Hi,
在 2023/07/29 11:36, Yu Kuai 写道:
Hi,
在 2023/07/29 11:29, Xueshi Hu 写道:
I think this is wrong, you should at least keep following:
set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
md_wakeup_thread(mddev->thread);
I fail to comprehend the rationale behind the kernel's need to invoke
raid1d() repeatedly despite the absence of any modifications.
It appears that raid1d() is responsible for invoking
md_check_recovery()
and resubmit the failed io.
No, the point here is to set MD_RECOVERY_NEEDED and MD_RECOVERY_RECOVER,
so that md_check_recovery() can start to reshape.
I apologize, but I am still unable to comprehend your idea.
If mmdev->delta_disks == 0 , what is the purpose of invoking
md_check_recovery() again?
Sounds like you think raid1_reshape() can only be called from
md_check_recovery(), please take a look at other callers.
And even if raid1_reshape() is called from md_check_recovery(),
which means reshape is interupted, then MD_RECOVERY_RECOVER
and MD_RECOVERY_RECOVER need to be set, so that reshape can
continue.
Thanks,
Kuai
Thanks,
Kuai
.
