1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * changes by Thomas Schoebel-Theuer
9 #include <linux/export.h>
10 #include <linux/time.h>
12 #include <linux/string.h>
13 #include <linux/sched/signal.h>
14 #include <linux/capability.h>
15 #include <linux/fsnotify.h>
16 #include <linux/fcntl.h>
17 #include <linux/security.h>
18 #include <linux/evm.h>
19 #include <linux/ima.h>
24 * setattr_should_drop_sgid - determine whether the setgid bit needs to be
26 * @mnt_userns: user namespace of the mount @inode was found from
27 * @inode: inode to check
29 * This function determines whether the setgid bit needs to be removed.
30 * We retain backwards compatibility and require setgid bit to be removed
31 * unconditionally if S_IXGRP is set. Otherwise we have the exact same
32 * requirements as setattr_prepare() and setattr_copy().
34 * Return: ATTR_KILL_SGID if setgid bit needs to be removed, 0 otherwise.
36 int setattr_should_drop_sgid(struct user_namespace
*mnt_userns
,
37 const struct inode
*inode
)
39 umode_t mode
= inode
->i_mode
;
41 if (!(mode
& S_ISGID
))
44 return ATTR_KILL_SGID
;
45 if (!in_group_or_capable(mnt_userns
, inode
,
46 i_gid_into_vfsgid(mnt_userns
, inode
)))
47 return ATTR_KILL_SGID
;
52 * setattr_should_drop_suidgid - determine whether the set{g,u}id bit needs to
54 * @mnt_userns: user namespace of the mount @inode was found from
55 * @inode: inode to check
57 * This function determines whether the set{g,u}id bits need to be removed.
58 * If the setuid bit needs to be removed ATTR_KILL_SUID is returned. If the
59 * setgid bit needs to be removed ATTR_KILL_SGID is returned. If both
60 * set{g,u}id bits need to be removed the corresponding mask of both flags is
63 * Return: A mask of ATTR_KILL_S{G,U}ID indicating which - if any - setid bits
64 * to remove, 0 otherwise.
66 int setattr_should_drop_suidgid(struct user_namespace
*mnt_userns
,
69 umode_t mode
= inode
->i_mode
;
72 /* suid always must be killed */
73 if (unlikely(mode
& S_ISUID
))
74 kill
= ATTR_KILL_SUID
;
76 kill
|= setattr_should_drop_sgid(mnt_userns
, inode
);
78 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
83 EXPORT_SYMBOL(setattr_should_drop_suidgid
);
86 * chown_ok - verify permissions to chown inode
87 * @mnt_userns: user namespace of the mount @inode was found from
88 * @inode: inode to check permissions on
89 * @ia_vfsuid: uid to chown @inode to
91 * If the inode has been found through an idmapped mount the user namespace of
92 * the vfsmount must be passed through @mnt_userns. This function will then
93 * take care to map the inode according to @mnt_userns before checking
94 * permissions. On non-idmapped mounts or if permission checking is to be
95 * performed on the raw inode simply passs init_user_ns.
97 static bool chown_ok(struct user_namespace
*mnt_userns
,
98 const struct inode
*inode
, vfsuid_t ia_vfsuid
)
100 vfsuid_t vfsuid
= i_uid_into_vfsuid(mnt_userns
, inode
);
101 if (vfsuid_eq_kuid(vfsuid
, current_fsuid()) &&
102 vfsuid_eq(ia_vfsuid
, vfsuid
))
104 if (capable_wrt_inode_uidgid(mnt_userns
, inode
, CAP_CHOWN
))
106 if (!vfsuid_valid(vfsuid
) &&
107 ns_capable(inode
->i_sb
->s_user_ns
, CAP_CHOWN
))
113 * chgrp_ok - verify permissions to chgrp inode
114 * @mnt_userns: user namespace of the mount @inode was found from
115 * @inode: inode to check permissions on
116 * @ia_vfsgid: gid to chown @inode to
118 * If the inode has been found through an idmapped mount the user namespace of
119 * the vfsmount must be passed through @mnt_userns. This function will then
120 * take care to map the inode according to @mnt_userns before checking
121 * permissions. On non-idmapped mounts or if permission checking is to be
122 * performed on the raw inode simply passs init_user_ns.
124 static bool chgrp_ok(struct user_namespace
*mnt_userns
,
125 const struct inode
*inode
, vfsgid_t ia_vfsgid
)
127 vfsgid_t vfsgid
= i_gid_into_vfsgid(mnt_userns
, inode
);
128 vfsuid_t vfsuid
= i_uid_into_vfsuid(mnt_userns
, inode
);
129 if (vfsuid_eq_kuid(vfsuid
, current_fsuid())) {
130 if (vfsgid_eq(ia_vfsgid
, vfsgid
))
132 if (vfsgid_in_group_p(ia_vfsgid
))
135 if (capable_wrt_inode_uidgid(mnt_userns
, inode
, CAP_CHOWN
))
137 if (!vfsgid_valid(vfsgid
) &&
138 ns_capable(inode
->i_sb
->s_user_ns
, CAP_CHOWN
))
144 * setattr_prepare - check if attribute changes to a dentry are allowed
145 * @mnt_userns: user namespace of the mount the inode was found from
146 * @dentry: dentry to check
147 * @attr: attributes to change
149 * Check if we are allowed to change the attributes contained in @attr
150 * in the given dentry. This includes the normal unix access permission
151 * checks, as well as checks for rlimits and others. The function also clears
152 * SGID bit from mode if user is not allowed to set it. Also file capabilities
153 * and IMA extended attributes are cleared if ATTR_KILL_PRIV is set.
155 * If the inode has been found through an idmapped mount the user namespace of
156 * the vfsmount must be passed through @mnt_userns. This function will then
157 * take care to map the inode according to @mnt_userns before checking
158 * permissions. On non-idmapped mounts or if permission checking is to be
159 * performed on the raw inode simply passs init_user_ns.
161 * Should be called as the first thing in ->setattr implementations,
162 * possibly after taking additional locks.
164 int setattr_prepare(struct user_namespace
*mnt_userns
, struct dentry
*dentry
,
167 struct inode
*inode
= d_inode(dentry
);
168 unsigned int ia_valid
= attr
->ia_valid
;
171 * First check size constraints. These can't be overriden using
174 if (ia_valid
& ATTR_SIZE
) {
175 int error
= inode_newsize_ok(inode
, attr
->ia_size
);
180 /* If force is set do it anyway. */
181 if (ia_valid
& ATTR_FORCE
)
184 /* Make sure a caller can chown. */
185 if ((ia_valid
& ATTR_UID
) &&
186 !chown_ok(mnt_userns
, inode
, attr
->ia_vfsuid
))
189 /* Make sure caller can chgrp. */
190 if ((ia_valid
& ATTR_GID
) &&
191 !chgrp_ok(mnt_userns
, inode
, attr
->ia_vfsgid
))
194 /* Make sure a caller can chmod. */
195 if (ia_valid
& ATTR_MODE
) {
198 if (!inode_owner_or_capable(mnt_userns
, inode
))
201 if (ia_valid
& ATTR_GID
)
202 vfsgid
= attr
->ia_vfsgid
;
204 vfsgid
= i_gid_into_vfsgid(mnt_userns
, inode
);
206 /* Also check the setgid bit! */
207 if (!in_group_or_capable(mnt_userns
, inode
, vfsgid
))
208 attr
->ia_mode
&= ~S_ISGID
;
211 /* Check for setting the inode time. */
212 if (ia_valid
& (ATTR_MTIME_SET
| ATTR_ATIME_SET
| ATTR_TIMES_SET
)) {
213 if (!inode_owner_or_capable(mnt_userns
, inode
))
218 /* User has permission for the change */
219 if (ia_valid
& ATTR_KILL_PRIV
) {
222 error
= security_inode_killpriv(mnt_userns
, dentry
);
229 EXPORT_SYMBOL(setattr_prepare
);
232 * inode_newsize_ok - may this inode be truncated to a given size
233 * @inode: the inode to be truncated
234 * @offset: the new size to assign to the inode
236 * inode_newsize_ok must be called with i_mutex held.
