patch, 2 partscommit: fix "author_ident" leak

This trivial series fixes a leak in "git commit", removes two UNLEAK() in favor of simply calling strbuf_release(), and marks two tests that previously hit that UNLEAK() being run in the linux-leaks CI job.

Ævar Arnfjörð Bjarmason (2):
  commit: fix "author_ident" leak
  commit: use strbuf_release() instead of UNLEAK()

 builtin/commit.c                 | 13 ++++++++-----
 t/t2203-add-intent.sh            |  1 +
 t/t7011-skip-worktree-reading.sh |  1 +
 3 files changed, 10 insertions(+), 5 deletions(-)

Fix a leak in cmd_commit(), since 4c28e4ada03 (commit: die before asking to edit the log message, 2010-12-20) we have been freeing the "author_ident" "struct strbuf", but not in the case where prepare_to_commit() returns non-zero.

This fixes a leak demonstrated by e.g. "t3505-cherry-pick-empty.sh", but unfortunately we cannot mark it or other affected tests as passing now with "TEST_PASSES_SANITIZE_LEAK=true" as we'll need to fix many other memory leaks before doing so.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
---
 builtin/commit.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

Ævar Arnfjörð Bjarmason  <avarab@gmail.com> writes:

It took me an actual reading of the code to see that the above refers to one and the same thing (i.e. "author_ident variable of type struct strbuf"). I think it should be sufficient and clearer to just mention "author_ident" here.

Good eyes.  

Hmph, if we hit one of the two die() after this point before we reach the "cleanup" label, author_ident will be left on the stack, which we may want to UNLEAK()?

I am wondering if prepare_to_commit(), which is the one that is responsible for allocating and using the information in the strbuf, should be the one who is responsible for cleaning it when it failed to do its thing, but I do not think it is a good idea, because the caller MUST release it in the success case anyway. So dealing with the releasing here does make sense.

By jumping to the cleanup label, we not just release author_ident, but we start unleak'ing err and sb as well, which shouldn't be a problem, hopefully.

Will queue.

Thanks.

Convert the UNLEAK() added in 0e5bba53af7 (add UNLEAK annotation for reducing leak false positives, 2017-09-08) to release the memory using strbuf_release() instead.

The tests being marked as passing with "TEST_PASSES_SANITIZE_LEAK=true" already passed before due to the UNLEAK(), but now they really don't leak memory, so let's mark them as such.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
---
 builtin/commit.c                 | 4 ++--
 t/t2203-add-intent.sh            | 1 +
 t/t7011-skip-worktree-reading.sh | 1 +
 3 files changed, 4 insertions(+), 2 deletions(-)

Ævar Arnfjörð Bjarmason  <avarab@gmail.com> writes:

That smells like a brave move.

Specifically, the cited commit turned an existing strbuf_release() on &err into UNLEAK(). If that and the other strbuf (sb) were so easily releasable, why didn't we do so back then already?

Junio C Hamano <gitster@pobox.com> writes:

I suspect that the answer to the above question is because these allocations are in the top-level cmd_commit() function, which is never called recursively or repeatedly as a subroutine. The only significant thing that happens after we return from it is to exit.

In such a code path, marking a variable as UNLEAK() is a better thing to do than calling strbuf_release(). Both will work as a way to squelch sanitizers from reporting a leak that does not matter, but calling strbuf_release() means we'd spend extra cycles to return pieces of memory to the pool, even though we know that the pool itself will be cleaned immediately later at exit.

We already have UNLEAK to tell sanitizers not to distract us from spotting and plugging real leaks by reporting these apparent leaks that do not matter. It is of somewhat dubious value to do a "we care too much about pleasing sanitizer and spend extra cycles at runtime while real users are doing real work" change.

On Wed, Feb 16 2022, Junio C Hamano wrote:

[CC-ing some people using/interested in UNLEAK()]

We've had several discussions about the utility of UNLEAK() as I've been submitting these patches, and I thought that if we weren't 100% on the same page it was at least clear what I was going for here.

Per https://lore.kernel.org/git/87a6k8daeu.fsf@evledraar.gmail.com/ the real goal I have in mind here is to use the built-ins as a stand-in for testing whether the underlying APIs are leak-free.

Because of that having to reason about UNLEAK() doing the right thing or not is just unneeded distraction. Yes for a "struct strbuf" it won't matter, but most of what we UNLEAK() is more complex stuff like "struct rev_info". We won't really make headway making revision.c not leak memory without using "git log" et al as the test subjects for whether that API leaks.

We are also freeing most of this already even for built-ins, e.g. see (not all of these are applicable obviously, but per the numbers enough are to make the point):

    $ git grep '\bstrbuf_release\(' -- builtin | wc -l
    456
    $ git grep '\bUNLEAK\(' -- builtin | wc -l
    29

So one goal I've got with these patches is to eventually get rid of UNLEAK() entirely. We only use it in these few instances:

    $ git grep '\bUNLEAK\(' -- '*.c' | wc -l
    31

Per the above we'd want to convert any that deal with the complex structures that are the big source of leaks to doing real releases for testing the APIs. That'll leave only a handful of remaining legitimate uses, which I don't think are worth keeping some UNLEAK() API around for, v.s. just freeing them.

I think we've also somewhat been talking past each other in past exchanges (including me with Jeff King) about what you call "real leaks".

When I'm referring to memory leaks I'm talking about them in the more inclusive sense explained in this valgrind documentation: https://valgrind.org/docs/manual/mc-manual.html#mc-manual.leaks; Basically "a malloc() not followed by a free() is a leak".

But you and Jeff King have (I think) commonly used that term to exclude what's called "still reachable" in that table. Those *are* memory leaks, but as explained in that table just ones that arguably don't matter. Or at least ones most people tracing leaks don't want reported by default.

UNLEAK() is just a mechanism for moving a memory leak from another category into "still reachable". That will make LSAN blind to it, and valgrind in many cases due to it being a heap tracer. But it & other memory leaking tools *will* report even on those if given the right options.

So I've found it useful to get rid of UNLEAK() in some cases for those, because even if LSAN runs clean it's useful to run valgrind in its more pedantic tracing modes to see even the "still unreachable", and if we should care about it or not.

Some of the leaks in those category *are* "real leaks". I.e. you can have an ever-growing structure with global reach that's always "still reachable", but when using the API would eventually OOM you.

So while it's a useful heuristic for spotting "will be cleaned up at exit anyway" you can't rely on it, so I've been looking at them anyway.

So being able to just free() them so I can permanently ignore them as I fix more leaks would be useful to me, so I hope you'll agree on just talking this as-is, thanks! :)

Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:

I have to say that you have a wrong goal and wrong priority. The number of UNLEAK in cmd_foo() functions is not even a poor approximation of our progress.

Imagine that a subsystem that are repeatedly set-up and used during a single program invocation was leaky. Let's say a program calls diff_setup() to prepare the diff_options structure, feeds two things to compare, and calls diff_flush() to show the comparison result, but let's assume this sequence leaks some resources.

Now cmd_diff() may be such a program that does a setup, feeds two trees, calls diff_flush() and exits. If we didn't do anything to it, diff_options may "leak". Marking it with UNLEAK may be a good measure, if we want to keep the leak checker from reporting a leak that does not matter in practice so that we can concentrate on plugging real leaks that matter.

But consider cmd_log(), running something like "git log -p". It iterates over commits, does the <setup, compare two trees, flush> repeatedly for each commit it encounters during the walk on the same diff_options structure. Now, the leak in the code path does matter. If diff_flush() is left leaky, it will show up in the leak checker's output, and that is reporting real leaks that matter. The thing is, the <setup, compare, flush> sequence whose leak matters is not in cmd_log(); it is much deeper in the revision walking machinery. We do not want to paper over such leaks with UNLEAK.

See the difference? The number of UNLEAK OUTside built-ins does matter. We do not want to have UNLEAK there to hide leaks in possible callees that may be called arbitrary number of times in arbitrary order. Compared to that, UNLEAK in cmd_foo() to mark an on-stack variable that was used only once without even a recursion is purely to squelch the leak checker from reporting leaks that does not matter. For simple things like strbuf on stack of the top-level cmd_foo() functions, we could even leave strbuf_release() not called and leave the resource reclamation to _exit(2). That would cause the leak checkers to report them and distract us, and UNLEAK is a better way to squelch the distraction without wasting extra cycles at runtime to actually free them.

So, don't look at "built-ins as a stand-in". It is a wrong direction to go in to let the "leak checker" tail wag the dog.

Thanks.

On Fri, Feb 18 2022, Junio C Hamano wrote:

To clarify I'm not saying it's an approximation of how close we are to getting rid of memory leaks. I think I'll know better than anyone what the current state of shoveling shit uphill that is.

