Add support for v4l2loopback

It allows to send the video stream to /dev/videoN, so that it can be
captured (like a webcam) by any v4l2-capable tool.

PR #2232 <https://github.com/Genymobile/scrcpy/pull/2232>
PR #2233 <https://github.com/Genymobile/scrcpy/pull/2233>
PR #2268 <https://github.com/Genymobile/scrcpy/pull/2268>

Co-authored-by: Romain Vimont <rom@rom1v.com>

app/meson.build

@@ -33,6 +33,11 @@ else
     src += [ 'src/sys/unix/process.c' ]
 endif
 
+v4l2_support = host_machine.system() == 'linux'
+if v4l2_support
+    src += [ 'src/v4l2_sink.c' ]
+endif
+
 check_functions = [
     'strdup'
 ]
@@ -49,6 +54,10 @@ if not get_option('crossbuild_windows')
         dependency('sdl2'),
     ]
 
+    if v4l2_support
+        dependencies += dependency('libavdevice')
+    endif
+
 else
 
     # cross-compile mingw32 build (from Linux to Windows)
@@ -124,6 +133,9 @@ conf.set('SERVER_DEBUGGER', get_option('server_debugger'))
 # select the debugger method ('old' for Android < 9, 'new' for Android >= 9)
 conf.set('SERVER_DEBUGGER_METHOD_NEW', get_option('server_debugger_method') == 'new')
 
+# enable V4L2 support (linux only)
+conf.set('HAVE_V4L2', v4l2_support)
+
 configure_file(configuration: conf, output: 'config.h')
 
 src_dir = include_directories('src')

app/scrcpy.1

@@ -185,6 +185,12 @@ Enable "show touches" on start, restore the initial value on exit.
 
 It only shows physical touches (not clicks from scrcpy).
 
+.TP
+.BI "\-\-v4l2_sink " /dev/videoN
+Output to v4l2loopback device.
+
+It requires to lock the video orientation (see --lock-video-orientation).
+
 .TP
 .BI "\-V, \-\-verbosity " value
 Set the log level ("debug", "info", "warn" or "error").

app/src/cli.c

@@ -176,6 +176,13 @@ scrcpy_print_usage(const char *arg0) {
         "        on exit.\n"
         "        It only shows physical touches (not clicks from scrcpy).\n"
         "\n"
+#ifdef HAVE_V4L2
+        "    --v4l2_sink /dev/videoN\n"
+        "        Output to v4l2loopback device.\n"
+        "        It requires to lock the video orientation (see\n"
+        "        --lock-video-orientation).\n"
+        "\n"
+#endif
         "    -V, --verbosity value\n"
         "        Set the log level (debug, info, warn or error).\n"
 #ifndef NDEBUG
@@ -676,6 +683,7 @@ guess_record_format(const char *filename) {
 #define OPT_LEGACY_PASTE           1024
 #define OPT_ENCODER_NAME           1025
 #define OPT_POWER_OFF_ON_CLOSE     1026
+#define OPT_V4L2_SINK              1027
 
 bool
 scrcpy_parse_args(struct scrcpy_cli_args *args, int argc, char *argv[]) {
@@ -717,6 +725,9 @@ scrcpy_parse_args(struct scrcpy_cli_args *args, int argc, char *argv[]) {
         {"show-touches",           no_argument,       NULL, 't'},
         {"stay-awake",             no_argument,       NULL, 'w'},
         {"turn-screen-off",        no_argument,       NULL, 'S'},
+#ifdef HAVE_V4L2
+        {"v4l2_sink",              required_argument, NULL, OPT_V4L2_SINK},
+#endif
         {"verbosity",              required_argument, NULL, 'V'},
         {"version",                no_argument,       NULL, 'v'},
         {"window-title",           required_argument, NULL, OPT_WINDOW_TITLE},
@@ -901,16 +912,36 @@ scrcpy_parse_args(struct scrcpy_cli_args *args, int argc, char *argv[]) {
             case OPT_POWER_OFF_ON_CLOSE:
                 opts->power_off_on_close = true;
                 break;
+#ifdef HAVE_V4L2
+            case OPT_V4L2_SINK:
+                opts->v4l2_device = optarg;
+                break;
+#endif
             default:
                 // getopt prints the error message on stderr
                 return false;
         }
     }
 
