]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
3e1c2515 | 2 | /* Common capabilities, needed by capability.o. |
1da177e4 LT |
3 | */ |
4 | ||
c59ede7b | 5 | #include <linux/capability.h> |
3fc689e9 | 6 | #include <linux/audit.h> |
1da177e4 LT |
7 | #include <linux/init.h> |
8 | #include <linux/kernel.h> | |
b1d9e6b0 | 9 | #include <linux/lsm_hooks.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
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> | |
b5376771 | 20 | #include <linux/mount.h> |
b460cbc5 | 21 | #include <linux/sched.h> |
3898b1b4 AM |
22 | #include <linux/prctl.h> |
23 | #include <linux/securebits.h> | |
3486740a | 24 | #include <linux/user_namespace.h> |
40401530 | 25 | #include <linux/binfmts.h> |
51b79bee | 26 | #include <linux/personality.h> |
72c2d582 | 27 | |
b5f22a59 SH |
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 | */ | |
d7627467 | 39 | static void warn_setuid_and_fcaps_mixed(const char *fname) |
b5f22a59 SH |
40 | { |
41 | static int warned; | |
42 | if (!warned) { | |
43 | printk(KERN_INFO "warning: `%s' has both setuid-root and" | |
44 | " effective capabilities. Therefore not raising all" | |
45 | " capabilities.\n", fname); | |
46 | warned = 1; | |
47 | } | |
48 | } | |
49 | ||
1d045980 DH |
50 | /** |
51 | * cap_capable - Determine whether a task has a particular effective capability | |
3699c53c | 52 | * @cred: The credentials to use |
3486740a | 53 | * @ns: The user namespace in which we need the capability |
1d045980 | 54 | * @cap: The capability to check for |
e88ed488 | 55 | * @opts: Bitmask of options defined in include/linux/security.h |
1d045980 DH |
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 | * | |
3699c53c DH |
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. | |
a6dbb1ef | 64 | */ |
6a9de491 | 65 | int cap_capable(const struct cred *cred, struct user_namespace *targ_ns, |
c1a85a00 | 66 | int cap, unsigned int opts) |
1da177e4 | 67 | { |
520d9eab | 68 | struct user_namespace *ns = targ_ns; |
3486740a | 69 | |
520d9eab EB |
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 | for (;;) { | |
3486740a | 75 | /* Do we have the necessary capabilities? */ |
520d9eab | 76 | if (ns == cred->user_ns) |
3486740a SH |
77 | return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM; |
78 | ||
64db4c7f KT |
79 | /* |
80 | * If we're already at a lower level than we're looking for, | |
81 | * we're done searching. | |
82 | */ | |
83 | if (ns->level <= cred->user_ns->level) | |
3486740a SH |
84 | return -EPERM; |
85 | ||
520d9eab EB |
86 | /* |
87 | * The owner of the user namespace in the parent of the | |
88 | * user namespace has all caps. | |
89 | */ | |
90 | if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid)) | |
91 | return 0; | |
92 | ||
3486740a | 93 | /* |
520d9eab | 94 | * If you have a capability in a parent user ns, then you have |
3486740a SH |
95 | * it over all children user namespaces as well. |
96 | */ | |
520d9eab | 97 | ns = ns->parent; |
3486740a SH |
98 | } |
99 | ||
100 | /* We never get here */ | |
1da177e4 LT |
101 | } |
102 | ||
1d045980 DH |
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 | */ | |
457db29b | 111 | int cap_settime(const struct timespec64 *ts, const struct timezone *tz) |
1da177e4 LT |
112 | { |
113 | if (!capable(CAP_SYS_TIME)) | |
114 | return -EPERM; | |
115 | return 0; | |
116 | } | |
117 | ||
1d045980 | 118 | /** |
9e48858f | 119 | * cap_ptrace_access_check - Determine whether the current process may access |
1d045980 DH |
120 | * another |
121 | * @child: The process to be accessed | |
122 | * @mode: The mode of attachment. | |
123 | * | |
8409cca7 SH |
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 | * | |
1d045980 DH |
130 | * Determine whether a process may access another, returning 0 if permission |
131 | * granted, -ve if denied. | |
132 | */ | |
9e48858f | 133 | int cap_ptrace_access_check(struct task_struct *child, unsigned int mode) |
1da177e4 | 134 | { |
c69e8d9c | 135 | int ret = 0; |
8409cca7 | 136 | const struct cred *cred, *child_cred; |
caaee623 | 137 | const kernel_cap_t *caller_caps; |
c69e8d9c DH |
138 | |
139 | rcu_read_lock(); | |
8409cca7 SH |
140 | cred = current_cred(); |
141 | child_cred = __task_cred(child); | |
caaee623 JH |
142 | if (mode & PTRACE_MODE_FSCREDS) |
143 | caller_caps = &cred->cap_effective; | |
144 | else | |
145 | caller_caps = &cred->cap_permitted; | |
c4a4d603 | 146 | if (cred->user_ns == child_cred->user_ns && |
caaee623 | 147 | cap_issubset(child_cred->cap_permitted, *caller_caps)) |
8409cca7 | 148 | goto out; |
c4a4d603 | 149 | if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE)) |
8409cca7 SH |
150 | goto out; |
151 | ret = -EPERM; | |
152 | out: | |
c69e8d9c DH |
153 | rcu_read_unlock(); |
154 | return ret; | |
5cd9c58f DH |
155 | } |
156 | ||
1d045980 DH |
157 | /** |
158 | * cap_ptrace_traceme - Determine whether another process may trace the current | |
159 | * @parent: The task proposed to be the tracer | |
160 | * | |
8409cca7 SH |
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 | * | |
1d045980 DH |
167 | * Determine whether the nominated task is permitted to trace the current |
168 | * process, returning 0 if permission is granted, -ve if denied. | |
169 | */ | |
5cd9c58f DH |
170 | int cap_ptrace_traceme(struct task_struct *parent) |
171 | { | |
c69e8d9c | 172 | int ret = 0; |
8409cca7 | 173 | const struct cred *cred, *child_cred; |
c69e8d9c DH |
174 | |
175 | rcu_read_lock(); | |
8409cca7 SH |
176 | cred = __task_cred(parent); |
177 | child_cred = current_cred(); | |
c4a4d603 | 178 | if (cred->user_ns == child_cred->user_ns && |
8409cca7 SH |
179 | cap_issubset(child_cred->cap_permitted, cred->cap_permitted)) |
180 | goto out; | |
c4a4d603 | 181 | if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE)) |
8409cca7 SH |
182 | goto out; |
183 | ret = -EPERM; | |
184 | out: | |
c69e8d9c DH |
185 | rcu_read_unlock(); |
186 | return ret; | |
1da177e4 LT |
187 | } |
188 | ||
1d045980 DH |
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 | int cap_capget(struct task_struct *target, kernel_cap_t *effective, | |
200 | kernel_cap_t *inheritable, kernel_cap_t *permitted) | |
1da177e4 | 201 | { |
c69e8d9c | 202 | const struct cred *cred; |
b6dff3ec | 203 | |
1da177e4 | 204 | /* Derived from kernel/capability.c:sys_capget. */ |
c69e8d9c DH |
205 | rcu_read_lock(); |
206 | cred = __task_cred(target); | |
b6dff3ec DH |
207 | *effective = cred->cap_effective; |
208 | *inheritable = cred->cap_inheritable; | |
209 | *permitted = cred->cap_permitted; | |
c69e8d9c | 210 | rcu_read_unlock(); |
1da177e4 LT |
211 | return 0; |
212 | } | |
213 | ||
1d045980 DH |
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 | */ | |
72c2d582 AM |
218 | static inline int cap_inh_is_capped(void) |
219 | { | |
1d045980 DH |
220 | /* they are so limited unless the current task has the CAP_SETPCAP |