238 * inode_newsize_ok will check filesystem limits and ulimits to check that the
239 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
240 * when necessary. Caller must not proceed with inode size change if failure is
241 * returned. @inode must be a file (not directory), with appropriate
242 * permissions to allow truncate (inode_newsize_ok does NOT check these
245 * Return: 0 on success, -ve errno on failure
247 int inode_newsize_ok(const struct inode
*inode
, loff_t offset
)
251 if (inode
->i_size
< offset
) {
254 limit
= rlimit(RLIMIT_FSIZE
);
255 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
257 if (offset
> inode
->i_sb
->s_maxbytes
)
261 * truncation of in-use swapfiles is disallowed - it would
262 * cause subsequent swapout to scribble on the now-freed
265 if (IS_SWAPFILE(inode
))
271 send_sig(SIGXFSZ
, current
, 0);
275 EXPORT_SYMBOL(inode_newsize_ok
);
278 * setattr_copy - copy simple metadata updates into the generic inode
279 * @mnt_userns: user namespace of the mount the inode was found from
280 * @inode: the inode to be updated
281 * @attr: the new attributes
283 * setattr_copy must be called with i_mutex held.
285 * setattr_copy updates the inode's metadata with that specified
286 * in attr on idmapped mounts. Necessary permission checks to determine
287 * whether or not the S_ISGID property needs to be removed are performed with
288 * the correct idmapped mount permission helpers.
289 * Noticeably missing is inode size update, which is more complex
290 * as it requires pagecache updates.
292 * If the inode has been found through an idmapped mount the user namespace of
293 * the vfsmount must be passed through @mnt_userns. This function will then
294 * take care to map the inode according to @mnt_userns before checking
295 * permissions. On non-idmapped mounts or if permission checking is to be
296 * performed on the raw inode simply passs init_user_ns.
298 * The inode is not marked as dirty after this operation. The rationale is
299 * that for "simple" filesystems, the struct inode is the inode storage.
300 * The caller is free to mark the inode dirty afterwards if needed.
302 void setattr_copy(struct user_namespace
*mnt_userns
, struct inode
*inode
,
303 const struct iattr
*attr
)
305 unsigned int ia_valid
= attr
->ia_valid
;
307 i_uid_update(mnt_userns
, attr
, inode
);
308 i_gid_update(mnt_userns
, attr
, inode
);
309 if (ia_valid
& ATTR_ATIME
)
310 inode
->i_atime
= attr
->ia_atime
;
311 if (ia_valid
& ATTR_MTIME
)
312 inode
->i_mtime
= attr
->ia_mtime
;
313 if (ia_valid
& ATTR_CTIME
)
314 inode
->i_ctime
= attr
->ia_ctime
;
315 if (ia_valid
& ATTR_MODE
) {
316 umode_t mode
= attr
->ia_mode
;
317 if (!in_group_or_capable(mnt_userns
, inode
,
318 i_gid_into_vfsgid(mnt_userns
, inode
)))
320 inode
->i_mode
= mode
;
323 EXPORT_SYMBOL(setattr_copy
);
325 int may_setattr(struct user_namespace
*mnt_userns
, struct inode
*inode
,
326 unsigned int ia_valid
)
330 if (ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
| ATTR_TIMES_SET
)) {
331 if (IS_IMMUTABLE(inode
) || IS_APPEND(inode
))
336 * If utimes(2) and friends are called with times == NULL (or both
337 * times are UTIME_NOW), then we need to check for write permission
339 if (ia_valid
& ATTR_TOUCH
) {
340 if (IS_IMMUTABLE(inode
))
343 if (!inode_owner_or_capable(mnt_userns
, inode
)) {
344 error
= inode_permission(mnt_userns
, inode
, MAY_WRITE
);
351 EXPORT_SYMBOL(may_setattr
);
354 * notify_change - modify attributes of a filesytem object
355 * @mnt_userns: user namespace of the mount the inode was found from
356 * @dentry: object affected
357 * @attr: new attributes
358 * @delegated_inode: returns inode, if the inode is delegated
360 * The caller must hold the i_mutex on the affected object.
362 * If notify_change discovers a delegation in need of breaking,
363 * it will return -EWOULDBLOCK and return a reference to the inode in
364 * delegated_inode. The caller should then break the delegation and
365 * retry. Because breaking a delegation may take a long time, the
366 * caller should drop the i_mutex before doing so.
368 * Alternatively, a caller may pass NULL for delegated_inode. This may
369 * be appropriate for callers that expect the underlying filesystem not
370 * to be NFS exported. Also, passing NULL is fine for callers holding
371 * the file open for write, as there can be no conflicting delegation in
374 * If the inode has been found through an idmapped mount the user namespace of
375 * the vfsmount must be passed through @mnt_userns. This function will then
376 * take care to map the inode according to @mnt_userns before checking
377 * permissions. On non-idmapped mounts or if permission checking is to be
378 * performed on the raw inode simply passs init_user_ns.
