> If I read the code correctly, than this patch is the culprit: > http://sources.debian.net/src/systemd/215-17%2Bdeb8u2/debian/patches/Skip-filesystem-check-if-already-done-by-the-initram.patch/?hl=29#L29 > > When systemd is run in the initramfs (which it is with dracut), there > is no /run/initramfs/fsck-usr.
The correct workaround is to mark the /usr entry in /etc/fstab with x-initrd.mount. The fact that this solves the problem confirms the analysis. Perhaps interesting side note: When dracut does not use systemd, its usrmount module will use the flag file /run/initramfs/usr-fsck to indicate that fsck has been performed [1]. Might be worth agreeing on a common file name for this? [1] https://sources.debian.net/src/dracut/044%2B3-2/modules.d/98usrmount/mount-usr.sh/#L95 When systemd is used in the initramfs, it is itself responsible for mounting /usr (and running fsck or not). The "culprit" patch breaks that. We need a solution that works both inside and outside of the initramfs.