xref: /openbmc/linux/security/commoncap.c (revision 80483c3a)
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 timespec64 *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 	const kernel_cap_t *caller_caps;
141 
142 	rcu_read_lock();
143 	cred = current_cred();
144 	child_cred = __task_cred(child);
145 	if (mode & PTRACE_MODE_FSCREDS)
146 		caller_caps = &cred->cap_effective;
147 	else
148 		caller_caps = &cred->cap_permitted;
149 	if (cred->user_ns == child_cred->user_ns &&
150 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
151 		goto out;
152 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
153 		goto out;
154 	ret = -EPERM;
155 out:
156 	rcu_read_unlock();
157 	return ret;
158 }
159 
160 /**
161  * cap_ptrace_traceme - Determine whether another process may trace the current
162  * @parent: The task proposed to be the tracer
163  *
164  * If parent is in the same or an ancestor user_ns and has all current's
165  * capabilities, then ptrace access is allowed.
166  * If parent has the ptrace capability to current's user_ns, then ptrace
167  * access is allowed.
168  * Else denied.
169  *
170  * Determine whether the nominated task is permitted to trace the current
171  * process, returning 0 if permission is granted, -ve if denied.
172  */
173 int cap_ptrace_traceme(struct task_struct *parent)
174 {
175 	int ret = 0;
176 	const struct cred *cred, *child_cred;
177 
178 	rcu_read_lock();
179 	cred = __task_cred(parent);
180 	child_cred = current_cred();
181 	if (cred->user_ns == child_cred->user_ns &&
182 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
183 		goto out;
184 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
185 		goto out;
186 	ret = -EPERM;
187 out:
188 	rcu_read_unlock();
189 	return ret;
190 }
191 
192 /**
193  * cap_capget - Retrieve a task's capability sets
194  * @target: The task from which to retrieve the capability sets
195  * @effective: The place to record the effective set
196  * @inheritable: The place to record the inheritable set
197  * @permitted: The place to record the permitted set
198  *
199  * This function retrieves the capabilities of the nominated task and returns
200  * them to the caller.
201  */
202 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
203 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
204 {
205 	const struct cred *cred;
206 
207 	/* Derived from kernel/capability.c:sys_capget. */
208 	rcu_read_lock();
209 	cred = __task_cred(target);
210 	*effective   = cred->cap_effective;
211 	*inheritable = cred->cap_inheritable;
212 	*permitted   = cred->cap_permitted;
213 	rcu_read_unlock();
214 	return 0;
215 }
216 
217 /*
218  * Determine whether the inheritable capabilities are limited to the old
219  * permitted set.  Returns 1 if they are limited, 0 if they are not.
220  */
221 static inline int cap_inh_is_capped(void)
222 {
223 
224 	/* they are so limited unless the current task has the CAP_SETPCAP
225 	 * capability
226 	 */
227 	if (cap_capable(current_cred(), current_cred()->user_ns,
228 			CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
229 		return 0;
230 	return 1;
231 }
232 
233 /**
234  * cap_capset - Validate and apply proposed changes to current's capabilities
235  * @new: The proposed new credentials; alterations should be made here
236  * @old: The current task's current credentials
237  * @effective: A pointer to the proposed new effective capabilities set
238  * @inheritable: A pointer to the proposed new inheritable capabilities set
239  * @permitted: A pointer to the proposed new permitted capabilities set
240  *
241  * This function validates and applies a proposed mass change to the current
242  * process's capability sets.  The changes are made to the proposed new
243  * credentials, and assuming no error, will be committed by the caller of LSM.
244  */
245 int cap_capset(struct cred *new,
246 	       const struct cred *old,
247 	       const kernel_cap_t *effective,
248 	       const kernel_cap_t *inheritable,
249 	       const kernel_cap_t *permitted)
250 {
251 	if (cap_inh_is_capped() &&
252 	    !cap_issubset(*inheritable,
253 			  cap_combine(old->cap_inheritable,
254 				      old->cap_permitted)))
255 		/* incapable of using this inheritable set */
256 		return -EPERM;
257 
258 	if (!cap_issubset(*inheritable,
259 			  cap_combine(old->cap_inheritable,
260 				      old->cap_bset)))
261 		/* no new pI capabilities outside bounding set */
262 		return -EPERM;
263 
264 	/* verify restrictions on target's new Permitted set */
265 	if (!cap_issubset(*permitted, old->cap_permitted))
266 		return -EPERM;
267 
268 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
269 	if (!cap_issubset(*effective, *permitted))
270 		return -EPERM;
271 
272 	new->cap_effective   = *effective;
273 	new->cap_inheritable = *inheritable;
274 	new->cap_permitted   = *permitted;
275 
276 	/*
277 	 * Mask off ambient bits that are no longer both permitted and
278 	 * inheritable.
279 	 */
280 	new->cap_ambient = cap_intersect(new->cap_ambient,
281 					 cap_intersect(*permitted,
282 						       *inheritable));
283 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
284 		return -EINVAL;
285 	return 0;
286 }
287 
288 /*
289  * Clear proposed capability sets for execve().
290  */
291 static inline void bprm_clear_caps(struct linux_binprm *bprm)
292 {
293 	cap_clear(bprm->cred->cap_permitted);
294 	bprm->cap_effective = false;
295 }
296 
297 /**
298  * cap_inode_need_killpriv - Determine if inode change affects privileges
299  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
300  *
301  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
302  * affects the security markings on that inode, and if it is, should
303  * inode_killpriv() be invoked or the change rejected?
304  *
305  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
306  * -ve to deny the change.
307  */
308 int cap_inode_need_killpriv(struct dentry *dentry)
309 {
310 	struct inode *inode = d_backing_inode(dentry);
311 	int error;
312 
313 	if (!inode->i_op->getxattr)
314 	       return 0;
315 
316 	error = inode->i_op->getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
317 	if (error <= 0)
318 		return 0;
319 	return 1;
320 }
321 
322 /**
323  * cap_inode_killpriv - Erase the security markings on an inode
324  * @dentry: The inode/dentry to alter
325  *
326  * Erase the privilege-enhancing security markings on an inode.
327  *
328  * Returns 0 if successful, -ve on error.
329  */
330 int cap_inode_killpriv(struct dentry *dentry)
331 {
332 	struct inode *inode = d_backing_inode(dentry);
333 
334 	if (!inode->i_op->removexattr)
335 	       return 0;
336 
337 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
338 }
339 
340 /*
341  * Calculate the new process capability sets from the capability sets attached
342  * to a file.
343  */
344 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
345 					  struct linux_binprm *bprm,
346 					  bool *effective,
347 					  bool *has_cap)
348 {
349 	struct cred *new = bprm->cred;
350 	unsigned i;
351 	int ret = 0;
352 
353 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
354 		*effective = true;
355 
356 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
357 		*has_cap = true;
358 
359 	CAP_FOR_EACH_U32(i) {
360 		__u32 permitted = caps->permitted.cap[i];
361 		__u32 inheritable = caps->inheritable.cap[i];
362 
363 		/*
364 		 * pP' = (X & fP) | (pI & fI)
365 		 * The addition of pA' is handled later.
366 		 */
367 		new->cap_permitted.cap[i] =
368 			(new->cap_bset.cap[i] & permitted) |
369 			(new->cap_inheritable.cap[i] & inheritable);
370 
371 		if (permitted & ~new->cap_permitted.cap[i])
372 			/* insufficient to execute correctly */
373 			ret = -EPERM;
374 	}
375 
376 	/*
377 	 * For legacy apps, with no internal support for recognizing they
378 	 * do not have enough capabilities, we return an error if they are
379 	 * missing some "forced" (aka file-permitted) capabilities.
380 	 */
381 	return *effective ? ret : 0;
382 }
383 
384 /*
385  * Extract the on-exec-apply capability sets for an executable file.
