Re: md /tmp and async mounts

On 05/31/06 20:20, Giorgos Keramidas wrote:
> The following diff should be sufficient for enabling -o async for
> the MFS /tmp and MFS /var filesystems we support in our current
> rc.d stuff:
> 
> %%%
> Index: share/man/man5/rc.conf.5
> ===================================================================
> --- share/man/man5/rc.conf.5	(revision 99)
> +++ share/man/man5/rc.conf.5	(revision 101)
> _at__at_ -24,7 +24,7 _at__at_
>  .\"
>  .\" $FreeBSD: src/share/man/man5/rc.conf.5,v 1.298 2006/05/30 16:20:48 matteo Exp $
>  .\"
> -.Dd May 29, 2006
> +.Dd Jun  1, 2006
>  .Dt RC.CONF 5
>  .Os
>  .Sh NAME
> _at__at_ -241,12 +241,13 _at__at_
>  .Pa /tmp
>  is created.
>  The default is
> -.Dq Li "-S -M" ,
> +.Dq Li "-S -M -o async" ,
>  which inhibits the use of softupdates on
>  .Pa /tmp
>  to waste as little space as possible
> -and creates a pure memory backed disk, which will never be swapped out,
> -for maximum performance and system stability at low memory conditions.
> +and creates a pure memory-backed disk, which uses asynchronous I/O
> +and will never be swapped out, for maximum performance and system
> +stability at low memory conditions.

Given recent posts about panics with malloc-backed mds under low-memory
conditions (see the "kmem leak in tmpmfs?" thread on stable_at_), I'm not
sure keeping '-M' as the default for tmpmfs and varmfs is advisable.
Here is the rather ominous note on using malloc-backed mds in mdconfig(8):
"Storage for this type of memory disk is allocated with malloc(9).  This
limits the size to the malloc bucket limit in the kernel.  If the -o
reserve option is not set, creating and filling a large malloc-backed
memory disk is a very easy way to panic a system."

-Jonathan

-- 
Jonathan Noack | noackjr_at_alumni.rice.edu | OpenPGP: 0x991D8195