blob: 1076608ff38f8383a063b32618ce6ca40b4a951b
1 | /* Common capabilities, needed by capability.o. |
2 | * |
3 | * This program is free software; you can redistribute it and/or modify |
4 | * it under the terms of the GNU General Public License as published by |
5 | * the Free Software Foundation; either version 2 of the License, or |
6 | * (at your option) any later version. |
7 | * |
8 | */ |
9 | |
10 | #include <linux/capability.h> |
11 | #include <linux/audit.h> |
12 | #include <linux/module.h> |
13 | #include <linux/init.h> |
14 | #include <linux/kernel.h> |
15 | #include <linux/lsm_hooks.h> |
16 | #include <linux/file.h> |
17 | #include <linux/mm.h> |
18 | #include <linux/mman.h> |
19 | #include <linux/pagemap.h> |
20 | #include <linux/swap.h> |
21 | #include <linux/skbuff.h> |
22 | #include <linux/netlink.h> |
23 | #include <linux/ptrace.h> |
24 | #include <linux/xattr.h> |
25 | #include <linux/hugetlb.h> |
26 | #include <linux/mount.h> |
27 | #include <linux/sched.h> |
28 | #include <linux/prctl.h> |
29 | #include <linux/securebits.h> |
30 | #include <linux/user_namespace.h> |
31 | #include <linux/binfmts.h> |
32 | #include <linux/personality.h> |
33 | |
34 | #ifdef CONFIG_ANDROID_PARANOID_NETWORK |
35 | #include <linux/android_aid.h> |
36 | #endif |
37 | |
38 | /* |
39 | * If a non-root user executes a setuid-root binary in |
40 | * !secure(SECURE_NOROOT) mode, then we raise capabilities. |
41 | * However if fE is also set, then the intent is for only |
42 | * the file capabilities to be applied, and the setuid-root |
43 | * bit is left on either to change the uid (plausible) or |
44 | * to get full privilege on a kernel without file capabilities |
45 | * support. So in that case we do not raise capabilities. |
46 | * |
47 | * Warn if that happens, once per boot. |
48 | */ |
49 | static void warn_setuid_and_fcaps_mixed(const char *fname) |
50 | { |
51 | static int warned; |
52 | if (!warned) { |
53 | printk(KERN_INFO "warning: `%s' has both setuid-root and" |
54 | " effective capabilities. Therefore not raising all" |
55 | " capabilities.\n", fname); |
56 | warned = 1; |
57 | } |
58 | } |
59 | |
60 | /** |
61 | * __cap_capable - Determine whether a task has a particular effective capability |
62 | * @cred: The credentials to use |
63 | * @ns: The user namespace in which we need the capability |
64 | * @cap: The capability to check for |
65 | * @audit: Whether to write an audit message or not |
66 | * |
67 | * Determine whether the nominated task has the specified capability amongst |
68 | * its effective set, returning 0 if it does, -ve if it does not. |
69 | * |
70 | * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable() |
71 | * and has_capability() functions. That is, it has the reverse semantics: |
72 | * cap_has_capability() returns 0 when a task has a capability, but the |
73 | * kernel's capable() and has_capability() returns 1 for this case. |
74 | */ |
75 | int __cap_capable(const struct cred *cred, struct user_namespace *targ_ns, |
76 | int cap, int audit) |
77 | { |
78 | struct user_namespace *ns = targ_ns; |
79 | |
80 | /* See if cred has the capability in the target user namespace |
81 | * by examining the target user namespace and all of the target |
82 | * user namespace's parents. |
83 | */ |
84 | for (;;) { |
85 | /* Do we have the necessary capabilities? */ |
86 | if (ns == cred->user_ns) |
87 | return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM; |
88 | |
89 | /* Have we tried all of the parent namespaces? */ |
90 | if (ns == &init_user_ns) |
91 | return -EPERM; |
92 | |
93 | /* |
94 | * The owner of the user namespace in the parent of the |
95 | * user namespace has all caps. |
96 | */ |
97 | if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid)) |
98 | return 0; |
99 | |
100 | /* |
101 | * If you have a capability in a parent user ns, then you have |
102 | * it over all children user namespaces as well. |
103 | */ |
104 | ns = ns->parent; |
105 | } |
106 | |
107 | /* We never get here */ |
108 | } |
109 | |
110 | int cap_capable(const struct cred *cred, struct user_namespace *targ_ns, |
111 | int cap, int audit) |
112 | { |
113 | int ret = __cap_capable(cred, targ_ns, cap, audit); |
114 | |
115 | #ifdef CONFIG_ANDROID_PARANOID_NETWORK |
116 | if (ret != 0 && cap == CAP_NET_RAW && in_egroup_p(AID_NET_RAW)) { |
117 | printk("Process %s granted CAP_NET_RAW from Android group net_raw.\n", current->comm); |
118 | printk(" Please update the .rc file to explictly set 'capabilities NET_RAW'\n"); |
119 | printk(" Implicit grants are deprecated and will be removed in the future.\n"); |
120 | return 0; |
121 | } |
122 | if (ret != 0 && cap == CAP_NET_ADMIN && in_egroup_p(AID_NET_ADMIN)) { |
123 | printk("Process %s granted CAP_NET_ADMIN from Android group net_admin.\n", current->comm); |
124 | printk(" Please update the .rc file to explictly set 'capabilities NET_ADMIN'\n"); |
125 | printk(" Implicit grants are deprecated and will be removed in the future.\n"); |
126 | return 0; |
