[FFmpeg-devel,03/14] pthread_frame: merge the functionality for normal decoder init and init_thread_copy

The current design, where
- proper init is called for the first per-thread context
- first thread's private data is copied into private data for all the
  other threads
- a "fixup" function is called for all the other threads to e.g.
  allocate dynamically allocated data
is very fragile and hard to follow, so it is abandoned. Instead, the
same init function is used to init each per-thread context. Where
necessary, AVCodecInternal.is_copy can be used to differentiate between
the first thread and the other ones (e.g. for decoding the extradata
just once).
---
 doc/multithreading.txt     |  3 ---
 libavcodec/aic.c           |  1 -
 libavcodec/alac.c          | 10 ---------
 libavcodec/avcodec.h       |  6 -----
 libavcodec/cfhd.c          |  6 ++---
 libavcodec/cllc.c          | 14 ------------
 libavcodec/dnxhddec.c      | 16 --------------
 libavcodec/exr.c           | 15 -------------
 libavcodec/ffv1dec.c       | 29 ------------------------
 libavcodec/flacdec.c       | 14 ------------
 libavcodec/h264dec.c       | 38 ++++++++------------------------
 libavcodec/hevcdec.c       | 29 ++++++------------------
 libavcodec/hqx.c           |  3 ---
 libavcodec/huffyuvdec.c    | 32 ---------------------------
 libavcodec/lagarith.c      | 11 ----------
 libavcodec/lcldec.c        |  9 --------
 libavcodec/magicyuv.c      | 16 --------------
 libavcodec/mdec.c          | 12 ----------
 libavcodec/mimic.c         | 22 +------------------
 libavcodec/mpeg4videodec.c | 10 ++++-----
 libavcodec/pixlet.c        | 16 --------------
 libavcodec/pngdec.c        |  6 +----
 libavcodec/proresdec2.c    | 12 ----------
 libavcodec/pthread_frame.c | 31 +++++++++++++++-----------
 libavcodec/rv30.c          |  1 -
 libavcodec/rv34.c          | 28 ------------------------
 libavcodec/rv34.h          |  1 -
 libavcodec/rv40.c          |  1 -
 libavcodec/sheervideo.c    | 14 ------------
 libavcodec/takdec.c        |  8 -------
 libavcodec/tiff.c          |  1 -
 libavcodec/tta.c           |  8 -------
 libavcodec/vble.c          |  1 -
 libavcodec/vp3.c           | 45 --------------------------------------
 libavcodec/vp8.c           | 16 --------------
 libavcodec/vp9.c           |  6 -----
 libavcodec/wavpack.c       | 15 -------------
 libavcodec/ylc.c           | 19 ----------------
 38 files changed, 42 insertions(+), 483 deletions(-)

On 3/27/20 5:57 AM, Anton Khirnov wrote:
> The current design, where
> - proper init is called for the first per-thread context
> - first thread's private data is copied into private data for all the
>   other threads
> - a "fixup" function is called for all the other threads to e.g.
>   allocate dynamically allocated data
> is very fragile and hard to follow, so it is abandoned. Instead, the
> same init function is used to init each per-thread context. Where
> necessary, AVCodecInternal.is_copy can be used to differentiate between
> the first thread and the other ones (e.g. for decoding the extradata
> just once).
> ---

I'm not sure I fully understand this change. You mention that AVCodecInternal.is_copy can be used where different treatment
might be necessary for subsequent threads, and I see that done in a couple places, but in most cases you have simply deleted the
init_thread_copy() function even when it's not at all obvious (to me anyway) that that won't break the codec.

Specifically this will break WavPack because now it will be allocating a new dsdctx for every thread instead of sharing one
between all threads. I am happy to fix and test this case once the patch goes in, but is the intent of this patch that the
respective authors will find and fix all the breakages? Or did I just happen to catch the one case you missed?

Cheers!

-David

Quoting David Bryant (2020-03-27 23:51:19)
> On 3/27/20 5:57 AM, Anton Khirnov wrote:
> > The current design, where
> > - proper init is called for the first per-thread context
> > - first thread's private data is copied into private data for all the
> >   other threads
> > - a "fixup" function is called for all the other threads to e.g.
> >   allocate dynamically allocated data
> > is very fragile and hard to follow, so it is abandoned. Instead, the
> > same init function is used to init each per-thread context. Where
> > necessary, AVCodecInternal.is_copy can be used to differentiate between
> > the first thread and the other ones (e.g. for decoding the extradata
> > just once).
> > ---
> 
> I'm not sure I fully understand this change. You mention that AVCodecInternal.is_copy can be used where different treatment
> might be necessary for subsequent threads, and I see that done in a couple places, but in most cases you have simply deleted the
> init_thread_copy() function even when it's not at all obvious (to me anyway) that that won't break the codec.

In most cases, just deleting init_thread_copy() is the right thing to
do. E.g. all decoders that do not implement update_thread_context() have
to be intra-only, so every frame thread is effectively a completely
standalone decoder. And in most of the more complex decoders (like h264)
the important parameters are dynamically changeable during decoding, so
not that much is done in decoder init beyond allocating some stuff that
does not depend on the bistream properties.

My intent is for each frame-thread worker to be treated inasmuch as
possible as a standalone decoder, and where it has to share data with
other threads to make this sharing explicit (rather than implicit as is
the case now).

