On 17/12/1 上午10:13, Michael Lyle wrote:
> On 11/30/2017 05:54 PM, 彭良彦 wrote:
>
>> Here is the issue reproduce step.
>> 1. create a block device sdb by iSCSI
>> 2. echo /dev/sdb > /sys/fs/bcache/register and succeed to register //sdb is the block dev create by iSCSI
>> 3. disconnect network and reboot // without network, SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE) will fail
>> 4. create a block device sdb by iSCSI
>> 5. echo /dev/sdb > /sys/fs/bcache/register and fail to register since bdev state is not reset to BDEV_STATE_NONE
>> 6. make-bcache -B /dev/sdb --writeback --wipe-bcache
>> the last step will fail in original version since bdev handle is not released during register and open() is failed in make-bcache, with this patch, it will succeed
>
> But it is supposed to fail / wait in this case, if the cache device is
> not registered. It needs the cache device metadata in order to know the
> correct state of each piece of data.
>
> That is, when you are using writeback mode, you need the cache device,
> because some of the writes will go to the cache device and some to the
> backing device.
>
> Mike
If the cache device is not persistent like ramdisk, all data will be lost after reboot. We have to use make-bcache to fix the inconsistent super block of backing device.��.n��������+%������w��{.n�����{���{ay� ʇڙ���f���h������_�(�階�ݢj"��������G����?���&��
