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/mount.h>
15 #include <linux/slab.h>
16 #include <linux/parser.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/statfs.h>
20 #include <linux/seq_file.h>
21 #include "overlayfs.h"
23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
24 MODULE_DESCRIPTION("Overlay filesystem");
25 MODULE_LICENSE("GPL");
27 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
35 /* private information held for overlayfs's superblock */
37 struct vfsmount *upper_mnt;
39 struct vfsmount **lower_mnt;
40 struct dentry *workdir;
42 /* pathnames of lower and upper dirs, for show_options */
43 struct ovl_config config;
48 /* private information held for every overlayfs dentry */
50 struct dentry *__upperdentry;
51 struct ovl_dir_cache *cache;
60 struct path lowerstack[];
63 const char *ovl_opaque_xattr = "trusted.overlay.opaque";
65 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
67 return oe->numlower ? oe->lowerstack[0].dentry : NULL;
70 enum ovl_path_type ovl_path_type(struct dentry *dentry)
72 struct ovl_entry *oe = dentry->d_fsdata;
73 enum ovl_path_type type = 0;
75 if (oe->__upperdentry) {
76 type = __OVL_PATH_UPPER;
79 if (S_ISDIR(dentry->d_inode->i_mode))
80 type |= __OVL_PATH_MERGE;
81 } else if (!oe->opaque) {
82 type |= __OVL_PATH_PURE;
86 type |= __OVL_PATH_MERGE;
91 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
93 return lockless_dereference(oe->__upperdentry);
96 void ovl_path_upper(struct dentry *dentry, struct path *path)
98 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
99 struct ovl_entry *oe = dentry->d_fsdata;
101 path->mnt = ofs->upper_mnt;
102 path->dentry = ovl_upperdentry_dereference(oe);
105 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
108 enum ovl_path_type type = ovl_path_type(dentry);
110 if (!OVL_TYPE_UPPER(type))
111 ovl_path_lower(dentry, path);
113 ovl_path_upper(dentry, path);
118 struct dentry *ovl_dentry_upper(struct dentry *dentry)
120 struct ovl_entry *oe = dentry->d_fsdata;
122 return ovl_upperdentry_dereference(oe);
125 struct dentry *ovl_dentry_lower(struct dentry *dentry)
127 struct ovl_entry *oe = dentry->d_fsdata;
129 return __ovl_dentry_lower(oe);
132 struct dentry *ovl_dentry_real(struct dentry *dentry)
134 struct ovl_entry *oe = dentry->d_fsdata;
135 struct dentry *realdentry;
137 realdentry = ovl_upperdentry_dereference(oe);
139 realdentry = __ovl_dentry_lower(oe);
144 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
146 struct dentry *realdentry;
148 realdentry = ovl_upperdentry_dereference(oe);
152 realdentry = __ovl_dentry_lower(oe);
158 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
160 struct ovl_entry *oe = dentry->d_fsdata;
165 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
167 struct ovl_entry *oe = dentry->d_fsdata;
172 void ovl_path_lower(struct dentry *dentry, struct path *path)
174 struct ovl_entry *oe = dentry->d_fsdata;
176 *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
179 int ovl_want_write(struct dentry *dentry)
181 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
182 return mnt_want_write(ofs->upper_mnt);
185 void ovl_drop_write(struct dentry *dentry)
187 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
188 mnt_drop_write(ofs->upper_mnt);
191 struct dentry *ovl_workdir(struct dentry *dentry)
193 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
197 bool ovl_dentry_is_opaque(struct dentry *dentry)
199 struct ovl_entry *oe = dentry->d_fsdata;
203 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
205 struct ovl_entry *oe = dentry->d_fsdata;
209 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
211 struct ovl_entry *oe = dentry->d_fsdata;
213 WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
214 WARN_ON(oe->__upperdentry);
215 BUG_ON(!upperdentry->d_inode);
217 * Make sure upperdentry is consistent before making it visible to
218 * ovl_upperdentry_dereference().
