| Message ID | 20220322191100.3071-1-jamrial@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series | [FFmpeg-devel,1/2] avcodec/av1: only set the private context pix_fmt field if get_pixel_format() succeeds | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/commit_msg_loongarch64 | warning | Please wrap lines in the body of the commit message between 60 and 72 characters. |
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/commit_msg_x86 | warning | Please wrap lines in the body of the commit message between 60 and 72 characters. |
| andriy/configure_x86 | warning | Failed to apply patch |
On Tue, Mar 22, 2022 at 04:10:59PM -0300, James Almer wrote: > Otherwise get_pixel_format() will not be called when parsing a subsequent Sequence > Header in non hwaccel enabled scenarios, allowing frame parsing when it shouldn't. > > This prevents the scenario seqhdr -> frame_hdr/redundant_frame_hdr -> seqhdr -> > redundant_frame_hdr from having the latter redundant frame header parsed as if it > was a frame header by the decoder because the former was discarded. > Since CBS did not discard it, the latter redundant frame header is output with a > zeroed AV1RawFrameHeader struct, which can have undesired results, like division > by zero with fields normally guaranteed to be anything else. > > Signed-off-by: James Almer <jamrial@gmail.com> > --- > libavcodec/av1dec.c | 7 +++---- > 1 file changed, 3 insertions(+), 4 deletions(-) This patchset fixes the 2 issues too thx [...]
diff --git a/libavcodec/av1dec.c b/libavcodec/av1dec.c index 613efc5d11..727d829e5b 100644 --- a/libavcodec/av1dec.c +++ b/libavcodec/av1dec.c @@ -506,9 +506,8 @@ static int get_pixel_format(AVCodecContext *avctx) if (pix_fmt == AV_PIX_FMT_NONE) return -1; - s->pix_fmt = pix_fmt; - switch (s->pix_fmt) { + switch (pix_fmt) { case AV_PIX_FMT_YUV420P: #if CONFIG_AV1_DXVA2_HWACCEL *fmtp++ = AV_PIX_FMT_DXVA2_VLD; @@ -551,7 +550,7 @@ static int get_pixel_format(AVCodecContext *avctx) break; } - *fmtp++ = s->pix_fmt; + *fmtp++ = pix_fmt; *fmtp = AV_PIX_FMT_NONE; ret = ff_thread_get_format(avctx, pix_fmts); @@ -569,7 +568,7 @@ static int get_pixel_format(AVCodecContext *avctx) return AVERROR(ENOSYS); } - avctx->pix_fmt = ret; + avctx->pix_fmt = s->pix_fmt = ret; return 0; }
Otherwise get_pixel_format() will not be called when parsing a subsequent Sequence Header in non hwaccel enabled scenarios, allowing frame parsing when it shouldn't. This prevents the scenario seqhdr -> frame_hdr/redundant_frame_hdr -> seqhdr -> redundant_frame_hdr from having the latter redundant frame header parsed as if it was a frame header by the decoder because the former was discarded. Since CBS did not discard it, the latter redundant frame header is output with a zeroed AV1RawFrameHeader struct, which can have undesired results, like division by zero with fields normally guaranteed to be anything else. Signed-off-by: James Almer <jamrial@gmail.com> --- libavcodec/av1dec.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)