xref: /linux/security/commoncap.c (revision 12871a0bd67dd4db4418e1daafcd46e9d329ef10)
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 
32 /*
33  * If a non-root user executes a setuid-root binary in
34  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
35  * However if fE is also set, then the intent is for only
36  * the file capabilities to be applied, and the setuid-root
37  * bit is left on either to change the uid (plausible) or
38  * to get full privilege on a kernel without file capabilities
39  * support.  So in that case we do not raise capabilities.
40  *
41  * Warn if that happens, once per boot.
42  */
43 static void warn_setuid_and_fcaps_mixed(const char *fname)
44 {
45 	static int warned;
46 	if (!warned) {
47 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
48 			" effective capabilities. Therefore not raising all"
49 			" capabilities.\n", fname);
50 		warned = 1;
51 	}
52 }
53 
54 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
55 {
56 	return 0;
57 }
58 
59 int cap_netlink_recv(struct sk_buff *skb, int cap)
60 {
61 	if (!cap_raised(current_cap(), cap))
62 		return -EPERM;
63 	return 0;
64 }
65 EXPORT_SYMBOL(cap_netlink_recv);
66 
67 /**
68  * cap_capable - Determine whether a task has a particular effective capability
69  * @tsk: The task to query
70  * @cred: The credentials to use
71  * @ns:  The user namespace in which we need the capability
72  * @cap: The capability to check for
73  * @audit: Whether to write an audit message or not
74  *
75  * Determine whether the nominated task has the specified capability amongst
76  * its effective set, returning 0 if it does, -ve if it does not.
77  *
78  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
79  * and has_capability() functions.  That is, it has the reverse semantics:
80  * cap_has_capability() returns 0 when a task has a capability, but the
81  * kernel's capable() and has_capability() returns 1 for this case.
82  */
83 int cap_capable(struct task_struct *tsk, const struct cred *cred,
84 		struct user_namespace *targ_ns, int cap, int audit)
85 {
86 	for (;;) {
87 		/* The creator of the user namespace has all caps. */
88 		if (targ_ns != &init_user_ns && targ_ns->creator == cred->user)
89 			return 0;
90 
91 		/* Do we have the necessary capabilities? */
92 		if (targ_ns == cred->user->user_ns)
93 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
94 
95 		/* Have we tried all of the parent namespaces? */
96 		if (targ_ns == &init_user_ns)
97 			return -EPERM;
98 
99 		/*
100 		 *If you have a capability in a parent user ns, then you have
101 		 * it over all children user namespaces as well.
102 		 */
103 		targ_ns = targ_ns->creator->user_ns;
104 	}
105 
106 	/* We never get here */
107 }
108 
109 /**
110  * cap_settime - Determine whether the current process may set the system clock
111  * @ts: The time to set
112  * @tz: The timezone to set
113  *
114  * Determine whether the current process may set the system clock and timezone
115  * information, returning 0 if permission granted, -ve if denied.
116  */
117 int cap_settime(const struct timespec *ts, const struct timezone *tz)
118 {
119 	if (!capable(CAP_SYS_TIME))
120 		return -EPERM;
121 	return 0;
122 }
123 
124 /**
125  * cap_ptrace_access_check - Determine whether the current process may access
126  *			   another
127  * @child: The process to be accessed
128  * @mode: The mode of attachment.
129  *
130  * If we are in the same or an ancestor user_ns and have all the target
131  * task's capabilities, then ptrace access is allowed.
132  * If we have the ptrace capability to the target user_ns, then ptrace
133  * access is allowed.
134  * Else denied.
135  *
136  * Determine whether a process may access another, returning 0 if permission
137  * granted, -ve if denied.
138  */
139 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
140 {
141 	int ret = 0;
142 	const struct cred *cred, *child_cred;
143 
144 	rcu_read_lock();
145 	cred = current_cred();
146 	child_cred = __task_cred(child);
147 	if (cred->user->user_ns == child_cred->user->user_ns &&
148 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
149 		goto out;
150 	if (ns_capable(child_cred->user->user_ns, CAP_SYS_PTRACE))
151 		goto out;
152 	ret = -EPERM;
153 out:
154 	rcu_read_unlock();
155 	return ret;
156 }
157 
158 /**
159  * cap_ptrace_traceme - Determine whether another process may trace the current
160  * @parent: The task proposed to be the tracer
161  *
162  * If parent is in the same or an ancestor user_ns and has all current's
163  * capabilities, then ptrace access is allowed.
164  * If parent has the ptrace capability to current's user_ns, then ptrace
165  * access is allowed.
166  * Else denied.
167  *
168  * Determine whether the nominated task is permitted to trace the current
169  * process, returning 0 if permission is granted, -ve if denied.
170  */
171 int cap_ptrace_traceme(struct task_struct *parent)
172 {
173 	int ret = 0;
174 	const struct cred *cred, *child_cred;
175 
176 	rcu_read_lock();
177 	cred = __task_cred(parent);
178 	child_cred = current_cred();
179 	if (cred->user->user_ns == child_cred->user->user_ns &&
180 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
181 		goto out;
182 	if (has_ns_capability(parent, child_cred->user->user_ns, CAP_SYS_PTRACE))
183 		goto out;
184 	ret = -EPERM;
185 out:
186 	rcu_read_unlock();
187 	return ret;
188 }
189 
190 /**
191  * cap_capget - Retrieve a task's capability sets
192  * @target: The task from which to retrieve the capability sets
193  * @effective: The place to record the effective set
194  * @inheritable: The place to record the inheritable set
195  * @permitted: The place to record the permitted set
196  *
197  * This function retrieves the capabilities of the nominated task and returns
198  * them to the caller.
199  */
200 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
201 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
202 {
203 	const struct cred *cred;
204 
205 	/* Derived from kernel/capability.c:sys_capget. */
206 	rcu_read_lock();
207 	cred = __task_cred(target);
208 	*effective   = cred->cap_effective;
209 	*inheritable = cred->cap_inheritable;
210 	*permitted   = cred->cap_permitted;
211 	rcu_read_unlock();
212 	return 0;
213 }
214 
215 /*
216  * Determine whether the inheritable capabilities are limited to the old
217  * permitted set.  Returns 1 if they are limited, 0 if they are not.
218  */
219 static inline int cap_inh_is_capped(void)
220 {
221 
222 	/* they are so limited unless the current task has the CAP_SETPCAP
223 	 * capability
224 	 */
225 	if (cap_capable(current, current_cred(),
226 			current_cred()->user->user_ns, CAP_SETPCAP,
227 			SECURITY_CAP_AUDIT) == 0)
228 		return 0;
229 	return 1;
230 }
231 
232 /**
233  * cap_capset - Validate and apply proposed changes to current's capabilities
234  * @new: The proposed new credentials; alterations should be made here
235  * @old: The current task's current credentials
236  * @effective: A pointer to the proposed new effective capabilities set
237  * @inheritable: A pointer to the proposed new inheritable capabilities set
238  * @permitted: A pointer to the proposed new permitted capabilities set
239  *
240  * This function validates and applies a proposed mass change to the current
241  * process's capability sets.  The changes are made to the proposed new
242  * credentials, and assuming no error, will be committed by the caller of LSM.
243  */
244 int cap_capset(struct cred *new,
245 	       const struct cred *old,
246 	       const kernel_cap_t *effective,
247 	       const kernel_cap_t *inheritable,
248 	       const kernel_cap_t *permitted)
249 {
250 	if (cap_inh_is_capped() &&
251 	    !cap_issubset(*inheritable,
252 			  cap_combine(old->cap_inheritable,
253 				      old->cap_permitted)))
254 		/* incapable of using this inheritable set */
255 		return -EPERM;
256 
257 	if (!cap_issubset(*inheritable,
258 			  cap_combine(old->cap_inheritable,
259 				      old->cap_bset)))
260 		/* no new pI capabilities outside bounding set */
261 		return -EPERM;
262 
263 	/* verify restrictions on target's new Permitted set */
264 	if (!cap_issubset(*permitted, old->cap_permitted))
265 		return -EPERM;
266 
267 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
268 	if (!cap_issubset(*effective, *permitted))
269 		return -EPERM;
270 
271 	new->cap_effective   = *effective;
272 	new->cap_inheritable = *inheritable;
273 	new->cap_permitted   = *permitted;
274 	return 0;
275 }
276 
277 /*
278  * Clear proposed capability sets for execve().
279  */
280 static inline void bprm_clear_caps(struct linux_binprm *bprm)
281 {
282 	cap_clear(bprm->cred->cap_permitted);
283 	bprm->cap_effective = false;
284 }
285 
286 /**
287  * cap_inode_need_killpriv - Determine if inode change affects privileges
288  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
289  *
290  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
291  * affects the security markings on that inode, and if it is, should
292  * inode_killpriv() be invoked or the change rejected?
293  *
294  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
295  * -ve to deny the change.
296  */
297 int cap_inode_need_killpriv(struct dentry *dentry)
298 {
299 	struct inode *inode = dentry->d_inode;
300 	int error;
301 
302 	if (!inode->i_op->getxattr)
303 	       return 0;
304 
305 	error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
306 	if (error <= 0)
307 		return 0;
308 	return 1;
309 }
310 
311 /**
312  * cap_inode_killpriv - Erase the security markings on an inode
313  * @dentry: The inode/dentry to alter
314  *
315  * Erase the privilege-enhancing security markings on an inode.
316  *
317  * Returns 0 if successful, -ve on error.
318  */
319 int cap_inode_killpriv(struct dentry *dentry)
320 {
321 	struct inode *inode = dentry->d_inode;
322 
323 	if (!inode->i_op->removexattr)
324 	       return 0;
325 
326 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
327 }
328 
329 /*
330  * Calculate the new process capability sets from the capability sets attached
331  * to a file.
332  */
333 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
334 					  struct linux_binprm *bprm,
335 					  bool *effective)
336 {
337 	struct cred *new = bprm->cred;
338 	unsigned i;
339 	int ret = 0;
340 
341 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
342 		*effective = 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 = dentry->d_inode;
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 	return 0;
420 }
421 
422 /*
423  * Attempt to get the on-exec apply capability sets for an executable file from
424  * its xattrs and, if present, apply them to the proposed credentials being
425  * constructed by execve().
426  */
427 static int get_file_caps(struct linux_binprm *bprm, bool *effective)
428 {
429 	struct dentry *dentry;
430 	int rc = 0;
431 	struct cpu_vfs_cap_data vcaps;
432 
433 	bprm_clear_caps(bprm);
434 
435 	if (!file_caps_enabled)
436 		return 0;
437 
438 	if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)
439 		return 0;
440 
441 	dentry = dget(bprm->file->f_dentry);
442 
443 	rc = get_vfs_caps_from_disk(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);
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 	dput(dentry);
460 	if (rc)
461 		bprm_clear_caps(bprm);
462 
463 	return rc;
464 }
465 
466 /**
467  * cap_bprm_set_creds - Set up the proposed credentials for execve().
468  * @bprm: The execution parameters, including the proposed creds
469  *
470  * Set up the proposed credentials for a new execution context being
471  * constructed by execve().  The proposed creds in @bprm->cred is altered,
472  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
473  */
474 int cap_bprm_set_creds(struct linux_binprm *bprm)
475 {
476 	const struct cred *old = current_cred();
477 	struct cred *new = bprm->cred;
478 	bool effective;
479 	int ret;
480 
481 	effective = false;
482 	ret = get_file_caps(bprm, &effective);
483 	if (ret < 0)
484 		return ret;
485 
486 	if (!issecure(SECURE_NOROOT)) {
487 		/*
488 		 * If the legacy file capability is set, then don't set privs
489 		 * for a setuid root binary run by a non-root user.  Do set it
490 		 * for a root user just to cause least surprise to an admin.
491 		 */
492 		if (effective && new->uid != 0 && new->euid == 0) {
493 			warn_setuid_and_fcaps_mixed(bprm->filename);
494 			goto skip;
495 		}
496 		/*
497 		 * To support inheritance of root-permissions and suid-root
498 		 * executables under compatibility mode, we override the
499 		 * capability sets for the file.
500 		 *
501 		 * If only the real uid is 0, we do not set the effective bit.
502 		 */
503 		if (new->euid == 0 || new->uid == 0) {
504 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
505 			new->cap_permitted = cap_combine(old->cap_bset,
506 							 old->cap_inheritable);
507 		}
508 		if (new->euid == 0)
509 			effective = true;
510 	}
511 skip:
512 
513 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
514 	 * credentials unless they have the appropriate permit
515 	 */
516 	if ((new->euid != old->uid ||
517 	     new->egid != old->gid ||
518 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
519 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
520 		/* downgrade; they get no more than they had, and maybe less */
521 		if (!capable(CAP_SETUID)) {
522 			new->euid = new->uid;
523 			new->egid = new->gid;
524 		}
525 		new->cap_permitted = cap_intersect(new->cap_permitted,
526 						   old->cap_permitted);
527 	}
528 
529 	new->suid = new->fsuid = new->euid;
530 	new->sgid = new->fsgid = new->egid;
531 
532 	if (effective)
533 		new->cap_effective = new->cap_permitted;
534 	else
535 		cap_clear(new->cap_effective);
536 	bprm->cap_effective = effective;
537 
538 	/*
539 	 * Audit candidate if current->cap_effective is set
540 	 *
541 	 * We do not bother to audit if 3 things are true:
542 	 *   1) cap_effective has all caps
543 	 *   2) we are root
544 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
545 	 * Since this is just a normal root execing a process.
546 	 *
547 	 * Number 1 above might fail if you don't have a full bset, but I think
548 	 * that is interesting information to audit.
549 	 */
550 	if (!cap_isclear(new->cap_effective)) {
551 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
552 		    new->euid != 0 || new->uid != 0 ||
553 		    issecure(SECURE_NOROOT)) {
554 			ret = audit_log_bprm_fcaps(bprm, new, old);
555 			if (ret < 0)
556 				return ret;
557 		}
558 	}
559 
560 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
561 	return 0;
562 }
563 
564 /**
565  * cap_bprm_secureexec - Determine whether a secure execution is required
566  * @bprm: The execution parameters
567  *
568  * Determine whether a secure execution is required, return 1 if it is, and 0
569  * if it is not.
570  *
571  * The credentials have been committed by this point, and so are no longer
572  * available through @bprm->cred.
573  */
574 int cap_bprm_secureexec(struct linux_binprm *bprm)
575 {
576 	const struct cred *cred = current_cred();
577 
578 	if (cred->uid != 0) {
579 		if (bprm->cap_effective)
580 			return 1;
581 		if (!cap_isclear(cred->cap_permitted))
582 			return 1;
583 	}
584 
585 	return (cred->euid != cred->uid ||
586 		cred->egid != cred->gid);
587 }
588 
589 /**
590  * cap_inode_setxattr - Determine whether an xattr may be altered
591  * @dentry: The inode/dentry being altered
592  * @name: The name of the xattr to be changed
593  * @value: The value that the xattr will be changed to
594  * @size: The size of value
595  * @flags: The replacement flag
596  *
597  * Determine whether an xattr may be altered or set on an inode, returning 0 if
598  * permission is granted, -ve if denied.
599  *
600  * This is used to make sure security xattrs don't get updated or set by those
601  * who aren't privileged to do so.
