Commit | Line | Data |
---|---|---|
acce292c CLG |
1 | /* |
2 | * This program is free software; you can redistribute it and/or | |
3 | * modify it under the terms of the GNU General Public License as | |
4 | * published by the Free Software Foundation, version 2 of the | |
5 | * License. | |
6 | */ | |
7 | ||
9984de1a | 8 | #include <linux/export.h> |
acce292c | 9 | #include <linux/nsproxy.h> |
1aeb272c | 10 | #include <linux/slab.h> |
acce292c | 11 | #include <linux/user_namespace.h> |
0bb80f24 | 12 | #include <linux/proc_ns.h> |
5c1469de | 13 | #include <linux/highuid.h> |
18b6e041 | 14 | #include <linux/cred.h> |
973c5914 | 15 | #include <linux/securebits.h> |
22d917d8 EB |
16 | #include <linux/keyctl.h> |
17 | #include <linux/key-type.h> | |
18 | #include <keys/user-type.h> | |
19 | #include <linux/seq_file.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/uaccess.h> | |
22 | #include <linux/ctype.h> | |
f76d207a | 23 | #include <linux/projid.h> |
e66eded8 | 24 | #include <linux/fs_struct.h> |
acce292c | 25 | |
6164281a | 26 | static struct kmem_cache *user_ns_cachep __read_mostly; |
6eaab787 | 27 | static DEFINE_MUTEX(userns_state_mutex); |
6164281a | 28 | |
6708075f EB |
29 | static bool new_idmap_permitted(const struct file *file, |
30 | struct user_namespace *ns, int cap_setid, | |
22d917d8 EB |
31 | struct uid_gid_map *map); |
32 | ||
cde1975b EB |
33 | static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns) |
34 | { | |
35 | /* Start with the same capabilities as init but useless for doing | |
36 | * anything as the capabilities are bound to the new user namespace. | |
37 | */ | |
38 | cred->securebits = SECUREBITS_DEFAULT; | |
39 | cred->cap_inheritable = CAP_EMPTY_SET; | |
40 | cred->cap_permitted = CAP_FULL_SET; | |
41 | cred->cap_effective = CAP_FULL_SET; | |
42 | cred->cap_bset = CAP_FULL_SET; | |
43 | #ifdef CONFIG_KEYS | |
44 | key_put(cred->request_key_auth); | |
45 | cred->request_key_auth = NULL; | |
46 | #endif | |
47 | /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */ | |
48 | cred->user_ns = user_ns; | |
49 | } | |
50 | ||
77ec739d | 51 | /* |
18b6e041 SH |
52 | * Create a new user namespace, deriving the creator from the user in the |
53 | * passed credentials, and replacing that user with the new root user for the | |
54 | * new namespace. | |
55 | * | |
56 | * This is called by copy_creds(), which will finish setting the target task's | |
57 | * credentials. | |
77ec739d | 58 | */ |
18b6e041 | 59 | int create_user_ns(struct cred *new) |
77ec739d | 60 | { |
0093ccb6 | 61 | struct user_namespace *ns, *parent_ns = new->user_ns; |
078de5f7 EB |
62 | kuid_t owner = new->euid; |
63 | kgid_t group = new->egid; | |
98f842e6 | 64 | int ret; |
783291e6 | 65 | |
5c5f9cd2 ON |
66 | if (parent_ns->level > 32) |
67 | return -EUSERS; | |
68 | ||
3151527e EB |
69 | /* |
70 | * Verify that we can not violate the policy of which files | |
71 | * may be accessed that is specified by the root directory, | |
72 | * by verifing that the root directory is at the root of the | |
73 | * mount namespace which allows all files to be accessed. | |
74 | */ | |
75 | if (current_chrooted()) | |
76 | return -EPERM; | |
77 | ||
783291e6 EB |
78 | /* The creator needs a mapping in the parent user namespace |
79 | * or else we won't be able to reasonably tell userspace who | |
80 | * created a user_namespace. | |
81 | */ | |
82 | if (!kuid_has_mapping(parent_ns, owner) || | |
83 | !kgid_has_mapping(parent_ns, group)) | |
84 | return -EPERM; | |
77ec739d | 85 | |
22d917d8 | 86 | ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL); |
77ec739d | 87 | if (!ns) |
18b6e041 | 88 | return -ENOMEM; |
77ec739d | 89 | |
98f842e6 EB |
90 | ret = proc_alloc_inum(&ns->proc_inum); |
91 | if (ret) { | |
92 | kmem_cache_free(user_ns_cachep, ns); | |
93 | return ret; | |
94 | } | |
95 | ||
c61a2810 | 96 | atomic_set(&ns->count, 1); |
cde1975b | 97 | /* Leave the new->user_ns reference with the new user namespace. */ |
aeb3ae9d | 98 | ns->parent = parent_ns; |
5c5f9cd2 | 99 | ns->level = parent_ns->level + 1; |
783291e6 EB |
100 | ns->owner = owner; |
101 | ns->group = group; | |
22d917d8 | 102 | |
1c587ee5 EB |
103 | /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */ |
104 | mutex_lock(&userns_state_mutex); | |
105 | ns->flags = parent_ns->flags; | |
106 | mutex_unlock(&userns_state_mutex); | |
107 | ||
cde1975b | 108 | set_cred_user_ns(new, ns); |
0093ccb6 | 109 | |
87a8ebd6 EB |
110 | update_mnt_policy(ns); |
111 | ||
