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
20#include "internal.h"
21
22/**
23 * setattr_should_drop_sgid - determine whether the setgid bit needs to be
24 * removed
25 * @idmap: idmap of the mount @inode was found from
26 * @inode: inode to check
27 *
28 * This function determines whether the setgid bit needs to be removed.
29 * We retain backwards compatibility and require setgid bit to be removed
30 * unconditionally if S_IXGRP is set. Otherwise we have the exact same
31 * requirements as setattr_prepare() and setattr_copy().
32 *
33 * Return: ATTR_KILL_SGID if setgid bit needs to be removed, 0 otherwise.
34 */
35int setattr_should_drop_sgid(struct mnt_idmap *idmap,
36 const struct inode *inode)
37{
38 umode_t mode = inode->i_mode;
39
40 if (!(mode & S_ISGID))
41 return 0;
42 if (mode & S_IXGRP)
43 return ATTR_KILL_SGID;
44 if (!in_group_or_capable(idmap, inode, vfsgid: i_gid_into_vfsgid(idmap, inode)))
45 return ATTR_KILL_SGID;
46 return 0;
47}
48EXPORT_SYMBOL(setattr_should_drop_sgid);
49
50/**
51 * setattr_should_drop_suidgid - determine whether the set{g,u}id bit needs to
52 * be dropped
53 * @idmap: idmap of the mount @inode was found from
54 * @inode: inode to check
55 *
56 * This function determines whether the set{g,u}id bits need to be removed.
57 * If the setuid bit needs to be removed ATTR_KILL_SUID is returned. If the
58 * setgid bit needs to be removed ATTR_KILL_SGID is returned. If both
59 * set{g,u}id bits need to be removed the corresponding mask of both flags is
60 * returned.
61 *
62 * Return: A mask of ATTR_KILL_S{G,U}ID indicating which - if any - setid bits
63 * to remove, 0 otherwise.
64 */
65int setattr_should_drop_suidgid(struct mnt_idmap *idmap,
66 struct inode *inode)
67{
68 umode_t mode = inode->i_mode;
69 int kill = 0;
70
71 /* suid always must be killed */
72 if (unlikely(mode & S_ISUID))
73 kill = ATTR_KILL_SUID;
74
75 kill |= setattr_should_drop_sgid(idmap, inode);
76
77 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
78 return kill;
79
80 return 0;
81}
82EXPORT_SYMBOL(setattr_should_drop_suidgid);
83
84/**
85 * chown_ok - verify permissions to chown inode
86 * @idmap: idmap of the mount @inode was found from
87 * @inode: inode to check permissions on
88 * @ia_vfsuid: uid to chown @inode to
89 *
90 * If the inode has been found through an idmapped mount the idmap of
91 * the vfsmount must be passed through @idmap. This function will then
92 * take care to map the inode according to @idmap before checking
93 * permissions. On non-idmapped mounts or if permission checking is to be
94 * performed on the raw inode simply pass @nop_mnt_idmap.
95 */
96static bool chown_ok(struct mnt_idmap *idmap,
97 const struct inode *inode, vfsuid_t ia_vfsuid)
98{
99 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
100 if (vfsuid_eq_kuid(vfsuid, current_fsuid()) &&
101 vfsuid_eq(left: ia_vfsuid, right: vfsuid))
102 return true;
103 if (capable_wrt_inode_uidgid(idmap, inode, CAP_CHOWN))
104 return true;
105 if (!vfsuid_valid(uid: vfsuid) &&
106 ns_capable(ns: inode->i_sb->s_user_ns, CAP_CHOWN))
107 return true;
108 return false;
109}
110
111/**
112 * chgrp_ok - verify permissions to chgrp inode
113 * @idmap: idmap of the mount @inode was found from
114 * @inode: inode to check permissions on
115 * @ia_vfsgid: gid to chown @inode to
116 *
117 * If the inode has been found through an idmapped mount the idmap of
118 * the vfsmount must be passed through @idmap. This function will then
119 * take care to map the inode according to @idmap before checking
120 * permissions. On non-idmapped mounts or if permission checking is to be
121 * performed on the raw inode simply pass @nop_mnt_idmap.
