On 8/27/25 1:30 PM, Lennart Poettering wrote: > Heya! > > Recent kernels fail if it is attempted to create a loopback block > device on top of a block device (rather than a regular file), which is a > feature long supported on Linux kernels, and that systemd relies on > (specifically, systemd-repart does). > > For example, this used to work: > > losetup --find --show --offset=4096 --sizelimit=409600 /dev/nvme0n1 > > But now it doesn't anymore... > > This is on 6.17rc3. My educated guess is that this is caused by > Rajeev's 8aa5a3b68ad144da49a3d17f165e6561255e3529, which tightened the > screws on validating the backing file's size, which now fails if the > backing file isn't actually a file, but a block device. > > (But I didn't spend more time tracking this down.) Yeah sorry about that, the problem is the followup to the commit you mentioned, and the fix for that is here: https://web.git.kernel.org/pub/scm/linux/kernel/git/axboe/linux-block.git/commit/?h=block-6.17&id=d14469ed7c00314fe8957b2841bda329e4eaf4ab and will land in -rc4. -- Jens Axboe