fs oddity when moving from 4.4 to 5.3

[Doug White]

On Fri, 4 Feb 2005, Jan Pechanec wrote:

>
> 	hi, after RW mounting UFS1 filesystem created using sysinstall under
> 4.4 to 5.3 and extracting 1/2 of the only file there (tarball), the file got
> lost. Now I got that disk to salvage as much as I can. I don't want to give
> much details since I think it would be too much, but is there any chance
> that some changes in the code could do that? Fsck doesn't complain, the
> directory chunk for the file is now part of larger free chunk etc., but the
> file was not manually deleted, bash history was checked.
>
> 	just one thing: 'dumpfs -m' on 5.3 and then 'newfs -N...' shows:
>
> /dev/ad1s1e: 38158.3MB (78148160 sectors) block size 8192, fragment size 1024
>         using 846 cylinder groups of 45.15MB, 5779 blks, 11584 inodes.
> super-block backups (for fsck -b #) at:
>  32, 92496, 184960, 277424, 369888, 462352, 554816, 647280, 739744, 832208,
>
> 	the same 'newfs -N' (without '-O 1') on 4.11 shows this:
>
> /dev/ad1s1e:    78148160 sectors in 19080 cylinders of 1 tracks, 4096 sectors
>         38158.3MB in 868 cyl groups (22 c/g, 44.00MB/g, 10944 i/g)
> super-block backups (for fsck -b #) at:
>  32, 90144, 180256, 270368, 360480, 450592, 540704, 630816, 720928, 811040,

Its possible between 4.4 and 4.11 (I haven't checked the CVS history) the
newfs default for the '-g' option, which sets the cylinder group size,
changed from the old BSD default to "the max possible" which is generally
better for performance. You would need the original FS parameters to feed
to newfs on 4.11 in order to calculate the correct superblock positions.
Or you could fish the fsck binary off a 4.4 cd 2 and run that to do the
calculation.

-- 
Doug White                    |  FreeBSD: The Power to Serve
dwhite at gumbysoft.com          |  www.FreeBSD.org