Compare commits

..
14 Commits
Author SHA1 Message Date
VallierDevandClaude Opus 5.5 50ed54d77a Release v0.2.23
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-22 16:34:37 +08:00
d6752e77c8 Fix SRTP rollover counter rewinding on delayed NACK retransmissions (#83) (#84)
webrtc-srtp 0.12.0 stored every sequence number and used a 100-packet window to
guess the rollover counter. A NACK retransmission of a pre-wrap packet sent after
the wrap (65000 after 85) rewound that state, so the next new packets were
protected with ROC+1 and the viewer failed authentication on all of them: frozen
video after the Windows viewer returned from the background.

Vendor webrtc-srtp 0.12.0 with the RFC 3711 Appendix A half-space estimate and a
high-water index that retransmissions cannot move back. Regression tests fail on
upstream (65000 != 85) and pass with the fix for AES-CM/HMAC and AES-GCM.

Co-authored-by: VallierDev <[email protected]>
Co-authored-by: AYssu <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
2026-10-22 01:34:07 -07:00
VallierDevandClaude Opus 5.5 9906dd4767 Release v0.2.22
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-22 15:14:55 +08:00
Jeff Zhang 9e2198b03d docs: AMD VAAPI verified on the in-process engine; CPU cost figures (#81)
README/README.en/AGENTS: AMD iGPUs (HawkPoint, Cezanne) are accepted on the inproc-av VAAPI engine with the official
Mac viewer; the FFmpeg process pipeline on AMD is not separately verified. inproc-av now covers VAAPI in process;
build headers listed. Thanks @ferstar.
2026-10-21 22:24:42 -07:00
Jeff Zhang 21c01d2e18 inproc-av: native VAAPI H.264/HEVC through DRI3 DMA-BUF, VPP and VCN (#80)
With --features inproc-av on X11 and UU_VIDEO_ENCODER=vaapi, VAAPI runs in process: a DRI3 DMA-BUF of the root
(XShm upload without DRI3) goes through VPP to NV12 and the hardware encoder. SPS/PPS(/VPS) and slice headers are
packed by the host because radeonsi (Mesa 26) does not write them. libva/libva-drm are dlopen'ed; release builds do
not enable inproc-av. The NVENC path is unchanged.

Accepted by the contributor on AMD HawkPoint1 (Mesa 26.2.3), 1920x1200@60, with the official Mac viewer
(HEVC, presets, fps, scaling). Intel iHD and 4K are not verified. Thanks @ferstar.
2026-10-21 15:25:13 -07:00
059a3b1bb8 inproc-av: 4:4:4 on the native NVENC engine (#76, #77)
H.264 High 4:4:4 / HEVC FRExt with chromaFormatIDC 3 and AQ off when 真彩 is requested; GPUs without YUV444 encode fail the capability check and fall back. full_chroma_available() no longer excludes the in-process path. UU_PROBE_CHROMA=444 for the probe.

Co-authored-by: ferstar <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-21 08:24:32 -07:00
76941459cc inproc-av: native in-process NVENC reading XShm directly (#74, #75)
Replaces the FFmpeg-based in-process encoder with a native NVENC engine (vendored nv-codec-headers n12.1.14.0, libcuda/libnvidia-encode dlopen'ed at runtime, primary context reuse). Fallback: native NVENC -> FFmpeg NVENC process -> libx264. inproc-av no longer links FFmpeg; VAAPI/CPU always use the process pipeline. Default builds are unchanged.

Co-authored-by: ferstar <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-21 07:26:45 -07:00
1fa346fd1f device_capability: list every supported format with its size, not 0x0 (#79)
The official viewer (libstreamer, Mac 4.43) treats a video_codec_capability entry
with width or height 0 as unsupported. The 4.37 Windows host we copied in #49 ran on
a GTX 660 without HEVC/4:4:4, so its 0x0 entries meant exactly that. Evidence by
@ferstar in #77.

Co-authored-by: VallierDev <[email protected]>
Co-authored-by: ferstar <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
2026-10-21 21:29:23 +08:00
VallierDevandClaude Opus 5.5 e09778cc94 Release v0.2.21: Wayland without virtual pointer/keyboard, systemd unit docs
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-21 21:05:58 +08:00
VallierDevandClaude Opus 5.5 b1da297e42 Wayland: only a missing virtual pointer falls back to uinput (#78)
uu_wl_move returns -2 only when the compositor has no wlr-virtual-pointer; other failures are returned as before. Document UU_WAYLAND_NO_VIRTUAL_KEYBOARD.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-21 21:05:11 +08:00
ossSBandClaude Opus 5.5 c739036301 Wayland: virtual pointer and keyboard are optional; capture still starts (#72, #73)
Compositors without wlr-virtual-pointer, or that restrict the virtual keyboard to their input method (Denial), now bring up capture and fall back to uinput for input. UU_WAYLAND_NO_VIRTUAL_KEYBOARD=1 skips the virtual keyboard.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-21 16:29:19 +08:00
25893cd4ab docs: align the user systemd unit with the deb package unit (#69)
Drop the hard-coded DISPLAY=:0 and install under graphical-session.target like packaging/deb; Xfce/i3, headless and the virtual desktop keep WantedBy=default.target. AGENTS.md recipe aligned.

Co-authored-by: ferstar <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-20 22:42:11 +08:00
VallierDevandClaude Opus 5.5 5089c0343c Release v0.2.20: QR sign-in matches the official client (#68)
Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-20 17:03:13 +08:00
VallierDevandClaude Opus 5.5 5fdb98b0f5 QR login follows the official client: status_query_ticket, 60 s long poll, stop on failure (#68)
From the official Mac client (UURemote 4.42; QRCodeGenLoginResponse decodes only qrcode_jump_url
and status_query_ticket, the status/by_qrcode bodies carry that ticket under "token"):
- gen/login is sent without a body; the status ticket is status_query_ticket (older token as fallback).
- The status long poll re-asks right after each answer with a 60 s request timeout (was 12 s).
- Status 3 regenerates at once, status 0 after 0.35 s, 4 logs in (3 no longer calls by_qrcode).
- Errors such as HTTP 400 code 1190 and timeouts stop polling until the user clicks the new
  刷新二维码 button (Enter in a terminal; one minute when neither exists). This replaces the
  v0.2.19 backoff and the re-register hint.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-20 16:58:00 +08:00
67 changed files with 12045 additions and 562 deletions

No files matched your search

+2 -2
View File
@@ -94,12 +94,12 @@ jobs:
runs-on: ubuntu-24.04 runs-on: ubuntu-24.04
steps: steps:
- uses: actions/checkout@v4 - uses: actions/checkout@v4
- name: Install build dependencies (X11 + FFmpeg 6 for the inproc-av feature) - name: Install build dependencies (X11; inproc-av vendors the NVENC headers; the VAAPI/DRI3 engines need libva/gbm/xcb-dri3 headers, libva itself is dlopen'ed at runtime)
run: | run: |
sudo apt-get update sudo apt-get update
sudo apt-get install -y build-essential pkg-config \ sudo apt-get install -y build-essential pkg-config \
libx11-dev libxext-dev libxtst-dev libxfixes-dev libgtk-3-dev \ libx11-dev libxext-dev libxtst-dev libxfixes-dev libgtk-3-dev \
libavcodec-dev libavformat-dev libavdevice-dev libswscale-dev libclang-dev libva-dev libgbm-dev libx11-xcb-dev libxcb-dri3-dev libxcb-damage0-dev
- uses: dtolnay/rust-toolchain@stable - uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2 - uses: Swatinem/rust-cache@v2
- run: cargo test --locked --features inproc-av - run: cargo test --locked --features inproc-av
+28 -13
View File
@@ -38,9 +38,13 @@ Dependencies (Ubuntu/Debian):
```sh ```sh
sudo apt install build-essential pkg-config \ sudo apt install build-essential pkg-config \
libx11-dev libxext-dev libxtst-dev libxfixes-dev libavcodec-dev libavformat-dev libavdevice-dev libswscale-dev libclang-dev libgtk-3-dev ffmpeg libx11-dev libxext-dev libxtst-dev libxfixes-dev libavcodec-dev libavformat-dev libavdevice-dev libswscale-dev libclang-dev libgtk-3-dev ffmpeg
# --features inproc-av: libavcodec-dev libavformat-dev libavdevice-dev libswscale-dev libclang-dev # --features inproc-av: X11 in-process capture + native NVENC/VAAPI. NVENC headers are vendored and
# (FFmpeg 6.x only; in-process capture/encoding, Ubuntu 24.04 ships it. FFmpeg >= 7 # libcuda/libnvidia-encode dlopen'ed at runtime (NVENC API 12.1, driver >= 530).
# fails in ffmpeg-sys-next with "avfft.h file not found", before our build.rs runs) # The VAAPI/DRI3 engines compile against
# libva-dev libgbm-dev libx11-xcb-dev libxcb-dri3-dev libxcb-damage-dev;
# libva/libva-drm themselves are dlopen'ed at runtime (no hard libva dependency)
# --features viewer-ffmpeg: libavcodec-dev libavformat-dev libavdevice-dev libswscale-dev libclang-dev
# (FFmpeg 6.x only. FFmpeg >= 7 fails in ffmpeg-sys-next with "avfft.h file not found")
# --features wayland: libwayland-dev libxkbcommon-dev wl-clipboard # --features wayland: libwayland-dev libxkbcommon-dev wl-clipboard
# --features wayland,portal: + libglib2.0-dev libpipewire-0.3-dev libei-dev (libei >= 1.1) # --features wayland,portal: + libglib2.0-dev libpipewire-0.3-dev libei-dev (libei >= 1.1)
``` ```
@@ -79,17 +83,21 @@ cargo build --release --features native-viewer,viewer-sdl --bin uuremote-viewer
[Unit] [Unit]
Description=UU Remote Linux host (unofficial) Description=UU Remote Linux host (unofficial)
After=graphical-session.target After=graphical-session.target
PartOf=graphical-session.target
[Service] [Service]
Environment=RUST_LOG=info Environment=RUST_LOG=info
Environment=DISPLAY=:0
# GNOME on Xorg usually needs: Environment=XAUTHORITY=/run/user/<uid>/gdm/Xauthority
ExecStart=%h/.local/bin/uuremote-server ExecStart=%h/.local/bin/uuremote-server
Restart=always Restart=always
RestartSec=5 RestartSec=5
[Install] [Install]
WantedBy=default.target WantedBy=graphical-session.target
``` ```
Same as packaging/deb. GNOME/KDE import DISPLAY / WAYLAND_DISPLAY and start graphical-session.target; do not hard-code
DISPLAY there (it overrides the session's real one, XWayland is often :1); under default.target the unit can
start before WAYLAND_DISPLAY is imported and fall back to X11. Desktops without that target (Xfce, i3) and the Xvfb
virtual desktop need `WantedBy=default.target` plus an explicit `Environment=DISPLAY=...`; headless
(`UU_DESKTOP_BACKEND=none` + linger) needs `WantedBy=default.target` only (GNOME on Xorg outside the session: `XAUTHORITY=/run/user/<uid>/gdm/Xauthority`).
Find the real display/Xauthority with `loginctl show-session <id> -p Display` / `ps -o args -C Xorg`. Find the real display/Xauthority with `loginctl show-session <id> -p Display` / `ps -o args -C Xorg`.
5. `systemctl --user daemon-reload && systemctl --user enable --now uuremote-native.service` 5. `systemctl --user daemon-reload && systemctl --user enable --now uuremote-native.service`
6. Verify: `journalctl --user -u uuremote-native.service -f` must show 6. Verify: `journalctl --user -u uuremote-native.service -f` must show
@@ -156,7 +164,7 @@ every 60 s.
Diagnostic subcommands: `doctor` (paste-safe report: session, selected backend, engine plugins, uinput, encoders, Diagnostic subcommands: `doctor` (paste-safe report: session, selected backend, engine plugins, uinput, encoders,
default codec, sign-in/service state; booleans and names only), `login [--no-gui]`, `unlock-check` (password on stdin, prints only accepted/rejected), default codec, sign-in/service state; booleans and names only), `login [--no-gui]`, `unlock-check` (password on stdin, prints only accepted/rejected),
`desktop-probe [new.png]`, `desktop-video-probe` (`UU_PROBE_FPS`, `UU_PROBE_FRAMES`, `UU_PROBE_QUALITY` 1-5 = viewer bitrate preset, 5 auto; `UU_PROBE_INPUT` fixed video/image sequence, `UU_PROBE_LOOP=1` repeats), `wayland-input-probe`, `clipboard-acceptance`, `desktop-probe [new.png]`, `desktop-video-probe` (`UU_PROBE_FPS`, `UU_PROBE_FRAMES`, `UU_PROBE_QUALITY` 1-5 = viewer bitrate preset, 5 auto; `UU_PROBE_CHROMA=444` 真彩; `UU_PROBE_INPUT` fixed video/image sequence, `UU_PROBE_LOOP=1` repeats), `wayland-input-probe`, `clipboard-acceptance`,
`portal-acceptance`, `help`. Unknown arguments exit with code 2. Builds up to v0.1.0 did **not** reject them and `portal-acceptance`, `help`. Unknown arguments exit with code 2. Builds up to v0.1.0 did **not** reject them and
started a second full host server instead, so check that a subcommand exists in the binary you run. started a second full host server instead, so check that a subcommand exists in the binary you run.
@@ -179,7 +187,7 @@ started a second full host server instead, so check that a subcommand exists in
5 Auto, 6 Custom. Resolution: 16/17 pixel size or 5/6, 10 resolution_type (3 = follow remote). 5 Auto, 6 Custom. Resolution: 16/17 pixel size or 5/6, 10 resolution_type (3 = follow remote).
Field 7 `chorma_format` (ChromaFormat 1 = 4:2:0, 3 = 4:4:4) is sent when 真彩 is toggled; ff_capture_setting versions Field 7 `chorma_format` (ChromaFormat 1 = 4:2:0, 3 = 4:4:4) is sent when 真彩 is toggled; ff_capture_setting versions
are cumulative (3 CHROMA_SAMPLING, 4 CUSTOM_BITRATE, 5 DPI_SCALE, 6 HDR). 4:4:4 = NVENC `high444p` / x264 `high444` are cumulative (3 CHROMA_SAMPLING, 4 CUSTOM_BITRATE, 5 DPI_SCALE, 6 HDR). 4:4:4 = NVENC `high444p` / x264 `high444`
with `yuv444p`; VAAPI and inproc stay 4:2:0. with `yuv444p`; the in-process NVENC engine uses High 4:4:4 / HEVC FRExt (AQ off); VAAPI stays 4:2:0.
Reply `RpcResponse{2 CaptureSettingResponse{}}` (no errors = applied) and send Message 25 CaptureSettingSync Reply `RpcResponse{2 CaptureSettingResponse{}}` (no errors = applied) and send Message 25 CaptureSettingSync
(field 3 = the chroma actually encoded). (field 3 = the chroma actually encoded).
- ConnectOptions field 4 `decoder_cap_list` = DecoderCap {1 fps, 2 codec_type (1 H264, 2 H265), 3 width, 4 height, - ConnectOptions field 4 `decoder_cap_list` = DecoderCap {1 fps, 2 codec_type (1 H264, 2 H265), 3 width, 4 height,
@@ -271,12 +279,19 @@ started a second full host server instead, so check that a subcommand exists in
`UU_VIDEO_ENCODER=vaapi` selects Intel/AMD H.264 VAAPI; `UU_VAAPI_DEVICE` defaults to `UU_VIDEO_ENCODER=vaapi` selects Intel/AMD H.264 VAAPI; `UU_VAAPI_DEVICE` defaults to
`/dev/dri/renderD128` (render-node permissions required). No automatic VAAPI selection. `/dev/dri/renderD128` (render-node permissions required). No automatic VAAPI selection.
Both FFmpeg process and `--features inproc-av` upload NV12 frames to a VAAPI frame pool. The FFmpeg process pipeline uploads NV12 frames to a VAAPI frame pool. With `--features inproc-av` on X11, VAAPI
H.264/HEVC Main runs in process: a DRI3 DMA-BUF of the root (XShm upload without DRI3) goes through VPP to NV12
and VCN, with SPS/PPS(/VPS) and slice headers packed by the host as radeonsi (Mesa 26) needs; parameters follow
FFmpeg's hevc_vaapi/h264_vaapi on that driver. Accepted on AMD HawkPoint1 (Mesa 26.2.3) and Cezanne
(Mesa 26.2.5; XShm-upload and DRI3 zero-copy probes plus a full HEVC Mac-viewer session) with the official
Mac viewer (presets, fps, scaling); Intel iHD is not verified. Measured on Cezanne + RTX 3050 Ti
(1920x1200, desktop-video-probe, 300 frames): in-process VAAPI ~1.5 ms CPU/frame vs in-process NVENC
~7.1 ms vs libx264 process pipeline ~19 ms.
VBR uses the viewer bitrate preset and CQ-derived minimum QP, no B-frames, async depth 1, VBR uses the viewer bitrate preset and CQ-derived minimum QP, no B-frames, async depth 1,
AUD and the existing GOP/VFR timing. Unsupported device/driver/initialization falls back to libx264 AUD and the existing GOP/VFR timing. Unsupported device/driver/initialization falls back to libx264
with a warning. Accepted on Intel UHD 770 (iHD) with the default FFmpeg process pipeline and the official Mac with a warning. Accepted on Intel UHD 770 (iHD) with the default FFmpeg process pipeline and the official Mac
viewer (presets, fps, scaling, reconnect; no fallback, ~40% less CPU than libx264). AMD and the `inproc-av` viewer (presets, fps, scaling, reconnect; no fallback, ~40% less CPU than libx264). The process pipeline on
VAAPI path are not hardware-verified. AMD is not separately verified (AMD acceptance is the in-process engine above).
### Virtual audio / 虚拟声卡 ### Virtual audio / 虚拟声卡
@@ -317,8 +332,8 @@ This permission can inject into the whole seat, including lock/greeter screens.
Optional VAAPI: `intel-media-driver` (modern Intel) or `libva-mesa-driver` (AMD), `libva-utils` Optional VAAPI: `intel-media-driver` (modern Intel) or `libva-mesa-driver` (AMD), `libva-utils`
for `vainfo`. Optional KMS: `libdrm`. Virtual Xfce: `xorg-server-xvfb xorg-xauth xorg-xdpyinfo for `vainfo`. Optional KMS: `libdrm`. Virtual Xfce: `xorg-server-xvfb xorg-xauth xorg-xdpyinfo
xfce4-session xfce4-panel xfdesktop xfwm4 gnome-terminal`. xfce4-session xfce4-panel xfdesktop xfwm4 gnome-terminal`.
Rolling Arch FFmpeg may be newer than supported `inproc-av` (FFmpeg 6.x); use the default process Rolling Arch FFmpeg may be newer than supported `viewer-ffmpeg` (FFmpeg 6.x); build it against a separate
pipeline or build against a separate FFmpeg 6 prefix. Do not downgrade the system FFmpeg. FFmpeg 6 prefix. Do not downgrade the system FFmpeg. The host's `inproc-av` does not link FFmpeg.
User-bus headless setup and the official Mac checklist: [virtual desktop acceptance](contrib/virtual-desktop/ACCEPTANCE.md). User-bus headless setup and the official Mac checklist: [virtual desktop acceptance](contrib/virtual-desktop/ACCEPTANCE.md).
### Experimental KMS / 实验后端 ### Experimental KMS / 实验后端
Generated
+1 -3
View File
@@ -2741,7 +2741,7 @@ checksum = "06abde3611657adf66d383f00b093d7faecc7fa57071cce2578660c9f1010821"
[[package]] [[package]]
name = "uu-remote-linux" name = "uu-remote-linux"
version = "0.2.19" version = "0.2.23"
dependencies = [ dependencies = [
"async-trait", "async-trait",
"base64 0.22.1", "base64 0.22.1",
@@ -3075,8 +3075,6 @@ dependencies = [
[[package]] [[package]]
name = "webrtc-srtp" name = "webrtc-srtp"
version = "0.12.0" version = "0.12.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "383b0f0f73ee6cce396bdbc4d54ec661861a59eae9fc988914c1a8d82c5ac272"
dependencies = [ dependencies = [
"aead", "aead",
"aes", "aes",
+11 -4
View File
@@ -1,6 +1,6 @@
[package] [package]
name = "uu-remote-linux" name = "uu-remote-linux"
version = "0.2.19" version = "0.2.23"
edition = "2021" edition = "2021"
default-run = "uuremote-server" default-run = "uuremote-server"
authors = ["uu-remote-linux contributors"] authors = ["uu-remote-linux contributors"]
@@ -30,9 +30,9 @@ kms = []
portal = [] portal = []
# Build the wayland/portal engines as runtime-loaded libuu-wayland.so / libuu-portal.so (release packages). # Build the wayland/portal engines as runtime-loaded libuu-wayland.so / libuu-portal.so (release packages).
engine-plugins = [] engine-plugins = []
# In-process capture and encoding through FFmpeg 6 libraries (libavcodec/libavdevice dev packages). # In-process X11 capture and NVENC encoding: vendored NVENC headers, NVIDIA driver libraries loaded at runtime.
# Off by default so release binaries do not link the build machine's libav. # Other encoders keep the FFmpeg process pipeline. Off by default.
inproc-av = ["dep:ffmpeg-next", "dep:x11rb"] inproc-av = ["dep:x11rb"]
[dependencies] [dependencies]
tokio = { version = "1", features = ["full"] } tokio = { version = "1", features = ["full"] }
@@ -64,6 +64,13 @@ sdl2 = { version = "0.37", optional = true, features = ["use-pkgconfig"] }
cc = "1.0" cc = "1.0"
[patch.crates-io] [patch.crates-io]
# RTP rollover counter survives delayed NACK retransmissions (vendor/webrtc-srtp/PATCHES.md).
webrtc-srtp = { path = "vendor/webrtc-srtp" }
webrtc-dtls = { path = "vendor/webrtc-dtls" } webrtc-dtls = { path = "vendor/webrtc-dtls" }
# Larger data-channel receive buffer (see vendor/webrtc/src/data_channel/mod.rs). # Larger data-channel receive buffer (see vendor/webrtc/src/data_channel/mod.rs).
webrtc = { path = "vendor/webrtc" } webrtc = { path = "vendor/webrtc" }
[[bench]]
name = "nvenc_pipeline_bench"
harness = false
required-features = ["inproc-av"]
+27 -14
View File
@@ -117,7 +117,7 @@ Switching reuses one video track and pauses input until the new source sends a f
| Feature | Status | | Feature | Status |
|---|---| |---|---|
| Switching between multiple monitors | Experimental on native Hyprland Wayland; see above | | Switching between multiple monitors | Experimental on native Hyprland Wayland; see above |
| True color (4:4:4) | Supported (v0.1.15): the default FFmpeg pipeline with NVENC / libx264 encodes H.264 High 4:4:4; VAAPI and `inproc-av` stay 4:2:0 and report that to the client | | True color (4:4:4) | Supported (v0.1.15): the default FFmpeg pipeline with NVENC / libx264 encodes H.264 High 4:4:4; `inproc-av` native NVENC encodes H.264 High 4:4:4 / HEVC FRExt; VAAPI stays 4:2:0 and reports that to the client |
| HDR | Not implemented | | HDR | Not implemented |
| Super screen / virtual display | Not implemented | | Super screen / virtual display | Not implemented |
| Privacy screen on the host | Not implemented | | Privacy screen on the host | Not implemented |
@@ -172,6 +172,8 @@ sudo apt install build-essential pkg-config \
sudo apt install libwayland-dev libxkbcommon-dev wl-clipboard sudo apt install libwayland-dev libxkbcommon-dev wl-clipboard
# optional: GNOME/KDE portal backend # optional: GNOME/KDE portal backend
sudo apt install libglib2.0-dev libpipewire-0.3-dev libei-dev sudo apt install libglib2.0-dev libpipewire-0.3-dev libei-dev
# optional: headers for the inproc-av VAAPI/DRI3 engines (libva itself is loaded at runtime)
sudo apt install libva-dev libgbm-dev libx11-xcb-dev libxcb-dri3-dev libxcb-damage0-dev
``` ```
- A Rust toolchain is required (<https://rustup.rs>). - A Rust toolchain is required (<https://rustup.rs>).
@@ -187,10 +189,8 @@ cargo build --release --features wayland # + Hyprland
cargo build --release --features wayland,portal # + GNOME/KDE on Wayland cargo build --release --features wayland,portal # + GNOME/KDE on Wayland
``` ```
The optional `--features inproc-av` (in-process encoding) and the viewer's `viewer-ffmpeg` support **FFmpeg 6.x only** The viewer's `viewer-ffmpeg` feature supports **FFmpeg 6.x only** (`pkg-config --modversion libavcodec` = 60.x); FFmpeg 7+ fails inside `ffmpeg-sys-next` with `avfft.h file not found` (#63).
(`pkg-config --modversion libavcodec` = 60.x). FFmpeg 7+ fails inside the `ffmpeg-sys-next` dependency with The host's optional `--features inproc-av` encodes with hardware inside the process on X11. It does not link FFmpeg development libraries; it loads the NVIDIA driver libraries (`libcuda.so.1`, `libnvidia-encode.so.1`) at runtime for NVENC (needs a driver with NVENC API 12.1, 530 or newer) and `libva.so.2`/`libva-drm.so.2` for VAAPI. When the capture and encoder sizes match, NVENC reads the XShm buffer directly, saving one host-to-device copy; VAAPI runs in process under the same feature (DRI3 DMA-BUF zero copy, or XShm upload without DRI3, then VPP to NV12 and VCN; H.264/HEVC). CPU encoding and `UU_CAPTURE=grab` keep the FFmpeg process; if the in-process hardware session fails it falls back to the matching FFmpeg process pipeline, then to the CPU. The default build and the release packages need neither feature.
`avfft.h file not found`; its build script runs before this project's, so no clearer check is possible (#63). The default
build and the release packages need neither feature.
### 3. Log in ### 3. Log in
@@ -211,17 +211,17 @@ machine, in `~/.config/uuremote/config.json` (mode 0600).
[Unit] [Unit]
Description=UU Remote Linux host (unofficial) Description=UU Remote Linux host (unofficial)
After=graphical-session.target After=graphical-session.target
PartOf=graphical-session.target
[Service] [Service]
Environment=RUST_LOG=info Environment=RUST_LOG=info
# X11 desktops need the display; on GNOME, XAUTHORITY is usually /run/user/<uid>/gdm/Xauthority # GNOME/KDE import DISPLAY / WAYLAND_DISPLAY into the user manager and start graphical-session.target
Environment=DISPLAY=:0
ExecStart=%h/.local/bin/uuremote-server ExecStart=%h/.local/bin/uuremote-server
Restart=always Restart=always
RestartSec=5 RestartSec=5
[Install] [Install]
WantedBy=default.target WantedBy=graphical-session.target
``` ```
```sh ```sh
@@ -233,6 +233,10 @@ journalctl --user -u uuremote-native.service -f # online once you see availabi
If you have not logged in yet, the service opens the QR login window on your desktop. If you have not logged in yet, the service opens the QR login window on your desktop.
Desktops that never start `graphical-session.target` (e.g. Xfce, i3) will not autostart this unit: change `[Install]` to
`WantedBy=default.target`, add `Environment=DISPLAY=:0` (your actual display) and `enable` again. Headless mode and the
virtual desktop below also need `default.target`.
### 5. uinput virtual keyboard/mouse on Wayland (recommended) ### 5. uinput virtual keyboard/mouse on Wayland (recommended)
On Wayland (especially GNOME/KDE via the portal) the default input path needs portal consent and some keyboard layouts On Wayland (especially GNOME/KDE via the portal) the default input path needs portal consent and some keyboard layouts
@@ -283,7 +287,7 @@ the same account cannot assist each other, an official restriction).
## Servers without a monitor ## Servers without a monitor
**Headless mode** (terminal, file transfer and port mapping only): add `Environment=UU_DESKTOP_BACKEND=none` to the **Headless mode** (terminal, file transfer and port mapping only): add `Environment=UU_DESKTOP_BACKEND=none` to the
service and run `loginctl enable-linger $USER`, so the systemd user manager (with `XDG_RUNTIME_DIR` and the user D-Bus) service, change `[Install]` to `WantedBy=default.target` (`graphical-session.target` never starts without a login), and run `loginctl enable-linger $USER`, so the systemd user manager (with `XDG_RUNTIME_DIR` and the user D-Bus)
starts at boot without a login. The viewer's remote desktop keeps showing "connecting" in this mode; that is expected. starts at boot without a login. The viewer's remote desktop keeps showing "connecting" in this mode; that is expected.
Log in once with `uuremote-server login --no-gui` (QR in the terminal). Log in once with `uuremote-server login --no-gui` (QR in the terminal).
@@ -328,6 +332,7 @@ Common environment variables:
| `UU_DESKTOP_BACKEND` | `auto` / `x11` / `wayland` / `portal` / `kms` (experimental feature) / `none` (headless) | | `UU_DESKTOP_BACKEND` | `auto` / `x11` / `wayland` / `portal` / `kms` (experimental feature) / `none` (headless) |
| `UU_WAYLAND_MULTISCREEN=1` | Enable Hyprland physical monitor switching (off by default) | | `UU_WAYLAND_MULTISCREEN=1` | Enable Hyprland physical monitor switching (off by default) |
| `UU_WAYLAND_OUTPUT` | Initial native Wayland output name, e.g. `HDMI-A-1` | | `UU_WAYLAND_OUTPUT` | Initial native Wayland output name, e.g. `HDMI-A-1` |
| `UU_WAYLAND_NO_VIRTUAL_KEYBOARD=1` | Native Wayland skips the virtual keyboard (e.g. Denial reserves it for its input method); keys go through uinput. Without wlr-virtual-pointer the pointer also uses uinput (#72) |
| `UU_INPUT_BACKEND` | `auto` (default: uinput on Wayland when accessible) / `uinput` / `native` | | `UU_INPUT_BACKEND` | `auto` (default: uinput on Wayland when accessible) / `uinput` / `native` |
| `UU_TERMINAL_SHORTCUTS=0` | Disable the Ctrl+V paste rewrite in terminal windows | | `UU_TERMINAL_SHORTCUTS=0` | Disable the Ctrl+V paste rewrite in terminal windows |
| `UU_VIDEO_ENCODER` | `cpu`: libx264; `vaapi`: opt-in Intel/AMD; `nvenc`: force NVENC; unset: NVENC only when an NVIDIA driver is loaded, else CPU | | `UU_VIDEO_ENCODER` | `cpu`: libx264; `vaapi`: opt-in Intel/AMD; `nvenc`: force NVENC; unset: NVENC only when an NVIDIA driver is loaded, else CPU |
@@ -424,12 +429,21 @@ sends about 12 kbit/s. `UU_VIDEO_CODEC=h265` selects H.265 explicitly for A/B di
`UU_VIDEO_ENCODER=vaapi` selects Intel/AMD H.264 VAAPI; `UU_VAAPI_DEVICE` defaults to `UU_VIDEO_ENCODER=vaapi` selects Intel/AMD H.264 VAAPI; `UU_VAAPI_DEVICE` defaults to
`/dev/dri/renderD128` (render-node permissions required). No automatic VAAPI selection. `/dev/dri/renderD128` (render-node permissions required). No automatic VAAPI selection.
Both FFmpeg process and `--features inproc-av` upload NV12 frames to a VAAPI frame pool. The FFmpeg process uploads NV12 frames to a VAAPI frame pool; a self-built `--features inproc-av`
binary instead runs VAAPI in process on X11 (a DRI3 DMA-BUF of the root window zero-copy, or XShm
upload without DRI3, then VPP to NV12 and VCN; H.264/HEVC).
VBR uses the viewer bitrate preset and CQ-derived minimum QP, no B-frames, async depth 1, VBR uses the viewer bitrate preset and CQ-derived minimum QP, no B-frames, async depth 1,
AUD and the existing GOP/VFR timing. Unsupported device/driver/initialization falls back to libx264 AUD and the existing GOP/VFR timing. Unsupported device/driver/initialization falls back to libx264
with a warning. Accepted on Intel UHD 770 (iHD) with the default FFmpeg process pipeline and the official Mac with a warning. Accepted on Intel UHD 770 (iHD) with the default FFmpeg process pipeline and the official Mac
viewer (presets, fps, scaling, reconnect; no fallback, ~40% less CPU than libx264). AMD and the `inproc-av` viewer (presets, fps, scaling, reconnect; no fallback, ~40% less CPU than libx264 — a figure from the
VAAPI path are not hardware-verified. FFmpeg-process-pipeline era). AMD iGPUs are verified on
the **in-process engine** (`--features inproc-av`): end-to-end sessions with the official Mac viewer passed
acceptance on HawkPoint (Mesa 26.2.3) and Cezanne (Mesa 26.2.5) — HEVC, presets, fps, scaling; on Cezanne,
H.264 and both capture paths were additionally probe-verified: the XShm upload (decodable bitstream, correct
picture) and the DRI3 zero-copy import. Measured on one machine (1920x1200@60, Cezanne iGPU + RTX 3050 Ti):
in-process VAAPI costs about **1/12** of the per-frame encode CPU of libx264 (FFmpeg pipeline, superfast)
and about **1/5** of in-process NVENC. The FFmpeg process pipeline (what the release packages use) is not
separately verified on AMD.
The default VAAPI VBV buffer (100-500 kbit) squeezes keyframes hard at high resolution: on a 2560x1440 The default VAAPI VBV buffer (100-500 kbit) squeezes keyframes hard at high resolution: on a 2560x1440
H.264 VAAPI stream capped at 30 Mbit/s the first IDR measured 59,910 bytes and the picture blurred at H.264 VAAPI stream capped at 30 Mbit/s the first IDR measured 59,910 bytes and the picture blurred at
@@ -485,8 +499,7 @@ Keep the unit name `uuremote-native.service` (not a template `[email protected]
Optional VAAPI: `intel-media-driver` (modern Intel) or `libva-mesa-driver` (AMD), `libva-utils` Optional VAAPI: `intel-media-driver` (modern Intel) or `libva-mesa-driver` (AMD), `libva-utils`
for `vainfo`. Optional KMS: `libdrm`. Virtual Xfce: `xorg-server-xvfb xorg-xauth xorg-xdpyinfo for `vainfo`. Optional KMS: `libdrm`. Virtual Xfce: `xorg-server-xvfb xorg-xauth xorg-xdpyinfo
xfce4-session xfce4-panel xfdesktop xfwm4 gnome-terminal`. xfce4-session xfce4-panel xfdesktop xfwm4 gnome-terminal`.
Rolling Arch FFmpeg may be newer than supported `inproc-av` (FFmpeg 6.x); use the default process If building `viewer-ffmpeg`, rolling Arch FFmpeg may be newer than supported 6.x; use a separate FFmpeg 6 prefix rather than downgrading system FFmpeg (`inproc-av` native NVENC on the host does not depend on FFmpeg).
pipeline or build against a separate FFmpeg 6 prefix. Do not downgrade the system FFmpeg.
User-bus headless setup and the official Mac checklist: [virtual desktop acceptance](contrib/virtual-desktop/ACCEPTANCE.md). User-bus headless setup and the official Mac checklist: [virtual desktop acceptance](contrib/virtual-desktop/ACCEPTANCE.md).
### Experimental KMS / 实验后端 ### Experimental KMS / 实验后端
+16 -12
View File
@@ -103,7 +103,7 @@ GNOME Portal 的控制通道不再等待授权窗口或第一帧画面:先通
| 功能 | 状态 | | 功能 | 状态 |
|---|---| |---|---|
| 多显示器切换 | Hyprland 原生 Wayland 后端实验性支持,见上文 | | 多显示器切换 | Hyprland 原生 Wayland 后端实验性支持,见上文 |
| 真彩(4:4:4) | 支持(v0.1.15):NVENC / libx264 的默认 FFmpeg 管线输出 H.264 High 4:4:4;VAAPI 和 `inproc-av` 管线仍为 4:2:0,并如实回报给客户端 | | 真彩(4:4:4) | 支持(v0.1.15):NVENC / libx264 的默认 FFmpeg 管线输出 H.264 High 4:4:4;`inproc-av` 原生 NVENC 输出 H.264 High 4:4:4 / HEVC FRExt;VAAPI 管线仍为 4:2:0,并如实回报给客户端 |
| HDR | 未实现 | | HDR | 未实现 |
| 超级屏 / 虚拟显示器 | 未实现 | | 超级屏 / 虚拟显示器 | 未实现 |
| 被控端防窥模式 | 未实现 | | 被控端防窥模式 | 未实现 |
@@ -157,6 +157,8 @@ sudo apt install build-essential pkg-config \
sudo apt install libwayland-dev libxkbcommon-dev wl-clipboard sudo apt install libwayland-dev libxkbcommon-dev wl-clipboard
# 可选:GNOME/KDE portal 后端 # 可选:GNOME/KDE portal 后端
sudo apt install libglib2.0-dev libpipewire-0.3-dev libei-dev sudo apt install libglib2.0-dev libpipewire-0.3-dev libei-dev
# 可选:inproc-av 的 VAAPI/DRI3 引擎编译所需(libva 本体运行时加载)
sudo apt install libva-dev libgbm-dev libx11-xcb-dev libxcb-dri3-dev libxcb-damage0-dev
``` ```
- 需要 Rust 工具链(<https://rustup.rs>)。 - 需要 Rust 工具链(<https://rustup.rs>)。
@@ -171,9 +173,8 @@ cargo build --release --features wayland # + Hyprland
cargo build --release --features wayland,portal # + GNOME/KDE Wayland cargo build --release --features wayland,portal # + GNOME/KDE Wayland
``` ```
可选的 `--features inproc-av`(进程内编码)和控制端的 `viewer-ffmpeg` 只支持 **FFmpeg 6.x**(`pkg-config --modversion libavcodec` 为 60.x)。 控制端的 `viewer-ffmpeg` 只支持 **FFmpeg 6.x**(`pkg-config --modversion libavcodec` 为 60.x);FFmpeg 7 及以上会在依赖 `ffmpeg-sys-next` 里报 `avfft.h file not found`(#63)。
FFmpeg 7 及以上会在依赖 `ffmpeg-sys-next` 里报 `avfft.h file not found`——它的构建脚本先于本项目运行,所以无法提前给出更清楚的提示(#63)。 服务端可选的 `--features inproc-av` 是 X11 下的进程内硬件编码:不链接 FFmpeg 开发库,NVENC 运行时加载 NVIDIA 驱动库(`libcuda.so.1`、`libnvidia-encode.so.1`,需要支持 NVENC API 12.1 的驱动,530 及以上),VAAPI 运行时加载 `libva.so.2`/`libva-drm.so.2`。截屏与编码尺寸一致时 NVENC 直接读取 XShm 缓冲,省去一次主机到显存的拷贝;VAAPI 在同一特性下也进程内运行(DRI3 DMA-BUF 零拷贝,无 DRI3 时 XShm 上传,VPP 转 NV12 后送 VCN,支持 H.264/H.265)。CPU 编码或 `UU_CAPTURE=grab` 仍走外部 FFmpeg 进程;进程内硬件编码启动失败时先回退到对应的 FFmpeg 进程管线,再回退到 CPU。默认构建和预编译包都不需要该特性。
默认构建和预编译包都不需要这两个特性。
### 3. 登录 ### 3. 登录
@@ -194,17 +195,17 @@ FFmpeg 7 及以上会在依赖 `ffmpeg-sys-next` 里报 `avfft.h file not found`
[Unit] [Unit]
Description=UU Remote Linux host (unofficial) Description=UU Remote Linux host (unofficial)
After=graphical-session.target After=graphical-session.target
PartOf=graphical-session.target
[Service] [Service]
Environment=RUST_LOG=info Environment=RUST_LOG=info
# X11 桌面需要指明显示器;GNOME 的 XAUTHORITY 一般在 /run/user/<uid>/gdm/Xauthority # GNOME/KDE 等桌面会把 DISPLAY / WAYLAND_DISPLAY 导入用户管理器,并启动 graphical-session.target
Environment=DISPLAY=:0
ExecStart=%h/.local/bin/uuremote-server ExecStart=%h/.local/bin/uuremote-server
Restart=always Restart=always
RestartSec=5 RestartSec=5
[Install] [Install]
WantedBy=default.target WantedBy=graphical-session.target
``` ```
```sh ```sh
@@ -216,6 +217,9 @@ journalctl --user -u uuremote-native.service -f # 看到 availability=Some("re
如果还没有登录,服务会自动在桌面上弹出扫码窗口。 如果还没有登录,服务会自动在桌面上弹出扫码窗口。
Xfce、i3 等不启动 `graphical-session.target` 的桌面,按上面的写法开机不会自启:把 `[Install]` 改成 `WantedBy=default.target`,
并加上 `Environment=DISPLAY=:0`(按实际显示器填写),再重新 `enable`。下文的无头模式、虚拟桌面也要用 `default.target`。
### 5. Wayland 下使用 uinput 虚拟键鼠(推荐) ### 5. Wayland 下使用 uinput 虚拟键鼠(推荐)
Wayland(尤其是 GNOME/KDE portal)下,默认的键鼠注入要经过 portal 授权,部分键盘布局下还会丢符号(比如下划线)。 Wayland(尤其是 GNOME/KDE portal)下,默认的键鼠注入要经过 portal 授权,部分键盘布局下还会丢符号(比如下划线)。
@@ -255,7 +259,7 @@ install -Dm644 contrib/desktop/uu-remote-linux-viewer.desktop ~/.local/share/app
## 没有显示器的服务器 ## 没有显示器的服务器
**无头模式**:只要终端、文件传输和端口映射,不需要图形界面。在服务里加 `Environment=UU_DESKTOP_BACKEND=none`, **无头模式**:只要终端、文件传输和端口映射,不需要图形界面。在服务里加 `Environment=UU_DESKTOP_BACKEND=none`,
再执行 `loginctl enable-linger $USER`,这样开机后不用登录,systemd 用户服务(以及 `XDG_RUNTIME_DIR`、用户 D-Bus)也会自动启动。 把 `[Install]` 改成 `WantedBy=default.target`(无人登录时 `graphical-session.target` 不会启动),再执行 `loginctl enable-linger $USER`,这样开机后不用登录,systemd 用户服务(以及 `XDG_RUNTIME_DIR`、用户 D-Bus)也会自动启动。
无头模式下控制端点「远程桌面」会一直显示正在连接,这是正常的。首次登录用 `uuremote-server login --no-gui` 在终端里扫码。 无头模式下控制端点「远程桌面」会一直显示正在连接,这是正常的。首次登录用 `uuremote-server login --no-gui` 在终端里扫码。
**群晖 / NAS(Docker)**:没有桌面的 NAS 可以用容器运行无头模式(文件传输、终端、端口映射)。 **群晖 / NAS(Docker)**:没有桌面的 NAS 可以用容器运行无头模式(文件传输、终端、端口映射)。
@@ -297,6 +301,7 @@ systemctl --user restart uuremote-native.service
| `UU_DESKTOP_BACKEND` | `auto` / `x11` / `wayland` / `portal` / `kms`(实验性特性) / `none`(无头模式) | | `UU_DESKTOP_BACKEND` | `auto` / `x11` / `wayland` / `portal` / `kms`(实验性特性) / `none`(无头模式) |
| `UU_WAYLAND_MULTISCREEN=1` | 启用 Hyprland 物理显示器切换(默认关闭) | | `UU_WAYLAND_MULTISCREEN=1` | 启用 Hyprland 物理显示器切换(默认关闭) |
| `UU_WAYLAND_OUTPUT` | 原生 Wayland 默认采集输出名称,例如 `HDMI-A-1` | | `UU_WAYLAND_OUTPUT` | 原生 Wayland 默认采集输出名称,例如 `HDMI-A-1` |
| `UU_WAYLAND_NO_VIRTUAL_KEYBOARD=1` | 原生 Wayland 不创建虚拟键盘(如 Denial 只把它给自己的输入法),键盘改走 uinput。合成器没有 wlr-virtual-pointer 时鼠标也自动改走 uinput(#72) |
| `UU_INPUT_BACKEND` | `auto`(默认,Wayland 下可用时用 uinput)/ `uinput` / `native` | | `UU_INPUT_BACKEND` | `auto`(默认,Wayland 下可用时用 uinput)/ `uinput` / `native` |
| `UU_TERMINAL_SHORTCUTS=0` | 关闭终端窗口里的 Ctrl+V 粘贴转换 | | `UU_TERMINAL_SHORTCUTS=0` | 关闭终端窗口里的 Ctrl+V 粘贴转换 |
| `UU_VIDEO_ENCODER` | `cpu` 强制 CPU;`vaapi` 显式启用 Intel/AMD 硬编;`nvenc` 强制 NVENC;未设置时检测到 NVIDIA 驱动才用 NVENC,否则用 CPU | | `UU_VIDEO_ENCODER` | `cpu` 强制 CPU;`vaapi` 显式启用 Intel/AMD 硬编;`nvenc` 强制 NVENC;未设置时检测到 NVIDIA 驱动才用 NVENC,否则用 CPU |
@@ -380,9 +385,9 @@ HEVC 的访问单元按 RFC 7798 的图片边界切分,不依赖 AUD:`hevc_v
设置 `UU_VIDEO_ENCODER=vaapi` 使用 Intel/AMD H.264 VAAPI;`UU_VAAPI_DEVICE` 默认 设置 `UU_VIDEO_ENCODER=vaapi` 使用 Intel/AMD H.264 VAAPI;`UU_VAAPI_DEVICE` 默认
`/dev/dri/renderD128`,运行用户需要 render 节点权限。不会自动选择 VAAPI;设备或驱动初始化 `/dev/dri/renderD128`,运行用户需要 render 节点权限。不会自动选择 VAAPI;设备或驱动初始化
失败会明确警告并回退 libx264。外部 FFmpeg 和 `--features inproc-av` 均先转 NV12 再上传硬件帧。 失败会明确警告并回退 libx264。外部 FFmpeg 先转 NV12 再上传硬件帧。
VBR 使用控制端码率预设及 CQ 对应的最小 QP,无 B 帧、async depth=1,并保留现有 GOP/VFR 规则。 VBR 使用控制端码率预设及 CQ 对应的最小 QP,无 B 帧、async depth=1,并保留现有 GOP/VFR 规则。
Intel 核显(UHD 770,iHD 驱动)已实机验收:默认的外部 FFmpeg 管线用官方 Mac 控制端测试了各档画质、帧率、缩放和重连,全程硬件编码、无回退,CPU 占用比 libx264 低约 40%。**AMD 显卡和 `--features inproc-av` 管线的 VAAPI 尚未实机验证。** Intel 核显(UHD 770,iHD 驱动)已实机验收:默认的外部 FFmpeg 管线用官方 Mac 控制端测试了各档画质、帧率、缩放和重连,全程硬件编码、无回退,CPU 占用比 libx264 低约 40%(该数字为 FFmpeg 进程管线时代的测量)。AMD 核显已在**进程内原生引擎**上实机验收(`--features inproc-av`):官方 Mac 客户端的端到端会话在 HawkPoint(Mesa 26.2.3)与 Cezanne(Mesa 26.2.5)上通过验收(HEVC,含画质档、帧率、缩放);Cezanne 上另以探测验证了 H.264 与两条采集路径——XShm 上传(码流可解码、画面正确)与 DRI3 DMA-BUF 零拷贝导入。同机实测(1920×1200@60,Cezanne iGPU + RTX 3050 Ti 双显卡):进程内 VAAPI 每帧编码 CPU 约为 libx264(FFmpeg 管线 superfast)的 **1/12**、进程内 NVENC 的 **1/5**;外部 FFmpeg 进程管线(预编译包使用)在 AMD 上未单独验证。
高分辨率下 VAAPI 的默认 VBV 缓冲(100–500 kbit)会把关键帧压得过狠:2560×1440 H.264 VAAPI、30M 高分辨率下 VAAPI 的默认 VBV 缓冲(100–500 kbit)会把关键帧压得过狠:2560×1440 H.264 VAAPI、30M
上限、静止画面上实测首个 IDR 只有 59,910 字节,画面每隔一个 GOP 糊一下再慢慢恢复。设 上限、静止画面上实测首个 IDR 只有 59,910 字节,画面每隔一个 GOP 糊一下再慢慢恢复。设
@@ -447,8 +452,7 @@ WantedBy=graphical-session.target
VAAPI 可选包:Intel 新显卡用 `intel-media-driver`,AMD 用 `libva-mesa-driver`,`libva-utils` 提供 VAAPI 可选包:Intel 新显卡用 `intel-media-driver`,AMD 用 `libva-mesa-driver`,`libva-utils` 提供
`vainfo`。KMS 依赖 `libdrm`。虚拟 Xfce 依赖 `xorg-server-xvfb xorg-xauth xorg-xdpyinfo `vainfo`。KMS 依赖 `libdrm`。虚拟 Xfce 依赖 `xorg-server-xvfb xorg-xauth xorg-xdpyinfo
xfce4-session xfce4-panel xfdesktop xfwm4 gnome-terminal`。 xfce4-session xfce4-panel xfdesktop xfwm4 gnome-terminal`。
滚动 Arch 的 FFmpeg 可能超出 `inproc-av` 支持的 6.x;请使用默认进程方案或独立的 FFmpeg 6 前缀, 若编译包含 `viewer-ffmpeg`,滚动 Arch 的 FFmpeg 可能超出其支持的 6.x;请使用独立的 FFmpeg 6 前缀,不要降级系统 FFmpeg(服务端的 `--features inproc-av` 原生 NVENC 不依赖 FFmpeg)。
不要降级系统 FFmpeg。
### 实验性 KMS 截屏 ### 实验性 KMS 截屏
+133
View File
@@ -0,0 +1,133 @@
//! Standalone benchmark tool measuring:
//! 1. Host Memory Copy Bandwidth & CPU Cost (What FFmpeg/Sway CPU swizzling does)
//! 2. NVENC H.265 Hardware Throughput & Latency (1920x1200 @ ultra-low-latency)
//! 3. NVENC reading a registered (pinned) host buffer
use std::time::Instant;
fn main() {
println!("=========================================================");
println!(" UU Remote Linux: Encoder Pipeline Benchmark Suite ");
println!("=========================================================");
println!();
let width: u32 = 1920;
let height: u32 = 1200;
let frames: usize = 300; // 5 seconds of 60fps video
let frame_bytes: usize = (width * height * 4) as usize;
println!("Test Configuration:");
println!(" Resolution : {}x{}", width, height);
println!(" Frame Size : {:.2} MB (ARGB32)", frame_bytes as f64 / 1024.0 / 1024.0);
println!(" Iterations : {} frames (Equivalent to 5.0s @ 60 FPS)", frames);
println!();
// Simulating XShm framebuffer
let mut host_pixels = vec![0x55u8; frame_bytes];
// Benchmark 1: CPU Memcpy (External FFmpeg stdin / Sway wlr-screencopy CPU copy baseline)
println!("[1] Baseline: CPU Frame Memcpy Overhead (IPC / Process Pipe Cost)");
let mut sink = vec![0u8; frame_bytes];
let start_memcpy = Instant::now();
for _ in 0..frames {
sink.copy_from_slice(&host_pixels);
std::hint::black_box(&sink);
}
let duration_memcpy = start_memcpy.elapsed();
let memcpy_per_frame = duration_memcpy.as_secs_f64() * 1000.0 / frames as f64;
let memcpy_gbps = (frame_bytes * frames) as f64 / duration_memcpy.as_secs_f64() / 1e9;
let cpu_load_60fps = (memcpy_per_frame / (1000.0 / 60.0)) * 100.0;
println!(" -> Total CPU copy time : {:.2?}", duration_memcpy);
println!(" -> Per-frame copy time : {:.3} ms", memcpy_per_frame);
println!(" -> Memory Bus Bandwidth : {:.2} GB/s", memcpy_gbps);
println!(" -> CPU Core Saturation : {:.1}% of 1 core consumed solely by memcpy at 60 FPS", cpu_load_60fps);
println!();
// Benchmark 2: in-process NVENC reading a registered host buffer (device copy when registration fails)
#[cfg(feature = "inproc-av")]
{
extern "C" {
fn nvenc_create(config: *const NvencConfigFFI) -> *mut std::ffi::c_void;
fn nvenc_register_host_buffer(
engine: *mut std::ffi::c_void,
host_ptr: *mut std::ffi::c_void,
size: usize,
) -> i32;
fn nvenc_encode_frame(
engine: *mut std::ffi::c_void,
input_argb: *const u8,
pitch: u32,
force_idr: i32,
out_data: *mut *const u8,
out_len: *mut u32,
) -> i32;
fn nvenc_destroy(engine: *mut std::ffi::c_void);
}
// Must match NvencConfig in c_engine/nvenc_engine.h.
