xref: /linux/security/commoncap.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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/security.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 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
57 {
58 	return 0;
59 }
60 
61 /**
62  * cap_capable - Determine whether a task has a particular effective capability
63  * @cred: The credentials to use
64  * @ns:  The user namespace in which we need the capability
65  * @cap: The capability to check for
66  * @audit: Whether to write an audit message or not
67  *
68  * Determine whether the nominated task has the specified capability amongst
69  * its effective set, returning 0 if it does, -ve if it does not.
70  *
71  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
72  * and has_capability() functions.  That is, it has the reverse semantics:
73  * cap_has_capability() returns 0 when a task has a capability, but the
74  * kernel's capable() and has_capability() returns 1 for this case.
75  */
76 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
77 		int cap, int audit)
78 {
79 	struct user_namespace *ns = targ_ns;
80 
81 	/* See if cred has the capability in the target user namespace
82 	 * by examining the target user namespace and all of the target
83 	 * user namespace's parents.
84 	 */
85 	for (;;) {
86 		/* Do we have the necessary capabilities? */
87 		if (ns == cred->user_ns)
88 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
89 
90 		/* Have we tried all of the parent namespaces? */
91 		if (ns == &init_user_ns)
92 			return -EPERM;
93 
94 		/*
95 		 * The owner of the user namespace in the parent of the
96 		 * user namespace has all caps.
97 		 */
98 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
99 			return 0;
100 
101 		/*
102 		 * If you have a capability in a parent user ns, then you have
103 		 * it over all children user namespaces as well.
104 		 */
105 		ns = ns->parent;
106 	}
107 
108 	/* We never get here */
109 }
110 
111 /**
112  * cap_settime - Determine whether the current process may set the system clock
113  * @ts: The time to set
114  * @tz: The timezone to set
115  *
116  * Determine whether the current process may set the system clock and timezone
117  * information, returning 0 if permission granted, -ve if denied.
118  */
119 int cap_settime(const struct timespec *ts, const struct timezone *tz)
120 {
121 	if (!capable(CAP_SYS_TIME))
122 		return -EPERM;
123 	return 0;
124 }
125 
126 /**
127  * cap_ptrace_access_check - Determine whether the current process may access
128  *			   another
129  * @child: The process to be accessed
130  * @mode: The mode of attachment.
131  *
132  * If we are in the same or an ancestor user_ns and have all the target
133  * task's capabilities, then ptrace access is allowed.
134  * If we have the ptrace capability to the target user_ns, then ptrace
135  * access is allowed.
136  * Else denied.
137  *
138  * Determine whether a process may access another, returning 0 if permission
139  * granted, -ve if denied.
140  */
141 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
142 {
143 	int ret = 0;
144 	const struct cred *cred, *child_cred;
145 
146 	rcu_read_lock();
147 	cred = current_cred();
148 	child_cred = __task_cred(child);
149 	if (cred->user_ns == child_cred->user_ns &&
150 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
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 	return 0;
276 }
277 
278 /*
279  * Clear proposed capability sets for execve().
280  */
281 static inline void bprm_clear_caps(struct linux_binprm *bprm)
282 {
283 	cap_clear(bprm->cred->cap_permitted);
284 	bprm->cap_effective = false;
285 }
286 
287 /**
288  * cap_inode_need_killpriv - Determine if inode change affects privileges
289  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
290  *
291  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
292  * affects the security markings on that inode, and if it is, should
293  * inode_killpriv() be invoked or the change rejected?
294  *
295  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
296  * -ve to deny the change.
297  */
298 int cap_inode_need_killpriv(struct dentry *dentry)
299 {
300 	struct inode *inode = dentry->d_inode;
301 	int error;
302 
303 	if (!inode->i_op->getxattr)
304 	       return 0;
305 
306 	error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
307 	if (error <= 0)
308 		return 0;
309 	return 1;
310 }
311 
312 /**
313  * cap_inode_killpriv - Erase the security markings on an inode
314  * @dentry: The inode/dentry to alter
315  *
316  * Erase the privilege-enhancing security markings on an inode.
317  *
318  * Returns 0 if successful, -ve on error.
