Re: [RFC] block: reject I/O in BLKRRPART if block size changed

On 20-12-27 03:02:32, Minwoo Im wrote:

Background:
  Let's say we have 2 LBA format for 4096B and 512B LBA size for a
NVMe namespace.  Assume that current LBA format is 4096B and in case
we convert namespace to 512B and 4096B back again:

  nvme format /dev/nvme0n1 --lbaf=1 --force  # to 512B LBA
  nvme format /dev/nvme0n1 --lbaf=0 --force  # to 4096B LBA

  Then we can see the following errors during the BLKRRPART ioctl from
the nvme-cli format subcommand:

  [   10.771740] blk_update_request: operation not supported error, dev nvme0n1, sector 0 op 0x0:(READ) flags 0x0 phys_seg 1 prio class 0
  [   10.780262] Buffer I/O error on dev nvme0n1, logical block 0, async page read
  ...

  Also, we can see the Read commands followed by the Format command due
to BLKRRPART ioctl with Number of LBAs to 65535(0xffff) which is
under-flowed because the request for the Read commands are coming with
512B and this is because it's playing around with i_blkbits from the
block_device inode which needs to be avoided as [1].

  kworker/0:1H-56      [000] ....   913.456922: nvme_setup_cmd: nvme0: disk=nvme0n1, qid=1, cmdid=216, nsid=1, flags=0x0, meta=0x0, cmd=(nvme_cmd_read slba=0, len=65535, ctrl=0x0, dsmgmt=0, reftag=0)
  ksoftirqd/0-9       [000] .Ns.   916.566351: nvme_complete_rq: nvme0: disk=nvme0n1, qid=1, cmdid=216, res=0x0, retries=0, flags=0x0, status=0x4002
  ...

  Before we have commit 5ff9f19231a0 ("block: simplify
set_init_blocksize"), block size used to be bumped up to the
4K(PAGE_SIZE) in this example and we have not seen these errors.  But
with this patch, we have to make sure that bdev->bd_inode->i_blkbits to
make sure that BLKRRPART ioctl pass proper request length based on the
changed logical block size.

Description:
  As the previous discussion [1], this patch introduced a gendisk flag
to indicate that block size has been changed in the runtime.  This flag
is set when logical block size is changed in the runtime with sector
capacity itself.  It will be cleared when the file descriptor for the
block devie is opened again which means __blkdev_get() updates the block
size via set_init_blocksize().
  This patch rejects I/O from the path of add_partitions() and
application should open the file descriptor again to update the block
size of the block device inode.

[1] https://lore.kernel.org/linux-nvme/20201223183143.GB13354@localhost.localdomain/T/#t (local)

Signed-off-by: Minwoo Im <redacted>

Hello,

Sorry for the noises here.  Please ignore this patch.  Will try to
prepare a new one for this issue.

Thanks,