Mark Thompson [Mon, 31 Aug 2020 21:00:57 +0000 (22:00 +0100)]
cbs_av1: Fill tile width/height values when uniform_tile_spacing_flag is set
They are not explicitly in the bitstream in this case, but it is helpful
to be able to use these values without always needing to check the flag
beforehand.
Since c6a63e11092c975b89d824f08682fe31948d3686, the parameter sets
modified as content of PPS units were references shared with the
CodedBitstreamH264Context, so modifying them alters the parsing process
of future access units which meant that frames often got discarded
because invalid values were parsed. This patch makes h264_redundant_pps
compatible with the reality of reference-counted parameter sets.
Fixes #7807.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: Mark Thompson <sw@jkqxz.net>
Mark Thompson [Mon, 27 Jul 2020 16:32:21 +0000 (17:32 +0100)]
cbs: Add support functions for handling unit content references
Use the unit type table to determine what we need to do to clone the
internals of the unit content when making copies for refcounting or
writeability. (This will still fail for units with complex content
if they do not have a defined clone function.)
Setup and naming from a patch by Andreas Rheinhardt
<andreas.rheinhardt@gmail.com>, but with the implementation changed
to use the unit type information if possible rather than requiring a
codec-specific function.
Mark Thompson [Mon, 27 Jul 2020 16:32:18 +0000 (17:32 +0100)]
cbs: Describe allocate/free methods in tabular form
Unit types are split into three categories, depending on how their
content is managed:
* POD structure - these require no special treatment.
* Structure containing references to refcounted buffers - these can use
a common free function when the offsets of all the internal references
are known.
* More complex structures - these still require ad-hoc treatment.
For each codec we can then maintain a table of descriptors for each set of
equivalent unit types, defining the mechanism needed to allocate/free that
unit content. This is not required to be used immediately - a new alloc
function supports this, but does not replace the old one which works without
referring to these tables.
James Almer [Mon, 31 Aug 2020 02:57:23 +0000 (23:57 -0300)]
avcodec/encode: unref the packet on AVCodec.receive_packet() failure
Fixes memleaks with some encoders that don't unref the packet before
returning.
This is consistent with the behavior of AVCodec.encode()
implementations in encode_simple_internal().
Found-by: mkver Reviewed-by: Anton Khirnov <anton@khirnov.net> Signed-off-by: James Almer <jamrial@gmail.com>
avcodec/magicyuv: Reuse array instead of using multiple arrays
The lengths of the VLC codes are implicitly contained in the VLC tables
itself; apart from that they are not used lateron. So it is unnecessary
to store them and the very same array can be reused to parse the Huffman
table for the next plane.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Up until now, there were three comparison functions depending upon
bitness. But they all are actually the same, namely a lexical ordering:
entry a > entry b iff a.len > b.len or a.len == b.len and a.sym < b.sym.
So they can be easily unified.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avcodec/magicyuv: Avoid copying values around pointlessly
When parsing Huffman tables, an array of HuffEntries (a struct
containing a code's bitlength, its bits and its symbol) is used as
intermediate tables in order to sort the entries (the order depends on
both the length of the entries as well as on their symbols). After sorting
them, the symbol and len components are copied into other arrays (the
HuffEntries' code has never been set or used, despite using quite a lot
of stack space) and the codes are generated. Afterwards, the VLC is
created.
Yet ff_init_vlc_sparse() can handle non-continuous arrays as input;
there is no need to copy the entries at all. This commit implements
this.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
When the MagicYUV decoder builds Huffman tables from an array of code
lengths, it proceeds as follows: First it copies the entries of the
array of lengths into an array of HuffEntries (a struct which contains
a length and a symbol field); it also sets the symbol field in
descending order from nb_elem - 1 to 0, where nb_elem is the common number
of elements of the length and HuffEntry arrays. Then the HuffEntry array
is sorted lexicographically: a > b iff a.len > b.len or a.len == b.len and
a.sym > b.sym. Afterwards the symbols of the so sorted array are
inverted again (i.e. each symbol sym is replaced by nb_elem - sym).
