• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Common capabilities, needed by capability.o.
2  *
3  *	This program is free software; you can redistribute it and/or modify
4  *	it under the terms of the GNU General Public License as published by
5  *	the Free Software Foundation; either version 2 of the License, or
6  *	(at your option) any later version.
7  *
8  */
9 
10 #include <linux/capability.h>
11 #include <linux/audit.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/security.h>
16 #include <linux/file.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/skbuff.h>
22 #include <linux/netlink.h>
23 #include <linux/ptrace.h>
24 #include <linux/xattr.h>
25 #include <linux/hugetlb.h>
26 #include <linux/mount.h>
27 #include <linux/sched.h>
28 #include <linux/prctl.h>
29 #include <linux/securebits.h>
30 #include <linux/user_namespace.h>
31 #include <linux/binfmts.h>
32 #include <linux/personality.h>
33 
34 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
35 #include <linux/android_aid.h>
36 #endif
37 
38 /*
39  * If a non-root user executes a setuid-root binary in
40  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
41  * However if fE is also set, then the intent is for only
42  * the file capabilities to be applied, and the setuid-root
43  * bit is left on either to change the uid (plausible) or
44  * to get full privilege on a kernel without file capabilities
45  * support.  So in that case we do not raise capabilities.
46  *
47  * Warn if that happens, once per boot.
48  */
warn_setuid_and_fcaps_mixed(const char * fname)49 static void warn_setuid_and_fcaps_mixed(const char *fname)
50 {
51 	static int warned;
52 	if (!warned) {
53 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
54 			" effective capabilities. Therefore not raising all"
55 			" capabilities.\n", fname);
56 		warned = 1;
57 	}
58 }
59 
cap_netlink_send(struct sock * sk,struct sk_buff * skb)60 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
61 {
62 	return 0;
63 }
64 
65 /**
66  * cap_capable - Determine whether a task has a particular effective capability
67  * @cred: The credentials to use
68  * @ns:  The user namespace in which we need the capability
69  * @cap: The capability to check for
70  * @audit: Whether to write an audit message or not
71  *
72  * Determine whether the nominated task has the specified capability amongst
73  * its effective set, returning 0 if it does, -ve if it does not.
74  *
75  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
76  * and has_capability() functions.  That is, it has the reverse semantics:
77  * cap_has_capability() returns 0 when a task has a capability, but the
78  * kernel's capable() and has_capability() returns 1 for this case.
79  */
cap_capable(const struct cred * cred,struct user_namespace * targ_ns,int cap,int audit)80 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
81 		int cap, int audit)
82 {
83 	struct user_namespace *ns = targ_ns;
84 
85 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
86 	if (cap == CAP_NET_RAW && in_egroup_p(AID_NET_RAW))
87 		return 0;
88 	if (cap == CAP_NET_ADMIN && in_egroup_p(AID_NET_ADMIN))
89 		return 0;
90 #endif
91 
92 	/* See if cred has the capability in the target user namespace
93 	 * by examining the target user namespace and all of the target
94 	 * user namespace's parents.
95 	 */
96 	for (;;) {
97 		/* Do we have the necessary capabilities? */
98 		if (ns == cred->user_ns)
99 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
100 
101 		/* Have we tried all of the parent namespaces? */
102 		if (ns == &init_user_ns)
103 			return -EPERM;
104 
105 		/*
106 		 * The owner of the user namespace in the parent of the
107 		 * user namespace has all caps.
108 		 */
109 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
110 			return 0;
111 
112 		/*
113 		 * If you have a capability in a parent user ns, then you have
114 		 * it over all children user namespaces as well.
115 		 */
116 		ns = ns->parent;
117 	}
118 
119 	/* We never get here */
120 }
121 
122 /**
123  * cap_settime - Determine whether the current process may set the system clock
124  * @ts: The time to set
125  * @tz: The timezone to set
126  *
127  * Determine whether the current process may set the system clock and timezone
128  * information, returning 0 if permission granted, -ve if denied.
129  */
cap_settime(const struct timespec * ts,const struct timezone * tz)130 int cap_settime(const struct timespec *ts, const struct timezone *tz)
131 {
132 	if (!capable(CAP_SYS_TIME))
133 		return -EPERM;
134 	return 0;
135 }
136 
137 /**
138  * cap_ptrace_access_check - Determine whether the current process may access
139  *			   another
140  * @child: The process to be accessed
141  * @mode: The mode of attachment.
