ovl: redirect on rename-dir

Current code returns EXDEV when a directory would need to be copied up to
move.  We could copy up the directory tree in this case, but there's
another, simpler solution: point to old lower directory from moved upper
directory.

This is achieved with a "trusted.overlay.redirect" xattr storing the path
relative to the root of the overlay.  After such attribute has been set,
the directory can be moved without further actions required.

This is a backward incompatible feature, old kernels won't be able to
correctly mount an overlay containing redirected directories.

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Documentation/filesystems/overlayfs.txt

@@ -130,6 +130,23 @@ directory.
 Readdir on directories that are not merged is simply handled by the
 underlying directory (upper or lower).
 
+renaming directories
+--------------------
+
+When renaming a directory that is on the lower layer or merged (i.e. the
+directory was not created on the upper layer to start with) overlayfs can
+handle it in two different ways:
+
+1) return EXDEV error: this error is returned by rename(2) when trying to
+   move a file or directory across filesystem boundaries.  Hence
+   applications are usually prepared to hande this error (mv(1) for example
+   recursively copies the directory tree).  This is the default behavior.
+
+2) If the "redirect_dir" feature is enabled, then the directory will be
+   copied up (but not the contents).  Then the "trusted.overlay.redirect"
+   extended attribute is set to the path of the original location from the
+   root of the overlay.  Finally the directory is moved to the new
+   location.
 
 Non-directories
 ---------------
@@ -189,8 +206,8 @@ If a file with multiple hard links is copied up, then this will
 "break" the link.  Changes will not be propagated to other names
 referring to the same inode.
 
-Directory trees are not copied up.  If rename(2) is performed on a directory
-which is on the lower layer or is merged, then -EXDEV will be returned.
+Unless "redirect_dir" feature is enabled, rename(2) on a lower or merged
+directory will fail with EXDEV.
 
 Changes to underlying filesystems
 ---------------------------------

fs/overlayfs/copy_up.c

@@ -324,17 +324,11 @@ out_cleanup:
 /*
  * Copy up a single dentry
  *
- * Directory renames only allowed on "pure upper" (already created on
- * upper filesystem, never copied up).  Directories which are on lower or
- * are merged may not be renamed.  For these -EXDEV is returned and
- * userspace has to deal with it.  This means, when copying up a
- * directory we can rely on it and ancestors being stable.
- *
- * Non-directory renames start with copy up of source if necessary.  The
- * actual rename will only proceed once the copy up was successful.  Copy
- * up uses upper parent i_mutex for exclusion.  Since rename can change
- * d_parent it is possible that the copy up will lock the old parent.  At
- * that point the file will have already been copied up anyway.
+ * All renames start with copy up of source if necessary.  The actual
+ * rename will only proceed once the copy up was successful.  Copy up uses
+ * upper parent i_mutex for exclusion.  Since rename can change d_parent it
+ * is possible that the copy up will lock the old parent.  At that point
+ * the file will have already been copied up anyway.
  */
 int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 		    struct path *lowerpath, struct kstat *stat)
@@ -346,7 +340,6 @@ int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 	struct path parentpath;
 	struct dentry *lowerdentry = lowerpath->dentry;
 	struct dentry *upperdir;
-	struct dentry *upperdentry;
 	const char *link = NULL;
 
 	if (WARN_ON(!workdir))
@@ -372,8 +365,7 @@ int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 		pr_err("overlayfs: failed to lock workdir+upperdir\n");
 		goto out_unlock;
 	}
-	upperdentry = ovl_dentry_upper(dentry);
-	if (upperdentry) {
+	if (ovl_dentry_upper(dentry)) {
 		/* Raced with another copy-up?  Nothing to do, then... */
 		err = 0;
 		goto out_unlock;

fs/overlayfs/dir.c

@@ -15,6 +15,7 @@
 #include <linux/posix_acl.h>
 #include <linux/posix_acl_xattr.h>
 #include <linux/atomic.h>
+#include <linux/ratelimit.h>
 #include "overlayfs.h"
 
 void ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
@@ -757,6 +758,104 @@ static bool ovl_type_merge_or_lower(struct dentry *dentry)
 	return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 }
 
