0
votes

I'm working on a buildroot image, and this image is using Busybox for init, along with devtmpfs and eudev for device initialization. There's no initramfs, the system boots directly into its permanent root. The fstab contains mountpoints that reference disks by partlabel, which require the udev rules for persistent disk naming to be run beforehand.

The buildroot eudev package creates a SysV init script at /etc/init.d/S10udev for this, and by default, it's started by rcS, after inittab handles the system's mountpoints. Allowing rcS to start udev won't work for this use case, because udev requires proc and sysfs to be mounted before it can work, which seems to indicate that fstab is a bad place for these mountpoints.

Is the proper way to handle this to move the mountpoints for proc and sysfs back into inittab, then manually start S10udev in inittab before mount -a?

1

1 Answers

3
votes

Either that, or mark these mount points as noauto (so mount -a won't try to mount them) and then mount them later (a later init script or from eudev)