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/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);
1// SPDX-License-Identifier: GPL-2.0-only
2
3#include <linux/export.h>
4#include <linux/nsproxy.h>
5#include <linux/slab.h>
6#include <linux/sched/signal.h>
7#include <linux/user_namespace.h>
8#include <linux/proc_ns.h>
9#include <linux/highuid.h>
10#include <linux/cred.h>
11#include <linux/securebits.h>
12#include <linux/security.h>
13#include <linux/keyctl.h>
14#include <linux/key-type.h>
15#include <keys/user-type.h>
16#include <linux/seq_file.h>
17#include <linux/fs.h>
18#include <linux/uaccess.h>
19#include <linux/ctype.h>
20#include <linux/projid.h>
21#include <linux/fs_struct.h>
22#include <linux/bsearch.h>
23#include <linux/sort.h>
24
25static struct kmem_cache *user_ns_cachep __ro_after_init;
26static DEFINE_MUTEX(userns_state_mutex);
27
28static bool new_idmap_permitted(const struct file *file,
29 struct user_namespace *ns, int cap_setid,
30 struct uid_gid_map *map);
31static void free_user_ns(struct work_struct *work);
32
33static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
34{
35 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
36}
37
38static void dec_user_namespaces(struct ucounts *ucounts)
39{
40 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
41}
42
43static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
44{
45 /* Start with the same capabilities as init but useless for doing
46 * anything as the capabilities are bound to the new user namespace.
47 */
48 cred->securebits = SECUREBITS_DEFAULT;
49 cred->cap_inheritable = CAP_EMPTY_SET;
50 cred->cap_permitted = CAP_FULL_SET;
51 cred->cap_effective = CAP_FULL_SET;
52 cred->cap_ambient = CAP_EMPTY_SET;
53 cred->cap_bset = CAP_FULL_SET;
54#ifdef CONFIG_KEYS
55 key_put(cred->request_key_auth);
56 cred->request_key_auth = NULL;
57#endif
58 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
59 cred->user_ns = user_ns;
60}
61
62static unsigned long enforced_nproc_rlimit(void)
63{
64 unsigned long limit = RLIM_INFINITY;
65
66 /* Is RLIMIT_NPROC currently enforced? */
67 if (!uid_eq(current_uid(), GLOBAL_ROOT_UID) ||
68 (current_user_ns() != &init_user_ns))
69 limit = rlimit(RLIMIT_NPROC);
70
71 return limit;
72}
73
74/*
75 * Create a new user namespace, deriving the creator from the user in the
76 * passed credentials, and replacing that user with the new root user for the
77 * new namespace.
78 *
79 * This is called by copy_creds(), which will finish setting the target task's
80 * credentials.
81 */
82int create_user_ns(struct cred *new)
83{
84 struct user_namespace *ns, *parent_ns = new->user_ns;
85 kuid_t owner = new->euid;
86 kgid_t group = new->egid;
87 struct ucounts *ucounts;
88 int ret, i;
89
90 ret = -ENOSPC;
91 if (parent_ns->level > 32)
92 goto fail;
93
94 ucounts = inc_user_namespaces(parent_ns, owner);
95 if (!ucounts)
96 goto fail;
97
98 /*
99 * Verify that we can not violate the policy of which files
100 * may be accessed that is specified by the root directory,
101 * by verifying that the root directory is at the root of the
102 * mount namespace which allows all files to be accessed.
103 */
104 ret = -EPERM;
105 if (current_chrooted())
106 goto fail_dec;
107
108 /* The creator needs a mapping in the parent user namespace
109 * or else we won't be able to reasonably tell userspace who
110 * created a user_namespace.
111 */
112 ret = -EPERM;
113 if (!kuid_has_mapping(parent_ns, owner) ||
114 !kgid_has_mapping(parent_ns, group))
115 goto fail_dec;
116
117 ret = security_create_user_ns(new);
118 if (ret < 0)
119 goto fail_dec;
120
121 ret = -ENOMEM;
122 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
123 if (!ns)
124 goto fail_dec;
125
126 ns->parent_could_setfcap = cap_raised(new->cap_effective, CAP_SETFCAP);
127 ret = ns_alloc_inum(&ns->ns);
128 if (ret)
129 goto fail_free;
130 ns->ns.ops = &userns_operations;
131
132 refcount_set(&ns->ns.count, 1);
133 /* Leave the new->user_ns reference with the new user namespace. */
134 ns->parent = parent_ns;
135 ns->level = parent_ns->level + 1;
136 ns->owner = owner;
137 ns->group = group;
138 INIT_WORK(&ns->work, free_user_ns);
139 for (i = 0; i < UCOUNT_COUNTS; i++) {
140 ns->ucount_max[i] = INT_MAX;
141 }
142 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_NPROC, enforced_nproc_rlimit());
143 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_MSGQUEUE, rlimit(RLIMIT_MSGQUEUE));
144 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_SIGPENDING, rlimit(RLIMIT_SIGPENDING));
145 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_MEMLOCK, rlimit(RLIMIT_MEMLOCK));
146 ns->ucounts = ucounts;
147
148 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
149 mutex_lock(&userns_state_mutex);
150 ns->flags = parent_ns->flags;
151 mutex_unlock(&userns_state_mutex);
152
153#ifdef CONFIG_KEYS
