Re: bcache & Fedora 30: massive corruption 100% reproducable

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 5/12/19 7:11 PM, Coly Li wrote:
 From the dmesg.lis file, it seems fc30 uses 5.0.11-300, so what is the
kernel version of fc29 ?

The reproducability extends to any kernel I tested for Fedora 30. And each of these is available for Fedora 29 too.

Kernels I tested so far (they all broke the FS):

- kernel-5.0.11-300.fc30 (gcc 9.0)

- kernel-5.0.13-300.fc30 (gcc 9.0)

- kernel-5.0.14-300.fc30 (gcc 9.1)

Kernels I tested so far which worked fine:

- kernel-5.0.11-200.fc29 (gcc 8)

- kernel-5.0.13-200.fc29 (gcc 8)

- kernel-5.0.14-200.fc29 (gcc 8)

So the pattern seems to be that they all break storage when built on Fedora 30, and they work fine when built on Fedora 29.

Reproducing is easy:

 * Download the kvm image by means of this torrent (part of the bug
   report below): https://bugzilla.redhat.com/attachment.cgi?id=1567288
 * Install the VM (have a look at the included libvirt xml, and tweak
   it if required)
 * There is a user testuser with password testuser.
 * Next do:
 * A) dnf update "kernel*", reboot. Do another reboot and enjoy a
   stable system.
 * B) dnf --release=30 update "kernel*", reboot. Try another reboot and
   have a look at dmesg, it contains ext4 errors. It may not even boot

On 5/11/19 10:06 AM, Rolf Fokkens wrote:
FYI:

https://bugzilla.kernel.org/show_bug.cgi?id=203573
https://bugzilla.redhat.com/show_bug.cgi?id=1708315

On 5/9/19 7:21 PM, Rolf Fokkens wrote:
Hi,

The reproducability is 100%. It's enough to only upgrade to a Fedora 30 kernel on a Fedora 29 system. The next reboot will probably be the last reboot ever.

My Fedora bug report is here:

If it's gcc9 related, the cause may be somewhere between "Fedora's decision to use gcc9" and "bcache needing a fix".

Rolf

On 5/6/19 7:45 PM, Rolf Fokkens wrote:

Hi,

I helped in 2013 to get bcache-tools integrated in Fedora 21 (https://lists.fedoraproject.org/archives/list/devel@xxxxxxxxxxxxxxxxxxxxxxx/thread/UEGAUSP377TB3KMUO7XK42KREHOUDZPG/).

Ever since it worked like a charm, and bcache laptops (we have several at work) survived upgrading to a next Fedora release flawlessly. Since Fedora 30 this has changed however: laptops using bcache mess up backing store big time. It seems as if the backing device is corrupted by random writes all over the place. It's hard to narrow down the cause of this issue, and I'm still in the process of trial and error. May be later on I'll have more info.

Some info:

  * The laptops are using writeback caching
  * The laptops have a bcache'd root file system
  * It seems like the issue is in the Fedora kernel 5.0.10 for Fedora
    30, but not kernel 5.0.10 for Fedora 29.
  * One notable difference between the Fedora 29 and Fedora 30 kernels
    is that Fedora 30 uses gcc 9 to build the kernel.

As mentioned i'm still in the process of narrowing down the cause of the issue. But any suggestions are welcome.

Rolf







[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [Linux ARM Kernel]     [Linux Filesystem Development]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux