1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/attr.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * changes by Thomas Schoebel-Theuer
7 */
8
9#include <linux/export.h>
10#include <linux/time.h>
11#include <linux/mm.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>
19#include <linux/evm.h>
20#include <linux/ima.h>
21
22#include "internal.h"
23
24/**
25 * setattr_should_drop_sgid - determine whether the setgid bit needs to be
26 * removed
27 * @idmap: idmap of the mount @inode was found from
28 * @inode: inode to check
29 *
30 * This function determines whether the setgid bit needs to be removed.
31 * We retain backwards compatibility and require setgid bit to be removed
32 * unconditionally if S_IXGRP is set. Otherwise we have the exact same
33 * requirements as setattr_prepare() and setattr_copy().
34 *
35 * Return: ATTR_KILL_SGID if setgid bit needs to be removed, 0 otherwise.
36 */
37int setattr_should_drop_sgid(struct mnt_idmap *idmap,
38 const struct inode *inode)
39{
40 umode_t mode = inode->i_mode;
41
42 if (!(mode & S_ISGID))
43 return 0;
44 if (mode & S_IXGRP)
45 return ATTR_KILL_SGID;
46 if (!in_group_or_capable(idmap, inode, vfsgid: i_gid_into_vfsgid(idmap, inode)))
47 return ATTR_KILL_SGID;
48 return 0;
49}
50EXPORT_SYMBOL(setattr_should_drop_sgid);
51
52/**
53 * setattr_should_drop_suidgid - determine whether the set{g,u}id bit needs to
54 * be dropped
55 * @idmap: idmap of the mount @inode was found from
56 * @inode: inode to check
57 *
58 * This function determines whether the set{g,u}id bits need to be removed.
59 * If the setuid bit needs to be removed ATTR_KILL_SUID is returned. If the
60 * setgid bit needs to be removed ATTR_KILL_SGID is returned. If both
61 * set{g,u}id bits need to be removed the corresponding mask of both flags is
62 * returned.
63 *
64 * Return: A mask of ATTR_KILL_S{G,U}ID indicating which - if any - setid bits
65 * to remove, 0 otherwise.
66 */
67int setattr_should_drop_suidgid(struct mnt_idmap *idmap,
68 struct inode *inode)
69{
70 umode_t mode = inode->i_mode;
71 int kill = 0;
72
73 /* suid always must be killed */
74 if (unlikely(mode & S_ISUID))
75 kill = ATTR_KILL_SUID;
76
77 kill |= setattr_should_drop_sgid(idmap, inode);
78
79 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
80 return kill;
81
82 return 0;
83}
84EXPORT_SYMBOL(setattr_should_drop_suidgid);
85
86/**
87 * chown_ok - verify permissions to chown inode
88 * @idmap: idmap of the mount @inode was found from
89 * @inode: inode to check permissions on
90 * @ia_vfsuid: uid to chown @inode to
91 *
92 * If the inode has been found through an idmapped mount the idmap of
93 * the vfsmount must be passed through @idmap. This function will then
94 * take care to map the inode according to @idmap before checking
95 * permissions. On non-idmapped mounts or if permission checking is to be
96 * performed on the raw inode simply pass @nop_mnt_idmap.
97 */
98static bool chown_ok(struct mnt_idmap *idmap,
99 const struct inode *inode, vfsuid_t ia_vfsuid)
100{
101 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
102 if (vfsuid_eq_kuid(vfsuid, current_fsuid()) &&
103 vfsuid_eq(left: ia_vfsuid, right: vfsuid))
104 return true;
105 if (capable_wrt_inode_uidgid(idmap, inode, CAP_CHOWN))
106 return true;
107 if (!vfsuid_valid(uid: vfsuid) &&
108 ns_capable(ns: inode->i_sb->s_user_ns, CAP_CHOWN))
109 return true;
110 return false;
111}
112
113/**
114 * chgrp_ok - verify permissions to chgrp inode
115 * @idmap: idmap of the mount @inode was found from
116 * @inode: inode to check permissions on
117 * @ia_vfsgid: gid to chown @inode to
118 *
119 * If the inode has been found through an idmapped mount the idmap of
120 * the vfsmount must be passed through @idmap. This function will then
121 * take care to map the inode according to @idmap before checking
122 * permissions. On non-idmapped mounts or if permission checking is to be
123 * performed on the raw inode simply pass @nop_mnt_idmap.