142  *
143  * If we are in the same or an ancestor user_ns and have all the target
144  * task's capabilities, then ptrace access is allowed.
145  * If we have the ptrace capability to the target user_ns, then ptrace
146  * access is allowed.
147  * Else denied.
148  *
149  * Determine whether a process may access another, returning 0 if permission
150  * granted, -ve if denied.
151  */
cap_ptrace_access_check(struct task_struct * child,unsigned int mode)152 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
153 {
154 	int ret = 0;
155 	const struct cred *cred, *child_cred;
156 	const kernel_cap_t *caller_caps;
157 
158 	rcu_read_lock();
159 	cred = current_cred();
160 	child_cred = __task_cred(child);
161 	if (mode & PTRACE_MODE_FSCREDS)
162 		caller_caps = &cred->cap_effective;
163 	else
164 		caller_caps = &cred->cap_permitted;
165 	if (cred->user_ns == child_cred->user_ns &&
166 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
167 		goto out;
168 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
169 		goto out;
170 	ret = -EPERM;
171 out:
172 	rcu_read_unlock();
173 	return ret;
174 }
175 
176 /**
177  * cap_ptrace_traceme - Determine whether another process may trace the current
178  * @parent: The task proposed to be the tracer
179  *
180  * If parent is in the same or an ancestor user_ns and has all current's
181  * capabilities, then ptrace access is allowed.
182  * If parent has the ptrace capability to current's user_ns, then ptrace
183  * access is allowed.
184  * Else denied.
185  *
186  * Determine whether the nominated task is permitted to trace the current
187  * process, returning 0 if permission is granted, -ve if denied.
188  */
cap_ptrace_traceme(struct task_struct * parent)189 int cap_ptrace_traceme(struct task_struct *parent)
190 {
191 	int ret = 0;
192 	const struct cred *cred, *child_cred;
193 
194 	rcu_read_lock();
195 	cred = __task_cred(parent);
196 	child_cred = current_cred();
197 	if (cred->user_ns == child_cred->user_ns &&
198 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
199 		goto out;
200 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
201 		goto out;
202 	ret = -EPERM;
203 out:
204 	rcu_read_unlock();
205 	return ret;
206 }
207 
208 /**
209  * cap_capget - Retrieve a task's capability sets
210  * @target: The task from which to retrieve the capability sets
211  * @effective: The place to record the effective set
212  * @inheritable: The place to record the inheritable set
213  * @permitted: The place to record the permitted set
214  *
215  * This function retrieves the capabilities of the nominated task and returns
216  * them to the caller.
217  */
cap_capget(struct task_struct * target,kernel_cap_t * effective,kernel_cap_t * inheritable,kernel_cap_t * permitted)218 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
219 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
220 {
221 	const struct cred *cred;
222 
223 	/* Derived from kernel/capability.c:sys_capget. */
224 	rcu_read_lock();
225 	cred = __task_cred(target);
226 	*effective   = cred->cap_effective;
227 	*inheritable = cred->cap_inheritable;
228 	*permitted   = cred->cap_permitted;
229 	rcu_read_unlock();
230 	return 0;
231 }
232 
233 /*
234  * Determine whether the inheritable capabilities are limited to the old
235  * permitted set.  Returns 1 if they are limited, 0 if they are not.
236  */
cap_inh_is_capped(void)237 static inline int cap_inh_is_capped(void)
238 {
239 
240 	/* they are so limited unless the current task has the CAP_SETPCAP
241 	 * capability
242 	 */
243 	if (cap_capable(current_cred(), current_cred()->user_ns,
244 			CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
245 		return 0;
246 	return 1;
247 }
248 
249 /**
250  * cap_capset - Validate and apply proposed changes to current's capabilities
251  * @new: The proposed new credentials; alterations should be made here
252  * @old: The current task's current credentials
253  * @effective: A pointer to the proposed new effective capabilities set
254  * @inheritable: A pointer to the proposed new inheritable capabilities set
255  * @permitted: A pointer to the proposed new permitted capabilities set
256  *
257  * This function validates and applies a proposed mass change to the current
258  * process's capability sets.  The changes are made to the proposed new
259  * credentials, and assuming no error, will be committed by the caller of LSM.