#[repr(C)]
struct NvencConfigFFI {
width: u32,
height: u32,
fps: u32,
gop: u32,
cq: u32,
target_bps: i64,
max_bps: i64,
buffer_bits: i64,
is_hevc: i32,
chroma_444: i32,
}
let cfg = NvencConfigFFI {
width,
height,
fps: 60,
gop: 120,
cq: 18,
target_bps: 16_000_000,
max_bps: 30_000_000,
buffer_bits: 500_000,
is_hevc: 1, // H.265 / HEVC
chroma_444: i32::from(std::env::var("UU_BENCH_CHROMA").as_deref() == Ok("444")),
};
println!("[2] In-process NVENC H.265 encoding");
unsafe {
let engine = nvenc_create(&cfg);
if engine.is_null() {
println!(" [SKIP] NVENC driver not available on this system.");
} else {
let registered = nvenc_register_host_buffer(engine, host_pixels.as_mut_ptr() as *mut _, frame_bytes) == 0;
println!(" -> Input path : {}", if registered { "registered host buffer (DMA)" } else { "cuMemcpyHtoD copy" });
let input = if registered { std::ptr::null() } else { host_pixels.as_ptr() };
let mut out_data: *const u8 = std::ptr::null();
let mut out_len: u32 = 0;
// Warmup
let _ = nvenc_encode_frame(engine, input, width * 4, 1, &mut out_data, &mut out_len);
let start = Instant::now();
let (mut encoded_bytes, mut encoded_frames) = (0u64, 0usize);
for i in 0..frames {
if nvenc_encode_frame(engine, input, width * 4, i32::from(i == 0), &mut out_data, &mut out_len) == 0 {
encoded_bytes += u64::from(out_len);
encoded_frames += 1;
}
}
let elapsed = start.elapsed();
if encoded_frames == 0 {
println!(" [FAIL] no frame was encoded");
} else {
let per_frame = elapsed.as_secs_f64() * 1000.0 / encoded_frames as f64;
println!(" -> Encoded frames : {encoded_frames}/{frames} in {elapsed:.2?}");
println!(" -> Per-frame latency : {per_frame:.3} ms ({:.1} FPS max)", 1000.0 / per_frame);
println!(" -> Output bitrate : {:.2} Mbps average", encoded_bytes as f64 * 8.0 / elapsed.as_secs_f64() / 1e6);
if registered {
println!(" -> Host-to-device copy : none ({:.0} MB/s at 60 fps avoided)", frame_bytes as f64 * 60.0 / 1024.0 / 1024.0);
}
}
nvenc_destroy(engine);
}
}
}
}
+24 -4
View File
@@ -1,6 +1,10 @@
fn main() { fn main() {
println!("cargo:rerun-if-changed=c_engine/portal_clipboard.inc"); println!("cargo:rerun-if-changed=c_engine/portal_clipboard.inc");
println!("cargo:rerun-if-changed=c_engine/x11_engine.c"); println!("cargo:rerun-if-changed=c_engine/x11_engine.c");
println!("cargo:rerun-if-changed=c_engine/nvenc_engine.c");
println!("cargo:rerun-if-changed=c_engine/nvenc_engine.h");
println!("cargo:rerun-if-changed=vendor/ffnvcodec/ffnvcodec/dynlink_cuda.h");
println!("cargo:rerun-if-changed=vendor/ffnvcodec/ffnvcodec/nvEncodeAPI.h");
println!("cargo:rerun-if-changed=c_engine/clipboard.c"); println!("cargo:rerun-if-changed=c_engine/clipboard.c");
println!("cargo:rerun-if-changed=c_engine/assist_window.c"); println!("cargo:rerun-if-changed=c_engine/assist_window.c");
println!("cargo:rerun-if-changed=c_engine/viewer_window.c"); println!("cargo:rerun-if-changed=c_engine/viewer_window.c");
@@ -21,11 +25,27 @@ fn main() {
build.compile("uu_kms_engine"); build.compile("uu_kms_engine");
println!("cargo:rustc-link-lib=drm"); println!("cargo:rustc-link-lib=drm");
} }
cc::Build::new() let mut x11_build = cc::Build::new();
.file("c_engine/x11_engine.c") x11_build.file("c_engine/x11_engine.c")
.file("c_engine/clipboard.c") .file("c_engine/clipboard.c")
.file("c_engine/pam_check.c") .file("c_engine/pam_check.c");
.compile("x11_engine"); if std::env::var_os("CARGO_FEATURE_INPROC_AV").is_some() {
x11_build.file("c_engine/nvenc_engine.c");
x11_build.file("c_engine/vaapi_engine.c");
x11_build.file("c_engine/dri3_capture.c");
for f in ["c_engine/vaapi_engine.c", "c_engine/vaapi_engine.h", "c_engine/dri3_capture.c", "c_engine/dri3_capture.h"] {
println!("cargo:rerun-if-changed={f}");
}
x11_build.include("vendor/ffnvcodec");
// libva/libva-drm are dlopen'ed at runtime (like NVENC's libcuda); these are
// link-time-only deps of the DRI3/VAAPI engines and must stay out of default builds.
println!("cargo:rustc-link-lib=gbm");
println!("cargo:rustc-link-lib=X11-xcb");
println!("cargo:rustc-link-lib=xcb");
println!("cargo:rustc-link-lib=xcb-dri3");
println!("cargo:rustc-link-lib=xcb-damage");
}
x11_build.compile("x11_engine");
println!("cargo:rustc-link-lib=X11"); println!("cargo:rustc-link-lib=X11");
println!("cargo:rustc-link-lib=Xext"); println!("cargo:rustc-link-lib=Xext");
+19 -1
View File
@@ -46,7 +46,10 @@ static pthread_cond_t init_cond = PTHREAD_COND_INITIALIZER;
static GtkWidget *window, *allow_switch, *id_label, *code_label, *eye_button, *share_button, *hint_label, *verify_combo, *note_label; static GtkWidget *window, *allow_switch, *id_label, *code_label, *eye_button, *share_button, *hint_label, *verify_combo, *note_label;
static GtkWidget *refresh_items[5], *footer_label; static GtkWidget *refresh_items[5], *footer_label;
// Account card: QR login before sign-in, user info (+ logout) after. // Account card: QR login before sign-in, user info (+ logout) after.
static GtkWidget *assist_card, *account_card, *account_stack, *qr_stack, *qr_image, *login_status, *user_label; static GtkWidget *assist_card, *account_card, *account_stack, *qr_stack, *qr_image, *login_status, *user_label, *qr_refresh;
/* Like the official client, a failed or timed-out QR stays until the user asks for a new one. */
static atomic_int qr_refresh_requested;
static void on_qr_refresh(GtkButton *b, gpointer d) { (void)b; (void)d; atomic_store(&qr_refresh_requested, 1); gtk_widget_hide(qr_refresh); }
static int show_assist = 1, show_account; static int show_assist = 1, show_account;
// Tray (host service only): AppIndicator via dlopen (no build dependency), else GtkStatusIcon. // Tray (host service only): AppIndicator via dlopen (no build dependency), else GtkStatusIcon.
static GtkWidget *tray_menu, *tray_allow; static GtkWidget *tray_menu, *tray_allow;
@@ -250,6 +253,11 @@ static GtkWidget *build_account_card(void) {
login_status = label("", "status"); login_status = label("", "status");
gtk_label_set_line_wrap(GTK_LABEL(login_status), TRUE); gtk_label_set_line_wrap(GTK_LABEL(login_status), TRUE);
gtk_box_pack_start(GTK_BOX(qr_text), login_status, FALSE, FALSE, 0); gtk_box_pack_start(GTK_BOX(qr_text), login_status, FALSE, FALSE, 0);
qr_refresh = gtk_button_new_with_label("刷新二维码");
gtk_widget_set_halign(qr_refresh, GTK_ALIGN_START);
gtk_widget_set_no_show_all(qr_refresh, TRUE);
g_signal_connect(qr_refresh, "clicked", G_CALLBACK(on_qr_refresh), NULL);
gtk_box_pack_start(GTK_BOX(qr_text), qr_refresh, FALSE, FALSE, 0);
gtk_box_pack_start(GTK_BOX(qr_page), qr_text, TRUE, TRUE, 0); gtk_box_pack_start(GTK_BOX(qr_page), qr_text, TRUE, TRUE, 0);
gtk_stack_add_named(GTK_STACK(account_stack), qr_page, "qr"); gtk_stack_add_named(GTK_STACK(account_stack), qr_page, "qr");
@@ -641,6 +649,16 @@ void uu_ui_set_user(const char *name) {
if (!ui_started || init_state != 1) return; if (!ui_started || init_state != 1) return;
g_idle_add(apply_user, g_strdup(name ? name : "")); g_idle_add(apply_user, g_strdup(name ? name : ""));
} }
static gboolean apply_qr_refresh(gpointer data) {
if (window && qr_refresh) { if (GPOINTER_TO_INT(data)) gtk_widget_show(qr_refresh); else gtk_widget_hide(qr_refresh); }
return G_SOURCE_REMOVE;
}
void uu_login_ui_offer_refresh(int visible) {
if (!ui_started || init_state != 1 || atomic_load(&closed)) return;
atomic_store(&qr_refresh_requested, 0);
g_idle_add(apply_qr_refresh, GINT_TO_POINTER(visible != 0));
}
int uu_login_ui_take_refresh(void) { return atomic_exchange(&qr_refresh_requested, 0); }
void uu_login_ui_set_done(const char *text) { uu_ui_set_user(text); } void uu_login_ui_set_done(const char *text) { uu_ui_set_user(text); }
int uu_login_ui_closed(void) { return atomic_load(&closed); } int uu_login_ui_closed(void) { return atomic_load(&closed); }
void uu_login_ui_close(void) { uu_assist_ui_close(); } void uu_login_ui_close(void) { uu_assist_ui_close(); }
+242
View File
@@ -0,0 +1,242 @@
#include "dri3_capture.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <poll.h>
#include <X11/Xlib.h>
#include <X11/Xlib-xcb.h>
#include <xcb/xcb.h>
#include <xcb/dri3.h>
#include <xcb/damage.h>
#include <gbm.h>
struct Dri3Capture {
Display *dpy;
xcb_connection_t *conn;
xcb_screen_t *screen;
uint32_t width;
uint32_t height;
int drm_fd;
struct gbm_device *gbm;
struct gbm_bo *bo;
int bo_fd;
uint32_t stride;
xcb_pixmap_t pix;
xcb_gcontext_t gc;
xcb_damage_damage_t damage;
uint8_t damage_event_code;
};
Dri3Capture* dri3_capture_create(const char* display_name, uint32_t width, uint32_t height) {
Display *dpy = XOpenDisplay(display_name);
if (!dpy) {
fprintf(stderr, "[dri3_capture] failed to open display %s\n", display_name ? display_name : "default");
return NULL;
}
// Events are read with xcb_poll_for_event below, so XCB, not Xlib, must own the queue
XSetEventQueueOwner(dpy, XCBOwnsEventQueue);
xcb_connection_t *conn = XGetXCBConnection(dpy);
const xcb_setup_t *setup = xcb_get_setup(conn);
xcb_screen_t *screen = xcb_setup_roots_iterator(setup).data;
if (!screen) {
XCloseDisplay(dpy);
return NULL;
}
// Check the extension first: a DRI3 request on a server without it breaks the connection,
// and the Xlib cleanup below then exits the process.
const xcb_query_extension_reply_t *dri3_ext = xcb_get_extension_data(conn, &xcb_dri3_id);
if (!dri3_ext || !dri3_ext->present) {
fprintf(stderr, "[dri3_capture] DRI3 extension not supported on display\n");
XCloseDisplay(dpy);
return NULL;
}
// Query DRI3
xcb_dri3_query_version_cookie_t qvc = xcb_dri3_query_version(conn, 1, 2);
xcb_dri3_query_version_reply_t *qvr = xcb_dri3_query_version_reply(conn, qvc, NULL);
if (!qvr) {
fprintf(stderr, "[dri3_capture] DRI3 extension not supported on display\n");
XCloseDisplay(dpy);
return NULL;
}
free(qvr);
// Open DRM device via DRI3
xcb_dri3_open_cookie_t oc = xcb_dri3_open(conn, screen->root, 0);
xcb_dri3_open_reply_t *or = xcb_dri3_open_reply(conn, oc, NULL);
if (!or) {
fprintf(stderr, "[dri3_capture] xcb_dri3_open failed\n");
XCloseDisplay(dpy);
return NULL;
}
int drm_fd = xcb_dri3_open_reply_fds(conn, or)[0];
free(or);
uint32_t w = width > 0 ? width : screen->width_in_pixels;
uint32_t h = height > 0 ? height : screen->height_in_pixels;
struct gbm_device *gbm = gbm_create_device(drm_fd);
if (!gbm) {
fprintf(stderr, "[dri3_capture] gbm_create_device failed\n");
close(drm_fd);
XCloseDisplay(dpy);
return NULL;
}
struct gbm_bo *bo = gbm_bo_create(gbm, w, h, GBM_FORMAT_XRGB8888, GBM_BO_USE_RENDERING | GBM_BO_USE_LINEAR);
if (!bo) {
fprintf(stderr, "[dri3_capture] gbm_bo_create failed\n");
gbm_device_destroy(gbm);
close(drm_fd);
XCloseDisplay(dpy);
return NULL;
}
int bo_fd = gbm_bo_get_fd(bo);
uint32_t stride = gbm_bo_get_stride(bo);
xcb_pixmap_t pix = xcb_generate_id(conn);
xcb_generic_error_t *err = NULL;
// XCB takes ownership of the fd it sends and closes it after transmission; only a
// duplicate goes to X11, bo_fd stays ours for the VAAPI import and teardown.
int send_fd = fcntl(bo_fd, F_DUPFD_CLOEXEC, 3);
if (send_fd < 0) {
close(bo_fd);
gbm_bo_destroy(bo);
gbm_device_destroy(gbm);
close(drm_fd);
XCloseDisplay(dpy);
return NULL;
}
xcb_void_cookie_t pc = xcb_dri3_pixmap_from_buffer_checked(
conn, pix, screen->root, stride * h, w, h, stride, screen->root_depth, 32, send_fd
);
err = xcb_request_check(conn, pc);
if (err) {
fprintf(stderr, "[dri3_capture] xcb_dri3_pixmap_from_buffer failed: %d\n", err->error_code);
free(err);
close(bo_fd);
gbm_bo_destroy(bo);
gbm_device_destroy(gbm);
close(drm_fd);
XCloseDisplay(dpy);
return NULL;
}
xcb_gcontext_t gc = xcb_generate_id(conn);
// The default ClipByChildren copies only the root background (solid black); XShmGetImage on the root
// includes child windows, so the blit must too.
uint32_t gc_mode = XCB_SUBWINDOW_MODE_INCLUDE_INFERIORS;
xcb_create_gc(conn, gc, pix, XCB_GC_SUBWINDOW_MODE, &gc_mode);
// Setup XDamage on root window
const xcb_query_extension_reply_t *ext = xcb_get_extension_data(conn, &xcb_damage_id);
uint8_t damage_event_code = 0;
xcb_damage_damage_t damage = 0;
if (ext && ext->present) {
damage_event_code = ext->first_event;
xcb_damage_query_version(conn, 1, 1);
damage = xcb_generate_id(conn);
xcb_damage_create(conn, damage, screen->root, XCB_DAMAGE_REPORT_LEVEL_NON_EMPTY);
}
Dri3Capture* cap = (Dri3Capture*)calloc(1, sizeof(Dri3Capture));
cap->dpy = dpy;
cap->conn = conn;
cap->screen = screen;
cap->width = w;
cap->height = h;
cap->drm_fd = drm_fd;
cap->gbm = gbm;
cap->bo = bo;
cap->bo_fd = bo_fd;
cap->stride = stride;
cap->pix = pix;
cap->gc = gc;
cap->damage = damage;
cap->damage_event_code = damage_event_code;
fprintf(stderr, "[dri3_capture] Initialized: %ux%u (stride: %u, dma_fd: %d) Pure GPU Blit ready!\n",
w, h, stride, bo_fd);
return cap;
}
int dri3_capture_blit(Dri3Capture* cap) {
if (!cap) return -1;
xcb_copy_area(cap->conn, cap->screen->root, cap->pix, cap->gc, 0, 0, 0, 0, cap->width, cap->height);
// Round trip: the server has executed (and flushed to the GPU) the copy before VAAPI reads the BO;
// xcb_flush alone only sends the request.
xcb_get_input_focus_reply_t *r = xcb_get_input_focus_reply(cap->conn, xcb_get_input_focus(cap->conn), NULL);
if (!r) return -1;
free(r);
return 0;
}
int dri3_capture_wait_damage(Dri3Capture* cap, int timeout_ms) {
if (!cap) return -1;
// Check queued events
xcb_generic_event_t *ev = NULL;
int damaged = 0;
while ((ev = xcb_poll_for_event(cap->conn)) != NULL) {
if ((ev->response_type & 0x7f) == (cap->damage_event_code + XCB_DAMAGE_NOTIFY)) {
damaged = 1;
}
free(ev);
}
if (damaged) {
xcb_damage_subtract(cap->conn, cap->damage, XCB_NONE, XCB_NONE);
xcb_flush(cap->conn);
return 1;
}
int fd = xcb_get_file_descriptor(cap->conn);
struct pollfd pfd = { .fd = fd, .events = POLLIN, .revents = 0 };
int ret = poll(&pfd, 1, timeout_ms);
if (ret > 0 && (pfd.revents & POLLIN)) {
// Readable but not yet a whole event: polling again would return at once and spin a core
ev = xcb_poll_for_event(cap->conn);
if (!ev) ev = xcb_wait_for_event(cap->conn);
if (!ev) return -1;
do {
if ((ev->response_type & 0x7f) == (cap->damage_event_code + XCB_DAMAGE_NOTIFY)) {
damaged = 1;
}
free(ev);
} while ((ev = xcb_poll_for_event(cap->conn)) != NULL);
if (damaged) {
xcb_damage_subtract(cap->conn, cap->damage, XCB_NONE, XCB_NONE);
xcb_flush(cap->conn);
return 1;
}
}
return ret < 0 ? -1 : 0;
}
int dri3_capture_get_dma_fd(Dri3Capture* cap) {
return cap ? cap->bo_fd : -1;
}
uint32_t dri3_capture_get_stride(Dri3Capture* cap) {
return cap ? cap->stride : 0;
}
void dri3_capture_destroy(Dri3Capture* cap) {
if (!cap) return;
if (cap->damage) xcb_damage_destroy(cap->conn, cap->damage);
if (cap->gc) xcb_free_gc(cap->conn, cap->gc);
if (cap->pix) xcb_free_pixmap(cap->conn, cap->pix);
if (cap->bo_fd >= 0) close(cap->bo_fd);
if (cap->bo) gbm_bo_destroy(cap->bo);
if (cap->gbm) gbm_device_destroy(cap->gbm);
if (cap->drm_fd >= 0) close(cap->drm_fd);
if (cap->dpy) XCloseDisplay(cap->dpy);
free(cap);
}
+33
View File
@@ -0,0 +1,33 @@
#ifndef UU_DRI3_CAPTURE_H
#define UU_DRI3_CAPTURE_H
#include <stdint.h>
#include <stddef.h>
#ifdef __cplusplus
extern "C" {
#endif
typedef struct Dri3Capture Dri3Capture;
Dri3Capture* dri3_capture_create(const char* display_name, uint32_t width, uint32_t height);
// Trigger GPU blit from Root Window to the DRI3 DMA-BUF Pixmap.
// Returns 0 on success, negative on error.
int dri3_capture_blit(Dri3Capture* cap);
// Wait for next XDamage event or timeout (ms).
// Returns 1 if damaged, 0 if timeout, negative on error.
int dri3_capture_wait_damage(Dri3Capture* cap, int timeout_ms);
// Get the exported DMA-BUF file descriptor and stride
int dri3_capture_get_dma_fd(Dri3Capture* cap);
uint32_t dri3_capture_get_stride(Dri3Capture* cap);
void dri3_capture_destroy(Dri3Capture* cap);
#ifdef __cplusplus
}
#endif
#endif // UU_DRI3_CAPTURE_H
+580
View File
@@ -0,0 +1,580 @@
#include "nvenc_engine.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <dlfcn.h>
#include <ffnvcodec/dynlink_cuda.h>
#include <ffnvcodec/nvEncodeAPI.h>
// Struct revisions come from the vendored nvEncodeAPI.h (*_VER macros), so the session must be
// opened with that header's NVENCAPI_VERSION; older drivers are rejected in nvenc_create.
#define CU_MEMHOSTREGISTER_DEVICEMAP 0x02
typedef CUresult CUDAAPI (*tcuMemHostRegister)(void *p, size_t bytesize, unsigned int Flags);
typedef CUresult CUDAAPI (*tcuMemHostGetDevicePointer)(CUdeviceptr *pdptr, void *p, unsigned int Flags);
typedef CUresult CUDAAPI (*tcuMemHostUnregister)(void *p);
typedef NVENCSTATUS NVENCAPI (*tNvEncodeAPICreateInstance)(NV_ENCODE_API_FUNCTION_LIST *functionList);
typedef NVENCSTATUS NVENCAPI (*tNvEncodeAPIGetMaxSupportedVersion)(uint32_t *version);
struct NvencEngine {
void *cuda_lib;
void *nvenc_lib;
tcuInit *cuInit;
tcuDeviceGet *cuDeviceGet;
tcuDevicePrimaryCtxRetain *cuDevicePrimaryCtxRetain;
tcuDevicePrimaryCtxRelease *cuDevicePrimaryCtxRelease;
tcuCtxCreate_v2 *cuCtxCreate;
tcuCtxDestroy_v2 *cuCtxDestroy;
tcuCtxPushCurrent_v2 *cuCtxPushCurrent;
tcuCtxPopCurrent_v2 *cuCtxPopCurrent;
tcuMemAlloc_v2 *cuMemAlloc;
tcuMemFree_v2 *cuMemFree;
tcuMemcpyHtoD_v2 *cuMemcpyHtoD;
tcuMemHostRegister cuMemHostRegister;
tcuMemHostGetDevicePointer cuMemHostGetDevicePointer;
tcuMemHostUnregister cuMemHostUnregister;
CUdevice dev;
CUcontext cu_ctx;
int is_standalone_ctx; // cuCtxCreate fallback (primary context unavailable, seen on GB10)
CUdeviceptr cu_dptr;
size_t frame_bytes;
// Pinned XShm buffer registered for direct DMA reads
void *registered_host_ptr;
int is_zero_copy;
NV_ENCODE_API_FUNCTION_LIST fn;
void *encoder;
NV_ENC_REGISTER_RESOURCE reg_res;
NV_ENC_OUTPUT_PTR bitstream_buf;
uint32_t width;
uint32_t height;
int is_hevc;
uint8_t *output_staging;
size_t output_staging_cap;
};
// Returns 1 when the context was pushed; pop only in that case.
static int ctx_push(NvencEngine *engine) {
return engine->cuCtxPushCurrent && engine->cu_ctx && engine->cuCtxPushCurrent(engine->cu_ctx) == CUDA_SUCCESS;
}
static void ctx_pop(NvencEngine *engine, int pushed) {
if (pushed && engine->cuCtxPopCurrent) {
CUcontext dummy = NULL;
engine->cuCtxPopCurrent(&dummy);
}
}
NvencEngine* nvenc_create(const NvencConfig* config) {
if (!config || config->width == 0 || config->height == 0) {
return NULL;
}
NvencEngine *engine = (NvencEngine*)calloc(1, sizeof(NvencEngine));
if (!engine) return NULL;
engine->width = config->width;
engine->height = config->height;
engine->is_hevc = config->is_hevc;
// 1. Load CUDA driver
engine->cuda_lib = dlopen("libcuda.so.1", RTLD_NOW);
if (!engine->cuda_lib) {
fprintf(stderr, "[nvenc] dlopen libcuda.so.1 failed\n");
free(engine);
return NULL;
}
engine->cuInit = (tcuInit*)dlsym(engine->cuda_lib, "cuInit");
engine->cuDeviceGet = (tcuDeviceGet*)dlsym(engine->cuda_lib, "cuDeviceGet");
engine->cuDevicePrimaryCtxRetain = (tcuDevicePrimaryCtxRetain*)dlsym(engine->cuda_lib, "cuDevicePrimaryCtxRetain");
engine->cuDevicePrimaryCtxRelease = (tcuDevicePrimaryCtxRelease*)dlsym(engine->cuda_lib, "cuDevicePrimaryCtxRelease_v2");
if (!engine->cuDevicePrimaryCtxRelease)
engine->cuDevicePrimaryCtxRelease = (tcuDevicePrimaryCtxRelease*)dlsym(engine->cuda_lib, "cuDevicePrimaryCtxRelease");
engine->cuCtxCreate = (tcuCtxCreate_v2*)dlsym(engine->cuda_lib, "cuCtxCreate_v2");
engine->cuCtxDestroy = (tcuCtxDestroy_v2*)dlsym(engine->cuda_lib, "cuCtxDestroy_v2");
engine->cuCtxPushCurrent = (tcuCtxPushCurrent_v2*)dlsym(engine->cuda_lib, "cuCtxPushCurrent_v2");
engine->cuCtxPopCurrent = (tcuCtxPopCurrent_v2*)dlsym(engine->cuda_lib, "cuCtxPopCurrent_v2");
engine->cuMemAlloc = (tcuMemAlloc_v2*)dlsym(engine->cuda_lib, "cuMemAlloc_v2");
engine->cuMemFree = (tcuMemFree_v2*)dlsym(engine->cuda_lib, "cuMemFree_v2");
engine->cuMemcpyHtoD = (tcuMemcpyHtoD_v2*)dlsym(engine->cuda_lib, "cuMemcpyHtoD_v2");
engine->cuMemHostRegister = (tcuMemHostRegister)dlsym(engine->cuda_lib, "cuMemHostRegister_v2");
engine->cuMemHostGetDevicePointer = (tcuMemHostGetDevicePointer)dlsym(engine->cuda_lib, "cuMemHostGetDevicePointer_v2");
engine->cuMemHostUnregister = (tcuMemHostUnregister)dlsym(engine->cuda_lib, "cuMemHostUnregister");
if (!engine->cuInit || !engine->cuDeviceGet || !engine->cuDevicePrimaryCtxRetain ||
!engine->cuDevicePrimaryCtxRelease || !engine->cuCtxPushCurrent || !engine->cuCtxPopCurrent ||
!engine->cuMemAlloc || !engine->cuMemFree || !engine->cuMemcpyHtoD) {
fprintf(stderr, "[nvenc] Missing CUDA driver symbols\n");
nvenc_destroy(engine);
return NULL;
}
CUresult cures = engine->cuInit(0);
if (cures != CUDA_SUCCESS) {
fprintf(stderr, "[nvenc] cuInit failed: %d\n", cures);
nvenc_destroy(engine);
return NULL;
}
cures = engine->cuDeviceGet(&engine->dev, 0);
if (cures != CUDA_SUCCESS) {
fprintf(stderr, "[nvenc] cuDeviceGet failed: %d\n", cures);
nvenc_destroy(engine);
return NULL;
}
cures = engine->cuDevicePrimaryCtxRetain(&engine->cu_ctx, engine->dev);
if ((cures != CUDA_SUCCESS || !engine->cu_ctx) && engine->cuCtxCreate && engine->cuCtxDestroy) {
engine->cu_ctx = NULL;
// cuCtxCreate makes the new context current; the push below makes it current again
cures = engine->cuCtxCreate(&engine->cu_ctx, 0, engine->dev);
if (cures == CUDA_SUCCESS && engine->cu_ctx) {
CUcontext dummy = NULL;
engine->cuCtxPopCurrent(&dummy);
engine->is_standalone_ctx = 1;
}
}
if (cures != CUDA_SUCCESS || !engine->cu_ctx) {
fprintf(stderr, "[nvenc] CUDA context creation failed: %d\n", cures);
engine->cu_ctx = NULL;
nvenc_destroy(engine);
return NULL;
}
int pushed = ctx_push(engine);
if (!pushed) {
fprintf(stderr, "[nvenc] cuCtxPushCurrent failed\n");
nvenc_destroy(engine);
return NULL;
}
// 2. Load NVENC
engine->nvenc_lib = dlopen("libnvidia-encode.so.1", RTLD_NOW);
if (!engine->nvenc_lib) {
fprintf(stderr, "[nvenc] dlopen libnvidia-encode.so.1 failed\n");
goto fail;
}
// Like FFmpeg: the driver must support the header's API version, otherwise fall back.
tNvEncodeAPIGetMaxSupportedVersion get_max_ver = (tNvEncodeAPIGetMaxSupportedVersion)dlsym(
engine->nvenc_lib, "NvEncodeAPIGetMaxSupportedVersion");
uint32_t max_supported_ver = 0;
if (!get_max_ver || get_max_ver(&max_supported_ver) != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] NvEncodeAPIGetMaxSupportedVersion failed\n");
goto fail;
}
if (max_supported_ver < ((NVENCAPI_MAJOR_VERSION << 4) | NVENCAPI_MINOR_VERSION)) {
fprintf(stderr, "[nvenc] Driver NVENC API %u.%u is older than required %d.%d; update the NVIDIA driver\n",
max_supported_ver >> 4, max_supported_ver & 0xf, NVENCAPI_MAJOR_VERSION, NVENCAPI_MINOR_VERSION);
goto fail;
}
tNvEncodeAPICreateInstance create_inst = (tNvEncodeAPICreateInstance)dlsym(engine->nvenc_lib, "NvEncodeAPICreateInstance");
if (!create_inst) {
fprintf(stderr, "[nvenc] dlsym NvEncodeAPICreateInstance failed\n");
goto fail;
}
memset(&engine->fn, 0, sizeof(engine->fn));
engine->fn.version = NV_ENCODE_API_FUNCTION_LIST_VER;
NVENCSTATUS nvres = create_inst(&engine->fn);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] NvEncodeAPICreateInstance failed: %d\n", nvres);
goto fail;
}
// 3. Open Encode Session
NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS open_params;
memset(&open_params, 0, sizeof(open_params));
open_params.version = NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS_VER;
open_params.deviceType = NV_ENC_DEVICE_TYPE_CUDA;
open_params.device = engine->cu_ctx;
open_params.apiVersion = NVENCAPI_VERSION;
nvres = engine->fn.nvEncOpenEncodeSessionEx(&open_params, &engine->encoder);
if (nvres != NV_ENC_SUCCESS || !engine->encoder) {
fprintf(stderr, "[nvenc] nvEncOpenEncodeSessionEx failed: %d\n", nvres);
engine->encoder = NULL;
goto fail;
}
// 4. Query Preset Config
GUID codec_guid = engine->is_hevc ? NV_ENC_CODEC_HEVC_GUID : NV_ENC_CODEC_H264_GUID;
NV_ENC_PRESET_CONFIG preset_cfg;
memset(&preset_cfg, 0, sizeof(preset_cfg));
preset_cfg.version = NV_ENC_PRESET_CONFIG_VER;
preset_cfg.presetCfg.version = NV_ENC_CONFIG_VER;
nvres = engine->fn.nvEncGetEncodePresetConfigEx(
engine->encoder,
codec_guid,
NV_ENC_PRESET_P4_GUID,
NV_ENC_TUNING_INFO_ULTRA_LOW_LATENCY,
&preset_cfg
);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncGetEncodePresetConfigEx failed: %d\n", nvres);
goto fail;
}
NV_ENC_CONFIG enc_cfg;
memcpy(&enc_cfg, &preset_cfg.presetCfg, sizeof(NV_ENC_CONFIG));
enc_cfg.gopLength = config->gop > 0 ? config->gop : NVENC_INFINITE_GOPLENGTH;
enc_cfg.frameIntervalP = 1; // Strict low-latency: zero B-frames
enc_cfg.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR;
enc_cfg.rcParams.averageBitRate = (uint32_t)(config->target_bps > 0 ? config->target_bps : 10000000);
enc_cfg.rcParams.maxBitRate = (uint32_t)(config->max_bps > 0 ? config->max_bps : 15000000);
enc_cfg.rcParams.vbvBufferSize = (uint32_t)(config->buffer_bits > 0 ? config->buffer_bits : (config->max_bps > 0 ? config->max_bps / 10 : 1500000));
enc_cfg.rcParams.targetQuality = (uint8_t)(config->cq > 0 ? config->cq : 28);
NV_ENC_CAPS_PARAM caps_param;
memset(&caps_param, 0, sizeof(caps_param));
caps_param.version = NV_ENC_CAPS_PARAM_VER;
int caps_val = 0;
if (config->chroma_444) {
// GPUs without 4:4:4 encode fail here; the caller falls back to FFmpeg/libx264, which keep 4:4:4.
caps_param.capsToQuery = NV_ENC_CAPS_SUPPORT_YUV444_ENCODE;
if (engine->fn.nvEncGetEncodeCaps(engine->encoder, codec_guid, &caps_param, &caps_val) != NV_ENC_SUCCESS || caps_val <= 0) {
fprintf(stderr, "[nvenc] 4:4:4 encoding not supported for this codec\n");
goto fail;
}
// NVENC rejects AQ / temporal AQ in 4:4:4 mode.
enc_cfg.rcParams.enableAQ = 0;
enc_cfg.rcParams.enableTemporalAQ = 0;
} else {
enc_cfg.rcParams.enableAQ = 1;
// Temporal AQ only where the hardware/codec reports it
caps_param.capsToQuery = NV_ENC_CAPS_SUPPORT_TEMPORAL_AQ;
caps_val = 0;
enc_cfg.rcParams.enableTemporalAQ =
engine->fn.nvEncGetEncodeCaps(engine->encoder, codec_guid, &caps_param, &caps_val) == NV_ENC_SUCCESS && caps_val > 0;
}
if (engine->is_hevc) {
enc_cfg.profileGUID = config->chroma_444 ? NV_ENC_HEVC_PROFILE_FREXT_GUID : NV_ENC_HEVC_PROFILE_MAIN_GUID;
enc_cfg.encodeCodecConfig.hevcConfig.chromaFormatIDC = config->chroma_444 ? 3 : 1;
enc_cfg.encodeCodecConfig.hevcConfig.idrPeriod = enc_cfg.gopLength;
enc_cfg.encodeCodecConfig.hevcConfig.repeatSPSPPS = 1;
enc_cfg.encodeCodecConfig.hevcConfig.outputAUD = 1;
} else {
enc_cfg.profileGUID = config->chroma_444 ? NV_ENC_H264_PROFILE_HIGH_444_GUID : NV_ENC_H264_PROFILE_BASELINE_GUID;
enc_cfg.encodeCodecConfig.h264Config.chromaFormatIDC = config->chroma_444 ? 3 : 1;
enc_cfg.encodeCodecConfig.h264Config.idrPeriod = enc_cfg.gopLength;
enc_cfg.encodeCodecConfig.h264Config.repeatSPSPPS = 1;
enc_cfg.encodeCodecConfig.h264Config.outputAUD = 1;
}
// 5. Initialize Encoder
NV_ENC_INITIALIZE_PARAMS init_params;
memset(&init_params, 0, sizeof(init_params));
init_params.version = NV_ENC_INITIALIZE_PARAMS_VER;
init_params.encodeGUID = codec_guid;
init_params.presetGUID = NV_ENC_PRESET_P4_GUID;
init_params.encodeWidth = engine->width;
init_params.encodeHeight = engine->height;
init_params.darWidth = engine->width;
init_params.darHeight = engine->height;
init_params.frameRateNum = config->fps > 0 ? config->fps : 60;
init_params.frameRateDen = 1;
init_params.enablePTD = 1;
init_params.tuningInfo = NV_ENC_TUNING_INFO_ULTRA_LOW_LATENCY;
init_params.encodeConfig = &enc_cfg;
nvres = engine->fn.nvEncInitializeEncoder(engine->encoder, &init_params);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncInitializeEncoder failed: %d\n", nvres);
goto fail;
}
// 6. Device input buffer (ARGB32), used until a host buffer is registered
engine->frame_bytes = (size_t)engine->width * engine->height * 4;
cures = engine->cuMemAlloc(&engine->cu_dptr, engine->frame_bytes);
if (cures != CUDA_SUCCESS) {
fprintf(stderr, "[nvenc] cuMemAlloc failed: %d\n", cures);
engine->cu_dptr = 0;
goto fail;
}
memset(&engine->reg_res, 0, sizeof(engine->reg_res));
engine->reg_res.version = NV_ENC_REGISTER_RESOURCE_VER;
engine->reg_res.resourceType = NV_ENC_INPUT_RESOURCE_TYPE_CUDADEVICEPTR;
engine->reg_res.width = engine->width;
engine->reg_res.height = engine->height;
engine->reg_res.pitch = engine->width * 4;
engine->reg_res.resourceToRegister = (void*)engine->cu_dptr;
engine->reg_res.bufferFormat = NV_ENC_BUFFER_FORMAT_ARGB;
nvres = engine->fn.nvEncRegisterResource(engine->encoder, &engine->reg_res);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncRegisterResource failed: %d\n", nvres);
engine->reg_res.registeredResource = NULL;
goto fail;
}
// 7. Create bitstream output buffer
NV_ENC_CREATE_BITSTREAM_BUFFER bs_buf;
memset(&bs_buf, 0, sizeof(bs_buf));
bs_buf.version = NV_ENC_CREATE_BITSTREAM_BUFFER_VER;
nvres = engine->fn.nvEncCreateBitstreamBuffer(engine->encoder, &bs_buf);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncCreateBitstreamBuffer failed: %d\n", nvres);
goto fail;
}
engine->bitstream_buf = bs_buf.bitstreamBuffer;
engine->output_staging_cap = 2 * 1024 * 1024;
engine->output_staging = (uint8_t*)malloc(engine->output_staging_cap);
if (!engine->output_staging) {
fprintf(stderr, "[nvenc] output buffer allocation failed\n");
goto fail;
}
ctx_pop(engine, pushed);
return engine;
fail:
// Pop the creation push first; nvenc_destroy pushes and pops on its own.