319  */
320 int cap_inode_killpriv(struct dentry *dentry)
321 {
322 	struct inode *inode = dentry->d_inode;
323 
324 	if (!inode->i_op->removexattr)
325 	       return 0;
326 
327 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
328 }
329 
330 /*
331  * Calculate the new process capability sets from the capability sets attached
332  * to a file.
333  */
334 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
335 					  struct linux_binprm *bprm,
336 					  bool *effective,
337 					  bool *has_cap)
338 {
339 	struct cred *new = bprm->cred;
340 	unsigned i;
341 	int ret = 0;
342 
343 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
344 		*effective = true;
345 
346 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
347 		*has_cap = true;
348 
349 	CAP_FOR_EACH_U32(i) {
350 		__u32 permitted = caps->permitted.cap[i];
351 		__u32 inheritable = caps->inheritable.cap[i];
352 
353 		/*
354 		 * pP' = (X & fP) | (pI & fI)
355 		 */
356 		new->cap_permitted.cap[i] =
357 			(new->cap_bset.cap[i] & permitted) |
358 			(new->cap_inheritable.cap[i] & inheritable);
359 
360 		if (permitted & ~new->cap_permitted.cap[i])
361 			/* insufficient to execute correctly */
362 			ret = -EPERM;
363 	}
364 
365 	/*
366 	 * For legacy apps, with no internal support for recognizing they
367 	 * do not have enough capabilities, we return an error if they are
368 	 * missing some "forced" (aka file-permitted) capabilities.
369 	 */
370 	return *effective ? ret : 0;
371 }
372 
373 /*
374  * Extract the on-exec-apply capability sets for an executable file.
375  */
376 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
377 {
378 	struct inode *inode = dentry->d_inode;
379 	__u32 magic_etc;
380 	unsigned tocopy, i;
381 	int size;
382 	struct vfs_cap_data caps;
383 
384 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
385 
386 	if (!inode || !inode->i_op->getxattr)
387 		return -ENODATA;
388 
389 	size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
390 				   XATTR_CAPS_SZ);
391 	if (size == -ENODATA || size == -EOPNOTSUPP)
392 		/* no data, that's ok */
393 		return -ENODATA;
394 	if (size < 0)
395 		return size;
396 
397 	if (size < sizeof(magic_etc))
398 		return -EINVAL;
399 
400 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
401 
402 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
403 	case VFS_CAP_REVISION_1:
404 		if (size != XATTR_CAPS_SZ_1)
405 			return -EINVAL;
406 		tocopy = VFS_CAP_U32_1;
407 		break;
408 	case VFS_CAP_REVISION_2:
409 		if (size != XATTR_CAPS_SZ_2)
410 			return -EINVAL;
411 		tocopy = VFS_CAP_U32_2;
412 		break;
413 	default:
414 		return -EINVAL;
415 	}
416 
417 	CAP_FOR_EACH_U32(i) {
418 		if (i >= tocopy)
419 			break;
420 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
421 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
422 	}
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 	struct dentry *dentry;
435 	int rc = 0;
436 	struct cpu_vfs_cap_data vcaps;
437 
438 	bprm_clear_caps(bprm);
439 
440 	if (!file_caps_enabled)
441 		return 0;
442 
443 	if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)
444 		return 0;
445 
446 	dentry = dget(bprm->file->f_dentry);
447 
448 	rc = get_vfs_caps_from_disk(dentry, &vcaps);
449 	if (rc < 0) {
450 		if (rc == -EINVAL)
451 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
452 				__func__, rc, bprm->filename);
453 		else if (rc == -ENODATA)
454 			rc = 0;
455 		goto out;
456 	}
457 
458 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
459 	if (rc == -EINVAL)
460 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
461 		       __func__, rc, bprm->filename);
462 
463 out:
464 	dput(dentry);
465 	if (rc)
466 		bprm_clear_caps(bprm);
467 
468 	return rc;
469 }
470 
471 /**
472  * cap_bprm_set_creds - Set up the proposed credentials for execve().
