xref: /openbmc/linux/security/commoncap.c (revision e2f1cf25)
1 /* Common capabilities, needed by capability.o.
2  *
3  *	This program is free software; you can redistribute it and/or modify
4  *	it under the terms of the GNU General Public License as published by
5  *	the Free Software Foundation; either version 2 of the License, or
6  *	(at your option) any later version.
7  *
8  */
9 
10 #include <linux/capability.h>
11 #include <linux/audit.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/lsm_hooks.h>
16 #include <linux/file.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/skbuff.h>
22 #include <linux/netlink.h>
23 #include <linux/ptrace.h>
24 #include <linux/xattr.h>
25 #include <linux/hugetlb.h>
26 #include <linux/mount.h>
27 #include <linux/sched.h>
28 #include <linux/prctl.h>
29 #include <linux/securebits.h>
30 #include <linux/user_namespace.h>
31 #include <linux/binfmts.h>
32 #include <linux/personality.h>
33 
34 /*
35  * If a non-root user executes a setuid-root binary in
36  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
37  * However if fE is also set, then the intent is for only
38  * the file capabilities to be applied, and the setuid-root
39  * bit is left on either to change the uid (plausible) or
40  * to get full privilege on a kernel without file capabilities
41  * support.  So in that case we do not raise capabilities.
42  *
43  * Warn if that happens, once per boot.
44  */
45 static void warn_setuid_and_fcaps_mixed(const char *fname)
46 {
47 	static int warned;
48 	if (!warned) {
49 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
50 			" effective capabilities. Therefore not raising all"
51 			" capabilities.\n", fname);
52 		warned = 1;
53 	}
54 }
55 
56 /**
57  * cap_capable - Determine whether a task has a particular effective capability
58  * @cred: The credentials to use
59  * @ns:  The user namespace in which we need the capability
60  * @cap: The capability to check for
61  * @audit: Whether to write an audit message or not
62  *
63  * Determine whether the nominated task has the specified capability amongst
64  * its effective set, returning 0 if it does, -ve if it does not.
65  *
66  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
67  * and has_capability() functions.  That is, it has the reverse semantics:
68  * cap_has_capability() returns 0 when a task has a capability, but the
69  * kernel's capable() and has_capability() returns 1 for this case.
70  */
71 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
72 		int cap, int audit)
73 {
74 	struct user_namespace *ns = targ_ns;
75 
76 	/* See if cred has the capability in the target user namespace
77 	 * by examining the target user namespace and all of the target
78 	 * user namespace's parents.
79 	 */
80 	for (;;) {
81 		/* Do we have the necessary capabilities? */
82 		if (ns == cred->user_ns)
83 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
84 
85 		/* Have we tried all of the parent namespaces? */
86 		if (ns == &init_user_ns)
87 			return -EPERM;
88 
89 		/*
90 		 * The owner of the user namespace in the parent of the
91 		 * user namespace has all caps.
92 		 */
93 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
94 			return 0;
95 
96 		/*
97 		 * If you have a capability in a parent user ns, then you have
98 		 * it over all children user namespaces as well.
99 		 */
100 		ns = ns->parent;
101 	}
102 
103 	/* We never get here */
104 }
105 
106 /**
107  * cap_settime - Determine whether the current process may set the system clock
108  * @ts: The time to set
109  * @tz: The timezone to set
110  *
111  * Determine whether the current process may set the system clock and timezone
112  * information, returning 0 if permission granted, -ve if denied.
113  */
114 int cap_settime(const struct timespec *ts, const struct timezone *tz)
115 {
116 	if (!capable(CAP_SYS_TIME))
117 		return -EPERM;
118 	return 0;
119 }
120 
121 /**
122  * cap_ptrace_access_check - Determine whether the current process may access
123  *			   another
124  * @child: The process to be accessed
125  * @mode: The mode of attachment.
