[PATCH] Fix cve-check false negative


Geoffrey GIRY
 

Fixes [YOCTO #14127]

NVD DB store version and update in the same value, separated by '_'.
The proposed patch check if the version from NVD DB contains a "_",
ie 9.2.0_p1 is convert to 9.2.0p1 before version comparison.

Reviewed-by: Yoann CONGAL <yoann.congal@...>
Signed-off-by: Geoffrey GIRY <geoffrey.giry@...>
---
meta/classes/cve-check.bbclass | 5 ++-
meta/lib/oe/cve_check.py | 39 +++++++++++++++++++++++
meta/lib/oeqa/selftest/cases/cve_check.py | 19 +++++++++++
3 files changed, 62 insertions(+), 1 deletion(-)

diff --git a/meta/classes/cve-check.bbclass b/meta/classes/cve-check.bbclass
index 41fdf8363f..5e2da56046 100644
--- a/meta/classes/cve-check.bbclass
+++ b/meta/classes/cve-check.bbclass
@@ -260,7 +260,7 @@ def check_cves(d, patched_cves):
"""
Connect to the NVD database and find unpatched cves.
"""
- from oe.cve_check import Version
+ from oe.cve_check import Version, convert_cve_version

pn = d.getVar("PN")
real_pv = d.getVar("PV")
@@ -324,6 +324,9 @@ def check_cves(d, patched_cves):
if cve in cve_ignore:
ignored = True

+ version_start = convert_cve_version(version_start)
+ version_end = convert_cve_version(version_end)
+
if (operator_start == '=' and pv == version_start) or version_start == '-':
vulnerable = True
else:
diff --git a/meta/lib/oe/cve_check.py b/meta/lib/oe/cve_check.py
index 4f1d80f050..dbaa0b373a 100644
--- a/meta/lib/oe/cve_check.py
+++ b/meta/lib/oe/cve_check.py
@@ -179,3 +179,42 @@ def update_symlinks(target_path, link_path):
if os.path.exists(os.path.realpath(link_path)):
os.remove(link_path)
os.symlink(os.path.basename(target_path), link_path)
+
+
+def convert_cve_version(version):
+ """
+ This function converts from CVE format to Yocto version format.
+ eg 8.3_p1 -> 8.3p1, 6.2_rc1 -> 6.2-rc1
+
+ Unless it is redefined using CVE_VERSION in the recipe,
+ cve_check uses the version in the name of the recipe (${PV})
+ to check vulnerabilities against a CVE in the database downloaded from NVD.
+
+ When the version has an update, i.e.
+ "p1" in OpenSSH 8.3p1,
+ "-rc1" in linux kernel 6.2-rc1,
+ the database stores the version as version_update (8.3_p1, 6.2_rc1).
+ Therefore, we must transform this version before comparing to the
+ recipe version.
+
+ In this case, the parameter of the function is 8.3_p1.
+ If the version uses the Release Candidate format, "rc",
+ this function replaces the '_' by '-'.
+ If the version uses the Update format, "p",
+ this function removes the '_' completely.
+ """
+ import re
+
+ matches = re.match('^([0-9.]+)_((p|rc)[0-9]+)$', version)
+
+ if not matches:
+ return version
+
+ version = matches.group(1)
+ update = matches.group(2)
+
+ if matches.group(3) == "rc":
+ return version + '-' + update
+
+ return version + update
+
diff --git a/meta/lib/oeqa/selftest/cases/cve_check.py b/meta/lib/oeqa/selftest/cases/cve_check.py
index ac47af1990..9534c9775c 100644
--- a/meta/lib/oeqa/selftest/cases/cve_check.py
+++ b/meta/lib/oeqa/selftest/cases/cve_check.py
@@ -54,6 +54,25 @@ class CVECheck(OESelftestTestCase):
self.assertTrue( result ,msg="Failed to compare version with suffix '1.0_patch2' < '1.0_patch3'")


+ def test_convert_cve_version(self):
+ from oe.cve_check import convert_cve_version
+
+ # Default format
+ self.assertEqual(convert_cve_version("8.3"), "8.3")
+ self.assertEqual(convert_cve_version(""), "")
+
+ # OpenSSL format version
+ self.assertEqual(convert_cve_version("1.1.1t"), "1.1.1t")
+
+ # OpenSSH format
+ self.assertEqual(convert_cve_version("8.3_p1"), "8.3p1")
+ self.assertEqual(convert_cve_version("8.3_p22"), "8.3p22")
+
+ # Linux kernel format
+ self.assertEqual(convert_cve_version("6.2_rc8"), "6.2-rc8")
+ self.assertEqual(convert_cve_version("6.2_rc31"), "6.2-rc31")
+
+
def test_recipe_report_json(self):
config = """
INHERIT += "cve-check"
--
2.30.2


Marta Rybczynska
 



On Tue, Mar 28, 2023 at 12:24 PM Geoffrey GIRY <geoffrey.giry@...> wrote:
Fixes [YOCTO #14127]

NVD DB store version and update in the same value, separated by '_'.
The proposed patch check if the version from NVD DB contains a "_",
ie 9.2.0_p1 is convert to 9.2.0p1 before version comparison.


