1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2011 Novell Inc.
4 * Copyright (C) 2016 Red Hat, Inc.
8 #include <linux/mount.h>
9 #include <linux/slab.h>
10 #include <linux/cred.h>
11 #include <linux/xattr.h>
12 #include <linux/exportfs.h>
13 #include <linux/fileattr.h>
14 #include <linux/uuid.h>
15 #include <linux/namei.h>
16 #include <linux/ratelimit.h>
17 #include "overlayfs.h"
19 int ovl_want_write(struct dentry *dentry)
21 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
22 return mnt_want_write(ovl_upper_mnt(ofs));
25 void ovl_drop_write(struct dentry *dentry)
27 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
28 mnt_drop_write(ovl_upper_mnt(ofs));
31 struct dentry *ovl_workdir(struct dentry *dentry)
33 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
37 const struct cred *ovl_override_creds(struct super_block *sb)
39 struct ovl_fs *ofs = sb->s_fs_info;
41 return override_creds(ofs->creator_cred);
45 * Check if underlying fs supports file handles and try to determine encoding
46 * type, in order to deduce maximum inode number used by fs.
48 * Return 0 if file handles are not supported.
49 * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding.
50 * Return -1 if fs uses a non default encoding with unknown inode size.
52 int ovl_can_decode_fh(struct super_block *sb)
54 if (!capable(CAP_DAC_READ_SEARCH))
57 if (!sb->s_export_op || !sb->s_export_op->fh_to_dentry)
60 return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
63 struct dentry *ovl_indexdir(struct super_block *sb)
65 struct ovl_fs *ofs = sb->s_fs_info;
70 /* Index all files on copy up. For now only enabled for NFS export */
71 bool ovl_index_all(struct super_block *sb)
73 struct ovl_fs *ofs = sb->s_fs_info;
75 return ofs->config.nfs_export && ofs->config.index;
78 /* Verify lower origin on lookup. For now only enabled for NFS export */
79 bool ovl_verify_lower(struct super_block *sb)
81 struct ovl_fs *ofs = sb->s_fs_info;
83 return ofs->config.nfs_export && ofs->config.index;
86 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
88 size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
89 struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
92 oe->numlower = numlower;
97 #define OVL_D_REVALIDATE (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE)
99 bool ovl_dentry_remote(struct dentry *dentry)
101 return dentry->d_flags & OVL_D_REVALIDATE;
104 void ovl_dentry_update_reval(struct dentry *dentry, struct dentry *realdentry)
106 if (!ovl_dentry_remote(realdentry))
109 spin_lock(&dentry->d_lock);
110 dentry->d_flags |= realdentry->d_flags & OVL_D_REVALIDATE;
111 spin_unlock(&dentry->d_lock);
114 void ovl_dentry_init_reval(struct dentry *dentry, struct dentry *upperdentry)
116 return ovl_dentry_init_flags(dentry, upperdentry, OVL_D_REVALIDATE);
119 void ovl_dentry_init_flags(struct dentry *dentry, struct dentry *upperdentry,
122 struct ovl_entry *oe = OVL_E(dentry);
123 unsigned int i, flags = 0;
126 flags |= upperdentry->d_flags;
127 for (i = 0; i < oe->numlower; i++)
128 flags |= oe->lowerstack[i].dentry->d_flags;
130 spin_lock(&dentry->d_lock);
131 dentry->d_flags &= ~mask;
132 dentry->d_flags |= flags & mask;
133 spin_unlock(&dentry->d_lock);
136 bool ovl_dentry_weird(struct dentry *dentry)
138 return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
139 DCACHE_MANAGE_TRANSIT |
144 enum ovl_path_type ovl_path_type(struct dentry *dentry)
146 struct ovl_entry *oe = dentry->d_fsdata;
147 enum ovl_path_type type = 0;
149 if (ovl_dentry_upper(dentry)) {
150 type = __OVL_PATH_UPPER;
153 * Non-dir dentry can hold lower dentry of its copy up origin.