> 
> Specifically this will break WavPack because now it will be allocating a new dsdctx for every thread instead of sharing one
> between all threads. I am happy to fix and test this case once the patch goes in, but is the intent of this patch that the
> respective authors will find and fix all the breakages? Or did I just happen to catch the one case you missed?

I certainly intended to convert the decoders correctly myself,
apparently I didn't pay enough attention to the recent wavpack changes.
Hopefully the other conversions are correct, but I will go through the
changes again to check.

Looking at wavpack, the code looks suspicious to me. You allocate one
dsdctx per channel at init, but AFAIU the number of channels can change
at each frame.

On 3/28/20 6:23 AM, Anton Khirnov wrote:
> Quoting David Bryant (2020-03-27 23:51:19)
>> On 3/27/20 5:57 AM, Anton Khirnov wrote:
>>> The current design, where
>>> - proper init is called for the first per-thread context
>>> - first thread's private data is copied into private data for all the
>>>   other threads
>>> - a "fixup" function is called for all the other threads to e.g.
>>>   allocate dynamically allocated data
>>> is very fragile and hard to follow, so it is abandoned. Instead, the
>>> same init function is used to init each per-thread context. Where
>>> necessary, AVCodecInternal.is_copy can be used to differentiate between
>>> the first thread and the other ones (e.g. for decoding the extradata
>>> just once).
>>> ---
>> I'm not sure I fully understand this change. You mention that AVCodecInternal.is_copy can be used where different treatment
>> might be necessary for subsequent threads, and I see that done in a couple places, but in most cases you have simply deleted the
>> init_thread_copy() function even when it's not at all obvious (to me anyway) that that won't break the codec.
> In most cases, just deleting init_thread_copy() is the right thing to
> do. E.g. all decoders that do not implement update_thread_context() have
> to be intra-only, so every frame thread is effectively a completely
> standalone decoder. And in most of the more complex decoders (like h264)
> the important parameters are dynamically changeable during decoding, so
> not that much is done in decoder init beyond allocating some stuff that
> does not depend on the bistream properties.
>
> My intent is for each frame-thread worker to be treated inasmuch as
> possible as a standalone decoder, and where it has to share data with
> other threads to make this sharing explicit (rather than implicit as is
> the case now).

Yes, this makes sense. The confusing part is when the decode_init() function looks completely different than the
init_thread_copy() function. This is often because the decode_init() function is generating things (tables, etc.) from scratch
and the init_thread_copy() is just copying the necessary things from the original. In cases where the original generation might
be time consuming this might make sense, but usually it's probably just making the code more complex and difficult to follow
(which I believe was your original point).

One possible interim solution for complex cases that break would be to leave the init_thread_copy() function there, but instead
of having it in the AVCodec struct and called from outside (which is no longer possible), it simply gets called first thing from
decode_init() if AVCodecInternal.is_copy is set. That way the architecture is cleaned up now, and the codec won't break and can
be cleaned up when time permits. Just a thought.

>
>> Specifically this will break WavPack because now it will be allocating a new dsdctx for every thread instead of sharing one
>> between all threads. I am happy to fix and test this case once the patch goes in, but is the intent of this patch that the
>> respective authors will find and fix all the breakages? Or did I just happen to catch the one case you missed?
> I certainly intended to convert the decoders correctly myself,
> apparently I didn't pay enough attention to the recent wavpack changes.
> Hopefully the other conversions are correct, but I will go through the
> changes again to check.
>
> Looking at wavpack, the code looks suspicious to me. You allocate one
> dsdctx per channel at init, but AFAIU the number of channels can change
> at each frame.
>
Multichannel WavPack consists of sequences of stereo or mono WavPack blocks that include flags indicating whether they are the
beginning or end of a "frame" or "packet". This allows the number of channels available to be essentially unlimited, however the
total number of channels in a stream may not change. When decoding, if a sequence of blocks generates more or less than the
correct number of channels, this is flagged as an error and overrun is prevented (see libavcodec/wavpack.c, line 1499 and line
1621).

If you are curious, the WavPack format is described in detail here:

https://github.com/dbry/WavPack/blob/master/doc/WavPack5FileFormat.pdf

Thanks!

-David

Quoting David Bryant (2020-03-28 21:22:40)
> On 3/28/20 6:23 AM, Anton Khirnov wrote:
> > Quoting David Bryant (2020-03-27 23:51:19)
> >> On 3/27/20 5:57 AM, Anton Khirnov wrote:
> >>> The current design, where
> >>> - proper init is called for the first per-thread context
> >>> - first thread's private data is copied into private data for all the
> >>>   other threads
> >>> - a "fixup" function is called for all the other threads to e.g.
> >>>   allocate dynamically allocated data
> >>> is very fragile and hard to follow, so it is abandoned. Instead, the
> >>> same init function is used to init each per-thread context. Where
> >>> necessary, AVCodecInternal.is_copy can be used to differentiate between
> >>> the first thread and the other ones (e.g. for decoding the extradata
> >>> just once).
> >>> ---
> >> I'm not sure I fully understand this change. You mention that AVCodecInternal.is_copy can be used where different treatment
> >> might be necessary for subsequent threads, and I see that done in a couple places, but in most cases you have simply deleted the
> >> init_thread_copy() function even when it's not at all obvious (to me anyway) that that won't break the codec.
> > In most cases, just deleting init_thread_copy() is the right thing to
> > do. E.g. all decoders that do not implement update_thread_context() have
> > to be intra-only, so every frame thread is effectively a completely
> > standalone decoder. And in most of the more complex decoders (like h264)
> > the important parameters are dynamically changeable during decoding, so
> > not that much is done in decoder init beyond allocating some stuff that
> > does not depend on the bistream properties.
> >
> > My intent is for each frame-thread worker to be treated inasmuch as
> > possible as a standalone decoder, and where it has to share data with
> > other threads to make this sharing explicit (rather than implicit as is
> > the case now).
> 
> Yes, this makes sense. The confusing part is when the decode_init() function looks completely different than the
> init_thread_copy() function. This is often because the decode_init() function is generating things (tables, etc.) from scratch
> and the init_thread_copy() is just copying the necessary things from the original. In cases where the original generation might
> be time consuming this might make sense, but usually it's probably just making the code more complex and difficult to follow
> (which I believe was your original point).

