docs: add AGENTS.md project guide

Capture sources of truth, build/test instructions, runtime architecture,
and unsafe/FFI notes for AI agents working in this repo.

AGENTS.md

@@ -0,0 +1,40 @@
+# AGENTS.md
+
+## Sources of truth
+
+- Trust current Rust source, `Cargo.toml`, `shell.nix`, and tests over `docs/superpowers/*` or `analysis.md`; those docs contain historical/aspirational modules that are not in the tree.
+- Crate edition is 2021, not the default Rust 2024. The package exposes a library, three binaries (`wl-webrtc`, `vaapi_import_bench`, `sw_encode_bench`), and two examples (`list_globals`, `test_portal`).
+
+## Setup and build
+
+- Native prerequisites are FFmpeg 6+ dev libs with VAAPI, Wayland protocols/libs, libdrm, PipeWire, and libclang. `shell.nix` provides FFmpeg/Wayland/libdrm/Mesa/libva/clang and `LIBCLANG_PATH`, but does not currently list PipeWire.
+- Normal build: `cargo build`. Release binary required by README and integration tests: `cargo build --release`.
+- `Cargo.toml` only warns on `clippy::undocumented_unsafe_blocks`; do not assume a broader clippy policy exists unless you add one.
+
+## Testing and verification
+
+- Unit/focused tests can run with filters, e.g. `cargo test transform`, `cargo test fps_limit`, `cargo test backend_detect`.
+- Full `cargo test` includes `tests/integration_test.rs`, which shells out to `target/release/wl-webrtc`; run `cargo build --release` first or those tests fail before reaching code behavior.
+- Hardware/live tests are marked ignored; run them only on a Wayland session with VAAPI-capable GPU and writable output: `cargo test -- --ignored`.
+- CLI smoke test surface: `target/release/wl-webrtc --help` and invalid-argument rejection. Real capture needs Wayland plus either wlr-screencopy or XDG Portal/PipeWire.
+- The README CLI table is stale for `--backend` and `--no-persist`; `src/args.rs` is authoritative for flags.
+
+## Runtime architecture
+
+- `src/main.rs` is the real entrypoint: parse `Args`, initialize tracing from `RUST_LOG` or `-v`, reject non-H.264, require `--output` or `--port`, detect backend, then run one of two loops.
+- Backend detection in `src/backend_detect.rs`: explicit `--backend portal|screencopy` wins; otherwise wlr-screencopy is preferred when the Wayland global `zwlr_screencopy_manager_v1` exists, else Portal/PipeWire is used if D-Bus ScreenCast is available.
+- Do not use `ashpd` for backend availability checks; `backend_detect.rs` intentionally uses raw `zbus` because `ashpd` caches a `zbus::Connection` in a global and can hang after its owning Tokio runtime is dropped.
+- `src/state.rs` drives the wlroots path using a mio Wayland fd loop and `State<CapWlrScreencopy>`; `src/state_portal.rs` drives the Portal/PipeWire path through `CapPortal` frame channels.
+- `src/webrtc.rs` is a small embedded HTTP/WebRTC signaling server using `str0m`; `--port 0` means file-output mode, `--port > 0` enables WebRTC mode.
+
+## Unsafe and FFI work
+
+- FFmpeg/VAAPI/PipeWire code relies on raw FFI and many `unsafe` blocks. Preserve nearby `// SAFETY:` explanations and add one for any new unsafe block.
+- `src/avhw.rs` owns FFmpeg `AVBufferRef`/frame contexts and has explicit `unsafe impl Send`; avoid moving those wrappers across threads without rechecking the documented exclusivity assumptions.
+- `CapPortal` stores the portal restore token under the user cache directory (`wl-webrtc/portal-restore-token`); use `--no-persist` when manually testing fresh authorization behavior.
+
+## Useful manual commands
+
+- List Wayland globals: `cargo run --example list_globals`.
+- Portal permission smoke test: `cargo run --example test_portal`.
+- Portal/VAAPI benchmarks require a screen-share dialog and hardware: `cargo run --bin vaapi_import_bench -- --output /tmp/vaapi_bench.mp4` and `cargo run --bin sw_encode_bench -- --output /tmp/bench_test.mp4`.

src/avhw.rs

@@ -3,10 +3,10 @@ use std::mem;
 use std::os::fd::{AsRawFd, RawFd};
 use std::os::raw::c_void;
 use std::path::Path;
+use std::ptr;
 use std::slice;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
-use std::ptr;
 
 use anyhow::{bail, Result};
 use ffmpeg_next as ff;
@@ -16,6 +16,23 @@ use ffmpeg_next::packet::Mut as _;
 use crate::cap_portal::PwDmaBufFrame;
 use crate::transform::{transpose_if_transform_transposed, Transform};
 
+// ---------------------------------------------------------------------------
+// Bitrate feedback command (WebRTC BWE → SW encoder)
+// ---------------------------------------------------------------------------
+
+/// Commands sent from the WebRTC thread to the SW encoder when the
+/// bandwidth estimate changes significantly.
+pub enum BitrateCommand {
+    UpdateBitrate { target_bps: u64 },
+    UpdateResolution { width: u32, height: u32 },
+}
+
+#[derive(Clone, Copy, Debug)]
+pub struct ResolutionChange {
+    pub width: u32,
+    pub height: u32,
+}
+
 // ---------------------------------------------------------------------------
 // AvHwDevCtx
 // ---------------------------------------------------------------------------
@@ -334,7 +351,9 @@ impl EncState {
             transform,
         )?;
 
-        let mut sink_ctx = video_filter.get("out").ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
+        let mut sink_ctx = video_filter
+            .get("out")
+            .ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
         // SAFETY: sink_ctx is a live buffersink; the returned hw_frames_ctx is
         // borrowed, so av_buffer_ref creates an owned reference.
         let sink_hw_frames = unsafe {
@@ -374,19 +393,19 @@ impl EncState {
         enc.set_width(enc_width);
         enc.set_height(enc_height);
         enc.set_format(ff::format::Pixel::VAAPI);
-    enc.set_bit_rate(bitrate as usize);
-    enc.set_gop(gop_size);
-    enc.set_time_base(ff::Rational::new(1, fps as i32));
-    enc.set_max_b_frames(0);
+        enc.set_bit_rate(bitrate as usize);
+        enc.set_gop(gop_size);
+        enc.set_time_base(ff::Rational::new(1, fps as i32));
+        enc.set_max_b_frames(0);
 
-    // VBV rate limiting: caps IDR burst size for WebRTC.  Without this a 4K
-    // scene change can produce a 256KB keyframe that overflows the UDP send
-    // buffer.  bufsize=bitrate/4 ≈ 250ms of video at the target bitrate.
-    unsafe {
-        let ctx_ptr = enc.as_mut_ptr();
-        (*ctx_ptr).rc_max_rate = bitrate as i64;
-        (*ctx_ptr).rc_buffer_size = (bitrate / 4) as i32;
-    }
+        // VBV rate limiting: caps IDR burst size for WebRTC.  Without this a 4K
+        // scene change can produce a 256KB keyframe that overflows the UDP send
+        // buffer.  bufsize=bitrate/4 ≈ 250ms of video at the target bitrate.
+        unsafe {
+            let ctx_ptr = enc.as_mut_ptr();
+            (*ctx_ptr).rc_max_rate = bitrate as i64;
+            (*ctx_ptr).rc_buffer_size = (bitrate / 4) as i32;
+        }
 
         // SAFETY: AV_CODEC_FLAG_GLOBAL_HEADER must be set BEFORE opening the encoder.
         // It triggers SPS/PPS extradata generation needed by the muxer for
@@ -461,7 +480,10 @@ impl EncState {
             ffi::avcodec_parameters_from_context((*stream_ptr).codecpar, enc_video.as_ptr())
         };
         if ret < 0 {
-            bail!("Failed to copy encoder parameters to stream: {}", ff_err(ret));
+            bail!(
+                "Failed to copy encoder parameters to stream: {}",
+                ff_err(ret)
+            );
         }
 
         // SAFETY: Copy encoder time_base to stream.
@@ -510,9 +532,15 @@ impl EncState {
     }
 
     pub fn encode_frame(&mut self, hw_frame: &ff::frame::Video) -> Result<()> {
-        let mut filter_src_ctx = self.video_filter.get("in").ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
+        let mut filter_src_ctx = self
+            .video_filter
+            .get("in")
+            .ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
         let mut filter_src = filter_src_ctx.source();
-        let mut filter_sink_ctx = self.video_filter.get("out").ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
+        let mut filter_sink_ctx = self
+            .video_filter
+            .get("out")
+            .ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
         let mut filter_sink = filter_sink_ctx.sink();
 
         // SAFETY: hw_frame is a valid VAAPI hardware frame from capture.
@@ -552,14 +580,20 @@ impl EncState {
 
     pub fn flush(&mut self) -> Result<()> {
         // Flush filter graph
-        let mut filter_src_ctx = self.video_filter.get("in").ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
+        let mut filter_src_ctx = self
+            .video_filter
+            .get("in")
+            .ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
         let mut filter_src = filter_src_ctx.source();
         if let Err(e) = filter_src.flush() {
             tracing::debug!("filter source flush error: {e}");
         }
 
         // Drain filter
-        let mut filter_sink_ctx = self.video_filter.get("out").ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
+        let mut filter_sink_ctx = self
+            .video_filter
+            .get("out")
+            .ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
         let mut filter_sink = filter_sink_ctx.sink();
         loop {
             let mut filtered = ff::frame::Video::empty();
@@ -667,8 +701,13 @@ pub struct SwEncImport {
     hw_dev: AvHwDevCtx,
     frames_rgb: AvHwFrameCtx,
     filter_graph: ff::filter::Graph,
+    width: u32,
+    height: u32,
     enc_width: u32,
     enc_height: u32,
+    fps: u32,
+    resolution_rx: Option<crossbeam_channel::Receiver<BitrateCommand>>,
+    encoder_resolution_tx: Option<crossbeam_channel::Sender<ResolutionChange>>,
 }
 
 impl SwEncImport {
@@ -698,20 +737,50 @@ impl SwEncImport {
             hw_dev,
             frames_rgb,
             filter_graph,
+            width,
+            height,
             enc_width,
             enc_height,
+            fps,
+            resolution_rx: None,
+            encoder_resolution_tx: None,
         })
     }
 