602  */
603 int cap_inode_setxattr(struct dentry *dentry, const char *name,
604 		       const void *value, size_t size, int flags)
605 {
606 	if (!strcmp(name, XATTR_NAME_CAPS)) {
607 		if (!capable(CAP_SETFCAP))
608 			return -EPERM;
609 		return 0;
610 	}
611 
612 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
613 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
614 	    !capable(CAP_SYS_ADMIN))
615 		return -EPERM;
616 	return 0;
617 }
618 
619 /**
620  * cap_inode_removexattr - Determine whether an xattr may be removed
621  * @dentry: The inode/dentry being altered
622  * @name: The name of the xattr to be changed
623  *
624  * Determine whether an xattr may be removed from an inode, returning 0 if
625  * permission is granted, -ve if denied.
626  *
627  * This is used to make sure security xattrs don't get removed by those who
628  * aren't privileged to remove them.
629  */
630 int cap_inode_removexattr(struct dentry *dentry, const char *name)
631 {
632 	if (!strcmp(name, XATTR_NAME_CAPS)) {
633 		if (!capable(CAP_SETFCAP))
634 			return -EPERM;
635 		return 0;
636 	}
637 
638 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
639 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
640 	    !capable(CAP_SYS_ADMIN))
641 		return -EPERM;
642 	return 0;
643 }
644 
645 /*
646  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
647  * a process after a call to setuid, setreuid, or setresuid.
648  *
649  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
650  *  {r,e,s}uid != 0, the permitted and effective capabilities are
651  *  cleared.
652  *
653  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
654  *  capabilities of the process are cleared.
655  *
656  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
657  *  capabilities are set to the permitted capabilities.
658  *
659  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
660  *  never happen.
661  *
662  *  -astor
663  *
664  * cevans - New behaviour, Oct '99
665  * A process may, via prctl(), elect to keep its capabilities when it
666  * calls setuid() and switches away from uid==0. Both permitted and
667  * effective sets will be retained.
668  * Without this change, it was impossible for a daemon to drop only some
669  * of its privilege. The call to setuid(!=0) would drop all privileges!
670  * Keeping uid 0 is not an option because uid 0 owns too many vital
671  * files..
672  * Thanks to Olaf Kirch and Peter Benie for spotting this.
673  */
674 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
675 {
676 	if ((old->uid == 0 || old->euid == 0 || old->suid == 0) &&
677 	    (new->uid != 0 && new->euid != 0 && new->suid != 0) &&
678 	    !issecure(SECURE_KEEP_CAPS)) {
679 		cap_clear(new->cap_permitted);
680 		cap_clear(new->cap_effective);
681 	}
682 	if (old->euid == 0 && new->euid != 0)
683 		cap_clear(new->cap_effective);
684 	if (old->euid != 0 && new->euid == 0)
685 		new->cap_effective = new->cap_permitted;
686 }
687 
688 /**
689  * cap_task_fix_setuid - Fix up the results of setuid() call
690  * @new: The proposed credentials
691  * @old: The current task's current credentials
692  * @flags: Indications of what has changed
693  *
694  * Fix up the results of setuid() call before the credential changes are
695  * actually applied, returning 0 to grant the changes, -ve to deny them.
696  */
697 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
698 {
699 	switch (flags) {
700 	case LSM_SETID_RE:
701 	case LSM_SETID_ID:
702 	case LSM_SETID_RES:
703 		/* juggle the capabilities to follow [RES]UID changes unless
704 		 * otherwise suppressed */
705 		if (!issecure(SECURE_NO_SETUID_FIXUP))
706 			cap_emulate_setxuid(new, old);
707 		break;
708 
709 	case LSM_SETID_FS:
710 		/* juggle the capabilties to follow FSUID changes, unless
711 		 * otherwise suppressed
712 		 *
713 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
714 		 *          if not, we might be a bit too harsh here.
