)
1141{
1142 struct seq_file *seq = file->private_data;
1143 struct user_namespace *ns = seq->private;
1144 char kbuf[8], *pos;
1145 bool setgroups_allowed;
1146 ssize_t ret;
1147
1148 /* Only allow a very narrow range of strings to be written */
1149 ret = -EINVAL;
1150 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1151 goto out;
1152
1153 /* What was written? */
1154 ret = -EFAULT;
1155 if (copy_from_user(kbuf, buf, count))
1156 goto out;
1157 kbuf[count] = '\0';
1158 pos = kbuf;
1159
1160 /* What is being requested? */
1161 ret = -EINVAL;
1162 if (strncmp(pos, "allow", 5) == 0) {
1163 pos += 5;
1164 setgroups_allowed = true;
1165 }
1166 else if (strncmp(pos, "deny", 4) == 0) {
1167 pos += 4;
1168 setgroups_allowed = false;
1169 }
1170 else
1171 goto out;
1172
1173 /* Verify there is not trailing junk on the line */
1174 pos = skip_spaces(pos);
1175 if (*pos != '\0')
1176 goto out;
1177
1178 ret = -EPERM;
1179 mutex_lock(&userns_state_mutex);
1180 if (setgroups_allowed) {
1181 /* Enabling setgroups after setgroups has been disabled
1182 * is not allowed.
1183 */
1184 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1185 goto out_unlock;
1186 } else {
1187 /* Permanently disabling setgroups after setgroups has
1188 * been enabled by writing the gid_map is not allowed.
1189 */
1190 if (ns->gid_map.nr_extents != 0)
1191 goto out_unlock;
1192 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1193 }
1194 mutex_unlock(&userns_state_mutex);
1195
1196 /* Report a successful write */
1197 *ppos = count;
1198 ret = count;
1199out:
1200 return ret;
1201out_unlock:
1202 mutex_unlock(&userns_state_mutex);
1203 goto out;
1204}
1205
1206bool userns_may_setgroups(const struct user_namespace *ns)
1207{
1208 bool allowed;
1209
1210 mutex_lock(&userns_state_mutex);
1211 /* It is not safe to use setgroups until a gid mapping in
1212 * the user namespace has been established.
1213 */
1214 allowed = ns->gid_map.nr_extents != 0;
1215 /* Is setgroups allowed? */
1216 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1217 mutex_unlock(&userns_state_mutex);
1218
1219 return allowed;
1220}
1221
1222/*
1223 * Returns true if @child is the same namespace or a descendant of
1224 * @ancestor.
1225 */
1226bool in_userns(const struct user_namespace *ancestor,
1227 const struct user_namespace *child)
1228{
1229 const struct user_namespace *ns;
1230 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1231 ;
1232 return (ns == ancestor);
1233}
1234
1235bool current_in_userns(const struct user_namespace *target_ns)
1236{
1237 return in_userns(target_ns, current_user_ns());
1238}
1239EXPORT_SYMBOL(current_in_userns);
1240
1241static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1242{
1243 return container_of(ns, struct user_namespace, ns);
1244}
1245
1246static struct ns_common *userns_get(struct task_struct *task)
1247{
1248 struct user_namespace *user_ns;
1249
1250 rcu_read_lock();
1251 user_ns = get_user_ns(__task_cred(task)->user_ns);
1252 rcu_read_unlock();
1253
1254 return user_ns ? &user_ns->ns : NULL;
1255}
1256
1257static void userns_put(struct ns_common *ns)
1258{
1259 put_user_ns(to_user_ns(ns));
1260}
1261
1262static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1263{
1264 struct user_namespace *user_ns = to_user_ns(ns);
1265 struct cred *cred;
1266
1267 /* Don't allow gaining capabilities by reentering
1268 * the same user namespace.
1269 */
1270 if (user_ns == current_user_ns())
1271 return -EINVAL;
1272
1273 /* Tasks that share a thread group must share a user namespace */
1274 if (!thread_group_empty(current))
1275 return -EINVAL;
1276
1277 if (current->fs->users != 1)
1278 return -EINVAL;
1279
1280 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1281 return -EPERM;
1282
1283 cred = prepare_creds();
1284 if (!cred)
1285 return -ENOMEM;
1286
1287 put_user_ns(cred->user_ns);
1288 set_cred_user_ns(cred, get_user_ns(user_ns));
1289
1290 return commit_creds(cred);
1291}
1292
1293struct ns_common *ns_get_owner(struct ns_common *ns)
1294{
1295 struct user_namespace *my_user_ns = current_user_ns();
1296 struct user_namespace *owner, *p;
1297
1298 /* See if the owner is in the current user namespace */
1299 owner = p = ns->ops->owner(ns);
1300 for (;;) {
1301 if (!p)
1302 return ERR_PTR(-EPERM);
1303 if (p == my_user_ns)
1304 break;
1305 p = p->parent;
1306 }
1307
1308 return &get_user_ns(owner)->ns;
1309}
1310
1311static struct user_namespace *userns_owner(struct ns_common *ns)
1312{
1313 return to_user_ns(ns)->parent;
1314}
1315
1316const struct proc_ns_operations userns_operations = {
1317 .name = "user",
1318 .type = CLONE_NEWUSER,
1319 .get = userns_get,
1320 .put = userns_put,
1321 .install = userns_install,
1322 .owner = userns_owner,
1323 .get_parent = ns_get_owner,
1324};
1325
1326static __init int user_namespaces_init(void)
1327{
1328 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1329 return 0;
1330}
1331subsys_initcall(user_namespaces_init);
1332