feat(local): 支持上传并播放本地视频文件，只保留音轨 (#149)

本地上传此前只接受音频。想放一段本地 mp4/mov/avi 里的音乐，四道关卡
挡着（前两道在服务端，后两道在浏览器端）：

1. src/music/local.ts 的 AUDIO_EXTENSIONS 只列了 12 种音频后缀；
2. src/web/api/music.ts 里 express.raw 的 type 只匹配 audio/*、
   video/webm、application/octet-stream —— 浏览器给 .mp4 打的
   Content-Type 是 video/mp4，请求体压根不会被解析，处理函数看到
   req.body === undefined，回 400「raw audio body is required」；
3. Search.vue 的 accept 属性让文件选择框把视频文件置灰；
4. isAudioFile() 把拖进来的视频文件静默丢掉。

ffmpeg 层不是瓶颈：s16le 输出格式不接受视频，ffmpeg 的自动选流本来
就只挑音轨。实测 mp4/mov/avi/mkv/flv/wmv/ts/m4v/mpg 九种容器用现有
参数全部正常出声，多音轨、带字幕、带 timecode 的也一样，所以
buildFfmpegArgs 一个字没动。

## 改动

- **打通四道关卡**：新增 VIDEO_EXTENSIONS（mp4/mov/avi/mkv/flv/wmv/
  m4v/mpg/mpeg/3gp/ts/m2ts/ogv），express.raw 收 video/*，前端 accept
  与过滤函数同步放宽。
- **上传时抽取音轨**（extractAudioTrack）：视频落盘后用
  `-vn -sn -dn -map 0:a:0 -c:a copy` 把音轨原样搬进 Matroska 音频容器
  （.mka）再删掉原视频。`-c:a copy` 不重编码，无损、快，且 Matroska
  几乎收所有音频编码，不用维护「编码→后缀」对照表。实测 720p 素材
  落盘体积降到原文件的 14%，这对 5 GiB 的上传目录配额很关键——否则
  十来个视频就把配额占满了。抽取失败（冷门编码、超时）则保留原容器
  继续播，只是占地方，绝不会因此上传失败。
- **拒绝没有音轨的视频**：上传时探测，直接回「这个视频里没有音轨，
  无法播放」，而不是等到播放时静默跳过。只在 ffmpeg 确实打开了容器
  （打印了 `Input #0,`）时才拒绝——认不出的字节一律放行，截断的 mp3
  一直是这个行为，不能因为这次改动开始被拒。
- **上限从 200mb 提到 500mb**，并把超限响应从 Express 默认的 HTML
  错误页（带堆栈和服务器绝对路径）换成和本路由一致的 JSON；前端也加
  了同样的预检，不再传完几百兆才被拒。
- **上传进度**：视频比音频大得多，原来那句静止的「正在上传 N 个文件」
  看着像卡死，现在按文件显示百分比，传完切到「服务端处理中」。

## 验证

- 全量 `npx vitest run`：136 个文件 / 2070 项，新增 24 项。
- 新增测试用 ffmpeg 现造真实容器跑端到端：mp4 上传后时长正确、原
  容器已删、剩下的 .mka 能被播放链路解码出 PCM；avi/mkv/flv 同样；
  无音轨视频被拒且不留残留文件；纯音频上传字节数不变、不被重封装。
- 变异测试（逐个改回旧实现，确认新测试真的会红）：后缀白名单 4 项失败、
  express.raw 的 type 5 项失败、抽取音轨 2 项失败、无音轨拒绝 2 项失败。
- `npx tsc --noEmit` 与 `npx vue-tsc --noEmit` 均 exit 0。

Reported-by: @LadenceE
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

src/music/local.test.ts

@@ -1,8 +1,11 @@
 import { describe, it, expect, beforeEach, afterEach } from "vitest";
-import { mkdtempSync, rmSync, existsSync, writeFileSync } from "node:fs";
+import { mkdtempSync, rmSync, existsSync, writeFileSync, readFileSync, readdirSync, statSync } from "node:fs";
+import { spawnSync } from "node:child_process";
+import { createRequire } from "node:module";
+import { buildFfmpegArgs } from "../audio/player.js";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
-import { LocalMusicProvider } from "./local.js";
+import { LocalMusicProvider, parseMediaProbe } from "./local.js";
 
 let dir: string;
 
