Loading...
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/keyctl.h>
13#include <linux/key-type.h>
14#include <keys/user-type.h>
15#include <linux/seq_file.h>
16#include <linux/fs.h>
17#include <linux/uaccess.h>
18#include <linux/ctype.h>
19#include <linux/projid.h>
20#include <linux/fs_struct.h>
21#include <linux/bsearch.h>
22#include <linux/sort.h>
23
24static struct kmem_cache *user_ns_cachep __read_mostly;
25static DEFINE_MUTEX(userns_state_mutex);
26
27static bool new_idmap_permitted(const struct file *file,
28 struct user_namespace *ns, int cap_setid,
29 struct uid_gid_map *map);
30static void free_user_ns(struct work_struct *work);
31
32static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
33{
34 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
35}
36
37static void dec_user_namespaces(struct ucounts *ucounts)
38{
39 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
40}
41
42static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
43{
44 /* Start with the same capabilities as init but useless for doing
45 * anything as the capabilities are bound to the new user namespace.
46 */
47 cred->securebits = SECUREBITS_DEFAULT;
48 cred->cap_inheritable = CAP_EMPTY_SET;
49 cred->cap_permitted = CAP_FULL_SET;
50 cred->cap_effective = CAP_FULL_SET;
51 cred->cap_ambient = CAP_EMPTY_SET;
52 cred->cap_bset = CAP_FULL_SET;
53#ifdef CONFIG_KEYS
54 key_put(cred->request_key_auth);
55 cred->request_key_auth = NULL;
56#endif
57 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
58 cred->user_ns = user_ns;
59}
60
61/*
62 * Create a new user namespace, deriving the creator from the user in the
63 * passed credentials, and replacing that user with the new root user for the
64 * new namespace.
65 *
66 * This is called by copy_creds(), which will finish setting the target task's
67 * credentials.
68 */
69int create_user_ns(struct cred *new)
70{
71 struct user_namespace *ns, *parent_ns = new->user_ns;
72 kuid_t owner = new->euid;
73 kgid_t group = new->egid;
74 struct ucounts *ucounts;
75 int ret, i;
76
77 ret = -ENOSPC;
78 if (parent_ns->level > 32)
79 goto fail;
80
81 ucounts = inc_user_namespaces(parent_ns, owner);
82 if (!ucounts)
83 goto fail;
84
85 /*
86 * Verify that we can not violate the policy of which files
87 * may be accessed that is specified by the root directory,
88 * by verifing that the root directory is at the root of the
89 * mount namespace which allows all files to be accessed.
90 */
91 ret = -EPERM;
92 if (current_chrooted())
93 goto fail_dec;
94
95 /* The creator needs a mapping in the parent user namespace
96 * or else we won't be able to reasonably tell userspace who
97 * created a user_namespace.
98 */
99 ret = -EPERM;
100 if (!kuid_has_mapping(parent_ns, owner) ||
101 !kgid_has_mapping(parent_ns, group))
102 goto fail_dec;
103
104 ret = -ENOMEM;
105 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
106 if (!ns)
107 goto fail_dec;
108
109 ret = ns_alloc_inum(&ns->ns);
110 if (ret)
111 goto fail_free;
112 ns->ns.ops = &userns_operations;
113
114 atomic_set(&ns->count, 1);
115 /* Leave the new->user_ns reference with the new user namespace. */
116 ns->parent = parent_ns;
117 ns->level = parent_ns->level + 1;
118 ns->owner = owner;
119 ns->group = group;
120 INIT_WORK(&ns->work, free_user_ns);
121 for (i = 0; i < UCOUNT_COUNTS; i++) {
122 ns->ucount_max[i] = INT_MAX;
123 }
124 ns->ucounts = ucounts;
125
126 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
127 mutex_lock(&userns_state_mutex);
128 ns->flags = parent_ns->flags;
129 mutex_unlock(&userns_state_mutex);
130
131#ifdef CONFIG_KEYS
132 INIT_LIST_HEAD(&ns->keyring_name_list);
133 init_rwsem(&ns->keyring_sem);
134#endif
135 ret = -ENOMEM;
136 if (!setup_userns_sysctls(ns))
137 goto fail_keyring;
138
139 set_cred_user_ns(new, ns);
140 return 0;
141fail_keyring:
142#ifdef CONFIG_PERSISTENT_KEYRINGS
143 key_put(ns->persistent_keyring_register);
144#endif
145 ns_free_inum(&ns->ns);
146fail_free:
147 kmem_cache_free(user_ns_cachep, ns);
148fail_dec:
149 dec_user_namespaces(ucounts);
150fail:
151 return ret;
152}
153
154int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
155{
156 struct cred *cred;
157 int err = -ENOMEM;
158
159 if (!(unshare_flags & CLONE_NEWUSER))
160 return 0;
161
162 cred = prepare_creds();
163 if (cred) {
164 err = create_user_ns(cred);
165 if (err)
166 put_cred(cred);
167 else
168 *new_cred = cred;
169 }
170
171 return err;
172}
173
174static void free_user_ns(struct work_struct *work)
175{
176 struct user_namespace *parent, *ns =
177 container_of(work, struct user_namespace, work);
178
179 do {
180 struct ucounts *ucounts = ns->ucounts;
181 parent = ns->parent;
182 if (ns->gid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
183 kfree(ns->gid_map.forward);
184 kfree(ns->gid_map.reverse);
185 }
186 if (ns->uid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
187 kfree(ns->uid_map.forward);
188 kfree(ns->uid_map.reverse);
189 }
190 if (ns->projid_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
191 kfree(ns->projid_map.forward);
192 kfree(ns->projid_map.reverse);
193 }
194 retire_userns_sysctls(ns);
195 key_free_user_ns(ns);
196 ns_free_inum(&ns->ns);
197 kmem_cache_free(user_ns_cachep, ns);
198 dec_user_namespaces(ucounts);
199 ns = parent;
200 } while (atomic_dec_and_test(&parent->count));
201}
202
203void __put_user_ns(struct user_namespace *ns)
204{
205 schedule_work(&ns->work);
206}
207EXPORT_SYMBOL(__put_user_ns);
208
209/**
210 * idmap_key struct holds the information necessary to find an idmapping in a
211 * sorted idmap array. It is passed to cmp_map_id() as first argument.
212 */
213struct idmap_key {
214 bool map_up; /* true -> id from kid; false -> kid from id */
215 u32 id; /* id to find */
216 u32 count; /* == 0 unless used with map_id_range_down() */
217};
218
219/**
220 * cmp_map_id - Function to be passed to bsearch() to find the requested
221 * idmapping. Expects struct idmap_key to be passed via @k.
