xref: /openbmc/linux/security/commoncap.c (revision 2572f00d)
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 /*
35  * If a non-root user executes a setuid-root binary in
36  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
37  * However if fE is also set, then the intent is for only
38  * the file capabilities to be applied, and the setuid-root
39  * bit is left on either to change the uid (plausible) or
40  * to get full privilege on a kernel without file capabilities
41  * support.  So in that case we do not raise capabilities.
42  *
43  * Warn if that happens, once per boot.
44  */
45 static void warn_setuid_and_fcaps_mixed(const char *fname)
46 {
47 	static int warned;
48 	if (!warned) {
49 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
50 			" effective capabilities. Therefore not raising all"
51 			" capabilities.\n", fname);
52 		warned = 1;
53 	}
54 }
55 
56 /**
57  * cap_capable - Determine whether a task has a particular effective capability
58  * @cred: The credentials to use
59  * @ns:  The user namespace in which we need the capability
60  * @cap: The capability to check for
61  * @audit: Whether to write an audit message or not
62  *
63  * Determine whether the nominated task has the specified capability amongst
64  * its effective set, returning 0 if it does, -ve if it does not.
65  *
66  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
67  * and has_capability() functions.  That is, it has the reverse semantics:
68  * cap_has_capability() returns 0 when a task has a capability, but the
69  * kernel's capable() and has_capability() returns 1 for this case.
70  */
71 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
72 		int cap, int audit)
73 {
74 	struct user_namespace *ns = targ_ns;
75 
76 	/* See if cred has the capability in the target user namespace
77 	 * by examining the target user namespace and all of the target
78 	 * user namespace's parents.
79 	 */
80 	for (;;) {
81 		/* Do we have the necessary capabilities? */
82 		if (ns == cred->user_ns)
83 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
84 
85 		/* Have we tried all of the parent namespaces? */
86 		if (ns == &init_user_ns)
87 			return -EPERM;
88 
89 		/*
90 		 * The owner of the user namespace in the parent of the
91 		 * user namespace has all caps.
92 		 */
93 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
94 			return 0;
95 
96 		/*
97 		 * If you have a capability in a parent user ns, then you have
98 		 * it over all children user namespaces as well.
99 		 */
100 		ns = ns->parent;
101 	}
102 
103 	/* We never get here */
104 }
105 
106 /**
107  * cap_settime - Determine whether the current process may set the system clock
108  * @ts: The time to set
109  * @tz: The timezone to set
110  *
111  * Determine whether the current process may set the system clock and timezone
112  * information, returning 0 if permission granted, -ve if denied.
113  */
114 int cap_settime(const struct timespec *ts, const struct timezone *tz)
115 {
116 	if (!capable(CAP_SYS_TIME))
117 		return -EPERM;
118 	return 0;
119 }
120 
121 /**
122  * cap_ptrace_access_check - Determine whether the current process may access
123  *			   another
124  * @child: The process to be accessed
125  * @mode: The mode of attachment.
126  *
127  * If we are in the same or an ancestor user_ns and have all the target
128  * task's capabilities, then ptrace access is allowed.
129  * If we have the ptrace capability to the target user_ns, then ptrace
130  * access is allowed.
131  * Else denied.
132  *
133  * Determine whether a process may access another, returning 0 if permission
134  * granted, -ve if denied.
135  */
136 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
137 {
138 	int ret = 0;
139 	const struct cred *cred, *child_cred;
140 
141 	rcu_read_lock();
142 	cred = current_cred();
143 	child_cred = __task_cred(child);
144 	if (cred->user_ns == child_cred->user_ns &&
145 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
146 		goto out;
147 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
148 		goto out;
149 	ret = -EPERM;
150 out:
151 	rcu_read_unlock();
152 	return ret;
153 }
154 
155 /**
156  * cap_ptrace_traceme - Determine whether another process may trace the current
157  * @parent: The task proposed to be the tracer
158  *
159  * If parent is in the same or an ancestor user_ns and has all current's
160  * capabilities, then ptrace access is allowed.
