Josef Sipek:
> Strange. Which kernel version are you using? Does it happen on shm only
> or any fs do this?

2.6.14.6.
all branches are loopback mounted ext2fs image.
I have tried real ext2 fs and got same result(rmdir failed).

As you read my patch, I am thinking the cause is in delete_whiteouts().
- writable_branch/a/b dir is owned by root and it is writable by owner
  only.
- there exists writable_branch/a/b/.wh.__dir_opaque. this entry can be
  deleted by the owner of the dir. i.e. root.
- current delete_whiteouts() doesn't modify current->fsuid and it should
  fail removing .wh.__dir_opaque.
- current unionfs_rmdir_all() calls delete_whiteouts() and ignores the
  return value (it might be another bug), and then calls vfs_rmdir(). it
  must fail removing the dir since .wh.__dir_opaque still exists.

If you succeed rmdir following the steps in my mail, my thoughts is
totally wrong.

Junjiro Okajima
_______________________________________________
unionfs mailing list
[email protected]
http://www.fsl.cs.sunysb.edu/mailman/listinfo/unionfs

Reply via email to