(In reply to Richard W.M. Jones from comment #0)
> Description of problem:
> 
> $ rm -f root
> $ truncate -s 4G root
> $ /usr/sbin/mke2fs -t ext2 -Fq root
> mke2fs: Device size reported to be zero.  Invalid partition specified, or
> 	partition table wasn't reread after running fdisk, due to
> 	a modified partition being busy and in use.  You may need to reboot
> 	to re-read your partition table.

I read your message on the devel mailing list.
Issuing the above commands on a Raspberry Pi 3 these are the results

$ uname -a
Linux rpi3 5.3.0-0.rc1.git3.1.fc31.armv7hl #1 SMP Thu Jul 25 15:54:45 UTC 2019 armv7l armv7l armv7l GNU/Linux
$ cat /etc/redhat-release 
Fedora release 31 (Rawhide)
$ rpm -qv e2fsprogs
e2fsprogs-1.45.2-1.fc31.armv7hl

$ rm -f root
$ truncate -s 4G root
$ /usr/sbin/mke2fs -t ext2 -Fq root
warning: Unable to get device geometry for root

Interesting thanks.  I wonder why this fails in Koji :-?  I guess I've got to do a bit more testing ...

FWIW

$ rm -f root 
$ touch root
$ /usr/sbin/mke2fs -t ext4 -Fq root
mke2fs: Device size reported to be zero.  Invalid partition specified, or
	partition table wasn't reread after running fdisk, due to
	a modified partition being busy and in use.  You may need to reboot
	to re-read your partition table.

I modified a Koji build to run supermin, but also print the kernel etc.

Linux buildvm-armv7-05.arm.fedoraproject.org 5.2.2-200.fc30.armv7hl+lpae #1 SMP Sun Jul 21 15:36:24 UTC 2019 armv7l armv7l armv7l GNU/Linux

e2fsprogs-1.45.2-1.fc31.armv7hl

supermin: ext2: creating empty ext2 filesystem 'd2.4r818cjh/root'
RPM build errors:
BUILDSTDERR: mke2fs 1.45.2 (27-May-2019)
BUILDSTDERR: mke2fs: Device size reported to be zero.  Invalid partition specified, or
BUILDSTDERR: 	partition table wasn't reread after running fdisk, due to
BUILDSTDERR: 	a modified partition being busy and in use.  You may need to reboot
BUILDSTDERR: 	to re-read your partition table.
BUILDSTDERR: supermin: /usr/sbin/mke2fs -t ext2 -F 'd2.4r818cjh/root': command failed, see earlier errors
BUILDSTDERR: error: Bad exit status from /var/tmp/rpm-tmp.hycBi4 (%check)
BUILDSTDERR:     Bad exit status from /var/tmp/rpm-tmp.hycBi4 (%check)

All that supermin is doing here is basically the commands as outlined above:

https://github.com/libguestfs/supermin/blob/c97b3917068597a0e68e88d9a905da766ade40da/src/format_ext2.ml#L40-L55

I wonder if there's some problem with the page cache, for example mke2fs is using O_DIRECT
but the previous write to the filesystem hasn't been committed yet.

(In reply to Alessio from comment #3)
> FWIW
> 
> $ rm -f root 
> $ touch root
> $ /usr/sbin/mke2fs -t ext4 -Fq root
> mke2fs: Device size reported to be zero.  Invalid partition specified, or
> 	partition table wasn't reread after running fdisk, due to
> 	a modified partition being busy and in use.  You may need to reboot
> 	to re-read your partition table.

This is expected because "touch" creates a zero length file.  See the code
linked in the previous comment for what we actually do to create a 4G file.

I tried to fsync the directory before calling mke2fs but it made no difference.

Oh lovely, this is actually a compiler bug.

$ cat test.ml
let size = Int64.shift_left 1L 32 in
Unix.LargeFile.ftruncate Unix.stdout size
$ ocamlopt unix.cmxa test.ml -o test
$ strace ./test > /tmp/file
...
ftruncate64(1, 0)                       = 0
               ^
               this should be 2^32

Proposed patch:
https://github.com/ocaml/ocaml/pull/8843

Will be included in OCaml 4.08.1 which we'll upgrade to when it is released.

Upstream fix:
https://github.com/ocaml/ocaml/commit/5e4b55d3bd3fdf3e7512c132ad36d103d7131e72