Nicolas Williams wrote: >> If we don't need to do the final sync before turning off the power, or >> if we can arrange for userspace involvement after the final sync, then I >> would hope that we could do all of the sequencing using SMF dependencies >> rather than by putting additional hooks into svc.startd. > > Well, yes; that implies strict shutdown ordering, at least for the last > one or two services.
Well, sure. Isn't sequencing startup and shutdown SMF's job? Ceri Davies wrote: > Special casing for the root file system is undesirable; requiring a ZFS > root just because of this special case would be a travesty, IMHO. I haven't thought about the question deeply, but it seems that there are many reasons for requiring a ZFS root, not the least of which is avoiding the need to ask the user which file system to use for the root. > As others have suggested, downgrading / to read-only sounds like the > best bet. Solving the question of what the hook mechanism would > actually look like might be the issue now. One presupposes that a > "magic" FMRI is out of the question? When possible, magic should be avoided. SMF is responsible for sequencing startup and shutdown; we should try to use it for that purpose.