ctx_pop(engine, pushed);
nvenc_destroy(engine);
return NULL;
}
int nvenc_register_host_buffer(NvencEngine* engine, void* host_ptr, size_t size) {
if (!engine || !host_ptr || size < engine->frame_bytes) {
return -1;
}
if (!engine->cuMemHostRegister || !engine->cuMemHostGetDevicePointer || !engine->cuMemHostUnregister) {
return -2;
}
int pushed = ctx_push(engine);
if (!pushed) return -2;
// If already registered, unregister first
nvenc_unregister_host_buffer(engine);
CUresult cures = engine->cuMemHostRegister(host_ptr, size, CU_MEMHOSTREGISTER_DEVICEMAP);
if (cures != CUDA_SUCCESS) {
fprintf(stderr, "[nvenc] cuMemHostRegister failed: %d\n", cures);
ctx_pop(engine, pushed);
return -3;
}
CUdeviceptr dev_ptr = 0;
cures = engine->cuMemHostGetDevicePointer(&dev_ptr, host_ptr, 0);
if (cures != CUDA_SUCCESS || dev_ptr == 0) {
fprintf(stderr, "[nvenc] cuMemHostGetDevicePointer failed: %d\n", cures);
engine->cuMemHostUnregister(host_ptr);
ctx_pop(engine, pushed);
return -4;
}
// Register the pinned buffer before releasing the device buffer, so a failure keeps the copy path working
NV_ENC_REGISTER_RESOURCE new_reg;
memset(&new_reg, 0, sizeof(new_reg));
new_reg.version = NV_ENC_REGISTER_RESOURCE_VER;
new_reg.resourceType = NV_ENC_INPUT_RESOURCE_TYPE_CUDADEVICEPTR;
new_reg.width = engine->width;
new_reg.height = engine->height;
new_reg.pitch = engine->width * 4;
new_reg.resourceToRegister = (void*)dev_ptr;
new_reg.bufferFormat = NV_ENC_BUFFER_FORMAT_ARGB;
NVENCSTATUS nvres = engine->fn.nvEncRegisterResource(engine->encoder, &new_reg);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncRegisterResource (pinned host) failed: %d\n", nvres);
engine->cuMemHostUnregister(host_ptr);
ctx_pop(engine, pushed);
return -5;
}
if (engine->reg_res.registeredResource) {
engine->fn.nvEncUnregisterResource(engine->encoder, engine->reg_res.registeredResource);
}
if (engine->cu_dptr) {
engine->cuMemFree(engine->cu_dptr);
engine->cu_dptr = 0;
}
engine->reg_res = new_reg;
engine->registered_host_ptr = host_ptr;
engine->is_zero_copy = 1;
ctx_pop(engine, pushed);
return 0;
}
// Caller has the context current.
void nvenc_unregister_host_buffer(NvencEngine* engine) {
if (!engine || !engine->is_zero_copy) return;
if (engine->reg_res.registeredResource && engine->encoder) {
engine->fn.nvEncUnregisterResource(engine->encoder, engine->reg_res.registeredResource);
memset(&engine->reg_res, 0, sizeof(engine->reg_res));
}
if (engine->registered_host_ptr && engine->cuMemHostUnregister) {
engine->cuMemHostUnregister(engine->registered_host_ptr);
engine->registered_host_ptr = NULL;
}
engine->is_zero_copy = 0;
}
int nvenc_encode_frame(
NvencEngine* engine,
const uint8_t* input_argb,
uint32_t pitch,
int force_idr,
const uint8_t** out_data,
uint32_t* out_len
) {
if (!engine || !out_data || !out_len || !engine->reg_res.registeredResource) {
return -1;
}
*out_data = NULL;
*out_len = 0;
int pushed = ctx_push(engine);
if (!pushed) return -1;
// Without a registered host buffer, copy the frame into the device buffer
if (!engine->is_zero_copy) {
if (!input_argb) {
ctx_pop(engine, pushed);
return -1;
}
CUresult cures = CUDA_SUCCESS;
if (pitch == engine->width * 4) {
cures = engine->cuMemcpyHtoD(engine->cu_dptr, input_argb, engine->frame_bytes);
} else {
for (uint32_t y = 0; y < engine->height && cures == CUDA_SUCCESS; y++) {
cures = engine->cuMemcpyHtoD(
engine->cu_dptr + (size_t)y * engine->width * 4,
input_argb + (size_t)y * pitch,
engine->width * 4
);
}
}
if (cures != CUDA_SUCCESS) {
fprintf(stderr, "[nvenc] cuMemcpyHtoD failed: %d\n", cures);
ctx_pop(engine, pushed);
return -2;
}
}
NV_ENC_MAP_INPUT_RESOURCE map_res;
memset(&map_res, 0, sizeof(map_res));
map_res.version = NV_ENC_MAP_INPUT_RESOURCE_VER;
map_res.registeredResource = engine->reg_res.registeredResource;
NVENCSTATUS nvres = engine->fn.nvEncMapInputResource(engine->encoder, &map_res);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncMapInputResource failed: %d\n", nvres);
ctx_pop(engine, pushed);
return -3;
}
NV_ENC_PIC_PARAMS pic_params;
memset(&pic_params, 0, sizeof(pic_params));
pic_params.version = NV_ENC_PIC_PARAMS_VER;
pic_params.inputWidth = engine->width;
pic_params.inputHeight = engine->height;
pic_params.inputPitch = engine->width * 4;
pic_params.inputBuffer = map_res.mappedResource;
pic_params.outputBitstream = engine->bitstream_buf;
pic_params.bufferFmt = NV_ENC_BUFFER_FORMAT_ARGB;
pic_params.pictureStruct = NV_ENC_PIC_STRUCT_FRAME;
if (force_idr) {
pic_params.encodePicFlags = NV_ENC_PIC_FLAG_FORCEIDR;
}
nvres = engine->fn.nvEncEncodePicture(engine->encoder, &pic_params);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncEncodePicture failed: %d\n", nvres);
engine->fn.nvEncUnmapInputResource(engine->encoder, map_res.mappedResource);
ctx_pop(engine, pushed);
return -4;
}
NV_ENC_LOCK_BITSTREAM lock_bs;
memset(&lock_bs, 0, sizeof(lock_bs));
lock_bs.version = NV_ENC_LOCK_BITSTREAM_VER;
lock_bs.outputBitstream = engine->bitstream_buf;
lock_bs.doNotWait = 0;
nvres = engine->fn.nvEncLockBitstream(engine->encoder, &lock_bs);
if (nvres != NV_ENC_SUCCESS) {
fprintf(stderr, "[nvenc] nvEncLockBitstream failed: %d\n", nvres);
engine->fn.nvEncUnmapInputResource(engine->encoder, map_res.mappedResource);
ctx_pop(engine, pushed);
return -5;
}
int rc = 0;
uint32_t size = lock_bs.bitstreamSizeInBytes;
if (size > engine->output_staging_cap) {
size_t cap = (size_t)size + 64 * 1024;
uint8_t *grown = (uint8_t*)realloc(engine->output_staging, cap);
if (grown) {
engine->output_staging = grown;
engine->output_staging_cap = cap;
} else {
fprintf(stderr, "[nvenc] output buffer allocation failed: %u bytes\n", size);
rc = -6;
}
}
if (rc == 0 && size > 0 && lock_bs.bitstreamBufferPtr) {
memcpy(engine->output_staging, lock_bs.bitstreamBufferPtr, size);
*out_data = engine->output_staging;
*out_len = size;
}
engine->fn.nvEncUnlockBitstream(engine->encoder, engine->bitstream_buf);
// The SDK allows unmapping only after the bitstream lock returned for this picture
engine->fn.nvEncUnmapInputResource(engine->encoder, map_res.mappedResource);
ctx_pop(engine, pushed);
return rc;
}
void nvenc_destroy(NvencEngine* engine) {
if (!engine) return;
int pushed = ctx_push(engine);
// Without a current context the NVENC/CUDA objects cannot be released safely; leak them.
if (pushed) {
if (engine->bitstream_buf && engine->encoder) {
engine->fn.nvEncDestroyBitstreamBuffer(engine->encoder, engine->bitstream_buf);
engine->bitstream_buf = NULL;
}
nvenc_unregister_host_buffer(engine);
if (engine->reg_res.registeredResource && engine->encoder) {
engine->fn.nvEncUnregisterResource(engine->encoder, engine->reg_res.registeredResource);
memset(&engine->reg_res, 0, sizeof(engine->reg_res));
}
if (engine->cu_dptr) {
engine->cuMemFree(engine->cu_dptr);
engine->cu_dptr = 0;
}
if (engine->encoder) {
engine->fn.nvEncDestroyEncoder(engine->encoder);
engine->encoder = NULL;
}
ctx_pop(engine, pushed);
}
if (engine->cu_ctx) {
if (engine->is_standalone_ctx) engine->cuCtxDestroy(engine->cu_ctx);
else engine->cuDevicePrimaryCtxRelease(engine->dev);
engine->cu_ctx = NULL;
}
free(engine->output_staging);
// libcuda/libnvidia-encode stay loaded: their driver threads outlive the session, and
// dlclose crashed them on the next encoder start (SEGV_MAPERR in libnvcuvid).
free(engine);
}
+59
View File
@@ -0,0 +1,59 @@
#ifndef UU_NVENC_ENGINE_H
#define UU_NVENC_ENGINE_H
#include <stdint.h>
#include <stddef.h>
#ifdef __cplusplus
extern "C" {
#endif
typedef struct NvencEngine NvencEngine;
typedef struct {
uint32_t width;
uint32_t height;
uint32_t fps;
uint32_t gop;
uint32_t cq;
int64_t target_bps;
int64_t max_bps;
int64_t buffer_bits;
int is_hevc; // 1 = H.265, 0 = H.264
int chroma_444; // 1 = 4:4:4 (H.264 High 4:4:4 / HEVC FRExt), 0 = 4:2:0
} NvencConfig;
// Initialize and open an NVENC encoding session. Returns NULL on error.
NvencEngine* nvenc_create(const NvencConfig* config);
// Register host shared memory (XShm buffer) as CUDA pinned memory that NVENC reads by DMA.
// Returns 0 on success, negative value on error.
int nvenc_register_host_buffer(NvencEngine* engine, void* host_ptr, size_t size);
// Unregister a previously registered host buffer (the CUDA context must be current).
void nvenc_unregister_host_buffer(NvencEngine* engine);
// Encode one frame.
// If host buffer was registered via nvenc_register_host_buffer, input_argb can be NULL
// and the encoder reads the registered host memory directly (no host-to-device copy).
// force_idr: 1 to force keyframe/IDR, 0 otherwise
// out_data: pointer to receive the bitstream buffer (valid until next encode call or destroy)
// out_len: pointer to receive the bitstream size in bytes
// Returns 0 on success, negative value on error.
int nvenc_encode_frame(
NvencEngine* engine,
const uint8_t* input_argb,
uint32_t pitch,
int force_idr,
const uint8_t** out_data,
uint32_t* out_len
);
// Destroy the NVENC engine and free all CUDA/NVENC resources.
void nvenc_destroy(NvencEngine* engine);
#ifdef __cplusplus
}
#endif
#endif // UU_NVENC_ENGINE_H
+835
View File
@@ -0,0 +1,835 @@
#include "vaapi_engine.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <dlfcn.h>
#include <va/va.h>
#include <va/va_drm.h>
#include <va/va_drmcommon.h>
#include <va/va_vpp.h>
#include <va/va_enc_h264.h>
// Runtime-loaded libva, like the NVENC engine's libcuda/libnvidia-encode: an inproc-av build
// must not add a hard DT_NEEDED on libva for users who never touch VAAPI.
// Signatures come from the installed headers via __typeof__ so they cannot drift.
static struct {
void *libva, *libva_drm;
__typeof__(vaGetDisplayDRM) *vaGetDisplayDRM;
__typeof__(vaInitialize) *vaInitialize;
__typeof__(vaTerminate) *vaTerminate;
__typeof__(vaErrorStr) *vaErrorStr;
__typeof__(vaCreateSurfaces) *vaCreateSurfaces;
__typeof__(vaDestroySurfaces) *vaDestroySurfaces;
__typeof__(vaCreateConfig) *vaCreateConfig;
__typeof__(vaDestroyConfig) *vaDestroyConfig;
__typeof__(vaCreateContext) *vaCreateContext;
__typeof__(vaDestroyContext) *vaDestroyContext;
__typeof__(vaCreateBuffer) *vaCreateBuffer;
__typeof__(vaDestroyBuffer) *vaDestroyBuffer;
__typeof__(vaMapBuffer) *vaMapBuffer;
__typeof__(vaUnmapBuffer) *vaUnmapBuffer;
__typeof__(vaCreateImage) *vaCreateImage;
__typeof__(vaDestroyImage) *vaDestroyImage;
__typeof__(vaPutImage) *vaPutImage;
__typeof__(vaBeginPicture) *vaBeginPicture;
__typeof__(vaRenderPicture) *vaRenderPicture;
__typeof__(vaEndPicture) *vaEndPicture;
__typeof__(vaSyncSurface) *vaSyncSurface;
} RT;
static int vaapi_load_runtime(void) {
if (RT.libva) return 0;
RT.libva = dlopen("libva.so.2", RTLD_NOW | RTLD_GLOBAL);
RT.libva_drm = dlopen("libva-drm.so.2", RTLD_NOW);
if (!RT.libva || !RT.libva_drm) {
fprintf(stderr, "[vaapi] dlopen libva/libva-drm failed\n");
return -1;
}
// libva-drm depends on libva, so one lookup covers both libraries
#define VA_LOAD(sym) do { \
*(void**)&RT.sym = dlsym(RT.libva_drm, #sym) ? dlsym(RT.libva_drm, #sym) : dlsym(RT.libva, #sym); \
if (!RT.sym) { dlclose(RT.libva_drm); dlclose(RT.libva); RT.libva = RT.libva_drm = NULL; return -1; } \
} while (0)
VA_LOAD(vaGetDisplayDRM);
VA_LOAD(vaInitialize);
VA_LOAD(vaTerminate);
VA_LOAD(vaErrorStr);
VA_LOAD(vaCreateSurfaces);
VA_LOAD(vaDestroySurfaces);
VA_LOAD(vaCreateConfig);
VA_LOAD(vaDestroyConfig);
VA_LOAD(vaCreateContext);
VA_LOAD(vaDestroyContext);
VA_LOAD(vaCreateBuffer);
VA_LOAD(vaDestroyBuffer);
VA_LOAD(vaMapBuffer);
VA_LOAD(vaUnmapBuffer);
VA_LOAD(vaCreateImage);
VA_LOAD(vaDestroyImage);
VA_LOAD(vaPutImage);
VA_LOAD(vaBeginPicture);
VA_LOAD(vaRenderPicture);
VA_LOAD(vaEndPicture);
VA_LOAD(vaSyncSurface);
#undef VA_LOAD
return 0;
}
struct VaapiEngine {
int drm_fd;
VADisplay va_dpy;
uint32_t cap_width; // capture (RGB input) size
uint32_t cap_height;
uint32_t width; // encode (output) size
uint32_t height;
uint32_t fps;
uint32_t gop;
int64_t target_bps;
int64_t max_bps;
int64_t buffer_bits;
int is_hevc;
uint32_t frame_count;
// frames since the last IDR (periodic GOP), the alternating idr_pic_id, and an IDR owed after a lost frame
uint32_t since_idr;
uint32_t idr_pic_id;
int idr_pending;
// frame_num / POC bookkeeping since the last IDR; H264 state the driver does not derive itself
uint32_t frame_num;
uint32_t rc_mode;
// DMA-BUF backing
int dma_fd;
uint32_t stride;
// Surfaces & Contexts
VASurfaceID in_rgb_surf;
VASurfaceID nv12_surf;
VASurfaceID recon_surfs[2];
VAConfigID vpp_cfg;
VAContextID vpp_ctx;
VABufferID vpp_buf;
VARectangle src_rect, dst_rect;
VAConfigID enc_cfg;
VAContextID enc_ctx;
VABufferID coded_buf;
uint8_t *output_staging;
uint32_t output_staging_cap;
// Host upload path: a VAImage the CPU writes, then vaPutImage into in_rgb_surf
VAImage upload_img;
int upload_ready;
};
// The VPP pipeline buffer points at the engine-owned rects, so VPP scales cap -> output
static int vaapi_make_vpp_buffer(VaapiEngine* engine) {
VAProcPipelineParameterBuffer vpp_param;
memset(&vpp_param, 0, sizeof(vpp_param));
vpp_param.surface = engine->in_rgb_surf;
vpp_param.surface_region = &engine->src_rect;
vpp_param.output_region = &engine->dst_rect;
return RT.vaCreateBuffer(engine->va_dpy, engine->vpp_ctx, VAProcPipelineParameterBufferType,
sizeof(vpp_param), 1, &vpp_param, &engine->vpp_buf) == VA_STATUS_SUCCESS ? 0 : -1;
}
VaapiEngine* vaapi_create(const VaapiConfig* config) {
if (!config || config->width == 0 || config->height == 0 || config->cap_width == 0 || config->cap_height == 0) return NULL;
if (vaapi_load_runtime() != 0) return NULL;
VaapiEngine* engine = (VaapiEngine*)calloc(1, sizeof(VaapiEngine));
if (!engine) return NULL;
engine->cap_width = config->cap_width;
engine->cap_height = config->cap_height;
engine->width = config->width;
engine->height = config->height;
engine->fps = config->fps > 0 ? config->fps : 60;
engine->gop = config->gop > 0 ? config->gop : 30;
engine->target_bps = config->target_bps > 0 ? config->target_bps : 6000000;
engine->max_bps = config->max_bps > engine->target_bps ? config->max_bps : engine->target_bps;
engine->buffer_bits = config->buffer_bits > 0 ? config->buffer_bits : engine->max_bps;
engine->is_hevc = config->is_hevc;
engine->dma_fd = -1;
// calloc leaves surface/context IDs at 0, which is a valid VA ID; mark them unset.
engine->in_rgb_surf = VA_INVALID_ID;
engine->nv12_surf = VA_INVALID_ID;
engine->recon_surfs[0] = VA_INVALID_ID;
engine->recon_surfs[1] = VA_INVALID_ID;
engine->vpp_cfg = VA_INVALID_ID;
engine->vpp_ctx = VA_INVALID_ID;
engine->vpp_buf = VA_INVALID_ID;
engine->enc_cfg = VA_INVALID_ID;
engine->enc_ctx = VA_INVALID_ID;
engine->coded_buf = VA_INVALID_ID;
engine->src_rect.width = config->cap_width;
engine->src_rect.height = config->cap_height;
engine->dst_rect.width = config->width;
engine->dst_rect.height = config->height;
engine->drm_fd = open(config->device ? config->device : "/dev/dri/renderD128", O_RDWR);
if (engine->drm_fd < 0) {
fprintf(stderr, "[vaapi] failed to open %s\n", config->device ? config->device : "/dev/dri/renderD128");
free(engine);
return NULL;
}
engine->va_dpy = RT.vaGetDisplayDRM(engine->drm_fd);
if (!engine->va_dpy) {
close(engine->drm_fd);
free(engine);
return NULL;
}
int maj, min;
VAStatus st = RT.vaInitialize(engine->va_dpy, &maj, &min);
if (st != VA_STATUS_SUCCESS) {
fprintf(stderr, "[vaapi] vaInitialize failed: %s\n", RT.vaErrorStr(st));
close(engine->drm_fd);
free(engine);
return NULL;
}
// 1. NV12 intermediate + two Recon surfaces at the encode size (ping-pong: a P frame must
// not reference the surface its own reconstruction is being written into)
st = RT.vaCreateSurfaces(engine->va_dpy, VA_RT_FORMAT_YUV420, engine->width, engine->height, &engine->nv12_surf, 1, NULL, 0);
if (st != VA_STATUS_SUCCESS) {
vaapi_destroy(engine);
return NULL;
}
st = RT.vaCreateSurfaces(engine->va_dpy, VA_RT_FORMAT_YUV420, engine->width, engine->height, engine->recon_surfs, 2, NULL, 0);
if (st != VA_STATUS_SUCCESS) {
vaapi_destroy(engine);
return NULL;
}
// 2. Setup VPP (Video Processing Pipeline for Hardware RGB->NV12 with scaling)
uint32_t vpp_w = engine->cap_width > engine->width ? engine->cap_width : engine->width;
uint32_t vpp_h = engine->cap_height > engine->height ? engine->cap_height : engine->height;
st = RT.vaCreateConfig(engine->va_dpy, VAProfileNone, VAEntrypointVideoProc, NULL, 0, &engine->vpp_cfg);
if (st != VA_STATUS_SUCCESS) {
vaapi_destroy(engine);
return NULL;
}
st = RT.vaCreateContext(engine->va_dpy, engine->vpp_cfg, vpp_w, vpp_h, 0, NULL, 0, &engine->vpp_ctx);
if (st != VA_STATUS_SUCCESS) {
vaapi_destroy(engine);
return NULL;
}
// 3. Setup H.264 / HEVC EncSlice
VAProfile profile = engine->is_hevc ? VAProfileHEVCMain : VAProfileH264Main;
VAConfigAttrib enc_attrib[3];
enc_attrib[0].type = VAConfigAttribRTFormat;
enc_attrib[0].value = VA_RT_FORMAT_YUV420;
enc_attrib[1].type = VAConfigAttribRateControl;
// The engine writes SPS, PPS and slice headers itself (see BitWriter)
enc_attrib[2].type = VAConfigAttribEncPackedHeaders;
enc_attrib[2].value = VA_ENC_PACKED_HEADER_SEQUENCE | VA_ENC_PACKED_HEADER_SLICE;
// VBR like the FFmpeg VAAPI pipeline (target + maxrate), then CBR, then CQP
enc_attrib[1].value = VA_RC_VBR;
engine->rc_mode = VA_RC_VBR;
st = RT.vaCreateConfig(engine->va_dpy, profile, VAEntrypointEncSlice, enc_attrib, 3, &engine->enc_cfg);
if (st != VA_STATUS_SUCCESS) {
enc_attrib[1].value = VA_RC_CBR;
engine->rc_mode = VA_RC_CBR;
st = RT.vaCreateConfig(engine->va_dpy, profile, VAEntrypointEncSlice, enc_attrib, 3, &engine->enc_cfg);
}
if (st != VA_STATUS_SUCCESS) {
// Fallback to CQP if CBR unsupported
enc_attrib[1].value = VA_RC_CQP;
st = RT.vaCreateConfig(engine->va_dpy, profile, VAEntrypointEncSlice, enc_attrib, 3, &engine->enc_cfg);
if (st != VA_STATUS_SUCCESS) {
fprintf(stderr, "[vaapi] vaCreateConfig EncSlice failed: %s\n", RT.vaErrorStr(st));
vaapi_destroy(engine);
return NULL;
}
engine->rc_mode = VA_RC_CQP;
}
st = RT.vaCreateContext(engine->va_dpy, engine->enc_cfg, engine->width, engine->height, VA_PROGRESSIVE, engine->recon_surfs, 2, &engine->enc_ctx);
if (st != VA_STATUS_SUCCESS) {
fprintf(stderr, "[vaapi] vaCreateContext EncSlice failed: %s\n", RT.vaErrorStr(st));
vaapi_destroy(engine);
return NULL;
}
// Coded bitstream buffer: generous, a 20M-preset intra frame can burst past 4MB
st = RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncCodedBufferType, 16 * 1024 * 1024, 1, NULL, &engine->coded_buf);
if (st != VA_STATUS_SUCCESS) {
vaapi_destroy(engine);
return NULL;
}
return engine;
}
int vaapi_import_dma_buffer(VaapiEngine* engine, int dma_fd, uint32_t stride) {
if (!engine || dma_fd < 0) return -1;
if (engine->in_rgb_surf != VA_INVALID_ID) {
RT.vaDestroySurfaces(engine->va_dpy, &engine->in_rgb_surf, 1);
engine->in_rgb_surf = VA_INVALID_ID;
}
if (engine->vpp_buf != VA_INVALID_ID) {
RT.vaDestroyBuffer(engine->va_dpy, engine->vpp_buf);
engine->vpp_buf = VA_INVALID_ID;
}
engine->dma_fd = dma_fd;
engine->stride = stride;
VADRMPRIMESurfaceDescriptor prime_desc;
memset(&prime_desc, 0, sizeof(prime_desc));
prime_desc.fourcc = VA_FOURCC_BGRX;
prime_desc.width = engine->cap_width;
prime_desc.height = engine->cap_height;
prime_desc.num_objects = 1;
prime_desc.objects[0].fd = dma_fd;
prime_desc.objects[0].size = stride * engine->cap_height;
prime_desc.objects[0].drm_format_modifier = 0;
prime_desc.num_layers = 1;
prime_desc.layers[0].drm_format = 0x34325258; // DRM_FORMAT_XRGB8888
prime_desc.layers[0].num_planes = 1;
prime_desc.layers[0].pitch[0] = stride;
VASurfaceAttrib attribs[2];
attribs[0].type = VASurfaceAttribMemoryType;
attribs[0].flags = VA_SURFACE_ATTRIB_SETTABLE;
attribs[0].value.type = VAGenericValueTypeInteger;
attribs[0].value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME_2;
attribs[1].type = VASurfaceAttribExternalBufferDescriptor;
attribs[1].flags = VA_SURFACE_ATTRIB_SETTABLE;
attribs[1].value.type = VAGenericValueTypePointer;
attribs[1].value.value.p = &prime_desc;
VAStatus st = RT.vaCreateSurfaces(engine->va_dpy, VA_RT_FORMAT_RGB32, engine->cap_width, engine->cap_height, &engine->in_rgb_surf, 1, attribs, 2);
if (st != VA_STATUS_SUCCESS) {
fprintf(stderr, "[vaapi] vaCreateSurfaces DRM_PRIME_2 failed: %s\n", RT.vaErrorStr(st));
return -1;
}
if (vaapi_make_vpp_buffer(engine) != 0) {
fprintf(stderr, "[vaapi] vaCreateBuffer VPP failed\n");
return -2;
}
return 0;
}
int vaapi_upload_host_buffer(VaapiEngine* engine, const uint8_t* host_ptr, uint32_t pitch) {
if (!engine || !host_ptr) return -1;
if (engine->in_rgb_surf == VA_INVALID_ID) {
VAStatus st = RT.vaCreateSurfaces(engine->va_dpy, VA_RT_FORMAT_RGB32, engine->cap_width, engine->cap_height, &engine->in_rgb_surf, 1, NULL, 0);
if (st != VA_STATUS_SUCCESS) return -2;
if (vaapi_make_vpp_buffer(engine) != 0) return -3;
}
if (!engine->upload_ready) {
VAImageFormat fmt;
memset(&fmt, 0, sizeof(fmt));
fmt.fourcc = VA_FOURCC_BGRX;
fmt.byte_order = VA_LSB_FIRST;
fmt.bits_per_pixel = 32;
fmt.depth = 24;
fmt.red_mask = 0x00ff0000;
fmt.green_mask = 0x0000ff00;
fmt.blue_mask = 0x000000ff;
// vaDeriveImage on the RGB32 surface is refused by some drivers
VAStatus st = RT.vaCreateImage(engine->va_dpy, &fmt, engine->cap_width, engine->cap_height, &engine->upload_img);
if (st != VA_STATUS_SUCCESS) return -4;
engine->upload_ready = 1;
}
void* p = NULL;
VAStatus st = RT.vaMapBuffer(engine->va_dpy, engine->upload_img.buf, &p);
if (st != VA_STATUS_SUCCESS || !p) return -5;
uint32_t copy_bytes = engine->cap_width * 4;
for (uint32_t y = 0; y < engine->cap_height; y++) {
memcpy((uint8_t*)p + y * engine->upload_img.pitches[0], host_ptr + y * pitch, copy_bytes);
}
RT.vaUnmapBuffer(engine->va_dpy, engine->upload_img.buf);
st = RT.vaPutImage(engine->va_dpy, engine->in_rgb_surf, engine->upload_img.image_id,
0, 0, engine->cap_width, engine->cap_height, 0, 0, engine->cap_width, engine->cap_height);
if (st != VA_STATUS_SUCCESS) return -6;
return 0;
}
// Sequence-level misc parameter. AMD's Mesa driver feeds the VCN ring message from these;
// a CBR config that never gets its rate-control/HRD parameters submits garbage and the
// encode job wedges the ring until a GPU reset.
// Packed H.264 / HEVC headers. Current radeonsi (Mesa 26) writes neither SPS/PPS nor the NAL/slice header from the
// parameter buffers alone (output began 00 00 00 01 00 ..., no SPS), so the host writes them like FFmpeg does.
// Every field here must match the sequence/picture/slice parameters submitted with it.
typedef struct { uint8_t buf[128]; uint32_t bits; } BitWriter;
static void bw_put(BitWriter* w, uint32_t value, int n) {
for (int i = n - 1; i >= 0; i--, w->bits++) {
if (w->bits / 8 >= sizeof(w->buf)) return;
if ((value >> i) & 1) w->buf[w->bits / 8] |= 0x80 >> (w->bits % 8);
}
}
static void bw_ue(BitWriter* w, uint32_t value) {
int len = 0;
for (uint32_t v = value + 1; v > 1; v >>= 1) len++;
bw_put(w, 0, len);
bw_put(w, value + 1, len + 1);
}
static void bw_se(BitWriter* w, int32_t value) { bw_ue(w, value > 0 ? 2 * (uint32_t)value - 1 : 2 * (uint32_t)(-value)); }
static void bw_trailing(BitWriter* w) { bw_put(w, 1, 1); if (w->bits % 8) bw_put(w, 0, 8 - w->bits % 8); }
// Start code + NAL header; the payload follows unescaped, emulation prevention is applied by bw_escape.
static void bw_nal(BitWriter* w, int ref_idc, int type) { bw_put(w, 1, 32); bw_put(w, (ref_idc << 5) | type, 8); }
// Insert emulation prevention bytes after the start code + NAL header (byte 5 on). Returns the byte length.
static uint32_t bw_escape(const BitWriter* w, uint8_t* out, uint32_t cap) {
uint32_t len = (w->bits + 7) / 8, n = 0, zeros = 0;
for (uint32_t i = 0; i < len && n < cap; i++) {
if (i >= 5 && zeros >= 2 && w->buf[i] <= 3) { out[n++] = 3; zeros = 0; if (n >= cap) break; }
zeros = w->buf[i] == 0 && i >= 5 ? zeros + 1 : 0;
out[n++] = w->buf[i];
}
return n;
}
// Each NAL is escaped on its own, so the next NAL's start code stays intact.
static int vaapi_add_packed(VaapiEngine* engine, VAEncPackedHeaderType type, const BitWriter* nals, int count,
VABufferID* ids, int* n) {
uint8_t data[320];
uint32_t bytes = 0;
for (int i = 0; i < count; i++) bytes += bw_escape(&nals[i], data + bytes, sizeof(data) - bytes);
VAEncPackedHeaderParameterBuffer param = { .type = type, .bit_length = bytes * 8, .has_emulation_bytes = 1 };
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncPackedHeaderParameterBufferType, sizeof(param), 1, &param, &ids[*n]) != VA_STATUS_SUCCESS)
return -1;
(*n)++;
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncPackedHeaderDataBufferType, bytes, 1, data, &ids[*n]) != VA_STATUS_SUCCESS)
return -1;
(*n)++;
return 0;
}
static VABufferID vaapi_create_misc_buffer(VaapiEngine* engine, VAEncMiscParameterType type, const void* param, uint32_t param_size) {
uint8_t buf[sizeof(VAEncMiscParameterBuffer) + 64];
VAEncMiscParameterBuffer* misc = (VAEncMiscParameterBuffer*)buf;
misc->type = type;
memcpy(misc->data, param, param_size);
VABufferID id = VA_INVALID_ID;
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncMiscParameterBufferType,
sizeof(VAEncMiscParameterBuffer) + param_size, 1, buf, &id) != VA_STATUS_SUCCESS) {
return VA_INVALID_ID;
}
return id;
}
// Buffers for one encoded picture; every created ID is destroyed after the picture, success or not.
typedef struct {
VABufferID seq, misc[3], pic, slice, packed[6];
int misc_n, packed_n;
} EncBuffers;
static int vaapi_h264_params(VaapiEngine* engine, int is_idr, VASurfaceID curr_recon, VASurfaceID prev_recon, EncBuffers* b) {
uint32_t mb_w = (engine->width + 15) / 16;
uint32_t mb_h = (engine->height + 15) / 16;
uint32_t frame_num = is_idr ? 0 : engine->frame_num;
uint32_t crop_right = (mb_w * 16 - engine->width) / 2, crop_bottom = (mb_h * 16 - engine->height) / 2;
if (is_idr) {
VAEncSequenceParameterBufferH264 seq_p;
memset(&seq_p, 0, sizeof(seq_p));
seq_p.seq_parameter_set_id = 0;
seq_p.level_idc = 51;
seq_p.intra_period = engine->gop;
seq_p.intra_idr_period = engine->gop;
seq_p.ip_period = 1;
seq_p.bits_per_second = (uint32_t)engine->max_bps;
seq_p.max_num_ref_frames = 1;
seq_p.picture_width_in_mbs = mb_w;
seq_p.picture_height_in_mbs = mb_h;
seq_p.seq_fields.bits.direct_8x8_inference_flag = 1;
seq_p.seq_fields.bits.frame_mbs_only_flag = 1;
seq_p.seq_fields.bits.pic_order_cnt_type = 0;
seq_p.frame_cropping_flag = crop_right || crop_bottom;
seq_p.frame_crop_right_offset = crop_right;
seq_p.frame_crop_bottom_offset = crop_bottom;
// An IDR without its SPS is undecodable: fail rather than submit pic + slice alone
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncSequenceParameterBufferType, sizeof(seq_p), 1, &seq_p, &b->seq) != VA_STATUS_SUCCESS)
return -1;
// SPS: Main, level 5.1, frame_num and POC lsb 4 bits (both wrap & 15 above), one reference
BitWriter sps = {0};
bw_nal(&sps, 3, 7);
bw_put(&sps, 77, 8); bw_put(&sps, 0x40, 8); bw_put(&sps, 51, 8); // constraint_set1: Main
bw_ue(&sps, 0); bw_ue(&sps, 0); bw_ue(&sps, 0); bw_ue(&sps, 0); // sps id, log2_max_frame_num-4, poc type 0, log2_max_poc_lsb-4
bw_ue(&sps, 1); bw_put(&sps, 0, 1); // max_num_ref_frames, gaps
bw_ue(&sps, mb_w - 1); bw_ue(&sps, mb_h - 1);
bw_put(&sps, 1, 1); bw_put(&sps, 1, 1); // frame_mbs_only, direct_8x8_inference
bw_put(&sps, seq_p.frame_cropping_flag, 1);
if (seq_p.frame_cropping_flag) { bw_ue(&sps, 0); bw_ue(&sps, crop_right); bw_ue(&sps, 0); bw_ue(&sps, crop_bottom); }
bw_put(&sps, 0, 1); // no VUI
bw_trailing(&sps);
// PPS: CABAC, one L0 reference, QP 26, default deblocking
BitWriter pps = {0};
bw_nal(&pps, 3, 8);
bw_ue(&pps, 0); bw_ue(&pps, 0); bw_put(&pps, 1, 1); bw_put(&pps, 0, 1);
bw_ue(&pps, 0); bw_ue(&pps, 0); bw_ue(&pps, 0); // slice groups, num_ref_idx l0/l1 - 1
bw_put(&pps, 0, 1); bw_put(&pps, 0, 2); // weighted pred / bipred
bw_se(&pps, 0); bw_se(&pps, 0); bw_se(&pps, 0); // pic_init_qp-26, qs, chroma offset
bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); // deblocking control, constrained intra, redundant
bw_trailing(&pps);
// SPS and PPS go out as one Sequence header, like FFmpeg
BitWriter headers[2] = { sps, pps };
if (vaapi_add_packed(engine, VAEncPackedHeaderSequence, headers, 2, b->packed, &b->packed_n) != 0) return -1;
}
VAEncPictureParameterBufferH264 pic_p;
memset(&pic_p, 0, sizeof(pic_p));
pic_p.CurrPic.picture_id = curr_recon;
pic_p.CurrPic.flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE;
pic_p.CurrPic.frame_idx = frame_num;
pic_p.CurrPic.TopFieldOrderCnt = pic_p.CurrPic.BottomFieldOrderCnt = (2 * frame_num) & 15;
for (int i = 0; i < 16; i++) pic_p.ReferenceFrames[i].picture_id = VA_INVALID_ID;
if (!is_idr) {
pic_p.ReferenceFrames[0].picture_id = prev_recon;
pic_p.ReferenceFrames[0].flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE;
pic_p.ReferenceFrames[0].frame_idx = (frame_num - 1) & 15;
pic_p.ReferenceFrames[0].TopFieldOrderCnt = pic_p.ReferenceFrames[0].BottomFieldOrderCnt = (2 * (frame_num - 1)) & 15;
}
pic_p.coded_buf = engine->coded_buf;
pic_p.frame_num = frame_num;
pic_p.pic_init_qp = 26;
pic_p.seq_parameter_set_id = 0;
pic_p.pic_parameter_set_id = 0;
pic_p.pic_fields.bits.idr_pic_flag = is_idr ? 1 : 0;
pic_p.pic_fields.bits.reference_pic_flag = 1;
pic_p.pic_fields.bits.entropy_coding_mode_flag = 1;
pic_p.pic_fields.bits.deblocking_filter_control_present_flag = 0;
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncPictureParameterBufferType, sizeof(pic_p), 1, &pic_p, &b->pic) != VA_STATUS_SUCCESS)
return -1;
VAEncSliceParameterBufferH264 slice_p;
memset(&slice_p, 0, sizeof(slice_p));
slice_p.macroblock_address = 0;
slice_p.num_macroblocks = mb_w * mb_h;
slice_p.macroblock_info = VA_INVALID_ID;
slice_p.slice_type = is_idr ? 2 : 0; // 2 = I-Slice, 0 = P-Slice
slice_p.direct_spatial_mv_pred_flag = 1;
// Consecutive IDRs must differ in idr_pic_id (7.4.3); flip it on every IDR
if (is_idr) engine->idr_pic_id ^= 1;
slice_p.idr_pic_id = engine->idr_pic_id;
slice_p.pic_order_cnt_lsb = pic_p.CurrPic.TopFieldOrderCnt;
for (int i = 0; i < 32; i++) slice_p.RefPicList0[i].picture_id = VA_INVALID_ID;
if (!is_idr) {
slice_p.RefPicList0[0].picture_id = prev_recon;
slice_p.RefPicList0[0].flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE;
slice_p.RefPicList0[0].frame_idx = (frame_num - 1) & 15;
}
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncSliceParameterBufferType, sizeof(slice_p), 1, &slice_p, &b->slice) != VA_STATUS_SUCCESS)
return -1;
// Slice header including cabac_alignment_one_bit, byte aligned and escaped as FFmpeg submits it
BitWriter sh = {0};
bw_nal(&sh, 3, is_idr ? 5 : 1);
bw_ue(&sh, 0); bw_ue(&sh, is_idr ? 7 : 5); bw_ue(&sh, 0); // first_mb, slice_type I/P (all slices), pps id
bw_put(&sh, frame_num & 15, 4);
if (is_idr) bw_ue(&sh, slice_p.idr_pic_id);
bw_put(&sh, slice_p.pic_order_cnt_lsb & 15, 4);
if (!is_idr) { bw_put(&sh, 0, 1); bw_put(&sh, 0, 1); } // num_ref_idx override, ref list modification
if (is_idr) { bw_put(&sh, 0, 1); bw_put(&sh, 0, 1); } // no_output_of_prior_pics, long_term_reference
else bw_put(&sh, 0, 1); // adaptive_ref_pic_marking
if (!is_idr) bw_ue(&sh, 0); // cabac_init_idc
bw_se(&sh, 0); // slice_qp_delta (no deblocking fields)
while (sh.bits % 8) bw_put(&sh, 1, 1);
return vaapi_add_packed(engine, VAEncPackedHeaderSlice, &sh, 1, b->packed, &b->packed_n);
}
// HEVC NAL header: forbidden 0, nal_unit_type, nuh_layer_id 0, nuh_temporal_id_plus1 1.
static void bw_hevc_nal(BitWriter* w, int type) { bw_put(w, 1, 32); bw_put(w, (type << 9) | 1, 16); }
// profile_tier_level(1, 0): Main, Main 10 compatible like FFmpeg, progressive frames only.
static void bw_hevc_ptl(BitWriter* w, uint32_t level) {
bw_put(w, 0, 2); bw_put(w, 0, 1); bw_put(w, 1, 5); // profile space, tier, general_profile_idc Main
bw_put(w, 0x60000000, 32); // compatibility flags 1 and 2
bw_put(w, 1, 1); bw_put(w, 0, 1); bw_put(w, 1, 1); bw_put(w, 1, 1); // progressive, interlaced, non_packed, frame_only
bw_put(w, 0, 32); bw_put(w, 0, 12); // reserved_zero_43bits + inbld_flag
bw_put(w, level, 8);
}
// HEVC Main with FFmpeg's hevc_vaapi parameters on radeonsi (LIBVA_TRACE + trace_headers of its output):
// 64x64 CTB, 8x8 min CB, 4x4..32x32 TB, AMP and SAO on, one short-term reference, POC lsb 12 bits.