221 | * capability | |
222 | */ | |
c4a4d603 | 223 | if (cap_capable(current_cred(), current_cred()->user_ns, |
c1a85a00 | 224 | CAP_SETPCAP, CAP_OPT_NONE) == 0) |
1d045980 | 225 | return 0; |
1d045980 | 226 | return 1; |
1209726c | 227 | } |
72c2d582 | 228 | |
1d045980 DH |
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 | */ | |
d84f4f99 DH |
241 | 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) | |
1da177e4 | 246 | { |
d84f4f99 DH |
247 | if (cap_inh_is_capped() && |
248 | !cap_issubset(*inheritable, | |
249 | cap_combine(old->cap_inheritable, | |
250 | old->cap_permitted))) | |
72c2d582 | 251 | /* incapable of using this inheritable set */ |
1da177e4 | 252 | return -EPERM; |
d84f4f99 | 253 | |
3b7391de | 254 | if (!cap_issubset(*inheritable, |
d84f4f99 DH |
255 | cap_combine(old->cap_inheritable, |
256 | old->cap_bset))) | |
3b7391de SH |
257 | /* no new pI capabilities outside bounding set */ |
258 | return -EPERM; | |
1da177e4 LT |
259 | |
260 | /* verify restrictions on target's new Permitted set */ | |
d84f4f99 | 261 | if (!cap_issubset(*permitted, old->cap_permitted)) |
1da177e4 | 262 | return -EPERM; |
1da177e4 LT |
263 | |
264 | /* verify the _new_Effective_ is a subset of the _new_Permitted_ */ | |
d84f4f99 | 265 | if (!cap_issubset(*effective, *permitted)) |
1da177e4 | 266 | return -EPERM; |
1da177e4 | 267 | |
d84f4f99 DH |
268 | new->cap_effective = *effective; |
269 | new->cap_inheritable = *inheritable; | |
270 | new->cap_permitted = *permitted; | |
58319057 AL |
271 | |
272 | /* | |
273 | * Mask off ambient bits that are no longer both permitted and | |
274 | * inheritable. | |
275 | */ | |
276 | new->cap_ambient = cap_intersect(new->cap_ambient, | |
277 | cap_intersect(*permitted, | |
278 | *inheritable)); | |
279 | if (WARN_ON(!cap_ambient_invariant_ok(new))) | |
280 | return -EINVAL; | |
1da177e4 LT |
281 | return 0; |
282 | } | |
283 | ||
1d045980 DH |
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 | |
ab5348c9 | 290 | * inode_killpriv() be invoked or the change rejected. |
1d045980 | 291 | * |
ab5348c9 SB |
292 | * Returns 1 if security.capability has a value, meaning inode_killpriv() |
293 | * is required, 0 otherwise, meaning inode_killpriv() is not required. | |
1d045980 | 294 | */ |
b5376771 SH |
295 | int cap_inode_need_killpriv(struct dentry *dentry) |
296 | { | |
c6f493d6 | 297 | struct inode *inode = d_backing_inode(dentry); |
b5376771 SH |
298 | int error; |
299 | ||
5d6c3191 AG |
300 | error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0); |
301 | return error > 0; | |
b5376771 SH |
302 | } |
303 | ||
1d045980 DH |
304 | /** |
305 | * cap_inode_killpriv - Erase the security markings on an inode | |
306 | * @dentry: The inode/dentry to alter | |
307 | * | |
308 | * Erase the privilege-enhancing security markings on an inode. | |
309 | * | |
310 | * Returns 0 if successful, -ve on error. | |
311 | */ | |
b5376771 SH |
312 | int cap_inode_killpriv(struct dentry *dentry) |
313 | { | |
5d6c3191 | 314 | int error; |
b5376771 | 315 | |
c7c7a1a1 | 316 | error = __vfs_removexattr(&init_user_ns, dentry, XATTR_NAME_CAPS); |
5d6c3191 AG |
317 | if (error == -EOPNOTSUPP) |
318 | error = 0; | |
319 | return error; | |
b5376771 SH |
320 | } |
321 | ||
8db6c34f SH |
322 | static bool rootid_owns_currentns(kuid_t kroot) |
323 | { | |
324 | struct user_namespace *ns; | |
325 | ||
326 | if (!uid_valid(kroot)) | |
327 | return false; | |
328 | ||
329 | for (ns = current_user_ns(); ; ns = ns->parent) { | |
330 | if (from_kuid(ns, kroot) == 0) | |
331 | return true; | |
332 | if (ns == &init_user_ns) | |
333 | break; | |
334 | } | |
335 | ||
336 | return false; | |
337 | } | |
338 | ||
339 | static __u32 sansflags(__u32 m) | |
340 | { | |
341 | return m & ~VFS_CAP_FLAGS_EFFECTIVE; | |
342 | } | |
343 | ||
dc32b5c3 | 344 | static bool is_v2header(size_t size, const struct vfs_cap_data *cap) |
8db6c34f | 345 | { |
8db6c34f SH |
346 | if (size != XATTR_CAPS_SZ_2) |
347 | return false; | |
dc32b5c3 | 348 | return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2; |
8db6c34f SH |
349 | } |
350 | ||
dc32b5c3 | 351 | static bool is_v3header(size_t size, const struct vfs_cap_data *cap) |
8db6c34f | 352 | { |
8db6c34f SH |
353 | if (size != XATTR_CAPS_SZ_3) |
354 | return false; | |
dc32b5c3 | 355 | return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3; |
8db6c34f SH |
356 | } |
357 | ||
358 | /* | |
359 | * getsecurity: We are called for security.* before any attempt to read the | |
360 | * xattr from the inode itself. | |
361 | * | |
362 | * This gives us a chance to read the on-disk value and convert it. If we | |
363 | * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler. | |
364 | * | |
365 | * Note we are not called by vfs_getxattr_alloc(), but that is only called | |
366 | * by the integrity subsystem, which really wants the unconverted values - | |
367 | * so that's good. | |
368 | */ | |
369 | int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer, | |
370 | bool alloc) | |
371 | { | |
372 | int size, ret; | |
373 | kuid_t kroot; | |
374 | uid_t root, mappedroot; | |
375 | char *tmpbuf = NULL; | |
376 | struct vfs_cap_data *cap; | |
377 | struct vfs_ns_cap_data *nscap; | |
378 | struct dentry *dentry; | |
379 | struct user_namespace *fs_ns; | |
380 | ||
381 | if (strcmp(name, "capability") != 0) | |
382 | return -EOPNOTSUPP; | |
383 | ||
355139a8 | 384 | dentry = d_find_any_alias(inode); |
8db6c34f SH |
385 | if (!dentry) |
386 | return -EINVAL; | |
387 | ||
388 | size = sizeof(struct vfs_ns_cap_data); | |
c7c7a1a1 TA |
389 | ret = (int)vfs_getxattr_alloc(&init_user_ns, dentry, XATTR_NAME_CAPS, |
390 | &tmpbuf, size, GFP_NOFS); | |
8db6c34f SH |
391 | dput(dentry); |
392 | ||
393 | if (ret < 0) | |
394 | return ret; | |
395 | ||
396 | fs_ns = inode->i_sb->s_user_ns; | |
397 | cap = (struct vfs_cap_data *) tmpbuf; | |
dc32b5c3 | 398 | if (is_v2header((size_t) ret, cap)) { |
8db6c34f SH |
399 | /* If this is sizeof(vfs_cap_data) then we're ok with the |
400 | * on-disk value, so return that. */ | |
401 | if (alloc) | |
402 | *buffer = tmpbuf; | |
403 | else | |
404 | kfree(tmpbuf); | |
405 | return ret; | |
dc32b5c3 | 406 | } else if (!is_v3header((size_t) ret, cap)) { |
8db6c34f SH |
407 | kfree(tmpbuf); |
408 | return -EINVAL; | |
409 | } | |
410 | ||
411 | nscap = (struct vfs_ns_cap_data *) tmpbuf; | |
412 | root = le32_to_cpu(nscap->rootid); | |
413 | kroot = make_kuid(fs_ns, root); | |
414 | ||
415 | /* If the root kuid maps to a valid uid in current ns, then return | |
416 | * this as a nscap. */ | |
417 | mappedroot = from_kuid(current_user_ns(), kroot); | |
418 | if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) { | |
419 | if (alloc) { | |
420 | *buffer = tmpbuf; | |
421 | nscap->rootid = cpu_to_le32(mappedroot); | |
422 | } else | |
423 | kfree(tmpbuf); | |
424 | return size; | |