156 if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry)))
157 type |= __OVL_PATH_ORIGIN;
158 if (d_is_dir(dentry) ||
159 !ovl_has_upperdata(d_inode(dentry)))
160 type |= __OVL_PATH_MERGE;
163 if (oe->numlower > 1)
164 type |= __OVL_PATH_MERGE;
169 void ovl_path_upper(struct dentry *dentry, struct path *path)
171 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
173 path->mnt = ovl_upper_mnt(ofs);
174 path->dentry = ovl_dentry_upper(dentry);
177 void ovl_path_lower(struct dentry *dentry, struct path *path)
179 struct ovl_entry *oe = dentry->d_fsdata;
182 path->mnt = oe->lowerstack[0].layer->mnt;
183 path->dentry = oe->lowerstack[0].dentry;
185 *path = (struct path) { };
189 void ovl_path_lowerdata(struct dentry *dentry, struct path *path)
191 struct ovl_entry *oe = dentry->d_fsdata;
194 path->mnt = oe->lowerstack[oe->numlower - 1].layer->mnt;
195 path->dentry = oe->lowerstack[oe->numlower - 1].dentry;
197 *path = (struct path) { };
201 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
203 enum ovl_path_type type = ovl_path_type(dentry);
205 if (!OVL_TYPE_UPPER(type))
206 ovl_path_lower(dentry, path);
208 ovl_path_upper(dentry, path);
213 enum ovl_path_type ovl_path_realdata(struct dentry *dentry, struct path *path)
215 enum ovl_path_type type = ovl_path_type(dentry);
217 WARN_ON_ONCE(d_is_dir(dentry));
219 if (!OVL_TYPE_UPPER(type) || OVL_TYPE_MERGE(type))
220 ovl_path_lowerdata(dentry, path);
222 ovl_path_upper(dentry, path);
227 struct dentry *ovl_dentry_upper(struct dentry *dentry)
229 return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
232 struct dentry *ovl_dentry_lower(struct dentry *dentry)
234 struct ovl_entry *oe = dentry->d_fsdata;
236 return oe->numlower ? oe->lowerstack[0].dentry : NULL;
239 const struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
241 struct ovl_entry *oe = dentry->d_fsdata;
243 return oe->numlower ? oe->lowerstack[0].layer : NULL;
247 * ovl_dentry_lower() could return either a data dentry or metacopy dentry
248 * depending on what is stored in lowerstack[0]. At times we need to find
249 * lower dentry which has data (and not metacopy dentry). This helper
250 * returns the lower data dentry.
252 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry)
254 struct ovl_entry *oe = dentry->d_fsdata;
256 return oe->numlower ? oe->lowerstack[oe->numlower - 1].dentry : NULL;
259 struct dentry *ovl_dentry_real(struct dentry *dentry)
261 return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
264 struct dentry *ovl_i_dentry_upper(struct inode *inode)
266 return ovl_upperdentry_dereference(OVL_I(inode));
269 void ovl_i_path_real(struct inode *inode, struct path *path)
271 path->dentry = ovl_i_dentry_upper(inode);
273 path->dentry = OVL_I(inode)->lowerpath.dentry;
274 path->mnt = OVL_I(inode)->lowerpath.layer->mnt;
276 path->mnt = ovl_upper_mnt(OVL_FS(inode->i_sb));
280 struct inode *ovl_inode_upper(struct inode *inode)
282 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
284 return upperdentry ? d_inode(upperdentry) : NULL;
287 struct inode *ovl_inode_lower(struct inode *inode)
289 struct dentry *lowerdentry = OVL_I(inode)->lowerpath.dentry;
291 return lowerdentry ? d_inode(lowerdentry) : NULL;
294 struct inode *ovl_inode_real(struct inode *inode)
296 return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
299 /* Return inode which contains lower data. Do not return metacopy */
300 struct inode *ovl_inode_lowerdata(struct inode *inode)
302 if (WARN_ON(!S_ISREG(inode->i_mode)))
305 return OVL_I(inode)->lowerdata ?: ovl_inode_lower(inode);
308 /* Return real inode which contains data. Does not return metacopy inode */
309 struct inode *ovl_inode_realdata(struct inode *inode)
311 struct inode *upperinode;
313 upperinode = ovl_inode_upper(inode);
314 if (upperinode && ovl_has_upperdata(inode))
317 return ovl_inode_lowerdata(inode);
320 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
322 return OVL_I(inode)->cache;
325 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
327 OVL_I(inode)->cache = cache;
330 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
332 set_bit(flag, &OVL_E(dentry)->flags);
335 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
337 clear_bit(flag, &OVL_E(dentry)->flags);
340 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
342 return test_bit(flag, &OVL_E(dentry)->flags);
345 bool ovl_dentry_is_opaque(struct dentry *dentry)
347 return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
350 bool ovl_dentry_is_whiteout(struct dentry *dentry)
352 return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
355 void ovl_dentry_set_opaque(struct dentry *dentry)
357 ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
361 * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
362 * to return positive, while there's no actual upper alias for the inode.
