projects/linux·blob master · refs

1444 lines · 36.5 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2
3#include <kunit/visibility.h>
4#include <linux/export.h>
5#include <linux/nsproxy.h>
6#include <linux/slab.h>
7#include <linux/sched/signal.h>
8#include <linux/user_namespace.h>
9#include <linux/proc_ns.h>
10#include <linux/highuid.h>
11#include <linux/cred.h>
12#include <linux/securebits.h>
13#include <linux/security.h>
14#include <linux/keyctl.h>
15#include <linux/key-type.h>
16#include <keys/user-type.h>
17#include <linux/seq_file.h>
18#include <linux/fs.h>
19#include <linux/uaccess.h>
20#include <linux/ctype.h>
21#include <linux/projid.h>
22#include <linux/fs_struct.h>
23#include <linux/bsearch.h>
24#include <linux/sort.h>
25#include <linux/nstree.h>
26
27static struct kmem_cache *user_ns_cachep __ro_after_init;
28static DEFINE_MUTEX(userns_state_mutex);
29
30static bool new_idmap_permitted(const struct file *file,
31 struct user_namespace *ns, int cap_setid,
32 struct uid_gid_map *map);
33static void free_user_ns(struct work_struct *work);
34
35static struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
36{
37 return inc_ucount(ns, uid, UCOUNT_USER_NAMESPACES);
38}
39
40static void dec_user_namespaces(struct ucounts *ucounts)
41{
42 return dec_ucount(ucounts, UCOUNT_USER_NAMESPACES);
43}
44
45static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
46{
47 /* Start with the same capabilities as init but useless for doing
48 * anything as the capabilities are bound to the new user namespace.
49 */
50 cred->securebits = SECUREBITS_DEFAULT;
51 cred->cap_inheritable = CAP_EMPTY_SET;
52 cred->cap_permitted = CAP_FULL_SET;
53 cred->cap_effective = CAP_FULL_SET;
54 cred->cap_ambient = CAP_EMPTY_SET;
55 cred->cap_bset = CAP_FULL_SET;
56#ifdef CONFIG_KEYS
57 key_put(cred->request_key_auth);
58 cred->request_key_auth = NULL;
59#endif
60 /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
61 cred->user_ns = user_ns;
62}
63
64static unsigned long enforced_nproc_rlimit(void)
65{
66 unsigned long limit = RLIM_INFINITY;
67
68 /* Is RLIMIT_NPROC currently enforced? */
69 if (!uid_eq(current_uid(), GLOBAL_ROOT_UID) ||
70 (current_user_ns() != &init_user_ns))
71 limit = rlimit(RLIMIT_NPROC);
72
73 return limit;
74}
75
76/*
77 * Create a new user namespace, deriving the creator from the user in the
78 * passed credentials, and replacing that user with the new root user for the
79 * new namespace.
80 *
81 * This is called by copy_creds(), which will finish setting the target task's
82 * credentials.
83 */
84int create_user_ns(struct cred *new)
85{
86 struct user_namespace *ns, *parent_ns = new->user_ns;
87 kuid_t owner = new->euid;
88 kgid_t group = new->egid;
89 struct ucounts *ucounts;
90 int ret, i;
91
92 ret = -ENOSPC;
93 if (parent_ns->level > 32)
94 goto fail;
95
96 ucounts = inc_user_namespaces(parent_ns, owner);
97 if (!ucounts)
98 goto fail;
99
100 /*
101 * Verify that we can not violate the policy of which files
102 * may be accessed that is specified by the root directory,
103 * by verifying that the root directory is at the root of the
104 * mount namespace which allows all files to be accessed.
105 */
106 ret = -EPERM;
107 if (current_chrooted())
108 goto fail_dec;
109
110 /* The creator needs a mapping in the parent user namespace
111 * or else we won't be able to reasonably tell userspace who
112 * created a user_namespace.
113 */
114 ret = -EPERM;
115 if (!kuid_has_mapping(parent_ns, owner) ||
116 !kgid_has_mapping(parent_ns, group))
117 goto fail_dec;
118
119 ret = security_create_user_ns(new);
120 if (ret < 0)
121 goto fail_dec;
122
123 ret = -ENOMEM;
124 ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
125 if (!ns)
126 goto fail_dec;
127
128 ns->parent_could_setfcap = cap_raised(new->cap_effective, CAP_SETFCAP);
129
130 ret = ns_common_init(ns);
131 if (ret)
132 goto fail_free;
133
134 /* Leave the new->user_ns reference with the new user namespace. */
135 ns->parent = parent_ns;
136 ns->level = parent_ns->level + 1;
137 ns->owner = owner;
138 ns->group = group;
139 INIT_WORK(&ns->work, free_user_ns);
140 for (i = 0; i < UCOUNT_COUNTS; i++) {
141 ns->ucount_max[i] = INT_MAX;
142 }
143 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_NPROC, enforced_nproc_rlimit());
144 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_MSGQUEUE, rlimit(RLIMIT_MSGQUEUE));
145 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_SIGPENDING, rlimit(RLIMIT_SIGPENDING));
146 set_userns_rlimit_max(ns, UCOUNT_RLIMIT_MEMLOCK, rlimit(RLIMIT_MEMLOCK));
147 ns->ucounts = ucounts;
148
149 /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */
150 mutex_lock(&userns_state_mutex);
151 ns->flags = parent_ns->flags;
152 mutex_unlock(&userns_state_mutex);
153
154#ifdef CONFIG_KEYS
155 INIT_LIST_HEAD(&ns->keyring_name_list);
156 init_rwsem(&ns->keyring_sem);
157#endif
158 ret = -ENOMEM;
159 if (!setup_userns_sysctls(ns))
160 goto fail_keyring;
161
162 set_cred_user_ns(new, ns);
163 ns_tree_add(ns);
164 return 0;
165fail_keyring:
166 ns_common_free(ns);
167fail_free:
168 kmem_cache_free(user_ns_cachep, ns);
169fail_dec:
170 dec_user_namespaces(ucounts);
171fail:
172 return ret;
173}
174
175int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
176{
177 struct cred *cred;
178 int err = -ENOMEM;
179
180 if (!(unshare_flags & CLONE_NEWUSER))
181 return 0;
182
183 cred = prepare_creds();
184 if (cred) {
185 err = create_user_ns(cred);
186 if (err)
187 put_cred(cred);
188 else
189 *new_cred = cred;
190 }
191
192 return err;
193}
194
195static void free_user_ns(struct work_struct *work)
196{
197 struct user_namespace *parent, *ns =
198 container_of(work, struct user_namespace, work);
199
200 do {
201 struct ucounts *ucounts = ns->ucounts;
202 parent = ns->parent;
203 ns_tree_remove(ns);
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_common_free(ns);
222 /* Concurrent nstree traversal depends on a grace period. */
223 kfree_rcu(ns, ns.ns_rcu);
224 dec_user_namespaces(ucounts);
225 ns = parent;
226 } while (ns_ref_put(parent));
227}
228
229void __put_user_ns(struct user_namespace *ns)
230{
231 schedule_work(&ns->work);
232}
233EXPORT_SYMBOL(__put_user_ns);
234
235/*
236 * struct idmap_key - holds the information necessary to find an idmapping in a
237 * sorted idmap array. It is passed to cmp_map_id() as first argument.
238 */
239struct idmap_key {
240 bool map_up; /* true -> id from kid; false -> kid from id */
241 u32 id; /* id to find */
242 u32 count;
243};
244
245/*
246 * cmp_map_id - Function to be passed to bsearch() to find the requested
247 * idmapping. Expects struct idmap_key to be passed via @k.
248 */
249static int cmp_map_id(const void *k, const void *e)
250{
251 u32 first, last, id2;
252 const struct idmap_key *key = k;
253 const struct uid_gid_extent *el = e;
254
255 id2 = key->id + key->count - 1;
256
257 /* handle map_id_{down,up}() */
258 if (key->map_up)
259 first = el->lower_first;
260 else
261 first = el->first;
262
263 last = first + el->count - 1;
264
265 if (key->id >= first && key->id <= last &&
266 (id2 >= first && id2 <= last))
267 return 0;
268
269 if (key->id < first || id2 < first)
270 return -1;
271
272 return 1;
273}
274
275/*
276 * map_id_range_down_max - Find idmap via binary search in ordered idmap array.
277 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
278 */
279static const struct uid_gid_extent *
280map_id_range_down_max(unsigned extents, const struct uid_gid_map *map, u32 id, u32 count)
281{
282 struct idmap_key key;
283
284 key.map_up = false;
285 key.count = count;
286 key.id = id;
287
288 return bsearch(&key, map->forward, extents,
289 sizeof(struct uid_gid_extent), cmp_map_id);
290}
291
292/*
293 * map_id_range_down_base - Find idmap via binary search in static extent array.
294 * Can only be called if number of mappings is equal or less than
295 * UID_GID_MAP_MAX_BASE_EXTENTS.
296 */
297static const struct uid_gid_extent *
298map_id_range_down_base(unsigned extents, const struct uid_gid_map *map, u32 id, u32 count)
299{
300 unsigned idx;
301 u32 first, last, id2;
302
303 id2 = id + count - 1;
304
305 /* Find the matching extent */
306 for (idx = 0; idx < extents; idx++) {
307 first = map->extent[idx].first;
308 last = first + map->extent[idx].count - 1;
309 if (id >= first && id <= last &&
310 (id2 >= first && id2 <= last))
311 return &map->extent[idx];
312 }
313 return NULL;
314}
315
316static u32 map_id_range_down(const struct uid_gid_map *map, u32 id, u32 count)
317{
318 const struct uid_gid_extent *extent;
319 unsigned extents = map->nr_extents;
320 smp_rmb();
321
322 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
323 extent = map_id_range_down_base(extents, map, id, count);
324 else
325 extent = map_id_range_down_max(extents, map, id, count);
326
327 /* Map the id or note failure */
328 if (extent)
329 id = (id - extent->first) + extent->lower_first;
330 else
331 id = (u32) -1;
332
333 return id;
334}
335
336u32 map_id_down(const struct uid_gid_map *map, u32 id)
337{
338 return map_id_range_down(map, id, 1);
339}
340
341/*
342 * map_id_up_base - Find idmap via binary search in static extent array.
343 * Can only be called if number of mappings is equal or less than
344 * UID_GID_MAP_MAX_BASE_EXTENTS.
345 */
346static const struct uid_gid_extent *
347map_id_range_up_base(unsigned extents, const struct uid_gid_map *map, u32 id, u32 count)
348{
349 unsigned idx;
350 u32 first, last, id2;
351
352 id2 = id + count - 1;
353
354 /* Find the matching extent */
355 for (idx = 0; idx < extents; idx++) {
356 first = map->extent[idx].lower_first;
357 last = first + map->extent[idx].count - 1;
358 if (id >= first && id <= last &&
359 (id2 >= first && id2 <= last))
360 return &map->extent[idx];
361 }
362 return NULL;
363}
364
365/*
366 * map_id_up_max - Find idmap via binary search in ordered idmap array.
