[Soekris] (no subject)

[Chuck Yerkes]

Quoting Shawn Mitchell (shawnm at iodamedia.net):
> Stuart Henderson wrote:
> > > Currently I am experimenting with mounting an mfs filesystem and then
> > > union mounting it /var/log on it.  If this works as I think it does,
> > > this means that /var/log files are stored in RAM (as well as an external
> > > syslog server)
> > 
> > OpenBSD 3.5 syslog can use RAM buffers for logging, which is useful
> > for this environment.

If it's RAM, why write to it at all?
Me? I log to another box, but I also syslog to /var/run/everything
(an MFS).  Newsyslog rotates it (1 generation) at 1MB.  Which is
enough for me to glance and see what's just happened recently for
minor nits and debugging.  Real logging goes elsewhere.

> > ISTR changes to tty handling and wondered if that
> > could mean that /dev can now be mounted RO too, but haven't had time
> > to try that yet...

> you mount /dev RO, nothing will work.  How else can it write output to the
> serial port?

By using it to call the routines in the kernel that are really
behind this pseudo "file" in /dev?

in other words:  works for me.


I changed libutil to move /dev/ptyp* to /dev/term/ptyp*  (and ttyp*).
It was pretty trivial.  Meant I could make /dev/term/ an MFS with /dev/
readonly.  I've changed since to just making /dev/ an MFS, but still
keep terminals in /dev/term.

What Sun did with their dev tree was wrong and evil and broken.
But the NOTION of dividing /dev/ a bit into functions makes sense.
Even just /dev/term/ (serial and terms) and /dev/disk/ (raid, disks,
floppies)  addresses around half of over 2000 devices I get (sorry,
no default install turned on right now).

When "ls $DIR" becomes useless, I look to moving it a bit into logical
divisions.  That ptyp/ttyp is only referenced in libutil made that one
easy to change.