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/filelock.h>
18 #include <linux/security.h>
21 * setattr_should_drop_sgid - determine whether the setgid bit needs to be
23 * @idmap: idmap of the mount @inode was found from
24 * @inode: inode to check
26 * This function determines whether the setgid bit needs to be removed.
27 * We retain backwards compatibility and require setgid bit to be removed
28 * unconditionally if S_IXGRP is set. Otherwise we have the exact same
29 * requirements as setattr_prepare() and setattr_copy().
31 * Return: ATTR_KILL_SGID if setgid bit needs to be removed, 0 otherwise.
33 int setattr_should_drop_sgid(struct mnt_idmap
*idmap
,
34 const struct inode
*inode
)
36 umode_t mode
= inode
->i_mode
;
38 if (!(mode
& S_ISGID
))
41 return ATTR_KILL_SGID
;
42 if (!in_group_or_capable(idmap
, inode
, i_gid_into_vfsgid(idmap
, inode
)))
43 return ATTR_KILL_SGID
;
46 EXPORT_SYMBOL(setattr_should_drop_sgid
);
49 * setattr_should_drop_suidgid - determine whether the set{g,u}id bit needs to
51 * @idmap: idmap of the mount @inode was found from
52 * @inode: inode to check
54 * This function determines whether the set{g,u}id bits need to be removed.
55 * If the setuid bit needs to be removed ATTR_KILL_SUID is returned. If the
56 * setgid bit needs to be removed ATTR_KILL_SGID is returned. If both
57 * set{g,u}id bits need to be removed the corresponding mask of both flags is
60 * Return: A mask of ATTR_KILL_S{G,U}ID indicating which - if any - setid bits
61 * to remove, 0 otherwise.
63 int setattr_should_drop_suidgid(struct mnt_idmap
*idmap
,
66 umode_t mode
= inode
->i_mode
;
69 /* suid always must be killed */
70 if (unlikely(mode
& S_ISUID
))
71 kill
= ATTR_KILL_SUID
;
73 kill
|= setattr_should_drop_sgid(idmap
, inode
);
75 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
80 EXPORT_SYMBOL(setattr_should_drop_suidgid
);
83 * chown_ok - verify permissions to chown inode
84 * @idmap: idmap of the mount @inode was found from
85 * @inode: inode to check permissions on
86 * @ia_vfsuid: uid to chown @inode to
88 * If the inode has been found through an idmapped mount the idmap of
89 * the vfsmount must be passed through @idmap. This function will then
90 * take care to map the inode according to @idmap before checking
91 * permissions. On non-idmapped mounts or if permission checking is to be
92 * performed on the raw inode simply pass @nop_mnt_idmap.
94 static bool chown_ok(struct mnt_idmap
*idmap
,
95 const struct inode
*inode
, vfsuid_t ia_vfsuid
)
97 vfsuid_t vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
98 if (vfsuid_eq_kuid(vfsuid
, current_fsuid()) &&
99 vfsuid_eq(ia_vfsuid
, vfsuid
))
101 if (capable_wrt_inode_uidgid(idmap
, inode
, CAP_CHOWN
))
103 if (!vfsuid_valid(vfsuid
) &&
104 ns_capable(inode
->i_sb
->s_user_ns
, CAP_CHOWN
))
110 * chgrp_ok - verify permissions to chgrp inode
111 * @idmap: idmap of the mount @inode was found from
112 * @inode: inode to check permissions on
113 * @ia_vfsgid: gid to chown @inode to
115 * If the inode has been found through an idmapped mount the idmap of
116 * the vfsmount must be passed through @idmap. This function will then
117 * take care to map the inode according to @idmap before checking
118 * permissions. On non-idmapped mounts or if permission checking is to be
119 * performed on the raw inode simply pass @nop_mnt_idmap.