715 		 */
716 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
717 			if (old->fsuid == 0 && new->fsuid != 0)
718 				new->cap_effective =
719 					cap_drop_fs_set(new->cap_effective);
720 
721 			if (old->fsuid != 0 && new->fsuid == 0)
722 				new->cap_effective =
723 					cap_raise_fs_set(new->cap_effective,
724 							 new->cap_permitted);
725 		}
726 		break;
727 
728 	default:
729 		return -EINVAL;
730 	}
731 
732 	return 0;
733 }
734 
735 /*
736  * Rationale: code calling task_setscheduler, task_setioprio, and
737  * task_setnice, assumes that
738  *   . if capable(cap_sys_nice), then those actions should be allowed
739  *   . if not capable(cap_sys_nice), but acting on your own processes,
740  *   	then those actions should be allowed
741  * This is insufficient now since you can call code without suid, but
742  * yet with increased caps.
743  * So we check for increased caps on the target process.
744  */
745 static int cap_safe_nice(struct task_struct *p)
746 {
747 	int is_subset;
748 
749 	rcu_read_lock();
750 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
751 				 current_cred()->cap_permitted);
752 	rcu_read_unlock();
753 
754 	if (!is_subset && !capable(CAP_SYS_NICE))
755 		return -EPERM;
756 	return 0;
757 }
758 
759 /**
760  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
761  * @p: The task to affect
762  *
763  * Detemine if the requested scheduler policy change is permitted for the
764  * specified task, returning 0 if permission is granted, -ve if denied.
765  */
766 int cap_task_setscheduler(struct task_struct *p)
767 {
768 	return cap_safe_nice(p);
769 }
770 
771 /**
772  * cap_task_ioprio - Detemine if I/O priority change is permitted
773  * @p: The task to affect
774  * @ioprio: The I/O priority to set
775  *
776  * Detemine if the requested I/O priority change is permitted for the specified
777  * task, returning 0 if permission is granted, -ve if denied.
778  */
779 int cap_task_setioprio(struct task_struct *p, int ioprio)
780 {
781 	return cap_safe_nice(p);
782 }
783 
784 /**
785  * cap_task_ioprio - Detemine if task priority change is permitted
786  * @p: The task to affect
787  * @nice: The nice value to set
788  *
789  * Detemine if the requested task priority change is permitted for the
790  * specified task, returning 0 if permission is granted, -ve if denied.
791  */
792 int cap_task_setnice(struct task_struct *p, int nice)
793 {
794 	return cap_safe_nice(p);
795 }
796 
797 /*
798  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
799  * the current task's bounding set.  Returns 0 on success, -ve on error.
800  */
801 static long cap_prctl_drop(struct cred *new, unsigned long cap)
802 {
803 	if (!capable(CAP_SETPCAP))
804 		return -EPERM;
805 	if (!cap_valid(cap))
806 		return -EINVAL;
807 
808 	cap_lower(new->cap_bset, cap);
809 	return 0;
810 }
811 
812 /**
813  * cap_task_prctl - Implement process control functions for this security module
814  * @option: The process control function requested
815  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
816  *
817  * Allow process control functions (sys_prctl()) to alter capabilities; may
818  * also deny access to other functions not otherwise implemented here.
819  *
820  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
821  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
822  * modules will consider performing the function.
823  */
824 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
825 		   unsigned long arg4, unsigned long arg5)
826 {
827 	struct cred *new;
828 	long error = 0;
829 
830 	new = prepare_creds();
831 	if (!new)
832 		return -ENOMEM;
833 
834 	switch (option) {
835 	case PR_CAPBSET_READ:
836 		error = -EINVAL;
837 		if (!cap_valid(arg2))
838 			goto error;
839 		error = !!cap_raised(new->cap_bset, arg2);
840 		goto no_change;
841 
842 	case PR_CAPBSET_DROP:
843 		error = cap_prctl_drop(new, arg2);
844 		if (error < 0)
845 			goto error;
846 		goto changed;
847 
848 	/*
849 	 * The next four prctl's remain to assist with transitioning a
850 	 * system from legacy UID=0 based privilege (when filesystem
851 	 * capabilities are not in use) to a system using filesystem
852 	 * capabilities only - as the POSIX.1e draft intended.
853 	 *
854 	 * Note:
855 	 *
856 	 *  PR_SET_SECUREBITS =
857 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
858 	 *    | issecure_mask(SECURE_NOROOT)
859 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
860 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
861 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
862 	 *
863 	 * will ensure that the current process and all of its
864 	 * children will be locked into a pure
865 	 * capability-based-privilege environment.
866 	 */
867 	case PR_SET_SECUREBITS:
868 		error = -EPERM;
869 		if ((((new->securebits & SECURE_ALL_LOCKS) >> 1)
870 		     & (new->securebits ^ arg2))			/*[1]*/
871 		    || ((new->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
872 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
873 		    || (cap_capable(current, current_cred(),
874 				    current_cred()->user->user_ns, CAP_SETPCAP,
875 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
876 			/*
877 			 * [1] no changing of bits that are locked
878 			 * [2] no unlocking of locks
879 			 * [3] no setting of unsupported bits
880 			 * [4] doing anything requires privilege (go read about
881 			 *     the "sendmail capabilities bug")
882 			 */
883 		    )
884 			/* cannot change a locked bit */
885 			goto error;
886 		new->securebits = arg2;
887 		goto changed;
888 
889 	case PR_GET_SECUREBITS:
890 		error = new->securebits;
891 		goto no_change;
892 
893 	case PR_GET_KEEPCAPS:
894 		if (issecure(SECURE_KEEP_CAPS))
895 			error = 1;
896 		goto no_change;
897 
898 	case PR_SET_KEEPCAPS:
899 		error = -EINVAL;
900 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
901 			goto error;
902 		error = -EPERM;
903 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
904 			goto error;
905 		if (arg2)
906 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
907 		else
908 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
909 		goto changed;
910 
911 	default:
912 		/* No functionality available - continue with default */
913 		error = -ENOSYS;
914 		goto error;
915 	}
916 
917 	/* Functionality provided */
918 changed:
919 	return commit_creds(new);
920 
921 no_change:
922 error:
923 	abort_creds(new);
924 	return error;
925 }
926 
927 /**
928  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
929  * @mm: The VM space in which the new mapping is to be made
930  * @pages: The size of the mapping
931  *
932  * Determine whether the allocation of a new virtual mapping by the current
933  * task is permitted, returning 0 if permission is granted, -ve if not.
934  */
935 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
936 {
937 	int cap_sys_admin = 0;
938 
939 	if (cap_capable(current, current_cred(), &init_user_ns, CAP_SYS_ADMIN,
940 			SECURITY_CAP_NOAUDIT) == 0)
941 		cap_sys_admin = 1;
942 	return __vm_enough_memory(mm, pages, cap_sys_admin);
943 }
944 
945 /*
946  * cap_file_mmap - check if able to map given addr
947  * @file: unused
948  * @reqprot: unused
949  * @prot: unused
950  * @flags: unused
951  * @addr: address attempting to be mapped
952  * @addr_only: unused
953  *
954  * If the process is attempting to map memory below dac_mmap_min_addr they need
955  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
956  * capability security module.  Returns 0 if this mapping should be allowed
957  * -EPERM if not.
958  */
959 int cap_file_mmap(struct file *file, unsigned long reqprot,
960 		  unsigned long prot, unsigned long flags,
961 		  unsigned long addr, unsigned long addr_only)
962 {
963 	int ret = 0;
964 
965 	if (addr < dac_mmap_min_addr) {
966 		ret = cap_capable(current, current_cred(), &init_user_ns, CAP_SYS_RAWIO,
967 				  SECURITY_CAP_AUDIT);
968 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
969 		if (ret == 0)
970 			current->flags |= PF_SUPERPRIV;
971 	}
972 	return ret;
973 }
974