Re: [PATCH v2 4/4] pack-check: use size_t for the size passed to verify callbacks
- From
Elia Pinto <gitter.spiros@gmail.com>
- Date
- Nov 19, 2026, 06:05 UTC
- Message-ID
- <20261119060524.775205-3-gitter.spiros@gmail.com>
- In-Reply-To
- <av2pUyAwTA7e2eOj@pks.im>
On Wed, Nov 18, 2026 at 02:34:11PM +0100, Patrick Steinhardt wrote:
> The more interesting question to me is: if they only inspect a prefix, > do they actually notice that something is off and report the error? If > not that certainly sounds like a bug to me.
Looking at it again, on master the callback is not even reached for such an object. Just before calling it, verify_packfile() checks the object with check_object_signature(), which also takes an "unsigned long". It hashes only the prefix, the hash does not match, and fsck reports "packed <oid> from <pack> is corrupt" for an object that is fine. So fsck does complain, but for the wrong reason, and the commit message is wrong to say that the callbacks look at a prefix today.
check_object_signature() is changed by 1/4 of "use size_t when parsing and checking objects" [1]. Only then is the callback reached, and it needs this patch to get the whole size, plus the rest of [1] for parse_object_buffer() and fsck_object(). So 4/4 belongs on top of [1], and I will resend it there with a corrected message.
Blobs are not affected: on these platforms core.bigFileThreshold, an "unsigned long" itself, is below 4GiB, so every blob that large is checked with stream_object_signature() and reaches the callback without its data.
[1] https://lore.kernel.org/git/cover.1793444420.git.gitter.spiros@gmail.com/
Thanks, Elia