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
25 static struct kmem_cache *user_ns_cachep __ro_after_init;
26 static DEFINE_MUTEX(userns_state_mutex);
27
28 static bool new_idmap_permitted(const struct file *file,
29 struct user_namespace *ns, int cap_setid,
30 struct uid_gid_map *map);
31 static void free_user_ns(struct work_struct *work);
32
33 static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
34 {
35 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
36 }
37
38 static void dec_user_namespaces(struct ucounts *ucounts)
39 {
40 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
41 }
42
43 static 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
62 static 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 */
82 int 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;
163 fail_keyring:
164 #ifdef CONFIG_PERSISTENT_KEYRINGS
165 key_put(ns->persistent_keyring_register);
166 #endif
167 ns_free_inum(&ns->ns);
168 fail_free:
169 kmem_cache_free(user_ns_cachep, ns);
170 fail_dec:
171 dec_user_namespaces(ucounts);
172 fail:
173 return ret;
174 }
175
176 int 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
196 static 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
228 void __put_user_ns(struct user_namespace *ns)
229 {
230 schedule_work(&ns->work);
231 }
232 EXPORT_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 */
238 struct 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 */
248 static 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 */
278 static struct uid_gid_extent *
279 map_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 */
296 static struct uid_gid_extent *
297 map_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
315 static 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
335 u32 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 */
345 static struct uid_gid_extent *
346 map_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 */
365 static struct uid_gid_extent *
366 map_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
378 u32 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 */
411 kuid_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 }
416 EXPORT_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 */
430 uid_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 }
435 EXPORT_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 */
455 uid_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 }
464 EXPORT_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 */
479 kgid_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 }
484 EXPORT_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 */
498 gid_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 }
503 EXPORT_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 */
522 gid_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 }
531 EXPORT_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 */
546 kprojid_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 }
551 EXPORT_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 */
565 projid_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 }
570 EXPORT_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 */
590 projid_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 }
599 EXPORT_SYMBOL(from_kprojid_munged);
600
601
602 static 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
623 static 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
644 static 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
665 static 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
681 static 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
688 static 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
695 static 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
702 static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
703 {
704 (*pos)++;
705 return seq->op->start(seq, pos);
706 }
707
708 static void m_stop(struct seq_file *seq, void *v)
709 {
710 return;
711 }
712
713 const 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
720 const 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
727 const 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
734 static 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 */
778 static 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 */
813 static 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 */
828 static 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 */
846 static 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 */
881 static 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
923 static 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;
1097 out:
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
1111 ssize_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
1128 ssize_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
1145 ssize_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
1163 static 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
1205 int 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
1216 ssize_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;
1276 out:
1277 return ret;
1278 out_unlock:
1279 mutex_unlock(&userns_state_mutex);
1280 goto out;
1281 }
1282
1283 bool 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 */
1303 bool 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
1312 bool current_in_userns(const struct user_namespace *target_ns)
1313 {
1314 return in_userns(target_ns, current_user_ns());
1315 }
1316 EXPORT_SYMBOL(current_in_userns);
1317
1318 static inline struct user_namespace *to_user_ns(struct ns_common *ns)
1319 {
1320 return container_of(ns, struct user_namespace, ns);
1321 }
1322
1323 static 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
1334 static void userns_put(struct ns_common *ns)
1335 {
1336 put_user_ns(to_user_ns(ns));
1337 }
1338
1339 static 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
1373 struct 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
1391 static struct user_namespace *userns_owner(struct ns_common *ns)
1392 {
1393 return to_user_ns(ns)->parent;
1394 }
1395
1396 const 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
1406 static __init int user_namespaces_init(void)
1407 {
1408 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC | SLAB_ACCOUNT);
1409 return 0;
1410 }
1411 subsys_initcall(user_namespaces_init);
1412
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.