+#ifdef HAVE_V4L2
+    if (!opts->display && !opts->record_filename && !opts->v4l2_device) {
+        LOGE("-N/--no-display requires either screen recording (-r/--record)"
+             " or sink to v4l2loopback device (--v4l2_sink)");
+        return false;
+    }
+
+    if (opts->v4l2_device && opts->lock_video_orientation
+                             == SC_LOCK_VIDEO_ORIENTATION_UNLOCKED) {
+        LOGI("Video orientation is locked for v4l2 sink. "
+             "See --lock-video-orientation.");
+        opts->lock_video_orientation = SC_LOCK_VIDEO_ORIENTATION_INITIAL;
+    }
+#else
     if (!opts->display && !opts->record_filename) {
         LOGE("-N/--no-display requires screen recording (-r/--record)");
         return false;
     }
+#endif
 
     int index = optind;
     if (index < argc) {

app/src/decoder.h

@@ -8,7 +8,7 @@
 #include <stdbool.h>
 #include <libavformat/avformat.h>
 
-#define DECODER_MAX_SINKS 1
+#define DECODER_MAX_SINKS 2
 
 struct decoder {
     struct sc_packet_sink packet_sink; // packet sink trait

app/src/main.c

@@ -6,6 +6,9 @@
 #include <stdbool.h>
 #include <unistd.h>
 #include <libavformat/avformat.h>
+#ifdef HAVE_V4L2
+# include <libavdevice/avdevice.h>
+#endif
 #define SDL_MAIN_HANDLED // avoid link error on Linux Windows Subsystem
 #include <SDL2/SDL.h>
 
@@ -28,6 +31,11 @@ print_version(void) {
     fprintf(stderr, " - libavutil %d.%d.%d\n", LIBAVUTIL_VERSION_MAJOR,
                                                LIBAVUTIL_VERSION_MINOR,
                                                LIBAVUTIL_VERSION_MICRO);
+#ifdef HAVE_V4L2
+    fprintf(stderr, " - libavdevice %d.%d.%d\n", LIBAVDEVICE_VERSION_MAJOR,
+                                                 LIBAVDEVICE_VERSION_MINOR,
+                                                 LIBAVDEVICE_VERSION_MICRO);
+#endif
 }
 
 static SDL_LogPriority
@@ -90,6 +98,12 @@ main(int argc, char *argv[]) {
     av_register_all();
 #endif
 
+#ifdef HAVE_V4L2
+    if (args.opts.v4l2_device) {
+        avdevice_register_all();
+    }
+#endif
+
     if (avformat_network_init()) {
         return 1;
     }

app/src/scrcpy.c

@@ -27,6 +27,9 @@
 #include "tiny_xpm.h"
 #include "util/log.h"
 #include "util/net.h"
+#ifdef HAVE_V4L2
+# include "v4l2_sink.h"
+#endif
 
 static struct server server;
 static struct screen screen;
@@ -34,6 +37,9 @@ static struct fps_counter fps_counter;
 static struct stream stream;
 static struct decoder decoder;
 static struct recorder recorder;
+#ifdef HAVE_V4L2
+static struct sc_v4l2_sink v4l2_sink;
+#endif
 static struct controller controller;
 static struct file_handler file_handler;
 
@@ -247,6 +253,9 @@ scrcpy(const struct scrcpy_options *options) {
     bool fps_counter_initialized = false;
     bool file_handler_initialized = false;
     bool recorder_initialized = false;
+#ifdef HAVE_V4L2
+    bool v4l2_sink_initialized = false;
+#endif
     bool stream_started = false;
     bool controller_initialized = false;
     bool controller_started = false;
@@ -295,7 +304,6 @@ scrcpy(const struct scrcpy_options *options) {
         goto end;
     }
 
