]>
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 | |
92e575e7 SH |
26 | /* |
27 | * sysctl determining whether unprivileged users may unshare a new | |
28 | * userns. Allowed by default | |
29 | */ | |
30 | int unprivileged_userns_clone = 1; | |
31 | ||
6164281a | 32 | static struct kmem_cache *user_ns_cachep __read_mostly; |
f0d62aec | 33 | static DEFINE_MUTEX(userns_state_mutex); |
6164281a | 34 | |
6708075f EB |
35 | static bool new_idmap_permitted(const struct file *file, |
36 | struct user_namespace *ns, int cap_setid, | |
22d917d8 EB |
37 | struct uid_gid_map *map); |
38 | ||
cde1975b EB |
39 | static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns) |
40 | { | |
41 | /* Start with the same capabilities as init but useless for doing | |
42 | * anything as the capabilities are bound to the new user namespace. | |
43 | */ | |
44 | cred->securebits = SECUREBITS_DEFAULT; | |
45 | cred->cap_inheritable = CAP_EMPTY_SET; | |
46 | cred->cap_permitted = CAP_FULL_SET; | |
47 | cred->cap_effective = CAP_FULL_SET; | |
58319057 | 48 | cred->cap_ambient = CAP_EMPTY_SET; |
cde1975b EB |
49 | cred->cap_bset = CAP_FULL_SET; |
50 | #ifdef CONFIG_KEYS | |
51 | key_put(cred->request_key_auth); | |
52 | cred->request_key_auth = NULL; | |
53 | #endif | |
54 | /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */ | |
55 | cred->user_ns = user_ns; | |
56 | } | |
57 | ||
77ec739d | 58 | /* |
18b6e041 SH |
59 | * Create a new user namespace, deriving the creator from the user in the |
60 | * passed credentials, and replacing that user with the new root user for the | |
61 | * new namespace. | |
62 | * | |
63 | * This is called by copy_creds(), which will finish setting the target task's | |
64 | * credentials. | |
77ec739d | 65 | */ |
18b6e041 | 66 | int create_user_ns(struct cred *new) |
77ec739d | 67 | { |
0093ccb6 | 68 | struct user_namespace *ns, *parent_ns = new->user_ns; |
078de5f7 EB |
69 | kuid_t owner = new->euid; |
70 | kgid_t group = new->egid; | |
98f842e6 | 71 | int ret; |
783291e6 | 72 | |
8742f229 ON |
73 | if (parent_ns->level > 32) |
74 | return -EUSERS; | |
75 | ||
3151527e EB |
76 | /* |
77 | * Verify that we can not violate the policy of which files | |
78 | * may be accessed that is specified by the root directory, | |
79 | * by verifing that the root directory is at the root of the | |
80 | * mount namespace which allows all files to be accessed. | |
81 | */ | |
82 | if (current_chrooted()) | |
83 | return -EPERM; | |
84 | ||
783291e6 EB |
85 | /* The creator needs a mapping in the parent user namespace |
86 | * or else we won't be able to reasonably tell userspace who | |
87 | * created a user_namespace. | |
88 | */ | |
89 | if (!kuid_has_mapping(parent_ns, owner) || | |
90 | !kgid_has_mapping(parent_ns, group)) | |
91 | return -EPERM; | |
77ec739d | 92 | |
22d917d8 | 93 | ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL); |
77ec739d | 94 | if (!ns) |
18b6e041 | 95 | return -ENOMEM; |
77ec739d | 96 | |
6344c433 | 97 | ret = ns_alloc_inum(&ns->ns); |
98f842e6 EB |
98 | if (ret) { |
99 | kmem_cache_free(user_ns_cachep, ns); | |
100 | return ret; | |
101 | } | |
33c42940 | 102 | ns->ns.ops = &userns_operations; |
98f842e6 | 103 | |
c61a2810 | 104 | atomic_set(&ns->count, 1); |
cde1975b | 105 | /* Leave the new->user_ns reference with the new user namespace. */ |
aeb3ae9d | 106 | ns->parent = parent_ns; |
8742f229 | 107 | ns->level = parent_ns->level + 1; |
783291e6 EB |
108 | ns->owner = owner; |
109 | ns->group = group; | |
22d917d8 | 110 | |
9cc46516 EB |
111 | /* Inherit USERNS_SETGROUPS_ALLOWED from our parent */ |
112 | mutex_lock(&userns_state_mutex); | |
113 | ns->flags = parent_ns->flags; | |
114 | mutex_unlock(&userns_state_mutex); | |
115 | ||
cde1975b | 116 | set_cred_user_ns(new, ns); |
0093ccb6 | 117 | |
f36f8c75 DH |
118 | #ifdef CONFIG_PERSISTENT_KEYRINGS |
119 | init_rwsem(&ns->persistent_keyring_register_sem); | |
120 | #endif | |
18b6e041 | 121 | return 0; |
acce292c CLG |
122 | } |
123 | ||
b2e0d987 EB |
124 | int unshare_userns(unsigned long unshare_flags, struct cred **new_cred) |
125 | { | |
126 | struct cred *cred; | |
6160968c | 127 | int err = -ENOMEM; |
b2e0d987 EB |
128 | |
129 | if (!(unshare_flags & CLONE_NEWUSER)) | |
130 | return 0; | |
131 | ||
132 | cred = prepare_creds(); | |
6160968c ON |
133 | if (cred) { |
134 | err = create_user_ns(cred); | |
135 | if (err) | |
136 | put_cred(cred); | |
137 | else | |
138 | *new_cred = cred; | |
139 | } | |
b2e0d987 | 140 | |
6160968c | 141 | return err; |