124 */
125static bool chgrp_ok(struct mnt_idmap *idmap,
126 const struct inode *inode, vfsgid_t ia_vfsgid)
127{
128 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
129 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
130 if (vfsuid_eq_kuid(vfsuid, current_fsuid())) {
131 if (vfsgid_eq(left: ia_vfsgid, right: vfsgid))
132 return true;
133 if (vfsgid_in_group_p(vfsgid: ia_vfsgid))
134 return true;
135 }
136 if (capable_wrt_inode_uidgid(idmap, inode, CAP_CHOWN))
137 return true;
138 if (!vfsgid_valid(gid: vfsgid) &&
139 ns_capable(ns: inode->i_sb->s_user_ns, CAP_CHOWN))
140 return true;
141 return false;
142}
143
144/**
145 * setattr_prepare - check if attribute changes to a dentry are allowed
146 * @idmap: idmap of the mount the inode was found from
147 * @dentry: dentry to check
148 * @attr: attributes to change
149 *
150 * Check if we are allowed to change the attributes contained in @attr
151 * in the given dentry. This includes the normal unix access permission
152 * checks, as well as checks for rlimits and others. The function also clears
153 * SGID bit from mode if user is not allowed to set it. Also file capabilities
154 * and IMA extended attributes are cleared if ATTR_KILL_PRIV is set.
155 *
156 * If the inode has been found through an idmapped mount the idmap of
157 * the vfsmount must be passed through @idmap. This function will then
158 * take care to map the inode according to @idmap before checking
159 * permissions. On non-idmapped mounts or if permission checking is to be
160 * performed on the raw inode simply passs @nop_mnt_idmap.
161 *
162 * Should be called as the first thing in ->setattr implementations,
163 * possibly after taking additional locks.
164 */
165int setattr_prepare(struct mnt_idmap *idmap, struct dentry *dentry,
166 struct iattr *attr)
167{
168 struct inode *inode = d_inode(dentry);
169 unsigned int ia_valid = attr->ia_valid;
170
171 /*
172 * First check size constraints. These can't be overriden using
173 * ATTR_FORCE.
174 */
175 if (ia_valid & ATTR_SIZE) {
176 int error = inode_newsize_ok(inode, offset: attr->ia_size);
177 if (error)
178 return error;
179 }
180
181 /* If force is set do it anyway. */
182 if (ia_valid & ATTR_FORCE)
183 goto kill_priv;
184
185 /* Make sure a caller can chown. */
186 if ((ia_valid & ATTR_UID) &&
187 !chown_ok(idmap, inode, ia_vfsuid: attr->ia_vfsuid))
188 return -EPERM;
189
190 /* Make sure caller can chgrp. */
191 if ((ia_valid & ATTR_GID) &&
192 !chgrp_ok(idmap, inode, ia_vfsgid: attr->ia_vfsgid))
193 return -EPERM;
194
195 /* Make sure a caller can chmod. */
196 if (ia_valid & ATTR_MODE) {
197 vfsgid_t vfsgid;
198
199 if (!inode_owner_or_capable(idmap, inode))
200 return -EPERM;
201
202 if (ia_valid & ATTR_GID)
203 vfsgid = attr->ia_vfsgid;
204 else
205 vfsgid = i_gid_into_vfsgid(idmap, inode);
206
207 /* Also check the setgid bit! */
208 if (!in_group_or_capable(idmap, inode, vfsgid))
209 attr->ia_mode &= ~S_ISGID;
210 }
211
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(idmap, inode))
215 return -EPERM;
216 }
217
218kill_priv:
219 /* User has permission for the change */
220 if (ia_valid & ATTR_KILL_PRIV) {
221 int error;
222
223 error = security_inode_killpriv(idmap, dentry);
224 if (error)
225 return error;
226 }
227
228 return 0;
229}
230EXPORT_SYMBOL(setattr_prepare);
231
232/**
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
236 *
237 * inode_newsize_ok must be called with i_mutex held.
238 *
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
244 * conditions).
245 *
246 * Return: 0 on success, -ve errno on failure
247 */
248int inode_newsize_ok(const struct inode *inode, loff_t offset)
249{
250 if (offset < 0)
251 return -EINVAL;
252 if (inode->i_size < offset) {
253 unsigned long limit;
254
255 limit = rlimit(RLIMIT_FSIZE);
256 if (limit != RLIM_INFINITY && offset > limit)
257 goto out_sig;
258 if (offset > inode->i_sb->s_maxbytes)
259 goto out_big;
260 } else {
261 /*
262 * truncation of in-use swapfiles is disallowed - it would
263 * cause subsequent swapout to scribble on the now-freed
264 * blocks.
