Date: Tue, 22 Oct 1996 14:49:20 -0700 From: Julian Elischer <julian@whistle.com> To: Julian Assange <proff@suburbia.net> Cc: hackers@FreeBSD.org Subject: Re: comments on this change please. Message-ID: <326D4160.2781E494@whistle.com> References: <199610221746.DAA00278@suburbia.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Julian Assange wrote: > > > please comment. > > I'm trying to get a production r/o root system out the door > > and we have to get around syslog somehow.. > > > > note that this system also works with devfs because devfs > > supports symlinks. > > > > julian > > Speaking of root only file systems, it would be nice to have a static > root file system - that is one that never needs to be modified i.e > over-ride hooks for various config files in /usr/local/etc, or a decent > union over-ride. yep. what we have is: /var/config (or equiv) with the real modifiable files files in /etc point there. before mounting /var, the directory /var/config contains some skeleton versions of the files to make the system useable in single user mode. > > Is anyone using devfs as a real /dev? It seems to me that devfs is flawed for > /dev replacement, insofar as the non-permanency of permissions/other inode > information. there are plans to provide this using the stackable vnodes feature. In the meanwhile till that's completed, 99.8% of users never change (by default)the permissions of a device. and .1% who do would be happy adding it to /etc/rc leaving 0.1% of crazed fanatics who wouldn't use devfs in it's present half completed state because of this. (my biased opinion) The design includes support for attribute persistance. but only a maniac would insist on it. It's more sucure to not have it in my opinion, (I'm of course aware that I'm in the minority :) > imho, the desired structure is to have symlinks in /dev pointing > to /devfs, with the permissions copied off the *symlinks* and into /devs at > boot (either that or symlinks that actually enforce their permissions) oh we can do much better than that with the current stackable vnode implimentation. And anyhow, remember that under BSD4.4 ffs, symlinks don't HAVE permissions and owners etc. because they don't have inodes.. oh yeah, YUK! that's the way sun did it..! SGI on the other hand I hear are doing it my way and have thrown away persisntance.. a wise move I think.. eventually I want ot to be illegal to create device nodes on a normal filesystem anyhow.. filesystems are persistant and devices are not. especially with major numbers being dynamically assigned.. next time you boot with a differnt set of peripherals, your tty device suddenly has the major number for the tape drive.. what fun.. remember that both were LKMs supplied by 3rd party developers. Thats what we are eventually aiming for. Totally dynamic configuration and linking. (in fact hopefully, eventually there won't BE major numbers so you won't be able to make dev nodes on persistant filesystems anyhow) >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?326D4160.2781E494>