386  */
387 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
388 {
389 	struct inode *inode = d_backing_inode(dentry);
390 	__u32 magic_etc;
391 	unsigned tocopy, i;
392 	int size;
393 	struct vfs_cap_data caps;
394 
395 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
396 
397 	if (!inode || !inode->i_op->getxattr)
398 		return -ENODATA;
399 
400 	size = inode->i_op->getxattr((struct dentry *)dentry, inode,
401 				     XATTR_NAME_CAPS, &caps, XATTR_CAPS_SZ);
402 	if (size == -ENODATA || size == -EOPNOTSUPP)
403 		/* no data, that's ok */
404 		return -ENODATA;
405 	if (size < 0)
406 		return size;
407 
408 	if (size < sizeof(magic_etc))
409 		return -EINVAL;
410 
411 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
412 
413 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
414 	case VFS_CAP_REVISION_1:
415 		if (size != XATTR_CAPS_SZ_1)
416 			return -EINVAL;
417 		tocopy = VFS_CAP_U32_1;
418 		break;
419 	case VFS_CAP_REVISION_2:
420 		if (size != XATTR_CAPS_SZ_2)
421 			return -EINVAL;
422 		tocopy = VFS_CAP_U32_2;
423 		break;
424 	default:
425 		return -EINVAL;
426 	}
427 
428 	CAP_FOR_EACH_U32(i) {
429 		if (i >= tocopy)
430 			break;
431 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
432 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
433 	}
434 
435 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
436 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
437 
438 	return 0;
439 }
440 
441 /*
442  * Attempt to get the on-exec apply capability sets for an executable file from
443  * its xattrs and, if present, apply them to the proposed credentials being
444  * constructed by execve().
445  */
446 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap)
447 {
448 	int rc = 0;
449 	struct cpu_vfs_cap_data vcaps;
450 
451 	bprm_clear_caps(bprm);
452 
453 	if (!file_caps_enabled)
454 		return 0;
455 
456 	if (!mnt_may_suid(bprm->file->f_path.mnt))
457 		return 0;
458 
459 	/*
460 	 * This check is redundant with mnt_may_suid() but is kept to make
461 	 * explicit that capability bits are limited to s_user_ns and its
462 	 * descendants.
463 	 */
464 	if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns))
465 		return 0;
466 
467 	rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
468 	if (rc < 0) {
469 		if (rc == -EINVAL)
470 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
471 				__func__, rc, bprm->filename);
472 		else if (rc == -ENODATA)
473 			rc = 0;
474 		goto out;
475 	}
476 
477 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
478 	if (rc == -EINVAL)
479 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
480 		       __func__, rc, bprm->filename);
481 
482 out:
483 	if (rc)
484 		bprm_clear_caps(bprm);
485 
486 	return rc;
487 }
488 
489 /**
490  * cap_bprm_set_creds - Set up the proposed credentials for execve().
491  * @bprm: The execution parameters, including the proposed creds
492  *
493  * Set up the proposed credentials for a new execution context being
494  * constructed by execve().  The proposed creds in @bprm->cred is altered,
495  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
496  */
497 int cap_bprm_set_creds(struct linux_binprm *bprm)
498 {
499 	const struct cred *old = current_cred();
500 	struct cred *new = bprm->cred;
501 	bool effective, has_cap = false, is_setid;
502 	int ret;
503 	kuid_t root_uid;
504 
505 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
506 		return -EPERM;
507 
508 	effective = false;
509 	ret = get_file_caps(bprm, &effective, &has_cap);
510 	if (ret < 0)
511 		return ret;
512 
513 	root_uid = make_kuid(new->user_ns, 0);
514 
515 	if (!issecure(SECURE_NOROOT)) {
516 		/*
517 		 * If the legacy file capability is set, then don't set privs
518 		 * for a setuid root binary run by a non-root user.  Do set it
519 		 * for a root user just to cause least surprise to an admin.
520 		 */
521 		if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) {
522 			warn_setuid_and_fcaps_mixed(bprm->filename);
523 			goto skip;
524 		}
525 		/*
526 		 * To support inheritance of root-permissions and suid-root
527 		 * executables under compatibility mode, we override the
528 		 * capability sets for the file.
529 		 *
530 		 * If only the real uid is 0, we do not set the effective bit.
531 		 */
532 		if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) {
533 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
534 			new->cap_permitted = cap_combine(old->cap_bset,
535 							 old->cap_inheritable);
536 		}
537 		if (uid_eq(new->euid, root_uid))
538 			effective = true;
539 	}
540 skip:
541 
542 	/* if we have fs caps, clear dangerous personality flags */
543 	if (!cap_issubset(new->cap_permitted, old->cap_permitted))
544 		bprm->per_clear |= PER_CLEAR_ON_SETID;
545 
546 
547 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
548 	 * credentials unless they have the appropriate permit.
