Comment 66 for bug 2044657

Revision history for this message
trackwitz (trackwitz) wrote :

So, I did test the -proposed package for mantic on a clean install by creating a test-pool and running the zhammer.sh linked #16.


As expected on an unpatched system the error occurred during the first iteration:

[zhammer::1858] zhammer_1858_0 differed from zhammer_1858_538!
[zhammer::1858] Hexdump diff follows
--- zhammer_1858_0.hex 2024-02-03 12:44:07.478205144 +0000
+++ zhammer_1858_538.hex 2024-02-03 12:44:07.478205144 +0000
@@ -1,3 +1,3 @@
-00000000 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff |................|
+00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
 *
 00004000
[zhammer::1858] Uname: Linux zfstest 6.5.0-15-generic #15-Ubuntu SMP PREEMPT_DYNAMIC Tue Jan 9 17:03:36 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux
[zhammer::1858] ZFS userspace: zfs-2.2.0-0ubuntu1~23.10.1
[zhammer::1858] ZFS kernel: zfs-kmod-2.2.0-0ubuntu1~23.10
[zhammer::1858] Module: /lib/modules/6.5.0-15-generic/kernel/zfs/zfs.ko.zst
[zhammer::1858] Srcversion: 92158472E32FE6AEEEC7201
[zhammer::1858] SHA256: 177442f43f4c94537f8b003ab28ed33d00240c175e500370ad5bdd5c50234655
parallel: This job failed: zhammer /test 10000000 16k 10000 7


After enabling the -proposed repository, installing the updates and restarting the system is looks like the userspace-tools are now on the patched version (zfs-2.2.0-0ubuntu1~23.10.1), however the kernel module is still on the old version (without the .1) and, as expected, the bug is still reproducible:

[zhammer::1706] zhammer_1706_0 differed from zhammer_1706_1204!
[zhammer::1706] Hexdump diff follows
--- zhammer_1706_0.hex 2024-02-04 14:29:28.296850257 +0000
+++ zhammer_1706_1204.hex 2024-02-04 14:29:28.296850257 +0000
@@ -1,3 +1,3 @@
-00000000 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff |................|
+00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
 *
 00004000
[zhammer::1706] Uname: Linux zfstest 6.5.0-17-generic #17-Ubuntu SMP PREEMPT_DYNAMIC Thu Jan 11 14:01:59 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux
[zhammer::1706] ZFS userspace: zfs-2.2.0-0ubuntu1~23.10.2
[zhammer::1706] ZFS kernel: zfs-kmod-2.2.0-0ubuntu1~23.10
[zhammer::1706] Module: /lib/modules/6.5.0-17-generic/kernel/zfs/zfs.ko.zst
[zhammer::1706] Srcversion: 92158472E32FE6AEEEC7201
[zhammer::1706] SHA256: 0f6a069f6c3045e7c86507d7c158691d4ace8c6785888579652236fbdf8c66c0
parallel: This job failed: zhammer /test 10000000 16k 10000 2

Only when I am explicitly using the zfs-dkms package instead of the build-in kernel module, the correct module is loaded and the bug can’t be triggered any more even after 5 iterations (x10.000 files).

Therefore, I can conclude, that the fix itself is working correctly, however the package distributed in the -proposed repository does not include the correct kernel module. However, as this is the way most people are using ZFS on Ubuntu (instead of using the dkms module) this fix also has to be introduced in the current kernel package to fix the problem.