122 */
123static bool chgrp_ok(struct mnt_idmap *idmap,
124 const struct inode *inode, vfsgid_t ia_vfsgid)
125{
126 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
127 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
128 if (vfsuid_eq_kuid(vfsuid, current_fsuid())) {
129 if (vfsgid_eq(left: ia_vfsgid, right: vfsgid))
130 return true;
131 if (vfsgid_in_group_p(vfsgid: ia_vfsgid))
132 return true;
133 }
134 if (capable_wrt_inode_uidgid(idmap, inode, CAP_CHOWN))
135 return true;
136 if (!vfsgid_valid(gid: vfsgid) &&
137 ns_capable(ns: inode->i_sb->s_user_ns, CAP_CHOWN))
138 return true;
139 return false;
140}
141
142/**
143 * setattr_prepare - check if attribute changes to a dentry are allowed
144 * @idmap: idmap of the mount the inode was found from
145 * @dentry: dentry to check
146 * @attr: attributes to change
147 *
148 * Check if we are allowed to change the attributes contained in @attr
149 * in the given dentry. This includes the normal unix access permission
150 * checks, as well as checks for rlimits and others. The function also clears
151 * SGID bit from mode if user is not allowed to set it. Also file capabilities
152 * and IMA extended attributes are cleared if ATTR_KILL_PRIV is set.
153 *
154 * If the inode has been found through an idmapped mount the idmap of
155 * the vfsmount must be passed through @idmap. This function will then
156 * take care to map the inode according to @idmap before checking
157 * permissions. On non-idmapped mounts or if permission checking is to be
158 * performed on the raw inode simply pass @nop_mnt_idmap.
159 *
160 * Should be called as the first thing in ->setattr implementations,
161 * possibly after taking additional locks.
162 */
163int setattr_prepare(struct mnt_idmap *idmap, struct dentry *dentry,
164 struct iattr *attr)
165{
166 struct inode *inode = d_inode(dentry);
167 unsigned int ia_valid = attr->ia_valid;
168
169 /*
170 * First check size constraints. These can't be overriden using
171 * ATTR_FORCE.
172 */
173 if (ia_valid & ATTR_SIZE) {
174 int error = inode_newsize_ok(inode, offset: attr->ia_size);
175 if (error)
176 return error;
177 }
178
179 /* If force is set do it anyway. */
180 if (ia_valid & ATTR_FORCE)
181 goto kill_priv;
182
183 /* Make sure a caller can chown. */
184 if ((ia_valid & ATTR_UID) &&
185 !chown_ok(idmap, inode, ia_vfsuid: attr->ia_vfsuid))
186 return -EPERM;
187
188 /* Make sure caller can chgrp. */
189 if ((ia_valid & ATTR_GID) &&
190 !chgrp_ok(idmap, inode, ia_vfsgid: attr->ia_vfsgid))
191 return -EPERM;
192
193 /* Make sure a caller can chmod. */
194 if (ia_valid & ATTR_MODE) {
195 vfsgid_t vfsgid;
196
197 if (!inode_owner_or_capable(idmap, inode))
198 return -EPERM;
199
200 if (ia_valid & ATTR_GID)
201 vfsgid = attr->ia_vfsgid;
202 else
203 vfsgid = i_gid_into_vfsgid(idmap, inode);
204
205 /* Also check the setgid bit! */
206 if (!in_group_or_capable(idmap, inode, vfsgid))
207 attr->ia_mode &= ~S_ISGID;
208 }
209
210 /* Check for setting the inode time. */
211 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)) {
212 if (!inode_owner_or_capable(idmap, inode))
213 return -EPERM;
214 }
215
216kill_priv:
217 /* User has permission for the change */
218 if (ia_valid & ATTR_KILL_PRIV) {
219 int error;
220
221 error = security_inode_killpriv(idmap, dentry);
222 if (error)
223 return error;
224 }
225
226 return 0;
227}
228EXPORT_SYMBOL(setattr_prepare);
229
230/**
231 * inode_newsize_ok - may this inode be truncated to a given size
232 * @inode: the inode to be truncated
233 * @offset: the new size to assign to the inode
234 *
235 * inode_newsize_ok must be called with i_mutex held.
236 *
237 * inode_newsize_ok will check filesystem limits and ulimits to check that the
238 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
239 * when necessary. Caller must not proceed with inode size change if failure is
240 * returned. @inode must be a file (not directory), with appropriate
241 * permissions to allow truncate (inode_newsize_ok does NOT check these
242 * conditions).
243 *
244 * Return: 0 on success, -ve errno on failure
245 */
246int inode_newsize_ok(const struct inode *inode, loff_t offset)
247{
248 if (offset < 0)
249 return -EINVAL;
250 if (inode->i_size < offset) {
251 unsigned long limit;
252
253 limit = rlimit(RLIMIT_FSIZE);
254 if (limit != RLIM_INFINITY && offset > limit)
255 goto out_sig;
256 if (offset > inode->i_sb->s_maxbytes)
257 goto out_big;
258 } else {
259 /*
260 * truncation of in-use swapfiles is disallowed - it would
261 * cause subsequent swapout to scribble on the now-freed
262 * blocks.
