Re: Best way forward for CVE-2021-22876/curl?

Hi Utkarsh, all

After reading the description of this CVE again I realize that I misunderstood the description last time.

The problem is that the "referrer" header is not stripped.

This changes my conclusion to some extent.

I see no problem with fixing this issue from a regression point of view (apart from what has already been expressed).

The amount of services that rely on the referrer field should be small, if any.

I still think we can ignore it though with the same reasoing as I expressed in the last email. The problem should be minor. There are other worse problem by providing sensitive data in the URL.

And again if the attacker can make a redirect, the attacker can most likely get the URL anyway so then nothing has leaked.

Cheers

// Ola

On Wed, 7 Apr 2021 at 13:19, Ola Lundqvist <ola@inguza.com> wrote:

Hi Utkarsh, all

Is this even a vulnerability?
The problem is that authentication information is not stripped if the browser is redirected to another place.

If you trust a site enough to provide authentication data, I guess you also trust that if that site happens to be relocated you should also trust the new place.
I mean if the attacker has the power to redirect I expect that it has the power to read the authentication data anyway. There could be cases when this is not the case, but in general it should not be possible for the attacker to redirect without also having more power.

We could of course consider to apply this fix, but it certainly will cause a regression since my expectation is that authentication information is forwarded.

I think it should be ignored. If we fix it, it should be with a configuration option, but I think that is a little too intrusive for (E)LTS.

Or have I missed something?

Best regards

// Ola

On Mon, 5 Apr 2021 at 02:20, Utkarsh Gupta <utkarsh@debian.org> wrote:
Hello,

[CCing the Security team in case they have some ideas or suggestions
for CVE-2021-22876/curl]

Whilst triaging and looking thoroughly for this CVE, affecting curl, I
noticed that the upstream patch uses elements like CURLU,
CURLUPART_{URL,FRAGMENT,USER,PASSWORD}. This comes from the URL API
which seems to be missing in both, stretch and jessie.

There seem to be two plausible options at this point:

1. Given that this CVE has been assigned low severity by upstream, we
could perhaps mark this as no-dsa or ignored, with an appropriate
comment; or

2. Backport the entire URL API (patch for that is attached; is
intrusive) and then apply the fix for CVE-2021-22876 (patch attached)
on top of that. Whilst this option makes sense, but backporting the
entire URL API could have an unforeseen effect (or chances of
potential regressions) and in any case, looks somewhat intrusive.

So for now, I've added curl to dla-needed and ela-needed but if we
decide to mark this as no-dsa or ignored, we could simply drop this
from there as this is the only CVE that needs working on.

Let me know what y'all think.

- u

--
--- Inguza Technology AB --- MSc in Information Technology ----
| ola@inguza.com opal@debian.org |
| http://inguza.com/ Mobile: +46 (0)70-332 1551 |
---------------------------------------------------------------