On 2/5/12 7:54 PM, Dave Chinner wrote: > On Sun, Feb 05, 2012 at 06:13:28PM -0600, Eric Sandeen wrote: >> On 2/5/12 6:11 PM, Dave Chinner wrote: >>> On Fri, Feb 03, 2012 at 12:50:17PM -0600, Eric Sandeen wrote: >>>> On 2/3/12 8:39 AM, Marcos Mello wrote: >>>>> Eric Sandeen <sandeen <at> sandeen.net> writes: >>>>> >>>>>> >>>>>> In general there is no problem with xfs on a root partition. However, the >>>>> installer >>>>>> may not make it easy or available for you. >>>>>> >>>>>> (I never use xfs for /boot though, I don't trust grub enough for that >>>>> honestly). >>>>>> >>>>>> -Eric >>>>> >>>>> Same thing on Fedora 16. Let's hope some day Anaconda will change that. >>>> >>>> F16 prevents it? I didn't see it in the upstream tree. That should >>>> not be so. :/ >>>> >>>>> About GRUB with a XFS /boot the problem was with GRUB Lagacy, wasn't it? >>>>> Or GRUB2 is still buggy? >>>> >>>> I have no idea, actually. I delved into grub a bit, it was disturbing >>>> enough that I have not tried to look at grub2. :) >>> >>> Certainly the problem exists with legacy grub - it assumes that it >>> can write to the first sector or any disk or partition which >>> overwrites the XFS superblock... >> >> well, it was worse than that. I can work around the grub-on-a-partition >> problem, but what I ran into was grub reading & writing to/from the block >> device under a mounted filesystem - corruption and hilarity ensued. > > Oh, grub2 does worse things than that - grub_probe mounts any device > that appears to have a valid superblock so it can find out what > devices it needs to list in it's boot menu. I found this out when I > built a big DM stripe out of disks that used to have filesystems on > them. grub-probe mounted every single device -inside- the stripe > and ran log recovery on them, corrupting the new filesystem and > quite a bit of data on the DM stripe they were part of.... Ok, your horror story wins from a pain POV ;) but sounds like someone should have zeroed out a bit more disk when setting up the dm stripes :( I've always thought maybe we needed a libzero.so to zap every known signature on disk.... -Eric > Cheers, > > Dave. _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