18b6e041 | 112 | return 0; |
acce292c CLG |
113 | } |
114 | ||
b2e0d987 EB |
115 | int unshare_userns(unsigned long unshare_flags, struct cred **new_cred) |
116 | { | |
117 | struct cred *cred; | |
c98ebcb6 | 118 | int err = -ENOMEM; |
b2e0d987 EB |
119 | |
120 | if (!(unshare_flags & CLONE_NEWUSER)) | |
121 | return 0; | |
122 | ||
123 | cred = prepare_creds(); | |
c98ebcb6 ON |
124 | if (cred) { |
125 | err = create_user_ns(cred); | |
126 | if (err) | |
127 | put_cred(cred); | |
128 | else | |
129 | *new_cred = cred; | |
130 | } | |
b2e0d987 | 131 | |
c98ebcb6 | 132 | return err; |
b2e0d987 EB |
133 | } |
134 | ||
c61a2810 | 135 | void free_user_ns(struct user_namespace *ns) |
acce292c | 136 | { |
c61a2810 | 137 | struct user_namespace *parent; |
783291e6 | 138 | |
c61a2810 EB |
139 | do { |
140 | parent = ns->parent; | |
141 | proc_free_inum(ns->proc_inum); | |
142 | kmem_cache_free(user_ns_cachep, ns); | |
143 | ns = parent; | |
144 | } while (atomic_dec_and_test(&parent->count)); | |
acce292c | 145 | } |
6a3fd92e | 146 | EXPORT_SYMBOL(free_user_ns); |
5c1469de | 147 | |
22d917d8 | 148 | static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count) |
5c1469de | 149 | { |
22d917d8 EB |
150 | unsigned idx, extents; |
151 | u32 first, last, id2; | |
5c1469de | 152 | |
22d917d8 | 153 | id2 = id + count - 1; |
5c1469de | 154 | |
22d917d8 EB |
155 | /* Find the matching extent */ |
156 | extents = map->nr_extents; | |
98afe6df | 157 | smp_rmb(); |
22d917d8 EB |
158 | for (idx = 0; idx < extents; idx++) { |
159 | first = map->extent[idx].first; | |
160 | last = first + map->extent[idx].count - 1; | |
161 | if (id >= first && id <= last && | |
162 | (id2 >= first && id2 <= last)) | |
163 | break; | |
164 | } | |
165 | /* Map the id or note failure */ | |
166 | if (idx < extents) | |
167 | id = (id - first) + map->extent[idx].lower_first; | |
168 | else | |
169 | id = (u32) -1; | |
170 | ||
171 | return id; | |
172 | } | |
173 | ||
174 | static u32 map_id_down(struct uid_gid_map *map, u32 id) | |
175 | { | |
176 | unsigned idx, extents; | |
177 | u32 first, last; | |
178 | ||
179 | /* Find the matching extent */ | |
180 | extents = map->nr_extents; | |
98afe6df | 181 | smp_rmb(); |
22d917d8 EB |
182 | for (idx = 0; idx < extents; idx++) { |
183 | first = map->extent[idx].first; | |
184 | last = first + map->extent[idx].count - 1; | |
185 | if (id >= first && id <= last) | |
186 | break; | |
187 | } | |
188 | /* Map the id or note failure */ | |
189 | if (idx < extents) | |
190 | id = (id - first) + map->extent[idx].lower_first; | |
191 | else | |
192 | id = (u32) -1; | |
193 | ||
194 | return id; | |
195 | } | |
196 | ||
197 | static u32 map_id_up(struct uid_gid_map *map, u32 id) | |
198 | { | |
199 | unsigned idx, extents; | |
200 | u32 first, last; | |
201 | ||
202 | /* Find the matching extent */ | |
203 | extents = map->nr_extents; | |
98afe6df | 204 | smp_rmb(); |
22d917d8 EB |
205 | for (idx = 0; idx < extents; idx++) { |
206 | first = map->extent[idx].lower_first; | |
207 | last = first + map->extent[idx].count - 1; | |
208 | if (id >= first && id <= last) | |
209 | break; | |
5c1469de | 210 | } |
22d917d8 EB |
211 | /* Map the id or note failure */ |
212 | if (idx < extents) | |
213 | id = (id - first) + map->extent[idx].first; | |
214 | else | |
215 | id = (u32) -1; | |
216 | ||
217 | return id; | |
218 | } | |
219 | ||
220 | /** | |
221 | * make_kuid - Map a user-namespace uid pair into a kuid. | |
222 | * @ns: User namespace that the uid is in | |
223 | * @uid: User identifier | |
224 | * | |
225 | * Maps a user-namespace uid pair into a kernel internal kuid, | |
226 | * and returns that kuid. | |
227 | * | |
228 | * When there is no mapping defined for the user-namespace uid | |
229 | * pair INVALID_UID is returned. Callers are expected to test | |
230 | * for and handle handle INVALID_UID being returned. INVALID_UID | |
231 | * may be tested for using uid_valid(). | |
232 | */ | |
233 | kuid_t make_kuid(struct user_namespace *ns, uid_t uid) | |
234 | { | |
235 | /* Map the uid to a global kernel uid */ | |
236 | return KUIDT_INIT(map_id_down(&ns->uid_map, uid)); | |
237 | } | |
238 | EXPORT_SYMBOL(make_kuid); | |
239 | ||
240 | /** | |
241 | * from_kuid - Create a uid from a kuid user-namespace pair. | |
242 | * @targ: The user namespace we want a uid in. | |
243 | * @kuid: The kernel internal uid to start with. | |
244 | * | |
245 | * Map @kuid into the user-namespace specified by @targ and | |