161  * If parent has the ptrace capability to current's user_ns, then ptrace
162  * access is allowed.
163  * Else denied.
164  *
165  * Determine whether the nominated task is permitted to trace the current
166  * process, returning 0 if permission is granted, -ve if denied.
167  */
168 int cap_ptrace_traceme(struct task_struct *parent)
169 {
170 	int ret = 0;
171 	const struct cred *cred, *child_cred;
172 
173 	rcu_read_lock();
174 	cred = __task_cred(parent);
175 	child_cred = current_cred();
176 	if (cred->user_ns == child_cred->user_ns &&
177 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
178 		goto out;
179 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
180 		goto out;
181 	ret = -EPERM;
182 out:
183 	rcu_read_unlock();
184 	return ret;
185 }
186 
187 /**
188  * cap_capget - Retrieve a task's capability sets
189  * @target: The task from which to retrieve the capability sets
190  * @effective: The place to record the effective set
191  * @inheritable: The place to record the inheritable set
192  * @permitted: The place to record the permitted set
193  *
194  * This function retrieves the capabilities of the nominated task and returns
195  * them to the caller.
196  */
197 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
198 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
199 {
200 	const struct cred *cred;
201 
202 	/* Derived from kernel/capability.c:sys_capget. */
203 	rcu_read_lock();
204 	cred = __task_cred(target);
205 	*effective   = cred->cap_effective;
206 	*inheritable = cred->cap_inheritable;
207 	*permitted   = cred->cap_permitted;
208 	rcu_read_unlock();
209 	return 0;
210 }
211 
212 /*
213  * Determine whether the inheritable capabilities are limited to the old
214  * permitted set.  Returns 1 if they are limited, 0 if they are not.
215  */
216 static inline int cap_inh_is_capped(void)
217 {
218 
219 	/* they are so limited unless the current task has the CAP_SETPCAP
220 	 * capability
221 	 */
222 	if (cap_capable(current_cred(), current_cred()->user_ns,
223 			CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
224 		return 0;
225 	return 1;
226 }
227 
228 /**
229  * cap_capset - Validate and apply proposed changes to current's capabilities
230  * @new: The proposed new credentials; alterations should be made here
231  * @old: The current task's current credentials
232  * @effective: A pointer to the proposed new effective capabilities set
233  * @inheritable: A pointer to the proposed new inheritable capabilities set
234  * @permitted: A pointer to the proposed new permitted capabilities set
235  *
236  * This function validates and applies a proposed mass change to the current
237  * process's capability sets.  The changes are made to the proposed new
238  * credentials, and assuming no error, will be committed by the caller of LSM.
239  */
240 int cap_capset(struct cred *new,
241 	       const struct cred *old,
242 	       const kernel_cap_t *effective,
243 	       const kernel_cap_t *inheritable,
244 	       const kernel_cap_t *permitted)
245 {
246 	if (cap_inh_is_capped() &&
247 	    !cap_issubset(*inheritable,
248 			  cap_combine(old->cap_inheritable,
249 				      old->cap_permitted)))
250 		/* incapable of using this inheritable set */
251 		return -EPERM;
252 
253 	if (!cap_issubset(*inheritable,
254 			  cap_combine(old->cap_inheritable,
255 				      old->cap_bset)))
256 		/* no new pI capabilities outside bounding set */
257 		return -EPERM;
258 
259 	/* verify restrictions on target's new Permitted set */
260 	if (!cap_issubset(*permitted, old->cap_permitted))
261 		return -EPERM;
262 
263 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
264 	if (!cap_issubset(*effective, *permitted))
265 		return -EPERM;
266 
267 	new->cap_effective   = *effective;
268 	new->cap_inheritable = *inheritable;
269 	new->cap_permitted   = *permitted;
270 
271 	/*
272 	 * Mask off ambient bits that are no longer both permitted and
273 	 * inheritable.
274 	 */
275 	new->cap_ambient = cap_intersect(new->cap_ambient,
276 					 cap_intersect(*permitted,
277 						       *inheritable));
278 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
279 		return -EINVAL;
280 	return 0;
281 }
282 
283 /*
284  * Clear proposed capability sets for execve().