b2e0d987 EB |
142 | } |
143 | ||
c61a2810 | 144 | void free_user_ns(struct user_namespace *ns) |
acce292c | 145 | { |
c61a2810 | 146 | struct user_namespace *parent; |
783291e6 | 147 | |
c61a2810 EB |
148 | do { |
149 | parent = ns->parent; | |
f36f8c75 DH |
150 | #ifdef CONFIG_PERSISTENT_KEYRINGS |
151 | key_put(ns->persistent_keyring_register); | |
152 | #endif | |
6344c433 | 153 | ns_free_inum(&ns->ns); |
c61a2810 EB |
154 | kmem_cache_free(user_ns_cachep, ns); |
155 | ns = parent; | |
156 | } while (atomic_dec_and_test(&parent->count)); | |
acce292c | 157 | } |
6a3fd92e | 158 | EXPORT_SYMBOL(free_user_ns); |
5c1469de | 159 | |
22d917d8 | 160 | static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count) |
5c1469de | 161 | { |
22d917d8 EB |
162 | unsigned idx, extents; |
163 | u32 first, last, id2; | |
5c1469de | 164 | |
22d917d8 | 165 | id2 = id + count - 1; |
5c1469de | 166 | |
22d917d8 EB |
167 | /* Find the matching extent */ |
168 | extents = map->nr_extents; | |
e79323bd | 169 | smp_rmb(); |
22d917d8 EB |
170 | for (idx = 0; idx < extents; idx++) { |
171 | first = map->extent[idx].first; | |
172 | last = first + map->extent[idx].count - 1; | |
173 | if (id >= first && id <= last && | |
174 | (id2 >= first && id2 <= last)) | |
175 | break; | |
176 | } | |
177 | /* Map the id or note failure */ | |
178 | if (idx < extents) | |
179 | id = (id - first) + map->extent[idx].lower_first; | |
180 | else | |
181 | id = (u32) -1; | |
182 | ||
183 | return id; | |
184 | } | |
185 | ||
186 | static u32 map_id_down(struct uid_gid_map *map, u32 id) | |
187 | { | |
188 | unsigned idx, extents; | |
189 | u32 first, last; | |
190 | ||
191 | /* Find the matching extent */ | |
192 | extents = map->nr_extents; | |
e79323bd | 193 | smp_rmb(); |
22d917d8 EB |
194 | for (idx = 0; idx < extents; idx++) { |
195 | first = map->extent[idx].first; | |
196 | last = first + map->extent[idx].count - 1; | |
197 | if (id >= first && id <= last) | |
198 | break; | |
199 | } | |
200 | /* Map the id or note failure */ | |
201 | if (idx < extents) | |
202 | id = (id - first) + map->extent[idx].lower_first; | |
203 | else | |
204 | id = (u32) -1; | |
205 | ||
206 | return id; | |
207 | } | |
208 | ||
209 | static u32 map_id_up(struct uid_gid_map *map, u32 id) | |
210 | { | |
211 | unsigned idx, extents; | |
212 | u32 first, last; | |
213 | ||
214 | /* Find the matching extent */ | |
215 | extents = map->nr_extents; | |
e79323bd | 216 | smp_rmb(); |
22d917d8 EB |
217 | for (idx = 0; idx < extents; idx++) { |
218 | first = map->extent[idx].lower_first; | |
219 | last = first + map->extent[idx].count - 1; | |
220 | if (id >= first && id <= last) | |
221 | break; | |
5c1469de | 222 | } |
22d917d8 EB |
223 | /* Map the id or note failure */ |
224 | if (idx < extents) | |
225 | id = (id - first) + map->extent[idx].first; | |
226 | else | |
227 | id = (u32) -1; | |
228 | ||
229 | return id; | |
230 | } | |
231 | ||
232 | /** | |
233 | * make_kuid - Map a user-namespace uid pair into a kuid. | |
234 | * @ns: User namespace that the uid is in | |
235 | * @uid: User identifier | |
236 | * | |
237 | * Maps a user-namespace uid pair into a kernel internal kuid, | |
238 | * and returns that kuid. | |
239 | * | |
240 | * When there is no mapping defined for the user-namespace uid | |
241 | * pair INVALID_UID is returned. Callers are expected to test | |
b080e047 | 242 | * for and handle INVALID_UID being returned. INVALID_UID |
22d917d8 EB |
243 | * may be tested for using uid_valid(). |
244 | */ | |
245 | kuid_t make_kuid(struct user_namespace *ns, uid_t uid) | |
246 | { | |
247 | /* Map the uid to a global kernel uid */ | |
248 | return KUIDT_INIT(map_id_down(&ns->uid_map, uid)); | |
249 | } | |
250 | EXPORT_SYMBOL(make_kuid); | |
251 | ||
252 | /** | |
253 | * from_kuid - Create a uid from a kuid user-namespace pair. | |
254 | * @targ: The user namespace we want a uid in. | |
255 | * @kuid: The kernel internal uid to start with. | |
256 | * | |
257 | * Map @kuid into the user-namespace specified by @targ and | |
258 | * return the resulting uid. | |
259 | * | |
260 | * There is always a mapping into the initial user_namespace. | |
261 | * | |
262 | * If @kuid has no mapping in @targ (uid_t)-1 is returned. | |
263 | */ | |
264 | uid_t from_kuid(struct user_namespace *targ, kuid_t kuid) | |
265 | { | |
266 | /* Map the uid from a global kernel uid */ | |
267 | return map_id_up(&targ->uid_map, __kuid_val(kuid)); | |
268 | } | |
269 | EXPORT_SYMBOL(from_kuid); | |
270 | ||
271 | /** | |