473  * @bprm: The execution parameters, including the proposed creds
474  *
475  * Set up the proposed credentials for a new execution context being
476  * constructed by execve().  The proposed creds in @bprm->cred is altered,
477  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
478  */
479 int cap_bprm_set_creds(struct linux_binprm *bprm)
480 {
481 	const struct cred *old = current_cred();
482 	struct cred *new = bprm->cred;
483 	bool effective, has_cap = false;
484 	int ret;
485 	kuid_t root_uid;
486 
487 	effective = false;
488 	ret = get_file_caps(bprm, &effective, &has_cap);
489 	if (ret < 0)
490 		return ret;
491 
492 	root_uid = make_kuid(new->user_ns, 0);
493 
494 	if (!issecure(SECURE_NOROOT)) {
495 		/*
496 		 * If the legacy file capability is set, then don't set privs
497 		 * for a setuid root binary run by a non-root user.  Do set it
498 		 * for a root user just to cause least surprise to an admin.
499 		 */
500 		if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) {
501 			warn_setuid_and_fcaps_mixed(bprm->filename);
502 			goto skip;
503 		}
504 		/*
505 		 * To support inheritance of root-permissions and suid-root
506 		 * executables under compatibility mode, we override the
507 		 * capability sets for the file.
508 		 *
509 		 * If only the real uid is 0, we do not set the effective bit.
510 		 */
511 		if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) {
512 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
513 			new->cap_permitted = cap_combine(old->cap_bset,
514 							 old->cap_inheritable);
515 		}
516 		if (uid_eq(new->euid, root_uid))
517 			effective = true;
518 	}
519 skip:
520 
521 	/* if we have fs caps, clear dangerous personality flags */
522 	if (!cap_issubset(new->cap_permitted, old->cap_permitted))
523 		bprm->per_clear |= PER_CLEAR_ON_SETID;
524 
525 
526 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
527 	 * credentials unless they have the appropriate permit.
528 	 *
529 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
530 	 */
531 	if ((!uid_eq(new->euid, old->uid) ||
532 	     !gid_eq(new->egid, old->gid) ||
533 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
534 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
535 		/* downgrade; they get no more than they had, and maybe less */
536 		if (!capable(CAP_SETUID) ||
537 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
538 			new->euid = new->uid;
539 			new->egid = new->gid;
540 		}
541 		new->cap_permitted = cap_intersect(new->cap_permitted,
542 						   old->cap_permitted);
543 	}
544 
545 	new->suid = new->fsuid = new->euid;
546 	new->sgid = new->fsgid = new->egid;
547 
548 	if (effective)
549 		new->cap_effective = new->cap_permitted;
550 	else
551 		cap_clear(new->cap_effective);
552 	bprm->cap_effective = effective;
553 
554 	/*
555 	 * Audit candidate if current->cap_effective is set
556 	 *
557 	 * We do not bother to audit if 3 things are true:
558 	 *   1) cap_effective has all caps
559 	 *   2) we are root
560 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
561 	 * Since this is just a normal root execing a process.
562 	 *
563 	 * Number 1 above might fail if you don't have a full bset, but I think
564 	 * that is interesting information to audit.
565 	 */
566 	if (!cap_isclear(new->cap_effective)) {
567 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
568 		    !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) ||
569 		    issecure(SECURE_NOROOT)) {
570 			ret = audit_log_bprm_fcaps(bprm, new, old);
571 			if (ret < 0)
572 				return ret;
573 		}
574 	}
575 
576 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
577 	return 0;
578 }
579 
580 /**
581  * cap_bprm_secureexec - Determine whether a secure execution is required
582  * @bprm: The execution parameters
583  *
584  * Determine whether a secure execution is required, return 1 if it is, and 0
585  * if it is not.
586  *
587  * The credentials have been committed by this point, and so are no longer
588  * available through @bprm->cred.
589  */
590 int cap_bprm_secureexec(struct linux_binprm *bprm)
591 {
592 	const struct cred *cred = current_cred();
593 	kuid_t root_uid = make_kuid(cred->user_ns, 0);
594 
595 	if (!uid_eq(cred->uid, root_uid)) {
596 		if (bprm->cap_effective)
597 			return 1;
598 		if (!cap_isclear(cred->cap_permitted))
599 			return 1;
600 	}
601 
602 	return (!uid_eq(cred->euid, cred->uid) ||
603 		!gid_eq(cred->egid, cred->gid));
604 }
605 
606 /**
607  * cap_inode_setxattr - Determine whether an xattr may be altered
608  * @dentry: The inode/dentry being altered
609  * @name: The name of the xattr to be changed
610  * @value: The value that the xattr will be changed to
611  * @size: The size of value
612  * @flags: The replacement flag
613  *
614  * Determine whether an xattr may be altered or set on an inode, returning 0 if
615  * permission is granted, -ve if denied.
616  *
617  * This is used to make sure security xattrs don't get updated or set by those
618  * who aren't privileged to do so.
619  */
620 int cap_inode_setxattr(struct dentry *dentry, const char *name,
621 		       const void *value, size_t size, int flags)
622 {
623 	if (!strcmp(name, XATTR_NAME_CAPS)) {
624 		if (!capable(CAP_SETFCAP))
625 			return -EPERM;
626 		return 0;
627 	}
628 
629 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
630 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
631 	    !capable(CAP_SYS_ADMIN))
632 		return -EPERM;
633 	return 0;
634 }
635 
636 /**
637  * cap_inode_removexattr - Determine whether an xattr may be removed
638  * @dentry: The inode/dentry being altered
639  * @name: The name of the xattr to be changed
640  *
641  * Determine whether an xattr may be removed from an inode, returning 0 if
642  * permission is granted, -ve if denied.
643  *
644  * This is used to make sure security xattrs don't get removed by those who
645  * aren't privileged to remove them.
646  */
647 int cap_inode_removexattr(struct dentry *dentry, const char *name)
648 {
649 	if (!strcmp(name, XATTR_NAME_CAPS)) {
650 		if (!capable(CAP_SETFCAP))
651 			return -EPERM;
652 		return 0;
653 	}
654 
655 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
656 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
657 	    !capable(CAP_SYS_ADMIN))
658 		return -EPERM;
659 	return 0;
660 }
661 
662 /*
663  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
664  * a process after a call to setuid, setreuid, or setresuid.
665  *
666  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
667  *  {r,e,s}uid != 0, the permitted and effective capabilities are
668  *  cleared.
669  *
670  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
671  *  capabilities of the process are cleared.
672  *
673  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
674  *  capabilities are set to the permitted capabilities.
675  *
676  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
677  *  never happen.
678  *
679  *  -astor
680  *
681  * cevans - New behaviour, Oct '99
682  * A process may, via prctl(), elect to keep its capabilities when it
683  * calls setuid() and switches away from uid==0. Both permitted and
684  * effective sets will be retained.
685  * Without this change, it was impossible for a daemon to drop only some
686  * of its privilege. The call to setuid(!=0) would drop all privileges!
687  * Keeping uid 0 is not an option because uid 0 owns too many vital
688  * files..
689  * Thanks to Olaf Kirch and Peter Benie for spotting this.
690  */
691 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
692 {
693 	kuid_t root_uid = make_kuid(old->user_ns, 0);
694 
695 	if ((uid_eq(old->uid, root_uid) ||
696 	     uid_eq(old->euid, root_uid) ||
697 	     uid_eq(old->suid, root_uid)) &&
698 	    (!uid_eq(new->uid, root_uid) &&
699 	     !uid_eq(new->euid, root_uid) &&
700 	     !uid_eq(new->suid, root_uid)) &&
701 	    !issecure(SECURE_KEEP_CAPS)) {
702 		cap_clear(new->cap_permitted);
703 		cap_clear(new->cap_effective);
704 	}
705 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
706 		cap_clear(new->cap_effective);
707 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
708 		new->cap_effective = new->cap_permitted;
709 }
710 
711 /**
712  * cap_task_fix_setuid - Fix up the results of setuid() call
713  * @new: The proposed credentials
714  * @old: The current task's current credentials
715  * @flags: Indications of what has changed
716  *
717  * Fix up the results of setuid() call before the credential changes are
718  * actually applied, returning 0 to grant the changes, -ve to deny them.
