Re: remove ->revalidate_disk (resend)

["chenxiang (M)" <chenxiang66@hisilicon.com>]

Hi Christoph,


在 2021/6/16 21:50, Christoph Hellwig 写道:
> On Wed, Jun 16, 2021 at 05:41:54PM +0800, chenxiang (M) wrote:
>> Hi,
>>
>> Before i reported a issue related to revalidate disk
>> (https://www.spinics.net/lists/linux-scsi/msg151610.html), and no one
>> replies, but the issue is still.
>>
>> And i plan to resend it, but i find that revalidate_disk interface is
>> completely removed in this patchset.
>>
>> Do you have any idea about the above issue?
> bdev_disk_changed still calls into sd_revalidate_disk through sd_open.
> How did bdev_disk_changed get called for you previously?  If it was
> through the BLKRRPART ioctl please try latest mainline, which ensures
> that ->open is called for that case.

I use the latest mainline (Linux Euler 5.13.0-rc6-next-20210616), and 
the issue is still.
It is through BLKRRPART ioctl, and the call stack is as follows:

BLKRRPART ->
         block_ioctl ->
                 blkdev_ioctl ->
                         blkdev_common_ioctl ->
                                 blkdev_get_by_dev ->
                                         __blkdev_get ->
                                                 ...
disk->fops->open() ->
                                                         sd_open()
                                                 ...
                                                 dev_disk_changed()
                                                 ...



In function sd_open(), it calls sd_revalidate_disk() when 
sdev->removable or sdkp-> write_prot is true, but for our disk, 
sdev->removable = 0 and
sdkp->write_prot = 0, so sd_revalidate_disk() is not called actually.
For previous code, it will call sd_revalidate_disk() in 
bdev_disk_changed() from here 
(https://elixir.bootlin.com/linux/v5.10-rc1/source/fs/block_dev.c#L1411).

>
> .
>