222 */
223static int cmp_map_id(const void *k, const void *e)
224{
225 u32 first, last, id2;
226 const struct idmap_key *key = k;
227 const struct uid_gid_extent *el = e;
228
229 id2 = key->id + key->count - 1;
230
231 /* handle map_id_{down,up}() */
232 if (key->map_up)
233 first = el->lower_first;
234 else
235 first = el->first;
236
237 last = first + el->count - 1;
238
239 if (key->id >= first && key->id <= last &&
240 (id2 >= first && id2 <= last))
241 return 0;
242
243 if (key->id < first || id2 < first)
244 return -1;
245
246 return 1;
247}
248
249/**
250 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
251 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
252 */
253static struct uid_gid_extent *
254map_id_range_down_max(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
255{
256 struct idmap_key key;
257
258 key.map_up = false;
259 key.count = count;
260 key.id = id;
261
262 return bsearch(&key, map->forward, extents,
263 sizeof(struct uid_gid_extent), cmp_map_id);
264}
265
266/**
267 * map_id_range_down_base - Find idmap via binary search in static extent array.
268 * Can only be called if number of mappings is equal or less than
269 * UID_GID_MAP_MAX_BASE_EXTENTS.
270 */
271static struct uid_gid_extent *
272map_id_range_down_base(unsigned extents, struct uid_gid_map *map, u32 id, u32 count)
273{
274 unsigned idx;
275 u32 first, last, id2;
276
277 id2 = id + count - 1;
278
279 /* Find the matching extent */
280 for (idx = 0; idx < extents; idx++) {
281 first = map->extent[idx].first;
282 last = first + map->extent[idx].count - 1;
283 if (id >= first && id <= last &&
284 (id2 >= first && id2 <= last))
285 return &map->extent[idx];
286 }
287 return NULL;
288}
289
290static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
291{
292 struct uid_gid_extent *extent;
293 unsigned extents = map->nr_extents;
294 smp_rmb();
295
296 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
297 extent = map_id_range_down_base(extents, map, id, count);
298 else
299 extent = map_id_range_down_max(extents, map, id, count);
300
301 /* Map the id or note failure */
302 if (extent)
303 id = (id - extent->first) + extent->lower_first;
304 else
305 id = (u32) -1;
306
307 return id;
308}
309
310static u32 map_id_down(struct uid_gid_map *map, u32 id)
311{
312 return map_id_range_down(map, id, 1);
313}
314
315/**
316 * map_id_up_base - Find idmap via binary search in static extent array.
317 * Can only be called if number of mappings is equal or less than
318 * UID_GID_MAP_MAX_BASE_EXTENTS.
319 */
320static struct uid_gid_extent *
321map_id_up_base(unsigned extents, struct uid_gid_map *map, u32 id)
322{
323 unsigned idx;
324 u32 first, last;
325
326 /* Find the matching extent */
327 for (idx = 0; idx < extents; idx++) {
328 first = map->extent[idx].lower_first;
329 last = first + map->extent[idx].count - 1;
330 if (id >= first && id <= last)
331 return &map->extent[idx];
332 }
333 return NULL;
334}
335
336/**
337 * map_id_up_max - Find idmap via binary search in ordered idmap array.
338 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
339 */
340static struct uid_gid_extent *
341map_id_up_max(unsigned extents, struct uid_gid_map *map, u32 id)
342{
343 struct idmap_key key;
344
345 key.map_up = true;
346 key.count = 1;
347 key.id = id;
348
349 return bsearch(&key, map->reverse, extents,
350 sizeof(struct uid_gid_extent), cmp_map_id);
351}
352
353static u32 map_id_up(struct uid_gid_map *map, u32 id)
354{
355 struct uid_gid_extent *extent;
356 unsigned extents = map->nr_extents;
357 smp_rmb();
358
359 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
360 extent = map_id_up_base(extents, map, id);
361 else
362 extent = map_id_up_max(extents, map, id);
363
364 /* Map the id or note failure */
365 if (extent)
366 id = (id - extent->lower_first) + extent->first;
367 else
368 id = (u32) -1;
369
370 return id;
371}
372
373/**
374 * make_kuid - Map a user-namespace uid pair into a kuid.
375 * @ns: User namespace that the uid is in
376 * @uid: User identifier
377 *
378 * Maps a user-namespace uid pair into a kernel internal kuid,
379 * and returns that kuid.
380 *
381 * When there is no mapping defined for the user-namespace uid
382 * pair INVALID_UID is returned. Callers are expected to test
383 * for and handle INVALID_UID being returned. INVALID_UID
384 * may be tested for using uid_valid().
385 */
386kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
387{
388 /* Map the uid to a global kernel uid */
389 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
390}
391EXPORT_SYMBOL(make_kuid);
392
393/**
394 * from_kuid - Create a uid from a kuid user-namespace pair.
395 * @targ: The user namespace we want a uid in.
396 * @kuid: The kernel internal uid to start with.
397 *
398 * Map @kuid into the user-namespace specified by @targ and
399 * return the resulting uid.
400 *
401 * There is always a mapping into the initial user_namespace.
402 *
403 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
404 */
405uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
406{
407 /* Map the uid from a global kernel uid */
408 return map_id_up(&targ->uid_map, __kuid_val(kuid));
409}
410EXPORT_SYMBOL(from_kuid);
411
412/**
413 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
414 * @targ: The user namespace we want a uid in.
415 * @kuid: The kernel internal uid to start with.
416 *
417 * Map @kuid into the user-namespace specified by @targ and
418 * return the resulting uid.
419 *
420 * There is always a mapping into the initial user_namespace.
421 *
422 * Unlike from_kuid from_kuid_munged never fails and always
423 * returns a valid uid. This makes from_kuid_munged appropriate
424 * for use in syscalls like stat and getuid where failing the
425 * system call and failing to provide a valid uid are not an
426 * options.
427 *
428 * If @kuid has no mapping in @targ overflowuid is returned.
429 */
430uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
431{
432 uid_t uid;
433 uid = from_kuid(targ, kuid);
434
435 if (uid == (uid_t) -1)
436 uid = overflowuid;
437 return uid;
438}
439EXPORT_SYMBOL(from_kuid_munged);
440
441/**
442 * make_kgid - Map a user-namespace gid pair into a kgid.
443 * @ns: User namespace that the gid is in
444 * @gid: group identifier
445 *
446 * Maps a user-namespace gid pair into a kernel internal kgid,
447 * and returns that kgid.
448 *
449 * When there is no mapping defined for the user-namespace gid
450 * pair INVALID_GID is returned. Callers are expected to test
451 * for and handle INVALID_GID being returned. INVALID_GID may be
452 * tested for using gid_valid().
453 */
454kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
455{
456 /* Map the gid to a global kernel gid */
457 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
458}
459EXPORT_SYMBOL(make_kgid);
460
461/**
462 * from_kgid - Create a gid from a kgid user-namespace pair.
463 * @targ: The user namespace we want a gid in.
464 * @kgid: The kernel internal gid to start with.
465 *
466 * Map @kgid into the user-namespace specified by @targ and
467 * return the resulting gid.
468 *
469 * There is always a mapping into the initial user_namespace.
470 *
471 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
472 */
473gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
474{
475 /* Map the gid from a global kernel gid */
476 return map_id_up(&targ->gid_map, __kgid_val(kgid));
477}
478EXPORT_SYMBOL(from_kgid);
479
480/**
481 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
482 * @targ: The user namespace we want a gid in.
483 * @kgid: The kernel internal gid to start with.
484 *
485 * Map @kgid into the user-namespace specified by @targ and
486 * return the resulting gid.
487 *
488 * There is always a mapping into the initial user_namespace.
489 *
490 * Unlike from_kgid from_kgid_munged never fails and always
491 * returns a valid gid. This makes from_kgid_munged appropriate
492 * for use in syscalls like stat and getgid where failing the
493 * system call and failing to provide a valid gid are not options.
494 *
495 * If @kgid has no mapping in @targ overflowgid is returned.
496 */
497gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
498{
499 gid_t gid;
500 gid = from_kgid(targ, kgid);
501
502 if (gid == (gid_t) -1)
503 gid = overflowgid;
504 return gid;
505}
506EXPORT_SYMBOL(from_kgid_munged);
507
508/**
509 * make_kprojid - Map a user-namespace projid pair into a kprojid.
510 * @ns: User namespace that the projid is in
511 * @projid: Project identifier
512 *
513 * Maps a user-namespace uid pair into a kernel internal kuid,
514 * and returns that kuid.
515 *
516 * When there is no mapping defined for the user-namespace projid
517 * pair INVALID_PROJID is returned. Callers are expected to test
518 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
519 * may be tested for using projid_valid().
520 */
521kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
522{
523 /* Map the uid to a global kernel uid */
524 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
525}
526EXPORT_SYMBOL(make_kprojid);
527
528/**
529 * from_kprojid - Create a projid from a kprojid user-namespace pair.
530 * @targ: The user namespace we want a projid in.
531 * @kprojid: The kernel internal project identifier to start with.
532 *
533 * Map @kprojid into the user-namespace specified by @targ and
534 * return the resulting projid.
535 *
536 * There is always a mapping into the initial user_namespace.
537 *
538 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
539 */
540projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
541{
542 /* Map the uid from a global kernel uid */
543 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
544}
545EXPORT_SYMBOL(from_kprojid);
546
547/**
548 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
549 * @targ: The user namespace we want a projid in.
550 * @kprojid: The kernel internal projid to start with.
551 *
552 * Map @kprojid into the user-namespace specified by @targ and
553 * return the resulting projid.
554 *
555 * There is always a mapping into the initial user_namespace.
556 *
557 * Unlike from_kprojid from_kprojid_munged never fails and always
558 * returns a valid projid. This makes from_kprojid_munged
559 * appropriate for use in syscalls like stat and where
560 * failing the system call and failing to provide a valid projid are
561 * not an options.
562 *
563 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
564 */
565projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
566{
567 projid_t projid;
568 projid = from_kprojid(targ, kprojid);
569
570 if (projid == (projid_t) -1)
571 projid = OVERFLOW_PROJID;
572 return projid;
573}
574EXPORT_SYMBOL(from_kprojid_munged);
575
576
577static int uid_m_show(struct seq_file *seq, void *v)
578{
579 struct user_namespace *ns = seq->private;
580 struct uid_gid_extent *extent = v;
581 struct user_namespace *lower_ns;
582 uid_t lower;
583
584 lower_ns = seq_user_ns(seq);
585 if ((lower_ns == ns) && lower_ns->parent)
586 lower_ns = lower_ns->parent;
587
588 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
589
590 seq_printf(seq, "%10u %10u %10u\n",
591 extent->first,
592 lower,
593 extent->count);
594
595 return 0;
596}
597
598static int gid_m_show(struct seq_file *seq, void *v)
599{
600 struct user_namespace *ns = seq->private;
601 struct uid_gid_extent *extent = v;
602 struct user_namespace *lower_ns;
603 gid_t lower;
604
605 lower_ns = seq_user_ns(seq);
606 if ((lower_ns == ns) && lower_ns->parent)
607 lower_ns = lower_ns->parent;
608
609 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
610
611 seq_printf(seq, "%10u %10u %10u\n",
612 extent->first,
613 lower,
614 extent->count);
615
616 return 0;
617}
618
619static int projid_m_show(struct seq_file *seq, void *v)
620{
621 struct user_namespace *ns = seq->private;
622 struct uid_gid_extent *extent = v;
623 struct user_namespace *lower_ns;
624 projid_t lower;
625
626 lower_ns = seq_user_ns(seq);
627 if ((lower_ns == ns) && lower_ns->parent)
628 lower_ns = lower_ns->parent;
629
630 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
631
632 seq_printf(seq, "%10u %10u %10u\n",
633 extent->first,
634 lower,
635 extent->count);
636
637 return 0;
638}
639
640static void *m_start(struct seq_file *seq, loff_t *ppos,
641 struct uid_gid_map *map)
642{
643 loff_t pos = *ppos;
644 unsigned extents = map->nr_extents;
645 smp_rmb();
646
647 if (pos >= extents)
648 return NULL;
649
650 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
651 return &map->extent[pos];
652
653 return &map->forward[pos];
654}
655
656static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
657{
658 struct user_namespace *ns = seq->private;
659
660 return m_start(seq, ppos, &ns->uid_map);
661}
662
663static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
664{
665 struct user_namespace *ns = seq->private;
666
667 return m_start(seq, ppos, &ns->gid_map);
668}
669
670static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
671{
672 struct user_namespace *ns = seq->private;
673
674 return m_start(seq, ppos, &ns->projid_map);
675}
676
677static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
678{
679 (*pos)++;
680 return seq->op->start(seq, pos);
681}
682
683static void m_stop(struct seq_file *seq, void *v)
684{
685 return;
686}
687
688const struct seq_operations proc_uid_seq_operations = {
689 .start = uid_m_start,
690 .stop = m_stop,
691 .next = m_next,
692 .show = uid_m_show,
693};
694
695const struct seq_operations proc_gid_seq_operations = {
696 .start = gid_m_start,
697 .stop = m_stop,
698 .next = m_next,
699 .show = gid_m_show,
700};
701
702const struct seq_operations proc_projid_seq_operations = {
703 .start = projid_m_start,
704 .stop = m_stop,
705 .next = m_next,
706 .show = projid_m_show,
707};
708
709static bool mappings_overlap(struct uid_gid_map *new_map,
710 struct uid_gid_extent *extent)
711{
712 u32 upper_first, lower_first, upper_last, lower_last;
713 unsigned idx;
714
715 upper_first = extent->first;
716 lower_first = extent->lower_first;
717 upper_last = upper_first + extent->count - 1;
718 lower_last = lower_first + extent->count - 1;
719
720 for (idx = 0; idx < new_map->nr_extents; idx++) {
721 u32 prev_upper_first, prev_lower_first;
722 u32 prev_upper_last, prev_lower_last;
723 struct uid_gid_extent *prev;
724
725 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
726 prev = &new_map->extent[idx];
727 else
728 prev = &new_map->forward[idx];
729
730 prev_upper_first = prev->first;
731 prev_lower_first = prev->lower_first;
732 prev_upper_last = prev_upper_first + prev->count - 1;
733 prev_lower_last = prev_lower_first + prev->count - 1;
734
735 /* Does the upper range intersect a previous extent? */
736 if ((prev_upper_first <= upper_last) &&
737 (prev_upper_last >= upper_first))
738 return true;
739
740 /* Does the lower range intersect a previous extent? */
741 if ((prev_lower_first <= lower_last) &&
742 (prev_lower_last >= lower_first))
743 return true;
744 }
745 return false;
746}
747
748/**
749 * insert_extent - Safely insert a new idmap extent into struct uid_gid_map.
750 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
751 * UID_GID_MAP_MAX_BASE_EXTENTS.
752 */
753static int insert_extent(struct uid_gid_map *map, struct uid_gid_extent *extent)
754{
755 struct uid_gid_extent *dest;
756
757 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
758 struct uid_gid_extent *forward;
759
760 /* Allocate memory for 340 mappings. */
761 forward = kmalloc_array(UID_GID_MAP_MAX_EXTENTS,
762 sizeof(struct uid_gid_extent),
763 GFP_KERNEL);
764 if (!forward)
765 return -ENOMEM;
766
767 /* Copy over memory. Only set up memory for the forward pointer.
768 * Defer the memory setup for the reverse pointer.
769 */
770 memcpy(forward, map->extent,
771 map->nr_extents * sizeof(map->extent[0]));
772
773 map->forward = forward;
774 map->reverse = NULL;
775 }
776
777 if (map->nr_extents < UID_GID_MAP_MAX_BASE_EXTENTS)
778 dest = &map->extent[map->nr_extents];
779 else
780 dest = &map->forward[map->nr_extents];
781
782 *dest = *extent;
783 map->nr_extents++;
784 return 0;
785}
786
787/* cmp function to sort() forward mappings */
788static int cmp_extents_forward(const void *a, const void *b)
789{
790 const struct uid_gid_extent *e1 = a;
791 const struct uid_gid_extent *e2 = b;
792
793 if (e1->first < e2->first)
794 return -1;
795
796 if (e1->first > e2->first)
797 return 1;
798
799 return 0;
800}
801
802/* cmp function to sort() reverse mappings */
803static int cmp_extents_reverse(const void *a, const void *b)
804{
805 const struct uid_gid_extent *e1 = a;
806 const struct uid_gid_extent *e2 = b;
807
808 if (e1->lower_first < e2->lower_first)
809 return -1;
810
811 if (e1->lower_first > e2->lower_first)
812 return 1;
813
814 return 0;
815}
816
817/**
818 * sort_idmaps - Sorts an array of idmap entries.
819 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
820 */
821static int sort_idmaps(struct uid_gid_map *map)
822{
823 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
824 return 0;
825
826 /* Sort forward array. */
827 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
828 cmp_extents_forward, NULL);
829
830 /* Only copy the memory from forward we actually need. */
831 map->reverse = kmemdup(map->forward,
832 map->nr_extents * sizeof(struct uid_gid_extent),
833 GFP_KERNEL);
834 if (!map->reverse)
835 return -ENOMEM;
836
837 /* Sort reverse array. */
838 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
839 cmp_extents_reverse, NULL);
840
841 return 0;
842}
843
844static ssize_t map_write(struct file *file, const char __user *buf,
845 size_t count, loff_t *ppos,
846 int cap_setid,
847 struct uid_gid_map *map,
848 struct uid_gid_map *parent_map)
849{
850 struct seq_file *seq = file->private_data;
851 struct user_namespace *ns = seq->private;
852 struct uid_gid_map new_map;
853 unsigned idx;
854 struct uid_gid_extent extent;
855 char *kbuf = NULL, *pos, *next_line;
856 ssize_t ret;
857
858 /* Only allow < page size writes at the beginning of the file */
859 if ((*ppos != 0) || (count >= PAGE_SIZE))
860 return -EINVAL;
861
862 /* Slurp in the user data */
863 kbuf = memdup_user_nul(buf, count);
864 if (IS_ERR(kbuf))
865 return PTR_ERR(kbuf);
866
867 /*
868 * The userns_state_mutex serializes all writes to any given map.
869 *
870 * Any map is only ever written once.
871 *
872 * An id map fits within 1 cache line on most architectures.
873 *
874 * On read nothing needs to be done unless you are on an
875 * architecture with a crazy cache coherency model like alpha.
876 *
877 * There is a one time data dependency between reading the
878 * count of the extents and the values of the extents. The
879 * desired behavior is to see the values of the extents that
880 * were written before the count of the extents.
881 *
882 * To achieve this smp_wmb() is used on guarantee the write
883 * order and smp_rmb() is guaranteed that we don't have crazy
884 * architectures returning stale data.
885 */
886 mutex_lock(&userns_state_mutex);
887
888 memset(&new_map, 0, sizeof(struct uid_gid_map));
889
890 ret = -EPERM;
891 /* Only allow one successful write to the map */
892 if (map->nr_extents != 0)
893 goto out;
894
895 /*
896 * Adjusting namespace settings requires capabilities on the target.
897 */
898 if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
899 goto out;
900
901 /* Parse the user data */
902 ret = -EINVAL;
903 pos = kbuf;
904 for (; pos; pos = next_line) {
905
906 /* Find the end of line and ensure I don't look past it */
907 next_line = strchr(pos, '\n');
908 if (next_line) {
909 *next_line = '\0';
910 next_line++;
911 if (*next_line == '\0')
912 next_line = NULL;
913 }
914
915 pos = skip_spaces(pos);
916 extent.first = simple_strtoul(pos, &pos, 10);
917 if (!isspace(*pos))
918 goto out;
919
920 pos = skip_spaces(pos);
921 extent.lower_first = simple_strtoul(pos, &pos, 10);
922 if (!isspace(*pos))
923 goto out;
924
925 pos = skip_spaces(pos);
926 extent.count = simple_strtoul(pos, &pos, 10);
927 if (*pos && !isspace(*pos))
928 goto out;
929
930 /* Verify there is not trailing junk on the line */
931 pos = skip_spaces(pos);
932 if (*pos != '\0')
933 goto out;
934
935 /* Verify we have been given valid starting values */
936 if ((extent.first == (u32) -1) ||
937 (extent.lower_first == (u32) -1))
938 goto out;
939
940 /* Verify count is not zero and does not cause the
941 * extent to wrap
942 */
943 if ((extent.first + extent.count) <= extent.first)
944 goto out;
945 if ((extent.lower_first + extent.count) <=
946 extent.lower_first)
947 goto out;
948
949 /* Do the ranges in extent overlap any previous extents? */
950 if (mappings_overlap(&new_map, &extent))
951 goto out;
952
953 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
954 (next_line != NULL))
955 goto out;
956
957 ret = insert_extent(&new_map, &extent);
958 if (ret < 0)
959 goto out;
960 ret = -EINVAL;
961 }
962 /* Be very certaint the new map actually exists */
963 if (new_map.nr_extents == 0)
964 goto out;
965
966 ret = -EPERM;
967 /* Validate the user is allowed to use user id's mapped to. */
968 if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
969 goto out;
970
971 ret = -EPERM;
972 /* Map the lower ids from the parent user namespace to the
973 * kernel global id space.
974 */
975 for (idx = 0; idx < new_map.nr_extents; idx++) {
976 struct uid_gid_extent *e;
977 u32 lower_first;
978
979 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
980 e = &new_map.extent[idx];
981 else
982 e = &new_map.forward[idx];
983
984 lower_first = map_id_range_down(parent_map,
985 e->lower_first,
986 e->count);
987
988 /* Fail if we can not map the specified extent to
989 * the kernel global id space.
990 */
991 if (lower_first == (u32) -1)
992 goto out;
993
994 e->lower_first = lower_first;
995 }
996
997 /*
998 * If we want to use binary search for lookup, this clones the extent
999 * array and sorts both copies.
1000 */
1001 ret = sort_idmaps(&new_map);
1002 if (ret < 0)
1003 goto out;
1004
1005 /* Install the map */
1006 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1007 memcpy(map->extent, new_map.extent,
1008 new_map.nr_extents * sizeof(new_map.extent[0]));
1009 } else {
1010 map->forward = new_map.forward;
1011 map->reverse = new_map.reverse;
1012 }
1013 smp_wmb();
1014 map->nr_extents = new_map.nr_extents;
1015
1016 *ppos = count;
1017 ret = count;
1018out:
1019 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1020 kfree(new_map.forward);
1021 kfree(new_map.reverse);
1022 map->forward = NULL;
1023 map->reverse = NULL;
1024 map->nr_extents = 0;
1025 }
1026
1027 mutex_unlock(&userns_state_mutex);
1028 kfree(kbuf);
1029 return ret;
1030}
1031
1032ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1033 size_t size, loff_t *ppos)
1034{
1035 struct seq_file *seq = file->private_data;
1036 struct user_namespace *ns = seq->private;
1037 struct user_namespace *seq_ns = seq_user_ns(seq);
1038
1039 if (!ns->parent)
1040 return -EPERM;
1041
1042 if ((seq_ns != ns) && (seq_ns != ns->parent))
1043 return -EPERM;
1044
1045 return map_write(file, buf, size, ppos, CAP_SETUID,
1046 &ns->uid_map, &ns->parent->uid_map);
1047}
1048
1049ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1050 size_t size, loff_t *ppos)
1051{
1052 struct seq_file *seq = file->private_data;
1053 struct user_namespace *ns = seq->private;
1054 struct user_namespace *seq_ns = seq_user_ns(seq);
1055
1056 if (!ns->parent)
1057 return -EPERM;
1058
1059 if ((seq_ns != ns) && (seq_ns != ns->parent))
1060 return -EPERM;
1061
1062 return map_write(file, buf, size, ppos, CAP_SETGID,
1063 &ns->gid_map, &ns->parent->gid_map);
1064}
1065
1066ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1067 size_t size, loff_t *ppos)
1068{
1069 struct seq_file *seq = file->private_data;
1070 struct user_namespace *ns = seq->private;
1071 struct user_namespace *seq_ns = seq_user_ns(seq);
1072
1073 if (!ns->parent)
1074 return -EPERM;
1075
1076 if ((seq_ns != ns) && (seq_ns != ns->parent))
1077 return -EPERM;
1078
1079 /* Anyone can set any valid project id no capability needed */
1080 return map_write(file, buf, size, ppos, -1,
1081 &ns->projid_map, &ns->parent->projid_map);
1082}
1083
1084static bool new_idmap_permitted(const struct file *file,
1085 struct user_namespace *ns, int cap_setid,
1086 struct uid_gid_map *new_map)
1087{
1088 const struct cred *cred = file->f_cred;
1089 /* Don't allow mappings that would allow anything that wouldn't
1090 * be allowed without the establishment of unprivileged mappings.
1091 */
1092 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1093 uid_eq(ns->owner, cred->euid)) {
1094 u32 id = new_map->extent[0].lower_first;
1095 if (cap_setid == CAP_SETUID) {
1096 kuid_t uid = make_kuid(ns->parent, id);
1097 if (uid_eq(uid, cred->euid))
1098 return true;
1099 } else if (cap_setid == CAP_SETGID) {
1100 kgid_t gid = make_kgid(ns->parent, id);
1101 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1102 gid_eq(gid, cred->egid))
1103 return true;
1104 }
1105 }
1106
1107 /* Allow anyone to set a mapping that doesn't require privilege */
1108 if (!cap_valid(cap_setid))
1109 return true;
1110
1111 /* Allow the specified ids if we have the appropriate capability
1112 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1113 * And the opener of the id file also had the approprpiate capability.
1114 */
1115 if (ns_capable(ns->parent, cap_setid) &&
1116 file_ns_capable(file, ns->parent, cap_setid))
1117 return true;
1118
1119 return false;
1120}
1121
1122int proc_setgroups_show(struct seq_file *seq, void *v)
1123{
1124 struct user_namespace *ns = seq->private;
1125 unsigned long userns_flags = READ_ONCE(ns->flags);
1126
1127 seq_printf(seq, "%s\n",
1128 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1129 "allow" : "deny");
1130 return 0;
1131}
1132
1133ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1134 size_t count, loff_t *ppos)
1135{
1136 struct seq_file *seq = file->private_data;
1137 struct user_namespace *ns = seq->private;
1138 char kbuf[8], *pos;
1139 bool setgroups_allowed;
1140 ssize_t ret;
1141
1142 /* Only allow a very narrow range of strings to be written */
1143 ret = -EINVAL;
1144 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1145 goto out;
1146
1147 /* What was written? */
1148 ret = -EFAULT;
1149 if (copy_from_user(kbuf, buf, count))
1150 goto out;
1151 kbuf[count] = '\0';
1152 pos = kbuf;
1153
1154 /* What is being requested? */
1155 ret = -EINVAL;
1156 if (strncmp(pos, "allow", 5) == 0) {
1157 pos += 5;
1158 setgroups_allowed = true;
1159 }
1160 else if (strncmp(pos, "deny", 4) == 0) {
1161 pos += 4;
1162 setgroups_allowed = false;
1163 }
1164 else
1165 goto out;
1166
1167 /* Verify there is not trailing junk on the line */
1168 pos = skip_spaces(pos);
1169 if (*pos != '\0')
1170 goto out;
1171
1172 ret = -EPERM;
1173 mutex_lock(&userns_state_mutex);
1174 if (setgroups_allowed) {
1175 /* Enabling setgroups after setgroups has been disabled
1176 * is not allowed.
1177 */
1178 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1179 goto out_unlock;
1180 } else {
1181 /* Permanently disabling setgroups after setgroups has
1182 * been enabled by writing the gid_map is not allowed.