121 static bool chgrp_ok(struct mnt_idmap
*idmap
,
122 const struct inode
*inode
, vfsgid_t ia_vfsgid
)
124 vfsgid_t vfsgid
= i_gid_into_vfsgid(idmap
, inode
);
125 vfsuid_t vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
126 if (vfsuid_eq_kuid(vfsuid
, current_fsuid())) {
127 if (vfsgid_eq(ia_vfsgid
, vfsgid
))
129 if (vfsgid_in_group_p(ia_vfsgid
))
132 if (capable_wrt_inode_uidgid(idmap
, inode
, CAP_CHOWN
))
134 if (!vfsgid_valid(vfsgid
) &&
135 ns_capable(inode
->i_sb
->s_user_ns
, CAP_CHOWN
))
141 * setattr_prepare - check if attribute changes to a dentry are allowed
142 * @idmap: idmap of the mount the inode was found from
143 * @dentry: dentry to check
144 * @attr: attributes to change
146 * Check if we are allowed to change the attributes contained in @attr
147 * in the given dentry. This includes the normal unix access permission
148 * checks, as well as checks for rlimits and others. The function also clears
149 * SGID bit from mode if user is not allowed to set it. Also file capabilities
150 * and IMA extended attributes are cleared if ATTR_KILL_PRIV is set.
152 * If the inode has been found through an idmapped mount the idmap of
153 * the vfsmount must be passed through @idmap. This function will then
154 * take care to map the inode according to @idmap before checking
155 * permissions. On non-idmapped mounts or if permission checking is to be
156 * performed on the raw inode simply pass @nop_mnt_idmap.
158 * Should be called as the first thing in ->setattr implementations,
159 * possibly after taking additional locks.
161 int setattr_prepare(struct mnt_idmap
*idmap
, struct dentry
*dentry
,
164 struct inode
*inode
= d_inode(dentry
);
165 unsigned int ia_valid
= attr
->ia_valid
;
168 * First check size constraints. These can't be overriden using
171 if (ia_valid
& ATTR_SIZE
) {
172 int error
= inode_newsize_ok(inode
, attr
->ia_size
);
177 /* If force is set do it anyway. */
178 if (ia_valid
& ATTR_FORCE
)
181 /* Make sure a caller can chown. */
182 if ((ia_valid
& ATTR_UID
) &&
183 !chown_ok(idmap
, inode
, attr
->ia_vfsuid
))
186 /* Make sure caller can chgrp. */
187 if ((ia_valid
& ATTR_GID
) &&
188 !chgrp_ok(idmap
, inode
, attr
->ia_vfsgid
))
191 /* Make sure a caller can chmod. */
192 if (ia_valid
& ATTR_MODE
) {
195 if (!inode_owner_or_capable(idmap
, inode
))
198 if (ia_valid
& ATTR_GID
)
199 vfsgid
= attr
->ia_vfsgid
;
201 vfsgid
= i_gid_into_vfsgid(idmap
, inode
);
203 /* Also check the setgid bit! */
204 if (!in_group_or_capable(idmap
, inode
, vfsgid
))
205 attr
->ia_mode
&= ~S_ISGID
;
208 /* Check for setting the inode time. */
209 if (ia_valid
& (ATTR_MTIME_SET
| ATTR_ATIME_SET
| ATTR_TIMES_SET
)) {
210 if (!inode_owner_or_capable(idmap
, inode
))
215 /* User has permission for the change */
216 if (ia_valid
& ATTR_KILL_PRIV
) {
219 error
= security_inode_killpriv(idmap
, dentry
);
226 EXPORT_SYMBOL(setattr_prepare
);
229 * inode_newsize_ok - may this inode be truncated to a given size
230 * @inode: the inode to be truncated
231 * @offset: the new size to assign to the inode
233 * inode_newsize_ok must be called with i_mutex held.
235 * inode_newsize_ok will check filesystem limits and ulimits to check that the
236 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
237 * when necessary. Caller must not proceed with inode size change if failure is
238 * returned. @inode must be a file (not directory), with appropriate
239 * permissions to allow truncate (inode_newsize_ok does NOT check these
242 * Return: 0 on success, -ve errno on failure
244 int inode_newsize_ok(const struct inode
*inode
, loff_t offset
)
248 if (inode
->i_size
< offset
) {
251 limit
= rlimit(RLIMIT_FSIZE
);
252 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
254 if (offset
> inode
->i_sb
->s_maxbytes
)
258 * truncation of in-use swapfiles is disallowed - it would
259 * cause subsequent swapout to scribble on the now-freed
262 if (IS_SWAPFILE(inode
))
268 send_sig(SIGXFSZ
, current
, 0);
272 EXPORT_SYMBOL(inode_newsize_ok
);
275 * setattr_copy - copy simple metadata updates into the generic inode
276 * @idmap: idmap of the mount the inode was found from
277 * @inode: the inode to be updated
278 * @attr: the new attributes
280 * setattr_copy must be called with i_mutex held.
