Verify secretstream TAG_FINAL and write downloads atomically (closes #1) #20

Merged
clawbot merged 4 commits from download-tag-final-atomic-write into main 2026-08-09 04:59:44 +02:00
6 changed files with 790 additions and 33 deletions

View File

@@ -20,6 +20,10 @@ downloads, cover it with mock-server tests, and update the README TODO checkbox.
# Completed Steps
- 2026-08-09: Downloads verify the secretstream terminated on `TAG_FINAL` and
write output atomically: a truncated body is rejected instead of landing on
disk as a short file, and plaintext is staged in a sibling temp file and
renamed into place, so a failed download leaves the destination untouched.
- 2026-07-07 Adopted scripts-to-rule-them-all: `script/` entrypoints, Makefile
shims, README Entrypoints section
- 2026-06-10: Decrypted collections shared by other users (sealed-box keys);

View File

@@ -14,5 +14,6 @@ export {
pullStreamChunk,
STREAM_CHUNK_OVERHEAD,
STREAM_CHUNK_SIZE,
streamTagFinal,
type StreamPullState,
} from "./stream.js";

View File

@@ -8,6 +8,27 @@ export const STREAM_CHUNK_SIZE = 4 * 1024 * 1024;
// 16 bytes of Poly1305 tag plus 1 byte of secretstream tag.
export const STREAM_CHUNK_OVERHEAD = 17;
// libsodium's crypto_secretstream_xchacha20poly1305_TAG_FINAL: the tag that
// marks the last chunk of a stream. Exported so callers (the download layer)
// can detect truncation without importing sodium themselves.
//
// This is a function rather than a constant, and that is load-bearing:
// libsodium attaches its constants to the module object inside
// `ready.then(...)`, which resolves long after this module is evaluated. A
// module-level read would bind `undefined`, every downstream comparison
// against it would then be false, and every valid download would be rejected
// as truncated. Reading at call time returns the library's own value, so
// there is also no second copy of a protocol constant to keep in sync.
//
// Under vitest sodium is already initialised in the worker before this module
// is evaluated, so an eager read would pick up a real value there and the
// ordinary tests could not tell the difference. The regression guard is
// "streamTagFinal() reads the constant at call time, not at import time" in
// test/crypto/stream.test.ts, which reproduces the plain-Node ESM ordering
// against a stand-in sodium module; it goes red if this becomes eager.
export const streamTagFinal = (): number =>
sodium.crypto_secretstream_xchacha20poly1305_TAG_FINAL;
// Encrypt a small blob as a single secretstream chunk with TAG_FINAL.
// Returns the header and ciphertext. Used for encrypting thumbnails
// and metadata before upload.
@@ -37,9 +58,6 @@ export const initStreamPull = (
): StreamPullState =>
sodium.crypto_secretstream_xchacha20poly1305_init_pull(header, key);
// Decrypt one ciphertext chunk. Returns the plaintext and the secretstream
// tag (0=MESSAGE, 1=PUSH, 2=REKEY, 3=FINAL). The caller should verify the
// stream ended on TAG_FINAL to detect truncation.
// Decrypt a small blob that was encrypted as a single secretstream chunk
// with TAG_FINAL. Ente uses this form ("blob") for file metadata and
// magic metadata — anything under ~1 MiB that isn't chunked.
@@ -50,12 +68,18 @@ export const decryptBlob = (
): Uint8Array => {
const state = initStreamPull(header, key);
const { plaintext, tag } = pullStreamChunk(state, ciphertext);
if (tag !== sodium.crypto_secretstream_xchacha20poly1305_TAG_FINAL) {
throw new Error(`decryptBlob: expected TAG_FINAL (3), got tag ${tag}`);
const tagFinal = streamTagFinal();
if (tag !== tagFinal) {
throw new Error(
`decryptBlob: expected TAG_FINAL (${tagFinal}), got tag ${tag}`,
);
}
return plaintext;
};
// Decrypt one ciphertext chunk. Returns the plaintext and the secretstream
// tag (0=MESSAGE, 1=PUSH, 2=REKEY, 3=FINAL). The caller must verify the
// stream ended on TAG_FINAL to detect truncation.
export const pullStreamChunk = (
state: StreamPullState,
ciphertext: Uint8Array,

