Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-or-later
2 : /* Common capabilities, needed by capability.o.
3 : */
4 :
5 : #include <linux/capability.h>
6 : #include <linux/audit.h>
7 : #include <linux/init.h>
8 : #include <linux/kernel.h>
9 : #include <linux/lsm_hooks.h>
10 : #include <linux/file.h>
11 : #include <linux/mm.h>
12 : #include <linux/mman.h>
13 : #include <linux/pagemap.h>
14 : #include <linux/swap.h>
15 : #include <linux/skbuff.h>
16 : #include <linux/netlink.h>
17 : #include <linux/ptrace.h>
18 : #include <linux/xattr.h>
19 : #include <linux/hugetlb.h>
20 : #include <linux/mount.h>
21 : #include <linux/sched.h>
22 : #include <linux/prctl.h>
23 : #include <linux/securebits.h>
24 : #include <linux/user_namespace.h>
25 : #include <linux/binfmts.h>
26 : #include <linux/personality.h>
27 :
28 : /*
29 : * If a non-root user executes a setuid-root binary in
30 : * !secure(SECURE_NOROOT) mode, then we raise capabilities.
31 : * However if fE is also set, then the intent is for only
32 : * the file capabilities to be applied, and the setuid-root
33 : * bit is left on either to change the uid (plausible) or
34 : * to get full privilege on a kernel without file capabilities
35 : * support. So in that case we do not raise capabilities.
36 : *
37 : * Warn if that happens, once per boot.
38 : */
39 0 : static void warn_setuid_and_fcaps_mixed(const char *fname)
40 : {
41 0 : static int warned;
42 0 : if (!warned) {
43 0 : printk(KERN_INFO "warning: `%s' has both setuid-root and"
44 : " effective capabilities. Therefore not raising all"
45 : " capabilities.\n", fname);
46 0 : warned = 1;
47 : }
48 0 : }
49 :
50 : /**
51 : * cap_capable - Determine whether a task has a particular effective capability
52 : * @cred: The credentials to use
53 : * @ns: The user namespace in which we need the capability
54 : * @cap: The capability to check for
55 : * @opts: Bitmask of options defined in include/linux/security.h
56 : *
57 : * Determine whether the nominated task has the specified capability amongst
58 : * its effective set, returning 0 if it does, -ve if it does not.
59 : *
60 : * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
61 : * and has_capability() functions. That is, it has the reverse semantics:
62 : * cap_has_capability() returns 0 when a task has a capability, but the
63 : * kernel's capable() and has_capability() returns 1 for this case.
64 : */
65 89346 : int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
66 : int cap, unsigned int opts)
67 : {
68 89346 : struct user_namespace *ns = targ_ns;
69 :
70 : /* See if cred has the capability in the target user namespace
71 : * by examining the target user namespace and all of the target
72 : * user namespace's parents.
73 : */
74 89346 : for (;;) {
75 : /* Do we have the necessary capabilities? */
76 89346 : if (ns == cred->user_ns)
77 89346 : return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
78 :
79 : /*
80 : * If we're already at a lower level than we're looking for,
81 : * we're done searching.
82 : */
83 0 : if (ns->level <= cred->user_ns->level)
84 : return -EPERM;
85 :
86 : /*
87 : * The owner of the user namespace in the parent of the
88 : * user namespace has all caps.
89 : */
90 0 : if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
91 : return 0;
92 :
93 : /*
94 : * If you have a capability in a parent user ns, then you have
95 : * it over all children user namespaces as well.
96 : */
97 : ns = ns->parent;
98 : }
99 :
100 : /* We never get here */
101 : }
102 :
103 : /**
104 : * cap_settime - Determine whether the current process may set the system clock
105 : * @ts: The time to set
106 : * @tz: The timezone to set
107 : *
108 : * Determine whether the current process may set the system clock and timezone
109 : * information, returning 0 if permission granted, -ve if denied.
110 : */
111 1 : int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
112 : {
113 1 : if (!capable(CAP_SYS_TIME))
114 0 : return -EPERM;
115 : return 0;
116 : }
117 :
118 : /**
119 : * cap_ptrace_access_check - Determine whether the current process may access
120 : * another
121 : * @child: The process to be accessed
122 : * @mode: The mode of attachment.
123 : *
124 : * If we are in the same or an ancestor user_ns and have all the target
125 : * task's capabilities, then ptrace access is allowed.
126 : * If we have the ptrace capability to the target user_ns, then ptrace
127 : * access is allowed.
128 : * Else denied.
129 : *
130 : * Determine whether a process may access another, returning 0 if permission
131 : * granted, -ve if denied.
132 : */
133 229 : int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
134 : {
135 229 : int ret = 0;
136 229 : const struct cred *cred, *child_cred;
137 229 : const kernel_cap_t *caller_caps;
138 :
139 229 : rcu_read_lock();
140 229 : cred = current_cred();
141 229 : child_cred = __task_cred(child);
142 229 : if (mode & PTRACE_MODE_FSCREDS)
143 213 : caller_caps = &cred->cap_effective;
144 : else
145 16 : caller_caps = &cred->cap_permitted;
146 229 : if (cred->user_ns == child_cred->user_ns &&
147 229 : cap_issubset(child_cred->cap_permitted, *caller_caps))
148 180 : goto out;
149 49 : if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
150 48 : goto out;
151 : ret = -EPERM;
152 229 : out:
153 229 : rcu_read_unlock();
154 229 : return ret;
155 : }
156 :
157 : /**
158 : * cap_ptrace_traceme - Determine whether another process may trace the current
159 : * @parent: The task proposed to be the tracer
160 : *
161 : * If parent is in the same or an ancestor user_ns and has all current's
162 : * capabilities, then ptrace access is allowed.
163 : * If parent has the ptrace capability to current's user_ns, then ptrace
164 : * access is allowed.
165 : * Else denied.
166 : *
167 : * Determine whether the nominated task is permitted to trace the current
168 : * process, returning 0 if permission is granted, -ve if denied.
169 : */
170 8 : int cap_ptrace_traceme(struct task_struct *parent)
171 : {
172 8 : int ret = 0;
173 8 : const struct cred *cred, *child_cred;
174 :
175 8 : rcu_read_lock();
176 8 : cred = __task_cred(parent);
177 8 : child_cred = current_cred();
178 8 : if (cred->user_ns == child_cred->user_ns &&
179 8 : cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
180 8 : goto out;
181 0 : if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
182 0 : goto out;
183 : ret = -EPERM;
184 8 : out:
185 8 : rcu_read_unlock();
186 8 : return ret;
187 : }
188 :
189 : /**
190 : * cap_capget - Retrieve a task's capability sets
191 : * @target: The task from which to retrieve the capability sets
192 : * @effective: The place to record the effective set
193 : * @inheritable: The place to record the inheritable set
194 : * @permitted: The place to record the permitted set
195 : *
196 : * This function retrieves the capabilities of the nominated task and returns
197 : * them to the caller.