127 | } |
128 | #endif |
129 | return ret; |
130 | } |
131 | /** |
132 | * cap_settime - Determine whether the current process may set the system clock |
133 | * @ts: The time to set |
134 | * @tz: The timezone to set |
135 | * |
136 | * Determine whether the current process may set the system clock and timezone |
137 | * information, returning 0 if permission granted, -ve if denied. |
138 | */ |
139 | int cap_settime(const struct timespec64 *ts, const struct timezone *tz) |
140 | { |
141 | if (!capable(CAP_SYS_TIME)) |
142 | return -EPERM; |
143 | return 0; |
144 | } |
145 | |
146 | /** |
147 | * cap_ptrace_access_check - Determine whether the current process may access |
148 | * another |
149 | * @child: The process to be accessed |
150 | * @mode: The mode of attachment. |
151 | * |
152 | * If we are in the same or an ancestor user_ns and have all the target |
153 | * task's capabilities, then ptrace access is allowed. |
154 | * If we have the ptrace capability to the target user_ns, then ptrace |
155 | * access is allowed. |
156 | * Else denied. |
157 | * |
158 | * Determine whether a process may access another, returning 0 if permission |
159 | * granted, -ve if denied. |
160 | */ |
161 | int cap_ptrace_access_check(struct task_struct *child, unsigned int mode) |
162 | { |
163 | int ret = 0; |
164 | const struct cred *cred, *child_cred; |
165 | const kernel_cap_t *caller_caps; |
166 | |
167 | rcu_read_lock(); |
168 | cred = current_cred(); |
169 | child_cred = __task_cred(child); |
170 | if (mode & PTRACE_MODE_FSCREDS) |
171 | caller_caps = &cred->cap_effective; |
172 | else |
173 | caller_caps = &cred->cap_permitted; |
174 | if (cred->user_ns == child_cred->user_ns && |
175 | cap_issubset(child_cred->cap_permitted, *caller_caps)) |
176 | goto out; |
177 | if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE)) |
178 | goto out; |
179 | ret = -EPERM; |
180 | out: |
181 | rcu_read_unlock(); |
182 | return ret; |
183 | } |
184 | |
185 | /** |
186 | * cap_ptrace_traceme - Determine whether another process may trace the current |
187 | * @parent: The task proposed to be the tracer |
188 | * |
189 | * If parent is in the same or an ancestor user_ns and has all current's |
190 | * capabilities, then ptrace access is allowed. |
191 | * If parent has the ptrace capability to current's user_ns, then ptrace |
192 | * access is allowed. |
193 | * Else denied. |
194 | * |
195 | * Determine whether the nominated task is permitted to trace the current |
196 | * process, returning 0 if permission is granted, -ve if denied. |
197 | */ |
198 | int cap_ptrace_traceme(struct task_struct *parent) |
199 | { |
200 | int ret = 0; |
201 | const struct cred *cred, *child_cred; |
202 | |
203 | rcu_read_lock(); |
204 | cred = __task_cred(parent); |
205 | child_cred = current_cred(); |
206 | if (cred->user_ns == child_cred->user_ns && |
207 | cap_issubset(child_cred->cap_permitted, cred->cap_permitted)) |
208 | goto out; |
209 | if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE)) |
210 | goto out; |
211 | ret = -EPERM; |
212 | out: |
213 | rcu_read_unlock(); |
214 | return ret; |
215 | } |
216 | |
217 | /** |
218 | * cap_capget - Retrieve a task's capability sets |
219 | * @target: The task from which to retrieve the capability sets |
220 | * @effective: The place to record the effective set |
221 | * @inheritable: The place to record the inheritable set |
222 | * @permitted: The place to record the permitted set |
223 | * |
224 | * This function retrieves the capabilities of the nominated task and returns |
225 | * them to the caller. |
226 | */ |
227 | int cap_capget(struct task_struct *target, kernel_cap_t *effective, |
228 | kernel_cap_t *inheritable, kernel_cap_t *permitted) |
229 | { |
230 | const struct cred *cred; |
231 | |
232 | /* Derived from kernel/capability.c:sys_capget. */ |
233 | rcu_read_lock(); |
234 | cred = __task_cred(target); |
235 | *effective = cred->cap_effective; |
236 | *inheritable = cred->cap_inheritable; |
237 | *permitted = cred->cap_permitted; |
238 | rcu_read_unlock(); |
239 | return 0; |
240 | } |
241 | |
242 | /* |
243 | * Determine whether the inheritable capabilities are limited to the old |
244 | * permitted set. Returns 1 if they are limited, 0 if they are not. |
245 | */ |
246 | static inline int cap_inh_is_capped(void) |
247 | { |
248 | |
249 | /* they are so limited unless the current task has the CAP_SETPCAP |
250 | * capability |
251 | */ |
252 | if (cap_capable(current_cred(), current_cred()->user_ns, |
253 | CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0) |
254 | return 0; |
255 | return 1; |
256 | } |
257 | |
258 | /** |
259 | * cap_capset - Validate and apply proposed changes to current's capabilities |
260 | * @new: The proposed new credentials; alterations should be made here |
261 | * @old: The current task's current credentials |
262 | * @effective: A pointer to the proposed new effective capabilities set |
263 | * @inheritable: A pointer to the proposed new inheritable capabilities set |
264 | * @permitted: A pointer to the proposed new permitted capabilities set |
265 | * |
266 | * This function validates and applies a proposed mass change to the current |
267 | * process's capability sets. The changes are made to the proposed new |
268 | * credentials, and assuming no error, will be committed by the caller of LSM. |
269 | */ |
270 | int cap_capset(struct cred *new, |
271 | const struct cred *old, |
272 | const kernel_cap_t *effective, |
273 | const kernel_cap_t *inheritable, |
274 | const kernel_cap_t *permitted) |
275 | { |
276 | if (cap_inh_is_capped() && |
277 | !cap_issubset(*inheritable, |
278 | cap_combine(old->cap_inheritable, |
279 | old->cap_permitted))) |
280 | /* incapable of using this inheritable set */ |
281 | return -EPERM; |
282 | |
283 | if (!cap_issubset(*inheritable, |
284 | cap_combine(old->cap_inheritable, |
285 | old->cap_bset))) |
286 | /* no new pI capabilities outside bounding set */ |
287 | return -EPERM; |
288 | |
289 | /* verify restrictions on target's new Permitted set */ |
290 | if (!cap_issubset(*permitted, old->cap_permitted)) |
291 | return -EPERM; |
292 | |
293 | /* verify the _new_Effective_ is a subset of the _new_Permitted_ */ |
294 | if (!cap_issubset(*effective, *permitted)) |
295 | return -EPERM; |
296 | |
297 | new->cap_effective = *effective; |
298 | new->cap_inheritable = *inheritable; |
299 | new->cap_permitted = *permitted; |
300 | |
301 | /* |
302 | * Mask off ambient bits that are no longer both permitted and |
303 | * inheritable. |
304 | */ |
305 | new->cap_ambient = cap_intersect(new->cap_ambient, |
306 | cap_intersect(*permitted, |
307 | *inheritable)); |
308 | if (WARN_ON(!cap_ambient_invariant_ok(new))) |
309 | return -EINVAL; |
310 | return 0; |
311 | } |
312 | |
313 | /* |
314 | * Clear proposed capability sets for execve(). |
315 | */ |
316 | static inline void bprm_clear_caps(struct linux_binprm *bprm) |
317 | { |
318 | cap_clear(bprm->cred->cap_permitted); |
319 | bprm->cap_effective = false; |
320 | } |
321 | |
322 | /** |
323 | * cap_inode_need_killpriv - Determine if inode change affects privileges |
324 | * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV |
325 | * |
326 | * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV |
327 | * affects the security markings on that inode, and if it is, should |
328 | * inode_killpriv() be invoked or the change rejected? |
329 | * |
330 | * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and |
331 | * -ve to deny the change. |
332 | */ |
333 | int cap_inode_need_killpriv(struct dentry *dentry) |
334 | { |
335 | struct inode *inode = d_backing_inode(dentry); |
336 | int error; |
337 | |
338 | error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0); |
339 | return error > 0; |
340 | } |
341 | |
342 | /** |
343 | * cap_inode_killpriv - Erase the security markings on an inode |
344 | * @dentry: The inode/dentry to alter |
345 | * |
346 | * Erase the privilege-enhancing security markings on an inode. |
347 | * |
348 | * Returns 0 if successful, -ve on error. |
349 | */ |
350 | int cap_inode_killpriv(struct dentry *dentry) |
351 | { |
352 | int error; |
353 | |
354 | error = __vfs_removexattr(dentry, XATTR_NAME_CAPS); |
355 | if (error == -EOPNOTSUPP) |
356 | error = 0; |
357 | return error; |
358 | } |
359 | |
360 | /* |
361 | * Calculate the new process capability sets from the capability sets attached |
362 | * to a file. |
363 | */ |
364 | static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps, |
365 | struct linux_binprm *bprm, |
366 | bool *effective, |
367 | bool *has_cap) |
368 | { |
369 | struct cred *new = bprm->cred; |
370 | unsigned i; |
371 | int ret = 0; |
372 | |
373 | if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE) |
374 | *effective = true; |
375 | |
376 | if (caps->magic_etc & VFS_CAP_REVISION_MASK) |
377 | *has_cap = true; |
378 | |
379 | CAP_FOR_EACH_U32(i) { |
380 | __u32 permitted = caps->permitted.cap[i]; |
381 | __u32 inheritable = caps->inheritable.cap[i]; |
382 | |
383 | /* |
384 | * pP' = (X & fP) | (pI & fI) |
385 | * The addition of pA' is handled later. |
386 | */ |
387 | new->cap_permitted.cap[i] = |
388 | (new->cap_bset.cap[i] & permitted) | |
389 | (new->cap_inheritable.cap[i] & inheritable); |
390 | |
391 | if (permitted & ~new->cap_permitted.cap[i]) |
392 | /* insufficient to execute correctly */ |
393 | ret = -EPERM; |
394 | } |
395 | |
396 | /* |
397 | * For legacy apps, with no internal support for recognizing they |
398 | * do not have enough capabilities, we return an error if they are |
399 | * missing some "forced" (aka file-permitted) capabilities. |
400 | */ |
401 | return *effective ? ret : 0; |
402 | } |
403 | |
404 | /* |
405 | * Extract the on-exec-apply capability sets for an executable file. |
406 | */ |
407 | int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps) |
408 | { |
409 | struct inode *inode = d_backing_inode(dentry); |
410 | __u32 magic_etc; |
411 | unsigned tocopy, i; |
412 | int size; |
413 | struct vfs_cap_data caps; |
414 | |
415 | memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data)); |
416 | |
417 | if (!inode) |
418 | return -ENODATA; |
419 | |
420 | size = __vfs_getxattr((struct dentry *)dentry, inode, |
421 | XATTR_NAME_CAPS, &caps, XATTR_CAPS_SZ); |
422 | if (size == -ENODATA || size == -EOPNOTSUPP) |
423 | /* no data, that's ok */ |
424 | return -ENODATA; |
425 | if (size < 0) |
426 | return size; |
427 | |
428 | if (size < sizeof(magic_etc)) |
429 | return -EINVAL; |
430 | |
431 | cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc); |
432 | |
433 | switch (magic_etc & VFS_CAP_REVISION_MASK) { |
434 | case VFS_CAP_REVISION_1: |
435 | if (size != XATTR_CAPS_SZ_1) |
436 | return -EINVAL; |
437 | tocopy = VFS_CAP_U32_1; |
438 | break; |
439 | case VFS_CAP_REVISION_2: |
440 | if (size != XATTR_CAPS_SZ_2) |
441 | return -EINVAL; |
442 | tocopy = VFS_CAP_U32_2; |
443 | break; |
444 | default: |
445 | return -EINVAL; |
446 | } |
447 | |
448 | CAP_FOR_EACH_U32(i) { |
449 | if (i >= tocopy) |
450 | break; |
451 | cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted); |
452 | cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable); |
453 | } |
454 | |
455 | cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; |
456 | cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; |
457 | |
458 | return 0; |
459 | } |
460 | |
461 | /* |
462 | * Attempt to get the on-exec apply capability sets for an executable file from |
463 | * its xattrs and, if present, apply them to the proposed credentials being |
464 | * constructed by execve(). |
465 | */ |
466 | static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap) |
467 | { |
468 | int rc = 0; |
469 | struct cpu_vfs_cap_data vcaps; |
470 | |
471 | bprm_clear_caps(bprm); |
472 | |
473 | if (!file_caps_enabled) |
474 | return 0; |
475 | |
476 | if (!mnt_may_suid(bprm->file->f_path.mnt)) |
477 | return 0; |
478 | |
479 | /* |
480 | * This check is redundant with mnt_may_suid() but is kept to make |
481 | * explicit that capability bits are limited to s_user_ns and its |
482 | * descendants. |
483 | */ |
484 | if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns)) |
485 | return 0; |
486 | |
487 | rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps); |
488 | if (rc < 0) { |
489 | if (rc == -EINVAL) |
490 | printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n", |
491 | __func__, rc, bprm->filename); |
492 | else if (rc == -ENODATA) |
493 | rc = 0; |
494 | goto out; |
495 | } |
496 | |
497 | rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap); |
498 | if (rc == -EINVAL) |
499 | printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n", |
500 | __func__, rc, bprm->filename); |
501 | |
502 | out: |
503 | if (rc) |
504 | bprm_clear_caps(bprm); |
505 | |
506 | return rc; |
507 | } |
508 | |
509 | /** |
510 | * cap_bprm_set_creds - Set up the proposed credentials for execve(). |
511 | * @bprm: The execution parameters, including the proposed creds |
512 | * |
513 | * Set up the proposed credentials for a new execution context being |
514 | * constructed by execve(). The proposed creds in @bprm->cred is altered, |
515 | * which won't take effect immediately. Returns 0 if successful, -ve on error. |
516 | */ |
517 | int cap_bprm_set_creds(struct linux_binprm *bprm) |
518 | { |
519 | const struct cred *old = current_cred(); |
520 | struct cred *new = bprm->cred; |
521 | bool effective, has_cap = false, is_setid; |
522 | int ret; |
523 | kuid_t root_uid; |
524 | |
525 | if (WARN_ON(!cap_ambient_invariant_ok(old))) |
526 | return -EPERM; |
527 | |
528 | effective = false; |
529 | ret = get_file_caps(bprm, &effective, &has_cap); |
530 | if (ret < 0) |
531 | return ret; |
532 | |
533 | root_uid = make_kuid(new->user_ns, 0); |
534 | |
535 | if (!issecure(SECURE_NOROOT)) { |
536 | /* |
537 | * If the legacy file capability is set, then don't set privs |
538 | * for a setuid root binary run by a non-root user. Do set it |
539 | * for a root user just to cause least surprise to an admin. |
540 | */ |
541 | if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) { |
542 | warn_setuid_and_fcaps_mixed(bprm->filename); |
543 | goto skip; |
544 | } |
545 | /* |
546 | * To support inheritance of root-permissions and suid-root |
547 | * executables under compatibility mode, we override the |
548 | * capability sets for the file. |
549 | * |
550 | * If only the real uid is 0, we do not set the effective bit. |
551 | */ |
552 | if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) { |
553 | /* pP' = (cap_bset & ~0) | (pI & ~0) */ |
554 | new->cap_permitted = cap_combine(old->cap_bset, |
555 | old->cap_inheritable); |
556 | } |
557 | if (uid_eq(new->euid, root_uid)) |
558 | effective = true; |
559 | } |
560 | skip: |
561 | |
562 | /* if we have fs caps, clear dangerous personality flags */ |
563 | if (!cap_issubset(new->cap_permitted, old->cap_permitted)) |
564 | bprm->per_clear |= PER_CLEAR_ON_SETID; |
565 | |
566 | |
