Re: CVS commit: src/sys/kern

[Andrew Doran <ad%netbsd.org@localhost>]

On Wed, Sep 24, 2008 at 01:46:34PM -0400, Thor Lancelot Simon wrote:

> On Wed, Sep 24, 2008 at 07:38:20PM +0700, Robert Elz wrote:
> >     Date:        Wed, 24 Sep 2008 10:07:19 +0000 (UTC)
> >     From:        Andrew Doran <ad%NetBSD.org@localhost>
> >     Message-ID:  <20080924100719.D78AE175D0%cvs.netbsd.org@localhost>
> > 
> >   | Prevent r/w to r/o downgrade until such time as someone has verified all
> >   | the relevant file system code.
> > 
> > Please don't do that - I do rw -> ro "down"grades (I'd call it an 
> > upgrade...)
> > all the time.   As long as you know what's up (that is, to make sure there
> > is no possibility of stale data) it is fine, and causes no problems.
> 
> This also breaks products which run from flash with their filesystems mounted
> read-only most of the time, but remount them read-write in order to write
> back configuration data when it changes.  That is not an uncommon model
> AFAIK and it poses a fairly major problem for me.
> 
> If anyone can give me an idea what actually has to be verified, I can (since
> I have to) put considerable time into it.

It would mean checking all uses of MNT_RDONLY, and thinking about what
happens if the flag changes while an operation is in-flight, and afterwards
if dirty buffers/pages have been generated. There might be barriers to
prevent flushing those after the flag is set. I don't know.

Andrew