+    #[allow(clippy::too_many_arguments)]
+    pub fn new_with_resolution_control(
+        drm_device: &Path,
+        width: u32,
+        height: u32,
+        enc_width: u32,
+        enc_height: u32,
+        fps: u32,
+        resolution_rx: crossbeam_channel::Receiver<BitrateCommand>,
+        encoder_resolution_tx: crossbeam_channel::Sender<ResolutionChange>,
+    ) -> Result<Self> {
+        let mut this = Self::new(drm_device, width, height, enc_width, enc_height, fps)?;
+        this.resolution_rx = Some(resolution_rx);
+        this.encoder_resolution_tx = Some(encoder_resolution_tx);
+        Ok(this)
+    }
+
     pub fn frames_rgb(&self) -> &AvHwFrameCtx {
         let _ = self.hw_dev.as_ptr();
         &self.frames_rgb
     }
 
     pub fn import_and_scale(&mut self, hw_frame: &ff::frame::Video) -> Result<CpuNv12Frame> {
-        let mut filter_src_ctx = self.filter_graph.get("in").ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
+        self.poll_resolution_commands()?;
+
+        let mut filter_src_ctx = self
+            .filter_graph
+            .get("in")
+            .ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
         let mut filter_src = filter_src_ctx.source();
-        let mut filter_sink_ctx = self.filter_graph.get("out").ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
+        let mut filter_sink_ctx = self
+            .filter_graph
+            .get("out")
+            .ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
         let mut filter_sink = filter_sink_ctx.sink();
 
         filter_src
@@ -740,20 +809,28 @@ impl SwEncImport {
         }
 
         if extra_count > 0 {
-            tracing::warn!("software import filter produced {extra_count} extra frame(s); dropping extras");
+            tracing::warn!(
+                "software import filter produced {extra_count} extra frame(s); dropping extras"
+            );
         }
 
         first.ok_or_else(|| anyhow::anyhow!("software pipeline produced no scaled frame"))
     }
 
     pub fn flush_import(&mut self) -> Result<Vec<CpuNv12Frame>> {
-        let mut filter_src_ctx = self.filter_graph.get("in").ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
+        let mut filter_src_ctx = self
+            .filter_graph
+            .get("in")
+            .ok_or_else(|| anyhow::anyhow!("filter 'in' not found"))?;
         let mut filter_src = filter_src_ctx.source();
         if let Err(e) = filter_src.flush() {
             tracing::debug!("filter source flush error: {e}");
         }
 
-        let mut filter_sink_ctx = self.filter_graph.get("out").ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
+        let mut filter_sink_ctx = self
+            .filter_graph
+            .get("out")
+            .ok_or_else(|| anyhow::anyhow!("filter 'out' not found"))?;
         let mut filter_sink = filter_sink_ctx.sink();
         let mut frames = Vec::new();
         loop {
@@ -767,6 +844,54 @@ impl SwEncImport {
         Ok(frames)
     }
 
+    fn poll_resolution_commands(&mut self) -> Result<()> {
+        let Some(rx) = self.resolution_rx.as_ref().cloned() else {
+            return Ok(());
+        };
+
+        let mut requested = None;
+        while let Ok(cmd) = rx.try_recv() {
+            match cmd {
+                BitrateCommand::UpdateResolution { width, height } => {
+                    requested = Some((width & !1, height & !1));
+                }
+                BitrateCommand::UpdateBitrate { .. } => {}
+            }
+        }
+
+        let Some((width, height)) = requested else {
+            return Ok(());
+        };
+        if width == self.enc_width && height == self.enc_height {
+            return Ok(());
+        }
+
+        tracing::info!(
+            from = format_args!("{}x{}", self.enc_width, self.enc_height),
+            to = format_args!("{}x{}", width, height),
+            "rebuilding software import filter graph for resolution change"
+        );
+        let _ = self.flush_import();
+        self.filter_graph = build_swenc_filter_graph(
+            &self.hw_dev,
+            &self.frames_rgb,
+            self.width,
+            self.height,
+            width,
+            height,
+            self.fps,
+        )?;
+        self.enc_width = width;
+        self.enc_height = height;
+
+        if let Some(tx) = &self.encoder_resolution_tx {
+            tx.send(ResolutionChange { width, height })
+                .map_err(|_| anyhow::anyhow!("encoder resolution channel disconnected"))?;
+        }
+
+        Ok(())
+    }
+
     fn transfer_filtered_to_cpu(&self, filtered: &ff::frame::Video) -> Result<CpuNv12Frame> {
         // SAFETY: av_frame_alloc returns a newly allocated AVFrame or null,
         // which is checked below.
@@ -798,7 +923,9 @@ impl SwEncImport {
             }
             let y_stride = (*sw_nv12).linesize[0] as usize;
             let uv_stride = (*sw_nv12).linesize[1] as usize;
-            if (*sw_nv12).width != self.enc_width as i32 || (*sw_nv12).height != self.enc_height as i32 {
+            if (*sw_nv12).width != self.enc_width as i32
+                || (*sw_nv12).height != self.enc_height as i32
+            {
                 ffi::av_frame_free(&mut sw_nv12);
                 bail!("NV12 transfer frame has unexpected dimensions");
             }
@@ -826,12 +953,38 @@ pub struct SwEncEncode {
     enc_video: ff::codec::encoder::video::Video,
     output: Option<FrameOutput>,
     yuv_frame: *mut ffi::AVFrame,
+    last_frame_hash: u64,
+    frame_count: u64,
     starting_timestamp: Option<i64>,
     frames_written: bool,
     webrtc_disconnected: bool,
     webrtc_paused: Option<Arc<AtomicBool>>,
+    bitrate_rx: crossbeam_channel::Receiver<BitrateCommand>,
+    resolution_rx: crossbeam_channel::Receiver<ResolutionChange>,
     enc_width: u32,
     enc_height: u32,
+    fps: u32,
+    bitrate: u64,
+    gop_size: u32,
+}
+
+const FNV1A_OFFSET_BASIS: u64 = 0xcbf29ce484222325;
+const FNV1A_PRIME: u64 = 0x100000001b3;
+const Y_PLANE_HASH_ROW_STEP: usize = 8;
+
+fn hash_sampled_y_plane(y_data: &[u8], width: usize, height: usize, stride: usize) -> u64 {
+    let mut hash = FNV1A_OFFSET_BASIS;
+
+    for row in (0..height).step_by(Y_PLANE_HASH_ROW_STEP) {
+        let row_start = row * stride;
+        let row_end = row_start + width;
+        for &byte in &y_data[row_start..row_end] {
+            hash ^= u64::from(byte);
+            hash = hash.wrapping_mul(FNV1A_PRIME);
+        }
+    }
+
+    hash
 }
 
 // SAFETY: SwEncEncode owns sws_ctx/yuv_frame/enc_video exclusively after construction.
@@ -852,18 +1005,29 @@ impl SwEncEncode {
         let (enc_video, octx) =
             create_software_h264_muxer(output_path, enc_width, enc_height, fps, bitrate, gop_size)?;
         let yuv_frame = alloc_yuv420p_frame(enc_width, enc_height)?;
+        let (dummy_tx, bitrate_rx) = crossbeam_channel::bounded(1);
+        drop(dummy_tx);
+        let (dummy_resolution_tx, resolution_rx) = crossbeam_channel::bounded(1);
+        drop(dummy_resolution_tx);
 
         Ok(Self {
             sws_ctx,
             enc_video,
             output: Some(FrameOutput::Muxer(octx)),
             yuv_frame,
+            last_frame_hash: 0,
+            frame_count: 0,
             starting_timestamp: None,
             frames_written: false,
             webrtc_disconnected: false,
             webrtc_paused: None,
+            bitrate_rx,
+            resolution_rx,
             enc_width,
             enc_height,
+            fps,
+            bitrate,
+            gop_size,
         })
     }
 
@@ -876,9 +1040,12 @@ impl SwEncEncode {
         gop_size: u32,
         tx: crossbeam_channel::Sender<Vec<u8>>,
         webrtc_paused: Arc<AtomicBool>,
+        bitrate_rx: crossbeam_channel::Receiver<BitrateCommand>,
+        resolution_rx: crossbeam_channel::Receiver<ResolutionChange>,
     ) -> Result<Self> {
         let sws_ctx = create_nv12_to_yuv420p_sws(enc_width, enc_height)?;
-        let enc_video = create_software_h264_encoder(enc_width, enc_height, fps, bitrate, gop_size)?;
+        let enc_video =
+            create_software_h264_encoder(enc_width, enc_height, fps, bitrate, gop_size)?;
         let yuv_frame = alloc_yuv420p_frame(enc_width, enc_height)?;
 
         Ok(Self {
@@ -886,12 +1053,19 @@ impl SwEncEncode {
             enc_video,
             output: Some(FrameOutput::Channel(tx)),
             yuv_frame,
+            last_frame_hash: 0,
+            frame_count: 0,
             starting_timestamp: None,
             frames_written: false,
             webrtc_disconnected: false,
             webrtc_paused: Some(webrtc_paused),
+            bitrate_rx,
+            resolution_rx,
             enc_width,
             enc_height,
+            fps,
+            bitrate,
+            gop_size,
         })
     }
 
@@ -914,6 +1088,28 @@ impl SwEncEncode {
         if self.webrtc_disconnected {
             return Ok(());
         }
+
+        // Must drain before the stride check: the import thread emits
+        // ResolutionChange before the new (smaller-stride) frame arrives.
+        while let Ok(cmd) = self.bitrate_rx.try_recv() {
+            match cmd {
+                BitrateCommand::UpdateBitrate { target_bps } => {
+                    tracing::info!(target_bps, "updating encoder bitrate from BWE feedback");
+                    self.bitrate = target_bps;
+                    // SAFETY: enc_video is an opened AVCodecContext exclusively owned by &mut self.
+                    unsafe {
+                        let ctx = self.enc_video.as_mut_ptr();
+                        (*ctx).bit_rate = target_bps as i64;
+                    }
+                }
+                BitrateCommand::UpdateResolution { .. } => {}
+            }
+        }
+
+        while let Ok(change) = self.resolution_rx.try_recv() {
+            self.recreate_encoder(change.width, change.height)?;
+        }
+
         if frame.y_stride < self.enc_width as usize || frame.uv_stride < self.enc_width as usize {
             bail!("CPU NV12 frame stride is smaller than encoder width");
         }
@@ -923,6 +1119,24 @@ impl SwEncEncode {
             }
         }
 