221 oe->__upperdentry = upperdentry;
224 void ovl_dentry_version_inc(struct dentry *dentry)
226 struct ovl_entry *oe = dentry->d_fsdata;
228 WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
232 u64 ovl_dentry_version_get(struct dentry *dentry)
234 struct ovl_entry *oe = dentry->d_fsdata;
236 WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
240 bool ovl_is_whiteout(struct dentry *dentry)
242 struct inode *inode = dentry->d_inode;
244 return inode && IS_WHITEOUT(inode);
247 static bool ovl_is_opaquedir(struct dentry *dentry)
251 struct inode *inode = dentry->d_inode;
253 if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
256 res = inode->i_op->getxattr(dentry, ovl_opaque_xattr, &val, 1);
257 if (res == 1 && val == 'y')
263 static void ovl_dentry_release(struct dentry *dentry)
265 struct ovl_entry *oe = dentry->d_fsdata;
270 dput(oe->__upperdentry);
271 for (i = 0; i < oe->numlower; i++)
272 dput(oe->lowerstack[i].dentry);
277 static const struct dentry_operations ovl_dentry_operations = {
278 .d_release = ovl_dentry_release,
281 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
283 size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
284 struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
287 oe->numlower = numlower;
292 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
295 struct dentry *dentry;
297 mutex_lock(&dir->d_inode->i_mutex);
298 dentry = lookup_one_len(name->name, dir, name->len);
299 mutex_unlock(&dir->d_inode->i_mutex);
301 if (IS_ERR(dentry)) {
302 if (PTR_ERR(dentry) == -ENOENT)
304 } else if (!dentry->d_inode) {
312 * Returns next layer in stack starting from top.
313 * Returns -1 if this is the last layer.
315 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
317 struct ovl_entry *oe = dentry->d_fsdata;
321 ovl_path_upper(dentry, path);
323 return oe->numlower ? 1 : -1;
326 BUG_ON(idx > oe->numlower);
327 *path = oe->lowerstack[idx - 1];
329 return (idx < oe->numlower) ? idx + 1 : -1;
332 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
335 struct ovl_entry *oe;
336 struct ovl_entry *poe = dentry->d_parent->d_fsdata;
337 struct path *stack = NULL;
338 struct dentry *upperdir, *upperdentry = NULL;
339 unsigned int ctr = 0;
340 struct inode *inode = NULL;
341 bool upperopaque = false;
342 struct dentry *this, *prev = NULL;
346 upperdir = ovl_upperdentry_dereference(poe);
348 this = ovl_lookup_real(upperdir, &dentry->d_name);
354 if (ovl_is_whiteout(this)) {
358 } else if (poe->numlower && ovl_is_opaquedir(this)) {
362 upperdentry = prev = this;
365 if (!upperopaque && poe->numlower) {
367 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
372 for (i = 0; !upperopaque && i < poe->numlower; i++) {
374 struct path lowerpath = poe->lowerstack[i];
377 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
381 * If it's positive, then treat ENAMETOOLONG as ENOENT.
383 if (err == -ENAMETOOLONG && (upperdentry || ctr))
389 if (ovl_is_whiteout(this)) {
394 * Only makes sense to check opaque dir if this is not the
397 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
400 * If this is a non-directory then stop here.
402 * FIXME: check for opaqueness maybe better done in remove code.
404 if (!S_ISDIR(this->d_inode->i_mode)) {
406 } else if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
407 !S_ISDIR(this->d_inode->i_mode))) {
408 if (prev == upperdentry)
413 stack[ctr].dentry = this;
414 stack[ctr].mnt = lowerpath.mnt;
421 oe = ovl_alloc_entry(ctr);
426 if (upperdentry || ctr) {
427 struct dentry *realdentry;
429 realdentry = upperdentry ? upperdentry : stack[0].dentry;
432 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
436 ovl_copyattr(realdentry->d_inode, inode);
439 oe->opaque = upperopaque;
440 oe->__upperdentry = upperdentry;
441 memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
443 dentry->d_fsdata = oe;
444 d_add(dentry, inode);
451 for (i = 0; i < ctr; i++)
452 dput(stack[i].dentry);
460 struct file *ovl_path_open(struct path *path, int flags)
462 return dentry_open(path, flags, current_cred());
465 static void ovl_put_super(struct super_block *sb)
467 struct ovl_fs *ufs = sb->s_fs_info;
471 mntput(ufs->upper_mnt);
472 for (i = 0; i < ufs->numlower; i++)
473 mntput(ufs->lower_mnt[i]);
475 kfree(ufs->config.lowerdir);
476 kfree(ufs->config.upperdir);
477 kfree(ufs->config.workdir);
483 * @sb: The overlayfs super block
484 * @buf: The struct kstatfs to fill in with stats
486 * Get the filesystem statistics. As writes always target the upper layer
487 * filesystem pass the statfs to the upper filesystem (if it exists)
489 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
491 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
492 struct dentry *root_dentry = dentry->d_sb->s_root;
496 ovl_path_real(root_dentry, &path);
498 err = vfs_statfs(&path, buf);
500 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
501 buf->f_type = OVERLAYFS_SUPER_MAGIC;
510 * Prints the mount options for a given superblock.
