Ben Hutchings: > I think it doesn't matter when the last reference is from a running > system call and not a vma. There is no way that the process could use > the aufs file structure at this point.
Basically you are right and I was wrong. The unbalance of the ref-count between vm_file and vm_prfile will not be a problem here, even there is a small window between releasing and re-acquiring mmap_sem in msync and madvise. *IF* msync and madvise touch vma or vm_prfile then it would be a problem. I will merge your patch and release after completing my local tests. Thank you. J. R. Okajima ------------------------------------------------------------------------------