// FFmpeg also sends VA max_transform_hierarchy_depth 0 / diff_cu_qp_delta_depth 3 while its SPS/PPS
// say 4 / 0; both copies are kept as FFmpeg has them, since that is what decodes on this driver.
static int vaapi_hevc_params(VaapiEngine* engine, int is_idr, VASurfaceID curr_recon, VASurfaceID prev_recon, EncBuffers* b) {
// pic size must be a multiple of the 8x8 min CB; the conformance window (in chroma units) crops the rest
uint32_t w = (engine->width + 7) & ~7u, h = (engine->height + 7) & ~7u;
uint32_t crop_right = (w - engine->width) / 2, crop_bottom = (h - engine->height) / 2;
uint32_t poc = is_idr ? 0 : engine->since_idr;
// Level from luma picture size and sample rate (Table A.8): 4.1, 5.0, 5.1, 5.2
uint64_t size = (uint64_t)w * h, rate = size * engine->fps;
uint32_t level = rate > 534773760 ? 156 : rate > 267386880 ? 153 : (size > 2228224 || rate > 133693440) ? 150 : 123;
if (is_idr) {
VAEncSequenceParameterBufferHEVC seq_p;
memset(&seq_p, 0, sizeof(seq_p));
seq_p.general_profile_idc = 1;
seq_p.general_level_idc = level;
seq_p.intra_period = engine->gop;
seq_p.intra_idr_period = engine->gop;
seq_p.ip_period = 1;
seq_p.bits_per_second = (uint32_t)engine->max_bps;
seq_p.pic_width_in_luma_samples = w;
seq_p.pic_height_in_luma_samples = h;
seq_p.seq_fields.bits.chroma_format_idc = 1;
seq_p.seq_fields.bits.amp_enabled_flag = 1;
seq_p.seq_fields.bits.sample_adaptive_offset_enabled_flag = 1;
seq_p.log2_diff_max_min_luma_coding_block_size = 3;
seq_p.log2_diff_max_min_transform_block_size = 3;
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncSequenceParameterBufferType, sizeof(seq_p), 1, &seq_p, &b->seq) != VA_STATUS_SUCCESS)
return -1;
BitWriter vps = {0};
bw_hevc_nal(&vps, 32);
bw_put(&vps, 0, 4); bw_put(&vps, 1, 1); bw_put(&vps, 1, 1); // vps id, base layer internal / available
bw_put(&vps, 0, 6); bw_put(&vps, 0, 3); bw_put(&vps, 1, 1); // max layers-1, max sub layers-1, temporal id nesting
bw_put(&vps, 0xffff, 16);
bw_hevc_ptl(&vps, level);
bw_put(&vps, 0, 1); bw_ue(&vps, 1); bw_ue(&vps, 0); bw_ue(&vps, 0); // ordering info: dec buffering-1, reorder, latency
bw_put(&vps, 0, 6); bw_ue(&vps, 0); // max layer id, num layer sets-1
bw_put(&vps, 0, 1); bw_put(&vps, 0, 1); // no timing info, no extension
bw_trailing(&vps);
BitWriter sps = {0};
bw_hevc_nal(&sps, 33);
bw_put(&sps, 0, 4); bw_put(&sps, 0, 3); bw_put(&sps, 1, 1); // vps id, max sub layers-1, temporal id nesting
bw_hevc_ptl(&sps, level);
bw_ue(&sps, 0); bw_ue(&sps, 1); bw_ue(&sps, w); bw_ue(&sps, h); // sps id, 4:2:0, size
bw_put(&sps, crop_right || crop_bottom, 1);
if (crop_right || crop_bottom) { bw_ue(&sps, 0); bw_ue(&sps, crop_right); bw_ue(&sps, 0); bw_ue(&sps, crop_bottom); }
bw_ue(&sps, 0); bw_ue(&sps, 0); bw_ue(&sps, 8); // 8-bit luma/chroma, log2_max_poc_lsb-4
bw_put(&sps, 0, 1); bw_ue(&sps, 1); bw_ue(&sps, 0); bw_ue(&sps, 0);
bw_ue(&sps, 0); bw_ue(&sps, 3); bw_ue(&sps, 0); bw_ue(&sps, 3); // CB 8..64, TB 4..32
bw_ue(&sps, 4); bw_ue(&sps, 4); // max transform hierarchy depth inter/intra
bw_put(&sps, 0, 1); bw_put(&sps, 1, 1); bw_put(&sps, 1, 1); bw_put(&sps, 0, 1); // scaling list, AMP, SAO, PCM
bw_ue(&sps, 0); bw_put(&sps, 0, 1); // RPS in slice headers, no long-term refs
bw_put(&sps, 0, 1); bw_put(&sps, 0, 1); // temporal MVP, strong intra smoothing
bw_put(&sps, 0, 1); bw_put(&sps, 0, 1); // no VUI, no extension
bw_trailing(&sps);
BitWriter pps = {0};
bw_hevc_nal(&pps, 34);
bw_ue(&pps, 0); bw_ue(&pps, 0);
bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); bw_put(&pps, 0, 3); // dependent slices, output flag, extra bits
bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); // sign data hiding, cabac init present
bw_ue(&pps, 0); bw_ue(&pps, 0); bw_se(&pps, 0); // num_ref_idx l0/l1 default-1, init_qp-26
bw_put(&pps, 0, 1); bw_put(&pps, 1, 1); bw_put(&pps, 1, 1); // constrained intra, transform skip, cu_qp_delta
bw_ue(&pps, 0); bw_se(&pps, 0); bw_se(&pps, 0); // diff_cu_qp_delta_depth, cb/cr offsets
bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); // slice chroma offsets, weighted pred/bipred
bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); // transquant bypass, tiles, entropy sync
bw_put(&pps, 1, 1); bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); // loop filter across slices, deblocking ctl, scaling
bw_put(&pps, 0, 1); bw_ue(&pps, 0); bw_put(&pps, 0, 1); bw_put(&pps, 0, 1); // list mod, merge level-2, ext
bw_trailing(&pps);
BitWriter headers[3] = { vps, sps, pps };
if (vaapi_add_packed(engine, VAEncPackedHeaderSequence, headers, 3, b->packed, &b->packed_n) != 0) return -1;
}
VAEncPictureParameterBufferHEVC pic_p;
memset(&pic_p, 0, sizeof(pic_p));
pic_p.decoded_curr_pic.picture_id = curr_recon;
pic_p.decoded_curr_pic.pic_order_cnt = poc;
for (int i = 0; i < 15; i++) {
pic_p.reference_frames[i].picture_id = VA_INVALID_ID;
pic_p.reference_frames[i].flags = VA_PICTURE_HEVC_INVALID;
}
if (!is_idr) {
pic_p.reference_frames[0].picture_id = prev_recon;
pic_p.reference_frames[0].pic_order_cnt = poc - 1;
pic_p.reference_frames[0].flags = VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE;
}
pic_p.coded_buf = engine->coded_buf;
pic_p.collocated_ref_pic_index = 0xff;
pic_p.pic_init_qp = 30;
pic_p.diff_cu_qp_delta_depth = 3;
pic_p.nal_unit_type = is_idr ? 19 : 1; // IDR_W_RADL / TRAIL_R
pic_p.pic_fields.bits.idr_pic_flag = is_idr;
pic_p.pic_fields.bits.coding_type = is_idr ? 1 : 2;
pic_p.pic_fields.bits.reference_pic_flag = 1;
pic_p.pic_fields.bits.transform_skip_enabled_flag = 1;
pic_p.pic_fields.bits.cu_qp_delta_enabled_flag = 1;
pic_p.pic_fields.bits.pps_loop_filter_across_slices_enabled_flag = 1;
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncPictureParameterBufferType, sizeof(pic_p), 1, &pic_p, &b->pic) != VA_STATUS_SUCCESS)
return -1;
VAEncSliceParameterBufferHEVC slice_p;
memset(&slice_p, 0, sizeof(slice_p));
slice_p.num_ctu_in_slice = ((w + 63) / 64) * ((h + 63) / 64);
slice_p.slice_type = is_idr ? 2 : 1; // I / P
for (int i = 0; i < 15; i++) {
slice_p.ref_pic_list0[i].picture_id = slice_p.ref_pic_list1[i].picture_id = VA_INVALID_ID;
slice_p.ref_pic_list0[i].flags = slice_p.ref_pic_list1[i].flags = VA_PICTURE_HEVC_INVALID;
}
if (!is_idr) slice_p.ref_pic_list0[0] = pic_p.reference_frames[0];
slice_p.max_num_merge_cand = 5;
slice_p.slice_fields.bits.last_slice_of_pic_flag = 1;
slice_p.slice_fields.bits.slice_sao_luma_flag = 1;
slice_p.slice_fields.bits.slice_sao_chroma_flag = 1;
slice_p.slice_fields.bits.collocated_from_l0_flag = !is_idr;
if (RT.vaCreateBuffer(engine->va_dpy, engine->enc_ctx, VAEncSliceParameterBufferType, sizeof(slice_p), 1, &slice_p, &b->slice) != VA_STATUS_SUCCESS)
return -1;
BitWriter sh = {0};
bw_hevc_nal(&sh, pic_p.nal_unit_type);
bw_put(&sh, 1, 1); // first slice segment
if (is_idr) bw_put(&sh, 0, 1); // no_output_of_prior_pics
bw_ue(&sh, 0); bw_ue(&sh, slice_p.slice_type); // pps id, slice type
if (!is_idr) {
bw_put(&sh, poc & 0xfff, 12);
bw_put(&sh, 0, 1); // RPS coded here: one previous picture
bw_ue(&sh, 1); bw_ue(&sh, 0); bw_ue(&sh, 0); bw_put(&sh, 1, 1); // negative 1, positive 0, delta-1 0, used
}
bw_put(&sh, 1, 1); bw_put(&sh, 1, 1); // SAO luma, chroma
if (!is_idr) { bw_put(&sh, 0, 1); bw_ue(&sh, 0); } // num_ref_idx override, five_minus_max_merge
bw_se(&sh, 0); // slice_qp_delta (the driver writes the real one)
bw_put(&sh, 0, 1); // loop filter across slices
bw_put(&sh, 1, 1); // byte_alignment()
while (sh.bits % 8) bw_put(&sh, 0, 1);
return vaapi_add_packed(engine, VAEncPackedHeaderSlice, &sh, 1, b->packed, &b->packed_n);
}
int vaapi_encode_frame(
VaapiEngine* engine,
int force_idr,
const uint8_t** out_data,
uint32_t* out_len
) {
if (out_data) *out_data = NULL;
if (out_len) *out_len = 0;
if (!engine || engine->in_rgb_surf == VA_INVALID_ID) return -1;
int is_idr = force_idr || engine->idr_pending || engine->frame_count == 0 || engine->since_idr >= engine->gop;
// Recon ping-pong: a P frame must reference the previous recon surface, not its own
VASurfaceID curr_recon = engine->recon_surfs[engine->frame_count & 1];
VASurfaceID prev_recon = engine->recon_surfs[(engine->frame_count + 1) & 1];
// 1. Hardware VPP Color Conversion + scaling (RGB -> NV12 entirely inside VRAM)
VAStatus st = RT.vaBeginPicture(engine->va_dpy, engine->vpp_ctx, engine->nv12_surf);
if (st != VA_STATUS_SUCCESS) return -10;
st = RT.vaRenderPicture(engine->va_dpy, engine->vpp_ctx, &engine->vpp_buf, 1);
if (st != VA_STATUS_SUCCESS) return -10;
st = RT.vaEndPicture(engine->va_dpy, engine->vpp_ctx);
if (st != VA_STATUS_SUCCESS) return -10;
// 2. Rate control (every IDR) and the codec's sequence / picture / slice parameters + packed headers
EncBuffers b = { .seq = VA_INVALID_ID, .pic = VA_INVALID_ID, .slice = VA_INVALID_ID };
int ret = -11;
if (is_idr) {
VAEncMiscParameterRateControl rc;
memset(&rc, 0, sizeof(rc));
// VBR: bits_per_second is the ceiling (maxrate) and target_percentage the average (#33)
rc.bits_per_second = (uint32_t)engine->max_bps;
rc.target_percentage = engine->rc_mode == VA_RC_VBR ? (uint32_t)(engine->target_bps * 100 / engine->max_bps) : 100;
rc.window_size = 1000;
rc.initial_qp = 28;
rc.min_qp = 16;
rc.max_qp = 44;
rc.rc_flags.bits.disable_frame_skip = 1; // a remote desktop must not drop frames silently
b.misc[b.misc_n++] = vaapi_create_misc_buffer(engine, VAEncMiscParameterTypeRateControl, &rc, sizeof(rc));
VAEncMiscParameterHRD hrd;
memset(&hrd, 0, sizeof(hrd));
hrd.buffer_size = (uint32_t)engine->buffer_bits; // the viewer preset's bufsize
hrd.initial_buffer_fullness = hrd.buffer_size / 2;
b.misc[b.misc_n++] = vaapi_create_misc_buffer(engine, VAEncMiscParameterTypeHRD, &hrd, sizeof(hrd));
// numerator in the low 16 bits, denominator in the high 16 bits
VAEncMiscParameterFrameRate fr = { .framerate = (1u << 16) | engine->fps };
b.misc[b.misc_n++] = vaapi_create_misc_buffer(engine, VAEncMiscParameterTypeFrameRate, &fr, sizeof(fr));
}
if ((engine->is_hevc ? vaapi_hevc_params : vaapi_h264_params)(engine, is_idr, curr_recon, prev_recon, &b) != 0) goto done;
// 3. Render and Encode on AMD VCN hardware
VABufferID render_b[12];
int render_n = 0;
// FFmpeg's order: sequence, rate control misc, picture, packed headers (parameter sets, slice header), slice.
// With misc first, radeonsi (Mesa 26) ignored the VBR target: the 1M preset ran at 1.6 Mbps, 0.96 after.
if (b.seq != VA_INVALID_ID) render_b[render_n++] = b.seq;
for (int i = 0; i < b.misc_n; i++) {
if (b.misc[i] != VA_INVALID_ID) render_b[render_n++] = b.misc[i];
}
render_b[render_n++] = b.pic;
for (int i = 0; i < b.packed_n; i++) render_b[render_n++] = b.packed[i];
render_b[render_n++] = b.slice;
st = RT.vaBeginPicture(engine->va_dpy, engine->enc_ctx, engine->nv12_surf);
if (st != VA_STATUS_SUCCESS) goto done;
st = RT.vaRenderPicture(engine->va_dpy, engine->enc_ctx, render_b, render_n);
if (st != VA_STATUS_SUCCESS) goto done;
st = RT.vaEndPicture(engine->va_dpy, engine->enc_ctx);
if (st != VA_STATUS_SUCCESS) goto done;
st = RT.vaSyncSurface(engine->va_dpy, curr_recon);
if (st != VA_STATUS_SUCCESS) goto done;
// 4. Map Bitstream Output; an access unit may arrive as several linked segments
VACodedBufferSegment *segment = NULL;
st = RT.vaMapBuffer(engine->va_dpy, engine->coded_buf, (void**)&segment);
if (st != VA_STATUS_SUCCESS || !segment) goto done;
uint32_t total = 0;
for (VACodedBufferSegment *seg = segment; seg; seg = seg->next) total += seg->size;
if (total > 0 && !(segment->status & VA_CODED_BUF_STATUS_SLICE_OVERFLOW_MASK)) {
if (total > engine->output_staging_cap) {
engine->output_staging_cap = total + 64 * 1024;
engine->output_staging = (uint8_t*)realloc(engine->output_staging, engine->output_staging_cap);
}
uint32_t off = 0;
for (VACodedBufferSegment *seg = segment; seg; seg = seg->next) {
memcpy(engine->output_staging + off, seg->buf, seg->size);
off += seg->size;
}
*out_data = engine->output_staging;
*out_len = total;
}
RT.vaUnmapBuffer(engine->va_dpy, engine->coded_buf);
engine->frame_num = is_idr ? 1 : ((engine->frame_num + 1) & 15);
engine->since_idr = is_idr ? 1 : engine->since_idr + 1;
engine->frame_count++;
ret = 0;
done:
// No bitstream (failure, empty or slice overflow): the next frame_num / POC would skip, so restart with an IDR
engine->idr_pending = ret != 0 || *out_len == 0;
for (int i = 0; i < b.misc_n; i++) {
if (b.misc[i] != VA_INVALID_ID) RT.vaDestroyBuffer(engine->va_dpy, b.misc[i]);
}
if (b.seq != VA_INVALID_ID) RT.vaDestroyBuffer(engine->va_dpy, b.seq);
if (b.pic != VA_INVALID_ID) RT.vaDestroyBuffer(engine->va_dpy, b.pic);
if (b.slice != VA_INVALID_ID) RT.vaDestroyBuffer(engine->va_dpy, b.slice);
for (int i = 0; i < b.packed_n; i++) RT.vaDestroyBuffer(engine->va_dpy, b.packed[i]);
return ret;
}
void vaapi_destroy(VaapiEngine* engine) {
if (!engine) return;
if (engine->coded_buf != VA_INVALID_ID) RT.vaDestroyBuffer(engine->va_dpy, engine->coded_buf);
if (engine->enc_ctx != VA_INVALID_ID) RT.vaDestroyContext(engine->va_dpy, engine->enc_ctx);
if (engine->enc_cfg != VA_INVALID_ID) RT.vaDestroyConfig(engine->va_dpy, engine->enc_cfg);
if (engine->vpp_buf != VA_INVALID_ID) RT.vaDestroyBuffer(engine->va_dpy, engine->vpp_buf);
if (engine->vpp_ctx != VA_INVALID_ID) RT.vaDestroyContext(engine->va_dpy, engine->vpp_ctx);
if (engine->vpp_cfg != VA_INVALID_ID) RT.vaDestroyConfig(engine->va_dpy, engine->vpp_cfg);
if (engine->upload_ready) RT.vaDestroyImage(engine->va_dpy, engine->upload_img.image_id);
if (engine->in_rgb_surf != VA_INVALID_ID) RT.vaDestroySurfaces(engine->va_dpy, &engine->in_rgb_surf, 1);
if (engine->nv12_surf != VA_INVALID_ID) RT.vaDestroySurfaces(engine->va_dpy, &engine->nv12_surf, 1);
if (engine->recon_surfs[0] != VA_INVALID_ID || engine->recon_surfs[1] != VA_INVALID_ID) {
RT.vaDestroySurfaces(engine->va_dpy, engine->recon_surfs, 2);
}
if (engine->va_dpy) RT.vaTerminate(engine->va_dpy);
if (engine->drm_fd >= 0) close(engine->drm_fd);
if (engine->output_staging) free(engine->output_staging);
free(engine);
}
+49
View File
@@ -0,0 +1,49 @@
#ifndef UU_VAAPI_ENGINE_H
#define UU_VAAPI_ENGINE_H
#include <stdint.h>
#include <stddef.h>
#ifdef __cplusplus
extern "C" {
#endif
typedef struct VaapiEngine VaapiEngine;
typedef struct {
uint32_t width; // encode (output) size
uint32_t height;
uint32_t cap_width; // capture (input) size; VPP scales cap -> width
uint32_t cap_height;
uint32_t fps;
uint32_t gop;
int64_t target_bps;
int64_t max_bps; // VBR ceiling (viewer preset maxrate)
int64_t buffer_bits; // HRD buffer size
int is_hevc;
const char *device; // DRM render node (UU_VAAPI_DEVICE); NULL = /dev/dri/renderD128
} VaapiConfig;
VaapiEngine* vaapi_create(const VaapiConfig* config);
// Import a DRI3 / GBM DMA-BUF fd as the zero-copy input surface
int vaapi_import_dma_buffer(VaapiEngine* engine, int dma_fd, uint32_t stride);
// Host fallback: upload ARGB host memory if DMA-BUF is unavailable
int vaapi_upload_host_buffer(VaapiEngine* engine, const uint8_t* host_ptr, uint32_t pitch);
// Encode one frame
int vaapi_encode_frame(
VaapiEngine* engine,
int force_idr,
const uint8_t** out_data,
uint32_t* out_len
);
void vaapi_destroy(VaapiEngine* engine);
#ifdef __cplusplus
}
#endif
#endif // UU_VAAPI_ENGINE_H
+54 -33
View File
@@ -535,7 +535,10 @@ int uu_wl_init(int32_t *w,int32_t *h) {
wl_registry_add_listener(registry,&registry_listener,NULL); wl_registry_add_listener(registry,&registry_listener,NULL);
if(sync_display()!=0 || sync_display()!=0) goto end; if(sync_display()!=0 || sync_display()!=0) goto end;
dma_init(); dma_init();
if(!shm || !seat || !copy_manager || !pointer_manager || !keyboard_manager || !output_count) goto end; /* wlr-virtual-pointer is an optional input path. When the compositor does not
advertise it, pointer injection is skipped and the caller may drive input
through uinput; screencopy + virtual keyboard are still enough to capture. */
if(!shm || !seat || !copy_manager || !output_count) goto end;
const char *name=getenv("UU_WAYLAND_OUTPUT"); selected=0; const char *name=getenv("UU_WAYLAND_OUTPUT"); selected=0;
if(name && *name) { if(name && *name) {
for(selected=0;selected<output_count;selected++) if(!strcmp(outputs[selected].name,name)) break; for(selected=0;selected<output_count;selected++) if(!strcmp(outputs[selected].name,name)) break;
@@ -549,23 +552,30 @@ int uu_wl_init(int32_t *w,int32_t *h) {
power=zwlr_output_power_manager_v1_get_output_power(power_manager,outputs[selected].handle); power=zwlr_output_power_manager_v1_get_output_power(power_manager,outputs[selected].handle);
zwlr_output_power_v1_add_listener(power,&power_listener,NULL); zwlr_output_power_v1_add_listener(power,&power_listener,NULL);
} }
pointer=zwlr_virtual_pointer_manager_v1_create_virtual_pointer_with_output(pointer_manager,seat,outputs[selected].handle); if(pointer_manager) pointer=zwlr_virtual_pointer_manager_v1_create_virtual_pointer_with_output(pointer_manager,seat,outputs[selected].handle);
keyboard=zwp_virtual_keyboard_manager_v1_create_virtual_keyboard(keyboard_manager,seat); /* Some compositors reject a virtual keyboard from anything that is not their active
xkb=xkb_context_new(XKB_CONTEXT_NO_FLAGS); input method (Denial does), and a rejected proxy poisons later requests. Allow the
if(!xkb) goto end; caller to opt out and rely on uinput for key input instead. */
struct xkb_rule_names names={.layout="us"}; const char *skip_kbd=getenv("UU_WAYLAND_NO_VIRTUAL_KEYBOARD");
keymap=xkb_keymap_new_from_names(xkb,&names,XKB_KEYMAP_COMPILE_NO_FLAGS); if(keyboard_manager && !(skip_kbd && *skip_kbd && strcmp(skip_kbd,"0")))
if(!keymap) goto end; keyboard=zwp_virtual_keyboard_manager_v1_create_virtual_keyboard(keyboard_manager,seat);
key_state=xkb_state_new(keymap); if(!key_state) goto end; if(keyboard) {
keymap_text=xkb_keymap_get_as_string(keymap,XKB_KEYMAP_FORMAT_TEXT_V1); xkb=xkb_context_new(XKB_CONTEXT_NO_FLAGS);
if(!keymap_text) goto end; if(!xkb) goto end;
keymap_text_len=strlen(keymap_text)+1; struct xkb_rule_names names={.layout="us"};
if(upload_keymap(keymap_text,keymap_text_len)!=0) goto end; keymap=xkb_keymap_new_from_names(xkb,&names,XKB_KEYMAP_COMPILE_NO_FLAGS);
if(sync_display()!=0) goto end; if(!keymap) goto end;
/* Text typing needs unmapped keycodes and a place to declare them. A keymap without either key_state=xkb_state_new(keymap); if(!key_state) goto end;
only costs that path - uu_wl_type_utf8 refuses and the caller keeps its own fallback. */ keymap_text=xkb_keymap_get_as_string(keymap,XKB_KEYMAP_FORMAT_TEXT_V1);
keymap_insert=symbols_close(keymap_text); if(!keymap_text) goto end;
spare_count=find_spare_codes(spare_codes,sizeof(spare_codes)/sizeof(spare_codes[0])); keymap_text_len=strlen(keymap_text)+1;
if(upload_keymap(keymap_text,keymap_text_len)!=0) goto end;
if(sync_display()!=0) goto end;
/* Text typing needs unmapped keycodes and a place to declare them. A keymap without either
only costs that path - uu_wl_type_utf8 refuses and the caller keeps its own fallback. */
keymap_insert=symbols_close(keymap_text);
spare_count=find_spare_codes(spare_codes,sizeof(spare_codes)/sizeof(spare_codes[0]));
}
initialized=1; *w=width; *h=height; result=0; initialized=1; *w=width; *h=height; result=0;
end: end:
if(result!=0) reset(); if(result!=0) reset();
@@ -581,9 +591,12 @@ int uu_wl_select_output(const char *name,uint32_t expected_w,uint32_t expected_h
if(target==output_count || outputs[target].transform!=WL_OUTPUT_TRANSFORM_NORMAL) goto end; if(target==output_count || outputs[target].transform!=WL_OUTPUT_TRANSFORM_NORMAL) goto end;
uint32_t w=outputs[target].width,h=outputs[target].height; uint32_t w=outputs[target].width,h=outputs[target].height;
if(w!=expected_w || h!=expected_h || !w || !h || w>16384 || h>16384 || (uint64_t)w*h*4>256*1024*1024) goto end; if(w!=expected_w || h!=expected_h || !w || !h || w>16384 || h>16384 || (uint64_t)w*h*4>256*1024*1024) goto end;
struct zwlr_virtual_pointer_v1 *next=zwlr_virtual_pointer_manager_v1_create_virtual_pointer_with_output(pointer_manager,seat,outputs[target].handle); struct zwlr_virtual_pointer_v1 *next=NULL;
if(!next) goto end; if(pointer_manager) {
if(sync_display()!=0) {zwlr_virtual_pointer_v1_destroy(next);goto end;} next=zwlr_virtual_pointer_manager_v1_create_virtual_pointer_with_output(pointer_manager,seat,outputs[target].handle);
if(!next) goto end;
if(sync_display()!=0) {zwlr_virtual_pointer_v1_destroy(next);goto end;}
}
for(unsigned i=0;i<2;i++) { frame_clear(&frames[i]); dma_clear(&frames[i]); shm_clear(&frames[i]); } for(unsigned i=0;i<2;i++) { frame_clear(&frames[i]); dma_clear(&frames[i]); shm_clear(&frames[i]); }
pending=0; pending=0;
if(pointer) zwlr_virtual_pointer_v1_destroy(pointer); if(pointer) zwlr_virtual_pointer_v1_destroy(pointer);
@@ -654,6 +667,8 @@ end:
} }
int uu_wl_move(int32_t x,int32_t y,int relative) { int uu_wl_move(int32_t x,int32_t y,int relative) {
pthread_mutex_lock(&lock); int result=-1; pthread_mutex_lock(&lock); int result=-1;
/* -2: no virtual pointer at all, so the caller may use uinput; -1 stays a real failure. */
if(initialized && !pointer) { result=-2; goto end; }
if(initialized) { if(initialized) {
if(relative) { if(relative) {
/* UU reports pixels (source DPR=1); Wayland motion is in logical points. */ /* UU reports pixels (source DPR=1); Wayland motion is in logical points. */
@@ -668,6 +683,7 @@ end: pthread_mutex_unlock(&lock); return result;
} }
int uu_wl_button(int button,int pressed) { int uu_wl_button(int button,int pressed) {
pthread_mutex_lock(&lock); int result=-1; pthread_mutex_lock(&lock); int result=-1;
if(initialized && !pointer) { goto end; } /* no virtual pointer: report failure so the caller falls back to uinput */
if(initialized) { if(initialized) {
if(button>=4 && button<=7) { if(button>=4 && button<=7) {
if(pressed) { if(pressed) {
@@ -687,6 +703,7 @@ end: pthread_mutex_unlock(&lock); return result;
} }
int uu_wl_key(uint64_t symbol,int pressed) { int uu_wl_key(uint64_t symbol,int pressed) {
pthread_mutex_lock(&lock); int result=-1; pthread_mutex_lock(&lock); int result=-1;
if(initialized && !keyboard) { pthread_mutex_unlock(&lock); return -1; } /* no virtual keyboard: caller falls back to uinput */
if(initialized) { if(initialized) {
for(xkb_keycode_t code=xkb_keymap_min_keycode(keymap);code<=xkb_keymap_max_keycode(keymap);code++) { for(xkb_keycode_t code=xkb_keymap_min_keycode(keymap);code<=xkb_keymap_max_keycode(keymap);code++) {
int matches=0; int matches=0;
@@ -717,20 +734,24 @@ int uu_wl_release_all(void) {
if(initialized) { if(initialized) {
/* Sweep this client's virtual keyboard directly: a keysym lookup can /* Sweep this client's virtual keyboard directly: a keysym lookup can
fail for unmapped OEM/keypad symbols before reaching held modifiers. */ fail for unmapped OEM/keypad symbols before reaching held modifiers. */
for(xkb_keycode_t code=xkb_keymap_min_keycode(keymap);code<=xkb_keymap_max_keycode(keymap);code++) { if(keyboard) {
if(code<8 || code>=sizeof(held_keys) || !xkb_keymap_key_get_name(keymap,code)) continue; for(xkb_keycode_t code=xkb_keymap_min_keycode(keymap);code<=xkb_keymap_max_keycode(keymap);code++) {
zwp_virtual_keyboard_v1_key(keyboard,timestamp(),code-8,WL_KEYBOARD_KEY_STATE_RELEASED); if(code<8 || code>=sizeof(held_keys) || !xkb_keymap_key_get_name(keymap,code)) continue;
if(held_keys[code]) xkb_state_update_key(key_state,code,XKB_KEY_UP); zwp_virtual_keyboard_v1_key(keyboard,timestamp(),code-8,WL_KEYBOARD_KEY_STATE_RELEASED);
if(held_keys[code]) xkb_state_update_key(key_state,code,XKB_KEY_UP);
}
zwp_virtual_keyboard_v1_modifiers(keyboard,
xkb_state_serialize_mods(key_state,XKB_STATE_MODS_DEPRESSED),
xkb_state_serialize_mods(key_state,XKB_STATE_MODS_LATCHED),
xkb_state_serialize_mods(key_state,XKB_STATE_MODS_LOCKED),
xkb_state_serialize_layout(key_state,XKB_STATE_LAYOUT_EFFECTIVE));
} }
memset(held_keys,0,sizeof(held_keys)); memset(held_keys,0,sizeof(held_keys));
zwp_virtual_keyboard_v1_modifiers(keyboard, if(pointer) {
xkb_state_serialize_mods(key_state,XKB_STATE_MODS_DEPRESSED), for(unsigned code=0x110;code<=0x114;code++)
xkb_state_serialize_mods(key_state,XKB_STATE_MODS_LATCHED), zwlr_virtual_pointer_v1_button(pointer,timestamp(),code,WL_POINTER_BUTTON_STATE_RELEASED);
xkb_state_serialize_mods(key_state,XKB_STATE_MODS_LOCKED), zwlr_virtual_pointer_v1_frame(pointer);
xkb_state_serialize_layout(key_state,XKB_STATE_LAYOUT_EFFECTIVE)); }
for(unsigned code=0x110;code<=0x114;code++)
zwlr_virtual_pointer_v1_button(pointer,timestamp(),code,WL_POINTER_BUTTON_STATE_RELEASED);
zwlr_virtual_pointer_v1_frame(pointer);
result=wl_display_flush(display)<0?-1:0; result=wl_display_flush(display)<0?-1:0;
} }
pthread_mutex_unlock(&lock);return result; pthread_mutex_unlock(&lock);return result;
+1 -1
View File
@@ -1,7 +1,7 @@
# Unofficial UU Remote Linux host in a container, for NAS boxes (Synology DSM, etc.) without a desktop. # Unofficial UU Remote Linux host in a container, for NAS boxes (Synology DSM, etc.) without a desktop.
# Headless mode: file transfer, terminal and port mapping; no remote desktop. Uses the published release binary. # Headless mode: file transfer, terminal and port mapping; no remote desktop. Uses the published release binary.
FROM ubuntu:24.04 FROM ubuntu:24.04
ARG VERSION=v0.2.19 ARG VERSION=v0.2.20
LABEL org.opencontainers.image.source="https://github.com/VallierDev/uu-remote-linux" \ LABEL org.opencontainers.image.source="https://github.com/VallierDev/uu-remote-linux" \
org.opencontainers.image.description="Unofficial UU Remote Linux host, headless (file transfer, terminal, port mapping)" \ org.opencontainers.image.description="Unofficial UU Remote Linux host, headless (file transfer, terminal, port mapping)" \
org.opencontainers.image.licenses="GPL-3.0" org.opencontainers.image.licenses="GPL-3.0"
+18
View File
@@ -0,0 +1,18 @@
扫码登录改为与官方客户端一致,修复二维码刚生成就失效。/ QR sign-in now matches the official client.
### 修复 / Fixed
- **登录二维码反复立即失效(1190)**(#68,感谢 @axlrose 的详细排查):逆向官方 Mac 客户端(4.42)后确认,新版服务端在生成二维码时
返回 `status_query_ticket`,官方客户端查询扫码状态和确认登录时都用它(放在 `token` 键下);我们此前优先发送旧的 `token` 字段。现在与官方一致:
- 生成二维码不带请求体;状态查询用 `status_query_ticket`(服务端没有时才用旧 `token`);
- 状态查询是长轮询,每次应答后立即再问,单次超时 60 秒(此前 12 秒);
- 状态 3 立即换新码、状态 0 在 0.35 秒后换新码、状态 4 登录(状态 3 不再去调确认登录);
- 出错(如 HTTP 400 / 1190)或超时时停止轮询,等用户点击登录窗口新增的「刷新二维码」按钮(终端里按回车;既无窗口也无终端时 60 秒后自动换码)。
取代 v0.2.19 的临时退避和「重新注册设备」提示。已用全新设备身份实测:状态查询正常挂起,60 秒后按官方行为停止并等待刷新;
未在 #68 那台被标记的设备上实测,「刷新二维码」按钮的界面尚未在桌面上目视验收。
- QR sign-in mirrors the official Mac client: `status_query_ticket` as the status ticket, 60 s long poll, regenerate on status 3/0,
stop on errors (e.g. code 1190) until the user clicks the new “refresh QR” button.
### 下载 / Download
`uu-remote-linux_0.2.20_<amd64|arm64>.deb`,或 `uu-remote-linux-v0.2.20-<x86_64|aarch64>-linux.tar.gz`,或 `docker pull ghcr.io/vallierdev/uu-remote-linux:v0.2.20`(x86_64)。
**非官方项目,与网易无关。** / **Unofficial, not affiliated with NetEase.**
+19
View File
@@ -0,0 +1,19 @@
Wayland 合成器缺少虚拟指针或虚拟键盘时,桌面仍可使用;systemd 单元示例与 deb 包一致。/ Wayland desktops without virtual pointer/keyboard protocols still come up.
### 修复 / Fixed
- **只缺 wlr-virtual-pointer 或虚拟键盘受限的合成器整个桌面不可用**(#72、#73,感谢 @clionertr):Denial 等合成器不提供
`zwlr_virtual_pointer_manager_v1`,或只把虚拟键盘给它自己的输入法。现在这两个协议都是可选的:画面采集照常启动,
缺指针时绝对移动改走 uinput(需能访问 `/dev/uinput`,见 README);`UU_WAYLAND_NO_VIRTUAL_KEYBOARD=1` 可跳过虚拟键盘。
只有「完全没有虚拟指针」才回退 uinput;坐标超出当前采集输出等其他失败照旧返回,不会把指针移到别的屏幕(#78)。
CI 构建与测试通过,**未在 Denial 上实机验证**;Hyprland 等已有虚拟指针的合成器行为不变。
- Virtual pointer and keyboard are optional on Wayland; capture still starts and pointer motion falls back to uinput only when
the compositor has no wlr-virtual-pointer. Not tested on Denial.
### 文档 / Docs
- README 的用户 systemd 单元与 deb 包一致:不再写死 `DISPLAY=:0`,随 `graphical-session.target` 启动;无头模式、虚拟桌面和
Xfce/i3 等不启动该 target 的桌面改用 `WantedBy=default.target`(#69,感谢 @ferstar)。
### 下载 / Download
`uu-remote-linux_0.2.21_<amd64|arm64>.deb`,或 `uu-remote-linux-v0.2.21-<x86_64|aarch64>-linux.tar.gz`,或 `docker pull ghcr.io/vallierdev/uu-remote-linux:v0.2.21`(x86_64)。
**非官方项目,与网易无关。** / **Unofficial, not affiliated with NetEase.**
+20
View File
@@ -0,0 +1,20 @@
官方控制端能选「真彩」(4:4:4) 了;自行编译的 `inproc-av` 新增进程内 NVENC 与 AMD/Intel VAAPI。/ The official viewer can select true color (4:4:4); self-built `inproc-av` gains in-process NVENC and VAAPI.
### 修复 / Fixed
- **官方控制端里「真彩」不可选或打开后马上被切回**(#77、#79):设备能力里 HEVC/4:4:4 等格式的宽高原来填的是 0x0,官方控制端会把这样的条目当作不支持。现在每种格式都带上最大尺寸 3840x2160。
已实机验收:官方 Mac 控制端连 X11 + NVIDIA 主机,打开真彩后切到 HEVC 4:4:4,持续 30 秒以上没有被切回,也没有请求关键帧(PLI)。
- The device capability listed HEVC/4:4:4 formats as 0x0, which the official viewer treats as unsupported; each format now
carries its maximum size. Verified with the official Mac viewer: true color stays on (HEVC 4:4:4) for 30+ s, no PLI.
### 新增(需自行编译,预编译包不受影响)/ Added (self-built only; release packages unchanged)
- `--features inproc-av`(仅 X11):
- **NVENC 进程内编码**(#75、#77,感谢 @ferstar):直接读取 XShm,支持 4:4:4,不再经过 FFmpeg 进程;失败时依次回退到 FFmpeg 的 NVENC 管线、CPU。我方 NVIDIA 机器未单独验收这条进程内路径。
- **VAAPI 进程内编码**(#80、#81,感谢 @ferstar):通过 DRI3 DMA-BUF 零拷贝取屏(没有 DRI3 时改用 XShm 上传),经 VPP 转成 NV12 后送入硬件编码,支持 H.264 和 H.265。贡献者已在 AMD HawkPoint、Cezanne 上用官方 Mac 控制端实机验收;Intel 与 4K 尚未验证。需设置 `UU_VIDEO_ENCODER=vaapi`。
- libcuda、libva 都在运行时加载;编译所需的头文件见 README。
- In-process NVENC (XShm, 4:4:4) and VAAPI (DRI3 DMA-BUF zero copy, H.264/H.265) under `--features inproc-av` on X11,
each falling back to the FFmpeg process pipeline. VAAPI accepted by the contributor on AMD HawkPoint/Cezanne; Intel and 4K untested.
### 下载 / Download
`uu-remote-linux_0.2.22_<amd64|arm64>.deb`,或 `uu-remote-linux-v0.2.22-<x86_64|aarch64>-linux.tar.gz`,或 `docker pull ghcr.io/vallierdev/uu-remote-linux:v0.2.22`(x86_64)。
**非官方项目,与网易无关。** / **Unofficial, not affiliated with NetEase.**
+13
View File
@@ -0,0 +1,13 @@
长时间连接后画面冻结的修复。/ Fixes video freezing after long sessions.
### 修复 / Fixed
- **控制端切到后台再恢复后画面冻结、延迟显示 0**(#83,感谢 @AYssu 定位并提供补丁):RTP 序号每 65536 包回绕一次;
回绕后如果控制端补要(NACK)回绕前的旧包,webrtc-srtp 0.12.0 会把加密状态退回旧值,之后的新包都用错误的回绕计数加密,
控制端全部认证失败,画面停住且不再恢复。现按 RFC 3711 附录 A 估算回绕计数,重传不再让状态倒退。加密算法和报文格式不变。
回归测试在原库上失败、修复后通过(AES-CM/HMAC 与 AES-GCM);报告者在 Windows 控制端实测后台放置恢复正常,跨过两次回绕。
- SRTP rollover counter no longer rewinds when a pre-wrap packet is retransmitted after the wrap (frozen video, #83).
### 下载 / Download
`uu-remote-linux_0.2.23_<amd64|arm64>.deb`,或 `uu-remote-linux-v0.2.23-<x86_64|aarch64>-linux.tar.gz`,或 `docker pull ghcr.io/vallierdev/uu-remote-linux:v0.2.23`(x86_64)。
**非官方项目,与网易无关。** / **Unofficial, not affiliated with NetEase.**
+1 -10
View File
@@ -1,7 +1,6 @@
//! Desktop sound for the viewer's 声音 menu: the default output's monitor (PulseAudio or //! Desktop sound for the viewer's 声音 menu: the default output's monitor (PulseAudio or
//! PipeWire-pulse) is encoded to Opus and sent on the negotiated audio m-line. //! PipeWire-pulse) is encoded to Opus and sent on the negotiated audio m-line.
//! Encoding runs in this process with the `inproc-av` feature (unless `UU_FFMPEG_ROOT` is set), //! An FFmpeg child process does the encoding.
//! otherwise an FFmpeg child process does it.
//! PipeWire monitors ignore the sink's mute by default, so 被控端静音运行 keeps the stream audible. //! PipeWire monitors ignore the sink's mute by default, so 被控端静音运行 keeps the stream audible.
//! `UU_AUDIO=0` disables it; `UU_AUDIO_SOURCE` picks another Pulse source. //! `UU_AUDIO=0` disables it; `UU_AUDIO_SOURCE` picks another Pulse source.
@@ -57,14 +56,6 @@ fn read_packets(stdout: impl Read, tx: &tokio::sync::mpsc::Sender<bytes::Bytes>)
fn start(source: &str, tx: tokio::sync::mpsc::Sender<bytes::Bytes>) fn start(source: &str, tx: tokio::sync::mpsc::Sender<bytes::Bytes>)
-> Result<(Stopper, tokio::task::JoinHandle<Result<(), String>>), String> { -> Result<(Stopper, tokio::task::JoinHandle<Result<(), String>>), String> {
let stop = Arc::new(AtomicBool::new(false)); let stop = Arc::new(AtomicBool::new(false));
#[cfg(feature = "inproc-av")]
if crate::inproc_encoder::selected() {
let job = tokio::task::spawn_blocking({
let (stop, source) = (stop.clone(), source.to_owned());
move || crate::inproc_encoder::run_audio(&source, tx, stop)
});
return Ok((Stopper { stop, child: None }, job));
}
let mut child = spawn_encoder(source)?; let mut child = spawn_encoder(source)?;
let stdout = child.stdout.take().ok_or("Encoder stdout missing")?; let stdout = child.stdout.take().ok_or("Encoder stdout missing")?;
let job = tokio::task::spawn_blocking(move || read_packets(stdout, &tx)); let job = tokio::task::spawn_blocking(move || read_packets(stdout, &tx));
+11
View File
@@ -251,6 +251,17 @@ impl UUAuthClient {
Ok((status, value)) Ok((status, value))
} }
/// Signed POST with an optional JSON body (None = empty body, like Moya `requestPlain`) and an
/// explicit timeout. The official clients long-poll the QR status with a 60 s request timeout.
pub async fn post_with_timeout(&self, path: &str, body: Option<&serde_json::Value>, timeout: std::time::Duration) -> Result<(u16, serde_json::Value), String> {
let body_str = body.map(serde_json::to_string).transpose().map_err(|e| e.to_string())?.unwrap_or_default();
let resp = self.http.post(format!("{}{}", BASE_URL, path)).timeout(timeout)
.headers(self.build_headers("POST", path, &body_str)).body(body_str)
.send().await.map_err(|e| format!("HTTP post {path} failed: {}", transport_error(&e)))?;
let status = resp.status().as_u16();
Ok((status, resp.json().await.unwrap_or(serde_json::Value::Null)))
}
/// Signed JSON POST to another API host (the signature covers the path only). /// Signed JSON POST to another API host (the signature covers the path only).
pub async fn post_json_url(&self, base: &str, path: &str, body: &serde_json::Value) -> Result<(u16, serde_json::Value), String> { pub async fn post_json_url(&self, base: &str, path: &str, body: &serde_json::Value) -> Result<(u16, serde_json::Value), String> {
let body_str = serde_json::to_string(body).map_err(|e| e.to_string())?; let body_str = serde_json::to_string(body).map_err(|e| e.to_string())?;
+4 -3
View File
@@ -35,12 +35,13 @@ pub fn h264_device_capability(
/// The payload sits under `data.device_capability` with `type` beside it. The earlier flat shape /// The payload sits under `data.device_capability` with `type` beside it. The earlier flat shape
/// (`type` at the top, fields directly in `data`) was not recognised by the iPhone client, which /// (`type` at the top, fields directly in `data`) was not recognised by the iPhone client, which
/// then kept quality and chroma locked at their defaults (#49). Only 8-bit and the chroma formats /// then kept quality and chroma locked at their defaults (#49). Only 8-bit and the chroma formats
/// the encoder really produces are listed; the first entry carries the maximum size as officially. /// the encoder really produces are listed, each with its maximum size: the official viewer treats a
/// 0x0 entry as unsupported (#77; the 4.37 host we copied had a GTX 660 without HEVC/4:4:4).
pub fn device_capability(client_id: &str, ice_id: &str, display_id: i32, fps: i32, codec_impl: i32, hevc: bool, full_chroma: bool) -> Value { pub fn device_capability(client_id: &str, ice_id: &str, display_id: i32, fps: i32, codec_impl: i32, hevc: bool, full_chroma: bool) -> Value {
let mut codecs = Vec::new(); let mut codecs = Vec::new();
for codec in [VIDEO_CODEC_H264].into_iter().chain(hevc.then_some(2)) { for codec in [VIDEO_CODEC_H264].into_iter().chain(hevc.then_some(2)) {
for chroma in [1].into_iter().chain(full_chroma.then_some(3)) { for chroma in [1].into_iter().chain(full_chroma.then_some(3)) {
let (width, height) = if codecs.is_empty() { (3840, 2160) } else { (0, 0) }; let (width, height) = (3840, 2160);
codecs.push(json!({"video_codec": codec, "chroma_sampling": chroma, "codec_impl": codec_impl, codecs.push(json!({"video_codec": codec, "chroma_sampling": chroma, "codec_impl": codec_impl,
"bit_depth": 8, "width": width, "height": height})); "bit_depth": 8, "width": width, "height": height}));
} }
@@ -86,7 +87,7 @@ mod tests {
let caps = &payload["data"]["device_capability"]["video_codec_capability"]; let caps = &payload["data"]["device_capability"]["video_codec_capability"];
assert_eq!(caps.as_array().unwrap().len(), 4); assert_eq!(caps.as_array().unwrap().len(), 4);
assert_eq!(caps[0], json!({"video_codec":1,"chroma_sampling":1,"codec_impl":5,"bit_depth":8,"width":3840,"height":2160})); assert_eq!(caps[0], json!({"video_codec":1,"chroma_sampling":1,"codec_impl":5,"bit_depth":8,"width":3840,"height":2160}));
assert_eq!(caps[3], json!({"video_codec":2,"chroma_sampling":3,"codec_impl":5,"bit_depth":8,"width":0,"height":0})); assert_eq!(caps[3], json!({"video_codec":2,"chroma_sampling":3,"codec_impl":5,"bit_depth":8,"width":3840,"height":2160}));
assert_eq!(payload["data"]["device_capability"]["display_info"][0]["hdr"], -1); assert_eq!(payload["data"]["device_capability"]["display_info"][0]["hdr"], -1);
let h264_420 = device_capability("c", "i", 0, 60, 5, false, false); let h264_420 = device_capability("c", "i", 0, 60, 5, false, false);
assert_eq!(h264_420["data"]["device_capability"]["video_codec_capability"].as_array().unwrap().len(), 1); assert_eq!(h264_420["data"]["device_capability"]["video_codec_capability"].as_array().unwrap().len(), 1);
+10
View File
@@ -70,6 +70,16 @@ impl DamageCapture {
pub fn size(&self) -> (u32, u32) { (self.width as u32, self.height as u32) } pub fn size(&self) -> (u32, u32) { (self.width as u32, self.height as u32) }
pub fn shm_raw_ptr(&self) -> *mut u8 { self.shm_ptr }
pub fn shm_size(&self) -> usize { self.len }
pub fn refresh(&mut self) -> Result<(), String> {
self.conn.shm_get_image(self.root, 0, 0, self.width, self.height, !0, Z_PIXMAP, self.seg, 0)
.map_err(text)?.reply().map_err(text)?;
Ok(())
}
/// Waits up to `timeout` for the screen to change. Returns `true` on change, `false` on timeout. /// Waits up to `timeout` for the screen to change. Returns `true` on change, `false` on timeout.
pub fn wait(&mut self, timeout: Duration) -> Result<bool, String> { pub fn wait(&mut self, timeout: Duration) -> Result<bool, String> {
let deadline = Instant::now() + timeout; let deadline = Instant::now() + timeout;
+671 -397
View File
File diff suppressed because it is too large. Load diff
+4 -1
View File
@@ -328,8 +328,11 @@ impl InputInjector {
// uinput ABS axes span the compositor's whole desktop. A captured output // uinput ABS axes span the compositor's whole desktop. A captured output
// uses its own pixel space; bind absolute motion to that Wayland output, // uses its own pixel space; bind absolute motion to that Wayland output,
// including when uinput remains in use for keyboard/buttons/relative input. // including when uinput remains in use for keyboard/buttons/relative input.
