On Tue, May 22, 2018 at 4:34 PM, Randy Dunlap [off-list ref] wrote:

On 05/22/2018 04:31 PM, Kees Cook wrote:

quoted

On Tue, May 22, 2018 at 12:16 PM, Jens Axboe [off-list ref] wrote:

quoted

On 5/22/18 1:13 PM, Christoph Hellwig wrote:

quoted

On Tue, May 22, 2018 at 01:09:41PM -0600, Jens Axboe wrote:

quoted

I think Martin and Christoph are objecting to moving the code to
block/scsi_ioctl.h. I don't care too much about where the code is, but
think it would be nice to have the definitions in a separate header. But
if they prefer just pulling in all of SCSI for it, well then I guess
it's pointless to move the header bits. Seems very heavy handed to me,
though.

It might be heavy handed for the 3 remaining users of drivers/ide,

Brutal :-)

Heh. I noticed a similar sense buffer use in drivers/cdrom/cdrom.c
too. Is this okay under the same considerations?

No.  Do not select an entire subsystem.  Use depends on it instead.

I looked at that first, but it seems it's designed for that. For
example, "config ATA" already has a "select SCSI".

It does look fishy, though, since "config SCSI" has a "depends" which
would be ignored by "select". Luckily, all these uses already do a
"depends on BLOCK" (directly or indirectly).

-Kees

-- 
Kees Cook
Pixel Security