425 | } | |
426 | ||
427 | if (!rootid_owns_currentns(kroot)) { | |
428 | kfree(tmpbuf); | |
429 | return -EOPNOTSUPP; | |
430 | } | |
431 | ||
432 | /* This comes from a parent namespace. Return as a v2 capability */ | |
433 | size = sizeof(struct vfs_cap_data); | |
434 | if (alloc) { | |
435 | *buffer = kmalloc(size, GFP_ATOMIC); | |
436 | if (*buffer) { | |
437 | struct vfs_cap_data *cap = *buffer; | |
438 | __le32 nsmagic, magic; | |
439 | magic = VFS_CAP_REVISION_2; | |
440 | nsmagic = le32_to_cpu(nscap->magic_etc); | |
441 | if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE) | |
442 | magic |= VFS_CAP_FLAGS_EFFECTIVE; | |
443 | memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32); | |
444 | cap->magic_etc = cpu_to_le32(magic); | |
1f578172 TH |
445 | } else { |
446 | size = -ENOMEM; | |
8db6c34f SH |
447 | } |
448 | } | |
449 | kfree(tmpbuf); | |
450 | return size; | |
451 | } | |
452 | ||
e65ce2a5 CB |
453 | /** |
454 | * rootid_from_xattr - translate root uid of vfs caps | |
455 | * | |
456 | * @value: vfs caps value which may be modified by this function | |
457 | * @size: size of @ivalue | |
458 | * @task_ns: user namespace of the caller | |
459 | * @mnt_userns: user namespace of the mount the inode was found from | |
460 | * | |
461 | * If the inode has been found through an idmapped mount the user namespace of | |
462 | * the vfsmount must be passed through @mnt_userns. This function will then | |
463 | * take care to map the inode according to @mnt_userns before checking | |
464 | * permissions. On non-idmapped mounts or if permission checking is to be | |
465 | * performed on the raw inode simply passs init_user_ns. | |
466 | */ | |
8db6c34f | 467 | static kuid_t rootid_from_xattr(const void *value, size_t size, |
e65ce2a5 CB |
468 | struct user_namespace *task_ns, |
469 | struct user_namespace *mnt_userns) | |
8db6c34f SH |
470 | { |
471 | const struct vfs_ns_cap_data *nscap = value; | |
e65ce2a5 | 472 | kuid_t rootkid; |
8db6c34f SH |
473 | uid_t rootid = 0; |
474 | ||
475 | if (size == XATTR_CAPS_SZ_3) | |
476 | rootid = le32_to_cpu(nscap->rootid); | |
477 | ||
e65ce2a5 CB |
478 | rootkid = make_kuid(task_ns, rootid); |
479 | return kuid_from_mnt(mnt_userns, rootkid); | |
8db6c34f SH |
480 | } |
481 | ||
dc32b5c3 | 482 | static bool validheader(size_t size, const struct vfs_cap_data *cap) |
8db6c34f | 483 | { |
dc32b5c3 | 484 | return is_v2header(size, cap) || is_v3header(size, cap); |
8db6c34f SH |
485 | } |
486 | ||
e65ce2a5 CB |
487 | /** |
488 | * cap_convert_nscap - check vfs caps | |
489 | * | |
490 | * @mnt_userns: user namespace of the mount the inode was found from | |
491 | * @dentry: used to retrieve inode to check permissions on | |
492 | * @ivalue: vfs caps value which may be modified by this function | |
493 | * @size: size of @ivalue | |
494 | * | |
8db6c34f SH |
495 | * User requested a write of security.capability. If needed, update the |
496 | * xattr to change from v2 to v3, or to fixup the v3 rootid. | |
497 | * | |
e65ce2a5 CB |
498 | * If the inode has been found through an idmapped mount the user namespace of |
499 | * the vfsmount must be passed through @mnt_userns. This function will then | |
500 | * take care to map the inode according to @mnt_userns before checking | |
501 | * permissions. On non-idmapped mounts or if permission checking is to be | |
502 | * performed on the raw inode simply passs init_user_ns. | |
503 | * | |
8db6c34f SH |
504 | * If all is ok, we return the new size, on error return < 0. |
505 | */ | |
e65ce2a5 CB |
506 | int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry, |
507 | const void **ivalue, size_t size) | |
8db6c34f SH |
508 | { |
509 | struct vfs_ns_cap_data *nscap; | |
510 | uid_t nsrootid; | |
511 | const struct vfs_cap_data *cap = *ivalue; | |
512 | __u32 magic, nsmagic; | |
513 | struct inode *inode = d_backing_inode(dentry); | |
514 | struct user_namespace *task_ns = current_user_ns(), | |
515 | *fs_ns = inode->i_sb->s_user_ns; | |
516 | kuid_t rootid; | |
517 | size_t newsize; | |
518 | ||
519 | if (!*ivalue) | |
520 | return -EINVAL; | |
dc32b5c3 | 521 | if (!validheader(size, cap)) |
8db6c34f | 522 | return -EINVAL; |
e65ce2a5 | 523 | if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP)) |
8db6c34f | 524 | return -EPERM; |
e65ce2a5 | 525 | if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns)) |
8db6c34f SH |
526 | if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP)) |
527 | /* user is privileged, just write the v2 */ | |
528 | return size; | |
529 | ||
e65ce2a5 | 530 | rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns); |
8db6c34f SH |
531 | if (!uid_valid(rootid)) |
532 | return -EINVAL; | |
533 | ||
534 | nsrootid = from_kuid(fs_ns, rootid); | |
535 | if (nsrootid == -1) | |
536 | return -EINVAL; | |
537 | ||
538 | newsize = sizeof(struct vfs_ns_cap_data); | |
539 | nscap = kmalloc(newsize, GFP_ATOMIC); | |
540 | if (!nscap) | |
541 | return -ENOMEM; | |
542 | nscap->rootid = cpu_to_le32(nsrootid); | |
543 | nsmagic = VFS_CAP_REVISION_3; | |
544 | magic = le32_to_cpu(cap->magic_etc); | |
545 | if (magic & VFS_CAP_FLAGS_EFFECTIVE) | |
546 | nsmagic |= VFS_CAP_FLAGS_EFFECTIVE; | |
547 | nscap->magic_etc = cpu_to_le32(nsmagic); | |
548 | memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32); | |
549 | ||
8db6c34f SH |
550 | *ivalue = nscap; |
551 | return newsize; | |
552 | } | |
553 | ||
1d045980 DH |
554 | /* |
555 | * Calculate the new process capability sets from the capability sets attached | |
556 | * to a file. | |
557 | */ | |
c0b00441 | 558 | static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps, |
a6f76f23 | 559 | struct linux_binprm *bprm, |
4d49f671 | 560 | bool *effective, |
fc7eadf7 | 561 | bool *has_fcap) |
b5376771 | 562 | { |
a6f76f23 | 563 | struct cred *new = bprm->cred; |
c0b00441 EP |
564 | unsigned i; |
565 | int ret = 0; | |
566 | ||
567 | if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE) | |
a6f76f23 | 568 | *effective = true; |
c0b00441 | 569 | |
4d49f671 | 570 | if (caps->magic_etc & VFS_CAP_REVISION_MASK) |
fc7eadf7 | 571 | *has_fcap = true; |
4d49f671 | 572 | |
c0b00441 EP |
573 | CAP_FOR_EACH_U32(i) { |
574 | __u32 permitted = caps->permitted.cap[i]; | |
575 | __u32 inheritable = caps->inheritable.cap[i]; | |
576 | ||
577 | /* | |
578 | * pP' = (X & fP) | (pI & fI) | |
58319057 | 579 | * The addition of pA' is handled later. |
c0b00441 | 580 | */ |
a6f76f23 DH |
581 | new->cap_permitted.cap[i] = |
582 | (new->cap_bset.cap[i] & permitted) | | |
583 | (new->cap_inheritable.cap[i] & inheritable); | |
c0b00441 | 584 | |
a6f76f23 DH |
585 | if (permitted & ~new->cap_permitted.cap[i]) |
586 | /* insufficient to execute correctly */ | |
c0b00441 | 587 | ret = -EPERM; |
c0b00441 EP |
588 | } |
589 | ||
590 | /* | |
591 | * For legacy apps, with no internal support for recognizing they | |
