xref: /linux/security/commoncap.c (revision 9ffc93f203c18a70623f21950f1dd473c9ec48cd)
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 
33 /*
34  * If a non-root user executes a setuid-root binary in
35  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
36  * However if fE is also set, then the intent is for only
37  * the file capabilities to be applied, and the setuid-root
38  * bit is left on either to change the uid (plausible) or
39  * to get full privilege on a kernel without file capabilities
40  * support.  So in that case we do not raise capabilities.
41  *
42  * Warn if that happens, once per boot.
43  */
44 static void warn_setuid_and_fcaps_mixed(const char *fname)
45 {
46 	static int warned;
47 	if (!warned) {
48 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
49 			" effective capabilities. Therefore not raising all"
50 			" capabilities.\n", fname);
51 		warned = 1;
52 	}
53 }
54 
55 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
56 {
57 	return 0;
58 }
59 
60 /**
61  * cap_capable - Determine whether a task has a particular effective capability
62  * @cred: The credentials to use
63  * @ns:  The user namespace in which we need the capability
64  * @cap: The capability to check for
65  * @audit: Whether to write an audit message or not
66  *
67  * Determine whether the nominated task has the specified capability amongst
68  * its effective set, returning 0 if it does, -ve if it does not.
69  *
70  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
71  * and has_capability() functions.  That is, it has the reverse semantics:
72  * cap_has_capability() returns 0 when a task has a capability, but the
73  * kernel's capable() and has_capability() returns 1 for this case.
74  */
75 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
76 		int cap, int audit)
77 {
78 	for (;;) {
79 		/* The creator of the user namespace has all caps. */
80 		if (targ_ns != &init_user_ns && targ_ns->creator == cred->user)
81 			return 0;
82 
83 		/* Do we have the necessary capabilities? */
84 		if (targ_ns == cred->user->user_ns)
85 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
86 
87 		/* Have we tried all of the parent namespaces? */
88 		if (targ_ns == &init_user_ns)
89 			return -EPERM;
90 
91 		/*
92 		 *If you have a capability in a parent user ns, then you have
93 		 * it over all children user namespaces as well.
94 		 */
95 		targ_ns = targ_ns->creator->user_ns;
96 	}
97 
98 	/* We never get here */
99 }
100 
101 /**
102  * cap_settime - Determine whether the current process may set the system clock
103  * @ts: The time to set
104  * @tz: The timezone to set
105  *
106  * Determine whether the current process may set the system clock and timezone
107  * information, returning 0 if permission granted, -ve if denied.
108  */
109 int cap_settime(const struct timespec *ts, const struct timezone *tz)
110 {
111 	if (!capable(CAP_SYS_TIME))
112 		return -EPERM;
113 	return 0;
114 }
115 
116 /**
117  * cap_ptrace_access_check - Determine whether the current process may access
118  *			   another
119  * @child: The process to be accessed
120  * @mode: The mode of attachment.
121  *
122  * If we are in the same or an ancestor user_ns and have all the target
123  * task's capabilities, then ptrace access is allowed.
124  * If we have the ptrace capability to the target user_ns, then ptrace
125  * access is allowed.
126  * Else denied.
127  *
128  * Determine whether a process may access another, returning 0 if permission
129  * granted, -ve if denied.
130  */
131 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
132 {
133 	int ret = 0;
134 	const struct cred *cred, *child_cred;
135 
136 	rcu_read_lock();
137 	cred = current_cred();
138 	child_cred = __task_cred(child);
139 	if (cred->user->user_ns == child_cred->user->user_ns &&
140 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
141 		goto out;
142 	if (ns_capable(child_cred->user->user_ns, CAP_SYS_PTRACE))
143 		goto out;
144 	ret = -EPERM;
145 out:
146 	rcu_read_unlock();
147 	return ret;
148 }
149 
150 /**
151  * cap_ptrace_traceme - Determine whether another process may trace the current
152  * @parent: The task proposed to be the tracer
153  *
154  * If parent is in the same or an ancestor user_ns and has all current's
155  * capabilities, then ptrace access is allowed.
156  * If parent has the ptrace capability to current's user_ns, then ptrace
157  * access is allowed.
158  * Else denied.
159  *
160  * Determine whether the nominated task is permitted to trace the current
161  * process, returning 0 if permission is granted, -ve if denied.
162  */
163 int cap_ptrace_traceme(struct task_struct *parent)
164 {
165 	int ret = 0;
166 	const struct cred *cred, *child_cred;
167 
168 	rcu_read_lock();
169 	cred = __task_cred(parent);
170 	child_cred = current_cred();
171 	if (cred->user->user_ns == child_cred->user->user_ns &&
172 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
173 		goto out;
174 	if (has_ns_capability(parent, child_cred->user->user_ns, CAP_SYS_PTRACE))
175 		goto out;
176 	ret = -EPERM;
177 out:
178 	rcu_read_unlock();
179 	return ret;
180 }
181 
182 /**
183  * cap_capget - Retrieve a task's capability sets
184  * @target: The task from which to retrieve the capability sets
185  * @effective: The place to record the effective set
186  * @inheritable: The place to record the inheritable set
187  * @permitted: The place to record the permitted set
188  *
189  * This function retrieves the capabilities of the nominated task and returns
190  * them to the caller.
191  */
192 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
193 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
194 {
195 	const struct cred *cred;
196 
197 	/* Derived from kernel/capability.c:sys_capget. */
198 	rcu_read_lock();
199 	cred = __task_cred(target);
200 	*effective   = cred->cap_effective;
201 	*inheritable = cred->cap_inheritable;
202 	*permitted   = cred->cap_permitted;
203 	rcu_read_unlock();
204 	return 0;
205 }
206 
207 /*
208  * Determine whether the inheritable capabilities are limited to the old
209  * permitted set.  Returns 1 if they are limited, 0 if they are not.
210  */
211 static inline int cap_inh_is_capped(void)
212 {
213 
214 	/* they are so limited unless the current task has the CAP_SETPCAP
215 	 * capability
216 	 */
217 	if (cap_capable(current_cred(), current_cred()->user->user_ns,
218 			CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
219 		return 0;
220 	return 1;
221 }
222 
223 /**
224  * cap_capset - Validate and apply proposed changes to current's capabilities
225  * @new: The proposed new credentials; alterations should be made here
226  * @old: The current task's current credentials
227  * @effective: A pointer to the proposed new effective capabilities set
228  * @inheritable: A pointer to the proposed new inheritable capabilities set
229  * @permitted: A pointer to the proposed new permitted capabilities set
230  *
231  * This function validates and applies a proposed mass change to the current
232  * process's capability sets.  The changes are made to the proposed new
233  * credentials, and assuming no error, will be committed by the caller of LSM.
234  */
235 int cap_capset(struct cred *new,
236 	       const struct cred *old,
237 	       const kernel_cap_t *effective,
238 	       const kernel_cap_t *inheritable,
239 	       const kernel_cap_t *permitted)
240 {
241 	if (cap_inh_is_capped() &&
242 	    !cap_issubset(*inheritable,
243 			  cap_combine(old->cap_inheritable,
244 				      old->cap_permitted)))
245 		/* incapable of using this inheritable set */
246 		return -EPERM;
247 
248 	if (!cap_issubset(*inheritable,
249 			  cap_combine(old->cap_inheritable,
250 				      old->cap_bset)))
251 		/* no new pI capabilities outside bounding set */
252 		return -EPERM;
253 
254 	/* verify restrictions on target's new Permitted set */
255 	if (!cap_issubset(*permitted, old->cap_permitted))
256 		return -EPERM;
257 
258 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
259 	if (!cap_issubset(*effective, *permitted))
260 		return -EPERM;
261 
262 	new->cap_effective   = *effective;
263 	new->cap_inheritable = *inheritable;
264 	new->cap_permitted   = *permitted;
265 	return 0;
266 }
267 
268 /*
269  * Clear proposed capability sets for execve().
270  */
271 static inline void bprm_clear_caps(struct linux_binprm *bprm)
272 {
273 	cap_clear(bprm->cred->cap_permitted);
274 	bprm->cap_effective = false;
275 }
276 
277 /**
278  * cap_inode_need_killpriv - Determine if inode change affects privileges
279  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
280  *
281  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
282  * affects the security markings on that inode, and if it is, should
283  * inode_killpriv() be invoked or the change rejected?