282 * setattr_copy updates the inode's metadata with that specified
283 * in attr on idmapped mounts. Necessary permission checks to determine
284 * whether or not the S_ISGID property needs to be removed are performed with
285 * the correct idmapped mount permission helpers.
286 * Noticeably missing is inode size update, which is more complex
287 * as it requires pagecache updates.
289 * If the inode has been found through an idmapped mount the idmap of
290 * the vfsmount must be passed through @idmap. This function will then
291 * take care to map the inode according to @idmap before checking
292 * permissions. On non-idmapped mounts or if permission checking is to be
293 * performed on the raw inode simply pass @nop_mnt_idmap.
295 * The inode is not marked as dirty after this operation. The rationale is
296 * that for "simple" filesystems, the struct inode is the inode storage.
297 * The caller is free to mark the inode dirty afterwards if needed.
299 void setattr_copy(struct mnt_idmap
*idmap
, struct inode
*inode
,
300 const struct iattr
*attr
)
302 unsigned int ia_valid
= attr
->ia_valid
;
304 i_uid_update(idmap
, attr
, inode
);
305 i_gid_update(idmap
, attr
, inode
);
306 if (ia_valid
& ATTR_ATIME
)
307 inode_set_atime_to_ts(inode
, attr
->ia_atime
);
308 if (ia_valid
& ATTR_MTIME
)
309 inode_set_mtime_to_ts(inode
, attr
->ia_mtime
);
310 if (ia_valid
& ATTR_CTIME
)
311 inode_set_ctime_to_ts(inode
, attr
->ia_ctime
);
312 if (ia_valid
& ATTR_MODE
) {
313 umode_t mode
= attr
->ia_mode
;
314 if (!in_group_or_capable(idmap
, inode
,
315 i_gid_into_vfsgid(idmap
, inode
)))
317 inode
->i_mode
= mode
;
320 EXPORT_SYMBOL(setattr_copy
);
322 int may_setattr(struct mnt_idmap
*idmap
, struct inode
*inode
,
323 unsigned int ia_valid
)
327 if (ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
| ATTR_TIMES_SET
)) {
328 if (IS_IMMUTABLE(inode
) || IS_APPEND(inode
))
333 * If utimes(2) and friends are called with times == NULL (or both
334 * times are UTIME_NOW), then we need to check for write permission
336 if (ia_valid
& ATTR_TOUCH
) {
337 if (IS_IMMUTABLE(inode
))
340 if (!inode_owner_or_capable(idmap
, inode
)) {
341 error
= inode_permission(idmap
, inode
, MAY_WRITE
);
348 EXPORT_SYMBOL(may_setattr
);
351 * notify_change - modify attributes of a filesystem object
352 * @idmap: idmap of the mount the inode was found from
353 * @dentry: object affected
354 * @attr: new attributes
355 * @delegated_inode: returns inode, if the inode is delegated
357 * The caller must hold the i_mutex on the affected object.
359 * If notify_change discovers a delegation in need of breaking,
360 * it will return -EWOULDBLOCK and return a reference to the inode in
361 * delegated_inode. The caller should then break the delegation and
362 * retry. Because breaking a delegation may take a long time, the
363 * caller should drop the i_mutex before doing so.
365 * Alternatively, a caller may pass NULL for delegated_inode. This may
366 * be appropriate for callers that expect the underlying filesystem not
367 * to be NFS exported. Also, passing NULL is fine for callers holding
368 * the file open for write, as there can be no conflicting delegation in
371 * If the inode has been found through an idmapped mount the idmap of
372 * the vfsmount must be passed through @idmap. This function will then
373 * take care to map the inode according to @idmap before checking
374 * permissions. On non-idmapped mounts or if permission checking is to be
375 * performed on the raw inode simply pass @nop_mnt_idmap.
377 int notify_change(struct mnt_idmap
*idmap
, struct dentry
*dentry
,
378 struct iattr
*attr
, struct inode
**delegated_inode
)
380 struct inode
*inode
= dentry
->d_inode
;
381 umode_t mode
= inode
->i_mode
;
383 struct timespec64 now
;
384 unsigned int ia_valid
= attr
->ia_valid
;
386 WARN_ON_ONCE(!inode_is_locked(inode
));
388 error
= may_setattr(idmap
, inode
, ia_valid
);
392 if ((ia_valid
& ATTR_MODE
)) {
394 * Don't allow changing the mode of symlinks:
396 * (1) The vfs doesn't take the mode of symlinks into account
397 * during permission checking.