592 | * do not have enough capabilities, we return an error if they are | |
593 | * missing some "forced" (aka file-permitted) capabilities. | |
594 | */ | |
a6f76f23 | 595 | return *effective ? ret : 0; |
c0b00441 EP |
596 | } |
597 | ||
1d045980 DH |
598 | /* |
599 | * Extract the on-exec-apply capability sets for an executable file. | |
600 | */ | |
c0b00441 EP |
601 | int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps) |
602 | { | |
c6f493d6 | 603 | struct inode *inode = d_backing_inode(dentry); |
b5376771 | 604 | __u32 magic_etc; |
e338d263 | 605 | unsigned tocopy, i; |
c0b00441 | 606 | int size; |
8db6c34f SH |
607 | struct vfs_ns_cap_data data, *nscaps = &data; |
608 | struct vfs_cap_data *caps = (struct vfs_cap_data *) &data; | |
609 | kuid_t rootkuid; | |
76ba89c7 | 610 | struct user_namespace *fs_ns; |
c0b00441 EP |
611 | |
612 | memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data)); | |
613 | ||
5d6c3191 | 614 | if (!inode) |
c0b00441 EP |
615 | return -ENODATA; |
616 | ||
76ba89c7 | 617 | fs_ns = inode->i_sb->s_user_ns; |
5d6c3191 | 618 | size = __vfs_getxattr((struct dentry *)dentry, inode, |
8db6c34f | 619 | XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ); |
a6f76f23 | 620 | if (size == -ENODATA || size == -EOPNOTSUPP) |
c0b00441 EP |
621 | /* no data, that's ok */ |
622 | return -ENODATA; | |
8db6c34f | 623 | |
c0b00441 EP |
624 | if (size < 0) |
625 | return size; | |
b5376771 | 626 | |
e338d263 | 627 | if (size < sizeof(magic_etc)) |
b5376771 SH |
628 | return -EINVAL; |
629 | ||
8db6c34f | 630 | cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc); |
b5376771 | 631 | |
8db6c34f | 632 | rootkuid = make_kuid(fs_ns, 0); |
a6f76f23 | 633 | switch (magic_etc & VFS_CAP_REVISION_MASK) { |
e338d263 AM |
634 | case VFS_CAP_REVISION_1: |
635 | if (size != XATTR_CAPS_SZ_1) | |
636 | return -EINVAL; | |
637 | tocopy = VFS_CAP_U32_1; | |
638 | break; | |
639 | case VFS_CAP_REVISION_2: | |
640 | if (size != XATTR_CAPS_SZ_2) | |
641 | return -EINVAL; | |
642 | tocopy = VFS_CAP_U32_2; | |
643 | break; | |
8db6c34f SH |
644 | case VFS_CAP_REVISION_3: |
645 | if (size != XATTR_CAPS_SZ_3) | |
646 | return -EINVAL; | |
647 | tocopy = VFS_CAP_U32_3; | |
648 | rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid)); | |
649 | break; | |
650 | ||
b5376771 SH |
651 | default: |
652 | return -EINVAL; | |
653 | } | |
8db6c34f SH |
654 | /* Limit the caps to the mounter of the filesystem |
655 | * or the more limited uid specified in the xattr. | |
656 | */ | |
657 | if (!rootid_owns_currentns(rootkuid)) | |
658 | return -ENODATA; | |
e338d263 | 659 | |
5459c164 | 660 | CAP_FOR_EACH_U32(i) { |
c0b00441 EP |
661 | if (i >= tocopy) |
662 | break; | |
8db6c34f SH |
663 | cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted); |
664 | cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable); | |
e338d263 | 665 | } |
a6f76f23 | 666 | |
7d8b6c63 EP |
667 | cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; |
668 | cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; | |
669 | ||
2fec30e2 RGB |
670 | cpu_caps->rootid = rootkuid; |
671 | ||
c0b00441 | 672 | return 0; |
b5376771 SH |
673 | } |
674 | ||
1d045980 DH |
675 | /* |
676 | * Attempt to get the on-exec apply capability sets for an executable file from | |
677 | * its xattrs and, if present, apply them to the proposed credentials being | |
678 | * constructed by execve(). | |
679 | */ | |
56305aa9 EB |
680 | static int get_file_caps(struct linux_binprm *bprm, struct file *file, |
681 | bool *effective, bool *has_fcap) | |
b5376771 | 682 | { |
b5376771 | 683 | int rc = 0; |
c0b00441 | 684 | struct cpu_vfs_cap_data vcaps; |
b5376771 | 685 | |
ee67ae7e | 686 | cap_clear(bprm->cred->cap_permitted); |
3318a386 | 687 | |
1f29fae2 SH |
688 | if (!file_caps_enabled) |
689 | return 0; | |
690 | ||
56305aa9 | 691 | if (!mnt_may_suid(file->f_path.mnt)) |
b5376771 | 692 | return 0; |
380cf5ba AL |
693 | |
694 | /* | |
695 | * This check is redundant with mnt_may_suid() but is kept to make | |
696 | * explicit that capability bits are limited to s_user_ns and its | |
697 | * descendants. | |
698 | */ | |
56305aa9 | 699 | if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns)) |
d07b846f | 700 | return 0; |
b5376771 | 701 | |
56305aa9 | 702 | rc = get_vfs_caps_from_disk(file->f_path.dentry, &vcaps); |
c0b00441 EP |
703 | if (rc < 0) { |
704 | if (rc == -EINVAL) | |
8db6c34f SH |
705 | printk(KERN_NOTICE "Invalid argument reading file caps for %s\n", |
706 | bprm->filename); | |
c0b00441 EP |
707 | else if (rc == -ENODATA) |
708 | rc = 0; | |
b5376771 SH |
709 | goto out; |
710 | } | |
b5376771 | 711 | |
fc7eadf7 | 712 | rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap); |
b5376771 SH |
713 | |
714 | out: | |
b5376771 | 715 | if (rc) |
ee67ae7e | 716 | cap_clear(bprm->cred->cap_permitted); |
b5376771 SH |
717 | |
718 | return rc; | |
719 | } | |
720 | ||
9304b46c RGB |
721 | static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); } |
722 | ||
81a6a012 RGB |
723 | static inline bool __is_real(kuid_t uid, struct cred *cred) |
724 | { return uid_eq(cred->uid, uid); } | |
725 | ||
726 | static inline bool __is_eff(kuid_t uid, struct cred *cred) | |
727 | { return uid_eq(cred->euid, uid); } | |
728 | ||
729 | static inline bool __is_suid(kuid_t uid, struct cred *cred) | |
730 | { return !__is_real(uid, cred) && __is_eff(uid, cred); } | |
731 | ||
db1a8922 RGB |
732 | /* |
733 | * handle_privileged_root - Handle case of privileged root | |
734 | * @bprm: The execution parameters, including the proposed creds | |
735 | * @has_fcap: Are any file capabilities set? | |
736 | * @effective: Do we have effective root privilege? | |
737 | * @root_uid: This namespace' root UID WRT initial USER namespace | |
738 | * | |
739 | * Handle the case where root is privileged and hasn't been neutered by | |
740 | * SECURE_NOROOT. If file capabilities are set, they won't be combined with | |
741 | * set UID root and nothing is changed. If we are root, cap_permitted is | |
742 | * updated. If we have become set UID root, the effective bit is set. | |
743 | */ | |
fc7eadf7 | 744 | static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap, |
db1a8922 RGB |
745 | bool *effective, kuid_t root_uid) |
746 | { | |
747 | const struct cred *old = current_cred(); | |
748 | struct cred *new = bprm->cred; | |
749 | ||
9304b46c | 750 | if (!root_privileged()) |
db1a8922 RGB |
751 | return; |
752 | /* | |
753 | * If the legacy file capability is set, then don't set privs | |
754 | * for a setuid root binary run by a non-root user. Do set it | |
755 | * for a root user just to cause least surprise to an admin. | |
756 | */ | |
81a6a012 | 757 | if (has_fcap && __is_suid(root_uid, new)) { |
db1a8922 RGB |
758 | warn_setuid_and_fcaps_mixed(bprm->filename); |
759 | return; | |
760 | } | |
761 | /* | |