272 | * from_kuid_munged - Create a uid from a kuid user-namespace pair. | |
273 | * @targ: The user namespace we want a uid in. | |
274 | * @kuid: The kernel internal uid to start with. | |
275 | * | |
276 | * Map @kuid into the user-namespace specified by @targ and | |
277 | * return the resulting uid. | |
278 | * | |
279 | * There is always a mapping into the initial user_namespace. | |
280 | * | |
281 | * Unlike from_kuid from_kuid_munged never fails and always | |
282 | * returns a valid uid. This makes from_kuid_munged appropriate | |
283 | * for use in syscalls like stat and getuid where failing the | |
284 | * system call and failing to provide a valid uid are not an | |
285 | * options. | |
286 | * | |
287 | * If @kuid has no mapping in @targ overflowuid is returned. | |
288 | */ | |
289 | uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid) | |
290 | { | |
291 | uid_t uid; | |
292 | uid = from_kuid(targ, kuid); | |
293 | ||
294 | if (uid == (uid_t) -1) | |
295 | uid = overflowuid; | |
296 | return uid; | |
297 | } | |
298 | EXPORT_SYMBOL(from_kuid_munged); | |
299 | ||
300 | /** | |
301 | * make_kgid - Map a user-namespace gid pair into a kgid. | |
302 | * @ns: User namespace that the gid is in | |
68a9a435 | 303 | * @gid: group identifier |
22d917d8 EB |
304 | * |
305 | * Maps a user-namespace gid pair into a kernel internal kgid, | |
306 | * and returns that kgid. | |
307 | * | |
308 | * When there is no mapping defined for the user-namespace gid | |
309 | * pair INVALID_GID is returned. Callers are expected to test | |
310 | * for and handle INVALID_GID being returned. INVALID_GID may be | |
311 | * tested for using gid_valid(). | |
312 | */ | |
313 | kgid_t make_kgid(struct user_namespace *ns, gid_t gid) | |
314 | { | |
315 | /* Map the gid to a global kernel gid */ | |
316 | return KGIDT_INIT(map_id_down(&ns->gid_map, gid)); | |
317 | } | |
318 | EXPORT_SYMBOL(make_kgid); | |
319 | ||
320 | /** | |
321 | * from_kgid - Create a gid from a kgid user-namespace pair. | |
322 | * @targ: The user namespace we want a gid in. | |
323 | * @kgid: The kernel internal gid to start with. | |
324 | * | |
325 | * Map @kgid into the user-namespace specified by @targ and | |
326 | * return the resulting gid. | |
327 | * | |
328 | * There is always a mapping into the initial user_namespace. | |
329 | * | |
330 | * If @kgid has no mapping in @targ (gid_t)-1 is returned. | |
331 | */ | |
332 | gid_t from_kgid(struct user_namespace *targ, kgid_t kgid) | |
333 | { | |
334 | /* Map the gid from a global kernel gid */ | |
335 | return map_id_up(&targ->gid_map, __kgid_val(kgid)); | |
336 | } | |
337 | EXPORT_SYMBOL(from_kgid); | |
338 | ||
339 | /** | |
340 | * from_kgid_munged - Create a gid from a kgid user-namespace pair. | |
341 | * @targ: The user namespace we want a gid in. | |
342 | * @kgid: The kernel internal gid to start with. | |
343 | * | |
344 | * Map @kgid into the user-namespace specified by @targ and | |
345 | * return the resulting gid. | |
346 | * | |
347 | * There is always a mapping into the initial user_namespace. | |
348 | * | |
349 | * Unlike from_kgid from_kgid_munged never fails and always | |
350 | * returns a valid gid. This makes from_kgid_munged appropriate | |
351 | * for use in syscalls like stat and getgid where failing the | |
352 | * system call and failing to provide a valid gid are not options. | |
353 | * | |
354 | * If @kgid has no mapping in @targ overflowgid is returned. | |
355 | */ | |
356 | gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid) | |
357 | { | |
358 | gid_t gid; | |
359 | gid = from_kgid(targ, kgid); | |
360 | ||
361 | if (gid == (gid_t) -1) | |
362 | gid = overflowgid; | |
363 | return gid; | |
364 | } | |
365 | EXPORT_SYMBOL(from_kgid_munged); | |
366 | ||
f76d207a EB |
367 | /** |
368 | * make_kprojid - Map a user-namespace projid pair into a kprojid. | |
369 | * @ns: User namespace that the projid is in | |
370 | * @projid: Project identifier | |
371 | * | |
372 | * Maps a user-namespace uid pair into a kernel internal kuid, | |
373 | * and returns that kuid. | |
374 | * | |
375 | * When there is no mapping defined for the user-namespace projid | |
376 | * pair INVALID_PROJID is returned. Callers are expected to test | |
377 | * for and handle handle INVALID_PROJID being returned. INVALID_PROJID | |
378 | * may be tested for using projid_valid(). | |
379 | */ | |
380 | kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid) | |
381 | { | |
382 | /* Map the uid to a global kernel uid */ | |
383 | return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid)); | |
384 | } | |
385 | EXPORT_SYMBOL(make_kprojid); | |
386 | ||
387 | /** | |
388 | * from_kprojid - Create a projid from a kprojid user-namespace pair. | |
389 | * @targ: The user namespace we want a projid in. | |
390 | * @kprojid: The kernel internal project identifier to start with. | |
391 | * | |
392 | * Map @kprojid into the user-namespace specified by @targ and | |
393 | * return the resulting projid. | |
394 | * | |
395 | * There is always a mapping into the initial user_namespace. | |
396 | * | |
397 | * If @kprojid has no mapping in @targ (projid_t)-1 is returned. | |
398 | */ | |
399 | projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid) | |
400 | { | |
401 | /* Map the uid from a global kernel uid */ | |
402 | return map_id_up(&targ->projid_map, __kprojid_val(kprojid)); | |
403 | } | |
404 | EXPORT_SYMBOL(from_kprojid); | |
405 | ||
406 | /** | |
407 | * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair. | |
408 | * @targ: The user namespace we want a projid in. | |
409 | * @kprojid: The kernel internal projid to start with. | |
410 | * | |
411 | * Map @kprojid into the user-namespace specified by @targ and | |
412 | * return the resulting projid. | |
413 | * | |
414 | * There is always a mapping into the initial user_namespace. | |
415 | * | |
416 | * Unlike from_kprojid from_kprojid_munged never fails and always | |
417 | * returns a valid projid. This makes from_kprojid_munged | |
418 | * appropriate for use in syscalls like stat and where | |
419 | * failing the system call and failing to provide a valid projid are | |
420 | * not an options. | |
421 | * | |
422 | * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned. | |
423 | */ | |
424 | projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid) | |
425 | { | |
426 | projid_t projid; | |
427 | projid = from_kprojid(targ, kprojid); | |
428 | ||
429 | if (projid == (projid_t) -1) | |
430 | projid = OVERFLOW_PROJID; | |
431 | return projid; | |
432 | } | |
433 | EXPORT_SYMBOL(from_kprojid_munged); | |
434 | ||
435 | ||
22d917d8 EB |
436 | static int uid_m_show(struct seq_file *seq, void *v) |
437 | { | |
438 | struct user_namespace *ns = seq->private; | |
439 | struct uid_gid_extent *extent = v; | |
440 | struct user_namespace *lower_ns; | |
441 | uid_t lower; | |
5c1469de | 442 | |
c450f371 | 443 | lower_ns = seq_user_ns(seq); |
22d917d8 EB |
444 | if ((lower_ns == ns) && lower_ns->parent) |
445 | lower_ns = lower_ns->parent; | |
446 | ||
447 | lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first)); | |
448 | ||
449 | seq_printf(seq, "%10u %10u %10u\n", | |
450 | extent->first, | |
451 | lower, | |
452 | extent->count); | |
453 | ||
454 | return 0; | |
5c1469de EB |
455 | } |
456 | ||
22d917d8 | 457 | static int gid_m_show(struct seq_file *seq, void *v) |
5c1469de | 458 | { |
22d917d8 EB |
459 | struct user_namespace *ns = seq->private; |
460 | struct uid_gid_extent *extent = v; | |
461 | struct user_namespace *lower_ns; | |
462 | gid_t lower; | |
5c1469de | 463 | |
c450f371 | 464 | lower_ns = seq_user_ns(seq); |
22d917d8 EB |
465 | if ((lower_ns == ns) && lower_ns->parent) |
466 | lower_ns = lower_ns->parent; | |
5c1469de | 467 | |
22d917d8 EB |
468 | lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first)); |
469 | ||
470 | seq_printf(seq, "%10u %10u %10u\n", | |
471 | extent->first, | |
472 | lower, | |
473 | extent->count); | |
474 | ||
475 | return 0; | |
476 | } | |
477 | ||
f76d207a EB |
478 | static int projid_m_show(struct seq_file *seq, void *v) |
479 | { | |
480 | struct user_namespace *ns = seq->private; | |
481 | struct uid_gid_extent *extent = v; | |
482 | struct user_namespace *lower_ns; | |
483 | projid_t lower; | |
484 | ||
485 | lower_ns = seq_user_ns(seq); | |
486 | if ((lower_ns == ns) && lower_ns->parent) | |
487 | lower_ns = lower_ns->parent; | |
488 | ||
489 | lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first)); | |
490 | ||
491 | seq_printf(seq, "%10u %10u %10u\n", | |
492 | extent->first, | |
493 | lower, | |
494 | extent->count); | |
495 | ||
496 | return 0; | |
497 | } | |
498 | ||
68a9a435 FF |
499 | static void *m_start(struct seq_file *seq, loff_t *ppos, |
500 | struct uid_gid_map *map) | |
22d917d8 EB |
501 | { |
502 | struct uid_gid_extent *extent = NULL; | |
503 | loff_t pos = *ppos; | |
504 | ||
505 | if (pos < map->nr_extents) | |
506 | extent = &map->extent[pos]; | |
507 | ||
508 | return extent; | |
509 | } | |
510 | ||
511 | static void *uid_m_start(struct seq_file *seq, loff_t *ppos) | |
512 | { | |
513 | struct user_namespace *ns = seq->private; | |
514 | ||
515 | return m_start(seq, ppos, &ns->uid_map); | |
516 | } | |
517 | ||
518 | static void *gid_m_start(struct seq_file *seq, loff_t *ppos) | |
519 | { | |
520 | struct user_namespace *ns = seq->private; | |
521 | ||
522 | return m_start(seq, ppos, &ns->gid_map); | |
523 | } | |
524 | ||
f76d207a EB |
525 | static void *projid_m_start(struct seq_file *seq, loff_t *ppos) |
526 | { | |
527 | struct user_namespace *ns = seq->private; | |
528 | ||
529 | return m_start(seq, ppos, &ns->projid_map); | |
530 | } | |
531 | ||
22d917d8 EB |
532 | static void *m_next(struct seq_file *seq, void *v, loff_t *pos) |
533 | { | |
534 | (*pos)++; | |
535 | return seq->op->start(seq, pos); | |
536 | } | |
537 | ||
538 | static void m_stop(struct seq_file *seq, void *v) | |
539 | { | |
540 | return; | |
541 | } | |
542 | ||
ccf94f1b | 543 | const struct seq_operations proc_uid_seq_operations = { |
22d917d8 EB |
544 | .start = uid_m_start, |
545 | .stop = m_stop, | |
546 | .next = m_next, | |
547 | .show = uid_m_show, | |
548 | }; | |
549 | ||
ccf94f1b | 550 | const struct seq_operations proc_gid_seq_operations = { |
22d917d8 EB |
551 | .start = gid_m_start, |
552 | .stop = m_stop, | |
553 | .next = m_next, | |
554 | .show = gid_m_show, | |
555 | }; | |
556 | ||
ccf94f1b | 557 | const struct seq_operations proc_projid_seq_operations = { |
f76d207a EB |
558 | .start = projid_m_start, |
559 | .stop = m_stop, | |
560 | .next = m_next, | |
561 | .show = projid_m_show, | |
562 | }; | |
563 | ||
68a9a435 FF |
564 | static bool mappings_overlap(struct uid_gid_map *new_map, |
565 | struct uid_gid_extent *extent) | |
0bd14b4f EB |
566 | { |
567 | u32 upper_first, lower_first, upper_last, lower_last; | |
568 | unsigned idx; | |
569 | ||
570 | upper_first = extent->first; | |
571 | lower_first = extent->lower_first; | |
572 | upper_last = upper_first + extent->count - 1; | |
573 | lower_last = lower_first + extent->count - 1; | |
574 | ||
575 | for (idx = 0; idx < new_map->nr_extents; idx++) { | |
576 | u32 prev_upper_first, prev_lower_first; | |
577 | u32 prev_upper_last, prev_lower_last; | |
578 | struct uid_gid_extent *prev; | |
579 | ||
580 | prev = &new_map->extent[idx]; | |
581 | ||
582 | prev_upper_first = prev->first; | |
583 | prev_lower_first = prev->lower_first; | |
584 | prev_upper_last = prev_upper_first + prev->count - 1; | |
585 | prev_lower_last = prev_lower_first + prev->count - 1; | |
586 | ||
587 | /* Does the upper range intersect a previous extent? */ | |
588 | if ((prev_upper_first <= upper_last) && | |
589 | (prev_upper_last >= upper_first)) | |
590 | return true; | |
591 | ||
592 | /* Does the lower range intersect a previous extent? */ | |
593 | if ((prev_lower_first <= lower_last) && | |
594 | (prev_lower_last >= lower_first)) | |
595 | return true; | |
596 | } | |
597 | return false; | |
598 | } | |
599 | ||
22d917d8 EB |
600 | static ssize_t map_write(struct file *file, const char __user *buf, |
601 | size_t count, loff_t *ppos, | |
602 | int cap_setid, | |
603 | struct uid_gid_map *map, | |
604 | struct uid_gid_map *parent_map) | |
605 | { | |
606 | struct seq_file *seq = file->private_data; | |
607 | struct user_namespace *ns = seq->private; | |
608 | struct uid_gid_map new_map; | |
609 | unsigned idx; | |
0bd14b4f | 610 | struct uid_gid_extent *extent = NULL; |
22d917d8 EB |
611 | unsigned long page = 0; |
612 | char *kbuf, *pos, *next_line; | |
613 | ssize_t ret = -EINVAL; | |
614 | ||
615 | /* | |
f0d62aec | 616 | * The userns_state_mutex serializes all writes to any given map. |
22d917d8 EB |
617 | * |
618 | * Any map is only ever written once. | |
619 | * | |
620 | * An id map fits within 1 cache line on most architectures. | |
621 | * | |
622 | * On read nothing needs to be done unless you are on an | |
623 | * architecture with a crazy cache coherency model like alpha. | |
624 | * | |
625 | * There is a one time data dependency between reading the | |
626 | * count of the extents and the values of the extents. The | |
627 | * desired behavior is to see the values of the extents that | |
628 | * were written before the count of the extents. | |
629 | * | |
630 | * To achieve this smp_wmb() is used on guarantee the write | |
e79323bd MP |
631 | * order and smp_rmb() is guaranteed that we don't have crazy |
632 | * architectures returning stale data. | |
22d917d8 | 633 | */ |
f0d62aec | 634 | mutex_lock(&userns_state_mutex); |
