xref: /linux/security/commoncap.c (revision 765532c8aaac624b5f8687af6d319c6a1138a257)
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 
31 /*
32  * If a non-root user executes a setuid-root binary in
33  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
34  * However if fE is also set, then the intent is for only
35  * the file capabilities to be applied, and the setuid-root
36  * bit is left on either to change the uid (plausible) or
37  * to get full privilege on a kernel without file capabilities
38  * support.  So in that case we do not raise capabilities.
39  *
40  * Warn if that happens, once per boot.
41  */
42 static void warn_setuid_and_fcaps_mixed(const char *fname)
43 {
44 	static int warned;
45 	if (!warned) {
46 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
47 			" effective capabilities. Therefore not raising all"
48 			" capabilities.\n", fname);
49 		warned = 1;
50 	}
51 }
52 
53 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
54 {
55 	NETLINK_CB(skb).eff_cap = current_cap();
56 	return 0;
57 }
58 
59 int cap_netlink_recv(struct sk_buff *skb, int cap)
60 {
61 	if (!cap_raised(NETLINK_CB(skb).eff_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  * @cap: The capability to check for
72  * @audit: Whether to write an audit message or not
73  *
74  * Determine whether the nominated task has the specified capability amongst
75  * its effective set, returning 0 if it does, -ve if it does not.
76  *
77  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
78  * and has_capability() functions.  That is, it has the reverse semantics:
79  * cap_has_capability() returns 0 when a task has a capability, but the
80  * kernel's capable() and has_capability() returns 1 for this case.
81  */
82 int cap_capable(struct task_struct *tsk, const struct cred *cred, int cap,
83 		int audit)
84 {
85 	return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
86 }
87 
88 /**
89  * cap_settime - Determine whether the current process may set the system clock
90  * @ts: The time to set
91  * @tz: The timezone to set
92  *
93  * Determine whether the current process may set the system clock and timezone
94  * information, returning 0 if permission granted, -ve if denied.
95  */
96 int cap_settime(struct timespec *ts, struct timezone *tz)
97 {
98 	if (!capable(CAP_SYS_TIME))
99 		return -EPERM;
100 	return 0;
101 }
102 
103 /**
104  * cap_ptrace_access_check - Determine whether the current process may access
105  *			   another
106  * @child: The process to be accessed
107  * @mode: The mode of attachment.
108  *
109  * Determine whether a process may access another, returning 0 if permission
110  * granted, -ve if denied.
111  */
112 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
113 {
114 	int ret = 0;
115 
116 	rcu_read_lock();
117 	if (!cap_issubset(__task_cred(child)->cap_permitted,
118 			  current_cred()->cap_permitted) &&
119 	    !capable(CAP_SYS_PTRACE))
120 		ret = -EPERM;
121 	rcu_read_unlock();
122 	return ret;
123 }
124 
125 /**
126  * cap_ptrace_traceme - Determine whether another process may trace the current
127  * @parent: The task proposed to be the tracer
128  *
129  * Determine whether the nominated task is permitted to trace the current
130  * process, returning 0 if permission is granted, -ve if denied.
131  */
132 int cap_ptrace_traceme(struct task_struct *parent)
133 {
134 	int ret = 0;
135 
136 	rcu_read_lock();
137 	if (!cap_issubset(current_cred()->cap_permitted,
138 			  __task_cred(parent)->cap_permitted) &&
139 	    !has_capability(parent, CAP_SYS_PTRACE))
140 		ret = -EPERM;
141 	rcu_read_unlock();
142 	return ret;
143 }
144 
145 /**
146  * cap_capget - Retrieve a task's capability sets
147  * @target: The task from which to retrieve the capability sets
148  * @effective: The place to record the effective set
149  * @inheritable: The place to record the inheritable set
150  * @permitted: The place to record the permitted set
151  *
152  * This function retrieves the capabilities of the nominated task and returns
153  * them to the caller.
154  */
155 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
156 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
157 {
158 	const struct cred *cred;
159 
160 	/* Derived from kernel/capability.c:sys_capget. */
161 	rcu_read_lock();
162 	cred = __task_cred(target);
163 	*effective   = cred->cap_effective;
164 	*inheritable = cred->cap_inheritable;
165 	*permitted   = cred->cap_permitted;
166 	rcu_read_unlock();
167 	return 0;
168 }
169 
170 /*
171  * Determine whether the inheritable capabilities are limited to the old
172  * permitted set.  Returns 1 if they are limited, 0 if they are not.
173  */
174 static inline int cap_inh_is_capped(void)
175 {
176 
177 	/* they are so limited unless the current task has the CAP_SETPCAP
178 	 * capability
179 	 */
180 	if (cap_capable(current, current_cred(), CAP_SETPCAP,
181 			SECURITY_CAP_AUDIT) == 0)
182 		return 0;
183 	return 1;
184 }
185 
186 /**
187  * cap_capset - Validate and apply proposed changes to current's capabilities
188  * @new: The proposed new credentials; alterations should be made here
189  * @old: The current task's current credentials
190  * @effective: A pointer to the proposed new effective capabilities set
191  * @inheritable: A pointer to the proposed new inheritable capabilities set
192  * @permitted: A pointer to the proposed new permitted capabilities set
193  *
194  * This function validates and applies a proposed mass change to the current
195  * process's capability sets.  The changes are made to the proposed new
196  * credentials, and assuming no error, will be committed by the caller of LSM.
197  */
198 int cap_capset(struct cred *new,
199 	       const struct cred *old,
200 	       const kernel_cap_t *effective,
201 	       const kernel_cap_t *inheritable,
202 	       const kernel_cap_t *permitted)
203 {
204 	if (cap_inh_is_capped() &&
205 	    !cap_issubset(*inheritable,
206 			  cap_combine(old->cap_inheritable,
207 				      old->cap_permitted)))
208 		/* incapable of using this inheritable set */
209 		return -EPERM;
210 
211 	if (!cap_issubset(*inheritable,
212 			  cap_combine(old->cap_inheritable,
213 				      old->cap_bset)))
214 		/* no new pI capabilities outside bounding set */
215 		return -EPERM;
216 
217 	/* verify restrictions on target's new Permitted set */
218 	if (!cap_issubset(*permitted, old->cap_permitted))
219 		return -EPERM;
220 
221 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
222 	if (!cap_issubset(*effective, *permitted))
223 		return -EPERM;
224 
225 	new->cap_effective   = *effective;
226 	new->cap_inheritable = *inheritable;
227 	new->cap_permitted   = *permitted;
228 	return 0;
229 }
230 
231 /*
232  * Clear proposed capability sets for execve().
233  */
234 static inline void bprm_clear_caps(struct linux_binprm *bprm)
235 {
236 	cap_clear(bprm->cred->cap_permitted);
237 	bprm->cap_effective = false;
238 }
239 
240 /**
241  * cap_inode_need_killpriv - Determine if inode change affects privileges
242  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
243  *
244  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
245  * affects the security markings on that inode, and if it is, should
246  * inode_killpriv() be invoked or the change rejected?
247  *
248  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
249  * -ve to deny the change.
250  */
251 int cap_inode_need_killpriv(struct dentry *dentry)
252 {
253 	struct inode *inode = dentry->d_inode;
254 	int error;
255 
256 	if (!inode->i_op->getxattr)
257 	       return 0;
258 
259 	error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
260 	if (error <= 0)
261 		return 0;
262 	return 1;
263 }
264 
265 /**
266  * cap_inode_killpriv - Erase the security markings on an inode
267  * @dentry: The inode/dentry to alter
268  *
269  * Erase the privilege-enhancing security markings on an inode.
270  *
271  * Returns 0 if successful, -ve on error.
272  */
273 int cap_inode_killpriv(struct dentry *dentry)
274 {
275 	struct inode *inode = dentry->d_inode;
276 
277 	if (!inode->i_op->removexattr)
278 	       return 0;
279 
280 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
281 }
282 
283 /*
284  * Calculate the new process capability sets from the capability sets attached
285  * to a file.
286  */
287 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
288 					  struct linux_binprm *bprm,
289 					  bool *effective)
290 {
291 	struct cred *new = bprm->cred;
292 	unsigned i;
293 	int ret = 0;
294 
295 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
296 		*effective = true;
297 
298 	CAP_FOR_EACH_U32(i) {
299 		__u32 permitted = caps->permitted.cap[i];
300 		__u32 inheritable = caps->inheritable.cap[i];
301 
302 		/*
303 		 * pP' = (X & fP) | (pI & fI)
304 		 */
305 		new->cap_permitted.cap[i] =
306 			(new->cap_bset.cap[i] & permitted) |
307 			(new->cap_inheritable.cap[i] & inheritable);
308 
309 		if (permitted & ~new->cap_permitted.cap[i])
310 			/* insufficient to execute correctly */
311 			ret = -EPERM;
312 	}
313 
314 	/*
315 	 * For legacy apps, with no internal support for recognizing they
316 	 * do not have enough capabilities, we return an error if they are
317 	 * missing some "forced" (aka file-permitted) capabilities.
318 	 */
319 	return *effective ? ret : 0;
320 }
321 
322 /*
323  * Extract the on-exec-apply capability sets for an executable file.
324  */
325 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
326 {
327 	struct inode *inode = dentry->d_inode;
328 	__u32 magic_etc;
329 	unsigned tocopy, i;
330 	int size;
331 	struct vfs_cap_data caps;
332 
333 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
334 
335 	if (!inode || !inode->i_op->getxattr)
336 		return -ENODATA;
337 
338 	size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
339 				   XATTR_CAPS_SZ);
340 	if (size == -ENODATA || size == -EOPNOTSUPP)
341 		/* no data, that's ok */
342 		return -ENODATA;
343 	if (size < 0)
344 		return size;
345 
346 	if (size < sizeof(magic_etc))
347 		return -EINVAL;
348 
349 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
350 
351 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
352 	case VFS_CAP_REVISION_1:
353 		if (size != XATTR_CAPS_SZ_1)
354 			return -EINVAL;
355 		tocopy = VFS_CAP_U32_1;
356 		break;
357 	case VFS_CAP_REVISION_2:
358 		if (size != XATTR_CAPS_SZ_2)
359 			return -EINVAL;
360 		tocopy = VFS_CAP_U32_2;
361 		break;
362 	default:
363 		return -EINVAL;
364 	}
365 
366 	CAP_FOR_EACH_U32(i) {
367 		if (i >= tocopy)
368 			break;
369 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
370 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
371 	}
372 
373 	return 0;
374 }
375 
376 /*
377  * Attempt to get the on-exec apply capability sets for an executable file from
378  * its xattrs and, if present, apply them to the proposed credentials being
379  * constructed by execve().
380  */
381 static int get_file_caps(struct linux_binprm *bprm, bool *effective)
382 {
383 	struct dentry *dentry;
384 	int rc = 0;
385 	struct cpu_vfs_cap_data vcaps;
386 
387 	bprm_clear_caps(bprm);
388 
389 	if (!file_caps_enabled)
390 		return 0;
391 
392 	if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)
393 		return 0;
394 
395 	dentry = dget(bprm->file->f_dentry);
396 
397 	rc = get_vfs_caps_from_disk(dentry, &vcaps);
398 	if (rc < 0) {
399 		if (rc == -EINVAL)
400 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
401 				__func__, rc, bprm->filename);
402 		else if (rc == -ENODATA)
403 			rc = 0;
404 		goto out;
405 	}
406 
407 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective);
408 	if (rc == -EINVAL)
409 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
410 		       __func__, rc, bprm->filename);
411 
412 out:
413 	dput(dentry);
414 	if (rc)
415 		bprm_clear_caps(bprm);
416 
417 	return rc;
418 }
419 
420 /**
421  * cap_bprm_set_creds - Set up the proposed credentials for execve().
