test/crypto/stream.test.ts

@@ -15,7 +15,8 @@
  * stream ended on a `TAG_FINAL` chunk and was therefore not truncated.
  *
  * These tests pin:
- *   - The chunk-size constants match Ente's expectations.
+ *   - The chunk-size constants match Ente's expectations, and the
+ *     re-exported `STREAM_TAG_FINAL` matches libsodium's own constant.
  *   - The pull state can decrypt a multi-chunk stream produced by
  *     sodium.crypto_secretstream_xchacha20poly1305_push, in order.
  *   - The tag byte is propagated to the caller.
@@ -30,6 +31,7 @@ import {
     pullStreamChunk,
     STREAM_CHUNK_OVERHEAD,
     STREAM_CHUNK_SIZE,
+    STREAM_TAG_FINAL,
 } from "../../src/crypto/index.js";
 
 describe("crypto stream constants", () => {
@@ -45,6 +47,22 @@ describe("crypto stream constants", () => {
     it("STREAM_CHUNK_OVERHEAD is 17 bytes", () => {
         expect(STREAM_CHUNK_OVERHEAD).toBe(17);
     });
+
+    /**
+     * `STREAM_TAG_FINAL` is re-exported so callers can detect a truncated
+     * stream (a body that ended on a non-final chunk) without importing
+     * libsodium themselves. It has to be declared as a literal, because
+     * libsodium only attaches its own constants to the module object after
+     * `sodium.ready` resolves — long after this library's modules are
+     * evaluated. This test is what keeps the literal honest.
+     */
+    it("STREAM_TAG_FINAL equals libsodium's TAG_FINAL", async () => {
+        await init();
+        await sodium.ready;
+        expect(STREAM_TAG_FINAL).toBe(
+            sodium.crypto_secretstream_xchacha20poly1305_TAG_FINAL,
+        );
+    });
 });
 
 describe("crypto.initStreamPull / pullStreamChunk", () => {

test/download/download.test.ts

@@ -9,20 +9,45 @@
  * secretstream ciphertext chunks. Each chunk is at most
  * `STREAM_CHUNK_SIZE + STREAM_CHUNK_OVERHEAD` bytes (4 MiB + 17 bytes).
  * The download function buffers incoming network data, splits it on the
- * chunk boundary, and feeds each piece to `pullStreamChunk`. The last
- * chunk carries `TAG_FINAL`; any truncation is detected because the tag
- * will be missing.
+ * chunk boundary, and feeds each piece to `pullStreamChunk`.
+ *
+ * Two contracts are load-bearing for anyone using this library as a backup
+ * tool, and both are documented by the tests below:
+ *
+ * 1. **Truncation is an error, never a short file.** Only the final chunk of
+ *    a secretstream carries `TAG_FINAL`. A download cut short by a dropped
+ *    connection still decrypts cleanly up to the last whole chunk, so without
+ *    an explicit `TAG_FINAL` check a truncated body is indistinguishable from
+ *    a complete one. `streamDecrypt` therefore refuses to return unless the
+ *    stream ended on `TAG_FINAL`, and the error says the stream was truncated.
+ *
+ * 2. **The destination path is written atomically.** Plaintext goes to a
+ *    temporary sibling file first and is `rename`d into place only after the
+ *    whole stream has decrypted and verified. A caller that sees no exception
+ *    can rely on the destination containing the complete, authenticated file;
+ *    a caller that sees an exception can rely on the destination being
+ *    untouched — whatever was there before is still there, byte for byte, and
+ *    no partial file has appeared. This matters because `runBackup` skips any
+ *    existing non-empty file, so a partial write would be treated as complete
+ *    forever after.
  *
  * These tests build synthetic encrypted files using sodium's push API,
  * serve them from a mock fetch, and verify the decrypted output on disk.
  */
 
-import { existsSync, readFileSync, rmSync, mkdtempSync } from "node:fs";
+import {
+    existsSync,
+    readdirSync,
+    readFileSync,
+    rmSync,
+    mkdtempSync,
+    writeFileSync,
+} from "node:fs";
 import { join } from "node:path";
 import { tmpdir } from "node:os";
 import sodium from "libsodium-wrappers-sumo";
 import { beforeAll, afterAll, describe, expect, it } from "vitest";
-import { init, toBase64 } from "../../src/crypto/index.js";
+import { init, toBase64, STREAM_CHUNK_SIZE } from "../../src/crypto/index.js";
 import { ApiClient } from "../../src/api/client.js";
 import { downloadFile, downloadThumbnail } from "../../src/download/index.js";
 import type { EnteFile, FileMetadata } from "../../src/model/types.js";
@@ -64,6 +89,79 @@ const encryptFileBody = (
     return { header: push.header, ciphertext };
 };
 