126  *
127  * If we are in the same or an ancestor user_ns and have all the target
128  * task's capabilities, then ptrace access is allowed.
129  * If we have the ptrace capability to the target user_ns, then ptrace
130  * access is allowed.
131  * Else denied.
132  *
133  * Determine whether a process may access another, returning 0 if permission
134  * granted, -ve if denied.
135  */
136 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
137 {
138 	int ret = 0;
139 	const struct cred *cred, *child_cred;
140 
141 	rcu_read_lock();
142 	cred = current_cred();
143 	child_cred = __task_cred(child);
144 	if (cred->user_ns == child_cred->user_ns &&
145 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
146 		goto out;
147 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
148 		goto out;
149 	ret = -EPERM;
150 out:
151 	rcu_read_unlock();
152 	return ret;
153 }
154 
155 /**
156  * cap_ptrace_traceme - Determine whether another process may trace the current
157  * @parent: The task proposed to be the tracer
158  *
159  * If parent is in the same or an ancestor user_ns and has all current's
160  * capabilities, then ptrace access is allowed.
161  * If parent has the ptrace capability to current's user_ns, then ptrace
162  * access is allowed.
163  * Else denied.
164  *
165  * Determine whether the nominated task is permitted to trace the current
166  * process, returning 0 if permission is granted, -ve if denied.
167  */
168 int cap_ptrace_traceme(struct task_struct *parent)
169 {
170 	int ret = 0;
171 	const struct cred *cred, *child_cred;
172 
173 	rcu_read_lock();
174 	cred = __task_cred(parent);
175 	child_cred = current_cred();
176 	if (cred->user_ns == child_cred->user_ns &&
177 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
178 		goto out;
179 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
180 		goto out;
181 	ret = -EPERM;
182 out:
183 	rcu_read_unlock();
184 	return ret;
185 }
186 
187 /**
188  * cap_capget - Retrieve a task's capability sets
189  * @target: The task from which to retrieve the capability sets
190  * @effective: The place to record the effective set
191  * @inheritable: The place to record the inheritable set
192  * @permitted: The place to record the permitted set
193  *
194  * This function retrieves the capabilities of the nominated task and returns
195  * them to the caller.
196  */
197 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
198 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
199 {
200 	const struct cred *cred;
201 
202 	/* Derived from kernel/capability.c:sys_capget. */
203 	rcu_read_lock();
204 	cred = __task_cred(target);
205 	*effective   = cred->cap_effective;
206 	*inheritable = cred->cap_inheritable;
207 	*permitted   = cred->cap_permitted;
208 	rcu_read_unlock();
209 	return 0;
210 }
211 
212 /*
213  * Determine whether the inheritable capabilities are limited to the old
214  * permitted set.  Returns 1 if they are limited, 0 if they are not.
215  */
216 static inline int cap_inh_is_capped(void)
217 {
218 
219 	/* they are so limited unless the current task has the CAP_SETPCAP
220 	 * capability
221 	 */
222 	if (cap_capable(current_cred(), current_cred()->user_ns,
223 			CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
224 		return 0;
225 	return 1;
226 }
227 
228 /**
229  * cap_capset - Validate and apply proposed changes to current's capabilities
230  * @new: The proposed new credentials; alterations should be made here
231  * @old: The current task's current credentials
232  * @effective: A pointer to the proposed new effective capabilities set
233  * @inheritable: A pointer to the proposed new inheritable capabilities set
234  * @permitted: A pointer to the proposed new permitted capabilities set
235  *
236  * This function validates and applies a proposed mass change to the current
237  * process's capability sets.  The changes are made to the proposed new
238  * credentials, and assuming no error, will be committed by the caller of LSM.
