1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 *
4 * Copyright (C) 2011 Novell Inc.
5 */
6
7 #include <linux/fs.h>
8 #include <linux/slab.h>
9 #include <linux/cred.h>
10 #include <linux/xattr.h>
11 #include <linux/ratelimit.h>
12 #include <linux/fiemap.h>
13 #include <linux/fileattr.h>
14 #include <linux/security.h>
15 #include <linux/namei.h>
16 #include <linux/posix_acl.h>
17 #include <linux/posix_acl_xattr.h>
18 #include "overlayfs.h"
19
20
21 int ovl_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
22 struct iattr *attr)
23 {
24 int err;
25 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
26 bool full_copy_up = false;
27 struct dentry *upperdentry;
28 const struct cred *old_cred;
29
30 err = setattr_prepare(&nop_mnt_idmap, dentry, attr);
31 if (err)
32 return err;
33
34 if (attr->ia_valid & ATTR_SIZE) {
35 /* Truncate should trigger data copy up as well */
36 full_copy_up = true;
37 }
38
39 if (!full_copy_up)
40 err = ovl_copy_up(dentry);
41 else
42 err = ovl_copy_up_with_data(dentry);
43 if (!err) {
44 struct inode *winode = NULL;
45
46 upperdentry = ovl_dentry_upper(dentry);
47
48 if (attr->ia_valid & ATTR_SIZE) {
49 winode = d_inode(upperdentry);
50 err = get_write_access(winode);
51 if (err)
52 goto out;
53 }
54
55 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
56 attr->ia_valid &= ~ATTR_MODE;
57
58 /*
59 * We might have to translate ovl file into real file object
60 * once use cases emerge. For now, simply don't let underlying
61 * filesystem rely on attr->ia_file
62 */
63 attr->ia_valid &= ~ATTR_FILE;
64
65 /*
66 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
67 * set. Overlayfs does not pass O_TRUNC flag to underlying
68 * filesystem during open -> do not pass ATTR_OPEN. This
69 * disables optimization in fuse which assumes open(O_TRUNC)
70 * already set file size to 0. But we never passed O_TRUNC to
71 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
72 * setattr request to server.
73 */
74 attr->ia_valid &= ~ATTR_OPEN;
75
76 err = ovl_want_write(dentry);
77 if (err)
78 goto out_put_write;
79
80 inode_lock(upperdentry->d_inode);
81 old_cred = ovl_override_creds(dentry->d_sb);
82 err = ovl_do_notify_change(ofs, upperdentry, attr);
83 revert_creds(old_cred);
84 if (!err)
85 ovl_copyattr(dentry->d_inode);
86 inode_unlock(upperdentry->d_inode);
87 ovl_drop_write(dentry);
88
89 out_put_write:
90 if (winode)
91 put_write_access(winode);
92 }
93 out:
94 return err;
95 }
96
97 static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
98 {
99 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
100 bool samefs = ovl_same_fs(ofs);
101 unsigned int xinobits = ovl_xino_bits(ofs);
102 unsigned int xinoshift = 64 - xinobits;
103
104 if (samefs) {
105 /*
106 * When all layers are on the same fs, all real inode
107 * number are unique, so we use the overlay st_dev,
108 * which is friendly to du -x.
109 */
110 stat->dev = dentry->d_sb->s_dev;
111 return;
112 } else if (xinobits) {
113 /*
114 * All inode numbers of underlying fs should not be using the
115 * high xinobits, so we use high xinobits to partition the
116 * overlay st_ino address space. The high bits holds the fsid
117 * (upper fsid is 0). The lowest xinobit is reserved for mapping
118 * the non-persistent inode numbers range in case of overflow.
119 * This way all overlay inode numbers are unique and use the
120 * overlay st_dev.
121 */
122 if (likely(!(stat->ino >> xinoshift))) {
123 stat->ino |= ((u64)fsid) << (xinoshift + 1);
124 stat->dev = dentry->d_sb->s_dev;
125 return;
126 } else if (ovl_xino_warn(ofs)) {
127 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
128 dentry, stat->ino, xinobits);
129 }
130 }
131
132 /* The inode could not be mapped to a unified st_ino address space */
133 if (S_ISDIR(dentry->d_inode->i_mode)) {
134 /*
135 * Always use the overlay st_dev for directories, so 'find
136 * -xdev' will scan the entire overlay mount and won't cross the
137 * overlay mount boundaries.
138 *
139 * If not all layers are on the same fs the pair {real st_ino;
140 * overlay st_dev} is not unique, so use the non persistent
141 * overlay st_ino for directories.
142 */
143 stat->dev = dentry->d_sb->s_dev;
144 stat->ino = dentry->d_inode->i_ino;
145 } else {
146 /*
147 * For non-samefs setup, if we cannot map all layers st_ino
148 * to a unified address space, we need to make sure that st_dev
149 * is unique per underlying fs, so we use the unique anonymous
150 * bdev assigned to the underlying fs.