+/**
+ * Encrypt a body that spans more than one secretstream chunk, the way the
+ * server does for files larger than the 4 MiB plaintext chunk size.
+ *
+ * Framing matters here: the downloader splits the byte stream on fixed
+ * `STREAM_CHUNK_SIZE + STREAM_CHUNK_OVERHEAD` boundaries, so every chunk
+ * except the last must carry exactly `STREAM_CHUNK_SIZE` plaintext bytes.
+ * Only the last chunk is tagged `TAG_FINAL`; the leading ones are
+ * `TAG_MESSAGE`.
+ *
+ * Returns the header, the concatenated body, the plaintext it decrypts to,
+ * and `finalChunkOffset` — the byte offset at which the `TAG_FINAL` chunk
+ * begins, so a test can slice it off to simulate a connection that dropped
+ * before the end of the file.
+ */
+const encryptMultiChunkBody = (
+    key: Uint8Array,
+    leadingChunks: number,
+    finalChunkPlainSize: number,
+): {
+    header: Uint8Array;
+    body: Uint8Array;
+    plaintext: Uint8Array;
+    finalChunkOffset: number;
+} => {
+    const push = sodium.crypto_secretstream_xchacha20poly1305_init_push(key);
+    const cipherParts: Uint8Array[] = [];
+    const plainParts: Uint8Array[] = [];
+
+    for (let i = 0; i < leadingChunks; i++) {
+        const plain = sodium.randombytes_buf(STREAM_CHUNK_SIZE);
+        plainParts.push(plain);
+        cipherParts.push(
+            sodium.crypto_secretstream_xchacha20poly1305_push(
+                push.state,
+                plain,
+                null,
+                sodium.crypto_secretstream_xchacha20poly1305_TAG_MESSAGE,
+            ),
+        );
+    }
+
+    const finalPlain = sodium.randombytes_buf(finalChunkPlainSize);
+    plainParts.push(finalPlain);
+    const finalCipher = sodium.crypto_secretstream_xchacha20poly1305_push(
+        push.state,
+        finalPlain,
+        null,
+        sodium.crypto_secretstream_xchacha20poly1305_TAG_FINAL,
+    );
+
+    const finalChunkOffset = cipherParts.reduce((n, c) => n + c.length, 0);
+    cipherParts.push(finalCipher);
+
+    return {
+        header: push.header,
+        body: concat(cipherParts),
+        plaintext: concat(plainParts),
+        finalChunkOffset,
+    };
+};
+
+const concat = (parts: Uint8Array[]): Uint8Array => {
+    const total = parts.reduce((n, p) => n + p.length, 0);
+    const out = new Uint8Array(total);
+    let offset = 0;
+    for (const p of parts) {
+        out.set(p, offset);
+        offset += p.length;
+    }
+    return out;
+};
+
 const buildMockEnteFile = (
     key: Uint8Array,
     fileHeader: Uint8Array,
@@ -90,6 +188,41 @@ const mockFetchForBody = (body: Uint8Array) => {
     return fake as typeof globalThis.fetch;
 };
 
+/**
+ * A multi-chunk fixture shared by the truncation tests. Building it costs a
+ * few MiB of encryption, so it is built once: one full 4 MiB `TAG_MESSAGE`
+ * chunk followed by a small `TAG_FINAL` chunk.
+ */
+let multiChunk: ReturnType<typeof encryptMultiChunkBody>;
+let multiChunkKey: Uint8Array;
+
+beforeAll(() => {
+    multiChunkKey = sodium.crypto_secretstream_xchacha20poly1305_keygen();
+    multiChunk = encryptMultiChunkBody(multiChunkKey, 1, 1024);
+});
+
+/**
+ * Build an EnteFile plus ApiClient whose file *and* thumbnail streams both
+ * serve `body` under `header`. The download path under test is otherwise
+ * identical for the two, so every truncation/atomicity case below runs
+ * against both entry points from a single fixture.
+ */
+const fixtureFor = (
+    key: Uint8Array,
+    header: Uint8Array,
+    body: Uint8Array,
+): { api: ApiClient; file: EnteFile } => ({
+    api: new ApiClient({ fetch: mockFetchForBody(body) }),
+    file: buildMockEnteFile(key, header, header),
+});
+
+// The two entry points share `streamDecrypt` and the atomic-write wrapper,
+// so the contract tests are written once and run against both.
+const entryPoints = [
+    { name: "downloadFile", download: downloadFile },
+    { name: "downloadThumbnail", download: downloadThumbnail },
+];
+
 // ---------------------------------------------------------------------------
 // Tests
 // ---------------------------------------------------------------------------
@@ -113,8 +246,14 @@ describe("downloadFile", () => {
 
         const result = await downloadFile(api, file, outPath);
 
