Fix CVE-2026-58010: off-by-one error in gvs_tuple_is_normal()

Backport upstream fix for CVE-2026-58010 from GNOME/glib MR !5129
to mingw-glib2 2.70.1. The fix corrects an off-by-one error in
gvs_tuple_is_normal() in glib/gvariant-serialiser.c that allowed
a single byte out-of-bounds read when checking a GVariant for
normal form. A regression test is included.

CVE: CVE-2026-58010
Upstream patches:
 - https://gitlab.gnome.org/GNOME/glib/-/merge_requests/5129.patch
Resolves: RHEL-212164

This commit was backported by Ymir, a Red Hat Enterprise Linux software maintenance AI agent.

Assisted-by: Ymir
This commit is contained in:
RHEL Packaging Agent 2026-07-22 16:24:15 +00:00
parent 16e08b276e
commit 99e6f79ece
2 changed files with 115 additions and 1 deletions

View File

@ -0,0 +1,106 @@
From eb63e89e237d35fb1d3cb757ea97b4c8806269a0 Mon Sep 17 00:00:00 2001
From: Philip Withnall <pwithnall@gnome.org>
Date: Sun, 29 Mar 2026 19:10:41 +0100
Subject: [PATCH] gvariant: Fix an off-by-one error in an offset comparison
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
This allows a single byte out-of-bounds read off the end of the
(potentially untrusted) byte array backing a `GVariant` when its
being checked for normal form.
I cant see how this could practically be exploited, but its certainly
a security bug as the `GVariant` normal form checking code is supposed
to be robust to malicious inputs.
Spotted by linhlhq as #YWH-PGM9867-190, and fix and reproducer provided
by them too, thanks. Confirmed and turned into a unit test by me.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Fixes: #3915
---
glib/gvariant-serialiser.c | 2 +-
glib/tests/gvariant.c | 48 ++++++++++++++++++++++++++++++++++++++
2 files changed, 49 insertions(+), 1 deletion(-)
diff --git a/glib/gvariant-serialiser.c b/glib/gvariant-serialiser.c
index 832a8fdc2..1bf3e4688 100644
--- a/glib/gvariant-serialiser.c
+++ b/glib/gvariant-serialiser.c
@@ -1079,7 +1079,7 @@ gvs_tuple_is_normal (GVariantSerialised value)
while (offset & alignment)
{
- if (offset > value.size || value.data[offset] != '\0')
+ if (offset >= value.size || value.data[offset] != '\0')
return FALSE;
offset++;
}
diff --git a/glib/tests/gvariant.c b/glib/tests/gvariant.c
index 0110f2664..dd3af6941 100644
--- a/glib/tests/gvariant.c
+++ b/glib/tests/gvariant.c
@@ -5047,6 +5047,52 @@ test_normal_checking_tuple_offsets (void)
g_variant_unref (variant);
}
+/* This is a regression test that looping over the padding bytes in a short
+ * (non-normal) tuple doesnt overflow the input data.
+ *
+ * See https://gitlab.gnome.org/GNOME/glib/-/issues/3915 */
+static void
+test_normal_checking_tuple_offsets6 (void)
+{
+ /*
+ * Type: (ynqiuxthdsog) — 12 members, first member 'y' (byte) has
+ * alignment 0, second 'n' (int16) has alignment 1.
+ * With 1 byte of data (0x28), after reading the first byte member,
+ * offset=1, alignment check for 'n' requires offset to be even,
+ * so the while loop checks value.data[1] — but size is only 1.
+ *
+ * Use heap allocation via GBytes so ASan reports heap-buffer-overflow.
+ */
+ uint8_t *heap_data = NULL;
+ GBytes *bytes = NULL;
+ const GVariantType *data_type = G_VARIANT_TYPE ("(ynqiuxthdsog)");
+ GVariant *variant = NULL;
+ GVariant *normal_variant = NULL;
+ GVariant *expected = NULL;
+
+ g_test_bug ("https://gitlab.gnome.org/GNOME/glib/-/issues/3915");
+
+ heap_data = g_malloc (1);
+ heap_data[0] = 0x28;
+ bytes = g_bytes_new_take (heap_data, 1);
+
+ variant = g_variant_new_from_bytes (data_type, bytes, FALSE);
+ g_assert_nonnull (variant);
+
+ g_assert_false (g_variant_is_normal_form (variant));
+
+ normal_variant = g_variant_get_normal_form (variant);
+ g_assert_nonnull (normal_variant);
+
+ expected = g_variant_new_parsed ("(byte 0x28, int16 0, uint16 0, 0, uint32 0, int64 0, uint64 0, handle 0, 0.0, '', objectpath '/', signature '')");
+ g_assert_cmpvariant (expected, variant);
+ g_assert_cmpvariant (expected, normal_variant);
+
+ g_variant_unref (expected);
+ g_variant_unref (normal_variant);
+ g_variant_unref (variant);
+}
+
/* Test that an empty object path is normalised successfully to the base object
* path, /. */
static void
@@ -5191,6 +5237,8 @@ main (int argc, char **argv)
test_normal_checking_array_offsets);
g_test_add_func ("/gvariant/normal-checking/tuple-offsets",
test_normal_checking_tuple_offsets);
+ g_test_add_func ("/gvariant/normal-checking/tuple-offsets6",
+ test_normal_checking_tuple_offsets6);
g_test_add_func ("/gvariant/normal-checking/empty-object-path",
test_normal_checking_empty_object_path);

View File

@ -5,7 +5,7 @@
Name: mingw-glib2
Version: 2.70.1
Release: 8%{?dist}
Release: 9%{?dist}
Summary: MinGW Windows GLib2 library
License: LGPLv2+
@ -77,6 +77,9 @@ Patch8: mingw-glib2-2.70.1-CVE-2026-58013.patch
# https://gitlab.gnome.org/GNOME/glib/-/issues/3917
Patch9: mingw-glib2-2.70.1-CVE-2026-58011.patch
# https://gitlab.gnome.org/GNOME/glib/-/issues/3915
Patch10: mingw-glib2-2.70.1-CVE-2026-58010.patch
%description
MinGW Windows Glib2 library.
@ -131,6 +134,7 @@ Static version of the MinGW Windows GLib2 library.
%patch7 -p1
%patch8 -p1
%patch9 -p1
%patch10 -p1
%build
%mingw_meson --default-library=both \
@ -308,6 +312,10 @@ find $RPM_BUILD_ROOT -name "*.la" -delete
%changelog
* Wed Jul 22 2026 RHEL Packaging Agent <redhat-ymir-agent@redhat.com> - 2.70.1-9
- Fix CVE-2026-58010: off-by-one error in gvs_tuple_is_normal()
Resolves: RHEL-212164
* Wed Jul 22 2026 RHEL Packaging Agent <redhat-ymir-agent@redhat.com> - 2.70.1-8
- Fix CVE-2026-58011: g_date_time_add_full() range validation
Resolves: RHEL-212184