@@ -149,6 +152,115 @@ describe("LocalMusicProvider upload validation", () => {
       p.uploadAudio({ buffer: Buffer.alloc(0), originalName: "a.mp3" }),
     ).rejects.toThrow();
   });
+
+  // #149: video containers are accepted; only their audio track is kept.
+  it("still rejects a non-media extension after video was allowed", async () => {
+    const p = new LocalMusicProvider(dir);
+    for (const name of ["evil.exe", "evil.html", "evil.mp4.txt", "notes.pdf"]) {
+      await expect(
+        p.uploadAudio({ buffer: Buffer.from("x"), originalName: name, mimeType: "video/mp4" }),
+      ).rejects.toThrow();
+    }
+  });
+
+  it("accepts every supported video extension at the extension gate", async () => {
+    const p = new LocalMusicProvider(dir);
+    // Junk content: ffmpeg cannot open it, so it is "unrecognised" rather than
+    // "no audio track" and must be accepted exactly like a truncated .mp3
+    // always has been. The extension allowlist is what is under test here.
+    // .m4v is excluded on purpose — see the next test.
+    for (const ext of [".mp4", ".mov", ".avi", ".mkv", ".flv", ".wmv", ".mpg", ".mpeg", ".3gp", ".ts", ".m2ts", ".ogv"]) {
+      const song = await p.uploadAudio({
+        buffer: Buffer.from("not really a video"),
+        originalName: `clip${ext}`,
+        mimeType: "video/mp4",
+      });
+      expect(song.platform).toBe("local");
+      expect(song.name).toBe("clip");
+    }
+  });
+
+  it("refuses a .m4v raw video elementary stream, which by definition has no audio", async () => {
+    // .m4v is not a container — ffmpeg's rawvideo demuxer opens arbitrary
+    // bytes as an MPEG-4 video elementary stream, so it IS recognised and
+    // genuinely carries no audio track. Refusing it is the correct outcome,
+    // and it is the one case that distinguishes `recognized` from `probed`.
+    const p = new LocalMusicProvider(dir);
+    await expect(
+      p.uploadAudio({
+        buffer: Buffer.from("not really a video"),
+        originalName: "clip.m4v",
+        mimeType: "video/x-m4v",
+      }),
+    ).rejects.toThrow(/音轨/);
+  });
+
+  it("the error message names both audio and video formats", async () => {
+    const p = new LocalMusicProvider(dir);
+    await expect(
+      p.uploadAudio({ buffer: Buffer.from("x"), originalName: "a.exe" }),
+    ).rejects.toThrow(/视频/);
+  });
+});
+
+describe("parseMediaProbe (#149)", () => {
+  const mp4Banner = `Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'clip.mp4':
+  Duration: 00:03:27.15, start: 0.000000, bitrate: 1105 kb/s
+  Stream #0:0[0x1](und): Video: h264 (High), yuv420p, 1280x720, 30 fps
+  Stream #0:1[0x2](und): Audio: aac (LC), 48000 Hz, stereo, fltp, 192 kb/s`;
+
+  it("reads duration and detects the audio stream in a video container", () => {
+    const r = parseMediaProbe(mp4Banner);
+    expect(r.durationSeconds).toBe(3 * 60 + 27);
+    expect(r.hasAudio).toBe(true);
+  });
+
+  it("reports hasAudio false for a video with only a video stream", () => {
+    const silent = `Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'silent.mp4':
+  Duration: 00:00:02.00, start: 0.000000, bitrate: 29 kb/s
+  Stream #0:0[0x1](und): Video: h264 (High 4:4:4 Predictive), yuv444p, 160x120, 10 fps`;
+    const r = parseMediaProbe(silent);
+    expect(r.durationSeconds).toBe(2);
+    expect(r.hasAudio).toBe(false);
+  });
+
+  it("detects a plain audio file", () => {
+    const r = parseMediaProbe(`Input #0, mp3, from 'a.mp3':
+  Duration: 00:00:30.02, start: 0.000000, bitrate: 128 kb/s
+  Stream #0:0: Audio: mp3, 44100 Hz, stereo, fltp, 128 kb/s`);
+    expect(r.durationSeconds).toBe(30);
+    expect(r.hasAudio).toBe(true);
+  });
+
+  it("does not mistake an attached cover image for an audio stream", () => {
+    const r = parseMediaProbe(`Input #0, mp3, from 'cover.mp3':
+  Duration: 00:00:10.00, start: 0.000000, bitrate: 130 kb/s
+  Stream #0:0: Audio: mp3, 44100 Hz, stereo, fltp, 128 kb/s
+  Stream #0:1: Video: mjpeg (Baseline), yuvj420p(pc), 100x100 [attached pic]`);
+    expect(r.hasAudio).toBe(true);
+  });
+
+  it("returns zeros on unparseable output rather than throwing", () => {
+    const r = parseMediaProbe("ffmpeg: command exploded");
+    expect(r.durationSeconds).toBe(0);
+    expect(r.hasAudio).toBe(false);
+    expect(r.recognized).toBe(false);
+  });
+
+  // The distinction that decides whether an upload is refused: ffmpeg opened
+  // the file and found no audio (refuse) vs ffmpeg could not open it at all
+  // (accept, as it always has for truncated audio).
+  it("marks a readable container recognized and unreadable bytes not", () => {
+    expect(parseMediaProbe(mp4Banner).recognized).toBe(true);
+    expect(parseMediaProbe(`[mov,mp4,m4a,3gp,3g2,mj2 @ 0x1] moov atom not found
+[in#0 @ 0x2] Error opening input: Invalid data found when processing input
+Error opening input file junk.mp4.`).recognized).toBe(false);
+  });
+
+  it("rounds fractional durations", () => {
+    expect(parseMediaProbe("Duration: 00:00:03.60,").durationSeconds).toBe(4);
+    expect(parseMediaProbe("Duration: 01:02:03.10,").durationSeconds).toBe(3723);
+  });
 });
 