422  * @bprm: The execution parameters, including the proposed creds
423  *
424  * Set up the proposed credentials for a new execution context being
425  * constructed by execve().  The proposed creds in @bprm->cred is altered,
426  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
427  */
428 int cap_bprm_set_creds(struct linux_binprm *bprm)
429 {
430 	const struct cred *old = current_cred();
431 	struct cred *new = bprm->cred;
432 	bool effective;
433 	int ret;
434 
435 	effective = false;
436 	ret = get_file_caps(bprm, &effective);
437 	if (ret < 0)
438 		return ret;
439 
440 	if (!issecure(SECURE_NOROOT)) {
441 		/*
442 		 * If the legacy file capability is set, then don't set privs
443 		 * for a setuid root binary run by a non-root user.  Do set it
444 		 * for a root user just to cause least surprise to an admin.
445 		 */
446 		if (effective && new->uid != 0 && new->euid == 0) {
447 			warn_setuid_and_fcaps_mixed(bprm->filename);
448 			goto skip;
449 		}
450 		/*
451 		 * To support inheritance of root-permissions and suid-root
452 		 * executables under compatibility mode, we override the
453 		 * capability sets for the file.
454 		 *
455 		 * If only the real uid is 0, we do not set the effective bit.
456 		 */
457 		if (new->euid == 0 || new->uid == 0) {
458 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
459 			new->cap_permitted = cap_combine(old->cap_bset,
460 							 old->cap_inheritable);
461 		}
462 		if (new->euid == 0)
463 			effective = true;
464 	}
465 skip:
466 
467 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
468 	 * credentials unless they have the appropriate permit
469 	 */
470 	if ((new->euid != old->uid ||
471 	     new->egid != old->gid ||
472 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
473 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
474 		/* downgrade; they get no more than they had, and maybe less */
475 		if (!capable(CAP_SETUID)) {
476 			new->euid = new->uid;
477 			new->egid = new->gid;
478 		}
479 		new->cap_permitted = cap_intersect(new->cap_permitted,
480 						   old->cap_permitted);
481 	}
482 
483 	new->suid = new->fsuid = new->euid;
484 	new->sgid = new->fsgid = new->egid;
485 
486 	/* For init, we want to retain the capabilities set in the initial
487 	 * task.  Thus we skip the usual capability rules
488 	 */
489 	if (!is_global_init(current)) {
490 		if (effective)
491 			new->cap_effective = new->cap_permitted;
492 		else
493 			cap_clear(new->cap_effective);
494 	}
495 	bprm->cap_effective = effective;
496 
497 	/*
498 	 * Audit candidate if current->cap_effective is set
499 	 *
500 	 * We do not bother to audit if 3 things are true:
501 	 *   1) cap_effective has all caps
502 	 *   2) we are root
503 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
504 	 * Since this is just a normal root execing a process.
505 	 *
506 	 * Number 1 above might fail if you don't have a full bset, but I think
507 	 * that is interesting information to audit.
508 	 */
509 	if (!cap_isclear(new->cap_effective)) {
510 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
511 		    new->euid != 0 || new->uid != 0 ||
512 		    issecure(SECURE_NOROOT)) {
513 			ret = audit_log_bprm_fcaps(bprm, new, old);
514 			if (ret < 0)
515 				return ret;
516 		}
517 	}
518 
519 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
520 	return 0;
521 }
522 
523 /**
524  * cap_bprm_secureexec - Determine whether a secure execution is required
525  * @bprm: The execution parameters
526  *
527  * Determine whether a secure execution is required, return 1 if it is, and 0
528  * if it is not.
529  *
530  * The credentials have been committed by this point, and so are no longer
531  * available through @bprm->cred.
532  */
533 int cap_bprm_secureexec(struct linux_binprm *bprm)
534 {
535 	const struct cred *cred = current_cred();
536 
537 	if (cred->uid != 0) {
538 		if (bprm->cap_effective)
539 			return 1;
540 		if (!cap_isclear(cred->cap_permitted))
541 			return 1;
542 	}
543 
544 	return (cred->euid != cred->uid ||
545 		cred->egid != cred->gid);
546 }
547 
548 /**
549  * cap_inode_setxattr - Determine whether an xattr may be altered
550  * @dentry: The inode/dentry being altered
551  * @name: The name of the xattr to be changed
552  * @value: The value that the xattr will be changed to
553  * @size: The size of value
554  * @flags: The replacement flag
555  *
556  * Determine whether an xattr may be altered or set on an inode, returning 0 if
557  * permission is granted, -ve if denied.
