Date: Wed, 20 Mar 1996 15:33:17 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: julian@ref.tfs.com (JULIAN Elischer) Cc: terry@lambert.org, scrappy@ki.net, current@FreeBSD.ORG Subject: Re: DEVFS vs "regular /dev" Message-ID: <199603202233.PAA27947@phaeton.artisoft.com> In-Reply-To: <199603202100.NAA17319@ref.tfs.com> from "JULIAN Elischer" at Mar 20, 96 01:00:45 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> all good but one question..... > > > > > > Just curious, but what is the advantage (and disadvantage, if any?) > > > of using devfs vs "regular /dev"? My first thought would be that its > > > memory based instead of file system based, giving better performance, but > > > that's just a guess. > > > [.... tons of reasons ] > > > > > This assumes that the file system abstractions currently in place > > change as well, since the /dev FS can't be mounted *after* the / FS > > has been mounted as an inferior FS -- > why not? > you don't need a mounted /dev to mount root. > that's done specially. So you don't need a mounted root to have a mounted /dev, of course! This makes it possible to remount root r/w without unmounting the devfs (and so still needing /dev). > > the /dev has to be there as > > an overlay mount (translucent FS), and that means FS changes to allow > > translucence and to divorce volume mappings from the need for a mount > > point (basically, a shadow / and /dev on which the devfs /dev is mounted > > and the real / is mounted over top of the shadow /). > > I think that's not needed Root remount, again. If you divorce the mappings, you can mount root over and over despite devfs being mounted by default. The other alternative is to put the device files in a POSIX name space escape. I felt this would be too "un-UNIX-like"; can you imagine having no /dev directory and using "//dev/..." to access devices? That's pretty much the alternative. It would mean you couldn't have a machine named "dev" if you used POSIX escapes for a network namespace (unless you shoved it down to "//net/dev", of course...). If you want to get more specific, it also lets us build a disk image in the kernel data segment for "sh" and/or "init" to get up with no disk and no NFS capability during a port. It would be nice to not have to have working drivers to run. 8-). Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199603202233.PAA27947>