TMPFS in kernels

[Warner Losh]

On Mar 2, 2014, at 11:11 PM, John Hay <jhay at meraka.org.za> wrote:

> On Sun, Mar 02, 2014 at 11:13:31PM -0500, George Rosamond wrote:
>> Ian Lepore:
>>> On Sun, 2014-03-02 at 19:46 -0500, George Rosamond wrote:
>>>> Is there a reason why TMPFS(5) is not in the 10-Stable's ALIX and RPI-B
>>>> kernels, yet in BEAGLEBONE?  Haven't checked the others.
>>>> 
>>>> Seems like a strong and mature replacement for md(4) from my
>>>> experiences, and probably belongs in the Crochet fstabs.  I have been
>>>> using on Alix, RPis, x86 for a long while, and never had any
>>>> (noticeable) issues.
>>>> 
>>>> g
>>> 
>>> I agree.  Should we add it to all configs, or just to armv6 configs?
>>> 
>> 
>> I'd assume we should have it all the kernels.. especially since most of
>> them are running off flash-type media, which makes memory-based disks
>> pretty vital.
> 
> Our embedded systems also use compact flash disks mounted ro. We have
> been (ab)using the nice etc/rc.initdiskless and etc/rc.d/var scripts
> by just having an etc/diskless and populated conf/base and conf/defaults
> directories. But it seems those scripts just work with md devices.

NANOBSD works that way too… tmpfs is a bit nicer since it doesn’t hard-wire the allocation,
but that can suffer a bit from the vagaries of system demand….

Warner