strategy for curl CVEs
Hello
Currently I am trying to fix some CVEs related to curl and figured out that the code base has evolved quite a bit. Which makes makes backporting patches a lot harder especially if used function in the patch are not even there in the 8.7.1 codebase. This in combination with a changed test framework (the last patches for CVE fixes were already without tests) makes it hard to backport and at the same time ensure that the CVE is really fixed.
Now my question would be if it still makes sense to try to backport some CVEs or if there will be a irregular major/minor upgrade anytime soon?
An example for such a CVE would be CVE-2026-11856 [1] and the upstream patch [2]. Here the structs like Curl_creds or Curl_peer and the corresponding functions do not even exist.
[1] https://curl.se/docs/CVE-2026-11856.html
[2] https://github.com/curl/curl/commit/5c6b4880357ab3e72967c1c45c
Patrick
Quoted 14 lines · Patrick Vogelaar via lists.openembedded.org
Hello
Currently I am trying to fix some CVEs related to curl and figured out that the code base has evolved quite a bit. Which makes makes backporting patches a lot harder especially if used function in the patch are not even there in the 8.7.1 codebase. This in combination with a changed test framework (the last patches for CVE fixes were already without tests) makes it hard to backport and at the same time ensure that the CVE is really fixed.
Now my question would be if it still makes sense to try to backport some CVEs or if there will be a irregular major/minor upgrade anytime soon?
Quoted 6 lines
An example for such a CVE would be CVE-2026-11856 [1] and the upstream patch [2]. Here the structs like
Curl_credsorCurl_peerand the corresponding functions do not even exist.[1] https://curl.se/docs/CVE-2026-11856.html
[2] https://github.com/curl/curl/commit/5c6b4880357ab3e72967c1c45c
Hello Patrick,
In OE-core, we do have a quite strict stable policy about what can or can't be included in stable branches:
https://wiki.yoctoproject.org/wiki/Stable_Release_and_LTS#Stable/LTS_Patch_Acceptance_Policies
Since the curl version after the scarthgap curl 8.7.1 is curl 8.8.0. But this version includes feature addition:
https://curl.se/ch/8.8.0.html
Quoted 9 lines
Changes:
curl_version_info: provide librtmp version
file: add support for directory listings
idn: add native AppleIDN (icucore) support for macOS/iOS
lib: addcurl_multi_waitfds
mbedTLS: implementCURLOPT_SSL_CIPHER_LISToption
NTLM_WB: drop support
TLS: add support for ECH (Encrypted Client Hello)
urlapi: addCURLU_GET_EMPTYfor empty queries and fragments
So, I can't accept this upgrade so we have to rely on backports.
IMHO, dropping tests during backport is fair.
So, yes, it still makes sense to try to backport CVE fixes.
Of course, at some point, the difference in codebases will be too much and the backport will be impossible under our stable policy. Then the fix will have to come from outside of oe-core (downstream layer, mixin layers, ...)
Regards,
Signature · 2 lines
Yoann Congal Smile ECS