-    struct decoder *dec = NULL;
     if (options->display) {
         if (!fps_counter_init(&fps_counter)) {
             goto end;
@@ -309,7 +317,14 @@ scrcpy(const struct scrcpy_options *options) {
             }
             file_handler_initialized = true;
         }
+    }
 
+    struct decoder *dec = NULL;
+    bool needs_decoder = options->display;
+#ifdef HAVE_V4L2
+    needs_decoder |= !!options->v4l2_device;
+#endif
+    if (needs_decoder) {
         decoder_init(&decoder);
         dec = &decoder;
     }
@@ -386,6 +401,18 @@ scrcpy(const struct scrcpy_options *options) {
         }
     }
 
+#ifdef HAVE_V4L2
+    if (options->v4l2_device) {
+        if (!sc_v4l2_sink_init(&v4l2_sink, options->v4l2_device, frame_size)) {
+            goto end;
+        }
+
+        decoder_add_sink(&decoder, &v4l2_sink.frame_sink);
+
+        v4l2_sink_initialized = true;
+    }
+#endif
+
     // now we consumed the header values, the socket receives the video stream
     // start the stream
     if (!stream_start(&stream)) {
@@ -426,6 +453,12 @@ end:
         stream_join(&stream);
     }
 
+#ifdef HAVE_V4L2
+    if (v4l2_sink_initialized) {
+        sc_v4l2_sink_destroy(&v4l2_sink);
+    }
+#endif
+
     // Destroy the screen only after the stream is guaranteed to be finished,
     // because otherwise the screen could receive new frames after destruction
     if (screen_initialized) {

app/src/scrcpy.h

@@ -62,6 +62,7 @@ struct scrcpy_options {
     const char *render_driver;
     const char *codec_options;
     const char *encoder_name;
+    const char *v4l2_device;
     enum sc_log_level log_level;
     enum sc_record_format record_format;
     struct sc_port_range port_range;
@@ -103,6 +104,7 @@ struct scrcpy_options {
     .render_driver = NULL, \
     .codec_options = NULL, \
     .encoder_name = NULL, \
+    .v4l2_device = NULL, \
     .log_level = SC_LOG_LEVEL_INFO, \
     .record_format = SC_RECORD_FORMAT_AUTO, \
     .port_range = { \

