[FFmpeg-devel] [PATCH] avformat/utils: Don't parse encrypted packets.

[Jacob Trimble]

On Tue, Sep 11, 2018 at 1:48 PM Michael Niedermayer
<michael at niedermayer.cc> wrote:
>
> On Thu, Aug 30, 2018 at 08:43:25AM -0700, Jacob Trimble wrote:
> > On Wed, Aug 29, 2018 at 4:37 PM Michael Niedermayer
> > <michael at niedermayer.cc> wrote:
> > >
> > > On Wed, Aug 29, 2018 at 03:30:39PM -0700, Jacob Trimble wrote:
> > > > On Wed, Aug 29, 2018 at 3:20 PM James Almer <jamrial at gmail.com> wrote:
> > > > >
> > > > > On 8/29/2018 7:07 PM, Michael Niedermayer wrote:
> > > > > > On Tue, Aug 28, 2018 at 10:58:43AM -0700, Jacob Trimble wrote:
> > > > > >> If a packet is full-sample encrypted, then packet data can't be parsed
> > > > > >> without decrypting it.  So this skips the packet parsing for those
> > > > > >> packets.  If the packet has sub-sample encryption, it is assumed that
> > > > > >> the headers are in the clear and the parser will only need that info.
> > > > > >>
> > > > > >> Signed-off-by: Jacob Trimble <modmaker at google.com>
> > > > > >> ---
> > > > > >>  libavformat/utils.c | 21 ++++++++++++++++++++-
> > > > > >>  1 file changed, 20 insertions(+), 1 deletion(-)
> > > > > >
> > > > > > Hmm, please correct me if iam wrong but IIUC
> > > > > > 1. The demuxer has set need_parsing, indicating that it _requires_ a parser
> > > >
> > > > There isn't a flag for "try parsing and ignore errors".  So my choice
> > > > (from an app) is to either require parsing or don't do parsing at all
> > > > (which can result in incorrect timestamps).
> > > >
> > > > > > 2. Parsing cannot be done before decrypting
> > > > > >
> > > > > > If all this and the set values are correct, logically, the fix would be
> > > > > > to decrypt the packet before the parser not to skip the parser.
> > > >
> > >
> > > > There are cases where the packet can't be decrypted before getting to
> > > > this point.
> > >
> > > Can you elaborate on these cases ? I dont think its obvious what these
> > > cases are, at least its not obvious to me what exactly you are thinking
> > > of here. And i think its not helpfull if i guess what you mean and then
> > > argue/comment on that because maybe you meant something entirely different ...
> >
> > Fair warning, I am working on a DRM solution.  I know many of you may
> > not agree with DRM, but it is a requirement imposed by content
> > creators.  FFmpeg doesn't have to support DRM itself, but you should
> > still consider its usages.
> >
> > At that point in the code, the packet can only be decrypted by the
> > demuxer.  For example, in MP4 this can be done by passing the
> > -decryption_key parameter.  But that requires that FFmpeg handle the
> > decryption.  In my case, we are passing the packet to a CDM (content
> > decryption module) to handle the decryption and that might be a
> > hardware-secure decryptor.  In most DRM situations, we can't get the
> > raw key at all, all we can do is say "decrypt this block of memory".
> > So the only way to have the packet decrypted before this point would
> > be to introduce a callback to the app to decrypt the packet.
>
> 1. If the demuxer sets need_parsing it requires parsing
> 2. If the demuxer requires parsing then parsing should be done
> 3. If parsing cannot be done but is required thats a error and should produce
>    some warning or error message. It should not be part of some intended
>    design
>

So would a log be enough?  That's what happens if the parser fails to
be created or can't be found.

>
> >
> > This is why I have been working on exposing the encryption info.  My
> > app (or more correctly the CDM) needs to handle the decryption and we
> > can't just give the key to libavformat so the demuxer can decrypt the
> > packet.
> >
> > >
> > >
> > > > This point is between the demuxer creating the packet and
> > > > giving to the app.  So the only way to decrypt the frame (as of now)
> > > > is for the demuxer to do this.  There is no way for the app to handle
> > > > the decryption before this point.
> > > >
> > >
> > > > From the app's perspective, I would expect the packet to remain
> > > > encrypted for as long as possible.
> > >
> > > why ?
> >
> > Because the point of encryption is to ensure that nefarious parties
>
> you call your customers, "nefarious" ? I guess its none of my buisness
> but i dont think this is a good mindset.
>

The people who bought the movie are our customers; the people trying
to get the decrypted frames without paying, or trying to distribute
the movie for free are the nefarious ones.

>
> > don't get access to the data.  Even though the packet data is stored
> > in memory, it could still be retrieved by a malicious user.  Usually
> > for protected media, the frames are only decrypted when needed and in
> > some cases are decrypted in a secure CPU and isn't even accessible to
> > the app.  There are platforms that support what is called "direct
> > render" where the app gives the platform the encrypted packets and a
> > discrete hardware unit will decrypt the packet, decode it, and render
> > it directly; this happens on some smart TVs.  There are also cases
> > which require decrypt-decode where we give the platform an encrypted
> > packet and it will decrypt and decode the packet, ensuring the
> > decrypted packet data is never in main memory.  So there are cases
> > where we can't even see the encoded packet and decoding is handled by
> > the platform.
>
> you arent yet using the cortex decryptor module with neuralink?
>
> seriously, i think without this your nefarious customers will still be able to
> easily record the data if they want. All this fancy stuff doesnt really
> provide any real protection. And it makes people hate you/your company.
>
> Anyway, if you want to skip the parser on random subsets of packets (encrypted)
> i think this requires more care. For example i think the parser could
> buffer some data, if some random set of packets bypass the parser
> this could loose data at the point of switchover.
> For which parsers is that working, do they need flushing, ...
> it should not just be done while totally ignoring if the code can handle
> being switched like this
> also the interrested user who hits a problem (aka failure due to lack of
> requested parsing) and increases verbosity should have a chance to
> find the cause. There should be some sort of debug or warning message
> that requested parsing is not done.

So how about, when we see an encrypted frame, we flush the parser
before skipping the frame?  Can we just flush the parser and then
reuse it later?  Or would we need to create a new parser if we saw
clear frames later?

>
> Still, personally i think it would be better to decrypt before the
> parser, its not really going to make a difference for someone who
> wants to access the data but it makes the code cleaner and more robust

Like I said, in some cases, this is impossible.  If we are running on
an Android device or a smart TV, we'll use a hardware secure decoder
where the OS will decrypt and decode the video in a secure chip to
avoid having the encoded frame in the clear at all.  So we literally
can't decrypt the frames ourselves.  Even if we could decrypt (for
software decoding), we can't under the current design since the app
needs to decrypt the data and the app can't get to the data before
this point.

>
> Thanks
>
> [...]
> --
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> The greatest way to live with honor in this world is to be what we pretend
> to be. -- Socrates
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel