Date: Thu, 18 Nov 1999 20:54:43 -0800 (PST) From: Matthew Dillon <dillon@apollo.backplane.com> To: David G Andersen <danderse@cs.utah.edu> Cc: barrett@phoenix.aye.net (Barrett Richardson), wes@softweyr.com, kris@hub.freebsd.org, trouble@netquick.net, danderse@cs.utah.edu, freebsd-security@FreeBSD.ORG Subject: Re: secure filesystem wiping Message-ID: <199911190454.UAA90603@apollo.backplane.com> References: <199911190446.VAA22206@faith.cs.utah.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
:Not if you do it right. Do it in blocks, and use madvise
:(try MADV_SEQUENTIAL on the whole region, or simply MADV_DONTNEED each
:block after you're done). I'd wager that MADV_SEQUENTIAL will give you
:exactly the behavior you're looking for if you do something like:
:
:Blocks in terms of pagesize would be particularly appropriate, yes. :)
:...
: -Dave
:
:--
MADV_SEQUENTIAL will definitely work. In fact, if you scan the file
sequentially you do not even need to call madvise(), it will figure it
out and do the right thing.
But there is no guarentee that the system will flush the pages to disk
prior to removal of the file, and removing the file may cancel the
pending I/O. Maybe if you msync() the area and then fsync() the file
prior to closing it.
I would simply use a write() loop, then fsync() and close() the file
when done. For this type of scribbling write() will be much faster
then mmap(). Don't bother with mmap() at all.
Both the VFS interface (read and write) and the mmap interface will
properly handle write-behind and reuse pages without loading the system
too badly. Current does a somewhat better job at it due to tuning work
and a change in the way the file position is cached in regards to
determining the file mode.
-Matt
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-security" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199911190454.UAA90603>