549 	 *
550 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
551 	 */
552 	is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid);
553 
554 	if ((is_setid ||
555 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
556 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
557 		/* downgrade; they get no more than they had, and maybe less */
558 		if (!capable(CAP_SETUID) ||
559 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
560 			new->euid = new->uid;
561 			new->egid = new->gid;
562 		}
563 		new->cap_permitted = cap_intersect(new->cap_permitted,
564 						   old->cap_permitted);
565 	}
566 
567 	new->suid = new->fsuid = new->euid;
568 	new->sgid = new->fsgid = new->egid;
569 
570 	/* File caps or setid cancels ambient. */
571 	if (has_cap || is_setid)
572 		cap_clear(new->cap_ambient);
573 
574 	/*
575 	 * Now that we've computed pA', update pP' to give:
576 	 *   pP' = (X & fP) | (pI & fI) | pA'
577 	 */
578 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
579 
580 	/*
581 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
582 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
583 	 */
584 	if (effective)
585 		new->cap_effective = new->cap_permitted;
586 	else
587 		new->cap_effective = new->cap_ambient;
588 
589 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
590 		return -EPERM;
591 
592 	bprm->cap_effective = effective;
593 
594 	/*
595 	 * Audit candidate if current->cap_effective is set
596 	 *
597 	 * We do not bother to audit if 3 things are true:
598 	 *   1) cap_effective has all caps
599 	 *   2) we are root
600 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
601 	 * Since this is just a normal root execing a process.
602 	 *
603 	 * Number 1 above might fail if you don't have a full bset, but I think
604 	 * that is interesting information to audit.
605 	 */
606 	if (!cap_issubset(new->cap_effective, new->cap_ambient)) {
607 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
608 		    !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) ||
609 		    issecure(SECURE_NOROOT)) {
610 			ret = audit_log_bprm_fcaps(bprm, new, old);
611 			if (ret < 0)
612 				return ret;
613 		}
614 	}
615 
616 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
617 
618 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
619 		return -EPERM;
620 
621 	return 0;
622 }
623 
624 /**
625  * cap_bprm_secureexec - Determine whether a secure execution is required
626  * @bprm: The execution parameters
627  *
628  * Determine whether a secure execution is required, return 1 if it is, and 0
629  * if it is not.
630  *
631  * The credentials have been committed by this point, and so are no longer
632  * available through @bprm->cred.
633  */
634 int cap_bprm_secureexec(struct linux_binprm *bprm)
635 {
636 	const struct cred *cred = current_cred();
637 	kuid_t root_uid = make_kuid(cred->user_ns, 0);
638 
639 	if (!uid_eq(cred->uid, root_uid)) {
640 		if (bprm->cap_effective)
641 			return 1;
642 		if (!cap_issubset(cred->cap_permitted, cred->cap_ambient))
643 			return 1;
644 	}
645 
646 	return (!uid_eq(cred->euid, cred->uid) ||
647 		!gid_eq(cred->egid, cred->gid));
648 }
649 
650 /**
651  * cap_inode_setxattr - Determine whether an xattr may be altered
652  * @dentry: The inode/dentry being altered
653  * @name: The name of the xattr to be changed
654  * @value: The value that the xattr will be changed to
655  * @size: The size of value
656  * @flags: The replacement flag
657  *
658  * Determine whether an xattr may be altered or set on an inode, returning 0 if
659  * permission is granted, -ve if denied.
660  *
661  * This is used to make sure security xattrs don't get updated or set by those
662  * who aren't privileged to do so.
663  */
664 int cap_inode_setxattr(struct dentry *dentry, const char *name,
665 		       const void *value, size_t size, int flags)
666 {
667 	if (!strcmp(name, XATTR_NAME_CAPS)) {
668 		if (!capable(CAP_SETFCAP))
669 			return -EPERM;
670 		return 0;
671 	}
672 
673 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
674 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
675 	    !capable(CAP_SYS_ADMIN))
676 		return -EPERM;
677 	return 0;
678 }
679 
680 /**
681  * cap_inode_removexattr - Determine whether an xattr may be removed
682  * @dentry: The inode/dentry being altered
683  * @name: The name of the xattr to be changed
684  *
685  * Determine whether an xattr may be removed from an inode, returning 0 if
686  * permission is granted, -ve if denied.
687  *
688  * This is used to make sure security xattrs don't get removed by those who
689  * aren't privileged to remove them.
690  */
691 int cap_inode_removexattr(struct dentry *dentry, const char *name)
692 {
693 	if (!strcmp(name, XATTR_NAME_CAPS)) {
694 		if (!capable(CAP_SETFCAP))
695 			return -EPERM;
696 		return 0;
697 	}
698 
699 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
700 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
701 	    !capable(CAP_SYS_ADMIN))
702 		return -EPERM;
703 	return 0;
704 }
705 
706 /*
707  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
708  * a process after a call to setuid, setreuid, or setresuid.
709  *
710  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
711  *  {r,e,s}uid != 0, the permitted and effective capabilities are
712  *  cleared.
713  *
714  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
715  *  capabilities of the process are cleared.
716  *
717  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
718  *  capabilities are set to the permitted capabilities.
719  *
720  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
721  *  never happen.
722  *
723  *  -astor
724  *
725  * cevans - New behaviour, Oct '99
726  * A process may, via prctl(), elect to keep its capabilities when it
727  * calls setuid() and switches away from uid==0. Both permitted and
728  * effective sets will be retained.
729  * Without this change, it was impossible for a daemon to drop only some
730  * of its privilege. The call to setuid(!=0) would drop all privileges!
731  * Keeping uid 0 is not an option because uid 0 owns too many vital
732  * files..
733  * Thanks to Olaf Kirch and Peter Benie for spotting this.
734  */
735 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
736 {
737 	kuid_t root_uid = make_kuid(old->user_ns, 0);
738 
739 	if ((uid_eq(old->uid, root_uid) ||
740 	     uid_eq(old->euid, root_uid) ||
741 	     uid_eq(old->suid, root_uid)) &&
742 	    (!uid_eq(new->uid, root_uid) &&
743 	     !uid_eq(new->euid, root_uid) &&
744 	     !uid_eq(new->suid, root_uid))) {
745 		if (!issecure(SECURE_KEEP_CAPS)) {
746 			cap_clear(new->cap_permitted);
747 			cap_clear(new->cap_effective);
748 		}
749 
750 		/*
751 		 * Pre-ambient programs expect setresuid to nonroot followed
752 		 * by exec to drop capabilities.  We should make sure that
753 		 * this remains the case.
754 		 */
755 		cap_clear(new->cap_ambient);
756 	}
757 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
758 		cap_clear(new->cap_effective);
759 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
760 		new->cap_effective = new->cap_permitted;
761 }
762 
763 /**
764  * cap_task_fix_setuid - Fix up the results of setuid() call
765  * @new: The proposed credentials
766  * @old: The current task's current credentials
767  * @flags: Indications of what has changed
768  *
769  * Fix up the results of setuid() call before the credential changes are
770  * actually applied, returning 0 to grant the changes, -ve to deny them.
771  */
772 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
773 {
774 	switch (flags) {
775 	case LSM_SETID_RE:
776 	case LSM_SETID_ID:
777 	case LSM_SETID_RES:
778 		/* juggle the capabilities to follow [RES]UID changes unless
779 		 * otherwise suppressed */
780 		if (!issecure(SECURE_NO_SETUID_FIXUP))
781 			cap_emulate_setxuid(new, old);
782 		break;
783 
784 	case LSM_SETID_FS:
785 		/* juggle the capabilties to follow FSUID changes, unless
786 		 * otherwise suppressed
787 		 *
788 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
789 		 *          if not, we might be a bit too harsh here.
790 		 */
791 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
792 			kuid_t root_uid = make_kuid(old->user_ns, 0);
793 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
794 				new->cap_effective =
795 					cap_drop_fs_set(new->cap_effective);
796 
797 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
798 				new->cap_effective =
799 					cap_raise_fs_set(new->cap_effective,
800 							 new->cap_permitted);
801 		}
802 		break;
803 
804 	default:
805 		return -EINVAL;
806 	}
807 
808 	return 0;
809 }
810 
811 /*
812  * Rationale: code calling task_setscheduler, task_setioprio, and
813  * task_setnice, assumes that
814  *   . if capable(cap_sys_nice), then those actions should be allowed
815  *   . if not capable(cap_sys_nice), but acting on your own processes,
816  *   	then those actions should be allowed
817  * This is insufficient now since you can call code without suid, but
818  * yet with increased caps.