Yes, precisely. I believe for obscure codecs it's better to waste a
little memory when doing threaded decoding, in return for simpler code.
And if people really care about being as memory-efficient as possible
then doing explicit refcounting on allocated buffers is still possible.

> 
> One possible interim solution for complex cases that break would be to leave the init_thread_copy() function there, but instead
> of having it in the AVCodec struct and called from outside (which is no longer possible), it simply gets called first thing from
> decode_init() if AVCodecInternal.is_copy is set. That way the architecture is cleaned up now, and the codec won't break and can
> be cleaned up when time permits. Just a thought.

I don't really see much point in any interim solutions, since no codecs
should break. You found a problem with wavpack, that's going to be fixed
in the next iteration of this patch.
Speaking of which, ideally we should have a FATE test for DSD files so
that I would have found the failure myself when running test.

> 
> >
> >> Specifically this will break WavPack because now it will be allocating a new dsdctx for every thread instead of sharing one
> >> between all threads. I am happy to fix and test this case once the patch goes in, but is the intent of this patch that the
> >> respective authors will find and fix all the breakages? Or did I just happen to catch the one case you missed?
> > I certainly intended to convert the decoders correctly myself,
> > apparently I didn't pay enough attention to the recent wavpack changes.
> > Hopefully the other conversions are correct, but I will go through the
> > changes again to check.
> >
> > Looking at wavpack, the code looks suspicious to me. You allocate one
> > dsdctx per channel at init, but AFAIU the number of channels can change
> > at each frame.
> >
> Multichannel WavPack consists of sequences of stereo or mono WavPack blocks that include flags indicating whether they are the
> beginning or end of a "frame" or "packet". This allows the number of channels available to be essentially unlimited, however the
> total number of channels in a stream may not change. When decoding, if a sequence of blocks generates more or less than the
> correct number of channels, this is flagged as an error and overrun is prevented (see libavcodec/wavpack.c, line 1499 and line
> 1621).

If my reading of the code is correct, line 1499 only checks for overflow
of the channel count read for the current sequence. That channel count
is exported to AVCodeContext in the block starting on line 1464 and
I don't see any checks for whether it is equal to the initial one.

On 3/31/20 2:47 AM, Anton Khirnov wrote:
> Quoting David Bryant (2020-03-28 21:22:40)
>> On 3/28/20 6:23 AM, Anton Khirnov wrote:
>>> Quoting David Bryant (2020-03-27 23:51:19)
>>>> On 3/27/20 5:57 AM, Anton Khirnov wrote:
>>>>> The current design, where
>>>>> - proper init is called for the first per-thread context
>>>>> - first thread's private data is copied into private data for all the
>>>>>   other threads
>>>>> - a "fixup" function is called for all the other threads to e.g.
>>>>>   allocate dynamically allocated data
>>>>> is very fragile and hard to follow, so it is abandoned. Instead, the
>>>>> same init function is used to init each per-thread context. Where
>>>>> necessary, AVCodecInternal.is_copy can be used to differentiate between
>>>>> the first thread and the other ones (e.g. for decoding the extradata
>>>>> just once).
>>>>> ---
>>>> I'm not sure I fully understand this change. You mention that AVCodecInternal.is_copy can be used where different treatment
>>>> might be necessary for subsequent threads, and I see that done in a couple places, but in most cases you have simply deleted the
>>>> init_thread_copy() function even when it's not at all obvious (to me anyway) that that won't break the codec.
>>> In most cases, just deleting init_thread_copy() is the right thing to
>>> do. E.g. all decoders that do not implement update_thread_context() have
>>> to be intra-only, so every frame thread is effectively a completely
>>> standalone decoder. And in most of the more complex decoders (like h264)
>>> the important parameters are dynamically changeable during decoding, so
>>> not that much is done in decoder init beyond allocating some stuff that
>>> does not depend on the bistream properties.
>>>
>>> My intent is for each frame-thread worker to be treated inasmuch as
>>> possible as a standalone decoder, and where it has to share data with
>>> other threads to make this sharing explicit (rather than implicit as is
>>> the case now).
>> Yes, this makes sense. The confusing part is when the decode_init() function looks completely different than the
>> init_thread_copy() function. This is often because the decode_init() function is generating things (tables, etc.) from scratch
>> and the init_thread_copy() is just copying the necessary things from the original. In cases where the original generation might
>> be time consuming this might make sense, but usually it's probably just making the code more complex and difficult to follow
>> (which I believe was your original point).
> Yes, precisely. I believe for obscure codecs it's better to waste a
> little memory when doing threaded decoding, in return for simpler code.
> And if people really care about being as memory-efficient as possible
> then doing explicit refcounting on allocated buffers is still possible.
>
>> One possible interim solution for complex cases that break would be to leave the init_thread_copy() function there, but instead
>> of having it in the AVCodec struct and called from outside (which is no longer possible), it simply gets called first thing from
>> decode_init() if AVCodecInternal.is_copy is set. That way the architecture is cleaned up now, and the codec won't break and can
>> be cleaned up when time permits. Just a thought.
> I don't really see much point in any interim solutions, since no codecs
> should break. You found a problem with wavpack, that's going to be fixed
> in the next iteration of this patch.
> Speaking of which, ideally we should have a FATE test for DSD files so
> that I would have found the failure myself when running test.