246 | * return the resulting uid. | |
247 | * | |
248 | * There is always a mapping into the initial user_namespace. | |
249 | * | |
250 | * If @kuid has no mapping in @targ (uid_t)-1 is returned. | |
251 | */ | |
252 | uid_t from_kuid(struct user_namespace *targ, kuid_t kuid) | |
253 | { | |
254 | /* Map the uid from a global kernel uid */ | |
255 | return map_id_up(&targ->uid_map, __kuid_val(kuid)); | |
256 | } | |
257 | EXPORT_SYMBOL(from_kuid); | |
258 | ||
259 | /** | |
260 | * from_kuid_munged - Create a uid from a kuid user-namespace pair. | |
261 | * @targ: The user namespace we want a uid in. | |
262 | * @kuid: The kernel internal uid to start with. | |
263 | * | |
264 | * Map @kuid into the user-namespace specified by @targ and | |
265 | * return the resulting uid. | |
266 | * | |
267 | * There is always a mapping into the initial user_namespace. | |
268 | * | |
269 | * Unlike from_kuid from_kuid_munged never fails and always | |
270 | * returns a valid uid. This makes from_kuid_munged appropriate | |
271 | * for use in syscalls like stat and getuid where failing the | |
272 | * system call and failing to provide a valid uid are not an | |
273 | * options. | |
274 | * | |
275 | * If @kuid has no mapping in @targ overflowuid is returned. | |
276 | */ | |
277 | uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid) | |
278 | { | |
279 | uid_t uid; | |
280 | uid = from_kuid(targ, kuid); | |
281 | ||
282 | if (uid == (uid_t) -1) | |
283 | uid = overflowuid; | |
284 | return uid; | |
285 | } | |
286 | EXPORT_SYMBOL(from_kuid_munged); | |
287 | ||
288 | /** | |
289 | * make_kgid - Map a user-namespace gid pair into a kgid. | |
290 | * @ns: User namespace that the gid is in | |
291 | * @uid: group identifier | |
292 | * | |
293 | * Maps a user-namespace gid pair into a kernel internal kgid, | |
294 | * and returns that kgid. | |
295 | * | |
296 | * When there is no mapping defined for the user-namespace gid | |
297 | * pair INVALID_GID is returned. Callers are expected to test | |
298 | * for and handle INVALID_GID being returned. INVALID_GID may be | |
299 | * tested for using gid_valid(). | |
300 | */ | |
301 | kgid_t make_kgid(struct user_namespace *ns, gid_t gid) | |
302 | { | |
303 | /* Map the gid to a global kernel gid */ | |
304 | return KGIDT_INIT(map_id_down(&ns->gid_map, gid)); | |
305 | } | |
306 | EXPORT_SYMBOL(make_kgid); | |
307 | ||
308 | /** | |
309 | * from_kgid - Create a gid from a kgid user-namespace pair. | |
310 | * @targ: The user namespace we want a gid in. | |
311 | * @kgid: The kernel internal gid to start with. | |
312 | * | |
313 | * Map @kgid into the user-namespace specified by @targ and | |
314 | * return the resulting gid. | |
315 | * | |
316 | * There is always a mapping into the initial user_namespace. | |
317 | * | |
318 | * If @kgid has no mapping in @targ (gid_t)-1 is returned. | |
319 | */ | |
320 | gid_t from_kgid(struct user_namespace *targ, kgid_t kgid) | |
321 | { | |
322 | /* Map the gid from a global kernel gid */ | |
323 | return map_id_up(&targ->gid_map, __kgid_val(kgid)); | |
324 | } | |
325 | EXPORT_SYMBOL(from_kgid); | |
326 | ||
327 | /** | |
328 | * from_kgid_munged - Create a gid from a kgid user-namespace pair. | |
329 | * @targ: The user namespace we want a gid in. | |
330 | * @kgid: The kernel internal gid to start with. | |
331 | * | |
332 | * Map @kgid into the user-namespace specified by @targ and | |
333 | * return the resulting gid. | |
334 | * | |
335 | * There is always a mapping into the initial user_namespace. | |
336 | * | |
337 | * Unlike from_kgid from_kgid_munged never fails and always | |
338 | * returns a valid gid. This makes from_kgid_munged appropriate | |
339 | * for use in syscalls like stat and getgid where failing the | |
340 | * system call and failing to provide a valid gid are not options. | |
341 | * | |
342 | * If @kgid has no mapping in @targ overflowgid is returned. | |
343 | */ | |
344 | gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid) | |
345 | { | |
346 | gid_t gid; | |
347 | gid = from_kgid(targ, kgid); | |
348 | ||
349 | if (gid == (gid_t) -1) | |
350 | gid = overflowgid; | |
351 | return gid; | |
352 | } | |
353 | EXPORT_SYMBOL(from_kgid_munged); | |
354 | ||
f76d207a EB |
355 | /** |
356 | * make_kprojid - Map a user-namespace projid pair into a kprojid. | |
357 | * @ns: User namespace that the projid is in | |
358 | * @projid: Project identifier | |
359 | * | |
360 | * Maps a user-namespace uid pair into a kernel internal kuid, | |