363 * Copy up code needs to know about the existence of the upper alias, so it
364 * can't use ovl_dentry_upper().
366 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
368 return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
371 void ovl_dentry_set_upper_alias(struct dentry *dentry)
373 ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
376 static bool ovl_should_check_upperdata(struct inode *inode)
378 if (!S_ISREG(inode->i_mode))
381 if (!ovl_inode_lower(inode))
387 bool ovl_has_upperdata(struct inode *inode)
389 if (!ovl_should_check_upperdata(inode))
392 if (!ovl_test_flag(OVL_UPPERDATA, inode))
395 * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of
396 * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure
397 * if setting of OVL_UPPERDATA is visible, then effects of writes
398 * before that are visible too.
404 void ovl_set_upperdata(struct inode *inode)
407 * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure
408 * if OVL_UPPERDATA flag is visible, then effects of write operations
409 * before it are visible as well.
412 ovl_set_flag(OVL_UPPERDATA, inode);
415 /* Caller should hold ovl_inode->lock */
416 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags)
418 if (!ovl_open_flags_need_copy_up(flags))
421 return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry));
424 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags)
426 if (!ovl_open_flags_need_copy_up(flags))
429 return !ovl_has_upperdata(d_inode(dentry));
432 bool ovl_redirect_dir(struct super_block *sb)
434 struct ovl_fs *ofs = sb->s_fs_info;
436 return ofs->config.redirect_dir && !ofs->noxattr;
439 const char *ovl_dentry_get_redirect(struct dentry *dentry)
441 return OVL_I(d_inode(dentry))->redirect;
444 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
446 struct ovl_inode *oi = OVL_I(d_inode(dentry));
449 oi->redirect = redirect;
452 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
454 struct inode *upperinode = d_inode(upperdentry);
456 WARN_ON(OVL_I(inode)->__upperdentry);
459 * Make sure upperdentry is consistent before making it visible
462 OVL_I(inode)->__upperdentry = upperdentry;
463 if (inode_unhashed(inode)) {
464 inode->i_private = upperinode;
465 __insert_inode_hash(inode, (unsigned long) upperinode);
469 static void ovl_dir_version_inc(struct dentry *dentry, bool impurity)
471 struct inode *inode = d_inode(dentry);
473 WARN_ON(!inode_is_locked(inode));
474 WARN_ON(!d_is_dir(dentry));
476 * Version is used by readdir code to keep cache consistent.
477 * For merge dirs (or dirs with origin) all changes need to be noted.
478 * For non-merge dirs, cache contains only impure entries (i.e. ones
479 * which have been copied up and have origins), so only need to note
480 * changes to impure entries.
482 if (!ovl_dir_is_real(dentry) || impurity)
483 OVL_I(inode)->version++;
486 void ovl_dir_modified(struct dentry *dentry, bool impurity)
488 /* Copy mtime/ctime */
489 ovl_copyattr(d_inode(dentry));
491 ovl_dir_version_inc(dentry, impurity);
494 u64 ovl_dentry_version_get(struct dentry *dentry)
496 struct inode *inode = d_inode(dentry);
498 WARN_ON(!inode_is_locked(inode));
499 return OVL_I(inode)->version;
502 bool ovl_is_whiteout(struct dentry *dentry)
504 struct inode *inode = dentry->d_inode;
506 return inode && IS_WHITEOUT(inode);
509 struct file *ovl_path_open(const struct path *path, int flags)
511 struct inode *inode = d_inode(path->dentry);
512 struct user_namespace *real_mnt_userns = mnt_user_ns(path->mnt);
515 if (flags & ~(O_ACCMODE | O_LARGEFILE))
518 switch (flags & O_ACCMODE) {
523 acc_mode = MAY_WRITE;
529 err = inode_permission(real_mnt_userns, inode, acc_mode | MAY_OPEN);
533 /* O_NOATIME is an optimization, don't fail if not permitted */
534 if (inode_owner_or_capable(real_mnt_userns, inode))
537 return dentry_open(path, flags, current_cred());
540 /* Caller should hold ovl_inode->lock */
541 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags)
543 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
545 if (ovl_dentry_upper(dentry) &&
546 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
547 !ovl_dentry_needs_data_copy_up_locked(dentry, flags))
553 bool ovl_already_copied_up(struct dentry *dentry, int flags)
555 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
558 * Check if copy-up has happened as well as for upper alias (in
559 * case of hard links) is there.
561 * Both checks are lockless:
562 * - false negatives: will recheck under oi->lock
564 * + ovl_dentry_upper() uses memory barriers to ensure the
565 * upper dentry is up-to-date
566 * + ovl_dentry_has_upper_alias() relies on locking of
567 * upper parent i_rwsem to prevent reordering copy-up
570 if (ovl_dentry_upper(dentry) &&
571 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
572 !ovl_dentry_needs_data_copy_up(dentry, flags))
578 int ovl_copy_up_start(struct dentry *dentry, int flags)
580 struct inode *inode = d_inode(dentry);
583 err = ovl_inode_lock_interruptible(inode);
584 if (!err && ovl_already_copied_up_locked(dentry, flags)) {
585 err = 1; /* Already copied up */
586 ovl_inode_unlock(inode);
592 void ovl_copy_up_end(struct dentry *dentry)
594 ovl_inode_unlock(d_inode(dentry));
597 bool ovl_path_check_origin_xattr(struct ovl_fs *ofs, const struct path *path)
601 res = ovl_path_getxattr(ofs, path, OVL_XATTR_ORIGIN, NULL, 0);
603 /* Zero size value means "copied up but origin unknown" */
610 bool ovl_path_check_dir_xattr(struct ovl_fs *ofs, const struct path *path,
616 if (!d_is_dir(path->dentry))
619 res = ovl_path_getxattr(ofs, path, ox, &val, 1);
620 if (res == 1 && val == 'y')
626 #define OVL_XATTR_OPAQUE_POSTFIX "opaque"
627 #define OVL_XATTR_REDIRECT_POSTFIX "redirect"
628 #define OVL_XATTR_ORIGIN_POSTFIX "origin"
629 #define OVL_XATTR_IMPURE_POSTFIX "impure"
630 #define OVL_XATTR_NLINK_POSTFIX "nlink"
631 #define OVL_XATTR_UPPER_POSTFIX "upper"
632 #define OVL_XATTR_METACOPY_POSTFIX "metacopy"
633 #define OVL_XATTR_PROTATTR_POSTFIX "protattr"
635 #define OVL_XATTR_TAB_ENTRY(x) \
636 [x] = { [false] = OVL_XATTR_TRUSTED_PREFIX x ## _POSTFIX, \
637 [true] = OVL_XATTR_USER_PREFIX x ## _POSTFIX }
639 const char *const ovl_xattr_table[][2] = {
640 OVL_XATTR_TAB_ENTRY(OVL_XATTR_OPAQUE),
641 OVL_XATTR_TAB_ENTRY(OVL_XATTR_REDIRECT),
642 OVL_XATTR_TAB_ENTRY(OVL_XATTR_ORIGIN),
643 OVL_XATTR_TAB_ENTRY(OVL_XATTR_IMPURE),
644 OVL_XATTR_TAB_ENTRY(OVL_XATTR_NLINK),
645 OVL_XATTR_TAB_ENTRY(OVL_XATTR_UPPER),
646 OVL_XATTR_TAB_ENTRY(OVL_XATTR_METACOPY),
647 OVL_XATTR_TAB_ENTRY(OVL_XATTR_PROTATTR),
650 int ovl_check_setxattr(struct ovl_fs *ofs, struct dentry *upperdentry,
651 enum ovl_xattr ox, const void *value, size_t size,
659 err = ovl_setxattr(ofs, upperdentry, ox, value, size);
661 if (err == -EOPNOTSUPP) {
662 pr_warn("cannot set %s xattr on upper\n", ovl_xattr(ofs, ox));
670 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
672 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
675 if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
679 * Do not fail when upper doesn't support xattrs.
