3 * Copyright (C) 2011 Novell Inc.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published by
7 * the Free Software Foundation.
11 #include <linux/namei.h>
12 #include <linux/xattr.h>
13 #include <linux/security.h>
14 #include <linux/cred.h>
15 #include "overlayfs.h"
17 void ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
22 if (S_ISDIR(wdentry->d_inode->i_mode))
23 err = ovl_do_rmdir(wdir, wdentry);
25 err = ovl_do_unlink(wdir, wdentry);
29 pr_err("overlayfs: cleanup of '%pd2' failed (%i)\n",
34 struct dentry *ovl_lookup_temp(struct dentry *workdir, struct dentry *dentry)
39 snprintf(name, sizeof(name), "#%lx", (unsigned long) dentry);
41 temp = lookup_one_len(name, workdir, strlen(name));
42 if (!IS_ERR(temp) && temp->d_inode) {
43 pr_err("overlayfs: workdir/%s already exists\n", name);
51 /* caller holds i_mutex on workdir */
52 static struct dentry *ovl_whiteout(struct dentry *workdir,
53 struct dentry *dentry)
56 struct dentry *whiteout;
57 struct inode *wdir = workdir->d_inode;
59 whiteout = ovl_lookup_temp(workdir, dentry);
63 err = ovl_do_whiteout(wdir, whiteout);
66 whiteout = ERR_PTR(err);
72 int ovl_create_real(struct inode *dir, struct dentry *newdentry,
73 struct kstat *stat, const char *link,
74 struct dentry *hardlink, bool debug)
78 if (newdentry->d_inode)
82 err = ovl_do_link(hardlink, dir, newdentry, debug);
84 switch (stat->mode & S_IFMT) {
86 err = ovl_do_create(dir, newdentry, stat->mode, debug);
90 err = ovl_do_mkdir(dir, newdentry, stat->mode, debug);
97 err = ovl_do_mknod(dir, newdentry,
98 stat->mode, stat->rdev, debug);
102 err = ovl_do_symlink(dir, newdentry, link, debug);
109 if (!err && WARN_ON(!newdentry->d_inode)) {
111 * Not quite sure if non-instantiated dentry is legal or not.
112 * VFS doesn't seem to care so check and warn here.
119 static int ovl_set_opaque(struct dentry *upperdentry)
121 return ovl_do_setxattr(upperdentry, ovl_opaque_xattr, "y", 1, 0);
124 static void ovl_remove_opaque(struct dentry *upperdentry)
128 err = ovl_do_removexattr(upperdentry, ovl_opaque_xattr);
130 pr_warn("overlayfs: failed to remove opaque from '%s' (%i)\n",
131 upperdentry->d_name.name, err);
135 static int ovl_dir_getattr(struct vfsmount *mnt, struct dentry *dentry,
139 enum ovl_path_type type;
140 struct path realpath;
142 type = ovl_path_real(dentry, &realpath);
143 err = vfs_getattr(&realpath, stat);
147 stat->dev = dentry->d_sb->s_dev;
148 stat->ino = dentry->d_inode->i_ino;
151 * It's probably not worth it to count subdirs to get the
152 * correct link count. nlink=1 seems to pacify 'find' and
155 if (type == OVL_PATH_MERGE)
161 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
162 struct kstat *stat, const char *link,
163 struct dentry *hardlink)
165 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
166 struct inode *udir = upperdir->d_inode;
167 struct dentry *newdentry;
170 mutex_lock_nested(&udir->i_mutex, I_MUTEX_PARENT);
171 newdentry = lookup_one_len(dentry->d_name.name, upperdir,
173 err = PTR_ERR(newdentry);
174 if (IS_ERR(newdentry))
176 err = ovl_create_real(udir, newdentry, stat, link, hardlink, false);
180 ovl_dentry_version_inc(dentry->d_parent);
181 ovl_dentry_update(dentry, newdentry);
182 ovl_copyattr(newdentry->d_inode, inode);
183 d_instantiate(dentry, inode);
188 mutex_unlock(&udir->i_mutex);
192 static int ovl_lock_rename_workdir(struct dentry *workdir,