Yet inverting can easily be avoided altogether: Just modify the order so
that smaller symbols correspond to bigger HuffEntries. This leads to the
same permutation as the current code does and given that the two
inversions just cancel each other out, the result is the same.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avdevice/lavfi: Don't set pkt->size to the value it already has
av_new_packet() already sets the size. And if the packet is not
allocated by av_new_packet() (which seems to be impossible atm), both
pkt->size as well as size are 0, so setting it again is unnecessary in
this scenario, too.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Chris Miceli [Mon, 31 Aug 2020 04:16:39 +0000 (14:16 +1000)]
libavformat/dashenc: Reduce confusion in par error message
In ticket #8754 there is discourse surrounding the error
message which is printed upon a mismatched aspect ratio in
derived encodings. This should make it clearer to the user
as to the issues which they are experiencing.
Reviewed-by: "Jeyapal, Karthick" <kjeyapal@akamai.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Mark Thompson [Fri, 28 Aug 2020 22:15:41 +0000 (23:15 +0100)]
hwcontext_vaapi: Don't require a render node when deriving from DRM
The V4L2 driver does not actually have an associated DRM device at all, so
users work around the requirement by giving libva an unrelated display-only
device instead (which is fine, because it doesn't actually do anything with
that device). This was broken by bc9b6358fb7315c0173de322472641766f6289da
forcing a render node, because the display-only device did not have an
associated render node to use. Fix that by just passing through the
original non-render DRM fd if we can't find a render node.
Reported-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com> Tested-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com>
avfilter/src_movie: Avoid intermediate buffer for writing string
Reviewed-by: Paul B Mahol <onemda@gmail.com> Reviewed-by: Nicolas George <george@nsup.org> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avcodec/roqvideodec: Move transient GetByteContext to the stack
This avoids keeping potentially dangling pointers in the context,
beautifies the code (by replacing "&ri->gb" by gb for every access to
the GetByteContext) and also highlights the GetByteContext's short-lived
nature better.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The implementation of the tag tree did not
set the correct reset value for the encoder.
This lead to inefficent tag tree being encoded.
This patch fixes the implementation of the
ff_tag_tree_zero() function.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This patch allows setting a compression ratio and to
set multiple layers. The user has to input a compression
ratio for each layer.
The per layer compression ration can be set as follows:
-layer_rates "r1,r2,...rn"
for to create 'n' layers.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The implementation of tag tree encoding was incorrect.
However, this error was not visible as the current j2k
encoder encodes only 1 layer.
This patch fixes tag tree coding for JPEG2000 such tag
tree coding would work for multi layer encoding.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: out of array access Fixes: 24823/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_CFHD_fuzzer-4855119863349248 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/mpegvideo_enc: Replace some s->avctx by avctx
Fixes issue reported by: Xu Guangxin <guangxin.xu@intel.com>
Original report:
Steps to reproduce:
1. ./configure --enable-debug=3 --disable-libx264 && make install
2. ffmpeg -i input.mp4 -profile:v baseline output.mp4 -y
you will see a crash like this:
[mpeg4 @ 0x5555575854c0] [Eval @ 0x7fffffffbf80] Undefined constant or missing '(' in 'baseline'
[mpeg4 @ 0x5555575854c0] Unable to parse option value "baseline"
[mpeg4 @ 0x5555575854c0] Error setting option profile to value baseline.
Thread 1 "ffmpeg" received signal SIGSEGV, Segmentation fault.
root cause:
If the codec has FF_CODEC_CAP_INIT_CLEANUP flag, and avcodec_open2 got an error before avctx->codec->init,
the ff_mpv_encode_end will face a null s->avctx.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/wnv1: Use LE bitstream reader, avoid copying packet, fix memleak
The Winnov WNV1 format is designed for a little-endian bitstream reader;
yet our decoder reversed every byte bitwise (in a buffer only
allocated for this purpose) to use a big-endian bitstream reader. This
commit stops this.
Two things needed to be done to achieve this: The codes in the table used
to initialize a VLC reader needed to be reversed bitwise (when
initializing a VLC in LE mode, it is expected that the first bit to be
read is in the least significant bit; with BE codes the first bit to be
read is the most significant bit of the code) and the following
expression needed to be adapted:
ff_reverse[get_bits(&w->gb, 8 - w->shift)]
But this is easy: When only the bits read are reversed, they coincide
with what a little-endian bitstream reader reads that reads the
original, not-reversed data. But ff_reverse always reverses the full
eight bits and this also performs a shift by (8 - (8 - w->shift)) on top
of reversing the bits read. So the above line needs to be changed to
get_bits(&w->gb, 8 - w->shift) << w->shift
and this also shows why the variable shift is named the way it is.