263 */
264 if (IS_SWAPFILE(inode))
265 return -ETXTBSY;
266 }
267
268 return 0;
269out_sig:
270 send_sig(SIGXFSZ, current, 0);
271out_big:
272 return -EFBIG;
273}
274EXPORT_SYMBOL(inode_newsize_ok);
275
276/**
277 * setattr_copy - copy simple metadata updates into the generic inode
278 * @idmap: idmap of the mount the inode was found from
279 * @inode: the inode to be updated
280 * @attr: the new attributes
281 *
282 * setattr_copy must be called with i_mutex held.
283 *
284 * setattr_copy updates the inode's metadata with that specified
285 * in attr on idmapped mounts. Necessary permission checks to determine
286 * whether or not the S_ISGID property needs to be removed are performed with
287 * the correct idmapped mount permission helpers.
288 * Noticeably missing is inode size update, which is more complex
289 * as it requires pagecache updates.
290 *
291 * If the inode has been found through an idmapped mount the idmap of
292 * the vfsmount must be passed through @idmap. This function will then
293 * take care to map the inode according to @idmap before checking
294 * permissions. On non-idmapped mounts or if permission checking is to be
295 * performed on the raw inode simply pass @nop_mnt_idmap.
296 *
297 * The inode is not marked as dirty after this operation. The rationale is
298 * that for "simple" filesystems, the struct inode is the inode storage.
299 * The caller is free to mark the inode dirty afterwards if needed.
300 */
301void setattr_copy(struct mnt_idmap *idmap, struct inode *inode,
302 const struct iattr *attr)
303{
304 unsigned int ia_valid = attr->ia_valid;
305
306 i_uid_update(idmap, attr, inode);
307 i_gid_update(idmap, attr, inode);
308 if (ia_valid & ATTR_ATIME)
309 inode_set_atime_to_ts(inode, ts: attr->ia_atime);
310 if (ia_valid & ATTR_MTIME)
311 inode_set_mtime_to_ts(inode, ts: attr->ia_mtime);
312 if (ia_valid & ATTR_CTIME)
313 inode_set_ctime_to_ts(inode, ts: attr->ia_ctime);
314 if (ia_valid & ATTR_MODE) {
315 umode_t mode = attr->ia_mode;
316 if (!in_group_or_capable(idmap, inode,
317 vfsgid: i_gid_into_vfsgid(idmap, inode)))
318 mode &= ~S_ISGID;
319 inode->i_mode = mode;
320 }
321}
322EXPORT_SYMBOL(setattr_copy);
323
324int may_setattr(struct mnt_idmap *idmap, struct inode *inode,
325 unsigned int ia_valid)
326{
327 int error;
328
329 if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_TIMES_SET)) {
330 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
331 return -EPERM;
332 }
333
334 /*
335 * If utimes(2) and friends are called with times == NULL (or both
336 * times are UTIME_NOW), then we need to check for write permission
337 */
338 if (ia_valid & ATTR_TOUCH) {
339 if (IS_IMMUTABLE(inode))
340 return -EPERM;
341
342 if (!inode_owner_or_capable(idmap, inode)) {
343 error = inode_permission(idmap, inode, MAY_WRITE);
344 if (error)
345 return error;
346 }
347 }
348 return 0;
349}
350EXPORT_SYMBOL(may_setattr);
351
352/**
353 * notify_change - modify attributes of a filesystem object
354 * @idmap: idmap of the mount the inode was found from
355 * @dentry: object affected
356 * @attr: new attributes
357 * @delegated_inode: returns inode, if the inode is delegated
358 *
359 * The caller must hold the i_mutex on the affected object.
360 *
361 * If notify_change discovers a delegation in need of breaking,
362 * it will return -EWOULDBLOCK and return a reference to the inode in
363 * delegated_inode. The caller should then break the delegation and
364 * retry. Because breaking a delegation may take a long time, the
365 * caller should drop the i_mutex before doing so.
366 *
367 * Alternatively, a caller may pass NULL for delegated_inode. This may
368 * be appropriate for callers that expect the underlying filesystem not
369 * to be NFS exported. Also, passing NULL is fine for callers holding
370 * the file open for write, as there can be no conflicting delegation in
371 * that case.
372 *
373 * If the inode has been found through an idmapped mount the idmap of
374 * the vfsmount must be passed through @idmap. This function will then
375 * take care to map the inode according to @idmap before checking
376 * permissions. On non-idmapped mounts or if permission checking is to be
377 * performed on the raw inode simply pass @nop_mnt_idmap.
378 */
379int notify_change(struct mnt_idmap *idmap, struct dentry *dentry,
380 struct iattr *attr, struct inode **delegated_inode)
381{
382 struct inode *inode = dentry->d_inode;
383 umode_t mode = inode->i_mode;
384 int error;
385 struct timespec64 now;
386 unsigned int ia_valid = attr->ia_valid;
387
388 WARN_ON_ONCE(!inode_is_locked(inode));
389
390 error = may_setattr(idmap, inode, ia_valid);
391 if (error)
392 return error;
393
394 if ((ia_valid & ATTR_MODE)) {
395 /*
396 * Don't allow changing the mode of symlinks:
397 *
398 * (1) The vfs doesn't take the mode of symlinks into account
399 * during permission checking.