193 struct dentry *upperdir)
195 /* Workdir should not be the same as upperdir */
196 if (workdir == upperdir)
199 /* Workdir should not be subdir of upperdir and vice versa */
200 if (lock_rename(workdir, upperdir) != NULL)
206 unlock_rename(workdir, upperdir);
208 pr_err("overlayfs: failed to lock workdir+upperdir\n");
212 static struct dentry *ovl_clear_empty(struct dentry *dentry,
213 struct list_head *list)
215 struct dentry *workdir = ovl_workdir(dentry);
216 struct inode *wdir = workdir->d_inode;
217 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
218 struct inode *udir = upperdir->d_inode;
219 struct path upperpath;
220 struct dentry *upper;
221 struct dentry *opaquedir;
225 err = ovl_lock_rename_workdir(workdir, upperdir);
229 ovl_path_upper(dentry, &upperpath);
230 err = vfs_getattr(&upperpath, &stat);
235 if (!S_ISDIR(stat.mode))
237 upper = upperpath.dentry;
238 if (upper->d_parent->d_inode != udir)
241 opaquedir = ovl_lookup_temp(workdir, dentry);
242 err = PTR_ERR(opaquedir);
243 if (IS_ERR(opaquedir))
246 err = ovl_create_real(wdir, opaquedir, &stat, NULL, NULL, true);
250 err = ovl_copy_xattr(upper, opaquedir);
254 err = ovl_set_opaque(opaquedir);
258 mutex_lock(&opaquedir->d_inode->i_mutex);
259 err = ovl_set_attr(opaquedir, &stat);
260 mutex_unlock(&opaquedir->d_inode->i_mutex);
264 err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
268 ovl_cleanup_whiteouts(upper, list);
269 ovl_cleanup(wdir, upper);
270 unlock_rename(workdir, upperdir);
272 /* dentry's upper doesn't match now, get rid of it */
278 ovl_cleanup(wdir, opaquedir);
282 unlock_rename(workdir, upperdir);
287 static struct dentry *ovl_check_empty_and_clear(struct dentry *dentry,
288 enum ovl_path_type type)
291 struct dentry *ret = NULL;
294 err = ovl_check_empty_dir(dentry, &list);
297 else if (type == OVL_PATH_MERGE)
298 ret = ovl_clear_empty(dentry, &list);
300 ovl_cache_free(&list);
305 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
306 struct kstat *stat, const char *link,
307 struct dentry *hardlink)
309 struct dentry *workdir = ovl_workdir(dentry);
310 struct inode *wdir = workdir->d_inode;
311 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
312 struct inode *udir = upperdir->d_inode;
313 struct dentry *upper;
314 struct dentry *newdentry;
317 err = ovl_lock_rename_workdir(workdir, upperdir);
321 newdentry = ovl_lookup_temp(workdir, dentry);
322 err = PTR_ERR(newdentry);
323 if (IS_ERR(newdentry))
326 upper = lookup_one_len(dentry->d_name.name, upperdir,
328 err = PTR_ERR(upper);
332 err = ovl_create_real(wdir, newdentry, stat, link, hardlink, true);
336 if (S_ISDIR(stat->mode)) {
337 err = ovl_set_opaque(newdentry);
341 err = ovl_do_rename(wdir, newdentry, udir, upper,
346 ovl_cleanup(wdir, upper);
348 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
352 ovl_dentry_version_inc(dentry->d_parent);
353 ovl_dentry_update(dentry, newdentry);
354 ovl_copyattr(newdentry->d_inode, inode);
355 d_instantiate(dentry, inode);
362 unlock_rename(workdir, upperdir);
367 ovl_cleanup(wdir, newdentry);
371 static int ovl_create_or_link(struct dentry *dentry, int mode, dev_t rdev,
372 const char *link, struct dentry *hardlink)
376 struct kstat stat = {
382 inode = ovl_new_inode(dentry->d_sb, mode, dentry->d_fsdata);
386 err = ovl_copy_up(dentry->d_parent);
390 if (!ovl_dentry_is_opaque(dentry)) {
391 err = ovl_create_upper(dentry, inode, &stat, link, hardlink);
