Here is a patch that fixes a rather nasty bug in hugetlbfs.  If root
tries to mount a hugetlbfs with invalid mount option,
hugetlbfs_fill_super incorrectly returns error code instead of NULL on
couple error conditions, causing generic vfs layer to segfault and took
some semaphore with it, the end result is kernel hang on any subsequent
file system mount/unmount.  There are some other cases that has memory
leak upon error conditions.

The first problem is 2.4 specific, while memleak problem exist in both
2.4 and 2.6.
Bjorn, again, would you please merge in your 2.4 tree?  I'll push the
rest for 2.6.

- Ken