154 INIT_LIST_HEAD(&ns->keyring_name_list);
155 init_rwsem(&ns->keyring_sem);
156#endif
157 ret = -ENOMEM;
158 if (!setup_userns_sysctls(ns))
159 goto fail_keyring;
160
161 set_cred_user_ns(new, ns);
162 return 0;
163fail_keyring:
164#ifdef CONFIG_PERSISTENT_KEYRINGS
165 key_put(ns->persistent_keyring_register);
166#endif
167 ns_free_inum(&ns->ns);
168fail_free:
169 kmem_cache_free(user_ns_cachep, ns);
170fail_dec:
171 dec_user_namespaces(ucounts);
172fail:
173 return ret;
174}
175
176int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
177{
178 struct cred *cred;
179 int err = -ENOMEM;
180
181 if (!(unshare_flags & CLONE_NEWUSER))
182 return 0;
183
184 cred = prepare_creds();
185 if (cred) {
186 err = create_user_ns(cred);
187 if (err)
188 put_cred(cred);
189 else
190 *new_cred = cred;
191 }
192
193 return err;
194}
195
196static void free_user_ns(struct work_struct *work)
197{
198 struct user_namespace *parent, *ns =
199 container_of(work, struct user_namespace, work);
200
201 do {
202 struct ucounts *ucounts = ns->ucounts;
203 parent = ns->parent;
204 if (ns->gid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
205 kfree(ns->gid_map.forward);
206 kfree(ns->gid_map.reverse);
207 }
208 if (ns->uid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
209 kfree(ns->uid_map.forward);
210 kfree(ns->uid_map.reverse);
211 }
212 if (ns->projid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
213 kfree(ns->projid_map.forward);
214 kfree(ns->projid_map.reverse);
215 }
216#if IS_ENABLED(CONFIG_BINFMT_MISC)
217 kfree(ns->binfmt_misc);
218#endif
219 retire_userns_sysctls(ns);
220 key_free_user_ns(ns);
221 ns_free_inum(&ns->ns);
222 kmem_cache_free(user_ns_cachep, ns);
223 dec_user_namespaces(ucounts);
224 ns = parent;
225 } while (refcount_dec_and_test(&parent->ns.count));
226}
227
228void __put_user_ns(struct user_namespace *ns)
229{
230 schedule_work(&ns->work);
231}
232EXPORT_SYMBOL(__put_user_ns);
233
234/*
235 * struct idmap_key - holds the information necessary to find an idmapping in a
236 * sorted idmap array. It is passed to cmp_map_id() as first argument.
237 */
238struct idmap_key {
239 bool map_up; /* true -> id from kid; false -> kid from id */
240 u32 id; /* id to find */
241 u32 count; /* == 0 unless used with map_id_range_down() */
242};
243
244/*
245 * cmp_map_id - Function to be passed to bsearch() to find the requested
246 * idmapping. Expects struct idmap_key to be passed via @k.
247 */
248static int cmp_map_id(const void *k, const void *e)
249{
250 u32 first, last, id2;
251 const struct idmap_key *key = k;
252 const struct uid_gid_extent *el = e;
253
254 id2 = key->id + key->count - 1;
255
256 /* handle map_id_{down,up}() */
257 if (key->map_up)
258 first = el->lower_first;
259 else
260 first = el->first;
261
262 last = first + el->count - 1;
263
264 if (key->id >= first && key->id <= last &&
265 (id2 >= first && id2 <= last))
266 return 0;
267
268 if (key->id < first || id2 < first)
269 return -1;
270
271 return 1;
272}
273
274/*
275 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
276 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
277 */
278static struct uid_gid_extent *
279map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
280{
281 struct idmap_key key;
282
283 key.map_up = false;
284 key.count = count;
285 key.id = id;
286
287 return bsearch(&key, map->forward, extents,
288 sizeof(struct uid_gid_extent), cmp_map_id);
289}
290
291/*
292 * map_id_range_down_base - Find idmap via binary search in static extent array.
293 * Can only be called if number of mappings is equal or less than
294 * UID_GID_MAP_MAX_BASE_EXTENTS.
295 */
296static struct uid_gid_extent *
297map_id_range_down_base(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
298{
299 unsigned idx;
300 u32 first, last, id2;
301
302 id2 = id + count - 1;
303
304 /* Find the matching extent */
305 for (idx = 0; idx < extents; idx++) {
306 first = map->extent[idx].first;
307 last = first + map->extent[idx].count - 1;
308 if (id >= first && id <= last &&
309 (id2 >= first && id2 <= last))
310 return &map->extent[idx];
311 }
312 return NULL;
313}
314
315static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
316{
317 struct uid_gid_extent *extent;
318 unsigned extents = map->nr_extents;
319 smp_rmb();
320
321 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
322 extent = map_id_range_down_base(extents, map, id, count);
323 else
324 extent = map_id_range_down_max(extents, map, id, count);
325
326 /* Map the id or note failure */
327 if (extent)
328 id = (id - extent->first) + extent->lower_first;
329 else
330 id = (u32) -1;
331
332 return id;
333}
334
335u32 map_id_down(struct uid_gid_map *map, u32 id)
336{
337 return map_id_range_down(map, id, 1);
338}
339
340/*
341 * map_id_up_base - Find idmap via binary search in static extent array.
342 * Can only be called if number of mappings is equal or less than
343 * UID_GID_MAP_MAX_BASE_EXTENTS.
344 */
345static struct uid_gid_extent *
346map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
347{
348 unsigned idx;
349 u32 first, last;
350
351 /* Find the matching extent */
352 for (idx = 0; idx < extents; idx++) {
353 first = map->extent[idx].lower_first;
354 last = first + map->extent[idx].count - 1;
355 if (id >= first && id <= last)
356 return &map->extent[idx];
357 }
358 return NULL;
359}
360
361/*
362 * map_id_up_max - Find idmap via binary search in ordered idmap array.
363 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
364 */
365static struct uid_gid_extent *
366map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
367{
368 struct idmap_key key;
369
370 key.map_up = true;
371 key.count = 1;
372 key.id = id;
373
374 return bsearch(&key, map->reverse, extents,
375 sizeof(struct uid_gid_extent), cmp_map_id);
376}
377
378u32 map_id_up(struct uid_gid_map *map, u32 id)
379{
380 struct uid_gid_extent *extent;
381 unsigned extents = map->nr_extents;
382 smp_rmb();
383
384 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
385 extent = map_id_up_base(extents, map, id);
386 else
387 extent = map_id_up_max(extents, map, id);
388
389 /* Map the id or note failure */
390 if (extent)
391 id = (id - extent->lower_first) + extent->first;
392 else
393 id = (u32) -1;
394
395 return id;
396}
397
398/**
399 * make_kuid - Map a user-namespace uid pair into a kuid.
400 * @ns: User namespace that the uid is in
401 * @uid: User identifier
402 *
403 * Maps a user-namespace uid pair into a kernel internal kuid,
404 * and returns that kuid.
405 *
406 * When there is no mapping defined for the user-namespace uid
407 * pair INVALID_UID is returned. Callers are expected to test
408 * for and handle INVALID_UID being returned. INVALID_UID
409 * may be tested for using uid_valid().
410 */
411kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
412{
413 /* Map the uid to a global kernel uid */
414 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
415}
416EXPORT_SYMBOL(make_kuid);
417
418/**
419 * from_kuid - 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 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
429 */
430uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
431{
432 /* Map the uid from a global kernel uid */
433 return map_id_up(&targ->uid_map, __kuid_val(kuid));
434}
435EXPORT_SYMBOL(from_kuid);
436
437/**
438 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
439 * @targ: The user namespace we want a uid in.
440 * @kuid: The kernel internal uid to start with.
441 *
442 * Map @kuid into the user-namespace specified by @targ and
443 * return the resulting uid.
444 *
445 * There is always a mapping into the initial user_namespace.
446 *
447 * Unlike from_kuid from_kuid_munged never fails and always
448 * returns a valid uid. This makes from_kuid_munged appropriate
449 * for use in syscalls like stat and getuid where failing the
450 * system call and failing to provide a valid uid are not an
451 * options.
452 *
453 * If @kuid has no mapping in @targ overflowuid is returned.
454 */
455uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
456{
457 uid_t uid;
458 uid = from_kuid(targ, kuid);
459
460 if (uid == (uid_t) -1)
461 uid = overflowuid;
462 return uid;
463}
464EXPORT_SYMBOL(from_kuid_munged);
465
466/**
467 * make_kgid - Map a user-namespace gid pair into a kgid.
468 * @ns: User namespace that the gid is in
469 * @gid: group identifier
470 *
471 * Maps a user-namespace gid pair into a kernel internal kgid,
472 * and returns that kgid.
473 *
474 * When there is no mapping defined for the user-namespace gid
475 * pair INVALID_GID is returned. Callers are expected to test
476 * for and handle INVALID_GID being returned. INVALID_GID may be
477 * tested for using gid_valid().
478 */
479kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
480{
481 /* Map the gid to a global kernel gid */
482 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
483}
484EXPORT_SYMBOL(make_kgid);
485
486/**
487 * from_kgid - 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 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
497 */
498gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
499{
500 /* Map the gid from a global kernel gid */
501 return map_id_up(&targ->gid_map, __kgid_val(kgid));
502}
503EXPORT_SYMBOL(from_kgid);
504
505/**
506 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
507 * @targ: The user namespace we want a gid in.
508 * @kgid: The kernel internal gid to start with.
509 *
510 * Map @kgid into the user-namespace specified by @targ and
511 * return the resulting gid.
512 *
513 * There is always a mapping into the initial user_namespace.
514 *
515 * Unlike from_kgid from_kgid_munged never fails and always
516 * returns a valid gid. This makes from_kgid_munged appropriate
517 * for use in syscalls like stat and getgid where failing the
518 * system call and failing to provide a valid gid are not options.
519 *
520 * If @kgid has no mapping in @targ overflowgid is returned.
521 */
522gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
523{
524 gid_t gid;
525 gid = from_kgid(targ, kgid);
526
527 if (gid == (gid_t) -1)
528 gid = overflowgid;
529 return gid;
530}
531EXPORT_SYMBOL(from_kgid_munged);
532
533/**
534 * make_kprojid - Map a user-namespace projid pair into a kprojid.