260  */
cap_capset(struct cred * new,const struct cred * old,const kernel_cap_t * effective,const kernel_cap_t * inheritable,const kernel_cap_t * permitted)261 int cap_capset(struct cred *new,
262 	       const struct cred *old,
263 	       const kernel_cap_t *effective,
264 	       const kernel_cap_t *inheritable,
265 	       const kernel_cap_t *permitted)
266 {
267 	if (cap_inh_is_capped() &&
268 	    !cap_issubset(*inheritable,
269 			  cap_combine(old->cap_inheritable,
270 				      old->cap_permitted)))
271 		/* incapable of using this inheritable set */
272 		return -EPERM;
273 
274 	if (!cap_issubset(*inheritable,
275 			  cap_combine(old->cap_inheritable,
276 				      old->cap_bset)))
277 		/* no new pI capabilities outside bounding set */
278 		return -EPERM;
279 
280 	/* verify restrictions on target's new Permitted set */
281 	if (!cap_issubset(*permitted, old->cap_permitted))
282 		return -EPERM;
283 
284 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
285 	if (!cap_issubset(*effective, *permitted))
286 		return -EPERM;
287 
288 	new->cap_effective   = *effective;
289 	new->cap_inheritable = *inheritable;
290 	new->cap_permitted   = *permitted;
291 
292 	/*
293 	 * Mask off ambient bits that are no longer both permitted and
294 	 * inheritable.
295 	 */
296 	new->cap_ambient = cap_intersect(new->cap_ambient,
297 					 cap_intersect(*permitted,
298 						       *inheritable));
299 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
300 		return -EINVAL;
301 	return 0;
302 }
303 
304 /*
305  * Clear proposed capability sets for execve().
306  */
bprm_clear_caps(struct linux_binprm * bprm)307 static inline void bprm_clear_caps(struct linux_binprm *bprm)
308 {
309 	cap_clear(bprm->cred->cap_permitted);
310 	bprm->cap_effective = false;
311 }
312 
313 /**
314  * cap_inode_need_killpriv - Determine if inode change affects privileges
315  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
316  *
317  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
318  * affects the security markings on that inode, and if it is, should
319  * inode_killpriv() be invoked or the change rejected?
320  *
321  * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and
322  * -ve to deny the change.
323  */
cap_inode_need_killpriv(struct dentry * dentry)324 int cap_inode_need_killpriv(struct dentry *dentry)
325 {
326 	struct inode *inode = dentry->d_inode;
327 	int error;
328 
329 	if (!inode->i_op->getxattr)
330 	       return 0;
331 
332 	error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
333 	if (error <= 0)
334 		return 0;
335 	return 1;
336 }
337 
338 /**
339  * cap_inode_killpriv - Erase the security markings on an inode
340  * @dentry: The inode/dentry to alter
341  *
342  * Erase the privilege-enhancing security markings on an inode.
343  *
344  * Returns 0 if successful, -ve on error.
345  */
cap_inode_killpriv(struct dentry * dentry)346 int cap_inode_killpriv(struct dentry *dentry)
347 {
348 	struct inode *inode = dentry->d_inode;
349 
350 	if (!inode->i_op->removexattr)
351 	       return 0;
352 
353 	return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
354 }
355 
356 /*
357  * Calculate the new process capability sets from the capability sets attached
358  * to a file.
359  */
bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data * caps,struct linux_binprm * bprm,bool * effective,bool * has_cap)360 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
361 					  struct linux_binprm *bprm,
362 					  bool *effective,
363 					  bool *has_cap)
364 {
365 	struct cred *new = bprm->cred;
366 	unsigned i;
367 	int ret = 0;
368 
369 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
370 		*effective = true;
371 
372 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
373 		*has_cap = true;
374 
375 	CAP_FOR_EACH_U32(i) {
376 		__u32 permitted = caps->permitted.cap[i];
377 		__u32 inheritable = caps->inheritable.cap[i];
378 
379 		/*
380 		 * pP' = (X & fP) | (pI & fI)
381 		 * The addition of pA' is handled later.
382 		 */
383 		new->cap_permitted.cap[i] =
384 			(new->cap_bset.cap[i] & permitted) |
385 			(new->cap_inheritable.cap[i] & inheritable);
386 
387 		if (permitted & ~new->cap_permitted.cap[i])
388 			/* insufficient to execute correctly */
389 			ret = -EPERM;
390 	}
391 
392 	/*
393 	 * For legacy apps, with no internal support for recognizing they
394 	 * do not have enough capabilities, we return an error if they are
395 	 * missing some "forced" (aka file-permitted) capabilities.
396 	 */
397 	return *effective ? ret : 0;
398 }
399 
400 /*
401  * Extract the on-exec-apply capability sets for an executable file.
