Backport of upstream commit 2d9ed64859c9887d0504cd71dbd5b2c15e14251a for 
CVE-2023-49528 to release/6.0.

The release/6.1 branch is still affected by CVE-2023-49528.
The upstream patch does not apply cleanly due to code differences,
so this patch manually adapts the fix while preserving the original logic.

A crafted input can trigger a heap-buffer-overflow in de_stereo()
in libavfilter/af_dialoguenhance.c. The issue is caused by copying
s->overlap samples without limiting the copy size to the actual
number of available input samples, which may lead to out-of-bounds
memory access.

This patch fixes the issue by using FFMIN(s->overlap, s->in->nb_samples)
for the memcpy() size and by adjusting the temporary audio buffer
allocation accordingly. Fix #22762.

Reproducer triggers a crash before the patch is applied, and works
as expected after applying the patch.

Associated issue: https://code.ffmpeg.org/FFmpeg/FFmpeg/issues/22762

CVE: CVE-2023-49528
Found-by: Zeng Yunxiang

Signed-off-by: lizhixuan.HUST <[email protected]>
---
 libavfilter/af_dialoguenhance.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/libavfilter/af_dialoguenhance.c b/libavfilter/af_dialoguenhance.c
index b613733faf..cc3f97b14d 100644
--- a/libavfilter/af_dialoguenhance.c
+++ b/libavfilter/af_dialoguenhance.c
@@ -95,12 +95,12 @@ static int config_input(AVFilterLink *inlink)
     if (!s->window)
         return AVERROR(ENOMEM);
 
-    s->in_frame       = ff_get_audio_buffer(inlink, s->fft_size * 4);
-    s->center_frame   = ff_get_audio_buffer(inlink, s->fft_size * 4);
-    s->out_dist_frame = ff_get_audio_buffer(inlink, s->fft_size * 4);
-    s->windowed_frame = ff_get_audio_buffer(inlink, s->fft_size * 4);
-    s->windowed_out   = ff_get_audio_buffer(inlink, s->fft_size * 4);
-    s->windowed_prev  = ff_get_audio_buffer(inlink, s->fft_size * 4);
+    s->in_frame       = ff_get_audio_buffer(inlink, (s->fft_size + 2) * 2);
+    s->center_frame   = ff_get_audio_buffer(inlink, (s->fft_size + 2) * 2);
+    s->out_dist_frame = ff_get_audio_buffer(inlink, (s->fft_size + 2) * 2);
+    s->windowed_frame = ff_get_audio_buffer(inlink, (s->fft_size + 2) * 2);
+    s->windowed_out   = ff_get_audio_buffer(inlink, (s->fft_size + 2) * 2);
+    s->windowed_prev  = ff_get_audio_buffer(inlink, (s->fft_size + 2) * 2);
     if (!s->in_frame || !s->windowed_out || !s->windowed_prev ||
         !s->out_dist_frame || !s->windowed_frame || !s->center_frame)
         return AVERROR(ENOMEM);
@@ -249,6 +249,7 @@ static int de_stereo(AVFilterContext *ctx, AVFrame *out)
     float *right_osamples  = (float *)out->extended_data[1];
     float *center_osamples = (float *)out->extended_data[2];
     const int offset = s->fft_size - s->overlap;
+    const int nb_samples = FFMIN(s->overlap, s->in->nb_samples);
     float vad;
 
     // shift in/out buffers
@@ -257,8 +258,8 @@ static int de_stereo(AVFilterContext *ctx, AVFrame *out)
     memmove(left_out, &left_out[s->overlap], offset * sizeof(float));
     memmove(right_out, &right_out[s->overlap], offset * sizeof(float));
 
-    memcpy(&left_in[offset], left_samples, s->overlap * sizeof(float));
-    memcpy(&right_in[offset], right_samples, s->overlap * sizeof(float));
+    memcpy(&left_in[offset], left_samples, nb_samples * sizeof(float));
+    memcpy(&right_in[offset], right_samples, nb_samples * sizeof(float));
     memset(&left_out[offset], 0, s->overlap * sizeof(float));
     memset(&right_out[offset], 0, s->overlap * sizeof(float));
 
-- 
2.34.1

_______________________________________________
ffmpeg-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to