239  */
240 int cap_capset(struct cred *new,
241 	       const struct cred *old,
242 	       const kernel_cap_t *effective,
243 	       const kernel_cap_t *inheritable,
244 	       const kernel_cap_t *permitted)
245 {
246 	if (cap_inh_is_capped() &&
247 	    !cap_issubset(*inheritable,
248 			  cap_combine(old->cap_inheritable,
249 				      old->cap_permitted)))
250 		/* incapable of using this inheritable set */
251 		return -EPERM;
252 
253 	if (!cap_issubset(*inheritable,
254 			  cap_combine(old->cap_inheritable,
255 				      old->cap_bset)))
256 		/* no new pI capabilities outside bounding set */
257 		return -EPERM;
258 
259 	/* verify restrictions on target's new Permitted set */
260 	if (!cap_issubset(*permitted, old->cap_permitted))
261 		return -EPERM;
262 
263 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
264 	if (!cap_issubset(*effective, *permitted))
265 		return -EPERM;
266 
267 	new->cap_effective   = *effective;
268 	new->cap_inheritable = *inheritable;
269 	new->cap_permitted   = *permitted;
270 	return 0;
271 }
272 
273 /*
274  * Clear proposed capability sets for execve().
275  */
276 static inline void bprm_clear_caps(struct linux_binprm *bprm)
277 {
278 	cap_clear(bprm->cred->cap_permitted);
279 	bprm->cap_effective = false;
280 }
281 
282 /**
283  * cap_inode_need_killpriv - Determine if inode change affects privileges
284  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
285  *
286  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
287  * affects the security markings on that inode, and if it is, should
288  * inode_killpriv() be invoked or the change rejected?
289  *
290  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
291  * -ve to deny the change.
292  */
293 int cap_inode_need_killpriv(struct dentry *dentry)
294 {
295 	struct inode *inode = d_backing_inode(dentry);
296 	int error;
297 
298 	if (!inode->i_op->getxattr)
299 	       return 0;
300 
301 	error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
302 	if (error <= 0)
303 		return 0;
304 	return 1;
305 }
306 
307 /**
308  * cap_inode_killpriv - Erase the security markings on an inode
309  * @dentry: The inode/dentry to alter
310  *
311  * Erase the privilege-enhancing security markings on an inode.
312  *
313  * Returns 0 if successful, -ve on error.
314  */
315 int cap_inode_killpriv(struct dentry *dentry)
316 {
317 	struct inode *inode = d_backing_inode(dentry);
318 
319 	if (!inode->i_op->removexattr)
320 	       return 0;
321 
322 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
323 }
324 
325 /*
326  * Calculate the new process capability sets from the capability sets attached
327  * to a file.
328  */
329 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
330 					  struct linux_binprm *bprm,
331 					  bool *effective,
332 					  bool *has_cap)
333 {
334 	struct cred *new = bprm->cred;
335 	unsigned i;
336 	int ret = 0;
337 
338 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
339 		*effective = true;
340 
341 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
342 		*has_cap = true;
343 
344 	CAP_FOR_EACH_U32(i) {
345 		__u32 permitted = caps->permitted.cap[i];
346 		__u32 inheritable = caps->inheritable.cap[i];
347 
348 		/*
349 		 * pP' = (X & fP) | (pI & fI)
350 		 */
351 		new->cap_permitted.cap[i] =
352 			(new->cap_bset.cap[i] & permitted) |
353 			(new->cap_inheritable.cap[i] & inheritable);
354 
355 		if (permitted & ~new->cap_permitted.cap[i])
356 			/* insufficient to execute correctly */
357 			ret = -EPERM;
358 	}
359 
360 	/*
361 	 * For legacy apps, with no internal support for recognizing they
362 	 * do not have enough capabilities, we return an error if they are
363 	 * missing some "forced" (aka file-permitted) capabilities.
364 	 */
365 	return *effective ? ret : 0;
366 }
367 
368 /*
369  * Extract the on-exec-apply capability sets for an executable file.