Thank you for the patch. Which layers (and world builds) have you verified it with?
I'm asking because versioning is always a complicated problems with frequent exceptions to all "rules".

Kind regards,
Marta


Geoffrey GIRY
 

Hello Marta,

We only tested core-image-minimal and some recipes that use the update
and release candidate formats (pX and -rcX)

Geoffrey GIRY
SMILE ECS - R&D Engineer

Le mer. 29 mars 2023 à 06:45, Marta Rybczynska <rybczynska@...> a écrit :

On Tue, Mar 28, 2023 at 12:24 PM Geoffrey GIRY <geoffrey.giry@...> wrote:

Fixes [YOCTO #14127]

NVD DB store version and update in the same value, separated by '_'.
The proposed patch check if the version from NVD DB contains a "_",
ie 9.2.0_p1 is convert to 9.2.0p1 before version comparison.
Thank you for the patch. Which layers (and world builds) have you verified it with?
I'm asking because versioning is always a complicated problems with frequent exceptions to all "rules".

Kind regards,
Marta


Marta Rybczynska
 

Hello Geoffrey,
Would it be possible to run it over the world build of oe-core and possibly meta-oe ?

My build farm will be available only next week and I would like to know if there are unexpected changes.

Kind regards,
Marta


On Wed, Mar 29, 2023 at 3:31 PM Geoffrey GIRY <geoffrey.giry@...> wrote:
Hello Marta,

We only tested core-image-minimal and some recipes that use the update
and release candidate formats (pX and -rcX)

Geoffrey GIRY
SMILE ECS - R&D Engineer

Le mer. 29 mars 2023 à 06:45, Marta Rybczynska <rybczynska@...> a écrit :
>
> On Tue, Mar 28, 2023 at 12:24 PM Geoffrey GIRY <geoffrey.giry@...> wrote:
>>
>> Fixes [YOCTO #14127]
>>
>> NVD DB store version and update in the same value, separated by '_'.
>> The proposed patch check if the version from NVD DB contains a "_",
>> ie 9.2.0_p1 is convert to 9.2.0p1 before version comparison.
>>
>
> Thank you for the patch. Which layers (and world builds) have you verified it with?
> I'm asking because versioning is always a complicated problems with frequent exceptions to all "rules".
>
> Kind regards,
> Marta


Alexandre Belloni
 

Hello Marta,

On 31/03/2023 09:48:27+0200, Marta Rybczynska wrote:
Hello Geoffrey,
Would it be possible to run it over the world build of oe-core and possibly
meta-oe ?
It has already run successfully and is already merged.

My build farm will be available only next week and I would like to know if
there are unexpected changes.

Kind regards,
Marta

On Wed, Mar 29, 2023 at 3:31 PM Geoffrey GIRY <geoffrey.giry@...>
wrote:

Hello Marta,

We only tested core-image-minimal and some recipes that use the update
and release candidate formats (pX and -rcX)

Geoffrey GIRY
SMILE ECS - R&D Engineer

Le mer. 29 mars 2023 à 06:45, Marta Rybczynska <rybczynska@...> a
écrit :

On Tue, Mar 28, 2023 at 12:24 PM Geoffrey GIRY <geoffrey.giry@...>
wrote:

Fixes [YOCTO #14127]

NVD DB store version and update in the same value, separated by '_'.
The proposed patch check if the version from NVD DB contains a "_",
ie 9.2.0_p1 is convert to 9.2.0p1 before version comparison.
Thank you for the patch. Which layers (and world builds) have you
verified it with?
I'm asking because versioning is always a complicated problems with
frequent exceptions to all "rules".

Kind regards,
Marta



--
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com


Richard Purdie
 

On Fri, 2023-03-31 at 10:59 +0200, Alexandre Belloni via
lists.openembedded.org wrote:
Hello Marta,

On 31/03/2023 09:48:27+0200, Marta Rybczynska wrote:
Hello Geoffrey,
Would it be possible to run it over the world build of oe-core and possibly
meta-oe ?
It has already run successfully and is already merged.
It has merged but I think Marta's question is a valid one. The
autobuilder doesn't test this.

I'd note that our patchmetrics do for OE-Core:

https://autobuilder.yocto.io/pub/non-release/patchmetrics/cve-status-master.txt

and those don't look worse as a result. That doesn't cover meta-oe
though.

Cheers,

Richard


Geoffrey GIRY
 

Hello,

Marta Rybczynska wrote:
Would it be possible to run it over the world build of oe-core and possibly meta-oe ?
I tried the following:

The command `bibake -c cve_check world` reports the same CVE with and
without the patch applied.
I did test for oe-core alone, and found the same results as the autobuilder.
I also did test with meta-oe and found the same result (many more CVE,
but the same appears with and without the patch).

Sincerely,
Geoffrey GIRY
SMILE ECS - R&D Engineer