402  */
get_vfs_caps_from_disk(const struct dentry * dentry,struct cpu_vfs_cap_data * cpu_caps)403 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
404 {
405 	struct inode *inode = dentry->d_inode;
406 	__u32 magic_etc;
407 	unsigned tocopy, i;
408 	int size;
409 	struct vfs_cap_data caps;
410 
411 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
412 
413 	if (!inode || !inode->i_op->getxattr)
414 		return -ENODATA;
415 
416 	size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
417 				   XATTR_CAPS_SZ);
418 	if (size == -ENODATA || size == -EOPNOTSUPP)
419 		/* no data, that's ok */
420 		return -ENODATA;
421 	if (size < 0)
422 		return size;
423 
424 	if (size < sizeof(magic_etc))
425 		return -EINVAL;
426 
427 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
428 
429 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
430 	case VFS_CAP_REVISION_1:
431 		if (size != XATTR_CAPS_SZ_1)
432 			return -EINVAL;
433 		tocopy = VFS_CAP_U32_1;
434 		break;
435 	case VFS_CAP_REVISION_2:
436 		if (size != XATTR_CAPS_SZ_2)
437 			return -EINVAL;
438 		tocopy = VFS_CAP_U32_2;
439 		break;
440 	default:
441 		return -EINVAL;
442 	}
443 
444 	CAP_FOR_EACH_U32(i) {
445 		if (i >= tocopy)
446 			break;
447 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
448 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
449 	}
450 
451 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
452 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
453 
454 	return 0;
455 }
456 
457 /*
458  * Attempt to get the on-exec apply capability sets for an executable file from
459  * its xattrs and, if present, apply them to the proposed credentials being
460  * constructed by execve().
461  */
get_file_caps(struct linux_binprm * bprm,bool * effective,bool * has_cap)462 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap)
463 {
464 	struct dentry *dentry;
465 	int rc = 0;
466 	struct cpu_vfs_cap_data vcaps;
467 
468 	bprm_clear_caps(bprm);
469 
470 	if (!file_caps_enabled)
471 		return 0;
472 
473 	if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
474 		return 0;
475 
476 	dentry = dget(bprm->file->f_dentry);
477 
478 	rc = get_vfs_caps_from_disk(dentry, &vcaps);
479 	if (rc < 0) {
480 		if (rc == -EINVAL)
481 			printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
482 				__func__, rc, bprm->filename);
483 		else if (rc == -ENODATA)
484 			rc = 0;
485 		goto out;
486 	}
487 
488 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
489 	if (rc == -EINVAL)
490 		printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
491 		       __func__, rc, bprm->filename);
492 
493 out:
494 	dput(dentry);
495 	if (rc)
496 		bprm_clear_caps(bprm);
497 
498 	return rc;
499 }
500 
501 /**
502  * cap_bprm_set_creds - Set up the proposed credentials for execve().
503  * @bprm: The execution parameters, including the proposed creds
504  *
505  * Set up the proposed credentials for a new execution context being
506  * constructed by execve().  The proposed creds in @bprm->cred is altered,
507  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
508  */
cap_bprm_set_creds(struct linux_binprm * bprm)509 int cap_bprm_set_creds(struct linux_binprm *bprm)
510 {
511 	const struct cred *old = current_cred();
512 	struct cred *new = bprm->cred;
513 	bool effective, has_cap = false, is_setid;
514 	int ret;
515 	kuid_t root_uid;
516 
517 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
518 		return -EPERM;
519 
520 	effective = false;
521 	ret = get_file_caps(bprm, &effective, &has_cap);
522 	if (ret < 0)
523 		return ret;
524 
525 	root_uid = make_kuid(new->user_ns, 0);
526 
527 	if (!issecure(SECURE_NOROOT)) {
528 		/*
529 		 * If the legacy file capability is set, then don't set privs
530 		 * for a setuid root binary run by a non-root user.  Do set it
531 		 * for a root user just to cause least surprise to an admin.
532 		 */
533 		if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) {
534 			warn_setuid_and_fcaps_mixed(bprm->filename);
535 			goto skip;
536 		}
537 		/*
538 		 * To support inheritance of root-permissions and suid-root
539 		 * executables under compatibility mode, we override the
540 		 * capability sets for the file.
541 		 *
542 		 * If only the real uid is 0, we do not set the effective bit.
