cvs commit: src/sbin/mdconfig mdconfig.8 mdconfig.c

[Kris Kennaway]

On Wed, Feb 21, 2007 at 11:13:09AM +0000, Robert Watson wrote:
> 
> On Wed, 21 Feb 2007, John Baldwin wrote:
> 
> >On Wednesday 21 February 2007 05:36, Robert Watson wrote:
> >>
> >>On Tue, 20 Feb 2007, Nick Hibma wrote:
> >>
> >>>n_hibma     2007-02-20 22:04:23 UTC
> >>>
> >>> FreeBSD src repository
> >>>
> >>> Modified files:
> >>>   sbin/mdconfig        mdconfig.8 mdconfig.c
> >>> Log:
> >>> Kris suggested that swap is a better choice as a default than malloc.
> >>>
> >>> MFC:    1 week
> >>
> >>Are you sure this should be MFC'd?  While I agree with the change in 
> >>principle, it will cause systems currently using malloc-backed md devices 
> >>to mysteriously switch to swap-backed.  How will this behave on systems 
> >>that don't have swap, such as pure diskless systems not using NFS-backed 
> >>swap? This isn't a strong objection per se, but it could be that rc.d 
> >>scripts need tweaking for diskless environments if they currently do use 
> >>malloc-back md devices based on that being the default.
> >
> >I don't think this changes the default for anything in 6.x.  He changed 
> >the default for his patch from yesterday that made it assume a type if you 
> >do 'mdconfig -s <size>'.  Currently you have to do:
> >
> >'mdconfig -a -t malloc -s <size>' or 'mdconfig -a -t swap -s <size>'
> >
> >but Nick changed it so just '-s <size>' implied '-a -t malloc', and now 
> >he's changed it to imply '-a -t swap'.
> 
> Ah, OK.  This sounds reasonable.

Anyway, swap-backing can't really have a worse failure mode than
malloc backing in the event that backing store becomes unavailable
(e.g. no swap is configured and ram is full): malloc will already
panic in that situation.

Kris