On Sun, Dec 06, 2020 at 02:44:16PM +0000, Colin Watson wrote:
> So, in the RESTORE_BACKUP case, shouldn't that be:
>
> char *dstf = grub_util_path_concat (2, di, de->d_name);
>
> ... rather than grub_util_path_concat_ext? Otherwise it seems to me
> that it's going to try to append an additional argument which doesn't
> exist, and may well add random uninitialised stuff from memory. Running
> grub-install under valgrind would probably show this up (I can't get it
> to do it for me so far, but most likely I just haven't set up quite the
> right initial conditions).
While I couldn't reproduce this on amd64 (and valgrind didn't show any
errors), I can reproduce it just fine on i386, which is what Paul is
using. I guess the va_list layout in memory is different enough between
the two ABIs to tickle this.
I'll apply this patch for Debian grub2 2.04-11, which I've confirmed
fixes it for me:
diff --git a/util/grub-install-common.c b/util/grub-install-common.c
index a883b6dae..61f9075bc 100644
--- a/util/grub-install-common.c
+++ b/util/grub-install-common.c
@@ -247,7 +247,7 @@ clean_grub_dir_real (const char *di, enum clean_grub_dir_mode mode)
}
else if (mode == RESTORE_BACKUP)
{
- char *dstf = grub_util_path_concat_ext (2, di, de->d_name);
+ char *dstf = grub_util_path_concat (2, di, de->d_name);
dstf[strlen (dstf) - 1] = 0;
if (grub_util_rename (srcf, dstf) < 0)
grub_util_error (_("cannot restore `%s': %s"), dstf,
Dimitri, I know Ubuntu isn't very interested in supporting i386, but IMO
you should apply this patch to Ubuntu in any case; it might affect other
architectures, and anyway leaving known undefined behaviour around isn't
a good idea.
Paul, note that you'll also need to run "sudo rm -f
/boot/grub/i386-pc/*.{img,lst,mo,mod,o,sh}-*" to clean up the stray
files created by this bug.
--
Colin Watson (he/him) [cjwatson@xxxxxxxxxx]