535 * @ns: User namespace that the projid is in
536 * @projid: Project identifier
537 *
538 * Maps a user-namespace uid pair into a kernel internal kuid,
539 * and returns that kuid.
540 *
541 * When there is no mapping defined for the user-namespace projid
542 * pair INVALID_PROJID is returned. Callers are expected to test
543 * for and handle INVALID_PROJID being returned. INVALID_PROJID
544 * may be tested for using projid_valid().
545 */
546kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
547{
548 /* Map the uid to a global kernel uid */
549 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
550}
551EXPORT_SYMBOL(make_kprojid);
552
553/**
554 * from_kprojid - Create a projid from a kprojid user-namespace pair.
555 * @targ: The user namespace we want a projid in.
556 * @kprojid: The kernel internal project identifier 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 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
564 */
565projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
566{
567 /* Map the uid from a global kernel uid */
568 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
569}
570EXPORT_SYMBOL(from_kprojid);
571
572/**
573 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
574 * @targ: The user namespace we want a projid in.
575 * @kprojid: The kernel internal projid to start with.
576 *
577 * Map @kprojid into the user-namespace specified by @targ and
578 * return the resulting projid.
579 *
580 * There is always a mapping into the initial user_namespace.
581 *
582 * Unlike from_kprojid from_kprojid_munged never fails and always
583 * returns a valid projid. This makes from_kprojid_munged
584 * appropriate for use in syscalls like stat and where
585 * failing the system call and failing to provide a valid projid are
586 * not an options.
587 *
588 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
589 */
590projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
591{
592 projid_t projid;
593 projid = from_kprojid(targ, kprojid);
594
595 if (projid == (projid_t) -1)
596 projid = OVERFLOW_PROJID;
597 return projid;
598}
599EXPORT_SYMBOL(from_kprojid_munged);
600
601
602static int uid_m_show(struct seq_file *seq, void *v)
603{
604 struct user_namespace *ns = seq->private;
605 struct uid_gid_extent *extent = v;
606 struct user_namespace *lower_ns;
607 uid_t lower;
608
609 lower_ns = seq_user_ns(seq);
610 if ((lower_ns == ns) && lower_ns->parent)
611 lower_ns = lower_ns->parent;
612
613 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
614
615 seq_printf(seq, "%10u %10u %10u\n",
616 extent->first,
617 lower,
618 extent->count);
619
620 return 0;
621}
622
623static int gid_m_show(struct seq_file *seq, void *v)
624{
625 struct user_namespace *ns = seq->private;
626 struct uid_gid_extent *extent = v;
627 struct user_namespace *lower_ns;
628 gid_t lower;
629
630 lower_ns = seq_user_ns(seq);
631 if ((lower_ns == ns) && lower_ns->parent)
632 lower_ns = lower_ns->parent;
633
634 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
635
636 seq_printf(seq, "%10u %10u %10u\n",
637 extent->first,
638 lower,
639 extent->count);
640
641 return 0;
642}
643
644static int projid_m_show(struct seq_file *seq, void *v)
645{
646 struct user_namespace *ns = seq->private;
647 struct uid_gid_extent *extent = v;
648 struct user_namespace *lower_ns;
649 projid_t lower;
650
651 lower_ns = seq_user_ns(seq);
652 if ((lower_ns == ns) && lower_ns->parent)
653 lower_ns = lower_ns->parent;
654
655 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
656
657 seq_printf(seq, "%10u %10u %10u\n",
658 extent->first,
659 lower,
660 extent->count);
661
662 return 0;
663}
664
665static void *m_start(struct seq_file *seq, loff_t *ppos,
666 struct uid_gid_map *map)
667{
668 loff_t pos = *ppos;
669 unsigned extents = map->nr_extents;
670 smp_rmb();
671
672 if (pos >= extents)
673 return NULL;
674
675 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
676 return &map->extent[pos];
677
678 return &map->forward[pos];
679}
680
681static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
682{
683 struct user_namespace *ns = seq->private;
684
685 return m_start(seq, ppos, &ns->uid_map);
686}
687
688static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
689{
690 struct user_namespace *ns = seq->private;
691
692 return m_start(seq, ppos, &ns->gid_map);
693}
694
695static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
696{
697 struct user_namespace *ns = seq->private;
698
699 return m_start(seq, ppos, &ns->projid_map);
700}
701
702static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
703{
704 (*pos)++;
705 return seq->op->start(seq, pos);
706}
707
708static void m_stop(struct seq_file *seq, void *v)
709{
710 return;
711}
712
713const struct seq_operations proc_uid_seq_operations = {
714 .start = uid_m_start,
715 .stop = m_stop,
716 .next = m_next,
717 .show = uid_m_show,
718};
719
720const struct seq_operations proc_gid_seq_operations = {
721 .start = gid_m_start,
722 .stop = m_stop,
723 .next = m_next,
724 .show = gid_m_show,
725};
726
727const struct seq_operations proc_projid_seq_operations = {
728 .start = projid_m_start,
729 .stop = m_stop,
730 .next = m_next,
731 .show = projid_m_show,
732};
733
734static bool mappings_overlap(struct uid_gid_map *new_map,
735 struct uid_gid_extent *extent)
736{
737 u32 upper_first, lower_first, upper_last, lower_last;
738 unsigned idx;
739
740 upper_first = extent->first;
741 lower_first = extent->lower_first;
742 upper_last = upper_first + extent->count - 1;
743 lower_last = lower_first + extent->count - 1;
744
745 for (idx = 0; idx < new_map->nr_extents; idx++) {
746 u32 prev_upper_first, prev_lower_first;
747 u32 prev_upper_last, prev_lower_last;
748 struct uid_gid_extent *prev;
749
750 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
751 prev = &new_map->extent[idx];
752 else
753 prev = &new_map->forward[idx];
754
755 prev_upper_first = prev->first;
756 prev_lower_first = prev->lower_first;
757 prev_upper_last = prev_upper_first + prev->count - 1;
758 prev_lower_last = prev_lower_first + prev->count - 1;
759
760 /* Does the upper range intersect a previous extent? */
761 if ((prev_upper_first <= upper_last) &&
762 (prev_upper_last >= upper_first))
763 return true;
764
765 /* Does the lower range intersect a previous extent? */
766 if ((prev_lower_first <= lower_last) &&
767 (prev_lower_last >= lower_first))
768 return true;
769 }
770 return false;
771}
772
773/*
774 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
775 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
776 * UID_GID_MAP_MAX_BASE_EXTENTS.
777 */
778static int insert_extent(struct uid_gid_map *map, struct uid_gid_extent *extent)
779{
780 struct uid_gid_extent *dest;
781
782 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
783 struct uid_gid_extent *forward;
784
785 /* Allocate memory for 340 mappings. */
786 forward = kmalloc_array(UID_GID_MAP_MAX_EXTENTS,
787 sizeof(struct uid_gid_extent),
788 GFP_KERNEL);
789 if (!forward)
790 return -ENOMEM;
791
792 /* Copy over memory. Only set up memory for the forward pointer.
793 * Defer the memory setup for the reverse pointer.
794 */
795 memcpy(forward, map->extent,
796 map->nr_extents * sizeof(map->extent[0]));
797
798 map->forward = forward;
799 map->reverse = NULL;
800 }
801
802 if (map->nr_extents < UID_GID_MAP_MAX_BASE_EXTENTS)
803 dest = &map->extent[map->nr_extents];
804 else
805 dest = &map->forward[map->nr_extents];
806
807 *dest = *extent;
808 map->nr_extents++;
809 return 0;
810}
811
812/* cmp function to sort() forward mappings */
813static int cmp_extents_forward(const void *a, const void *b)
814{
815 const struct uid_gid_extent *e1 = a;
816 const struct uid_gid_extent *e2 = b;
817
818 if (e1->first < e2->first)
819 return -1;
820
821 if (e1->first > e2->first)
822 return 1;
823
824 return 0;
825}
826
827/* cmp function to sort() reverse mappings */
828static int cmp_extents_reverse(const void *a, const void *b)
829{
830 const struct uid_gid_extent *e1 = a;
831 const struct uid_gid_extent *e2 = b;
832
833 if (e1->lower_first < e2->lower_first)
834 return -1;
835
836 if (e1->lower_first > e2->lower_first)
837 return 1;
838
839 return 0;
840}
841
842/*
843 * sort_idmaps - Sorts an array of idmap entries.
844 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
845 */
846static int sort_idmaps(struct uid_gid_map *map)
847{
848 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
849 return 0;
850
851 /* Sort forward array. */
852 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
853 cmp_extents_forward, NULL);
854
855 /* Only copy the memory from forward we actually need. */
856 map->reverse = kmemdup_array(map->forward, map->nr_extents,
857 sizeof(struct uid_gid_extent), GFP_KERNEL);
858 if (!map->reverse)
859 return -ENOMEM;
860
861 /* Sort reverse array. */
862 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
863 cmp_extents_reverse, NULL);
864
865 return 0;
866}
867
868/**
869 * verify_root_map() - check the uid 0 mapping
870 * @file: idmapping file
871 * @map_ns: user namespace of the target process
872 * @new_map: requested idmap
873 *
874 * If a process requests mapping parent uid 0 into the new ns, verify that the
875 * process writing the map had the CAP_SETFCAP capability as the target process
876 * will be able to write fscaps that are valid in ancestor user namespaces.
877 *
878 * Return: true if the mapping is allowed, false if not.
879 */
880static bool verify_root_map(const struct file *file,
881 struct user_namespace *map_ns,
882 struct uid_gid_map *new_map)
883{
884 int idx;
885 const struct user_namespace *file_ns = file->f_cred->user_ns;
886 struct uid_gid_extent *extent0 = NULL;
887
888 for (idx = 0; idx < new_map->nr_extents; idx++) {
889 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
890 extent0 = &new_map->extent[idx];
891 else
892 extent0 = &new_map->forward[idx];
893 if (extent0->lower_first == 0)
894 break;
895
896 extent0 = NULL;
897 }
898
899 if (!extent0)
900 return true;
901
902 if (map_ns == file_ns) {
903 /* The process unshared its ns and is writing to its own
904 * /proc/self/uid_map. User already has full capabilites in
905 * the new namespace. Verify that the parent had CAP_SETFCAP
906 * when it unshared.