151 */
152 stat->dev = ofs->fs[fsid].pseudo_dev;
153 }
154 }
155
156 int ovl_getattr(struct mnt_idmap *idmap, const struct path *path,
157 struct kstat *stat, u32 request_mask, unsigned int flags)
158 {
159 struct dentry *dentry = path->dentry;
160 enum ovl_path_type type;
161 struct path realpath;
162 const struct cred *old_cred;
163 struct inode *inode = d_inode(dentry);
164 bool is_dir = S_ISDIR(inode->i_mode);
165 int fsid = 0;
166 int err;
167 bool metacopy_blocks = false;
168
169 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
170
171 type = ovl_path_real(dentry, &realpath);
172 old_cred = ovl_override_creds(dentry->d_sb);
173 err = ovl_do_getattr(&realpath, stat, request_mask, flags);
174 if (err)
175 goto out;
176
177 /* Report the effective immutable/append-only STATX flags */
178 generic_fill_statx_attr(inode, stat);
179
180 /*
181 * For non-dir or same fs, we use st_ino of the copy up origin.
182 * This guaranties constant st_dev/st_ino across copy up.
183 * With xino feature and non-samefs, we use st_ino of the copy up
184 * origin masked with high bits that represent the layer id.
185 *
186 * If lower filesystem supports NFS file handles, this also guaranties
187 * persistent st_ino across mount cycle.
188 */
189 if (!is_dir || ovl_same_dev(OVL_FS(dentry->d_sb))) {
190 if (!OVL_TYPE_UPPER(type)) {
191 fsid = ovl_layer_lower(dentry)->fsid;
192 } else if (OVL_TYPE_ORIGIN(type)) {
193 struct kstat lowerstat;
194 u32 lowermask = STATX_INO | STATX_BLOCKS |
195 (!is_dir ? STATX_NLINK : 0);
196
197 ovl_path_lower(dentry, &realpath);
198 err = ovl_do_getattr(&realpath, &lowerstat, lowermask,
199 flags);
200 if (err)
201 goto out;
202
203 /*
204 * Lower hardlinks may be broken on copy up to different
205 * upper files, so we cannot use the lower origin st_ino
206 * for those different files, even for the same fs case.
207 *
208 * Similarly, several redirected dirs can point to the
209 * same dir on a lower layer. With the "verify_lower"
210 * feature, we do not use the lower origin st_ino, if
211 * we haven't verified that this redirect is unique.
212 *
213 * With inodes index enabled, it is safe to use st_ino
214 * of an indexed origin. The index validates that the
215 * upper hardlink is not broken and that a redirected
216 * dir is the only redirect to that origin.
217 */
218 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
219 (!ovl_verify_lower(dentry->d_sb) &&
220 (is_dir || lowerstat.nlink == 1))) {
221 fsid = ovl_layer_lower(dentry)->fsid;
222 stat->ino = lowerstat.ino;
223 }
224
225 /*
226 * If we are querying a metacopy dentry and lower
227 * dentry is data dentry, then use the blocks we
228 * queried just now. We don't have to do additional
229 * vfs_getattr(). If lower itself is metacopy, then
230 * additional vfs_getattr() is unavoidable.
231 */
232 if (metacopy_blocks &&
233 realpath.dentry == ovl_dentry_lowerdata(dentry)) {
234 stat->blocks = lowerstat.blocks;
235 metacopy_blocks = false;
236 }
237 }
238
239 if (metacopy_blocks) {
240 /*
241 * If lower is not same as lowerdata or if there was
242 * no origin on upper, we can end up here.
243 * With lazy lowerdata lookup, guess lowerdata blocks
244 * from size to avoid lowerdata lookup on stat(2).
245 */
246 struct kstat lowerdatastat;
247 u32 lowermask = STATX_BLOCKS;
248
249 ovl_path_lowerdata(dentry, &realpath);
250 if (realpath.dentry) {
251 err = ovl_do_getattr(&realpath, &lowerdatastat,
252 lowermask, flags);
253 if (err)
254 goto out;
255 } else {
256 lowerdatastat.blocks =
257 round_up(stat->size, stat->blksize) >> 9;
258 }
259 stat->blocks = lowerdatastat.blocks;
260 }
261 }
262
263 ovl_map_dev_ino(dentry, stat, fsid);
264
265 /*
266 * It's probably not worth it to count subdirs to get the
267 * correct link count. nlink=1 seems to pacify 'find' and
268 * other utilities.
269 */
270 if (is_dir && OVL_TYPE_MERGE(type))
271 stat->nlink = 1;
272
273 /*
274 * Return the overlay inode nlinks for indexed upper inodes.
275 * Overlay inode nlink counts the union of the upper hardlinks
276 * and non-covered lower hardlinks. It does not include the upper
277 * index hardlink.