284  *
285  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
286  * -ve to deny the change.
287  */
288 int cap_inode_need_killpriv(struct dentry *dentry)
289 {
290 	struct inode *inode = dentry->d_inode;
291 	int error;
292 
293 	if (!inode->i_op->getxattr)
294 	       return 0;
295 
296 	error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
297 	if (error <= 0)
298 		return 0;
299 	return 1;
300 }
301 
302 /**
303  * cap_inode_killpriv - Erase the security markings on an inode
304  * @dentry: The inode/dentry to alter
305  *
306  * Erase the privilege-enhancing security markings on an inode.
307  *
308  * Returns 0 if successful, -ve on error.
309  */
310 int cap_inode_killpriv(struct dentry *dentry)
311 {
312 	struct inode *inode = dentry->d_inode;
313 
314 	if (!inode->i_op->removexattr)
315 	       return 0;
316 
317 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
318 }
319 
320 /*
321  * Calculate the new process capability sets from the capability sets attached
322  * to a file.
323  */
324 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
325 					  struct linux_binprm *bprm,
326 					  bool *effective,
327 					  bool *has_cap)
328 {
329 	struct cred *new = bprm->cred;
330 	unsigned i;
331 	int ret = 0;
332 
333 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
334 		*effective = true;
335 
336 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
337 		*has_cap = true;
338 
339 	CAP_FOR_EACH_U32(i) {
340 		__u32 permitted = caps->permitted.cap[i];
341 		__u32 inheritable = caps->inheritable.cap[i];
342 
343 		/*
344 		 * pP' = (X & fP) | (pI & fI)
345 		 */
346 		new->cap_permitted.cap[i] =
347 			(new->cap_bset.cap[i] & permitted) |
348 			(new->cap_inheritable.cap[i] & inheritable);
349 
350 		if (permitted & ~new->cap_permitted.cap[i])
351 			/* insufficient to execute correctly */
352 			ret = -EPERM;
353 	}
354 
355 	/*
356 	 * For legacy apps, with no internal support for recognizing they
357 	 * do not have enough capabilities, we return an error if they are
358 	 * missing some "forced" (aka file-permitted) capabilities.
359 	 */
360 	return *effective ? ret : 0;
361 }
362 
363 /*
364  * Extract the on-exec-apply capability sets for an executable file.
365  */
366 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
367 {
368 	struct inode *inode = dentry->d_inode;
369 	__u32 magic_etc;
370 	unsigned tocopy, i;
371 	int size;
372 	struct vfs_cap_data caps;
373 
374 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
375 
376 	if (!inode || !inode->i_op->getxattr)
377 		return -ENODATA;
378 
379 	size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
380 				   XATTR_CAPS_SZ);
381 	if (size == -ENODATA || size == -EOPNOTSUPP)
382 		/* no data, that's ok */
383 		return -ENODATA;
384 	if (size < 0)
385 		return size;
386 
387 	if (size < sizeof(magic_etc))
388 		return -EINVAL;
389 
390 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
391 
392 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
393 	case VFS_CAP_REVISION_1:
394 		if (size != XATTR_CAPS_SZ_1)
395 			return -EINVAL;
396 		tocopy = VFS_CAP_U32_1;
397 		break;
398 	case VFS_CAP_REVISION_2:
399 		if (size != XATTR_CAPS_SZ_2)
400 			return -EINVAL;
401 		tocopy = VFS_CAP_U32_2;
402 		break;
403 	default:
404 		return -EINVAL;
405 	}
406 
407 	CAP_FOR_EACH_U32(i) {
408 		if (i >= tocopy)
409 			break;
410 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
411 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
412 	}
413 
414 	return 0;
415 }
416 
417 /*
418  * Attempt to get the on-exec apply capability sets for an executable file from
419  * its xattrs and, if present, apply them to the proposed credentials being
420  * constructed by execve().
