On Thu, Oct 19, 2017 at 2:14 AM, Al Viro [off-list ref] wrote:

On Wed, Oct 18, 2017 at 02:38:38PM +0200, Ilya Dryomov wrote:

quoted

Check for CAP_SYS_ADMIN before calling into the driver, similar to
blkdev_flushbuf().  This is safer and can spare a check in the driver.

(Currently BLKROSET is overridden by md and rbd, rbd is missing the
check.  md has the check, but it covers a lot more than BLKROSET.)

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
---
Al, this appears to go back to your "[PATCH] block ioctl cleanup",
history commit c6973580141c.  2002 was a long time ago, but still ;)
Was there a reason you made BLKFLSBUF check for CAP_SYS_ADMIN before
->ioctl() and BLKROSET after?

It was a long time ago, indeed...  The funny part is, at the time
there had been no ->ioctl() instances with unusual BLKROSET handling
left; I really don't remember what had left to the override for
those remaining and (assuming it hadn't been a plain and simple braino)
the reasons for leaving the check to drivers that might eventually
want to add such overrides would be in whatever discussion that
had lead to leaving that override...

There was a *lot* of patch series (semi)manual reordering/rebasing, so
it might have easily been braindamage on conflict resolution during
rebase.

gendisk work had been literally hundreds of patches all over the
drivers/* over the summer and autumn of 2002; I have bits and pieces of
email archives from back then, but quick grep doesn't catch any
discussions along those lines and they are incomplete ;-/

Anyway,
        a) I don't see any reason for drivers to relax the checks on
BLKROSET and rbd lacking those is almost certainly a bug
        b) Acked-by: Al Viro [off-list ref]
        c) I can push it through vfs tree, but it would probably make
more sense block one.

Jens, can you pick this up for 4.15?

Thanks,

                Ilya