+static bool ovl_can_move(struct dentry *dentry)
+{
+	return ovl_redirect_dir(dentry->d_sb) ||
+		!d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
+}
+
+#define OVL_REDIRECT_MAX 256
+
+static char *ovl_get_redirect(struct dentry *dentry, bool samedir)
+{
+	char *buf, *ret;
+	struct dentry *d, *tmp;
+	int buflen = OVL_REDIRECT_MAX + 1;
+
+	if (samedir) {
+		ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
+			       GFP_KERNEL);
+		goto out;
+	}
+
+	buf = ret = kmalloc(buflen, GFP_TEMPORARY);
+	if (!buf)
+		goto out;
+
+	buflen--;
+	buf[buflen] = '\0';
+	for (d = dget(dentry); !IS_ROOT(d);) {
+		const char *name;
+		int thislen;
+
+		spin_lock(&d->d_lock);
+		name = ovl_dentry_get_redirect(d);
+		if (name) {
+			thislen = strlen(name);
+		} else {
+			name = d->d_name.name;
+			thislen = d->d_name.len;
+		}
+
+		/* If path is too long, fall back to userspace move */
+		if (thislen + (name[0] != '/') > buflen) {
+			ret = ERR_PTR(-EXDEV);
+			spin_unlock(&d->d_lock);
+			goto out_put;
+		}
+
+		buflen -= thislen;
+		memcpy(&buf[buflen], name, thislen);
+		tmp = dget_dlock(d->d_parent);
+		spin_unlock(&d->d_lock);
+
+		dput(d);
+		d = tmp;
+
+		/* Absolute redirect: finished */
+		if (buf[buflen] == '/')
+			break;
+		buflen--;
+		buf[buflen] = '/';
+	}
+	ret = kstrdup(&buf[buflen], GFP_KERNEL);
+out_put:
+	dput(d);
+	kfree(buf);
+out:
+	return ret ? ret : ERR_PTR(-ENOMEM);
+}
+
+static int ovl_set_redirect(struct dentry *dentry, bool samedir)
+{
+	int err;
+	const char *redirect = ovl_dentry_get_redirect(dentry);
+
+	if (redirect && (samedir || redirect[0] == '/'))
+		return 0;
+
+	redirect = ovl_get_redirect(dentry, samedir);
+	if (IS_ERR(redirect))
+		return PTR_ERR(redirect);
+
+	err = ovl_do_setxattr(ovl_dentry_upper(dentry), OVL_XATTR_REDIRECT,
+			      redirect, strlen(redirect), 0);
+	if (!err) {
+		spin_lock(&dentry->d_lock);
+		ovl_dentry_set_redirect(dentry, redirect);
+		spin_unlock(&dentry->d_lock);
+	} else {
+		kfree(redirect);
+		if (err == -EOPNOTSUPP)
+			ovl_clear_redirect_dir(dentry->d_sb);
+		else
+			pr_warn_ratelimited("overlay: failed to set redirect (%i)\n", err);
+		/* Fall back to userspace copy-up */
+		err = -EXDEV;
+	}
+	return err;
+}
+
 static int ovl_rename(struct inode *olddir, struct dentry *old,
 		      struct inode *newdir, struct dentry *new,
 		      unsigned int flags)
@@ -773,6 +872,7 @@ static int ovl_rename(struct inode *olddir, struct dentry *old,
 	bool overwrite = !(flags & RENAME_EXCHANGE);
 	bool is_dir = d_is_dir(old);
 	bool new_is_dir = d_is_dir(new);
+	bool samedir = olddir == newdir;
 	struct dentry *opaquedir = NULL;
 	const struct cred *old_cred = NULL;
 
@@ -784,9 +884,9 @@ static int ovl_rename(struct inode *olddir, struct dentry *old,
 
 	/* Don't copy up directory trees */
 	err = -EXDEV;
-	if (is_dir && ovl_type_merge_or_lower(old))
+	if (!ovl_can_move(old))
 		goto out;
-	if (!overwrite && new_is_dir && ovl_type_merge_or_lower(new))
+	if (!overwrite && !ovl_can_move(new))
 		goto out;
 
 	err = ovl_want_write(old);
@@ -837,7 +937,6 @@ static int ovl_rename(struct inode *olddir, struct dentry *old,
 
 	trap = lock_rename(new_upperdir, old_upperdir);
 
-
 	olddentry = lookup_one_len(old->d_name.name, old_upperdir,
 				   old->d_name.len);
 	err = PTR_ERR(olddentry);
@@ -880,18 +979,29 @@ static int ovl_rename(struct inode *olddir, struct dentry *old,
 	if (WARN_ON(olddentry->d_inode == newdentry->d_inode))
 		goto out_dput;
 
