Messages in this thread | | | From | "Jeff V. Merkey" <> | Subject | Re: FENRIS & 2.0/2.2 problems | Date | Wed, 23 Jun 1999 20:22:01 -0600 |
| |
One of the great things about the Linux Community is that everyone has a great perspective and gets to voice it. These are some excellent comments.
Jeff
-----Original Message----- From: Peter Waltenberg <peterw@dascom.com> To: Alexander Viro <viro@math.psu.edu> Cc: alan@lxorguk.ukuu.org.uk <alan@lxorguk.ukuu.org.uk>; Jeff Merkey <jmerkey@timpanogas.com> Date: Wednesday, June 23, 1999 8:19 PM Subject: Re: FENRIS & 2.0/2.2 problems
> >On 24-Jun-99 Alexander Viro wrote: >> >> >> On Wed, 23 Jun 1999, Jeff Merkey wrote: >> >>> >>> Peter, >>> >>> David Goebel and I have been discussing writing a "universal" interface that >>> will allow linux file systems and NT filesystems to be used interchangably >>> between the two platforms. The benfits to Windows NT are obvious, and David >> >> No. Not because of anti-Microsoft idiocy, but because that will introduce >> a lot of cruft into the system. If NT folks want to do it - it's their >> kernel and their choice. As far as I'm concerned fixing the common >> interface for two *very* different kernels is idiocy. If anything, we >> might look for unification with other Unices, but even that is unfeasible. >> Too different VFS designs. Linux is unique in that respect (post-2.0 >> branches). Linux interface will change - some things should be done to fix >> NFS interaction with braindead (inodeless) filesystems, etc. Linux VFS is >> *good* thing, but it lacks some very nice stuff (stackable filesystems, >> etc.) It *will* change. Not because somebody wants to break the stuff >> left, right and center - there is more than enough in the main tree to >> make such changes PITA for ourselves. But there are legitimate reasons for >> changes. Sometimes it's inevitable. >> >>> be beneficial for both platforms moving forward. An unchanging "universal" >>> file system interface would be very hot in Linux. >> >> Yes. As in "red-hot soldering iron in the place where the Sun never >> shines". The only reasonable way, IMO, being to write a proxy fs that >> would look as foo filesystem from one side and as bar VFS from another >> (and the pair to it - other way round). Notice that native filesystems >> would not use such thing. Said that, I'm pretty sceptical on NT future. >> Yes, UNIX bigot. So sue me. > >We aren't criticizing the VFS design, or the fact that it changes. >What we are saying is that it'd be nice if there was an interface layer >between the whatever the vfs of the day was and "external" filesystems >which provided an alternate filesystem layer with a consistant interface. > >The *ONLY* downside I can see with this is that it means that there isn't >the pressure to update all that older code. Thats a major downside, and >Alan probably has a better idea than I do of how many problems that would >cause (or solve). > >There are certainly some filesystems where the vfs changes have been >nightmarish to deal with. MSDOS/UMSDOS for example. Performance isn't an >issue for "legacy" filesystems. Having to find someone who can update them >when they break and where only a few % of Linux users need them is. > >I've helped break some more *cough* er fix a couple of filesystems going >from 2.0->2.2 and it isn't fun. I doubt anyone would care much if the UMSDOS >filesystem sucked speedwise, it does anyway. But it's taken about a year to >get that anything like functional again. > >That "proxy" filesystem is pretty much what I'm suggesting, someone writes >support to allow fs "X" to work with that once, and it stays working forever >rather than being broken for a year after every internal change. > >If someone needs performance, fine, they can write a version which works with >the "real kernel interface", but there'd be a fallback available for when >the next upgrade broke everything. > >Things like MSDOS, UFS, MINIX, probably even NTFS largely fall into this >category. >It's really nice to be able to interoperate with (i.e. read and write) those >filesystems, and performance is a secondary issue compared with it just >working. > > > >Peter > > > >---------------------------------- >E-Mail: Peter Waltenberg <peterw@surf.dascom.com> >Date: 24-Jun-99 >Time: 11:55:45 > >This message was sent by XFMail >---------------------------------- >
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |