Ext4 encryption

http://austingroupbugs.net/view.php?id=251... oh my god that bug has so much concentrated stupid in it, it hurts. From dwheeler's perfectly reasonable proposal -- I can't see any sane programs being hurt by an optional, default-on flag that bars creation of files with linefeeds, esc, escape, backspace and the like in them -- we get to a mention (in Geoff Clare's comment 1732) of an old committee decision which has been catastrophically infected with second-system syndrome involving a new readdir() flag, a readdir() that fails when encountering such filenames (!!!) and oh dear oh dear I can't see this ever getting implemented. I'm frankly surprised Al Viro didn't turn up on the thread with some choice words for the people proposing such an overengineered disaster.

Even ignoring everything else, readdir() doesn't actually take flags, so what are they going to do? changing readdir()'s prototype is impossible and introducing a readdir2() with an extra flags argument is more or less useless because only those programs whose authors think about the problem of weird filenames would ever use it, but if people *did* think about that problem, weird filenames wouldn't cause trouble in the first place because everyone would handle them fine. If you had the flag default on unless readdir2() was used to turn it off, this would just turn filenames containing linefeeds into an attack vector on *everything*, not just on shell scripts, as others on the same bug pointed out. It's not like anything is really prepared for readdir() to fail, either -- people tend to assume readdir() failure means disk damage or some other catastrophic problem, so any directory with a filename in it inducing such a failure would be more or less entirely unusable. Bingo, instant DoS attack vector.