]> www.infradead.org Git - users/hch/block.git/commit
block: propagate BLKROSET on the whole device to all partitions
authorChristoph Hellwig <hch@lst.de>
Sat, 5 Dec 2020 09:11:02 +0000 (10:11 +0100)
committerChristoph Hellwig <hch@lst.de>
Sat, 23 Jan 2021 17:38:21 +0000 (18:38 +0100)
commit4d0e6989cfd3d22569e79584042ad79de0ebf00c
tree7df4dd31ccec18bcbb1e4970f3e68b38a3e650f6
parentc40f4179231800d043b555bb92cd1d8b968a0bd8
block: propagate BLKROSET on the whole device to all partitions

Change the policy so that a BLKROSET on the whole device also affects
partitions.  To quote Martin K. Petersen:

It's very common for database folks to twiddle the read-only state of
block devices and partitions. I know that our users will find it very
counter-intuitive that setting /dev/sda read-only won't prevent writes
to /dev/sda1.

The existing behavior is inconsistent in the sense that doing:

  # blockdev --setro /dev/sda
  # echo foo > /dev/sda1

permits writes. But:

  # blockdev --setro /dev/sda
  <something triggers revalidate>
  # echo foo > /dev/sda1

doesn't.

And a subsequent:

  # blockdev --setrw /dev/sda
  # echo foo > /dev/sda1

doesn't work either since sda1's read-only policy has been inherited
from the whole-disk device.

You need to do:

  # blockdev --rereadpt

after setting the whole-disk device rw to effectuate the same change on
the partitions, otherwise they are stuck being read-only indefinitely.

However, setting the read-only policy on a partition does *not* require
the revalidate step. As a matter of fact, doing the revalidate will blow
away the policy setting you just made.

So the user needs to take different actions depending on whether they
are trying to read-protect a whole-disk device or a partition. Despite
using the same ioctl. That is really confusing.

I have lost count how many times our customers have had data clobbered
because of ambiguity of the existing whole-disk device policy. The
current behavior violates the principle of least surprise by letting the
user think they write protected the whole disk when they actually
didn't.

Suggested-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
block/genhd.c