278 */
279 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
280 stat->nlink = dentry->d_inode->i_nlink;
281
282 out:
283 revert_creds(old_cred);
284
285 return err;
286 }
287
288 int ovl_permission(struct mnt_idmap *idmap,
289 struct inode *inode, int mask)
290 {
291 struct inode *upperinode = ovl_inode_upper(inode);
292 struct inode *realinode;
293 struct path realpath;
294 const struct cred *old_cred;
295 int err;
296
297 /* Careful in RCU walk mode */
298 realinode = ovl_i_path_real(inode, &realpath);
299 if (!realinode) {
300 WARN_ON(!(mask & MAY_NOT_BLOCK));
301 return -ECHILD;
302 }
303
304 /*
305 * Check overlay inode with the creds of task and underlying inode
306 * with creds of mounter
307 */
308 err = generic_permission(&nop_mnt_idmap, inode, mask);
309 if (err)
310 return err;
311
312 old_cred = ovl_override_creds(inode->i_sb);
313 if (!upperinode &&
314 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
315 mask &= ~(MAY_WRITE | MAY_APPEND);
316 /* Make sure mounter can read file for copy up later */
317 mask |= MAY_READ;
318 }
319 err = inode_permission(mnt_idmap(realpath.mnt), realinode, mask);
320 revert_creds(old_cred);
321
322 return err;
323 }
324
325 static const char *ovl_get_link(struct dentry *dentry,
326 struct inode *inode,
327 struct delayed_call *done)
328 {
329 const struct cred *old_cred;
330 const char *p;
331
332 if (!dentry)
333 return ERR_PTR(-ECHILD);
334
335 old_cred = ovl_override_creds(dentry->d_sb);
336 p = vfs_get_link(ovl_dentry_real(dentry), done);
337 revert_creds(old_cred);
338 return p;
339 }
340
341 #ifdef CONFIG_FS_POSIX_ACL
342 /*
343 * Apply the idmapping of the layer to POSIX ACLs. The caller must pass a clone
344 * of the POSIX ACLs retrieved from the lower layer to this function to not
345 * alter the POSIX ACLs for the underlying filesystem.
346 */
347 static void ovl_idmap_posix_acl(const struct inode *realinode,
348 struct mnt_idmap *idmap,
349 struct posix_acl *acl)
350 {
351 struct user_namespace *fs_userns = i_user_ns(realinode);
352
353 for (unsigned int i = 0; i < acl->a_count; i++) {
354 vfsuid_t vfsuid;
355 vfsgid_t vfsgid;
356
357 struct posix_acl_entry *e = &acl->a_entries[i];
358 switch (e->e_tag) {
359 case ACL_USER:
360 vfsuid = make_vfsuid(idmap, fs_userns, e->e_uid);
361 e->e_uid = vfsuid_into_kuid(vfsuid);
362 break;
363 case ACL_GROUP:
364 vfsgid = make_vfsgid(idmap, fs_userns, e->e_gid);
365 e->e_gid = vfsgid_into_kgid(vfsgid);
366 break;
367 }
368 }
369 }
370
371 /*
372 * The @noperm argument is used to skip permission checking and is a temporary
373 * measure. Quoting Miklos from an earlier discussion:
374 *
375 * > So there are two paths to getting an acl:
376 * > 1) permission checking and 2) retrieving the value via getxattr(2).
377 * > This is a similar situation as reading a symlink vs. following it.
378 * > When following a symlink overlayfs always reads the link on the
379 * > underlying fs just as if it was a readlink(2) call, calling
380 * > security_inode_readlink() instead of security_inode_follow_link().
381 * > This is logical: we are reading the link from the underlying storage,
382 * > and following it on overlayfs.
383 * >
384 * > Applying the same logic to acl: we do need to call the
385 * > security_inode_getxattr() on the underlying fs, even if just want to
386 * > check permissions on overlay. This is currently not done, which is an
387 * > inconsistency.
388 * >
389 * > Maybe adding the check to ovl_get_acl() is the right way to go, but
390 * > I'm a little afraid of a performance regression. Will look into that.
391 *
392 * Until we have made a decision allow this helper to take the @noperm
393 * argument. We should hopefully be able to remove it soon.
394 */
395 struct posix_acl *ovl_get_acl_path(const struct path *path,
396 const char *acl_name, bool noperm)
397 {
398 struct posix_acl *real_acl, *clone;
399 struct mnt_idmap *idmap;
400 struct inode *realinode = d_inode(path->dentry);
401
402 idmap = mnt_idmap(path->mnt);
403
404 if (noperm)
405 real_acl = get_inode_acl(realinode, posix_acl_type(acl_name));
406 else
407 real_acl = vfs_get_acl(idmap, path->dentry, acl_name);
408 if (IS_ERR_OR_NULL(real_acl))
409 return real_acl;
410
411 if (!is_idmapped_mnt(path->mnt))
412 return real_acl;
413
414 /*
415 * We cannot alter the ACLs returned from the relevant layer as that
416 * would alter the cached values filesystem wide for the lower
417 * filesystem. Instead we can clone the ACLs and then apply the
418 * relevant idmapping of the layer.
419 */
420 clone = posix_acl_clone(real_acl, GFP_KERNEL);
421 posix_acl_release(real_acl); /* release original acl */
422 if (!clone)
423 return ERR_PTR(-ENOMEM);
424
425 ovl_idmap_posix_acl(realinode, idmap, clone);
426 return clone;
427 }
428
429 /*
430 * When the relevant layer is an idmapped mount we need to take the idmapping
431 * of the layer into account and translate any ACL_{GROUP,USER} values
432 * according to the idmapped mount.
433 *
434 * We cannot alter the ACLs returned from the relevant layer as that would
435 * alter the cached values filesystem wide for the lower filesystem. Instead we
436 * can clone the ACLs and then apply the relevant idmapping of the layer.
437 *
438 * This is obviously only relevant when idmapped layers are used.
439 */
440 struct posix_acl *do_ovl_get_acl(struct mnt_idmap *idmap,
441 struct inode *inode, int type,
442 bool rcu, bool noperm)
443 {
444 struct inode *realinode;
445 struct posix_acl *acl;
446 struct path realpath;
447
448 /* Careful in RCU walk mode */
449 realinode = ovl_i_path_real(inode, &realpath);
450 if (!realinode) {
451 WARN_ON(!rcu);
452 return ERR_PTR(-ECHILD);
453 }
454
455 if (!IS_POSIXACL(realinode))
456 return NULL;
457
458 if (rcu) {
459 /*
460 * If the layer is idmapped drop out of RCU path walk
461 * so we can clone the ACLs.
462 */
463 if (is_idmapped_mnt(realpath.mnt))
464 return ERR_PTR(-ECHILD);
465
466 acl = get_cached_acl_rcu(realinode, type);
467 } else {
468 const struct cred *old_cred;
469
470 old_cred = ovl_override_creds(inode->i_sb);
471 acl = ovl_get_acl_path(&realpath, posix_acl_xattr_name(type), noperm);
472 revert_creds(old_cred);
473 }
474
475 return acl;
476 }
477
478 static int ovl_set_or_remove_acl(struct dentry *dentry, struct inode *inode,
479 struct posix_acl *acl, int type)
480 {
481 int err;
482 struct path realpath;
483 const char *acl_name;
484 const struct cred *old_cred;
485 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
486 struct dentry *upperdentry = ovl_dentry_upper(dentry);
487 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
488
489 /*
490 * If ACL is to be removed from a lower file, check if it exists in
491 * the first place before copying it up.
492 */
493 acl_name = posix_acl_xattr_name(type);
494 if (!acl && !upperdentry) {
495 struct posix_acl *real_acl;
496
497 ovl_path_lower(dentry, &realpath);
498 old_cred = ovl_override_creds(dentry->d_sb);
499 real_acl = vfs_get_acl(mnt_idmap(realpath.mnt), realdentry,
500 acl_name);
501 revert_creds(old_cred);
502 if (IS_ERR(real_acl)) {
503 err = PTR_ERR(real_acl);
504 goto out;
505 }
506 posix_acl_release(real_acl);
507 }
508
509 if (!upperdentry) {
510 err = ovl_copy_up(dentry);
511 if (err)
512 goto out;
513
514 realdentry = ovl_dentry_upper(dentry);
515 }
516
517 err = ovl_want_write(dentry);
518 if (err)
519 goto out;
520
521 old_cred = ovl_override_creds(dentry->d_sb);
522 if (acl)
523 err = ovl_do_set_acl(ofs, realdentry, acl_name, acl);
524 else
525 err = ovl_do_remove_acl(ofs, realdentry, acl_name);
526 revert_creds(old_cred);
527 ovl_drop_write(dentry);
528
529 /* copy c/mtime */
530 ovl_copyattr(inode);
531 out:
532 return err;
533 }
534
535 int ovl_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
536 struct posix_acl *acl, int type)
537 {
538 int err;
539 struct inode *inode = d_inode(dentry);
540 struct dentry *workdir = ovl_workdir(dentry);
541 struct inode *realinode = ovl_inode_real(inode);
542
543 if (!IS_POSIXACL(d_inode(workdir)))
544 return -EOPNOTSUPP;
545 if (!realinode->i_op->set_acl)
546 return -EOPNOTSUPP;
547 if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode))
548 return acl ? -EACCES : 0;
549 if (!inode_owner_or_capable(&nop_mnt_idmap, inode))
550 return -EPERM;
551
552 /*
553 * Check if sgid bit needs to be cleared (actual setacl operation will
554 * be done with mounter's capabilities and so that won't do it for us).
555 */
556 if (unlikely(inode->i_mode & S_ISGID) && type == ACL_TYPE_ACCESS &&
557 !in_group_p(inode->i_gid) &&
558 !capable_wrt_inode_uidgid(&nop_mnt_idmap, inode, CAP_FSETID)) {
559 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
560
561 err = ovl_setattr(&nop_mnt_idmap, dentry, &iattr);
562 if (err)
563 return err;
564 }
565
566 return ovl_set_or_remove_acl(dentry, inode, acl, type);
567 }
568 #endif
569
570 int ovl_update_time(struct inode *inode, int flags)
571 {
572 if (flags & S_ATIME) {
573 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
574 struct path upperpath = {
575 .mnt = ovl_upper_mnt(ofs),
576 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
577 };
578
579 if (upperpath.dentry) {
580 touch_atime(&upperpath);
581 inode_set_atime_to_ts(inode,
582 inode_get_atime(d_inode(upperpath.dentry)));
583 }
584 }
585 return 0;
586 }
587
588 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
589 u64 start, u64 len)
590 {
591 int err;
592 struct inode *realinode = ovl_inode_realdata(inode);
593 const struct cred *old_cred;
594
595 if (!realinode)
596 return -EIO;
597
598 if (!realinode->i_op->fiemap)
599 return -EOPNOTSUPP;
600
601 old_cred = ovl_override_creds(inode->i_sb);
602 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
603 revert_creds(old_cred);
604
605 return err;
606 }
607
608 /*
609 * Work around the fact that security_file_ioctl() takes a file argument.
610 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
611 * properly.