// Only a compositor without wlr-virtual-pointer (-2) falls through to the uinput
// absolute pointer; an out-of-range point must not land on another output.
if !relative && self.backend==crate::desktop_backend::DesktopBackend::Wayland { if !relative && self.backend==crate::desktop_backend::DesktopBackend::Wayland {
return crate::wayland::move_pointer(x,y,false); let result=crate::wayland::move_pointer(x,y,false);
if result!=-2 { return result; }
} }
if let Some(uinput)=&self.uinput { if let Some(uinput)=&self.uinput {
if relative { return Self::uinput_result(uinput.move_relative(x,y)); } if relative { return Self::uinput_result(uinput.move_relative(x,y)); }
+65 -50
View File
@@ -268,14 +268,22 @@ pub async fn prepare_device() -> Result<(Value, Credentials), String> {
pub(crate) async fn flow(ui: &LoginUi) -> Result<String, String> { pub(crate) async fn flow(ui: &LoginUi) -> Result<String, String> {
let (mut cfg, creds) = prepare_device().await?; let (mut cfg, creds) = prepare_device().await?;
let client = UUAuthClient::new(creds.clone()); let client = UUAuthClient::new(creds.clone());
let mut instant_expiries = 0u32; // Mirrors the official Mac client (LoginViewController / LoginManager, UURemote 4.42): gen/login has
// no body; the status long poll sends gen's `status_query_ticket` under `token` with the last
// status, re-asks as soon as an answer arrives (60 s request timeout), regenerates on status 3
// at once and on status 0 after 0.35 s, logs in on 4, and on any error (e.g. HTTP 400 code 1190)
// stops until the user asks for a new QR. Sending gen's older `token` field made the cloud reject
// every QR within seconds (#68).
let timeout = Duration::from_secs(60);
loop { loop {
let (status, resp) = client.post_json("/api/v1/qrcode/gen/login", &json!({})).await?; let (status, resp) = client.post_with_timeout("/api/v1/qrcode/gen/login", None, timeout).await?;
let d = data("qrcode gen", status, &resp)?; let d = match data("qrcode gen", status, &resp) {
Ok(d) => d.clone(),
Err(e) => { println!("QR code could not be loaded ({e})"); wait_for_refresh(ui, "二维码加载失败").await; continue; }
};
let url = d.get("qrcode_jump_url").and_then(|v| v.as_str()).ok_or("qrcode gen returned no qrcode_jump_url")?.to_string(); let url = d.get("qrcode_jump_url").and_then(|v| v.as_str()).ok_or("qrcode gen returned no qrcode_jump_url")?.to_string();
let ticket = d.get("token").or_else(|| d.get("status_query_ticket")).and_then(|v| v.as_str()) let ticket = qr_ticket(&d).ok_or("qrcode gen returned no status ticket")?.to_string();
.ok_or("qrcode gen returned no token")?.to_string(); ui.offer_refresh(false);
if ui.shown() { if ui.shown() {
let modules = qrcode::QrCode::new(url.as_bytes()).map_err(|e| e.to_string())?.width() + 2; let modules = qrcode::QrCode::new(url.as_bytes()).map_err(|e| e.to_string())?.width() + 2;
let (rgba, edge) = qr_rgba(&url, (224 / modules).max(2), 1)?; let (rgba, edge) = qr_rgba(&url, (224 / modules).max(2), 1)?;
@@ -290,48 +298,31 @@ pub(crate) async fn flow(ui: &LoginUi) -> Result<String, String> {
file.write_all(&crate::cursor::encode_png(edge as u32,edge as u32,&rgba)).map_err(|e|e.to_string())?; file.write_all(&crate::cursor::encode_png(edge as u32,edge as u32,&rgba)).map_err(|e|e.to_string())?;
} }
if !ui.quiet() { println!("\nScan with the UU Remote mobile app (网易UU远程 App 扫码登录):\n{}", render_qr(&url)?); } if !ui.quiet() { println!("\nScan with the UU Remote mobile app (网易UU远程 App 扫码登录):\n{}", render_qr(&url)?); }
// Status is a long poll: send the state already shown (server enum 1..4; observed:
// 1 waiting, 2 scanned, 4 after the phone confirmed); the server answers when it changes.
let mut login_status = 1i64; let mut login_status = 1i64;
let mut expired = false; let failure = loop {
let generated = std::time::Instant::now();
for _ in 0..150 {
let query = json!({"token": ticket, "qrcode_jump_url": url, "login_status": login_status}); let query = json!({"token": ticket, "qrcode_jump_url": url, "login_status": login_status});
let asked = std::time::Instant::now(); let (status, resp) = match client.post_with_timeout("/api/v1/qrcode/login/status", Some(&query), timeout).await {
let (status, resp) = match client.post_json("/api/v1/qrcode/login/status", &query).await {
Ok(r) => r, Ok(r) => r,
// Long-poll timeout: ask again, but never spin on transport errors (#68). Err(_) => break "网络超时,二维码已停止刷新",
Err(_) => { if asked.elapsed() < Duration::from_secs(1) { tokio::time::sleep(Duration::from_secs(2)).await; } continue; }
}; };
// Code 1190 (HTTP 400, "二维码已失效"): the QR expired; generate a new one. let code = resp.get("code").and_then(|c| c.as_i64());
if resp.get("code").and_then(|c| c.as_i64()) == Some(QR_EXPIRED) { if code == Some(QR_EXPIRED) { break "二维码已失效"; }
ui.status("二维码已过期,正在刷新…", Tone::Info); let Ok(d) = data("qrcode status", status, &resp) else { break "二维码状态查询失败"; };
expired = true; break; let st = d.get("status").and_then(|v| v.as_i64()).unwrap_or(login_status);
}
let st = data("qrcode status", status, &resp)?.get("status").and_then(|v| v.as_i64()).unwrap_or(login_status);
if st == login_status {
// The server normally holds this request; an immediate unchanged answer must not
// turn into a tight loop of dozens of requests per second (#68).
if asked.elapsed() < Duration::from_secs(1) { tokio::time::sleep(Duration::from_secs(2)).await; }
continue;
}
login_status = st;
match st { match st {
2 => { 1 => {}
2 => if login_status != 2 {
println!("已扫码,请在手机上确认 (scanned, confirm on the phone)"); println!("已扫码,请在手机上确认 (scanned, confirm on the phone)");
ui.status("已扫码,请在手机上确认登录", Tone::Ok); ui.status("已扫码,请在手机上确认登录", Tone::Ok);
} },
3 | 4 => { 3 => { println!("QR code expired; generating a new one"); break ""; }
// A final state: confirmed logs in; cancelled/expired is refused and a new QR follows. 0 => { tokio::time::sleep(Duration::from_millis(350)).await; break ""; }
4 => {
let body = json!({"token": ticket, "qrcode_jump_url": url}); let body = json!({"token": ticket, "qrcode_jump_url": url});
let (status, resp) = client.post_json("/api/v1/login/by_qrcode", &body).await?; let (status, resp) = client.post_with_timeout("/api/v1/login/by_qrcode", Some(&body), timeout).await?;
let d = match data("qrcode login", status, &resp) { let d = match data("qrcode login", status, &resp) {
Ok(d) => d, Ok(d) => d,
Err(e) => { Err(e) => { println!("QR login not granted ({e})"); break "登录失败"; }
println!("QR status {st}: login not granted ({e})");
ui.status("手机端未确认登录,正在刷新二维码…", Tone::Error);
expired = true; break;
}
}; };
let token = d.get("token").and_then(|v| v.as_str()).ok_or("login returned no token")?; let token = d.get("token").and_then(|v| v.as_str()).ok_or("login returned no token")?;
let user = d.get("user_id").and_then(|v| v.as_str()).ok_or("login returned no user_id")?; let user = d.get("user_id").and_then(|v| v.as_str()).ok_or("login returned no user_id")?;
@@ -348,25 +339,49 @@ pub(crate) async fn flow(ui: &LoginUi) -> Result<String, String> {
println!("Logged in as {} ({}); saved to {}", nickname, user, path.display()); println!("Logged in as {} ({}); saved to {}", nickname, user, path.display());
return Ok(if nickname.is_empty() { user.to_string() } else { nickname.to_string() }); return Ok(if nickname.is_empty() { user.to_string() } else { nickname.to_string() });
} }
_ => { println!("QR status: {st}"); ui.status("二维码已失效,正在刷新…", Tone::Error); expired = true; break; } other => { println!("QR status: {other}"); break "二维码已失效"; }
} }
} login_status = st;
println!("{}", if expired { "QR code expired; generating a new one" } else { "QR code timed out; generating a new one" }); };
// A QR rejected within seconds of being generated, again and again, was seen when the cloud if !failure.is_empty() { wait_for_refresh(ui, failure).await; }
// flagged this device identity (#68): back off and say how to recover.
if expired && generated.elapsed() < Duration::from_secs(10) { instant_expiries += 1; } else { instant_expiries = 0; }
if instant_expiries == 3 {
eprintln!("The QR code keeps expiring immediately. If this persists, the cloud may reject this device identity: \
back up and remove ~/.config/uuremote/config.json, then run `uuremote-server login` again to register a new device.");
ui.status("二维码反复立即失效:可备份并删除 ~/.config/uuremote/config.json 后重新登录(会注册新设备)", Tone::Error);
}
tokio::time::sleep(Duration::from_secs(1u64 << instant_expiries.min(5))).await; // never spin if the server keeps refusing
} }
} }
/// gen/login's status ticket: `status_query_ticket` (current cloud and official clients), else the
/// older `token` field for servers that do not send it.
fn qr_ticket(gen: &Value) -> Option<&str> {
gen.get("status_query_ticket").or_else(|| gen.get("token")).and_then(|v| v.as_str()).filter(|t| !t.is_empty())
}
/// Stop like the official client and wait for the user: the 刷新二维码 button, Enter in a terminal,
/// or (no window and no terminal, e.g. a service) one minute.
async fn wait_for_refresh(ui: &LoginUi, reason: &str) {
use std::io::IsTerminal;
ui.status(&format!("{reason},请点击「刷新二维码」"), Tone::Error);
ui.offer_refresh(true);
let terminal = !ui.quiet() && std::io::stdin().is_terminal();
if terminal { println!("{reason}. Press Enter to show a new QR code (按回车刷新二维码)"); }
else { println!("{reason}; waiting for 刷新二维码{}", if ui.shown() { "" } else { " (a new QR code follows in 60 s)" }); }
let mut line = terminal.then(|| tokio::task::spawn_blocking(|| { let mut s = String::new(); let _ = std::io::stdin().read_line(&mut s); }));
let started = std::time::Instant::now();
loop {
if ui.take_refresh() || (ui.shown() && ui.closed()) { break; }
if let Some(handle) = line.as_mut() { if handle.is_finished() { break; } }
if !ui.shown() && !terminal && started.elapsed() >= Duration::from_secs(60) { break; }
tokio::time::sleep(Duration::from_millis(200)).await;
}
ui.offer_refresh(false);
}
#[cfg(test)] #[cfg(test)]
mod tests { mod tests {
use super::*; use super::*;
#[test]
fn status_ticket_prefers_status_query_ticket() {
assert_eq!(qr_ticket(&json!({"token": "old", "status_query_ticket": "new"})), Some("new"));
assert_eq!(qr_ticket(&json!({"token": "old"})), Some("old"));
assert_eq!(qr_ticket(&json!({"status_query_ticket": ""})), None);
}
#[tokio::test] #[tokio::test]
async fn service_waits_until_a_new_token_is_saved() { async fn service_waits_until_a_new_token_is_saved() {
let dir = std::env::temp_dir().join(format!("uu-login-wait-{}", std::process::id())); let dir = std::env::temp_dir().join(format!("uu-login-wait-{}", std::process::id()));
+15
View File
@@ -10,6 +10,8 @@ extern "C" {
fn uu_login_ui_set_done(text: *const libc::c_char); fn uu_login_ui_set_done(text: *const libc::c_char);
fn uu_login_ui_closed() -> libc::c_int; fn uu_login_ui_closed() -> libc::c_int;
fn uu_login_ui_close(); fn uu_login_ui_close();
fn uu_login_ui_offer_refresh(visible: libc::c_int);
fn uu_login_ui_take_refresh() -> libc::c_int;
} }
#[derive(Clone, Copy)] #[derive(Clone, Copy)]
@@ -56,6 +58,19 @@ impl LoginUi {
} }
#[allow(unused_variables)] #[allow(unused_variables)]
/// Show or hide the 刷新二维码 button (a failed QR waits for the user, like the official client).
pub fn offer_refresh(&self, visible: bool) {
#[cfg(uu_gtk)]
if self.shown { unsafe { uu_login_ui_offer_refresh(libc::c_int::from(visible)) } }
#[cfg(not(uu_gtk))]
let _ = visible;
}
/// The user clicked 刷新二维码 since the last call.
pub fn take_refresh(&self) -> bool {
#[cfg(uu_gtk)]
if self.shown { return unsafe { uu_login_ui_take_refresh() } != 0; }
false
}
pub fn status(&self, text: &str, tone: Tone) { pub fn status(&self, text: &str, tone: Tone) {
#[cfg(uu_gtk)] #[cfg(uu_gtk)]
if self.shown { let t = cstr(text); unsafe { uu_login_ui_set_status(t.as_ptr(), tone as libc::c_int) } } if self.shown { let t = cstr(text); unsafe { uu_login_ui_set_status(t.as_ptr(), tone as libc::c_int) } }
+33 -13
View File
@@ -198,11 +198,9 @@ impl Encoder {
fn nvidia_present() -> bool { fn nvidia_present() -> bool {
std::path::Path::new("/proc/driver/nvidia/version").exists() || std::path::Path::new("/dev/nvidia0").exists() std::path::Path::new("/proc/driver/nvidia/version").exists() || std::path::Path::new("/dev/nvidia0").exists()
} }
/// 真彩 (4:4:4) needs the FFmpeg process pipeline with NVENC or libx264 (h264_vaapi and the /// 真彩 (4:4:4): NVENC (in-process or FFmpeg) and libx264; h264_vaapi is 4:2:0 only.
/// in-process encoder are 4:2:0 only). /// A GPU without 4:4:4 NVENC falls back to libx264, which keeps 4:4:4.
pub fn full_chroma_available() -> bool { pub fn full_chroma_available() -> bool {
#[cfg(feature = "inproc-av")]
if crate::inproc_encoder::selected() { return false; }
Encoder::requested() != Encoder::Vaapi Encoder::requested() != Encoder::Vaapi
} }
@@ -506,17 +504,33 @@ async fn read_encoder(
) -> Result<(), String> { ) -> Result<(), String> {
// The sequence is shared so a CPU fallback or settings restart continues it (skip statistics stay correct). // The sequence is shared so a CPU fallback or settings restart continues it (skip statistics stay correct).
let encoder = Encoder::requested(); let encoder = Encoder::requested();
if encoder == Encoder::Cpu { return read_encoder_with(size,display,portal,latest,Encoder::Cpu,sequence,settings,idr,budget).await; } if encoder == Encoder::Cpu { return read_encoder_with(size,display,portal,latest,Encoder::Cpu,true,sequence,settings,idr,budget).await; }
match read_encoder_with(size,display.clone(),portal.clone(),latest.clone(),encoder,sequence.clone(),settings,idr.clone(),budget.clone()).await { #[cfg_attr(not(feature = "inproc-av"), allow(unused_mut))]
let mut result = read_encoder_with(size,display.clone(),portal.clone(),latest.clone(),encoder,true,sequence.clone(),settings,idr.clone(),budget.clone()).await;
// A failed native NVENC session (old driver, unsupported mode) still leaves FFmpeg's NVENC process pipeline.
#[cfg(feature = "inproc-av")]
if let Err(error) = &result {
if !latest.is_closed() && native_inproc(&display, portal.is_some(), encoder) {
log::warn!("In-process {} stream unavailable; switching to the FFmpeg process pipeline: {}",encoder.name_for(settings.codec),error);
result = read_encoder_with(size,display.clone(),portal.clone(),latest.clone(),encoder,false,sequence.clone(),settings,idr.clone(),budget.clone()).await;
}
}
match result {
Ok(())=>Ok(()), Ok(())=>Ok(()),
Err(error) if !latest.is_closed()=>{ Err(error) if !latest.is_closed()=>{
log::warn!("{} stream unavailable; switching scoped encoder to CPU: {}",encoder.name_for(settings.codec),error); log::warn!("{} stream unavailable; switching scoped encoder to CPU: {}",encoder.name_for(settings.codec),error);
read_encoder_with(size,display,portal,latest,Encoder::Cpu,sequence,settings,idr,budget).await read_encoder_with(size,display,portal,latest,Encoder::Cpu,true,sequence,settings,idr,budget).await
}, },
Err(error)=>Err(error), Err(error)=>Err(error),
} }
} }
/// The native in-process NVENC path serves X11 only (not Wayland/KMS capture or a portal session).
#[cfg(feature = "inproc-av")]
fn native_inproc(display: &str, portal: bool, encoder: Encoder) -> bool {
!display.starts_with("wayland:") && !display.starts_with("kms:") && !portal && crate::inproc_encoder::selected(encoder)
}
/// Capture and encode inside this process; each packet is one access unit. /// Capture and encode inside this process; each packet is one access unit.
#[cfg(feature = "inproc-av")] #[cfg(feature = "inproc-av")]
async fn read_inprocess( async fn read_inprocess(
@@ -528,9 +542,11 @@ async fn read_inprocess(
let stop = Arc::new(AtomicBool::new(false)); let stop = Arc::new(AtomicBool::new(false));
// If this future is dropped (reader task aborted), the blocking encoder stops at its next packet. // If this future is dropped (reader task aborted), the blocking encoder stops at its next packet.
let _stop_on_drop = StopOnDrop(stop.clone()); let _stop_on_drop = StopOnDrop(stop.clone());
// A restart aborts the previous reader without waiting for its blocking encoder; wait for it here.
let slot = crate::inproc_encoder::stream_slot(&idr).await;
let job = tokio::task::spawn_blocking({ let job = tokio::task::spawn_blocking({
let stop = stop.clone(); let stop = stop.clone();
move || crate::inproc_encoder::run(config, units_tx, stop, idr, budget) move || { let _slot = slot; crate::inproc_encoder::run(config, units_tx, stop, idr, budget) }
}); });
let mut gate = DependencyGate::default(); let mut gate = DependencyGate::default();
while let Some(unit) = units.recv().await { while let Some(unit) = units.recv().await {
@@ -579,7 +595,8 @@ fn rate_args(settings: &VideoSettings, encoder: Encoder) -> (String, String) {
} }
async fn read_encoder_with( async fn read_encoder_with(
size: (u32,u32),display:String,portal:Option<Arc<crate::portal::PortalSession>>,latest:mpsc::Sender<Arc<CompleteFrame>>,encoder:Encoder,sequence:Arc<std::sync::atomic::AtomicU64>, size: (u32,u32),display:String,portal:Option<Arc<crate::portal::PortalSession>>,latest:mpsc::Sender<Arc<CompleteFrame>>,encoder:Encoder,
#[cfg_attr(not(feature = "inproc-av"), allow(unused_variables))] allow_inproc: bool,sequence:Arc<std::sync::atomic::AtomicU64>,
settings: VideoSettings, #[cfg_attr(not(feature = "inproc-av"), allow(unused_variables))] idr: Arc<AtomicBool>, budget: Arc<Semaphore>, settings: VideoSettings, #[cfg_attr(not(feature = "inproc-av"), allow(unused_variables))] idr: Arc<AtomicBool>, budget: Arc<Semaphore>,
)->Result<(),String>{ )->Result<(),String>{
let selected_output=if display.starts_with("wayland:") {crate::wayland::source(settings.source_id)} else {None}; let selected_output=if display.starts_with("wayland:") {crate::wayland::source(settings.source_id)} else {None};
@@ -593,14 +610,14 @@ async fn read_encoder_with(
Some(Arc::new(capture)) Some(Arc::new(capture))
} else { None }; } else { None };
#[cfg(feature = "inproc-av")] #[cfg(feature = "inproc-av")]
if !wayland && portal.is_none() && crate::inproc_encoder::selected() { if allow_inproc && native_inproc(&display, portal.is_some(), encoder) {
let max_kbps = settings.max_kbps().unwrap_or(30_000) as i64; let max_kbps = settings.max_kbps().unwrap_or(30_000) as i64;
let config = crate::inproc_encoder::Config { let config = crate::inproc_encoder::Config {
display: display.clone(), input: size, output: settings.output, fps: u32::from(settings.fps), display: display.clone(), input: size, output: settings.output, fps: u32::from(settings.fps),
gop: gop_frames(settings.fps, true) as u32, cq: settings.cq() as u32, gop: gop_frames(settings.fps, true) as u32, cq: settings.cq() as u32,
target_bps: settings.max_kbps().map_or(16_000, |k| i64::from(k) * 3 / 4) * 1000, target_bps: settings.max_kbps().map_or(16_000, |k| i64::from(k) * 3 / 4) * 1000,
max_bps: max_kbps * 1000, buffer_bits: i64::from(buffer_for(&settings, encoder)) * 1000, encoder, codec: settings.codec, max_bps: max_kbps * 1000, buffer_bits: i64::from(buffer_for(&settings, encoder)) * 1000, codec: settings.codec, chroma: settings.chroma,
damage: crate::inproc_encoder::damage_enabled(), vaapi: encoder == Encoder::Vaapi,
}; };
log::info!("Desktop encoder: display={} source={}x{} output={:?} fps={} cq={} maxrate={}k encoder={} pipeline=in-process", log::info!("Desktop encoder: display={} source={}x{} output={:?} fps={} cq={} maxrate={}k encoder={} pipeline=in-process",
display,size.0,size.1,settings.output,settings.fps,config.cq,max_kbps,encoder.name_for(settings.codec)); display,size.0,size.1,settings.output,settings.fps,config.cq,max_kbps,encoder.name_for(settings.codec));
@@ -775,7 +792,10 @@ fn probe_settings() -> Result<VideoSettings,String> {
None=>crate::capture_setting::QUALITY_AUTO, None=>crate::capture_setting::QUALITY_AUTO,
Some(v)=>v.parse::<u64>().ok().filter(|q|(1..=5).contains(q)).ok_or("UU_PROBE_QUALITY must be 1-5")?, Some(v)=>v.parse::<u64>().ok().filter(|q|(1..=5).contains(q)).ok_or("UU_PROBE_QUALITY must be 1-5")?,
}; };
Ok(VideoSettings {fps,quality,codec:probe_codec(),..VideoSettings::default()}) let mut settings=VideoSettings {fps,quality,codec:probe_codec(),..VideoSettings::default()};
// UU_PROBE_CHROMA=444 requests 真彩 like the viewer does (only where full_chroma_available()).
if std::env::var("UU_PROBE_CHROMA").as_deref()==Ok("444") && full_chroma_available() { settings.chroma=crate::capture_setting::CHROMA_444; }
Ok(settings)
} }
/// The probe follows `UU_VIDEO_CODEC` so the HEVC path can be measured at all (H.264 otherwise), /// The probe follows `UU_VIDEO_CODEC` so the HEVC path can be measured at all (H.264 otherwise),
+1 -1
View File
@@ -1,4 +1,4 @@
//! Uses the same optional FFmpeg dependency as the host inproc-av feature. //! Uses the optional FFmpeg 6 libraries of the `viewer-ffmpeg` feature.
use super::desktop::{Decoder,Frame}; use super::desktop::{Decoder,Frame};
use ffmpeg_next as av; use ffmpeg_next as av;
use super::media::Codec; use super::media::Codec;
+4
View File
@@ -0,0 +1,4 @@
`ffnvcodec/dynlink_cuda.h` and `ffnvcodec/nvEncodeAPI.h` are unmodified copies from
[nv-codec-headers](https://github.com/FFmpeg/nv-codec-headers) tag `n12.1.14.0`
(NVENC API 12.1, the version FFmpeg 6.1 builds against; NVIDIA Linux driver 530.41 or newer).
MIT licensed, see the header comments. Only the in-process NVENC engine (`--features inproc-av`) uses them.
+518
View File
@@ -0,0 +1,518 @@
/*
* This copyright notice applies to this header file only:
*
* Copyright (c) 2016
*
* Permission is hereby granted, free of charge, to any person
* obtaining a copy of this software and associated documentation
* files (the "Software"), to deal in the Software without
* restriction, including without limitation the rights to use,
* copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the software, and to permit persons to whom the
* software is furnished to do so, subject to the following
* conditions:
*
* The above copyright notice and this permission notice shall be
* included in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
* OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
* HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
* WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#if !defined(FFNV_DYNLINK_CUDA_H) && !defined(CUDA_VERSION)
#define FFNV_DYNLINK_CUDA_H
#include <stddef.h>
#include <stdint.h>
#define CUDA_VERSION 7050
#if defined(_WIN32) || defined(__CYGWIN__)
#define CUDAAPI __stdcall
#else
#define CUDAAPI
#endif
#define CU_CTX_SCHED_BLOCKING_SYNC 4
typedef int CUdevice;
#if defined(__x86_64) || defined(AMD64) || defined(_M_AMD64) || defined(__LP64__) || defined(__aarch64__)
typedef unsigned long long CUdeviceptr;
#else
typedef unsigned int CUdeviceptr;
#endif
typedef unsigned long long CUtexObject;
typedef struct CUarray_st *CUarray;
typedef struct CUctx_st *CUcontext;
typedef struct CUstream_st *CUstream;
typedef struct CUevent_st *CUevent;
typedef struct CUfunc_st *CUfunction;
typedef struct CUmod_st *CUmodule;
typedef struct CUmipmappedArray_st *CUmipmappedArray;
typedef struct CUgraphicsResource_st *CUgraphicsResource;
typedef struct CUextMemory_st *CUexternalMemory;
typedef struct CUextSemaphore_st *CUexternalSemaphore;
typedef struct CUeglStreamConnection_st *CUeglStreamConnection;
typedef struct CUlinkState_st *CUlinkState;
typedef enum cudaError_enum {
CUDA_SUCCESS = 0,
CUDA_ERROR_NOT_READY = 600,
CUDA_ERROR_LAUNCH_TIMEOUT = 702,
CUDA_ERROR_UNKNOWN = 999
} CUresult;
/**
* Device properties (subset)
*/
typedef enum CUdevice_attribute_enum {
CU_DEVICE_ATTRIBUTE_CLOCK_RATE = 13,
CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT = 14,
CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT = 16,
CU_DEVICE_ATTRIBUTE_INTEGRATED = 18,
CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY = 19,
CU_DEVICE_ATTRIBUTE_COMPUTE_MODE = 20,
CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS = 31,
CU_DEVICE_ATTRIBUTE_PCI_BUS_ID = 33,
CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID = 34,
CU_DEVICE_ATTRIBUTE_TCC_DRIVER = 35,
CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE = 36,
CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH = 37,
CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT = 40,
CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING = 41,
CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID = 50,
CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT = 51,
CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = 75,
CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR = 76,
CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY = 83,
CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD = 84,
CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID = 85,
} CUdevice_attribute;
typedef enum CUarray_format_enum {
CU_AD_FORMAT_UNSIGNED_INT8 = 0x01,
CU_AD_FORMAT_UNSIGNED_INT16 = 0x02,
CU_AD_FORMAT_UNSIGNED_INT32 = 0x03,
CU_AD_FORMAT_SIGNED_INT8 = 0x08,
CU_AD_FORMAT_SIGNED_INT16 = 0x09,
CU_AD_FORMAT_SIGNED_INT32 = 0x0a,
CU_AD_FORMAT_HALF = 0x10,
CU_AD_FORMAT_FLOAT = 0x20
} CUarray_format;
typedef enum CUmemorytype_enum {
CU_MEMORYTYPE_HOST = 1,
CU_MEMORYTYPE_DEVICE = 2,
CU_MEMORYTYPE_ARRAY = 3
} CUmemorytype;
typedef enum CUlimit_enum {
CU_LIMIT_STACK_SIZE = 0,
CU_LIMIT_PRINTF_FIFO_SIZE = 1,
CU_LIMIT_MALLOC_HEAP_SIZE = 2,
CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH = 3,
CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT = 4
} CUlimit;
typedef enum CUresourcetype_enum {
CU_RESOURCE_TYPE_ARRAY = 0x00,
CU_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01,
CU_RESOURCE_TYPE_LINEAR = 0x02,
CU_RESOURCE_TYPE_PITCH2D = 0x03
} CUresourcetype;
typedef enum CUaddress_mode_enum {
CU_TR_ADDRESS_MODE_WRAP = 0,
CU_TR_ADDRESS_MODE_CLAMP = 1,
CU_TR_ADDRESS_MODE_MIRROR = 2,
CU_TR_ADDRESS_MODE_BORDER = 3
} CUaddress_mode;
typedef enum CUfilter_mode_enum {
CU_TR_FILTER_MODE_POINT = 0,
CU_TR_FILTER_MODE_LINEAR = 1
} CUfilter_mode;
typedef enum CUgraphicsRegisterFlags_enum {
CU_GRAPHICS_REGISTER_FLAGS_NONE = 0,
CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY = 1,
CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD = 2,
CU_GRAPHICS_REGISTER_FLAGS_SURFACE_LDST = 4,
CU_GRAPHICS_REGISTER_FLAGS_TEXTURE_GATHER = 8
} CUgraphicsRegisterFlags;
typedef enum CUexternalMemoryHandleType_enum {
CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD = 1,
CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 = 2,
CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3,
CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP = 4,
CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE = 5,
} CUexternalMemoryHandleType;
typedef enum CUexternalSemaphoreHandleType_enum {
CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD = 1,
CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 = 2,
CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3,
CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE = 4,
CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_TIMELINE_SEMAPHORE_FD = 9,
CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_TIMELINE_SEMAPHORE_WIN32 = 10,
} CUexternalSemaphoreHandleType;
typedef enum CUjit_option_enum
{
CU_JIT_MAX_REGISTERS = 0,
CU_JIT_THREADS_PER_BLOCK = 1,
CU_JIT_WALL_TIME = 2,
CU_JIT_INFO_LOG_BUFFER = 3,
CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES = 4,
CU_JIT_ERROR_LOG_BUFFER = 5,
CU_JIT_ERROR_LOG_BUFFER_SIZE_BYTES = 6,
CU_JIT_OPTIMIZATION_LEVEL = 7,
CU_JIT_TARGET_FROM_CUCONTEXT = 8,
CU_JIT_TARGET = 9,
CU_JIT_FALLBACK_STRATEGY = 10,
CU_JIT_GENERATE_DEBUG_INFO = 11,
CU_JIT_LOG_VERBOSE = 12,
CU_JIT_GENERATE_LINE_INFO = 13,
CU_JIT_CACHE_MODE = 14,
CU_JIT_NEW_SM3X_OPT = 15,
CU_JIT_FAST_COMPILE = 16,
CU_JIT_GLOBAL_SYMBOL_NAMES = 17,
CU_JIT_GLOBAL_SYMBOL_ADDRESSES = 18,
CU_JIT_GLOBAL_SYMBOL_COUNT = 19,
CU_JIT_NUM_OPTIONS
} CUjit_option;
typedef enum CUjitInputType_enum
{
CU_JIT_INPUT_CUBIN = 0,
CU_JIT_INPUT_PTX = 1,
CU_JIT_INPUT_FATBINARY = 2,
CU_JIT_INPUT_OBJECT = 3,
CU_JIT_INPUT_LIBRARY = 4,
CU_JIT_NUM_INPUT_TYPES
} CUjitInputType;
typedef enum CUeglFrameType
{
CU_EGL_FRAME_TYPE_ARRAY = 0,
CU_EGL_FRAME_TYPE_PITCH = 1,
} CUeglFrameType;
typedef enum CUeglColorFormat
{
CU_EGL_COLOR_FORMAT_YUV420_PLANAR = 0x00,
CU_EGL_COLOR_FORMAT_YUV420_SEMIPLANAR = 0x01,
CU_EGL_COLOR_FORMAT_YVU420_SEMIPLANAR = 0x15,
CU_EGL_COLOR_FORMAT_Y10V10U10_420_SEMIPLANAR = 0x17,
CU_EGL_COLOR_FORMAT_Y12V12U12_420_SEMIPLANAR = 0x19,
} CUeglColorFormat;
typedef enum CUd3d11DeviceList_enum
{
CU_D3D11_DEVICE_LIST_ALL = 1,
CU_D3D11_DEVICE_LIST_CURRENT_FRAME = 2,
CU_D3D11_DEVICE_LIST_NEXT_FRAME = 3,
} CUd3d11DeviceList;
#ifndef CU_UUID_HAS_BEEN_DEFINED
#define CU_UUID_HAS_BEEN_DEFINED
typedef struct CUuuid_st {
char bytes[16];
} CUuuid;
#endif
typedef struct CUDA_MEMCPY2D_st {
size_t srcXInBytes;
size_t srcY;
CUmemorytype srcMemoryType;
const void *srcHost;
CUdeviceptr srcDevice;
CUarray srcArray;
size_t srcPitch;
size_t dstXInBytes;
size_t dstY;
CUmemorytype dstMemoryType;
void *dstHost;
CUdeviceptr dstDevice;
CUarray dstArray;
size_t dstPitch;
size_t WidthInBytes;
size_t Height;
} CUDA_MEMCPY2D;
typedef struct CUDA_RESOURCE_DESC_st {
CUresourcetype resType;
union {
struct {
CUarray hArray;
} array;
struct {
CUmipmappedArray hMipmappedArray;
} mipmap;
struct {
CUdeviceptr devPtr;
CUarray_format format;
unsigned int numChannels;
size_t sizeInBytes;
} linear;
struct {
CUdeviceptr devPtr;
CUarray_format format;
unsigned int numChannels;
size_t width;
size_t height;
size_t pitchInBytes;
} pitch2D;
struct {
int reserved[32];
} reserved;
} res;
unsigned int flags;
} CUDA_RESOURCE_DESC;
typedef struct CUDA_TEXTURE_DESC_st {
CUaddress_mode addressMode[3];
CUfilter_mode filterMode;
unsigned int flags;
unsigned int maxAnisotropy;
CUfilter_mode mipmapFilterMode;
float mipmapLevelBias;
float minMipmapLevelClamp;
float maxMipmapLevelClamp;
float borderColor[4];
int reserved[12];
} CUDA_TEXTURE_DESC;
/* Unused type */
typedef struct CUDA_RESOURCE_VIEW_DESC_st CUDA_RESOURCE_VIEW_DESC;
typedef unsigned int GLenum;
typedef unsigned int GLuint;
/*
* Prefix type name to avoid collisions. Clients using these types
* will include the real headers with real definitions.
