On Tue, 16 Sept 2025 at 17:04, Jan Kara jack@suse.cz wrote:
On Tue 16-09-25 13:04:42, Jan Kara wrote:
On Tue 16-09-25 12:57:26, Naresh Kamboju wrote:
The following LTP chdir01 df01_sh and stat04 tests failed on the rock-pi-4b qemu-arm64 on the Linux next-20250915 tag build.
First seen on next-20250915 Good: next-20250912 Bad: next-20250915
Regression Analysis:
- New regression? yes
- Reproducibility? yes
- rk3399-rock-pi-4b, ltp-smoke
- qemu-arm64, ltp-smoke
- qemu-arm64-compat, ltp-smoke
- chdir01
- df01_sh
- stat04
Test regression: next-20250915: LTP chdir01 df01_sh stat04 tst_device.c:97: TBROK: Could not stat loop device 0
This is really strange. Those failing tests all start to complain that /dev/loop0 doesn't exist (open gets ENOENT)... The fact that this is limited to only a few archs suggests it's some race somewhere but I don't see any relevant changes in linux-block in last three days...
Ha, Mark Brown tracked this [1] to changes in VFS tree in extensible_ioctl_valid(). More discussion there I guess.
That right, Ander’s bisection confirmed the same commit that Mark Brown reported.
# first bad commit: [60949057a2e71c9244e82608adf269e62e6ac443] block: use extensible_ioctl_valid()
[1] https://lore.kernel.org/all/02da33e3-6583-4344-892f-a9784b9c5b1b@sirena.org....
Honza-- Jan Kara jack@suse.com SUSE Labs, CR
- Naresh