On Dec 8, 2018, at 10:45 AM, Theodore Y. Ts'o <tytso@xxxxxxx> wrote: > > On Mon, Dec 03, 2018 at 04:00:12PM -0500, Gabriel Krisman Bertazi wrote: >> I didn't want to load the table in those functions because I didn't want >> to fail there if the nls_table wasn't found. If the user has some new >> unsupported encoding, e2fsprogs could provide some functionality, even >> if it can't deal with the file tree itself. Failing during open seemed >> too harsh. > > Unfortunately, I think the only thing we can do is to fail at open or > mount if the encoding is unknown. The problem is that we can't > correctly handle case-folded directories which have htree enabled. > The problem is that we might have encoding-oblivious applications, > such as fuse2fs (for example) where if they use the high-level > libext2fs interfaces, they don't *need* to be encoding aware. > > But if we don't fail the open, then what do we do if the library > routine to calculate a directory hash is called? Most applications > (or callers in libext2fs for that matter) won't gracefully handle an > error there. > > It's one thing if we only support Unicode version N, and the file > system is Unicode version N+1. So long as the user isn't trying to > use the new scripts, things are mostly OK. > > But what if the alternate encoding is something completely different? > Say, EBCDIC, or UTF-EBCDIC[1]? :-) There really is nothing we can do > sanely but to fail the mount. > > This also effectively means that new encodings are effectively > incompatible features, but I think that's OK. Doesn't this meant that if opening a filesystem with an unknown encoding fails, that any kind of corruption to the encoding type would cause problems, and opening the filesystem to _remove_ the encoding would also fail? I think that it makes more sense to allow the filesystem to be opened in this case, and only fail operations that need the specific encoding (e.g. "e2fsck -fD" that needs to know the hashing). It shouldn't hurt tune2fs, debugfs, resize2fs, etc. that probably do not need to know the encoding. At worst, they could revert the htree directory to a flat directory if they need to insert an entry and don't understand how to hash it. Cheers, Andreas
Attachment:
signature.asc
Description: Message signed with OpenPGP