-        expect(result.path).toBe(outPath);
-        expect(result.bytesWritten).toBe(plaintext.length);
+        // The whole DownloadResult shape is asserted, not just its fields:
+        // callers depend on `path` being the destination they asked for
+        // (never the temporary file used along the way) and on
+        // `bytesWritten` being the plaintext length.
+        expect(result).toEqual({
+            path: outPath,
+            bytesWritten: plaintext.length,
+        });
         expect(readFileSync(outPath)).toEqual(Buffer.from(plaintext));
     });
 
@@ -131,6 +270,7 @@ describe("downloadFile", () => {
         const result = await downloadFile(api, file);
 
         expect(result.path).toBe("fallback-name.png");
+        expect(readFileSync(result.path)).toEqual(Buffer.from(plaintext));
         // Clean up since it writes to cwd
         if (existsSync(result.path)) rmSync(result.path);
     });
@@ -139,8 +279,6 @@ describe("downloadFile", () => {
         // Most photos are under 4 MiB and therefore a single secretstream
         // chunk. This test exercises a non-trivial payload size with
         // random binary data (not just ASCII) to verify no encoding bugs.
-        // Multi-chunk (>4 MiB) decryption is verified by the live
-        // integration test against real photos from the dev account.
         const plaintext = sodium.randombytes_buf(100_000);
         const key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
         const { header, ciphertext } = encryptFileBody(plaintext, key);
@@ -155,6 +293,28 @@ describe("downloadFile", () => {
         expect(result.bytesWritten).toBe(100_000);
         expect(readFileSync(outPath)).toEqual(Buffer.from(plaintext));
     });
+
+    it("decrypts a body that spans several secretstream chunks", async () => {
+        // Files over 4 MiB arrive as several ciphertext chunks concatenated
+        // into one HTTP body. The downloader has to re-split them on the
+        // exact chunk boundary; getting that wrong corrupts every large
+        // photo in an account. This is also the positive control for the
+        // truncation tests below: it proves the multi-chunk fixture itself
+        // decrypts cleanly when nothing has been removed from it.
+        const { api, file } = fixtureFor(
+            multiChunkKey,
+            multiChunk.header,
+            multiChunk.body,
+        );
+        const outPath = join(testDir, "multi-chunk.bin");
+
+        const result = await downloadFile(api, file, outPath);
+
+        expect(result.bytesWritten).toBe(multiChunk.plaintext.length);
+        expect(readFileSync(outPath)).toEqual(
+            Buffer.from(multiChunk.plaintext),
+        );
+    });
 });
 
 describe("downloadThumbnail", () => {
@@ -173,7 +333,172 @@ describe("downloadThumbnail", () => {
 
         const result = await downloadThumbnail(api, file, outPath);
 