380 int notify_change(struct user_namespace
*mnt_userns
, struct dentry
*dentry
,
381 struct iattr
*attr
, struct inode
**delegated_inode
)
383 struct inode
*inode
= dentry
->d_inode
;
384 umode_t mode
= inode
->i_mode
;
386 struct timespec64 now
;
387 unsigned int ia_valid
= attr
->ia_valid
;
389 WARN_ON_ONCE(!inode_is_locked(inode
));
391 error
= may_setattr(mnt_userns
, inode
, ia_valid
);
395 if ((ia_valid
& ATTR_MODE
)) {
396 umode_t amode
= attr
->ia_mode
;
397 /* Flag setting protected by i_mutex */
399 inode
->i_flags
&= ~S_NOSEC
;
402 now
= current_time(inode
);
404 attr
->ia_ctime
= now
;
405 if (!(ia_valid
& ATTR_ATIME_SET
))
406 attr
->ia_atime
= now
;
408 attr
->ia_atime
= timestamp_truncate(attr
->ia_atime
, inode
);
409 if (!(ia_valid
& ATTR_MTIME_SET
))
410 attr
->ia_mtime
= now
;
412 attr
->ia_mtime
= timestamp_truncate(attr
->ia_mtime
, inode
);
414 if (ia_valid
& ATTR_KILL_PRIV
) {
415 error
= security_inode_need_killpriv(dentry
);
419 ia_valid
= attr
->ia_valid
&= ~ATTR_KILL_PRIV
;
423 * We now pass ATTR_KILL_S*ID to the lower level setattr function so
424 * that the function has the ability to reinterpret a mode change
425 * that's due to these bits. This adds an implicit restriction that
426 * no function will ever call notify_change with both ATTR_MODE and
427 * ATTR_KILL_S*ID set.
429 if ((ia_valid
& (ATTR_KILL_SUID
|ATTR_KILL_SGID
)) &&
430 (ia_valid
& ATTR_MODE
))
433 if (ia_valid
& ATTR_KILL_SUID
) {
434 if (mode
& S_ISUID
) {
435 ia_valid
= attr
->ia_valid
|= ATTR_MODE
;
436 attr
->ia_mode
= (inode
->i_mode
& ~S_ISUID
);
439 if (ia_valid
& ATTR_KILL_SGID
) {
440 if (mode
& S_ISGID
) {
441 if (!(ia_valid
& ATTR_MODE
)) {
442 ia_valid
= attr
->ia_valid
|= ATTR_MODE
;
443 attr
->ia_mode
= inode
->i_mode
;
445 attr
->ia_mode
&= ~S_ISGID
;
448 if (!(attr
->ia_valid
& ~(ATTR_KILL_SUID
| ATTR_KILL_SGID
)))
452 * Verify that uid/gid changes are valid in the target
453 * namespace of the superblock.
455 if (ia_valid
& ATTR_UID
&&
456 !vfsuid_has_fsmapping(mnt_userns
, inode
->i_sb
->s_user_ns
,
459 if (ia_valid
& ATTR_GID
&&
460 !vfsgid_has_fsmapping(mnt_userns
, inode
->i_sb
->s_user_ns
,
464 /* Don't allow modifications of files with invalid uids or
465 * gids unless those uids & gids are being made valid.
467 if (!(ia_valid
& ATTR_UID
) &&
468 !vfsuid_valid(i_uid_into_vfsuid(mnt_userns
, inode
)))
470 if (!(ia_valid
& ATTR_GID
) &&
471 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns
, inode
)))
474 error
= security_inode_setattr(mnt_userns
, dentry
, attr
);
477 error
= try_break_deleg(inode
, delegated_inode
);
481 if (inode
->i_op
->setattr
)
482 error
= inode
->i_op
->setattr(mnt_userns
, dentry
, attr
);
484 error
= simple_setattr(mnt_userns
, dentry
, attr
);
487 fsnotify_change(dentry
, ia_valid
);
488 ima_inode_post_setattr(mnt_userns
, dentry
);
489 evm_inode_post_setattr(dentry
, ia_valid
);
494 EXPORT_SYMBOL(notify_change
);