app/src/v4l2_sink.c

@@ -0,0 +1,341 @@
+#include "v4l2_sink.h"
+
+#include "util/log.h"
+#include "util/str_util.h"
+
+/** Downcast frame_sink to sc_v4l2_sink */
+#define DOWNCAST(SINK) container_of(SINK, struct sc_v4l2_sink, frame_sink)
+
+static const AVRational SCRCPY_TIME_BASE = {1, 1000000}; // timestamps in us
+
+static const AVOutputFormat *
+find_muxer(const char *name) {
+#ifdef SCRCPY_LAVF_HAS_NEW_MUXER_ITERATOR_API
+    void *opaque = NULL;
+#endif
+    const AVOutputFormat *oformat = NULL;
+    do {
+#ifdef SCRCPY_LAVF_HAS_NEW_MUXER_ITERATOR_API
+        oformat = av_muxer_iterate(&opaque);
+#else
+        oformat = av_oformat_next(oformat);
+#endif
+        // until null or containing the requested name
+    } while (oformat && !strlist_contains(oformat->name, ',', name));
+    return oformat;
+}
+
+static bool
+write_header(struct sc_v4l2_sink *vs, const AVPacket *packet) {
+    AVStream *ostream = vs->format_ctx->streams[0];
+
+    uint8_t *extradata = av_malloc(packet->size * sizeof(uint8_t));
+    if (!extradata) {
+        LOGC("Could not allocate extradata");
+        return false;
+    }
+
+    // copy the first packet to the extra data
+    memcpy(extradata, packet->data, packet->size);
+
+    ostream->codecpar->extradata = extradata;
+    ostream->codecpar->extradata_size = packet->size;
+
+    int ret = avformat_write_header(vs->format_ctx, NULL);
+    if (ret < 0) {
+        LOGE("Failed to write header to %s", vs->device_name);
+        return false;
+    }
+
+    return true;
+}
+
+static void
+rescale_packet(struct sc_v4l2_sink *vs, AVPacket *packet) {
+    AVStream *ostream = vs->format_ctx->streams[0];
+    av_packet_rescale_ts(packet, SCRCPY_TIME_BASE, ostream->time_base);
+}
+
+static bool
+write_packet(struct sc_v4l2_sink *vs, AVPacket *packet) {
+    if (!vs->header_written) {
+        bool ok = write_header(vs, packet);
+        if (!ok) {
+            return false;
+        }
+        vs->header_written = true;
+        return true;
+    }
+
+    rescale_packet(vs, packet);
+
+    bool ok = av_write_frame(vs->format_ctx, packet) >= 0;
+
+    // Failing to write the last frame is not very serious, no future frame may
+    // depend on it, so the resulting file will still be valid
+    (void) ok;
+
+    return true;
+}
+
+static bool
+encode_and_write_frame(struct sc_v4l2_sink *vs, const AVFrame *frame) {
+    int ret = avcodec_send_frame(vs->encoder_ctx, frame);
+    if (ret < 0 && ret != AVERROR(EAGAIN)) {
+        LOGE("Could not send v4l2 video frame: %d", ret);
+        return false;
+    }
+
+    AVPacket *packet = &vs->packet;
+    ret = avcodec_receive_packet(vs->encoder_ctx, packet);
+    if (ret == 0) {
+        // A packet was received
+
+        bool ok = write_packet(vs, packet);
+        if (!ok) {
+            LOGW("Could not send packet to v4l2 sink");
+            return false;
+        }
+        av_packet_unref(packet);
+    } else if (ret != AVERROR(EAGAIN)) {
+        LOGE("Could not receive v4l2 video packet: %d", ret);
+        return false;
+    }
+
+    return true;
+}
+
+static int
+run_v4l2_sink(void *data) {
+    struct sc_v4l2_sink *vs = data;
+
+    for (;;) {
+        sc_mutex_lock(&vs->mutex);
+
+        while (!vs->stopped && vs->vb.pending_frame_consumed) {
+            sc_cond_wait(&vs->cond, &vs->mutex);
+        }
+
+        if (vs->stopped) {
+            sc_mutex_unlock(&vs->mutex);
+            break;
+        }
+
+        sc_mutex_unlock(&vs->mutex);
+
+        video_buffer_consume(&vs->vb, vs->frame);
+        bool ok = encode_and_write_frame(vs, vs->frame);
+        if (!ok) {
+            LOGE("Could not send frame to v4l2 sink");
+            break;
+        }
+    }
+
+    LOGD("V4l2 thread ended");
+
+    return 0;
+}
+
+static bool
+sc_v4l2_sink_open(struct sc_v4l2_sink *vs) {
+    bool ok = video_buffer_init(&vs->vb);
+    if (!ok) {
+        return false;
+    }
+
+    ok = sc_mutex_init(&vs->mutex);
+    if (!ok) {
+        LOGC("Could not create mutex");
+        goto error_video_buffer_destroy;
+    }
+
+    ok = sc_cond_init(&vs->cond);
+    if (!ok) {
+        LOGC("Could not create cond");
+        goto error_mutex_destroy;
+    }
+
+    // FIXME
+    const AVOutputFormat *format = find_muxer("video4linux2,v4l2");
+    if (!format) {
+        LOGE("Could not find v4l2 muxer");
+        goto error_cond_destroy;
+    }
+
+    const AVCodec *encoder = avcodec_find_encoder(AV_CODEC_ID_RAWVIDEO);
+    if (!encoder) {
+        LOGE("Raw video encoder not found");
+        return false;
+    }
+
+    vs->format_ctx = avformat_alloc_context();
+    if (!vs->format_ctx) {
+        LOGE("Could not allocate v4l2 output context");
+        return false;
+    }
+
+    // contrary to the deprecated API (av_oformat_next()), av_muxer_iterate()
+    // returns (on purpose) a pointer-to-const, but AVFormatContext.oformat
+    // still expects a pointer-to-non-const (it has not be updated accordingly)
+    // <https://github.com/FFmpeg/FFmpeg/commit/0694d8702421e7aff1340038559c438b61bb30dd>
+    vs->format_ctx->oformat = (AVOutputFormat *) format;
+    vs->format_ctx->url = strdup(vs->device_name);
+    if (!vs->format_ctx->url) {
+        LOGE("Could not strdup v4l2 device name");
+        goto error_avformat_free_context;
+        return false;
+    }
+
+    AVStream *ostream = avformat_new_stream(vs->format_ctx, encoder);
+    if (!ostream) {
+        LOGE("Could not allocate new v4l2 stream");
+        goto error_avformat_free_context;
+        return false;
+    }
+
+    ostream->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
+    ostream->codecpar->codec_id = encoder->id;
+    ostream->codecpar->format = AV_PIX_FMT_YUV420P;
+    ostream->codecpar->width = vs->frame_size.width;
+    ostream->codecpar->height = vs->frame_size.height;
+
+    int ret = avio_open(&vs->format_ctx->pb, vs->device_name, AVIO_FLAG_WRITE);
+    if (ret < 0) {
+        LOGE("Failed to open output device: %s", vs->device_name);
+        // ostream will be cleaned up during context cleaning
+        goto error_avformat_free_context;
+    }
+
+    vs->encoder_ctx = avcodec_alloc_context3(encoder);
+    if (!vs->encoder_ctx) {
+        LOGC("Could not allocate codec context for v4l2");
+        goto error_avio_close;
+    }
+
+    vs->encoder_ctx->width = vs->frame_size.width;
+    vs->encoder_ctx->height = vs->frame_size.height;
+    vs->encoder_ctx->pix_fmt = AV_PIX_FMT_YUV420P;
+    vs->encoder_ctx->time_base.num = 1;
+    vs->encoder_ctx->time_base.den = 1;
+
+    if (avcodec_open2(vs->encoder_ctx, encoder, NULL) < 0) {
+        LOGE("Could not open codec for v4l2");
+        goto error_avcodec_free_context;
+    }
+
+    vs->frame = av_frame_alloc();
+    if (!vs->frame) {
+        LOGE("Could not create v4l2 frame");
+        goto error_avcodec_close;
+    }
+
+    LOGD("Starting v4l2 thread");
+    ok = sc_thread_create(&vs->thread, run_v4l2_sink, "v4l2", vs);
+    if (!ok) {
+        LOGC("Could not start v4l2 thread");
