Backport six commits from upstream glibc to fix CVE-2026-6238.
4ba0b79b95 resolv: Add test case tst-ns_sprintrr (bug 34033, bug 34069)
a7b60d23bb resolv: Fix buffer overreads in ns_sprintrrf (CVE-2026-6238)
cd0db208d5 resolv: Check for inet_ntop failure in ns_sprintrrf
d58415eb17 resolv: Improve formatting of unknown records in ns_sprintrrf
f69b7f95e3 resolv: Fix ns_sprintrrf formatting of class, type values (bug 34289)
360f352c9a resolv: Declare __p_class_syms, __p_type_syms for internal use
The upstream patch series [PATCH 0/5] contains five commits:
1/5: Update GLIBC-SA-2026-0012 to mention A6 records (doc only)
2/5: resolv: Check for inet_ntop failure in ns_sprintrrf
3/5: resolv: Remove incorrect parts of TSIG handling from ns_sprintrrf
(CVE-2026-5435)
4/5: resolv: Fix buffer overreads in ns_sprintrrf (CVE-2026-6238)
5/5: resolv: Add test case tst-ns_sprintrr (bug 34033, bug 34069)
For this backport:
- Patch 1/5 is skipped (documentation-only change to advisories,
upstream glibc itself does not backport this to older releases)
- Patch 3/5 (CVE-2026-5435) is already patched in scarthgap sources
- Patches 2/5, 4/5, and 5/5 are backported as:
0028-CVE-2026-6238-0004.patch (inet_ntop failure check)
0029-CVE-2026-6238-0005.patch (buffer overread fix - CVE-2026-6238)
0030-CVE-2026-6238-0006.patch (test case for bug 34033, bug 34069)
However, the test case (tst-ns_sprintrr) from patch 5/5 failed on
scarthgap's glibc 2.39 due to missing prerequisite commits. Three
additional patches were backported to resolve the test failure:
0025-CVE-2026-6238-0001.patch (Declare __p_class_syms, __p_type_syms for internal)
0026-CVE-2026-6238-0002.patch (Fix ns_sprintrrf formatting of class, type values)
0027-CVE-2026-6238-0003.patch (Improve formatting of unknown records in ns_sprintrrf)
CVE-2026-6238 fixes buffer overreads in ns_sprintrrf affecting A6 and
LOC record handling. The vulnerable LOC record handling was introduced
before glibc 2.0, while A6 record handling was added in glibc 2.7.
Reference:
https://inbox.sourceware.org/libc-alpha/cover.1777546194.git.fweimer@redhat.com/https://nvd.nist.gov/vuln/detail/CVE-2026-6238https://sourceware.org/bugzilla/show_bug.cgi?id=34069
Testing Results:
Before After Diff
PASS 4896 4897 +1
XPASS 4 4 0
FAIL 372 372 0
XFAIL 16 16 0
UNSUPPORTED 224 224 0
Changes in testcases:
testcase-name before after
resolv/tst-ns_sprintrr(new) - PASS
commit - 4ba0b79b95 resolv: Add test case tst-ns_sprintrr (bug 34033, bug 34069)
+PASS: resolv/tst-ns_sprintrr
(From OE-Core rev: 7761a8aaac0703a738a2a4979ddb015157374dcb)
Signed-off-by: Deepesh Varatharajan <Deepesh.Varatharajan@windriver.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
[YC: fixed patch format]
Signed-off-by: Paul Barker <paul@pbarker.dev>
Using bash to handle execution parameters isn't a great idea, particularly using
os.system. In theory this could be exploited by injecting shell code through the
placeholders although whether this can be done without security tokens is unclear.
Regardless, we have been cleaning up shell=True and similar calls in the codebase.
Instead use shlex to parse the callback command, then we can iterate the parameters
to perform any expansions needed and avoid the need to call any shell expansion.
Any invalid data will then just become an invalid image target to bitbake.
