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 * @idmap: idmap 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 idmap of
156 * the vfsmount must be passed through @idmap. This function will then
157 * take care to map the inode according to @idmap before checking
158 * permissions. On non-idmapped mounts or if permission checking is to be
159 * performed on the raw inode simply passs @nop_mnt_idmap.
161 * Should be called as the first thing in ->setattr implementations,
162 * possibly after taking additional locks.
164 int setattr_prepare(struct mnt_idmap *idmap, struct dentry *dentry,
167 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
168 struct inode *inode = d_inode(dentry);
169 unsigned int ia_valid = attr->ia_valid;
172 * First check size constraints. These can't be overriden using
175 if (ia_valid & ATTR_SIZE) {
176 int error = inode_newsize_ok(inode, attr->ia_size);
181 /* If force is set do it anyway. */
182 if (ia_valid & ATTR_FORCE)
185 /* Make sure a caller can chown. */
186 if ((ia_valid & ATTR_UID) &&
187 !chown_ok(mnt_userns, inode, attr->ia_vfsuid))
190 /* Make sure caller can chgrp. */
191 if ((ia_valid & ATTR_GID) &&
192 !chgrp_ok(mnt_userns, inode, attr->ia_vfsgid))
195 /* Make sure a caller can chmod. */
196 if (ia_valid & ATTR_MODE) {
199 if (!inode_owner_or_capable(mnt_userns, inode))
202 if (ia_valid & ATTR_GID)
203 vfsgid = attr->ia_vfsgid;
205 vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
207 /* Also check the setgid bit! */
208 if (!in_group_or_capable(mnt_userns, inode, vfsgid))
209 attr->ia_mode &= ~S_ISGID;
212 /* Check for setting the inode time. */
213 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)) {
214 if (!inode_owner_or_capable(mnt_userns, inode))
219 /* User has permission for the change */
220 if (ia_valid & ATTR_KILL_PRIV) {
223 error = security_inode_killpriv(mnt_userns, dentry);
230 EXPORT_SYMBOL(setattr_prepare);
233 * inode_newsize_ok - may this inode be truncated to a given size
234 * @inode: the inode to be truncated
235 * @offset: the new size to assign to the inode
237 * inode_newsize_ok must be called with i_mutex held.
239 * inode_newsize_ok will check filesystem limits and ulimits to check that the
240 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
241 * when necessary. Caller must not proceed with inode size change if failure is
242 * returned. @inode must be a file (not directory), with appropriate
243 * permissions to allow truncate (inode_newsize_ok does NOT check these
246 * Return: 0 on success, -ve errno on failure
248 int inode_newsize_ok(const struct inode *inode, loff_t offset)
252 if (inode->i_size < offset) {
255 limit = rlimit(RLIMIT_FSIZE);
256 if (limit != RLIM_INFINITY && offset > limit)
258 if (offset > inode->i_sb->s_maxbytes)
262 * truncation of in-use swapfiles is disallowed - it would
263 * cause subsequent swapout to scribble on the now-freed
266 if (IS_SWAPFILE(inode))
272 send_sig(SIGXFSZ, current, 0);
276 EXPORT_SYMBOL(inode_newsize_ok);
279 * setattr_copy - copy simple metadata updates into the generic inode
280 * @idmap: idmap of the mount the inode was found from
281 * @inode: the inode to be updated
282 * @attr: the new attributes
284 * setattr_copy must be called with i_mutex held.
286 * setattr_copy updates the inode's metadata with that specified
287 * in attr on idmapped mounts. Necessary permission checks to determine
288 * whether or not the S_ISGID property needs to be removed are performed with
289 * the correct idmapped mount permission helpers.
290 * Noticeably missing is inode size update, which is more complex
291 * as it requires pagecache updates.
293 * If the inode has been found through an idmapped mount the idmap of
294 * the vfsmount must be passed through @idmap. This function will then
295 * take care to map the inode according to @idmap before checking
296 * permissions. On non-idmapped mounts or if permission checking is to be
297 * performed on the raw inode simply pass @nop_mnt_idmap.
299 * The inode is not marked as dirty after this operation. The rationale is
300 * that for "simple" filesystems, the struct inode is the inode storage.
301 * The caller is free to mark the inode dirty afterwards if needed.
303 void setattr_copy(struct mnt_idmap *idmap, struct inode *inode,
304 const struct iattr *attr)
306 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
307 unsigned int ia_valid = attr->ia_valid;
309 i_uid_update(mnt_userns, attr, inode);
310 i_gid_update(mnt_userns, attr, inode);
311 if (ia_valid & ATTR_ATIME)
312 inode->i_atime = attr->ia_atime;
313 if (ia_valid & ATTR_MTIME)
314 inode->i_mtime = attr->ia_mtime;
315 if (ia_valid & ATTR_CTIME)
316 inode->i_ctime = attr->ia_ctime;
317 if (ia_valid & ATTR_MODE) {
318 umode_t mode = attr->ia_mode;
319 if (!in_group_or_capable(mnt_userns, inode,
320 i_gid_into_vfsgid(mnt_userns, inode)))
322 inode->i_mode = mode;
325 EXPORT_SYMBOL(setattr_copy);
327 int may_setattr(struct mnt_idmap *idmap, struct inode *inode,
328 unsigned int ia_valid)
331 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
333 if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_TIMES_SET)) {
334 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
339 * If utimes(2) and friends are called with times == NULL (or both
340 * times are UTIME_NOW), then we need to check for write permission
342 if (ia_valid & ATTR_TOUCH) {
343 if (IS_IMMUTABLE(inode))
346 if (!inode_owner_or_capable(mnt_userns, inode)) {
347 error = inode_permission(idmap, inode, MAY_WRITE);
354 EXPORT_SYMBOL(may_setattr);
357 * notify_change - modify attributes of a filesytem object
358 * @idmap: idmap of the mount the inode was found from
359 * @dentry: object affected
360 * @attr: new attributes
361 * @delegated_inode: returns inode, if the inode is delegated
363 * The caller must hold the i_mutex on the affected object.