198 : */
199 522 : int cap_capget(struct task_struct *target, kernel_cap_t *effective,
200 : kernel_cap_t *inheritable, kernel_cap_t *permitted)
201 : {
202 522 : const struct cred *cred;
203 :
204 : /* Derived from kernel/capability.c:sys_capget. */
205 522 : rcu_read_lock();
206 522 : cred = __task_cred(target);
207 522 : *effective = cred->cap_effective;
208 522 : *inheritable = cred->cap_inheritable;
209 522 : *permitted = cred->cap_permitted;
210 522 : rcu_read_unlock();
211 522 : return 0;
212 : }
213 :
214 : /*
215 : * Determine whether the inheritable capabilities are limited to the old
216 : * permitted set. Returns 1 if they are limited, 0 if they are not.
217 : */
218 477 : static inline int cap_inh_is_capped(void)
219 : {
220 : /* they are so limited unless the current task has the CAP_SETPCAP
221 : * capability
222 : */
223 477 : if (cap_capable(current_cred(), current_cred()->user_ns,
224 : CAP_SETPCAP, CAP_OPT_NONE) == 0)
225 63 : return 0;
226 : return 1;
227 : }
228 :
229 : /**
230 : * cap_capset - Validate and apply proposed changes to current's capabilities
231 : * @new: The proposed new credentials; alterations should be made here
232 : * @old: The current task's current credentials
233 : * @effective: A pointer to the proposed new effective capabilities set
234 : * @inheritable: A pointer to the proposed new inheritable capabilities set
235 : * @permitted: A pointer to the proposed new permitted capabilities set
236 : *
237 : * This function validates and applies a proposed mass change to the current
238 : * process's capability sets. The changes are made to the proposed new
239 : * credentials, and assuming no error, will be committed by the caller of LSM.
240 : */
241 477 : int cap_capset(struct cred *new,
242 : const struct cred *old,
243 : const kernel_cap_t *effective,
244 : const kernel_cap_t *inheritable,
245 : const kernel_cap_t *permitted)
246 : {
247 477 : if (cap_inh_is_capped() &&
248 414 : !cap_issubset(*inheritable,
249 : cap_combine(old->cap_inheritable,
250 : old->cap_permitted)))
251 : /* incapable of using this inheritable set */
252 : return -EPERM;
253 :
254 954 : if (!cap_issubset(*inheritable,
255 : cap_combine(old->cap_inheritable,
256 : old->cap_bset)))
257 : /* no new pI capabilities outside bounding set */
258 : return -EPERM;
259 :
260 : /* verify restrictions on target's new Permitted set */
261 477 : if (!cap_issubset(*permitted, old->cap_permitted))
262 : return -EPERM;
263 :
264 : /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
265 477 : if (!cap_issubset(*effective, *permitted))
266 : return -EPERM;
267 :
268 477 : new->cap_effective = *effective;
269 477 : new->cap_inheritable = *inheritable;
270 477 : new->cap_permitted = *permitted;
271 :
272 : /*
273 : * Mask off ambient bits that are no longer both permitted and
274 : * inheritable.
275 : */
276 954 : new->cap_ambient = cap_intersect(new->cap_ambient,
277 : cap_intersect(*permitted,
278 : *inheritable));
279 477 : if (WARN_ON(!cap_ambient_invariant_ok(new)))
280 0 : return -EINVAL;
281 : return 0;
282 : }
283 :
284 : /**
285 : * cap_inode_need_killpriv - Determine if inode change affects privileges
286 : * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
287 : *
288 : * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
289 : * affects the security markings on that inode, and if it is, should
290 : * inode_killpriv() be invoked or the change rejected.
291 : *
292 : * Returns 1 if security.capability has a value, meaning inode_killpriv()
293 : * is required, 0 otherwise, meaning inode_killpriv() is not required.
294 : */
295 845 : int cap_inode_need_killpriv(struct dentry *dentry)
296 : {
297 845 : struct inode *inode = d_backing_inode(dentry);
298 845 : int error;
299 :
300 845 : error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
301 845 : return error > 0;
302 : }
303 :
304 : /**
305 : * cap_inode_killpriv - Erase the security markings on an inode
306 : *
307 : * @mnt_userns: user namespace of the mount the inode was found from
308 : * @dentry: The inode/dentry to alter
309 : *
310 : * Erase the privilege-enhancing security markings on an inode.
311 : *
312 : * If the inode has been found through an idmapped mount the user namespace of
313 : * the vfsmount must be passed through @mnt_userns. This function will then
314 : * take care to map the inode according to @mnt_userns before checking
315 : * permissions. On non-idmapped mounts or if permission checking is to be
316 : * performed on the raw inode simply passs init_user_ns.
317 : *
318 : * Returns 0 if successful, -ve on error.
319 : */
320 0 : int cap_inode_killpriv(struct user_namespace *mnt_userns, struct dentry *dentry)
321 : {
322 0 : int error;
323 :
324 0 : error = __vfs_removexattr(mnt_userns, dentry, XATTR_NAME_CAPS);
325 0 : if (error == -EOPNOTSUPP)
326 0 : error = 0;
327 0 : return error;
328 : }
329 :
330 0 : static bool rootid_owns_currentns(kuid_t kroot)
331 : {
332 0 : struct user_namespace *ns;
333 :
334 0 : if (!uid_valid(kroot))
335 : return false;
336 :
337 0 : for (ns = current_user_ns(); ; ns = ns->parent) {
338 0 : if (from_kuid(ns, kroot) == 0)
339 0 : return true;
340 : if (ns == &init_user_ns)
341 : break;
342 : }
343 :
344 : return false;
345 : }
346 :
347 0 : static __u32 sansflags(__u32 m)
348 : {
349 0 : return m & ~VFS_CAP_FLAGS_EFFECTIVE;
350 : }
351 :
352 0 : static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
353 : {
354 0 : if (size != XATTR_CAPS_SZ_2)
355 : return false;
356 0 : return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
357 : }
358 :
359 0 : static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
360 : {
361 0 : if (size != XATTR_CAPS_SZ_3)
362 : return false;
363 0 : return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
364 : }
365 :
366 : /*
367 : * getsecurity: We are called for security.* before any attempt to read the
368 : * xattr from the inode itself.
369 : *
370 : * This gives us a chance to read the on-disk value and convert it. If we
371 : * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
372 : *
373 : * Note we are not called by vfs_getxattr_alloc(), but that is only called
374 : * by the integrity subsystem, which really wants the unconverted values -
375 : * so that's good.