907 * */
908 if (!file_ns->parent_could_setfcap)
909 return false;
910 } else {
911 /* Process p1 is writing to uid_map of p2, who is in a child
912 * user namespace to p1's. Verify that the opener of the map
913 * file has CAP_SETFCAP against the parent of the new map
914 * namespace */
915 if (!file_ns_capable(file, map_ns->parent, CAP_SETFCAP))
916 return false;
917 }
918
919 return true;
920}
921
922static ssize_t map_write(struct file *file, const char __user *buf,
923 size_t count, loff_t *ppos,
924 int cap_setid,
925 struct uid_gid_map *map,
926 struct uid_gid_map *parent_map)
927{
928 struct seq_file *seq = file->private_data;
929 struct user_namespace *map_ns = seq->private;
930 struct uid_gid_map new_map;
931 unsigned idx;
932 struct uid_gid_extent extent;
933 char *kbuf, *pos, *next_line;
934 ssize_t ret;
935
936 /* Only allow < page size writes at the beginning of the file */
937 if ((*ppos != 0) || (count >= PAGE_SIZE))
938 return -EINVAL;
939
940 /* Slurp in the user data */
941 kbuf = memdup_user_nul(buf, count);
942 if (IS_ERR(kbuf))
943 return PTR_ERR(kbuf);
944
945 /*
946 * The userns_state_mutex serializes all writes to any given map.
947 *
948 * Any map is only ever written once.
949 *
950 * An id map fits within 1 cache line on most architectures.
951 *
952 * On read nothing needs to be done unless you are on an
953 * architecture with a crazy cache coherency model like alpha.
954 *
955 * There is a one time data dependency between reading the
956 * count of the extents and the values of the extents. The
957 * desired behavior is to see the values of the extents that
958 * were written before the count of the extents.
959 *
960 * To achieve this smp_wmb() is used on guarantee the write
961 * order and smp_rmb() is guaranteed that we don't have crazy
962 * architectures returning stale data.
963 */
964 mutex_lock(&userns_state_mutex);
965
966 memset(&new_map, 0, sizeof(struct uid_gid_map));
967
968 ret = -EPERM;
969 /* Only allow one successful write to the map */
970 if (map->nr_extents != 0)
971 goto out;
972
973 /*
974 * Adjusting namespace settings requires capabilities on the target.
975 */
976 if (cap_valid(cap_setid) && !file_ns_capable(file, map_ns, CAP_SYS_ADMIN))
977 goto out;
978
979 /* Parse the user data */
980 ret = -EINVAL;
981 pos = kbuf;
982 for (; pos; pos = next_line) {
983
984 /* Find the end of line and ensure I don't look past it */
985 next_line = strchr(pos, '\n');
986 if (next_line) {
987 *next_line = '\0';
988 next_line++;
989 if (*next_line == '\0')
990 next_line = NULL;
991 }
992
993 pos = skip_spaces(pos);
994 extent.first = simple_strtoul(pos, &pos, 10);
995 if (!isspace(*pos))
996 goto out;
997
998 pos = skip_spaces(pos);
999 extent.lower_first = simple_strtoul(pos, &pos, 10);
1000 if (!isspace(*pos))
1001 goto out;
1002
1003 pos = skip_spaces(pos);
1004 extent.count = simple_strtoul(pos, &pos, 10);
1005 if (*pos && !isspace(*pos))
1006 goto out;
1007
1008 /* Verify there is not trailing junk on the line */
1009 pos = skip_spaces(pos);
1010 if (*pos != '\0')
1011 goto out;
1012
1013 /* Verify we have been given valid starting values */
1014 if ((extent.first == (u32) -1) ||
1015 (extent.lower_first == (u32) -1))
1016 goto out;
1017
1018 /* Verify count is not zero and does not cause the
1019 * extent to wrap
1020 */
1021 if ((extent.first + extent.count) <= extent.first)
1022 goto out;
1023 if ((extent.lower_first + extent.count) <=
1024 extent.lower_first)
1025 goto out;
1026
1027 /* Do the ranges in extent overlap any previous extents? */
1028 if (mappings_overlap(&new_map, &extent))
1029 goto out;
1030
1031 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
1032 (next_line != NULL))
1033 goto out;
1034
1035 ret = insert_extent(&new_map, &extent);
1036 if (ret < 0)
1037 goto out;
1038 ret = -EINVAL;
1039 }
1040 /* Be very certain the new map actually exists */
1041 if (new_map.nr_extents == 0)
1042 goto out;
1043
1044 ret = -EPERM;
1045 /* Validate the user is allowed to use user id's mapped to. */
1046 if (!new_idmap_permitted(file, map_ns, cap_setid, &new_map))
1047 goto out;
1048
1049 ret = -EPERM;
1050 /* Map the lower ids from the parent user namespace to the
1051 * kernel global id space.
1052 */
1053 for (idx = 0; idx < new_map.nr_extents; idx++) {
1054 struct uid_gid_extent *e;
1055 u32 lower_first;
1056
1057 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
1058 e = &new_map.extent[idx];
1059 else
1060 e = &new_map.forward[idx];
1061
1062 lower_first = map_id_range_down(parent_map,
1063 e->lower_first,
1064 e->count);
1065
1066 /* Fail if we can not map the specified extent to
1067 * the kernel global id space.
