On Tue, Dec 15, 2020 at 06:34:53PM +0100, Daniel C. Klauer wrote:
> I'm getting the following crash from "git pack-redundant --all" (output
> from valgrind):
>
> ==14070== Invalid read of size 8
> ==14070== at 0x18F165: minimize (pack-redundant.c:399)
> ==14070== by 0x18F165: cmd_pack_redundant (pack-redundant.c:622)
> ==14070== by 0x1242D3: run_builtin (git.c:444)
> ==14070== by 0x1242D3: handle_builtin (git.c:674)
> ==14070== by 0x125393: run_argv (git.c:741)
> ==14070== by 0x125393: cmd_main (git.c:872)
> ==14070== by 0x123E7D: main (common-main.c:52)
> ==14070== Address 0x10 is not stack'd, malloc'd or (recently) free'd
>
> Commands to reproduce:
>
> mkdir new
> cd new
> git init
> touch foo.txt
> git add foo.txt
> git commit -m "first commit"
> git gc
> git pack-redundant --all
Thanks for an easy reproduction. This bisects to 3011177640
(pack-redundant: delay creation of unique_objects, 2019-02-02).
I suspect the fix is just:
diff --git a/builtin/pack-redundant.c b/builtin/pack-redundant.c
index 3e70f2a4c1..68afcfeb7b 100644
--- a/builtin/pack-redundant.c
+++ b/builtin/pack-redundant.c
@@ -396,7 +396,7 @@ static void minimize(struct pack_list **min)
pl = local_packs;
while (pl) {
- if (pl->unique_objects->size)
+ if (pl->unique_objects && pl->unique_objects->size)
pack_list_insert(&unique, pl);
else
pack_list_insert(&non_unique, pl);
but I didn't look closely (author cc'd).
> The reason this showed up is that Yocto's build tool, bitbake, sometimes
> invokes this git command on some of the repositories it downloaded. This
> issue was apparently also reported here:
> https://bugzilla.redhat.com/show_bug.cgi?id=1803506
We had a discussion not too long ago[1] about whether we should
deprecate and remove pack-redundant, as it's not generally useful. I'm
curious if you have any more context on why it's used in that tool.
-Peff
[1] https://lore.kernel.org/git/20200825172214.GC1414394@xxxxxxxxxxxxxxxxxxxxxxx/
