Hi,

I can confirm this bug and would like to add some data points from our 
environment.

Environment: 9 Debian 12 amd64 servers, all VMware guests with ext4 root, all 
upgraded to linux-image-6.1.0-55-amd64 (6.1.192-1). 5 of the 9 have crashed so 
far.

Time until the oops varies widely:
- On some machines it occurs immediately after boot (13 and 20 seconds of 
uptime in two cases).
- On others it takes several hours (about 4.3 hours / 15569 s in one case).
So a system that survives the first minutes on 6.1.192-1 should not be 
considered unaffected.

Triggering syscalls seen so far: mkdir (ORIG_RAX 0x53) and fsync (ORIG_RAX 
0x4a), so it is not limited to file creation.

Oops signature, identical on the machines where we captured it:

RIP: 0010:ext4_mb_load_buddy_gfp+0xbc/0x380 [ext4]
CR2: 0000000000000000 RAX: 0000000000000000
Code: ... 49 89 c7 48 3d 00 f0 ff ff 0f 87 28 01 00 00 <48> 8b 00 a8 04 ...

The pointer is checked with IS_ERR() and then dereferenced while being NULL.

After the oops the systems do not recover: systemd-journald and 
systemd-timesyncd hit their watchdog timeouts and cannot be killed even with 
SIGKILL ("Processes still around after final SIGKILL"). A hard reset is 
required.

The 6.1.192-1 changelog lists these two commits, which are not in the 6.1.190-1 
changelog:

- ext4: convert bd_bitmap_page to bd_bitmap_folio
- ext4: convert bd_buddy_page to bd_buddy_folio

The same backports caused an identical oops during the 6.1.167-rc1 stable 
review in March 2026 and were dropped then:
https://lkml.iu.edu/2603.3/02704.html

Workaround in use: booting linux-image-6.1.0-54-amd64 (6.1.190-1), which runs 
stable on the same machines.

Console screenshots of three of the crashes are attached. I am happy to provide 
further information or test a fixed package.

Regards,
Walter

Reply via email to