393 const struct cred *old_cred;
394 struct cred *override_cred;
397 override_cred = prepare_creds();
402 * CAP_SYS_ADMIN for setting opaque xattr
403 * CAP_DAC_OVERRIDE for create in workdir, rename
404 * CAP_FOWNER for removing whiteout from sticky dir
406 cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
407 cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
408 cap_raise(override_cred->cap_effective, CAP_FOWNER);
409 old_cred = override_creds(override_cred);
411 err = ovl_create_over_whiteout(dentry, inode, &stat, link,
414 revert_creds(old_cred);
415 put_cred(override_cred);
426 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
431 err = ovl_want_write(dentry);
433 err = ovl_create_or_link(dentry, mode, rdev, link, NULL);
434 ovl_drop_write(dentry);
440 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
443 return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
446 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
448 return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
451 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
454 /* Don't allow creation of "whiteout" on overlay */
455 if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
458 return ovl_create_object(dentry, mode, rdev, NULL);
461 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
464 return ovl_create_object(dentry, S_IFLNK, 0, link);
467 static int ovl_link(struct dentry *old, struct inode *newdir,
471 struct dentry *upper;
473 err = ovl_want_write(old);
477 err = ovl_copy_up(old);
481 upper = ovl_dentry_upper(old);
482 err = ovl_create_or_link(new, upper->d_inode->i_mode, 0, NULL, upper);
490 static int ovl_remove_and_whiteout(struct dentry *dentry,
491 enum ovl_path_type type, bool is_dir)
493 struct dentry *workdir = ovl_workdir(dentry);
494 struct inode *wdir = workdir->d_inode;
495 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
496 struct inode *udir = upperdir->d_inode;
497 struct dentry *whiteout;
498 struct dentry *upper;
499 struct dentry *opaquedir = NULL;
503 opaquedir = ovl_check_empty_and_clear(dentry, type);
504 err = PTR_ERR(opaquedir);
505 if (IS_ERR(opaquedir))
509 err = ovl_lock_rename_workdir(workdir, upperdir);
513 whiteout = ovl_whiteout(workdir, dentry);
514 err = PTR_ERR(whiteout);
515 if (IS_ERR(whiteout))
518 if (type == OVL_PATH_LOWER) {
519 upper = lookup_one_len(dentry->d_name.name, upperdir,
521 err = PTR_ERR(upper);
525 err = ovl_do_rename(wdir, whiteout, udir, upper, 0);
532 upper = ovl_dentry_upper(dentry);
536 if (upper->d_parent != upperdir)
540 flags |= RENAME_EXCHANGE;
542 err = ovl_do_rename(wdir, whiteout, udir, upper, flags);
547 ovl_cleanup(wdir, upper);
549 ovl_dentry_version_inc(dentry->d_parent);
554 unlock_rename(workdir, upperdir);
561 ovl_cleanup(wdir, whiteout);
565 static int ovl_remove_upper(struct dentry *dentry, bool is_dir)
567 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
568 struct inode *dir = upperdir->d_inode;
569 struct dentry *upper = ovl_dentry_upper(dentry);
572 mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
574 if (upper->d_parent == upperdir) {
575 /* Don't let d_delete() think it can reset d_inode */
578 err = vfs_rmdir(dir, upper);
580 err = vfs_unlink(dir, upper, NULL);
582 ovl_dentry_version_inc(dentry->d_parent);
586 * Keeping this dentry hashed would mean having to release