762 | * To support inheritance of root-permissions and suid-root | |
763 | * executables under compatibility mode, we override the | |
764 | * capability sets for the file. | |
765 | */ | |
81a6a012 | 766 | if (__is_eff(root_uid, new) || __is_real(root_uid, new)) { |
db1a8922 RGB |
767 | /* pP' = (cap_bset & ~0) | (pI & ~0) */ |
768 | new->cap_permitted = cap_combine(old->cap_bset, | |
769 | old->cap_inheritable); | |
770 | } | |
771 | /* | |
772 | * If only the real uid is 0, we do not set the effective bit. | |
773 | */ | |
81a6a012 | 774 | if (__is_eff(root_uid, new)) |
db1a8922 RGB |
775 | *effective = true; |
776 | } | |
777 | ||
4c7e715f RGB |
778 | #define __cap_gained(field, target, source) \ |
779 | !cap_issubset(target->cap_##field, source->cap_##field) | |
780 | #define __cap_grew(target, source, cred) \ | |
781 | !cap_issubset(cred->cap_##target, cred->cap_##source) | |
782 | #define __cap_full(field, cred) \ | |
783 | cap_issubset(CAP_FULL_SET, cred->cap_##field) | |
81a6a012 RGB |
784 | |
785 | static inline bool __is_setuid(struct cred *new, const struct cred *old) | |
786 | { return !uid_eq(new->euid, old->uid); } | |
787 | ||
788 | static inline bool __is_setgid(struct cred *new, const struct cred *old) | |
789 | { return !gid_eq(new->egid, old->gid); } | |
790 | ||
9fbc2c79 | 791 | /* |
dbbbe110 | 792 | * 1) Audit candidate if current->cap_effective is set |
9fbc2c79 RGB |
793 | * |
794 | * We do not bother to audit if 3 things are true: | |
795 | * 1) cap_effective has all caps | |
588fb2c7 | 796 | * 2) we became root *OR* are were already root |
9fbc2c79 RGB |
797 | * 3) root is supposed to have all caps (SECURE_NOROOT) |
798 | * Since this is just a normal root execing a process. | |
799 | * | |
800 | * Number 1 above might fail if you don't have a full bset, but I think | |
801 | * that is interesting information to audit. | |
dbbbe110 RGB |
802 | * |
803 | * A number of other conditions require logging: | |
804 | * 2) something prevented setuid root getting all caps | |
805 | * 3) non-setuid root gets fcaps | |
806 | * 4) non-setuid root gets ambient | |
9fbc2c79 | 807 | */ |
dbbbe110 RGB |
808 | static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old, |
809 | kuid_t root, bool has_fcap) | |
9fbc2c79 RGB |
810 | { |
811 | bool ret = false; | |
812 | ||
dbbbe110 RGB |
813 | if ((__cap_grew(effective, ambient, new) && |
814 | !(__cap_full(effective, new) && | |
815 | (__is_eff(root, new) || __is_real(root, new)) && | |
816 | root_privileged())) || | |
817 | (root_privileged() && | |
818 | __is_suid(root, new) && | |
819 | !__cap_full(effective, new)) || | |
820 | (!__is_setuid(new, old) && | |
821 | ((has_fcap && | |
822 | __cap_gained(permitted, new, old)) || | |
823 | __cap_gained(ambient, new, old)))) | |
824 | ||
02ebbaf4 | 825 | ret = true; |
dbbbe110 | 826 | |
9fbc2c79 RGB |
827 | return ret; |
828 | } | |
829 | ||
1d045980 | 830 | /** |
56305aa9 | 831 | * cap_bprm_creds_from_file - Set up the proposed credentials for execve(). |
1d045980 | 832 | * @bprm: The execution parameters, including the proposed creds |
56305aa9 | 833 | * @file: The file to pull the credentials from |
1d045980 DH |
834 | * |
835 | * Set up the proposed credentials for a new execution context being | |
836 | * constructed by execve(). The proposed creds in @bprm->cred is altered, | |
837 | * which won't take effect immediately. Returns 0 if successful, -ve on error. | |
a6f76f23 | 838 | */ |
56305aa9 | 839 | int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file) |
1da177e4 | 840 | { |
56305aa9 | 841 | /* Process setpcap binaries and capabilities for uid 0 */ |
a6f76f23 DH |
842 | const struct cred *old = current_cred(); |
843 | struct cred *new = bprm->cred; | |
fc7eadf7 | 844 | bool effective = false, has_fcap = false, is_setid; |
b5376771 | 845 | int ret; |
18815a18 | 846 | kuid_t root_uid; |
1da177e4 | 847 | |
58319057 AL |
848 | if (WARN_ON(!cap_ambient_invariant_ok(old))) |
849 | return -EPERM; | |
850 | ||
56305aa9 | 851 | ret = get_file_caps(bprm, file, &effective, &has_fcap); |
a6f76f23 DH |
852 | if (ret < 0) |
853 | return ret; | |
1da177e4 | 854 | |
18815a18 EB |
855 | root_uid = make_kuid(new->user_ns, 0); |
856 | ||
fc7eadf7 | 857 | handle_privileged_root(bprm, has_fcap, &effective, root_uid); |
b5376771 | 858 | |
d52fc5dd | 859 | /* if we have fs caps, clear dangerous personality flags */ |
4c7e715f | 860 | if (__cap_gained(permitted, new, old)) |
56305aa9 | 861 | bprm->per_clear |= PER_CLEAR_ON_SETID; |
d52fc5dd | 862 | |
a6f76f23 | 863 | /* Don't let someone trace a set[ug]id/setpcap binary with the revised |
259e5e6c AL |
864 | * credentials unless they have the appropriate permit. |
865 | * | |
866 | * In addition, if NO_NEW_PRIVS, then ensure we get no new privs. | |
a6f76f23 | 867 | */ |
81a6a012 | 868 | is_setid = __is_setuid(new, old) || __is_setgid(new, old); |
58319057 | 869 | |
4c7e715f | 870 | if ((is_setid || __cap_gained(permitted, new, old)) && |
9227dd2a | 871 | ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) || |
20523132 | 872 | !ptracer_capable(current, new->user_ns))) { |
a6f76f23 | 873 | /* downgrade; they get no more than they had, and maybe less */ |
70169420 | 874 | if (!ns_capable(new->user_ns, CAP_SETUID) || |
259e5e6c | 875 | (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) { |
a6f76f23 DH |
876 | new->euid = new->uid; |
877 | new->egid = new->gid; | |
1da177e4 | 878 | } |
b3a222e5 SH |
879 | new->cap_permitted = cap_intersect(new->cap_permitted, |
880 | old->cap_permitted); | |
1da177e4 LT |
881 | } |
882 | ||
a6f76f23 DH |
883 | new->suid = new->fsuid = new->euid; |
884 | new->sgid = new->fsgid = new->egid; | |
1da177e4 | 885 | |
58319057 | 886 | /* File caps or setid cancels ambient. */ |
fc7eadf7 | 887 | if (has_fcap || is_setid) |
58319057 AL |
888 | cap_clear(new->cap_ambient); |
889 | ||
890 | /* | |
891 | * Now that we've computed pA', update pP' to give: | |
892 | * pP' = (X & fP) | (pI & fI) | pA' | |
893 | */ | |
894 | new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient); | |
895 | ||
896 | /* | |
897 | * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set, | |
898 | * this is the same as pE' = (fE ? pP' : 0) | pA'. | |
899 | */ | |
4bf2ea77 EP |
900 | if (effective) |
901 | new->cap_effective = new->cap_permitted; | |
902 | else | |
58319057 AL |
903 | new->cap_effective = new->cap_ambient; |
904 | ||
905 | if (WARN_ON(!cap_ambient_invariant_ok(new))) | |
906 | return -EPERM; | |
907 | ||
dbbbe110 | 908 | if (nonroot_raised_pE(new, old, root_uid, has_fcap)) { |
9fbc2c79 RGB |
909 | ret = audit_log_bprm_fcaps(bprm, new, old); |
910 | if (ret < 0) | |
911 | return ret; | |
3fc689e9 | 912 | } |
1da177e4 | 913 | |
d84f4f99 | 914 | new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); |
58319057 AL |
915 | |
916 | if (WARN_ON(!cap_ambient_invariant_ok(new))) | |
917 | return -EPERM; | |