376 : */
377 0 : int cap_inode_getsecurity(struct user_namespace *mnt_userns,
378 : struct inode *inode, const char *name, void **buffer,
379 : bool alloc)
380 : {
381 0 : int size, ret;
382 0 : kuid_t kroot;
383 0 : u32 nsmagic, magic;
384 0 : uid_t root, mappedroot;
385 0 : char *tmpbuf = NULL;
386 0 : struct vfs_cap_data *cap;
387 0 : struct vfs_ns_cap_data *nscap = NULL;
388 0 : struct dentry *dentry;
389 0 : struct user_namespace *fs_ns;
390 :
391 0 : if (strcmp(name, "capability") != 0)
392 : return -EOPNOTSUPP;
393 :
394 0 : dentry = d_find_any_alias(inode);
395 0 : if (!dentry)
396 : return -EINVAL;
397 :
398 0 : size = sizeof(struct vfs_ns_cap_data);
399 0 : ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS,
400 : &tmpbuf, size, GFP_NOFS);
401 0 : dput(dentry);
402 :
403 0 : if (ret < 0)
404 : return ret;
405 :
406 0 : fs_ns = inode->i_sb->s_user_ns;
407 0 : cap = (struct vfs_cap_data *) tmpbuf;
408 0 : if (is_v2header((size_t) ret, cap)) {
409 : root = 0;
410 0 : } else if (is_v3header((size_t) ret, cap)) {
411 0 : nscap = (struct vfs_ns_cap_data *) tmpbuf;
412 0 : root = le32_to_cpu(nscap->rootid);
413 : } else {
414 0 : size = -EINVAL;
415 0 : goto out_free;
416 : }
417 :
418 0 : kroot = make_kuid(fs_ns, root);
419 :
420 : /* If this is an idmapped mount shift the kuid. */
421 0 : kroot = kuid_into_mnt(mnt_userns, kroot);
422 :
423 : /* If the root kuid maps to a valid uid in current ns, then return
424 : * this as a nscap. */
425 0 : mappedroot = from_kuid(current_user_ns(), kroot);
426 0 : if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
427 0 : size = sizeof(struct vfs_ns_cap_data);
428 0 : if (alloc) {
429 0 : if (!nscap) {
430 : /* v2 -> v3 conversion */
431 0 : nscap = kzalloc(size, GFP_ATOMIC);
432 0 : if (!nscap) {
433 0 : size = -ENOMEM;
434 0 : goto out_free;
435 : }
436 0 : nsmagic = VFS_CAP_REVISION_3;
437 0 : magic = le32_to_cpu(cap->magic_etc);
438 0 : if (magic & VFS_CAP_FLAGS_EFFECTIVE)
439 : nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
440 0 : memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
441 0 : nscap->magic_etc = cpu_to_le32(nsmagic);
442 : } else {
443 : /* use allocated v3 buffer */
444 0 : tmpbuf = NULL;
445 : }
446 0 : nscap->rootid = cpu_to_le32(mappedroot);
447 0 : *buffer = nscap;
448 : }
449 0 : goto out_free;
450 : }
451 :
452 0 : if (!rootid_owns_currentns(kroot)) {
453 0 : size = -EOVERFLOW;
454 0 : goto out_free;
455 : }
456 :
457 : /* This comes from a parent namespace. Return as a v2 capability */
458 0 : size = sizeof(struct vfs_cap_data);
459 0 : if (alloc) {
460 0 : if (nscap) {
461 : /* v3 -> v2 conversion */
462 0 : cap = kzalloc(size, GFP_ATOMIC);
463 0 : if (!cap) {
464 0 : size = -ENOMEM;
465 0 : goto out_free;
466 : }
467 0 : magic = VFS_CAP_REVISION_2;
468 0 : nsmagic = le32_to_cpu(nscap->magic_etc);
469 0 : if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
470 : magic |= VFS_CAP_FLAGS_EFFECTIVE;
471 0 : memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
472 0 : cap->magic_etc = cpu_to_le32(magic);
473 : } else {
474 : /* use unconverted v2 */
475 0 : tmpbuf = NULL;
476 : }
477 0 : *buffer = cap;
478 : }
479 0 : out_free:
480 0 : kfree(tmpbuf);
481 0 : return size;
482 : }
483 :
484 : /**
485 : * rootid_from_xattr - translate root uid of vfs caps
486 : *
487 : * @value: vfs caps value which may be modified by this function
488 : * @size: size of @ivalue
489 : * @task_ns: user namespace of the caller
490 : * @mnt_userns: user namespace of the mount the inode was found from
491 : *
492 : * If the inode has been found through an idmapped mount the user namespace of
493 : * the vfsmount must be passed through @mnt_userns. This function will then
494 : * take care to map the inode according to @mnt_userns before checking
495 : * permissions. On non-idmapped mounts or if permission checking is to be
496 : * performed on the raw inode simply passs init_user_ns.
497 : */
498 0 : static kuid_t rootid_from_xattr(const void *value, size_t size,
499 : struct user_namespace *task_ns,
500 : struct user_namespace *mnt_userns)
501 : {
502 0 : const struct vfs_ns_cap_data *nscap = value;
503 0 : kuid_t rootkid;
504 0 : uid_t rootid = 0;
505 :
506 0 : if (size == XATTR_CAPS_SZ_3)
507 0 : rootid = le32_to_cpu(nscap->rootid);
508 :
509 0 : rootkid = make_kuid(task_ns, rootid);
510 0 : return kuid_from_mnt(mnt_userns, rootkid);
511 : }
512 :
513 0 : static bool validheader(size_t size, const struct vfs_cap_data *cap)
514 : {
515 0 : return is_v2header(size, cap) || is_v3header(size, cap);
516 : }
517 :
518 : /**
519 : * cap_convert_nscap - check vfs caps
520 : *
521 : * @mnt_userns: user namespace of the mount the inode was found from
522 : * @dentry: used to retrieve inode to check permissions on
523 : * @ivalue: vfs caps value which may be modified by this function
524 : * @size: size of @ivalue
525 : *
526 : * User requested a write of security.capability. If needed, update the
527 : * xattr to change from v2 to v3, or to fixup the v3 rootid.
528 : *
529 : * If the inode has been found through an idmapped mount the user namespace of
530 : * the vfsmount must be passed through @mnt_userns. This function will then
531 : * take care to map the inode according to @mnt_userns before checking
532 : * permissions. On non-idmapped mounts or if permission checking is to be
533 : * performed on the raw inode simply passs init_user_ns.
534 : *
535 : * If all is ok, we return the new size, on error return < 0.
