Russell Coker writes:
> This one was first discovered on my SE Linux play box, but has been
> reproduced on normal machines too.
> Log into a devfs-enabled box remotely with ssh, or with gnome-terminal,
> or possibly others (haven't checked)
>
> Do not issue any comands in that shell
>
> Open another shell, and look at "w" output. Idle time is absurd.
>
> use stat on the pts from the first shell, notice the atime is 1970.
OK, that's because the atime is initialised to zero (the Epoch,
aka. 1970). This is the case for all device nodes in devfs, and is a
convenient way of seeing if something has been changed from the
default. The old "tar kludge" method of saving permissions depends on
this (well, it looks at the ctime, not the atime).
> People say this doesn't happen with devpts.
I guess devpts initialises the atime to the current time when the
device node is created. So devpts pretends that the user was active in
that login when the login was created. That's not strictly true, since
the user was never active in that terminal.
Devfs is more honest, and shows (effectively) that the user was never
active in that login. Which is true!
Either way, does it really matter? I could change this, but it would
require more code, and in some ways, what is there now is more
"correct".
Regards,
Richard....
Permanent: rgooch@xxxxxxxxxxxxx
Current: rgooch@xxxxxxxxxxxxxxx
|