421  */
422 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap)
423 {
424 	struct dentry *dentry;
425 	int rc = 0;
426 	struct cpu_vfs_cap_data vcaps;
427 
428 	bprm_clear_caps(bprm);
429 
430 	if (!file_caps_enabled)
431 		return 0;
432 
433 	if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)
434 		return 0;
435 
436 	dentry = dget(bprm->file->f_dentry);
437 
438 	rc = get_vfs_caps_from_disk(dentry, &vcaps);
439 	if (rc < 0) {
440 		if (rc == -EINVAL)
441 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
442 				__func__, rc, bprm->filename);
443 		else if (rc == -ENODATA)
444 			rc = 0;
445 		goto out;
446 	}
447 
448 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
449 	if (rc == -EINVAL)
450 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
451 		       __func__, rc, bprm->filename);
452 
453 out:
454 	dput(dentry);
455 	if (rc)
456 		bprm_clear_caps(bprm);
457 
458 	return rc;
459 }
460 
461 /**
462  * cap_bprm_set_creds - Set up the proposed credentials for execve().
463  * @bprm: The execution parameters, including the proposed creds
464  *
465  * Set up the proposed credentials for a new execution context being
466  * constructed by execve().  The proposed creds in @bprm->cred is altered,
467  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
468  */
469 int cap_bprm_set_creds(struct linux_binprm *bprm)
470 {
471 	const struct cred *old = current_cred();
472 	struct cred *new = bprm->cred;
473 	bool effective, has_cap = false;
474 	int ret;
475 
476 	effective = false;
477 	ret = get_file_caps(bprm, &effective, &has_cap);
478 	if (ret < 0)
479 		return ret;
480 
481 	if (!issecure(SECURE_NOROOT)) {
482 		/*
483 		 * If the legacy file capability is set, then don't set privs
484 		 * for a setuid root binary run by a non-root user.  Do set it
485 		 * for a root user just to cause least surprise to an admin.
486 		 */
487 		if (has_cap && new->uid != 0 && new->euid == 0) {
488 			warn_setuid_and_fcaps_mixed(bprm->filename);
489 			goto skip;
490 		}
491 		/*
492 		 * To support inheritance of root-permissions and suid-root
493 		 * executables under compatibility mode, we override the
494 		 * capability sets for the file.
495 		 *
496 		 * If only the real uid is 0, we do not set the effective bit.
497 		 */
498 		if (new->euid == 0 || new->uid == 0) {
499 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
500 			new->cap_permitted = cap_combine(old->cap_bset,
501 							 old->cap_inheritable);
502 		}
503 		if (new->euid == 0)
504 			effective = true;
505 	}
506 skip:
507 
508 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
509 	 * credentials unless they have the appropriate permit
510 	 */
511 	if ((new->euid != old->uid ||
512 	     new->egid != old->gid ||
513 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
514 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
515 		/* downgrade; they get no more than they had, and maybe less */
516 		if (!capable(CAP_SETUID)) {
517 			new->euid = new->uid;
518 			new->egid = new->gid;
519 		}
520 		new->cap_permitted = cap_intersect(new->cap_permitted,
521 						   old->cap_permitted);
522 	}
523 
524 	new->suid = new->fsuid = new->euid;
525 	new->sgid = new->fsgid = new->egid;
526 
527 	if (effective)
528 		new->cap_effective = new->cap_permitted;
529 	else
530 		cap_clear(new->cap_effective);
531 	bprm->cap_effective = effective;
532 
533 	/*
534 	 * Audit candidate if current->cap_effective is set
535 	 *
536 	 * We do not bother to audit if 3 things are true:
537 	 *   1) cap_effective has all caps
538 	 *   2) we are root
539 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
540 	 * Since this is just a normal root execing a process.
541 	 *
542 	 * Number 1 above might fail if you don't have a full bset, but I think
543 	 * that is interesting information to audit.
544 	 */
545 	if (!cap_isclear(new->cap_effective)) {
546 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
547 		    new->euid != 0 || new->uid != 0 ||
548 		    issecure(SECURE_NOROOT)) {
549 			ret = audit_log_bprm_fcaps(bprm, new, old);
550 			if (ret < 0)
551 				return ret;
552 		}
553 	}
554 
555 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
556 	return 0;
557 }
558 
559 /**
560  * cap_bprm_secureexec - Determine whether a secure execution is required
561  * @bprm: The execution parameters
562  *
563  * Determine whether a secure execution is required, return 1 if it is, and 0
564  * if it is not.