361 | * and returns that kuid. | |
362 | * | |
363 | * When there is no mapping defined for the user-namespace projid | |
364 | * pair INVALID_PROJID is returned. Callers are expected to test | |
365 | * for and handle handle INVALID_PROJID being returned. INVALID_PROJID | |
366 | * may be tested for using projid_valid(). | |
367 | */ | |
368 | kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid) | |
369 | { | |
370 | /* Map the uid to a global kernel uid */ | |
371 | return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid)); | |
372 | } | |
373 | EXPORT_SYMBOL(make_kprojid); | |
374 | ||
375 | /** | |
376 | * from_kprojid - Create a projid from a kprojid user-namespace pair. | |
377 | * @targ: The user namespace we want a projid in. | |
378 | * @kprojid: The kernel internal project identifier to start with. | |
379 | * | |
380 | * Map @kprojid into the user-namespace specified by @targ and | |
381 | * return the resulting projid. | |
382 | * | |
383 | * There is always a mapping into the initial user_namespace. | |
384 | * | |
385 | * If @kprojid has no mapping in @targ (projid_t)-1 is returned. | |
386 | */ | |
387 | projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid) | |
388 | { | |
389 | /* Map the uid from a global kernel uid */ | |
390 | return map_id_up(&targ->projid_map, __kprojid_val(kprojid)); | |
391 | } | |
392 | EXPORT_SYMBOL(from_kprojid); | |
393 | ||
394 | /** | |
395 | * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair. | |
396 | * @targ: The user namespace we want a projid in. | |
397 | * @kprojid: The kernel internal projid to start with. | |
398 | * | |
399 | * Map @kprojid into the user-namespace specified by @targ and | |
400 | * return the resulting projid. | |
401 | * | |
402 | * There is always a mapping into the initial user_namespace. | |
403 | * | |
404 | * Unlike from_kprojid from_kprojid_munged never fails and always | |
405 | * returns a valid projid. This makes from_kprojid_munged | |
406 | * appropriate for use in syscalls like stat and where | |
407 | * failing the system call and failing to provide a valid projid are | |
408 | * not an options. | |
409 | * | |
410 | * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned. | |
411 | */ | |
412 | projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid) | |
413 | { | |
414 | projid_t projid; | |
415 | projid = from_kprojid(targ, kprojid); | |
416 | ||
417 | if (projid == (projid_t) -1) | |
418 | projid = OVERFLOW_PROJID; | |
419 | return projid; | |
420 | } | |
421 | EXPORT_SYMBOL(from_kprojid_munged); | |
422 | ||
423 | ||
22d917d8 EB |
424 | static int uid_m_show(struct seq_file *seq, void *v) |
425 | { | |
426 | struct user_namespace *ns = seq->private; | |
427 | struct uid_gid_extent *extent = v; | |
428 | struct user_namespace *lower_ns; | |
429 | uid_t lower; | |
5c1469de | 430 | |
c450f371 | 431 | lower_ns = seq_user_ns(seq); |
22d917d8 EB |
432 | if ((lower_ns == ns) && lower_ns->parent) |
433 | lower_ns = lower_ns->parent; | |
434 | ||
435 | lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first)); | |
436 | ||
437 | seq_printf(seq, "%10u %10u %10u\n", | |
438 | extent->first, | |
439 | lower, | |
440 | extent->count); | |
441 | ||
442 | return 0; | |
5c1469de EB |
443 | } |
444 | ||
22d917d8 | 445 | static int gid_m_show(struct seq_file *seq, void *v) |
5c1469de | 446 | { |
22d917d8 EB |
447 | struct user_namespace *ns = seq->private; |
448 | struct uid_gid_extent *extent = v; | |
449 | struct user_namespace *lower_ns; | |
450 | gid_t lower; | |
5c1469de | 451 | |
c450f371 | 452 | lower_ns = seq_user_ns(seq); |
22d917d8 EB |
453 | if ((lower_ns == ns) && lower_ns->parent) |
454 | lower_ns = lower_ns->parent; | |
5c1469de | 455 | |
22d917d8 EB |
456 | lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first)); |
457 | ||
458 | seq_printf(seq, "%10u %10u %10u\n", | |
459 | extent->first, | |
460 | lower, | |
461 | extent->count); | |
462 | ||
463 | return 0; | |
464 | } | |
465 | ||
f76d207a EB |
466 | static int projid_m_show(struct seq_file *seq, void *v) |
467 | { | |
468 | struct user_namespace *ns = seq->private; | |
469 | struct uid_gid_extent *extent = v; | |
470 | struct user_namespace *lower_ns; | |
471 | projid_t lower; | |
472 | ||
473 | lower_ns = seq_user_ns(seq); | |
474 | if ((lower_ns == ns) && lower_ns->parent) | |
475 | lower_ns = lower_ns->parent; | |
476 | ||
477 | lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first)); | |
478 | ||
479 | seq_printf(seq, "%10u %10u %10u\n", | |
480 | extent->first, | |
481 | lower, | |
482 | extent->count); | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
22d917d8 EB |
487 | static void *m_start(struct seq_file *seq, loff_t *ppos, struct uid_gid_map *map) |
488 | { | |
489 | struct uid_gid_extent *extent = NULL; | |
490 | loff_t pos = *ppos; | |
491 | ||
492 | if (pos < map->nr_extents) | |
493 | extent = &map->extent[pos]; | |
494 | ||
495 | return extent; | |
496 | } | |
497 | ||
498 | static void *uid_m_start(struct seq_file *seq, loff_t *ppos) | |
499 | { | |
500 | struct user_namespace *ns = seq->private; | |
501 | ||
502 | return m_start(seq, ppos, &ns->uid_map); | |
503 | } | |
504 | ||
505 | static void *gid_m_start(struct seq_file *seq, loff_t *ppos) | |
506 | { | |
507 | struct user_namespace *ns = seq->private; | |
508 | ||
509 | return m_start(seq, ppos, &ns->gid_map); | |
510 | } | |
511 | ||
f76d207a EB |
512 | static void *projid_m_start(struct seq_file *seq, loff_t *ppos) |
513 | { | |
514 | struct user_namespace *ns = seq->private; | |
515 | ||
516 | return m_start(seq, ppos, &ns->projid_map); | |
517 | } | |
518 | ||
22d917d8 EB |
519 | static void *m_next(struct seq_file *seq, void *v, loff_t *pos) |
520 | { | |
521 | (*pos)++; | |
522 | return seq->op->start(seq, pos); | |
523 | } | |
524 | ||
525 | static void m_stop(struct seq_file *seq, void *v) | |
526 | { | |
527 | return; | |
528 | } | |
529 | ||
530 | struct seq_operations proc_uid_seq_operations = { | |
531 | .start = uid_m_start, | |
532 | .stop = m_stop, | |
533 | .next = m_next, | |
534 | .show = uid_m_show, | |
535 | }; | |
536 | ||
537 | struct seq_operations proc_gid_seq_operations = { | |
538 | .start = gid_m_start, | |
539 | .stop = m_stop, | |
540 | .next = m_next, | |
541 | .show = gid_m_show, | |
542 | }; | |
543 | ||
f76d207a EB |
544 | struct seq_operations proc_projid_seq_operations = { |
545 | .start = projid_m_start, | |
546 | .stop = m_stop, | |
547 | .next = m_next, | |
548 | .show = projid_m_show, | |
549 | }; | |
550 | ||
0bd14b4f EB |
551 | static bool mappings_overlap(struct uid_gid_map *new_map, struct uid_gid_extent *extent) |
552 | { | |
553 | u32 upper_first, lower_first, upper_last, lower_last; | |
554 | unsigned idx; | |
555 | ||
556 | upper_first = extent->first; | |
557 | lower_first = extent->lower_first; | |
558 | upper_last = upper_first + extent->count - 1; | |
559 | lower_last = lower_first + extent->count - 1; | |
560 | ||
561 | for (idx = 0; idx < new_map->nr_extents; idx++) { | |
562 | u32 prev_upper_first, prev_lower_first; | |
563 | u32 prev_upper_last, prev_lower_last; | |
564 | struct uid_gid_extent *prev; | |
565 | ||
566 | prev = &new_map->extent[idx]; | |
567 | ||
568 | prev_upper_first = prev->first; | |
569 | prev_lower_first = prev->lower_first; | |
570 | prev_upper_last = prev_upper_first + prev->count - 1; | |
571 | prev_lower_last = prev_lower_first + prev->count - 1; | |
572 | ||
573 | /* Does the upper range intersect a previous extent? */ | |
574 | if ((prev_upper_first <= upper_last) && | |
575 | (prev_upper_last >= upper_first)) | |
576 | return true; | |
577 | ||
578 | /* Does the lower range intersect a previous extent? */ | |
579 | if ((prev_lower_first <= lower_last) && | |
580 | (prev_lower_last >= lower_first)) | |
581 | return true; | |
582 | } | |
583 | return false; | |
584 | } | |
585 | ||
22d917d8 EB |
586 | static ssize_t map_write(struct file *file, const char __user *buf, |
587 | size_t count, loff_t *ppos, | |
588 | int cap_setid, | |
589 | struct uid_gid_map *map, | |
590 | struct uid_gid_map *parent_map) | |
591 | { | |
592 | struct seq_file *seq = file->private_data; | |
593 | struct user_namespace *ns = seq->private; | |
594 | struct uid_gid_map new_map; | |
595 | unsigned idx; | |
0bd14b4f | 596 | struct uid_gid_extent *extent = NULL; |
22d917d8 EB |
597 | unsigned long page = 0; |
598 | char *kbuf, *pos, *next_line; | |
599 | ssize_t ret = -EINVAL; | |
600 | ||
601 | /* | |
6eaab787 | 602 | * The userns_state_mutex serializes all writes to any given map. |
22d917d8 EB |
603 | * |
604 | * Any map is only ever written once. | |
605 | * | |
606 | * An id map fits within 1 cache line on most architectures. | |
607 | * | |
608 | * On read nothing needs to be done unless you are on an | |
609 | * architecture with a crazy cache coherency model like alpha. | |
610 | * | |
611 | * There is a one time data dependency between reading the | |
612 | * count of the extents and the values of the extents. The | |