543 		 */
544 		if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) {
545 			/* pP' = (cap_bset & ~0) | (pI & ~0) */
546 			new->cap_permitted = cap_combine(old->cap_bset,
547 							 old->cap_inheritable);
548 		}
549 		if (uid_eq(new->euid, root_uid))
550 			effective = true;
551 	}
552 skip:
553 
554 	/* if we have fs caps, clear dangerous personality flags */
555 	if (!cap_issubset(new->cap_permitted, old->cap_permitted))
556 		bprm->per_clear |= PER_CLEAR_ON_SETID;
557 
558 
559 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
560 	 * credentials unless they have the appropriate permit.
561 	 *
562 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
563 	 */
564 	is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid);
565 
566 	if ((is_setid ||
567 	     !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
568 	    bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
569 		/* downgrade; they get no more than they had, and maybe less */
570 		if (!capable(CAP_SETUID) ||
571 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
572 			new->euid = new->uid;
573 			new->egid = new->gid;
574 		}
575 		new->cap_permitted = cap_intersect(new->cap_permitted,
576 						   old->cap_permitted);
577 	}
578 
579 	new->suid = new->fsuid = new->euid;
580 	new->sgid = new->fsgid = new->egid;
581 
582 	/* File caps or setid cancels ambient. */
583 	if (has_cap || is_setid)
584 		cap_clear(new->cap_ambient);
585 
586 	/*
587 	 * Now that we've computed pA', update pP' to give:
588 	 *   pP' = (X & fP) | (pI & fI) | pA'
589 	 */
590 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
591 
592 	/*
593 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
594 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
595 	 */
596 	if (effective)
597 		new->cap_effective = new->cap_permitted;
598 	else
599 		new->cap_effective = new->cap_ambient;
600 
601 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
602 		return -EPERM;
603 
604 	bprm->cap_effective = effective;
605 
606 	/*
607 	 * Audit candidate if current->cap_effective is set
608 	 *
609 	 * We do not bother to audit if 3 things are true:
610 	 *   1) cap_effective has all caps
611 	 *   2) we are root
612 	 *   3) root is supposed to have all caps (SECURE_NOROOT)
613 	 * Since this is just a normal root execing a process.
614 	 *
615 	 * Number 1 above might fail if you don't have a full bset, but I think
616 	 * that is interesting information to audit.
617 	 */
618 	if (!cap_issubset(new->cap_effective, new->cap_ambient)) {
619 		if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
620 		    !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) ||
621 		    issecure(SECURE_NOROOT)) {
622 			ret = audit_log_bprm_fcaps(bprm, new, old);
623 			if (ret < 0)
624 				return ret;
625 		}
626 	}
627 
628 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
629 
630 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
631 		return -EPERM;
632 
633 	return 0;
634 }
635 
636 /**
637  * cap_bprm_secureexec - Determine whether a secure execution is required
638  * @bprm: The execution parameters
639  *
640  * Determine whether a secure execution is required, return 1 if it is, and 0
641  * if it is not.
642  *
643  * The credentials have been committed by this point, and so are no longer
644  * available through @bprm->cred.
645  */
cap_bprm_secureexec(struct linux_binprm * bprm)646 int cap_bprm_secureexec(struct linux_binprm *bprm)
647 {
648 	const struct cred *cred = current_cred();
649 	kuid_t root_uid = make_kuid(cred->user_ns, 0);
650 
651 	if (!uid_eq(cred->uid, root_uid)) {
652 		if (bprm->cap_effective)
653 			return 1;
654 		if (!cap_issubset(cred->cap_permitted, cred->cap_ambient))
655 			return 1;
656 	}
657 
658 	return (!uid_eq(cred->euid, cred->uid) ||
659 		!gid_eq(cred->egid, cred->gid));
660 }
661 
662 /**
663  * cap_inode_setxattr - Determine whether an xattr may be altered
664  * @dentry: The inode/dentry being altered
665  * @name: The name of the xattr to be changed
666  * @value: The value that the xattr will be changed to
667  * @size: The size of value
668  * @flags: The replacement flag
669  *
670  * Determine whether an xattr may be altered or set on an inode, returning 0 if
671  * permission is granted, -ve if denied.
672  *
673  * This is used to make sure security xattrs don't get updated or set by those
674  * who aren't privileged to do so.