819  * So we check for increased caps on the target process.
820  */
821 static int cap_safe_nice(struct task_struct *p)
822 {
823 	int is_subset, ret = 0;
824 
825 	rcu_read_lock();
826 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
827 				 current_cred()->cap_permitted);
828 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
829 		ret = -EPERM;
830 	rcu_read_unlock();
831 
832 	return ret;
833 }
834 
835 /**
836  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
837  * @p: The task to affect
838  *
839  * Detemine if the requested scheduler policy change is permitted for the
840  * specified task, returning 0 if permission is granted, -ve if denied.
841  */
842 int cap_task_setscheduler(struct task_struct *p)
843 {
844 	return cap_safe_nice(p);
845 }
846 
847 /**
848  * cap_task_ioprio - Detemine if I/O priority change is permitted
849  * @p: The task to affect
850  * @ioprio: The I/O priority to set
851  *
852  * Detemine if the requested I/O priority change is permitted for the specified
853  * task, returning 0 if permission is granted, -ve if denied.
854  */
855 int cap_task_setioprio(struct task_struct *p, int ioprio)
856 {
857 	return cap_safe_nice(p);
858 }
859 
860 /**
861  * cap_task_ioprio - Detemine if task priority change is permitted
862  * @p: The task to affect
863  * @nice: The nice value to set
864  *
865  * Detemine if the requested task priority change is permitted for the
866  * specified task, returning 0 if permission is granted, -ve if denied.
867  */
868 int cap_task_setnice(struct task_struct *p, int nice)
869 {
870 	return cap_safe_nice(p);
871 }
872 
873 /*
874  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
875  * the current task's bounding set.  Returns 0 on success, -ve on error.
876  */
877 static int cap_prctl_drop(unsigned long cap)
878 {
879 	struct cred *new;
880 
881 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
882 		return -EPERM;
883 	if (!cap_valid(cap))
884 		return -EINVAL;
885 
886 	new = prepare_creds();
887 	if (!new)
888 		return -ENOMEM;
889 	cap_lower(new->cap_bset, cap);
890 	return commit_creds(new);
891 }
892 
893 /**
894  * cap_task_prctl - Implement process control functions for this security module
895  * @option: The process control function requested
896  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
897  *
898  * Allow process control functions (sys_prctl()) to alter capabilities; may
899  * also deny access to other functions not otherwise implemented here.
900  *
901  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
902  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
903  * modules will consider performing the function.
904  */
905 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
906 		   unsigned long arg4, unsigned long arg5)
907 {
908 	const struct cred *old = current_cred();
909 	struct cred *new;
910 
911 	switch (option) {
912 	case PR_CAPBSET_READ:
913 		if (!cap_valid(arg2))
914 			return -EINVAL;
915 		return !!cap_raised(old->cap_bset, arg2);
916 
917 	case PR_CAPBSET_DROP:
918 		return cap_prctl_drop(arg2);
919 
920 	/*
921 	 * The next four prctl's remain to assist with transitioning a
922 	 * system from legacy UID=0 based privilege (when filesystem
923 	 * capabilities are not in use) to a system using filesystem
924 	 * capabilities only - as the POSIX.1e draft intended.
925 	 *
926 	 * Note:
927 	 *
928 	 *  PR_SET_SECUREBITS =
929 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
930 	 *    | issecure_mask(SECURE_NOROOT)
931 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
932 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
933 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
934 	 *
935 	 * will ensure that the current process and all of its
936 	 * children will be locked into a pure
937 	 * capability-based-privilege environment.