398 * (2) This has never worked correctly. Most major filesystems
399 * did return EOPNOTSUPP due to interactions with POSIX ACLs
400 * but did still updated the mode of the symlink.
401 * This inconsistency led system call wrapper providers such
402 * as libc to block changing the mode of symlinks with
403 * EOPNOTSUPP already.
404 * (3) To even do this in the first place one would have to use
405 * specific file descriptors and quite some effort.
407 if (S_ISLNK(inode
->i_mode
))
410 /* Flag setting protected by i_mutex */
411 if (is_sxid(attr
->ia_mode
))
412 inode
->i_flags
&= ~S_NOSEC
;
415 now
= current_time(inode
);
417 attr
->ia_ctime
= now
;
418 if (!(ia_valid
& ATTR_ATIME_SET
))
419 attr
->ia_atime
= now
;
421 attr
->ia_atime
= timestamp_truncate(attr
->ia_atime
, inode
);
422 if (!(ia_valid
& ATTR_MTIME_SET
))
423 attr
->ia_mtime
= now
;
425 attr
->ia_mtime
= timestamp_truncate(attr
->ia_mtime
, inode
);
427 if (ia_valid
& ATTR_KILL_PRIV
) {
428 error
= security_inode_need_killpriv(dentry
);
432 ia_valid
= attr
->ia_valid
&= ~ATTR_KILL_PRIV
;
436 * We now pass ATTR_KILL_S*ID to the lower level setattr function so
437 * that the function has the ability to reinterpret a mode change
438 * that's due to these bits. This adds an implicit restriction that
439 * no function will ever call notify_change with both ATTR_MODE and
440 * ATTR_KILL_S*ID set.
442 if ((ia_valid
& (ATTR_KILL_SUID
|ATTR_KILL_SGID
)) &&
443 (ia_valid
& ATTR_MODE
))
446 if (ia_valid
& ATTR_KILL_SUID
) {
447 if (mode
& S_ISUID
) {
448 ia_valid
= attr
->ia_valid
|= ATTR_MODE
;
449 attr
->ia_mode
= (inode
->i_mode
& ~S_ISUID
);
452 if (ia_valid
& ATTR_KILL_SGID
) {
453 if (mode
& S_ISGID
) {
454 if (!(ia_valid
& ATTR_MODE
)) {
455 ia_valid
= attr
->ia_valid
|= ATTR_MODE
;
456 attr
->ia_mode
= inode
->i_mode
;
458 attr
->ia_mode
&= ~S_ISGID
;
461 if (!(attr
->ia_valid
& ~(ATTR_KILL_SUID
| ATTR_KILL_SGID
)))
465 * Verify that uid/gid changes are valid in the target
466 * namespace of the superblock.
468 if (ia_valid
& ATTR_UID
&&
469 !vfsuid_has_fsmapping(idmap
, inode
->i_sb
->s_user_ns
,
472 if (ia_valid
& ATTR_GID
&&
473 !vfsgid_has_fsmapping(idmap
, inode
->i_sb
->s_user_ns
,
477 /* Don't allow modifications of files with invalid uids or
478 * gids unless those uids & gids are being made valid.
480 if (!(ia_valid
& ATTR_UID
) &&
481 !vfsuid_valid(i_uid_into_vfsuid(idmap
, inode
)))
483 if (!(ia_valid
& ATTR_GID
) &&
484 !vfsgid_valid(i_gid_into_vfsgid(idmap
, inode
)))
487 error
= security_inode_setattr(idmap
, dentry
, attr
);
492 * If ATTR_DELEG is set, then these attributes are being set on
493 * behalf of the holder of a write delegation. We want to avoid
494 * breaking the delegation in this case.
496 if (!(ia_valid
& ATTR_DELEG
)) {
497 error
= try_break_deleg(inode
, delegated_inode
);
502 if (inode
->i_op
->setattr
)
503 error
= inode
->i_op
->setattr(idmap
, dentry
, attr
);
505 error
= simple_setattr(idmap
, dentry
, attr
);
508 fsnotify_change(dentry
, ia_valid
);
509 security_inode_post_setattr(idmap
, dentry
, ia_valid
);
514 EXPORT_SYMBOL(notify_change
);