612 */
613 static int ovl_security_fileattr(const struct path *realpath, struct fileattr *fa,
614 bool set)
615 {
616 struct file *file;
617 unsigned int cmd;
618 int err;
619
620 file = dentry_open(realpath, O_RDONLY, current_cred());
621 if (IS_ERR(file))
622 return PTR_ERR(file);
623
624 if (set)
625 cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
626 else
627 cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
628
629 err = security_file_ioctl(file, cmd, 0);
630 fput(file);
631
632 return err;
633 }
634
635 int ovl_real_fileattr_set(const struct path *realpath, struct fileattr *fa)
636 {
637 int err;
638
639 err = ovl_security_fileattr(realpath, fa, true);
640 if (err)
641 return err;
642
643 return vfs_fileattr_set(mnt_idmap(realpath->mnt), realpath->dentry, fa);
644 }
645
646 int ovl_fileattr_set(struct mnt_idmap *idmap,
647 struct dentry *dentry, struct fileattr *fa)
648 {
649 struct inode *inode = d_inode(dentry);
650 struct path upperpath;
651 const struct cred *old_cred;
652 unsigned int flags;
653 int err;
654
655 err = ovl_copy_up(dentry);
656 if (!err) {
657 ovl_path_real(dentry, &upperpath);
658
659 err = ovl_want_write(dentry);
660 if (err)
661 goto out;
662
663 old_cred = ovl_override_creds(inode->i_sb);
664 /*
665 * Store immutable/append-only flags in xattr and clear them
666 * in upper fileattr (in case they were set by older kernel)
667 * so children of "ovl-immutable" directories lower aliases of
668 * "ovl-immutable" hardlinks could be copied up.
669 * Clear xattr when flags are cleared.
670 */
671 err = ovl_set_protattr(inode, upperpath.dentry, fa);
672 if (!err)
673 err = ovl_real_fileattr_set(&upperpath, fa);
674 revert_creds(old_cred);
675 ovl_drop_write(dentry);
676
677 /*
678 * Merge real inode flags with inode flags read from
679 * overlay.protattr xattr
680 */
681 flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK;
682
683 BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK);
684 flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK;
685 inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK);
686
687 /* Update ctime */
688 ovl_copyattr(inode);
689 }
690 out:
691 return err;
692 }
693
694 /* Convert inode protection flags to fileattr flags */
695 static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa)
696 {
697 BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL);
698 BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
699
700 if (inode->i_flags & S_APPEND) {
701 fa->flags |= FS_APPEND_FL;
702 fa->fsx_xflags |= FS_XFLAG_APPEND;
703 }
704 if (inode->i_flags & S_IMMUTABLE) {
705 fa->flags |= FS_IMMUTABLE_FL;
706 fa->fsx_xflags |= FS_XFLAG_IMMUTABLE;
707 }
708 }
709
710 int ovl_real_fileattr_get(const struct path *realpath, struct fileattr *fa)
711 {
712 int err;
713
714 err = ovl_security_fileattr(realpath, fa, false);
715 if (err)
716 return err;
717
718 err = vfs_fileattr_get(realpath->dentry, fa);
719 if (err == -ENOIOCTLCMD)
720 err = -ENOTTY;
721 return err;
722 }
723
724 int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa)
725 {
726 struct inode *inode = d_inode(dentry);
727 struct path realpath;
728 const struct cred *old_cred;
729 int err;
730
731 ovl_path_real(dentry, &realpath);
732
733 old_cred = ovl_override_creds(inode->i_sb);
734 err = ovl_real_fileattr_get(&realpath, fa);
735 ovl_fileattr_prot_flags(inode, fa);
736 revert_creds(old_cred);
737
738 return err;
739 }
740
741 static const struct inode_operations ovl_file_inode_operations = {
742 .setattr = ovl_setattr,
743 .permission = ovl_permission,
744 .getattr = ovl_getattr,
745 .listxattr = ovl_listxattr,
746 .get_inode_acl = ovl_get_inode_acl,
747 .get_acl = ovl_get_acl,
748 .set_acl = ovl_set_acl,
749 .update_time = ovl_update_time,
750 .fiemap = ovl_fiemap,
751 .fileattr_get = ovl_fileattr_get,
752 .fileattr_set = ovl_fileattr_set,
753 };
754
755 static const struct inode_operations ovl_symlink_inode_operations = {
756 .setattr = ovl_setattr,
757 .get_link = ovl_get_link,
758 .getattr = ovl_getattr,
759 .listxattr = ovl_listxattr,
760 .update_time = ovl_update_time,
761 };
762
763 static const struct inode_operations ovl_special_inode_operations = {
764 .setattr = ovl_setattr,
765 .permission = ovl_permission,
766 .getattr = ovl_getattr,
767 .listxattr = ovl_listxattr,
768 .get_inode_acl = ovl_get_inode_acl,
769 .get_acl = ovl_get_acl,
770 .set_acl = ovl_set_acl,
771 .update_time = ovl_update_time,
772 };
773
774 static const struct address_space_operations ovl_aops = {
775 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
776 .direct_IO = noop_direct_IO,
777 };
778
779 /*
780 * It is possible to stack overlayfs instance on top of another
781 * overlayfs instance as lower layer. We need to annotate the
782 * stackable i_mutex locks according to stack level of the super
783 * block instance. An overlayfs instance can never be in stack
784 * depth 0 (there is always a real fs below it). An overlayfs
785 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
786 *
787 * For example, here is a snip from /proc/lockdep_chains after
788 * dir_iterate of nested overlayfs:
789 *
790 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
791 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
792 * [...] &type->i_mutex_dir_key (stack_depth=0)
793 *
794 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
795 *
796 * This chain is valid:
797 * - inode->i_rwsem (inode_lock[2])
798 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
799 * - OVL_I(inode)->lock (ovl_inode_lock[2])
800 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
801 *
802 * And this chain is valid:
803 * - inode->i_rwsem (inode_lock[2])
804 * - OVL_I(inode)->lock (ovl_inode_lock[2])
805 * - lowerinode->i_rwsem (inode_lock[1])
806 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
807 *
808 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
809 * held, because it is in reverse order of the non-nested case using the same
810 * upper fs:
811 * - inode->i_rwsem (inode_lock[1])
812 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
813 * - OVL_I(inode)->lock (ovl_inode_lock[1])
814 */
815 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
816
817 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
818 {
819 #ifdef CONFIG_LOCKDEP
820 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
821 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
822 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
823
824 int depth = inode->i_sb->s_stack_depth - 1;
825
826 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
827 depth = 0;
828
829 if (S_ISDIR(inode->i_mode))
830 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
831 else
832 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
833
834 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
835 #endif
836 }
837
838 static void ovl_next_ino(struct inode *inode)
839 {
840 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
841
842 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
843 if (unlikely(!inode->i_ino))
844 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
845 }
846
847 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
848 {
849 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
850 int xinobits = ovl_xino_bits(ofs);
851 unsigned int xinoshift = 64 - xinobits;
852
853 /*
854 * When d_ino is consistent with st_ino (samefs or i_ino has enough
855 * bits to encode layer), set the same value used for st_ino to i_ino,
856 * so inode number exposed via /proc/locks and a like will be
857 * consistent with d_ino and st_ino values. An i_ino value inconsistent
858 * with d_ino also causes nfsd readdirplus to fail.
859 */
860 inode->i_ino = ino;
861 if (ovl_same_fs(ofs)) {
862 return;
863 } else if (xinobits && likely(!(ino >> xinoshift))) {
864 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
865 return;
866 }
867
868 /*
869 * For directory inodes on non-samefs with xino disabled or xino
870 * overflow, we allocate a non-persistent inode number, to be used for
871 * resolving st_ino collisions in ovl_map_dev_ino().
872 *
873 * To avoid ino collision with legitimate xino values from upper
874 * layer (fsid 0), use the lowest xinobit to map the non
875 * persistent inode numbers to the unified st_ino address space.
876 */
877 if (S_ISDIR(inode->i_mode)) {
878 ovl_next_ino(inode);
879 if (xinobits) {
880 inode->i_ino &= ~0UL >> xinobits;
881 inode->i_ino |= 1UL << xinoshift;
882 }
883 }
884 }
885
886 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
887 unsigned long ino, int fsid)
888 {
889 struct inode *realinode;
890 struct ovl_inode *oi = OVL_I(inode);
891
892 oi->__upperdentry = oip->upperdentry;
893 oi->oe = oip->oe;
894 oi->redirect = oip->redirect;
895 oi->lowerdata_redirect = oip->lowerdata_redirect;
896
897 realinode = ovl_inode_real(inode);
898 ovl_copyattr(inode);
899 ovl_copyflags(realinode, inode);
900 ovl_map_ino(inode, ino, fsid);
901 }
902
903 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
904 {
905 inode->i_mode = mode;
906 inode->i_flags |= S_NOCMTIME;
907 #ifdef CONFIG_FS_POSIX_ACL
908 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
909 #endif
910
911 ovl_lockdep_annotate_inode_mutex_key(inode);
912
913 switch (mode & S_IFMT) {
914 case S_IFREG:
915 inode->i_op = &ovl_file_inode_operations;
916 inode->i_fop = &ovl_file_operations;
917 inode->i_mapping->a_ops = &ovl_aops;
918 break;
919
920 case S_IFDIR:
921 inode->i_op = &ovl_dir_inode_operations;
922 inode->i_fop = &ovl_dir_operations;
923 break;
924
925 case S_IFLNK:
926 inode->i_op = &ovl_symlink_inode_operations;
927 break;
928
929 default:
930 inode->i_op = &ovl_special_inode_operations;
931 init_special_inode(inode, mode, rdev);
932 break;
933 }
934 }
935
936 /*
937 * With inodes index enabled, an overlay inode nlink counts the union of upper
938 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
939 * upper inode, the following nlink modifying operations can happen:
940 *
941 * 1. Lower hardlink copy up
942 * 2. Upper hardlink created, unlinked or renamed over
943 * 3. Lower hardlink whiteout or renamed over
944 *
945 * For the first, copy up case, the union nlink does not change, whether the
946 * operation succeeds or fails, but the upper inode nlink may change.
947 * Therefore, before copy up, we store the union nlink value relative to the
948 * lower inode nlink in the index inode xattr .overlay.nlink.
949 *
950 * For the second, upper hardlink case, the union nlink should be incremented
951 * or decremented IFF the operation succeeds, aligned with nlink change of the
952 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
953 * value relative to the upper inode nlink in the index inode.