918 | ||
46d98eb4 | 919 | /* Check for privilege-elevated exec. */ |
02ebbaf4 RGB |
920 | if (is_setid || |
921 | (!__is_real(root_uid, new) && | |
922 | (effective || | |
923 | __cap_grew(permitted, ambient, new)))) | |
56305aa9 | 924 | bprm->secureexec = 1; |
b5376771 | 925 | |
ee67ae7e | 926 | return 0; |
1da177e4 LT |
927 | } |
928 | ||
1d045980 DH |
929 | /** |
930 | * cap_inode_setxattr - Determine whether an xattr may be altered | |
931 | * @dentry: The inode/dentry being altered | |
932 | * @name: The name of the xattr to be changed | |
933 | * @value: The value that the xattr will be changed to | |
934 | * @size: The size of value | |
935 | * @flags: The replacement flag | |
936 | * | |
937 | * Determine whether an xattr may be altered or set on an inode, returning 0 if | |
938 | * permission is granted, -ve if denied. | |
939 | * | |
940 | * This is used to make sure security xattrs don't get updated or set by those | |
941 | * who aren't privileged to do so. | |
942 | */ | |
8f0cfa52 DH |
943 | int cap_inode_setxattr(struct dentry *dentry, const char *name, |
944 | const void *value, size_t size, int flags) | |
1da177e4 | 945 | { |
b1d749c5 EB |
946 | struct user_namespace *user_ns = dentry->d_sb->s_user_ns; |
947 | ||
8db6c34f SH |
948 | /* Ignore non-security xattrs */ |
949 | if (strncmp(name, XATTR_SECURITY_PREFIX, | |
c5eaab1d | 950 | XATTR_SECURITY_PREFIX_LEN) != 0) |
8db6c34f SH |
951 | return 0; |
952 | ||
953 | /* | |
954 | * For XATTR_NAME_CAPS the check will be done in | |
955 | * cap_convert_nscap(), called by setxattr() | |
956 | */ | |
957 | if (strcmp(name, XATTR_NAME_CAPS) == 0) | |
b5376771 | 958 | return 0; |
1d045980 | 959 | |
b1d749c5 | 960 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
1da177e4 LT |
961 | return -EPERM; |
962 | return 0; | |
963 | } | |
964 | ||
1d045980 DH |
965 | /** |
966 | * cap_inode_removexattr - Determine whether an xattr may be removed | |
967 | * @dentry: The inode/dentry being altered | |
968 | * @name: The name of the xattr to be changed | |
969 | * | |
970 | * Determine whether an xattr may be removed from an inode, returning 0 if | |
971 | * permission is granted, -ve if denied. | |
972 | * | |
973 | * This is used to make sure security xattrs don't get removed by those who | |
974 | * aren't privileged to remove them. | |
975 | */ | |
8f0cfa52 | 976 | int cap_inode_removexattr(struct dentry *dentry, const char *name) |
1da177e4 | 977 | { |
b1d749c5 EB |
978 | struct user_namespace *user_ns = dentry->d_sb->s_user_ns; |
979 | ||
8db6c34f SH |
980 | /* Ignore non-security xattrs */ |
981 | if (strncmp(name, XATTR_SECURITY_PREFIX, | |
c5eaab1d | 982 | XATTR_SECURITY_PREFIX_LEN) != 0) |
8db6c34f SH |
983 | return 0; |
984 | ||
985 | if (strcmp(name, XATTR_NAME_CAPS) == 0) { | |
986 | /* security.capability gets namespaced */ | |
987 | struct inode *inode = d_backing_inode(dentry); | |
988 | if (!inode) | |
989 | return -EINVAL; | |
0558c1bf CB |
990 | if (!capable_wrt_inode_uidgid(&init_user_ns, inode, |
991 | CAP_SETFCAP)) | |
b5376771 SH |
992 | return -EPERM; |
993 | return 0; | |
1d045980 DH |
994 | } |
995 | ||
b1d749c5 | 996 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
1da177e4 LT |
997 | return -EPERM; |
998 | return 0; | |
999 | } | |
1000 | ||
a6f76f23 | 1001 | /* |
1da177e4 LT |
1002 | * cap_emulate_setxuid() fixes the effective / permitted capabilities of |
1003 | * a process after a call to setuid, setreuid, or setresuid. | |
1004 | * | |
1005 | * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of | |
1006 | * {r,e,s}uid != 0, the permitted and effective capabilities are | |
1007 | * cleared. | |
1008 | * | |
1009 | * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective | |
1010 | * capabilities of the process are cleared. | |
1011 | * | |
1012 | * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective | |
1013 | * capabilities are set to the permitted capabilities. | |
1014 | * | |
a6f76f23 | 1015 | * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should |
1da177e4 LT |
1016 | * never happen. |
1017 | * | |
a6f76f23 | 1018 | * -astor |
1da177e4 LT |
1019 | * |
1020 | * cevans - New behaviour, Oct '99 | |
1021 | * A process may, via prctl(), elect to keep its capabilities when it | |
1022 | * calls setuid() and switches away from uid==0. Both permitted and | |
1023 | * effective sets will be retained. | |
1024 | * Without this change, it was impossible for a daemon to drop only some | |
1025 | * of its privilege. The call to setuid(!=0) would drop all privileges! | |
1026 | * Keeping uid 0 is not an option because uid 0 owns too many vital | |
1027 | * files.. | |
1028 | * Thanks to Olaf Kirch and Peter Benie for spotting this. | |
1029 | */ | |
d84f4f99 | 1030 | static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old) |
1da177e4 | 1031 | { |
18815a18 EB |
1032 | kuid_t root_uid = make_kuid(old->user_ns, 0); |
1033 | ||
1034 | if ((uid_eq(old->uid, root_uid) || | |
1035 | uid_eq(old->euid, root_uid) || | |
1036 | uid_eq(old->suid, root_uid)) && | |
1037 | (!uid_eq(new->uid, root_uid) && | |
1038 | !uid_eq(new->euid, root_uid) && | |
58319057 AL |
1039 | !uid_eq(new->suid, root_uid))) { |
1040 | if (!issecure(SECURE_KEEP_CAPS)) { | |
1041 | cap_clear(new->cap_permitted); | |
1042 | cap_clear(new->cap_effective); | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * Pre-ambient programs expect setresuid to nonroot followed | |
1047 | * by exec to drop capabilities. We should make sure that | |
1048 | * this remains the case. | |
1049 | */ | |
1050 | cap_clear(new->cap_ambient); | |
1da177e4 | 1051 | } |
18815a18 | 1052 | if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid)) |
d84f4f99 | 1053 | cap_clear(new->cap_effective); |
18815a18 | 1054 | if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid)) |
d84f4f99 | 1055 | new->cap_effective = new->cap_permitted; |
1da177e4 LT |
1056 | } |
1057 | ||
1d045980 DH |
1058 | /** |
1059 | * cap_task_fix_setuid - Fix up the results of setuid() call | |
1060 | * @new: The proposed credentials | |
1061 | * @old: The current task's current credentials | |
1062 | * @flags: Indications of what has changed | |
1063 | * | |
1064 | * Fix up the results of setuid() call before the credential changes are | |
1065 | * actually applied, returning 0 to grant the changes, -ve to deny them. | |
1066 | */ | |
d84f4f99 | 1067 | int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags) |
1da177e4 LT |
1068 | { |
1069 | switch (flags) { | |
1070 | case LSM_SETID_RE: | |
1071 | case LSM_SETID_ID: | |
1072 | case LSM_SETID_RES: | |
1d045980 DH |
1073 | /* juggle the capabilities to follow [RES]UID changes unless |
1074 | * otherwise suppressed */ | |
d84f4f99 DH |
1075 | if (!issecure(SECURE_NO_SETUID_FIXUP)) |
1076 | cap_emulate_setxuid(new, old); | |
1da177e4 | 1077 | break; |
1da177e4 | 1078 | |
1d045980 DH |