680 * Upper inodes won't have origin nor redirect xattr anyway.
682 err = ovl_check_setxattr(ofs, upperdentry, OVL_XATTR_IMPURE, "y", 1, 0);
684 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
690 #define OVL_PROTATTR_MAX 32 /* Reserved for future flags */
692 void ovl_check_protattr(struct inode *inode, struct dentry *upper)
694 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
695 u32 iflags = inode->i_flags & OVL_PROT_I_FLAGS_MASK;
696 char buf[OVL_PROTATTR_MAX+1];
699 res = ovl_getxattr_upper(ofs, upper, OVL_XATTR_PROTATTR, buf,
705 * Initialize inode flags from overlay.protattr xattr and upper inode
706 * flags. If upper inode has those fileattr flags set (i.e. from old
707 * kernel), we do not clear them on ovl_get_inode(), but we will clear
708 * them on next fileattr_set().
710 for (n = 0; n < res; n++) {
713 else if (buf[n] == 'i')
714 iflags |= S_IMMUTABLE;
719 if (!res || n < res) {
720 pr_warn_ratelimited("incompatible overlay.protattr format (%pd2, len=%d)\n",
723 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
727 int ovl_set_protattr(struct inode *inode, struct dentry *upper,
730 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
731 char buf[OVL_PROTATTR_MAX];
732 int len = 0, err = 0;
735 BUILD_BUG_ON(HWEIGHT32(OVL_PROT_FS_FLAGS_MASK) > OVL_PROTATTR_MAX);
737 if (fa->flags & FS_APPEND_FL) {
741 if (fa->flags & FS_IMMUTABLE_FL) {
743 iflags |= S_IMMUTABLE;
747 * Do not allow to set protection flags when upper doesn't support
748 * xattrs, because we do not set those fileattr flags on upper inode.
749 * Remove xattr if it exist and all protection flags are cleared.
752 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_PROTATTR,
754 } else if (inode->i_flags & OVL_PROT_I_FLAGS_MASK) {
755 err = ovl_removexattr(ofs, upper, OVL_XATTR_PROTATTR);
756 if (err == -EOPNOTSUPP || err == -ENODATA)
762 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
764 /* Mask out the fileattr flags that should not be set in upper inode */
765 fa->flags &= ~OVL_PROT_FS_FLAGS_MASK;
766 fa->fsx_xflags &= ~OVL_PROT_FSX_FLAGS_MASK;
772 * Caller must hold a reference to inode to prevent it from being freed while
773 * it is marked inuse.
775 bool ovl_inuse_trylock(struct dentry *dentry)
777 struct inode *inode = d_inode(dentry);
780 spin_lock(&inode->i_lock);
781 if (!(inode->i_state & I_OVL_INUSE)) {
782 inode->i_state |= I_OVL_INUSE;
785 spin_unlock(&inode->i_lock);
790 void ovl_inuse_unlock(struct dentry *dentry)
793 struct inode *inode = d_inode(dentry);
795 spin_lock(&inode->i_lock);
796 WARN_ON(!(inode->i_state & I_OVL_INUSE));
797 inode->i_state &= ~I_OVL_INUSE;
798 spin_unlock(&inode->i_lock);
802 bool ovl_is_inuse(struct dentry *dentry)
804 struct inode *inode = d_inode(dentry);
807 spin_lock(&inode->i_lock);
808 inuse = (inode->i_state & I_OVL_INUSE);
809 spin_unlock(&inode->i_lock);
815 * Does this overlay dentry need to be indexed on copy up?