675  */
cap_inode_setxattr(struct dentry * dentry,const char * name,const void * value,size_t size,int flags)676 int cap_inode_setxattr(struct dentry *dentry, const char *name,
677 		       const void *value, size_t size, int flags)
678 {
679 	if (!strcmp(name, XATTR_NAME_CAPS)) {
680 		if (!capable(CAP_SETFCAP))
681 			return -EPERM;
682 		return 0;
683 	}
684 
685 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
686 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
687 	    !capable(CAP_SYS_ADMIN))
688 		return -EPERM;
689 	return 0;
690 }
691 
692 /**
693  * cap_inode_removexattr - Determine whether an xattr may be removed
694  * @dentry: The inode/dentry being altered
695  * @name: The name of the xattr to be changed
696  *
697  * Determine whether an xattr may be removed from an inode, returning 0 if
698  * permission is granted, -ve if denied.
699  *
700  * This is used to make sure security xattrs don't get removed by those who
701  * aren't privileged to remove them.
702  */
cap_inode_removexattr(struct dentry * dentry,const char * name)703 int cap_inode_removexattr(struct dentry *dentry, const char *name)
704 {
705 	if (!strcmp(name, XATTR_NAME_CAPS)) {
706 		if (!capable(CAP_SETFCAP))
707 			return -EPERM;
708 		return 0;
709 	}
710 
711 	if (!strncmp(name, XATTR_SECURITY_PREFIX,
712 		     sizeof(XATTR_SECURITY_PREFIX) - 1) &&
713 	    !capable(CAP_SYS_ADMIN))
714 		return -EPERM;
715 	return 0;
716 }
717 
718 /*
719  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
720  * a process after a call to setuid, setreuid, or setresuid.
721  *
722  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
723  *  {r,e,s}uid != 0, the permitted and effective capabilities are
724  *  cleared.
725  *
726  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
727  *  capabilities of the process are cleared.
728  *
729  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
730  *  capabilities are set to the permitted capabilities.
731  *
732  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
733  *  never happen.
734  *
735  *  -astor
736  *
737  * cevans - New behaviour, Oct '99
738  * A process may, via prctl(), elect to keep its capabilities when it
739  * calls setuid() and switches away from uid==0. Both permitted and
740  * effective sets will be retained.
741  * Without this change, it was impossible for a daemon to drop only some
742  * of its privilege. The call to setuid(!=0) would drop all privileges!
743  * Keeping uid 0 is not an option because uid 0 owns too many vital
744  * files..
745  * Thanks to Olaf Kirch and Peter Benie for spotting this.
746  */
cap_emulate_setxuid(struct cred * new,const struct cred * old)747 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
748 {
749 	kuid_t root_uid = make_kuid(old->user_ns, 0);
750 
751 	if ((uid_eq(old->uid, root_uid) ||
752 	     uid_eq(old->euid, root_uid) ||
753 	     uid_eq(old->suid, root_uid)) &&
754 	    (!uid_eq(new->uid, root_uid) &&
755 	     !uid_eq(new->euid, root_uid) &&
756 	     !uid_eq(new->suid, root_uid))) {
757 		if (!issecure(SECURE_KEEP_CAPS)) {
758 			cap_clear(new->cap_permitted);
759 			cap_clear(new->cap_effective);
760 		}
761 
762 		/*
763 		 * Pre-ambient programs expect setresuid to nonroot followed
764 		 * by exec to drop capabilities.  We should make sure that
765 		 * this remains the case.
766 		 */
767 		cap_clear(new->cap_ambient);
768 	}
769 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
770 		cap_clear(new->cap_effective);
771 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
772 		new->cap_effective = new->cap_permitted;
773 }
774 
775 /**
776  * cap_task_fix_setuid - Fix up the results of setuid() call
777  * @new: The proposed credentials
778  * @old: The current task's current credentials
779  * @flags: Indications of what has changed
780  *
781  * Fix up the results of setuid() call before the credential changes are
782  * actually applied, returning 0 to grant the changes, -ve to deny them.
783  */
cap_task_fix_setuid(struct cred * new,const struct cred * old,int flags)784 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
785 {
786 	switch (flags) {
787 	case LSM_SETID_RE:
788 	case LSM_SETID_ID:
789 	case LSM_SETID_RES:
790 		/* juggle the capabilities to follow [RES]UID changes unless
791 		 * otherwise suppressed */
792 		if (!issecure(SECURE_NO_SETUID_FIXUP))
793 			cap_emulate_setxuid(new, old);
794 		break;
795 
796 	case LSM_SETID_FS:
797 		/* juggle the capabilties to follow FSUID changes, unless
798 		 * otherwise suppressed
799 		 *
800 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
801 		 *          if not, we might be a bit too harsh here.
