Greg Troxel <g...@lexort.com> writes: > Brad Spencer <b...@anduin.eldar.org> writes: > >>> I'm now thinking, would it make sense to do the layering the other >>> way around, i.e. have cgd on top of a zvol? I wonder if there would >>> be any resilience (and possibly performance) advantage to having zfs >>> directly access the hard drives (which, being old and bought off >>> ebay, I don't quite trust) rather than go through cgd. >> >> To be honest about it, I don't remember if I tested in that direction. >> It would be simple enough to try. You just might have the same problem >> with missing IOCTLs that keeps swapping from working. > > If you put cgd on zvol, then what filesystem are you putting in the cgd? > Another zfs pool? ufs2? To me, zfs on cgd makes more sense, because > then you actually get zfs, and can have raidz2 or whatever, but all > media is encrypted.
I was speaking only of the attempt to do this, not the reason why someone might want to do this. As to what you may want to use.. I honestly would hope that anything would be possible with stackable filesystems.... As you suggest the question of why and what?? Well... one I can think of is to make a portable, but fully encrypted, container that would allow for 'zfs export' to be a trival thing or at the very least allow you to write out just a part of a pool or drive and be assured that it is encrypted and contained. I do something like this a bit between MacOS, NetBSD and sometimes FreeBSD, of course without encryption (I do understand that cgd isn't avaliable everywhere). Of course, you could also put a FFS on it. I have tested FFS on a zvol and it works, but I didn't try cgd, but it isn't hard to imagine layering in encryption too. But I do agree with you that zfs on cgd probably makes the most sense in the common case if you don't have to move anything around too much or slice out a contained portion of the storage. -- Brad Spencer - b...@anduin.eldar.org - KC8VKS - http://anduin.eldar.org