370  */
371 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
372 {
373 	struct inode *inode = d_backing_inode(dentry);
374 	__u32 magic_etc;
375 	unsigned tocopy, i;
376 	int size;
377 	struct vfs_cap_data caps;
378 
379 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
380 
381 	if (!inode || !inode->i_op->getxattr)
382 		return -ENODATA;
383 
384 	size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
385 				   XATTR_CAPS_SZ);
386 	if (size == -ENODATA || size == -EOPNOTSUPP)
387 		/* no data, that's ok */
388 		return -ENODATA;
389 	if (size < 0)
390 		return size;
391 
392 	if (size < sizeof(magic_etc))
393 		return -EINVAL;
394 
395 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
396 
397 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
398 	case VFS_CAP_REVISION_1:
399 		if (size != XATTR_CAPS_SZ_1)
400 			return -EINVAL;
401 		tocopy = VFS_CAP_U32_1;
402 		break;
403 	case VFS_CAP_REVISION_2:
404 		if (size != XATTR_CAPS_SZ_2)
405 			return -EINVAL;
406 		tocopy = VFS_CAP_U32_2;
407 		break;
408 	default:
409 		return -EINVAL;
410 	}
411 
412 	CAP_FOR_EACH_U32(i) {
413 		if (i >= tocopy)
414 			break;
415 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
416 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
417 	}
418 
419 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
420 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
421 
422 	return 0;
423 }
424 
425 /*
426  * Attempt to get the on-exec apply capability sets for an executable file from
427  * its xattrs and, if present, apply them to the proposed credentials being
428  * constructed by execve().
429  */
430 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap)
431 {
432 	int rc = 0;
433 	struct cpu_vfs_cap_data vcaps;
434 
435 	bprm_clear_caps(bprm);
436 
437 	if (!file_caps_enabled)
438 		return 0;
439 
440 	if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
441 		return 0;
442 
443 	rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
444 	if (rc < 0) {
445 		if (rc == -EINVAL)
446 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
447 				__func__, rc, bprm->filename);
448 		else if (rc == -ENODATA)
449 			rc = 0;
450 		goto out;
451 	}
452 
453 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
454 	if (rc == -EINVAL)
455 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
456 		       __func__, rc, bprm->filename);
457 
458 out:
459 	if (rc)
460 		bprm_clear_caps(bprm);
461 
462 	return rc;
463 }
464 
465 /**
466  * cap_bprm_set_creds - Set up the proposed credentials for execve().
467  * @bprm: The execution parameters, including the proposed creds
468  *
469  * Set up the proposed credentials for a new execution context being
470  * constructed by execve().  The proposed creds in @bprm->cred is altered,
471  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
472  */
473 int cap_bprm_set_creds(struct linux_binprm *bprm)
474 {
475 	const struct cred *old = current_cred();
476 	struct cred *new = bprm->cred;
477 	bool effective, has_cap = false;
478 	int ret;
479 	kuid_t root_uid;
480 
481 	effective = false;
482 	ret = get_file_caps(bprm, &effective, &has_cap);
483 	if (ret < 0)
484 		return ret;
485 
486 	root_uid = make_kuid(new->user_ns, 0);
487 
488 	if (!issecure(SECURE_NOROOT)) {
489 		/*
490 		 * If the legacy file capability is set, then don't set privs
491 		 * for a setuid root binary run by a non-root user.  Do set it
492 		 * for a root user just to cause least surprise to an admin.
493 		 */
494 		if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) {
495 			warn_setuid_and_fcaps_mixed(bprm->filename);
496 			goto skip;
497 		}
498 		/*
499 		 * To support inheritance of root-permissions and suid-root
500 		 * executables under compatibility mode, we override the
501 		 * capability sets for the file.
502 		 *
503 		 * If only the real uid is 0, we do not set the effective bit.
504 		 */
505 		if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) {
506 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
507 			new->cap_permitted = cap_combine(old->cap_bset,
508 							 old->cap_inheritable);
509 		}
510 		if (uid_eq(new->euid, root_uid))
511 			effective = true;
512 	}
513 skip:
514 
515 	/* if we have fs caps, clear dangerous personality flags */
516 	if (!cap_issubset(new->cap_permitted, old->cap_permitted))
517 		bprm->per_clear |= PER_CLEAR_ON_SETID;
518 
519 
520 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
521 	 * credentials unless they have the appropriate permit.
522 	 *
523 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
524 	 */
525 	if ((!uid_eq(new->euid, old->uid) ||
526 	     !gid_eq(new->egid, old->gid) ||
527 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
528 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
529 		/* downgrade; they get no more than they had, and maybe less */
530 		if (!capable(CAP_SETUID) ||
531 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
532 			new->euid = new->uid;
533 			new->egid = new->gid;
534 		}
535 		new->cap_permitted = cap_intersect(new->cap_permitted,
536 						   old->cap_permitted);
537 	}
538 
539 	new->suid = new->fsuid = new->euid;
540 	new->sgid = new->fsgid = new->egid;
541 
542 	if (effective)
543 		new->cap_effective = new->cap_permitted;
544 	else
545 		cap_clear(new->cap_effective);
546 	bprm->cap_effective = effective;
547 
548 	/*
549 	 * Audit candidate if current->cap_effective is set
550 	 *
551 	 * We do not bother to audit if 3 things are true:
552 	 *   1) cap_effective has all caps
553 	 *   2) we are root
554 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
555 	 * Since this is just a normal root execing a process.
556 	 *
557 	 * Number 1 above might fail if you don't have a full bset, but I think
558 	 * that is interesting information to audit.
559 	 */
560 	if (!cap_isclear(new->cap_effective)) {
561 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
562 		    !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) ||
563 		    issecure(SECURE_NOROOT)) {
564 			ret = audit_log_bprm_fcaps(bprm, new, old);
565 			if (ret < 0)
566 				return ret;
567 		}
568 	}
569 
570 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
571 	return 0;
572 }
573 
574 /**
575  * cap_bprm_secureexec - Determine whether a secure execution is required
576  * @bprm: The execution parameters
577  *
578  * Determine whether a secure execution is required, return 1 if it is, and 0
579  * if it is not.
580  *
581  * The credentials have been committed by this point, and so are no longer
582  * available through @bprm->cred.
583  */
584 int cap_bprm_secureexec(struct linux_binprm *bprm)
585 {
586 	const struct cred *cred = current_cred();
587 	kuid_t root_uid = make_kuid(cred->user_ns, 0);
588 
589 	if (!uid_eq(cred->uid, root_uid)) {
590 		if (bprm->cap_effective)
591 			return 1;
592 		if (!cap_isclear(cred->cap_permitted))
593 			return 1;
594 	}
595 
596 	return (!uid_eq(cred->euid, cred->uid) ||
597 		!gid_eq(cred->egid, cred->gid));
598 }
599 
600 /**
601  * cap_inode_setxattr - Determine whether an xattr may be altered
602  * @dentry: The inode/dentry being altered
603  * @name: The name of the xattr to be changed
604  * @value: The value that the xattr will be changed to
605  * @size: The size of value
606  * @flags: The replacement flag
607  *
608  * Determine whether an xattr may be altered or set on an inode, returning 0 if
609  * permission is granted, -ve if denied.
610  *
611  * This is used to make sure security xattrs don't get updated or set by those
612  * who aren't privileged to do so.
613  */
614 int cap_inode_setxattr(struct dentry *dentry, const char *name,
615 		       const void *value, size_t size, int flags)
616 {
617 	if (!strcmp(name, XATTR_NAME_CAPS)) {
618 		if (!capable(CAP_SETFCAP))
619 			return -EPERM;
620 		return 0;
621 	}
622 
623 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
624 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
625 	    !capable(CAP_SYS_ADMIN))
626 		return -EPERM;
627 	return 0;
628 }
629 
630 /**
631  * cap_inode_removexattr - Determine whether an xattr may be removed
632  * @dentry: The inode/dentry being altered
633  * @name: The name of the xattr to be changed
634  *
635  * Determine whether an xattr may be removed from an inode, returning 0 if
636  * permission is granted, -ve if denied.
