On Fri, Jul 25, 2025 at 09:37:53PM +0800, Zorro Lang wrote: > About why this case need "shutdown": > https://lore.kernel.org/fstests/PUZPR04MB63169A8C1008035BB2D104568166A@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/ > > So looks like we'd better to notrun if shutdown isn't supported by fs on > SCSI_DEBUG_DEV. But if think "_require_scratch_shutdown" is a problem, maybe > we can avoid using _scratch_ things, use a local function to check shutdown > on SCSI_DEBUG_DEV manually, and notrun if it's not supported? These are two almost entirely unrelated, except usually sharing some code for the implementation: _require_scratch_shutdown checks for the IOC_SHUDOWN ioctl, while this test coverd behavior of the file system when the underlying block device is removed. I can't really think of a feature test for shutting down the file system on device removal, because it really should not be an optional feature.. > BTW, can I ask why do you need to test with TEST_DEV only :) Although > SCRATCH_DEV is optional (README says), I think nearly all testers test > with SCRATCH_DEV. I'm working on an experimental xfs change that will need a lot of tooling changes to handle the various scratch mkfs use cases. So while it's still bleeding edge I'd rather get all the TEST_DEV testing than debuggіng/fixing xfstests.