1079 | case LSM_SETID_FS: |
1080 | /* juggle the capabilties to follow FSUID changes, unless | |
1081 | * otherwise suppressed | |
1082 | * | |
d84f4f99 DH |
1083 | * FIXME - is fsuser used for all CAP_FS_MASK capabilities? |
1084 | * if not, we might be a bit too harsh here. | |
1085 | */ | |
1086 | if (!issecure(SECURE_NO_SETUID_FIXUP)) { | |
18815a18 EB |
1087 | kuid_t root_uid = make_kuid(old->user_ns, 0); |
1088 | if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid)) | |
d84f4f99 DH |
1089 | new->cap_effective = |
1090 | cap_drop_fs_set(new->cap_effective); | |
1d045980 | 1091 | |
18815a18 | 1092 | if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid)) |
d84f4f99 DH |
1093 | new->cap_effective = |
1094 | cap_raise_fs_set(new->cap_effective, | |
1095 | new->cap_permitted); | |
1da177e4 | 1096 | } |
d84f4f99 | 1097 | break; |
1d045980 | 1098 | |
1da177e4 LT |
1099 | default: |
1100 | return -EINVAL; | |
1101 | } | |
1102 | ||
1103 | return 0; | |
1104 | } | |
1105 | ||
b5376771 SH |
1106 | /* |
1107 | * Rationale: code calling task_setscheduler, task_setioprio, and | |
1108 | * task_setnice, assumes that | |
1109 | * . if capable(cap_sys_nice), then those actions should be allowed | |
1110 | * . if not capable(cap_sys_nice), but acting on your own processes, | |
1111 | * then those actions should be allowed | |
1112 | * This is insufficient now since you can call code without suid, but | |
1113 | * yet with increased caps. | |
1114 | * So we check for increased caps on the target process. | |
1115 | */ | |
de45e806 | 1116 | static int cap_safe_nice(struct task_struct *p) |
b5376771 | 1117 | { |
f54fb863 | 1118 | int is_subset, ret = 0; |
c69e8d9c DH |
1119 | |
1120 | rcu_read_lock(); | |
1121 | is_subset = cap_issubset(__task_cred(p)->cap_permitted, | |
1122 | current_cred()->cap_permitted); | |
f54fb863 SH |
1123 | if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) |
1124 | ret = -EPERM; | |
c69e8d9c DH |
1125 | rcu_read_unlock(); |
1126 | ||
f54fb863 | 1127 | return ret; |
b5376771 SH |
1128 | } |
1129 | ||
1d045980 DH |
1130 | /** |
1131 | * cap_task_setscheduler - Detemine if scheduler policy change is permitted | |
1132 | * @p: The task to affect | |
1d045980 DH |
1133 | * |
1134 | * Detemine if the requested scheduler policy change is permitted for the | |
1135 | * specified task, returning 0 if permission is granted, -ve if denied. | |
1136 | */ | |
b0ae1981 | 1137 | int cap_task_setscheduler(struct task_struct *p) |
b5376771 SH |
1138 | { |
1139 | return cap_safe_nice(p); | |
1140 | } | |
1141 | ||
1d045980 DH |
1142 | /** |
1143 | * cap_task_ioprio - Detemine if I/O priority change is permitted | |
1144 | * @p: The task to affect | |
1145 | * @ioprio: The I/O priority to set | |
1146 | * | |
1147 | * Detemine if the requested I/O priority change is permitted for the specified | |
1148 | * task, returning 0 if permission is granted, -ve if denied. | |
1149 | */ | |
1150 | int cap_task_setioprio(struct task_struct *p, int ioprio) | |
b5376771 SH |
1151 | { |
1152 | return cap_safe_nice(p); | |
1153 | } | |
1154 | ||
1d045980 DH |
1155 | /** |
1156 | * cap_task_ioprio - Detemine if task priority change is permitted | |
1157 | * @p: The task to affect | |
1158 | * @nice: The nice value to set | |
1159 | * | |
1160 | * Detemine if the requested task priority change is permitted for the | |
1161 | * specified task, returning 0 if permission is granted, -ve if denied. | |
1162 | */ | |
1163 | int cap_task_setnice(struct task_struct *p, int nice) | |
b5376771 SH |
1164 | { |
1165 | return cap_safe_nice(p); | |
1166 | } | |
1167 | ||
3b7391de | 1168 | /* |
1d045980 DH |
1169 | * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from |
1170 | * the current task's bounding set. Returns 0 on success, -ve on error. | |
3b7391de | 1171 | */ |
6d6f3328 | 1172 | static int cap_prctl_drop(unsigned long cap) |
3b7391de | 1173 | { |
6d6f3328 TH |
1174 | struct cred *new; |
1175 | ||
160da84d | 1176 | if (!ns_capable(current_user_ns(), CAP_SETPCAP)) |
3b7391de SH |
1177 | return -EPERM; |
1178 | if (!cap_valid(cap)) | |
1179 | return -EINVAL; | |
d84f4f99 | 1180 | |
6d6f3328 TH |
1181 | new = prepare_creds(); |
1182 | if (!new) | |
1183 | return -ENOMEM; | |
d84f4f99 | 1184 | cap_lower(new->cap_bset, cap); |
6d6f3328 | 1185 | return commit_creds(new); |
3b7391de | 1186 | } |
3898b1b4 | 1187 | |
1d045980 DH |
1188 | /** |
1189 | * cap_task_prctl - Implement process control functions for this security module | |
1190 | * @option: The process control function requested | |
1191 | * @arg2, @arg3, @arg4, @arg5: The argument data for this function | |
1192 | * | |
1193 | * Allow process control functions (sys_prctl()) to alter capabilities; may | |
1194 | * also deny access to other functions not otherwise implemented here. | |
1195 | * | |
1196 | * Returns 0 or +ve on success, -ENOSYS if this function is not implemented | |
1197 | * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM | |
1198 | * modules will consider performing the function. | |
1199 | */ | |
3898b1b4 | 1200 | int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3, |
d84f4f99 | 1201 | unsigned long arg4, unsigned long arg5) |
3898b1b4 | 1202 | { |
6d6f3328 | 1203 | const struct cred *old = current_cred(); |
d84f4f99 | 1204 | struct cred *new; |
d84f4f99 | 1205 | |
3898b1b4 AM |
1206 | switch (option) { |
1207 | case PR_CAPBSET_READ: | |
1208 | if (!cap_valid(arg2)) | |
6d6f3328 TH |
1209 | return -EINVAL; |
1210 | return !!cap_raised(old->cap_bset, arg2); | |
d84f4f99 | 1211 | |
3898b1b4 | 1212 | case PR_CAPBSET_DROP: |
6d6f3328 | 1213 | return cap_prctl_drop(arg2); |
3898b1b4 AM |
1214 | |
1215 | /* | |
1216 | * The next four prctl's remain to assist with transitioning a | |
1217 | * system from legacy UID=0 based privilege (when filesystem | |
1218 | * capabilities are not in use) to a system using filesystem | |
1219 | * capabilities only - as the POSIX.1e draft intended. | |
1220 | * | |
1221 | * Note: | |
1222 | * | |
1223 | * PR_SET_SECUREBITS = | |
1224 | * issecure_mask(SECURE_KEEP_CAPS_LOCKED) | |
1225 | * | issecure_mask(SECURE_NOROOT) | |
1226 | * | issecure_mask(SECURE_NOROOT_LOCKED) | |
1227 | * | issecure_mask(SECURE_NO_SETUID_FIXUP) | |
1228 | * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED) | |
1229 | * | |
1230 | * will ensure that the current process and all of its | |
1231 | * children will be locked into a pure | |
1232 | * capability-based-privilege environment. | |
1233 | */ | |
1234 | case PR_SET_SECUREBITS: | |
6d6f3328 TH |
1235 | if ((((old->securebits & SECURE_ALL_LOCKS) >> 1) |
1236 | & (old->securebits ^ arg2)) /*[1]*/ | |
1237 | || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/ | |
d84f4f99 | 1238 | || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/ |
6a9de491 | 1239 | || (cap_capable(current_cred(), |
c1a85a00 MM |
1240 | current_cred()->user_ns, |