637  *
638  * This is used to make sure security xattrs don't get removed by those who
639  * aren't privileged to remove them.
640  */
641 int cap_inode_removexattr(struct dentry *dentry, const char *name)
642 {
643 	if (!strcmp(name, XATTR_NAME_CAPS)) {
644 		if (!capable(CAP_SETFCAP))
645 			return -EPERM;
646 		return 0;
647 	}
648 
649 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
650 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
651 	    !capable(CAP_SYS_ADMIN))
652 		return -EPERM;
653 	return 0;
654 }
655 
656 /*
657  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
658  * a process after a call to setuid, setreuid, or setresuid.
659  *
660  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
661  *  {r,e,s}uid != 0, the permitted and effective capabilities are
662  *  cleared.
663  *
664  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
665  *  capabilities of the process are cleared.
666  *
667  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
668  *  capabilities are set to the permitted capabilities.
669  *
670  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
671  *  never happen.
672  *
673  *  -astor
674  *
675  * cevans - New behaviour, Oct '99
676  * A process may, via prctl(), elect to keep its capabilities when it
677  * calls setuid() and switches away from uid==0. Both permitted and
678  * effective sets will be retained.
679  * Without this change, it was impossible for a daemon to drop only some
680  * of its privilege. The call to setuid(!=0) would drop all privileges!
681  * Keeping uid 0 is not an option because uid 0 owns too many vital
682  * files..
683  * Thanks to Olaf Kirch and Peter Benie for spotting this.
684  */
685 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
686 {
687 	kuid_t root_uid = make_kuid(old->user_ns, 0);
688 
689 	if ((uid_eq(old->uid, root_uid) ||
690 	     uid_eq(old->euid, root_uid) ||
691 	     uid_eq(old->suid, root_uid)) &&
692 	    (!uid_eq(new->uid, root_uid) &&
693 	     !uid_eq(new->euid, root_uid) &&
694 	     !uid_eq(new->suid, root_uid)) &&
695 	    !issecure(SECURE_KEEP_CAPS)) {
696 		cap_clear(new->cap_permitted);
697 		cap_clear(new->cap_effective);
698 	}
699 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
700 		cap_clear(new->cap_effective);
701 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
702 		new->cap_effective = new->cap_permitted;
703 }
704 
705 /**
706  * cap_task_fix_setuid - Fix up the results of setuid() call
707  * @new: The proposed credentials
708  * @old: The current task's current credentials
709  * @flags: Indications of what has changed
710  *
711  * Fix up the results of setuid() call before the credential changes are
712  * actually applied, returning 0 to grant the changes, -ve to deny them.
713  */
714 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
715 {
716 	switch (flags) {
717 	case LSM_SETID_RE:
718 	case LSM_SETID_ID:
719 	case LSM_SETID_RES:
720 		/* juggle the capabilities to follow [RES]UID changes unless
721 		 * otherwise suppressed */
722 		if (!issecure(SECURE_NO_SETUID_FIXUP))
723 			cap_emulate_setxuid(new, old);
724 		break;
725 
726 	case LSM_SETID_FS:
727 		/* juggle the capabilties to follow FSUID changes, unless
728 		 * otherwise suppressed
729 		 *
730 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
731 		 *          if not, we might be a bit too harsh here.