802 		 */
803 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
804 			kuid_t root_uid = make_kuid(old->user_ns, 0);
805 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
806 				new->cap_effective =
807 					cap_drop_fs_set(new->cap_effective);
808 
809 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
810 				new->cap_effective =
811 					cap_raise_fs_set(new->cap_effective,
812 							 new->cap_permitted);
813 		}
814 		break;
815 
816 	default:
817 		return -EINVAL;
818 	}
819 
820 	return 0;
821 }
822 
823 /*
824  * Rationale: code calling task_setscheduler, task_setioprio, and
825  * task_setnice, assumes that
826  *   . if capable(cap_sys_nice), then those actions should be allowed
827  *   . if not capable(cap_sys_nice), but acting on your own processes,
828  *   	then those actions should be allowed
829  * This is insufficient now since you can call code without suid, but
830  * yet with increased caps.
831  * So we check for increased caps on the target process.
832  */
cap_safe_nice(struct task_struct * p)833 static int cap_safe_nice(struct task_struct *p)
834 {
835 	int is_subset, ret = 0;
836 
837 	rcu_read_lock();
838 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
839 				 current_cred()->cap_permitted);
840 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
841 		ret = -EPERM;
842 	rcu_read_unlock();
843 
844 	return ret;
845 }
846 
847 /**
848  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
849  * @p: The task to affect
850  *
851  * Detemine if the requested scheduler policy change is permitted for the
852  * specified task, returning 0 if permission is granted, -ve if denied.
853  */
cap_task_setscheduler(struct task_struct * p)854 int cap_task_setscheduler(struct task_struct *p)
855 {
856 	return cap_safe_nice(p);
857 }
858 
859 /**
860  * cap_task_ioprio - Detemine if I/O priority change is permitted
861  * @p: The task to affect
862  * @ioprio: The I/O priority to set
863  *
864  * Detemine if the requested I/O priority change is permitted for the specified
865  * task, returning 0 if permission is granted, -ve if denied.
866  */
cap_task_setioprio(struct task_struct * p,int ioprio)867 int cap_task_setioprio(struct task_struct *p, int ioprio)
868 {
869 	return cap_safe_nice(p);
870 }
871 
872 /**
873  * cap_task_ioprio - Detemine if task priority change is permitted
874  * @p: The task to affect
875  * @nice: The nice value to set
876  *
877  * Detemine if the requested task priority change is permitted for the
878  * specified task, returning 0 if permission is granted, -ve if denied.
879  */
cap_task_setnice(struct task_struct * p,int nice)880 int cap_task_setnice(struct task_struct *p, int nice)
881 {
882 	return cap_safe_nice(p);
883 }
884 
885 /*
886  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
887  * the current task's bounding set.  Returns 0 on success, -ve on error.
888  */
cap_prctl_drop(unsigned long cap)889 static int cap_prctl_drop(unsigned long cap)
890 {
891 	struct cred *new;
892 
893 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
894 		return -EPERM;
895 	if (!cap_valid(cap))
896 		return -EINVAL;
897 
898 	new = prepare_creds();
899 	if (!new)
900 		return -ENOMEM;
901 	cap_lower(new->cap_bset, cap);
902 	return commit_creds(new);
903 }
904 
905 /**
906  * cap_task_prctl - Implement process control functions for this security module
907  * @option: The process control function requested
908  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
909  *
910  * Allow process control functions (sys_prctl()) to alter capabilities; may
911  * also deny access to other functions not otherwise implemented here.
912  *
913  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
914  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
915  * modules will consider performing the function.
916  */
cap_task_prctl(int option,unsigned long arg2,unsigned long arg3,unsigned long arg4,unsigned long arg5)917 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
918 		   unsigned long arg4, unsigned long arg5)
919 {
920 	const struct cred *old = current_cred();
921 	struct cred *new;
922 
923 	switch (option) {
924 	case PR_CAPBSET_READ:
925 		if (!cap_valid(arg2))
926 			return -EINVAL;
927 		return !!cap_raised(old->cap_bset, arg2);
928 
929 	case PR_CAPBSET_DROP:
930 		return cap_prctl_drop(arg2);
931 
932 	/*
933 	 * The next four prctl's remain to assist with transitioning a
934 	 * system from legacy UID=0 based privilege (when filesystem
935 	 * capabilities are not in use) to a system using filesystem
936 	 * capabilities only - as the POSIX.1e draft intended.
