On Sat, May 13, 2000 at 10:24:22PM +0200, Norbert Nemec wrote: > OK, that would basically be a way for read-only shadowfs. If that's all we > aim at, there should not be a problem for the specifications. Just layer the > whole tree (ignoring contents of non-readable directories) If a file exists > in more than one versions, choose the "topmost" one. Everything beyond that > (like filename patterns etc.) would be additional features that do not > change the basic concept. > > Question is, whether read-only is the only thing we want to go for. It would > be absolutly neat, if we had a way to layer filesystems in a absolutely > transparent way, so one could do things like have a read-only system as base > (maybe NFS-shared by several computers?) and a read-write system on top that > takes any changes you make and perhaps copies of frequently used files for > better performance.
When we'll have r/o shadow, it won't be very hard to make it simple r/w. At first, there could be 1 r/w + 1 or more r/o subfs. This is easy for modifying files from r/w fs and creating new files. You only have to create subdirs that exists on one of r/o's and not yet on r/w, and forward all calls to r/w fs. More problems arise with modifying files that exist only on r/o, deleting files, renaming, etc. > Imagine a pre-installed system distributed on DVD (several GB in size) that > you can use just as any normal system, since it writes only the changes on a > fairly small partition (or even a ramdisk) Problems with ramdisk is that some stuff /etc/* /var/mail/* should be available between reboots, so such system would need NFS or smal hd/zip/... to do anything useful.