1183 */
1184 if (ns->gid_map.nr_extents != 0)
1185 goto out_unlock;
1186 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1187 }
1188 mutex_unlock(&userns_state_mutex);
1189
1190 /* Report a successful write */
1191 *ppos = count;
1192 ret = count;
1193out:
1194 return ret;
1195out_unlock:
1196 mutex_unlock(&userns_state_mutex);
1197 goto out;
1198}
1199
1200bool userns_may_setgroups(const struct user_namespace *ns)
1201{
1202 bool allowed;
1203
1204 mutex_lock(&userns_state_mutex);
1205 /* It is not safe to use setgroups until a gid mapping in
1206 * the user namespace has been established.
1207 */
1208 allowed = ns->gid_map.nr_extents != 0;
1209 /* Is setgroups allowed? */
1210 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1211 mutex_unlock(&userns_state_mutex);
1212
1213 return allowed;
1214}
1215
1216/*
1217 * Returns true if @child is the same namespace or a descendant of
1218 * @ancestor.
1219 */
1220bool in_userns(const struct user_namespace *ancestor,
1221 const struct user_namespace *child)
1222{
1223 const struct user_namespace *ns;
1224 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1225 ;
1226 return (ns == ancestor);
1227}
1228
1229bool current_in_userns(const struct user_namespace *target_ns)
1230{
1231 return in_userns(target_ns, current_user_ns());
1232}
1233EXPORT_SYMBOL(current_in_userns);
1234
1235static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1236{
1237 return container_of(ns, struct user_namespace, ns);
1238}
1239
1240static struct ns_common *userns_get(struct task_struct *task)
1241{
1242 struct user_namespace *user_ns;
1243
1244 rcu_read_lock();
1245 user_ns = get_user_ns(__task_cred(task)->user_ns);
1246 rcu_read_unlock();
1247
1248 return user_ns ? &user_ns->ns : NULL;
1249}
1250
1251static void userns_put(struct ns_common *ns)
1252{
1253 put_user_ns(to_user_ns(ns));
1254}
1255
1256static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1257{
1258 struct user_namespace *user_ns = to_user_ns(ns);
1259 struct cred *cred;
1260
1261 /* Don't allow gaining capabilities by reentering
1262 * the same user namespace.
1263 */
1264 if (user_ns == current_user_ns())
1265 return -EINVAL;
1266
1267 /* Tasks that share a thread group must share a user namespace */
1268 if (!thread_group_empty(current))
1269 return -EINVAL;
1270
1271 if (current->fs->users != 1)
1272 return -EINVAL;
1273
1274 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1275 return -EPERM;
1276
1277 cred = prepare_creds();
1278 if (!cred)
1279 return -ENOMEM;
1280
1281 put_user_ns(cred->user_ns);
1282 set_cred_user_ns(cred, get_user_ns(user_ns));
1283
1284 return commit_creds(cred);
1285}
1286
1287struct ns_common *ns_get_owner(struct ns_common *ns)
1288{
1289 struct user_namespace *my_user_ns = current_user_ns();
1290 struct user_namespace *owner, *p;
1291
1292 /* See if the owner is in the current user namespace */
1293 owner = p = ns->ops->owner(ns);
1294 for (;;) {
1295 if (!p)
1296 return ERR_PTR(-EPERM);
1297 if (p == my_user_ns)
1298 break;
1299 p = p->parent;
1300 }
1301
1302 return &get_user_ns(owner)->ns;
1303}
1304
1305static struct user_namespace *userns_owner(struct ns_common *ns)
1306{
1307 return to_user_ns(ns)->parent;
1308}
1309
1310const struct proc_ns_operations userns_operations = {
1311 .name = "user",
1312 .type = CLONE_NEWUSER,
1313 .get = userns_get,
1314 .put = userns_put,
1315 .install = userns_install,
1316 .owner = userns_owner,
1317 .get_parent = ns_get_owner,
1318};
1319
1320static __init int user_namespaces_init(void)
1321{
1322 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
1323 return 0;
1324}
1325subsys_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(map->forward,
857 map->nr_extents * sizeof(struct uid_gid_extent),
858 GFP_KERNEL);
859 if (!map->reverse)
860 return -ENOMEM;
861
862 /* Sort reverse array. */
863 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
864 cmp_extents_reverse, NULL);
865
866 return 0;
867}
868
869/**
870 * verify_root_map() - check the uid 0 mapping
871 * @file: idmapping file
872 * @map_ns: user namespace of the target process
873 * @new_map: requested idmap
874 *
875 * If a process requests mapping parent uid 0 into the new ns, verify that the
876 * process writing the map had the CAP_SETFCAP capability as the target process
877 * will be able to write fscaps that are valid in ancestor user namespaces.
878 *
879 * Return: true if the mapping is allowed, false if not.
880 */
881static bool verify_root_map(const struct file *file,
882 struct user_namespace *map_ns,
883 struct uid_gid_map *new_map)
884{
885 int idx;
886 const struct user_namespace *file_ns = file->f_cred->user_ns;
887 struct uid_gid_extent *extent0 = NULL;
888
889 for (idx = 0; idx < new_map->nr_extents; idx++) {
890 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
891 extent0 = &new_map->extent[idx];
892 else
893 extent0 = &new_map->forward[idx];
894 if (extent0->lower_first == 0)
895 break;
896
897 extent0 = NULL;
898 }
899
900 if (!extent0)
901 return true;
902
903 if (map_ns == file_ns) {
904 /* The process unshared its ns and is writing to its own
905 * /proc/self/uid_map. User already has full capabilites in
906 * the new namespace. Verify that the parent had CAP_SETFCAP
907 * when it unshared.
908 * */
909 if (!file_ns->parent_could_setfcap)
910 return false;
911 } else {
912 /* Process p1 is writing to uid_map of p2, who is in a child
913 * user namespace to p1's. Verify that the opener of the map
914 * file has CAP_SETFCAP against the parent of the new map
915 * namespace */
916 if (!file_ns_capable(file, map_ns->parent, CAP_SETFCAP))
917 return false;
918 }
919
920 return true;
921}
922
923static ssize_t map_write(struct file *file, const char __user *buf,
924 size_t count, loff_t *ppos,
925 int cap_setid,
926 struct uid_gid_map *map,
927 struct uid_gid_map *parent_map)
928{
929 struct seq_file *seq = file->private_data;
930 struct user_namespace *map_ns = seq->private;
931 struct uid_gid_map new_map;
932 unsigned idx;
933 struct uid_gid_extent extent;
934 char *kbuf, *pos, *next_line;
935 ssize_t ret;
936
937 /* Only allow < page size writes at the beginning of the file */
938 if ((*ppos != 0) || (count >= PAGE_SIZE))
939 return -EINVAL;
940
941 /* Slurp in the user data */
942 kbuf = memdup_user_nul(buf, count);
943 if (IS_ERR(kbuf))
944 return PTR_ERR(kbuf);
945
946 /*
947 * The userns_state_mutex serializes all writes to any given map.
948 *
949 * Any map is only ever written once.
950 *
951 * An id map fits within 1 cache line on most architectures.
952 *
953 * On read nothing needs to be done unless you are on an
954 * architecture with a crazy cache coherency model like alpha.
955 *
956 * There is a one time data dependency between reading the
957 * count of the extents and the values of the extents. The
958 * desired behavior is to see the values of the extents that
959 * were written before the count of the extents.
