On 2019/5/12 1:30 上午, Pierre JUHEN wrote: > Hi, > > I use bcache extensively on 3 PC, and I lost data on 2 of them after an > attempt to migrate to Fedora 30. > > My configuration is almost the same on the 3 PCs : > > /boot/efi and /boot are native on the SSD, and there is a bache frontend. > > bcache backend is on a HDD or on a raid1 array. > > bcache device is the physical volume of an LVM volume group. > > Here is how to reproduce the problem : > > 1/ Create the storage configuration as explained above. > > 2/ Install Fedora 29 on a logical volume (ext4) and a swap logical volume. > > 3/ Update the installation (dnf update --refresh) > > 4/ Migrate to Fedora 30 in download mode (dnf system-upgrade > --releasever=30 --allowerasing donwnload, then dnf system-upgrade reboot) > > 5/ try to prevent automatic reboot in Fedora 30 (for example in > commenting out /boot/efi in /etc/fstab) > > 6/ reboot using Fedora 29 kernel and initramfs -> Everything is fine > > 7/ reboot using Fedora 30 kernel and initramfs -> Everything is > corrupted, even unmounted volumes of the volume group > > I did the test case twice, the second time in downgrading bcache-tools > to Fedora 29 -> same issue > > This means that's the problem is located in the bcache kernel module ; > but since I guess it's the same code, the problem is probably linked to > the building environment (gcc version ?) > > I reported the bug : https://bugzilla.redhat.com/show_bug.cgi?id=1707822 > > But I thought it was not a kernel problem. On my development machine the GCC is still v7.3.1, for now I don't know how to upgrade to GCC 9.1 yet. >From the dmesg.lis file, it seems fc30 uses 5.0.11-300, so what is the kernel version of fc29 ? (And from dmesg.lis I don't see anything suspicious on bcache message, no clue yet). Coly Li -- Coly Li