367 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
368 */
369static const struct uid_gid_extent *
370map_id_range_up_max(unsigned extents, const struct uid_gid_map *map, u32 id, u32 count)
371{
372 struct idmap_key key;
373
374 key.map_up = true;
375 key.count = count;
376 key.id = id;
377
378 return bsearch(&key, map->reverse, extents,
379 sizeof(struct uid_gid_extent), cmp_map_id);
380}
381
382u32 map_id_range_up(const struct uid_gid_map *map, u32 id, u32 count)
383{
384 const struct uid_gid_extent *extent;
385 unsigned extents = map->nr_extents;
386 smp_rmb();
387
388 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
389 extent = map_id_range_up_base(extents, map, id, count);
390 else
391 extent = map_id_range_up_max(extents, map, id, count);
392
393 /* Map the id or note failure */
394 if (extent)
395 id = (id - extent->lower_first) + extent->first;
396 else
397 id = (u32) -1;
398
399 return id;
400}
401
402u32 map_id_up(const struct uid_gid_map *map, u32 id)
403{
404 return map_id_range_up(map, id, 1);
405}
406
407/**
408 * make_kuid - Map a user-namespace uid pair into a kuid.
409 * @ns: User namespace that the uid is in
410 * @uid: User identifier
411 *
412 * Maps a user-namespace uid pair into a kernel internal kuid,
413 * and returns that kuid.
414 *
415 * When there is no mapping defined for the user-namespace uid
416 * pair INVALID_UID is returned. Callers are expected to test
417 * for and handle INVALID_UID being returned. INVALID_UID
418 * may be tested for using uid_valid().
419 */
420kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
421{
422 /* Map the uid to a global kernel uid */
423 return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
424}
425EXPORT_SYMBOL(make_kuid);
426
427/**
428 * from_kuid - Create a uid from a kuid user-namespace pair.
429 * @targ: The user namespace we want a uid in.
430 * @kuid: The kernel internal uid to start with.
431 *
432 * Map @kuid into the user-namespace specified by @targ and
433 * return the resulting uid.
434 *
435 * There is always a mapping into the initial user_namespace.
436 *
437 * If @kuid has no mapping in @targ (uid_t)-1 is returned.
438 */
439uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
440{
441 /* Map the uid from a global kernel uid */
442 return map_id_up(&targ->uid_map, __kuid_val(kuid));
443}
444EXPORT_SYMBOL(from_kuid);
445
446/**
447 * from_kuid_munged - Create a uid from a kuid user-namespace pair.
448 * @targ: The user namespace we want a uid in.
449 * @kuid: The kernel internal uid to start with.
450 *
451 * Map @kuid into the user-namespace specified by @targ and
452 * return the resulting uid.
453 *
454 * There is always a mapping into the initial user_namespace.
455 *
456 * Unlike from_kuid from_kuid_munged never fails and always
457 * returns a valid uid. This makes from_kuid_munged appropriate
458 * for use in syscalls like stat and getuid where failing the
459 * system call and failing to provide a valid uid are not an
460 * options.
461 *
462 * If @kuid has no mapping in @targ overflowuid is returned.
463 */
464uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
465{
466 uid_t uid;
467 uid = from_kuid(targ, kuid);
468
469 if (uid == (uid_t) -1)
470 uid = overflowuid;
471 return uid;
472}
473EXPORT_SYMBOL(from_kuid_munged);
474
475/**
476 * make_kgid - Map a user-namespace gid pair into a kgid.
477 * @ns: User namespace that the gid is in
478 * @gid: group identifier
479 *
480 * Maps a user-namespace gid pair into a kernel internal kgid,
481 * and returns that kgid.
482 *
483 * When there is no mapping defined for the user-namespace gid
484 * pair INVALID_GID is returned. Callers are expected to test
485 * for and handle INVALID_GID being returned. INVALID_GID may be
486 * tested for using gid_valid().
487 */
488kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
489{
490 /* Map the gid to a global kernel gid */
491 return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
492}
493EXPORT_SYMBOL(make_kgid);
494
495/**
496 * from_kgid - Create a gid from a kgid user-namespace pair.
497 * @targ: The user namespace we want a gid in.
498 * @kgid: The kernel internal gid to start with.
499 *
500 * Map @kgid into the user-namespace specified by @targ and
501 * return the resulting gid.
502 *
503 * There is always a mapping into the initial user_namespace.
504 *
505 * If @kgid has no mapping in @targ (gid_t)-1 is returned.
506 */
507gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
508{
509 /* Map the gid from a global kernel gid */
510 return map_id_up(&targ->gid_map, __kgid_val(kgid));
511}
512EXPORT_SYMBOL(from_kgid);
513
514/**
515 * from_kgid_munged - Create a gid from a kgid user-namespace pair.
516 * @targ: The user namespace we want a gid in.
517 * @kgid: The kernel internal gid to start with.
518 *
519 * Map @kgid into the user-namespace specified by @targ and
520 * return the resulting gid.
521 *
522 * There is always a mapping into the initial user_namespace.
523 *
524 * Unlike from_kgid from_kgid_munged never fails and always
525 * returns a valid gid. This makes from_kgid_munged appropriate
526 * for use in syscalls like stat and getgid where failing the
527 * system call and failing to provide a valid gid are not options.
528 *
529 * If @kgid has no mapping in @targ overflowgid is returned.
530 */
531gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
532{
533 gid_t gid;
534 gid = from_kgid(targ, kgid);
535
536 if (gid == (gid_t) -1)
537 gid = overflowgid;
538 return gid;
539}
540EXPORT_SYMBOL(from_kgid_munged);
541
542/**
543 * make_kprojid - Map a user-namespace projid pair into a kprojid.
544 * @ns: User namespace that the projid is in
545 * @projid: Project identifier
546 *
547 * Maps a user-namespace uid pair into a kernel internal kuid,
548 * and returns that kuid.
549 *
550 * When there is no mapping defined for the user-namespace projid
551 * pair INVALID_PROJID is returned. Callers are expected to test
552 * for and handle INVALID_PROJID being returned. INVALID_PROJID
553 * may be tested for using projid_valid().
554 */
555kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
556{
557 /* Map the uid to a global kernel uid */
558 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
559}
560EXPORT_SYMBOL(make_kprojid);
561
562/**
563 * from_kprojid - Create a projid from a kprojid user-namespace pair.
564 * @targ: The user namespace we want a projid in.
565 * @kprojid: The kernel internal project identifier to start with.
566 *
567 * Map @kprojid into the user-namespace specified by @targ and
568 * return the resulting projid.
569 *
570 * There is always a mapping into the initial user_namespace.
571 *
572 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
573 */
574projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
575{
576 /* Map the uid from a global kernel uid */
577 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
578}
579EXPORT_SYMBOL(from_kprojid);
580
581/**
582 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
583 * @targ: The user namespace we want a projid in.
584 * @kprojid: The kernel internal projid to start with.
585 *
586 * Map @kprojid into the user-namespace specified by @targ and
587 * return the resulting projid.
588 *
589 * There is always a mapping into the initial user_namespace.
590 *
591 * Unlike from_kprojid from_kprojid_munged never fails and always
592 * returns a valid projid. This makes from_kprojid_munged
593 * appropriate for use in syscalls like stat and where
594 * failing the system call and failing to provide a valid projid are
595 * not an options.
596 *
597 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
598 */
599projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
600{
601 projid_t projid;
602 projid = from_kprojid(targ, kprojid);
603
604 if (projid == (projid_t) -1)
605 projid = OVERFLOW_PROJID;
606 return projid;
607}
608EXPORT_SYMBOL(from_kprojid_munged);
609
610
611static int uid_m_show(struct seq_file *seq, void *v)
612{
613 struct user_namespace *ns = seq->private;
614 struct uid_gid_extent *extent = v;
615 struct user_namespace *lower_ns;
616 uid_t lower;
617
618 lower_ns = seq_user_ns(seq);
619 if ((lower_ns == ns) && lower_ns->parent)
620 lower_ns = lower_ns->parent;
621
622 lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
623
624 seq_printf(seq, "%10u %10u %10u\n",
625 extent->first,
626 lower,
627 extent->count);
628
629 return 0;
630}
631
632static int gid_m_show(struct seq_file *seq, void *v)
633{
634 struct user_namespace *ns = seq->private;
635 struct uid_gid_extent *extent = v;
636 struct user_namespace *lower_ns;
637 gid_t lower;
638
639 lower_ns = seq_user_ns(seq);
640 if ((lower_ns == ns) && lower_ns->parent)
641 lower_ns = lower_ns->parent;
642
643 lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
644
645 seq_printf(seq, "%10u %10u %10u\n",
646 extent->first,
647 lower,
648 extent->count);
649
650 return 0;
651}
652
653static int projid_m_show(struct seq_file *seq, void *v)
654{
655 struct user_namespace *ns = seq->private;
656 struct uid_gid_extent *extent = v;
657 struct user_namespace *lower_ns;
658 projid_t lower;
659
660 lower_ns = seq_user_ns(seq);
661 if ((lower_ns == ns) && lower_ns->parent)
662 lower_ns = lower_ns->parent;
663
664 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
665
666 seq_printf(seq, "%10u %10u %10u\n",
667 extent->first,
668 lower,
669 extent->count);
670
671 return 0;
672}
673
674static void *m_start(struct seq_file *seq, loff_t *ppos,
675 struct uid_gid_map *map)
676{
677 loff_t pos = *ppos;
678 unsigned extents = map->nr_extents;
679 smp_rmb();
680
681 if (pos >= extents)
682 return NULL;
683
684 if (extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
685 return &map->extent[pos];
686
687 return &map->forward[pos];
688}
689
690static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
691{
692 struct user_namespace *ns = seq->private;
693
694 return m_start(seq, ppos, &ns->uid_map);
695}
696
697static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
698{
699 struct user_namespace *ns = seq->private;
700
701 return m_start(seq, ppos, &ns->gid_map);
702}
703
704static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
705{
706 struct user_namespace *ns = seq->private;
707
708 return m_start(seq, ppos, &ns->projid_map);
709}
710
711static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
712{
713 (*pos)++;
714 return seq->op->start(seq, pos);
715}
716
717static void m_stop(struct seq_file *seq, void *v)
718{
719 return;
720}
721
722const struct seq_operations proc_uid_seq_operations = {
723 .start = uid_m_start,
724 .stop = m_stop,
725 .next = m_next,
726 .show = uid_m_show,
727};
728
729const struct seq_operations proc_gid_seq_operations = {
730 .start = gid_m_start,
731 .stop = m_stop,
732 .next = m_next,
733 .show = gid_m_show,
734};
735
736const struct seq_operations proc_projid_seq_operations = {
737 .start = projid_m_start,
738 .stop = m_stop,
739 .next = m_next,
740 .show = projid_m_show,
741};
742
743static bool mappings_overlap(struct uid_gid_map *new_map,
744 struct uid_gid_extent *extent)
745{
746 u32 upper_first, lower_first, upper_last, lower_last;
747 unsigned idx;
748
749 upper_first = extent->first;
750 lower_first = extent->lower_first;
751 upper_last = upper_first + extent->count - 1;
752 lower_last = lower_first + extent->count - 1;
753
754 for (idx = 0; idx < new_map->nr_extents; idx++) {
755 u32 prev_upper_first, prev_lower_first;
756 u32 prev_upper_last, prev_lower_last;
757 struct uid_gid_extent *prev;
758
759 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
760 prev = &new_map->extent[idx];
761 else
762 prev = &new_map->forward[idx];
763
764 prev_upper_first = prev->first;
765 prev_lower_first = prev->lower_first;
766 prev_upper_last = prev_upper_first + prev->count - 1;
767 prev_lower_last = prev_lower_first + prev->count - 1;
768
769 /* Does the upper range intersect a previous extent? */
770 if ((prev_upper_first <= upper_last) &&
771 (prev_upper_last >= upper_first))
772 return true;
773
774 /* Does the lower range intersect a previous extent? */
775 if ((prev_lower_first <= lower_last) &&
776 (prev_lower_last >= lower_first))
777 return true;
778 }
779 return false;
780}
781
782/*
783 * uid_gid_map_insert_extent - Safely insert a new idmap extent into
784 * struct uid_gid_map.
785 * Takes care to allocate a 4K block of memory if the number of mappings exceeds
786 * UID_GID_MAP_MAX_BASE_EXTENTS.
787 */
788VISIBLE_IF_KUNIT int uid_gid_map_insert_extent(struct uid_gid_map *map,
789 struct uid_gid_extent *extent)
790{
791 struct uid_gid_extent *dest;
792
793 if (map->nr_extents == UID_GID_MAP_MAX_BASE_EXTENTS) {
794 struct uid_gid_extent *forward;
795
796 /* Allocate memory for 340 mappings. */
797 forward = kmalloc_objs(struct uid_gid_extent,
798 UID_GID_MAP_MAX_EXTENTS);
799 if (!forward)
800 return -ENOMEM;
801
802 /* Copy over memory. Only set up memory for the forward pointer.
803 * Defer the memory setup for the reverse pointer.
804 */
805 memcpy(forward, map->extent,
806 map->nr_extents * sizeof(map->extent[0]));
807
808 map->forward = forward;
809 map->reverse = NULL;
810 }
811
812 /*
813 * nr_extents must be updated before the extent and forward arrays are
814 * accessed, otherwise KSAN will assert an out-of-bounds error.
815 */
816 map->nr_extents++;
817 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
818 dest = &map->extent[map->nr_extents - 1];
819 else
820 dest = &map->forward[map->nr_extents - 1];
821
822 *dest = *extent;
823 return 0;
824}
825EXPORT_SYMBOL_IF_KUNIT(uid_gid_map_insert_extent);
826
827/* cmp function to sort() forward mappings */
828static int cmp_extents_forward(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->first < e2->first)
834 return -1;
835
836 if (e1->first > e2->first)
837 return 1;
838
839 return 0;
840}
841
842/* cmp function to sort() reverse mappings */
843static int cmp_extents_reverse(const void *a, const void *b)
844{
845 const struct uid_gid_extent *e1 = a;
846 const struct uid_gid_extent *e2 = b;
847
848 if (e1->lower_first < e2->lower_first)
849 return -1;
850
851 if (e1->lower_first > e2->lower_first)
852 return 1;
853
854 return 0;
855}
856
857/*
858 * uid_gid_map_sort - Sorts an array of idmap entries.
859 * Can only be called if number of mappings exceeds UID_GID_MAP_MAX_BASE_EXTENTS.
860 */
861VISIBLE_IF_KUNIT int uid_gid_map_sort(struct uid_gid_map *map)
862{
863 if (map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
864 return 0;
865
866 /* Sort forward array. */
867 sort(map->forward, map->nr_extents, sizeof(struct uid_gid_extent),
868 cmp_extents_forward, NULL);
869
870 /* Only copy the memory from forward we actually need. */
871 map->reverse = kmemdup_array(map->forward, map->nr_extents,
872 sizeof(struct uid_gid_extent), GFP_KERNEL);
873 if (!map->reverse)
874 return -ENOMEM;
875
876 /* Sort reverse array. */
877 sort(map->reverse, map->nr_extents, sizeof(struct uid_gid_extent),
878 cmp_extents_reverse, NULL);
879
880 return 0;
881}
882EXPORT_SYMBOL_IF_KUNIT(uid_gid_map_sort);
883
884/**
885 * verify_root_map() - check the uid 0 mapping
886 * @file: idmapping file
887 * @map_ns: user namespace of the target process
888 * @new_map: requested idmap
889 *
890 * If a process requests mapping parent uid 0 into the new ns, verify that the
891 * process writing the map had the CAP_SETFCAP capability as the target process
892 * will be able to write fscaps that are valid in ancestor user namespaces.
893 *
894 * Return: true if the mapping is allowed, false if not.
895 */
896static bool verify_root_map(const struct file *file,
897 struct user_namespace *map_ns,
898 struct uid_gid_map *new_map)
899{
900 int idx;
901 const struct user_namespace *file_ns = file->f_cred->user_ns;
902 struct uid_gid_extent *extent0 = NULL;
903
904 for (idx = 0; idx < new_map->nr_extents; idx++) {
905 if (new_map->nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
906 extent0 = &new_map->extent[idx];
907 else
908 extent0 = &new_map->forward[idx];
909 if (extent0->lower_first == 0)
910 break;
911
912 extent0 = NULL;
913 }
914
915 if (!extent0)
916 return true;
917
918 if (map_ns == file_ns) {
919 /* The process unshared its ns and is writing to its own
920 * /proc/self/uid_map. User already has full capabilites in
921 * the new namespace. Verify that the parent had CAP_SETFCAP
922 * when it unshared.
923 * */
924 if (!file_ns->parent_could_setfcap)
925 return false;
926 } else {
927 /* Process p1 is writing to uid_map of p2, who is in a child
928 * user namespace to p1's. Verify that the opener of the map
929 * file has CAP_SETFCAP against the parent of the new map
930 * namespace */
931 if (!file_ns_capable(file, map_ns->parent, CAP_SETFCAP))
932 return false;
933 }
934
935 return true;
936}
937
938/*
939 * Parse the decimal u32 at *pos, up to the next whitespace or NUL, and
940 * leave *pos pointing at that character.
941 */
942static int map_parse_u32(char **pos, u32 *res)
943{
944 char *end = *pos;
945 char c;
946 int ret;
947
948 while (*end && !isspace(*end))
949 end++;
950 c = *end;
951 *end = '\0';
952 ret = kstrtou32(*pos, 10, res);
953 *end = c;
954 *pos = end;
955 return ret;
956}
957
958static ssize_t map_write(struct file *file, const char __user *buf,
959 size_t count, loff_t *ppos,
960 int cap_setid,
961 struct uid_gid_map *map,
962 struct uid_gid_map *parent_map)
963{
964 struct seq_file *seq = file->private_data;
965 struct user_namespace *map_ns = seq->private;
966 struct uid_gid_map new_map;
967 unsigned idx;
968 struct uid_gid_extent extent;
969 char *kbuf, *pos, *next_line;
970 ssize_t ret;
971
972 /* Only allow < page size writes at the beginning of the file */
973 if ((*ppos != 0) || (count >= PAGE_SIZE))
974 return -EINVAL;
975
976 /* Slurp in the user data */
977 kbuf = memdup_user_nul(buf, count);
978 if (IS_ERR(kbuf))
979 return PTR_ERR(kbuf);
980
981 /*
982 * The userns_state_mutex serializes all writes to any given map.
983 *
984 * Any map is only ever written once.
985 *
986 * An id map fits within 1 cache line on most architectures.
987 *
988 * On read nothing needs to be done unless you are on an
989 * architecture with a crazy cache coherency model like alpha.
990 *
991 * There is a one time data dependency between reading the
992 * count of the extents and the values of the extents. The
993 * desired behavior is to see the values of the extents that
994 * were written before the count of the extents.
995 *
996 * To achieve this smp_wmb() is used on guarantee the write
997 * order and smp_rmb() is guaranteed that we don't have crazy
998 * architectures returning stale data.
999 */
1000 mutex_lock(&userns_state_mutex);
1001
1002 memset(&new_map, 0, sizeof(struct uid_gid_map));
1003
1004 ret = -EPERM;
1005 /* Only allow one successful write to the map */
1006 if (map->nr_extents != 0)
1007 goto out;
1008
1009 /*
1010 * Adjusting namespace settings requires capabilities on the target.
1011 */
1012 if (cap_valid(cap_setid) && !file_ns_capable(file, map_ns, CAP_SYS_ADMIN))
1013 goto out;
1014
1015 /* Parse the user data */
1016 ret = -EINVAL;
1017 pos = kbuf;
1018 for (; pos; pos = next_line) {
1019
1020 /* Find the end of line and ensure I don't look past it */
1021 next_line = strchr(pos, '\n');
1022 if (next_line) {
1023 *next_line = '\0';
1024 next_line++;
1025 if (*next_line == '\0')
1026 next_line = NULL;
1027 }
1028
1029 pos = skip_spaces(pos);
1030 if (map_parse_u32(&pos, &extent.first))
1031 goto out;
1032 if (!isspace(*pos))
1033 goto out;
1034
1035 pos = skip_spaces(pos);
1036 if (map_parse_u32(&pos, &extent.lower_first))
1037 goto out;
1038 if (!isspace(*pos))
1039 goto out;
1040
1041 pos = skip_spaces(pos);
1042 if (map_parse_u32(&pos, &extent.count))
1043 goto out;
1044 if (*pos && !isspace(*pos))
1045 goto out;
1046
1047 /* Verify there is not trailing junk on the line */
1048 pos = skip_spaces(pos);
1049 if (*pos != '\0')
1050 goto out;
1051
1052 /* Verify we have been given valid starting values */
1053 if ((extent.first == (u32) -1) ||
1054 (extent.lower_first == (u32) -1))
1055 goto out;
1056
1057 /* Verify count is not zero and does not cause the
1058 * extent to wrap
1059 */
1060 if ((extent.first + extent.count) <= extent.first)
1061 goto out;
1062 if ((extent.lower_first + extent.count) <=
1063 extent.lower_first)
1064 goto out;
1065
1066 /* Do the ranges in extent overlap any previous extents? */
1067 if (mappings_overlap(&new_map, &extent))
1068 goto out;
1069
1070 if ((new_map.nr_extents + 1) == UID_GID_MAP_MAX_EXTENTS &&
1071 (next_line != NULL))
1072 goto out;
1073
1074 ret = uid_gid_map_insert_extent(&new_map, &extent);
1075 if (ret < 0)
1076 goto out;
1077 ret = -EINVAL;
1078 }
1079 /* Be very certain the new map actually exists */
1080 if (new_map.nr_extents == 0)
1081 goto out;
1082
1083 ret = -EPERM;
1084 /* Validate the user is allowed to use user id's mapped to. */
1085 if (!new_idmap_permitted(file, map_ns, cap_setid, &new_map))
1086 goto out;
1087
1088 ret = -EPERM;
1089 /* Map the lower ids from the parent user namespace to the
1090 * kernel global id space.
1091 */
1092 for (idx = 0; idx < new_map.nr_extents; idx++) {
1093 struct uid_gid_extent *e;
1094 u32 lower_first;
1095
1096 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS)
1097 e = &new_map.extent[idx];
1098 else
1099 e = &new_map.forward[idx];
1100
1101 lower_first = map_id_range_down(parent_map,
1102 e->lower_first,
1103 e->count);
1104
1105 /* Fail if we can not map the specified extent to
1106 * the kernel global id space.