+        let width = self.enc_width as usize;
+        let height = self.enc_height as usize;
+        let required_y_len = frame.y_stride * height.saturating_sub(1) + width;
+        if frame.y_data.len() < required_y_len {
+            bail!("CPU NV12 frame Y plane is smaller than encoder dimensions");
+        }
+
+        let frame_index = self.frame_count;
+        self.frame_count = self.frame_count.saturating_add(1);
+        let current_hash = hash_sampled_y_plane(&frame.y_data, width, height, frame.y_stride);
+        let force_gop_frame = self.gop_size > 0 && frame_index % u64::from(self.gop_size) == 0;
+        if frame_index > 0 && !force_gop_frame && current_hash == self.last_frame_hash {
+            tracing::debug!(frame_index, "skipping duplicate frame");
+            self.last_frame_hash = current_hash;
+            return Ok(());
+        }
+        self.last_frame_hash = current_hash;
+
         // SAFETY: yuv_frame is an owned reusable YUV420P frame at the same dimensions as sw_nv12;
         // sws_ctx was created for NV12 -> YUV420P with no resize, so sws_scale only converts format.
         unsafe {
@@ -930,7 +1144,12 @@ impl SwEncEncode {
             if ret < 0 {
                 bail!("av_frame_make_writable failed: {}", ff_err(ret));
             }
-            let src_slices = [frame.y_data.as_ptr(), frame.uv_data.as_ptr(), ptr::null(), ptr::null()];
+            let src_slices = [
+                frame.y_data.as_ptr(),
+                frame.uv_data.as_ptr(),
+                ptr::null(),
+                ptr::null(),
+            ];
             let src_strides = [frame.y_stride as i32, frame.uv_stride as i32, 0, 0];
             let scaled = ffi::sws_scale(
                 self.sws_ctx,
@@ -957,13 +1176,49 @@ impl SwEncEncode {
             (*self.yuv_frame).pts = pts;
             let ret = ffi::avcodec_send_frame(self.enc_video.as_mut_ptr(), self.yuv_frame);
             if ret < 0 {
-                bail!("avcodec_send_frame failed for software encoder: {}", ff_err(ret));
+                bail!(
+                    "avcodec_send_frame failed for software encoder: {}",
+                    ff_err(ret)
+                );
             }
         }
 
         self.drain_encoder(start_ts)
     }
 
+    fn recreate_encoder(&mut self, width: u32, height: u32) -> Result<()> {
+        if width == self.enc_width && height == self.enc_height {
+            return Ok(());
+        }
+
+        tracing::info!(
+            from = format_args!("{}x{}", self.enc_width, self.enc_height),
+            to = format_args!("{}x{}", width, height),
+            "recreating WebRTC software encoder for resolution change"
+        );
+
+        if !self.sws_ctx.is_null() {
+            // SAFETY: sws_ctx is owned exclusively by self and will be replaced below.
+            unsafe { ffi::sws_freeContext(self.sws_ctx) };
+            self.sws_ctx = ptr::null_mut();
+        }
+        if !self.yuv_frame.is_null() {
+            // SAFETY: yuv_frame is owned exclusively by self and will be replaced below.
+            unsafe { ffi::av_frame_free(&mut self.yuv_frame) };
+        }
+
+        self.sws_ctx = create_nv12_to_yuv420p_sws(width, height)?;
+        self.enc_video =
+            create_software_h264_encoder(width, height, self.fps, self.bitrate, self.gop_size)?;
+        self.yuv_frame = alloc_yuv420p_frame(width, height)?;
+        self.enc_width = width;
+        self.enc_height = height;
+        self.last_frame_hash = 0;
+        self.frame_count = 0;
+
+        Ok(())
+    }
+
     fn write_trailer_if_needed(&mut self) -> Result<()> {
         if self.frames_written {
             if let Some(FrameOutput::Muxer(ref mut octx)) = self.output {
@@ -1026,9 +1281,8 @@ impl SwEncEncode {
                         // `data` is non-null, so `data` points to `size` initialized
                         // bytes owned by the packet. `u8` has alignment 1, and the
                         // slice is copied into a Vec before the packet is unreffed.
-                        let data: &[u8] = unsafe {
-                            std::slice::from_raw_parts(raw.data, raw.size as usize)
-                        };
+                        let data: &[u8] =
+                            unsafe { std::slice::from_raw_parts(raw.data, raw.size as usize) };
                         match tx.try_send(data.to_vec()) {
                             Ok(()) => {}
                             Err(crossbeam_channel::TrySendError::Full(frame)) => {
@@ -1095,7 +1349,8 @@ impl SwEncState {
             "SwEncState::new: GPU downscale {width}x{height} BGRA -> {enc_width}x{enc_height} NV12, software H.264"
         );
         let import = SwEncImport::new(drm_device, width, height, enc_width, enc_height, fps)?;
-        let encode = SwEncEncode::new_muxer(output_path, enc_width, enc_height, fps, bitrate, gop_size)?;
+        let encode =
+            SwEncEncode::new_muxer(output_path, enc_width, enc_height, fps, bitrate, gop_size)?;
         Ok(Self { import, encode })
     }
 
@@ -1116,6 +1371,10 @@ impl SwEncState {
             "SwEncState::new_webrtc: GPU downscale {width}x{height} BGRA -> {enc_width}x{enc_height} NV12, software H.264 -> WebRTC"
         );
         let import = SwEncImport::new(drm_device, width, height, enc_width, enc_height, fps)?;
+        let (dummy_tx, bitrate_rx) = crossbeam_channel::bounded(1);
+        drop(dummy_tx);
+        let (dummy_resolution_tx, resolution_rx) = crossbeam_channel::bounded(1);
+        drop(dummy_resolution_tx);
         let encode = SwEncEncode::new_webrtc(
             enc_width,
             enc_height,
@@ -1124,6 +1383,8 @@ impl SwEncState {
             gop_size,
             tx,
             webrtc_paused,
+            bitrate_rx,
+            resolution_rx,
         )?;
         Ok(Self { import, encode })
     }
@@ -1348,6 +1609,10 @@ fn create_software_h264_muxer(
             let val = CString::new("6").unwrap();
             ffi::av_opt_set((*enc.as_mut_ptr()).priv_data, key.as_ptr(), val.as_ptr(), 0);
             (*enc.as_mut_ptr()).profile = ffi::AV_PROFILE_H264_HIGH as i32;
+            // SAFETY: enc is a valid, initialized AVCodecContext from
+            // avcodec_alloc_context3. Setting level is a simple i32 field
+            // assignment on a properly aligned struct.
+            (*enc.as_mut_ptr()).level = 40; // H.264 Level 4.0 (up to 1080p@30)
         }
     }
 
@@ -1472,8 +1737,17 @@ fn create_software_h264_encoder(
             // High profile via AVCodecContext.profile (not x264opts — x264 rejects it there).
             // High enables CABAC + 8x8dct automatically.
             (*enc.as_mut_ptr()).profile = ffi::AV_PROFILE_H264_HIGH as i32;
+            // SAFETY: enc is a valid, initialized AVCodecContext from
+            // avcodec_alloc_context3. Setting level is a simple i32 field
+            // assignment on a properly aligned struct.
+            (*enc.as_mut_ptr()).level = 42; // H.264 Level 4.2 (up to 1440p@30)
             let key = CString::new("x264opts").unwrap();
-            let val = CString::new("repeat_headers=1").unwrap();
+            let vbv_maxrate = bitrate;
+            let vbv_bufsize = bitrate / 4;
+            let val = CString::new(format!(
+                "repeat_headers=1:vbv-maxrate={vbv_maxrate}:vbv-bufsize={vbv_bufsize}"
+            ))
+            .unwrap();
             ffi::av_opt_set((*enc.as_mut_ptr()).priv_data, key.as_ptr(), val.as_ptr(), 0);
         }
     }
@@ -1590,3 +1864,129 @@ fn build_filter_graph(
 