I have created a WavPack DSD test file that exercises all three of the DSD modes (fast, high, and copy). I have also verified
that it would have caught this change, although I had to specify fate multithreading once I discovered it wasn't the default. If
someone can add this to the fate suite (in the wavpack/lossless folder) then I can supply a patch to add the test:

file: http://www.wavpack.com/dsd.wv (MD5: 74b2181f3e9829d9a5b98edd037984ac)
decoded MD5 (f32le): a3a88bba95f809025dce01ecb9064091

>>>> Specifically this will break WavPack because now it will be allocating a new dsdctx for every thread instead of sharing one
>>>> between all threads. I am happy to fix and test this case once the patch goes in, but is the intent of this patch that the
>>>> respective authors will find and fix all the breakages? Or did I just happen to catch the one case you missed?
>>> I certainly intended to convert the decoders correctly myself,
>>> apparently I didn't pay enough attention to the recent wavpack changes.
>>> Hopefully the other conversions are correct, but I will go through the
>>> changes again to check.
>>>
>>> Looking at wavpack, the code looks suspicious to me. You allocate one
>>> dsdctx per channel at init, but AFAIU the number of channels can change
>>> at each frame.
>>>
>> Multichannel WavPack consists of sequences of stereo or mono WavPack blocks that include flags indicating whether they are the
>> beginning or end of a "frame" or "packet". This allows the number of channels available to be essentially unlimited, however the
>> total number of channels in a stream may not change. When decoding, if a sequence of blocks generates more or less than the
>> correct number of channels, this is flagged as an error and overrun is prevented (see libavcodec/wavpack.c, line 1499 and line
>> 1621).
> If my reading of the code is correct, line 1499 only checks for overflow
> of the channel count read for the current sequence. That channel count
> is exported to AVCodeContext in the block starting on line 1464 and
> I don't see any checks for whether it is equal to the initial one.
>
Full disclosure: I did not write the WavPack decoder. I understand _how_ it works, but I don't necessarily understand _why_ it
was written the way it was due to gaps in my knowledge of the FFmpeg internals, and there are certainly things in there that
don't look quite right to me.

That said, I have tested it thoroughly and it handles everything I throw at it, and I have extended it to handle the DSD mode
without anything unexpected happening. And of course it seems to handle extensive fuzzing, and the recent error was related to
my DSD change, suggesting that the core it pretty robust. My inclination has been to not "fix" anything that wasn't broken.

Yes, that channel count stuff is a little difficult to follow. I could experiment with streams that abruptly change channel
configuration, but again I'm hesitant to change code that's worked so well so long.

Kind regards,

David

Quoting David Bryant (2020-04-01 23:35:13)
> On 3/31/20 2:47 AM, Anton Khirnov wrote:
> >>> Looking at wavpack, the code looks suspicious to me. You allocate one
> >>> dsdctx per channel at init, but AFAIU the number of channels can change
> >>> at each frame.
> >>>
> >> Multichannel WavPack consists of sequences of stereo or mono WavPack blocks that include flags indicating whether they are the
> >> beginning or end of a "frame" or "packet". This allows the number of channels available to be essentially unlimited, however the
> >> total number of channels in a stream may not change. When decoding, if a sequence of blocks generates more or less than the
> >> correct number of channels, this is flagged as an error and overrun is prevented (see libavcodec/wavpack.c, line 1499 and line
> >> 1621).
> > If my reading of the code is correct, line 1499 only checks for overflow
> > of the channel count read for the current sequence. That channel count
> > is exported to AVCodeContext in the block starting on line 1464 and
> > I don't see any checks for whether it is equal to the initial one.
> >
> Full disclosure: I did not write the WavPack decoder. I understand _how_ it works, but I don't necessarily understand _why_ it
> was written the way it was due to gaps in my knowledge of the FFmpeg internals, and there are certainly things in there that
> don't look quite right to me.
> 
> That said, I have tested it thoroughly and it handles everything I throw at it, and I have extended it to handle the DSD mode
> without anything unexpected happening. And of course it seems to handle extensive fuzzing, and the recent error was related to
> my DSD change, suggesting that the core it pretty robust. My inclination has been to not "fix" anything that wasn't broken.
> 
> Yes, that channel count stuff is a little difficult to follow. I could experiment with streams that abruptly change channel
> configuration, but again I'm hesitant to change code that's worked so well so long.