+        goto error_av_frame_free;
+    }
+
+    vs->header_written = false;
+    vs->stopped = false;
+
+    LOGI("v4l2 sink started to device: %s", vs->device_name);
+
+    return true;
+
+error_av_frame_free:
+    av_frame_free(&vs->frame);
+error_avcodec_close:
+    avcodec_close(vs->encoder_ctx);
+error_avcodec_free_context:
+    avcodec_free_context(&vs->encoder_ctx);
+error_avio_close:
+    avio_close(vs->format_ctx->pb);
+error_avformat_free_context:
+    avformat_free_context(vs->format_ctx);
+error_cond_destroy:
+    sc_cond_destroy(&vs->cond);
+error_mutex_destroy:
+    sc_mutex_destroy(&vs->mutex);
+error_video_buffer_destroy:
+    video_buffer_destroy(&vs->vb);
+
+    return false;
+}
+
+static void
+sc_v4l2_sink_close(struct sc_v4l2_sink *vs) {
+    sc_mutex_lock(&vs->mutex);
+    vs->stopped = true;
+    sc_cond_signal(&vs->cond);
+    sc_mutex_unlock(&vs->mutex);
+
+    sc_thread_join(&vs->thread, NULL);
+
+    av_frame_free(&vs->frame);
+    avcodec_close(vs->encoder_ctx);
+    avcodec_free_context(&vs->encoder_ctx);
+    avio_close(vs->format_ctx->pb);
+    avformat_free_context(vs->format_ctx);
+    sc_cond_destroy(&vs->cond);
+    sc_mutex_destroy(&vs->mutex);
+    video_buffer_destroy(&vs->vb);
+}
+
+static bool
+sc_v4l2_sink_push(struct sc_v4l2_sink *vs, const AVFrame *frame) {
+    bool ok = video_buffer_push(&vs->vb, frame, NULL);
+    if (!ok) {
+        return false;
+    }
+
+    // signal possible change of vs->vb.pending_frame_consumed
+    sc_cond_signal(&vs->cond);
+
+    return true;
+}
+
+static bool
+sc_v4l2_frame_sink_open(struct sc_frame_sink *sink) {
+    struct sc_v4l2_sink *vs = DOWNCAST(sink);
+    return sc_v4l2_sink_open(vs);
+}
+
+static void
+sc_v4l2_frame_sink_close(struct sc_frame_sink *sink) {
+    struct sc_v4l2_sink *vs = DOWNCAST(sink);
+    sc_v4l2_sink_close(vs);
+}
+
+static bool
+sc_v4l2_frame_sink_push(struct sc_frame_sink *sink, const AVFrame *frame) {
+    struct sc_v4l2_sink *vs = DOWNCAST(sink);
+    return sc_v4l2_sink_push(vs, frame);
+}
+
+bool
+sc_v4l2_sink_init(struct sc_v4l2_sink *vs, const char *device_name,
+                  struct size frame_size) {
+    vs->device_name = strdup(device_name);
+    if (!vs->device_name) {
+        LOGE("Could not strdup v4l2 device name");
+        return false;
+    }
+
+    vs->frame_size = frame_size;
+
+    static const struct sc_frame_sink_ops ops = {
+        .open = sc_v4l2_frame_sink_open,
+        .close = sc_v4l2_frame_sink_close,
+        .push = sc_v4l2_frame_sink_push,
+    };
+
+    vs->frame_sink.ops = &ops;
+
+    return true;
+}
+
+void
+sc_v4l2_sink_destroy(struct sc_v4l2_sink *vs) {
+    free(vs->device_name);
+}

app/src/v4l2_sink.h

@@ -0,0 +1,39 @@
+#ifndef SC_V4L2_SINK_H
+#define SC_V4L2_SINK_H
+
+#include "common.h"
+
+#include "coords.h"
+#include "trait/frame_sink.h"
+#include "video_buffer.h"
+
+#include <libavformat/avformat.h>
+
+struct sc_v4l2_sink {
+    struct sc_frame_sink frame_sink; // frame sink trait
+
+    struct video_buffer vb;
+    AVFormatContext *format_ctx;
+    AVCodecContext *encoder_ctx;
+
+    char *device_name;
+    struct size frame_size;
+
+    sc_thread thread;
+    sc_mutex mutex;
+    sc_cond cond;
+    bool stopped;
+    bool header_written;
+
+    AVFrame *frame;
+    AVPacket packet;
+};
+
+bool
+sc_v4l2_sink_init(struct sc_v4l2_sink *vs, const char *device_name,
+                  struct size frame_size);
+
+void
+sc_v4l2_sink_destroy(struct sc_v4l2_sink *vs);
+
+#endif