     Ok(graph)
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    // ── Task 1: VBV x264opts formatting ──
+
+    #[test]
+    fn vbv_x264opts_format() {
+        let bitrate: u64 = 5_000_000;
+        let vbv_maxrate = bitrate;
+        let vbv_bufsize = bitrate / 4;
+        let opts = format!("repeat_headers=1:vbv-maxrate={vbv_maxrate}:vbv-bufsize={vbv_bufsize}");
+        assert!(opts.contains("vbv-maxrate=5000000"));
+        assert!(opts.contains("vbv-bufsize=1250000"));
+    }
+
+    #[test]
+    fn vbv_bufsize_is_quarter_of_maxrate() {
+        for bitrate in [1_000_000, 5_000_000, 10_000_000] {
+            let maxrate = bitrate;
+            let bufsize = bitrate / 4;
+            assert_eq!(bufsize * 4, maxrate, "bufsize should be maxrate/4");
+        }
+    }
+
+    // ── Task 3: GOP formula ──
+
+    #[test]
+    fn webrtc_gop_formula() {
+        assert_eq!((15u32 * 2).max(20), 30); // 15fps -> 30
+        assert_eq!((30u32 * 2).max(20), 60); // 30fps -> 60
+        assert_eq!((60u32 * 2).max(20), 120); // 60fps -> 120
+        assert_eq!((5u32 * 2).max(20), 20); // 5fps -> 20 (floor)
+    }
+
+    #[test]
+    fn h264_level_values() {
+        // Level 4.0 supports up to 1080p@30fps (used for file muxer)
+        assert_eq!(40i32, 40);
+        // Level 4.2 supports up to 1440p@30fps (used for WebRTC low-latency encoder)
+        assert_eq!(42i32, 42);
+    }
+
+    // ── Task 4: Duplicate frame hash detection ──
+
+    #[test]
+    fn hash_sampled_y_plane_first_frame_consistent() {
+        let width = 64;
+        let height = 64;
+        let stride = 64;
+        let y_data = vec![0u8; stride * height];
+        let hash1 = hash_sampled_y_plane(&y_data, width, height, stride);
+        let hash2 = hash_sampled_y_plane(&y_data, width, height, stride);
+        assert_eq!(hash1, hash2, "same input should produce same hash");
+    }
+
+    #[test]
+    fn hash_sampled_y_plane_detects_different_frames() {
+        let width = 64;
+        let height = 64;
+        let stride = 64;
+        let y_data1 = vec![0u8; stride * height];
+        let y_data2 = vec![128u8; stride * height];
+        let hash1 = hash_sampled_y_plane(&y_data1, width, height, stride);
+        let hash2 = hash_sampled_y_plane(&y_data2, width, height, stride);
+        assert_ne!(
+            hash1, hash2,
+            "different frame data should produce different hashes"
+        );
+    }
+
+    #[test]
+    fn hash_sampled_y_plane_samples_every_8th_row() {
+        // Changing a non-sampled row (e.g., row 1) should NOT change the hash
+        let width = 64;
+        let height = 64;
+        let stride = 64;
+        let y_data1 = vec![0u8; stride * height];
+        let mut y_data2 = vec![0u8; stride * height];
+        // Row 1 is NOT sampled (sampling is every 8th row: 0, 8, 16, ...)
+        y_data2[stride * 1..stride * 1 + width].fill(255);
+        let hash1 = hash_sampled_y_plane(&y_data1, width, height, stride);
+        let hash2 = hash_sampled_y_plane(&y_data2, width, height, stride);
+        assert_eq!(
+            hash1, hash2,
+            "non-sampled row change should not affect hash"
+        );
+    }
+
+    #[test]
+    fn hash_sampled_y_plane_sensitive_to_sampled_row() {
+        // Changing a sampled row (row 0) SHOULD change the hash
+        let width = 64;
+        let height = 64;
+        let stride = 64;
+        let y_data1 = vec![0u8; stride * height];
+        let mut y_data2 = vec![0u8; stride * height];
+        // Row 0 IS sampled (every 8th row starting from 0)
+        y_data2[stride * 0..stride * 0 + width].fill(255);
+        let hash1 = hash_sampled_y_plane(&y_data1, width, height, stride);
+        let hash2 = hash_sampled_y_plane(&y_data2, width, height, stride);
+        assert_ne!(
+            hash1, hash2,
+            "sampled row change should produce different hash"
+        );
+    }
+
+    #[test]
+    fn hash_sampled_y_plane_handles_stride_greater_than_width() {
+        // Stride can be larger than width due to alignment; unused padding should not affect hash
+        let width = 32;
+        let height = 16;
+        let stride = 64; // padded stride
+        let y_data1 = vec![0u8; stride * height];
+        let mut y_data2 = vec![0u8; stride * height];
+        // Fill the padding area (columns 32..63) of row 0 with garbage
+        y_data2[width..stride].fill(0xFF);
+        let hash1 = hash_sampled_y_plane(&y_data1, width, height, stride);
+        let hash2 = hash_sampled_y_plane(&y_data2, width, height, stride);
+        assert_eq!(
+            hash1, hash2,
+            "padding bytes beyond width should not affect hash"
+        );
+    }
+}

src/cap_portal.rs

@@ -225,9 +225,7 @@ impl CapPortal {
         proxy
             .select_sources(&session, options)
             .await
-            .map_err(|e| {
-                anyhow::anyhow!("Screen sharing permission denied: {e}")
-            })?;
+            .map_err(|e| anyhow::anyhow!("Screen sharing permission denied: {e}"))?;
 
         let response = proxy
             .start(&session, None, Default::default())
@@ -272,7 +270,10 @@ fn verify_secure_dir(path: &std::path::Path) -> bool {
     match std::fs::symlink_metadata(path) {
         Ok(meta) => {
             if meta.file_type().is_symlink() {
-                tracing::warn!("Token parent dir is a symlink, rejecting: {}", path.display());
+                tracing::warn!(
+                    "Token parent dir is a symlink, rejecting: {}",
+                    path.display()
+                );
                 return false;
             }
             // Must be a directory
@@ -282,7 +283,10 @@ fn verify_secure_dir(path: &std::path::Path) -> bool {
             }
             // Must be owned by current user
             if meta.uid() != unsafe { libc::getuid() } {
-                tracing::warn!("Token parent dir not owned by current user: {}", path.display());
+                tracing::warn!(
+                    "Token parent dir not owned by current user: {}",
+                    path.display()
+                );
                 return false;
             }
             // No group or other permissions (mode must be 0o700 exactly within the 0o777 mask)
@@ -346,7 +350,10 @@ fn load_restore_token_from(path: PathBuf) -> Option<String> {
     };
 
     if meta.file_type().is_symlink() {
-        tracing::warn!("Token file is a symlink, refusing to read: {}", path.display());
+        tracing::warn!(
+            "Token file is a symlink, refusing to read: {}",
+            path.display()
+        );
         return None;
     }
     if !meta.is_file() {
@@ -369,7 +376,11 @@ fn load_restore_token_from(path: PathBuf) -> Option<String> {
 
     let token = std::fs::read_to_string(&path).ok()?;
     let trimmed = token.trim().to_string();
-    if trimmed.is_empty() { None } else { Some(trimmed) }
+    if trimmed.is_empty() {
+        None
+    } else {
+        Some(trimmed)
+    }
 }
 
 fn save_restore_token(token: &str) {
@@ -490,7 +501,9 @@ fn pipewire_thread(ctx: PwThreadCtx) {
     let mainloop = match pw::main_loop::MainLoopBox::new(None) {
         Ok(ml) => ml,
         Err(e) => {
-            if let Err(e) = event_tx.try_send(PwCtrlEvent::Error(format!("MainLoop::new failed: {e}"))) {
+            if let Err(e) =
+                event_tx.try_send(PwCtrlEvent::Error(format!("MainLoop::new failed: {e}")))
+            {
                 tracing::error!("MainLoop::new failed and error channel also failed: {e}");
             }
             return;
@@ -500,7 +513,9 @@ fn pipewire_thread(ctx: PwThreadCtx) {
     let context = match pw::context::ContextBox::new(mainloop.loop_(), None) {
         Ok(c) => c,
         Err(e) => {
-            if let Err(e) = event_tx.try_send(PwCtrlEvent::Error(format!("Context::new failed: {e}"))) {
+            if let Err(e) =
+                event_tx.try_send(PwCtrlEvent::Error(format!("Context::new failed: {e}")))
+            {
                 tracing::error!("Context::new failed and error channel also failed: {e}");
             }
             return;
@@ -510,7 +525,8 @@ fn pipewire_thread(ctx: PwThreadCtx) {
     let core = match context.connect_fd(pw_fd, None) {
         Ok(c) => c,
         Err(e) => {
-            if let Err(e) = event_tx.try_send(PwCtrlEvent::Error(format!("connect_fd failed: {e}"))) {
+            if let Err(e) = event_tx.try_send(PwCtrlEvent::Error(format!("connect_fd failed: {e}")))
+            {
                 tracing::error!("connect_fd failed and error channel also failed: {e}");
             }
             return;
@@ -534,7 +550,9 @@ fn pipewire_thread(ctx: PwThreadCtx) {
     ) {
         Ok(s) => s,
         Err(e) => {
-            if let Err(e) = event_tx.try_send(PwCtrlEvent::Error(format!("Stream::new failed: {e}"))) {
+            if let Err(e) =
+                event_tx.try_send(PwCtrlEvent::Error(format!("Stream::new failed: {e}")))
+            {
                 tracing::error!("Stream::new failed and error channel also failed: {e}");
             }
             return;
@@ -608,8 +626,10 @@ fn pipewire_thread(ctx: PwThreadCtx) {
                     "PipeWire format negotiated: {width}x{height}, \
                      drm_format={drm_format:#010x}, modifier={modifier:#x}, \
                      framerate={}/{}, max_framerate={}/{}",
-                    framerate.num, framerate.denom,
-                    max_framerate.num, max_framerate.denom,
+                    framerate.num,
+                    framerate.denom,
+                    max_framerate.num,
+                    max_framerate.denom,
                 );
             }
         })
@@ -621,7 +641,6 @@ fn pipewire_thread(ctx: PwThreadCtx) {
             let frame_tx = frame_tx.clone();
             let dropped = dropped;
             move |stream, _| {
-
                 let raw_buf = unsafe { stream.dequeue_raw_buffer() };
                 if raw_buf.is_null() {
                     tracing::trace!("process: null raw_buf");
@@ -742,7 +761,8 @@ fn pipewire_thread(ctx: PwThreadCtx) {
         StreamFlags::AUTOCONNECT | StreamFlags::MAP_BUFFERS,
         &mut params,
     ) {
-        if let Err(e) = event_tx.try_send(PwCtrlEvent::Error(format!("stream.connect failed: {e}"))) {
+        if let Err(e) = event_tx.try_send(PwCtrlEvent::Error(format!("stream.connect failed: {e}")))
+        {
             tracing::error!("stream.connect failed and error channel also failed: {e}");
         }
         return;
@@ -918,7 +938,10 @@ mod tests {
 
         let meta = std::fs::symlink_metadata(&new_dir).unwrap();
         let mode = meta.permissions().mode() & 0o777;
-        assert_eq!(mode, 0o700, "created directory should be 0700, got {mode:o}");
+        assert_eq!(
+            mode, 0o700,
+            "created directory should be 0700, got {mode:o}"
+        );
     }
 
     #[test]
@@ -933,7 +956,10 @@ mod tests {
 
         let meta = std::fs::symlink_metadata(path).unwrap();
         let mode = meta.permissions().mode() & 0o777;
-        assert_eq!(mode, 0o700, "tightened directory should be 0700, got {mode:o}");
+        assert_eq!(
+            mode, 0o700,
+            "tightened directory should be 0700, got {mode:o}"
+        );
     }
 
     #[test]
@@ -947,7 +973,10 @@ mod tests {
         let meta = std::fs::symlink_metadata(&token_path).unwrap();
         let mode = meta.permissions().mode() & 0o777;
         assert_eq!(mode, 0o600, "token file should be 0600, got {mode:o}");
-        assert_eq!(std::fs::read_to_string(&token_path).unwrap(), "secret-token-123");
+        assert_eq!(
+            std::fs::read_to_string(&token_path).unwrap(),
+            "secret-token-123"
+        );
     }
 
