Overall I think it is wonderful to see support for read-only root being worked on.
On Wed, 2003-04-09 at 15:41, Matthew Garrett wrote: > Jeremy Jackson wrote: > > (doing this with bind mounts) > > >2.2 kernels are out though. > > As are BSDs. I have no idea whether the Hurd supports bind mounting. > Can it be assumed that all systems that may use this support ramdisks? What other schemes would allow a read-only root? Mounting a small fs on /run (although I hope it's not in the root directory)? > In most cases, just moving the files somewhere more sensible like > /var/run is sensible. The only argument should be over files that need > to be written before /var has been mounted, of which there should be a > very small number. It should be possible to deal with this in a simple > fashion. The technique involving bind mounts can be done with a regular mount, it would just take a few more steps: Mount ramdisk or whatever other rw fs on /etc/run very early in boot. Whatever needs to scribble in it does so, and everything works. If later the root FS is mounted rw, copy contents of /run to temp dir, unmount /etc/run, clear it out, then copy it all back. I think it's the least invasive technique, although I agree that some things that are messing about in /etc and can do it in /var should be fixed. -- Jeremy Jackson <[EMAIL PROTECTED]>