567 | /* Don't let someone trace a set[ug]id/setpcap binary with the revised |
568 | * credentials unless they have the appropriate permit. |
569 | * |
570 | * In addition, if NO_NEW_PRIVS, then ensure we get no new privs. |
571 | */ |
572 | is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid); |
573 | |
574 | if ((is_setid || |
575 | !cap_issubset(new->cap_permitted, old->cap_permitted)) && |
576 | bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) { |
577 | /* downgrade; they get no more than they had, and maybe less */ |
578 | if (!capable(CAP_SETUID) || |
579 | (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) { |
580 | new->euid = new->uid; |
581 | new->egid = new->gid; |
582 | } |
583 | new->cap_permitted = cap_intersect(new->cap_permitted, |
584 | old->cap_permitted); |
585 | } |
586 | |
587 | new->suid = new->fsuid = new->euid; |
588 | new->sgid = new->fsgid = new->egid; |
589 | |
590 | /* File caps or setid cancels ambient. */ |
591 | if (has_cap || is_setid) |
592 | cap_clear(new->cap_ambient); |
593 | |
594 | /* |
595 | * Now that we've computed pA', update pP' to give: |
596 | * pP' = (X & fP) | (pI & fI) | pA' |
597 | */ |
598 | new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient); |
599 | |
600 | /* |
601 | * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set, |
602 | * this is the same as pE' = (fE ? pP' : 0) | pA'. |
603 | */ |
604 | if (effective) |
605 | new->cap_effective = new->cap_permitted; |
606 | else |
607 | new->cap_effective = new->cap_ambient; |
608 | |
609 | if (WARN_ON(!cap_ambient_invariant_ok(new))) |
610 | return -EPERM; |
611 | |
612 | bprm->cap_effective = effective; |
613 | |
614 | /* |
615 | * Audit candidate if current->cap_effective is set |
616 | * |
617 | * We do not bother to audit if 3 things are true: |
618 | * 1) cap_effective has all caps |
619 | * 2) we are root |
620 | * 3) root is supposed to have all caps (SECURE_NOROOT) |
621 | * Since this is just a normal root execing a process. |
622 | * |
623 | * Number 1 above might fail if you don't have a full bset, but I think |
624 | * that is interesting information to audit. |
625 | */ |
626 | if (!cap_issubset(new->cap_effective, new->cap_ambient)) { |
627 | if (!cap_issubset(CAP_FULL_SET, new->cap_effective) || |
628 | !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) || |
629 | issecure(SECURE_NOROOT)) { |
630 | ret = audit_log_bprm_fcaps(bprm, new, old); |
631 | if (ret < 0) |
632 | return ret; |
633 | } |
634 | } |
635 | |
636 | new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); |
637 | |
638 | if (WARN_ON(!cap_ambient_invariant_ok(new))) |
639 | return -EPERM; |
640 | |
641 | return 0; |
642 | } |
643 | |
644 | /** |
645 | * cap_bprm_secureexec - Determine whether a secure execution is required |
646 | * @bprm: The execution parameters |
647 | * |
648 | * Determine whether a secure execution is required, return 1 if it is, and 0 |
649 | * if it is not. |
650 | * |
651 | * The credentials have been committed by this point, and so are no longer |
652 | * available through @bprm->cred. |
653 | */ |
654 | int cap_bprm_secureexec(struct linux_binprm *bprm) |
655 | { |
656 | const struct cred *cred = current_cred(); |
657 | kuid_t root_uid = make_kuid(cred->user_ns, 0); |
658 | |
659 | if (!uid_eq(cred->uid, root_uid)) { |
660 | if (bprm->cap_effective) |
661 | return 1; |
662 | if (!cap_issubset(cred->cap_permitted, cred->cap_ambient)) |
663 | return 1; |
664 | } |
665 | |
666 | return (!uid_eq(cred->euid, cred->uid) || |
667 | !gid_eq(cred->egid, cred->gid)); |
668 | } |
669 | |
670 | /** |
671 | * cap_inode_setxattr - Determine whether an xattr may be altered |
672 | * @dentry: The inode/dentry being altered |
673 | * @name: The name of the xattr to be changed |
674 | * @value: The value that the xattr will be changed to |
675 | * @size: The size of value |
676 | * @flags: The replacement flag |
677 | * |
678 | * Determine whether an xattr may be altered or set on an inode, returning 0 if |
679 | * permission is granted, -ve if denied. |
680 | * |
681 | * This is used to make sure security xattrs don't get updated or set by those |
682 | * who aren't privileged to do so. |
683 | */ |
684 | int cap_inode_setxattr(struct dentry *dentry, const char *name, |
685 | const void *value, size_t size, int flags) |
686 | { |
687 | if (!strcmp(name, XATTR_NAME_CAPS)) { |
688 | if (!capable(CAP_SETFCAP)) |
689 | return -EPERM; |
690 | return 0; |
691 | } |
692 | |
693 | if (!strncmp(name, XATTR_SECURITY_PREFIX, |
694 | sizeof(XATTR_SECURITY_PREFIX) - 1) && |
695 | !capable(CAP_SYS_ADMIN)) |
696 | return -EPERM; |
697 | return 0; |
698 | } |
699 | |
700 | /** |
701 | * cap_inode_removexattr - Determine whether an xattr may be removed |
702 | * @dentry: The inode/dentry being altered |
703 | * @name: The name of the xattr to be changed |
704 | * |
705 | * Determine whether an xattr may be removed from an inode, returning 0 if |
706 | * permission is granted, -ve if denied. |
707 | * |
