On Wed, Jan 07, 2026 at 07:25:18AM -0700, Theo de Raadt wrote:
> Job Snijders <job@bsd.nl> wrote:
> > On Wed, Jan 07, 2026 at 11:51:42AM +0000, Crystal Kolipe wrote:
> > > Whilst I like the general idea, the use of 1000 as a divisor instead
> > > of 1024 seems very wrong.
> > 
> > I also tried using a 1024 divisor, but the results just drifted too far
> > away from reality for my liking. For example, a power-of-2-based 'mega'
> > is ~ 5% larger than a power-of-10-based 'mega'. To me it seems a 1024
> > divisor might mislead the operator about inode availability.
> 
> I don't understand this. How many people are requiring precision for
> the inode count? This is a scheme which intentionally drops 3 digits
> of precision.

I use inode count to estimate how much runway I have on certain
machines, between rounding up and the power-of-2/power-of-10 difference
for large file systems it adds up.

-current:

	$ df -hi /var/www/htdocs/rpki
	Filesystem     Size    Used   Avail Capacity  iused    ifree %iused  Mounted on
	/dev/sd0m      2.8T   62.7G    2.6T     3%  8630581 803356617    2%   /var/www/htdocs/rpki

With my patch:

	$ ./obj/df -hi /var/www/htdocs/rpki
	Filesystem     Size    Used   Avail Capacity iused  ifree %iused  Mounted on
	/dev/sd0m      2.8T   62.6G    2.6T     3%    8.6M   803M    2%   /var/www/htdocs/rpki

With fmt_scaled(1):

	$ ./obj/df -hi /var/www/htdocs/rpki
	Filesystem     Size    Used   Avail Capacity iused  ifree %iused  Mounted on
	/dev/sd0m      2.8T   62.7G    2.6T     3%    8.2M   766M    2%   /var/www/htdocs/rpki

With fmt_scaled(1) the df utility reports 37 million fewer inodes than
actually are available, and worse, it reports 400K more inodes than
actually are in use. In larger file system the deviation matters.