*/
typedef int32_t ffnv_EGLint;
typedef void *ffnv_EGLStreamKHR;
typedef enum CUGLDeviceList_enum {
CU_GL_DEVICE_LIST_ALL = 1,
CU_GL_DEVICE_LIST_CURRENT_FRAME = 2,
CU_GL_DEVICE_LIST_NEXT_FRAME = 3,
} CUGLDeviceList;
typedef struct CUDA_EXTERNAL_MEMORY_HANDLE_DESC_st {
CUexternalMemoryHandleType type;
union {
int fd;
struct {
void *handle;
const void *name;
} win32;
} handle;
unsigned long long size;
unsigned int flags;
unsigned int reserved[16];
} CUDA_EXTERNAL_MEMORY_HANDLE_DESC;
typedef struct CUDA_EXTERNAL_MEMORY_BUFFER_DESC_st {
unsigned long long offset;
unsigned long long size;
unsigned int flags;
unsigned int reserved[16];
} CUDA_EXTERNAL_MEMORY_BUFFER_DESC;
typedef struct CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC_st {
CUexternalSemaphoreHandleType type;
union {
int fd;
struct {
void *handle;
const void *name;
} win32;
} handle;
unsigned int flags;
unsigned int reserved[16];
} CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC;
typedef struct CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS_st {
struct {
struct {
unsigned long long value;
} fence;
unsigned int reserved[16];
} params;
unsigned int flags;
unsigned int reserved[16];
} CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS;
typedef CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS;
typedef struct CUDA_ARRAY_DESCRIPTOR_st {
size_t Width;
size_t Height;
CUarray_format Format;
unsigned int NumChannels;
} CUDA_ARRAY_DESCRIPTOR;
typedef struct CUDA_ARRAY3D_DESCRIPTOR_st {
size_t Width;
size_t Height;
size_t Depth;
CUarray_format Format;
unsigned int NumChannels;
unsigned int Flags;
} CUDA_ARRAY3D_DESCRIPTOR;
typedef struct CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC_st {
unsigned long long offset;
CUDA_ARRAY3D_DESCRIPTOR arrayDesc;
unsigned int numLevels;
unsigned int reserved[16];
} CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC;
#define CU_EGL_FRAME_MAX_PLANES 3
typedef struct CUeglFrame_st {
union {
CUarray pArray[CU_EGL_FRAME_MAX_PLANES];
void* pPitch[CU_EGL_FRAME_MAX_PLANES];
} frame;
unsigned int width;
unsigned int height;
unsigned int depth;
unsigned int pitch;
unsigned int planeCount;
unsigned int numChannels;
CUeglFrameType frameType;
CUeglColorFormat eglColorFormat;
CUarray_format cuFormat;
} CUeglFrame;
#define CU_STREAM_DEFAULT 0
#define CU_STREAM_NON_BLOCKING 1
#define CU_EVENT_DEFAULT 0
#define CU_EVENT_BLOCKING_SYNC 1
#define CU_EVENT_DISABLE_TIMING 2
#define CU_EVENT_WAIT_DEFAULT 0
#define CU_EVENT_WAIT_EXTERNAL 1
#define CU_TRSF_READ_AS_INTEGER 1
typedef void CUDAAPI CUstreamCallback(CUstream hStream, CUresult status, void *userdata);
typedef CUresult CUDAAPI tcuInit(unsigned int Flags);
typedef CUresult CUDAAPI tcuDriverGetVersion(int *driverVersion);
typedef CUresult CUDAAPI tcuDeviceGetCount(int *count);
typedef CUresult CUDAAPI tcuDeviceGet(CUdevice *device, int ordinal);
typedef CUresult CUDAAPI tcuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, CUdevice dev);
typedef CUresult CUDAAPI tcuDeviceGetName(char *name, int len, CUdevice dev);
typedef CUresult CUDAAPI tcuDeviceGetUuid(CUuuid *uuid, CUdevice dev);
typedef CUresult CUDAAPI tcuDeviceGetUuid_v2(CUuuid *uuid, CUdevice dev);
typedef CUresult CUDAAPI tcuDeviceGetLuid(char* luid, unsigned int* deviceNodeMask, CUdevice dev);
typedef CUresult CUDAAPI tcuDeviceGetByPCIBusId(CUdevice* dev, const char* pciBusId);
typedef CUresult CUDAAPI tcuDeviceGetPCIBusId(char* pciBusId, int len, CUdevice dev);
typedef CUresult CUDAAPI tcuDeviceComputeCapability(int *major, int *minor, CUdevice dev);
typedef CUresult CUDAAPI tcuCtxCreate_v2(CUcontext *pctx, unsigned int flags, CUdevice dev);
typedef CUresult CUDAAPI tcuCtxGetCurrent(CUcontext *pctx);
typedef CUresult CUDAAPI tcuCtxSetLimit(CUlimit limit, size_t value);
typedef CUresult CUDAAPI tcuCtxPushCurrent_v2(CUcontext pctx);
typedef CUresult CUDAAPI tcuCtxPopCurrent_v2(CUcontext *pctx);
typedef CUresult CUDAAPI tcuCtxDestroy_v2(CUcontext ctx);
typedef CUresult CUDAAPI tcuMemAlloc_v2(CUdeviceptr *dptr, size_t bytesize);
typedef CUresult CUDAAPI tcuMemAllocPitch_v2(CUdeviceptr *dptr, size_t *pPitch, size_t WidthInBytes, size_t Height, unsigned int ElementSizeBytes);
typedef CUresult CUDAAPI tcuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, unsigned int flags);
typedef CUresult CUDAAPI tcuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream);
typedef CUresult CUDAAPI tcuMemFree_v2(CUdeviceptr dptr);
typedef CUresult CUDAAPI tcuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t bytesize);
typedef CUresult CUDAAPI tcuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t bytesize, CUstream hStream);
typedef CUresult CUDAAPI tcuMemcpy2D_v2(const CUDA_MEMCPY2D *pcopy);
typedef CUresult CUDAAPI tcuMemcpy2DAsync_v2(const CUDA_MEMCPY2D *pcopy, CUstream hStream);
typedef CUresult CUDAAPI tcuMemcpyHtoD_v2(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount);
typedef CUresult CUDAAPI tcuMemcpyHtoDAsync_v2(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream);
typedef CUresult CUDAAPI tcuMemcpyDtoH_v2(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount);
typedef CUresult CUDAAPI tcuMemcpyDtoHAsync_v2(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
typedef CUresult CUDAAPI tcuMemcpyDtoD_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount);
typedef CUresult CUDAAPI tcuMemcpyDtoDAsync_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
typedef CUresult CUDAAPI tcuGetErrorName(CUresult error, const char** pstr);
typedef CUresult CUDAAPI tcuGetErrorString(CUresult error, const char** pstr);
typedef CUresult CUDAAPI tcuCtxGetDevice(CUdevice *device);
typedef CUresult CUDAAPI tcuDevicePrimaryCtxRetain(CUcontext *pctx, CUdevice dev);
typedef CUresult CUDAAPI tcuDevicePrimaryCtxRelease(CUdevice dev);
typedef CUresult CUDAAPI tcuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags);
typedef CUresult CUDAAPI tcuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags, int *active);
typedef CUresult CUDAAPI tcuDevicePrimaryCtxReset(CUdevice dev);
typedef CUresult CUDAAPI tcuStreamCreate(CUstream *phStream, unsigned int flags);
typedef CUresult CUDAAPI tcuStreamQuery(CUstream hStream);
typedef CUresult CUDAAPI tcuStreamSynchronize(CUstream hStream);
typedef CUresult CUDAAPI tcuStreamDestroy_v2(CUstream hStream);
typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream hStream, CUstreamCallback *callback, void *userdata, unsigned int flags);
typedef CUresult CUDAAPI tcuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int flags);
typedef CUresult CUDAAPI tcuEventCreate(CUevent *phEvent, unsigned int flags);
typedef CUresult CUDAAPI tcuEventDestroy_v2(CUevent hEvent);
typedef CUresult CUDAAPI tcuEventSynchronize(CUevent hEvent);
typedef CUresult CUDAAPI tcuEventQuery(CUevent hEvent);
typedef CUresult CUDAAPI tcuEventRecord(CUevent hEvent, CUstream hStream);
typedef CUresult CUDAAPI tcuLaunchKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void** kernelParams, void** extra);
typedef CUresult CUDAAPI tcuLinkCreate(unsigned int numOptions, CUjit_option* options, void** optionValues, CUlinkState* stateOut);
typedef CUresult CUDAAPI tcuLinkAddData(CUlinkState state, CUjitInputType type, void* data, size_t size, const char* name, unsigned int numOptions, CUjit_option* options, void** optionValues);
typedef CUresult CUDAAPI tcuLinkComplete(CUlinkState state, void** cubinOut, size_t* sizeOut);
typedef CUresult CUDAAPI tcuLinkDestroy(CUlinkState state);
typedef CUresult CUDAAPI tcuModuleLoadData(CUmodule* module, const void* image);
typedef CUresult CUDAAPI tcuModuleUnload(CUmodule hmod);
typedef CUresult CUDAAPI tcuModuleGetFunction(CUfunction* hfunc, CUmodule hmod, const char* name);
typedef CUresult CUDAAPI tcuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, CUmodule hmod, const char* name);
typedef CUresult CUDAAPI tcuTexObjectCreate(CUtexObject* pTexObject, const CUDA_RESOURCE_DESC* pResDesc, const CUDA_TEXTURE_DESC* pTexDesc, const CUDA_RESOURCE_VIEW_DESC* pResViewDesc);
typedef CUresult CUDAAPI tcuTexObjectDestroy(CUtexObject texObject);
typedef CUresult CUDAAPI tcuGLGetDevices_v2(unsigned int* pCudaDeviceCount, CUdevice* pCudaDevices, unsigned int cudaDeviceCount, CUGLDeviceList deviceList);
typedef CUresult CUDAAPI tcuGraphicsGLRegisterImage(CUgraphicsResource* pCudaResource, GLuint image, GLenum target, unsigned int Flags);
typedef CUresult CUDAAPI tcuGraphicsUnregisterResource(CUgraphicsResource resource);
typedef CUresult CUDAAPI tcuGraphicsMapResources(unsigned int count, CUgraphicsResource* resources, CUstream hStream);
typedef CUresult CUDAAPI tcuGraphicsUnmapResources(unsigned int count, CUgraphicsResource* resources, CUstream hStream);
typedef CUresult CUDAAPI tcuGraphicsSubResourceGetMappedArray(CUarray* pArray, CUgraphicsResource resource, unsigned int arrayIndex, unsigned int mipLevel);
typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr *devPtrOut, size_t *sizeOut, CUgraphicsResource resource);
typedef CUresult CUDAAPI tcuImportExternalMemory(CUexternalMemory* extMem_out, const CUDA_EXTERNAL_MEMORY_HANDLE_DESC* memHandleDesc);
typedef CUresult CUDAAPI tcuDestroyExternalMemory(CUexternalMemory extMem);
typedef CUresult CUDAAPI tcuExternalMemoryGetMappedBuffer(CUdeviceptr* devPtr, CUexternalMemory extMem, const CUDA_EXTERNAL_MEMORY_BUFFER_DESC* bufferDesc);
typedef CUresult CUDAAPI tcuExternalMemoryGetMappedMipmappedArray(CUmipmappedArray* mipmap, CUexternalMemory extMem, const CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC* mipmapDesc);
typedef CUresult CUDAAPI tcuMipmappedArrayGetLevel(CUarray* pLevelArray, CUmipmappedArray hMipmappedArray, unsigned int level);
typedef CUresult CUDAAPI tcuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray);
typedef CUresult CUDAAPI tcuImportExternalSemaphore(CUexternalSemaphore* extSem_out, const CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC* semHandleDesc);
typedef CUresult CUDAAPI tcuDestroyExternalSemaphore(CUexternalSemaphore extSem);
typedef CUresult CUDAAPI tcuSignalExternalSemaphoresAsync(const CUexternalSemaphore* extSemArray, const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS* paramsArray, unsigned int numExtSems, CUstream stream);
typedef CUresult CUDAAPI tcuWaitExternalSemaphoresAsync(const CUexternalSemaphore* extSemArray, const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS* paramsArray, unsigned int numExtSems, CUstream stream);
typedef CUresult CUDAAPI tcuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR* pAllocateArray);
typedef CUresult CUDAAPI tcuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR* pAllocateArray);
typedef CUresult CUDAAPI tcuArrayDestroy(CUarray hArray);
typedef CUresult CUDAAPI tcuEGLStreamProducerConnect(CUeglStreamConnection* conn, ffnv_EGLStreamKHR stream, ffnv_EGLint width, ffnv_EGLint height);
typedef CUresult CUDAAPI tcuEGLStreamProducerDisconnect(CUeglStreamConnection* conn);
typedef CUresult CUDAAPI tcuEGLStreamConsumerDisconnect(CUeglStreamConnection* conn);
typedef CUresult CUDAAPI tcuEGLStreamProducerPresentFrame(CUeglStreamConnection* conn, CUeglFrame eglframe, CUstream* pStream);
typedef CUresult CUDAAPI tcuEGLStreamProducerReturnFrame(CUeglStreamConnection* conn, CUeglFrame* eglframe, CUstream* pStream);
typedef CUresult CUDAAPI tcuD3D11GetDevice(CUdevice *device, void *dxgiAdapter);
typedef CUresult CUDAAPI tcuD3D11GetDevices(unsigned int *deviceCountOut, CUdevice *devices, unsigned int deviceCount, void *d3d11device, CUd3d11DeviceList listType);
typedef CUresult CUDAAPI tcuGraphicsD3D11RegisterResource(CUgraphicsResource *cudaResourceOut, void *d3d11Resource, unsigned int flags);
#endif
File diff suppressed because it is too large. Load diff
+11
View File
@@ -0,0 +1,11 @@
# Generated by Cargo
# will have compiled files and executables
/target/
/.idea/
# Remove Cargo.lock from gitignore if creating an executable, leave it for libraries
# More information here https://doc.rust-lang.org/cargo/guide/cargo-toml-vs-cargo-lock.html
Cargo.lock
# These are backup files generated by rustfmt
**/*.rs.bk
+19
View File
@@ -0,0 +1,19 @@
# webrtc-srtp changelog
## Unreleased
## v0.9.1
* Increased minimum support rust version to `1.60.0`.
* Increased required `webrtc-util` version to `0.7.0`.
## v0.9.0
* [#8 update deps + loosen some requirements](https://github.com/webrtc-rs/srtp/pull/8) by [@melekes](https://github.com/melekes).
* Increased min version of `log` dependency to `0.4.16`. [#250 Fix log at ^0.4.16 to make tests compile](https://github.com/webrtc-rs/webrtc/pull/250) by [@k0nserv](https://github.com/k0nserv).
## Prior to 0.8.9
Before 0.8.9 there was no changelog, previous changes are sometimes, but not always, available in the [GitHub Releases](https://github.com/webrtc-rs/srtp/releases).
+100
View File
@@ -0,0 +1,100 @@
# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO
#
# When uploading crates to the registry Cargo will automatically
# "normalize" Cargo.toml files for maximal compatibility
# with all versions of Cargo and also rewrite `path` dependencies
# to registry (e.g., crates.io) dependencies.
#
# If you are reading this file be aware that the original Cargo.toml
# will likely look very different (and much more reasonable).
# See Cargo.toml.orig for the original contents.
[package]
edition = "2021"
name = "webrtc-srtp"
version = "0.12.0"
authors = ["Rain Liu <[email protected]>"]
description = "A pure Rust implementation of SRTP"
homepage = "https://webrtc.rs"
documentation = "https://docs.rs/webrtc-srtp"
readme = "README.md"
license = "MIT OR Apache-2.0"
repository = "https://github.com/webrtc-rs/srtp"
[[bench]]
name = "srtp_bench"
harness = false
[dependencies.aead]
version = "0.5"
features = ["std"]
[dependencies.aes]
version = "0.8"
[dependencies.aes-gcm]
version = "0.10"
features = ["std"]
[dependencies.byteorder]
version = "1"
[dependencies.bytes]
version = "1"
[dependencies.ctr]
version = "0.9"
[dependencies.hmac]
version = "0.12"
features = ["std"]
[dependencies.log]
version = "0.4"
[dependencies.openssl]
version = "0.10.57"
optional = true
[dependencies.rtcp]
version = "0.10.1"
[dependencies.rtp]
version = "0.10.0"
[dependencies.sha1]
version = "0.10"
[dependencies.subtle]
version = "2"
[dependencies.thiserror]
version = "1"
[dependencies.tokio]
version = "1.32.0"
features = ["full"]
[dependencies.util]
version = "0.8.1"
features = [
"conn",
"buffer",
"marshal",
]
default-features = false
package = "webrtc-util"
[dev-dependencies.criterion]
version = "0.5"
features = ["async_futures"]
[dev-dependencies.lazy_static]
version = "1"
[dev-dependencies.tokio-test]
version = "0.4"
[features]
openssl = ["dep:openssl"]
vendored-openssl = ["openssl/vendored"]
+46
View File
@@ -0,0 +1,46 @@
[package]
name = "webrtc-srtp"
version = "0.12.0"
authors = ["Rain Liu <[email protected]>"]
edition = "2021"
description = "A pure Rust implementation of SRTP"
license = "MIT OR Apache-2.0"
documentation = "https://docs.rs/webrtc-srtp"
homepage = "https://webrtc.rs"
repository = "https://github.com/webrtc-rs/srtp"
[features]
openssl = ["dep:openssl"]
vendored-openssl = ["openssl/vendored"]
[dependencies]
util = { version = "0.8.1", path = "../util", package = "webrtc-util", default-features = false, features = [
"conn",
"buffer",
"marshal",
] }
rtp = { version = "0.10.0", path = "../rtp" }
rtcp = { version = "0.10.1", path = "../rtcp" }
byteorder = "1"
bytes = "1"
thiserror = "1"
hmac = { version = "0.12", features = ["std"] }
sha1 = "0.10"
ctr = "0.9"
aes = "0.8"
subtle = "2"
tokio = { version = "1.32.0", features = ["full"] }
log = "0.4"
aead = { version = "0.5", features = ["std"] }
aes-gcm = { version = "0.10", features = ["std"] }
openssl = { version = "0.10.57", optional = true }
[dev-dependencies]
criterion = { version = "0.5", features = ["async_futures"] }
tokio-test = "0.4"
lazy_static = "1"
[[bench]]
name = "srtp_bench"
harness = false
+201
View File
@@ -0,0 +1,201 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
+21
View File
@@ -0,0 +1,21 @@
MIT License
Copyright (c) 2021 WebRTC.rs
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
+14
View File
@@ -0,0 +1,14 @@
# Local patch to webrtc-srtp 0.12.0
Source: crates.io webrtc-srtp 0.12.0 (MIT / Apache-2.0; upstream licenses retained).
`SrtpSsrcState` estimates the rollover counter with the half-sequence-space rule from
RFC 3711 Appendix A and only advances to the highest packet index seen. Upstream used a
100-packet window and stored every sequence number, so a NACK retransmission of a
pre-wrap packet sent after the wrap (e.g. 65000 after 85) moved the state back and the
next new packet was protected with ROC+1; the receiver then failed authentication for
every following packet (frozen video, #83). Fix and regression tests contributed by
@AYssu. Cipher code, wire format and SRTCP are unchanged.
The upstream lifecycle test jumped from 5004 to 65535, outside the half-space window;
bridge packets keep it consecutive and the known-answer vectors are unchanged.
+30
View File
@@ -0,0 +1,30 @@
<h1 align="center">
<a href="https://webrtc.rs"><img src="./doc/webrtc.rs.png" alt="WebRTC.rs"></a>
<br>
</h1>
<p align="center">
<a href="https://github.com/webrtc-rs/srtp/actions">
<img src="https://github.com/webrtc-rs/srtp/workflows/cargo/badge.svg">
</a>
<a href="https://codecov.io/gh/webrtc-rs/srtp">
<img src="https://codecov.io/gh/webrtc-rs/srtp/branch/main/graph/badge.svg">
</a>
<a href="https://deps.rs/repo/github/webrtc-rs/srtp">
<img src="https://deps.rs/repo/github/webrtc-rs/srtp/status.svg">
</a>
<a href="https://crates.io/crates/webrtc-srtp">
<img src="https://img.shields.io/crates/v/webrtc-srtp.svg">
</a>
<a href="https://docs.rs/webrtc-srtp">
<img src="https://docs.rs/webrtc-srtp/badge.svg">
</a>
<a href="https://doc.rust-lang.org/1.6.0/complement-project-faq.html#why-dual-mitasl2-license">
<img src="https://img.shields.io/badge/license-MIT%2FApache--2.0-blue" alt="License: MIT/Apache 2.0">
</a>
<a href="https://discord.gg/4Ju8UHdXMs">
<img src="https://img.shields.io/discord/800204819540869120?logo=discord" alt="Discord">
</a>
</p>
<p align="center">
A pure Rust implementation of SRTP. Rewrite <a href="https://github.com/pion/srtp/releases/tag/v1.5.2">Pion</a> SRTP in Rust
</p>
+158
View File
@@ -0,0 +1,158 @@
use bytes::BytesMut;
use criterion::{criterion_group, criterion_main, Criterion};
use util::Marshal;
use webrtc_srtp::{context::Context, protection_profile::ProtectionProfile};
const MASTER_KEY: &[u8] = &[
96, 180, 31, 4, 119, 137, 128, 252, 75, 194, 252, 44, 63, 56, 61, 55,
];
const MASTER_SALT: &[u8] = &[247, 26, 49, 94, 99, 29, 79, 94, 5, 111, 252, 216, 62, 195];
const RAW_RTCP: &[u8] = &[
0x81, 0xc8, 0x00, 0x0b, 0xca, 0xfe, 0xba, 0xbe, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab,
0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab,
];
fn benchmark_encrypt_rtp_aes_128_cm_hmac_sha1(c: &mut Criterion) {
let mut ctx = Context::new(
MASTER_KEY,
MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)
.unwrap();
let mut pld = BytesMut::new();
for i in 0..1200 {
pld.extend_from_slice(&[i as u8]);
}
c.bench_function("Benchmark RTP encrypt", |b| {
let mut seq = 1;
b.iter_batched(
|| {
let pkt = rtp::packet::Packet {
header: rtp::header::Header {
sequence_number: seq,
timestamp: seq.into(),
extension_profile: 48862,
marker: true,
padding: false,
extension: true,
payload_type: 96,
..Default::default()
},
payload: pld.clone().into(),
};
seq += 1;
pkt.marshal().unwrap()
},
|pkt_raw| {
ctx.encrypt_rtp(&pkt_raw).unwrap();
},
criterion::BatchSize::LargeInput,
);
});
}
fn benchmark_decrypt_rtp_aes_128_cm_hmac_sha1(c: &mut Criterion) {
let mut setup_ctx = Context::new(
MASTER_KEY,
MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)
.unwrap();
let mut ctx = Context::new(
MASTER_KEY,
MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)
.unwrap();
let mut pld = BytesMut::new();
for i in 0..1200 {
pld.extend_from_slice(&[i as u8]);
}
c.bench_function("Benchmark RTP decrypt", |b| {
let mut seq = 1;
b.iter_batched(
|| {
let pkt = rtp::packet::Packet {
header: rtp::header::Header {
sequence_number: seq,
timestamp: seq.into(),
extension_profile: 48862,
marker: true,
padding: false,
extension: true,
payload_type: 96,
..Default::default()
},
payload: pld.clone().into(),
};
seq += 1;
setup_ctx.encrypt_rtp(&pkt.marshal().unwrap()).unwrap()
},
|encrypted| ctx.decrypt_rtp(&encrypted).unwrap(),
criterion::BatchSize::LargeInput,
);
});
}
fn benchmark_encrypt_rtcp_aes_128_cm_hmac_sha1(c: &mut Criterion) {
let mut ctx = Context::new(
MASTER_KEY,
MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)
.unwrap();
c.bench_function("Benchmark RTCP encrypt", |b| {
b.iter(|| {
ctx.encrypt_rtcp(RAW_RTCP).unwrap();
});
});
}
fn benchmark_decrypt_rtcp_aes_128_cm_hmac_sha1(c: &mut Criterion) {
let encrypted = Context::new(
MASTER_KEY,
MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)
.unwrap()
.encrypt_rtcp(RAW_RTCP)
.unwrap();
let mut ctx = Context::new(
MASTER_KEY,
MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)
.unwrap();
c.bench_function("Benchmark RTCP decrypt", |b| {
b.iter(|| ctx.decrypt_rtcp(&encrypted).unwrap());
});
}
criterion_group!(
benches,
benchmark_encrypt_rtp_aes_128_cm_hmac_sha1,
benchmark_decrypt_rtp_aes_128_cm_hmac_sha1,
benchmark_encrypt_rtcp_aes_128_cm_hmac_sha1,
benchmark_decrypt_rtcp_aes_128_cm_hmac_sha1
);
criterion_main!(benches);
+23
View File
@@ -0,0 +1,23 @@
codecov:
require_ci_to_pass: yes
max_report_age: off
token: d65de923-7c3d-4836-8d9a-5183b356be4f
coverage:
precision: 2
round: down
range: 50..90
status:
project:
default:
enabled: no
threshold: 0.2
if_not_found: success
patch:
default:
enabled: no
if_not_found: success
changes:
default:
enabled: no
if_not_found: success
Binary file not shown.

After

Width:  |  Height:  |  Size: 32 KiB

+247
View File
@@ -0,0 +1,247 @@
use aes_gcm::aead::generic_array::GenericArray;
use aes_gcm::aead::{Aead, Payload};
use aes_gcm::{Aes128Gcm, KeyInit, Nonce};
use byteorder::{BigEndian, ByteOrder};
use bytes::{Bytes, BytesMut};
use util::marshal::*;
use super::Cipher;
use crate::error::{Error, Result};
use crate::key_derivation::*;
pub const CIPHER_AEAD_AES_GCM_AUTH_TAG_LEN: usize = 16;
const RTCP_ENCRYPTION_FLAG: u8 = 0x80;
/// AEAD Cipher based on AES.
pub(crate) struct CipherAeadAesGcm {
srtp_cipher: aes_gcm::Aes128Gcm,
srtcp_cipher: aes_gcm::Aes128Gcm,
srtp_session_salt: Vec<u8>,
srtcp_session_salt: Vec<u8>,
}
impl Cipher for CipherAeadAesGcm {
fn auth_tag_len(&self) -> usize {
CIPHER_AEAD_AES_GCM_AUTH_TAG_LEN
}
fn encrypt_rtp(
&mut self,
payload: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes> {
// Grow the given buffer to fit the output.
let header_len = header.marshal_size();
let mut writer = BytesMut::with_capacity(payload.len() + self.auth_tag_len());
// Copy header unencrypted.
writer.extend_from_slice(&payload[..header_len]);
let nonce = self.rtp_initialization_vector(header, roc);
let encrypted = self.srtp_cipher.encrypt(
Nonce::from_slice(&nonce),
Payload {
msg: &payload[header_len..],
aad: &writer,
},
)?;
writer.extend(encrypted);
Ok(writer.freeze())
}
fn decrypt_rtp(
&mut self,
ciphertext: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes> {
if ciphertext.len() < self.auth_tag_len() {
return Err(Error::ErrFailedToVerifyAuthTag);
}
let nonce = self.rtp_initialization_vector(header, roc);
let payload_offset = header.marshal_size();
let decrypted_msg: Vec<u8> = self.srtp_cipher.decrypt(
Nonce::from_slice(&nonce),
Payload {
msg: &ciphertext[payload_offset..],
aad: &ciphertext[..payload_offset],
},
)?;
let mut writer = BytesMut::with_capacity(payload_offset + decrypted_msg.len());
writer.extend_from_slice(&ciphertext[..payload_offset]);
writer.extend(decrypted_msg);
Ok(writer.freeze())
}
fn encrypt_rtcp(&mut self, decrypted: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes> {
let iv = self.rtcp_initialization_vector(srtcp_index, ssrc);
let aad = self.rtcp_additional_authenticated_data(decrypted, srtcp_index);
let encrypted_data = self.srtcp_cipher.encrypt(
Nonce::from_slice(&iv),
Payload {
msg: &decrypted[8..],
aad: &aad,
},
)?;
let mut writer = BytesMut::with_capacity(encrypted_data.len() + aad.len());
writer.extend_from_slice(&decrypted[..8]);
writer.extend(encrypted_data);
writer.extend_from_slice(&aad[8..]);
Ok(writer.freeze())
}
fn decrypt_rtcp(&mut self, encrypted: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes> {
if encrypted.len() < self.auth_tag_len() + SRTCP_INDEX_SIZE {
return Err(Error::ErrFailedToVerifyAuthTag);
}
let nonce = self.rtcp_initialization_vector(srtcp_index, ssrc);
let aad = self.rtcp_additional_authenticated_data(encrypted, srtcp_index);
let decrypted_data = self.srtcp_cipher.decrypt(
Nonce::from_slice(&nonce),
Payload {
msg: &encrypted[8..(encrypted.len() - SRTCP_INDEX_SIZE)],
aad: &aad,
},
)?;
let mut writer = BytesMut::with_capacity(8 + decrypted_data.len());
writer.extend_from_slice(&encrypted[..8]);
writer.extend(decrypted_data);
Ok(writer.freeze())
}
fn get_rtcp_index(&self, input: &[u8]) -> usize {
let pos = input.len() - 4;
let val = BigEndian::read_u32(&input[pos..]);
(val & !((RTCP_ENCRYPTION_FLAG as u32) << 24)) as usize
}
}
impl CipherAeadAesGcm {
/// Create a new AEAD instance.
pub(crate) fn new(master_key: &[u8], master_salt: &[u8]) -> Result<CipherAeadAesGcm> {
let srtp_session_key = aes_cm_key_derivation(
LABEL_SRTP_ENCRYPTION,
master_key,
master_salt,
0,
master_key.len(),
)?;
let srtp_block = GenericArray::from_slice(&srtp_session_key);
let srtp_cipher = Aes128Gcm::new(srtp_block);
let srtcp_session_key = aes_cm_key_derivation(
LABEL_SRTCP_ENCRYPTION,
master_key,
master_salt,
0,
master_key.len(),
)?;
let srtcp_block = GenericArray::from_slice(&srtcp_session_key);
let srtcp_cipher = Aes128Gcm::new(srtcp_block);
let srtp_session_salt = aes_cm_key_derivation(
LABEL_SRTP_SALT,
master_key,
master_salt,
0,
master_key.len(),
)?;
let srtcp_session_salt = aes_cm_key_derivation(
LABEL_SRTCP_SALT,
master_key,
master_salt,
0,
master_key.len(),
)?;
Ok(CipherAeadAesGcm {
srtp_cipher,
srtcp_cipher,
srtp_session_salt,
srtcp_session_salt,
})
}
/// The 12-octet IV used by AES-GCM SRTP is formed by first concatenating
/// 2 octets of zeroes, the 4-octet SSRC, the 4-octet rollover counter
/// (ROC), and the 2-octet sequence number (SEQ). The resulting 12-octet
/// value is then XORed to the 12-octet salt to form the 12-octet IV.
///
/// https://tools.ietf.org/html/rfc7714#section-8.1
pub(crate) fn rtp_initialization_vector(
&self,
header: &rtp::header::Header,
roc: u32,
) -> Vec<u8> {
let mut iv = vec![0u8; 12];
BigEndian::write_u32(&mut iv[2..], header.ssrc);
BigEndian::write_u32(&mut iv[6..], roc);
BigEndian::write_u16(&mut iv[10..], header.sequence_number);
for (i, v) in iv.iter_mut().enumerate() {
*v ^= self.srtp_session_salt[i];
}
iv
}
/// The 12-octet IV used by AES-GCM SRTCP is formed by first
/// concatenating 2 octets of zeroes, the 4-octet SSRC identifier,
/// 2 octets of zeroes, a single "0" bit, and the 31-bit SRTCP index.
/// The resulting 12-octet value is then XORed to the 12-octet salt to
/// form the 12-octet IV.
///
/// https://tools.ietf.org/html/rfc7714#section-9.1
pub(crate) fn rtcp_initialization_vector(&self, srtcp_index: usize, ssrc: u32) -> Vec<u8> {
let mut iv = vec![0u8; 12];
BigEndian::write_u32(&mut iv[2..], ssrc);
BigEndian::write_u32(&mut iv[8..], srtcp_index as u32);
for (i, v) in iv.iter_mut().enumerate() {
*v ^= self.srtcp_session_salt[i];
}
iv
}
/// In an SRTCP packet, a 1-bit Encryption flag is prepended to the
/// 31-bit SRTCP index to form a 32-bit value we shall call the
/// "ESRTCP word"
///
/// https://tools.ietf.org/html/rfc7714#section-17
pub(crate) fn rtcp_additional_authenticated_data(
&self,
rtcp_packet: &[u8],
srtcp_index: usize,
) -> Vec<u8> {
let mut aad = vec![0u8; 12];
aad[..8].copy_from_slice(&rtcp_packet[..8]);
BigEndian::write_u32(&mut aad[8..], srtcp_index as u32);
aad[8] |= RTCP_ENCRYPTION_FLAG;
aad
}
}
@@ -0,0 +1,220 @@
use aes::cipher::generic_array::GenericArray;
use aes::cipher::{KeyIvInit, StreamCipher, StreamCipherSeek};
use bytes::{BufMut, Bytes};
use rtcp::header::{HEADER_LENGTH, SSRC_LENGTH};
use subtle::ConstantTimeEq;
use util::marshal::*;
use super::{Cipher, CipherInner};
use crate::error::{Error, Result};
use crate::key_derivation::*;
type Aes128Ctr = ctr::Ctr128BE<aes::Aes128>;
pub(crate) struct CipherAesCmHmacSha1 {
inner: CipherInner,
srtp_session_key: Vec<u8>,
srtcp_session_key: Vec<u8>,
}
impl CipherAesCmHmacSha1 {
pub fn new(master_key: &[u8], master_salt: &[u8]) -> Result<Self> {
let inner = CipherInner::new(master_key, master_salt)?;
let srtp_session_key = aes_cm_key_derivation(
LABEL_SRTP_ENCRYPTION,
master_key,
master_salt,
0,
master_key.len(),
)?;
let srtcp_session_key = aes_cm_key_derivation(
LABEL_SRTCP_ENCRYPTION,
master_key,
master_salt,
0,
master_key.len(),
)?;
Ok(CipherAesCmHmacSha1 {
inner,
srtp_session_key,
srtcp_session_key,
})
}
}
impl Cipher for CipherAesCmHmacSha1 {
fn auth_tag_len(&self) -> usize {
self.inner.auth_tag_len()
}
fn get_rtcp_index(&self, input: &[u8]) -> usize {
self.inner.get_rtcp_index(input)
}
fn encrypt_rtp(
&mut self,
plaintext: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes> {
let mut writer = Vec::with_capacity(plaintext.len() + self.auth_tag_len());
// Write the plaintext to the destination buffer.
writer.extend_from_slice(plaintext);
// Encrypt the payload
let counter = generate_counter(
header.sequence_number,
roc,
header.ssrc,
&self.inner.srtp_session_salt,
);
let key = GenericArray::from_slice(&self.srtp_session_key);
let nonce = GenericArray::from_slice(&counter);
let mut stream = Aes128Ctr::new(key, nonce);
stream.apply_keystream(&mut writer[header.marshal_size()..]);
// Generate the auth tag.
let auth_tag = &self.inner.generate_srtp_auth_tag(&writer, roc)[..self.auth_tag_len()];
writer.extend(auth_tag);
Ok(Bytes::from(writer))
}
fn decrypt_rtp(
&mut self,
encrypted: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes> {
let encrypted_len = encrypted.len();
if encrypted_len < self.auth_tag_len() {
return Err(Error::SrtpTooSmall(encrypted_len, self.auth_tag_len()));
}
let mut writer = Vec::with_capacity(encrypted_len - self.auth_tag_len());
// Split the auth tag and the cipher text into two parts.
let actual_tag = &encrypted[encrypted_len - self.auth_tag_len()..];
let cipher_text = &encrypted[..encrypted_len - self.auth_tag_len()];
// Generate the auth tag we expect to see from the ciphertext.
let expected_tag =
&self.inner.generate_srtp_auth_tag(cipher_text, roc)[..self.auth_tag_len()];
// See if the auth tag actually matches.
// We use a constant time comparison to prevent timing attacks.
if actual_tag.ct_eq(expected_tag).unwrap_u8() != 1 {
return Err(Error::RtpFailedToVerifyAuthTag);
}
// Write cipher_text to the destination buffer.
writer.extend_from_slice(cipher_text);
// Decrypt the ciphertext for the payload.
let counter = generate_counter(
header.sequence_number,
roc,
header.ssrc,
&self.inner.srtp_session_salt,
);
let key = GenericArray::from_slice(&self.srtp_session_key);
let nonce = GenericArray::from_slice(&counter);
let mut stream = Aes128Ctr::new(key, nonce);
stream.seek(0);
stream.apply_keystream(&mut writer[header.marshal_size()..]);
Ok(Bytes::from(writer))
}
fn encrypt_rtcp(&mut self, decrypted: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes> {
let mut writer =
Vec::with_capacity(decrypted.len() + SRTCP_INDEX_SIZE + self.auth_tag_len());
// Write the decrypted to the destination buffer.
writer.extend_from_slice(decrypted);
// Encrypt everything after header
let counter = generate_counter(
(srtcp_index & 0xFFFF) as u16,
(srtcp_index >> 16) as u32,
ssrc,
&self.inner.srtcp_session_salt,
);
let key = GenericArray::from_slice(&self.srtcp_session_key);
let nonce = GenericArray::from_slice(&counter);
let mut stream = Aes128Ctr::new(key, nonce);
stream.apply_keystream(&mut writer[HEADER_LENGTH + SSRC_LENGTH..]);
// Add SRTCP index and set Encryption bit
writer.put_u32(srtcp_index as u32 | (1u32 << 31));
// Generate the auth tag.
let auth_tag = &self.inner.generate_srtcp_auth_tag(&writer)[..self.auth_tag_len()];
writer.extend(auth_tag);
Ok(Bytes::from(writer))
}
fn decrypt_rtcp(&mut self, encrypted: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes> {
let encrypted_len = encrypted.len();
if encrypted_len < self.auth_tag_len() + SRTCP_INDEX_SIZE {
return Err(Error::SrtcpTooSmall(
encrypted_len,
self.auth_tag_len() + SRTCP_INDEX_SIZE,
));
}
let tail_offset = encrypted_len - (self.auth_tag_len() + SRTCP_INDEX_SIZE);
let mut writer = Vec::with_capacity(tail_offset);
writer.extend_from_slice(&encrypted[0..tail_offset]);
let is_encrypted = encrypted[tail_offset] >> 7;
if is_encrypted == 0 {
return Ok(Bytes::from(writer));
}
// Split the auth tag and the cipher text into two parts.
let actual_tag = &encrypted[encrypted_len - self.auth_tag_len()..];
if actual_tag.len() != self.auth_tag_len() {
return Err(Error::RtcpInvalidLengthAuthTag(
actual_tag.len(),
self.auth_tag_len(),
));
}
let cipher_text = &encrypted[..encrypted_len - self.auth_tag_len()];
// Generate the auth tag we expect to see from the ciphertext.
let expected_tag = &self.inner.generate_srtcp_auth_tag(cipher_text)[..self.auth_tag_len()];
// See if the auth tag actually matches.
// We use a constant time comparison to prevent timing attacks.
if actual_tag.ct_eq(expected_tag).unwrap_u8() != 1 {
return Err(Error::RtcpFailedToVerifyAuthTag);
}
let counter = generate_counter(
(srtcp_index & 0xFFFF) as u16,
(srtcp_index >> 16) as u32,
ssrc,
&self.inner.srtcp_session_salt,
);
let key = GenericArray::from_slice(&self.srtcp_session_key);
let nonce = GenericArray::from_slice(&counter);
let mut stream = Aes128Ctr::new(key, nonce);
stream.seek(0);
stream.apply_keystream(&mut writer[HEADER_LENGTH + SSRC_LENGTH..]);
Ok(Bytes::from(writer))
}
}
@@ -0,0 +1,133 @@
use byteorder::{BigEndian, ByteOrder};
use hmac::{Hmac, Mac};
use sha1::Sha1;
use super::Cipher;
use crate::error::{Error, Result};
use crate::key_derivation::*;
use crate::protection_profile::*;
#[cfg(not(feature = "openssl"))]
mod ctrcipher;
#[cfg(feature = "openssl")]
mod opensslcipher;
#[cfg(not(feature = "openssl"))]
pub(crate) use ctrcipher::CipherAesCmHmacSha1;
#[cfg(feature = "openssl")]
pub(crate) use opensslcipher::CipherAesCmHmacSha1;
type HmacSha1 = Hmac<Sha1>;
pub const CIPHER_AES_CM_HMAC_SHA1AUTH_TAG_LEN: usize = 10;
pub(crate) struct CipherInner {
srtp_session_salt: Vec<u8>,
srtp_session_auth: HmacSha1,
srtcp_session_salt: Vec<u8>,
srtcp_session_auth: HmacSha1,
}
impl CipherInner {
pub fn new(master_key: &[u8], master_salt: &[u8]) -> Result<Self> {
let srtp_session_salt = aes_cm_key_derivation(
LABEL_SRTP_SALT,
master_key,
master_salt,
0,
master_salt.len(),
)?;
let srtcp_session_salt = aes_cm_key_derivation(
LABEL_SRTCP_SALT,
master_key,
master_salt,
0,
master_salt.len(),
)?;
let auth_key_len = ProtectionProfile::Aes128CmHmacSha1_80.auth_key_len();
let srtp_session_auth_tag = aes_cm_key_derivation(
LABEL_SRTP_AUTHENTICATION_TAG,
master_key,
master_salt,
0,
auth_key_len,
)?;
let srtcp_session_auth_tag = aes_cm_key_derivation(
LABEL_SRTCP_AUTHENTICATION_TAG,
master_key,
master_salt,
0,
auth_key_len,
)?;
let srtp_session_auth = HmacSha1::new_from_slice(&srtp_session_auth_tag)
.map_err(|e| Error::Other(e.to_string()))?;
let srtcp_session_auth = HmacSha1::new_from_slice(&srtcp_session_auth_tag)
.map_err(|e| Error::Other(e.to_string()))?;
Ok(Self {
srtp_session_salt,
srtp_session_auth,
srtcp_session_salt,
srtcp_session_auth,
})
}
/// https://tools.ietf.org/html/rfc3711#section-4.2
/// In the case of SRTP, M SHALL consist of the Authenticated
/// Portion of the packet (as specified in Figure 1) concatenated with
/// the roc, M = Authenticated Portion || roc;
///
/// The pre-defined authentication transform for SRTP is HMAC-SHA1
/// [RFC2104]. With HMAC-SHA1, the SRTP_PREFIX_LENGTH (Figure 3) SHALL
/// be 0. For SRTP (respectively SRTCP), the HMAC SHALL be applied to
/// the session authentication key and M as specified above, i.e.,
/// HMAC(k_a, M). The HMAC output SHALL then be truncated to the n_tag
/// left-most bits.
/// - Authenticated portion of the packet is everything BEFORE MKI
/// - k_a is the session message authentication key
/// - n_tag is the bit-length of the output authentication tag
fn generate_srtp_auth_tag(&self, buf: &[u8], roc: u32) -> [u8; 20] {
let mut signer = self.srtp_session_auth.clone();
signer.update(buf);
// For SRTP only, we need to hash the rollover counter as well.
signer.update(&roc.to_be_bytes());
signer.finalize().into_bytes().into()
}
/// https://tools.ietf.org/html/rfc3711#section-4.2
///
/// The pre-defined authentication transform for SRTP is HMAC-SHA1
/// [RFC2104]. With HMAC-SHA1, the SRTP_PREFIX_LENGTH (Figure 3) SHALL
/// be 0. For SRTP (respectively SRTCP), the HMAC SHALL be applied to
/// the session authentication key and M as specified above, i.e.,
/// HMAC(k_a, M). The HMAC output SHALL then be truncated to the n_tag
/// left-most bits.
/// - Authenticated portion of the packet is everything BEFORE MKI
/// - k_a is the session message authentication key
/// - n_tag is the bit-length of the output authentication tag
fn generate_srtcp_auth_tag(&self, buf: &[u8]) -> [u8; 20] {
let mut signer = self.srtcp_session_auth.clone();
signer.update(buf);
signer.finalize().into_bytes().into()
}
fn auth_tag_len(&self) -> usize {
CIPHER_AES_CM_HMAC_SHA1AUTH_TAG_LEN
}
fn get_rtcp_index(&self, input: &[u8]) -> usize {
let tail_offset = input.len() - (self.auth_tag_len() + SRTCP_INDEX_SIZE);
(BigEndian::read_u32(&input[tail_offset..tail_offset + SRTCP_INDEX_SIZE]) & !(1 << 31))
as usize
}
}
@@ -0,0 +1,261 @@
use bytes::{BufMut, Bytes};
use openssl::cipher_ctx::CipherCtx;
use rtcp::header::{HEADER_LENGTH, SSRC_LENGTH};
use subtle::ConstantTimeEq;
use util::marshal::*;
use super::{Cipher, CipherInner};
use crate::{
error::{Error, Result},
key_derivation::*,
};
pub(crate) struct CipherAesCmHmacSha1 {
inner: CipherInner,
rtp_ctx: CipherCtx,
rtcp_ctx: CipherCtx,
}
impl CipherAesCmHmacSha1 {
pub fn new(master_key: &[u8], master_salt: &[u8]) -> Result<Self> {
let inner = CipherInner::new(master_key, master_salt)?;
let srtp_session_key = aes_cm_key_derivation(
LABEL_SRTP_ENCRYPTION,
master_key,
master_salt,
0,
master_key.len(),
)?;
let srtcp_session_key = aes_cm_key_derivation(
LABEL_SRTCP_ENCRYPTION,
master_key,
master_salt,
0,
master_key.len(),
)?;
let t = openssl::cipher::Cipher::aes_128_ctr();
let mut rtp_ctx = CipherCtx::new().map_err(|e| Error::Other(e.to_string()))?;
rtp_ctx
.encrypt_init(Some(t), Some(&srtp_session_key[..]), None)
.map_err(|e| Error::Other(e.to_string()))?;
let t = openssl::cipher::Cipher::aes_128_ctr();
let mut rtcp_ctx = CipherCtx::new().map_err(|e| Error::Other(e.to_string()))?;
rtcp_ctx
.encrypt_init(Some(t), Some(&srtcp_session_key[..]), None)
.map_err(|e| Error::Other(e.to_string()))?;
Ok(Self {
inner,
rtp_ctx,
rtcp_ctx,
})
}
}
impl Cipher for CipherAesCmHmacSha1 {
fn auth_tag_len(&self) -> usize {
self.inner.auth_tag_len()
}
fn get_rtcp_index(&self, input: &[u8]) -> usize {
self.inner.get_rtcp_index(input)
}
fn encrypt_rtp(
&mut self,
plaintext: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes> {
let header_len = header.marshal_size();
let mut writer = Vec::with_capacity(plaintext.len() + self.auth_tag_len());
// Copy the header unencrypted.
writer.extend_from_slice(&plaintext[..header_len]);
// Encrypt the payload
let nonce = generate_counter(
header.sequence_number,
roc,
header.ssrc,
&self.inner.srtp_session_salt,
);
writer.resize(plaintext.len(), 0);
self.rtp_ctx.encrypt_init(None, None, Some(&nonce)).unwrap();
let count = self
.rtp_ctx
.cipher_update(&plaintext[header_len..], Some(&mut writer[header_len..]))
.unwrap();
self.rtp_ctx
.cipher_final(&mut writer[header_len + count..])
.unwrap();
// Generate and write the auth tag.
let auth_tag = &self.inner.generate_srtp_auth_tag(&writer, roc)[..self.auth_tag_len()];
writer.extend_from_slice(auth_tag);
Ok(Bytes::from(writer))
}
fn decrypt_rtp(
&mut self,
encrypted: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes> {
let encrypted_len = encrypted.len();
if encrypted_len < self.auth_tag_len() {
return Err(Error::SrtpTooSmall(encrypted_len, self.auth_tag_len()));
}
let header_len = header.marshal_size();
let mut writer = Vec::with_capacity(encrypted_len - self.auth_tag_len());
// Split the auth tag and the cipher text into two parts.
let actual_tag = &encrypted[encrypted_len - self.auth_tag_len()..];
let cipher_text = &encrypted[..encrypted_len - self.auth_tag_len()];
// Generate the auth tag we expect to see from the ciphertext.
let expected_tag =
&self.inner.generate_srtp_auth_tag(cipher_text, roc)[..self.auth_tag_len()];
// See if the auth tag actually matches.
// We use a constant time comparison to prevent timing attacks.
if actual_tag.ct_eq(expected_tag).unwrap_u8() != 1 {
return Err(Error::RtpFailedToVerifyAuthTag);
}
// Write cipher_text to the destination buffer.
writer.extend_from_slice(&cipher_text[..header_len]);
// Decrypt the ciphertext for the payload.
let nonce = generate_counter(
header.sequence_number,
roc,
header.ssrc,
&self.inner.srtp_session_salt,
);
writer.resize(encrypted_len - self.auth_tag_len(), 0);
self.rtp_ctx.decrypt_init(None, None, Some(&nonce)).unwrap();
let count = self
.rtp_ctx
.cipher_update(&cipher_text[header_len..], Some(&mut writer[header_len..]))
.unwrap();
self.rtp_ctx
.cipher_final(&mut writer[header_len + count..])
.unwrap();
Ok(Bytes::from(writer))
}
fn encrypt_rtcp(&mut self, decrypted: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes> {
let decrypted_len = decrypted.len();
let mut writer = Vec::with_capacity(decrypted_len + SRTCP_INDEX_SIZE + self.auth_tag_len());
// Write the decrypted to the destination buffer.
writer.extend_from_slice(&decrypted[..HEADER_LENGTH + SSRC_LENGTH]);
// Encrypt everything after header
let nonce = generate_counter(
(srtcp_index & 0xFFFF) as u16,
(srtcp_index >> 16) as u32,
ssrc,
&self.inner.srtcp_session_salt,
);
writer.resize(decrypted_len, 0);
self.rtcp_ctx
.encrypt_init(None, None, Some(&nonce))
.unwrap();
let count = self
.rtcp_ctx
.cipher_update(
&decrypted[HEADER_LENGTH + SSRC_LENGTH..],
Some(&mut writer[HEADER_LENGTH + SSRC_LENGTH..]),
)
.unwrap();
self.rtcp_ctx
.cipher_final(&mut writer[HEADER_LENGTH + SSRC_LENGTH + count..])