400 * (2) This has never worked correctly. Most major filesystems
401 * did return EOPNOTSUPP due to interactions with POSIX ACLs
402 * but did still updated the mode of the symlink.
403 * This inconsistency led system call wrapper providers such
404 * as libc to block changing the mode of symlinks with
405 * EOPNOTSUPP already.
406 * (3) To even do this in the first place one would have to use
407 * specific file descriptors and quite some effort.
408 */
409 if (S_ISLNK(inode->i_mode))
410 return -EOPNOTSUPP;
411
412 /* Flag setting protected by i_mutex */
413 if (is_sxid(mode: attr->ia_mode))
414 inode->i_flags &= ~S_NOSEC;
415 }
416
417 now = current_time(inode);
418
419 attr->ia_ctime = now;
420 if (!(ia_valid & ATTR_ATIME_SET))
421 attr->ia_atime = now;
422 else
423 attr->ia_atime = timestamp_truncate(t: attr->ia_atime, inode);
424 if (!(ia_valid & ATTR_MTIME_SET))
425 attr->ia_mtime = now;
426 else
427 attr->ia_mtime = timestamp_truncate(t: attr->ia_mtime, inode);
428
429 if (ia_valid & ATTR_KILL_PRIV) {
430 error = security_inode_need_killpriv(dentry);
431 if (error < 0)
432 return error;
433 if (error == 0)
434 ia_valid = attr->ia_valid &= ~ATTR_KILL_PRIV;
435 }
436
437 /*
438 * We now pass ATTR_KILL_S*ID to the lower level setattr function so
439 * that the function has the ability to reinterpret a mode change
440 * that's due to these bits. This adds an implicit restriction that
441 * no function will ever call notify_change with both ATTR_MODE and
442 * ATTR_KILL_S*ID set.
443 */
444 if ((ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) &&
445 (ia_valid & ATTR_MODE))
446 BUG();
447
448 if (ia_valid & ATTR_KILL_SUID) {
449 if (mode & S_ISUID) {
450 ia_valid = attr->ia_valid |= ATTR_MODE;
451 attr->ia_mode = (inode->i_mode & ~S_ISUID);
452 }
453 }
454 if (ia_valid & ATTR_KILL_SGID) {
455 if (mode & S_ISGID) {
456 if (!(ia_valid & ATTR_MODE)) {
457 ia_valid = attr->ia_valid |= ATTR_MODE;
458 attr->ia_mode = inode->i_mode;
459 }
460 attr->ia_mode &= ~S_ISGID;
461 }
462 }
463 if (!(attr->ia_valid & ~(ATTR_KILL_SUID | ATTR_KILL_SGID)))
464 return 0;
465
466 /*
467 * Verify that uid/gid changes are valid in the target
468 * namespace of the superblock.
469 */
470 if (ia_valid & ATTR_UID &&
471 !vfsuid_has_fsmapping(idmap, fs_userns: inode->i_sb->s_user_ns,
472 vfsuid: attr->ia_vfsuid))
473 return -EOVERFLOW;
474 if (ia_valid & ATTR_GID &&
475 !vfsgid_has_fsmapping(idmap, fs_userns: inode->i_sb->s_user_ns,
476 vfsgid: attr->ia_vfsgid))
477 return -EOVERFLOW;
478
479 /* Don't allow modifications of files with invalid uids or
480 * gids unless those uids & gids are being made valid.
481 */
482 if (!(ia_valid & ATTR_UID) &&
483 !vfsuid_valid(uid: i_uid_into_vfsuid(idmap, inode)))
484 return -EOVERFLOW;
485 if (!(ia_valid & ATTR_GID) &&
486 !vfsgid_valid(gid: i_gid_into_vfsgid(idmap, inode)))
487 return -EOVERFLOW;
488
489 error = security_inode_setattr(idmap, dentry, attr);
490 if (error)
491 return error;
492 error = try_break_deleg(inode, delegated_inode);
493 if (error)
494 return error;
495
496 if (inode->i_op->setattr)
497 error = inode->i_op->setattr(idmap, dentry, attr);
498 else
499 error = simple_setattr(idmap, dentry, attr);
500
501 if (!error) {
502 fsnotify_change(dentry, ia_valid);
503 security_inode_post_setattr(idmap, dentry, ia_valid);
504 }
505
506 return error;
507}
508EXPORT_SYMBOL(notify_change);
509

source code of linux/fs/attr.c