613 | * desired behavior is to see the values of the extents that | |
614 | * were written before the count of the extents. | |
615 | * | |
616 | * To achieve this smp_wmb() is used on guarantee the write | |
98afe6df MP |
617 | * order and smp_rmb() is guaranteed that we don't have crazy |
618 | * architectures returning stale data. | |
22d917d8 | 619 | */ |
6eaab787 | 620 | mutex_lock(&userns_state_mutex); |
22d917d8 EB |
621 | |
622 | ret = -EPERM; | |
623 | /* Only allow one successful write to the map */ | |
624 | if (map->nr_extents != 0) | |
625 | goto out; | |
626 | ||
41c21e35 AL |
627 | /* |
628 | * Adjusting namespace settings requires capabilities on the target. | |
5c1469de | 629 | */ |
41c21e35 | 630 | if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN)) |
22d917d8 EB |
631 | goto out; |
632 | ||
633 | /* Get a buffer */ | |
634 | ret = -ENOMEM; | |
635 | page = __get_free_page(GFP_TEMPORARY); | |
636 | kbuf = (char *) page; | |
637 | if (!page) | |
638 | goto out; | |
639 | ||
640 | /* Only allow <= page size writes at the beginning of the file */ | |
641 | ret = -EINVAL; | |
642 | if ((*ppos != 0) || (count >= PAGE_SIZE)) | |
643 | goto out; | |
644 | ||
645 | /* Slurp in the user data */ | |
646 | ret = -EFAULT; | |
647 | if (copy_from_user(kbuf, buf, count)) | |
648 | goto out; | |
649 | kbuf[count] = '\0'; | |
650 | ||
651 | /* Parse the user data */ | |
652 | ret = -EINVAL; | |
653 | pos = kbuf; | |
654 | new_map.nr_extents = 0; | |
655 | for (;pos; pos = next_line) { | |
656 | extent = &new_map.extent[new_map.nr_extents]; | |
657 | ||
658 | /* Find the end of line and ensure I don't look past it */ | |
659 | next_line = strchr(pos, '\n'); | |
660 | if (next_line) { | |
661 | *next_line = '\0'; | |
662 | next_line++; | |
663 | if (*next_line == '\0') | |
664 | next_line = NULL; | |
5c1469de | 665 | } |
22d917d8 EB |
666 | |
667 | pos = skip_spaces(pos); | |
668 | extent->first = simple_strtoul(pos, &pos, 10); | |
669 | if (!isspace(*pos)) | |
670 | goto out; | |
671 | ||
672 | pos = skip_spaces(pos); | |
673 | extent->lower_first = simple_strtoul(pos, &pos, 10); | |
674 | if (!isspace(*pos)) | |
675 | goto out; | |
676 | ||
677 | pos = skip_spaces(pos); | |
678 | extent->count = simple_strtoul(pos, &pos, 10); | |
679 | if (*pos && !isspace(*pos)) | |
680 | goto out; | |
681 | ||
682 | /* Verify there is not trailing junk on the line */ | |
683 | pos = skip_spaces(pos); | |
684 | if (*pos != '\0') | |
685 | goto out; | |
686 | ||
687 | /* Verify we have been given valid starting values */ | |
688 | if ((extent->first == (u32) -1) || | |
689 | (extent->lower_first == (u32) -1 )) | |
690 | goto out; | |
691 | ||
692 | /* Verify count is not zero and does not cause the extent to wrap */ | |
693 | if ((extent->first + extent->count) <= extent->first) | |
694 | goto out; | |
695 | if ((extent->lower_first + extent->count) <= extent->lower_first) | |
696 | goto out; | |
697 | ||
0bd14b4f EB |
698 | /* Do the ranges in extent overlap any previous extents? */ |
699 | if (mappings_overlap(&new_map, extent)) | |
22d917d8 EB |
700 | goto out; |
701 | ||
702 | new_map.nr_extents++; | |
22d917d8 EB |
703 | |
704 | /* Fail if the file contains too many extents */ | |
705 | if ((new_map.nr_extents == UID_GID_MAP_MAX_EXTENTS) && | |
706 | (next_line != NULL)) | |
707 | goto out; | |
5c1469de | 708 | } |
22d917d8 EB |
709 | /* Be very certaint the new map actually exists */ |
710 | if (new_map.nr_extents == 0) | |
711 | goto out; | |
712 | ||
713 | ret = -EPERM; | |
714 | /* Validate the user is allowed to use user id's mapped to. */ | |
6708075f | 715 | if (!new_idmap_permitted(file, ns, cap_setid, &new_map)) |
22d917d8 EB |
716 | goto out; |
717 | ||
718 | /* Map the lower ids from the parent user namespace to the | |
719 | * kernel global id space. | |
720 | */ | |
721 | for (idx = 0; idx < new_map.nr_extents; idx++) { | |
722 | u32 lower_first; | |
723 | extent = &new_map.extent[idx]; | |
724 | ||
725 | lower_first = map_id_range_down(parent_map, | |
726 | extent->lower_first, | |
727 | extent->count); | |
728 | ||
729 | /* Fail if we can not map the specified extent to | |
730 | * the kernel global id space. | |
731 | */ | |
732 | if (lower_first == (u32) -1) | |
733 | goto out; | |
734 | ||
735 | extent->lower_first = lower_first; | |
736 | } | |
737 | ||
738 | /* Install the map */ | |
739 | memcpy(map->extent, new_map.extent, | |
740 | new_map.nr_extents*sizeof(new_map.extent[0])); | |
741 | smp_wmb(); | |