565  *
566  * The credentials have been committed by this point, and so are no longer
567  * available through @bprm->cred.
568  */
569 int cap_bprm_secureexec(struct linux_binprm *bprm)
570 {
571 	const struct cred *cred = current_cred();
572 
573 	if (cred->uid != 0) {
574 		if (bprm->cap_effective)
575 			return 1;
576 		if (!cap_isclear(cred->cap_permitted))
577 			return 1;
578 	}
579 
580 	return (cred->euid != cred->uid ||
581 		cred->egid != cred->gid);
582 }
583 
584 /**
585  * cap_inode_setxattr - Determine whether an xattr may be altered
586  * @dentry: The inode/dentry being altered
587  * @name: The name of the xattr to be changed
588  * @value: The value that the xattr will be changed to
589  * @size: The size of value
590  * @flags: The replacement flag
591  *
592  * Determine whether an xattr may be altered or set on an inode, returning 0 if
593  * permission is granted, -ve if denied.
594  *
595  * This is used to make sure security xattrs don't get updated or set by those
596  * who aren't privileged to do so.
597  */
598 int cap_inode_setxattr(struct dentry *dentry, const char *name,
599 		       const void *value, size_t size, int flags)
600 {
601 	if (!strcmp(name, XATTR_NAME_CAPS)) {
602 		if (!capable(CAP_SETFCAP))
603 			return -EPERM;
604 		return 0;
605 	}
606 
607 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
608 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
609 	    !capable(CAP_SYS_ADMIN))
610 		return -EPERM;
611 	return 0;
612 }
613 
614 /**
615  * cap_inode_removexattr - Determine whether an xattr may be removed
616  * @dentry: The inode/dentry being altered
617  * @name: The name of the xattr to be changed
618  *
619  * Determine whether an xattr may be removed from an inode, returning 0 if
620  * permission is granted, -ve if denied.
621  *
622  * This is used to make sure security xattrs don't get removed by those who
623  * aren't privileged to remove them.
624  */
625 int cap_inode_removexattr(struct dentry *dentry, const char *name)
626 {
627 	if (!strcmp(name, XATTR_NAME_CAPS)) {
628 		if (!capable(CAP_SETFCAP))
629 			return -EPERM;
630 		return 0;
631 	}
632 
633 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
634 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
635 	    !capable(CAP_SYS_ADMIN))
636 		return -EPERM;
637 	return 0;
638 }
639 
640 /*
641  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
642  * a process after a call to setuid, setreuid, or setresuid.
643  *
644  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
645  *  {r,e,s}uid != 0, the permitted and effective capabilities are
646  *  cleared.
647  *
648  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
649  *  capabilities of the process are cleared.
650  *
651  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
652  *  capabilities are set to the permitted capabilities.
653  *
654  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
655  *  never happen.
656  *
657  *  -astor
658  *
659  * cevans - New behaviour, Oct '99
660  * A process may, via prctl(), elect to keep its capabilities when it
661  * calls setuid() and switches away from uid==0. Both permitted and
662  * effective sets will be retained.
663  * Without this change, it was impossible for a daemon to drop only some
664  * of its privilege. The call to setuid(!=0) would drop all privileges!
665  * Keeping uid 0 is not an option because uid 0 owns too many vital
666  * files..
667  * Thanks to Olaf Kirch and Peter Benie for spotting this.
668  */
669 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
670 {
671 	if ((old->uid == 0 || old->euid == 0 || old->suid == 0) &&
672 	    (new->uid != 0 && new->euid != 0 && new->suid != 0) &&
673 	    !issecure(SECURE_KEEP_CAPS)) {
674 		cap_clear(new->cap_permitted);
675 		cap_clear(new->cap_effective);
676 	}
677 	if (old->euid == 0 && new->euid != 0)
678 		cap_clear(new->cap_effective);
679 	if (old->euid != 0 && new->euid == 0)
680 		new->cap_effective = new->cap_permitted;
681 }
682 
683 /**
684  * cap_task_fix_setuid - Fix up the results of setuid() call
685  * @new: The proposed credentials
686  * @old: The current task's current credentials
687  * @flags: Indications of what has changed
688  *
689  * Fix up the results of setuid() call before the credential changes are
690  * actually applied, returning 0 to grant the changes, -ve to deny them.
691  */
692 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
693 {
694 	switch (flags) {
695 	case LSM_SETID_RE:
696 	case LSM_SETID_ID:
697 	case LSM_SETID_RES:
698 		/* juggle the capabilities to follow [RES]UID changes unless
699 		 * otherwise suppressed */
700 		if (!issecure(SECURE_NO_SETUID_FIXUP))
701 			cap_emulate_setxuid(new, old);
702 		break;
703 
704 	case LSM_SETID_FS:
705 		/* juggle the capabilties to follow FSUID changes, unless
706 		 * otherwise suppressed
707 		 *
708 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
709 		 *          if not, we might be a bit too harsh here.
710 		 */
711 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
712 			if (old->fsuid == 0 && new->fsuid != 0)
713 				new->cap_effective =
714 					cap_drop_fs_set(new->cap_effective);
715 
716 			if (old->fsuid != 0 && new->fsuid == 0)
717 				new->cap_effective =
718 					cap_raise_fs_set(new->cap_effective,
719 							 new->cap_permitted);
720 		}
721 		break;
722 
723 	default:
724 		return -EINVAL;
725 	}
726 
727 	return 0;
728 }
729 
730 /*
731  * Rationale: code calling task_setscheduler, task_setioprio, and
732  * task_setnice, assumes that
733  *   . if capable(cap_sys_nice), then those actions should be allowed
734  *   . if not capable(cap_sys_nice), but acting on your own processes,
735  *   	then those actions should be allowed
736  * This is insufficient now since you can call code without suid, but
737  * yet with increased caps.
738  * So we check for increased caps on the target process.
739  */
740 static int cap_safe_nice(struct task_struct *p)
741 {
742 	int is_subset;
743 
744 	rcu_read_lock();
745 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
746 				 current_cred()->cap_permitted);
747 	rcu_read_unlock();
748 
749 	if (!is_subset && !capable(CAP_SYS_NICE))
750 		return -EPERM;
751 	return 0;
752 }
753 
754 /**
755  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
756  * @p: The task to affect
757  *
758  * Detemine if the requested scheduler policy change is permitted for the
759  * specified task, returning 0 if permission is granted, -ve if denied.
760  */
761 int cap_task_setscheduler(struct task_struct *p)
762 {
763 	return cap_safe_nice(p);
764 }
765 
766 /**
767  * cap_task_ioprio - Detemine if I/O priority change is permitted
768  * @p: The task to affect
769  * @ioprio: The I/O priority to set
770  *
771  * Detemine if the requested I/O priority change is permitted for the specified
772  * task, returning 0 if permission is granted, -ve if denied.
773  */
774 int cap_task_setioprio(struct task_struct *p, int ioprio)
775 {
776 	return cap_safe_nice(p);
777 }
778 
779 /**
780  * cap_task_ioprio - Detemine if task priority change is permitted
781  * @p: The task to affect
782  * @nice: The nice value to set
783  *
784  * Detemine if the requested task priority change is permitted for the
785  * specified task, returning 0 if permission is granted, -ve if denied.
786  */
787 int cap_task_setnice(struct task_struct *p, int nice)
788 {
789 	return cap_safe_nice(p);
790 }
791 
792 /*
793  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
794  * the current task's bounding set.  Returns 0 on success, -ve on error.
795  */
796 static long cap_prctl_drop(struct cred *new, unsigned long cap)
797 {
798 	if (!capable(CAP_SETPCAP))
799 		return -EPERM;
800 	if (!cap_valid(cap))
801 		return -EINVAL;
802 
803 	cap_lower(new->cap_bset, cap);
804 	return 0;
805 }
806 
807 /**
808  * cap_task_prctl - Implement process control functions for this security module
809  * @option: The process control function requested
810  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
811  *
812  * Allow process control functions (sys_prctl()) to alter capabilities; may
813  * also deny access to other functions not otherwise implemented here.
814  *
815  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
816  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
817  * modules will consider performing the function.