558  *
559  * This is used to make sure security xattrs don't get updated or set by those
560  * who aren't privileged to do so.
561  */
562 int cap_inode_setxattr(struct dentry *dentry, const char *name,
563 		       const void *value, size_t size, int flags)
564 {
565 	if (!strcmp(name, XATTR_NAME_CAPS)) {
566 		if (!capable(CAP_SETFCAP))
567 			return -EPERM;
568 		return 0;
569 	}
570 
571 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
572 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
573 	    !capable(CAP_SYS_ADMIN))
574 		return -EPERM;
575 	return 0;
576 }
577 
578 /**
579  * cap_inode_removexattr - Determine whether an xattr may be removed
580  * @dentry: The inode/dentry being altered
581  * @name: The name of the xattr to be changed
582  *
583  * Determine whether an xattr may be removed from an inode, returning 0 if
584  * permission is granted, -ve if denied.
585  *
586  * This is used to make sure security xattrs don't get removed by those who
587  * aren't privileged to remove them.
588  */
589 int cap_inode_removexattr(struct dentry *dentry, const char *name)
590 {
591 	if (!strcmp(name, XATTR_NAME_CAPS)) {
592 		if (!capable(CAP_SETFCAP))
593 			return -EPERM;
594 		return 0;
595 	}
596 
597 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
598 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
599 	    !capable(CAP_SYS_ADMIN))
600 		return -EPERM;
601 	return 0;
602 }
603 
604 /*
605  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
606  * a process after a call to setuid, setreuid, or setresuid.
607  *
608  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
609  *  {r,e,s}uid != 0, the permitted and effective capabilities are
610  *  cleared.
611  *
612  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
613  *  capabilities of the process are cleared.
614  *
615  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
616  *  capabilities are set to the permitted capabilities.
617  *
618  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
619  *  never happen.
620  *
621  *  -astor
622  *
623  * cevans - New behaviour, Oct '99
624  * A process may, via prctl(), elect to keep its capabilities when it
625  * calls setuid() and switches away from uid==0. Both permitted and
626  * effective sets will be retained.
627  * Without this change, it was impossible for a daemon to drop only some
628  * of its privilege. The call to setuid(!=0) would drop all privileges!
629  * Keeping uid 0 is not an option because uid 0 owns too many vital
630  * files..
631  * Thanks to Olaf Kirch and Peter Benie for spotting this.
632  */
633 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
634 {
635 	if ((old->uid == 0 || old->euid == 0 || old->suid == 0) &&
636 	    (new->uid != 0 && new->euid != 0 && new->suid != 0) &&
637 	    !issecure(SECURE_KEEP_CAPS)) {
638 		cap_clear(new->cap_permitted);
639 		cap_clear(new->cap_effective);
640 	}
641 	if (old->euid == 0 && new->euid != 0)
642 		cap_clear(new->cap_effective);
643 	if (old->euid != 0 && new->euid == 0)
644 		new->cap_effective = new->cap_permitted;
645 }
646 
647 /**
648  * cap_task_fix_setuid - Fix up the results of setuid() call
649  * @new: The proposed credentials
650  * @old: The current task's current credentials
651  * @flags: Indications of what has changed
652  *
653  * Fix up the results of setuid() call before the credential changes are
654  * actually applied, returning 0 to grant the changes, -ve to deny them.
655  */
656 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
657 {
658 	switch (flags) {
659 	case LSM_SETID_RE:
660 	case LSM_SETID_ID:
661 	case LSM_SETID_RES:
662 		/* juggle the capabilities to follow [RES]UID changes unless
663 		 * otherwise suppressed */
664 		if (!issecure(SECURE_NO_SETUID_FIXUP))
665 			cap_emulate_setxuid(new, old);
666 		break;
667 
668 	case LSM_SETID_FS:
669 		/* juggle the capabilties to follow FSUID changes, unless
670 		 * otherwise suppressed
671 		 *
672 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
673 		 *          if not, we might be a bit too harsh here.