719  */
720 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
721 {
722 	switch (flags) {
723 	case LSM_SETID_RE:
724 	case LSM_SETID_ID:
725 	case LSM_SETID_RES:
726 		/* juggle the capabilities to follow [RES]UID changes unless
727 		 * otherwise suppressed */
728 		if (!issecure(SECURE_NO_SETUID_FIXUP))
729 			cap_emulate_setxuid(new, old);
730 		break;
731 
732 	case LSM_SETID_FS:
733 		/* juggle the capabilties to follow FSUID changes, unless
734 		 * otherwise suppressed
735 		 *
736 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
737 		 *          if not, we might be a bit too harsh here.
738 		 */
739 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
740 			kuid_t root_uid = make_kuid(old->user_ns, 0);
741 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
742 				new->cap_effective =
743 					cap_drop_fs_set(new->cap_effective);
744 
745 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
746 				new->cap_effective =
747 					cap_raise_fs_set(new->cap_effective,
748 							 new->cap_permitted);
749 		}
750 		break;
751 
752 	default:
753 		return -EINVAL;
754 	}
755 
756 	return 0;
757 }
758 
759 /*
760  * Rationale: code calling task_setscheduler, task_setioprio, and
761  * task_setnice, assumes that
762  *   . if capable(cap_sys_nice), then those actions should be allowed
763  *   . if not capable(cap_sys_nice), but acting on your own processes,
764  *   	then those actions should be allowed
765  * This is insufficient now since you can call code without suid, but
766  * yet with increased caps.
767  * So we check for increased caps on the target process.
768  */
769 static int cap_safe_nice(struct task_struct *p)
770 {
771 	int is_subset;
772 
773 	rcu_read_lock();
774 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
775 				 current_cred()->cap_permitted);
776 	rcu_read_unlock();
777 
778 	if (!is_subset && !capable(CAP_SYS_NICE))
779 		return -EPERM;
780 	return 0;
781 }
782 
783 /**
784  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
785  * @p: The task to affect
786  *
787  * Detemine if the requested scheduler policy change is permitted for the
788  * specified task, returning 0 if permission is granted, -ve if denied.
789  */
790 int cap_task_setscheduler(struct task_struct *p)
791 {
792 	return cap_safe_nice(p);
793 }
794 
795 /**
796  * cap_task_ioprio - Detemine if I/O priority change is permitted
797  * @p: The task to affect
798  * @ioprio: The I/O priority to set
799  *
800  * Detemine if the requested I/O priority change is permitted for the specified
801  * task, returning 0 if permission is granted, -ve if denied.
802  */
803 int cap_task_setioprio(struct task_struct *p, int ioprio)
804 {
805 	return cap_safe_nice(p);
806 }
807 
808 /**
809  * cap_task_ioprio - Detemine if task priority change is permitted
810  * @p: The task to affect
811  * @nice: The nice value to set
812  *
813  * Detemine if the requested task priority change is permitted for the
814  * specified task, returning 0 if permission is granted, -ve if denied.
815  */
816 int cap_task_setnice(struct task_struct *p, int nice)
817 {
818 	return cap_safe_nice(p);
819 }
820 
821 /*
822  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
823  * the current task's bounding set.  Returns 0 on success, -ve on error.
824  */
825 static long cap_prctl_drop(struct cred *new, unsigned long cap)
826 {
827 	if (!capable(CAP_SETPCAP))
828 		return -EPERM;
829 	if (!cap_valid(cap))
830 		return -EINVAL;
831 
832 	cap_lower(new->cap_bset, cap);
833 	return 0;
834 }
835 
836 /**
837  * cap_task_prctl - Implement process control functions for this security module
838  * @option: The process control function requested
839  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
840  *
841  * Allow process control functions (sys_prctl()) to alter capabilities; may
842  * also deny access to other functions not otherwise implemented here.
843  *
844  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
845  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
846  * modules will consider performing the function.
