]> Kevux Git Server - rit/commitdiff
rebase: don't leak on "--abort"
authorÆvar Arnfjörð Bjarmason <avarab@gmail.com>
Tue, 8 Nov 2022 18:17:48 +0000 (19:17 +0100)
committerJunio C Hamano <gitster@pobox.com>
Mon, 21 Nov 2022 03:32:48 +0000 (12:32 +0900)
Fix a leak in the recent 6159e7add49 (rebase --abort: improve reflog
message, 2022-10-12). Before that commit we'd strbuf_release() the
reflog message we were formatting, but when that code was refactored
to use "ropts.head_msg" the strbuf_release() was omitted.

Ideally the three users of "ropts" in cmd_rebase() should use
different "ropts" variables, in practice they're completely separate,
as this and the other user in the "switch" statement will "goto
cleanup", which won't touch "ropts".

The third caller after the "switch" is then unreachable if we take
these two branches, so all of them are getting a "{ 0 }" init'd
"ropts".

So it's OK that we're leaving a stale pointer in "ropts.head_msg",
cleaning it up was our responsibility, and it won't be used again.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
builtin/rebase.c
t/t7517-per-repo-email.sh

index 6a83132..3f360eb 100644 (file)
@@ -1322,6 +1322,7 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)
                if (reset_head(the_repository, &ropts) < 0)
                        die(_("could not move back to %s"),
                            oid_to_hex(&options.orig_head->object.oid));
+               strbuf_release(&head_msg);
                remove_branch_state(the_repository, 0);
                ret = finish_rebase(&options);
                goto cleanup;
index 163ae80..efc6496 100755 (executable)
@@ -9,6 +9,7 @@ test_description='per-repo forced setting of email address'
 GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME=main
 export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME
 
+TEST_PASSES_SANITIZE_LEAK=true
 . ./test-lib.sh
 
 test_expect_success 'setup a likely user.useConfigOnly use case' '