[FFmpeg-devel] [PATCH] avcodec/rl2: set dimensions

[Paul B Mahol]

On 7/24/19, James Almer <jamrial at gmail.com> wrote:
> On 7/23/2019 9:39 PM, Kieran Kunhya wrote:
>>>
>>> What was the cause of the slow decoding? Does this actually fix it, or
>>> does it just swipe it "under the carpet"?
>>> If someone ever found a sample with a different solution, how would they
>>> know that they shouldn't just remove this again? Without any kind of
>>> comment on the point of this call, people might assume it's pointless
>>> nonsense.
>>>
>>
>> A significant proportion of these patches sweep the issue under the
>> carpet.
>> Not to mention the swathes of annoyed developers
>>
>> Kieran
>
> The only very vocal people so far that i could see have been you and
> Lynne. Everyone else at most didn't find these patches useful, but Lynne
> especially has been very aggressive to communicate their displeasure
> about fuzzing related fixes, and unjustifiably so.
> If some of these patches affect code you're interested into, then
> suggest nicer looking alternatives, like i did for the recent mpc8 patch.
>
> Patches like this one that fixes timeouts during fuzzing are usually
> validation checks to abort on invalid data early, and in this case they
> actually force an spec constrain in the correct place. They are hardly
> clutter. And if they use what you consider magic numbers or heuristics,
> or "sweep the issue under the carpet", you can always just request a
> comment to be placed next to them, like Reimar just did, or use some
> configurable value, like the recently added
> AVCodecContext.discard_damaged_percentage
>
> My point is, instead of complaining about UB fixes, try to be
> constructive about how to implement them better. Campaigning to migrate
> the codebase to something else than C and cut the issue from the root is
> more productive than replying to every other fuzzing fix like this.

This are not UB fixes, this are bandaids for some cases of too big
frame width & height sizes given as invalid input to decoder.
Instead of adding broken heuristic for every and each decoder one
should simply limit video sizes which are being fuzzed.