 describe("LocalMusicProvider quota", () => {
@@ -233,3 +345,114 @@ describe("LocalMusicProvider filename handling", () => {
     expect(await p.getSongUrl(song.id)).not.toBeNull();
   });
 });
+
+// #149 end-to-end: build real containers with the bundled ffmpeg and push
+// them through the actual upload path. Skipped automatically if the binary is
+// unavailable, so the suite still runs on a machine without it.
+describe("LocalMusicProvider video upload, end to end (#149)", () => {
+  const ffmpeg: string | null = (() => {
+    try {
+      return createRequire(import.meta.url)("ffmpeg-static") as string;
+    } catch {
+      return null;
+    }
+  })();
+
+  const have = !!ffmpeg && spawnSync(ffmpeg, ["-version"], { stdio: "ignore" }).status === 0;
+
+  /** Render a real container into the temp dir and return its bytes. */
+  function render(name: string, args: string[]): Buffer {
+    const out = join(dir, name);
+    const r = spawnSync(ffmpeg!, ["-y", "-hide_banner", "-loglevel", "error", ...args, out], {
+      stdio: "ignore",
+    });
+    if (r.status !== 0) throw new Error(`fixture render failed: ${name}`);
+    const buf = readFileSync(out);
+    rmSync(out, { force: true }); // upload writes its own copy
+    return buf;
+  }
+
+  const withAudio = (dur: number, vcodec: string, acodec: string) => [
+    "-f", "lavfi", "-i", `testsrc=s=160x120:r=10:d=${dur}`,
+    "-f", "lavfi", "-i", `sine=f=440:d=${dur}`,
+    "-c:v", vcodec, "-c:a", acodec, "-shortest",
+  ];
+
+  it.runIf(have)("accepts an mp4, reads its duration, and keeps only the audio", async () => {
+    const p = new LocalMusicProvider(dir);
+    const mp4 = render("src.mp4", withAudio(3, "libx264", "aac"));
+    const song = await p.uploadAudio({
+      buffer: mp4, originalName: "My Clip.mp4", mimeType: "video/mp4",
+    });
+
+    expect(song.name).toBe("My Clip");
+    expect(song.platform).toBe("local");
+    expect(song.duration).toBe(3);
+
+    const resolved = await p.getSongUrl(song.id);
+    expect(resolved).not.toBeNull();
+    // The video container is gone; what remains is the extracted audio track.
+    expect(resolved!.url.endsWith(".mka")).toBe(true);
+    expect(existsSync(join(dir, `${song.id}.mp4`))).toBe(false);
+    expect(existsSync(resolved!.url)).toBe(true);
+    expect(statSync(resolved!.url).size).toBeGreaterThan(0);
+    expect(statSync(resolved!.url).size).toBeLessThan(mp4.length);
+  }, 60000);
+
+  it.runIf(have)("extracted audio is still decodable by the player's ffmpeg args", async () => {
+    const p = new LocalMusicProvider(dir);
+    const song = await p.uploadAudio({
+      buffer: render("src2.mp4", withAudio(2, "libx264", "aac")),
+      originalName: "clip.mp4",
+      mimeType: "video/mp4",
+    });
+    const url = (await p.getSongUrl(song.id))!.url;
+
+    const decoded = spawnSync(
+      ffmpeg!,
+      [...buildFfmpegArgs(url, 0).slice(0, -1), "-"],
+      { maxBuffer: 64 * 1024 * 1024 },
+    );
+    expect(decoded.status).toBe(0);
+    // 2s of 48 kHz stereo s16le ≈ 384000 bytes; allow codec priming slack.
+    expect(decoded.stdout.length).toBeGreaterThan(300000);
+  }, 60000);
+
+  it.runIf(have)("refuses a video that genuinely has no audio track", async () => {
+    const p = new LocalMusicProvider(dir);
+    const silent = render("silent.mp4", [
+      "-f", "lavfi", "-i", "testsrc=s=160x120:r=10:d=2", "-an",
+    ]);
+    await expect(
+      p.uploadAudio({ buffer: silent, originalName: "silent.mp4", mimeType: "video/mp4" }),
+    ).rejects.toThrow(/音轨/);
+    // The rejected upload must not leave its bytes behind.
+    expect(readdirSync(dir).filter((f) => f.endsWith(".mp4"))).toEqual([]);
+  }, 60000);
+
+  it.runIf(have)("extracts losslessly from avi/mkv/flv too, not just mp4", async () => {
+    const p = new LocalMusicProvider(dir);
+    const cases: Array<[string, string[]]> = [
+      ["a.avi", withAudio(2, "mpeg4", "libmp3lame")],
+      ["a.mkv", withAudio(2, "libx264", "libopus")],
+      ["a.flv", withAudio(2, "flv", "libmp3lame")],
+    ];
+    for (const [name, args] of cases) {
+      const song = await p.uploadAudio({
+        buffer: render(`src-${name}`, args), originalName: name, mimeType: "video/x-msvideo",
+      });
+      const url = (await p.getSongUrl(song.id))!.url;
+      expect(url.endsWith(".mka")).toBe(true);
+      expect(statSync(url).size).toBeGreaterThan(0);
+    }
+  }, 120000);
+
+  it.runIf(have)("a plain audio upload is untouched — no extraction, original extension kept", async () => {
+    const p = new LocalMusicProvider(dir);
+    const mp3 = render("src.mp3", ["-f", "lavfi", "-i", "sine=f=440:d=2", "-c:a", "libmp3lame"]);
+    const song = await p.uploadAudio({ buffer: mp3, originalName: "tune.mp3", mimeType: "audio/mpeg" });
+    const url = (await p.getSongUrl(song.id))!.url;
+    expect(url.endsWith(".mp3")).toBe(true);
+    expect(statSync(url).size).toBe(mp3.length); // byte-identical, not remuxed
+  }, 60000);
+});

src/music/local.ts

@@ -1,5 +1,5 @@
 import { spawn } from "node:child_process";
-import { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, readFileSync, rmSync, statSync, writeFileSync } from "node:fs";
 import { createRequire } from "node:module";
 import path from "node:path";
 import crypto from "node:crypto";
@@ -34,6 +34,37 @@ const AUDIO_EXTENSIONS = new Set([
   ".ape",
 ]);
 
+/** Video containers accepted for upload (#149). Only the audio track is ever
+ *  used — the bot has no video output. Playback would work straight from the
+ *  container (ffmpeg selects the audio stream), but we extract the audio on
+ *  upload so a 200 MB clip does not sit on disk for a 3 MB song; see
+ *  extractAudioTrack. `.webm` is deliberately absent: it is already in
+ *  AUDIO_EXTENSIONS and both audio-only and video .webm are handled there. */
+const VIDEO_EXTENSIONS = new Set([
+  ".mp4",
+  ".mov",
+  ".avi",
+  ".mkv",
+  ".flv",
+  ".wmv",
+  ".m4v",
+  ".mpg",
+  ".mpeg",
+  ".3gp",
+  ".ts",
+  ".m2ts",
+  ".ogv",
+]);
+
+/** Container the extracted audio track is remuxed into. Matroska takes
+ *  essentially any audio codec, so `-c:a copy` works without knowing what the
+ *  source used — no re-encode, no quality loss, no codec/extension table. */
+const EXTRACTED_AUDIO_EXT = ".mka";
+
+function isSupportedUploadExt(ext: string): boolean {
+  return AUDIO_EXTENSIONS.has(ext) || VIDEO_EXTENSIONS.has(ext);
+}
+
 const DEFAULT_MAX_FILES = 200;
 const DEFAULT_MAX_TOTAL_BYTES = 5 * 1024 * 1024 * 1024; // 5 GiB
 
@@ -71,39 +102,125 @@ function titleFromFileName(name: string): string {
   return safeFileName(name).replace(/\.[^.]+$/, "") || "本地音频";
 }
 