I was suggesting that these's a trivial number of these, and mostly we strbuf_release() already, and that on modern libc's free() is pretty much free anyway, so worrying about using UNLEAK() v.s. just freeing before exit is some combination of a premature optimization and needlessly retaining a special-case for the built-ins.

Yes, I see the difference and I fully accept the point you're making.

I just don't find it to be a useful distinction in practice, because for e.g. "git log" we do have certain uses of the API which are only performed by the top-level part of a "git log --whatever".

So if we take the view that e.g. the "struct rev_info rev" should be UNLEAK()'d because "we're about to exit anyway" then as soon as we *do* use that part of the API we'll run into the previously unfixed and hidden leak.

Which is why as noted in the linked-to-above https://lore.kernel.org/git/87a6k8daeu.fsf@evledraar.gmail.com/ I think it's useful to fix leaks even in built-ins, because they're useful canaries for those APIs.

Now, in this case does it really matter? No, it doesn't currently. But I think the direction we should be heading in is turning more of these cmd_whatever() into properly functioning libraries, as e.g. John Cai's just-sent series to do that for "reflog delete" does: https://lore.kernel.org/git/pull.1218.git.git.1645209647.gitgitgadget@gmail.com/

Once we do that the UNLEAK() here will absolutely need to be changed to a strbuf_release() or equivalent, since if you make two commits with that new library we'll have leaked memory.

So I think it makes sense to just fix leaks everywhere if it's easy without worrying about the distinction, particularly since I haven't seen that "wasting cycles" concern matter in practice.

If it did I'd think adding a "GIT_DESTRUCT_LEVEL" as I suggested in the linked E-Mail would make more sense, since e.g. you could also use that if you knew you were walking 100 commits & exiting, which can be useful to further reduce free() churn, and is how a similar global in the Perl API (PERL_DESTRUCT_LEVEL) works.

Since 4c28e4ada03 (commit: die before asking to edit the log message, 2010-12-20), we have been "leaking" the "author_ident" when prepare_to_commit() fails. Instead of returning from right there, introduce an exit status variable and jump to the clean-up label at the end.

Instead of explicitly releasing the resource with strbuf_release(), mark the variable with UNLEAK() at the end, together with two other variables that are already marked as such. If this were in a utility function that is called number of times, but these are different, we should explicitly release resources that grow proportionally to the size of the problem being solved, but cmd_commit() is like main() and there is no point in spending extra cycles to release individual pieces of resource at the end, just before process exit will clean everything for us for free anyway.

This fixes a leak demonstrated by e.g. "t3505-cherry-pick-empty.sh", but unfortunately we cannot mark it or other affected tests as passing now with "TEST_PASSES_SANITIZE_LEAK=true" as we'll need to fix many other memory leaks before doing so.

Incidentally there are two tests that always passes the leak checker with or without this change. Mark them as such.

This is based on an earlier patch by Ævar, but takes a different approach that is more maintainable.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
---
 builtin/commit.c                 | 9 ++++++---
 t/t2203-add-intent.sh            | 1 +
 t/t7011-skip-worktree-reading.sh | 1 +
 3 files changed, 8 insertions(+), 3 deletions(-)

On Thu, May 12 2022, Junio C Hamano wrote:

We've talked about UNLEAK() v.s. strbuf_release() elsewhere, so let's leave that aside. I know your preferences in that area.

But even accounting for that, I don't see what the "more maintainable" here refers to. The approach I suggested would s/UNLEAK/strbuf_release/ in the 4th hunk, but otherwise be equivalent.

Surely both are about as easy to maintain. To the extent that there's any difference at all I'd think the strbuf_release() would pull ahead, as it's guaranteed to do the right thing with all of our memory analysis tooling (some of which will have a noop UNLEAK()).

Just a small question, I see this is in "next" already, and I'm fine with this change either way.

Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:

Judicious use of UNLEAK() has documentation value to tell readers which use of pointer variables need to be explicitly released, and which pointer variables can just implicitly released by going out of scope at the end. There are also a few other small added benefits (they do not have to be changed when the helper needed to do the real release changes, and not doing an explicit real release on resources when there no need to is conceptually cleaner). But they are icing on the cake.

Sorry for a late reply---I go offline every other Tuesday and yesterday was such a Tuesday.