285  */
286 static inline void bprm_clear_caps(struct linux_binprm *bprm)
287 {
288 	cap_clear(bprm->cred->cap_permitted);
289 	bprm->cap_effective = false;
290 }
291 
292 /**
293  * cap_inode_need_killpriv - Determine if inode change affects privileges
294  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
295  *
296  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
297  * affects the security markings on that inode, and if it is, should
298  * inode_killpriv() be invoked or the change rejected?
299  *
300  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
301  * -ve to deny the change.
302  */
303 int cap_inode_need_killpriv(struct dentry *dentry)
304 {
305 	struct inode *inode = d_backing_inode(dentry);
306 	int error;
307 
308 	if (!inode->i_op->getxattr)
309 	       return 0;
310 
311 	error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
312 	if (error <= 0)
313 		return 0;
314 	return 1;
315 }
316 
317 /**
318  * cap_inode_killpriv - Erase the security markings on an inode
319  * @dentry: The inode/dentry to alter
320  *
321  * Erase the privilege-enhancing security markings on an inode.
322  *
323  * Returns 0 if successful, -ve on error.
324  */
325 int cap_inode_killpriv(struct dentry *dentry)
326 {
327 	struct inode *inode = d_backing_inode(dentry);
328 
329 	if (!inode->i_op->removexattr)
330 	       return 0;
331 
332 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
333 }
334 
335 /*
336  * Calculate the new process capability sets from the capability sets attached
337  * to a file.
338  */
339 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
340 					  struct linux_binprm *bprm,
341 					  bool *effective,
342 					  bool *has_cap)
343 {
344 	struct cred *new = bprm->cred;
345 	unsigned i;
346 	int ret = 0;
347 
348 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
349 		*effective = true;
350 
351 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
352 		*has_cap = true;
353 
354 	CAP_FOR_EACH_U32(i) {
355 		__u32 permitted = caps->permitted.cap[i];
356 		__u32 inheritable = caps->inheritable.cap[i];
357 
358 		/*
359 		 * pP' = (X & fP) | (pI & fI)
360 		 * The addition of pA' is handled later.
361 		 */
362 		new->cap_permitted.cap[i] =
363 			(new->cap_bset.cap[i] & permitted) |
364 			(new->cap_inheritable.cap[i] & inheritable);
365 
366 		if (permitted & ~new->cap_permitted.cap[i])
367 			/* insufficient to execute correctly */
368 			ret = -EPERM;
369 	}
370 
371 	/*
372 	 * For legacy apps, with no internal support for recognizing they
373 	 * do not have enough capabilities, we return an error if they are
374 	 * missing some "forced" (aka file-permitted) capabilities.
375 	 */
376 	return *effective ? ret : 0;
377 }
378 
379 /*
380  * Extract the on-exec-apply capability sets for an executable file.
381  */
382 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
383 {
384 	struct inode *inode = d_backing_inode(dentry);
385 	__u32 magic_etc;
386 	unsigned tocopy, i;
387 	int size;
388 	struct vfs_cap_data caps;
389 
390 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
391 
392 	if (!inode || !inode->i_op->getxattr)
393 		return -ENODATA;
394 
395 	size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
396 				   XATTR_CAPS_SZ);
397 	if (size == -ENODATA || size == -EOPNOTSUPP)
398 		/* no data, that's ok */
399 		return -ENODATA;
400 	if (size < 0)
401 		return size;
402 
403 	if (size < sizeof(magic_etc))
404 		return -EINVAL;
405 
406 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
407 
408 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
409 	case VFS_CAP_REVISION_1:
410 		if (size != XATTR_CAPS_SZ_1)
411 			return -EINVAL;
412 		tocopy = VFS_CAP_U32_1;
413 		break;
414 	case VFS_CAP_REVISION_2:
415 		if (size != XATTR_CAPS_SZ_2)
416 			return -EINVAL;
417 		tocopy = VFS_CAP_U32_2;
418 		break;
419 	default:
420 		return -EINVAL;
421 	}
422 
423 	CAP_FOR_EACH_U32(i) {
424 		if (i >= tocopy)
425 			break;
426 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
427 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
428 	}
429 
430 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
431 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
432 
433 	return 0;
434 }
435 
436 /*
437  * Attempt to get the on-exec apply capability sets for an executable file from
438  * its xattrs and, if present, apply them to the proposed credentials being
439  * constructed by execve().