Finally, this also fixes a hypothetical memleak: For gigantic packets,
initializing a GetBitContext can fail and in this case, the buffer
containing the reversed data would leak.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
TrueMotion 2.0 uses Huffmann trees. To parse them, the decoder allocates
arrays for the codes, their lengths and their value; afterwards a VLC
table is initialized using these values. If everything up to this point
succeeds, a new buffer of the same size as the already allocated arrays
for the values is allocated and upon success the values are copied into
the new array; all the old arrays are then freed. Yet if allocating the
new array fails, the old arrays get freed, but the VLC table doesn't.
This leak is fixed by not allocating a new array at all; instead the old
array is simply reused, ensuring that nothing can fail after the
creation of the VLC table.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Dale Curtis [Thu, 13 Aug 2020 22:02:02 +0000 (15:02 -0700)]
avformat/mov: See if mfra makes up the difference for an incomplete sidx.
A few popular sites have started generating MP4 files which have a
sidx plus an mfra. The sidx accounts for all size except the mfra,
so the old code did not mark the fragment index as complete.
Instead we can just check if there's an mfra and if its size makes
up the difference we can mark the index as complete.
avfilter/af_headphone: Fix leak of channel layouts list on error
In case the multichannel HRIR mode was enabled, an error could happen
between allocating a channel layouts list and attaching it to its target
destination. If an error happened, the list would leak. This is fixed by
attaching the list to its target directly after its allocation.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avfilter/af_headphone: Fix segfault upon allocation failure
The headphone filter uses a variable number of inpads and allocates them
in its init function; if all goes well, the number of inpads coincides
with a number stored in the filter's private context. Yet if allocating a
subsequent inpad fails, the uninit function nevertheless uses the number
stored in the private context to determine the number of inpads to free
and not the AVFilterContext's nb_inputs. This will lead to an access
beyond the end of the allocated AVFilterContext.input_pads array and
an invalid free.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avfilter/vf_signature: Avoid cast from function pointer to void*
The signature filter uses qsort, but its compare function doesn't have
the signature required of such a function; therefore it casts the
function pointer to void. Yet this is wrong:
C90 only guarantees that one can convert a pointer to any incomplete
type or object type to void* and back with the result comparing equal
to the original which makes pointers to void generic pointers to
incomplete or object type. Yet C90 lacks a generic function pointer
type.
C99 additionally guarantees that a pointer to a function of one type may
be converted to a pointer to a function of another type with the result
and the original comparing equal when converting back.
This makes any function pointer type a generic function pointer type.
Yet even this does not make pointers to void generic function pointers.
Both GCC and Clang emit warnings for this when in pedantic mode.
This commit fixes this by modifying the compare function to comply with
the expected signature.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avfilter/vf_signature: Fix leak of string upon error
If an error happens between allocating a string intended to be used as
an inpad's name and attaching it to its input pad, the string leaks.
Fix this by inserting the inpad directly after allocating its string.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avfilter/af_aiir: Fix segfault and leak upon allocation failure
The aiir filter adds output pads in its init function. Each of these
output pads had a name which was allocated and to be freed in the uninit
function. Given that the aiir filter has between one and two outputs,
one output pad's name was freed unconditionally and a second was freed
conditionally.
Yet if adding output pads fails, there are no output pads at all and
trying to free a nonexistent pad's name will lead to a segfault.
Furthermore, if the name could be successfully allocated, yet adding the
new pad fails, the name would leak.
This commit fixes this by not allocating the pads' names at all any
more: They are constant anyway. This allows to remove the code to free
them and hence fixes the aforementioned bugs.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avfilter/vf_premultiply: Fix leak of names of inpads
These names leak because freeing them in the uninit function has been
forgotten. Instead of adding the freeing code, this commit stops
allocating these names. They are constants anyway.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
avfilter/af_anequalizer: Fix memleak when inserting pad fails
It has been forgotten to free the name of the second outpad if attaching
the first one to the AVFilterContext fails. Fixing this is easy: Only
prepare the second outpad after (and if) the first outpad has been
successfully attached to the AVFilterContext.
Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
projects / ffmpeg / log