1068 */
1069 if (lower_first == (u32) -1)
1070 goto out;
1071
1072 e->lower_first = lower_first;
1073 }
1074
1075 /*
1076 * If we want to use binary search for lookup, this clones the extent
1077 * array and sorts both copies.
1078 */
1079 ret = sort_idmaps(&new_map);
1080 if (ret < 0)
1081 goto out;
1082
1083 /* Install the map */
1084 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1085 memcpy(map->extent, new_map.extent,
1086 new_map.nr_extents * sizeof(new_map.extent[0]));
1087 } else {
1088 map->forward = new_map.forward;
1089 map->reverse = new_map.reverse;
1090 }
1091 smp_wmb();
1092 map->nr_extents = new_map.nr_extents;
1093
1094 *ppos = count;
1095 ret = count;
1096out:
1097 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1098 kfree(new_map.forward);
1099 kfree(new_map.reverse);
1100 map->forward = NULL;
1101 map->reverse = NULL;
1102 map->nr_extents = 0;
1103 }
1104
1105 mutex_unlock(&userns_state_mutex);
1106 kfree(kbuf);
1107 return ret;
1108}
1109
1110ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1111 size_t size, loff_t *ppos)
1112{
1113 struct seq_file *seq = file->private_data;
1114 struct user_namespace *ns = seq->private;
1115 struct user_namespace *seq_ns = seq_user_ns(seq);
1116
1117 if (!ns->parent)
1118 return -EPERM;
1119
1120 if ((seq_ns != ns) && (seq_ns != ns->parent))
1121 return -EPERM;
1122
1123 return map_write(file, buf, size, ppos, CAP_SETUID,
1124 &ns->uid_map, &ns->parent->uid_map);
1125}
1126
1127ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1128 size_t size, loff_t *ppos)
1129{
1130 struct seq_file *seq = file->private_data;
1131 struct user_namespace *ns = seq->private;
1132 struct user_namespace *seq_ns = seq_user_ns(seq);
1133
1134 if (!ns->parent)
1135 return -EPERM;
1136
1137 if ((seq_ns != ns) && (seq_ns != ns->parent))
1138 return -EPERM;
1139
1140 return map_write(file, buf, size, ppos, CAP_SETGID,
1141 &ns->gid_map, &ns->parent->gid_map);
1142}
1143
1144ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1145 size_t size, loff_t *ppos)
1146{
1147 struct seq_file *seq = file->private_data;
1148 struct user_namespace *ns = seq->private;
1149 struct user_namespace *seq_ns = seq_user_ns(seq);
1150
1151 if (!ns->parent)
1152 return -EPERM;
1153
1154 if ((seq_ns != ns) && (seq_ns != ns->parent))
1155 return -EPERM;
1156
1157 /* Anyone can set any valid project id no capability needed */
1158 return map_write(file, buf, size, ppos, -1,
1159 &ns->projid_map, &ns->parent->projid_map);
1160}
1161
1162static bool new_idmap_permitted(const struct file *file,
1163 struct user_namespace *ns, int cap_setid,
1164 struct uid_gid_map *new_map)
1165{
1166 const struct cred *cred = file->f_cred;
1167
1168 if (cap_setid == CAP_SETUID && !verify_root_map(file, ns, new_map))
1169 return false;
1170
1171 /* Don't allow mappings that would allow anything that wouldn't
1172 * be allowed without the establishment of unprivileged mappings.
1173 */
1174 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1175 uid_eq(ns->owner, cred->euid)) {
1176 u32 id = new_map->extent[0].lower_first;
1177 if (cap_setid == CAP_SETUID) {
1178 kuid_t uid = make_kuid(ns->parent, id);
1179 if (uid_eq(uid, cred->euid))
1180 return true;
1181 } else if (cap_setid == CAP_SETGID) {
1182 kgid_t gid = make_kgid(ns->parent, id);
1183 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1184 gid_eq(gid, cred->egid))
1185 return true;
1186 }
1187 }
1188
1189 /* Allow anyone to set a mapping that doesn't require privilege */
1190 if (!cap_valid(cap_setid))
1191 return true;
1192
1193 /* Allow the specified ids if we have the appropriate capability
1194 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1195 * And the opener of the id file also has the appropriate capability.
1196 */
1197 if (ns_capable(ns->parent, cap_setid) &&
1198 file_ns_capable(file, ns->parent, cap_setid))
1199 return true;
1200
1201 return false;
1202}
1203
1204int proc_setgroups_show(struct seq_file *seq, void *v)
1205{
1206 struct user_namespace *ns = seq->private;
1207 unsigned long userns_flags = READ_ONCE(ns->flags);
1208
1209 seq_printf(seq, "%s\n",
1210 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1211 "allow" : "deny");
1212 return 0;
1213}
1214
1215ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1216 size_t count, loff_t *ppos)
1217{
1218 struct seq_file *seq = file->private_data;
1219 struct user_namespace *ns = seq->private;
1220 char kbuf[8], *pos;
1221 bool setgroups_allowed;
1222 ssize_t ret;
1223
1224 /* Only allow a very narrow range of strings to be written */
1225 ret = -EINVAL;
1226 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1227 goto out;
1228
1229 /* What was written? */
1230 ret = -EFAULT;
1231 if (copy_from_user(kbuf, buf, count))
1232 goto out;
1233 kbuf[count] = '\0';
1234 pos = kbuf;
1235
1236 /* What is being requested? */
1237 ret = -EINVAL;
1238 if (strncmp(pos, "allow", 5) == 0) {
1239 pos += 5;
1240 setgroups_allowed = true;
1241 }
1242 else if (strncmp(pos, "deny", 4) == 0) {
1243 pos += 4;
1244 setgroups_allowed = false;
1245 }
1246 else
1247 goto out;
1248
1249 /* Verify there is not trailing junk on the line */
1250 pos = skip_spaces(pos);
1251 if (*pos != '\0')
1252 goto out;
1253
1254 ret = -EPERM;
1255 mutex_lock(&userns_state_mutex);
1256 if (setgroups_allowed) {
1257 /* Enabling setgroups after setgroups has been disabled
1258 * is not allowed.
