Andrei, I've found that running udevadm test /sys/block/sda/sda4 will create the correct link after having initialized a filesystem on /dev/sda4 with a label using mkfs -t xfs -L root -q -f /dev/sda4 Tried udevadm control -S WORKED! And found my bug!. I did stop the exec queue to prevent old md device to appear, then only restarted it if I create lvm (forgot to restat itin normal conditions).... Thanks for your help and tips. (udev was running, but it was stopped...) Olivier. Le 04/02/2019 19:02, « Andrei Borzenkov » <arvidjaar@xxxxxxxxx> a écrit : 04.02.2019 19:15, LAHAYE Olivier пишет: > Progressing in diagnostic, still need more help. > A temporary script named devexists-\x2fdev\x2fdisk\x2fby-label\x2froot.sh has been created in /lib/dracut/hook/initqueue/finished by rootfs-generator.sh. > It tests the existence of /dev/disk/by-label/root (normal behavior) > > PROBLEM: /dev/disk/by-label doesn't exists (thus the script always return 1, thus the finish hook loops forever) > > blkid lists all partitions including the LABEL=root one (after reporting an error on fd0) > /dev/sda4: LABEL="root" UUID="xydfkjhzkfghzv" TYPE="xfs" PARTLABEL="primary" PARTUUID="ssdfighuskdjfhb" > > I can mount LABEL=root /sysroot with success. > > In the dmesg I have blk_update_request: I/O error, dev fd0, sector 0 > > I tried udevadm trigger, but this didn't help. /dev/disk/by-label still missing... > Is udevd running? Do you have any other link for other devices? Do you have udev rules in initrd? > (I'm using CentOS-7, kernel 3.10.0-957.1.3.el7.x86_64) > > Am I facing a kernel bug or is there something I'm doing wrong? (I'm creating the partition table and I format them in the initqueue). Is there something I need to do to inform the kernel that it has to update /dev/disk ? >