| Message ID | 20240331114019.3801-1-michael@niedermayer.cc |
|---|---|
| State | Accepted |
| Commit | d88c284c18bf6cd3dd24a7c86b5e496dd3037405 |
| Headers | show |
| Series | [FFmpeg-devel] avformat/movenc: Check that cts fits in 32bit | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
On 3/31/2024 8:40 AM, Michael Niedermayer wrote: > Fixes: Assertion av_rescale_rnd(start_dts, mov->movie_timescale, track->timescale, AV_ROUND_DOWN) <= 0 failed at libavformat/movenc.c:3694 > Fixes: poc2 > > Found-by: Wang Dawei and Zhou Geng, from Zhongguancun Laboratory > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> > --- > libavformat/movenc.c | 6 ++++++ > 1 file changed, 6 insertions(+) > > diff --git a/libavformat/movenc.c b/libavformat/movenc.c > index ae94d8d5959..5617a2620c5 100644 > --- a/libavformat/movenc.c > +++ b/libavformat/movenc.c > @@ -6194,6 +6194,12 @@ int ff_mov_write_packet(AVFormatContext *s, AVPacket *pkt) > if (ret < 0) > return ret; > > + if (pkt->pts != AV_NOPTS_VALUE && > + (uint64_t)pkt->dts - pkt->pts != (int32_t)((uint64_t)pkt->dts - pkt->pts)) { > + av_log(s, AV_LOG_WARNING, "pts/dts pair unsupported\n"); > + return AVERROR_PATCHWELCOME; > + } Any such check should happen in check_pkt(), called directly above. And afaict there's no reason to not support 64bit cts. Even in mov_write_edts_tag() we check for it and write a version 1 of the box that supports 64bit values. Maybe the problem is that MOVIentry.cts is an int, when it should be an int64_t like start_cts? Can you test the following? > diff --git a/libavformat/movenc.h b/libavformat/movenc.h > index 08d580594d..e736c92750 100644 > --- a/libavformat/movenc.h > +++ b/libavformat/movenc.h > @@ -49,11 +49,11 @@ typedef struct MOVIentry { > uint64_t pos; > int64_t dts; > int64_t pts; > + int64_t cts; > unsigned int size; > unsigned int samples_in_chunk; > unsigned int chunkNum; ///< Chunk number if the current entry is a chunk start otherwise 0 > unsigned int entries; > - int cts; > #define MOV_SYNC_SAMPLE 0x0001 > #define MOV_PARTIAL_SYNC_SAMPLE 0x0002 > #define MOV_DISPOSABLE_SAMPLE 0x0004
On Sun, Mar 31, 2024 at 01:30:26PM -0300, James Almer wrote: > On 3/31/2024 8:40 AM, Michael Niedermayer wrote: > > Fixes: Assertion av_rescale_rnd(start_dts, mov->movie_timescale, track->timescale, AV_ROUND_DOWN) <= 0 failed at libavformat/movenc.c:3694 > > Fixes: poc2 > > > > Found-by: Wang Dawei and Zhou Geng, from Zhongguancun Laboratory > > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> > > --- > > libavformat/movenc.c | 6 ++++++ > > 1 file changed, 6 insertions(+) > > > > diff --git a/libavformat/movenc.c b/libavformat/movenc.c > > index ae94d8d5959..5617a2620c5 100644 > > --- a/libavformat/movenc.c > > +++ b/libavformat/movenc.c > > @@ -6194,6 +6194,12 @@ int ff_mov_write_packet(AVFormatContext *s, AVPacket *pkt) > > if (ret < 0) > > return ret; > > + if (pkt->pts != AV_NOPTS_VALUE && > > + (uint64_t)pkt->dts - pkt->pts != (int32_t)((uint64_t)pkt->dts - pkt->pts)) { > > + av_log(s, AV_LOG_WARNING, "pts/dts pair unsupported\n"); > > + return AVERROR_PATCHWELCOME; > > + } > > Any such check should happen in check_pkt(), called directly above. And > afaict there's no reason to not support 64bit cts. Even in > mov_write_edts_tag() we check for it and write a version 1 of the box that > supports 64bit values. > > Maybe the problem is that MOVIentry.cts is an int, when it should be an > int64_t like start_cts? Can you test the following? changing cts to 64bit does avoid the assert with the test sample but If you chaneg cts to 64bit consider cts is assigned to MOVCtts duration (32bit) in mov_write_ctts_tag() and also compared to it. its also written with avio_wb32() later theres also avio_wb32(pb, track->cluster[i].cts); in mov_write_trun_tag() so the suggestion would avoid the assert but the code is not correct, now one can maybe add support for this by switching to 64bit variants. But this needs to be backported too. and "64bit cts support" seems not ideal both because of complexity and beacause its not a bugfix I would prefer to keep a simple (easy backportable) bugfix for the releases. thx [...]