954 *
955 * For the last, lower cover up case, we simplify things by preceding the
956 * whiteout or cover up with copy up. This makes sure that there is an index
957 * upper inode where the nlink xattr can be stored before the copied up upper
958 * entry is unlink.
959 */
960 #define OVL_NLINK_ADD_UPPER (1 << 0)
961
962 /*
963 * On-disk format for indexed nlink:
964 *
965 * nlink relative to the upper inode - "U[+-]NUM"
966 * nlink relative to the lower inode - "L[+-]NUM"
967 */
968
969 static int ovl_set_nlink_common(struct dentry *dentry,
970 struct dentry *realdentry, const char *format)
971 {
972 struct inode *inode = d_inode(dentry);
973 struct inode *realinode = d_inode(realdentry);
974 char buf[13];
975 int len;
976
977 len = snprintf(buf, sizeof(buf), format,
978 (int) (inode->i_nlink - realinode->i_nlink));
979
980 if (WARN_ON(len >= sizeof(buf)))
981 return -EIO;
982
983 return ovl_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
984 OVL_XATTR_NLINK, buf, len);
985 }
986
987 int ovl_set_nlink_upper(struct dentry *dentry)
988 {
989 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
990 }
991
992 int ovl_set_nlink_lower(struct dentry *dentry)
993 {
994 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
995 }
996
997 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
998 struct dentry *upperdentry,
999 unsigned int fallback)
1000 {
1001 int nlink_diff;
1002 int nlink;
1003 char buf[13];
1004 int err;
1005
1006 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
1007 return fallback;
1008
1009 err = ovl_getxattr_upper(ofs, upperdentry, OVL_XATTR_NLINK,
1010 &buf, sizeof(buf) - 1);
1011 if (err < 0)
1012 goto fail;
1013
1014 buf[err] = '\0';
1015 if ((buf[0] != 'L' && buf[0] != 'U') ||
1016 (buf[1] != '+' && buf[1] != '-'))
1017 goto fail;
1018
1019 err = kstrtoint(buf + 1, 10, &nlink_diff);
1020 if (err < 0)
1021 goto fail;
1022
1023 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
1024 nlink += nlink_diff;
1025
1026 if (nlink <= 0)
1027 goto fail;
1028
1029 return nlink;
1030
1031 fail:
1032 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
1033 upperdentry, err);
1034 return fallback;
1035 }
1036
1037 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
1038 {
1039 struct inode *inode;
1040
1041 inode = new_inode(sb);
1042 if (inode)
1043 ovl_fill_inode(inode, mode, rdev);
1044
1045 return inode;
1046 }
1047
1048 static int ovl_inode_test(struct inode *inode, void *data)
1049 {
1050 return inode->i_private == data;
1051 }
1052
1053 static int ovl_inode_set(struct inode *inode, void *data)
1054 {
1055 inode->i_private = data;
1056 return 0;
1057 }
1058
1059 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
1060 struct dentry *upperdentry, bool strict)
1061 {
1062 /*
1063 * For directories, @strict verify from lookup path performs consistency
1064 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
1065 * inode. Non @strict verify from NFS handle decode path passes NULL for
1066 * 'unknown' lower/upper.
1067 */
1068 if (S_ISDIR(inode->i_mode) && strict) {
1069 /* Real lower dir moved to upper layer under us? */
1070 if (!lowerdentry && ovl_inode_lower(inode))
1071 return false;
1072
1073 /* Lookup of an uncovered redirect origin? */
1074 if (!upperdentry && ovl_inode_upper(inode))
1075 return false;
1076 }
1077
1078 /*
1079 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
1080 * This happens when finding a copied up overlay inode for a renamed
1081 * or hardlinked overlay dentry and lower dentry cannot be followed
1082 * by origin because lower fs does not support file handles.
1083 */
1084 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
1085 return false;
1086
1087 /*
1088 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
1089 * This happens when finding a lower alias for a copied up hard link.
1090 */
1091 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
1092 return false;
1093
1094 return true;
1095 }
1096
1097 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
1098 bool is_upper)
1099 {
1100 struct inode *inode, *key = d_inode(real);
1101
1102 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1103 if (!inode)
1104 return NULL;
1105
1106 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
1107 is_upper ? real : NULL, false)) {
1108 iput(inode);
1109 return ERR_PTR(-ESTALE);
1110 }
1111
1112 return inode;
1113 }
1114
1115 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
1116 {
1117 struct inode *key = d_inode(dir);
1118 struct inode *trap;
1119 bool res;
1120
1121 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1122 if (!trap)
1123 return false;
1124
1125 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
1126 !ovl_inode_lower(trap);
1127
1128 iput(trap);
1129 return res;
1130 }
1131
1132 /*
1133 * Create an inode cache entry for layer root dir, that will intentionally
1134 * fail ovl_verify_inode(), so any lookup that will find some layer root
1135 * will fail.
