Fixes: index 33280 out of bounds for type 'float [32768]'
Fixes: 18718/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_XMA2_fuzzer-5635373899710464

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegSigned-off-by: Michael Niedermayer <michael@niedermayer.cc>

Fixes: signed integer overflow: 1778647621 + 574372924 cannot be represented in type 'int'
Fixes: 18692/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_TRUEMOTION2_fuzzer-6248679635943424

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegSigned-off-by: Michael Niedermayer <michael@niedermayer.cc>

Fixes: NULL pointer dereference
Fixes: 18689/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_XMA1_fuzzer-5715114640015360

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegSigned-off-by: Michael Niedermayer <michael@niedermayer.cc>

Fixes: signed integer overflow: 2147483188 + 2048 cannot be represented in type 'int'
Fixes: 18741/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_TRUESPEECH_fuzzer-5748950460268544

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegSigned-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

ff_id3v2_parse_priv_dict() uses av_dict_set() with the flags
AV_DICT_DONT_STRDUP_KEY and AV_DICT_DONT_STRDUP_VAL. In this case both
key and value are freed on error (and owned by the destination
dictionary on success), so that freeing them again on error is a
double-free and therefore forbidden. But it nevertheless happened.

Fixes CID 1452489 and 1452421.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

scale.c is too generic; scale_eval is more representative

contained in Vorbis comments in the CodecPrivate of flac tracks.
Moreover, it also tests header removal compression.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

This should improve coverage
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Adjustment of evaluated values shifted to ff_adjust_scale_dimensions
Shifted code for force_original_aspect_ratio and force_divisble_by from
vf_scale so it is now available for scale_cuda, scale_npp and
scale_vaapi as well.

This test contains a track with zlib compressed CodecPrivate in addition
to compressed frames; the former was unchecked before.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

avcodec/atrac9dec: Check q_unit_cnt more completely before using it to access at9_tab_band_ext_group

Fixes: index 8 out of bounds for type 'const uint8_t [8][3]'
Fixes: 19127/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_ATRAC9_fuzzer-5709394985091072

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegReviewed-by: Lynne <dev@lynne.ee>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: hwrenx <hwrenx@126.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: James Almer <jamrial@gmail.com>

Signed-off-by: James Almer <jamrial@gmail.com>

ProRes in Matroska is supposed to not contain the first atom header
(containing a size field and the tag "icpf") and therefore the Matroska
demuxer has to recreate it; this involves an allocation and copy, of
course. Whether the old buffer (containing the data without the atom
header) needs to be freed or not depends upon whether it is what was
directly read (in which case it is owned by an AVBuffer) or whether it
has been allocated when reversing the track's content compression (e.g.
zlib compression) that Matroska supports.

So there are three pointers involved: The one pointing to the directly
read data (owned by the AVBuffer), the one pointing to the currently
valid data (which coincides with the former if no content compression
needed to be reverted) and the one pointing to the new data with the
first atom header. The check for whether to free the second of these is
simply whether the first two are different.

This works mostly, but there is a complication: Some muxers don't strip
the first atom header away and in this case, it is also not reinserted
and no new buffer is allocated; instead, the second and the third
pointers agree. In this case, one must never free the second buffer.
Yet it is currently done if the track is e.g. zlib compressed.
This commit fixes this.

This is a regression since b8e75a2a.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

This happens if ffurl_open_whitelist fails and stream is unset.
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

Fix the memory leak in error handle path.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

In the worst case the startcode prefix has 4 bytes.

This fixes a trigerred assertion:
Assertion dp <= max_size failed at libavcodec/cbs_h2645.c:1451

Found-by:libFuzzer
Reviewed-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Andriy Gelman <andriy.gelman@gmail.com>

Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Fixes: Infinite loop
Fixes: 19183/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MVHA_fuzzer-5666216765292544

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegReviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: leozhang <leozhang@qiyi.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: Ting Fu <ting.fu@intel.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

This avoids a memcpy improving performance if SHM is not used.
Signed-off-by: Marton Balint <cus@passwd.hu>

Width and height expressions can refer to each other. Width is
evaluated twice to allow for reference to output height. So we
should not error out upon failure of first evaluation of width.

No functional changes. ref/unref vs add/sub is symmetrical.
Signed-off-by: James Almer <jamrial@gmail.com>

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

Fixes: fate-fitsdec-bitpix-64

Possibly Fixes: -nan is outside the range of representable values of type 'unsigned short'
Possibly Fixes: 17769/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_FITS_fuzzer-5678314672357376

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegSigned-off-by: Michael Niedermayer <michael@niedermayer.cc>

If an error happens in vobsub_read_header() after allocating the
AVFormatContext intended to read the sub-file, both the AVFormatContext
as well as the data in the subtitles queues leaks. This has been fixed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

vobsub_read_header() uses an AVBPrint to write a string and up until
now, it collected the string stored in the AVBPrint via
av_bprint_finalize(), which might involve an allocation and copy of the
string. But this is unnecessary, as the lifetime of the returned string
does not exceed the lifetime of the AVBPrint. So use the string in the
AVBPrint directly.

This also makes it possible to easily fix a memleak: In certain error
situations, the string stored in the AVBPrint would not be freed (if it
was dynamically allocated). This has been fixed, too.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

When the VobSub demuxer was added, the fields it required were simply
added to the MpegDemuxContext (if the VobSub demuxer was selected at
all). The mpeg demuxer of course doesn't use these fields even if they
are there; and the VobSub demuxer doesn't use the old ones: It opens an
mpeg subdemuxer of its own and uses this where a mpeg demuxer is
required. Hence the two contexts can be split, saving memory.

Furthermore several headers can now be moved to the section that is
guarded by #if CONFIG_VOBSUB_DEMUXER (this even includes avassert.h
which was unguarded and has been added in 9cde9f70 despite not being
used in that patch).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

When parsing EBML lacing, for every number read, a new AVIOContext has
been initialized (via ffio_init_context()) just for this number. This
has been changed: The context is kept now.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

When parsing the sizes of the frames in a lace fails, sometimes no
error message was raised (e.g. when using xiph or fixed-size lacing).
Only EBML lacing generated error messages (which were wrongly declared
as AV_LOG_INFO), but even here not all errors resulted in an error
message. So add a generic error message to catch them all.

Moreover, if parsing one of the EBML numbers fails, ebml_read_num already
emits its own error messages, so that all that is needed is a generic error
message to indicate that this happened during parsing the sizes of the
frames in a block; in other words, the error messages specific to
parsing EBML lace numbers can be and have been removed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

870e7552 introduced validating the lace sizes when they are parsed and
removed the old check; yet when merging this libav commit in 6902c3ac,
the old check for whether the frame extends beyond the frame has been kept.
It is unnecessary and has been removed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

Up until now, when an error happened in one of the inner loops in
matroska_parse_laces, a variable designated for the return value has
been set to an error value and break has been used to exit the
current loop/case. This was done so that the end of matroska_parse_laces
is reached, because said function allocated memory which is later used
and freed in the calling function and passed at the end of
matroska_parse_laces.

But given that there is no allocation any more, one can now return
immediately. And this commit does this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

The maximal number of frames in a lace can be 256; hence one has a not
excessive upper bound on the size of an array that can hold the sizes of
all the frames in a lace. Yet up until now, said array has been
dynamically allocated. This has been changed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

It avoids the overhead of function calls.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>