On Wed, Jan 17, 2024 at 08:21:25PM -0800, Christoph Hellwig wrote: > On Wed, Jan 17, 2024 at 05:32:50PM -0800, Darrick J. Wong wrote: > > > > > > For xfs/002 that is the libxfs_buf_read in __set_cur, when setting the > > > type to data, but I haven't looked at the other test in detail. > > > > Hmm. Perhaps the userspace buftarg setup should go find the physical > > sector size of the device? That "bb_count = 1" in set_iocur_type looks > > a bit smelly. > > Yes, that should fix this particular issue. > > > > Should I look into finding all these assumptions in xfs_db, or > > > just make the direct I/O enablement conditional n a 612 byte sector > > > size? > > > > Let me go run a lbasize=4k fstests run overnight and see what happens. > > IIRC zorro told me last year that it wasn't pretty. > > There's a few failures, but I've been slowly trying to fix this. The > libxfs/mkfs log sector size detection series in one part of that, > and this: > > https://lore.kernel.org/linux-block/20240117175901.871796-1-hch@xxxxxx/T/#u Hmm well I didn't manage to add your loop device patch before I sent this out last night, but here's the fstest results: https://djwong.org/fstests/output/.67c2f90f0a1bb329a1b895c50285b0d23c1bd2bb44b7839f3543f82281665db1/.4a10533d4dd2085d3f996649e0886284f557617c94e604189448672e6009b9e8/ Looks like there were a lot of weird problems. OFC now the second ice storm has started and the lights are flickering so that might be all from me for now. --D > is another >