-async function probeDurationSeconds(filePath: string): Promise<number> {
+export interface MediaProbe {
+  /** Rounded seconds, 0 when the probe failed or the container has no duration. */
+  durationSeconds: number;
+  /** True when ffmpeg reported at least one audio stream. Only meaningful
+   *  together with `recognized` — see the comment there. */
+  hasAudio: boolean;
+  /**
+   * True when ffmpeg actually opened the container and printed its
+   * `Input #0, <format>, from '...'` header.
+   *
+   * This is what separates "ffmpeg looked inside and there is genuinely no
+   * audio track" from "ffmpeg could not make sense of these bytes at all".
+   * Both produce hasAudio === false, but only the first is a file we should
+   * refuse. Unreadable bytes have always been accepted here (a truncated mp3
+   * uploads fine and simply reports duration 0), and that stays true.
+   */
+  recognized: boolean;
+  /** False when ffmpeg could not be run or timed out, so nothing else in this
+   *  object is meaningful and the caller must not reject the file on it. */
+  probed: boolean;
+}
+
+/** Parse `Duration: HH:MM:SS.ss`, the `Input #0,` header and
+ *  `Stream #0:N...: Audio:` out of the banner ffmpeg prints on stderr when
+ *  asked to open a file with no output. */
+export function parseMediaProbe(stderr: string): Omit<MediaProbe, "probed"> {
+  const match = stderr.match(/Duration:\s*(\d+):(\d+):(\d+(?:\.\d+)?)/);
+  let durationSeconds = 0;
+  if (match) {
+    const total = Number(match[1]) * 3600 + Number(match[2]) * 60 + Number(match[3]);
+    durationSeconds = Number.isFinite(total) ? Math.round(total) : 0;
+  }
+  // e.g. "  Stream #0:1[0x2](und): Audio: aac (LC) ..." — the stream index and
+  // the bracketed id/language vary, so match on the "Audio:" tag itself. An
+  // embedded cover image is a separate "Video: mjpeg ... [attached pic]" line
+  // and never matches this.
+  const hasAudio = /Stream #\d+:\d+[^\n]*:\s*Audio:/.test(stderr);
+  // "Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'clip.mp4':" — absent entirely
+  // when ffmpeg bails with "Error opening input: Invalid data found ...".
+  const recognized = /^Input #\d+,/m.test(stderr);
+  return { durationSeconds, hasAudio, recognized };
+}
+
+async function probeMedia(filePath: string): Promise<MediaProbe> {
   return new Promise((resolve) => {
     const ffmpeg = spawn(ffmpegPath || "ffmpeg", ["-hide_banner", "-i", filePath], {
       stdio: ["ignore", "ignore", "pipe"],
     });
     let stderr = "";
+    let settled = false;
+    const done = (probe: MediaProbe) => {
+      if (settled) return;
+      settled = true;
+      resolve(probe);
+    };
+    // Video containers are much larger than the audio files this used to see,
+    // and the probe only reads headers — but a network/USB path can still be
+    // slow, so allow more than the old 5s before giving up.
     const timeout = setTimeout(() => {
       ffmpeg.kill("SIGKILL");
-      resolve(0);
-    }, 5000);
+      done({ durationSeconds: 0, hasAudio: false, recognized: false, probed: false });
+    }, 20000);
     ffmpeg.stderr.on("data", (chunk) => {
       stderr += chunk.toString("utf8");
     });
     ffmpeg.on("error", () => {
       clearTimeout(timeout);
-      resolve(0);
+      done({ durationSeconds: 0, hasAudio: false, recognized: false, probed: false });
     });
     ffmpeg.on("close", () => {
       clearTimeout(timeout);
-      const match = stderr.match(/Duration:\s*(\d+):(\d+):(\d+(?:\.\d+)?)/);
-      if (!match) {
-        resolve(0);
-        return;
-      }
-      const hours = Number(match[1]);
-      const minutes = Number(match[2]);
-      const seconds = Number(match[3]);
-      const total = hours * 3600 + minutes * 60 + seconds;
-      resolve(Number.isFinite(total) ? Math.round(total) : 0);
+      done({ ...parseMediaProbe(stderr), probed: true });
     });
   });
 }
 