674 		 */
675 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
676 			if (old->fsuid == 0 && new->fsuid != 0)
677 				new->cap_effective =
678 					cap_drop_fs_set(new->cap_effective);
679 
680 			if (old->fsuid != 0 && new->fsuid == 0)
681 				new->cap_effective =
682 					cap_raise_fs_set(new->cap_effective,
683 							 new->cap_permitted);
684 		}
685 		break;
686 
687 	default:
688 		return -EINVAL;
689 	}
690 
691 	return 0;
692 }
693 
694 /*
695  * Rationale: code calling task_setscheduler, task_setioprio, and
696  * task_setnice, assumes that
697  *   . if capable(cap_sys_nice), then those actions should be allowed
698  *   . if not capable(cap_sys_nice), but acting on your own processes,
699  *   	then those actions should be allowed
700  * This is insufficient now since you can call code without suid, but
701  * yet with increased caps.
702  * So we check for increased caps on the target process.
703  */
704 static int cap_safe_nice(struct task_struct *p)
705 {
706 	int is_subset;
707 
708 	rcu_read_lock();
709 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
710 				 current_cred()->cap_permitted);
711 	rcu_read_unlock();
712 
713 	if (!is_subset && !capable(CAP_SYS_NICE))
714 		return -EPERM;
715 	return 0;
716 }
717 
718 /**
719  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
720  * @p: The task to affect
721  *
722  * Detemine if the requested scheduler policy change is permitted for the
723  * specified task, returning 0 if permission is granted, -ve if denied.
724  */
725 int cap_task_setscheduler(struct task_struct *p)
726 {
727 	return cap_safe_nice(p);
728 }
729 
730 /**
731  * cap_task_ioprio - Detemine if I/O priority change is permitted
732  * @p: The task to affect
733  * @ioprio: The I/O priority to set
734  *
735  * Detemine if the requested I/O priority change is permitted for the specified
736  * task, returning 0 if permission is granted, -ve if denied.
737  */
738 int cap_task_setioprio(struct task_struct *p, int ioprio)
739 {
740 	return cap_safe_nice(p);
741 }
742 
743 /**
744  * cap_task_ioprio - Detemine if task priority change is permitted
745  * @p: The task to affect
746  * @nice: The nice value to set
747  *
748  * Detemine if the requested task priority change is permitted for the
749  * specified task, returning 0 if permission is granted, -ve if denied.
750  */
751 int cap_task_setnice(struct task_struct *p, int nice)
752 {
753 	return cap_safe_nice(p);
754 }
755 
756 /*
757  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
758  * the current task's bounding set.  Returns 0 on success, -ve on error.
759  */
760 static long cap_prctl_drop(struct cred *new, unsigned long cap)
761 {
762 	if (!capable(CAP_SETPCAP))
763 		return -EPERM;
764 	if (!cap_valid(cap))
765 		return -EINVAL;
766 
767 	cap_lower(new->cap_bset, cap);
768 	return 0;
769 }
770 
771 /**
772  * cap_task_prctl - Implement process control functions for this security module
773  * @option: The process control function requested
774  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
775  *
776  * Allow process control functions (sys_prctl()) to alter capabilities; may
777  * also deny access to other functions not otherwise implemented here.
778  *
779  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
780  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
781  * modules will consider performing the function.
782  */
783 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
784 		   unsigned long arg4, unsigned long arg5)
785 {
786 	struct cred *new;
787 	long error = 0;
788 
789 	new = prepare_creds();
790 	if (!new)
791 		return -ENOMEM;
792 
793 	switch (option) {
794 	case PR_CAPBSET_READ:
795 		error = -EINVAL;
796 		if (!cap_valid(arg2))
797 			goto error;
798 		error = !!cap_raised(new->cap_bset, arg2);
799 		goto no_change;
800 
801 	case PR_CAPBSET_DROP:
802 		error = cap_prctl_drop(new, arg2);
803 		if (error < 0)
804 			goto error;
805 		goto changed;
806 
807 	/*
808 	 * The next four prctl's remain to assist with transitioning a
809 	 * system from legacy UID=0 based privilege (when filesystem
810 	 * capabilities are not in use) to a system using filesystem
811 	 * capabilities only - as the POSIX.1e draft intended.