587 * upperpath/lowerpath, which could only be done if we are the
588 * sole user of this dentry. Too tricky... Just unhash for
592 mutex_unlock(&dir->i_mutex);
597 static inline int ovl_check_sticky(struct dentry *dentry)
599 struct inode *dir = ovl_dentry_real(dentry->d_parent)->d_inode;
600 struct inode *inode = ovl_dentry_real(dentry)->d_inode;
602 if (check_sticky(dir, inode))
608 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
610 enum ovl_path_type type;
613 err = ovl_check_sticky(dentry);
617 err = ovl_want_write(dentry);
621 err = ovl_copy_up(dentry->d_parent);
625 type = ovl_path_type(dentry);
626 if (type == OVL_PATH_PURE_UPPER) {
627 err = ovl_remove_upper(dentry, is_dir);
629 const struct cred *old_cred;
630 struct cred *override_cred;
633 override_cred = prepare_creds();
638 * CAP_SYS_ADMIN for setting xattr on whiteout, opaque dir
639 * CAP_DAC_OVERRIDE for create in workdir, rename
640 * CAP_FOWNER for removing whiteout from sticky dir
641 * CAP_FSETID for chmod of opaque dir
642 * CAP_CHOWN for chown of opaque dir
644 cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
645 cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
646 cap_raise(override_cred->cap_effective, CAP_FOWNER);
647 cap_raise(override_cred->cap_effective, CAP_FSETID);
648 cap_raise(override_cred->cap_effective, CAP_CHOWN);
649 old_cred = override_creds(override_cred);
651 err = ovl_remove_and_whiteout(dentry, type, is_dir);
653 revert_creds(old_cred);
654 put_cred(override_cred);
657 ovl_drop_write(dentry);
662 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
664 return ovl_do_remove(dentry, false);
667 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
669 return ovl_do_remove(dentry, true);
672 static int ovl_rename2(struct inode *olddir, struct dentry *old,
673 struct inode *newdir, struct dentry *new,
677 enum ovl_path_type old_type;
678 enum ovl_path_type new_type;
679 struct dentry *old_upperdir;
680 struct dentry *new_upperdir;
681 struct dentry *olddentry;
682 struct dentry *newdentry;
686 bool new_create = false;
687 bool cleanup_whiteout = false;
688 bool overwrite = !(flags & RENAME_EXCHANGE);
689 bool is_dir = S_ISDIR(old->d_inode->i_mode);
690 bool new_is_dir = false;
691 struct dentry *opaquedir = NULL;
692 const struct cred *old_cred = NULL;
693 struct cred *override_cred = NULL;
696 if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
699 flags &= ~RENAME_NOREPLACE;
701 err = ovl_check_sticky(old);
705 /* Don't copy up directory trees */
706 old_type = ovl_path_type(old);
708 if ((old_type == OVL_PATH_LOWER || old_type == OVL_PATH_MERGE) && is_dir)
712 err = ovl_check_sticky(new);
716 if (S_ISDIR(new->d_inode->i_mode))
719 new_type = ovl_path_type(new);
721 if (!overwrite && (new_type == OVL_PATH_LOWER || new_type == OVL_PATH_MERGE) && new_is_dir)
725 if (new_type == OVL_PATH_LOWER && old_type == OVL_PATH_LOWER) {
726 if (ovl_dentry_lower(old)->d_inode ==
727 ovl_dentry_lower(new)->d_inode)
730 if (new_type != OVL_PATH_LOWER && old_type != OVL_PATH_LOWER) {
731 if (ovl_dentry_upper(old)->d_inode ==
732 ovl_dentry_upper(new)->d_inode)
736 if (ovl_dentry_is_opaque(new))
737 new_type = OVL_PATH_UPPER;
739 new_type = OVL_PATH_PURE_UPPER;
742 err = ovl_want_write(old);
746 err = ovl_copy_up(old);
750 err = ovl_copy_up(new->d_parent);
754 err = ovl_copy_up(new);
759 old_opaque = old_type != OVL_PATH_PURE_UPPER;
760 new_opaque = new_type != OVL_PATH_PURE_UPPER;