440  */
441 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap)
442 {
443 	int rc = 0;
444 	struct cpu_vfs_cap_data vcaps;
445 
446 	bprm_clear_caps(bprm);
447 
448 	if (!file_caps_enabled)
449 		return 0;
450 
451 	if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
452 		return 0;
453 
454 	rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
455 	if (rc < 0) {
456 		if (rc == -EINVAL)
457 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
458 				__func__, rc, bprm->filename);
459 		else if (rc == -ENODATA)
460 			rc = 0;
461 		goto out;
462 	}
463 
464 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
465 	if (rc == -EINVAL)
466 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
467 		       __func__, rc, bprm->filename);
468 
469 out:
470 	if (rc)
471 		bprm_clear_caps(bprm);
472 
473 	return rc;
474 }
475 
476 /**
477  * cap_bprm_set_creds - Set up the proposed credentials for execve().
478  * @bprm: The execution parameters, including the proposed creds
479  *
480  * Set up the proposed credentials for a new execution context being
481  * constructed by execve().  The proposed creds in @bprm->cred is altered,
482  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
483  */
484 int cap_bprm_set_creds(struct linux_binprm *bprm)
485 {
486 	const struct cred *old = current_cred();
487 	struct cred *new = bprm->cred;
488 	bool effective, has_cap = false, is_setid;
489 	int ret;
490 	kuid_t root_uid;
491 
492 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
493 		return -EPERM;
494 
495 	effective = false;
496 	ret = get_file_caps(bprm, &effective, &has_cap);
497 	if (ret < 0)
498 		return ret;
499 
500 	root_uid = make_kuid(new->user_ns, 0);
501 
502 	if (!issecure(SECURE_NOROOT)) {
503 		/*
504 		 * If the legacy file capability is set, then don't set privs
505 		 * for a setuid root binary run by a non-root user.  Do set it
506 		 * for a root user just to cause least surprise to an admin.
507 		 */
508 		if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) {
509 			warn_setuid_and_fcaps_mixed(bprm->filename);
510 			goto skip;
511 		}
512 		/*
513 		 * To support inheritance of root-permissions and suid-root
514 		 * executables under compatibility mode, we override the
515 		 * capability sets for the file.
516 		 *
517 		 * If only the real uid is 0, we do not set the effective bit.
518 		 */
519 		if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) {
520 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
521 			new->cap_permitted = cap_combine(old->cap_bset,
522 							 old->cap_inheritable);
523 		}
524 		if (uid_eq(new->euid, root_uid))
525 			effective = true;
526 	}
527 skip:
528 
529 	/* if we have fs caps, clear dangerous personality flags */
530 	if (!cap_issubset(new->cap_permitted, old->cap_permitted))
531 		bprm->per_clear |= PER_CLEAR_ON_SETID;
532 
533 
534 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
535 	 * credentials unless they have the appropriate permit.
536 	 *
537 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
538 	 */
539 	is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid);
540 
541 	if ((is_setid ||
542 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
543 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
544 		/* downgrade; they get no more than they had, and maybe less */
545 		if (!capable(CAP_SETUID) ||
546 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
547 			new->euid = new->uid;
548 			new->egid = new->gid;
549 		}
550 		new->cap_permitted = cap_intersect(new->cap_permitted,
551 						   old->cap_permitted);
552 	}
553 
554 	new->suid = new->fsuid = new->euid;
555 	new->sgid = new->fsgid = new->egid;
556 
557 	/* File caps or setid cancels ambient. */
558 	if (has_cap || is_setid)
559 		cap_clear(new->cap_ambient);
560 
561 	/*
562 	 * Now that we've computed pA', update pP' to give:
563 	 *   pP' = (X & fP) | (pI & fI) | pA'
564 	 */
565 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
566 
567 	/*
568 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
569 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
570 	 */
571 	if (effective)
572 		new->cap_effective = new->cap_permitted;
573 	else
574 		new->cap_effective = new->cap_ambient;
575 
576 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
577 		return -EPERM;
578 
579 	bprm->cap_effective = effective;
580 
581 	/*
582 	 * Audit candidate if current->cap_effective is set
583 	 *
584 	 * We do not bother to audit if 3 things are true:
585 	 *   1) cap_effective has all caps
586 	 *   2) we are root
587 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
588 	 * Since this is just a normal root execing a process.
589 	 *
590 	 * Number 1 above might fail if you don't have a full bset, but I think
591 	 * that is interesting information to audit.
592 	 */
593 	if (!cap_issubset(new->cap_effective, new->cap_ambient)) {
594 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
595 		    !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) ||
596 		    issecure(SECURE_NOROOT)) {
597 			ret = audit_log_bprm_fcaps(bprm, new, old);
598 			if (ret < 0)
599 				return ret;
600 		}
601 	}
602 
603 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
604 
605 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
606 		return -EPERM;
607 
608 	return 0;
609 }
610 
611 /**
612  * cap_bprm_secureexec - Determine whether a secure execution is required
613  * @bprm: The execution parameters
614  *
615  * Determine whether a secure execution is required, return 1 if it is, and 0
616  * if it is not.
617  *
618  * The credentials have been committed by this point, and so are no longer
619  * available through @bprm->cred.
620  */
621 int cap_bprm_secureexec(struct linux_binprm *bprm)
622 {
623 	const struct cred *cred = current_cred();
624 	kuid_t root_uid = make_kuid(cred->user_ns, 0);
625 
626 	if (!uid_eq(cred->uid, root_uid)) {
627 		if (bprm->cap_effective)
628 			return 1;
629 		if (!cap_issubset(cred->cap_permitted, cred->cap_ambient))
630 			return 1;
631 	}
632 
633 	return (!uid_eq(cred->euid, cred->uid) ||
634 		!gid_eq(cred->egid, cred->gid));
635 }
636 
637 /**
638  * cap_inode_setxattr - Determine whether an xattr may be altered
639  * @dentry: The inode/dentry being altered
640  * @name: The name of the xattr to be changed
641  * @value: The value that the xattr will be changed to
642  * @size: The size of value
643  * @flags: The replacement flag
644  *
645  * Determine whether an xattr may be altered or set on an inode, returning 0 if
646  * permission is granted, -ve if denied.
647  *
648  * This is used to make sure security xattrs don't get updated or set by those
649  * who aren't privileged to do so.
650  */
651 int cap_inode_setxattr(struct dentry *dentry, const char *name,
652 		       const void *value, size_t size, int flags)
653 {
654 	if (!strcmp(name, XATTR_NAME_CAPS)) {
655 		if (!capable(CAP_SETFCAP))
656 			return -EPERM;
657 		return 0;
658 	}
659 
660 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
661 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
662 	    !capable(CAP_SYS_ADMIN))
663 		return -EPERM;
664 	return 0;
665 }
666 
667 /**
668  * cap_inode_removexattr - Determine whether an xattr may be removed
669  * @dentry: The inode/dentry being altered
670  * @name: The name of the xattr to be changed
671  *
672  * Determine whether an xattr may be removed from an inode, returning 0 if
673  * permission is granted, -ve if denied.
674  *
675  * This is used to make sure security xattrs don't get removed by those who
676  * aren't privileged to remove them.
677  */
678 int cap_inode_removexattr(struct dentry *dentry, const char *name)
679 {
680 	if (!strcmp(name, XATTR_NAME_CAPS)) {
681 		if (!capable(CAP_SETFCAP))
682 			return -EPERM;
683 		return 0;
684 	}
685 
686 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
687 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
688 	    !capable(CAP_SYS_ADMIN))
689 		return -EPERM;
690 	return 0;
691 }
692 
693 /*
694  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
695  * a process after a call to setuid, setreuid, or setresuid.
