On Fri, May 05, 2000 at 12:38:26AM +0200, David Weinehall wrote: > > there is no permission handling, not without really ugly kludges. > > I don't consider the DevFSd to be a horrible kludge; in fact, for > hot-pluggable devices, such a solution in necessary anyway. i mean things like chmod/chown are broken on devfs, well they work but the change is not persistent. instead you have to edit a massive conf file. and IMNSHO storing permissions in a ordinary file is broken. there is talk of making devfs into a union mount so if permissions are changed a real node is created in the filesystem which shows through no matter what. this seems to be a generally acceptable solution but then you just end up with a /dev/ full of real nodes anyway so why bother with the fake filesystem? > Yes, I can agree with this one. This was however something Linus brought > upon us. The namespace Richard Gooch originally created was imho pretty > perfect. iirc Richard pretty much left the current namespace intact, with a few tweaks here and there. i don't know what linus' problem is with the current namespace is. this new bastardization is pure evil, its gratuitously incompatible and breaks everything unless you run devfsd to create a huge symlink farm to simulate the old one! talk about a mess! how many times have you heard "i turned on devfs and now my system is totally fscked!!" why? because /dev/tty* are gone and getty/init get pissed. i don't see what is SOOOO broken about the current namespace to warrent such a gratuitously incompatible change that WILL break literally everything. > > If I'm not all wrong, Alexander Viro and Richard are working together on > solving this particular one. more like flaming each other... > > Because you don't want to create a couple of 100 extra device-nodes. i don't buy this "1000+ nodes in dev is soo horrible!" thing at all: [eb@plato eb]$ ls -l /usr/bin/ | wc -l 2105 [eb@plato eb]$ oh my i have more then twice as many files in /usr/bin we better move to a MS filesystem layout immediatly! i agree that 1394, usb fibrchannel etc would be out of hand but for all the stuff we have had forever, hard disks, ttys ie whats you see in /dev/ now is not a problem. > And as I said, where not talking static hardware. We're talking > hot-plugable hardware. USB/Firewire/Fibrechannel/PCMCIA/PCI-hotplug > can all be hotplugged. If I'm not all wrong, Fibrechannel can handle > 65535 devices on the bus. so the problem is we don't want 65000 files in /dev, that is reasonable (that would close to exhaust the available inodes). some want nodes created for them (i personally hate that) why exactly does there ahve to be a fake filesystem for that? you still get stuck wtih a userland daemon doing everything anyway so why not use a userland daemon with real device nodes? chroot problems go away, permissions problems go away. the only thing left is this retarded NT style namespace. anyway, i could learn to live with the fake filesystem provided chroot() is CLEANLY taken care of, and permissions are persistent as they are now (i run chmod 600 /dev/scd0 and that change sticks, no fscking with a devfs conf file, and no tar nonsense at shutdown). but the main fatal thing wtih devfs that would make me switch OSes before using it is that crappy namespace, that is the one thing about devfs i WILL NOT accept, ever. -- Ethan Benson http://www.alaska.net/~erbenson/
Attachment:
pgp6g2m1X8VON.pgp
Description: PGP signature