22d917d8 EB |
635 | |
636 | ret = -EPERM; | |
637 | /* Only allow one successful write to the map */ | |
638 | if (map->nr_extents != 0) | |
639 | goto out; | |
640 | ||
41c21e35 AL |
641 | /* |
642 | * Adjusting namespace settings requires capabilities on the target. | |
5c1469de | 643 | */ |
41c21e35 | 644 | if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN)) |
22d917d8 EB |
645 | goto out; |
646 | ||
647 | /* Get a buffer */ | |
648 | ret = -ENOMEM; | |
649 | page = __get_free_page(GFP_TEMPORARY); | |
650 | kbuf = (char *) page; | |
651 | if (!page) | |
652 | goto out; | |
653 | ||
36476bea | 654 | /* Only allow < page size writes at the beginning of the file */ |
22d917d8 EB |
655 | ret = -EINVAL; |
656 | if ((*ppos != 0) || (count >= PAGE_SIZE)) | |
657 | goto out; | |
658 | ||
659 | /* Slurp in the user data */ | |
660 | ret = -EFAULT; | |
661 | if (copy_from_user(kbuf, buf, count)) | |
662 | goto out; | |
663 | kbuf[count] = '\0'; | |
664 | ||
665 | /* Parse the user data */ | |
666 | ret = -EINVAL; | |
667 | pos = kbuf; | |
668 | new_map.nr_extents = 0; | |
68a9a435 | 669 | for (; pos; pos = next_line) { |
22d917d8 EB |
670 | extent = &new_map.extent[new_map.nr_extents]; |
671 | ||
672 | /* Find the end of line and ensure I don't look past it */ | |
673 | next_line = strchr(pos, '\n'); | |
674 | if (next_line) { | |
675 | *next_line = '\0'; | |
676 | next_line++; | |
677 | if (*next_line == '\0') | |
678 | next_line = NULL; | |
5c1469de | 679 | } |
22d917d8 EB |
680 | |
681 | pos = skip_spaces(pos); | |
682 | extent->first = simple_strtoul(pos, &pos, 10); | |
683 | if (!isspace(*pos)) | |
684 | goto out; | |
685 | ||
686 | pos = skip_spaces(pos); | |
687 | extent->lower_first = simple_strtoul(pos, &pos, 10); | |
688 | if (!isspace(*pos)) | |
689 | goto out; | |
690 | ||
691 | pos = skip_spaces(pos); | |
692 | extent->count = simple_strtoul(pos, &pos, 10); | |
693 | if (*pos && !isspace(*pos)) | |
694 | goto out; | |
695 | ||
696 | /* Verify there is not trailing junk on the line */ | |
697 | pos = skip_spaces(pos); | |
698 | if (*pos != '\0') | |
699 | goto out; | |
700 | ||
701 | /* Verify we have been given valid starting values */ | |
702 | if ((extent->first == (u32) -1) || | |
68a9a435 | 703 | (extent->lower_first == (u32) -1)) |
22d917d8 EB |
704 | goto out; |
705 | ||
68a9a435 FF |
706 | /* Verify count is not zero and does not cause the |
707 | * extent to wrap | |
708 | */ | |
22d917d8 EB |
709 | if ((extent->first + extent->count) <= extent->first) |
710 | goto out; | |
68a9a435 FF |
711 | if ((extent->lower_first + extent->count) <= |
712 | extent->lower_first) | |
22d917d8 EB |
713 | goto out; |
714 | ||
0bd14b4f EB |
715 | /* Do the ranges in extent overlap any previous extents? */ |
716 | if (mappings_overlap(&new_map, extent)) | |
22d917d8 EB |
717 | goto out; |
718 | ||
719 | new_map.nr_extents++; | |
22d917d8 EB |
720 | |
721 | /* Fail if the file contains too many extents */ | |
722 | if ((new_map.nr_extents == UID_GID_MAP_MAX_EXTENTS) && | |
723 | (next_line != NULL)) | |
724 | goto out; | |
5c1469de | 725 | } |
22d917d8 EB |
726 | /* Be very certaint the new map actually exists */ |
727 | if (new_map.nr_extents == 0) | |
728 | goto out; | |
729 | ||
730 | ret = -EPERM; | |
731 | /* Validate the user is allowed to use user id's mapped to. */ | |
6708075f | 732 | if (!new_idmap_permitted(file, ns, cap_setid, &new_map)) |
22d917d8 EB |
733 | goto out; |
734 | ||
735 | /* Map the lower ids from the parent user namespace to the | |
736 | * kernel global id space. | |
737 | */ | |
738 | for (idx = 0; idx < new_map.nr_extents; idx++) { | |
739 | u32 lower_first; | |
740 | extent = &new_map.extent[idx]; | |
741 | ||
742 | lower_first = map_id_range_down(parent_map, | |
743 | extent->lower_first, | |
744 | extent->count); | |
745 | ||
746 | /* Fail if we can not map the specified extent to | |
747 | * the kernel global id space. | |
748 | */ | |
749 | if (lower_first == (u32) -1) | |
750 | goto out; | |
751 | ||
752 | extent->lower_first = lower_first; | |
753 | } | |
754 | ||
755 | /* Install the map */ | |
756 | memcpy(map->extent, new_map.extent, | |
757 | new_map.nr_extents*sizeof(new_map.extent[0])); | |
758 | smp_wmb(); | |
759 | map->nr_extents = new_map.nr_extents; | |
760 | ||
761 | *ppos = count; | |
762 | ret = count; | |
763 | out: | |
f0d62aec | 764 | mutex_unlock(&userns_state_mutex); |
22d917d8 EB |
765 | if (page) |
766 | free_page(page); | |
767 | return ret; | |
768 | } | |
769 | ||
68a9a435 FF |
770 | ssize_t proc_uid_map_write(struct file *file, const char __user *buf, |
771 | size_t size, loff_t *ppos) | |
22d917d8 EB |
772 | { |
773 | struct seq_file *seq = file->private_data; | |
774 | struct user_namespace *ns = seq->private; | |
c450f371 | 775 | struct user_namespace *seq_ns = seq_user_ns(seq); |
22d917d8 EB |
776 | |
777 | if (!ns->parent) | |
778 | return -EPERM; | |
779 | ||
c450f371 EB |
780 | if ((seq_ns != ns) && (seq_ns != ns->parent)) |
781 | return -EPERM; | |
782 | ||
22d917d8 EB |
783 | return map_write(file, buf, size, ppos, CAP_SETUID, |
784 | &ns->uid_map, &ns->parent->uid_map); | |
785 | } | |
786 | ||
68a9a435 FF |
787 | ssize_t proc_gid_map_write(struct file *file, const char __user *buf, |
788 | size_t size, loff_t *ppos) | |
22d917d8 EB |
789 | { |
790 | struct seq_file *seq = file->private_data; | |
791 | struct user_namespace *ns = seq->private; | |
c450f371 | 792 | struct user_namespace *seq_ns = seq_user_ns(seq); |
22d917d8 EB |
793 | |
794 | if (!ns->parent) | |
795 | return -EPERM; | |
796 | ||
c450f371 EB |
797 | if ((seq_ns != ns) && (seq_ns != ns->parent)) |
798 | return -EPERM; | |
799 | ||
22d917d8 EB |
800 | return map_write(file, buf, size, ppos, CAP_SETGID, |
801 | &ns->gid_map, &ns->parent->gid_map); | |
802 | } | |
803 | ||
68a9a435 FF |
804 | ssize_t proc_projid_map_write(struct file *file, const char __user *buf, |
805 | size_t size, loff_t *ppos) | |
f76d207a EB |
806 | { |
807 | struct seq_file *seq = file->private_data; | |
808 | struct user_namespace *ns = seq->private; | |
809 | struct user_namespace *seq_ns = seq_user_ns(seq); | |
810 | ||
811 | if (!ns->parent) | |
812 | return -EPERM; | |
813 | ||
814 | if ((seq_ns != ns) && (seq_ns != ns->parent)) | |
815 | return -EPERM; | |
816 | ||
817 | /* Anyone can set any valid project id no capability needed */ | |
818 | return map_write(file, buf, size, ppos, -1, | |
819 | &ns->projid_map, &ns->parent->projid_map); | |
820 | } | |
821 | ||
68a9a435 | 822 | static bool new_idmap_permitted(const struct file *file, |
6708075f | 823 | struct user_namespace *ns, int cap_setid, |
22d917d8 EB |
824 | struct uid_gid_map *new_map) |
825 | { | |
f95d7918 | 826 | const struct cred *cred = file->f_cred; |
0542f17b EB |
827 | /* Don't allow mappings that would allow anything that wouldn't |
828 | * be allowed without the establishment of unprivileged mappings. | |
829 | */ | |
f95d7918 EB |
830 | if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1) && |
831 | uid_eq(ns->owner, cred->euid)) { | |
37657da3 EB |
832 | u32 id = new_map->extent[0].lower_first; |
833 | if (cap_setid == CAP_SETUID) { | |
834 | kuid_t uid = make_kuid(ns->parent, id); | |
f95d7918 | 835 | if (uid_eq(uid, cred->euid)) |
37657da3 | 836 | return true; |
68a9a435 | 837 | } else if (cap_setid == CAP_SETGID) { |
37657da3 | 838 | kgid_t gid = make_kgid(ns->parent, id); |
66d2f338 EB |
839 | if (!(ns->flags & USERNS_SETGROUPS_ALLOWED) && |
840 | gid_eq(gid, cred->egid)) | |
37657da3 EB |
841 | return true; |
842 | } | |
843 | } | |
844 | ||
f76d207a EB |
845 | /* Allow anyone to set a mapping that doesn't require privilege */ |
846 | if (!cap_valid(cap_setid)) | |
847 | return true; | |
848 | ||
22d917d8 EB |
849 | /* Allow the specified ids if we have the appropriate capability |
850 | * (CAP_SETUID or CAP_SETGID) over the parent user namespace. | |
6708075f | 851 | * And the opener of the id file also had the approprpiate capability. |
22d917d8 | 852 | */ |
6708075f EB |
853 | if (ns_capable(ns->parent, cap_setid) && |
854 | file_ns_capable(file, ns->parent, cap_setid)) | |
22d917d8 | 855 | return true; |
5c1469de | 856 | |
22d917d8 | 857 | return false; |
5c1469de | 858 | } |
6164281a | 859 | |
9cc46516 EB |
860 | int proc_setgroups_show(struct seq_file *seq, void *v) |
861 | { | |
862 | struct user_namespace *ns = seq->private; | |
863 | unsigned long userns_flags = ACCESS_ONCE(ns->flags); | |
864 | ||
865 | seq_printf(seq, "%s\n", | |
866 | (userns_flags & USERNS_SETGROUPS_ALLOWED) ? | |
867 | "allow" : "deny"); | |
868 | return 0; | |
869 | } | |
870 | ||
871 | ssize_t proc_setgroups_write(struct file *file, const char __user *buf, | |
872 | size_t count, loff_t *ppos) | |
873 | { | |
874 | struct seq_file *seq = file->private_data; | |
875 | struct user_namespace *ns = seq->private; | |
876 | char kbuf[8], *pos; | |
877 | bool setgroups_allowed; | |
878 | ssize_t ret; | |
879 | ||
880 | /* Only allow a very narrow range of strings to be written */ | |
881 | ret = -EINVAL; | |
