xref: /linux/security/commoncap.c (revision 66559a691a5b4a3d79852a82c8c2f8532077b301)
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  * cap_inode_need_killpriv - Determine if inode change affects privileges
290  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
291  *
292  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
293  * affects the security markings on that inode, and if it is, should
294  * inode_killpriv() be invoked or the change rejected?
295  *
296  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
297  * -ve to deny the change.
298  */
299 int cap_inode_need_killpriv(struct dentry *dentry)
300 {
301 	struct inode *inode = d_backing_inode(dentry);
302 	int error;
303 
304 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
305 	return error > 0;
306 }
307 
308 /**
309  * cap_inode_killpriv - Erase the security markings on an inode
310  * @dentry: The inode/dentry to alter
311  *
312  * Erase the privilege-enhancing security markings on an inode.
313  *
314  * Returns 0 if successful, -ve on error.
315  */
316 int cap_inode_killpriv(struct dentry *dentry)
317 {
318 	int error;
319 
320 	error = __vfs_removexattr(dentry, XATTR_NAME_CAPS);
321 	if (error == -EOPNOTSUPP)
322 		error = 0;
323 	return error;
324 }
325 
326 /*
327  * Calculate the new process capability sets from the capability sets attached
328  * to a file.
329  */
330 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
331 					  struct linux_binprm *bprm,
332 					  bool *effective,
333 					  bool *has_cap)
334 {
335 	struct cred *new = bprm->cred;
336 	unsigned i;
337 	int ret = 0;
338 
339 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
340 		*effective = true;
341 
342 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
343 		*has_cap = true;
344 
345 	CAP_FOR_EACH_U32(i) {
346 		__u32 permitted = caps->permitted.cap[i];
347 		__u32 inheritable = caps->inheritable.cap[i];
348 
349 		/*
350 		 * pP' = (X & fP) | (pI & fI)
351 		 * The addition of pA' is handled later.
352 		 */
353 		new->cap_permitted.cap[i] =
354 			(new->cap_bset.cap[i] & permitted) |
355 			(new->cap_inheritable.cap[i] & inheritable);
356 
357 		if (permitted & ~new->cap_permitted.cap[i])
358 			/* insufficient to execute correctly */
359 			ret = -EPERM;
360 	}
361 
362 	/*
363 	 * For legacy apps, with no internal support for recognizing they
364 	 * do not have enough capabilities, we return an error if they are
365 	 * missing some "forced" (aka file-permitted) capabilities.
366 	 */
367 	return *effective ? ret : 0;
368 }
369 
370 /*
371  * Extract the on-exec-apply capability sets for an executable file.
372  */
373 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
374 {
375 	struct inode *inode = d_backing_inode(dentry);
376 	__u32 magic_etc;
377 	unsigned tocopy, i;
378 	int size;
379 	struct vfs_cap_data caps;
380 
381 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
382 
383 	if (!inode)
384 		return -ENODATA;
385 
386 	size = __vfs_getxattr((struct dentry *)dentry, inode,
387 			      XATTR_NAME_CAPS, &caps, XATTR_CAPS_SZ);
388 	if (size == -ENODATA || size == -EOPNOTSUPP)
389 		/* no data, that's ok */
390 		return -ENODATA;
391 	if (size < 0)
392 		return size;
393 
394 	if (size < sizeof(magic_etc))
395 		return -EINVAL;
396 
397 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
398 
399 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
400 	case VFS_CAP_REVISION_1:
401 		if (size != XATTR_CAPS_SZ_1)
402 			return -EINVAL;
403 		tocopy = VFS_CAP_U32_1;
404 		break;
405 	case VFS_CAP_REVISION_2:
406 		if (size != XATTR_CAPS_SZ_2)
407 			return -EINVAL;
408 		tocopy = VFS_CAP_U32_2;
409 		break;
410 	default:
411 		return -EINVAL;
412 	}
413 
414 	CAP_FOR_EACH_U32(i) {
415 		if (i >= tocopy)
416 			break;
417 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
418 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
419 	}
420 
421 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
422 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
423 
424 	return 0;
425 }
426 
427 /*
428  * Attempt to get the on-exec apply capability sets for an executable file from
429  * its xattrs and, if present, apply them to the proposed credentials being
430  * constructed by execve().
431  */
432 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap)
433 {
434 	int rc = 0;
435 	struct cpu_vfs_cap_data vcaps;
436 
437 	cap_clear(bprm->cred->cap_permitted);
438 
439 	if (!file_caps_enabled)
440 		return 0;
441 
442 	if (!mnt_may_suid(bprm->file->f_path.mnt))
443 		return 0;
444 
445 	/*
446 	 * This check is redundant with mnt_may_suid() but is kept to make
447 	 * explicit that capability bits are limited to s_user_ns and its
448 	 * descendants.
449 	 */
450 	if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns))
451 		return 0;
452 
453 	rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
454 	if (rc < 0) {
455 		if (rc == -EINVAL)
456 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
457 				__func__, rc, bprm->filename);
458 		else if (rc == -ENODATA)
459 			rc = 0;
460 		goto out;
461 	}
462 
463 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
464 	if (rc == -EINVAL)
465 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
466 		       __func__, rc, bprm->filename);
467 
468 out:
469 	if (rc)
470 		cap_clear(bprm->cred->cap_permitted);
471 
472 	return rc;
473 }
474 
475 /**
476  * cap_bprm_set_creds - Set up the proposed credentials for execve().
477  * @bprm: The execution parameters, including the proposed creds
478  *
479  * Set up the proposed credentials for a new execution context being
480  * constructed by execve().  The proposed creds in @bprm->cred is altered,
481  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
482  */
483 int cap_bprm_set_creds(struct linux_binprm *bprm)
484 {
485 	const struct cred *old = current_cred();
486 	struct cred *new = bprm->cred;
487 	bool effective, has_cap = false, is_setid;
488 	int ret;
489 	kuid_t root_uid;
490 
491 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
492 		return -EPERM;
493 
494 	effective = false;
495 	ret = get_file_caps(bprm, &effective, &has_cap);
496 	if (ret < 0)
497 		return ret;
498 
499 	root_uid = make_kuid(new->user_ns, 0);
500 
501 	if (!issecure(SECURE_NOROOT)) {
502 		/*
503 		 * If the legacy file capability is set, then don't set privs
504 		 * for a setuid root binary run by a non-root user.  Do set it
505 		 * for a root user just to cause least surprise to an admin.
506 		 */
507 		if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) {
508 			warn_setuid_and_fcaps_mixed(bprm->filename);
509 			goto skip;
510 		}
511 		/*
512 		 * To support inheritance of root-permissions and suid-root
513 		 * executables under compatibility mode, we override the
514 		 * capability sets for the file.
515 		 *
516 		 * If only the real uid is 0, we do not set the effective bit.
517 		 */
518 		if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) {
519 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
520 			new->cap_permitted = cap_combine(old->cap_bset,
521 							 old->cap_inheritable);
522 		}
523 		if (uid_eq(new->euid, root_uid))
524 			effective = true;
525 	}
526 skip:
527 
528 	/* if we have fs caps, clear dangerous personality flags */
529 	if (!cap_issubset(new->cap_permitted, old->cap_permitted))
530 		bprm->per_clear |= PER_CLEAR_ON_SETID;
531 
532 
533 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
534 	 * credentials unless they have the appropriate permit.
535 	 *
536 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
537 	 */
538 	is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid);
539 
540 	if ((is_setid ||
541 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
542 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
543 	     !ptracer_capable(current, new->user_ns))) {
544 		/* downgrade; they get no more than they had, and maybe less */
545 		if (!ns_capable(new->user_ns, 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 	/*
580 	 * Audit candidate if current->cap_effective is set
581 	 *
582 	 * We do not bother to audit if 3 things are true:
583 	 *   1) cap_effective has all caps
584 	 *   2) we are root
585 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
586 	 * Since this is just a normal root execing a process.
587 	 *
588 	 * Number 1 above might fail if you don't have a full bset, but I think
589 	 * that is interesting information to audit.
590 	 */
591 	if (!cap_issubset(new->cap_effective, new->cap_ambient)) {
592 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
593 		    !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) ||
594 		    issecure(SECURE_NOROOT)) {
595 			ret = audit_log_bprm_fcaps(bprm, new, old);
596 			if (ret < 0)
597 				return ret;
598 		}
599 	}
600 
601 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
602 
603 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
604 		return -EPERM;
605 
606 	/* Check for privilege-elevated exec. */
607 	bprm->cap_elevated = 0;
608 	if (is_setid) {
609 		bprm->cap_elevated = 1;
610 	} else if (!uid_eq(new->uid, root_uid)) {
611 		if (effective ||
612 		    !cap_issubset(new->cap_permitted, new->cap_ambient))
613 			bprm->cap_elevated = 1;
614 	}
615 
616 	return 0;
617 }
618 
619 /**
620  * cap_inode_setxattr - Determine whether an xattr may be altered
621  * @dentry: The inode/dentry being altered
622  * @name: The name of the xattr to be changed
623  * @value: The value that the xattr will be changed to
624  * @size: The size of value
625  * @flags: The replacement flag
626  *
627  * Determine whether an xattr may be altered or set on an inode, returning 0 if
628  * permission is granted, -ve if denied.