960 *
961 * To achieve this smp_wmb() is used on guarantee the write
962 * order and smp_rmb() is guaranteed that we don't have crazy
963 * architectures returning stale data.
964 */
965 mutex_lock(&userns_state_mutex);
966
967 memset(&new_map, 0, sizeof(struct uid_gid_map));
968
969 ret = -EPERM;
970 /* Only allow one successful write to the map */
971 if (map->nr_extents != 0)
972 goto out;
973
974 /*
975 * Adjusting namespace settings requires capabilities on the target.
976 */
977 if (cap_valid(cap_setid) && !file_ns_capable(file, map_ns, CAP_SYS_ADMIN))
978 goto out;
979
980 /* Parse the user data */
981 ret = -EINVAL;
982 pos = kbuf;
983 for (; pos; pos = next_line) {
984
985 /* Find the end of line and ensure I don't look past it */
986 next_line = strchr(pos, '\n');
987 if (next_line) {
988 *next_line = '\0';
989 next_line++;
990 if (*next_line == '\0')
991 next_line = NULL;
992 }
993
994 pos = skip_spaces(pos);
995 extent.first = simple_strtoul(pos, &pos, 10);
996 if (!isspace(*pos))
997 goto out;
998
999 pos = skip_spaces(pos);
1000 extent.lower_first = simple_strtoul(pos, &pos, 10);
1001 if (!isspace(*pos))
1002 goto out;
1003
1004 pos = skip_spaces(pos);
1005 extent.count = simple_strtoul(pos, &pos, 10);
1006 if (*pos && !isspace(*pos))
1007 goto out;
1008
1009 /* Verify there is not trailing junk on the line */
1010 pos = skip_spaces(pos);
1011 if (*pos != '\0')
1012 goto out;
1013
1014 /* Verify we have been given valid starting values */
1015 if ((extent.first == (u32) -1) ||
1016 (extent.lower_first == (u32) -1))
1017 goto out;
1018
1019 /* Verify count is not zero and does not cause the
1020 * extent to wrap
1021 */
1022 if ((extent.first + extent.count) <= extent.first)
1023 goto out;
1024 if ((extent.lower_first + extent.count) <=
1025 extent.lower_first)
1026 goto out;
1027
1028 /* Do the ranges in extent overlap any previous extents? */
1029 if (mappings_overlap(&new_map, &extent))
1030 goto out;
1031
1032 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
1033 (next_line != NULL))
1034 goto out;
1035
1036 ret = insert_extent(&new_map, &extent);
1037 if (ret < 0)
1038 goto out;
1039 ret = -EINVAL;
1040 }
1041 /* Be very certain the new map actually exists */
1042 if (new_map.nr_extents == 0)
1043 goto out;
1044
1045 ret = -EPERM;
1046 /* Validate the user is allowed to use user id's mapped to. */
1047 if (!new_idmap_permitted(file, map_ns, cap_setid, &new_map))
1048 goto out;
1049
1050 ret = -EPERM;
1051 /* Map the lower ids from the parent user namespace to the
1052 * kernel global id space.
1053 */
1054 for (idx = 0; idx < new_map.nr_extents; idx++) {
1055 struct uid_gid_extent *e;
1056 u32 lower_first;
1057
1058 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
1059 e = &new_map.extent[idx];
1060 else
1061 e = &new_map.forward[idx];
1062
1063 lower_first = map_id_range_down(parent_map,
1064 e->lower_first,
1065 e->count);
1066
1067 /* Fail if we can not map the specified extent to
1068 * the kernel global id space.
1069 */
1070 if (lower_first == (u32) -1)
1071 goto out;
1072
1073 e->lower_first = lower_first;
1074 }
1075
1076 /*
1077 * If we want to use binary search for lookup, this clones the extent
1078 * array and sorts both copies.
1079 */
1080 ret = sort_idmaps(&new_map);
1081 if (ret < 0)
1082 goto out;
1083
1084 /* Install the map */
1085 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1086 memcpy(map->extent, new_map.extent,
1087 new_map.nr_extents * sizeof(new_map.extent[0]));
1088 } else {
1089 map->forward = new_map.forward;
1090 map->reverse = new_map.reverse;
1091 }
1092 smp_wmb();
1093 map->nr_extents = new_map.nr_extents;
1094
1095 *ppos = count;
1096 ret = count;
1097out:
1098 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1099 kfree(new_map.forward);
1100 kfree(new_map.reverse);
1101 map->forward = NULL;
1102 map->reverse = NULL;
1103 map->nr_extents = 0;
1104 }
1105
1106 mutex_unlock(&userns_state_mutex);
1107 kfree(kbuf);
1108 return ret;
1109}
1110
1111ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1112 size_t size, loff_t *ppos)
1113{
1114 struct seq_file *seq = file->private_data;
1115 struct user_namespace *ns = seq->private;
1116 struct user_namespace *seq_ns = seq_user_ns(seq);
1117
1118 if (!ns->parent)
1119 return -EPERM;
1120
1121 if ((seq_ns != ns) && (seq_ns != ns->parent))
1122 return -EPERM;
1123
1124 return map_write(file, buf, size, ppos, CAP_SETUID,
1125 &ns->uid_map, &ns->parent->uid_map);
1126}
1127
1128ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1129 size_t size, loff_t *ppos)
1130{
1131 struct seq_file *seq = file->private_data;
1132 struct user_namespace *ns = seq->private;
1133 struct user_namespace *seq_ns = seq_user_ns(seq);
1134
1135 if (!ns->parent)
1136 return -EPERM;
1137
1138 if ((seq_ns != ns) && (seq_ns != ns->parent))
1139 return -EPERM;
1140
1141 return map_write(file, buf, size, ppos, CAP_SETGID,
1142 &ns->gid_map, &ns->parent->gid_map);
1143}
1144
1145ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1146 size_t size, loff_t *ppos)
1147{
1148 struct seq_file *seq = file->private_data;
1149 struct user_namespace *ns = seq->private;
1150 struct user_namespace *seq_ns = seq_user_ns(seq);
1151
1152 if (!ns->parent)
1153 return -EPERM;
1154
1155 if ((seq_ns != ns) && (seq_ns != ns->parent))
1156 return -EPERM;
1157
1158 /* Anyone can set any valid project id no capability needed */
1159 return map_write(file, buf, size, ppos, -1,
1160 &ns->projid_map, &ns->parent->projid_map);
1161}
1162
1163static bool new_idmap_permitted(const struct file *file,
1164 struct user_namespace *ns, int cap_setid,
1165 struct uid_gid_map *new_map)
1166{
1167 const struct cred *cred = file->f_cred;
1168
1169 if (cap_setid == CAP_SETUID && !verify_root_map(file, ns, new_map))
1170 return false;
1171
1172 /* Don't allow mappings that would allow anything that wouldn't
1173 * be allowed without the establishment of unprivileged mappings.