265 */
266 if (IS_SWAPFILE(inode))
267 return -ETXTBSY;
268 }
269
270 return 0;
271out_sig:
272 send_sig(SIGXFSZ, current, 0);
273out_big:
274 return -EFBIG;
275}
276EXPORT_SYMBOL(inode_newsize_ok);
277
278/**
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
283 *
284 * setattr_copy must be called with i_mutex held.
285 *
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.
292 *
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.
298 *
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.
302 */
303void setattr_copy(struct mnt_idmap *idmap, struct inode *inode,
304 const struct iattr *attr)
305{
306 unsigned int ia_valid = attr->ia_valid;
307
308 i_uid_update(idmap, attr, inode);
309 i_gid_update(idmap, attr, inode);
310 if (ia_valid & ATTR_ATIME)
311 inode_set_atime_to_ts(inode, ts: attr->ia_atime);
312 if (ia_valid & ATTR_MTIME)
313 inode_set_mtime_to_ts(inode, ts: attr->ia_mtime);
314 if (ia_valid & ATTR_CTIME)
315 inode_set_ctime_to_ts(inode, ts: attr->ia_ctime);
316 if (ia_valid & ATTR_MODE) {
317 umode_t mode = attr->ia_mode;
318 if (!in_group_or_capable(idmap, inode,
319 vfsgid: i_gid_into_vfsgid(idmap, inode)))
320 mode &= ~S_ISGID;
321 inode->i_mode = mode;
322 }
323}
324EXPORT_SYMBOL(setattr_copy);
325
326int may_setattr(struct mnt_idmap *idmap, struct inode *inode,
327 unsigned int ia_valid)
328{
329 int error;
330
331 if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_TIMES_SET)) {
332 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
333 return -EPERM;
334 }
335
336 /*
337 * If utimes(2) and friends are called with times == NULL (or both
338 * times are UTIME_NOW), then we need to check for write permission
339 */
340 if (ia_valid & ATTR_TOUCH) {
341 if (IS_IMMUTABLE(inode))
342 return -EPERM;
343
344 if (!inode_owner_or_capable(idmap, inode)) {
345 error = inode_permission(idmap, inode, MAY_WRITE);
346 if (error)
347 return error;
348 }
349 }
350 return 0;
351}
352EXPORT_SYMBOL(may_setattr);
353
354/**
355 * notify_change - modify attributes of a filesytem object
356 * @idmap: idmap of the mount the inode was found from
357 * @dentry: object affected
358 * @attr: new attributes
359 * @delegated_inode: returns inode, if the inode is delegated
360 *
361 * The caller must hold the i_mutex on the affected object.
362 *
363 * If notify_change discovers a delegation in need of breaking,
364 * it will return -EWOULDBLOCK and return a reference to the inode in
365 * delegated_inode. The caller should then break the delegation and
366 * retry. Because breaking a delegation may take a long time, the
367 * caller should drop the i_mutex before doing so.
368 *
369 * Alternatively, a caller may pass NULL for delegated_inode. This may
370 * be appropriate for callers that expect the underlying filesystem not
371 * to be NFS exported. Also, passing NULL is fine for callers holding
372 * the file open for write, as there can be no conflicting delegation in
373 * that case.
374 *
375 * If the inode has been found through an idmapped mount the idmap of
376 * the vfsmount must be passed through @idmap. This function will then
377 * take care to map the inode according to @idmap before checking
378 * permissions. On non-idmapped mounts or if permission checking is to be
379 * performed on the raw inode simply pass @nop_mnt_idmap.
380 */
381int notify_change(struct mnt_idmap *idmap, struct dentry *dentry,
382 struct iattr *attr, struct inode **delegated_inode)
383{
384 struct inode *inode = dentry->d_inode;
385 umode_t mode = inode->i_mode;
386 int error;
387 struct timespec64 now;
388 unsigned int ia_valid = attr->ia_valid;
389
390 WARN_ON_ONCE(!inode_is_locked(inode));
391
392 error = may_setattr(idmap, inode, ia_valid);
393 if (error)
394 return error;
395
396 if ((ia_valid & ATTR_MODE)) {
397 /*
398 * Don't allow changing the mode of symlinks:
399 *
400 * (1) The vfs doesn't take the mode of symlinks into account
401 * during permission checking.
402 * (2) This has never worked correctly. Most major filesystems
403 * did return EOPNOTSUPP due to interactions with POSIX ACLs
404 * but did still updated the mode of the symlink.
405 * This inconsistency led system call wrapper providers such
406 * as libc to block changing the mode of symlinks with
407 * EOPNOTSUPP already.
408 * (3) To even do this in the first place one would have to use
409 * specific file descriptors and quite some effort.
410 */
411 if (S_ISLNK(inode->i_mode))
412 return -EOPNOTSUPP;
413
414 /* Flag setting protected by i_mutex */
415 if (is_sxid(mode: attr->ia_mode))
416 inode->i_flags &= ~S_NOSEC;
417 }
418
419 now = current_time(inode);
420
421 attr->ia_ctime = now;
422 if (!(ia_valid & ATTR_ATIME_SET))
423 attr->ia_atime = now;
424 else
425 attr->ia_atime = timestamp_truncate(t: attr->ia_atime, inode);
426 if (!(ia_valid & ATTR_MTIME_SET))
427 attr->ia_mtime = now;
428 else
429 attr->ia_mtime = timestamp_truncate(t: attr->ia_mtime, inode);
430
431 if (ia_valid & ATTR_KILL_PRIV) {
432 error = security_inode_need_killpriv(dentry);
433 if (error < 0)
434 return error;
435 if (error == 0)
436 ia_valid = attr->ia_valid &= ~ATTR_KILL_PRIV;
437 }
438
439 /*
440 * We now pass ATTR_KILL_S*ID to the lower level setattr function so
441 * that the function has the ability to reinterpret a mode change
442 * that's due to these bits. This adds an implicit restriction that
443 * no function will ever call notify_change with both ATTR_MODE and
444 * ATTR_KILL_S*ID set.
445 */
446 if ((ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) &&
447 (ia_valid & ATTR_MODE))
448 BUG();
449
450 if (ia_valid & ATTR_KILL_SUID) {
451 if (mode & S_ISUID) {
452 ia_valid = attr->ia_valid |= ATTR_MODE;
453 attr->ia_mode = (inode->i_mode & ~S_ISUID);
454 }
455 }
456 if (ia_valid & ATTR_KILL_SGID) {
457 if (mode & S_ISGID) {
458 if (!(ia_valid & ATTR_MODE)) {
459 ia_valid = attr->ia_valid |= ATTR_MODE;
460 attr->ia_mode = inode->i_mode;
461 }
462 attr->ia_mode &= ~S_ISGID;
463 }
464 }
465 if (!(attr->ia_valid & ~(ATTR_KILL_SUID | ATTR_KILL_SGID)))
466 return 0;
467
468 /*
469 * Verify that uid/gid changes are valid in the target
470 * namespace of the superblock.
471 */
472 if (ia_valid & ATTR_UID &&
473 !vfsuid_has_fsmapping(idmap, fs_userns: inode->i_sb->s_user_ns,
474 vfsuid: attr->ia_vfsuid))
475 return -EOVERFLOW;
476 if (ia_valid & ATTR_GID &&
477 !vfsgid_has_fsmapping(idmap, fs_userns: inode->i_sb->s_user_ns,
478 vfsgid: attr->ia_vfsgid))
479 return -EOVERFLOW;
480
481 /* Don't allow modifications of files with invalid uids or
482 * gids unless those uids & gids are being made valid.
483 */
484 if (!(ia_valid & ATTR_UID) &&
485 !vfsuid_valid(uid: i_uid_into_vfsuid(idmap, inode)))
486 return -EOVERFLOW;
487 if (!(ia_valid & ATTR_GID) &&
488 !vfsgid_valid(gid: i_gid_into_vfsgid(idmap, inode)))
489 return -EOVERFLOW;
490
491 error = security_inode_setattr(idmap, dentry, attr);
492 if (error)
493 return error;
494 error = try_break_deleg(inode, delegated_inode);
495 if (error)
496 return error;
497
498 if (inode->i_op->setattr)
499 error = inode->i_op->setattr(idmap, dentry, attr);
500 else
501 error = simple_setattr(idmap, dentry, attr);
502
503 if (!error) {
504 fsnotify_change(dentry, ia_valid);
505 ima_inode_post_setattr(idmap, dentry);
506 evm_inode_post_setattr(dentry, ia_valid);
507 }
508
509 return error;
510}
511EXPORT_SYMBOL(notify_change);
512

source code of linux/fs/attr.c