708 | * This is used to make sure security xattrs don't get removed by those who |
709 | * aren't privileged to remove them. |
710 | */ |
711 | int cap_inode_removexattr(struct dentry *dentry, const char *name) |
712 | { |
713 | if (!strcmp(name, XATTR_NAME_CAPS)) { |
714 | if (!capable(CAP_SETFCAP)) |
715 | return -EPERM; |
716 | return 0; |
717 | } |
718 | |
719 | if (!strncmp(name, XATTR_SECURITY_PREFIX, |
720 | sizeof(XATTR_SECURITY_PREFIX) - 1) && |
721 | !capable(CAP_SYS_ADMIN)) |
722 | return -EPERM; |
723 | return 0; |
724 | } |
725 | |
726 | /* |
727 | * cap_emulate_setxuid() fixes the effective / permitted capabilities of |
728 | * a process after a call to setuid, setreuid, or setresuid. |
729 | * |
730 | * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of |
731 | * {r,e,s}uid != 0, the permitted and effective capabilities are |
732 | * cleared. |
733 | * |
734 | * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective |
735 | * capabilities of the process are cleared. |
736 | * |
737 | * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective |
738 | * capabilities are set to the permitted capabilities. |
739 | * |
740 | * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should |
741 | * never happen. |
742 | * |
743 | * -astor |
744 | * |
745 | * cevans - New behaviour, Oct '99 |
746 | * A process may, via prctl(), elect to keep its capabilities when it |
747 | * calls setuid() and switches away from uid==0. Both permitted and |
748 | * effective sets will be retained. |
749 | * Without this change, it was impossible for a daemon to drop only some |
750 | * of its privilege. The call to setuid(!=0) would drop all privileges! |
751 | * Keeping uid 0 is not an option because uid 0 owns too many vital |
752 | * files.. |
753 | * Thanks to Olaf Kirch and Peter Benie for spotting this. |
754 | */ |
755 | static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old) |
756 | { |
757 | kuid_t root_uid = make_kuid(old->user_ns, 0); |
758 | |
759 | if ((uid_eq(old->uid, root_uid) || |
760 | uid_eq(old->euid, root_uid) || |
761 | uid_eq(old->suid, root_uid)) && |
762 | (!uid_eq(new->uid, root_uid) && |
763 | !uid_eq(new->euid, root_uid) && |
764 | !uid_eq(new->suid, root_uid))) { |
765 | if (!issecure(SECURE_KEEP_CAPS)) { |
766 | cap_clear(new->cap_permitted); |
767 | cap_clear(new->cap_effective); |
768 | } |
769 | |
770 | /* |
771 | * Pre-ambient programs expect setresuid to nonroot followed |
772 | * by exec to drop capabilities. We should make sure that |
773 | * this remains the case. |
774 | */ |
775 | cap_clear(new->cap_ambient); |
776 | } |
777 | if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid)) |
778 | cap_clear(new->cap_effective); |
779 | if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid)) |
780 | new->cap_effective = new->cap_permitted; |
781 | } |
782 | |
783 | /** |
784 | * cap_task_fix_setuid - Fix up the results of setuid() call |
785 | * @new: The proposed credentials |
786 | * @old: The current task's current credentials |
787 | * @flags: Indications of what has changed |
788 | * |
789 | * Fix up the results of setuid() call before the credential changes are |
790 | * actually applied, returning 0 to grant the changes, -ve to deny them. |
791 | */ |
792 | int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags) |
793 | { |
794 | switch (flags) { |
795 | case LSM_SETID_RE: |
796 | case LSM_SETID_ID: |
797 | case LSM_SETID_RES: |
798 | /* juggle the capabilities to follow [RES]UID changes unless |
799 | * otherwise suppressed */ |
800 | if (!issecure(SECURE_NO_SETUID_FIXUP)) |
801 | cap_emulate_setxuid(new, old); |
802 | break; |
803 | |
804 | case LSM_SETID_FS: |
805 | /* juggle the capabilties to follow FSUID changes, unless |
806 | * otherwise suppressed |
807 | * |
808 | * FIXME - is fsuser used for all CAP_FS_MASK capabilities? |
809 | * if not, we might be a bit too harsh here. |
810 | */ |
811 | if (!issecure(SECURE_NO_SETUID_FIXUP)) { |
812 | kuid_t root_uid = make_kuid(old->user_ns, 0); |
813 | if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid)) |
814 | new->cap_effective = |
815 | cap_drop_fs_set(new->cap_effective); |
816 | |
817 | if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid)) |
818 | new->cap_effective = |
819 | cap_raise_fs_set(new->cap_effective, |
820 | new->cap_permitted); |
821 | } |
822 | break; |
823 | |
824 | default: |
825 | return -EINVAL; |
826 | } |
827 | |
828 | return 0; |
829 | } |
830 | |
831 | /* |
832 | * Rationale: code calling task_setscheduler, task_setioprio, and |
833 | * task_setnice, assumes that |
834 | * . if capable(cap_sys_nice), then those actions should be allowed |
835 | * . if not capable(cap_sys_nice), but acting on your own processes, |
836 | * then those actions should be allowed |
837 | * This is insufficient now since you can call code without suid, but |
838 | * yet with increased caps. |
839 | * So we check for increased caps on the target process. |
840 | */ |
841 | static int cap_safe_nice(struct task_struct *p) |