817 bool ovl_need_index(struct dentry *dentry)
819 struct dentry *lower = ovl_dentry_lower(dentry);
821 if (!lower || !ovl_indexdir(dentry->d_sb))
824 /* Index all files for NFS export and consistency verification */
825 if (ovl_index_all(dentry->d_sb))
828 /* Index only lower hardlinks on copy up */
829 if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
835 /* Caller must hold OVL_I(inode)->lock */
836 static void ovl_cleanup_index(struct dentry *dentry)
838 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
839 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
840 struct inode *dir = indexdir->d_inode;
841 struct dentry *lowerdentry = ovl_dentry_lower(dentry);
842 struct dentry *upperdentry = ovl_dentry_upper(dentry);
843 struct dentry *index = NULL;
845 struct qstr name = { };
848 err = ovl_get_index_name(ofs, lowerdentry, &name);
852 inode = d_inode(upperdentry);
853 if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
854 pr_warn_ratelimited("cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
855 upperdentry, inode->i_ino, inode->i_nlink);
857 * We either have a bug with persistent union nlink or a lower
858 * hardlink was added while overlay is mounted. Adding a lower
859 * hardlink and then unlinking all overlay hardlinks would drop
860 * overlay nlink to zero before all upper inodes are unlinked.
861 * As a safety measure, when that situation is detected, set
862 * the overlay nlink to the index inode nlink minus one for the
863 * index entry itself.
865 set_nlink(d_inode(dentry), inode->i_nlink - 1);
866 ovl_set_nlink_upper(dentry);
870 inode_lock_nested(dir, I_MUTEX_PARENT);
871 index = ovl_lookup_upper(ofs, name.name, indexdir, name.len);
872 err = PTR_ERR(index);
875 } else if (ovl_index_all(dentry->d_sb)) {
876 /* Whiteout orphan index to block future open by handle */
877 err = ovl_cleanup_and_whiteout(OVL_FS(dentry->d_sb),
880 /* Cleanup orphan index entries */
881 err = ovl_cleanup(ofs, dir, index);
894 pr_err("cleanup index of '%pd2' failed (%i)\n", dentry, err);
899 * Operations that change overlay inode and upper inode nlink need to be
900 * synchronized with copy up for persistent nlink accounting.
902 int ovl_nlink_start(struct dentry *dentry)
904 struct inode *inode = d_inode(dentry);
905 const struct cred *old_cred;
912 * With inodes index is enabled, we store the union overlay nlink
913 * in an xattr on the index inode. When whiting out an indexed lower,
914 * we need to decrement the overlay persistent nlink, but before the
915 * first copy up, we have no upper index inode to store the xattr.
917 * As a workaround, before whiteout/rename over an indexed lower,
918 * copy up to create the upper index. Creating the upper index will
919 * initialize the overlay nlink, so it could be dropped if unlink
920 * or rename succeeds.
922 * TODO: implement metadata only index copy up when called with
923 * ovl_copy_up_flags(dentry, O_PATH).
925 if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
926 err = ovl_copy_up(dentry);
931 err = ovl_inode_lock_interruptible(inode);
935 if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode))
938 old_cred = ovl_override_creds(dentry->d_sb);
940 * The overlay inode nlink should be incremented/decremented IFF the
941 * upper operation succeeds, along with nlink change of upper inode.
942 * Therefore, before link/unlink/rename, we store the union nlink
943 * value relative to the upper inode nlink in an upper inode xattr.
945 err = ovl_set_nlink_upper(dentry);
946 revert_creds(old_cred);
950 ovl_inode_unlock(inode);
955 void ovl_nlink_end(struct dentry *dentry)
957 struct inode *inode = d_inode(dentry);
959 if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) {
960 const struct cred *old_cred;
962 old_cred = ovl_override_creds(dentry->d_sb);
963 ovl_cleanup_index(dentry);
964 revert_creds(old_cred);
967 ovl_inode_unlock(inode);
970 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
972 /* Workdir should not be the same as upperdir */
973 if (workdir == upperdir)
976 /* Workdir should not be subdir of upperdir and vice versa */
977 if (lock_rename(workdir, upperdir) != NULL)
983 unlock_rename(workdir, upperdir);
985 pr_err("failed to lock workdir+upperdir\n");
989 /* err < 0, 0 if no metacopy xattr, 1 if metacopy xattr found */
990 int ovl_check_metacopy_xattr(struct ovl_fs *ofs, const struct path *path)
994 /* Only regular files can have metacopy xattr */
995 if (!S_ISREG(d_inode(path->dentry)->i_mode))
998 res = ovl_path_getxattr(ofs, path, OVL_XATTR_METACOPY, NULL, 0);
1000 if (res == -ENODATA || res == -EOPNOTSUPP)
1003 * getxattr on user.* may fail with EACCES in case there's no
1004 * read permission on the inode. Not much we can do, other than
1005 * tell the caller that this is not a metacopy inode.