937 	 *
938 	 * Note:
939 	 *
940 	 *  PR_SET_SECUREBITS =
941 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
942 	 *    | issecure_mask(SECURE_NOROOT)
943 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
944 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
945 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
946 	 *
947 	 * will ensure that the current process and all of its
948 	 * children will be locked into a pure
949 	 * capability-based-privilege environment.
950 	 */
951 	case PR_SET_SECUREBITS:
952 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
953 		     & (old->securebits ^ arg2))			/*[1]*/
954 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
955 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
956 		    || (cap_capable(current_cred(),
957 				    current_cred()->user_ns, CAP_SETPCAP,
958 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
959 			/*
960 			 * [1] no changing of bits that are locked
961 			 * [2] no unlocking of locks
962 			 * [3] no setting of unsupported bits
963 			 * [4] doing anything requires privilege (go read about
964 			 *     the "sendmail capabilities bug")
965 			 */
966 		    )
967 			/* cannot change a locked bit */
968 			return -EPERM;
969 
970 		new = prepare_creds();
971 		if (!new)
972 			return -ENOMEM;
973 		new->securebits = arg2;
974 		return commit_creds(new);
975 
976 	case PR_GET_SECUREBITS:
977 		return old->securebits;
978 
979 	case PR_GET_KEEPCAPS:
980 		return !!issecure(SECURE_KEEP_CAPS);
981 
982 	case PR_SET_KEEPCAPS:
983 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
984 			return -EINVAL;
985 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
986 			return -EPERM;
987 
988 		new = prepare_creds();
989 		if (!new)
990 			return -ENOMEM;
991 		if (arg2)
992 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
993 		else
994 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
995 		return commit_creds(new);
996 
997 	case PR_CAP_AMBIENT:
998 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
999 			if (arg3 | arg4 | arg5)
1000 				return -EINVAL;
1001 
1002 			new = prepare_creds();
1003 			if (!new)
1004 				return -ENOMEM;
1005 			cap_clear(new->cap_ambient);
1006 			return commit_creds(new);
1007 		}
1008 
1009 		if (((!cap_valid(arg3)) | arg4 | arg5))
1010 			return -EINVAL;
1011 
1012 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1013 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1014 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1015 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1016 			return -EINVAL;
1017 		} else {
1018 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1019 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1020 			     !cap_raised(current_cred()->cap_inheritable,
1021 					 arg3)))
1022 				return -EPERM;
1023 
1024 			new = prepare_creds();
1025 			if (!new)
1026 				return -ENOMEM;
1027 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1028 				cap_raise(new->cap_ambient, arg3);
1029 			else
1030 				cap_lower(new->cap_ambient, arg3);
1031 			return commit_creds(new);
1032 		}
1033 
1034 	default:
1035 		/* No functionality available - continue with default */
1036 		return -ENOSYS;
1037 	}
1038 }
1039 
1040 /**
1041  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1042  * @mm: The VM space in which the new mapping is to be made
1043  * @pages: The size of the mapping
1044  *
1045  * Determine whether the allocation of a new virtual mapping by the current
1046  * task is permitted, returning 0 if permission is granted, -ve if not.
1047  */
cap_vm_enough_memory(struct mm_struct * mm,long pages)1048 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1049 {
1050 	int cap_sys_admin = 0;
1051 
1052 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
1053 			SECURITY_CAP_NOAUDIT) == 0)
1054 		cap_sys_admin = 1;
1055 	return __vm_enough_memory(mm, pages, cap_sys_admin);
1056 }
1057 
1058 /*
1059  * cap_mmap_addr - check if able to map given addr
1060  * @addr: address attempting to be mapped
1061  *
1062  * If the process is attempting to map memory below dac_mmap_min_addr they need
1063  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1064  * capability security module.  Returns 0 if this mapping should be allowed
1065  * -EPERM if not.
1066  */
cap_mmap_addr(unsigned long addr)1067 int cap_mmap_addr(unsigned long addr)
1068 {
1069 	int ret = 0;
1070 
1071 	if (addr < dac_mmap_min_addr) {
1072 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1073 				  SECURITY_CAP_AUDIT);
1074 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1075 		if (ret == 0)
1076 			current->flags |= PF_SUPERPRIV;
1077 	}
1078 	return ret;
1079 }
1080 
cap_mmap_file(struct file * file,unsigned long reqprot,unsigned long prot,unsigned long flags)1081 int cap_mmap_file(struct file *file, unsigned long reqprot,
1082 		  unsigned long prot, unsigned long flags)
1083 {
1084 	return 0;
1085 }
1086