-	if (is_dir && !old_opaque && ovl_lower_positive(new)) {
-		err = ovl_set_opaque(olddentry);
-		if (err)
-			goto out_dput;
-		ovl_dentry_set_opaque(old, true);
+	if (is_dir) {
+		if (ovl_type_merge_or_lower(old)) {
+			err = ovl_set_redirect(old, samedir);
+			if (err)
+				goto out_dput;
+		} else if (!old_opaque && ovl_lower_positive(new)) {
+			err = ovl_set_opaque(olddentry);
+			if (err)
+				goto out_dput;
+			ovl_dentry_set_opaque(old, true);
+		}
 	}
-	if (!overwrite &&
-	    new_is_dir && !new_opaque && ovl_lower_positive(old)) {
-		err = ovl_set_opaque(newdentry);
-		if (err)
-			goto out_dput;
-		ovl_dentry_set_opaque(new, true);
+	if (!overwrite && new_is_dir) {
+		if (ovl_type_merge_or_lower(new)) {
+			err = ovl_set_redirect(new, samedir);
+			if (err)
+				goto out_dput;
+		} else if (!new_opaque && ovl_lower_positive(old)) {
+			err = ovl_set_opaque(newdentry);
+			if (err)
+				goto out_dput;
+			ovl_dentry_set_opaque(new, true);
+		}
 	}
 
 	err = ovl_do_rename(old_upperdir->d_inode, olddentry,

fs/overlayfs/overlayfs.h

@@ -157,6 +157,10 @@ void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache);
 bool ovl_dentry_is_opaque(struct dentry *dentry);
 bool ovl_dentry_is_whiteout(struct dentry *dentry);
 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque);
+bool ovl_redirect_dir(struct super_block *sb);
+void ovl_clear_redirect_dir(struct super_block *sb);
+const char *ovl_dentry_get_redirect(struct dentry *dentry);
+void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect);
 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry);
 void ovl_inode_init(struct inode *inode, struct inode *realinode,
 		    bool is_upper);

fs/overlayfs/ovl_entry.h

@@ -13,6 +13,7 @@ struct ovl_config {
 	char *upperdir;
 	char *workdir;
 	bool default_permissions;
+	bool redirect_dir;
 };
 
 /* private information held for overlayfs's superblock */

fs/overlayfs/super.c

@@ -226,6 +226,8 @@ enum {
 	OPT_UPPERDIR,
 	OPT_WORKDIR,
 	OPT_DEFAULT_PERMISSIONS,
+	OPT_REDIRECT_DIR_ON,
+	OPT_REDIRECT_DIR_OFF,
 	OPT_ERR,
 };
 
@@ -234,6 +236,8 @@ static const match_table_t ovl_tokens = {
 	{OPT_UPPERDIR,			"upperdir=%s"},
 	{OPT_WORKDIR,			"workdir=%s"},
 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
+	{OPT_REDIRECT_DIR_ON,		"redirect_dir=on"},
+	{OPT_REDIRECT_DIR_OFF,		"redirect_dir=off"},
 	{OPT_ERR,			NULL}
 };
 
@@ -298,6 +302,14 @@ static int ovl_parse_opt(char *opt, struct ovl_config *config)
 			config->default_permissions = true;
 			break;
 
+		case OPT_REDIRECT_DIR_ON:
+			config->redirect_dir = true;
+			break;
+
+		case OPT_REDIRECT_DIR_OFF:
+			config->redirect_dir = false;
+			break;
+
 		default:
 			pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 			return -EINVAL;

fs/overlayfs/util.c

@@ -176,6 +176,35 @@ void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 	oe->opaque = opaque;
 }
 
+bool ovl_redirect_dir(struct super_block *sb)
+{
+	struct ovl_fs *ofs = sb->s_fs_info;
+
+	return ofs->config.redirect_dir;
+}
+
+void ovl_clear_redirect_dir(struct super_block *sb)
+{
+	struct ovl_fs *ofs = sb->s_fs_info;
+
+	ofs->config.redirect_dir = false;
+}
+
+const char *ovl_dentry_get_redirect(struct dentry *dentry)
+{
+	struct ovl_entry *oe = dentry->d_fsdata;
+
+	return oe->redirect;
+}
+
+void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
+{
+	struct ovl_entry *oe = dentry->d_fsdata;
+
+	kfree(oe->redirect);
+	oe->redirect = redirect;
+}
+
 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 {
 	struct ovl_entry *oe = dentry->d_fsdata;