I can easily get the decoder to crash even in single-threaded mode. Just
make the demuxer always say the stream is mono, then try decoding your
DSD sample.

Also, I looked at the specification PDF you linked in the previous email
and I see nothing in it that explicitly forbids the channel count from
changing from one "superblock" to the next. It seems to me like you
could just concatenate wavpack files with different channel counts and
get a valid wavpack file.

On 4/2/20 2:13 AM, Anton Khirnov wrote:
> Quoting David Bryant (2020-04-01 23:35:13)
>> On 3/31/20 2:47 AM, Anton Khirnov wrote:
>>>>> Looking at wavpack, the code looks suspicious to me. You allocate one
>>>>> dsdctx per channel at init, but AFAIU the number of channels can change
>>>>> at each frame.
>>>>>
>>>> Multichannel WavPack consists of sequences of stereo or mono WavPack blocks that include flags indicating whether they are the
>>>> beginning or end of a "frame" or "packet". This allows the number of channels available to be essentially unlimited, however the
>>>> total number of channels in a stream may not change. When decoding, if a sequence of blocks generates more or less than the
>>>> correct number of channels, this is flagged as an error and overrun is prevented (see libavcodec/wavpack.c, line 1499 and line
>>>> 1621).
>>> If my reading of the code is correct, line 1499 only checks for overflow
>>> of the channel count read for the current sequence. That channel count
>>> is exported to AVCodeContext in the block starting on line 1464 and
>>> I don't see any checks for whether it is equal to the initial one.
>>>
>> Full disclosure: I did not write the WavPack decoder. I understand _how_ it works, but I don't necessarily understand _why_ it
>> was written the way it was due to gaps in my knowledge of the FFmpeg internals, and there are certainly things in there that
>> don't look quite right to me.
>>
>> That said, I have tested it thoroughly and it handles everything I throw at it, and I have extended it to handle the DSD mode
>> without anything unexpected happening. And of course it seems to handle extensive fuzzing, and the recent error was related to
>> my DSD change, suggesting that the core it pretty robust. My inclination has been to not "fix" anything that wasn't broken.
>>
>> Yes, that channel count stuff is a little difficult to follow. I could experiment with streams that abruptly change channel
>> configuration, but again I'm hesitant to change code that's worked so well so long.
> I can easily get the decoder to crash even in single-threaded mode. Just
> make the demuxer always say the stream is mono, then try decoding your
> DSD sample.

Okay, I reproduced that. However, I'm not sure it's really that serious because the demuxer and the decoder use basically the
same logic to determine that number of channels, so it would be tricky to craft a file that caused the demuxer to calculate 1
channel and the decoder to reach a different result, which is probably why the fuzzers have not been able to do so.

That said, it would not be hard to have the decoder save away the initial channel count from avctx, never overwrite it, and
verify that all superblocks match that count. I don't know off hand whether that would break anything though. Is it guaranteed
that avctx->channels is correct during the call to decode_init(), even if it's being demuxed from Matroska, for example? If so,
why is the WavPack decoder _ever_ writing to the avctx channel count?

>
> Also, I looked at the specification PDF you linked in the previous email
> and I see nothing in it that explicitly forbids the channel count from
> changing from one "superblock" to the next. It seems to me like you
> could just concatenate wavpack files with different channel counts and
> get a valid wavpack file.
>
You're right, I could not find that being explicitly forbidden. Nevertheless, it _is_ forbidden for any stream parameters to
change (e.g. data format, sampling rate, channel count or mapping) and both libwavpack and the FFmpeg demuxer enforce this (see
libavformat/wvdec.c starting at line 211), or at least gracefully fail without crashing.

The reason it's never mentioned is probably just because it never occurred to me. None of the source file formats for WavPack
(e.g., wav, caf, dsdiff) allow format changes and neither do any of the other lossless compressors I am aware of. I can't
imagine how that would work, and extensive fuzzing has ensured that pathological files like you suggest are unlikely to be the
source of exploits.

Kind regards,

David

Quoting David Bryant (2020-04-03 07:14:21)
> On 4/2/20 2:13 AM, Anton Khirnov wrote:
> > Quoting David Bryant (2020-04-01 23:35:13)
> >> On 3/31/20 2:47 AM, Anton Khirnov wrote:
> >>>>> Looking at wavpack, the code looks suspicious to me. You allocate one
> >>>>> dsdctx per channel at init, but AFAIU the number of channels can change
> >>>>> at each frame.
> >>>>>
> >>>> Multichannel WavPack consists of sequences of stereo or mono WavPack blocks that include flags indicating whether they are the
> >>>> beginning or end of a "frame" or "packet". This allows the number of channels available to be essentially unlimited, however the
> >>>> total number of channels in a stream may not change. When decoding, if a sequence of blocks generates more or less than the
> >>>> correct number of channels, this is flagged as an error and overrun is prevented (see libavcodec/wavpack.c, line 1499 and line
> >>>> 1621).
> >>> If my reading of the code is correct, line 1499 only checks for overflow
> >>> of the channel count read for the current sequence. That channel count
> >>> is exported to AVCodeContext in the block starting on line 1464 and
> >>> I don't see any checks for whether it is equal to the initial one.
> >>>
> >> Full disclosure: I did not write the WavPack decoder. I understand _how_ it works, but I don't necessarily understand _why_ it
> >> was written the way it was due to gaps in my knowledge of the FFmpeg internals, and there are certainly things in there that
> >> don't look quite right to me.
> >>
> >> That said, I have tested it thoroughly and it handles everything I throw at it, and I have extended it to handle the DSD mode
> >> without anything unexpected happening. And of course it seems to handle extensive fuzzing, and the recent error was related to
> >> my DSD change, suggesting that the core it pretty robust. My inclination has been to not "fix" anything that wasn't broken.
> >>
> >> Yes, that channel count stuff is a little difficult to follow. I could experiment with streams that abruptly change channel
> >> configuration, but again I'm hesitant to change code that's worked so well so long.
> > I can easily get the decoder to crash even in single-threaded mode. Just
> > make the demuxer always say the stream is mono, then try decoding your
> > DSD sample.
> 
> Okay, I reproduced that. However, I'm not sure it's really that serious because the demuxer and the decoder use basically the
> same logic to determine that number of channels, so it would be tricky to craft a file that caused the demuxer to calculate 1
> channel and the decoder to reach a different result, which is probably why the fuzzers have not been able to do so.

I disagree - this is just as serious as any other crash on invalid data.

The decoder must not ever crash for arbitrary input data. It cannot
assume that it's always going to be provided valid input or that it's
always used with some specific demuxer. The caller might not be using
libavformat at all, libavcodec should work with any source of data.

Besides, I can reproduce the crash with no code modification necessary
by muxing your sample into Matroska and then hex-editing the file to say
it has only one channel.

> 
> That said, it would not be hard to have the decoder save away the initial channel count from avctx, never overwrite it, and
> verify that all superblocks match that count. I don't know off hand whether that would break anything though. Is it guaranteed
> that avctx->channels is correct during the call to decode_init(), even if it's being demuxed from Matroska, for example? If so,
> why is the WavPack decoder _ever_ writing to the avctx channel count?

The question is what you define as correct. The initial value provided
by the caller is supposed to be the caller's best guess at the correct
(initial) value. It may come from a demuxer, or from the Delphic Oracle.
The decoder doesn't know.

For some codecs, the relevant parameters are not specified in the
bistream, so they have no choice but to use the caller's value. That is
not the case for WavPack, since the channel count IS written in the
bistream. So I think it's best to ignore the caller-provided value and
always use the one we get from the data.

> 
> >
> > Also, I looked at the specification PDF you linked in the previous email
> > and I see nothing in it that explicitly forbids the channel count from
> > changing from one "superblock" to the next. It seems to me like you
> > could just concatenate wavpack files with different channel counts and
> > get a valid wavpack file.
> >
> You're right, I could not find that being explicitly forbidden. Nevertheless, it _is_ forbidden for any stream parameters to
> change (e.g. data format, sampling rate, channel count or mapping) and both libwavpack and the FFmpeg demuxer enforce this (see
> libavformat/wvdec.c starting at line 211), or at least gracefully fail without crashing.

Forbidden by whom? From what I can see:
- sample format: the wv demuxer does not set it at all, the decoder sets
  it in wavpack_decode_frame() from the block header; also, Michael sent
  a patch recently to deal with a wavpack decoder crash related to
  format changes
- sample rate: set by wv demuxer and then checked that it does not
  change; but the caller is not necessarily using the wv demuxer and the
  decoder will happily change it the same way it does with channel count

> 
> The reason it's never mentioned is probably just because it never occurred to me. None of the source file formats for WavPack
> (e.g., wav, caf, dsdiff) allow format changes and neither do any of the other lossless compressors I am aware of.

Correct, I cannot see any other lossless codecs that would allow this.
But then again all these codecs have global headers. So it's not that
they artificially forbid format changes, but the bistream format is not
able to represent them.

That is different for WavPack - it goes out of its way to make its
blocks independent with no global headers. All other codecs that work
like this allow format changes, so it seems natural that WavPack should
as well.

As I said before - it seems to me that you could simply concatenate two
WavPack files and get a valid WavPack file.

> I can't imagine how that would work,

Why not? We have plenty of decoders where the format or channel count
can change whenever.

> and extensive fuzzing has ensured that pathological files like you
> suggest are unlikely to be the source of exploits.

See the crash above. I think fuzzing is very overhyped these days. It's
a useful tool for sure, but one should not rely purely on fuzzing too
much. It may not cover important parts of the input space because of the
way it is set up. Or there could be exploitable codepaths that require
too specific inputs for a random process to find, but a determined
attacker could construct them from reading the code.

On 4/2/20 11:32 PM, Anton Khirnov wrote:
> Quoting David Bryant (2020-04-03 07:14:21)
>> On 4/2/20 2:13 AM, Anton Khirnov wrote:
>>> Quoting David Bryant (2020-04-01 23:35:13)
>>>> On 3/31/20 2:47 AM, Anton Khirnov wrote:
>>>>>>> Looking at wavpack, the code looks suspicious to me. You allocate one
>>>>>>> dsdctx per channel at init, but AFAIU the number of channels can change
>>>>>>> at each frame.
>>>>>>>
>>>>>> Multichannel WavPack consists of sequences of stereo or mono WavPack blocks that include flags indicating whether they are the
>>>>>> beginning or end of a "frame" or "packet". This allows the number of channels available to be essentially unlimited, however the
>>>>>> total number of channels in a stream may not change. When decoding, if a sequence of blocks generates more or less than the
>>>>>> correct number of channels, this is flagged as an error and overrun is prevented (see libavcodec/wavpack.c, line 1499 and line
>>>>>> 1621).
>>>>> If my reading of the code is correct, line 1499 only checks for overflow
>>>>> of the channel count read for the current sequence. That channel count
>>>>> is exported to AVCodeContext in the block starting on line 1464 and
>>>>> I don't see any checks for whether it is equal to the initial one.
>>>>>
>>>> Full disclosure: I did not write the WavPack decoder. I understand _how_ it works, but I don't necessarily understand _why_ it
>>>> was written the way it was due to gaps in my knowledge of the FFmpeg internals, and there are certainly things in there that
>>>> don't look quite right to me.
>>>>
>>>> That said, I have tested it thoroughly and it handles everything I throw at it, and I have extended it to handle the DSD mode
>>>> without anything unexpected happening. And of course it seems to handle extensive fuzzing, and the recent error was related to
>>>> my DSD change, suggesting that the core it pretty robust. My inclination has been to not "fix" anything that wasn't broken.
>>>>
>>>> Yes, that channel count stuff is a little difficult to follow. I could experiment with streams that abruptly change channel
>>>> configuration, but again I'm hesitant to change code that's worked so well so long.
>>> I can easily get the decoder to crash even in single-threaded mode. Just
>>> make the demuxer always say the stream is mono, then try decoding your
>>> DSD sample.
>> Okay, I reproduced that. However, I'm not sure it's really that serious because the demuxer and the decoder use basically the
>> same logic to determine that number of channels, so it would be tricky to craft a file that caused the demuxer to calculate 1
>> channel and the decoder to reach a different result, which is probably why the fuzzers have not been able to do so.
> I disagree - this is just as serious as any other crash on invalid data.
>
> The decoder must not ever crash for arbitrary input data. It cannot
> assume that it's always going to be provided valid input or that it's
> always used with some specific demuxer. The caller might not be using
> libavformat at all, libavcodec should work with any source of data.
>
> Besides, I can reproduce the crash with no code modification necessary
> by muxing your sample into Matroska and then hex-editing the file to say
> it has only one channel.

Yes, I have reproduced this crash even without Matroska. I was working on a solution but you beat me to it. Thanks!


>
>> That said, it would not be hard to have the decoder save away the initial channel count from avctx, never overwrite it, and
>> verify that all superblocks match that count. I don't know off hand whether that would break anything though. Is it guaranteed
>> that avctx->channels is correct during the call to decode_init(), even if it's being demuxed from Matroska, for example? If so,
>> why is the WavPack decoder _ever_ writing to the avctx channel count?
> The question is what you define as correct. The initial value provided
> by the caller is supposed to be the caller's best guess at the correct
> (initial) value. It may come from a demuxer, or from the Delphic Oracle.
> The decoder doesn't know.
>
> For some codecs, the relevant parameters are not specified in the
> bistream, so they have no choice but to use the caller's value. That is
> not the case for WavPack, since the channel count IS written in the
> bistream. So I think it's best to ignore the caller-provided value and
> always use the one we get from the data.

Agreed.


>
>>> Also, I looked at the specification PDF you linked in the previous email
>>> and I see nothing in it that explicitly forbids the channel count from
>>> changing from one "superblock" to the next. It seems to me like you
>>> could just concatenate wavpack files with different channel counts and
>>> get a valid wavpack file.
>>>
>> You're right, I could not find that being explicitly forbidden. Nevertheless, it _is_ forbidden for any stream parameters to
>> change (e.g. data format, sampling rate, channel count or mapping) and both libwavpack and the FFmpeg demuxer enforce this (see
>> libavformat/wvdec.c starting at line 211), or at least gracefully fail without crashing.
> Forbidden by whom? From what I can see:
> - sample format: the wv demuxer does not set it at all, the decoder sets
>   it in wavpack_decode_frame() from the block header; also, Michael sent
>   a patch recently to deal with a wavpack decoder crash related to
>   format changes
> - sample rate: set by wv demuxer and then checked that it does not
>   change; but the caller is not necessarily using the wv demuxer and the
>   decoder will happily change it the same way it does with channel count
>
>> The reason it's never mentioned is probably just because it never occurred to me. None of the source file formats for WavPack
>> (e.g., wav, caf, dsdiff) allow format changes and neither do any of the other lossless compressors I am aware of.
> Correct, I cannot see any other lossless codecs that would allow this.
> But then again all these codecs have global headers. So it's not that
> they artificially forbid format changes, but the bistream format is not
> able to represent them.

I disagree with that. It would be trivial to add a new STREAMINFO block to a FLAC file and change the format after that. This is
not specifically prohibited in the FLAC specs, but it's sort of implied when it says that the initial STREAMINFO block has
"information about the whole stream".

WAV files could certainly have multiple <fmt> and <data> pairs in a file, and I could not find any mention of this being
prohibited, but you would be hard pressed to find a program that would deal with that correctly.

And WavPack actually does have a global header of sorts. The first block that contains audio determines the format of the entire
file. I could not find that explicitly stated anywhere, but it is how it works, and changing the format in the middle of the
file is undefined behavior (although it should not crash, obviously).


>
> That is different for WavPack - it goes out of its way to make its
> blocks independent with no global headers. All other codecs that work
> like this allow format changes, so it seems natural that WavPack should
> as well.
>
> As I said before - it seems to me that you could simply concatenate two
> WavPack files and get a valid WavPack file.

Well, you actually can't produce a valid file that way because the WavPack blocks include information required for seeking. But
if you got around that it would still not be a valid file, by convention.

>
>> I can't imagine how that would work,
> Why not? We have plenty of decoders where the format or channel count
> can change whenever.

Well, there are all sorts of reasons why this would not be appropriate for native WavPack files, and what they are used for
today. For example, if the sample rate changed it would be impossible to determine the duration without scanning the whole file.
Also, libwavpack does not have the ability to generate such files nor the APIs that would allow decoding them in any reasonable way.

That said, I am only referring to the native WavPack container file format (.wv and .wvc). If WavPack streams are embedded in
another container (e.g., Matroska) then this is fine (assuming the container format is happy with it) because, as you say, each
WavPack audio block is independent.


>
>> and extensive fuzzing has ensured that pathological files like you
>> suggest are unlikely to be the source of exploits.
> See the crash above. I think fuzzing is very overhyped these days. It's
> a useful tool for sure, but one should not rely purely on fuzzing too
> much. It may not cover important parts of the input space because of the
> way it is set up. Or there could be exploitable codepaths that require
> too specific inputs for a random process to find, but a determined
> attacker could construct them from reading the code.
>
Of course fuzzing cannot find everything, but my experience has been that fuzzing finds things that I would _never_ find by just
looking at the code.

And I was certainly not suggesting that one should just write sloppy code knowing that all the bugs will be found with fuzzing.

-David

Quoting David Bryant (2020-04-06 06:11:17)
> On 4/2/20 11:32 PM, Anton Khirnov wrote:
> >
> >>> Also, I looked at the specification PDF you linked in the previous email
> >>> and I see nothing in it that explicitly forbids the channel count from
> >>> changing from one "superblock" to the next. It seems to me like you
> >>> could just concatenate wavpack files with different channel counts and
> >>> get a valid wavpack file.
> >>>
> >> You're right, I could not find that being explicitly forbidden. Nevertheless, it _is_ forbidden for any stream parameters to
> >> change (e.g. data format, sampling rate, channel count or mapping) and both libwavpack and the FFmpeg demuxer enforce this (see
> >> libavformat/wvdec.c starting at line 211), or at least gracefully fail without crashing.
> > Forbidden by whom? From what I can see:
> > - sample format: the wv demuxer does not set it at all, the decoder sets
> >   it in wavpack_decode_frame() from the block header; also, Michael sent
> >   a patch recently to deal with a wavpack decoder crash related to
> >   format changes
> > - sample rate: set by wv demuxer and then checked that it does not
> >   change; but the caller is not necessarily using the wv demuxer and the
> >   decoder will happily change it the same way it does with channel count
> >
> >> The reason it's never mentioned is probably just because it never occurred to me. None of the source file formats for WavPack
> >> (e.g., wav, caf, dsdiff) allow format changes and neither do any of the other lossless compressors I am aware of.
> > Correct, I cannot see any other lossless codecs that would allow this.
> > But then again all these codecs have global headers. So it's not that
> > they artificially forbid format changes, but the bistream format is not
> > able to represent them.
> 
> I disagree with that. It would be trivial to add a new STREAMINFO block to a FLAC file and change the format after that. This is
> not specifically prohibited in the FLAC specs, but it's sort of implied when it says that the initial STREAMINFO block has
> "information about the whole stream".

From a quick glance at the spec, metadata blocks are only allowed at the
beginning and are followed by audio frames. Once you see a single audio
frame, there should be no further metadata blocks.

That said, the frame header does contain stream parameters and those can
conceivably change and after looking at our decoder, it does support
those changes. It was added explicitly in
commit 90fcac0e95b7d266c148a86506f301a2072d9de3
Author: Justin Ruggles <justin.ruggles@gmail.com>
Date:   Sun Oct 21 17:02:28 2012 -0400

    flacdec: allow mid-stream channel layout change
    
    Although the libFLAC decoder cannot handle such a change, it is allowed by the
    spec and could potentially occur with live streams.

> 
> >
> >> and extensive fuzzing has ensured that pathological files like you
> >> suggest are unlikely to be the source of exploits.
> > See the crash above. I think fuzzing is very overhyped these days. It's
> > a useful tool for sure, but one should not rely purely on fuzzing too
> > much. It may not cover important parts of the input space because of the
> > way it is set up. Or there could be exploitable codepaths that require
> > too specific inputs for a random process to find, but a determined
> > attacker could construct them from reading the code.
> >
> Of course fuzzing cannot find everything, but my experience has been that fuzzing finds things that I would _never_ find by just
> looking at the code.
> 
> And I was certainly not suggesting that one should just write sloppy code knowing that all the bugs will be found with fuzzing.

Yeah, I certainly agree that fuzzing is a very useful tool. My point is
just that it shouldn't be your _only_ tool. There can be plenty of bugs
fuzzing won't find.