882 | if ((*ppos != 0) || (count >= sizeof(kbuf))) | |
883 | goto out; | |
884 | ||
885 | /* What was written? */ | |
886 | ret = -EFAULT; | |
887 | if (copy_from_user(kbuf, buf, count)) | |
888 | goto out; | |
889 | kbuf[count] = '\0'; | |
890 | pos = kbuf; | |
891 | ||
892 | /* What is being requested? */ | |
893 | ret = -EINVAL; | |
894 | if (strncmp(pos, "allow", 5) == 0) { | |
895 | pos += 5; | |
896 | setgroups_allowed = true; | |
897 | } | |
898 | else if (strncmp(pos, "deny", 4) == 0) { | |
899 | pos += 4; | |
900 | setgroups_allowed = false; | |
901 | } | |
902 | else | |
903 | goto out; | |
904 | ||
905 | /* Verify there is not trailing junk on the line */ | |
906 | pos = skip_spaces(pos); | |
907 | if (*pos != '\0') | |
908 | goto out; | |
909 | ||
910 | ret = -EPERM; | |
911 | mutex_lock(&userns_state_mutex); | |
912 | if (setgroups_allowed) { | |
913 | /* Enabling setgroups after setgroups has been disabled | |
914 | * is not allowed. | |
915 | */ | |
916 | if (!(ns->flags & USERNS_SETGROUPS_ALLOWED)) | |
917 | goto out_unlock; | |
918 | } else { | |
919 | /* Permanently disabling setgroups after setgroups has | |
920 | * been enabled by writing the gid_map is not allowed. | |
921 | */ | |
922 | if (ns->gid_map.nr_extents != 0) | |
923 | goto out_unlock; | |
924 | ns->flags &= ~USERNS_SETGROUPS_ALLOWED; | |
925 | } | |
926 | mutex_unlock(&userns_state_mutex); | |
927 | ||
928 | /* Report a successful write */ | |
929 | *ppos = count; | |
930 | ret = count; | |
931 | out: | |
932 | return ret; | |
933 | out_unlock: | |
934 | mutex_unlock(&userns_state_mutex); | |
935 | goto out; | |
936 | } | |
937 | ||
273d2c67 EB |
938 | bool userns_may_setgroups(const struct user_namespace *ns) |
939 | { | |
940 | bool allowed; | |
941 | ||
f0d62aec | 942 | mutex_lock(&userns_state_mutex); |
273d2c67 EB |
943 | /* It is not safe to use setgroups until a gid mapping in |
944 | * the user namespace has been established. | |
945 | */ | |
946 | allowed = ns->gid_map.nr_extents != 0; | |
9cc46516 EB |
947 | /* Is setgroups allowed? */ |
948 | allowed = allowed && (ns->flags & USERNS_SETGROUPS_ALLOWED); | |
f0d62aec | 949 | mutex_unlock(&userns_state_mutex); |
273d2c67 EB |
950 | |
951 | return allowed; | |
952 | } | |
953 | ||
3c041184 AV |
954 | static inline struct user_namespace *to_user_ns(struct ns_common *ns) |
955 | { | |
956 | return container_of(ns, struct user_namespace, ns); | |
957 | } | |
958 | ||
64964528 | 959 | static struct ns_common *userns_get(struct task_struct *task) |
cde1975b EB |
960 | { |
961 | struct user_namespace *user_ns; | |
962 | ||
963 | rcu_read_lock(); | |
964 | user_ns = get_user_ns(__task_cred(task)->user_ns); | |
965 | rcu_read_unlock(); | |
966 | ||
3c041184 | 967 | return user_ns ? &user_ns->ns : NULL; |
cde1975b EB |
968 | } |
969 | ||
64964528 | 970 | static void userns_put(struct ns_common *ns) |
cde1975b | 971 | { |
3c041184 | 972 | put_user_ns(to_user_ns(ns)); |
cde1975b EB |
973 | } |
974 | ||
64964528 | 975 | static int userns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
cde1975b | 976 | { |
3c041184 | 977 | struct user_namespace *user_ns = to_user_ns(ns); |
cde1975b EB |
978 | struct cred *cred; |
979 | ||
980 | /* Don't allow gaining capabilities by reentering | |
981 | * the same user namespace. | |
982 | */ | |
983 | if (user_ns == current_user_ns()) | |
984 | return -EINVAL; | |
985 | ||
faf00da5 EB |
986 | /* Tasks that share a thread group must share a user namespace */ |
987 | if (!thread_group_empty(current)) | |
cde1975b EB |
988 | return -EINVAL; |
989 | ||
e66eded8 EB |
990 | if (current->fs->users != 1) |
991 | return -EINVAL; | |
992 | ||
cde1975b EB |
993 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
994 | return -EPERM; | |
995 | ||
996 | cred = prepare_creds(); | |
997 | if (!cred) | |
998 | return -ENOMEM; | |
999 | ||
1000 | put_user_ns(cred->user_ns); | |
1001 | set_cred_user_ns(cred, get_user_ns(user_ns)); | |
1002 | ||
1003 | return commit_creds(cred); | |
1004 | } | |
1005 | ||
1006 | const struct proc_ns_operations userns_operations = { | |
1007 | .name = "user", | |
1008 | .type = CLONE_NEWUSER, | |
1009 | .get = userns_get, | |
1010 | .put = userns_put, | |
1011 | .install = userns_install, | |
1012 | }; | |
1013 | ||
6164281a PE |
1014 | static __init int user_namespaces_init(void) |
1015 | { | |
1016 | user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC); | |
1017 | return 0; | |
1018 | } | |
c96d6660 | 1019 | subsys_initcall(user_namespaces_init); |