+/**
+ * Remux the first audio stream of `source` into `target` (#149).
+ *
+ * `-c:a copy` — the audio is moved bit-for-bit into a Matroska audio
+ * container, so this is fast, lossless, and codec-agnostic. Nothing is
+ * re-encoded, so a 200 MB .mp4 becomes a few MB .mka with the original audio
+ * intact. Video, subtitle and data streams are dropped.
+ *
+ * Returns true only if ffmpeg exited 0 AND produced a non-empty file, so a
+ * partial/zero-byte result can never be mistaken for a successful extraction.
+ * Callers fall back to keeping the original container, which plays fine.
+ */
+async function extractAudioTrack(source: string, target: string): Promise<boolean> {
+  const ok = await new Promise<boolean>((resolve) => {
+    const ffmpeg = spawn(
+      ffmpegPath || "ffmpeg",
+      ["-hide_banner", "-loglevel", "error", "-y", "-i", source,
+       "-vn", "-sn", "-dn", "-map", "0:a:0", "-c:a", "copy", target],
+      { stdio: ["ignore", "ignore", "ignore"] },
+    );
+    let settled = false;
+    const done = (v: boolean) => {
+      if (settled) return;
+      settled = true;
+      resolve(v);
+    };
+    // Remuxing is I/O bound, but a multi-GB input on a slow disk still takes
+    // a while. Cap it so a pathological file cannot wedge the upload request.
+    const timeout = setTimeout(() => {
+      ffmpeg.kill("SIGKILL");
+      done(false);
+    }, 120000);
+    ffmpeg.on("error", () => { clearTimeout(timeout); done(false); });
+    ffmpeg.on("close", (code) => { clearTimeout(timeout); done(code === 0); });
+  });
+  if (!ok) return false;
+  try {
+    return statSync(target).size > 0;
+  } catch {
+    return false;
+  }
+}
+
 export class LocalMusicProvider implements MusicProvider {
   readonly platform = "local" as const;
   private readonly uploadDir: string;
@@ -171,33 +288,75 @@ export class LocalMusicProvider implements MusicProvider {
     const ext = path.extname(originalName).toLowerCase();
     // Validate by the (sanitised) file extension only — never trust the
     // client-supplied Content-Type. This also guarantees the STORED extension
-    // is one of the known audio types, so a spoofed header cannot persist an
-    // arbitrary-extension blob on disk.
-    if (!AUDIO_EXTENSIONS.has(ext)) {
-      throw new Error("只支持常见音频文件，如 mp3、flac、wav、m4a、ogg、opus、aac、webm 等");
+    // is one of the known audio/video types, so a spoofed header cannot
+    // persist an arbitrary-extension blob on disk.
+    if (!isSupportedUploadExt(ext)) {
+      throw new Error(
+        "只支持常见音频文件（mp3、flac、wav、m4a、ogg、opus、aac、webm 等）" +
+        "和视频文件（mp4、mov、avi、mkv、flv、wmv 等，仅取其中的音轨播放）",
+      );
     }
     if (!input.buffer || input.buffer.length === 0) {
       throw new Error("上传文件为空");
     }
 
     const id = crypto.randomUUID();
-    const storedName = `${id}${ext}`;
-    const filePath = path.join(this.uploadDir, storedName);
+    const isVideo = VIDEO_EXTENSIONS.has(ext);
+    let filePath = path.join(this.uploadDir, `${id}${ext}`);
     writeFileSync(filePath, input.buffer);
 
-    const duration = await probeDurationSeconds(filePath);
+    let probe: MediaProbe;
+    try {
+      probe = await probeMedia(filePath);
+    } catch {
+      probe = { durationSeconds: 0, hasAudio: false, recognized: false, probed: false };
+    }
+
+    // Reject a video with no audio track up front (#149). Left to playback it
+    // would produce a silent, zero-byte stream that just looks like a broken
+    // song. Require `recognized` as well as `probed`: bytes ffmpeg cannot open
+    // at all report hasAudio false for a different reason, and those have
+    // always been accepted (a truncated upload lands with duration 0) — this
+    // change must not start rejecting them.
+    if (isVideo && probe.probed && probe.recognized && !probe.hasAudio) {
+      rmSync(filePath, { force: true });
+      throw new Error("这个视频里没有音轨，无法播放");
+    }
+
+    let size = input.buffer.length;
+    if (isVideo) {
+      // Keep only the audio. The video bytes are dead weight against the
+      // upload-directory quota and would never be used.
+      const extracted = path.join(this.uploadDir, `${id}${EXTRACTED_AUDIO_EXT}`);
+      if (await extractAudioTrack(filePath, extracted)) {
+        try {
+          size = statSync(extracted).size;
+          rmSync(filePath, { force: true });
+          filePath = extracted;
+        } catch {
+          // Could not stat/remove (Windows lock) — keep playing the original
+          // container and drop the half-finished extract.
+          rmSync(extracted, { force: true });
+        }
+      } else {
+        // Extraction failed (exotic codec Matroska won't take, timeout, …).
+        // The original container still plays: ffmpeg picks its audio stream.
+        rmSync(extracted, { force: true });
+      }
+    }
+
     const song: LocalSongRecord = {
       id,
       name: titleFromFileName(originalName),
       artist: "本地上传",
       album: "本地音乐",
-      duration,
+      duration: probe.durationSeconds,
       coverUrl: "",
       platform: "local",
       filePath,
       originalName,
       uploadedAt: new Date().toISOString(),
-      size: input.buffer.length,
+      size,
       mimeType: input.mimeType || "application/octet-stream",
     };
 

src/web/api/music.test.ts

@@ -14,7 +14,7 @@ import { createSessionStore } from "../../data/sessions.js";
 import { createPermissionStore } from "../../data/permissions.js";
 import { createRequireAuth } from "../middleware/requireAuth.js";
 import { SESSION_COOKIE_NAME } from "../auth/validateSession.js";
-import { createMusicRouter } from "./music.js";
+import { createMusicRouter, createLocalUploadBody } from "./music.js";
 
 const empty: SearchResult = { songs: [], albums: [], playlists: [] };
 