938 	 */
939 	case PR_SET_SECUREBITS:
940 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
941 		     & (old->securebits ^ arg2))			/*[1]*/
942 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
943 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
944 		    || (cap_capable(current_cred(),
945 				    current_cred()->user_ns, CAP_SETPCAP,
946 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
947 			/*
948 			 * [1] no changing of bits that are locked
949 			 * [2] no unlocking of locks
950 			 * [3] no setting of unsupported bits
951 			 * [4] doing anything requires privilege (go read about
952 			 *     the "sendmail capabilities bug")
953 			 */
954 		    )
955 			/* cannot change a locked bit */
956 			return -EPERM;
957 
958 		new = prepare_creds();
959 		if (!new)
960 			return -ENOMEM;
961 		new->securebits = arg2;
962 		return commit_creds(new);
963 
964 	case PR_GET_SECUREBITS:
965 		return old->securebits;
966 
967 	case PR_GET_KEEPCAPS:
968 		return !!issecure(SECURE_KEEP_CAPS);
969 
970 	case PR_SET_KEEPCAPS:
971 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
972 			return -EINVAL;
973 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
974 			return -EPERM;
975 
976 		new = prepare_creds();
977 		if (!new)
978 			return -ENOMEM;
979 		if (arg2)
980 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
981 		else
982 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
983 		return commit_creds(new);
984 
985 	case PR_CAP_AMBIENT:
986 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
987 			if (arg3 | arg4 | arg5)
988 				return -EINVAL;
989 
990 			new = prepare_creds();
991 			if (!new)
992 				return -ENOMEM;
993 			cap_clear(new->cap_ambient);
994 			return commit_creds(new);
995 		}
996 
997 		if (((!cap_valid(arg3)) | arg4 | arg5))
998 			return -EINVAL;
999 
1000 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1001 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1002 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1003 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1004 			return -EINVAL;
1005 		} else {
1006 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1007 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1008 			     !cap_raised(current_cred()->cap_inheritable,
1009 					 arg3) ||
1010 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1011 				return -EPERM;
1012 
1013 			new = prepare_creds();
1014 			if (!new)
1015 				return -ENOMEM;
1016 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1017 				cap_raise(new->cap_ambient, arg3);
1018 			else
1019 				cap_lower(new->cap_ambient, arg3);
1020 			return commit_creds(new);
1021 		}
1022 
1023 	default:
1024 		/* No functionality available - continue with default */
1025 		return -ENOSYS;
1026 	}
1027 }
1028 
1029 /**
1030  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1031  * @mm: The VM space in which the new mapping is to be made
1032  * @pages: The size of the mapping
1033  *
1034  * Determine whether the allocation of a new virtual mapping by the current
1035  * task is permitted, returning 1 if permission is granted, 0 if not.
1036  */
1037 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1038 {
1039 	int cap_sys_admin = 0;
1040 
1041 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
1042 			SECURITY_CAP_NOAUDIT) == 0)
1043 		cap_sys_admin = 1;
1044 	return cap_sys_admin;
1045 }
1046 
1047 /*
1048  * cap_mmap_addr - check if able to map given addr
1049  * @addr: address attempting to be mapped
1050  *
1051  * If the process is attempting to map memory below dac_mmap_min_addr they need
1052  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1053  * capability security module.  Returns 0 if this mapping should be allowed
1054  * -EPERM if not.
1055  */
1056 int cap_mmap_addr(unsigned long addr)
1057 {
1058 	int ret = 0;
1059 
1060 	if (addr < dac_mmap_min_addr) {
1061 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1062 				  SECURITY_CAP_AUDIT);
1063 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1064 		if (ret == 0)
1065 			current->flags |= PF_SUPERPRIV;
1066 	}
1067 	return ret;
1068 }
1069 
1070 int cap_mmap_file(struct file *file, unsigned long reqprot,
1071 		  unsigned long prot, unsigned long flags)
1072 {
1073 	return 0;
1074 }
1075 
1076 #ifdef CONFIG_SECURITY
1077 
1078 struct security_hook_list capability_hooks[] = {
1079 	LSM_HOOK_INIT(capable, cap_capable),
1080 	LSM_HOOK_INIT(settime, cap_settime),
1081 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1082 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1083 	LSM_HOOK_INIT(capget, cap_capget),
1084 	LSM_HOOK_INIT(capset, cap_capset),
1085 	LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
1086 	LSM_HOOK_INIT(bprm_secureexec, cap_bprm_secureexec),
1087 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1088 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1089 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1090 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1091 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1092 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1093 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1094 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1095 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1096 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1097 };
1098 
1099 void __init capability_add_hooks(void)
1100 {
1101 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks));
1102 }
1103 
1104 #endif /* CONFIG_SECURITY */
1105