-        expect(result.bytesWritten).toBe(4);
+        expect(result).toEqual({ path: outPath, bytesWritten: 4 });
         expect(readFileSync(outPath)).toEqual(Buffer.from(plaintext));
     });
 });
+
+// ---------------------------------------------------------------------------
+// Truncation detection and atomic writes
+//
+// Everything below is the failure contract. It is deliberately written once
+// per entry point via `entryPoints`, because `downloadFile` and
+// `downloadThumbnail` must behave identically here: a corrupt thumbnail is
+// just as unacceptable as a corrupt original, and `runBackup` trusts both.
+// ---------------------------------------------------------------------------
+
+describe.each(entryPoints)(
+    "$name truncation handling",
+    ({ name, download }) => {
+        /** A fresh, empty directory so leftover-file assertions are meaningful. */
+        const freshDir = (): string => {
+            const dir = mkdtempSync(join(testDir, `${name}-`));
+            return dir;
+        };
+
+        it("rejects a body whose final TAG_FINAL chunk never arrived", async () => {
+            // Simulate a connection that dropped after the first 4 MiB chunk.
+            // Every byte that did arrive decrypts and authenticates perfectly —
+            // that is precisely the danger. The only signal that the file is
+            // incomplete is the absence of a chunk tagged TAG_FINAL, so the
+            // downloader must treat "stream ended on TAG_MESSAGE" as a hard
+            // error rather than returning a short file.
+            const truncatedBody = multiChunk.body.slice(
+                0,
+                multiChunk.finalChunkOffset,
+            );
+            const { api, file } = fixtureFor(
+                multiChunkKey,
+                multiChunk.header,
+                truncatedBody,
+            );
+            const outPath = join(freshDir(), "truncated.bin");
+
+            await expect(download(api, file, outPath)).rejects.toThrow(
+                /truncated/i,
+            );
+        });
+
+        it("rejects an empty body instead of writing a zero-byte file", async () => {
+            // Ente always emits at least one chunk, even for empty content:
+            // `encryptBlob` shows that a zero-length plaintext still produces a
+            // TAG_FINAL chunk. A body with no chunks at all therefore means the
+            // transfer failed, not that the file is empty. Writing a zero-byte
+            // file here would be the worst outcome, because `runBackup` would
+            // then see a file it considers present and never retry it.
+            const { api, file } = fixtureFor(
+                multiChunkKey,
+                multiChunk.header,
+                new Uint8Array(0),
+            );
+            const outPath = join(freshDir(), "empty.bin");
+
+            await expect(download(api, file, outPath)).rejects.toThrow(
+                /truncated/i,
+            );
+        });
+
+        it("leaves no file at the destination after a truncated download", async () => {
+            // The caller's contract: if the promise rejects, the destination
+            // path does not exist. Nothing downstream should ever have to guess
+            // whether a leftover file is complete.
+            const truncatedBody = multiChunk.body.slice(
+                0,
+                multiChunk.finalChunkOffset,
+            );
+            const { api, file } = fixtureFor(
+                multiChunkKey,
+                multiChunk.header,
+                truncatedBody,
+            );
+            const dir = freshDir();
+            const outPath = join(dir, "absent.bin");
+
+            await expect(download(api, file, outPath)).rejects.toThrow(
+                /truncated/i,
+            );
+
+            expect(existsSync(outPath)).toBe(false);
+            // And no temporary scratch file is left behind either: the download
+            // stages plaintext in a sibling temp file, which must be removed on
+            // the failure path so repeated failures cannot fill the disk.
+            expect(readdirSync(dir)).toEqual([]);
+        });
+
+        it("leaves no file at the destination when a chunk fails authentication", async () => {
+            // The same guarantee has to hold for every failure mode, not just
+            // truncation. Here a byte of ciphertext is flipped, so Poly1305
+            // verification fails inside `pullStreamChunk`. The error must
+            // propagate unchanged (it is the real diagnosis) and the destination
+            // must still be untouched.
+            const plaintext = sodium.randombytes_buf(256);
+            const key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
+            const { header, ciphertext } = encryptFileBody(plaintext, key);
+            const corrupted = Uint8Array.from(ciphertext);
+            corrupted[10] ^= 0xff;
+
+            const { api, file } = fixtureFor(key, header, corrupted);
+            const dir = freshDir();
+            const outPath = join(dir, "corrupt.bin");
+
+            await expect(download(api, file, outPath)).rejects.toThrow(
+                /authentication failed/i,
+            );
+
+            expect(existsSync(outPath)).toBe(false);
+            expect(readdirSync(dir)).toEqual([]);
+        });
+
+        it("does not clobber an existing file when the download fails", async () => {
+            // The repair case. A user re-running a backup over a directory that
+            // already holds good originals must never end up worse off: a failed
+            // download leaves the previous contents exactly as they were, so the
+            // old good copy survives until a complete new one is available to
+            // replace it in a single rename.
+            const existing = new TextEncoder().encode(
+                "previously downloaded, known-good contents",
+            );
+            const truncatedBody = multiChunk.body.slice(
+                0,
+                multiChunk.finalChunkOffset,
+            );
+            const { api, file } = fixtureFor(
+                multiChunkKey,
+                multiChunk.header,
+                truncatedBody,
+            );
+            const dir = freshDir();
+            const outPath = join(dir, "existing.bin");
+            writeFileSync(outPath, existing);
+
+            await expect(download(api, file, outPath)).rejects.toThrow(
+                /truncated/i,
+            );
+
+            expect(readFileSync(outPath)).toEqual(Buffer.from(existing));
+            expect(readdirSync(dir)).toEqual(["existing.bin"]);
+        });
+
+        it("replaces an existing file when the download succeeds", async () => {
+            // The mirror image of the previous test: a complete download does
+            // overwrite whatever was at the destination, atomically, via rename.
+            const existing = new TextEncoder().encode("stale contents");
+            const plaintext = sodium.randombytes_buf(512);
+            const key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
+            const { header, ciphertext } = encryptFileBody(plaintext, key);
+
+            const { api, file } = fixtureFor(key, header, ciphertext);
+            const dir = freshDir();
+            const outPath = join(dir, "replaced.bin");
+            writeFileSync(outPath, existing);
+
+            const result = await download(api, file, outPath);
+
+            expect(result).toEqual({ path: outPath, bytesWritten: 512 });
+            expect(readFileSync(outPath)).toEqual(Buffer.from(plaintext));
+            // The temp file is gone once the rename has happened, so a
+            // successful download leaves exactly one file behind.
+            expect(readdirSync(dir)).toEqual(["replaced.bin"]);
+        });
+    },
+);