732 		 */
733 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
734 			kuid_t root_uid = make_kuid(old->user_ns, 0);
735 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
736 				new->cap_effective =
737 					cap_drop_fs_set(new->cap_effective);
738 
739 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
740 				new->cap_effective =
741 					cap_raise_fs_set(new->cap_effective,
742 							 new->cap_permitted);
743 		}
744 		break;
745 
746 	default:
747 		return -EINVAL;
748 	}
749 
750 	return 0;
751 }
752 
753 /*
754  * Rationale: code calling task_setscheduler, task_setioprio, and
755  * task_setnice, assumes that
756  *   . if capable(cap_sys_nice), then those actions should be allowed
757  *   . if not capable(cap_sys_nice), but acting on your own processes,
758  *   	then those actions should be allowed
759  * This is insufficient now since you can call code without suid, but
760  * yet with increased caps.
761  * So we check for increased caps on the target process.
762  */
763 static int cap_safe_nice(struct task_struct *p)
764 {
765 	int is_subset, ret = 0;
766 
767 	rcu_read_lock();
768 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
769 				 current_cred()->cap_permitted);
770 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
771 		ret = -EPERM;
772 	rcu_read_unlock();
773 
774 	return ret;
775 }
776 
777 /**
778  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
779  * @p: The task to affect
780  *
781  * Detemine if the requested scheduler policy change is permitted for the
782  * specified task, returning 0 if permission is granted, -ve if denied.
783  */
784 int cap_task_setscheduler(struct task_struct *p)
785 {
786 	return cap_safe_nice(p);
787 }
788 
789 /**
790  * cap_task_ioprio - Detemine if I/O priority change is permitted
791  * @p: The task to affect
792  * @ioprio: The I/O priority to set
793  *
794  * Detemine if the requested I/O priority change is permitted for the specified
795  * task, returning 0 if permission is granted, -ve if denied.
796  */
797 int cap_task_setioprio(struct task_struct *p, int ioprio)
798 {
799 	return cap_safe_nice(p);
800 }
801 
802 /**
803  * cap_task_ioprio - Detemine if task priority change is permitted
804  * @p: The task to affect
805  * @nice: The nice value to set
806  *
807  * Detemine if the requested task priority change is permitted for the
808  * specified task, returning 0 if permission is granted, -ve if denied.
809  */
810 int cap_task_setnice(struct task_struct *p, int nice)
811 {
812 	return cap_safe_nice(p);
813 }
814 
815 /*
816  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
817  * the current task's bounding set.  Returns 0 on success, -ve on error.
818  */
819 static int cap_prctl_drop(unsigned long cap)
820 {
821 	struct cred *new;
822 
823 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
824 		return -EPERM;
825 	if (!cap_valid(cap))
826 		return -EINVAL;
827 
828 	new = prepare_creds();
829 	if (!new)
830 		return -ENOMEM;
831 	cap_lower(new->cap_bset, cap);
832 	return commit_creds(new);
833 }
834 
835 /**
836  * cap_task_prctl - Implement process control functions for this security module
837  * @option: The process control function requested
838  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
839  *
840  * Allow process control functions (sys_prctl()) to alter capabilities; may
841  * also deny access to other functions not otherwise implemented here.
842  *
843  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
844  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
845  * modules will consider performing the function.
846  */
847 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
848 		   unsigned long arg4, unsigned long arg5)
849 {
850 	const struct cred *old = current_cred();
851 	struct cred *new;
852 
853 	switch (option) {
854 	case PR_CAPBSET_READ:
855 		if (!cap_valid(arg2))
856 			return -EINVAL;
857 		return !!cap_raised(old->cap_bset, arg2);
858 
859 	case PR_CAPBSET_DROP:
860 		return cap_prctl_drop(arg2);
861 
862 	/*
863 	 * The next four prctl's remain to assist with transitioning a
864 	 * system from legacy UID=0 based privilege (when filesystem
865 	 * capabilities are not in use) to a system using filesystem
866 	 * capabilities only - as the POSIX.1e draft intended.