536 : */
537 0 : int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
538 : const void **ivalue, size_t size)
539 : {
540 0 : struct vfs_ns_cap_data *nscap;
541 0 : uid_t nsrootid;
542 0 : const struct vfs_cap_data *cap = *ivalue;
543 0 : __u32 magic, nsmagic;
544 0 : struct inode *inode = d_backing_inode(dentry);
545 0 : struct user_namespace *task_ns = current_user_ns(),
546 0 : *fs_ns = inode->i_sb->s_user_ns;
547 0 : kuid_t rootid;
548 0 : size_t newsize;
549 :
550 0 : if (!*ivalue)
551 : return -EINVAL;
552 0 : if (!validheader(size, cap))
553 : return -EINVAL;
554 0 : if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
555 : return -EPERM;
556 0 : if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns))
557 0 : if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
558 : /* user is privileged, just write the v2 */
559 : return size;
560 :
561 0 : rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns);
562 0 : if (!uid_valid(rootid))
563 : return -EINVAL;
564 :
565 0 : nsrootid = from_kuid(fs_ns, rootid);
566 0 : if (nsrootid == -1)
567 : return -EINVAL;
568 :
569 0 : newsize = sizeof(struct vfs_ns_cap_data);
570 0 : nscap = kmalloc(newsize, GFP_ATOMIC);
571 0 : if (!nscap)
572 : return -ENOMEM;
573 0 : nscap->rootid = cpu_to_le32(nsrootid);
574 0 : nsmagic = VFS_CAP_REVISION_3;
575 0 : magic = le32_to_cpu(cap->magic_etc);
576 0 : if (magic & VFS_CAP_FLAGS_EFFECTIVE)
577 : nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
578 0 : nscap->magic_etc = cpu_to_le32(nsmagic);
579 0 : memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
580 :
581 0 : *ivalue = nscap;
582 0 : return newsize;
583 : }
584 :
585 : /*
586 : * Calculate the new process capability sets from the capability sets attached
587 : * to a file.
588 : */
589 0 : static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
590 : struct linux_binprm *bprm,
591 : bool *effective,
592 : bool *has_fcap)
593 : {
594 0 : struct cred *new = bprm->cred;
595 0 : unsigned i;
596 0 : int ret = 0;
597 :
598 0 : if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
599 0 : *effective = true;
600 :
601 0 : if (caps->magic_etc & VFS_CAP_REVISION_MASK)
602 0 : *has_fcap = true;
603 :
604 0 : CAP_FOR_EACH_U32(i) {
605 0 : __u32 permitted = caps->permitted.cap[i];
606 0 : __u32 inheritable = caps->inheritable.cap[i];
607 :
608 : /*
609 : * pP' = (X & fP) | (pI & fI)
610 : * The addition of pA' is handled later.
611 : */
612 0 : new->cap_permitted.cap[i] =
613 0 : (new->cap_bset.cap[i] & permitted) |
614 0 : (new->cap_inheritable.cap[i] & inheritable);
615 :
616 0 : if (permitted & ~new->cap_permitted.cap[i])
617 : /* insufficient to execute correctly */
618 0 : ret = -EPERM;
619 : }
620 :
621 : /*
622 : * For legacy apps, with no internal support for recognizing they
623 : * do not have enough capabilities, we return an error if they are
624 : * missing some "forced" (aka file-permitted) capabilities.
625 : */
626 0 : return *effective ? ret : 0;
627 : }
628 :
629 : /**
630 : * get_vfs_caps_from_disk - retrieve vfs caps from disk
631 : *
632 : * @mnt_userns: user namespace of the mount the inode was found from
633 : * @dentry: dentry from which @inode is retrieved
634 : * @cpu_caps: vfs capabilities
635 : *
636 : * Extract the on-exec-apply capability sets for an executable file.
637 : *
638 : * If the inode has been found through an idmapped mount the user namespace of
639 : * the vfsmount must be passed through @mnt_userns. This function will then
640 : * take care to map the inode according to @mnt_userns before checking
641 : * permissions. On non-idmapped mounts or if permission checking is to be
642 : * performed on the raw inode simply passs init_user_ns.
643 : */
644 1367 : int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
645 : const struct dentry *dentry,
646 : struct cpu_vfs_cap_data *cpu_caps)
647 : {
648 1367 : struct inode *inode = d_backing_inode(dentry);
649 1367 : __u32 magic_etc;
650 1367 : unsigned tocopy, i;
651 1367 : int size;
652 1367 : struct vfs_ns_cap_data data, *nscaps = &data;
653 1367 : struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
654 1367 : kuid_t rootkuid;
655 1367 : struct user_namespace *fs_ns;
656 :
657 1367 : memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
658 :
659 1367 : if (!inode)
660 : return -ENODATA;
661 :
662 1367 : fs_ns = inode->i_sb->s_user_ns;
663 1367 : size = __vfs_getxattr((struct dentry *)dentry, inode,
664 : XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
665 1367 : if (size == -ENODATA || size == -EOPNOTSUPP)
666 : /* no data, that's ok */
667 : return -ENODATA;
668 :
669 0 : if (size < 0)
670 : return size;
671 :
672 0 : if (size < sizeof(magic_etc))
673 : return -EINVAL;
674 :
675 0 : cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
676 :
677 0 : rootkuid = make_kuid(fs_ns, 0);
678 0 : switch (magic_etc & VFS_CAP_REVISION_MASK) {
679 0 : case VFS_CAP_REVISION_1:
680 0 : if (size != XATTR_CAPS_SZ_1)
681 : return -EINVAL;
682 : tocopy = VFS_CAP_U32_1;
683 : break;
684 0 : case VFS_CAP_REVISION_2:
685 0 : if (size != XATTR_CAPS_SZ_2)
686 : return -EINVAL;
687 : tocopy = VFS_CAP_U32_2;
688 : break;
689 0 : case VFS_CAP_REVISION_3:
690 0 : if (size != XATTR_CAPS_SZ_3)
691 : return -EINVAL;
692 0 : tocopy = VFS_CAP_U32_3;
693 0 : rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
694 : break;
695 :
696 : default:
697 : return -EINVAL;
698 : }
699 : /* Limit the caps to the mounter of the filesystem
700 : * or the more limited uid specified in the xattr.
701 : */
702 0 : rootkuid = kuid_into_mnt(mnt_userns, rootkuid);
703 0 : if (!rootid_owns_currentns(rootkuid))
704 : return -ENODATA;
705 :
706 0 : CAP_FOR_EACH_U32(i) {
707 0 : if (i >= tocopy)
708 : break;
709 0 : cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
710 0 : cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
711 : }
712 :
713 0 : cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
714 0 : cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
715 :
716 0 : cpu_caps->rootid = rootkuid;
717 :
718 0 : return 0;
719 : }
720 :
721 : /*
722 : * Attempt to get the on-exec apply capability sets for an executable file from
723 : * its xattrs and, if present, apply them to the proposed credentials being
724 : * constructed by execve().
725 : */
726 1367 : static int get_file_caps(struct linux_binprm *bprm, struct file *file,
727 : bool *effective, bool *has_fcap)
728 : {
729 1367 : int rc = 0;
730 1367 : struct cpu_vfs_cap_data vcaps;
731 :
732 1367 : cap_clear(bprm->cred->cap_permitted);
733 :
734 1367 : if (!file_caps_enabled)
735 : return 0;
736 :
737 1367 : if (!mnt_may_suid(file->f_path.mnt))
738 : return 0;
739 :
740 : /*
741 : * This check is redundant with mnt_may_suid() but is kept to make
742 : * explicit that capability bits are limited to s_user_ns and its
743 : * descendants.
744 : */
745 1367 : if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
746 : return 0;
747 :
748 1367 : rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
749 1367 : file->f_path.dentry, &vcaps);
750 1367 : if (rc < 0) {
751 1367 : if (rc == -EINVAL)
752 0 : printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
753 : bprm->filename);
754 1367 : else if (rc == -ENODATA)
755 1367 : rc = 0;
756 0 : goto out;
757 : }
758 :
759 0 : rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
760 :
761 0 : out:
762 1367 : if (rc)
763 0 : cap_clear(bprm->cred->cap_permitted);
764 :
765 : return rc;
766 : }
767 :
768 4054 : static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
769 :
770 2715 : static inline bool __is_real(kuid_t uid, struct cred *cred)
771 1370 : { return uid_eq(cred->uid, uid); }
772 :
773 4079 : static inline bool __is_eff(kuid_t uid, struct cred *cred)
774 1345 : { return uid_eq(cred->euid, uid); }
775 :
776 1345 : static inline bool __is_suid(kuid_t uid, struct cred *cred)
777 1345 : { return !__is_real(uid, cred) && __is_eff(uid, cred); }
778 :
779 : /*
780 : * handle_privileged_root - Handle case of privileged root
781 : * @bprm: The execution parameters, including the proposed creds
782 : * @has_fcap: Are any file capabilities set?
783 : * @effective: Do we have effective root privilege?
784 : * @root_uid: This namespace' root UID WRT initial USER namespace
785 : *
786 : * Handle the case where root is privileged and hasn't been neutered by
787 : * SECURE_NOROOT. If file capabilities are set, they won't be combined with
788 : * set UID root and nothing is changed. If we are root, cap_permitted is
789 : * updated. If we have become set UID root, the effective bit is set.
790 : */
791 1367 : static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
792 : bool *effective, kuid_t root_uid)
793 : {
794 1367 : const struct cred *old = current_cred();
795 1367 : struct cred *new = bprm->cred;
796 :
797 1367 : if (!root_privileged())
798 : return;
799 : /*
800 : * If the legacy file capability is set, then don't set privs
801 : * for a setuid root binary run by a non-root user. Do set it
802 : * for a root user just to cause least surprise to an admin.
803 : */
804 1367 : if (has_fcap && __is_suid(root_uid, new)) {
805 0 : warn_setuid_and_fcaps_mixed(bprm->filename);
806 0 : return;
807 : }
808 : /*
809 : * To support inheritance of root-permissions and suid-root
810 : * executables under compatibility mode, we override the
811 : * capability sets for the file.
812 : */
813 1367 : if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
814 : /* pP' = (cap_bset & ~0) | (pI & ~0) */
815 2728 : new->cap_permitted = cap_combine(old->cap_bset,
816 : old->cap_inheritable);
817 : }
818 : /*
819 : * If only the real uid is 0, we do not set the effective bit.
820 : */
821 1367 : if (__is_eff(root_uid, new))
822 1364 : *effective = true;
823 : }
824 :
825 : #define __cap_gained(field, target, source) \
826 : !cap_issubset(target->cap_##field, source->cap_##field)
827 : #define __cap_grew(target, source, cred) \
828 : !cap_issubset(cred->cap_##target, cred->cap_##source)
829 : #define __cap_full(field, cred) \
830 : cap_issubset(CAP_FULL_SET, cred->cap_##field)
831 :
832 2712 : static inline bool __is_setuid(struct cred *new, const struct cred *old)
833 1345 : { return !uid_eq(new->euid, old->uid); }
834 :
835 1367 : static inline bool __is_setgid(struct cred *new, const struct cred *old)
836 1367 : { return !gid_eq(new->egid, old->gid); }
837 :
838 : /*
839 : * 1) Audit candidate if current->cap_effective is set
840 : *
841 : * We do not bother to audit if 3 things are true:
842 : * 1) cap_effective has all caps
843 : * 2) we became root *OR* are were already root
844 : * 3) root is supposed to have all caps (SECURE_NOROOT)
845 : * Since this is just a normal root execing a process.
846 : *
847 : * Number 1 above might fail if you don't have a full bset, but I think
848 : * that is interesting information to audit.
849 : *
850 : * A number of other conditions require logging:
851 : * 2) something prevented setuid root getting all caps
852 : * 3) non-setuid root gets fcaps
853 : * 4) non-setuid root gets ambient
854 : */
855 1367 : static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
856 : kuid_t root, bool has_fcap)
857 : {
858 1367 : bool ret = false;
859 :
860 1367 : if ((__cap_grew(effective, ambient, new) &&
861 2706 : !(__cap_full(effective, new) &&
862 1342 : (__is_eff(root, new) || __is_real(root, new)) &&
863 2687 : root_privileged())) ||
864 1345 : (root_privileged() &&
865 1345 : __is_suid(root, new) &&
866 0 : !__cap_full(effective, new)) ||
867 1345 : (!__is_setuid(new, old) &&
868 0 : ((has_fcap &&
869 0 : __cap_gained(permitted, new, old)) ||
870 1345 : __cap_gained(ambient, new, old))))
871 :
872 : ret = true;
873 :
874 1367 : return ret;
875 : }
876 :
877 : /**
878 : * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
879 : * @bprm: The execution parameters, including the proposed creds
880 : * @file: The file to pull the credentials from
881 : *
882 : * Set up the proposed credentials for a new execution context being
883 : * constructed by execve(). The proposed creds in @bprm->cred is altered,
884 : * which won't take effect immediately. Returns 0 if successful, -ve on error.
885 : */
886 1367 : int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
887 : {
888 : /* Process setpcap binaries and capabilities for uid 0 */
889 1367 : const struct cred *old = current_cred();
890 1367 : struct cred *new = bprm->cred;
891 1367 : bool effective = false, has_fcap = false, is_setid;
892 1367 : int ret;
893 1367 : kuid_t root_uid;
894 :
895 1367 : if (WARN_ON(!cap_ambient_invariant_ok(old)))
896 : return -EPERM;
897 :
898 1367 : ret = get_file_caps(bprm, file, &effective, &has_fcap);
899 1367 : if (ret < 0)
900 : return ret;
901 :
902 1367 : root_uid = make_kuid(new->user_ns, 0);
903 :
904 1367 : handle_privileged_root(bprm, has_fcap, &effective, root_uid);
905 :
906 : /* if we have fs caps, clear dangerous personality flags */
907 1367 : if (__cap_gained(permitted, new, old))
908 2 : bprm->per_clear |= PER_CLEAR_ON_SETID;
909 :
910 : /* Don't let someone trace a set[ug]id/setpcap binary with the revised
911 : * credentials unless they have the appropriate permit.