Reported-by: Lei Zhang <18792670849@163.com>
(Bitbake rev: d3b32587c2314febf96897f97e0172790b4fc1cf)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Paul Barker <paul@pbarker.dev>
When SDKMACHINE is set to i686 or i586, nativesdk binaries are compiled
as 32-bit. Without -D_TIME_BITS=64 and -D_FILE_OFFSET_BITS=64, stat()
and time-related syscalls use 32-bit types, causing EOVERFLOW on
filesystems with large inode numbers (e.g. container overlay filesystems)
and Y2038 issues.
Add SDK_CC_ARCH appends for class-nativesdk:i686 and class-nativesdk:i586
using GLIBC_64BIT_TIME_FLAGS, mirroring how target architectures are
handled.
The existing GLIBC_64BIT_TIME_FLAGS:pn-glibc override only matches the
target recipe, not the class extended nativesdk-glibc, so an additional
pn-nativesdk-glibc override is needed here. This is required on
scarthgap but not on master/wrynose, because scarthgap carries glibc
2.39, which predates glibc commit a4ed0471d717 ("Always define
__USE_TIME_BITS64 when 64 bit time_t is used"). In 2.39 the header
redirect turns sigtimedwait into an alias of __sigtimedwait64, which
then collides with glibc's own weak_alias() and fails to assemble:
Error: symbol '__sigtimedwait64' is already defined
Also exclude nativesdk-mingw-w64-runtime. The MinGW runtime provides the
Windows CRT rather than glibc, and _FILE_OFFSET_BITS=64 changes its
stat declarations. That causes conflicting declarations for wstat when
building the CRT itself.
(From OE-Core rev: 07ca97baad7b182849ee50f0bd629b68dc9f53b3)
Signed-off-by: Peter Tatrai <peter.tatrai.ext@siemens.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit d9f62a45555673842021d5746437e66c40d3f3cc)
Signed-off-by: Peter Tatrai <peter.tatrai.ext@siemens.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Analysis:
- The problem only exists when curl is built with libpsl support enabled.[1]
- The recipe is built with "--without-libpsl" option.
- Hence, ignoring the CVE for this recipe.
Reference:
[1] https://curl.se/docs/CVE-2026-82209.html
(From OE-Core rev: a3804a343732b601c06167ec6414b69850992897)
Signed-off-by: Siddharth Doshi <sdoshi@mvista.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
ROOTFS_BOOTSTRAP_INSTALL in image.bbclass unconditionally stages
run-postinsts into every image. When no delayed postinsts remain,
_uninstall_unneeded() is supposed to remove it again, but the removal
call used the bare "run-postinsts" package name without applying
MLPREFIX.
On a multilib image (e.g. lib32-core-image-minimal), the package that
is actually installed is lib32-run-postinsts, so the unprefixed
removal silently matches nothing and run-postinsts leaks into the
final image manifest, along with its now-pointless init script /
systemd unit.
Expand MLPREFIX before removing the package, matching the pattern
already used elsewhere in oe-core (e.g. oe/package.py) for
package-name lookups that need to work across multilib variants.
(From OE-Core rev: 2522a1ad4fb66af0a9ca146ca7a88aece8f471da)
Signed-off-by: Kyungjik Min <dp.min@lge.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit dfa5673907bdc5211671ecdfce8ab1fb332eae48)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Adds suport for a working close_range() intercept.
(From OE-Core rev: 2d50a4b12798e57a6d7900de2cb2b6b4edcb77e9)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit 9a68d3f4bd0e80ea7e1ea745b6a10acd341dc7d8)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Pulls in various important fixes:
Makefile.in: Bump to 1.9.10
test: various: Move to makefile compilation
test-bash-exec-env: Add bash env test case
test/test-openat2-func.c: Remove unusuaed saved_errno
pseudo_util.c: strchr now returns const char
pseudo_client: remove the unused pseudo_prefix_dir_fd
pseudo_client: step around all of pseudo's own fds in closefrom
pseudo_client: step fully past pseudo's own fds when computing startfd
makewrappers: Avoid efault workaround if using AT_EMPTY_PATH
ports/linux/guts: Add __open64_2 wrapper
Makefile.in: Bump to 1.9.9
(From OE-Core rev: fc1a18d4da57dcee29861b3b5efd1924541ac71a)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit 4232a7cc5fa28828fdf3223b3c0a6432c8c1e670)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Pull in buffering fix:
"""
When running slow ptests, we can see an issue where the tests are writing output
but it doesn't make it to the process (e.g. ssh connection) running ptest-runner.
The issue is that the standard buffering for non-interactive terminals is 8kb and
some ptests don't output enough data to trigger a write. This can lead to
the controlling connection timing out.
This change forces the output streams to be line buffered in all cases. Most
ptest output would contain newlines so this should work well. stderr can be
unbuffered by default but making it consistent here seems the best approach.
Testing with this change on slow ptest runs (e.g. qemu emulated) showed much
more consistent data with this change.
"""
(From OE-Core rev: b969018b72abd96e4b504a23cb7054db4f661761)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit b57105311f1c5fd879dd58fa1de8862ebad989c1)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Changelog:
- utils.c: Fix dirname() handling in run_ptests()
- ptest-runner-collect-system-data: add info to logs
- main.c: Set PYTHONUNBUFFERED in the environment
- utils.c: print a message when a timeout occurs
- main.c: Add print_helptext() for detailed help
- main.print_usage: Remove unused argument list of -l
(From OE-Core rev: 62d96e61678773686773203216f924d34a47160b)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit 869d70d46c8e31cfc380b5143059cab3713cae26)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
This is 2 commits squashed:
| ptest-runner: upgrade 2.4.5 -> 2.5.1
|
| Signed-off-by: Wang Mingyu <wangmy@fujitsu.com>
| Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
| Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
| (cherry picked from commit d4497ba3fed93807f59c00925056aa21e72c2189)
|ptest-runner: correct PV from 2.5.1 to 2.4.5.1
|
|2.4.5.1 is an actual version, 2.5.1 was a typo mistake.
|That merged less than an hour ago, so no one should
|trip over version going backwards hopefully.
|
|Signed-off-by: Alexander Kanavin <alex@linutronix.de>
|Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|(cherry picked from commit d5e0f51bccae7410b10c98a93fa5853155c2f59e)
Changelog: https://git.yoctoproject.org/ptest-runner2/log/?h=v2.4.5.1
* Makefile: use pkg-config to obtain libcheck flags
* Makefile: remove redundant rule
* README.md: Add my cc to be aware of patches
* README: Update mailing list address for patches + add git command
(From OE-Core rev: 6688fc0df342dfb75061647462e36590dc5175ca)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Fix incorrect function call when processing SPDX_CUSTOM_ANNOTATION_VARS.
The code was calling new_annotation() as a standalone function, but it
should be called as a method on the build_objset object.
Error:
new_annotation(d, build_objset, build, ...)
Corrected to:
build_objset.new_annotation(d, build_objset, build, ...)
This bug would cause a NameError at runtime if SPDX_CUSTOM_ANNOTATION_VARS
was set to a non-empty value, preventing SPDX document generation.
The fix aligns with how new_annotation() is called elsewhere in the
codebase and matches the SBOMObjset class method signature.
(From OE-Core rev: de5c85b2c4c5a62d12aa5b9ab328d9a51d831c41)
Signed-off-by: Stefano Tondo <stefano.tondo.ext@siemens.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
(cherry picked from commit 52ab3b640c6bb7ece34cb4ea6026fd6375f17af4)
Cc: Joshua Watt <jpewhacker@gmail.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
[YC:
Link: https://lore.kernel.org/all/CAJdd5GaqcWujQae-GXQpcgAGtosHe7_X0T%2B8LyOfrr8-mOgzLA@mail.gmail.com/
]
Signed-off-by: Paul Barker <paul@pbarker.dev>
Update to the 5.0.20 release of the 5.0 series for buildtools
(From OE-Core rev: 955e0ed1c6651f0a1d18d0c9bee5f6ef494cd189)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Leading space leads to SPDX document validation errors in some tools.
Example:
SchemaError: \" http://certifi.io/\" is not valid under any of the
schemas listed in the 'anyOf' keyword
(components -> ... -> externalReferences -> 0 -> url)
(From OE-Core rev: 758899639692fb76e28f3914f75249957c0b3600)
Signed-off-by: Peter Marko <peter.marko@siemens.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(From OE-Core rev: ee556d1e1cc16327e0a11207e90ac61d9f1577f3)
Signed-off-by: Peter Marko <peter.marko@siemens.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Leading space leads to SPDX document validation errors in some tools.
Example:
SchemaError: \" http://golang.org/\" is not valid under any of the
schemas listed in the 'anyOf' keyword
(components -> ... -> externalReferences -> 0 -> url)
(From OE-Core rev: e859061f8f00937e5bb737c029ca2da3837d551c)
Signed-off-by: Peter Marko <peter.marko@siemens.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(From OE-Core rev: cbf36f436b477d81b58ff605846a2482308aefa)
Signed-off-by: Peter Marko <peter.marko@siemens.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
The guard added in 80ff4903ea tests "detail" in cve_data, but cve_data is
keyed by CVE id, so it asks whether a CVE literally named "detail" was
scanned. That is never true, the condition short-circuits, and the guard
never runs: a CVE_STATUS[CVE-...] = "backported-patch" set for a vendor
cherry-picked patch is silently overwritten to Unpatched by the CNA.
Use .get() on the entry instead. Guard the fallthrough warning the same
way, it makes the same assumption.
Tested against a qemuarm64 linux-yocto report (6.6.142+git, 16221 entries,
4313 of them with no detail). Current master and this version produce
identical output, 18773 entries with no difference. Adding
CVE_STATUS[CVE-2024-42067] = "backported-patch" to that report then makes
the only difference between them: master overwrites it to Unpatched, this
version keeps it Patched. The pre-80ff4903ea code aborts on the same
report with "KeyError: 'detail'".
AI-Generated: Uses Claude (claude-opus-5)
Fixes: 80ff4903ea1b ("improve_kernel_cve_report: validate that cve details field exists")
(From OE-Core rev: ff8709c6ea17ef64e660c12f225c839865b3facb)
Signed-off-by: Hiago De Franco <hfranco@baylibre.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit f5da16b0d3c8f889dab061ba1d8808aba95d4c67)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
CVE-2026-8458 allows a Negotiate-authenticated connection to be
incorrectly reused for a request using a different SASL service name.
Scarthgap uses curl 8.7.1, which is within the affected version range.
The vulnerable code path on Linux requires both Negotiate authentication
and GSSAPI support, represented by the negotiate-auth and krb5
PACKAGECONFIG options.
The upstream fix [1] stores the SASL service name in struct Curl_creds
and includes it in connection-reuse comparisons. However, struct
Curl_creds was introduced by the credential-management rework in [2],
after curl 8.7.1. Therefore, the security fix cannot be cleanly
backported without introducing a substantial credential-management
refactor.
Use a conditional CVE_STATUS as the least invasive solution. Report the
CVE as unpatched when both krb5 and negotiate-auth are enabled. Otherwise,
mark it not-applicable-config because the vulnerable GSSAPI-backed
Negotiate implementation is not built. The default Scarthgap
configuration enables negotiate-auth but does not enable krb5.
References:
[1] https://github.com/curl/curl/commit/5e99b73cf441d9c369768b9cd48b5389b9a2503d
[2] https://github.com/curl/curl/commit/8f71d0fde515aa4c68002477356c35bd79927729
[3] https://curl.se/docs/CVE-2026-8458.html
(From OE-Core rev: e986fa2c8d74c1b943bb9069e47d1a9ae7dee9e0)
Signed-off-by: Devansh Patel <devanshp@cisco.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
The inherited "python:Mako" mapping is not used for the packaged Mako
source and causes its vulnerability records to be missed.
