nilfs-related kernel crashes

[mbb78]

Hi!

I'm not sure if I'm posting this in the right place, but the crashes seem to be related with nilfs driver. I personally use Manjaro Linux. Some reports can be seen on the following posts:

(https://forum.manjaro.org/t/stable-update-2019-11-14-kernels-plasma-kde-apps-kde-frameworks-mesa-nvidia/111239/134?u=mbb)

(https://forum.manjaro.org/t/kernel-bug-related-to-nilfs/112783?u=mbb)

To sumarize:

• this issue was first noted on kernel 4.14-154;
• the previous I was using (and still use) is 4.14.151;
• I don't know if the problem occurs between the two kernels above, because I may have skipped updates or Manjaro may not have issued them on the stable branch;
• so far I have tried kernels 4.14-154, 4.14-155, 4.19-85 and 5.3-12 (all of them fail);
• if I disable pti or mitigations with a kernel parameter the crash doesn't occur upon boot, but it eventually happens later, while using the nilfs partition (which I use for data);
• I use the kernel parameter scsi_mod.use_blk_mq=1 but removing it doesn't change anything.

Thank you very much for reading this and for your work on the nilfs filesystem. I really like it. Please tell me if I should report the issue on the kernel bugtracker instead.

[konis]

This looks the same as following issue reported in mailing list:

• https://marc.info/?t=157413851100002
• https://marc.info/?l=linux-nilfs&m=157401249403466

Thank you for reporting details.

[mbb78]

Yes, it appears to be the same issue. Apparently the commit was identified, so I hope it gets fixed soon.

Thanks!

[daiyy]

A workaround could be disable CONFIG_CGROUP_WRITEBACK in your kernel config, rebuild kernel and reboot.

[mbb78]

Thank you. I will try this when I have time and report back.

[konis]

This issue was fixed by commit 8301c719a2bd ("nilfs2: fix null pointer dereference at nilfs_segctor_do_construct()") which is merged to kernel 5.8-rc1. It was back ported to the following stable kernels:

v5.7.3 (linux-5.7.y) v5.6.19 (linux-5.6.y) v5.4.47 (linux-5.4.y)

For former stable kernels that back-ported the causal commit c3aab9a0bd91 ("mm/filemap.c: don't initiate writeback if mapping has no dirty pages"), back porting the above commit may resolve the issue.

[mbb78]

Thank you very much. I'm not currently using nilfs, but I'll get back to it.