That's good to know. Since we're completely in the dark on how to
produce a reliable test case that exercises this crash, the hope was
that fuzzing could inch us towards a solution.
There's also the problem that we're observing these crashes on Windows 7
and x86-64 (at least for now), as
It might be possible that even though RKV was built in debug mode, the
backing sys crate isn't building LMDB itself this way. On a very first
glance it looks like this is the case, always going for opt level 2:
https://github.com/danburkert/lmdb-rs/blob/master/lmdb-sys/build.rs#L23
I'll have a
Thanks for getting back!
The files at https://github.com/mozkeeler/rkv-fuzz/tree/trunk/crashes
are fuzzed variants of the input data seeded from the mdb input here:
They're generated when running the `american fuzzy lop` fuzzer:
http://lcamtuf.coredump.cx/afl/.
Victor
Quick update on this.
Potentially useful to the efforts here, Dana Keeler has implemented a =
fuzzing harness for RKV (our typed Rust interface to LMDB): =
https://github.com/mozkeeler/rkv-fuzz
We weren't able to trigger the `mdb_cursor_put` crashes reported at =
Apologies for the previous botched message, I was unfamiliar with the
formatting rules for this list.
To reiterate:
Since the link to the crash stats from my earlier reply now expired,
here's an up to date place to see all recent crash reports involving
`mdb_cursor_put`:
--Apple-Mail=_F8BCAB64-A19B-4B00-B41B-03BE06C12A1F
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
charset=utf-8
> On 6. Aug 2019, at 13:43, Victor Porof wrote:
>=20
> Here's everything we know about: =
> On 1. Aug 2019, at 19:47, Howard Chu wrote:
>=20
> vpo...@mozilla.com wrote:
>> Hey folks.
>>=20
>> =3D46rom Myk=3DE2=3D80=3D99s investigations in the previous followup, =
it seems =3D
>> that the suggested changes to `mdb_cursor_init` to avoid using an =3D
>> invalid DBI might not be
Hey folks.
=46rom Myk=E2=80=99s investigations in the previous followup, it seems =
that the suggested changes to `mdb_cursor_init` to avoid using an =
invalid DBI might not be solving the actual issue, given the behaviour =
of `mdb_page_search`.
It=E2=80=99s also causing the seemingly correct
