bug#40142: CVE checker return false positives

bug#40142: CVE checker return false positives

[Brice Waegeneire]

Hello,

The CVE checker of “guix lint” returns false positives:
┌────
│ LANGUAGE=C guix lint git 2>&1
├───
│ gnu/packages/version-control.scm:149:2: git@2.25.1: probably 
vulnerable to CVE-2020-2136, CVE-2019-1003010, CVE-2018-1000110, 
CVE-2018-1000182
│ 
/gnu/store/8q0nfd6vnc6lnjh13rwl7fyimwlv7fml-guix-module-union/share/guile/site/3.0/gnu/packages/version-control.scm:153:12: 
git@2.25.1: can be upgraded to 2.25.2
│ 
/gnu/store/8q0nfd6vnc6lnjh13rwl7fyimwlv7fml-guix-module-union/share/guile/site/3.0/gnu/packages/version-control.scm:154:11: 
git@2.25.1: source not archived on Software Heritage
└────


• [CVE-2020-2136]: “Jenkins Git Plugin 4.2.0 and earlier […]”
• [CVE-2019-1003010]: “[…] Jenkins Git Plugin 3.9.1 and earlier […]”
• [CVE-2018-1000110]: “[…] Jenkins Git Plugin version 3.7.0 and earlier
   […]”
• [CVE-2018-1000182]: “[…] Jenkins Git Plugin 3.9.0 and older […]”

Also note the missing / on the first line and it output on `stderr'
instead of `stdout'.

[CVE-2020-2136] <https://nvd.nist.gov/vuln/detail/CVE-2020-2136>

[CVE-2019-1003010] <https://nvd.nist.gov/vuln/detail/CVE-2019-1003010>

[CVE-2018-1000110] <https://nvd.nist.gov/vuln/detail/CVE-2018-1000110>

[CVE-2018-1000182] <https://nvd.nist.gov/vuln/detail/CVE-2018-1000182>

Brice.

bug#40142: CVE checker return false positives

[Ludovic Courtès]

Hi,

Brice Waegeneire <brice@waegenei.re> skribis:

> The CVE checker of “guix lint” returns false positives:
> ┌────
> │ LANGUAGE=C guix lint git 2>&1
> ├───
> │ gnu/packages/version-control.scm:149:2: git@2.25.1: probably
> vulnerable to CVE-2020-2136, CVE-2019-1003010, CVE-2018-1000110,
> CVE-2018-1000182

[...]

> • [CVE-2020-2136]: “Jenkins Git Plugin 4.2.0 and earlier […]”
> • [CVE-2019-1003010]: “[…] Jenkins Git Plugin 3.9.1 and earlier […]”
> • [CVE-2018-1000110]: “[…] Jenkins Git Plugin version 3.7.0 and earlier
>   […]”
> • [CVE-2018-1000182]: “[…] Jenkins Git Plugin 3.9.0 and older […]”

(guix cve) reports it as applying to “git”:

--8<---------------cut here---------------start------------->8---
scheme@(guix cve)> (define items
		     (call-with-decompressed-port 'gzip (http-fetch (yearly-feed-uri 2020))
		       json->cve-items))
scheme@(guix cve)> (find (lambda (item)
			   (string=? (cve-id (cve-item-cve item)) "CVE-2020-2136"))
			 items)
$130 = #<<cve-item> cve: #<<cve> id: "CVE-2020-2136" data-type: CVE data-format: MITRE references: (#<<cve-reference> url: "http://www.openwall.com/lists/oss-security/2020/03/09/1" tags: ("Third Party Advisory")> #<<cve-reference> url: "https://jenkins.io/security/advisory/2020-03-09/#SECURITY-1723" tags: ("Vendor Advisory")>)> configurations: (("git" (<= "4.2.0"))) published-date: #<date nanosecond: 0 second: 0 minute: 15 hour: 16 day: 9 month: 3 year: 2020 zone-offset: 0> last-modified-date: #<date nanosecond: 0 second: 0 minute: 4 hour: 20 day: 9 month: 3 year: 2020 zone-offset: 0>>
--8<---------------cut here---------------end--------------->8---

I think the problem stems from the fact that the CVE configuration
specify “jenkins:git” (where “jenkins” is the “vendor” and “git” is the
“product”), but we just strip the vendor part:

--8<---------------cut here---------------start------------->8---
$ wget -O - -q https://nvd.nist.gov/feeds/json/cve/1.1/nvdcve-1.1-2020.json.gz| gunzip | jq

[…]

      "configurations": {
        "CVE_data_version": "4.0",
        "nodes": [
          {
            "operator": "OR",
            "cpe_match": [
              {
                "vulnerable": true,
                "cpe23Uri": "cpe:2.3:a:jenkins:git:*:*:*:*:*:jenkins:*:*",
                "versionEndIncluding": "4.2.0"
              }
            ]
          }
        ]
--8<---------------cut here---------------end--------------->8---

It’s usually the case that the vendor part has little relevance for free
software packages, but in this case it does make a difference.

Probably the fix would be to preserve the vendor part in the API and to
somehow use it meaningfully.

Ideas & patches welcome!

> Also note the missing / on the first line and it output on `stderr'
> instead of `stdout'.

What do you mean?

Thanks,
Ludo’.

bug#40142: CVE checker return false positives

[Brice Waegeneire]

Hello,

On 2020-03-21 16:25, Ludovic Courtès wrote:
> Probably the fix would be to preserve the vendor part in the API and to
> somehow use it meaningfully.
> 
> Ideas & patches welcome!

I'll see what I can write a patch to fix it then.

>> Also note the missing / on the first line and it output on `stderr'
>> instead of `stdout'.
> 
> What do you mean?

I misunderstood the meaning of “gnu/packages/version-control.scm:149:2:”
and thought there was a missing / before “gnu/”; this is irrelevant. 
About
the output stream of “guix lint” I think it should output to `stdout', 
not
`stderr' as it's currently the case.

Brice.

bug#40142: (guix cve) discards configuration "vendor", leading to false positives

[Brice Waegeneire]

Hello,

I have thought of a way to improve on those false positives. And I have
submitted a patch to solve the stderr situation at
https://issues.guix.info/issue/40367.

> Probably the fix would be to preserve the vendor part in the API and to
> somehow use it meaningfully

It looks like, for most free software the name of the software is used 
as
  the vendor too, but I'm guessing that's not always the case in 
particular
  when two project are using the same name. So we can't just filter the
  entries where the vendor name isn't the name of the package or we could
  end up with false negatives which seems worse than false positive for a
  vulnerability checker.

One solution would be to display the name of the vendor when it doesn't
correspond to the name of the package. Such solution would still output
false positives but at least it will be quicker to identify then as 
such,
compared to looking up and reading trough each CVE.

- Brice

bug#40142: (guix cve) discards configuration "vendor", leading to false positives

[Ludovic Courtès]

Hi,

Brice Waegeneire <brice@waegenei.re> skribis:

> It looks like, for most free software the name of the software is used
> as
>  the vendor too, but I'm guessing that's not always the case in
> particular
>  when two project are using the same name. So we can't just filter the
>  entries where the vendor name isn't the name of the package or we could
>  end up with false negatives which seems worse than false positive for a
>  vulnerability checker.

Yeah.

> One solution would be to display the name of the vendor when it doesn't
> correspond to the name of the package. Such solution would still output
> false positives but at least it will be quicker to identify then as
> such,
> compared to looking up and reading trough each CVE.

Yes, though I think that (guix cve) should simply preserve the vendor
part, and leave it up to its user, ‘guix lint’, to display vendor
mismatches.

Thanks,
Ludo’.