     #[test]

src/fps_limit.rs

@@ -82,7 +82,12 @@ mod tests {
             }
         }
 
-        assert!(outputs.len() >= 3, "expected at least 3 outputs, got {} ({:?})", outputs.len(), outputs);
+        assert!(
+            outputs.len() >= 3,
+            "expected at least 3 outputs, got {} ({:?})",
+            outputs.len(),
+            outputs
+        );
         assert_eq!(outputs[0], 0);
     }
 

src/lib.rs

@@ -4,8 +4,8 @@ pub mod backend_detect;
 pub mod cap_portal;
 pub mod cap_wlr_screencopy;
 pub mod fps_limit;
-pub mod stats;
 pub mod state;
 pub mod state_portal;
+pub mod stats;
 pub mod transform;
 pub mod webrtc;

src/main.rs

@@ -15,9 +15,9 @@ mod backend_detect; // 截屏后端自动检测（wlroots vs Portal/PipeWire）
 mod cap_portal; // XDG Portal 屏幕捕获
 mod cap_wlr_screencopy; // wlroots wlr-screencopy 截屏协议
 mod fps_limit; // 帧率限制器
-mod stats; // 管道性能统计（卡顿诊断）
 mod state; // wlr-screencopy 后端的主状态机
 mod state_portal; // Portal/PipeWire 后端的主状态机
+mod stats; // 管道性能统计（卡顿诊断）
 mod transform; // 图像变换（旋转/翻转）
 mod webrtc; // WebRTC 传输（str0m Sans-IO）
 
@@ -46,21 +46,27 @@ fn main() -> Result<()> {
 
     // 根据 verbose 模式或 RUST_LOG 环境变量设置日志级别
     // 支持 RUST_LOG 粒度控制（如 RUST_LOG=wl_webrtc::webrtc=trace）
-    let env_filter = tracing_subscriber::EnvFilter::try_from_default_env()
-        .unwrap_or_else(|_| {
-            if args.verbose {
-                tracing_subscriber::EnvFilter::new("debug")
-            } else {
-                tracing_subscriber::EnvFilter::new("info")
-            }
-        });
+    let env_filter = tracing_subscriber::EnvFilter::try_from_default_env().unwrap_or_else(|_| {
+        if args.verbose {
+            tracing_subscriber::EnvFilter::new("debug")
+        } else {
+            tracing_subscriber::EnvFilter::new("info")
+        }
+    });
     tracing_subscriber::fmt()
         .with_env_filter(env_filter)
         .with_writer(std::io::stderr)
         .init();
 
     tracing::info!("wl-webrtc starting");
-    tracing::debug!("Args: output={:?} fps={} codec={} port={} verbose={}", args.output, args.fps, args.codec, args.port, args.verbose);
+    tracing::debug!(
+        "Args: output={:?} fps={} codec={} port={} verbose={}",
+        args.output,
+        args.fps,
+        args.codec,
+        args.port,
+        args.verbose
+    );
 
     // MVP 阶段仅支持 H.264 编码，不支持 HEVC
     if args.codec != "h264" {
@@ -352,7 +358,9 @@ fn run_portal_pipewire(args: Args) -> Result<()> {
 
         // Portal 状态机遇到致命错误时退出
         if state.is_errored() {
-            tracing::error!("Fatal error in portal state machine (check preceding error logs), exiting");
+            tracing::error!(
+                "Fatal error in portal state machine (check preceding error logs), exiting"
+            );
             running = false;
         }
     }

src/state.rs

@@ -539,7 +539,10 @@ impl<S: CaptureSource> State<S> {
                 // takes ownership of the fd, and the original fd is owned by map_frame.
                 let fd_dup = unsafe { libc::dup(obj.fd) };
                 if fd_dup < 0 {
-                    tracing::error!("failed to dup dma-buf fd: {}", std::io::Error::last_os_error());
+                    tracing::error!(
+                        "failed to dup dma-buf fd: {}",
+                        std::io::Error::last_os_error()
+                    );
                     // wayland-client does not auto-destroy params on Drop.
                     params.destroy();
                     self.errored = true;
@@ -641,7 +644,11 @@ impl<S: CaptureSource> State<S> {
             if last.elapsed() >= std::time::Duration::from_secs(10) {
                 let delta = self.stats_frames;
                 let fps = delta as f64 / last.elapsed().as_secs_f64();
-                tracing::info!(frames = self.stats_frames, fps = format!("{fps:.1}"), "encoding stats");
+                tracing::info!(
+                    frames = self.stats_frames,
+                    fps = format!("{fps:.1}"),
+                    "encoding stats"
+                );
                 self.stats_last_time = Some(std::time::Instant::now());
                 self.stats_frames = 0;
             }
@@ -672,7 +679,9 @@ impl<S: CaptureSource> State<S> {
     }
 
     pub fn poll_webrtc(&mut self) -> Result<()> {
-        let Some(ref mut wrtc) = self.webrtc else { return Ok(()) };
+        let Some(ref mut wrtc) = self.webrtc else {
+            return Ok(());
+        };
 
         wrtc.handle_signaling()?;
         wrtc.poll_and_feed()?;
@@ -697,7 +706,8 @@ impl<S: CaptureSource> State<S> {
                     continue;
                 }
                 count += 1;
-                if let Err(e) = wrtc.write_h264_frame(&data, self.webrtc_frames_sent, self.args.fps) {
+                if let Err(e) = wrtc.write_h264_frame(&data, self.webrtc_frames_sent, self.args.fps)
+                {
                     tracing::debug!("WebRTC write frame error: {e}");
                 }
                 self.stats.record_send(0.0, None);
@@ -709,10 +719,8 @@ impl<S: CaptureSource> State<S> {
         }
 
         if self.args.stats && self.stats.should_snapshot() {
-            self.stats.set_queue_depths(
-                0,
-                self.webrtc_rx.as_ref().map(|r| r.len()).unwrap_or(0),
-            );
+            self.stats
+                .set_queue_depths(0, self.webrtc_rx.as_ref().map(|r| r.len()).unwrap_or(0));
             let snap = self.stats.snapshot_and_reset();
             tracing::info!("stats: {snap}");
         }
@@ -752,9 +760,12 @@ impl<S: CaptureSource> State<S> {
             .unwrap_or_else(|| 2 * (width as u64) * (height as u64) * (fps as u64) / 100);
 
         let enc = if let Some(ref tx) = self.webrtc_tx {
-            let (enc_w, enc_h) =
-                transpose_if_transform_transposed(output_info.transform, width as i32, height as i32);
-            let actual_gop_size = self.args.gop_size.unwrap_or((fps / 2).max(10));
+            let (enc_w, enc_h) = transpose_if_transform_transposed(
+                output_info.transform,
+                width as i32,
+                height as i32,
+            );
+            let actual_gop_size = self.args.gop_size.unwrap_or((fps * 2).max(20));
             match SwEncState::new_webrtc(
                 &drm_path,
                 width,
@@ -765,7 +776,10 @@ impl<S: CaptureSource> State<S> {
                 bitrate,
                 actual_gop_size,
                 tx.clone(),
-                self.webrtc_paused.as_ref().expect("webrtc_paused must exist when webrtc_tx exists").clone(),
+                self.webrtc_paused
+                    .as_ref()
+                    .expect("webrtc_paused must exist when webrtc_tx exists")
+                    .clone(),
             ) {
                 Ok(enc) => StreamingEncoder::WebRtc(enc),
                 Err(e) => {
@@ -775,7 +789,11 @@ impl<S: CaptureSource> State<S> {
                 }
             }
         } else {
-            let output_path = self.args.output.as_deref().expect("output required for MP4 mode");
+            let output_path = self
+                .args
+                .output
+                .as_deref()
+                .expect("output required for MP4 mode");
             match crate::avhw::create_encoder(
                 &drm_path,
                 Path::new(output_path),
@@ -1035,7 +1053,7 @@ impl<S: CaptureSource> Dispatch<WlRegistry, GlobalListContents> for State<S> {
         qhandle: &QueueHandle<State<S>>,
     ) {
         use wayland_client::protocol::wl_registry::Event as RegistryEvent;
-        
+
         match event {
             RegistryEvent::Global {
                 name,

src/state_portal.rs

@@ -5,10 +5,12 @@ use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
 use std::time::{Duration, Instant};
 
-use anyhow::{bail, Result};     // 错误处理工具
+use anyhow::{bail, Result}; // 错误处理工具
 
-use crate::args::Args;          // 命令行参数
-use crate::avhw::{self, CpuNv12Frame, SwEncEncode, SwEncImport, SwEncState}; // 软件编码器状态（VAAPI 导入 + H.264 编码）
+use crate::args::Args; // 命令行参数
+use crate::avhw::{
+    self, BitrateCommand, CpuNv12Frame, ResolutionChange, SwEncEncode, SwEncImport, SwEncState,
+}; // 软件编码器状态（VAAPI 导入 + H.264 编码）
 use crate::cap_portal::{CapPortal, PwCtrlEvent, PwDmaBufFrame}; // PipeWire 屏幕采集端点
 use crate::stats::{FrameTimings, PipelineStats}; // 管道统计（帧计时、每秒快照）
 use crate::webrtc::WebRtcState; // WebRTC 信令与媒体传输
@@ -43,26 +45,26 @@ struct WebrtcThread {
 /// 负责管理从 PipeWire 采集屏幕帧、通过 VAAPI 硬件编码的完整生命周期。
 /// 工作流程：等待第一帧 → 创建编码器 → 持续编码帧数据。
 pub struct StatePortal {
-    stage: PortalStage,                             // 当前采集阶段（等待首帧 / 流式编码中）
-    enc: Option<SwEncState>,                        // 软件编码器，首帧到达后初始化
+    stage: PortalStage,      // 当前采集阶段（等待首帧 / 流式编码中）
+    enc: Option<SwEncState>, // 软件编码器，首帧到达后初始化
     enc_import: Option<SwEncImport>,
     enc_thread: Option<EncodeThread>,
-    cap: CapPortal,                                 // PipeWire 屏幕采集端点
-    args: Args,                                     // 用户命令行参数
-    errored: bool,                                  // 是否遇到不可恢复的错误
-    drm_device: Option<PathBuf>,                    // DRM 渲染设备路径（可自动检测）
-    frames_encoded: u64,                            // 已编码帧数（用于 PTS 编号）
-    start_time: Option<Instant>,                    // 编码开始时间
-    stats: PipelineStats,                           // 管道统计（窗口化帧计时 + 每秒快照）
-    pw_dropped_prev: u64,                           // 上一窗口的 PipeWire 丢弃帧数（用于增量计算）
+    cap: CapPortal,              // PipeWire 屏幕采集端点
+    args: Args,                  // 用户命令行参数
+    errored: bool,               // 是否遇到不可恢复的错误
+    drm_device: Option<PathBuf>, // DRM 渲染设备路径（可自动检测）
+    frames_encoded: u64,         // 已编码帧数（用于 PTS 编号）
+    start_time: Option<Instant>, // 编码开始时间
+    stats: PipelineStats,        // 管道统计（窗口化帧计时 + 每秒快照）
+    pw_dropped_prev: u64,        // 上一窗口的 PipeWire 丢弃帧数（用于增量计算）
     webrtc: Option<WebRtcState>,
     webrtc_thread: Option<WebrtcThread>,
     webrtc_paused: Option<Arc<AtomicBool>>,
-    last_capture_arrival: Option<Instant>,  // timestamp of last real frame arrival
-    stall_start: Option<Instant>,           // when current stall began
-    last_stall_log: Option<Instant>,        // rate-limiting for stall warnings
-    last_fillable_frame: Option<CpuNv12Frame>,  // cached last frame for filler duplication
-    next_filler_at: Option<Instant>,            // when to send next filler frame
+    last_capture_arrival: Option<Instant>, // timestamp of last real frame arrival
+    stall_start: Option<Instant>,          // when current stall began
+    last_stall_log: Option<Instant>,       // rate-limiting for stall warnings
+    last_fillable_frame: Option<CpuNv12Frame>, // cached last frame for filler duplication
+    next_filler_at: Option<Instant>,       // when to send next filler frame
     filler_frames_sent: u64,
 }
 
@@ -147,7 +149,11 @@ impl StatePortal {
         // 根据阻塞模式选择不同的帧接收策略
         let frame = if block {
             // 阻塞模式：最多等待 2ms 接收帧
-            match self.cap.frame_receiver().recv_timeout(std::time::Duration::from_millis(2)) {
+            match self
+                .cap
+                .frame_receiver()
+                .recv_timeout(std::time::Duration::from_millis(2))
+            {
                 Ok(frame) => frame,
                 Err(_) => {
                     self.record_capture_timeout();
@@ -194,10 +200,10 @@ impl StatePortal {
                 let actual_bitrate = self.args.bitrate.unwrap_or_else(|| {
                     5 * (enc_width as u64) * (enc_height as u64) * (self.args.fps as u64) / 100
                 });
-                // GOP 大小：WebRTC 模式使用更小的 GOP（fps/2，最低10），MP4 模式使用 fps
+                // GOP 大小：WebRTC 模式使用较大的 GOP（fps*2，最低20），MP4 模式使用 fps
                 let actual_gop_size = self.args.gop_size.unwrap_or_else(|| {
                     if self.webrtc.is_some() {
-                        (self.args.fps / 2).max(10)
+                        (self.args.fps * 2).max(20)
                     } else {
                         self.args.fps
                     }
@@ -207,17 +213,25 @@ impl StatePortal {
                 if self.webrtc.is_some() {
                     let paused = self.webrtc_paused.as_ref()
                         .ok_or_else(|| anyhow::anyhow!("internal invariant broken: webrtc_paused missing while WebRTC mode is active"))?;
-                    let import = SwEncImport::new(
+                    let (resolution_tx, resolution_rx) =
+                        crossbeam_channel::bounded::<BitrateCommand>(4);
+                    let (encoder_resolution_tx, encoder_resolution_rx) =
+                        crossbeam_channel::bounded::<ResolutionChange>(4);
+                    let import = SwEncImport::new_with_resolution_control(
                         &drm_path,
                         frame.width,
                         frame.height,
                         enc_width,
                         enc_height,
                         self.args.fps,
+                        resolution_rx,
+                        encoder_resolution_tx,
                     )?;
                     let (webrtc_tx, webrtc_rx) = crossbeam_channel::bounded(2);
                     let (input_tx, input_rx) = crossbeam_channel::bounded::<CpuNv12Frame>(1);
-                    let (timing_tx, timing_rx) = crossbeam_channel::bounded::<EncodeThreadTiming>(32);
+                    let (timing_tx, timing_rx) =
+                        crossbeam_channel::bounded::<EncodeThreadTiming>(32);
+                    let (bitrate_tx, bitrate_rx) = crossbeam_channel::bounded::<BitrateCommand>(4);
                     let encode = SwEncEncode::new_webrtc(
                         enc_width,
                         enc_height,
@@ -226,24 +240,48 @@ impl StatePortal {
                         actual_gop_size,
                         webrtc_tx,
                         paused.clone(),
+                        bitrate_rx,
+                        encoder_resolution_rx,
                     )?;
                     let handle = std::thread::Builder::new()
                         .name("wl-webrtc-encode".into())
                         .spawn(move || encode_thread_loop(encode, input_rx, timing_tx))?;
                     self.enc_import = Some(import);
-                    self.enc_thread = Some(EncodeThread { handle: Some(handle), input_tx, timing_rx });
+                    self.enc_thread = Some(EncodeThread {
+                        handle: Some(handle),
+                        input_tx,
+                        timing_rx,
+                    });
 
-                    let wrtc = self.webrtc.take()
-                        .ok_or_else(|| anyhow::anyhow!("internal: WebRtcState missing during init"))?;
-                    let paused = self.webrtc_paused.as_ref()
+                    let wrtc = self.webrtc.take().ok_or_else(|| {
+                        anyhow::anyhow!("internal: WebRtcState missing during init")
+                    })?;
+                    let paused = self
+                        .webrtc_paused
+                        .as_ref()
                         .ok_or_else(|| anyhow::anyhow!("internal: webrtc_paused missing"))?
                         .clone();
                     let fps = self.args.fps;
                     let (sent_gap_tx, sent_gap_rx) = crossbeam_channel::bounded(64);
                     let webrtc_handle = std::thread::Builder::new()
                         .name("wl-webrtc-webrtc".into())
-                        .spawn(move || webrtc_thread_loop(wrtc, webrtc_rx, fps, paused, sent_gap_tx))?;
-                    self.webrtc_thread = Some(WebrtcThread { handle: Some(webrtc_handle), sent_gap_rx });
+                        .spawn(move || {
+                            webrtc_thread_loop(
+                                wrtc,
+                                webrtc_rx,
+                                fps,
+                                enc_width,
+                                enc_height,
+                                paused,
+                                sent_gap_tx,
+                                bitrate_tx,
+                                resolution_tx,
+                            )
+                        })?;
+                    self.webrtc_thread = Some(WebrtcThread {
+                        handle: Some(webrtc_handle),
+                        sent_gap_rx,
+                    });
                 } else {
                     // MP4 模式：编码输出写入文件
                     let output_path = self.args.output.as_deref()
@@ -263,7 +301,9 @@ impl StatePortal {
                 };
                 self.stage = PortalStage::Streaming; // 切换到流式编码阶段
                 self.start_time = Some(Instant::now());
-                tracing::info!("First frame processed, encoder initialized, transitioning to Streaming");
+                tracing::info!(
+                    "First frame processed, encoder initialized, transitioning to Streaming"
+                );
                 drop(frame); // 首帧仅用于初始化，不参与编码
             }
             PortalStage::Streaming => {
@@ -295,7 +335,10 @@ impl StatePortal {
             }
             let snap = self.stats.snapshot_and_reset();
             if self.filler_frames_sent > 0 {
-                tracing::info!("stats: {snap} filler_frames_sent={}", self.filler_frames_sent);
+                tracing::info!(
+                    "stats: {snap} filler_frames_sent={}",
+                    self.filler_frames_sent
+                );
             } else {
                 tracing::info!("stats: {snap}");
             }
@@ -321,9 +364,9 @@ impl StatePortal {
             self.last_stall_log = Some(now);
             tracing::warn!("compositor frame delivery stalled");
         } else {
-            let should_log = self
-                .last_stall_log
-                .map_or(true, |last_log| now.duration_since(last_log) >= Duration::from_secs(1));
+            let should_log = self.last_stall_log.map_or(true, |last_log| {
+                now.duration_since(last_log) >= Duration::from_secs(1)
+            });
             if should_log {
                 self.last_stall_log = Some(now);
                 tracing::warn!("compositor frame delivery stalled");
@@ -437,13 +480,11 @@ impl StatePortal {
         }
 
         // 所有候选设备均失败，返回详细错误信息
-        bail!(
-            failures
-                .into_iter()
-                .map(|(p, e)| format!("{} ({e})", p.display()))
-                .collect::<Vec<_>>()
-                .join(", ")
-        );
+        bail!(failures
+            .into_iter()
+            .map(|(p, e)| format!("{} ({e})", p.display()))
+            .collect::<Vec<_>>()
+            .join(", "));
     }
 
     /// 处理单帧 DMA-BUF 数据
@@ -513,8 +554,11 @@ impl StatePortal {
             let import_us = t_import_start.elapsed().as_micros() as u64;
             self.stats.record_import(import_us);
 
-            let enc_thread = self.enc_thread.as_ref()
-                .ok_or_else(|| anyhow::anyhow!("internal invariant broken: encode thread missing while async import is active"))?;
+            let enc_thread = self.enc_thread.as_ref().ok_or_else(|| {
+                anyhow::anyhow!(
+                    "internal invariant broken: encode thread missing while async import is active"
+                )
+            })?;
             let fillable_frame = CpuNv12Frame {
                 y_data: cpu_nv12.y_data.clone(),
                 uv_data: cpu_nv12.uv_data.clone(),
@@ -632,11 +676,20 @@ fn webrtc_thread_loop(
     mut wrtc: WebRtcState,
     webrtc_rx: crossbeam_channel::Receiver<Vec<u8>>,
     fps: u32,
+    enc_width: u32,
+    enc_height: u32,
     paused: Arc<AtomicBool>,
     sent_gap_tx: crossbeam_channel::Sender<f64>,
+    bitrate_tx: crossbeam_channel::Sender<BitrateCommand>,
+    resolution_tx: crossbeam_channel::Sender<BitrateCommand>,
 ) {
     let mut frames_sent: u64 = 0;
     let mut last_send: Option<std::time::Instant> = None;
+    let mut last_sent_bitrate: Option<u64> = None;
+    let initial_tier = (enc_width, enc_height);
+    let mut current_tier = initial_tier;
+    let mut upscale_counter = 0u32;
+    let mut last_resolution_eval = Instant::now();
     let timeout = Duration::from_millis(1);
 
     loop {
@@ -659,6 +712,52 @@ fn webrtc_thread_loop(
         }
         paused.store(now_paused, Ordering::Relaxed);
 
