On Wed, Apr 28, 2021 at 09:50:56AM -0400, Theodore Ts'o wrote: > On Tue, Apr 27, 2021 at 11:11:16AM -0700, Darrick J. Wong wrote: > > > > TBH I think these tests (g/453 and g/454) are probably only useful for > > filesystems that allow unrestricted byte streams for names. > > I'm actually a little puzzled about why these tests should exist: > > # Create a directory with multiple filenames that all appear the same > # (in unicode, anyway) but point to different inodes. In theory all > # Linux filesystems should allow this (filenames are a sequence of > # arbitrary bytes) even if the user implications are horrifying. > > Why do we care about testing this? The assertion "In all theory all > Linux filesystems should allow this" is clearly not true --- if you > enable unicode support for ext4 or f2fs, this will no longer be true, > and this is considered by some a _feature_ not a bug --- precisely > _because_ the user implications are horrifying. > > So why does these tests exist? Darrick, I see you added them in 2017 > to test whether or not xfs_scrub will warn about confuable names, if > _check_xfs_scrub_does_unicode is true. So we already understand that > it's possible for a file system checker to complain that these file > names are bad. Yes, that's exactly why this test (and generic/454) were created -- as a functional test for xfs_scrub's unicode checking. > It's not at all clear to me that asserting that all Linux file systems > _must_ treat file names as "bag of bits" and not apply any kind of > unicode normalization or strict unicode validation is a valid thing to > test for in 2021. Perhaps not. These two tests do have the interesting side effect of catching filesystems that don't hew to the "names are bytestreams" philosophy. In 2017, fstests usage seemed like it pretty narrowly included only the big three filesystems, so it amuses me to no end that four years went by before this discussion started. :P Nowadays with wider testing of other filesystems (thanks, Red Hat!) we should hide these behind _require_names_are_bytes or move them to tests/xfs/. Question -- the unicode case folding doesn't apply to xattr names, right? --D > > - Ted