696  *
697  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
698  *  {r,e,s}uid != 0, the permitted and effective capabilities are
699  *  cleared.
700  *
701  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
702  *  capabilities of the process are cleared.
703  *
704  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
705  *  capabilities are set to the permitted capabilities.
706  *
707  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
708  *  never happen.
709  *
710  *  -astor
711  *
712  * cevans - New behaviour, Oct '99
713  * A process may, via prctl(), elect to keep its capabilities when it
714  * calls setuid() and switches away from uid==0. Both permitted and
715  * effective sets will be retained.
716  * Without this change, it was impossible for a daemon to drop only some
717  * of its privilege. The call to setuid(!=0) would drop all privileges!
718  * Keeping uid 0 is not an option because uid 0 owns too many vital
719  * files..
720  * Thanks to Olaf Kirch and Peter Benie for spotting this.
721  */
722 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
723 {
724 	kuid_t root_uid = make_kuid(old->user_ns, 0);
725 
726 	if ((uid_eq(old->uid, root_uid) ||
727 	     uid_eq(old->euid, root_uid) ||
728 	     uid_eq(old->suid, root_uid)) &&
729 	    (!uid_eq(new->uid, root_uid) &&
730 	     !uid_eq(new->euid, root_uid) &&
731 	     !uid_eq(new->suid, root_uid))) {
732 		if (!issecure(SECURE_KEEP_CAPS)) {
733 			cap_clear(new->cap_permitted);
734 			cap_clear(new->cap_effective);
735 		}
736 
737 		/*
738 		 * Pre-ambient programs expect setresuid to nonroot followed
739 		 * by exec to drop capabilities.  We should make sure that
740 		 * this remains the case.
741 		 */
742 		cap_clear(new->cap_ambient);
743 	}
744 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
745 		cap_clear(new->cap_effective);
746 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
747 		new->cap_effective = new->cap_permitted;
748 }
749 
750 /**
751  * cap_task_fix_setuid - Fix up the results of setuid() call
752  * @new: The proposed credentials
753  * @old: The current task's current credentials
754  * @flags: Indications of what has changed
755  *
756  * Fix up the results of setuid() call before the credential changes are
757  * actually applied, returning 0 to grant the changes, -ve to deny them.
758  */
759 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
760 {
761 	switch (flags) {
762 	case LSM_SETID_RE:
763 	case LSM_SETID_ID:
764 	case LSM_SETID_RES:
765 		/* juggle the capabilities to follow [RES]UID changes unless
766 		 * otherwise suppressed */
767 		if (!issecure(SECURE_NO_SETUID_FIXUP))
768 			cap_emulate_setxuid(new, old);
769 		break;
770 
771 	case LSM_SETID_FS:
772 		/* juggle the capabilties to follow FSUID changes, unless
773 		 * otherwise suppressed
774 		 *
775 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
776 		 *          if not, we might be a bit too harsh here.
777 		 */
778 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
779 			kuid_t root_uid = make_kuid(old->user_ns, 0);
780 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
781 				new->cap_effective =
782 					cap_drop_fs_set(new->cap_effective);
783 
784 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
785 				new->cap_effective =
786 					cap_raise_fs_set(new->cap_effective,
787 							 new->cap_permitted);
788 		}
789 		break;
790 
791 	default:
792 		return -EINVAL;
793 	}
794 
795 	return 0;
796 }
797 
798 /*
799  * Rationale: code calling task_setscheduler, task_setioprio, and
800  * task_setnice, assumes that
801  *   . if capable(cap_sys_nice), then those actions should be allowed
802  *   . if not capable(cap_sys_nice), but acting on your own processes,
803  *   	then those actions should be allowed
804  * This is insufficient now since you can call code without suid, but
805  * yet with increased caps.
806  * So we check for increased caps on the target process.
