Re: [PATCH blktests] nvme/{033,036,037}: skip when the test target namespace has metadata

[Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>]

On Jun 30, 2026 / 20:37, Shin'ichiro Kawasaki wrote:
> The kernel commit c8cdecdb47d3 ("nvme: core: reject invalid LBA data
> size from Identify Namespace") in the v7.2-rc1 tag triggered failures of
> blktests test cases nvme/033, 036 and 037. The failures are observed
> when TEST_DEV specifies a NVMe device with metadata. The commit
> introduced a new check in nvme_update_ns_info_block() to reject invalid
> LBA data size. After this commit, namespaces with zero LBA data size are
> rejected.
> 
> According to the comment in nvmet_passthru_override_id_ns() in the
> kernel code drivers/nvme/target/passthru.c, "Presently the NVMEoF target
> code does not support sending metadata, so we must disable it here". For
> that purpose, the function sets the LBA data size to zero for all LBA
> format (lbaf) entries with metadata. After the trigger commit, these
> entries become invalid for NVMEoF target passthru, then namespaces with
> metadata are not visible as block devices through the target setup.
> 
> These test cases set up NVMEoF target passthru assuming the namespace
> devices are visible, so they cannot find the namespaces and fail.
> 
> To avoid the failures, skip the test cases when TEST_DEV has metadata.

FYI, I applied this patch.