1174 */
1175 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1176 uid_eq(ns->owner, cred->euid)) {
1177 u32 id = new_map->extent[0].lower_first;
1178 if (cap_setid == CAP_SETUID) {
1179 kuid_t uid = make_kuid(ns->parent, id);
1180 if (uid_eq(uid, cred->euid))
1181 return true;
1182 } else if (cap_setid == CAP_SETGID) {
1183 kgid_t gid = make_kgid(ns->parent, id);
1184 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1185 gid_eq(gid, cred->egid))
1186 return true;
1187 }
1188 }
1189
1190 /* Allow anyone to set a mapping that doesn't require privilege */
1191 if (!cap_valid(cap_setid))
1192 return true;
1193
1194 /* Allow the specified ids if we have the appropriate capability
1195 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1196 * And the opener of the id file also has the appropriate capability.
1197 */
1198 if (ns_capable(ns->parent, cap_setid) &&
1199 file_ns_capable(file, ns->parent, cap_setid))
1200 return true;
1201
1202 return false;
1203}
1204
1205int proc_setgroups_show(struct seq_file *seq, void *v)
1206{
1207 struct user_namespace *ns = seq->private;
1208 unsigned long userns_flags = READ_ONCE(ns->flags);
1209
1210 seq_printf(seq, "%s\n",
1211 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1212 "allow" : "deny");
1213 return 0;
1214}
1215
1216ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1217 size_t count, loff_t *ppos)
1218{
1219 struct seq_file *seq = file->private_data;
1220 struct user_namespace *ns = seq->private;
1221 char kbuf[8], *pos;
1222 bool setgroups_allowed;
1223 ssize_t ret;
1224
1225 /* Only allow a very narrow range of strings to be written */
1226 ret = -EINVAL;
1227 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1228 goto out;
1229
1230 /* What was written? */
1231 ret = -EFAULT;
1232 if (copy_from_user(kbuf, buf, count))
1233 goto out;
1234 kbuf[count] = '\0';
1235 pos = kbuf;
1236
1237 /* What is being requested? */
1238 ret = -EINVAL;
1239 if (strncmp(pos, "allow", 5) == 0) {
1240 pos += 5;
1241 setgroups_allowed = true;
1242 }
1243 else if (strncmp(pos, "deny", 4) == 0) {
1244 pos += 4;
1245 setgroups_allowed = false;
1246 }
1247 else
1248 goto out;
1249
1250 /* Verify there is not trailing junk on the line */
1251 pos = skip_spaces(pos);
1252 if (*pos != '\0')
1253 goto out;
1254
1255 ret = -EPERM;
1256 mutex_lock(&userns_state_mutex);
1257 if (setgroups_allowed) {
1258 /* Enabling setgroups after setgroups has been disabled
1259 * is not allowed.
1260 */
1261 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1262 goto out_unlock;
1263 } else {
1264 /* Permanently disabling setgroups after setgroups has
1265 * been enabled by writing the gid_map is not allowed.
1266 */
1267 if (ns->gid_map.nr_extents != 0)
1268 goto out_unlock;
1269 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1270 }
1271 mutex_unlock(&userns_state_mutex);
1272
1273 /* Report a successful write */
1274 *ppos = count;
1275 ret = count;
1276out:
1277 return ret;
1278out_unlock:
1279 mutex_unlock(&userns_state_mutex);
1280 goto out;
1281}
1282
1283bool userns_may_setgroups(const struct user_namespace *ns)
1284{
1285 bool allowed;
1286
1287 mutex_lock(&userns_state_mutex);
1288 /* It is not safe to use setgroups until a gid mapping in
1289 * the user namespace has been established.
1290 */
1291 allowed = ns->gid_map.nr_extents != 0;
1292 /* Is setgroups allowed? */
1293 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1294 mutex_unlock(&userns_state_mutex);
1295
1296 return allowed;
1297}
1298
1299/*
1300 * Returns true if @child is the same namespace or a descendant of
1301 * @ancestor.
1302 */
1303bool in_userns(const struct user_namespace *ancestor,
1304 const struct user_namespace *child)
1305{
1306 const struct user_namespace *ns;
1307 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1308 ;
1309 return (ns == ancestor);
1310}
1311
1312bool current_in_userns(const struct user_namespace *target_ns)
1313{
1314 return in_userns(target_ns, current_user_ns());
1315}
1316EXPORT_SYMBOL(current_in_userns);
1317
1318static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1319{
1320 return container_of(ns, struct user_namespace, ns);
1321}
1322
1323static struct ns_common *userns_get(struct task_struct *task)
1324{
1325 struct user_namespace *user_ns;
1326
1327 rcu_read_lock();
1328 user_ns = get_user_ns(__task_cred(task)->user_ns);
1329 rcu_read_unlock();
1330
1331 return user_ns ? &user_ns->ns : NULL;
1332}
1333
1334static void userns_put(struct ns_common *ns)
1335{
1336 put_user_ns(to_user_ns(ns));
1337}
1338
1339static int userns_install(struct nsset *nsset, struct ns_common *ns)
1340{
1341 struct user_namespace *user_ns = to_user_ns(ns);
1342 struct cred *cred;
1343
1344 /* Don't allow gaining capabilities by reentering
1345 * the same user namespace.
1346 */
1347 if (user_ns == current_user_ns())
1348 return -EINVAL;
1349
1350 /* Tasks that share a thread group must share a user namespace */
1351 if (!thread_group_empty(current))
1352 return -EINVAL;
1353
1354 if (current->fs->users != 1)
1355 return -EINVAL;
1356
1357 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1358 return -EPERM;
1359
1360 cred = nsset_cred(nsset);
1361 if (!cred)
1362 return -EINVAL;
1363
1364 put_user_ns(cred->user_ns);
1365 set_cred_user_ns(cred, get_user_ns(user_ns));
1366
1367 if (set_cred_ucounts(cred) < 0)
1368 return -EINVAL;
1369
1370 return 0;
1371}
1372
1373struct ns_common *ns_get_owner(struct ns_common *ns)
1374{
1375 struct user_namespace *my_user_ns = current_user_ns();
1376 struct user_namespace *owner, *p;
1377
1378 /* See if the owner is in the current user namespace */
1379 owner = p = ns->ops->owner(ns);
1380 for (;;) {
1381 if (!p)
1382 return ERR_PTR(-EPERM);
1383 if (p == my_user_ns)
1384 break;
1385 p = p->parent;
1386 }
1387
1388 return &get_user_ns(owner)->ns;
1389}
1390
1391static struct user_namespace *userns_owner(struct ns_common *ns)
1392{
1393 return to_user_ns(ns)->parent;
1394}
1395
1396const struct proc_ns_operations userns_operations = {
1397 .name = "user",
1398 .type = CLONE_NEWUSER,
1399 .get = userns_get,
1400 .put = userns_put,
1401 .install = userns_install,
1402 .owner = userns_owner,
1403 .get_parent = ns_get_owner,
1404};
1405
1406static __init int user_namespaces_init(void)
1407{
1408 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC | SLAB_ACCOUNT);
1409 return 0;
1410}
1411subsys_initcall(user_namespaces_init);