| Message ID | 20240322103502.GA2045297@coredump.intra.peff.net (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 647e870a08e31984f5f42703f3618ff378a72932 |
| Headers | show |
| Series | rebase: use child_process_clear() to clean | expand |
Jeff King <peff@peff.net> writes: > In the run_am() function, we set up a child_process struct to run > "git-am", allocating memory for its args and env strvecs. These are > normally cleaned up when we call run_command(). But if we encounter > certain errors, we exit the function early and try to clean up ourselves > by clearing the am.args field. This leaks the "env" strvec. > > We should use child_process_clear() instead, which covers both. And more > importantly, it future proofs us against the struct ever growing more > allocated fields. When 21853626 (built-in rebase: call `git am` directly, 2019-01-18) started using run_command() API to drive "am", there already was child_process API and child_process_clear() did clear both .args and .env_array members but we used argv_array_clear() only to clear am.args, leaking am.env_array. > These are unlikely errors to happen in practice, so they don't actually > trigger the leak sanitizer in the tests. But we can add a new test which > does exercise one of the paths (and fails SANITIZE=leak without this > patch). > ... > Not sure if the test is overkill. It really does nothing useful in a > non-leak-checking build. I wonder what are in .env array at this point, though, but that is mere curiosity and not a problem with this patch. Thanks. Will queue. > diff --git a/t/t3438-rebase-broken-files.sh b/t/t3438-rebase-broken-files.sh > index c614c4f2e4..821f08e5af 100755 > --- a/t/t3438-rebase-broken-files.sh > +++ b/t/t3438-rebase-broken-files.sh > @@ -58,4 +58,13 @@ test_expect_success 'unknown key in author-script' ' > check_resolve_fails > ' > > +test_expect_success POSIXPERM,SANITY 'unwritable rebased-patches does not leak' ' > + >.git/rebased-patches && > + chmod a-w .git/rebased-patches && > + > + git checkout -b side HEAD^ && > + test_commit unrelated && > + test_must_fail git rebase --apply --onto tmp HEAD^ > +' > + > test_done
On Fri, Mar 22, 2024 at 10:21:23AM -0700, Junio C Hamano wrote: > > We should use child_process_clear() instead, which covers both. And more > > importantly, it future proofs us against the struct ever growing more > > allocated fields. > > When 21853626 (built-in rebase: call `git am` directly, 2019-01-18) > started using run_command() API to drive "am", there already was > child_process API and child_process_clear() did clear both .args and > .env_array members but we used argv_array_clear() only to clear > am.args, leaking am.env_array. Ah, interesting. I didn't bother to dig into the history. Back then, though, I think that "args" was the only allocated thing. The env call was added much later in be0d29d301 (rebase --apply: make reflog messages match rebase --merge, 2022-10-12). > I wonder what are in .env array at this point, though, but that is > mere curiosity and not a problem with this patch. It's $GIT_REFLOG_ACTION for the sub-process. -Peff
Jeff King <peff@peff.net> writes: >> I wonder what are in .env array at this point, though, but that is >> mere curiosity and not a problem with this patch. > > It's $GIT_REFLOG_ACTION for the sub-process. Ah, OK.
diff --git a/builtin/rebase.c b/builtin/rebase.c index e444ab102d..b9d0fb3269 100644 --- a/builtin/rebase.c +++ b/builtin/rebase.c @@ -610,7 +610,7 @@ static int run_am(struct rebase_options *opts) status = error_errno(_("could not open '%s' for writing"), rebased_patches); free(rebased_patches); - strvec_clear(&am.args); + child_process_clear(&am); return status; } @@ -638,7 +638,7 @@ static int run_am(struct rebase_options *opts) struct reset_head_opts ropts = { 0 }; unlink(rebased_patches); free(rebased_patches); - strvec_clear(&am.args); + child_process_clear(&am); ropts.oid = &opts->orig_head->object.oid; ropts.branch = opts->head_name; @@ -659,7 +659,7 @@ static int run_am(struct rebase_options *opts) status = error_errno(_("could not open '%s' for reading"), rebased_patches); free(rebased_patches); - strvec_clear(&am.args); + child_process_clear(&am); return status; } diff --git a/t/t3438-rebase-broken-files.sh b/t/t3438-rebase-broken-files.sh index c614c4f2e4..821f08e5af 100755 --- a/t/t3438-rebase-broken-files.sh +++ b/t/t3438-rebase-broken-files.sh @@ -58,4 +58,13 @@ test_expect_success 'unknown key in author-script' ' check_resolve_fails ' +test_expect_success POSIXPERM,SANITY 'unwritable rebased-patches does not leak' ' + >.git/rebased-patches && + chmod a-w .git/rebased-patches && + + git checkout -b side HEAD^ && + test_commit unrelated && + test_must_fail git rebase --apply --onto tmp HEAD^ +' + test_done
In the run_am() function, we set up a child_process struct to run "git-am", allocating memory for its args and env strvecs. These are normally cleaned up when we call run_command(). But if we encounter certain errors, we exit the function early and try to clean up ourselves by clearing the am.args field. This leaks the "env" strvec. We should use child_process_clear() instead, which covers both. And more importantly, it future proofs us against the struct ever growing more allocated fields. These are unlikely errors to happen in practice, so they don't actually trigger the leak sanitizer in the tests. But we can add a new test which does exercise one of the paths (and fails SANITIZE=leak without this patch). Signed-off-by: Jeff King <peff@peff.net> --- I noticed this when investigating a different leak, since that other one caused format-patch to crash, hitting one of the error paths. ;) The results of "git grep 'strvec_clear.*\.args'" imply the problem doesn't exist elsewhere (arguably the hit in test-trace2.c should use child_process_clear() to set a good example, but it's already a bit of an oddball use of the struct). Not sure if the test is overkill. It really does nothing useful in a non-leak-checking build. builtin/rebase.c | 6 +++--- t/t3438-rebase-broken-files.sh | 9 +++++++++ 2 files changed, 12 insertions(+), 3 deletions(-)