511 * Returns zero; does not fail.
513 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
515 struct super_block *sb = dentry->d_sb;
516 struct ovl_fs *ufs = sb->s_fs_info;
518 seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
519 seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
520 seq_printf(m, ",workdir=%s", ufs->config.workdir);
524 static const struct super_operations ovl_super_operations = {
525 .put_super = ovl_put_super,
526 .statfs = ovl_statfs,
527 .show_options = ovl_show_options,
537 static const match_table_t ovl_tokens = {
538 {OPT_LOWERDIR, "lowerdir=%s"},
539 {OPT_UPPERDIR, "upperdir=%s"},
540 {OPT_WORKDIR, "workdir=%s"},
544 static char *ovl_next_opt(char **s)
552 for (p = sbegin; *p; p++) {
557 } else if (*p == ',') {
567 static int ovl_parse_opt(char *opt, struct ovl_config *config)
571 while ((p = ovl_next_opt(&opt)) != NULL) {
573 substring_t args[MAX_OPT_ARGS];
578 token = match_token(p, ovl_tokens, args);
581 kfree(config->upperdir);
582 config->upperdir = match_strdup(&args[0]);
583 if (!config->upperdir)
588 kfree(config->lowerdir);
589 config->lowerdir = match_strdup(&args[0]);
590 if (!config->lowerdir)
595 kfree(config->workdir);
596 config->workdir = match_strdup(&args[0]);
597 if (!config->workdir)
608 #define OVL_WORKDIR_NAME "work"
610 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
611 struct dentry *dentry)
613 struct inode *dir = dentry->d_inode;
616 bool retried = false;
618 err = mnt_want_write(mnt);
622 mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
624 work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
625 strlen(OVL_WORKDIR_NAME));
628 struct kstat stat = {
638 ovl_cleanup(dir, work);
643 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
648 mutex_unlock(&dir->i_mutex);
659 static void ovl_unescape(char *s)
672 static bool ovl_is_allowed_fs_type(struct dentry *root)
674 const struct dentry_operations *dop = root->d_op;
678 * - automount filesystems
679 * - filesystems with revalidate (FIXME for lower layer)
680 * - filesystems with case insensitive names
683 (dop->d_manage || dop->d_automount ||
684 dop->d_revalidate || dop->d_weak_revalidate ||
685 dop->d_compare || dop->d_hash)) {
691 static int ovl_mount_dir_noesc(const char *name, struct path *path)
695 err = kern_path(name, LOOKUP_FOLLOW, path);
697 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
701 if (!ovl_is_allowed_fs_type(path->dentry)) {
702 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
705 if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
706 pr_err("overlayfs: '%s' not a directory\n", name);
717 static int ovl_mount_dir(const char *name, struct path *path)
720 char *tmp = kstrdup(name, GFP_KERNEL);
724 err = ovl_mount_dir_noesc(tmp, path);
730 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
734 struct kstatfs statfs;
736 err = ovl_mount_dir(name, path);
740 err = vfs_statfs(path, &statfs);
742 pr_err("overlayfs: statfs failed on '%s'\n", name);
745 *namelen = max(*namelen, statfs.f_namelen);
746 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
756 /* Workdir should not be subdir of upperdir and vice versa */
757 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
761 if (workdir != upperdir) {
762 ok = (lock_rename(workdir, upperdir) == NULL);
763 unlock_rename(workdir, upperdir);
768 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
770 struct path lowerpath;
771 struct path upperpath;
772 struct path workpath;
773 struct dentry *root_dentry;
774 struct ovl_entry *oe;
