[Libreoffice-bugs] [Bug 114878] Add option to CSV import to disable formula injection

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=114878

--- Comment #11 from Martin Häcker <spamfaenger at gmx.de> ---
I certainly didn't find this bug report before reporting mine. I'd like to
bring over some of my comments there so they are not lost in the duplicate.

-- snip --
Steps to Reproduce:
1. Create data only css file

-- snip --
,0
0,=SUMME(1;1)
1,=WEBDIENST("http://localhost:8000")
-- snap --

2. Open with  German version of Calc (Screenshot 'import dialog with
preview.png'). Observe that the preview renders all the formulas as _DATA_ as
it should be.
3. Click 'import'
4. Observe Screenshot 'imported.png'

Actual Results:
The two fields are not rendered as previewed, instead they are assumed to be
formulas and are executed. Luckily there seems to be a security safeguard that
at least blocks the http call from immediate execution. However even this block
is removed by a single click on the notice at the top of the window.

Expected Results:
I have imported a CSV file (which is a data only format), watched the file
beforehand in a text editor to see what I will be getting, watched the preview
for correctness and am still not getting the import that was previewed. This is
highly surprising and als a huge enabler for a full class of security problems.

If I want the data to be interpreted and changed by Libre Office Calc, that
needs to be a separate (off by default) check box that warns about the problems
and security risks this poses - especially if the preview is not complete and
therefore does not allow me to assess what checking this box would exactly do.

Several problems I see here:

a) The preview should match the actual imported data
b) It is highly surprising that importing a data only format will suddenly
interpret that data and not display what is in the file. This is especially
problematic if a web application exports data, that contains user controlled
inputs to exchange it to other applications and it gets imported in Calc at
some stage. The only workaround available is to know at export time, where the
file will be imported in later, so the export can be sanitised for the
importing application. This is highly unpractical and has a high likelihood of
data loss / unintended data changes if the file is imported in the wrong
application.
c) This is also highly surprising when one investigates the RFC for CSV:
<https://datatracker.ietf.org/doc/html/rfc4180> which states:

   Security considerations:

      CSV files contain passive text data that should not pose any
      risks.  However, it is possible in theory that malicious binary
      data may be included in order to exploit potential buffer overruns
      in the program processing CSV data.  Additionally, private data
      may be shared via this format (which of course applies to any text
      data).

This has many and quite surprising security considerations - so much so, that
OWASP maintains it as it's own category of security problem:
<https://owasp.org/www-community/attacks/CSV_Injection>.

I learned of this because the German Corona Tracing App Luca was attacked
through this vector - but also users of web applications I maintain are
attackable by this problem.

I understand that this Is probably a long running convention for CSV import and
has an aspect of compatibility with other spreadsheet applications. However
this is a problematic behaviour for which there is no workaround when importing
data into Calc, and there needs to be a strategy for fixing - but at least
allowing a workaround for this.

I would like to suggest going at this in a multi step process - quite possibly
stretched out over a long period. Maybe even 5-10 years - but of course I would
like a faster transition period.

My suggestion is:

1. Add a setting on import that at least allows forcing Libre Office Calc to
interpret all imported data literally so there is at least a workaround
available immediately.
2. After some time, start warning on the import preview if the imported data
contains anything that LibreOffice would like to interpret (At least formulas,
but probably also data that could be auto formatted). This should explain the
problem and/or link to a website that explains the problem and the security
concerns.
3. After some more time, switch on this option by default and instead warn if
the imported data contains interpretable data. Maybe show a preview of what the
interpretation would change to allow the user to understand what this would do.

That way impact on existing users of that feature can be minimised, while still
there is at least an immediate workaround available. The time bought by this
measures can then be used to create the other suggested import features to make
the transition to not interpreting imported CSV data by default safe for
everyone.
-- snap --

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20210528/17751dab/attachment.htm>