847  */
848 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
849 		   unsigned long arg4, unsigned long arg5)
850 {
851 	struct cred *new;
852 	long error = 0;
853 
854 	new = prepare_creds();
855 	if (!new)
856 		return -ENOMEM;
857 
858 	switch (option) {
859 	case PR_CAPBSET_READ:
860 		error = -EINVAL;
861 		if (!cap_valid(arg2))
862 			goto error;
863 		error = !!cap_raised(new->cap_bset, arg2);
864 		goto no_change;
865 
866 	case PR_CAPBSET_DROP:
867 		error = cap_prctl_drop(new, arg2);
868 		if (error < 0)
869 			goto error;
870 		goto changed;
871 
872 	/*
873 	 * The next four prctl's remain to assist with transitioning a
874 	 * system from legacy UID=0 based privilege (when filesystem
875 	 * capabilities are not in use) to a system using filesystem
876 	 * capabilities only - as the POSIX.1e draft intended.
877 	 *
878 	 * Note:
879 	 *
880 	 *  PR_SET_SECUREBITS =
881 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
882 	 *    | issecure_mask(SECURE_NOROOT)
883 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
884 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
885 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
886 	 *
887 	 * will ensure that the current process and all of its
888 	 * children will be locked into a pure
889 	 * capability-based-privilege environment.
890 	 */
891 	case PR_SET_SECUREBITS:
892 		error = -EPERM;
893 		if ((((new->securebits & SECURE_ALL_LOCKS) >> 1)
894 		     & (new->securebits ^ arg2))			/*[1]*/
895 		    || ((new->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
896 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
897 		    || (cap_capable(current_cred(),
898 				    current_cred()->user_ns, CAP_SETPCAP,
899 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
900 			/*
901 			 * [1] no changing of bits that are locked
902 			 * [2] no unlocking of locks
903 			 * [3] no setting of unsupported bits
904 			 * [4] doing anything requires privilege (go read about
905 			 *     the "sendmail capabilities bug")
906 			 */
907 		    )
908 			/* cannot change a locked bit */
909 			goto error;
910 		new->securebits = arg2;
911 		goto changed;
912 
913 	case PR_GET_SECUREBITS:
914 		error = new->securebits;
915 		goto no_change;
916 
917 	case PR_GET_KEEPCAPS:
918 		if (issecure(SECURE_KEEP_CAPS))
919 			error = 1;
920 		goto no_change;
921 
922 	case PR_SET_KEEPCAPS:
923 		error = -EINVAL;
924 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
925 			goto error;
926 		error = -EPERM;
927 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
928 			goto error;
929 		if (arg2)
930 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
931 		else
932 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
933 		goto changed;
934 
935 	default:
936 		/* No functionality available - continue with default */
937 		error = -ENOSYS;
938 		goto error;
939 	}
940 
941 	/* Functionality provided */
942 changed:
943 	return commit_creds(new);
944 
945 no_change:
946 error:
947 	abort_creds(new);
948 	return error;
949 }
950 
951 /**
952  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
953  * @mm: The VM space in which the new mapping is to be made
954  * @pages: The size of the mapping
955  *
956  * Determine whether the allocation of a new virtual mapping by the current
957  * task is permitted, returning 0 if permission is granted, -ve if not.
958  */
959 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
960 {
961 	int cap_sys_admin = 0;
962 
963 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
964 			SECURITY_CAP_NOAUDIT) == 0)
965 		cap_sys_admin = 1;
966 	return __vm_enough_memory(mm, pages, cap_sys_admin);
967 }
968 
969 /*
970  * cap_mmap_addr - check if able to map given addr
971  * @addr: address attempting to be mapped
972  *
973  * If the process is attempting to map memory below dac_mmap_min_addr they need
974  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
975  * capability security module.  Returns 0 if this mapping should be allowed
976  * -EPERM if not.
977  */
978 int cap_mmap_addr(unsigned long addr)
979 {
980 	int ret = 0;
981 
982 	if (addr < dac_mmap_min_addr) {
983 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
984 				  SECURITY_CAP_AUDIT);
985 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
986 		if (ret == 0)
987 			current->flags |= PF_SUPERPRIV;
988 	}
989 	return ret;
990 }
991 
992 int cap_mmap_file(struct file *file, unsigned long reqprot,
993 		  unsigned long prot, unsigned long flags)
994 {
995 	return 0;
996 }
997