912 : *
913 : * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
914 : */
915 1367 : is_setid = __is_setuid(new, old) || __is_setgid(new, old);
916 :
917 1367 : if ((is_setid || __cap_gained(permitted, new, old)) &&
918 2 : ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
919 0 : !ptracer_capable(current, new->user_ns))) {
920 : /* downgrade; they get no more than they had, and maybe less */
921 2 : if (!ns_capable(new->user_ns, CAP_SETUID) ||
922 0 : (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
923 2 : new->euid = new->uid;
924 2 : new->egid = new->gid;
925 : }
926 4 : new->cap_permitted = cap_intersect(new->cap_permitted,
927 : old->cap_permitted);
928 : }
929 :
930 1367 : new->suid = new->fsuid = new->euid;
931 1367 : new->sgid = new->fsgid = new->egid;
932 :
933 : /* File caps or setid cancels ambient. */
934 1367 : if (has_fcap || is_setid)
935 0 : cap_clear(new->cap_ambient);
936 :
937 : /*
938 : * Now that we've computed pA', update pP' to give:
939 : * pP' = (X & fP) | (pI & fI) | pA'
940 : */
941 1367 : new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
942 :
943 : /*
944 : * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
945 : * this is the same as pE' = (fE ? pP' : 0) | pA'.
946 : */
947 1367 : if (effective)
948 1364 : new->cap_effective = new->cap_permitted;
949 : else
950 3 : new->cap_effective = new->cap_ambient;
951 :
952 1367 : if (WARN_ON(!cap_ambient_invariant_ok(new)))
953 : return -EPERM;
954 :
955 1367 : if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
956 1367 : ret = audit_log_bprm_fcaps(bprm, new, old);
957 : if (ret < 0)
958 : return ret;
959 : }
960 :
961 1367 : new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
962 :
963 1367 : if (WARN_ON(!cap_ambient_invariant_ok(new)))
964 : return -EPERM;
965 :
966 : /* Check for privilege-elevated exec. */
967 1367 : if (is_setid ||
968 1367 : (!__is_real(root_uid, new) &&
969 3 : (effective ||
970 3 : __cap_grew(permitted, ambient, new))))
971 0 : bprm->secureexec = 1;
972 :
973 : return 0;
974 : }
975 :
976 : /**
977 : * cap_inode_setxattr - Determine whether an xattr may be altered
978 : * @dentry: The inode/dentry being altered
979 : * @name: The name of the xattr to be changed
980 : * @value: The value that the xattr will be changed to
981 : * @size: The size of value
982 : * @flags: The replacement flag
983 : *
984 : * Determine whether an xattr may be altered or set on an inode, returning 0 if
985 : * permission is granted, -ve if denied.
986 : *
987 : * This is used to make sure security xattrs don't get updated or set by those
988 : * who aren't privileged to do so.
989 : */
990 61 : int cap_inode_setxattr(struct dentry *dentry, const char *name,
991 : const void *value, size_t size, int flags)
992 : {
993 61 : struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
994 :
995 : /* Ignore non-security xattrs */
996 61 : if (strncmp(name, XATTR_SECURITY_PREFIX,
997 : XATTR_SECURITY_PREFIX_LEN) != 0)
998 : return 0;
999 :
1000 : /*
1001 : * For XATTR_NAME_CAPS the check will be done in
1002 : * cap_convert_nscap(), called by setxattr()
1003 : */
1004 0 : if (strcmp(name, XATTR_NAME_CAPS) == 0)
1005 : return 0;
1006 :
1007 0 : if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1008 0 : return -EPERM;
1009 : return 0;
1010 : }
1011 :
1012 : /**
1013 : * cap_inode_removexattr - Determine whether an xattr may be removed
1014 : *
1015 : * @mnt_userns: User namespace of the mount the inode was found from
1016 : * @dentry: The inode/dentry being altered
1017 : * @name: The name of the xattr to be changed
1018 : *
1019 : * Determine whether an xattr may be removed from an inode, returning 0 if
1020 : * permission is granted, -ve if denied.
1021 : *
1022 : * If the inode has been found through an idmapped mount the user namespace of
1023 : * the vfsmount must be passed through @mnt_userns. This function will then
1024 : * take care to map the inode according to @mnt_userns before checking
1025 : * permissions. On non-idmapped mounts or if permission checking is to be
1026 : * performed on the raw inode simply passs init_user_ns.
1027 : *
1028 : * This is used to make sure security xattrs don't get removed by those who
1029 : * aren't privileged to remove them.
1030 : */
1031 8 : int cap_inode_removexattr(struct user_namespace *mnt_userns,
1032 : struct dentry *dentry, const char *name)
1033 : {
1034 8 : struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1035 :
1036 : /* Ignore non-security xattrs */
1037 8 : if (strncmp(name, XATTR_SECURITY_PREFIX,
1038 : XATTR_SECURITY_PREFIX_LEN) != 0)
1039 : return 0;
1040 :
1041 0 : if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1042 : /* security.capability gets namespaced */
1043 0 : struct inode *inode = d_backing_inode(dentry);
1044 0 : if (!inode)
1045 : return -EINVAL;
1046 0 : if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1047 : return -EPERM;
1048 0 : return 0;
1049 : }
1050 :
1051 0 : if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1052 0 : return -EPERM;
1053 : return 0;
1054 : }
1055 :
1056 : /*
1057 : * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1058 : * a process after a call to setuid, setreuid, or setresuid.
1059 : *
1060 : * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1061 : * {r,e,s}uid != 0, the permitted and effective capabilities are
1062 : * cleared.
1063 : *
1064 : * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1065 : * capabilities of the process are cleared.
1066 : *
1067 : * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1068 : * capabilities are set to the permitted capabilities.
1069 : *
1070 : * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1071 : * never happen.
1072 : *
1073 : * -astor
1074 : *
1075 : * cevans - New behaviour, Oct '99
1076 : * A process may, via prctl(), elect to keep its capabilities when it
1077 : * calls setuid() and switches away from uid==0. Both permitted and
1078 : * effective sets will be retained.
1079 : * Without this change, it was impossible for a daemon to drop only some
1080 : * of its privilege. The call to setuid(!=0) would drop all privileges!
1081 : * Keeping uid 0 is not an option because uid 0 owns too many vital
1082 : * files..
