Skip to contentSkip to conversation

libarchive: mark CVE-2026-14164 as not-applicable-platform

PATCH In review [email protected] 2 participants 3 messages Aug 17 – Sep 2

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 libarchive: mark CVE-2026-14164 as not-applicable-platform

From: Daniel Turull <[email protected]>

- The RAR5 double-free in init_unpack() is a regression introduced upstream by commit 620bdafa on 2026-05-16 and existed only on the git master branch until it was fixed by PR #3071 (commit 1c914cdf) on 2026-05-24. It was never part of an upstream release.
- The upstream release tarballs (3.6.x/3.7.x/3.8.6) use the older init_unpack() with unchecked calloc and no early-return path, so the freed window_buf/filtered_buf pointers are never left dangling and the double-free cannot occur.

References:
https://nvd.nist.gov/vuln/detail/CVE-2026-14164

---
 meta/recipes-extended/libarchive/libarchive_3.8.7.bb | 4 ++++
 1 file changed, 4 insertions(+)
meta/recipes-extended/libarchive/libarchive_3.8.7.bb +4 -0
index e8c3a3bfe3..0926acd8f9 100644
@@ -92,3 +92,7 @@ RDEPENDS:${PN}-ptest += "bsdtar bsdcpio"
 CVE_STATUS[CVE-2026-4426] = "fixed-version: fixed since 3.8.7"
 CVE_STATUS[CVE-2026-5121] = "fixed-version: fixed since 3.8.7"
 CVE_STATUS[CVE-2026-5745] = "fixed-version: fixed since 3.8.6"
+CVE_STATUS[CVE-2026-14164] = "not-applicable-platform: Double-free regression in the RAR5\
+ reader's init_unpack() was introduced upstream by commit 620bdafa (2026-05-16) and existed\
+ only on the git master branch until the fix in PR #3071 (commit 1c914cdf, 2026-05-24). It was\
+ never part of an upstream release tarball."
#1 lore raw
Quoted 24 lines · Daniel Turull via lists.openembedded.org

From: Daniel Turull <[email protected]>

- The RAR5 double-free in init_unpack() is a regression introduced upstream by commit 620bdafa on 2026-05-16 and existed only on the git master branch until it was fixed by PR #3071 (commit 1c914cdf) on 2026-05-24. It was never part of an upstream release.
- The upstream release tarballs (3.6.x/3.7.x/3.8.6) use the older init_unpack() with unchecked calloc and no early-return path, so the freed window_buf/filtered_buf pointers are never left dangling and the double-free cannot occur.

References:
https://nvd.nist.gov/vuln/detail/CVE-2026-14164

---
 meta/recipes-extended/libarchive/libarchive_3.8.7.bb | 4 ++++
 1 file changed, 4 insertions(+)
meta/recipes-extended/libarchive/libarchive_3.8.7.bb +0 -0
index e8c3a3bfe3..0926acd8f9 100644

meta/recipes-extended/libarchive/libarchive_3.8.7.bb RDEPENDS:${PN}-ptest += "bsdtar bsdcpio" · line 98

Quoted 7 lines
meta/recipes-extended/libarchive/libarchive_3.8.7.bb +4 -0
@@ -92,3 +92,7 @@ RDEPENDS:${PN}-ptest += "bsdtar bsdcpio"
 CVE_STATUS[CVE-2026-4426] = "fixed-version: fixed since 3.8.7"
 CVE_STATUS[CVE-2026-5121] = "fixed-version: fixed since 3.8.7"
 CVE_STATUS[CVE-2026-5745] = "fixed-version: fixed since 3.8.6"
+CVE_STATUS[CVE-2026-14164] = "not-applicable-platform: Double-free regression in the RAR5\
+ reader's init_unpack() was introduced upstream by commit 620bdafa (2026-05-16) and existed\
+ only on the git master branch until the fix in PR #3071 (commit 1c914cdf, 2026-05-24). It was\
+ never part of an upstream release tarball."

Hello,

I don't think not-applicable-platform is the right flag for this CVE_STATUS. See cve-check-map.conf[0]:

Quoted 1 line

[not-applicable-platform] use when vulnerability affects other platform (e.g. Windows or Debian)

How about "fixed-version"?

Regards,

[0]: https://git.openembedded.org/openembedded-core/tree/meta/conf/cve-check-map.conf?h=wrynose

Signature · 2 lines
Yoann Congal
Smile ECS
#2 lore raw
Quoted 36 lines · Yoann Congal
Quoted 25 lines · Daniel Turull via lists.openembedded.org

From: Daniel Turull <[email protected]>

- The RAR5 double-free in init_unpack() is a regression introduced upstream by commit 620bdafa on 2026-05-16 and existed only on the git master branch until it was fixed by PR #3071 (commit 1c914cdf) on 2026-05-24. It was never part of an upstream release.
- The upstream release tarballs (3.6.x/3.7.x/3.8.6) use the older init_unpack() with unchecked calloc and no early-return path, so the freed window_buf/filtered_buf pointers are never left dangling and the double-free cannot occur.

References:

---
 meta/recipes-extended/libarchive/libarchive_3.8.7.bb | 4 ++++

 1 file changed, 4 insertions(+)
meta/recipes- renamed from meta/recipes-extended/libarchive/libarchive_3.8.7.bb +0 -0

extended/libarchive/libarchive_3.8.7.bb
index e8c3a3bfe3..0926acd8f9 100644

meta/recipes-extended/libarchive/libarchive_3.8.7.bb +0 -0
@@ -92,3 +92,7 @@ RDEPENDS:${PN}-ptest += "bsdtar bsdcpio"
Quoted 7 lines

 CVE_STATUS[CVE-2026-4426] = "fixed-version: fixed since 3.8.7" CVE_STATUS[CVE-2026-5121] = "fixed-version: fixed since 3.8.7" CVE_STATUS[CVE-2026-5745] = "fixed-version: fixed since 3.8.6" +CVE_STATUS[CVE-2026-14164] = "not-applicable-platform: Double-free regression in the RAR5\
+ reader's init_unpack() was introduced upstream by commit 620bdafa (2026-05-16) and existed\
+ only on the git master branch until the fix in PR #3071 (commit 1c914cdf, 2026-05-24). It was\
+ never part of an upstream release tarball."

Hello,

Quoted 5 lines

I don't think not-applicable-platform is the right flag for this CVE_STATUS. See cve-check-map.conf[0]:

Quoted 1 line

[not-applicable-platform] use when vulnerability affects other platform (e.g. Windows or Debian)

How about "fixed-version"?

Sure. I'll send a v2 with it. also for scarthgap

Daniel