ext4: don't access the source subdirectory content on same-directory rename
authorAl Viro <viro@zeniv.linux.org.uk>
Tue, 17 Oct 2023 18:55:42 +0000 (14:55 -0400)
committerAl Viro <viro@zeniv.linux.org.uk>
Sat, 25 Nov 2023 07:53:20 +0000 (02:53 -0500)
We can't really afford locking the source on same-directory rename;
currently vfs_rename() tries to do that, but it will have to be changed.
The logics in ext4 is lazy and goes looking for ".." in source even in
same-directory case.  It's not hard to get rid of that, leaving that
behaviour only for cross-directory case; that VFS can get locks safely
(and will keep doing that after the coming changes).

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/ext4/namei.c

index d252935f9c8abbf72217451398287919b69d4c09..467ba47a691c443ce57e1f4608f0986890b38915 100644 (file)
@@ -3591,10 +3591,14 @@ struct ext4_renament {
        int dir_inlined;
 };
 
-static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
+static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
 {
        int retval;
 
+       ent->is_dir = true;
+       if (!is_cross)
+               return 0;
+
        ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
                                              &retval, &ent->parent_de,
                                              &ent->dir_inlined);
@@ -3612,6 +3616,9 @@ static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
 {
        int retval;
 
+       if (!ent->dir_bh)
+               return 0;
+
        ent->parent_de->inode = cpu_to_le32(dir_ino);
        BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
        if (!ent->dir_inlined) {
@@ -3900,7 +3907,7 @@ static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
                        if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
                                goto end_rename;
                }
-               retval = ext4_rename_dir_prepare(handle, &old);
+               retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
                if (retval)
                        goto end_rename;
        }
@@ -3964,7 +3971,7 @@ static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
        }
        inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
        ext4_update_dx_flag(old.dir);
-       if (old.dir_bh) {
+       if (old.is_dir) {
                retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
                if (retval)
                        goto end_rename;
@@ -3987,7 +3994,7 @@ static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
        if (unlikely(retval))
                goto end_rename;
 
-       if (S_ISDIR(old.inode->i_mode)) {
+       if (old.is_dir) {
                /*
                 * We disable fast commits here that's because the
                 * replay code is not yet capable of changing dot dot
@@ -4114,14 +4121,12 @@ static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
                ext4_handle_sync(handle);
 
        if (S_ISDIR(old.inode->i_mode)) {
-               old.is_dir = true;
-               retval = ext4_rename_dir_prepare(handle, &old);
+               retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
                if (retval)
                        goto end_rename;
        }
        if (S_ISDIR(new.inode->i_mode)) {
-               new.is_dir = true;
-               retval = ext4_rename_dir_prepare(handle, &new);
+               retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
                if (retval)
                        goto end_rename;
        }