On 4/1/2024 12:38 PM, Michael Niedermayer wrote: > On Sun, Mar 31, 2024 at 01:30:26PM -0300, James Almer wrote: >> On 3/31/2024 8:40 AM, Michael Niedermayer wrote: >>> Fixes: Assertion av_rescale_rnd(start_dts, mov->movie_timescale, track->timescale, AV_ROUND_DOWN) <= 0 failed at libavformat/movenc.c:3694 >>> Fixes: poc2 >>> >>> Found-by: Wang Dawei and Zhou Geng, from Zhongguancun Laboratory >>> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> >>> --- >>> libavformat/movenc.c | 6 ++++++ >>> 1 file changed, 6 insertions(+) >>> >>> diff --git a/libavformat/movenc.c b/libavformat/movenc.c >>> index ae94d8d5959..5617a2620c5 100644 >>> --- a/libavformat/movenc.c >>> +++ b/libavformat/movenc.c >>> @@ -6194,6 +6194,12 @@ int ff_mov_write_packet(AVFormatContext *s, AVPacket *pkt) >>> if (ret < 0) >>> return ret; >>> + if (pkt->pts != AV_NOPTS_VALUE && >>> + (uint64_t)pkt->dts - pkt->pts != (int32_t)((uint64_t)pkt->dts - pkt->pts)) { >>> + av_log(s, AV_LOG_WARNING, "pts/dts pair unsupported\n"); >>> + return AVERROR_PATCHWELCOME; >>> + } >> >> Any such check should happen in check_pkt(), called directly above. And >> afaict there's no reason to not support 64bit cts. Even in >> mov_write_edts_tag() we check for it and write a version 1 of the box that >> supports 64bit values. >> >> Maybe the problem is that MOVIentry.cts is an int, when it should be an >> int64_t like start_cts? Can you test the following? > > changing cts to 64bit does avoid the assert with the test sample but > > If you chaneg cts to 64bit consider cts is assigned to MOVCtts duration (32bit) > in mov_write_ctts_tag() and also compared to it. > its also written with avio_wb32() later > > theres also > avio_wb32(pb, track->cluster[i].cts); > in mov_write_trun_tag() You're right, there's no 64bit version for these, i guess they will not define one either in a future revision of the spec. So your patch should be fine. > > so the suggestion would avoid the assert but the code is not correct, now one can > maybe add support for this by switching to 64bit variants. But this needs to be > backported too. > and "64bit cts support" seems not ideal both because of complexity and beacause > its not a bugfix > > I would prefer to keep a simple (easy backportable) bugfix for the releases. > > thx > > [...] > > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
On Mon, Apr 01, 2024 at 12:45:36PM -0300, James Almer wrote: > On 4/1/2024 12:38 PM, Michael Niedermayer wrote: > > On Sun, Mar 31, 2024 at 01:30:26PM -0300, James Almer wrote: > > > On 3/31/2024 8:40 AM, Michael Niedermayer wrote: > > > > Fixes: Assertion av_rescale_rnd(start_dts, mov->movie_timescale, track->timescale, AV_ROUND_DOWN) <= 0 failed at libavformat/movenc.c:3694 > > > > Fixes: poc2 > > > > > > > > Found-by: Wang Dawei and Zhou Geng, from Zhongguancun Laboratory > > > > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> > > > > --- > > > > libavformat/movenc.c | 6 ++++++ > > > > 1 file changed, 6 insertions(+) > > > > > > > > diff --git a/libavformat/movenc.c b/libavformat/movenc.c > > > > index ae94d8d5959..5617a2620c5 100644 > > > > --- a/libavformat/movenc.c > > > > +++ b/libavformat/movenc.c > > > > @@ -6194,6 +6194,12 @@ int ff_mov_write_packet(AVFormatContext *s, AVPacket *pkt) > > > > if (ret < 0) > > > > return ret; > > > > + if (pkt->pts != AV_NOPTS_VALUE && > > > > + (uint64_t)pkt->dts - pkt->pts != (int32_t)((uint64_t)pkt->dts - pkt->pts)) { > > > > + av_log(s, AV_LOG_WARNING, "pts/dts pair unsupported\n"); > > > > + return AVERROR_PATCHWELCOME; > > > > + } > > > > > > Any such check should happen in check_pkt(), called directly above. And > > > afaict there's no reason to not support 64bit cts. Even in > > > mov_write_edts_tag() we check for it and write a version 1 of the box that > > > supports 64bit values. > > > > > > Maybe the problem is that MOVIentry.cts is an int, when it should be an > > > int64_t like start_cts? Can you test the following? > > > > changing cts to 64bit does avoid the assert with the test sample but > > > > If you chaneg cts to 64bit consider cts is assigned to MOVCtts duration (32bit) > > in mov_write_ctts_tag() and also compared to it. > > its also written with avio_wb32() later > > > > theres also > > avio_wb32(pb, track->cluster[i].cts); > > in mov_write_trun_tag() > > You're right, there's no 64bit version for these, i guess they will not > define one either in a future revision of the spec. So your patch should be > fine. ok, will apply thx [...]
diff --git a/libavformat/movenc.c b/libavformat/movenc.c index ae94d8d5959..5617a2620c5 100644 --- a/libavformat/movenc.c +++ b/libavformat/movenc.c @@ -6194,6 +6194,12 @@ int ff_mov_write_packet(AVFormatContext *s, AVPacket *pkt) if (ret < 0) return ret; + if (pkt->pts != AV_NOPTS_VALUE && + (uint64_t)pkt->dts - pkt->pts != (int32_t)((uint64_t)pkt->dts - pkt->pts)) { + av_log(s, AV_LOG_WARNING, "pts/dts pair unsupported\n"); + return AVERROR_PATCHWELCOME; + } + if (mov->flags & FF_MOV_FLAG_FRAGMENT || mov->mode == MODE_AVIF) { int ret; if (mov->moov_written || mov->flags & FF_MOV_FLAG_EMPTY_MOOV) {
Fixes: Assertion av_rescale_rnd(start_dts, mov->movie_timescale, track->timescale, AV_ROUND_DOWN) <= 0 failed at libavformat/movenc.c:3694 Fixes: poc2 Found-by: Wang Dawei and Zhou Geng, from Zhongguancun Laboratory Signed-off-by: Michael Niedermayer <michael@niedermayer.cc> --- libavformat/movenc.c | 6 ++++++ 1 file changed, 6 insertions(+)