1136 */
1137 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
1138 {
1139 struct inode *key = d_inode(dir);
1140 struct inode *trap;
1141
1142 if (!d_is_dir(dir))
1143 return ERR_PTR(-ENOTDIR);
1144
1145 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
1146 ovl_inode_set, key);
1147 if (!trap)
1148 return ERR_PTR(-ENOMEM);
1149
1150 if (!(trap->i_state & I_NEW)) {
1151 /* Conflicting layer roots? */
1152 iput(trap);
1153 return ERR_PTR(-ELOOP);
1154 }
1155
1156 trap->i_mode = S_IFDIR;
1157 trap->i_flags = S_DEAD;
1158 unlock_new_inode(trap);
1159
1160 return trap;
1161 }
1162
1163 /*
1164 * Does overlay inode need to be hashed by lower inode?
1165 */
1166 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
1167 struct dentry *lower, bool index)
1168 {
1169 struct ovl_fs *ofs = OVL_FS(sb);
1170
1171 /* No, if pure upper */
1172 if (!lower)
1173 return false;
1174
1175 /* Yes, if already indexed */
1176 if (index)
1177 return true;
1178
1179 /* Yes, if won't be copied up */
1180 if (!ovl_upper_mnt(ofs))
1181 return true;
1182
1183 /* No, if lower hardlink is or will be broken on copy up */
1184 if ((upper || !ovl_indexdir(sb)) &&
1185 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1186 return false;
1187
1188 /* No, if non-indexed upper with NFS export */
1189 if (ofs->config.nfs_export && upper)
1190 return false;
1191
1192 /* Otherwise, hash by lower inode for fsnotify */
1193 return true;
1194 }
1195
1196 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1197 struct inode *key)
1198 {
1199 return newinode ? inode_insert5(newinode, (unsigned long) key,
1200 ovl_inode_test, ovl_inode_set, key) :
1201 iget5_locked(sb, (unsigned long) key,
1202 ovl_inode_test, ovl_inode_set, key);
1203 }
1204
1205 struct inode *ovl_get_inode(struct super_block *sb,
1206 struct ovl_inode_params *oip)
1207 {
1208 struct ovl_fs *ofs = OVL_FS(sb);
1209 struct dentry *upperdentry = oip->upperdentry;
1210 struct ovl_path *lowerpath = ovl_lowerpath(oip->oe);
1211 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1212 struct inode *inode;
1213 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1214 struct path realpath = {
1215 .dentry = upperdentry ?: lowerdentry,
1216 .mnt = upperdentry ? ovl_upper_mnt(ofs) : lowerpath->layer->mnt,
1217 };
1218 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1219 oip->index);
1220 int fsid = bylower ? lowerpath->layer->fsid : 0;
1221 bool is_dir;
1222 unsigned long ino = 0;
1223 int err = oip->newinode ? -EEXIST : -ENOMEM;
1224
1225 if (!realinode)
1226 realinode = d_inode(lowerdentry);
1227
1228 /*
1229 * Copy up origin (lower) may exist for non-indexed upper, but we must
1230 * not use lower as hash key if this is a broken hardlink.
1231 */
1232 is_dir = S_ISDIR(realinode->i_mode);
1233 if (upperdentry || bylower) {
1234 struct inode *key = d_inode(bylower ? lowerdentry :
1235 upperdentry);
1236 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1237
1238 inode = ovl_iget5(sb, oip->newinode, key);
1239 if (!inode)
1240 goto out_err;
1241 if (!(inode->i_state & I_NEW)) {
1242 /*
1243 * Verify that the underlying files stored in the inode
1244 * match those in the dentry.
1245 */
1246 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1247 true)) {
1248 iput(inode);
1249 err = -ESTALE;
1250 goto out_err;
1251 }
1252
1253 dput(upperdentry);
1254 ovl_free_entry(oip->oe);
1255 kfree(oip->redirect);
1256 kfree(oip->lowerdata_redirect);
1257 goto out;
1258 }
1259
1260 /* Recalculate nlink for non-dir due to indexing */
1261 if (!is_dir)
1262 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1263 nlink);
1264 set_nlink(inode, nlink);
1265 ino = key->i_ino;
1266 } else {
1267 /* Lower hardlink that will be broken on copy up */
1268 inode = new_inode(sb);
1269 if (!inode) {
1270 err = -ENOMEM;
1271 goto out_err;
1272 }
1273 ino = realinode->i_ino;
1274 fsid = lowerpath->layer->fsid;
1275 }
1276 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1277 ovl_inode_init(inode, oip, ino, fsid);
1278
1279 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1280 ovl_set_flag(OVL_IMPURE, inode);
1281
1282 if (oip->index)
1283 ovl_set_flag(OVL_INDEX, inode);
1284
1285 if (bylower)
1286 ovl_set_flag(OVL_CONST_INO, inode);
1287
1288 /* Check for non-merge dir that may have whiteouts */
1289 if (is_dir) {
1290 if (((upperdentry && lowerdentry) || ovl_numlower(oip->oe) > 1) ||
1291 ovl_path_check_origin_xattr(ofs, &realpath)) {
1292 ovl_set_flag(OVL_WHITEOUTS, inode);
1293 }
1294 }
1295
1296 /* Check for immutable/append-only inode flags in xattr */
1297 if (upperdentry)
1298 ovl_check_protattr(inode, upperdentry);
1299
1300 if (inode->i_state & I_NEW)
1301 unlock_new_inode(inode);
1302 out:
1303 return inode;
1304
1305 out_err:
1306 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1307 inode = ERR_PTR(err);
1308 goto out;
1309 }
1310
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.