807  */
808 static int cap_safe_nice(struct task_struct *p)
809 {
810 	int is_subset, ret = 0;
811 
812 	rcu_read_lock();
813 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
814 				 current_cred()->cap_permitted);
815 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
816 		ret = -EPERM;
817 	rcu_read_unlock();
818 
819 	return ret;
820 }
821 
822 /**
823  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
824  * @p: The task to affect
825  *
826  * Detemine if the requested scheduler policy change is permitted for the
827  * specified task, returning 0 if permission is granted, -ve if denied.
828  */
829 int cap_task_setscheduler(struct task_struct *p)
830 {
831 	return cap_safe_nice(p);
832 }
833 
834 /**
835  * cap_task_ioprio - Detemine if I/O priority change is permitted
836  * @p: The task to affect
837  * @ioprio: The I/O priority to set
838  *
839  * Detemine if the requested I/O priority change is permitted for the specified
840  * task, returning 0 if permission is granted, -ve if denied.
841  */
842 int cap_task_setioprio(struct task_struct *p, int ioprio)
843 {
844 	return cap_safe_nice(p);
845 }
846 
847 /**
848  * cap_task_ioprio - Detemine if task priority change is permitted
849  * @p: The task to affect
850  * @nice: The nice value to set
851  *
852  * Detemine if the requested task priority change is permitted for the
853  * specified task, returning 0 if permission is granted, -ve if denied.
854  */
855 int cap_task_setnice(struct task_struct *p, int nice)
856 {
857 	return cap_safe_nice(p);
858 }
859 
860 /*
861  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
862  * the current task's bounding set.  Returns 0 on success, -ve on error.
863  */
864 static int cap_prctl_drop(unsigned long cap)
865 {
866 	struct cred *new;
867 
868 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
869 		return -EPERM;
870 	if (!cap_valid(cap))
871 		return -EINVAL;
872 
873 	new = prepare_creds();
874 	if (!new)
875 		return -ENOMEM;
876 	cap_lower(new->cap_bset, cap);
877 	return commit_creds(new);
878 }
879 
880 /**
881  * cap_task_prctl - Implement process control functions for this security module
882  * @option: The process control function requested
883  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
884  *
885  * Allow process control functions (sys_prctl()) to alter capabilities; may
886  * also deny access to other functions not otherwise implemented here.
887  *
888  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
889  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
890  * modules will consider performing the function.
891  */
892 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
893 		   unsigned long arg4, unsigned long arg5)
894 {
895 	const struct cred *old = current_cred();
896 	struct cred *new;
897 
898 	switch (option) {
899 	case PR_CAPBSET_READ:
900 		if (!cap_valid(arg2))
901 			return -EINVAL;
902 		return !!cap_raised(old->cap_bset, arg2);
903 
904 	case PR_CAPBSET_DROP:
905 		return cap_prctl_drop(arg2);
906 
907 	/*
908 	 * The next four prctl's remain to assist with transitioning a
909 	 * system from legacy UID=0 based privilege (when filesystem
910 	 * capabilities are not in use) to a system using filesystem
911 	 * capabilities only - as the POSIX.1e draft intended.
912 	 *
913 	 * Note:
914 	 *
915 	 *  PR_SET_SECUREBITS =
916 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
917 	 *    | issecure_mask(SECURE_NOROOT)
918 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
919 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
920 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
921 	 *
922 	 * will ensure that the current process and all of its
923 	 * children will be locked into a pure
924 	 * capability-based-privilege environment.