762 if (old_opaque || new_opaque) {
764 override_cred = prepare_creds();
769 * CAP_SYS_ADMIN for setting xattr on whiteout, opaque dir
770 * CAP_DAC_OVERRIDE for create in workdir
771 * CAP_FOWNER for removing whiteout from sticky dir
772 * CAP_FSETID for chmod of opaque dir
773 * CAP_CHOWN for chown of opaque dir
775 cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
776 cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
777 cap_raise(override_cred->cap_effective, CAP_FOWNER);
778 cap_raise(override_cred->cap_effective, CAP_FSETID);
779 cap_raise(override_cred->cap_effective, CAP_CHOWN);
780 old_cred = override_creds(override_cred);
783 if (overwrite && (new_type == OVL_PATH_LOWER || new_type == OVL_PATH_MERGE) && new_is_dir) {
784 opaquedir = ovl_check_empty_and_clear(new, new_type);
785 err = PTR_ERR(opaquedir);
786 if (IS_ERR(opaquedir)) {
788 goto out_revert_creds;
794 if (new->d_inode || !new_opaque) {
795 /* Whiteout source */
796 flags |= RENAME_WHITEOUT;
798 /* Switch whiteouts */
799 flags |= RENAME_EXCHANGE;
801 } else if (is_dir && !new->d_inode && new_opaque) {
802 flags |= RENAME_EXCHANGE;
803 cleanup_whiteout = true;
807 old_upperdir = ovl_dentry_upper(old->d_parent);
808 new_upperdir = ovl_dentry_upper(new->d_parent);
810 trap = lock_rename(new_upperdir, old_upperdir);
812 olddentry = ovl_dentry_upper(old);
813 newdentry = ovl_dentry_upper(new);
816 newdentry = opaquedir;
823 newdentry = lookup_one_len(new->d_name.name, new_upperdir,
825 err = PTR_ERR(newdentry);
826 if (IS_ERR(newdentry))
831 if (olddentry->d_parent != old_upperdir)
833 if (newdentry->d_parent != new_upperdir)
835 if (olddentry == trap)
837 if (newdentry == trap)
840 if (is_dir && !old_opaque && new_opaque) {
841 err = ovl_set_opaque(olddentry);
845 if (!overwrite && new_is_dir && old_opaque && !new_opaque) {
846 err = ovl_set_opaque(newdentry);
851 if (old_opaque || new_opaque) {
852 err = ovl_do_rename(old_upperdir->d_inode, olddentry,
853 new_upperdir->d_inode, newdentry,
856 /* No debug for the plain case */
857 BUG_ON(flags & ~RENAME_EXCHANGE);
858 err = vfs_rename(old_upperdir->d_inode, olddentry,
859 new_upperdir->d_inode, newdentry,
864 if (is_dir && !old_opaque && new_opaque)
865 ovl_remove_opaque(olddentry);
866 if (!overwrite && new_is_dir && old_opaque && !new_opaque)
867 ovl_remove_opaque(newdentry);
871 if (is_dir && old_opaque && !new_opaque)
872 ovl_remove_opaque(olddentry);
873 if (!overwrite && new_is_dir && !old_opaque && new_opaque)
874 ovl_remove_opaque(newdentry);
876 if (old_opaque != new_opaque) {
877 ovl_dentry_set_opaque(old, new_opaque);
879 ovl_dentry_set_opaque(new, old_opaque);
882 if (cleanup_whiteout)
883 ovl_cleanup(old_upperdir->d_inode, newdentry);
885 ovl_dentry_version_inc(old->d_parent);
886 ovl_dentry_version_inc(new->d_parent);
891 unlock_rename(new_upperdir, old_upperdir);
893 if (old_opaque || new_opaque) {
894 revert_creds(old_cred);
895 put_cred(override_cred);
904 const struct inode_operations ovl_dir_inode_operations = {
905 .lookup = ovl_lookup,
907 .symlink = ovl_symlink,
908 .unlink = ovl_unlink,
910 .rename2 = ovl_rename2,
912 .setattr = ovl_setattr,
913 .create = ovl_create,
915 .permission = ovl_permission,
916 .getattr = ovl_dir_getattr,
917 .setxattr = ovl_setxattr,
918 .getxattr = ovl_getxattr,
919 .listxattr = ovl_listxattr,
920 .removexattr = ovl_removexattr,