@@ -295,3 +295,146 @@ describe("music router POST /quality — persistence (#125)", () => {
     expect(config.audioQuality.netease).toBe("lossless");
   });
 });
+
+// #149: video containers must survive the transport layer. Before this the
+// express.raw type filter only matched audio/*, video/webm and
+// application/octet-stream, so a browser-sent video/mp4 body was never parsed
+// and the handler answered 400 "raw audio body is required".
+describe("music router POST /local/upload — content types and size cap (#149)", () => {
+  let app: express.Express;
+  let botDb: BotDatabase;
+  let cookie: string;
+  let uploadAudio: ReturnType<typeof vi.fn>;
+
+  beforeEach(async () => {
+    botDb = createDatabase(":memory:");
+    const users = createUserStore(botDb.db);
+    const sessions = createSessionStore(botDb.db);
+    const admin = await users.createUser("admin", "pw-admin", "admin");
+    cookie = `${SESSION_COOKIE_NAME}=${sessions.createSession(admin.id).token}`;
+
+    uploadAudio = vi.fn(async (input: { originalName: string }) => ({
+      id: "local-1", name: input.originalName, artist: "本地上传", album: "本地音乐",
+      duration: 1, coverUrl: "", platform: "local",
+    }));
+    const local = { platform: "local", search: vi.fn().mockResolvedValue(empty), uploadAudio } as unknown as MusicProvider;
+
+    app = express();
+    app.use(express.json());
+    app.use(cookieParser());
+    app.use("/api", createRequireAuth(sessions, createPermissionStore(botDb.db), () => getDefaultConfig().guestMode));
+    app.use("/api/music", createMusicRouter(
+      fakeProvider("netease"), fakeProvider("qq"), fakeProvider("bilibili"),
+      pino({ level: "silent" }), local, getDefaultConfig(),
+    ));
+  });
+
+  afterEach(() => botDb.close());
+
+  const post = (contentType: string, body: Buffer, name = "clip.mp4") =>
+    request(app)
+      .post("/api/music/local/upload")
+      .set("Cookie", cookie)
+      .set("Content-Type", contentType)
+      .set("X-Filename", encodeURIComponent(name))
+      .send(body);
+
+  it("accepts the video MIME types browsers actually send", async () => {
+    // These are what Chrome/Firefox put on a File for .mp4/.mov/.avi/.mkv.
+    for (const ct of ["video/mp4", "video/quicktime", "video/x-msvideo", "video/x-matroska", "video/webm"]) {
+      uploadAudio.mockClear();
+      const res = await post(ct, Buffer.from("fake video bytes"));
+      expect(res.status, `content-type ${ct}`).toBe(200);
+      expect(uploadAudio).toHaveBeenCalledOnce();
+      expect(res.body.song.platform).toBe("local");
+    }
+  });
+
+  it("still accepts audio and octet-stream bodies", async () => {
+    for (const ct of ["audio/mpeg", "audio/flac", "application/octet-stream"]) {
+      uploadAudio.mockClear();
+      const res = await post(ct, Buffer.from("fake audio"), "tune.mp3");
+      expect(res.status, `content-type ${ct}`).toBe(200);
+      expect(uploadAudio).toHaveBeenCalledOnce();
+    }
+  });
+
+  it("passes the decoded filename and the content type through to the provider", async () => {
+    await post("video/mp4", Buffer.from("bytes"), "我的 视频.mp4");
+    expect(uploadAudio).toHaveBeenCalledWith(
+      expect.objectContaining({ originalName: "我的 视频.mp4", mimeType: "video/mp4" }),
+    );
+  });
+
+  it("surfaces a provider rejection as a 400 with its message", async () => {
+    uploadAudio.mockRejectedValueOnce(new Error("这个视频里没有音轨，无法播放"));
+    const res = await post("video/mp4", Buffer.from("bytes"));
+    expect(res.status).toBe(400);
+    expect(res.body.error).toBe("这个视频里没有音轨，无法播放");
+  });
+
+  it("requires authentication", async () => {
+    const res = await request(app)
+      .post("/api/music/local/upload")
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.from("bytes"));
+    expect(res.status).toBe(401);
+  });
+
+  it("rejects an oversize body as JSON, not an HTML stack trace", async () => {
+    // Same middleware the route mounts, built with a small limit so the test
+    // does not have to allocate half a gigabyte to reach the cap.
+    const tiny = express();
+    const reached = vi.fn();
+    tiny.post("/u", createLocalUploadBody("1kb"), (_req, res) => { reached(); res.json({ ok: true }); });
+
+    const res = await request(tiny)
+      .post("/u")
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.alloc(4096, 1));
+
+    expect(res.status).toBe(413);
+    expect(res.headers["content-type"]).toMatch(/application\/json/);
+    expect(res.body.error).toContain("文件太大");
+    // The HTML default handler leaked absolute server paths and a stack.
+    expect(res.text).not.toMatch(/node_modules|<\/pre>|at read/);
+    expect(reached).not.toHaveBeenCalled();
+  });
+
+  it("lets a body under the cap through the same middleware", async () => {
+    const tiny = express();
+    tiny.post("/u", createLocalUploadBody("1kb"), (req, res) => {
+      res.json({ bytes: (req.body as Buffer).length });
+    });
+    const res = await request(tiny)
+      .post("/u")
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.alloc(512, 1));
+    expect(res.status).toBe(200);
+    expect(res.body.bytes).toBe(512);
+  });
+
+  it("rejects local uploads when the feature is switched off", async () => {
+    const off = getDefaultConfig();
+    off.localAudioEnabled = false;
+    const users = createUserStore(botDb.db);
+    const sessions = createSessionStore(botDb.db);
+    const a2 = await users.createUser("admin2", "pw-admin2", "admin");
+    const c2 = `${SESSION_COOKIE_NAME}=${sessions.createSession(a2.id).token}`;
+    const app2 = express();
+    app2.use(cookieParser());
+    app2.use("/api", createRequireAuth(sessions, createPermissionStore(botDb.db), () => getDefaultConfig().guestMode));
+    app2.use("/api/music", createMusicRouter(
+      fakeProvider("netease"), fakeProvider("qq"), fakeProvider("bilibili"),
+      pino({ level: "silent" }),
+      { platform: "local", search: vi.fn(), uploadAudio } as unknown as MusicProvider, off,
+    ));
+    const res = await request(app2)
+      .post("/api/music/local/upload")
+      .set("Cookie", c2)
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.from("bytes"));
+    expect(res.status).toBe(403);
+    expect(uploadAudio).not.toHaveBeenCalled();
+  });
+});