629  *
630  * This is used to make sure security xattrs don't get updated or set by those
631  * who aren't privileged to do so.
632  */
633 int cap_inode_setxattr(struct dentry *dentry, const char *name,
634 		       const void *value, size_t size, int flags)
635 {
636 	if (!strcmp(name, XATTR_NAME_CAPS)) {
637 		if (!capable(CAP_SETFCAP))
638 			return -EPERM;
639 		return 0;
640 	}
641 
642 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
643 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
644 	    !capable(CAP_SYS_ADMIN))
645 		return -EPERM;
646 	return 0;
647 }
648 
649 /**
650  * cap_inode_removexattr - Determine whether an xattr may be removed
651  * @dentry: The inode/dentry being altered
652  * @name: The name of the xattr to be changed
653  *
654  * Determine whether an xattr may be removed from an inode, returning 0 if
655  * permission is granted, -ve if denied.
656  *
657  * This is used to make sure security xattrs don't get removed by those who
658  * aren't privileged to remove them.
659  */
660 int cap_inode_removexattr(struct dentry *dentry, const char *name)
661 {
662 	if (!strcmp(name, XATTR_NAME_CAPS)) {
663 		if (!capable(CAP_SETFCAP))
664 			return -EPERM;
665 		return 0;
666 	}
667 
668 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
669 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
670 	    !capable(CAP_SYS_ADMIN))
671 		return -EPERM;
672 	return 0;
673 }
674 
675 /*
676  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
677  * a process after a call to setuid, setreuid, or setresuid.
678  *
679  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
680  *  {r,e,s}uid != 0, the permitted and effective capabilities are
681  *  cleared.
682  *
683  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
684  *  capabilities of the process are cleared.
685  *
686  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
687  *  capabilities are set to the permitted capabilities.
688  *
689  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
690  *  never happen.
691  *
692  *  -astor
693  *
694  * cevans - New behaviour, Oct '99
695  * A process may, via prctl(), elect to keep its capabilities when it
696  * calls setuid() and switches away from uid==0. Both permitted and
697  * effective sets will be retained.
698  * Without this change, it was impossible for a daemon to drop only some
699  * of its privilege. The call to setuid(!=0) would drop all privileges!
700  * Keeping uid 0 is not an option because uid 0 owns too many vital
701  * files..
702  * Thanks to Olaf Kirch and Peter Benie for spotting this.
703  */
704 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
705 {
706 	kuid_t root_uid = make_kuid(old->user_ns, 0);
707 
708 	if ((uid_eq(old->uid, root_uid) ||
709 	     uid_eq(old->euid, root_uid) ||
710 	     uid_eq(old->suid, root_uid)) &&
711 	    (!uid_eq(new->uid, root_uid) &&
712 	     !uid_eq(new->euid, root_uid) &&
713 	     !uid_eq(new->suid, root_uid))) {
714 		if (!issecure(SECURE_KEEP_CAPS)) {
715 			cap_clear(new->cap_permitted);
716 			cap_clear(new->cap_effective);
717 		}
718 
719 		/*
720 		 * Pre-ambient programs expect setresuid to nonroot followed
721 		 * by exec to drop capabilities.  We should make sure that
722 		 * this remains the case.
723 		 */
724 		cap_clear(new->cap_ambient);
725 	}
726 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
727 		cap_clear(new->cap_effective);
728 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
729 		new->cap_effective = new->cap_permitted;
730 }
731 
732 /**
733  * cap_task_fix_setuid - Fix up the results of setuid() call
734  * @new: The proposed credentials
735  * @old: The current task's current credentials
736  * @flags: Indications of what has changed
737  *
738  * Fix up the results of setuid() call before the credential changes are
739  * actually applied, returning 0 to grant the changes, -ve to deny them.
740  */
741 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
742 {
743 	switch (flags) {
744 	case LSM_SETID_RE:
745 	case LSM_SETID_ID:
746 	case LSM_SETID_RES:
747 		/* juggle the capabilities to follow [RES]UID changes unless
748 		 * otherwise suppressed */
749 		if (!issecure(SECURE_NO_SETUID_FIXUP))
750 			cap_emulate_setxuid(new, old);
751 		break;
752 
753 	case LSM_SETID_FS:
754 		/* juggle the capabilties to follow FSUID changes, unless
755 		 * otherwise suppressed
756 		 *
757 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
758 		 *          if not, we might be a bit too harsh here.
759 		 */
760 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
761 			kuid_t root_uid = make_kuid(old->user_ns, 0);
762 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
763 				new->cap_effective =
764 					cap_drop_fs_set(new->cap_effective);
765 
766 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
767 				new->cap_effective =
768 					cap_raise_fs_set(new->cap_effective,
769 							 new->cap_permitted);
770 		}
771 		break;
772 
773 	default:
774 		return -EINVAL;
775 	}
776 
777 	return 0;
778 }
779 
780 /*
781  * Rationale: code calling task_setscheduler, task_setioprio, and
782  * task_setnice, assumes that
783  *   . if capable(cap_sys_nice), then those actions should be allowed
784  *   . if not capable(cap_sys_nice), but acting on your own processes,
785  *   	then those actions should be allowed
786  * This is insufficient now since you can call code without suid, but
787  * yet with increased caps.
788  * So we check for increased caps on the target process.
789  */
790 static int cap_safe_nice(struct task_struct *p)
791 {
792 	int is_subset, ret = 0;
793 
794 	rcu_read_lock();
795 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
796 				 current_cred()->cap_permitted);
797 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
798 		ret = -EPERM;
799 	rcu_read_unlock();
800 
801 	return ret;
802 }
803 
804 /**
805  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
806  * @p: The task to affect
807  *
808  * Detemine if the requested scheduler policy change is permitted for the
809  * specified task, returning 0 if permission is granted, -ve if denied.
810  */
811 int cap_task_setscheduler(struct task_struct *p)
812 {
813 	return cap_safe_nice(p);
814 }
815 
816 /**
817  * cap_task_ioprio - Detemine if I/O priority change is permitted
818  * @p: The task to affect
819  * @ioprio: The I/O priority to set
820  *
821  * Detemine if the requested I/O priority change is permitted for the specified
822  * task, returning 0 if permission is granted, -ve if denied.
823  */
824 int cap_task_setioprio(struct task_struct *p, int ioprio)
825 {
826 	return cap_safe_nice(p);
827 }
828 
829 /**
830  * cap_task_ioprio - Detemine if task priority change is permitted
831  * @p: The task to affect
832  * @nice: The nice value to set
833  *
834  * Detemine if the requested task priority change is permitted for the
835  * specified task, returning 0 if permission is granted, -ve if denied.
836  */
837 int cap_task_setnice(struct task_struct *p, int nice)
838 {
839 	return cap_safe_nice(p);
840 }
841 
842 /*
843  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
844  * the current task's bounding set.  Returns 0 on success, -ve on error.
845  */
846 static int cap_prctl_drop(unsigned long cap)
847 {
848 	struct cred *new;
849 
850 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
851 		return -EPERM;
852 	if (!cap_valid(cap))
853 		return -EINVAL;
854 
855 	new = prepare_creds();
856 	if (!new)
857 		return -ENOMEM;
858 	cap_lower(new->cap_bset, cap);
859 	return commit_creds(new);
860 }
861 
862 /**
863  * cap_task_prctl - Implement process control functions for this security module
864  * @option: The process control function requested
865  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
866  *
867  * Allow process control functions (sys_prctl()) to alter capabilities; may
868  * also deny access to other functions not otherwise implemented here.
869  *
870  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
871  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
872  * modules will consider performing the function.
873  */
874 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
875 		   unsigned long arg4, unsigned long arg5)
876 {
877 	const struct cred *old = current_cred();
878 	struct cred *new;
879 
880 	switch (option) {
881 	case PR_CAPBSET_READ:
882 		if (!cap_valid(arg2))
883 			return -EINVAL;
884 		return !!cap_raised(old->cap_bset, arg2);
885 
886 	case PR_CAPBSET_DROP:
887 		return cap_prctl_drop(arg2);
888 
889 	/*
890 	 * The next four prctl's remain to assist with transitioning a
891 	 * system from legacy UID=0 based privilege (when filesystem
892 	 * capabilities are not in use) to a system using filesystem
893 	 * capabilities only - as the POSIX.1e draft intended.
894 	 *
895 	 * Note:
896 	 *
897 	 *  PR_SET_SECUREBITS =
898 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
899 	 *    | issecure_mask(SECURE_NOROOT)
900 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
901 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
902 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
903 	 *
904 	 * will ensure that the current process and all of its
905 	 * children will be locked into a pure
906 	 * capability-based-privilege environment.