1241 | CAP_SETPCAP, | |
1242 | CAP_OPT_NONE) != 0) /*[4]*/ | |
3898b1b4 AM |
1243 | /* |
1244 | * [1] no changing of bits that are locked | |
1245 | * [2] no unlocking of locks | |
1246 | * [3] no setting of unsupported bits | |
1247 | * [4] doing anything requires privilege (go read about | |
1248 | * the "sendmail capabilities bug") | |
1249 | */ | |
d84f4f99 DH |
1250 | ) |
1251 | /* cannot change a locked bit */ | |
6d6f3328 TH |
1252 | return -EPERM; |
1253 | ||
1254 | new = prepare_creds(); | |
1255 | if (!new) | |
1256 | return -ENOMEM; | |
d84f4f99 | 1257 | new->securebits = arg2; |
6d6f3328 | 1258 | return commit_creds(new); |
d84f4f99 | 1259 | |
3898b1b4 | 1260 | case PR_GET_SECUREBITS: |
6d6f3328 | 1261 | return old->securebits; |
3898b1b4 | 1262 | |
3898b1b4 | 1263 | case PR_GET_KEEPCAPS: |
6d6f3328 | 1264 | return !!issecure(SECURE_KEEP_CAPS); |
d84f4f99 | 1265 | |
3898b1b4 AM |
1266 | case PR_SET_KEEPCAPS: |
1267 | if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */ | |
6d6f3328 | 1268 | return -EINVAL; |
d84f4f99 | 1269 | if (issecure(SECURE_KEEP_CAPS_LOCKED)) |
6d6f3328 TH |
1270 | return -EPERM; |
1271 | ||
1272 | new = prepare_creds(); | |
1273 | if (!new) | |
1274 | return -ENOMEM; | |
d84f4f99 DH |
1275 | if (arg2) |
1276 | new->securebits |= issecure_mask(SECURE_KEEP_CAPS); | |
3898b1b4 | 1277 | else |
d84f4f99 | 1278 | new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); |
6d6f3328 | 1279 | return commit_creds(new); |
3898b1b4 | 1280 | |
58319057 AL |
1281 | case PR_CAP_AMBIENT: |
1282 | if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) { | |
1283 | if (arg3 | arg4 | arg5) | |
1284 | return -EINVAL; | |
1285 | ||
1286 | new = prepare_creds(); | |
1287 | if (!new) | |
1288 | return -ENOMEM; | |
1289 | cap_clear(new->cap_ambient); | |
1290 | return commit_creds(new); | |
1291 | } | |
1292 | ||
1293 | if (((!cap_valid(arg3)) | arg4 | arg5)) | |
1294 | return -EINVAL; | |
1295 | ||
1296 | if (arg2 == PR_CAP_AMBIENT_IS_SET) { | |
1297 | return !!cap_raised(current_cred()->cap_ambient, arg3); | |
1298 | } else if (arg2 != PR_CAP_AMBIENT_RAISE && | |
1299 | arg2 != PR_CAP_AMBIENT_LOWER) { | |
1300 | return -EINVAL; | |
1301 | } else { | |
1302 | if (arg2 == PR_CAP_AMBIENT_RAISE && | |
1303 | (!cap_raised(current_cred()->cap_permitted, arg3) || | |
1304 | !cap_raised(current_cred()->cap_inheritable, | |
746bf6d6 AL |
1305 | arg3) || |
1306 | issecure(SECURE_NO_CAP_AMBIENT_RAISE))) | |
58319057 AL |
1307 | return -EPERM; |
1308 | ||
1309 | new = prepare_creds(); | |
1310 | if (!new) | |
1311 | return -ENOMEM; | |
1312 | if (arg2 == PR_CAP_AMBIENT_RAISE) | |
1313 | cap_raise(new->cap_ambient, arg3); | |
1314 | else | |
1315 | cap_lower(new->cap_ambient, arg3); | |
1316 | return commit_creds(new); | |
1317 | } | |
1318 | ||
3898b1b4 AM |
1319 | default: |
1320 | /* No functionality available - continue with default */ | |
6d6f3328 | 1321 | return -ENOSYS; |
3898b1b4 | 1322 | } |
1da177e4 LT |
1323 | } |
1324 | ||
1d045980 DH |
1325 | /** |
1326 | * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted | |
1327 | * @mm: The VM space in which the new mapping is to be made | |
1328 | * @pages: The size of the mapping | |
1329 | * | |
1330 | * Determine whether the allocation of a new virtual mapping by the current | |
b1d9e6b0 | 1331 | * task is permitted, returning 1 if permission is granted, 0 if not. |
1d045980 | 1332 | */ |
34b4e4aa | 1333 | int cap_vm_enough_memory(struct mm_struct *mm, long pages) |
1da177e4 LT |
1334 | { |
1335 | int cap_sys_admin = 0; | |
1336 | ||
c1a85a00 MM |
1337 | if (cap_capable(current_cred(), &init_user_ns, |
1338 | CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0) | |
1da177e4 | 1339 | cap_sys_admin = 1; |
c1a85a00 | 1340 | |
b1d9e6b0 | 1341 | return cap_sys_admin; |
1da177e4 | 1342 | } |
7c73875e EP |
1343 | |
1344 | /* | |
d007794a | 1345 | * cap_mmap_addr - check if able to map given addr |
7c73875e | 1346 | * @addr: address attempting to be mapped |
7c73875e | 1347 | * |
6f262d8e | 1348 | * If the process is attempting to map memory below dac_mmap_min_addr they need |
7c73875e EP |
1349 | * CAP_SYS_RAWIO. The other parameters to this function are unused by the |
1350 | * capability security module. Returns 0 if this mapping should be allowed | |
1351 | * -EPERM if not. | |
1352 | */ | |
d007794a | 1353 | int cap_mmap_addr(unsigned long addr) |
7c73875e EP |
1354 | { |
1355 | int ret = 0; | |
1356 | ||
a2551df7 | 1357 | if (addr < dac_mmap_min_addr) { |
6a9de491 | 1358 | ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO, |
c1a85a00 | 1359 | CAP_OPT_NONE); |
7c73875e EP |
1360 | /* set PF_SUPERPRIV if it turns out we allow the low mmap */ |
1361 | if (ret == 0) | |
1362 | current->flags |= PF_SUPERPRIV; | |
1363 | } | |
1364 | return ret; | |
1365 | } | |
d007794a | 1366 | |
e5467859 AV |
1367 | int cap_mmap_file(struct file *file, unsigned long reqprot, |
1368 | unsigned long prot, unsigned long flags) | |
d007794a | 1369 | { |
e5467859 | 1370 | return 0; |
d007794a | 1371 | } |
b1d9e6b0 CS |
1372 | |
1373 | #ifdef CONFIG_SECURITY | |
1374 | ||
d1c5947e | 1375 | static struct security_hook_list capability_hooks[] __lsm_ro_after_init = { |
b1d9e6b0 CS |
1376 | LSM_HOOK_INIT(capable, cap_capable), |
1377 | LSM_HOOK_INIT(settime, cap_settime), | |
1378 | LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check), | |
1379 | LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme), | |
1380 | LSM_HOOK_INIT(capget, cap_capget), | |
1381 | LSM_HOOK_INIT(capset, cap_capset), | |
56305aa9 | 1382 | LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file), |
b1d9e6b0 CS |
1383 | LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv), |
1384 | LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv), | |
8db6c34f | 1385 | LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity), |
b1d9e6b0 CS |
1386 | LSM_HOOK_INIT(mmap_addr, cap_mmap_addr), |
1387 | LSM_HOOK_INIT(mmap_file, cap_mmap_file), | |
1388 | LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid), | |
1389 | LSM_HOOK_INIT(task_prctl, cap_task_prctl), | |
1390 | LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler), | |
1391 | LSM_HOOK_INIT(task_setioprio, cap_task_setioprio), | |
1392 | LSM_HOOK_INIT(task_setnice, cap_task_setnice), | |
1393 | LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory), | |
1394 | }; | |
1395 | ||
d117a154 | 1396 | static int __init capability_init(void) |
b1d9e6b0 | 1397 | { |
d69dece5 CS |
1398 | security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks), |
1399 | "capability"); | |
d117a154 | 1400 | return 0; |
b1d9e6b0 CS |
1401 | } |
1402 | ||
d117a154 KC |
1403 | DEFINE_LSM(capability) = { |
1404 | .name = "capability", | |
1405 | .order = LSM_ORDER_FIRST, | |
1406 | .init = capability_init, | |
1407 | }; | |
1408 | ||
b1d9e6b0 | 1409 | #endif /* CONFIG_SECURITY */ |