.unwrap();
// Add SRTCP index and set Encryption bit
writer.put_u32(srtcp_index as u32 | (1u32 << 31));
// Generate the auth tag.
let auth_tag = &self.inner.generate_srtcp_auth_tag(&writer)[..self.auth_tag_len()];
writer.extend(auth_tag);
Ok(Bytes::from(writer))
}
fn decrypt_rtcp(&mut self, encrypted: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes> {
let encrypted_len = encrypted.len();
if encrypted_len < self.auth_tag_len() + SRTCP_INDEX_SIZE {
return Err(Error::SrtcpTooSmall(
encrypted_len,
self.auth_tag_len() + SRTCP_INDEX_SIZE,
));
}
let tail_offset = encrypted_len - (self.auth_tag_len() + SRTCP_INDEX_SIZE);
let mut writer = Vec::with_capacity(tail_offset);
writer.extend_from_slice(&encrypted[..HEADER_LENGTH + SSRC_LENGTH]);
let is_encrypted = encrypted[tail_offset] >> 7;
if is_encrypted == 0 {
return Ok(Bytes::from(writer));
}
// Split the auth tag and the cipher text into two parts.
let actual_tag = &encrypted[encrypted_len - self.auth_tag_len()..];
if actual_tag.len() != self.auth_tag_len() {
return Err(Error::RtcpInvalidLengthAuthTag(
actual_tag.len(),
self.auth_tag_len(),
));
}
let cipher_text = &encrypted[..encrypted_len - self.auth_tag_len()];
// Generate the auth tag we expect to see from the ciphertext.
let expected_tag = &self.inner.generate_srtcp_auth_tag(cipher_text)[..self.auth_tag_len()];
// See if the auth tag actually matches.
// We use a constant time comparison to prevent timing attacks.
if actual_tag.ct_eq(expected_tag).unwrap_u8() != 1 {
return Err(Error::RtcpFailedToVerifyAuthTag);
}
let nonce = generate_counter(
(srtcp_index & 0xFFFF) as u16,
(srtcp_index >> 16) as u32,
ssrc,
&self.inner.srtcp_session_salt,
);
writer.resize(tail_offset, 0);
self.rtcp_ctx
.decrypt_init(None, None, Some(&nonce))
.unwrap();
let count = self
.rtcp_ctx
.cipher_update(
&encrypted[HEADER_LENGTH + SSRC_LENGTH..tail_offset],
Some(&mut writer[HEADER_LENGTH + SSRC_LENGTH..]),
)
.unwrap();
self.rtcp_ctx
.cipher_final(&mut writer[HEADER_LENGTH + SSRC_LENGTH + count..])
.unwrap();
Ok(Bytes::from(writer))
}
}
+61
View File
@@ -0,0 +1,61 @@
pub mod cipher_aead_aes_gcm;
pub mod cipher_aes_cm_hmac_sha1;
use bytes::Bytes;
use crate::error::Result;
///NOTE: Auth tag and AEAD auth tag are placed at the different position in SRTCP
///
///In non-AEAD cipher, the authentication tag is placed *after* the ESRTCP word
///(Encrypted-flag and SRTCP index).
///
///> AES_128_CM_HMAC_SHA1_80
///> | RTCP Header | Encrypted payload |E| SRTCP Index | Auth tag |
///> ^ |----------|
///> | ^
///> | authTagLen=10
///> aeadAuthTagLen=0
///
///In AEAD cipher, the AEAD authentication tag is embedded in the ciphertext.
///It is *before* the ESRTCP word (Encrypted-flag and SRTCP index).
///
///> AEAD_AES_128_GCM
///> | RTCP Header | Encrypted payload | AEAD auth tag |E| SRTCP Index |
///> |---------------| ^
///> ^ authTagLen=0
///> aeadAuthTagLen=16
///
///See https://tools.ietf.org/html/rfc7714 for the full specifications.
/// Cipher represents a implementation of one
/// of the SRTP Specific ciphers.
pub(crate) trait Cipher {
/// Get authenticated tag length.
fn auth_tag_len(&self) -> usize;
/// Retrieved RTCP index.
fn get_rtcp_index(&self, input: &[u8]) -> usize;
/// Encrypt RTP payload.
fn encrypt_rtp(
&mut self,
payload: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes>;
/// Decrypt RTP payload.
fn decrypt_rtp(
&mut self,
payload: &[u8],
header: &rtp::header::Header,
roc: u32,
) -> Result<Bytes>;
/// Encrypt RTCP payload.
fn encrypt_rtcp(&mut self, payload: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes>;
/// Decrypt RTCP payload.
fn decrypt_rtcp(&mut self, payload: &[u8], srtcp_index: usize, ssrc: u32) -> Result<Bytes>;
}
+83
View File
@@ -0,0 +1,83 @@
use util::KeyingMaterialExporter;
use crate::error::Result;
use crate::option::*;
use crate::protection_profile::*;
const LABEL_EXTRACTOR_DTLS_SRTP: &str = "EXTRACTOR-dtls_srtp";
/// SessionKeys bundles the keys required to setup an SRTP session
#[derive(Default, Debug, Clone)]
pub struct SessionKeys {
pub local_master_key: Vec<u8>,
pub local_master_salt: Vec<u8>,
pub remote_master_key: Vec<u8>,
pub remote_master_salt: Vec<u8>,
}
/// Config is used to configure a session.
/// You can provide either a KeyingMaterialExporter to export keys
/// or directly pass the keys themselves.
/// After a Config is passed to a session it must not be modified.
#[derive(Default)]
pub struct Config {
pub keys: SessionKeys,
pub profile: ProtectionProfile,
//LoggerFactory: logging.LoggerFactory
/// List of local/remote context options.
/// ReplayProtection is enabled on remote context by default.
/// Default replay protection window size is 64.
pub local_rtp_options: Option<ContextOption>,
pub remote_rtp_options: Option<ContextOption>,
pub local_rtcp_options: Option<ContextOption>,
pub remote_rtcp_options: Option<ContextOption>,
}
impl Config {
/// ExtractSessionKeysFromDTLS allows setting the Config SessionKeys by
/// extracting them from DTLS. This behavior is defined in RFC5764:
/// https://tools.ietf.org/html/rfc5764
pub async fn extract_session_keys_from_dtls(
&mut self,
exporter: impl KeyingMaterialExporter,
is_client: bool,
) -> Result<()> {
let key_len = self.profile.key_len();
let salt_len = self.profile.salt_len();
let keying_material = exporter
.export_keying_material(
LABEL_EXTRACTOR_DTLS_SRTP,
&[],
(key_len * 2) + (salt_len * 2),
)
.await?;
let mut offset = 0;
let client_write_key = keying_material[offset..offset + key_len].to_vec();
offset += key_len;
let server_write_key = keying_material[offset..offset + key_len].to_vec();
offset += key_len;
let client_write_salt = keying_material[offset..offset + salt_len].to_vec();
offset += salt_len;
let server_write_salt = keying_material[offset..offset + salt_len].to_vec();
if is_client {
self.keys.local_master_key = client_write_key;
self.keys.local_master_salt = client_write_salt;
self.keys.remote_master_key = server_write_key;
self.keys.remote_master_salt = server_write_salt;
} else {
self.keys.local_master_key = server_write_key;
self.keys.local_master_salt = server_write_salt;
self.keys.remote_master_key = client_write_key;
self.keys.remote_master_salt = client_write_salt;
}
Ok(())
}
}
+341
View File
@@ -0,0 +1,341 @@
use bytes::Bytes;
use lazy_static::lazy_static;
use super::*;
use crate::key_derivation::*;
const CIPHER_CONTEXT_ALGO: ProtectionProfile = ProtectionProfile::Aes128CmHmacSha1_80;
const DEFAULT_SSRC: u32 = 0;
#[test]
fn test_context_roc() -> Result<()> {
let key_len = CIPHER_CONTEXT_ALGO.key_len();
let salt_len = CIPHER_CONTEXT_ALGO.salt_len();
let mut c = Context::new(
&vec![0; key_len],
&vec![0; salt_len],
CIPHER_CONTEXT_ALGO,
None,
None,
)?;
let roc = c.get_roc(123);
assert!(roc.is_none(), "ROC must return None for unused SSRC");
c.set_roc(123, 100);
let roc = c.get_roc(123);
if let Some(r) = roc {
assert_eq!(r, 100, "ROC is set to 100, but returned {r}")
} else {
panic!("ROC must return value for used SSRC");
}
Ok(())
}
#[test]
fn test_context_index() -> Result<()> {
let key_len = CIPHER_CONTEXT_ALGO.key_len();
let salt_len = CIPHER_CONTEXT_ALGO.salt_len();
let mut c = Context::new(
&vec![0; key_len],
&vec![0; salt_len],
CIPHER_CONTEXT_ALGO,
None,
None,
)?;
let index = c.get_index(123);
assert!(index.is_none(), "Index must return None for unused SSRC");
c.set_index(123, 100);
let index = c.get_index(123);
if let Some(i) = index {
assert_eq!(i, 100, "Index is set to 100, but returned {i}");
} else {
panic!("Index must return true for used SSRC")
}
Ok(())
}
#[test]
fn test_key_len() -> Result<()> {
let key_len = CIPHER_CONTEXT_ALGO.key_len();
let salt_len = CIPHER_CONTEXT_ALGO.salt_len();
let result = Context::new(&[], &vec![0; salt_len], CIPHER_CONTEXT_ALGO, None, None);
assert!(result.is_err(), "CreateContext accepted a 0 length key");
let result = Context::new(&vec![0; key_len], &[], CIPHER_CONTEXT_ALGO, None, None);
assert!(result.is_err(), "CreateContext accepted a 0 length salt");
let result = Context::new(
&vec![0; key_len],
&vec![0; salt_len],
CIPHER_CONTEXT_ALGO,
None,
None,
);
assert!(
result.is_ok(),
"CreateContext failed with a valid length key and salt"
);
Ok(())
}
#[test]
fn test_valid_packet_counter() -> Result<()> {
let master_key = vec![
0x0d, 0xcd, 0x21, 0x3e, 0x4c, 0xbc, 0xf2, 0x8f, 0x01, 0x7f, 0x69, 0x94, 0x40, 0x1e, 0x28,
0x89,
];
let master_salt = vec![
0x62, 0x77, 0x60, 0x38, 0xc0, 0x6d, 0xc9, 0x41, 0x9f, 0x6d, 0xd9, 0x43, 0x3e, 0x7c,
];
let srtp_session_salt = aes_cm_key_derivation(
LABEL_SRTP_SALT,
&master_key,
&master_salt,
0,
master_salt.len(),
)?;
let s = SrtpSsrcState {
ssrc: 4160032510,
..Default::default()
};
let expected_counter = [
0xcf, 0x90, 0x1e, 0xa5, 0xda, 0xd3, 0x2c, 0x15, 0x00, 0xa2, 0x24, 0xae, 0xae, 0xaf, 0x00,
0x00,
];
let counter = generate_counter(32846, s.rollover_counter, s.ssrc, &srtp_session_salt);
assert_eq!(
counter, expected_counter,
"Session Key {counter:?} does not match expected {expected_counter:?}",
);
Ok(())
}
#[test]
fn test_rollover_count() -> Result<()> {
let mut s = SrtpSsrcState {
ssrc: DEFAULT_SSRC,
..Default::default()
};
// Set initial seqnum
let roc = s.next_rollover_count(65530);
assert_eq!(roc, 0, "Initial rolloverCounter must be 0");
s.update_rollover_count(65530);
// Invalid packets never update ROC
s.next_rollover_count(0);
s.next_rollover_count(0x4000);
s.next_rollover_count(0x8000);
s.next_rollover_count(0xFFFF);
s.next_rollover_count(0);
// We rolled over to 0
let roc = s.next_rollover_count(0);
assert_eq!(roc, 1, "rolloverCounter was not updated after it crossed 0");
s.update_rollover_count(0);
let roc = s.next_rollover_count(65530);
assert_eq!(
roc, 0,
"rolloverCounter was not updated when it rolled back, failed to handle out of order"
);
s.update_rollover_count(65530);
let roc = s.next_rollover_count(5);
assert_eq!(
roc, 1,
"rolloverCounter was not updated when it rolled over initial, to handle out of order"
);
s.update_rollover_count(5);
s.next_rollover_count(6);
s.update_rollover_count(6);
s.next_rollover_count(7);
s.update_rollover_count(7);
let roc = s.next_rollover_count(8);
assert_eq!(
roc, 1,
"rolloverCounter was improperly updated for non-significant packets"
);
s.update_rollover_count(8);
// valid packets never update ROC
let roc = s.next_rollover_count(0x4000);
assert_eq!(
roc, 1,
"rolloverCounter was improperly updated for non-significant packets"
);
s.update_rollover_count(0x4000);
let roc = s.next_rollover_count(0x8000);
assert_eq!(
roc, 1,
"rolloverCounter was improperly updated for non-significant packets"
);
s.update_rollover_count(0x8000);
let roc = s.next_rollover_count(0xFFFF);
assert_eq!(
roc, 1,
"rolloverCounter was improperly updated for non-significant packets"
);
s.update_rollover_count(0xFFFF);
let roc = s.next_rollover_count(0);
assert_eq!(
roc, 2,
"rolloverCounter must be incremented after wrapping, got {roc}"
);
Ok(())
}
lazy_static! {
static ref MASTER_KEY: Bytes = Bytes::from_static(&[
0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e,
0x0f,
]);
static ref MASTER_SALT: Bytes = Bytes::from_static(&[
0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xab,
]);
static ref DECRYPTED_RTP_PACKET: Bytes = Bytes::from_static(&[
0x80, 0x0f, 0x12, 0x34, 0xde, 0xca, 0xfb, 0xad, 0xca, 0xfe, 0xba, 0xbe, 0xab, 0xab, 0xab,
0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab,
]);
static ref ENCRYPTED_RTP_PACKET: Bytes = Bytes::from_static(&[
0x80, 0x0f, 0x12, 0x34, 0xde, 0xca, 0xfb, 0xad, 0xca, 0xfe, 0xba, 0xbe, 0xc5, 0x00, 0x2e,
0xde, 0x04, 0xcf, 0xdd, 0x2e, 0xb9, 0x11, 0x59, 0xe0, 0x88, 0x0a, 0xa0, 0x6e, 0xd2, 0x97,
0x68, 0x26, 0xf7, 0x96, 0xb2, 0x01, 0xdf, 0x31, 0x31, 0xa1, 0x27, 0xe8, 0xa3, 0x92,
]);
static ref DECRYPTED_RTCP_PACKET: Bytes = Bytes::from_static(&[
0x81, 0xc8, 0x00, 0x0b, 0xca, 0xfe, 0xba, 0xbe, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab,
0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab, 0xab,
]);
static ref ENCRYPTED_RTCP_PACKET: Bytes = Bytes::from_static(&[
0x81, 0xc8, 0x00, 0x0b, 0xca, 0xfe, 0xba, 0xbe, 0xc9, 0x8b, 0x8b, 0x5d, 0xf0, 0x39, 0x2a,
0x55, 0x85, 0x2b, 0x6c, 0x21, 0xac, 0x8e, 0x70, 0x25, 0xc5, 0x2c, 0x6f, 0xbe, 0xa2, 0xb3,
0xb4, 0x46, 0xea, 0x31, 0x12, 0x3b, 0xa8, 0x8c, 0xe6, 0x1e, 0x80, 0x00, 0x00, 0x01,
]);
}
#[test]
fn test_encrypt_rtp() {
let mut ctx = Context::new(
&MASTER_KEY,
&MASTER_SALT,
ProtectionProfile::AeadAes128Gcm,
None,
None,
)
.expect("Error creating srtp context");
let gotten_encrypted_rtp_packet = ctx
.encrypt_rtp(&DECRYPTED_RTP_PACKET)
.expect("Error encrypting rtp payload");
assert_eq!(gotten_encrypted_rtp_packet, *ENCRYPTED_RTP_PACKET)
}
#[test]
fn test_decrypt_rtp() {
let mut ctx = Context::new(
&MASTER_KEY,
&MASTER_SALT,
ProtectionProfile::AeadAes128Gcm,
None,
None,
)
.expect("Error creating srtp context");
let gotten_decrypted_rtp_packet = ctx
.decrypt_rtp(&ENCRYPTED_RTP_PACKET)
.expect("Error decrypting rtp payload");
assert_eq!(gotten_decrypted_rtp_packet, *DECRYPTED_RTP_PACKET)
}
#[test]
fn test_encrypt_rtcp() {
let mut ctx = Context::new(
&MASTER_KEY,
&MASTER_SALT,
ProtectionProfile::AeadAes128Gcm,
None,
None,
)
.expect("Error creating srtp context");
let gotten_encrypted_rtcp_packet = ctx
.encrypt_rtcp(&DECRYPTED_RTCP_PACKET)
.expect("Error encrypting rtcp payload");
assert_eq!(gotten_encrypted_rtcp_packet, *ENCRYPTED_RTCP_PACKET)
}
#[test]
fn test_decrypt_rtcp() {
let mut ctx = Context::new(
&MASTER_KEY,
&MASTER_SALT,
ProtectionProfile::AeadAes128Gcm,
None,
None,
)
.expect("Error creating srtp context");
let gotten_decrypted_rtcp_packet = ctx
.decrypt_rtcp(&ENCRYPTED_RTCP_PACKET)
.expect("Error decrypting rtcp payload");
assert_eq!(gotten_decrypted_rtcp_packet, *DECRYPTED_RTCP_PACKET)
}
#[test]
fn test_rollover_high_water_mark() {
let mut state = super::SrtpSsrcState::default();
for seq in [65000, 65535, 85] {
state.update_rollover_count(seq);
}
for seq in [65000, 65530, 84, 0] {
state.update_rollover_count(seq);
assert_eq!(state.rollover_counter, 1);
assert_eq!(state.last_sequence_number, 85);
}
assert_eq!(state.next_rollover_count(86), 1);
state.update_rollover_count(1000);
state.update_rollover_count(32000);
state.update_rollover_count(64000);
assert_eq!(state.next_rollover_count(1000), 2);
state.update_rollover_count(1000);
assert_eq!(state.rollover_counter, 2);
assert_eq!(state.last_sequence_number, 1000);
}
#[test]
fn test_rollover_initial_zero_and_counter_wrap() {
let mut state = super::SrtpSsrcState::default();
state.update_rollover_count(0);
assert_eq!(state.next_rollover_count(65535), u32::MAX);
state.update_rollover_count(65535);
assert_eq!(state.rollover_counter, 0);
assert_eq!(state.last_sequence_number, 0);
state.rollover_counter = u32::MAX;
state.last_sequence_number = 65535;
state.update_rollover_count(0);
assert_eq!(state.rollover_counter, 0);
assert_eq!(state.last_sequence_number, 0);
}
+177
View File
@@ -0,0 +1,177 @@
#[cfg(test)]
mod context_test;
#[cfg(test)]
mod rollover_regression_test;
#[cfg(test)]
mod srtcp_test;
#[cfg(test)]
mod srtp_test;
use std::collections::HashMap;
use util::replay_detector::*;
use crate::cipher::cipher_aead_aes_gcm::*;
use crate::cipher::cipher_aes_cm_hmac_sha1::*;
use crate::cipher::*;
use crate::error::{Error, Result};
use crate::option::*;
use crate::protection_profile::*;
pub mod srtcp;
pub mod srtp;
const SEQUENCE_HALF_RANGE: i32 = 1 << 15;
/// Encrypt/Decrypt state for a single SRTP SSRC
#[derive(Default)]
pub(crate) struct SrtpSsrcState {
ssrc: u32,
rollover_counter: u32,
rollover_has_processed: bool,
last_sequence_number: u16,
replay_detector: Option<Box<dyn ReplayDetector + Send + 'static>>,
}
/// Encrypt/Decrypt state for a single SRTCP SSRC
#[derive(Default)]
pub(crate) struct SrtcpSsrcState {
srtcp_index: usize,
ssrc: u32,
replay_detector: Option<Box<dyn ReplayDetector + Send + 'static>>,
}
impl SrtpSsrcState {
pub fn next_rollover_count(&self, sequence_number: u16) -> u32 {
let roc = self.rollover_counter;
if !self.rollover_has_processed {
return roc;
}
// RFC 3711 Appendix A: estimate the index nearest the highest packet.
let last = i32::from(self.last_sequence_number);
let seq = i32::from(sequence_number);
if last < SEQUENCE_HALF_RANGE {
if seq - last > SEQUENCE_HALF_RANGE {
return roc.wrapping_sub(1);
}
} else if last - SEQUENCE_HALF_RANGE > seq {
return roc.wrapping_add(1);
}
roc
}
pub fn update_rollover_count(&mut self, sequence_number: u16) {
let roc = self.next_rollover_count(sequence_number);
// Retransmissions must never move the highest authenticated index back.
if !self.rollover_has_processed
|| (roc == self.rollover_counter && sequence_number > self.last_sequence_number)
|| roc == self.rollover_counter.wrapping_add(1)
{
self.rollover_counter = roc;
self.last_sequence_number = sequence_number;
self.rollover_has_processed = true;
}
}
}
/// Context represents a SRTP cryptographic context
/// Context can only be used for one-way operations
/// it must either used ONLY for encryption or ONLY for decryption
pub struct Context {
cipher: Box<dyn Cipher + Send>,
srtp_ssrc_states: HashMap<u32, SrtpSsrcState>,
srtcp_ssrc_states: HashMap<u32, SrtcpSsrcState>,
new_srtp_replay_detector: ContextOption,
new_srtcp_replay_detector: ContextOption,
}
impl Context {
/// CreateContext creates a new SRTP Context
pub fn new(
master_key: &[u8],
master_salt: &[u8],
profile: ProtectionProfile,
srtp_ctx_opt: Option<ContextOption>,
srtcp_ctx_opt: Option<ContextOption>,
) -> Result<Context> {
let key_len = profile.key_len();
let salt_len = profile.salt_len();
if master_key.len() != key_len {
return Err(Error::SrtpMasterKeyLength(key_len, master_key.len()));
} else if master_salt.len() != salt_len {
return Err(Error::SrtpSaltLength(salt_len, master_salt.len()));
}
let cipher: Box<dyn Cipher + Send> = match profile {
ProtectionProfile::Aes128CmHmacSha1_80 => {
Box::new(CipherAesCmHmacSha1::new(master_key, master_salt)?)
}
ProtectionProfile::AeadAes128Gcm => {
Box::new(CipherAeadAesGcm::new(master_key, master_salt)?)
}
};
let srtp_ctx_opt = if let Some(ctx_opt) = srtp_ctx_opt {
ctx_opt
} else {
srtp_no_replay_protection()
};
let srtcp_ctx_opt = if let Some(ctx_opt) = srtcp_ctx_opt {
ctx_opt
} else {
srtcp_no_replay_protection()
};
Ok(Context {
cipher,
srtp_ssrc_states: HashMap::new(),
srtcp_ssrc_states: HashMap::new(),
new_srtp_replay_detector: srtp_ctx_opt,
new_srtcp_replay_detector: srtcp_ctx_opt,
})
}
fn get_srtp_ssrc_state(&mut self, ssrc: u32) -> &mut SrtpSsrcState {
let s = SrtpSsrcState {
ssrc,
replay_detector: Some((self.new_srtp_replay_detector)()),
..Default::default()
};
self.srtp_ssrc_states.entry(ssrc).or_insert(s)
}
fn get_srtcp_ssrc_state(&mut self, ssrc: u32) -> &mut SrtcpSsrcState {
let s = SrtcpSsrcState {
ssrc,
replay_detector: Some((self.new_srtcp_replay_detector)()),
..Default::default()
};
self.srtcp_ssrc_states.entry(ssrc).or_insert(s)
}
/// roc returns SRTP rollover counter value of specified SSRC.
fn get_roc(&self, ssrc: u32) -> Option<u32> {
self.srtp_ssrc_states.get(&ssrc).map(|s| s.rollover_counter)
}
/// set_roc sets SRTP rollover counter value of specified SSRC.
fn set_roc(&mut self, ssrc: u32, roc: u32) {
self.get_srtp_ssrc_state(ssrc).rollover_counter = roc;
}
/// index returns SRTCP index value of specified SSRC.
fn get_index(&self, ssrc: u32) -> Option<usize> {
self.srtcp_ssrc_states.get(&ssrc).map(|s| s.srtcp_index)
}
/// set_index sets SRTCP index value of specified SSRC.
fn set_index(&mut self, ssrc: u32, index: usize) {
self.get_srtcp_ssrc_state(ssrc).srtcp_index = index;
}
}
@@ -0,0 +1,73 @@
use crate::{context::Context, protection_profile::ProtectionProfile};
fn packet(seq: u16) -> (Vec<u8>, rtp::header::Header) {
let h = rtp::header::Header {
version: 2,
sequence_number: seq,
ssrc: 1,
..Default::default()
};
let mut b = vec![0x80, 96];
b.extend(seq.to_be_bytes());
b.extend(0u32.to_be_bytes());
b.extend(1u32.to_be_bytes());
b.extend([1, 2, 3, 4]);
(b, h)
}
fn run(profile: ProtectionProfile, retransmits: bool) -> usize {
let mut tx = Context::new(
&vec![0; 16],
&vec![
0;
match profile {
ProtectionProfile::Aes128CmHmacSha1_80 => 14,
_ => 12,
}
],
profile,
None,
None,
)
.unwrap();
let mut rx = Context::new(
&vec![0; 16],
&vec![
0;
match profile {
ProtectionProfile::Aes128CmHmacSha1_80 => 14,
_ => 12,
}
],
profile,
None,
None,
)
.unwrap();
for seq in (65000u16..=65535).chain(0..=85) {
let (b, h) = packet(seq);
let e = tx.encrypt_rtp_with_header(&b, &h).unwrap();
rx.decrypt_rtp(&e).unwrap();
}
if retransmits {
for seq in [65000, 65530] {
let (b, h) = packet(seq);
let _ = tx.encrypt_rtp_with_header(&b, &h).unwrap();
}
}
let mut failures = 0;
for seq in 86..=1000 {
let (b, h) = packet(seq);
let e = tx.encrypt_rtp_with_header(&b, &h).unwrap();
failures += usize::from(rx.decrypt_rtp(&e).is_err());
}
failures
}
#[test]
fn delayed_retransmissions_preserve_authentication() {
for profile in [
ProtectionProfile::Aes128CmHmacSha1_80,
ProtectionProfile::AeadAes128Gcm,
] {
assert_eq!(run(profile, false), 0);
assert_eq!(run(profile, true), 0);
}
}
+50
View File
@@ -0,0 +1,50 @@
use bytes::Bytes;
use util::marshal::*;
use super::*;
use crate::error::Result;
impl Context {
/// DecryptRTCP decrypts a RTCP packet with an encrypted payload
pub fn decrypt_rtcp(&mut self, encrypted: &[u8]) -> Result<Bytes> {
let mut buf = encrypted;
rtcp::header::Header::unmarshal(&mut buf)?;
let index = self.cipher.get_rtcp_index(encrypted);
let ssrc = u32::from_be_bytes([encrypted[4], encrypted[5], encrypted[6], encrypted[7]]);
if let Some(replay_detector) = &mut self.get_srtcp_ssrc_state(ssrc).replay_detector {
if !replay_detector.check(index as u64) {
return Err(Error::SrtcpSsrcDuplicated(ssrc, index));
}
}
let dst = self.cipher.decrypt_rtcp(encrypted, index, ssrc)?;
if let Some(replay_detector) = &mut self.get_srtcp_ssrc_state(ssrc).replay_detector {
replay_detector.accept();
}
Ok(dst)
}
/// EncryptRTCP marshals and encrypts an RTCP packet, writing to the dst buffer provided.
/// If the dst buffer does not have the capacity to hold `len(plaintext) + 14` bytes, a new one will be allocated and returned.
pub fn encrypt_rtcp(&mut self, decrypted: &[u8]) -> Result<Bytes> {
let mut buf = decrypted;
rtcp::header::Header::unmarshal(&mut buf)?;
let ssrc = u32::from_be_bytes([decrypted[4], decrypted[5], decrypted[6], decrypted[7]]);
let index = {
let state = self.get_srtcp_ssrc_state(ssrc);
state.srtcp_index += 1;
if state.srtcp_index > MAX_SRTCP_INDEX {
state.srtcp_index = 0;
}
state.srtcp_index
};
self.cipher.encrypt_rtcp(decrypted, index, ssrc)
}
}
+257
View File
@@ -0,0 +1,257 @@
use bytes::{Buf, Bytes, BytesMut};
use lazy_static::lazy_static;
use super::*;
use crate::key_derivation::*;
pub struct RTCPTestCase {
ssrc: u32,
index: usize,
encrypted: Bytes,
decrypted: Bytes,
}
lazy_static! {
static ref RTCP_TEST_MASTER_KEY: Bytes = Bytes::from_static(&[
0xfd, 0xa6, 0x25, 0x95, 0xd7, 0xf6, 0x92, 0x6f, 0x7d, 0x9c, 0x02, 0x4c, 0xc9, 0x20, 0x9f,
0x34
]);
static ref RTCP_TEST_MASTER_SALT: Bytes = Bytes::from_static(&[
0xa9, 0x65, 0x19, 0x85, 0x54, 0x0b, 0x47, 0xbe, 0x2f, 0x27, 0xa8, 0xb8, 0x81, 0x23
]);
static ref RTCP_TEST_CASES: Vec<RTCPTestCase> = vec![
RTCPTestCase {
ssrc: 0x66ef91ff,
index: 0,
encrypted: Bytes::from_static(&[
0x80, 0xc8, 0x00, 0x06, 0x66, 0xef, 0x91, 0xff, 0xcd, 0x34, 0xc5, 0x78, 0xb2, 0x8b,
0xe1, 0x6b, 0xc5, 0x09, 0xd5, 0x77, 0xe4, 0xce, 0x5f, 0x20, 0x80, 0x21, 0xbd, 0x66,
0x74, 0x65, 0xe9, 0x5f, 0x49, 0xe5, 0xf5, 0xc0, 0x68, 0x4e, 0xe5, 0x6a, 0x78, 0x07,
0x75, 0x46, 0xed, 0x90, 0xf6, 0xdc, 0x9d, 0xef, 0x3b, 0xdf, 0xf2, 0x79, 0xa9, 0xd8,
0x80, 0x00, 0x00, 0x01, 0x60, 0xc0, 0xae, 0xb5, 0x6f, 0x40, 0x88, 0x0e, 0x28, 0xba
]),
decrypted: Bytes::from_static(&[
0x80, 0xc8, 0x00, 0x06, 0x66, 0xef, 0x91, 0xff, 0xdf, 0x48, 0x80, 0xdd, 0x61, 0xa6,
0x2e, 0xd3, 0xd8, 0xbc, 0xde, 0xbe, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x16, 0x04,
0x81, 0xca, 0x00, 0x06, 0x66, 0xef, 0x91, 0xff, 0x01, 0x10, 0x52, 0x6e, 0x54, 0x35,
0x43, 0x6d, 0x4a, 0x68, 0x7a, 0x79, 0x65, 0x74, 0x41, 0x78, 0x77, 0x2b, 0x00, 0x00
]),
},
RTCPTestCase{
ssrc: 0x11111111,
index: 0,
encrypted: Bytes::from_static(&[
0x80, 0xc8, 0x00, 0x06, 0x11, 0x11, 0x11, 0x11, 0x17, 0x8c, 0x15, 0xf1, 0x4b, 0x11,
0xda, 0xf5, 0x74, 0x53, 0x86, 0x2b, 0xc9, 0x07, 0x29, 0x40, 0xbf, 0x22, 0xf6, 0x46,
0x11, 0xa4, 0xc1, 0x3a, 0xff, 0x5a, 0xbd, 0xd0, 0xf8, 0x8b, 0x38, 0xe4, 0x95, 0x38,
0x5d, 0xcf, 0x1b, 0xf5, 0x27, 0x77, 0xfb, 0xdb, 0x3f, 0x10, 0x68, 0x99, 0xd8, 0xad,
0x80, 0x00, 0x00, 0x01, 0x34, 0x3c, 0x2e, 0x83, 0x17, 0x13, 0x93, 0x69, 0xcf, 0xc0
]),
decrypted: Bytes::from_static(&[
0x80, 0xc8, 0x00, 0x06, 0x11, 0x11, 0x11, 0x11, 0xdf, 0x48, 0x80, 0xdd, 0x61, 0xa6,
0x2e, 0xd3, 0xd8, 0xbc, 0xde, 0xbe, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x16, 0x04,
0x81, 0xca, 0x00, 0x06, 0x66, 0xef, 0x91, 0xff, 0x01, 0x10, 0x52, 0x6e, 0x54, 0x35,
0x43, 0x6d, 0x4a, 0x68, 0x7a, 0x79, 0x65, 0x74, 0x41, 0x78, 0x77, 0x2b, 0x00, 0x00
]),
},
RTCPTestCase{
ssrc: 0x11111111,
index: 0x7ffffffe, // Upper boundary of index
encrypted: Bytes::from_static(&[
0x80, 0xc8, 0x00, 0x06, 0x11, 0x11, 0x11, 0x11, 0x17, 0x8c, 0x15, 0xf1, 0x4b, 0x11,
0xda, 0xf5, 0x74, 0x53, 0x86, 0x2b, 0xc9, 0x07, 0x29, 0x40, 0xbf, 0x22, 0xf6, 0x46,
0x11, 0xa4, 0xc1, 0x3a, 0xff, 0x5a, 0xbd, 0xd0, 0xf8, 0x8b, 0x38, 0xe4, 0x95, 0x38,
0x5d, 0xcf, 0x1b, 0xf5, 0x27, 0x77, 0xfb, 0xdb, 0x3f, 0x10, 0x68, 0x99, 0xd8, 0xad,
0xff, 0xff, 0xff, 0xff, 0x5a, 0x99, 0xce, 0xed, 0x9f, 0x2e, 0x4d, 0x9d, 0xfa, 0x97
]),
decrypted: Bytes::from_static(&[
0x80, 0xc8, 0x0, 0x6, 0x11, 0x11, 0x11, 0x11, 0x4, 0x99, 0x47, 0x53, 0xc4, 0x1e,
0xb9, 0xde, 0x52, 0xa3, 0x1d, 0x77, 0x2f, 0xff, 0xcc, 0x75, 0xbb, 0x6a, 0x29, 0xb8,
0x1, 0xb7, 0x2e, 0x4b, 0x4e, 0xcb, 0xa4, 0x81, 0x2d, 0x46, 0x4, 0x5e, 0x86, 0x90,
0x17, 0x4f, 0x4d, 0x78, 0x2f, 0x58, 0xb8, 0x67, 0x91, 0x89, 0xe3, 0x61, 0x1, 0x7d
]),
},
RTCPTestCase{
ssrc: 0x11111111,
index: 0x7fffffff, // Will be wrapped to 0
encrypted: Bytes::from_static(&[
0x80, 0xc8, 0x00, 0x06, 0x11, 0x11, 0x11, 0x11, 0x17, 0x8c, 0x15, 0xf1, 0x4b, 0x11,
0xda, 0xf5, 0x74, 0x53, 0x86, 0x2b, 0xc9, 0x07, 0x29, 0x40, 0xbf, 0x22, 0xf6, 0x46,
0x11, 0xa4, 0xc1, 0x3a, 0xff, 0x5a, 0xbd, 0xd0, 0xf8, 0x8b, 0x38, 0xe4, 0x95, 0x38,
0x5d, 0xcf, 0x1b, 0xf5, 0x27, 0x77, 0xfb, 0xdb, 0x3f, 0x10, 0x68, 0x99, 0xd8, 0xad,
0x80, 0x00, 0x00, 0x00, 0x7d, 0x51, 0xf8, 0x0e, 0x56, 0x40, 0x72, 0x7b, 0x9e, 0x02
]),
decrypted: Bytes::from_static(&[
0x80, 0xc8, 0x0, 0x6, 0x11, 0x11, 0x11, 0x11, 0xda, 0xb5, 0xe0, 0x56, 0x9a, 0x4a,
0x74, 0xed, 0x8a, 0x54, 0xc, 0xcf, 0xd5, 0x9, 0xb1, 0x40, 0x1, 0x42, 0xc3, 0x9a,
0x76, 0x0, 0xa9, 0xd4, 0xf7, 0x29, 0x9e, 0x51, 0xfb, 0x3c, 0xc1, 0x74, 0x72, 0xf9,
0x52, 0xb1, 0x92, 0x31, 0xca, 0x22, 0xab, 0x3e, 0xc5, 0x5f, 0x83, 0x34, 0xf0, 0x28
]),
},
];
}
#[test]
fn test_rtcp_lifecycle() -> Result<()> {
let mut encrypt_context = Context::new(
&RTCP_TEST_MASTER_KEY,
&RTCP_TEST_MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)?;
let mut decrypt_context = Context::new(
&RTCP_TEST_MASTER_KEY,
&RTCP_TEST_MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)?;
for test_case in &*RTCP_TEST_CASES {
let decrypt_result = decrypt_context.decrypt_rtcp(&test_case.encrypted)?;
assert_eq!(
decrypt_result, test_case.decrypted,
"RTCP failed to decrypt"
);
encrypt_context.set_index(test_case.ssrc, test_case.index);
let encrypt_result = encrypt_context.encrypt_rtcp(&test_case.decrypted)?;
assert_eq!(
encrypt_result, test_case.encrypted,
"RTCP failed to encrypt"
);
}
Ok(())
}
#[test]
fn test_rtcp_invalid_auth_tag() -> Result<()> {
let auth_tag_len = ProtectionProfile::Aes128CmHmacSha1_80.auth_tag_len();
let mut decrypt_context = Context::new(
&RTCP_TEST_MASTER_KEY,
&RTCP_TEST_MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)?;
let decrypt_result = decrypt_context.decrypt_rtcp(&RTCP_TEST_CASES[0].encrypted)?;
assert_eq!(
decrypt_result, RTCP_TEST_CASES[0].decrypted,
"RTCP failed to decrypt"
);
// Zero out auth tag
let mut rtcp_packet = BytesMut::new();
rtcp_packet.extend_from_slice(&RTCP_TEST_CASES[0].encrypted);
let rtcp_packet_len = rtcp_packet.len();
rtcp_packet[rtcp_packet_len - auth_tag_len..].copy_from_slice(&vec![0; auth_tag_len]);
let rtcp_packet = rtcp_packet.freeze();
let decrypt_result = decrypt_context.decrypt_rtcp(&rtcp_packet);
assert!(
decrypt_result.is_err(),
"Was able to decrypt RTCP packet with invalid Auth Tag"
);
Ok(())
}
#[test]
fn test_rtcp_replay_detector_separation() -> Result<()> {
let mut decrypt_context = Context::new(
&RTCP_TEST_MASTER_KEY,
&RTCP_TEST_MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
Some(srtcp_replay_protection(10)),
)?;
let rtcp_packet1 = RTCP_TEST_CASES[0].encrypted.clone();
let decrypt_result1 = decrypt_context.decrypt_rtcp(&rtcp_packet1)?;
assert_eq!(
decrypt_result1, RTCP_TEST_CASES[0].decrypted,
"RTCP failed to decrypt"
);
let rtcp_packet2 = RTCP_TEST_CASES[1].encrypted.clone();
let decrypt_result2 = decrypt_context.decrypt_rtcp(&rtcp_packet2)?;
assert_eq!(
decrypt_result2, RTCP_TEST_CASES[1].decrypted,
"RTCP failed to decrypt"
);
let result = decrypt_context.decrypt_rtcp(&rtcp_packet1);
assert!(
result.is_err(),
"Was able to decrypt duplicated RTCP packet"
);
let result = decrypt_context.decrypt_rtcp(&rtcp_packet2);
assert!(
result.is_err(),
"Was able to decrypt duplicated RTCP packet"
);
Ok(())
}
fn get_rtcp_index(encrypted: &Bytes, auth_tag_len: usize) -> u32 {
let tail_offset = encrypted.len() - (auth_tag_len + SRTCP_INDEX_SIZE);
let reader = &mut encrypted.slice(tail_offset..tail_offset + SRTCP_INDEX_SIZE);
//^(1 << 31)
reader.get_u32() & 0x7FFFFFFF
}
#[test]
fn test_encrypt_rtcp_separation() -> Result<()> {
let mut encrypt_context = Context::new(
&RTCP_TEST_MASTER_KEY,
&RTCP_TEST_MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)?;
let auth_tag_len = ProtectionProfile::Aes128CmHmacSha1_80.auth_tag_len();
let mut decrypt_context = Context::new(
&RTCP_TEST_MASTER_KEY,
&RTCP_TEST_MASTER_SALT,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
Some(srtcp_replay_protection(10)),
)?;
let inputs = vec![
RTCP_TEST_CASES[0].decrypted.clone(),
RTCP_TEST_CASES[1].decrypted.clone(),
RTCP_TEST_CASES[0].decrypted.clone(),
RTCP_TEST_CASES[1].decrypted.clone(),
];
let mut encrypted_rctps = vec![];
for input in &inputs {
let encrypted = encrypt_context.encrypt_rtcp(input)?;
encrypted_rctps.push(encrypted);
}
for (i, expected_index) in [1, 1, 2, 2].iter().enumerate() {
assert_eq!(
*expected_index,
get_rtcp_index(&encrypted_rctps[i], auth_tag_len),
"RTCP index does not match"
);
}
for (i, output) in encrypted_rctps.iter().enumerate() {
let decrypted = decrypt_context.decrypt_rtcp(output)?;
assert_eq!(inputs[i], decrypted);
}
Ok(())
}
+70
View File
@@ -0,0 +1,70 @@
use bytes::Bytes;
use util::marshal::*;
use super::*;
use crate::error::Result;
impl Context {
pub fn decrypt_rtp_with_header(
&mut self,
encrypted: &[u8],
header: &rtp::header::Header,
) -> Result<Bytes> {
let roc = {
let state = self.get_srtp_ssrc_state(header.ssrc);
if let Some(replay_detector) = &mut state.replay_detector {
if !replay_detector.check(header.sequence_number as u64) {
return Err(Error::SrtpSsrcDuplicated(
header.ssrc,
header.sequence_number,
));
}
}
state.next_rollover_count(header.sequence_number)
};
let dst = self.cipher.decrypt_rtp(encrypted, header, roc)?;
{
let state = self.get_srtp_ssrc_state(header.ssrc);
if let Some(replay_detector) = &mut state.replay_detector {
replay_detector.accept();
}
state.update_rollover_count(header.sequence_number);
}
Ok(dst)
}
/// DecryptRTP decrypts a RTP packet with an encrypted payload
pub fn decrypt_rtp(&mut self, encrypted: &[u8]) -> Result<Bytes> {
let mut buf = encrypted;
let header = rtp::header::Header::unmarshal(&mut buf)?;
self.decrypt_rtp_with_header(encrypted, &header)
}
pub fn encrypt_rtp_with_header(
&mut self,
payload: &[u8],
header: &rtp::header::Header,
) -> Result<Bytes> {
let roc = self
.get_srtp_ssrc_state(header.ssrc)
.next_rollover_count(header.sequence_number);
let dst = self.cipher.encrypt_rtp(payload, header, roc)?;
self.get_srtp_ssrc_state(header.ssrc)
.update_rollover_count(header.sequence_number);
Ok(dst)
}
/// EncryptRTP marshals and encrypts an RTP packet, writing to the dst buffer provided.
