Subject: Re: Partial fix for the NetBSD VM problem
To: Greg Hudson <>
From: Phil Knaack <>
List: tech-kern
Date: 03/04/1996 14:16:21
>After this patch, the following problems will remain:

>	* If you modify a file through the filesystem, the changes
>	  will not necessarily affect mmap()'d regions until an
>	  msync().

>	* If you modify an mmap()'d region with MAP_SHARE, the changes
>	  will not necessarily affect the filesystem until an msync()
>	  or munmap().

>   [ vm_mmap.c patch ]

	Just at a cursory glance, and with the understanding that I don't
know a heck of a lot about what POSIX.1 says, it seems to me that the second
point is one which seems more than reasonable, and not a bug.

	By the same token, the first one seems almost *correct* to me.

	Perhaps I'm thinking in a different paradigm, but it seems to me 
to be dangerous to "assume" that these two problems do not exist in any 
given system a package you're designing may be run on.

	It just seems to me as a design _flaw_ for a design solution to
use _both_ read/write and mmap on the same file. Sort of a lack of 
purity of design thing.

	My 2 pfennig.

Phillip F Knaack     
Database Programmer, NCREMP    Student Development Group
ISU Extension                  Project Vincent, Iowa State University