867 	 *
868 	 * Note:
869 	 *
870 	 *  PR_SET_SECUREBITS =
871 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
872 	 *    | issecure_mask(SECURE_NOROOT)
873 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
874 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
875 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
876 	 *
877 	 * will ensure that the current process and all of its
878 	 * children will be locked into a pure
879 	 * capability-based-privilege environment.
880 	 */
881 	case PR_SET_SECUREBITS:
882 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
883 		     & (old->securebits ^ arg2))			/*[1]*/
884 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
885 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
886 		    || (cap_capable(current_cred(),
887 				    current_cred()->user_ns, CAP_SETPCAP,
888 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
889 			/*
890 			 * [1] no changing of bits that are locked
891 			 * [2] no unlocking of locks
892 			 * [3] no setting of unsupported bits
893 			 * [4] doing anything requires privilege (go read about
894 			 *     the "sendmail capabilities bug")
895 			 */
896 		    )
897 			/* cannot change a locked bit */
898 			return -EPERM;
899 
900 		new = prepare_creds();
901 		if (!new)
902 			return -ENOMEM;
903 		new->securebits = arg2;
904 		return commit_creds(new);
905 
906 	case PR_GET_SECUREBITS:
907 		return old->securebits;
908 
909 	case PR_GET_KEEPCAPS:
910 		return !!issecure(SECURE_KEEP_CAPS);
911 
912 	case PR_SET_KEEPCAPS:
913 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
914 			return -EINVAL;
915 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
916 			return -EPERM;
917 
918 		new = prepare_creds();
919 		if (!new)
920 			return -ENOMEM;
921 		if (arg2)
922 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
923 		else
924 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
925 		return commit_creds(new);
926 
927 	default:
928 		/* No functionality available - continue with default */
929 		return -ENOSYS;
930 	}
931 }
932 
933 /**
934  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
935  * @mm: The VM space in which the new mapping is to be made
936  * @pages: The size of the mapping
937  *
938  * Determine whether the allocation of a new virtual mapping by the current
939  * task is permitted, returning 1 if permission is granted, 0 if not.
940  */
941 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
942 {
943 	int cap_sys_admin = 0;
944 
945 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
946 			SECURITY_CAP_NOAUDIT) == 0)
947 		cap_sys_admin = 1;
948 	return cap_sys_admin;
949 }
950 
951 /*
952  * cap_mmap_addr - check if able to map given addr
953  * @addr: address attempting to be mapped
954  *
955  * If the process is attempting to map memory below dac_mmap_min_addr they need
956  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
957  * capability security module.  Returns 0 if this mapping should be allowed
958  * -EPERM if not.
959  */
960 int cap_mmap_addr(unsigned long addr)
961 {
962 	int ret = 0;
963 
964 	if (addr < dac_mmap_min_addr) {
965 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
966 				  SECURITY_CAP_AUDIT);
967 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
968 		if (ret == 0)
969 			current->flags |= PF_SUPERPRIV;
970 	}
971 	return ret;
972 }
973 
974 int cap_mmap_file(struct file *file, unsigned long reqprot,
975 		  unsigned long prot, unsigned long flags)
976 {
977 	return 0;
978 }
979 
980 #ifdef CONFIG_SECURITY
981 
982 struct security_hook_list capability_hooks[] = {
983 	LSM_HOOK_INIT(capable, cap_capable),
984 	LSM_HOOK_INIT(settime, cap_settime),
985 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
986 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
987 	LSM_HOOK_INIT(capget, cap_capget),
988 	LSM_HOOK_INIT(capset, cap_capset),
989 	LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
990 	LSM_HOOK_INIT(bprm_secureexec, cap_bprm_secureexec),
991 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
992 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
993 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
994 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
995 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
996 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
997 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
998 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
999 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1000 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1001 };
1002 
1003 void __init capability_add_hooks(void)
1004 {
1005 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks));
1006 }
1007 
1008 #endif /* CONFIG_SECURITY */
1009