On Tue, Jul 17, 2012 at 08:19:14AM +0100, James Bottomley wrote:

On Mon, 2012-07-16 at 16:48 -0700, Greg KH wrote:

quoted

On Sat, Jul 07, 2012 at 11:04:45PM -0400, Namjae Jeon wrote:

quoted

From: Namjae Jeon <redacted>

The numbers of USB HDDs(All USB HDD I checked) does not respond
correctly to scsi mode sense command for retrieving the write cache
page status. Even though write cache is enabled by default, due to
scsi driver assume that cache is not enabled which in turn might lead
to loss of data since data still will be in cache.
This result that all filesystems is not stable on USB HDD when the
device is unplugged abruptly, even though these are having journaling
feature. Our first trying is that scsi driver send ATA command
(ATA Pass through, #85) to USB HDD after failure from normal routine to
know write cache enable.
We have known it is dangerous after testing several USB HDD. some of
HDD is stalled by this command(A-DATA HDD). So we tried to make the
patch James Bottomley's suggestion(usb quirk) on version 2 that add
product ID and verdor ID of USB HDD to USB quirk list after checking
write cache.
All filesystem will be stable on USB HDD registered in quirk list.
And it will be updated continuously.

Now applied to the usb-next branch.

It's been in scsi#misc for ten days with no problems.  Lets leave it
there rather than create merge and rebase issues.

No need for rebasing.  It's already in my tree, so we can handle the
merge when this hits Linus's tree.

thanks,

greg k-h