src/web/api/music.ts

@@ -7,6 +7,56 @@ import { requirePermission } from "../middleware/requirePermission.js";
 import { requireNotGuest } from "../middleware/requireNotGuest.js";
 import { authorize } from "../middleware/authorize.js";
 
+/**
+ * Body cap for a local upload. express.raw buffers the whole body in memory,
+ * so this is also the peak RAM one upload can cost — raised from 200mb for
+ * video (#149), which is far bigger than audio for the same song, but kept
+ * well short of "any video file at all" for that reason. Only the audio track
+ * survives to disk.
+ */
+export const LOCAL_UPLOAD_LIMIT = "500mb";
+
+/**
+ * Body parser for the local-upload route.
+ *
+ * `type` includes "video/*" (#149): the browser sends the File's own MIME
+ * type, so an .mp4 arrives as video/mp4 and used to be rejected by this
+ * filter before ever reaching the provider. Only the audio track is kept —
+ * uploadAudio remuxes it out on the way in.
+ *
+ * express.raw hands an oversize body to the default error handler, which
+ * answers with an HTML page carrying a stack trace and absolute server paths
+ * (unless NODE_ENV=production, which this project never sets). Video makes
+ * hitting the cap far more likely than audio did, so that one case is
+ * translated into the same JSON shape the rest of this route returns. Any
+ * other body-parser error is passed on untouched.
+ *
+ * Exported as a factory so tests can drive the identical path with a small
+ * limit instead of allocating half a gigabyte.
+ */
+export function createLocalUploadBody(limit: string): express.RequestHandler {
+  const raw = express.raw({
+    type: ["audio/*", "video/*", "application/octet-stream"],
+    limit,
+  });
+  return (req, res, next) => {
+    raw(req, res, (err?: unknown) => {
+      if (!err) {
+        next();
+        return;
+      }
+      const e = err as { type?: string; status?: number };
+      if (e?.type === "entity.too.large" || e?.status === 413) {
+        res.status(413).json({ error: `文件太大，单个文件上限 ${limit}` });
+        return;
+      }
+      next(err);
+    });
+  };
+}
+
+const localUploadBody = createLocalUploadBody(LOCAL_UPLOAD_LIMIT);
+
 export function createMusicRouter(
   neteaseProvider: MusicProvider,
   qqProvider: MusicProvider,
@@ -65,10 +115,7 @@ export function createMusicRouter(
       }
       next();
     },
-    express.raw({
-      type: ["audio/*", "video/webm", "application/octet-stream"],
-      limit: "200mb",
-    }),
+    localUploadBody,
     async (req, res) => {
       try {
         if (!localProvider) {