[BUG] git rev-list --missing=allow-promisor

Steps to reproduce:
1. Create a git repository with missing+promised blobs.
For example, I did this:
git clone git@github.com:git/git.git git-no-blobs --filter=blob:none
--depth=1 --bare

--filter=blob:none - don't fetch any blobs --depth=1 - we don't need git's entire history, just one commit is fine --bare - this avoids git fetching any blobs as part of the checkout operation

2. In this repository, run:
git rev-list HEAD --objects --missing=allow-promisor

Expected outcome: git rev-list prints the paths and OIDs of all the objects

Actual outcome: Any of the following: a) git rev-list prints the paths and OIDs of all the objects b) git rev-list fails with "fatal: malformed mode in tree entry" c) git rev-list fails with "fatal: too-short tree object"

Diagnosis: With missing=allow-promisor, when git encounters a missing object, it calls is_promisor on it, which in turn calls add_promisor_object on lots of things. In this line in add_promisor_object, the buffer of the tree that we are currently traversing is freed: https://github.com/git/git/blob/master/packfile.c#L2242 This clearly causes problems but not in a deterministic way.

On Thu, Aug 11, 2022 at 09:33:54AM +1200, Andrew Olsen wrote:

Thanks for a clear report. I can reproduce this easily.

Even better is to compile with SANITIZE=address, which reliably shows that this is indeed a use-after-free.

Ugh, yes. This is due to my fcc07e980b (is_promisor_object(): free tree buffer after parsing, 2021-04-13). In most cases we'll have just allocated memory for the tree buffer in parse_object(), but occasionally we'll jump into is_promisor_object() while looking at an existing tree, and parse_object() will just reuse the buffer cached in the struct.

It would be really nice to have some kind of reference-counting for these cached buffers, but doing it everywhere may be rather complicated. An easier hack to fix this area is for it to check ahead of time whether there's a cached buffer we'll reuse. That only fixes this spot, but this function is rather unlike most others in that it starts looking at new random trees in the middle of an existing tree traversal (this would be possible in a regular tree walk if you could have circular references, but you can't easily because of sha1).

So something like this makes the bug go away, though it does feel a little dirty:

---

On Thu, Aug 11, 2022 at 04:12:01AM -0400, Jeff King wrote:

Having slept on this, I actually think it's the right approach, for the reasons given in the commit message below. And we definitely need to do _something_ to fix the bug, and preferably something small and un-risky, since this should probably go to 'maint'. The other obvious option is reverting the original patch, but I think the memory savings it brought would be hard to give up, and this patch should reliably fix the bug.

Thanks again for a clear bug report.

-- >8 --
Subject: is_promisor_object(): fix use-after-free of tree buffer

Since commit fcc07e980b (is_promisor_object(): free tree buffer after parsing, 2021-04-13), we'll always free the buffers attached to a "struct tree" after searching them for promisor links. But there's an important case where we don't want to do so: if somebody else is already using the tree!

This can happen during a "rev-list --missing=allow-promisor" traversal in a partial clone that is missing one or more trees or blobs. The backtrace for the free looks like this:

      #1 free_tree_buffer tree.c:147
      #2 add_promisor_object packfile.c:2250
      #3 for_each_object_in_pack packfile.c:2190
      #4 for_each_packed_object packfile.c:2215
      #5 is_promisor_object packfile.c:2272
      #6 finish_object__ma builtin/rev-list.c:245
      #7 finish_object builtin/rev-list.c:261
      #8 show_object builtin/rev-list.c:274
      #9 process_blob list-objects.c:63
      #10 process_tree_contents list-objects.c:145
      #11 process_tree list-objects.c:201
      #12 traverse_trees_and_blobs list-objects.c:344
      [...]

We're in the middle of walking through the entries of a tree object via process_tree_contents(). We see a blob (or it could even be another tree entry) that we don't have, so we call is_promisor_object() to check it. That function loops over all of the objects in the promisor packfile, including the tree we're currently walking. When we're done with it there, we free the tree buffer. But as we return to the walk in process_tree_contents(), it's still holding on to a pointer to that buffer, via its tree_desc iterator, and it accesses the freed memory.

Even a trivial use of "--missing=allow-promisor" triggers this problem, as the included test demonstrates (it's just a vanilla --blob:none clone).

We can detect this case by only freeing the tree buffer if it was allocated on our behalf. This is a little tricky since that happens inside parse_object(), and it doesn't tell us whether the object was already parsed, or whether it allocated the buffer itself. But by checking for an already-parsed tree beforehand, we can distinguish the two cases.

That feels a little hacky, and does incur an extra lookup in the object-hash table. But that cost is fairly minimal compared to actually loading objects (and since we're iterating the whole pack here, we're likely to be loading most objects, rather than reusing cached results).

It may also be a good direction for this function in general, as there are other possible optimizations that rely on doing some analysis before parsing:

  - we could detect blobs and avoid reading their contents; they can't
    link to other objects, but parse_object() doesn't know that we don't
    care about checking their hashes.

  - we could avoid allocating object structs entirely for most objects
    (since we really only need them in the oidset), which would save
    some memory.

  - promisor commits could use the commit-graph rather than loading the
    object from disk

This commit doesn't do any of those optimizations, but I think it argues that this direction is reasonable, rather than relying on parse_object() and trying to teach it to give us more information about whether it parsed.

The included test fails reliably under SANITIZE=address just when running "rev-list --missing=allow-promisor". Checking the output isn't strictly necessary to detect the bug, but it seems like a reasonable addition given the general lack of coverage for "allow-promisor" in the test suite.

Reported-by: Andrew Olsen <andrew.olsen@koordinates.com>
Signed-off-by: Jeff King <peff@peff.net>
---
The bug is in v2.32.0. I prepared this directly on top of fcc07e980b,
but it should apply cleanly to any recent maint or master tips.

 packfile.c               | 15 +++++++++++++--
 t/t5616-partial-clone.sh |  7 +++++++
 2 files changed, 20 insertions(+), 2 deletions(-)

Jeff King <peff@peff.net> writes:

I forgot that the above "loops over" happens only once to populate the oidset hashtable, and briefly wondered if we are being grossly inefficient by scanning pack .idx file each time we encounter a missing object. "Upon first call, that function loops over ... walking, to prepare a hashtable to answer if any object id is referred to by an object in promisor packs" would have helped ;-).

Yeah, all of the future bits sound sensible. 

Will queue.

Thanks.

On Sun, Aug 14, 2022 at 10:32:12PM -0700, Junio C Hamano wrote:

Right. When you have worked in an area, sometimes it is easy to forget which things are common knowledge and which are not. :) I don't mind at all if you want to amend the commit message as you apply.

I very intentionally didn't work on those things yet, because I wanted to make sure we got a simple fix in as quickly as possible. That said, I don't have immediate plans for them. They are perhaps not quite small enough for #leftoverbits, but I think they might also be nice bite-sized chunks for somebody wanting to get their feet wet in that part of the code.

-Peff