842 | { |
843 | int is_subset, ret = 0; |
844 | |
845 | rcu_read_lock(); |
846 | is_subset = cap_issubset(__task_cred(p)->cap_permitted, |
847 | current_cred()->cap_permitted); |
848 | if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) |
849 | ret = -EPERM; |
850 | rcu_read_unlock(); |
851 | |
852 | return ret; |
853 | } |
854 | |
855 | /** |
856 | * cap_task_setscheduler - Detemine if scheduler policy change is permitted |
857 | * @p: The task to affect |
858 | * |
859 | * Detemine if the requested scheduler policy change is permitted for the |
860 | * specified task, returning 0 if permission is granted, -ve if denied. |
861 | */ |
862 | int cap_task_setscheduler(struct task_struct *p) |
863 | { |
864 | return cap_safe_nice(p); |
865 | } |
866 | |
867 | /** |
868 | * cap_task_ioprio - Detemine if I/O priority change is permitted |
869 | * @p: The task to affect |
870 | * @ioprio: The I/O priority to set |
871 | * |
872 | * Detemine if the requested I/O priority change is permitted for the specified |
873 | * task, returning 0 if permission is granted, -ve if denied. |
874 | */ |
875 | int cap_task_setioprio(struct task_struct *p, int ioprio) |
876 | { |
877 | return cap_safe_nice(p); |
878 | } |
879 | |
880 | /** |
881 | * cap_task_ioprio - Detemine if task priority change is permitted |
882 | * @p: The task to affect |
883 | * @nice: The nice value to set |
884 | * |
885 | * Detemine if the requested task priority change is permitted for the |
886 | * specified task, returning 0 if permission is granted, -ve if denied. |
887 | */ |
888 | int cap_task_setnice(struct task_struct *p, int nice) |
889 | { |
890 | return cap_safe_nice(p); |
891 | } |
892 | |
893 | /* |
894 | * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from |
895 | * the current task's bounding set. Returns 0 on success, -ve on error. |
896 | */ |
897 | static int cap_prctl_drop(unsigned long cap) |
898 | { |
899 | struct cred *new; |
900 | |
901 | if (!ns_capable(current_user_ns(), CAP_SETPCAP)) |
902 | return -EPERM; |
903 | if (!cap_valid(cap)) |
904 | return -EINVAL; |
905 | |
906 | new = prepare_creds(); |
907 | if (!new) |
908 | return -ENOMEM; |
909 | cap_lower(new->cap_bset, cap); |
910 | return commit_creds(new); |
911 | } |
912 | |
913 | /** |
914 | * cap_task_prctl - Implement process control functions for this security module |
915 | * @option: The process control function requested |
916 | * @arg2, @arg3, @arg4, @arg5: The argument data for this function |
917 | * |
918 | * Allow process control functions (sys_prctl()) to alter capabilities; may |
919 | * also deny access to other functions not otherwise implemented here. |
920 | * |
921 | * Returns 0 or +ve on success, -ENOSYS if this function is not implemented |
922 | * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM |
923 | * modules will consider performing the function. |
924 | */ |
925 | int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3, |
926 | unsigned long arg4, unsigned long arg5) |
927 | { |
928 | const struct cred *old = current_cred(); |
929 | struct cred *new; |
930 | |
931 | switch (option) { |
932 | case PR_CAPBSET_READ: |
933 | if (!cap_valid(arg2)) |
934 | return -EINVAL; |
935 | return !!cap_raised(old->cap_bset, arg2); |
936 | |
937 | case PR_CAPBSET_DROP: |
938 | return cap_prctl_drop(arg2); |
939 | |
940 | /* |
941 | * The next four prctl's remain to assist with transitioning a |
942 | * system from legacy UID=0 based privilege (when filesystem |
943 | * capabilities are not in use) to a system using filesystem |
944 | * capabilities only - as the POSIX.1e draft intended. |
945 | * |
946 | * Note: |
947 | * |
948 | * PR_SET_SECUREBITS = |
949 | * issecure_mask(SECURE_KEEP_CAPS_LOCKED) |
950 | * | issecure_mask(SECURE_NOROOT) |
951 | * | issecure_mask(SECURE_NOROOT_LOCKED) |
952 | * | issecure_mask(SECURE_NO_SETUID_FIXUP) |
953 | * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED) |
954 | * |
955 | * will ensure that the current process and all of its |
956 | * children will be locked into a pure |
957 | * capability-based-privilege environment. |
958 | */ |
959 | case PR_SET_SECUREBITS: |
960 | if ((((old->securebits & SECURE_ALL_LOCKS) >> 1) |
961 | & (old->securebits ^ arg2)) /*[1]*/ |
962 | || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/ |
963 | || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/ |
964 | || (cap_capable(current_cred(), |
965 | current_cred()->user_ns, CAP_SETPCAP, |
966 | SECURITY_CAP_AUDIT) != 0) /*[4]*/ |
967 | /* |
968 | * [1] no changing of bits that are locked |
969 | * [2] no unlocking of locks |
970 | * [3] no setting of unsupported bits |
971 | * [4] doing anything requires privilege (go read about |
972 | * the "sendmail capabilities bug") |
973 | */ |
974 | ) |
975 | /* cannot change a locked bit */ |
976 | return -EPERM; |
977 | |
978 | new = prepare_creds(); |
979 | if (!new) |
980 | return -ENOMEM; |
981 | new->securebits = arg2; |
982 | return commit_creds(new); |
983 | |
984 | case PR_GET_SECUREBITS: |