1007 if (ofs->config.userxattr && res == -EACCES)
1014 pr_warn_ratelimited("failed to get metacopy (%i)\n", res);
1018 bool ovl_is_metacopy_dentry(struct dentry *dentry)
1020 struct ovl_entry *oe = dentry->d_fsdata;
1022 if (!d_is_reg(dentry))
1025 if (ovl_dentry_upper(dentry)) {
1026 if (!ovl_has_upperdata(d_inode(dentry)))
1031 return (oe->numlower > 1);
1034 char *ovl_get_redirect_xattr(struct ovl_fs *ofs, const struct path *path, int padding)
1037 char *s, *next, *buf = NULL;
1039 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, NULL, 0);
1040 if (res == -ENODATA || res == -EOPNOTSUPP)
1047 buf = kzalloc(res + padding + 1, GFP_KERNEL);
1049 return ERR_PTR(-ENOMEM);
1051 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, buf, res);
1057 if (buf[0] == '/') {
1058 for (s = buf; *s++ == '/'; s = next) {
1059 next = strchrnul(s, '/');
1064 if (strchr(buf, '/') != NULL)
1070 pr_warn_ratelimited("invalid redirect (%s)\n", buf);
1074 pr_warn_ratelimited("failed to get redirect (%i)\n", res);
1077 return ERR_PTR(res);
1081 * ovl_sync_status() - Check fs sync status for volatile mounts
1083 * Returns 1 if this is not a volatile mount and a real sync is required.
1085 * Returns 0 if syncing can be skipped because mount is volatile, and no errors
1086 * have occurred on the upperdir since the mount.
1088 * Returns -errno if it is a volatile mount, and the error that occurred since
1089 * the last mount. If the error code changes, it'll return the latest error
1093 int ovl_sync_status(struct ovl_fs *ofs)
1095 struct vfsmount *mnt;
1097 if (ovl_should_sync(ofs))
1100 mnt = ovl_upper_mnt(ofs);
1104 return errseq_check(&mnt->mnt_sb->s_wb_err, ofs->errseq);
1108 * ovl_copyattr() - copy inode attributes from layer to ovl inode
1110 * When overlay copies inode information from an upper or lower layer to the
1111 * relevant overlay inode it will apply the idmapping of the upper or lower
1112 * layer when doing so ensuring that the ovl inode ownership will correctly
1113 * reflect the ownership of the idmapped upper or lower layer. For example, an
1114 * idmapped upper or lower layer mapping id 1001 to id 1000 will take care to
1115 * map any lower or upper inode owned by id 1001 to id 1000. These mapping
1116 * helpers are nops when the relevant layer isn't idmapped.
1118 void ovl_copyattr(struct inode *inode)
1120 struct path realpath;
1121 struct inode *realinode;
1122 struct user_namespace *real_mnt_userns;
1124 ovl_i_path_real(inode, &realpath);
1125 realinode = d_inode(realpath.dentry);
1126 real_mnt_userns = mnt_user_ns(realpath.mnt);
1128 inode->i_uid = i_uid_into_mnt(real_mnt_userns, realinode);
1129 inode->i_gid = i_gid_into_mnt(real_mnt_userns, realinode);
1130 inode->i_mode = realinode->i_mode;
1131 inode->i_atime = realinode->i_atime;
1132 inode->i_mtime = realinode->i_mtime;
1133 inode->i_ctime = realinode->i_ctime;
1134 i_size_write(inode, i_size_read(realinode));