1107 */
1108 if (lower_first == (u32) -1)
1109 goto out;
1110
1111 e->lower_first = lower_first;
1112 }
1113
1114 /*
1115 * If we want to use binary search for lookup, this clones the extent
1116 * array and sorts both copies.
1117 */
1118 ret = uid_gid_map_sort(&new_map);
1119 if (ret < 0)
1120 goto out;
1121
1122 /* Install the map */
1123 if (new_map.nr_extents <= UID_GID_MAP_MAX_BASE_EXTENTS) {
1124 memcpy(map->extent, new_map.extent,
1125 new_map.nr_extents * sizeof(new_map.extent[0]));
1126 } else {
1127 map->forward = new_map.forward;
1128 map->reverse = new_map.reverse;
1129 }
1130 smp_wmb();
1131 map->nr_extents = new_map.nr_extents;
1132
1133 *ppos = count;
1134 ret = count;
1135out:
1136 if (ret < 0 && new_map.nr_extents > UID_GID_MAP_MAX_BASE_EXTENTS) {
1137 kfree(new_map.forward);
1138 kfree(new_map.reverse);
1139 map->forward = NULL;
1140 map->reverse = NULL;
1141 map->nr_extents = 0;
1142 }
1143
1144 mutex_unlock(&userns_state_mutex);
1145 kfree(kbuf);
1146 return ret;
1147}
1148
1149ssize_t proc_uid_map_write(struct file *file, const char __user *buf,
1150 size_t size, loff_t *ppos)
1151{
1152 struct seq_file *seq = file->private_data;
1153 struct user_namespace *ns = seq->private;
1154 struct user_namespace *seq_ns = seq_user_ns(seq);
1155
1156 if (!ns->parent)
1157 return -EPERM;
1158
1159 if ((seq_ns != ns) && (seq_ns != ns->parent))
1160 return -EPERM;
1161
1162 return map_write(file, buf, size, ppos, CAP_SETUID,
1163 &ns->uid_map, &ns->parent->uid_map);
1164}
1165
1166ssize_t proc_gid_map_write(struct file *file, const char __user *buf,
1167 size_t size, loff_t *ppos)
1168{
1169 struct seq_file *seq = file->private_data;
1170 struct user_namespace *ns = seq->private;
1171 struct user_namespace *seq_ns = seq_user_ns(seq);
1172
1173 if (!ns->parent)
1174 return -EPERM;
1175
1176 if ((seq_ns != ns) && (seq_ns != ns->parent))
1177 return -EPERM;
1178
1179 return map_write(file, buf, size, ppos, CAP_SETGID,
1180 &ns->gid_map, &ns->parent->gid_map);
1181}
1182
1183ssize_t proc_projid_map_write(struct file *file, const char __user *buf,
1184 size_t size, loff_t *ppos)
1185{
1186 struct seq_file *seq = file->private_data;
1187 struct user_namespace *ns = seq->private;
1188 struct user_namespace *seq_ns = seq_user_ns(seq);
1189
1190 if (!ns->parent)
1191 return -EPERM;
1192
1193 if ((seq_ns != ns) && (seq_ns != ns->parent))
1194 return -EPERM;
1195
1196 /* Anyone can set any valid project id no capability needed */
1197 return map_write(file, buf, size, ppos, -1,
1198 &ns->projid_map, &ns->parent->projid_map);
1199}
1200
1201static bool new_idmap_permitted(const struct file *file,
1202 struct user_namespace *ns, int cap_setid,
1203 struct uid_gid_map *new_map)
1204{
1205 const struct cred *cred = file->f_cred;
1206
1207 if (cap_setid == CAP_SETUID && !verify_root_map(file, ns, new_map))
1208 return false;
1209
1210 /* Don't allow mappings that would allow anything that wouldn't
1211 * be allowed without the establishment of unprivileged mappings.
1212 */
1213 if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) &&
1214 uid_eq(ns->owner, cred->euid)) {
1215 u32 id = new_map->extent[0].lower_first;
1216 if (cap_setid == CAP_SETUID) {
1217 kuid_t uid = make_kuid(ns->parent, id);
1218 if (uid_eq(uid, cred->euid))
1219 return true;
1220 } else if (cap_setid == CAP_SETGID) {
1221 kgid_t gid = make_kgid(ns->parent, id);
1222 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) &&
1223 gid_eq(gid, cred->egid))
1224 return true;
1225 }
1226 }
1227
1228 /* Allow anyone to set a mapping that doesn't require privilege */
1229 if (!cap_valid(cap_setid))
1230 return true;
1231
1232 /* Allow the specified ids if we have the appropriate capability
1233 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
1234 * And the opener of the id file also has the appropriate capability.
1235 */
1236 if (ns_capable(ns->parent, cap_setid) &&
1237 file_ns_capable(file, ns->parent, cap_setid))
1238 return true;
1239
1240 return false;
1241}
1242
1243int proc_setgroups_show(struct seq_file *seq, void *v)
1244{
1245 struct user_namespace *ns = seq->private;
1246 unsigned long userns_flags = READ_ONCE(ns->flags);
1247
1248 seq_printf(seq, "%s\n",
1249 (userns_flags & USERNS_SETGROUPS_ALLOWED) ?