925 	 */
926 	case PR_SET_SECUREBITS:
927 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
928 		     & (old->securebits ^ arg2))			/*[1]*/
929 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
930 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
931 		    || (cap_capable(current_cred(),
932 				    current_cred()->user_ns, CAP_SETPCAP,
933 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
934 			/*
935 			 * [1] no changing of bits that are locked
936 			 * [2] no unlocking of locks
937 			 * [3] no setting of unsupported bits
938 			 * [4] doing anything requires privilege (go read about
939 			 *     the "sendmail capabilities bug")
940 			 */
941 		    )
942 			/* cannot change a locked bit */
943 			return -EPERM;
944 
945 		new = prepare_creds();
946 		if (!new)
947 			return -ENOMEM;
948 		new->securebits = arg2;
949 		return commit_creds(new);
950 
951 	case PR_GET_SECUREBITS:
952 		return old->securebits;
953 
954 	case PR_GET_KEEPCAPS:
955 		return !!issecure(SECURE_KEEP_CAPS);
956 
957 	case PR_SET_KEEPCAPS:
958 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
959 			return -EINVAL;
960 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
961 			return -EPERM;
962 
963 		new = prepare_creds();
964 		if (!new)
965 			return -ENOMEM;
966 		if (arg2)
967 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
968 		else
969 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
970 		return commit_creds(new);
971 
972 	case PR_CAP_AMBIENT:
973 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
974 			if (arg3 | arg4 | arg5)
975 				return -EINVAL;
976 
977 			new = prepare_creds();
978 			if (!new)
979 				return -ENOMEM;
980 			cap_clear(new->cap_ambient);
981 			return commit_creds(new);
982 		}
983 
984 		if (((!cap_valid(arg3)) | arg4 | arg5))
985 			return -EINVAL;
986 
987 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
988 			return !!cap_raised(current_cred()->cap_ambient, arg3);
989 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
990 			   arg2 != PR_CAP_AMBIENT_LOWER) {
991 			return -EINVAL;
992 		} else {
993 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
994 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
995 			     !cap_raised(current_cred()->cap_inheritable,
996 					 arg3) ||
997 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
998 				return -EPERM;
999 
1000 			new = prepare_creds();
1001 			if (!new)
1002 				return -ENOMEM;
1003 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1004 				cap_raise(new->cap_ambient, arg3);
1005 			else
1006 				cap_lower(new->cap_ambient, arg3);
1007 			return commit_creds(new);
1008 		}
1009 
1010 	default:
1011 		/* No functionality available - continue with default */
1012 		return -ENOSYS;
1013 	}
1014 }
1015 
1016 /**
1017  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1018  * @mm: The VM space in which the new mapping is to be made
1019  * @pages: The size of the mapping
1020  *
1021  * Determine whether the allocation of a new virtual mapping by the current
1022  * task is permitted, returning 1 if permission is granted, 0 if not.
1023  */
1024 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1025 {
1026 	int cap_sys_admin = 0;
1027 
1028 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
1029 			SECURITY_CAP_NOAUDIT) == 0)
1030 		cap_sys_admin = 1;
1031 	return cap_sys_admin;
1032 }
1033 
1034 /*
1035  * cap_mmap_addr - check if able to map given addr
1036  * @addr: address attempting to be mapped
1037  *
1038  * If the process is attempting to map memory below dac_mmap_min_addr they need
1039  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1040  * capability security module.  Returns 0 if this mapping should be allowed
1041  * -EPERM if not.
1042  */
1043 int cap_mmap_addr(unsigned long addr)
1044 {
1045 	int ret = 0;
1046 
1047 	if (addr < dac_mmap_min_addr) {
1048 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1049 				  SECURITY_CAP_AUDIT);
1050 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1051 		if (ret == 0)
1052 			current->flags |= PF_SUPERPRIV;
1053 	}
1054 	return ret;
1055 }
1056 
1057 int cap_mmap_file(struct file *file, unsigned long reqprot,
1058 		  unsigned long prot, unsigned long flags)
1059 {
1060 	return 0;
1061 }
1062 
1063 #ifdef CONFIG_SECURITY
1064 
1065 struct security_hook_list capability_hooks[] = {
1066 	LSM_HOOK_INIT(capable, cap_capable),
1067 	LSM_HOOK_INIT(settime, cap_settime),
1068 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1069 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1070 	LSM_HOOK_INIT(capget, cap_capget),
1071 	LSM_HOOK_INIT(capset, cap_capset),
1072 	LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
1073 	LSM_HOOK_INIT(bprm_secureexec, cap_bprm_secureexec),
1074 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1075 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1076 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1077 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1078 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1079 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1080 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1081 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1082 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1083 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1084 };
1085 
1086 void __init capability_add_hooks(void)
1087 {
1088 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks));
1089 }
1090 
1091 #endif /* CONFIG_SECURITY */
1092