742 | map->nr_extents = new_map.nr_extents; | |
743 | ||
744 | *ppos = count; | |
745 | ret = count; | |
746 | out: | |
6eaab787 | 747 | mutex_unlock(&userns_state_mutex); |
22d917d8 EB |
748 | if (page) |
749 | free_page(page); | |
750 | return ret; | |
751 | } | |
752 | ||
753 | ssize_t proc_uid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos) | |
754 | { | |
755 | struct seq_file *seq = file->private_data; | |
756 | struct user_namespace *ns = seq->private; | |
c450f371 | 757 | struct user_namespace *seq_ns = seq_user_ns(seq); |
22d917d8 EB |
758 | |
759 | if (!ns->parent) | |
760 | return -EPERM; | |
761 | ||
c450f371 EB |
762 | if ((seq_ns != ns) && (seq_ns != ns->parent)) |
763 | return -EPERM; | |
764 | ||
22d917d8 EB |
765 | return map_write(file, buf, size, ppos, CAP_SETUID, |
766 | &ns->uid_map, &ns->parent->uid_map); | |
767 | } | |
768 | ||
769 | ssize_t proc_gid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos) | |
770 | { | |
771 | struct seq_file *seq = file->private_data; | |
772 | struct user_namespace *ns = seq->private; | |
c450f371 | 773 | struct user_namespace *seq_ns = seq_user_ns(seq); |
22d917d8 EB |
774 | |
775 | if (!ns->parent) | |
776 | return -EPERM; | |
777 | ||
c450f371 EB |
778 | if ((seq_ns != ns) && (seq_ns != ns->parent)) |
779 | return -EPERM; | |
780 | ||
22d917d8 EB |
781 | return map_write(file, buf, size, ppos, CAP_SETGID, |
782 | &ns->gid_map, &ns->parent->gid_map); | |
783 | } | |
784 | ||
f76d207a EB |
785 | ssize_t proc_projid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos) |
786 | { | |
787 | struct seq_file *seq = file->private_data; | |
788 | struct user_namespace *ns = seq->private; | |
789 | struct user_namespace *seq_ns = seq_user_ns(seq); | |
790 | ||
791 | if (!ns->parent) | |
792 | return -EPERM; | |
793 | ||
794 | if ((seq_ns != ns) && (seq_ns != ns->parent)) | |
795 | return -EPERM; | |
796 | ||
797 | /* Anyone can set any valid project id no capability needed */ | |
798 | return map_write(file, buf, size, ppos, -1, | |
799 | &ns->projid_map, &ns->parent->projid_map); | |
800 | } | |
801 | ||
6708075f EB |
802 | static bool new_idmap_permitted(const struct file *file, |
803 | struct user_namespace *ns, int cap_setid, | |
22d917d8 EB |
804 | struct uid_gid_map *new_map) |
805 | { | |
a1821391 | 806 | const struct cred *cred = file->f_cred; |
b8a0441b EB |
807 | /* Don't allow mappings that would allow anything that wouldn't |
808 | * be allowed without the establishment of unprivileged mappings. | |
809 | */ | |
a1821391 EB |
810 | if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) && |
811 | uid_eq(ns->owner, cred->euid)) { | |
37657da3 EB |
812 | u32 id = new_map->extent[0].lower_first; |
813 | if (cap_setid == CAP_SETUID) { | |
814 | kuid_t uid = make_kuid(ns->parent, id); | |
a1821391 | 815 | if (uid_eq(uid, cred->euid)) |
37657da3 | 816 | return true; |
afb52853 EB |
817 | } else if (cap_setid == CAP_SETGID) { |
818 | kgid_t gid = make_kgid(ns->parent, id); | |
819 | if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) && | |
820 | gid_eq(gid, cred->egid)) | |
821 | return true; | |
37657da3 | 822 | } |
37657da3 EB |
823 | } |
824 | ||
f76d207a EB |
825 | /* Allow anyone to set a mapping that doesn't require privilege */ |
826 | if (!cap_valid(cap_setid)) | |
827 | return true; | |
828 | ||
22d917d8 EB |
829 | /* Allow the specified ids if we have the appropriate capability |
830 | * (CAP_SETUID or CAP_SETGID) over the parent user namespace. | |
6708075f | 831 | * And the opener of the id file also had the approprpiate capability. |
22d917d8 | 832 | */ |
6708075f EB |
833 | if (ns_capable(ns->parent, cap_setid) && |
834 | file_ns_capable(file, ns->parent, cap_setid)) | |
22d917d8 | 835 | return true; |
5c1469de | 836 | |
22d917d8 | 837 | return false; |
5c1469de | 838 | } |
6164281a | 839 | |
1c587ee5 EB |
840 | int proc_setgroups_show(struct seq_file *seq, void *v) |
841 | { | |
842 | struct user_namespace *ns = seq->private; | |
843 | unsigned long userns_flags = ACCESS_ONCE(ns->flags); | |
844 | ||
845 | seq_printf(seq, "%s\n", | |
846 | (userns_flags & USERNS_SETGROUPS_ALLOWED) ? | |
847 | "allow" : "deny"); | |
848 | return 0; | |
849 | } | |
850 | ||
851 | ssize_t proc_setgroups_write(struct file *file, const char __user *buf, | |
852 | size_t count, loff_t *ppos) | |
853 | { | |
854 | struct seq_file *seq = file->private_data; | |
855 | struct user_namespace *ns = seq->private; | |
856 | char kbuf[8], *pos; | |
857 | bool setgroups_allowed; | |