1250 "allow" : "deny");
1251 return 0;
1252}
1253
1254ssize_t proc_setgroups_write(struct file *file, const char __user *buf,
1255 size_t count, loff_t *ppos)
1256{
1257 struct seq_file *seq = file->private_data;
1258 struct user_namespace *ns = seq->private;
1259 char kbuf[8], *pos;
1260 bool setgroups_allowed;
1261 ssize_t ret;
1262
1263 /* Only allow a very narrow range of strings to be written */
1264 ret = -EINVAL;
1265 if ((*ppos != 0) || (count >= sizeof(kbuf)))
1266 goto out;
1267
1268 /* What was written? */
1269 ret = -EFAULT;
1270 if (copy_from_user(kbuf, buf, count))
1271 goto out;
1272 kbuf[count] = '\0';
1273 pos = kbuf;
1274
1275 /* What is being requested? */
1276 ret = -EINVAL;
1277 if (strncmp(pos, "allow", 5) == 0) {
1278 pos += 5;
1279 setgroups_allowed = true;
1280 }
1281 else if (strncmp(pos, "deny", 4) == 0) {
1282 pos += 4;
1283 setgroups_allowed = false;
1284 }
1285 else
1286 goto out;
1287
1288 /* Verify there is not trailing junk on the line */
1289 pos = skip_spaces(pos);
1290 if (*pos != '\0')
1291 goto out;
1292
1293 ret = -EPERM;
1294 mutex_lock(&userns_state_mutex);
1295 if (setgroups_allowed) {
1296 /* Enabling setgroups after setgroups has been disabled
1297 * is not allowed.
1298 */
1299 if (!(ns->flags & USERNS_SETGROUPS_ALLOWED))
1300 goto out_unlock;
1301 } else {
1302 /* Permanently disabling setgroups after setgroups has
1303 * been enabled by writing the gid_map is not allowed.
1304 */
1305 if (ns->gid_map.nr_extents != 0)
1306 goto out_unlock;
1307 ns->flags &= ~USERNS_SETGROUPS_ALLOWED;
1308 }
1309 mutex_unlock(&userns_state_mutex);
1310
1311 /* Report a successful write */
1312 *ppos = count;
1313 ret = count;
1314out:
1315 return ret;
1316out_unlock:
1317 mutex_unlock(&userns_state_mutex);
1318 goto out;
1319}
1320
1321bool userns_may_setgroups(const struct user_namespace *ns)
1322{
1323 bool allowed;
1324
1325 mutex_lock(&userns_state_mutex);
1326 /* It is not safe to use setgroups until a gid mapping in
1327 * the user namespace has been established.
1328 */
1329 allowed = ns->gid_map.nr_extents != 0;
1330 /* Is setgroups allowed? */
1331 allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED);
1332 mutex_unlock(&userns_state_mutex);
1333
1334 return allowed;
1335}
1336
1337/*
1338 * Returns true if @child is the same namespace or a descendant of
1339 * @ancestor.
1340 */
1341bool in_userns(const struct user_namespace *ancestor,
1342 const struct user_namespace *child)
1343{
1344 const struct user_namespace *ns;
1345 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
1346 ;
1347 return (ns == ancestor);
1348}
1349
1350bool current_in_userns(const struct user_namespace *target_ns)
1351{
1352 return in_userns(target_ns, current_user_ns());
1353}
1354EXPORT_SYMBOL(current_in_userns);
1355
1356static struct ns_common *userns_get(struct task_struct *task)
1357{
1358 struct user_namespace *user_ns;
1359
1360 rcu_read_lock();
1361 user_ns = get_user_ns(__task_cred(task)->user_ns);
1362 rcu_read_unlock();
1363
1364 return user_ns ? &user_ns->ns : NULL;
1365}
1366
1367static void userns_put(struct ns_common *ns)
1368{
1369 put_user_ns(to_user_ns(ns));
1370}
1371
1372static int userns_install(struct nsset *nsset, struct ns_common *ns)
1373{
1374 struct user_namespace *user_ns = to_user_ns(ns);
1375 struct cred *cred;
1376
1377 /* Don't allow gaining capabilities by reentering
1378 * the same user namespace.
1379 */
1380 if (user_ns == current_user_ns())
1381 return -EINVAL;
1382
1383 /* Tasks that share a thread group must share a user namespace */
1384 if (!thread_group_empty(current))
1385 return -EINVAL;
1386
1387 if (current->fs->users != 1)
1388 return -EINVAL;
1389
1390 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1391 return -EPERM;
1392
1393 cred = nsset_cred(nsset);
1394 if (!cred)
1395 return -EINVAL;
1396
1397 put_user_ns(cred->user_ns);
1398 set_cred_user_ns(cred, get_user_ns(user_ns));
1399
1400 if (set_cred_ucounts(cred) < 0)
1401 return -EINVAL;
1402
1403 return 0;
1404}
1405
1406struct ns_common *ns_get_owner(struct ns_common *ns)
1407{
1408 struct user_namespace *my_user_ns = current_user_ns();
1409 struct user_namespace *owner, *p;
1410
1411 /* See if the owner is in the current user namespace */
1412 owner = p = ns->ops->owner(ns);
1413 for (;;) {
1414 if (!p)
1415 return ERR_PTR(-EPERM);
1416 if (p == my_user_ns)
1417 break;
1418 p = p->parent;
1419 }
1420
1421 return &get_user_ns(owner)->ns;
1422}
1423
1424static struct user_namespace *userns_owner(struct ns_common *ns)
1425{
1426 return to_user_ns(ns)->parent;
1427}
1428
1429const struct proc_ns_operations userns_operations = {
1430 .name = "user",
1431 .get = userns_get,
1432 .put = userns_put,
1433 .install = userns_install,
1434 .owner = userns_owner,
1435 .get_parent = ns_get_owner,
1436};
1437
1438static __init int user_namespaces_init(void)
1439{
1440 user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC | SLAB_ACCOUNT);
1441 ns_tree_add(&init_user_ns);
1442 return 0;
1443}
1444subsys_initcall(user_namespaces_init);