Let's go with Nathan's hunch - what version of the kernel are you using?
the BLKGETSIZE64 iocttl was broken for a while, and returned bad values.
That could be causing your problem...
THe default internal log grows with the size of the fs, to a point. But
aside from some overall min/max log sizes (probably documented in the
mkfs.xfs man page) you can have a wide range of log sizes for a given fs.
Log v1 is still default, logv2 could probably use more testing before
it becomes the default.
On Fri, 21 Feb 2003, James Rich wrote:
> On Thu, 20 Feb 2003, Eric Sandeen wrote:
> > On Thu, 2003-02-20 at 20:32, James Rich wrote:
> > > what should I do?
> > Please try a newer xfsprogs version, for starters - 2.0.3 is pretty old
> > by now. 2.3.9 is on the ftp site now, or grab a slightly older version
> > from the 1.2 release.
> I did so and found that a problem exists when I specify the size as in:
> size=10000b. If I leave the size off it works. Here is the failing
> mkfs.xfs -f -l logdev=/dev/sdb6,size=10000b /dev/sda2
> If I just leave the size parameter off it works. But how much of the
> /dev/sdb6 partition is used? /proc/partitions shows this for sdb6:
> 8 22 16033 sdb6
> Is the partition big enough (I made it really small thinking that the log
> is small)? /proc/partitions for /dev/sda2 looks like:
> 8 2 208845 sda2
> Does the log size depend on the partition being logged (i.e. if sda2 was
> 10 times bigger would the log need to be ten times bigger)? I know I
> could just use an internal log, but I have two disks and the itch to
> experiment :)
> Futher, by default mkfs.xfs creates logs with version 1. Shouldn't it
> default to version 2?
> Now using mkfs.xfs version 2.3.9
> James Rich