On 2025/12/26 15:54, zhaoyifan (H) wrote:
On 2025/12/26 14:51, Gao Xiang wrote:
Hi Yifan,
On 2025/12/23 18:04, Yifan Zhao wrote:
The `current erofsmount_nbd()` implementation verifies that the value in
`/sys/block/nbdX/pid`` matches the PID of the process executing
`erofsmount_nbd_loopfn()`, using this as an indicator that the NBD
device is ready. This check is incorrect, as the PID recorded in the
aforementioned sysfs file may belong to a kernel thread rather than a
userspace process (see [1]).
Do you have a way to reproduce that?
This issue is consistently reproducible in my WSL2 environment (kernel version
6.6.87.2-microsoft-standard-WSL2),
I guess you could use bpftrace or similiar to confirm the kthread
stack if possible?
but works correctly in other environments (e.g., openEuler 24.03 SP2). The
finer-grained difference
between these 2 environments remains unclear.
Thanks,
Gao Xiang