+        if let Some(bwe) = wrtc.get_bwe_estimate() {
+            let should_send = match last_sent_bitrate {
+                None => true,
+                Some(last) => {
+                    let diff = if bwe > last { bwe - last } else { last - bwe };
+                    diff * 10 > last
+                }
+            };
+            if should_send {
+                let _ = bitrate_tx.try_send(BitrateCommand::UpdateBitrate { target_bps: bwe });
+                last_sent_bitrate = Some(bwe);
+            }
+
+            if last_resolution_eval.elapsed() >= Duration::from_secs(1) {
+                last_resolution_eval = Instant::now();
+                let selected = select_resolution(current_tier.0, current_tier.1, bwe, fps);
+                if selected != current_tier {
+                    current_tier = selected;
+                    upscale_counter = 0;
+                    let _ = resolution_tx.try_send(BitrateCommand::UpdateResolution {
+                        width: current_tier.0,
+                        height: current_tier.1,
+                    });
+                    wrtc.set_need_keyframe();
+                } else if let Some(next_tier) = next_upscale_tier(current_tier, initial_tier) {
+                    let needed = resolution_bitrate_bps(next_tier.0, next_tier.1, fps);
+                    if bwe > needed.saturating_mul(120) / 100 {
+                        upscale_counter = upscale_counter.saturating_add(1);
+                        if upscale_counter >= 10 {
+                            current_tier = next_tier;
+                            upscale_counter = 0;
+                            let _ = resolution_tx.try_send(BitrateCommand::UpdateResolution {
+                                width: current_tier.0,
+                                height: current_tier.1,
+                            });
+                            wrtc.set_need_keyframe();
+                        }
+                    } else {
+                        upscale_counter = 0;
+                    }
+                } else {
+                    upscale_counter = 0;
+                }
+            }
+        }
+
         if connected {
             while let Ok(data) = webrtc_rx.try_recv() {
                 if let Err(e) = wrtc.write_h264_frame(&data, frames_sent, fps) {
@@ -700,6 +799,43 @@ fn webrtc_thread_loop(
     tracing::info!("WebRTC thread exiting");
 }
 
+const RESOLUTION_TIERS: &[(u32, u32)] = &[(2560, 1440), (1920, 1080), (1280, 720)];
+
+fn resolution_bitrate_bps(width: u32, height: u32, fps: u32) -> u64 {
+    5 * u64::from(width) * u64::from(height) * u64::from(fps) / 100
+}
+
+/// Select resolution tier based on BWE estimate.
+/// Returns (width, height) for the selected tier.
+fn select_resolution(current_w: u32, current_h: u32, bwe_bps: u64, fps: u32) -> (u32, u32) {
+    let current = (current_w, current_h);
+    let current_bitrate = resolution_bitrate_bps(current_w, current_h, fps);
+    if bwe_bps >= current_bitrate.saturating_mul(60) / 100 {
+        return current;
+    }
+
+    let current_index = RESOLUTION_TIERS
+        .iter()
+        .position(|&tier| tier == current)
+        .unwrap_or_else(|| {
+            RESOLUTION_TIERS
+                .iter()
+                .position(|&(w, h)| w <= current_w && h <= current_h)
+                .unwrap_or(RESOLUTION_TIERS.len() - 1)
+        });
+    let next_index = (current_index + 1).min(RESOLUTION_TIERS.len() - 1);
+    RESOLUTION_TIERS[next_index]
+}
+
+fn next_upscale_tier(current: (u32, u32), ceiling: (u32, u32)) -> Option<(u32, u32)> {
+    let current_index = RESOLUTION_TIERS.iter().position(|&tier| tier == current)?;
+    if current_index == 0 {
+        return None;
+    }
+    let next = RESOLUTION_TIERS[current_index - 1];
+    (next.0 <= ceiling.0 && next.1 <= ceiling.1).then_some(next)
+}
+
 impl Drop for StatePortal {
     // 析构时自动调用 shutdown，确保编码器被刷新、资源被释放
     fn drop(&mut self) {
@@ -712,8 +848,8 @@ impl Drop for StatePortal {
 /// 将原始分辨率等比缩放至不超过 2560×1440（2K），并确保宽高为偶数
 /// （H.264 编码要求偶数尺寸）。
 fn portal_encode_dimensions(width: u32, height: u32) -> (u32, u32) {
-    const TARGET_W: u32 = 2560;   // 目标最大宽度
-    const TARGET_H: u32 = 1440;   // 目标最大高度
+    const TARGET_W: u32 = 2560; // 目标最大宽度
+    const TARGET_H: u32 = 1440; // 目标最大高度
 
     // 原始分辨率已在 2K 以内，直接对齐偶数
     if width <= TARGET_W && height <= TARGET_H {
@@ -748,16 +884,16 @@ fn resolve_drm_device(args: &Args) -> Result<Option<PathBuf>> {
 #[cfg(test)]
 fn build_drm_descriptor(frame: &PwDmaBufFrame) -> ffmpeg_next::ffi::AVDRMFrameDescriptor {
     let mut desc: ffmpeg_next::ffi::AVDRMFrameDescriptor = unsafe { std::mem::zeroed() };
-    desc.nb_objects = 1;                              // 单个 DMA-BUF 对象
-    desc.objects[0].fd = frame.fd.as_raw_fd();        // DMA-BUF 文件描述符
-    desc.objects[0].size = 0;                         // 大小设为 0（内核自动确定）
+    desc.nb_objects = 1; // 单个 DMA-BUF 对象
+    desc.objects[0].fd = frame.fd.as_raw_fd(); // DMA-BUF 文件描述符
+    desc.objects[0].size = 0; // 大小设为 0（内核自动确定）
     desc.objects[0].format_modifier = frame.modifier; // DRM 格式修饰符（如线性、tiled）
-    desc.nb_layers = 1;                               // 单层
-    desc.layers[0].format = frame.format;             // 像素格式（如 XR24）
-    desc.layers[0].nb_planes = 1;                     // 单平面
-    desc.layers[0].planes[0].object_index = 0;        // 指向第 0 个对象
-    desc.layers[0].planes[0].offset = frame.offset as isize;  // 帧数据偏移
-    desc.layers[0].planes[0].pitch = frame.stride as isize;   // 行跨度（stride）
+    desc.nb_layers = 1; // 单层
+    desc.layers[0].format = frame.format; // 像素格式（如 XR24）
+    desc.layers[0].nb_planes = 1; // 单平面
+    desc.layers[0].planes[0].object_index = 0; // 指向第 0 个对象
+    desc.layers[0].planes[0].offset = frame.offset as isize; // 帧数据偏移
+    desc.layers[0].planes[0].pitch = frame.stride as isize; // 行跨度（stride）
     desc
 }
 
@@ -774,11 +910,11 @@ mod tests {
         PwDmaBufFrame {
             fd,
             offset: 0,
-            stride: 1920 * 4,          // 每行 1920 像素 × 4 字节（XRGB）
-            modifier: 0,               // DRM_FORMAT_MOD_LINEAR（线性布局）
+            stride: 1920 * 4, // 每行 1920 像素 × 4 字节（XRGB）
+            modifier: 0,      // DRM_FORMAT_MOD_LINEAR（线性布局）
             width: 1920,
             height: 1080,
-            format: 0x34325258,        // XR24 little-endian（XRGB8888）
+            format: 0x34325258, // XR24 little-endian（XRGB8888）
             pts: 12345,
         }
     }
@@ -845,14 +981,48 @@ mod tests {
         assert_eq!(result, None);
     }
 
+    #[test]
+    fn select_resolution_downscales_one_tier_below_sixty_percent() {
+        let fps = 30;
+        let current = resolution_bitrate_bps(1920, 1080, fps);
+        assert_eq!(
+            select_resolution(1920, 1080, current * 59 / 100, fps),
+            (1280, 720)
+        );
+    }
+
+    #[test]
+    fn select_resolution_keeps_tier_at_sixty_percent() {
+        let fps = 30;
+        let current = resolution_bitrate_bps(1920, 1080, fps);
+        assert_eq!(
+            select_resolution(1920, 1080, current * 60 / 100, fps),
+            (1920, 1080)
+        );
+    }
+
+    #[test]
+    fn select_resolution_never_goes_below_720p() {
+        assert_eq!(select_resolution(1280, 720, 1, 30), (1280, 720));
+    }
+
+    #[test]
+    fn next_upscale_tier_respects_initial_ceiling() {
+        assert_eq!(
+            next_upscale_tier((1280, 720), (1920, 1080)),
+            Some((1920, 1080))
+        );
+        assert_eq!(next_upscale_tier((1920, 1080), (1920, 1080)), None);
+    }
+
     /// 测试：使用自定义偏移量和 stride 构建 DRM 描述符
     #[test]
     fn build_drm_descriptor_custom_offset_and_stride() {
         let frame = PwDmaBufFrame {
             fd: unsafe { OwnedFd::from_raw_fd(libc::dup(2)) },
-            offset: 4096,                           // 4KB 对齐偏移
-            stride: 3840 * 4,                       // 4K 宽度 × 4 字节
-            modifier: 0x0100000000000001,            // AMD modifiers
+            offset: 4096,                 // 4KB 对齐偏移
+            stride: 3840 * 4,             // 4K 宽度 × 4 字节
+            modifier: 0x0100000000000001, // AMD modifiers
             width: 3840,
             height: 2160,
             format: 0x34325258,
@@ -908,4 +1078,48 @@ mod tests {
         ));
     }
 