776 struct vfsmount *mnt;
781 ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
785 err = ovl_parse_opt((char *) data, &ufs->config);
787 goto out_free_config;
789 /* FIXME: workdir is not needed for a R/O mount */
791 if (!ufs->config.upperdir || !ufs->config.lowerdir ||
792 !ufs->config.workdir) {
793 pr_err("overlayfs: missing upperdir or lowerdir or workdir\n");
794 goto out_free_config;
797 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
799 goto out_free_config;
801 err = ovl_mount_dir(ufs->config.workdir, &workpath);
803 goto out_put_upperpath;
805 if (upperpath.mnt != workpath.mnt) {
806 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
807 goto out_put_workpath;
809 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
810 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
811 goto out_put_workpath;
813 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
815 err = ovl_lower_dir(ufs->config.lowerdir, &lowerpath,
816 &ufs->lower_namelen, &sb->s_stack_depth);
818 goto out_put_workpath;
822 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
823 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
824 goto out_put_lowerpath;
827 ufs->upper_mnt = clone_private_mount(&upperpath);
828 err = PTR_ERR(ufs->upper_mnt);
829 if (IS_ERR(ufs->upper_mnt)) {
830 pr_err("overlayfs: failed to clone upperpath\n");
831 goto out_put_lowerpath;
834 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
835 err = PTR_ERR(ufs->workdir);
836 if (IS_ERR(ufs->workdir)) {
837 pr_err("overlayfs: failed to create directory %s/%s\n",
838 ufs->config.workdir, OVL_WORKDIR_NAME);
839 goto out_put_upper_mnt;
842 ufs->lower_mnt = kcalloc(1, sizeof(struct vfsmount *), GFP_KERNEL);
843 if (ufs->lower_mnt == NULL)
844 goto out_put_workdir;
846 mnt = clone_private_mount(&lowerpath);
849 pr_err("overlayfs: failed to clone lowerpath\n");
850 goto out_put_lower_mnt;
853 * Make lower_mnt R/O. That way fchmod/fchown on lower file
854 * will fail instead of modifying lower fs.
856 mnt->mnt_flags |= MNT_READONLY;
858 ufs->lower_mnt[0] = mnt;
861 /* If the upper fs is r/o, we mark overlayfs r/o too */
862 if (ufs->upper_mnt->mnt_sb->s_flags & MS_RDONLY)
863 sb->s_flags |= MS_RDONLY;
865 sb->s_d_op = &ovl_dentry_operations;
868 oe = ovl_alloc_entry(1);
870 goto out_put_lower_mnt;
872 root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
876 mntput(upperpath.mnt);
877 mntput(lowerpath.mnt);
880 oe->__upperdentry = upperpath.dentry;
881 oe->lowerstack[0].dentry = lowerpath.dentry;
882 oe->lowerstack[0].mnt = ufs->lower_mnt[0];
884 root_dentry->d_fsdata = oe;
886 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
887 sb->s_op = &ovl_super_operations;
888 sb->s_root = root_dentry;
896 for (i = 0; i < ufs->numlower; i++)
897 mntput(ufs->lower_mnt[i]);
898 kfree(ufs->lower_mnt);
902 mntput(ufs->upper_mnt);
904 path_put(&lowerpath);
908 path_put(&upperpath);
910 kfree(ufs->config.lowerdir);
911 kfree(ufs->config.upperdir);
912 kfree(ufs->config.workdir);
918 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
919 const char *dev_name, void *raw_data)
921 return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
924 static struct file_system_type ovl_fs_type = {
925 .owner = THIS_MODULE,
928 .kill_sb = kill_anon_super,
930 MODULE_ALIAS_FS("overlay");
932 static int __init ovl_init(void)
934 return register_filesystem(&ovl_fs_type);
937 static void __exit ovl_exit(void)
939 unregister_filesystem(&ovl_fs_type);
942 module_init(ovl_init);
943 module_exit(ovl_exit);