365 * If notify_change discovers a delegation in need of breaking,
366 * it will return -EWOULDBLOCK and return a reference to the inode in
367 * delegated_inode. The caller should then break the delegation and
368 * retry. Because breaking a delegation may take a long time, the
369 * caller should drop the i_mutex before doing so.
371 * Alternatively, a caller may pass NULL for delegated_inode. This may
372 * be appropriate for callers that expect the underlying filesystem not
373 * to be NFS exported. Also, passing NULL is fine for callers holding
374 * the file open for write, as there can be no conflicting delegation in
377 * If the inode has been found through an idmapped mount the idmap of
378 * the vfsmount must be passed through @idmap. This function will then
379 * take care to map the inode according to @idmap before checking
380 * permissions. On non-idmapped mounts or if permission checking is to be
381 * performed on the raw inode simply pass @nop_mnt_idmap.
383 int notify_change(struct mnt_idmap *idmap, struct dentry *dentry,
384 struct iattr *attr, struct inode **delegated_inode)
386 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
387 struct inode *inode = dentry->d_inode;
388 umode_t mode = inode->i_mode;
390 struct timespec64 now;
391 unsigned int ia_valid = attr->ia_valid;
393 WARN_ON_ONCE(!inode_is_locked(inode));
395 error = may_setattr(idmap, inode, ia_valid);
399 if ((ia_valid & ATTR_MODE)) {
400 umode_t amode = attr->ia_mode;
401 /* Flag setting protected by i_mutex */
403 inode->i_flags &= ~S_NOSEC;
406 now = current_time(inode);
408 attr->ia_ctime = now;
409 if (!(ia_valid & ATTR_ATIME_SET))
410 attr->ia_atime = now;
412 attr->ia_atime = timestamp_truncate(attr->ia_atime, inode);
413 if (!(ia_valid & ATTR_MTIME_SET))
414 attr->ia_mtime = now;
416 attr->ia_mtime = timestamp_truncate(attr->ia_mtime, inode);
418 if (ia_valid & ATTR_KILL_PRIV) {
419 error = security_inode_need_killpriv(dentry);
423 ia_valid = attr->ia_valid &= ~ATTR_KILL_PRIV;
427 * We now pass ATTR_KILL_S*ID to the lower level setattr function so
428 * that the function has the ability to reinterpret a mode change
429 * that's due to these bits. This adds an implicit restriction that
430 * no function will ever call notify_change with both ATTR_MODE and
431 * ATTR_KILL_S*ID set.
433 if ((ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) &&
434 (ia_valid & ATTR_MODE))
437 if (ia_valid & ATTR_KILL_SUID) {
438 if (mode & S_ISUID) {
439 ia_valid = attr->ia_valid |= ATTR_MODE;
440 attr->ia_mode = (inode->i_mode & ~S_ISUID);
443 if (ia_valid & ATTR_KILL_SGID) {
444 if (mode & S_ISGID) {
445 if (!(ia_valid & ATTR_MODE)) {
446 ia_valid = attr->ia_valid |= ATTR_MODE;
447 attr->ia_mode = inode->i_mode;
449 attr->ia_mode &= ~S_ISGID;
452 if (!(attr->ia_valid & ~(ATTR_KILL_SUID | ATTR_KILL_SGID)))
456 * Verify that uid/gid changes are valid in the target
457 * namespace of the superblock.
459 if (ia_valid & ATTR_UID &&
460 !vfsuid_has_fsmapping(mnt_userns, inode->i_sb->s_user_ns,
463 if (ia_valid & ATTR_GID &&
464 !vfsgid_has_fsmapping(mnt_userns, inode->i_sb->s_user_ns,
468 /* Don't allow modifications of files with invalid uids or
469 * gids unless those uids & gids are being made valid.
471 if (!(ia_valid & ATTR_UID) &&
472 !vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)))
474 if (!(ia_valid & ATTR_GID) &&
475 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
478 error = security_inode_setattr(idmap, dentry, attr);
481 error = try_break_deleg(inode, delegated_inode);
485 if (inode->i_op->setattr)
486 error = inode->i_op->setattr(idmap, dentry, attr);
488 error = simple_setattr(idmap, dentry, attr);
491 fsnotify_change(dentry, ia_valid);
492 ima_inode_post_setattr(mnt_userns, dentry);
493 evm_inode_post_setattr(dentry, ia_valid);
498 EXPORT_SYMBOL(notify_change);