985 | return old->securebits; |
986 | |
987 | case PR_GET_KEEPCAPS: |
988 | return !!issecure(SECURE_KEEP_CAPS); |
989 | |
990 | case PR_SET_KEEPCAPS: |
991 | if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */ |
992 | return -EINVAL; |
993 | if (issecure(SECURE_KEEP_CAPS_LOCKED)) |
994 | return -EPERM; |
995 | |
996 | new = prepare_creds(); |
997 | if (!new) |
998 | return -ENOMEM; |
999 | if (arg2) |
1000 | new->securebits |= issecure_mask(SECURE_KEEP_CAPS); |
1001 | else |
1002 | new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); |
1003 | return commit_creds(new); |
1004 | |
1005 | case PR_CAP_AMBIENT: |
1006 | if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) { |
1007 | if (arg3 | arg4 | arg5) |
1008 | return -EINVAL; |
1009 | |
1010 | new = prepare_creds(); |
1011 | if (!new) |
1012 | return -ENOMEM; |
1013 | cap_clear(new->cap_ambient); |
1014 | return commit_creds(new); |
1015 | } |
1016 | |
1017 | if (((!cap_valid(arg3)) | arg4 | arg5)) |
1018 | return -EINVAL; |
1019 | |
1020 | if (arg2 == PR_CAP_AMBIENT_IS_SET) { |
1021 | return !!cap_raised(current_cred()->cap_ambient, arg3); |
1022 | } else if (arg2 != PR_CAP_AMBIENT_RAISE && |
1023 | arg2 != PR_CAP_AMBIENT_LOWER) { |
1024 | return -EINVAL; |
1025 | } else { |
1026 | if (arg2 == PR_CAP_AMBIENT_RAISE && |
1027 | (!cap_raised(current_cred()->cap_permitted, arg3) || |
1028 | !cap_raised(current_cred()->cap_inheritable, |
1029 | arg3) || |
1030 | issecure(SECURE_NO_CAP_AMBIENT_RAISE))) |
1031 | return -EPERM; |
1032 | |
1033 | new = prepare_creds(); |
1034 | if (!new) |
1035 | return -ENOMEM; |
1036 | if (arg2 == PR_CAP_AMBIENT_RAISE) |
1037 | cap_raise(new->cap_ambient, arg3); |
1038 | else |
1039 | cap_lower(new->cap_ambient, arg3); |
1040 | return commit_creds(new); |
1041 | } |
1042 | |
1043 | default: |
1044 | /* No functionality available - continue with default */ |
1045 | return -ENOSYS; |
1046 | } |
1047 | } |
1048 | |
1049 | /** |
1050 | * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted |
1051 | * @mm: The VM space in which the new mapping is to be made |
1052 | * @pages: The size of the mapping |
1053 | * |
1054 | * Determine whether the allocation of a new virtual mapping by the current |
1055 | * task is permitted, returning 1 if permission is granted, 0 if not. |
1056 | */ |
1057 | int cap_vm_enough_memory(struct mm_struct *mm, long pages) |
1058 | { |
1059 | int cap_sys_admin = 0; |
1060 | |
1061 | if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN, |
1062 | SECURITY_CAP_NOAUDIT) == 0) |
1063 | cap_sys_admin = 1; |
1064 | return cap_sys_admin; |
1065 | } |
1066 | |
1067 | /* |
1068 | * cap_mmap_addr - check if able to map given addr |
1069 | * @addr: address attempting to be mapped |
1070 | * |
1071 | * If the process is attempting to map memory below dac_mmap_min_addr they need |
1072 | * CAP_SYS_RAWIO. The other parameters to this function are unused by the |
1073 | * capability security module. Returns 0 if this mapping should be allowed |
1074 | * -EPERM if not. |
1075 | */ |
1076 | int cap_mmap_addr(unsigned long addr) |
1077 | { |
1078 | int ret = 0; |
1079 | |
1080 | if (addr < dac_mmap_min_addr) { |
1081 | ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO, |
1082 | SECURITY_CAP_AUDIT); |
1083 | /* set PF_SUPERPRIV if it turns out we allow the low mmap */ |
1084 | if (ret == 0) |
1085 | current->flags |= PF_SUPERPRIV; |
1086 | } |
1087 | return ret; |
1088 | } |
1089 | |
1090 | int cap_mmap_file(struct file *file, unsigned long reqprot, |
1091 | unsigned long prot, unsigned long flags) |
1092 | { |
1093 | return 0; |
1094 | } |
1095 | |
1096 | #ifdef CONFIG_SECURITY |
1097 | |
1098 | struct security_hook_list capability_hooks[] = { |
1099 | LSM_HOOK_INIT(capable, cap_capable), |
1100 | LSM_HOOK_INIT(settime, cap_settime), |
1101 | LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check), |
1102 | LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme), |
1103 | LSM_HOOK_INIT(capget, cap_capget), |
1104 | LSM_HOOK_INIT(capset, cap_capset), |
1105 | LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds), |
1106 | LSM_HOOK_INIT(bprm_secureexec, cap_bprm_secureexec), |
1107 | LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv), |
1108 | LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv), |
1109 | LSM_HOOK_INIT(mmap_addr, cap_mmap_addr), |
1110 | LSM_HOOK_INIT(mmap_file, cap_mmap_file), |
1111 | LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid), |
1112 | LSM_HOOK_INIT(task_prctl, cap_task_prctl), |
1113 | LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler), |
1114 | LSM_HOOK_INIT(task_setioprio, cap_task_setioprio), |
1115 | LSM_HOOK_INIT(task_setnice, cap_task_setnice), |
1116 | LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory), |
1117 | }; |
1118 | |
1119 | void __init capability_add_hooks(void) |
1120 | { |
1121 | security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks)); |
1122 | } |
1123 | |
1124 | #endif /* CONFIG_SECURITY */ |
1125 |