907 	 */
908 	case PR_SET_SECUREBITS:
909 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
910 		     & (old->securebits ^ arg2))			/*[1]*/
911 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
912 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
913 		    || (cap_capable(current_cred(),
914 				    current_cred()->user_ns, CAP_SETPCAP,
915 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
916 			/*
917 			 * [1] no changing of bits that are locked
918 			 * [2] no unlocking of locks
919 			 * [3] no setting of unsupported bits
920 			 * [4] doing anything requires privilege (go read about
921 			 *     the "sendmail capabilities bug")
922 			 */
923 		    )
924 			/* cannot change a locked bit */
925 			return -EPERM;
926 
927 		new = prepare_creds();
928 		if (!new)
929 			return -ENOMEM;
930 		new->securebits = arg2;
931 		return commit_creds(new);
932 
933 	case PR_GET_SECUREBITS:
934 		return old->securebits;
935 
936 	case PR_GET_KEEPCAPS:
937 		return !!issecure(SECURE_KEEP_CAPS);
938 
939 	case PR_SET_KEEPCAPS:
940 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
941 			return -EINVAL;
942 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
943 			return -EPERM;
944 
945 		new = prepare_creds();
946 		if (!new)
947 			return -ENOMEM;
948 		if (arg2)
949 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
950 		else
951 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
952 		return commit_creds(new);
953 
954 	case PR_CAP_AMBIENT:
955 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
956 			if (arg3 | arg4 | arg5)
957 				return -EINVAL;
958 
959 			new = prepare_creds();
960 			if (!new)
961 				return -ENOMEM;
962 			cap_clear(new->cap_ambient);
963 			return commit_creds(new);
964 		}
965 
966 		if (((!cap_valid(arg3)) | arg4 | arg5))
967 			return -EINVAL;
968 
969 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
970 			return !!cap_raised(current_cred()->cap_ambient, arg3);
971 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
972 			   arg2 != PR_CAP_AMBIENT_LOWER) {
973 			return -EINVAL;
974 		} else {
975 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
976 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
977 			     !cap_raised(current_cred()->cap_inheritable,
978 					 arg3) ||
979 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
980 				return -EPERM;
981 
982 			new = prepare_creds();
983 			if (!new)
984 				return -ENOMEM;
985 			if (arg2 == PR_CAP_AMBIENT_RAISE)
986 				cap_raise(new->cap_ambient, arg3);
987 			else
988 				cap_lower(new->cap_ambient, arg3);
989 			return commit_creds(new);
990 		}
991 
992 	default:
993 		/* No functionality available - continue with default */
994 		return -ENOSYS;
995 	}
996 }
997 
998 /**
999  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1000  * @mm: The VM space in which the new mapping is to be made
1001  * @pages: The size of the mapping
1002  *
1003  * Determine whether the allocation of a new virtual mapping by the current
1004  * task is permitted, returning 1 if permission is granted, 0 if not.
1005  */
1006 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1007 {
1008 	int cap_sys_admin = 0;
1009 
1010 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
1011 			SECURITY_CAP_NOAUDIT) == 0)
1012 		cap_sys_admin = 1;
1013 	return cap_sys_admin;
1014 }
1015 
1016 /*
1017  * cap_mmap_addr - check if able to map given addr
1018  * @addr: address attempting to be mapped
1019  *
1020  * If the process is attempting to map memory below dac_mmap_min_addr they need
1021  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1022  * capability security module.  Returns 0 if this mapping should be allowed
1023  * -EPERM if not.
1024  */
1025 int cap_mmap_addr(unsigned long addr)
1026 {
1027 	int ret = 0;
1028 
1029 	if (addr < dac_mmap_min_addr) {
1030 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1031 				  SECURITY_CAP_AUDIT);
1032 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1033 		if (ret == 0)
1034 			current->flags |= PF_SUPERPRIV;
1035 	}
1036 	return ret;
1037 }
1038 
1039 int cap_mmap_file(struct file *file, unsigned long reqprot,
1040 		  unsigned long prot, unsigned long flags)
1041 {
1042 	return 0;
1043 }
1044 
1045 #ifdef CONFIG_SECURITY
1046 
1047 struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1048 	LSM_HOOK_INIT(capable, cap_capable),
1049 	LSM_HOOK_INIT(settime, cap_settime),
1050 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1051 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1052 	LSM_HOOK_INIT(capget, cap_capget),
1053 	LSM_HOOK_INIT(capset, cap_capset),
1054 	LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
1055 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1056 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1057 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1058 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1059 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1060 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1061 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1062 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1063 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1064 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1065 };
1066 
1067 void __init capability_add_hooks(void)
1068 {
1069 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1070 				"capability");
1071 }
1072 
1073 #endif /* CONFIG_SECURITY */
1074