Re: ????mux and firmlink

[Neal H Walfield]

>  What I'm wondering is how to set a translator on a file when
> the file is created.

You would need to write a translator that acts when a file is
created.  For instance for a libdiskfs based translator, you would do
this in diskfs_lookup when TYPE is CREATE.

> So here is an example /hurd/crash could dump
> into /var/crash, but a mux translator would be set on /var/crash so
> a dump of the proc server would be /var/crash/core-proc but 
> core-proc would be a multiple firmlink to
> /var/db/crash/core-proc-<timestamp> .  This would also be the same
> sort of setup as /var/log so /var/log/gopherd.log would be firmlink
> to /var/db/log/gopherd.log-<timestamp> .  Setting a system up this
> way seems to obsolete logrotate.

No.  Consider what happens when you mv a file which processes have
open: the inode stays the same and any users continue using it is if
nothing happened.  Only new users (i.e. those who call open after the
mv) will see that the file no longer exists or has been replaced,
etc.

>  More mux rant: Would a use of usermux be for /proc, in that
> usermux is set on /proc starting a procfs for each user, so
> /proc/jim contains a procfs of only my processes. Does the user
> actually have to exist or could the procfs assume if the user doesn't
> and is called system or GNU_Hurd_rulz or whatever, that it should act
> like a system wide procfs like neals procfs acts now?

Right now if the user does not exist, ENOENT is returned.