Re: org-crypt leaking data when encryption password is not entered twice (was: Please document the caching and its user options)

[Daniel Clemente]

>
> As for not typing the same password twice and not using
> org-crypt-use-before-save-magic, we should somehow fix this.
> (I am starting a new thread branch.)
>

„Not using org-crypt-use-before-save-magic“ is currently a user
decision, not a bug.
For instance, I don't use it because it adds around 5 seconds to each
saving of a large file. If it were instantaneous I would enable it.
With it disabled, this explains why I often find unencrypted sections
at the end of the day… I have to rely on myself to reencrypt them
again.



> One simple idea is to disable backups if encryption fails.
> Or use `write-contents-functions' instead of `before-save-hook' - that
> way, Emacs will not ignore errors thrown by org-crypt and will not
> actually save anything if encryption fails.
>

Disabling backups makes sense too, if we decide that unencrypted
private data shouldn't end up in backups.
I don't have an absolute opinion. Some people may prefer having
backups of all data (including private unencrypted data).

If it's possible to detect whether encryption failed in this buffer,
there could be a warning saying „Last encryption failed. Really
save?“.
Or just a message in the style of „Encryption failed. Saving the file
may store unencrypted data in disk, and in backups and cache if
enabled“.

Totally preventing the user from saving a file seems harsh but it also
seems safer. Since users have different safety preferences, Emacs can
let the user decide what the do, through a question or optional
setting.


> > At the end of the day when I do "git diff" + "git commit" sometimes I
> > realize there's unencrypted data and then I have to reencrypt it. In
> > the meantime I might have killed and reopened the buffer, thus
> > updating the file cache.
> > That may be a problem by org-encrypt and something to document in
> > org-crypt itself. The point is that users of org-encrypt should take
> > extra precautions when enabling org-element-cache-persistent. Like:
> > not closing buffers while the sections are unencrypted.
>
> These things should be considered bugs. And we should fix them. Cache and
> other libraries should not be responsible for special treatment of
> optional org-crypt library.
>

You can't fix all bugs all the time, so you can't base security on „we
strongly believe there are no more bugs“. If doing an extra
verification (to avoid storing private data on disk in unencrypted
form) is fast, it's better with the verification.

In addition, „leaving some encrypted sections unencrypted for a short
amount of time, and closing and reopening the buffer during that time“
isn't a bug, it's a possible user behaviour that we can't control. But
org-crypt can mention that that behaviour is unsafe when using on-disk
cache. Or detect it (if it's fast) and warn the user.

> Cache and
> other libraries should not be responsible for special treatment of
> optional org-crypt library.

That's arbitrary. Both persistent cache and org-crypt are optional,
but any of them can check whether the other is enabled and try to do
what the user wants.
I know they both have separate responsibilities, but if there are only
these 2 parts, one of them must be the one caring about „unencrypted
data leaking into disk caches“.

It would be different If we had a third component… E.g. imagine we had
a component/overlay/text property/… in Emacs that could tell whether a
buffer's region contains very private information or not; then all
other components could just obey that setting (that section won't be
backed up, it won't end up in disk cache, … It can even be displayed
in a different face). Then org-crypt just needs to set that flag when
encryption fails. Does something like that exist? Anyway this is a bit
utopic or overengineered. Simpler ways of improving things are with
documentation (e.g. „Don't do this, it's unsafe“), with messages
(„You're doing this, which may be unsafe“), or with questions („Really
do this unsafe thing?“)