858 | ssize_t ret; | |
859 | ||
860 | /* Only allow a very narrow range of strings to be written */ | |
861 | ret = -EINVAL; | |
862 | if ((*ppos != 0) || (count >= sizeof(kbuf))) | |
863 | goto out; | |
864 | ||
865 | /* What was written? */ | |
866 | ret = -EFAULT; | |
867 | if (copy_from_user(kbuf, buf, count)) | |
868 | goto out; | |
869 | kbuf[count] = '\0'; | |
870 | pos = kbuf; | |
871 | ||
872 | /* What is being requested? */ | |
873 | ret = -EINVAL; | |
874 | if (strncmp(pos, "allow", 5) == 0) { | |
875 | pos += 5; | |
876 | setgroups_allowed = true; | |
877 | } | |
878 | else if (strncmp(pos, "deny", 4) == 0) { | |
879 | pos += 4; | |
880 | setgroups_allowed = false; | |
881 | } | |
882 | else | |
883 | goto out; | |
884 | ||
885 | /* Verify there is not trailing junk on the line */ | |
886 | pos = skip_spaces(pos); | |
887 | if (*pos != '\0') | |
888 | goto out; | |
889 | ||
890 | ret = -EPERM; | |
891 | mutex_lock(&userns_state_mutex); | |
892 | if (setgroups_allowed) { | |
893 | /* Enabling setgroups after setgroups has been disabled | |
894 | * is not allowed. | |
895 | */ | |
896 | if (!(ns->flags & USERNS_SETGROUPS_ALLOWED)) | |
897 | goto out_unlock; | |
898 | } else { | |
899 | /* Permanently disabling setgroups after setgroups has | |
900 | * been enabled by writing the gid_map is not allowed. | |
901 | */ | |
902 | if (ns->gid_map.nr_extents != 0) | |
903 | goto out_unlock; | |
904 | ns->flags &= ~USERNS_SETGROUPS_ALLOWED; | |
905 | } | |
906 | mutex_unlock(&userns_state_mutex); | |
907 | ||
908 | /* Report a successful write */ | |
909 | *ppos = count; | |
910 | ret = count; | |
911 | out: | |
912 | return ret; | |
913 | out_unlock: | |
914 | mutex_unlock(&userns_state_mutex); | |
915 | goto out; | |
916 | } | |
917 | ||
fc9b65e3 EB |
918 | bool userns_may_setgroups(const struct user_namespace *ns) |
919 | { | |
920 | bool allowed; | |
921 | ||
6eaab787 | 922 | mutex_lock(&userns_state_mutex); |
fc9b65e3 EB |
923 | /* It is not safe to use setgroups until a gid mapping in |
924 | * the user namespace has been established. | |
925 | */ | |
926 | allowed = ns->gid_map.nr_extents != 0; | |
1c587ee5 EB |
927 | /* Is setgroups allowed? */ |
928 | allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED); | |
6eaab787 | 929 | mutex_unlock(&userns_state_mutex); |
fc9b65e3 EB |
930 | |
931 | return allowed; | |
932 | } | |
933 | ||
cde1975b EB |
934 | static void *userns_get(struct task_struct *task) |
935 | { | |
936 | struct user_namespace *user_ns; | |
937 | ||
938 | rcu_read_lock(); | |
939 | user_ns = get_user_ns(__task_cred(task)->user_ns); | |
940 | rcu_read_unlock(); | |
941 | ||
942 | return user_ns; | |
943 | } | |
944 | ||
945 | static void userns_put(void *ns) | |
946 | { | |
947 | put_user_ns(ns); | |
948 | } | |
949 | ||
950 | static int userns_install(struct nsproxy *nsproxy, void *ns) | |
951 | { | |
952 | struct user_namespace *user_ns = ns; | |
953 | struct cred *cred; | |
954 | ||
955 | /* Don't allow gaining capabilities by reentering | |
956 | * the same user namespace. | |
957 | */ | |
958 | if (user_ns == current_user_ns()) | |
959 | return -EINVAL; | |
960 | ||
5155040e | 961 | /* Threaded processes may not enter a different user namespace */ |
cde1975b EB |
962 | if (atomic_read(¤t->mm->mm_users) > 1) |
963 | return -EINVAL; | |
964 | ||
e66eded8 EB |
965 | if (current->fs->users != 1) |
966 | return -EINVAL; | |
967 | ||
cde1975b EB |
968 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
969 | return -EPERM; | |
970 | ||
971 | cred = prepare_creds(); | |
972 | if (!cred) | |
973 | return -ENOMEM; | |
974 | ||
975 | put_user_ns(cred->user_ns); | |
976 | set_cred_user_ns(cred, get_user_ns(user_ns)); | |
977 | ||
978 | return commit_creds(cred); | |
979 | } | |
980 | ||
98f842e6 EB |
981 | static unsigned int userns_inum(void *ns) |
982 | { | |
983 | struct user_namespace *user_ns = ns; | |
984 | return user_ns->proc_inum; | |
985 | } | |
986 | ||
cde1975b EB |
987 | const struct proc_ns_operations userns_operations = { |
988 | .name = "user", | |
989 | .type = CLONE_NEWUSER, | |
990 | .get = userns_get, | |
991 | .put = userns_put, | |
992 | .install = userns_install, | |
98f842e6 | 993 | .inum = userns_inum, |
cde1975b EB |
994 | }; |
995 | ||
6164281a PE |
996 | static __init int user_namespaces_init(void) |
997 | { | |
998 | user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC); | |
999 | return 0; | |
1000 | } | |
1001 | module_init(user_namespaces_init); |