RFR: 8365609: Null pointer dereference in src/java.desktop/share/native/common/java2d/opengl/OGLBlitLoops.c OGLBlitToSurfaceViaTexture()
Phil Race
prr at openjdk.org
Fri Aug 22 04:36:50 UTC 2025
On Fri, 22 Aug 2025 00:23:03 GMT, Alexander Zvegintsev <azvegint at openjdk.org> wrote:
>> src/java.desktop/unix/native/libawt_xawt/awt/gtk3_interface.c line 618:
>>
>>> 616: fp_g_string_free = dl_symbol("g_string_free");
>>> 617:
>>> 618: if (fp_glib_check_version != NULL) {
>>
>> I don't know why we don't treat failing to find this symbol as a fatal error like we do for others such as this first one. dl_symbol will do a longjmp
>> fp_gtk_check_version = dl_symbol("gtk_check_version");
>>
>> but for this one we just clear the error.
>> /* GLib */
>> fp_glib_check_version = dlsym(gtk3_libhandle, "glib_check_version");
>> if (!fp_glib_check_version) {
>> dlerror();
>> }
>>
>> @azvegint any idea ?
>
>> I don't know why we don't treat failing to find this symbol as a fatal error like we do for others such as this first one.
>
> It looks like the `glib_check_version` was added but never used. The first instance of its use appears to be the added check for the methods used for Screencast. And it was overlooked that the `fp_glib_check_version` can be `null`.
>
> The `glib_check_version` is [available since 2.6](https://docs.gtk.org/glib/func.check_version.html), and the 2.6.0 [was released in 2004](https://mail.gnome.org/archives/gnome-announce-list/2004-December/msg00036.html), so I guess we can safely replace the `dlsym` with the `dl_symbol`.
> This will be better than remembering to protect it with a null check if we decide to use `glib_check_version` later somewhere else.
>
>
>
> - fp_glib_check_version = dlsym(gtk3_libhandle, "glib_check_version");
> - if (!fp_glib_check_version) {
> - dlerror();
> - }
> + fp_glib_check_version = dl_symbol("glib_check_version");
So that would be the proper fix here, not the proposed one.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/26799#discussion_r2292666097
More information about the client-libs-dev
mailing list