Use "makotemplates:mako" for its historical NVD configuration identity
and "sqlalchemy:mako" for the current NVD dictionary CPE, NVD
configuration, and CNA affected-data identity.
Backport note: this applies the metadata to Scarthgap Mako 1.3.2 rather
than master 1.4.1; the older release exposes applicable unpatched records.
(From OE-Core rev: d419d38614ad7dae149e005ac6183c0290f54b2d)
Signed-off-by: Devansh Patel <devanshp@cisco.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit 76fc2046d3f251af34dd04f8fdcfc0c1d6016380)
Signed-off-by: Devansh Patel <devanshp@cisco.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
PLY 3.11 contains the picklefile parameter described by
CVE-2025-56005, but exploitation requires an application to
explicitly pass an attacker-controlled pickle file to yacc().
No OE-Core consumer uses this parameter. Ubuntu ignores the issue for
the same reason [1], and NVD records the CVE as disputed [2].
[1] https://ubuntu.com/security/CVE-2025-56005
[2] https://nvd.nist.gov/vuln/detail/CVE-2025-56005
(From OE-Core rev: 9bce52b90a6783e0ba23ff048209ea88922fd1da)
Signed-off-by: Hetvi Thakar <hthakar@cisco.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Analysis:
- NVD marks CVE-2018-20225 as disputed. [1]
- pip searches all configured package indexes without priority and selects
the highest matching version. --extra-index-url adds an equal-priority
package source. [2]
- Exploitation requires use of --extra-index-url and an attacker-published,
higher-version package with the name of a private package. [1]
- Upstream closed the related report as not planned, and Red Hat classified
the issue as WONTFIX because this behavior is intentional. [3][4]
- Record the disputed status; no source patch is available.
Reference:
[1] https://nvd.nist.gov/vuln/detail/CVE-2018-20225
[2] https://pip.pypa.io/en/stable/cli/pip_install/
[3] https://github.com/pypa/pip/issues/12874
[4] https://bugzilla.redhat.com/show_bug.cgi?id=1835736
(From OE-Core rev: 34c0972ffea7dcf4888e3bcfe0ce17206b32e408)
Signed-off-by: Hetvi Thakar <hthakar@cisco.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
When qemu test runs only produce suffixed bootlog variants, the base
bootlog path may not exist. os.symlink() creates a dangling symlink
in the oeqa result log directory.
Avoid advertising a boot log in the collected test results when that
path does not exist. Guard the symlink creation and log a note instead.
This is a partial backport of master commit
7a6596925cb0eb8dd48a0362a51875cdd60152bc. Only the symlink guard is
picked here; the bootlog variant globbing from master is not applicable
to scarthgap.
(From OE-Core rev: 6672769e251ab0dd42fc5f5cde99800ed5ba8289)
Signed-off-by: Peter Tatrai <peter.tatrai.ext@siemens.com>
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
This was introduced as a test for the fix in 6ec70d5d2e330b41b932b0a655b838a5f37df01e
however the way github works has changed in the decade since and this
test is no longer testing what was intended.
The test is failing intermittently on our infrastructure thanks to spam
and IP block throttling restrictions which lead to occasional failures.
At this point it is probably simplest just to drop the url from testing.
(Bitbake rev: 1b619d136c01c97ab1b76a75e8845cddc5e49735)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit b59bdcf8134e6f5935e8f7b0ed8033cbb745e0ff)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Paul Barker <paul@pbarker.dev>
Deepesh Varatharajan
Richard Purdie
Peter Tatrai
Siddharth Doshi
Vijay Anusuri
Kyungjik Min
Peter Marko
Yoann Congal
Wang Mingyu
Stefano Tondo
Darsh Kelaiya
Hemanth Kumar M D
Hiago De Franco
Daniel Turull
Devansh Patel
Jakub Szczudlo (Nokia)
Hetvi Thakar
Hitendra Prajapati
Jaipaul Cheernam
Adarsh Jagadish Kamini