Hi all, this is really quite an old issue which I came across a lot of times in many different scenarios.
Having read this thread, my feeling is that what is being suggested right now would solve the lofs of UFS and lofs of zfs issues, but what about... ... a lofs mount of an nfs mount? ... a lofs mount of something on a metaset ... a lofs mountpoint in an nfs mountpoint in a zfs? This list is not complete at all, but, IMHO, the point really is that one could think of arbitrary dependencies between filesystems which are being mounted a different times during bootup. So adding yet another layer by delaying lfos mounts will probably solve this issue, but not the general case. Once we are at it and having a proper framework to declare dependencies at hand (named SMF ;-), why don't we use it to solve this "once and for all"? Here's my suggestion: * Make all these mount-some-filesystem methods (like fs-local and nfs-client) configurable with respect to which vfstab to use, which fstypes to mount, which mountpoints to mount etc. * Move the complexity to resolve dependencies to SMF itself by creating multiple instances of the "new general mount method" with proper dependencies between them * What is fs-local, nfs-client etc. now, could become milestones What remains is the task to configure the required dependencies, but I am pretty sure that this is something we can achieve with some vfstab parser. Nils This message posted from opensolaris.org