+    // ── Task 7: Additional resolution tier edge cases ──
+
+    #[test]
+    fn select_resolution_keeps_720p_when_bwe_sufficient() {
+        let fps = 30;
+        let bitrate_720 = resolution_bitrate_bps(1280, 720, fps);
+        assert_eq!(
+            select_resolution(1280, 720, bitrate_720, fps),
+            (1280, 720)
+        );
+    }
+
+    #[test]
+    fn select_resolution_downscales_1440p_to_1080p() {
+        let fps = 30;
+        let bitrate_1440 = resolution_bitrate_bps(2560, 1440, fps);
+        assert_eq!(
+            select_resolution(2560, 1440, bitrate_1440 * 59 / 100, fps),
+            (1920, 1080)
+        );
+    }
+
+    #[test]
+    fn select_resolution_1080p_to_720p_at_very_low_bwe() {
+        let fps = 30;
+        let bitrate_1080 = resolution_bitrate_bps(1920, 1080, fps);
+        assert_eq!(
+            select_resolution(1920, 1080, bitrate_1080 / 10, fps),
+            (1280, 720)
+        );
+    }
+
+    #[test]
+    fn next_upscale_tier_from_720p_to_1080p() {
+        assert_eq!(
+            next_upscale_tier((1280, 720), (2560, 1440)),
+            Some((1920, 1080))
+        );
+    }
+
+    #[test]
+    fn next_upscale_tier_returns_none_at_highest() {
+        assert_eq!(next_upscale_tier((2560, 1440), (2560, 1440)), None);
+    }
 }

src/stats.rs

@@ -461,14 +461,20 @@ mod tests {
         // 100 values: 0.0 through 99.0
         let data: Vec<f64> = (0..100).map(|i| i as f64).collect();
         let result = p95_f64(&data);
-        assert!((result - 95.0).abs() < 1.0, "p95 of 0..100 should be ~95, got {result}");
+        assert!(
+            (result - 95.0).abs() < 1.0,
+            "p95 of 0..100 should be ~95, got {result}"
+        );
     }
 
     #[test]
     fn p95_ms_microseconds() {
         let data: Vec<u64> = (0..100).map(|i| i * 1000).collect(); // 0ms..99ms
         let result = p95_ms(&data);
-        assert!((result - 95.0).abs() < 1.0, "p95_ms should be ~95ms, got {result}");
+        assert!(
+            (result - 95.0).abs() < 1.0,
+            "p95_ms should be ~95ms, got {result}"
+        );
     }
 
     #[test]

src/webrtc.rs

@@ -4,6 +4,7 @@ use std::net::{SocketAddr, TcpListener, UdpSocket};
 use std::time::Instant;
 
 use anyhow::{bail, Result};
+use str0m::bwe::{Bitrate, BweKind};
 use str0m::change::SdpOffer;
 use str0m::format::Codec;
 use str0m::media::{Frequency, MediaKind, MediaTime, Mid, Pt};
@@ -199,6 +200,7 @@ struct WebRtcInner {
     video_pt: Option<Pt>,
     connected: bool,
     need_keyframe: bool,
+    current_bwe_estimate: Option<Bitrate>,
     rtp_clock: u32,
     buf: Vec<u8>,
 }
@@ -260,11 +262,10 @@ impl WebRtcState {
                     continue;
                 }
 
-                match WebRtcInner::new(self.fps)
-                    .and_then(|mut new_inner| {
-                        let answer_json = new_inner.handle_sdp_offer(body.as_bytes())?;
-                        Ok((new_inner, answer_json))
-                    }) {
+                match WebRtcInner::new(self.fps).and_then(|mut new_inner| {
+                    let answer_json = new_inner.handle_sdp_offer(body.as_bytes())?;
+                    Ok((new_inner, answer_json))
+                }) {
                     Ok((new_inner, answer_json)) => {
                         let replacing = self.inner.is_some();
                         self.inner = Some(new_inner);
@@ -285,7 +286,8 @@ impl WebRtcState {
                     }
                     Err(e) => {
                         tracing::error!("SDP offer handling failed: {e}");
-                        let resp = "HTTP/1.1 500 Internal Server Error\r\nConnection: close\r\n\r\n";
+                        let resp =
+                            "HTTP/1.1 500 Internal Server Error\r\nConnection: close\r\n\r\n";
                         if let Err(e) = stream.write_all(resp.as_bytes()) {
                             tracing::debug!("HTTP write error: {e}");
                         }
@@ -340,12 +342,27 @@ impl WebRtcState {
     pub fn is_connected(&self) -> bool {
         self.inner.as_ref().is_some_and(WebRtcInner::is_connected)
     }
+
+    /// Returns the latest bandwidth estimation estimate in bits per second, if available.
+    pub fn get_bwe_estimate(&self) -> Option<u64> {
+        self.inner
+            .as_ref()
+            .and_then(|inner| inner.current_bwe_estimate.map(|b| b.as_u64()))
+    }
+
+    pub fn set_need_keyframe(&mut self) {
+        if let Some(inner) = self.inner.as_mut() {
+            inner.need_keyframe = true;
+        }
+    }
 }
 
 impl WebRtcInner {
     fn new(fps: u32) -> Result<Self> {
         let _ = fps;
-        let mut rtc = RtcConfig::new().build(Instant::now());
+        let mut rtc = RtcConfig::new()
+            .enable_bwe(Some(Bitrate::mbps(5)))
+            .build(Instant::now());
 
         let socket = UdpSocket::bind("0.0.0.0:0")?;
         socket.set_nonblocking(true)?;
@@ -368,10 +385,14 @@ impl WebRtcInner {
                 std::mem::size_of::<libc::c_int>() as libc::socklen_t,
             );
             if ret < 0 {
-                tracing::warn!("setsockopt SO_SNDBUF failed (errno {})", std::io::Error::last_os_error());
+                tracing::warn!(
+                    "setsockopt SO_SNDBUF failed (errno {})",
+                    std::io::Error::last_os_error()
+                );
             }
             let mut actual: libc::c_int = 0;
-            let mut actual_len: libc::socklen_t = std::mem::size_of::<libc::c_int>() as libc::socklen_t;
+            let mut actual_len: libc::socklen_t =
+                std::mem::size_of::<libc::c_int>() as libc::socklen_t;
             let gret = libc::getsockopt(
                 fd,
                 libc::SOL_SOCKET,
@@ -408,14 +429,15 @@ impl WebRtcInner {
             video_pt: None,
             connected: false,
             need_keyframe: false,
+            current_bwe_estimate: None,
             rtp_clock: 0,
             buf: vec![0u8; 65535],
         })
     }
 
     fn handle_sdp_offer(&mut self, body: &[u8]) -> Result<String> {
-        let offer: SdpOffer = serde_json::from_slice(body)
-            .map_err(|e| anyhow::anyhow!("parse SDP offer: {e}"))?;
+        let offer: SdpOffer =
+            serde_json::from_slice(body).map_err(|e| anyhow::anyhow!("parse SDP offer: {e}"))?;
 
         let answer = self
             .rtc
@@ -492,9 +514,7 @@ impl WebRtcInner {
                             tracing::info!("Media added: mid={} kind={:?}", ma.mid, ma.kind);
                             if ma.kind == MediaKind::Video {
                                 if let Some(media) = self.rtc.media(ma.mid) {
-                                    if media.direction().is_sending()
-                                        && self.video_mid.is_none()
-                                    {
+                                    if media.direction().is_sending() && self.video_mid.is_none() {
                                         self.video_mid = Some(ma.mid);
                                         tracing::info!("Captured video mid: {}", ma.mid);
                                         self.discover_video_params();
@@ -502,6 +522,22 @@ impl WebRtcInner {
                                 }
                             }
                         }
+                        Event::KeyframeRequest(_) => {
+                            tracing::info!("received keyframe request from viewer");
+                            self.need_keyframe = true;
+                        }
+                        Event::EgressBitrateEstimate(est) => {
+                            let bitrate = match est {
+                                BweKind::Twcc(b) => *b,
+                                BweKind::Remb(_, b) => *b,
+                                _ => {
+                                    tracing::debug!("BWE estimate: unrecognized kind, skipping");
+                                    continue;
+                                }
+                            };
+                            tracing::info!("BWE estimate: {bitrate}");
+                            self.current_bwe_estimate = Some(bitrate);
+                        }
                         _ => {
                             tracing::debug!("WebRTC event: {:?}", e);
                         }
@@ -538,9 +574,9 @@ impl WebRtcInner {
                                 .map_err(|e| anyhow::anyhow!("receive contents: {e}"))?,
                         },
                     );
-                    self.rtc
-                        .handle_input(input)
-                        .map_err(|e| anyhow::anyhow!("handle_input({n} bytes from {source}): {e}"))?;
+                    self.rtc.handle_input(input).map_err(|e| {
+                        anyhow::anyhow!("handle_input({n} bytes from {source}): {e}")
+                    })?;
                 }
                 Err(ref e) if e.kind() == std::io::ErrorKind::WouldBlock => break,
                 Err(ref e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
@@ -636,18 +672,16 @@ fn extract_body(req: &str) -> &str {
 }
 
 fn local_ip() -> Option<String> {
-    std::net::UdpSocket::bind("0.0.0.0:0")
-        .ok()
-        .and_then(|s| {
-            s.connect("1.1.1.1:80").ok()?;
-            let addr = s.local_addr().ok()?;
-            drop(s);
-            let ip = addr.ip().to_string();
-            if ip == "0.0.0.0" || ip.starts_with("127.") {
-                return None;
-            }
-            Some(ip)
-        })
+    std::net::UdpSocket::bind("0.0.0.0:0").ok().and_then(|s| {
+        s.connect("1.1.1.1:80").ok()?;
+        let addr = s.local_addr().ok()?;
+        drop(s);
+        let ip = addr.ip().to_string();
+        if ip == "0.0.0.0" || ip.starts_with("127.") {
+            return None;
+        }
+        Some(ip)
+    })
 }
 
 fn is_idr_nalu(data: &[u8]) -> bool {
@@ -746,4 +780,26 @@ mod tests {
     fn all_zeros() {
         assert!(!is_idr_nalu(&[0, 0, 0, 0, 0, 0, 0, 0]));
     }
+
+    // ── Task 5: BWE estimate handling ──
+
+    #[test]
+    fn bitrate_mbps_conversion() {
+        let b = Bitrate::mbps(5);
+        assert!(b.as_u64() > 0, "5 Mbps should be > 0 bps");
+        assert_eq!(b.as_u64(), 5_000_000, "5 Mbps should be 5,000,000 bps");
+    }
+
+    #[test]
+    fn bitrate_bps_conversion() {
+        let b = Bitrate::bps(1_234_567);
+        assert_eq!(b.as_u64(), 1_234_567);
+    }
+
+    #[test]
+    fn default_bwe_is_5mbps() {
+        let default = Bitrate::mbps(5);
+        let bps = default.as_u64();
+        assert_eq!(bps, 5_000_000);
+    }
 }