812 	 *
813 	 * Note:
814 	 *
815 	 *  PR_SET_SECUREBITS =
816 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
817 	 *    | issecure_mask(SECURE_NOROOT)
818 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
819 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
820 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
821 	 *
822 	 * will ensure that the current process and all of its
823 	 * children will be locked into a pure
824 	 * capability-based-privilege environment.
825 	 */
826 	case PR_SET_SECUREBITS:
827 		error = -EPERM;
828 		if ((((new->securebits & SECURE_ALL_LOCKS) >> 1)
829 		     & (new->securebits ^ arg2))			/*[1]*/
830 		    || ((new->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
831 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
832 		    || (cap_capable(current, current_cred(), CAP_SETPCAP,
833 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
834 			/*
835 			 * [1] no changing of bits that are locked
836 			 * [2] no unlocking of locks
837 			 * [3] no setting of unsupported bits
838 			 * [4] doing anything requires privilege (go read about
839 			 *     the "sendmail capabilities bug")
840 			 */
841 		    )
842 			/* cannot change a locked bit */
843 			goto error;
844 		new->securebits = arg2;
845 		goto changed;
846 
847 	case PR_GET_SECUREBITS:
848 		error = new->securebits;
849 		goto no_change;
850 
851 	case PR_GET_KEEPCAPS:
852 		if (issecure(SECURE_KEEP_CAPS))
853 			error = 1;
854 		goto no_change;
855 
856 	case PR_SET_KEEPCAPS:
857 		error = -EINVAL;
858 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
859 			goto error;
860 		error = -EPERM;
861 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
862 			goto error;
863 		if (arg2)
864 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
865 		else
866 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
867 		goto changed;
868 
869 	default:
870 		/* No functionality available - continue with default */
871 		error = -ENOSYS;
872 		goto error;
873 	}
874 
875 	/* Functionality provided */
876 changed:
877 	return commit_creds(new);
878 
879 no_change:
880 error:
881 	abort_creds(new);
882 	return error;
883 }
884 
885 /**
886  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
887  * @mm: The VM space in which the new mapping is to be made
888  * @pages: The size of the mapping
889  *
890  * Determine whether the allocation of a new virtual mapping by the current
891  * task is permitted, returning 0 if permission is granted, -ve if not.
892  */
893 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
894 {
895 	int cap_sys_admin = 0;
896 
897 	if (cap_capable(current, current_cred(), CAP_SYS_ADMIN,
898 			SECURITY_CAP_NOAUDIT) == 0)
899 		cap_sys_admin = 1;
900 	return __vm_enough_memory(mm, pages, cap_sys_admin);
901 }
902 
903 /*
904  * cap_file_mmap - check if able to map given addr
905  * @file: unused
906  * @reqprot: unused
907  * @prot: unused
908  * @flags: unused
909  * @addr: address attempting to be mapped
910  * @addr_only: unused
911  *
912  * If the process is attempting to map memory below dac_mmap_min_addr they need
913  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
914  * capability security module.  Returns 0 if this mapping should be allowed
915  * -EPERM if not.
916  */
917 int cap_file_mmap(struct file *file, unsigned long reqprot,
918 		  unsigned long prot, unsigned long flags,
919 		  unsigned long addr, unsigned long addr_only)
920 {
921 	int ret = 0;
922 
923 	if (addr < dac_mmap_min_addr) {
924 		ret = cap_capable(current, current_cred(), CAP_SYS_RAWIO,
925 				  SECURITY_CAP_AUDIT);
926 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
927 		if (ret == 0)
928 			current->flags |= PF_SUPERPRIV;
929 	}
930 	return ret;
931 }
932