1259 */
1260 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1261 goto out_unlock;
1262 } else {
1263 /* Permanently disabling setgroups after setgroups has
1264 * been enabled by writing the gid_map is not allowed.
1265 */
1266 if (ns->gid_map.nr_extents != 0)
1267 goto out_unlock;
1268 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1269 }
1270 mutex_unlock(&userns_state_mutex);
1271
1272 /* Report a successful write */
1273 *ppos = count;
1274 ret = count;
1275out:
1276 return ret;
1277out_unlock:
1278 mutex_unlock(&userns_state_mutex);
1279 goto out;
1280}
1281
1282bool userns_may_setgroups(const struct user_namespace *ns)
1283{
1284 bool allowed;
1285
1286 mutex_lock(&userns_state_mutex);
1287 /* It is not safe to use setgroups until a gid mapping in
1288 * the user namespace has been established.
1289 */
1290 allowed = ns->gid_map.nr_extents != 0;
1291 /* Is setgroups allowed? */
1292 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1293 mutex_unlock(&userns_state_mutex);
1294
1295 return allowed;
1296}
1297
1298/*
1299 * Returns true if @child is the same namespace or a descendant of
1300 * @ancestor.
1301 */
1302bool in_userns(const struct user_namespace *ancestor,
1303 const struct user_namespace *child)
1304{
1305 const struct user_namespace *ns;
1306 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1307 ;
1308 return (ns == ancestor);
1309}
1310
1311bool current_in_userns(const struct user_namespace *target_ns)
1312{
1313 return in_userns(target_ns, current_user_ns());
1314}
1315EXPORT_SYMBOL(current_in_userns);
1316
1317static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1318{
1319 return container_of(ns, struct user_namespace, ns);
1320}
1321
1322static struct ns_common *userns_get(struct task_struct *task)
1323{
1324 struct user_namespace *user_ns;
1325
1326 rcu_read_lock();
1327 user_ns = get_user_ns(__task_cred(task)->user_ns);
1328 rcu_read_unlock();
1329
1330 return user_ns ? &user_ns->ns : NULL;
1331}
1332
1333static void userns_put(struct ns_common *ns)
1334{
1335 put_user_ns(to_user_ns(ns));
1336}
1337
1338static int userns_install(struct nsset *nsset, struct ns_common *ns)
1339{
1340 struct user_namespace *user_ns = to_user_ns(ns);
1341 struct cred *cred;
1342
1343 /* Don't allow gaining capabilities by reentering
1344 * the same user namespace.
1345 */
1346 if (user_ns == current_user_ns())
1347 return -EINVAL;
1348
1349 /* Tasks that share a thread group must share a user namespace */
1350 if (!thread_group_empty(current))
1351 return -EINVAL;
1352
1353 if (current->fs->users != 1)
1354 return -EINVAL;
1355
1356 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1357 return -EPERM;
1358
1359 cred = nsset_cred(nsset);
1360 if (!cred)
1361 return -EINVAL;
1362
1363 put_user_ns(cred->user_ns);
1364 set_cred_user_ns(cred, get_user_ns(user_ns));
1365
1366 if (set_cred_ucounts(cred) < 0)
1367 return -EINVAL;
1368
1369 return 0;
1370}
1371
1372struct ns_common *ns_get_owner(struct ns_common *ns)
1373{
1374 struct user_namespace *my_user_ns = current_user_ns();
1375 struct user_namespace *owner, *p;
1376
1377 /* See if the owner is in the current user namespace */
1378 owner = p = ns->ops->owner(ns);
1379 for (;;) {
1380 if (!p)
1381 return ERR_PTR(-EPERM);
1382 if (p == my_user_ns)
1383 break;
1384 p = p->parent;
1385 }
1386
1387 return &get_user_ns(owner)->ns;
1388}
1389
1390static struct user_namespace *userns_owner(struct ns_common *ns)
1391{
1392 return to_user_ns(ns)->parent;
1393}
1394
1395const struct proc_ns_operations userns_operations = {
1396 .name = "user",
1397 .type = CLONE_NEWUSER,
1398 .get = userns_get,
1399 .put = userns_put,
1400 .install = userns_install,
1401 .owner = userns_owner,
1402 .get_parent = ns_get_owner,
1403};
1404
1405static __init int user_namespaces_init(void)
1406{
1407 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC | SLAB_ACCOUNT);
1408 return 0;
1409}
1410subsys_initcall(user_namespaces_init);