1083 : * Thanks to Olaf Kirch and Peter Benie for spotting this.
1084 : */
1085 61 : static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1086 : {
1087 61 : kuid_t root_uid = make_kuid(old->user_ns, 0);
1088 :
1089 61 : if ((uid_eq(old->uid, root_uid) ||
1090 5 : uid_eq(old->euid, root_uid) ||
1091 0 : uid_eq(old->suid, root_uid)) &&
1092 61 : (!uid_eq(new->uid, root_uid) &&
1093 16 : !uid_eq(new->euid, root_uid) &&
1094 11 : !uid_eq(new->suid, root_uid))) {
1095 11 : if (!issecure(SECURE_KEEP_CAPS)) {
1096 9 : cap_clear(new->cap_permitted);
1097 9 : cap_clear(new->cap_effective);
1098 : }
1099 :
1100 : /*
1101 : * Pre-ambient programs expect setresuid to nonroot followed
1102 : * by exec to drop capabilities. We should make sure that
1103 : * this remains the case.
1104 : */
1105 11 : cap_clear(new->cap_ambient);
1106 : }
1107 61 : if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1108 17 : cap_clear(new->cap_effective);
1109 61 : if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1110 5 : new->cap_effective = new->cap_permitted;
1111 61 : }
1112 :
1113 : /**
1114 : * cap_task_fix_setuid - Fix up the results of setuid() call
1115 : * @new: The proposed credentials
1116 : * @old: The current task's current credentials
1117 : * @flags: Indications of what has changed
1118 : *
1119 : * Fix up the results of setuid() call before the credential changes are
1120 : * actually applied, returning 0 to grant the changes, -ve to deny them.
1121 : */
1122 61 : int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1123 : {
1124 61 : switch (flags) {
1125 61 : case LSM_SETID_RE:
1126 : case LSM_SETID_ID:
1127 : case LSM_SETID_RES:
1128 : /* juggle the capabilities to follow [RES]UID changes unless
1129 : * otherwise suppressed */
1130 61 : if (!issecure(SECURE_NO_SETUID_FIXUP))
1131 61 : cap_emulate_setxuid(new, old);
1132 : break;
1133 :
1134 0 : case LSM_SETID_FS:
1135 : /* juggle the capabilties to follow FSUID changes, unless
1136 : * otherwise suppressed
1137 : *
1138 : * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1139 : * if not, we might be a bit too harsh here.
1140 : */
1141 0 : if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1142 0 : kuid_t root_uid = make_kuid(old->user_ns, 0);
1143 0 : if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1144 0 : new->cap_effective =
1145 0 : cap_drop_fs_set(new->cap_effective);
1146 :
1147 0 : if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1148 0 : new->cap_effective =
1149 0 : cap_raise_fs_set(new->cap_effective,
1150 : new->cap_permitted);
1151 : }
1152 : break;
1153 :
1154 : default:
1155 : return -EINVAL;
1156 : }
1157 :
1158 : return 0;
1159 : }
1160 :
1161 : /*
1162 : * Rationale: code calling task_setscheduler, task_setioprio, and
1163 : * task_setnice, assumes that
1164 : * . if capable(cap_sys_nice), then those actions should be allowed
1165 : * . if not capable(cap_sys_nice), but acting on your own processes,
1166 : * then those actions should be allowed
1167 : * This is insufficient now since you can call code without suid, but
1168 : * yet with increased caps.
1169 : * So we check for increased caps on the target process.
1170 : */
1171 18 : static int cap_safe_nice(struct task_struct *p)
1172 : {
1173 18 : int is_subset, ret = 0;
1174 :
1175 18 : rcu_read_lock();
1176 18 : is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1177 18 : current_cred()->cap_permitted);
1178 18 : if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1179 0 : ret = -EPERM;
1180 18 : rcu_read_unlock();
1181 :
1182 18 : return ret;
1183 : }
1184 :
1185 : /**
1186 : * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1187 : * @p: The task to affect
1188 : *
1189 : * Detemine if the requested scheduler policy change is permitted for the
1190 : * specified task, returning 0 if permission is granted, -ve if denied.
1191 : */
1192 3 : int cap_task_setscheduler(struct task_struct *p)
1193 : {
1194 3 : return cap_safe_nice(p);
1195 : }
1196 :
1197 : /**
1198 : * cap_task_ioprio - Detemine if I/O priority change is permitted
1199 : * @p: The task to affect
1200 : * @ioprio: The I/O priority to set
1201 : *
1202 : * Detemine if the requested I/O priority change is permitted for the specified
1203 : * task, returning 0 if permission is granted, -ve if denied.
1204 : */
1205 6 : int cap_task_setioprio(struct task_struct *p, int ioprio)
1206 : {
1207 6 : return cap_safe_nice(p);
1208 : }
1209 :
1210 : /**
1211 : * cap_task_ioprio - Detemine if task priority change is permitted
1212 : * @p: The task to affect
1213 : * @nice: The nice value to set
1214 : *
1215 : * Detemine if the requested task priority change is permitted for the
1216 : * specified task, returning 0 if permission is granted, -ve if denied.
1217 : */
1218 9 : int cap_task_setnice(struct task_struct *p, int nice)
1219 : {
1220 9 : return cap_safe_nice(p);
1221 : }
1222 :
1223 : /*
1224 : * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1225 : * the current task's bounding set. Returns 0 on success, -ve on error.
1226 : */
1227 104 : static int cap_prctl_drop(unsigned long cap)
1228 : {
1229 104 : struct cred *new;
1230 :
1231 104 : if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1232 : return -EPERM;
1233 104 : if (!cap_valid(cap))
1234 : return -EINVAL;
1235 :
1236 104 : new = prepare_creds();
1237 104 : if (!new)
1238 : return -ENOMEM;
1239 104 : cap_lower(new->cap_bset, cap);
1240 104 : return commit_creds(new);
1241 : }
1242 :
1243 : /**
1244 : * cap_task_prctl - Implement process control functions for this security module
1245 : * @option: The process control function requested
1246 : * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1247 : *
1248 : * Allow process control functions (sys_prctl()) to alter capabilities; may
1249 : * also deny access to other functions not otherwise implemented here.
1250 : *
1251 : * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1252 : * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1253 : * modules will consider performing the function.
1254 : */
1255 696 : int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1256 : unsigned long arg4, unsigned long arg5)
1257 : {
1258 696 : const struct cred *old = current_cred();
1259 696 : struct cred *new;
1260 :
1261 696 : switch (option) {
1262 0 : case PR_CAPBSET_READ:
1263 0 : if (!cap_valid(arg2))
1264 : return -EINVAL;
1265 0 : return !!cap_raised(old->cap_bset, arg2);
1266 :
1267 104 : case PR_CAPBSET_DROP:
1268 104 : return cap_prctl_drop(arg2);
1269 :
1270 : /*
1271 : * The next four prctl's remain to assist with transitioning a
1272 : * system from legacy UID=0 based privilege (when filesystem
1273 : * capabilities are not in use) to a system using filesystem
1274 : * capabilities only - as the POSIX.1e draft intended.
1275 : *
1276 : * Note:
1277 : *
1278 : * PR_SET_SECUREBITS =
1279 : * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1280 : * | issecure_mask(SECURE_NOROOT)
1281 : * | issecure_mask(SECURE_NOROOT_LOCKED)
1282 : * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1283 : * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1284 : *
1285 : * will ensure that the current process and all of its
1286 : * children will be locked into a pure
1287 : * capability-based-privilege environment.
1288 : */
1289 1 : case PR_SET_SECUREBITS:
1290 1 : if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1291 1 : & (old->securebits ^ arg2)) /*[1]*/
1292 1 : || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1293 1 : || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1294 1 : || (cap_capable(current_cred(),
1295 1 : current_cred()->user_ns,
1296 : CAP_SETPCAP,
1297 : CAP_OPT_NONE) != 0) /*[4]*/
1298 : /*
1299 : * [1] no changing of bits that are locked
1300 : * [2] no unlocking of locks
1301 : * [3] no setting of unsupported bits
1302 : * [4] doing anything requires privilege (go read about
1303 : * the "sendmail capabilities bug")
1304 : */
1305 : )
1306 : /* cannot change a locked bit */
1307 0 : return -EPERM;
1308 :
1309 1 : new = prepare_creds();
1310 1 : if (!new)
1311 : return -ENOMEM;
1312 1 : new->securebits = arg2;
1313 1 : return commit_creds(new);
1314 :
1315 47 : case PR_GET_SECUREBITS:
1316 47 : return old->securebits;
1317 :
1318 0 : case PR_GET_KEEPCAPS:
1319 0 : return !!issecure(SECURE_KEEP_CAPS);
1320 :
1321 2 : case PR_SET_KEEPCAPS:
1322 2 : if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1323 : return -EINVAL;
1324 2 : if (issecure(SECURE_KEEP_CAPS_LOCKED))
1325 : return -EPERM;
1326 :
1327 2 : new = prepare_creds();
1328 2 : if (!new)
1329 : return -ENOMEM;
1330 2 : if (arg2)
1331 1 : new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1332 : else
1333 1 : new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1334 2 : return commit_creds(new);
1335 :
1336 3 : case PR_CAP_AMBIENT:
1337 3 : if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1338 0 : if (arg3 | arg4 | arg5)
1339 : return -EINVAL;
1340 :
1341 0 : new = prepare_creds();
1342 0 : if (!new)
1343 : return -ENOMEM;
1344 0 : cap_clear(new->cap_ambient);
1345 0 : return commit_creds(new);
1346 : }
1347 :
1348 3 : if (((!cap_valid(arg3)) | arg4 | arg5))
1349 : return -EINVAL;
1350 :
1351 3 : if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1352 1 : return !!cap_raised(current_cred()->cap_ambient, arg3);
1353 2 : } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1354 : arg2 != PR_CAP_AMBIENT_LOWER) {
1355 : return -EINVAL;
1356 : } else {
1357 4 : if (arg2 == PR_CAP_AMBIENT_RAISE &&
1358 4 : (!cap_raised(current_cred()->cap_permitted, arg3) ||
1359 2 : !cap_raised(current_cred()->cap_inheritable,
1360 2 : arg3) ||
1361 2 : issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1362 0 : return -EPERM;
1363 :
1364 2 : new = prepare_creds();
1365 2 : if (!new)
1366 : return -ENOMEM;
1367 2 : if (arg2 == PR_CAP_AMBIENT_RAISE)
1368 2 : cap_raise(new->cap_ambient, arg3);
1369 : else
1370 0 : cap_lower(new->cap_ambient, arg3);
1371 2 : return commit_creds(new);
1372 : }
1373 :
1374 : default:
1375 : /* No functionality available - continue with default */
1376 : return -ENOSYS;
1377 : }
1378 : }
1379 :
1380 : /**
1381 : * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1382 : * @mm: The VM space in which the new mapping is to be made
1383 : * @pages: The size of the mapping
1384 : *
1385 : * Determine whether the allocation of a new virtual mapping by the current
1386 : * task is permitted, returning 1 if permission is granted, 0 if not.
1387 : */
1388 58420 : int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1389 : {
1390 58420 : int cap_sys_admin = 0;
1391 :
1392 58420 : if (cap_capable(current_cred(), &init_user_ns,
1393 : CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1394 56854 : cap_sys_admin = 1;
1395 :
1396 58422 : return cap_sys_admin;
1397 : }
1398 :
1399 : /*
1400 : * cap_mmap_addr - check if able to map given addr
1401 : * @addr: address attempting to be mapped
1402 : *
1403 : * If the process is attempting to map memory below dac_mmap_min_addr they need
1404 : * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1405 : * capability security module. Returns 0 if this mapping should be allowed
1406 : * -EPERM if not.
1407 : */
1408 40701 : int cap_mmap_addr(unsigned long addr)
1409 : {
1410 40701 : int ret = 0;
1411 :
1412 40701 : if (addr < dac_mmap_min_addr) {
1413 0 : ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1414 : CAP_OPT_NONE);
1415 : /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1416 0 : if (ret == 0)
1417 0 : current->flags |= PF_SUPERPRIV;
1418 : }
1419 40701 : return ret;
1420 : }
1421 :
1422 35741 : int cap_mmap_file(struct file *file, unsigned long reqprot,
1423 : unsigned long prot, unsigned long flags)
1424 : {
1425 35741 : return 0;
1426 : }
1427 :
1428 : #ifdef CONFIG_SECURITY
1429 :
1430 : static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1431 : LSM_HOOK_INIT(capable, cap_capable),
1432 : LSM_HOOK_INIT(settime, cap_settime),
1433 : LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1434 : LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1435 : LSM_HOOK_INIT(capget, cap_capget),
1436 : LSM_HOOK_INIT(capset, cap_capset),
1437 : LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1438 : LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1439 : LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1440 : LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1441 : LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1442 : LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1443 : LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1444 : LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1445 : LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1446 : LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1447 : LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1448 : LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1449 : };
1450 :
1451 1 : static int __init capability_init(void)
1452 : {
1453 1 : security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1454 : "capability");
1455 1 : return 0;
1456 : }
1457 :
1458 : DEFINE_LSM(capability) = {
1459 : .name = "capability",
1460 : .order = LSM_ORDER_FIRST,
1461 : .init = capability_init,
1462 : };
1463 :
1464 : #endif /* CONFIG_SECURITY */
|
