vfs: fix vfs_rename_dir for FS_RENAME_DOES_D_MOVE filesystems
authorMiklos Szeredi <mszeredi@suse.cz>
Tue, 24 May 2011 20:06:10 +0000 (13:06 -0700)
committerAl Viro <viro@zeniv.linux.org.uk>
Thu, 26 May 2011 11:26:50 +0000 (07:26 -0400)
commitb5afd2c406f5c6272d916fd705f44f070fbbc0ba
tree93378abaf5120d5fe449ee77234368291331615a
parent5c5d3f3b871a719e2c6413b85009c723adeb27e1
vfs: fix vfs_rename_dir for FS_RENAME_DOES_D_MOVE filesystems

vfs_rename_dir() doesn't properly account for filesystems with
FS_RENAME_DOES_D_MOVE.  If new_dentry has a target inode attached, it
unhashes the new_dentry prior to the rename() iop and rehashes it after,
but doesn't account for the possibility that rename() may have swapped
{old,new}_dentry.  For FS_RENAME_DOES_D_MOVE filesystems, it rehashes
new_dentry (now the old renamed-from name, which d_move() expected to go
away), such that a subsequent lookup will find it.  Currently all
FS_RENAME_DOES_D_MOVE filesystems compensate for this by failing in
d_revalidate.

The bug was introduced by: commit 349457ccf2592c14bdf13b6706170ae2e94931b1
"[PATCH] Allow file systems to manually d_move() inside of ->rename()"

Fix by not rehashing the new dentry.  Rehashing used to be needed by
d_move() but isn't anymore.

Reported-by: Sage Weil <sage@newdream.net>
Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/namei.c