818  */
819 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
820 		   unsigned long arg4, unsigned long arg5)
821 {
822 	struct cred *new;
823 	long error = 0;
824 
825 	new = prepare_creds();
826 	if (!new)
827 		return -ENOMEM;
828 
829 	switch (option) {
830 	case PR_CAPBSET_READ:
831 		error = -EINVAL;
832 		if (!cap_valid(arg2))
833 			goto error;
834 		error = !!cap_raised(new->cap_bset, arg2);
835 		goto no_change;
836 
837 	case PR_CAPBSET_DROP:
838 		error = cap_prctl_drop(new, arg2);
839 		if (error < 0)
840 			goto error;
841 		goto changed;
842 
843 	/*
844 	 * The next four prctl's remain to assist with transitioning a
845 	 * system from legacy UID=0 based privilege (when filesystem
846 	 * capabilities are not in use) to a system using filesystem
847 	 * capabilities only - as the POSIX.1e draft intended.
848 	 *
849 	 * Note:
850 	 *
851 	 *  PR_SET_SECUREBITS =
852 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
853 	 *    | issecure_mask(SECURE_NOROOT)
854 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
855 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
856 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
857 	 *
858 	 * will ensure that the current process and all of its
859 	 * children will be locked into a pure
860 	 * capability-based-privilege environment.
861 	 */
862 	case PR_SET_SECUREBITS:
863 		error = -EPERM;
864 		if ((((new->securebits & SECURE_ALL_LOCKS) >> 1)
865 		     & (new->securebits ^ arg2))			/*[1]*/
866 		    || ((new->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
867 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
868 		    || (cap_capable(current_cred(),
869 				    current_cred()->user->user_ns, CAP_SETPCAP,
870 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
871 			/*
872 			 * [1] no changing of bits that are locked
873 			 * [2] no unlocking of locks
874 			 * [3] no setting of unsupported bits
875 			 * [4] doing anything requires privilege (go read about
876 			 *     the "sendmail capabilities bug")
877 			 */
878 		    )
879 			/* cannot change a locked bit */
880 			goto error;
881 		new->securebits = arg2;
882 		goto changed;
883 
884 	case PR_GET_SECUREBITS:
885 		error = new->securebits;
886 		goto no_change;
887 
888 	case PR_GET_KEEPCAPS:
889 		if (issecure(SECURE_KEEP_CAPS))
890 			error = 1;
891 		goto no_change;
892 
893 	case PR_SET_KEEPCAPS:
894 		error = -EINVAL;
895 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
896 			goto error;
897 		error = -EPERM;
898 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
899 			goto error;
900 		if (arg2)
901 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
902 		else
903 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
904 		goto changed;
905 
906 	default:
907 		/* No functionality available - continue with default */
908 		error = -ENOSYS;
909 		goto error;
910 	}
911 
912 	/* Functionality provided */
913 changed:
914 	return commit_creds(new);
915 
916 no_change:
917 error:
918 	abort_creds(new);
919 	return error;
920 }
921 
922 /**
923  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
924  * @mm: The VM space in which the new mapping is to be made
925  * @pages: The size of the mapping
926  *
927  * Determine whether the allocation of a new virtual mapping by the current
928  * task is permitted, returning 0 if permission is granted, -ve if not.
929  */
930 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
931 {
932 	int cap_sys_admin = 0;
933 
934 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
935 			SECURITY_CAP_NOAUDIT) == 0)
936 		cap_sys_admin = 1;
937 	return __vm_enough_memory(mm, pages, cap_sys_admin);
938 }
939 
940 /*
941  * cap_file_mmap - check if able to map given addr
942  * @file: unused
943  * @reqprot: unused
944  * @prot: unused
945  * @flags: unused
946  * @addr: address attempting to be mapped
947  * @addr_only: unused
948  *
949  * If the process is attempting to map memory below dac_mmap_min_addr they need
950  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
951  * capability security module.  Returns 0 if this mapping should be allowed
952  * -EPERM if not.
953  */
954 int cap_file_mmap(struct file *file, unsigned long reqprot,
955 		  unsigned long prot, unsigned long flags,
956 		  unsigned long addr, unsigned long addr_only)
957 {
958 	int ret = 0;
959 
960 	if (addr < dac_mmap_min_addr) {
961 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
962 				  SECURITY_CAP_AUDIT);
963 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
964 		if (ret == 0)
965 			current->flags |= PF_SUPERPRIV;
966 	}
967 	return ret;
968 }
969