/// If the dst buffer does not have the capacity to hold `len(plaintext) + 10` bytes, a new one will be allocated and returned.
pub fn encrypt_rtp(&mut self, plaintext: &[u8]) -> Result<Bytes> {
let mut buf = plaintext;
let header = rtp::header::Header::unmarshal(&mut buf)?;
self.encrypt_rtp_with_header(plaintext, &header)
}
}
+188
View File
@@ -0,0 +1,188 @@
use bytes::Bytes;
use lazy_static::lazy_static;
use util::marshal::*;
use super::*;
struct RTPTestCase {
sequence_number: u16,
encrypted: Bytes,
}
lazy_static! {
static ref RTP_TEST_CASE_DECRYPTED: Bytes = Bytes::from_static(&[0x00, 0x01, 0x02, 0x03, 0x04, 0x05]);
static ref RTP_TEST_CASES: Vec<RTPTestCase> = vec![
RTPTestCase {
sequence_number: 5000,
encrypted: Bytes::from_static(&[
0x6d, 0xd3, 0x7e, 0xd5, 0x99, 0xb7, 0x2d, 0x28, 0xb1, 0xf3, 0xa1, 0xf0, 0xc, 0xfb,
0xfd, 0x8
]),
},
RTPTestCase {
sequence_number: 5001,
encrypted: Bytes::from_static(&[
0xda, 0x47, 0xb, 0x2a, 0x74, 0x53, 0x65, 0xbd, 0x2f, 0xeb, 0xdc, 0x4b, 0x6d, 0x23,
0xf3, 0xde
]),
},
RTPTestCase {
sequence_number: 5002,
encrypted: Bytes::from_static(&[
0x6e, 0xa7, 0x69, 0x8d, 0x24, 0x6d, 0xdc, 0xbf, 0xec, 0x2, 0x1c, 0xd1, 0x60, 0x76,
0xc1, 0x0e
]),
},
RTPTestCase {
sequence_number: 5003,
encrypted: Bytes::from_static(&[
0x24, 0x7e, 0x96, 0xc8, 0x7d, 0x33, 0xa2, 0x92, 0x8d, 0x13, 0x8d, 0xe0, 0x76, 0x9f,
0x08, 0xdc
]),
},
RTPTestCase {
sequence_number: 5004,
encrypted: Bytes::from_static(&[
0x75, 0x43, 0x28, 0xe4, 0x3a, 0x77, 0x59, 0x9b, 0x2e, 0xdf, 0x7b, 0x12, 0x68, 0x0b,
0x57, 0x49
]),
},
RTPTestCase{
sequence_number: 65535, // upper boundary
encrypted: Bytes::from_static(&[
0xaf, 0xf7, 0xc2, 0x70, 0x37, 0x20, 0x83, 0x9c, 0x2c, 0x63, 0x85, 0x15, 0x0e, 0x44,
0xca, 0x36
]),
},
];
}
fn build_test_context() -> Result<Context> {
let master_key = Bytes::from_static(&[
0x0d, 0xcd, 0x21, 0x3e, 0x4c, 0xbc, 0xf2, 0x8f, 0x01, 0x7f, 0x69, 0x94, 0x40, 0x1e, 0x28,
0x89,
]);
let master_salt = Bytes::from_static(&[
0x62, 0x77, 0x60, 0x38, 0xc0, 0x6d, 0xc9, 0x41, 0x9f, 0x6d, 0xd9, 0x43, 0x3e, 0x7c,
]);
Context::new(
&master_key,
&master_salt,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)
}
#[test]
fn test_rtp_invalid_auth() -> Result<()> {
let master_key = Bytes::from_static(&[
0x0d, 0xcd, 0x21, 0x3e, 0x4c, 0xbc, 0xf2, 0x8f, 0x01, 0x7f, 0x69, 0x94, 0x40, 0x1e, 0x28,
0x89,
]);
let invalid_salt = Bytes::from_static(&[
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
]);
let mut encrypt_context = build_test_context()?;
let mut invalid_context = Context::new(
&master_key,
&invalid_salt,
ProtectionProfile::Aes128CmHmacSha1_80,
None,
None,
)?;
for test_case in &*RTP_TEST_CASES {
let pkt = rtp::packet::Packet {
header: rtp::header::Header {
sequence_number: test_case.sequence_number,
..Default::default()
},
payload: RTP_TEST_CASE_DECRYPTED.clone(),
};
let pkt_raw = pkt.marshal()?;
let out = encrypt_context.encrypt_rtp(&pkt_raw)?;
let result = invalid_context.decrypt_rtp(&out);
assert!(
result.is_err(),
"Managed to decrypt with incorrect salt for packet with SeqNum: {}",
test_case.sequence_number
);
}
Ok(())
}
#[test]
fn test_rtp_lifecycle() -> Result<()> {
let mut encrypt_context = build_test_context()?;
let mut decrypt_context = build_test_context()?;
let auth_tag_len = ProtectionProfile::Aes128CmHmacSha1_80.auth_tag_len();
for test_case in RTP_TEST_CASES.iter() {
// Keep consecutive indices within RFC 3711's half-sequence-space window.
// The known-answer vector at 65535 still uses ROC 0, as before.
if test_case.sequence_number == 65535 {
for sequence_number in [32000, 64000] {
let bridge = rtp::packet::Packet {
header: rtp::header::Header {
sequence_number,
..Default::default()
},
payload: RTP_TEST_CASE_DECRYPTED.clone(),
}
.marshal()?;
let encrypted = encrypt_context.encrypt_rtp(&bridge)?;
assert_eq!(decrypt_context.decrypt_rtp(&encrypted)?, bridge);
}
}
let decrypted_pkt = rtp::packet::Packet {
header: rtp::header::Header {
sequence_number: test_case.sequence_number,
..Default::default()
},
payload: RTP_TEST_CASE_DECRYPTED.clone(),
};
let decrypted_raw = decrypted_pkt.marshal()?;
let encrypted_pkt = rtp::packet::Packet {
header: rtp::header::Header {
sequence_number: test_case.sequence_number,
..Default::default()
},
payload: test_case.encrypted.clone(),
};
let encrypted_raw = encrypted_pkt.marshal()?;
let actual_encrypted = encrypt_context.encrypt_rtp(&decrypted_raw)?;
assert_eq!(
actual_encrypted, encrypted_raw,
"RTP packet with SeqNum invalid encryption: {}",
test_case.sequence_number
);
let actual_decrypted = decrypt_context.decrypt_rtp(&encrypted_raw)?;
assert_ne!(
encrypted_raw[..encrypted_raw.len() - auth_tag_len].to_vec(),
actual_decrypted,
"DecryptRTP improperly encrypted in place"
);
assert_eq!(
actual_decrypted, decrypted_raw,
"RTP packet with SeqNum invalid decryption: {}",
test_case.sequence_number,
)
}
Ok(())
}
//TODO: BenchmarkEncryptRTP
//TODO: BenchmarkEncryptRTPInPlace
//TODO: BenchmarkDecryptRTP
+120
View File
@@ -0,0 +1,120 @@
use std::io;
use thiserror::Error;
use tokio::sync::mpsc::error::SendError as MpscSendError;
pub type Result<T> = std::result::Result<T, Error>;
#[derive(Error, Debug, PartialEq)]
#[non_exhaustive]
pub enum Error {
#[error("duplicated packet")]
ErrDuplicated,
#[error("SRTP master key is not long enough")]
ErrShortSrtpMasterKey,
#[error("SRTP master salt is not long enough")]
ErrShortSrtpMasterSalt,
#[error("no such SRTP Profile")]
ErrNoSuchSrtpProfile,
#[error("indexOverKdr > 0 is not supported yet")]
ErrNonZeroKdrNotSupported,
#[error("exporter called with wrong label")]
ErrExporterWrongLabel,
#[error("no config provided")]
ErrNoConfig,
#[error("no conn provided")]
ErrNoConn,
#[error("failed to verify auth tag")]
ErrFailedToVerifyAuthTag,
#[error("packet is too short to be rtcp packet")]
ErrTooShortRtcp,
#[error("payload differs")]
ErrPayloadDiffers,
#[error("started channel used incorrectly, should only be closed")]
ErrStartedChannelUsedIncorrectly,
#[error("stream has not been inited, unable to close")]
ErrStreamNotInited,
#[error("stream is already closed")]
ErrStreamAlreadyClosed,
#[error("stream is already inited")]
ErrStreamAlreadyInited,
#[error("failed to cast child")]
ErrFailedTypeAssertion,
#[error("index_over_kdr > 0 is not supported yet")]
UnsupportedIndexOverKdr,
#[error("SRTP Master Key must be len {0}, got {1}")]
SrtpMasterKeyLength(usize, usize),
#[error("SRTP Salt must be len {0}, got {1}")]
SrtpSaltLength(usize, usize),
#[error("SyntaxError: {0}")]
ExtMapParse(String),
#[error("srtp ssrc={0} index={1}: duplicated")]
SrtpSsrcDuplicated(u32, u16),
#[error("srtcp ssrc={0} index={1}: duplicated")]
SrtcpSsrcDuplicated(u32, usize),
#[error("ssrc {0} not exist in srtcp_ssrc_state")]
SsrcMissingFromSrtcp(u32),
#[error("Stream with ssrc {0} exists")]
StreamWithSsrcExists(u32),
#[error("Session RTP/RTCP type must be same as input buffer")]
SessionRtpRtcpTypeMismatch,
#[error("Session EOF")]
SessionEof,
#[error("too short SRTP packet: only {0} bytes, expected > {1} bytes")]
SrtpTooSmall(usize, usize),
#[error("too short SRTCP packet: only {0} bytes, expected > {1} bytes")]
SrtcpTooSmall(usize, usize),
#[error("failed to verify rtp auth tag")]
RtpFailedToVerifyAuthTag,
#[error("too short auth tag: only {0} bytes, expected > {1} bytes")]
RtcpInvalidLengthAuthTag(usize, usize),
#[error("failed to verify rtcp auth tag")]
RtcpFailedToVerifyAuthTag,
#[error("SessionSRTP has been closed")]
SessionSrtpAlreadyClosed,
#[error("this stream is not a RTPStream")]
InvalidRtpStream,
#[error("this stream is not a RTCPStream")]
InvalidRtcpStream,
#[error("{0}")]
Io(#[source] IoError),
#[error("{0}")]
KeyingMaterial(#[from] util::KeyingMaterialExporterError),
#[error("mpsc send: {0}")]
MpscSend(String),
#[error("{0}")]
Util(#[from] util::Error),
#[error("{0}")]
Rtcp(#[from] rtcp::Error),
#[error("aes gcm: {0}")]
AesGcm(#[from] aes_gcm::Error),
#[error("{0}")]
Other(String),
}
#[derive(Debug, Error)]
#[error("io error: {0}")]
pub struct IoError(#[from] pub io::Error);
// Workaround for wanting PartialEq for io::Error.
impl PartialEq for IoError {
fn eq(&self, other: &Self) -> bool {
self.0.kind() == other.0.kind()
}
}
impl From<io::Error> for Error {
fn from(e: io::Error) -> Self {
Error::Io(IoError(e))
}
}
// Because Tokio SendError is parameterized, we sadly lose the backtrace.
impl<T> From<MpscSendError<T>> for Error {
fn from(e: MpscSendError<T>) -> Self {
Error::MpscSend(e.to_string())
}
}
+173
View File
@@ -0,0 +1,173 @@
use aes::cipher::generic_array::GenericArray;
use aes::cipher::BlockEncrypt;
use aes::Aes128;
use aes_gcm::KeyInit;
use crate::error::{Error, Result};
pub const LABEL_SRTP_ENCRYPTION: u8 = 0x00;
pub const LABEL_SRTP_AUTHENTICATION_TAG: u8 = 0x01;
pub const LABEL_SRTP_SALT: u8 = 0x02;
pub const LABEL_SRTCP_ENCRYPTION: u8 = 0x03;
pub const LABEL_SRTCP_AUTHENTICATION_TAG: u8 = 0x04;
pub const LABEL_SRTCP_SALT: u8 = 0x05;
pub(crate) const SRTCP_INDEX_SIZE: usize = 4;
pub(crate) fn aes_cm_key_derivation(
label: u8,
master_key: &[u8],
master_salt: &[u8],
index_over_kdr: usize,
out_len: usize,
) -> Result<Vec<u8>> {
if index_over_kdr != 0 {
// 24-bit "index DIV kdr" must be xored to prf input.
return Err(Error::UnsupportedIndexOverKdr);
}
// https://tools.ietf.org/html/rfc3711#appendix-B.3
// The input block for AES-CM is generated by exclusive-oring the master salt with the
// concatenation of the encryption key label 0x00 with (index DIV kdr),
// - index is 'rollover count' and DIV is 'divided by'
let n_master_key = master_key.len();
let n_master_salt = master_salt.len();
let mut prf_in = vec![0u8; n_master_key];
prf_in[..n_master_salt].copy_from_slice(master_salt);
prf_in[7] ^= label;
//The resulting value is then AES encrypted using the master key to get the cipher key.
let key = GenericArray::from_slice(master_key);
let block = Aes128::new(key);
let mut out = vec![0u8; ((out_len + n_master_key) / n_master_key) * n_master_key];
for (i, n) in (0..out_len).step_by(n_master_key).enumerate() {
//BigEndian.PutUint16(prfIn[nMasterKey-2:], i)
prf_in[n_master_key - 2] = ((i >> 8) & 0xFF) as u8;
prf_in[n_master_key - 1] = (i & 0xFF) as u8;
out[n..n + n_master_key].copy_from_slice(&prf_in);
let out_key = GenericArray::from_mut_slice(&mut out[n..n + n_master_key]);
block.encrypt_block(out_key);
}
Ok(out[..out_len].to_vec())
}
/// Generate IV https://tools.ietf.org/html/rfc3711#section-4.1.1
/// where the 128-bit integer value IV SHALL be defined by the SSRC, the
/// SRTP packet index i, and the SRTP session salting key k_s, as below.
/// ROC = a 32-bit unsigned rollover counter (roc), which records how many
/// times the 16-bit RTP sequence number has been reset to zero after
/// passing through 65,535
/// ```nobuild
/// i = 2^16 * roc + SEQ
/// IV = (salt*2 ^ 16) | (ssrc*2 ^ 64) | (i*2 ^ 16)
/// ```
pub(crate) fn generate_counter(
sequence_number: u16,
rollover_counter: u32,
ssrc: u32,
session_salt: &[u8],
) -> [u8; 16] {
assert!(session_salt.len() <= 16);
let mut counter = [0; 16];
let ssrc_be = ssrc.to_be_bytes();
let rollover_be = rollover_counter.to_be_bytes();
let seq_be = ((sequence_number as u32) << 16).to_be_bytes();
counter[4..8].copy_from_slice(&ssrc_be);
counter[8..12].copy_from_slice(&rollover_be);
counter[12..16].copy_from_slice(&seq_be);
for i in 0..session_salt.len() {
counter[i] ^= session_salt[i];
}
counter
}
#[cfg(test)]
mod test {
use super::*;
use crate::protection_profile::*;
#[test]
fn test_valid_session_keys() -> Result<()> {
// Key Derivation Test Vectors from https://tools.ietf.org/html/rfc3711#appendix-B.3
let master_key = vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
];
let master_salt = vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
];
let expected_session_key = vec![
0xC6, 0x1E, 0x7A, 0x93, 0x74, 0x4F, 0x39, 0xEE, 0x10, 0x73, 0x4A, 0xFE, 0x3F, 0xF7,
0xA0, 0x87,
];
let expected_session_salt = vec![
0x30, 0xCB, 0xBC, 0x08, 0x86, 0x3D, 0x8C, 0x85, 0xD4, 0x9D, 0xB3, 0x4A, 0x9A, 0xE1,
];
let expected_session_auth_tag = vec![
0xCE, 0xBE, 0x32, 0x1F, 0x6F, 0xF7, 0x71, 0x6B, 0x6F, 0xD4, 0xAB, 0x49, 0xAF, 0x25,
0x6A, 0x15, 0x6D, 0x38, 0xBA, 0xA4,
];
let session_key = aes_cm_key_derivation(
LABEL_SRTP_ENCRYPTION,
&master_key,
&master_salt,
0,
master_key.len(),
)?;
assert_eq!(
session_key, expected_session_key,
"Session Key:\n{session_key:?} \ndoes not match expected:\n{expected_session_key:?}\nMaster Key:\n{master_key:?}\nMaster Salt:\n{master_salt:?}\n",
);
let session_salt = aes_cm_key_derivation(
LABEL_SRTP_SALT,
&master_key,
&master_salt,
0,
master_salt.len(),
)?;
assert_eq!(
session_salt, expected_session_salt,
"Session Salt {session_salt:?} does not match expected {expected_session_salt:?}"
);
let auth_key_len = ProtectionProfile::Aes128CmHmacSha1_80.auth_key_len();
let session_auth_tag = aes_cm_key_derivation(
LABEL_SRTP_AUTHENTICATION_TAG,
&master_key,
&master_salt,
0,
auth_key_len,
)?;
assert_eq!(
session_auth_tag, expected_session_auth_tag,
"Session Auth Tag {session_auth_tag:?} does not match expected {expected_session_auth_tag:?}",
);
Ok(())
}
// This test asserts that calling aesCmKeyDerivation with a non-zero indexOverKdr fails
// Currently this isn't supported, but the API makes sure we can add this in the future
#[test]
fn test_index_over_kdr() -> Result<()> {
let result = aes_cm_key_derivation(LABEL_SRTP_AUTHENTICATION_TAG, &[], &[], 1, 0);
assert!(result.is_err());
Ok(())
}
}
+14
View File
@@ -0,0 +1,14 @@
#![warn(rust_2018_idioms)]
#![allow(dead_code)]
mod cipher;
pub mod config;
pub mod context;
mod error;
mod key_derivation;
pub mod option;
pub mod protection_profile;
pub mod session;
pub mod stream;
pub use error::Error;
+36
View File
@@ -0,0 +1,36 @@
use util::replay_detector::*;
pub type ContextOption = Box<dyn (Fn() -> Box<dyn ReplayDetector + Send + 'static>) + Send + Sync>;
pub(crate) const MAX_SEQUENCE_NUMBER: u16 = 65535;
pub(crate) const MAX_SRTCP_INDEX: usize = 0x7FFFFFFF;
/// srtp_replay_protection sets SRTP replay protection window size.
pub fn srtp_replay_protection(window_size: usize) -> ContextOption {
Box::new(move || -> Box<dyn ReplayDetector + Send> {
Box::new(WrappedSlidingWindowDetector::new(
window_size,
MAX_SEQUENCE_NUMBER as u64,
))
})
}
/// Sets SRTCP replay protection window size.
pub fn srtcp_replay_protection(window_size: usize) -> ContextOption {
Box::new(move || -> Box<dyn ReplayDetector + Send> {
Box::new(WrappedSlidingWindowDetector::new(
window_size,
MAX_SRTCP_INDEX as u64,
))
})
}
/// srtp_no_replay_protection disables SRTP replay protection.
pub fn srtp_no_replay_protection() -> ContextOption {
Box::new(|| -> Box<dyn ReplayDetector + Send> { Box::<NoOpReplayDetector>::default() })
}
/// srtcp_no_replay_protection disables SRTCP replay protection.
pub fn srtcp_no_replay_protection() -> ContextOption {
Box::new(|| -> Box<dyn ReplayDetector + Send> { Box::<NoOpReplayDetector>::default() })
}
+37
View File
@@ -0,0 +1,37 @@
/// ProtectionProfile specifies Cipher and AuthTag details, similar to TLS cipher suite
#[derive(Default, Debug, Clone, Copy)]
#[repr(u8)]
pub enum ProtectionProfile {
#[default]
Aes128CmHmacSha1_80 = 0x0001,
AeadAes128Gcm = 0x0007,
}
impl ProtectionProfile {
pub(crate) fn key_len(&self) -> usize {
match *self {
ProtectionProfile::Aes128CmHmacSha1_80 | ProtectionProfile::AeadAes128Gcm => 16,
}
}
pub(crate) fn salt_len(&self) -> usize {
match *self {
ProtectionProfile::Aes128CmHmacSha1_80 => 14,
ProtectionProfile::AeadAes128Gcm => 12,
}
}
pub(crate) fn auth_tag_len(&self) -> usize {
match *self {
ProtectionProfile::Aes128CmHmacSha1_80 => 10, //CIPHER_AES_CM_HMAC_SHA1AUTH_TAG_LEN,
ProtectionProfile::AeadAes128Gcm => 16, //CIPHER_AEAD_AES_GCM_AUTH_TAG_LEN,
}
}
pub(crate) fn auth_key_len(&self) -> usize {
match *self {
ProtectionProfile::Aes128CmHmacSha1_80 => 20,
ProtectionProfile::AeadAes128Gcm => 0,
}
}
}
+271
View File
@@ -0,0 +1,271 @@
#[cfg(test)]
mod session_rtcp_test;
#[cfg(test)]
mod session_rtp_test;
use std::collections::{HashMap, HashSet};
use std::marker::{Send, Sync};
use std::sync::Arc;
use bytes::Bytes;
use tokio::sync::{mpsc, Mutex};
use util::conn::Conn;
use util::marshal::*;
use crate::config::*;
use crate::context::*;
use crate::error::{Error, Result};
use crate::option::*;
use crate::stream::*;
const DEFAULT_SESSION_SRTP_REPLAY_PROTECTION_WINDOW: usize = 64;
const DEFAULT_SESSION_SRTCP_REPLAY_PROTECTION_WINDOW: usize = 64;
/// Session implements io.ReadWriteCloser and provides a bi-directional SRTP session
/// SRTP itself does not have a design like this, but it is common in most applications
/// for local/remote to each have their own keying material. This provides those patterns
/// instead of making everyone re-implement
pub struct Session {
local_context: Arc<Mutex<Context>>,
streams_map: Arc<Mutex<HashMap<u32, Arc<Stream>>>>,
new_stream_rx: Arc<Mutex<mpsc::Receiver<Arc<Stream>>>>,
close_stream_tx: mpsc::Sender<u32>,
close_session_tx: mpsc::Sender<()>,
pub(crate) udp_tx: Arc<dyn Conn + Send + Sync>,
is_rtp: bool,
}
impl Session {
pub async fn new(
conn: Arc<dyn Conn + Send + Sync>,
config: Config,
is_rtp: bool,
) -> Result<Self> {
let local_context = Context::new(
&config.keys.local_master_key,
&config.keys.local_master_salt,
config.profile,
config.local_rtp_options,
config.local_rtcp_options,
)?;
let mut remote_context = Context::new(
&config.keys.remote_master_key,
&config.keys.remote_master_salt,
config.profile,
if config.remote_rtp_options.is_none() {
Some(srtp_replay_protection(
DEFAULT_SESSION_SRTP_REPLAY_PROTECTION_WINDOW,
))
} else {
config.remote_rtp_options
},
if config.remote_rtcp_options.is_none() {
Some(srtcp_replay_protection(
DEFAULT_SESSION_SRTCP_REPLAY_PROTECTION_WINDOW,
))
} else {
config.remote_rtcp_options
},
)?;
let streams_map = Arc::new(Mutex::new(HashMap::new()));
let (mut new_stream_tx, new_stream_rx) = mpsc::channel(8);
let (close_stream_tx, mut close_stream_rx) = mpsc::channel(8);
let (close_session_tx, mut close_session_rx) = mpsc::channel(8);
let udp_tx = Arc::clone(&conn);
let udp_rx = Arc::clone(&conn);
let cloned_streams_map = Arc::clone(&streams_map);
let cloned_close_stream_tx = close_stream_tx.clone();
tokio::spawn(async move {
let mut buf = vec![0u8; 8192];
loop {
let incoming_stream = Session::incoming(
&udp_rx,
&mut buf,
&cloned_streams_map,
&cloned_close_stream_tx,
&mut new_stream_tx,
&mut remote_context,
is_rtp,
);
let close_stream = close_stream_rx.recv();
let close_session = close_session_rx.recv();
tokio::select! {
result = incoming_stream => match result{
Ok(()) => {},
Err(err) => log::info!("{}", err),
},
opt = close_stream => if let Some(ssrc) = opt {
Session::close_stream(&cloned_streams_map, ssrc).await
},
_ = close_session => break
}
}
});
Ok(Session {
local_context: Arc::new(Mutex::new(local_context)),
streams_map,
new_stream_rx: Arc::new(Mutex::new(new_stream_rx)),
close_stream_tx,
close_session_tx,
udp_tx,
is_rtp,
})
}
async fn close_stream(streams_map: &Arc<Mutex<HashMap<u32, Arc<Stream>>>>, ssrc: u32) {
let mut streams = streams_map.lock().await;
streams.remove(&ssrc);
}
async fn incoming(
udp_rx: &Arc<dyn Conn + Send + Sync>,
buf: &mut [u8],
streams_map: &Arc<Mutex<HashMap<u32, Arc<Stream>>>>,
close_stream_tx: &mpsc::Sender<u32>,
new_stream_tx: &mut mpsc::Sender<Arc<Stream>>,
remote_context: &mut Context,
is_rtp: bool,
) -> Result<()> {
let n = udp_rx.recv(buf).await?;
if n == 0 {
return Err(Error::SessionEof);
}
let decrypted = if is_rtp {
remote_context.decrypt_rtp(&buf[0..n])?
} else {
remote_context.decrypt_rtcp(&buf[0..n])?
};
let mut buf = &decrypted[..];
let ssrcs = if is_rtp {
vec![rtp::header::Header::unmarshal(&mut buf)?.ssrc]
} else {
let pkts = rtcp::packet::unmarshal(&mut buf)?;
destination_ssrc(&pkts)
};
for ssrc in ssrcs {
let (stream, is_new) =
Session::get_or_create_stream(streams_map, close_stream_tx.clone(), is_rtp, ssrc)
.await;
if is_new {
log::trace!(
"srtp session got new {} stream {}",
if is_rtp { "rtp" } else { "rtcp" },
ssrc
);
new_stream_tx.send(Arc::clone(&stream)).await?;
}
match stream.buffer.write(&decrypted).await {
Ok(_) => {}
Err(err) => {
// Silently drop data when the buffer is full.
if util::Error::ErrBufferFull != err {
return Err(err.into());
}
}
}
}
Ok(())
}
async fn get_or_create_stream(
streams_map: &Arc<Mutex<HashMap<u32, Arc<Stream>>>>,
close_stream_tx: mpsc::Sender<u32>,
is_rtp: bool,
ssrc: u32,
) -> (Arc<Stream>, bool) {
let mut streams = streams_map.lock().await;
if let Some(stream) = streams.get(&ssrc) {
(Arc::clone(stream), false)
} else {
let stream = Arc::new(Stream::new(ssrc, close_stream_tx, is_rtp));
streams.insert(ssrc, Arc::clone(&stream));
(stream, true)
}
}
/// open on the given SSRC to create a stream, it can be used
/// if you want a certain SSRC, but don't want to wait for Accept
pub async fn open(&self, ssrc: u32) -> Arc<Stream> {
let (stream, _) = Session::get_or_create_stream(
&self.streams_map,
self.close_stream_tx.clone(),
self.is_rtp,
ssrc,
)
.await;
stream
}
/// accept returns a stream to handle RTCP for a single SSRC
pub async fn accept(&self) -> Result<Arc<Stream>> {
let mut new_stream_rx = self.new_stream_rx.lock().await;
let result = new_stream_rx.recv().await;
if let Some(stream) = result {
Ok(stream)
} else {
Err(Error::SessionSrtpAlreadyClosed)
}
}
pub async fn close(&self) -> Result<()> {
self.close_session_tx.send(()).await?;
Ok(())
}
pub async fn write(&self, buf: &Bytes, is_rtp: bool) -> Result<usize> {
if self.is_rtp != is_rtp {
return Err(Error::SessionRtpRtcpTypeMismatch);
}
let encrypted = {
let mut local_context = self.local_context.lock().await;
if is_rtp {
local_context.encrypt_rtp(buf)?
} else {
local_context.encrypt_rtcp(buf)?
}
};
Ok(self.udp_tx.send(&encrypted).await?)
}
pub async fn write_rtp(&self, pkt: &rtp::packet::Packet) -> Result<usize> {
let raw = pkt.marshal()?;
self.write(&raw, true).await
}
pub async fn write_rtcp(
&self,
pkt: &(dyn rtcp::packet::Packet + Send + Sync),
) -> Result<usize> {
let raw = pkt.marshal()?;
self.write(&raw, false).await
}
}
/// create a list of Destination SSRCs
/// that's a superset of all Destinations in the slice.
fn destination_ssrc(pkts: &[Box<dyn rtcp::packet::Packet + Send + Sync>]) -> Vec<u32> {
let mut ssrc_set = HashSet::new();
for p in pkts {
for ssrc in p.destination_ssrc() {
ssrc_set.insert(ssrc);
}
}
ssrc_set.into_iter().collect()
}
+239
View File
@@ -0,0 +1,239 @@
use std::sync::Arc;
use bytes::{Bytes, BytesMut};
use rtcp::payload_feedbacks::*;
use tokio::sync::{mpsc, Mutex};
use util::conn::conn_pipe::*;
use super::*;
use crate::error::Result;
use crate::protection_profile::*;
async fn build_session_srtcp_pair() -> Result<(Session, Session)> {
let (ua, ub) = pipe();
let ca = Config {
profile: ProtectionProfile::Aes128CmHmacSha1_80,
keys: SessionKeys {
local_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
local_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
remote_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
remote_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
},
local_rtp_options: None,
remote_rtp_options: None,
local_rtcp_options: None,
remote_rtcp_options: None,
};
let cb = Config {
profile: ProtectionProfile::Aes128CmHmacSha1_80,
keys: SessionKeys {
local_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
local_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
remote_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
remote_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
},
local_rtp_options: None,
remote_rtp_options: None,
local_rtcp_options: None,
remote_rtcp_options: None,
};
let sa = Session::new(Arc::new(ua), ca, false).await?;
let sb = Session::new(Arc::new(ub), cb, false).await?;
Ok((sa, sb))
}
const TEST_SSRC: u32 = 5000;
#[tokio::test]
async fn test_session_srtcp_accept() -> Result<()> {
let (sa, sb) = build_session_srtcp_pair().await?;
let rtcp_packet = picture_loss_indication::PictureLossIndication {
media_ssrc: TEST_SSRC,
..Default::default()
};
let test_payload = rtcp_packet.marshal()?;
sa.write_rtcp(&rtcp_packet).await?;
let read_stream = sb.accept().await?;
let ssrc = read_stream.get_ssrc();
assert_eq!(
ssrc, TEST_SSRC,
"SSRC mismatch during accept exp({TEST_SSRC}) actual({ssrc})"
);
let mut read_buffer = BytesMut::with_capacity(test_payload.len());
read_buffer.resize(test_payload.len(), 0u8);
read_stream.read(&mut read_buffer).await?;
assert_eq!(
&test_payload[..],
&read_buffer[..],
"Sent buffer does not match the one received exp({:?}) actual({:?})",
&test_payload[..],
&read_buffer[..]
);
sa.close().await?;
sb.close().await?;
Ok(())
}
#[tokio::test]
async fn test_session_srtcp_listen() -> Result<()> {
let (sa, sb) = build_session_srtcp_pair().await?;
let rtcp_packet = picture_loss_indication::PictureLossIndication {
media_ssrc: TEST_SSRC,
..Default::default()
};
let test_payload = rtcp_packet.marshal()?;
let read_stream = sb.open(TEST_SSRC).await;
sa.write_rtcp(&rtcp_packet).await?;
let mut read_buffer = BytesMut::with_capacity(test_payload.len());
read_buffer.resize(test_payload.len(), 0u8);
read_stream.read(&mut read_buffer).await?;
assert_eq!(
&test_payload[..],
&read_buffer[..],
"Sent buffer does not match the one received exp({:?}) actual({:?})",
&test_payload[..],
&read_buffer[..]
);
sa.close().await?;
sb.close().await?;
Ok(())
}
fn encrypt_srtcp(
context: &mut Context,
pkt: &(dyn rtcp::packet::Packet + Send + Sync),
) -> Result<Bytes> {
let decrypted = pkt.marshal()?;
let encrypted = context.encrypt_rtcp(&decrypted)?;
Ok(encrypted)
}
const PLI_PACKET_SIZE: usize = 8;
async fn get_sender_ssrc(read_stream: &Arc<Stream>) -> Result<u32> {
let auth_tag_size = ProtectionProfile::Aes128CmHmacSha1_80.auth_tag_len();
let mut read_buffer = BytesMut::with_capacity(PLI_PACKET_SIZE + auth_tag_size);
read_buffer.resize(PLI_PACKET_SIZE + auth_tag_size, 0u8);
let pkts = read_stream.read_rtcp(&mut read_buffer).await?;
let mut bytes = &pkts[0].marshal()?[..];
let pli = picture_loss_indication::PictureLossIndication::unmarshal(&mut bytes)?;
Ok(pli.sender_ssrc)
}
#[tokio::test]
async fn test_session_srtcp_replay_protection() -> Result<()> {
let (sa, sb) = build_session_srtcp_pair().await?;
let read_stream = sb.open(TEST_SSRC).await;
// Generate test packets
let mut packets = vec![];
let mut expected_ssrc = vec![];
{
let mut local_context = sa.local_context.lock().await;
for i in 0..0x10u32 {
expected_ssrc.push(i);
let packet = picture_loss_indication::PictureLossIndication {
media_ssrc: TEST_SSRC,
sender_ssrc: i,
};
let encrypted = encrypt_srtcp(&mut local_context, &packet)?;
packets.push(encrypted);
}
}
let (done_tx, mut done_rx) = mpsc::channel::<()>(1);
let received_ssrc = Arc::new(Mutex::new(vec![]));
let cloned_received_ssrc = Arc::clone(&received_ssrc);
let count = expected_ssrc.len();
tokio::spawn(async move {
let mut i = 0;
while i < count {
match get_sender_ssrc(&read_stream).await {
Ok(ssrc) => {
let mut r = cloned_received_ssrc.lock().await;
r.push(ssrc);
i += 1;
}
Err(_) => break,
}
}
drop(done_tx);
});
// Write with replay attack
for packet in &packets {
sa.udp_tx.send(packet).await?;
// Immediately replay
sa.udp_tx.send(packet).await?;
}
for packet in &packets {
// Delayed replay
sa.udp_tx.send(packet).await?;
}
done_rx.recv().await;
sa.close().await?;
sb.close().await?;
{
let received_ssrc = received_ssrc.lock().await;
assert_eq!(&expected_ssrc[..], &received_ssrc[..]);
}
Ok(())
}
+308
View File
@@ -0,0 +1,308 @@
use std::collections::HashMap;
use std::sync::Arc;
use bytes::{Bytes, BytesMut};
use tokio::net::UdpSocket;
use tokio::sync::{mpsc, Mutex};
use super::*;
use crate::error::Result;
use crate::protection_profile::*;
async fn build_session_srtp_pair() -> Result<(Session, Session)> {
let ua = UdpSocket::bind("127.0.0.1:0").await?;
let ub = UdpSocket::bind("127.0.0.1:0").await?;
ua.connect(ub.local_addr()?).await?;
ub.connect(ua.local_addr()?).await?;
let ca = Config {
profile: ProtectionProfile::Aes128CmHmacSha1_80,
keys: SessionKeys {
local_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
local_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
remote_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
remote_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
},
local_rtp_options: None,
remote_rtp_options: None,
local_rtcp_options: None,
remote_rtcp_options: None,
};
let cb = Config {
profile: ProtectionProfile::Aes128CmHmacSha1_80,
keys: SessionKeys {
local_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
local_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
remote_master_key: vec![
0xE1, 0xF9, 0x7A, 0x0D, 0x3E, 0x01, 0x8B, 0xE0, 0xD6, 0x4F, 0xA3, 0x2C, 0x06, 0xDE,
0x41, 0x39,
],
remote_master_salt: vec![
0x0E, 0xC6, 0x75, 0xAD, 0x49, 0x8A, 0xFE, 0xEB, 0xB6, 0x96, 0x0B, 0x3A, 0xAB, 0xE6,
],
},
local_rtp_options: None,
remote_rtp_options: None,
local_rtcp_options: None,
remote_rtcp_options: None,
};
let sa = Session::new(Arc::new(ua), ca, true).await?;
let sb = Session::new(Arc::new(ub), cb, true).await?;
Ok((sa, sb))
}
const TEST_SSRC: u32 = 5000;
const RTP_HEADER_SIZE: usize = 12;
#[tokio::test]
async fn test_session_srtp_accept() -> Result<()> {
let test_payload = Bytes::from_static(&[0x00, 0x01, 0x03, 0x04]);
let mut read_buffer = BytesMut::with_capacity(RTP_HEADER_SIZE + test_payload.len());
read_buffer.resize(RTP_HEADER_SIZE + test_payload.len(), 0u8);
let (sa, sb) = build_session_srtp_pair().await?;
let packet = rtp::packet::Packet {
header: rtp::header::Header {
ssrc: TEST_SSRC,
..Default::default()
},
payload: test_payload.clone(),
};
sa.write_rtp(&packet).await?;
let read_stream = sb.accept().await?;
let ssrc = read_stream.get_ssrc();
assert_eq!(
ssrc, TEST_SSRC,
"SSRC mismatch during accept exp({TEST_SSRC}) actual({ssrc})"
);
read_stream.read(&mut read_buffer).await?;
assert_eq!(
&test_payload[..],
&read_buffer[RTP_HEADER_SIZE..],
"Sent buffer does not match the one received exp({:?}) actual({:?})",
&test_payload[..],
&read_buffer[RTP_HEADER_SIZE..]
);
sa.close().await?;
sb.close().await?;
Ok(())
}
#[tokio::test]
async fn test_session_srtp_listen() -> Result<()> {
let test_payload = Bytes::from_static(&[0x00, 0x01, 0x03, 0x04]);
let mut read_buffer = BytesMut::with_capacity(RTP_HEADER_SIZE + test_payload.len());
read_buffer.resize(RTP_HEADER_SIZE + test_payload.len(), 0u8);
let (sa, sb) = build_session_srtp_pair().await?;
let packet = rtp::packet::Packet {
header: rtp::header::Header {
ssrc: TEST_SSRC,
..Default::default()
},
payload: test_payload.clone(),
};
let read_stream = sb.open(TEST_SSRC).await;
sa.write_rtp(&packet).await?;
read_stream.read(&mut read_buffer).await?;
assert_eq!(
&test_payload[..],
&read_buffer[RTP_HEADER_SIZE..],
"Sent buffer does not match the one received exp({:?}) actual({:?})",
&test_payload[..],
&read_buffer[RTP_HEADER_SIZE..]
);
sa.close().await?;
sb.close().await?;
Ok(())
}
#[tokio::test]
async fn test_session_srtp_multi_ssrc() -> Result<()> {
let ssrcs = vec![5000, 5001, 5002];
let test_payload = Bytes::from_static(&[0x00, 0x01, 0x03, 0x04]);
let mut read_buffer = BytesMut::with_capacity(RTP_HEADER_SIZE + test_payload.len());
read_buffer.resize(RTP_HEADER_SIZE + test_payload.len(), 0u8);
let (sa, sb) = build_session_srtp_pair().await?;
let mut read_streams = HashMap::new();
for ssrc in &ssrcs {
let read_stream = sb.open(*ssrc).await;
read_streams.insert(*ssrc, read_stream);
}
for ssrc in &ssrcs {
let packet = rtp::packet::Packet {
header: rtp::header::Header {
ssrc: *ssrc,
..Default::default()
},
payload: test_payload.clone(),
};
sa.write_rtp(&packet).await?;
if let Some(read_stream) = read_streams.get_mut(ssrc) {
read_stream.read(&mut read_buffer).await?;
assert_eq!(
&test_payload[..],
&read_buffer[RTP_HEADER_SIZE..],
"Sent buffer does not match the one received exp({:?}) actual({:?})",
&test_payload[..],
&read_buffer[RTP_HEADER_SIZE..]
);
} else {
panic!("ssrc {} not found", *ssrc);
}
}
sa.close().await?;
sb.close().await?;
Ok(())
}
fn encrypt_srtp(context: &mut Context, pkt: &rtp::packet::Packet) -> Result<Bytes> {
let decrypted = pkt.marshal()?;
let encrypted = context.encrypt_rtp(&decrypted)?;
Ok(encrypted)
}
async fn payload_srtp(
read_stream: &Arc<Stream>,
header_size: usize,
expected_payload: &[u8],
) -> Result<u16> {
let mut read_buffer = BytesMut::with_capacity(header_size + expected_payload.len());
read_buffer.resize(header_size + expected_payload.len(), 0u8);
let pkt = read_stream.read_rtp(&mut read_buffer).await?;
assert_eq!(
expected_payload,
&pkt.payload[..],
"Sent buffer does not match the one received exp({:?}) actual({:?})",
expected_payload,
&pkt.payload[..]
);
Ok(pkt.header.sequence_number)
}
#[tokio::test]
async fn test_session_srtp_replay_protection() -> Result<()> {
let test_payload = Bytes::from_static(&[0x00, 0x01, 0x03, 0x04]);
let (sa, sb) = build_session_srtp_pair().await?;
let read_stream = sb.open(TEST_SSRC).await;
// Generate test packets
let mut packets = vec![];
let mut expected_sequence_number = vec![];
{
let mut local_context = sa.local_context.lock().await;
let mut i = 0xFFF0u16;
while i != 0x10 {
expected_sequence_number.push(i);
let packet = rtp::packet::Packet {
header: rtp::header::Header {
ssrc: TEST_SSRC,
sequence_number: i,
..Default::default()
},
payload: test_payload.clone(),
};
let encrypted = encrypt_srtp(&mut local_context, &packet)?;
packets.push(encrypted);
if i == 0xFFFF {
i = 0;
} else {
i += 1;
}
}
}
let (done_tx, mut done_rx) = mpsc::channel::<()>(1);
let received_sequence_number = Arc::new(Mutex::new(vec![]));
let cloned_received_sequence_number = Arc::clone(&received_sequence_number);
let count = expected_sequence_number.len();
tokio::spawn(async move {
let mut i = 0;
while i < count {
let seq = payload_srtp(&read_stream, RTP_HEADER_SIZE, &test_payload)
.await
.unwrap();
let mut r = cloned_received_sequence_number.lock().await;
r.push(seq);
i += 1;
}
drop(done_tx);
});
// Write with replay attack
for packet in &packets {
sa.udp_tx.send(packet).await?;
// Immediately replay
sa.udp_tx.send(packet).await?;
}
for packet in &packets {
// Delayed replay
sa.udp_tx.send(packet).await?;
}
done_rx.recv().await;
sa.close().await?;
sb.close().await?;
{
let received_sequence_number = received_sequence_number.lock().await;
assert_eq!(&received_sequence_number[..], &expected_sequence_number[..]);
}
Ok(())
}
+91
View File
@@ -0,0 +1,91 @@
use tokio::sync::mpsc;
use util::marshal::*;
use util::Buffer;
use crate::error::{Error, Result};
/// Limit the buffer size to 1MB
pub const SRTP_BUFFER_SIZE: usize = 1000 * 1000;
/// Limit the buffer size to 100KB
pub const SRTCP_BUFFER_SIZE: usize = 100 * 1000;
/// Stream handles decryption for a single RTP/RTCP SSRC
#[derive(Debug)]
pub struct Stream {
ssrc: u32,
tx: mpsc::Sender<u32>,
pub(crate) buffer: Buffer,
is_rtp: bool,
}
impl Stream {
/// Create a new stream
pub fn new(ssrc: u32, tx: mpsc::Sender<u32>, is_rtp: bool) -> Self {
Stream {
ssrc,
tx,
// Create a buffer with a 1MB limit
buffer: Buffer::new(
0,
if is_rtp {
SRTP_BUFFER_SIZE
} else {
SRTCP_BUFFER_SIZE
},
),
is_rtp,
}
}
/// GetSSRC returns the SSRC we are demuxing for
pub fn get_ssrc(&self) -> u32 {
self.ssrc
}
/// Check if RTP is a stream.
pub fn is_rtp_stream(&self) -> bool {
self.is_rtp
}
/// Read reads and decrypts full RTP packet from the nextConn
pub async fn read(&self, buf: &mut [u8]) -> Result<usize> {
Ok(self.buffer.read(buf, None).await?)
}
/// ReadRTP reads and decrypts full RTP packet and its header from the nextConn
pub async fn read_rtp(&self, buf: &mut [u8]) -> Result<rtp::packet::Packet> {
if !self.is_rtp {
return Err(Error::InvalidRtpStream);
}
let n = self.buffer.read(buf, None).await?;
let mut b = &buf[..n];
let pkt = rtp::packet::Packet::unmarshal(&mut b)?;
Ok(pkt)
}
/// read_rtcp reads and decrypts full RTP packet and its header from the nextConn
pub async fn read_rtcp(
&self,
buf: &mut [u8],
) -> Result<Vec<Box<dyn rtcp::packet::Packet + Send + Sync>>> {
if self.is_rtp {
return Err(Error::InvalidRtcpStream);
}
let n = self.buffer.read(buf, None).await?;
let mut b = &buf[..n];
let pkt = rtcp::packet::unmarshal(&mut b)?;
Ok(pkt)
}
/// Close removes the ReadStream from the session and cleans up any associated state
pub async fn close(&self) -> Result<()> {
self.buffer.close().await;
let _ = self.tx.send(self.ssrc).await;
Ok(())
}
}