View File

@@ -1,10 +1,13 @@
import { writeFile } from "node:fs/promises";
import { randomUUID } from "node:crypto";
import { rename, rm, writeFile } from "node:fs/promises";
import { dirname, join } from "node:path";
import {
fromBase64,
initStreamPull,
pullStreamChunk,
STREAM_CHUNK_OVERHEAD,
STREAM_CHUNK_SIZE,
streamTagFinal,
} from "../crypto/index.js";
import type { ApiClient } from "../api/client.js";
import type { EnteFile } from "../model/types.js";
@@ -26,6 +29,8 @@ const streamDecrypt = async (
let buffer = new Uint8Array(0);
const plainChunks: Uint8Array[] = [];
let totalPlain = 0;
let chunksPulled = 0;
let lastTag = -1;
for (;;) {
const { done, value } = await reader.read();
@@ -39,21 +44,58 @@ const streamDecrypt = async (
while (buffer.length >= ENC_CHUNK_SIZE) {
const encChunk = buffer.slice(0, ENC_CHUNK_SIZE);
buffer = buffer.slice(ENC_CHUNK_SIZE);
const { plaintext } = pullStreamChunk(state, encChunk);
const { plaintext, tag } = pullStreamChunk(state, encChunk);
plainChunks.push(plaintext);
totalPlain += plaintext.length;
chunksPulled++;
lastTag = tag;
}
if (done) {
if (buffer.length > 0) {
const { plaintext } = pullStreamChunk(state, buffer);
plainChunks.push(plaintext);
totalPlain += plaintext.length;
// Whatever is left over once every whole chunk has been
// consumed must be the stream's final chunk, and a final
// chunk that actually arrived in full authenticates. If it
// does not, the body stopped part-way through a chunk — the
// ordinary shape of a dropped connection. Poly1305 cannot
// tell a partial chunk from a corrupt one, so this is
// reported as the truncation it almost always is, with the
// authentication failure kept as the error's cause.
let pulled;
try {
pulled = pullStreamChunk(state, buffer);
} catch (err) {
throw new Error(
`download: stream truncated: response body ended with ${buffer.length} trailing bytes that did not authenticate as a final chunk (transfer stopped mid-chunk, or the data is corrupt)`,
{ cause: err },
);
}
plainChunks.push(pulled.plaintext);
totalPlain += pulled.plaintext.length;
chunksPulled++;
lastTag = pulled.tag;
}
break;
}
}
// Only the last chunk of a secretstream carries TAG_FINAL. Everything a
// dropped connection did deliver still decrypts and authenticates, so the
// absence of TAG_FINAL is the only evidence that the body was cut short.
// Returning a short plaintext here would put a corrupt file on disk that
// later backup runs would treat as complete.
if (chunksPulled === 0) {
throw new Error(
"download: stream truncated: response body contained no secretstream chunks",
);
}
const tagFinal = streamTagFinal();
if (lastTag !== tagFinal) {
throw new Error(
`download: stream truncated: last chunk tag ${lastTag}, expected TAG_FINAL (${tagFinal})`,
);
}
const result = new Uint8Array(totalPlain);
let offset = 0;
for (const chunk of plainChunks) {
@@ -63,6 +105,29 @@ const streamDecrypt = async (
return result;
};
// Write `plaintext` to `destination` atomically: stage it in a temporary
// sibling file (same directory, so the rename cannot cross a filesystem
// boundary) and rename it into place. Callers therefore never observe a
// partially written destination, and a pre-existing file at that path is
// replaced only once the new contents are complete on disk.
const writeAtomic = async (
destination: string,
plaintext: Uint8Array,
): Promise<void> => {
// The random suffix keeps concurrent downloads of the same destination
// from stepping on each other's temporary file.
const tmpPath = join(dirname(destination), `.quak-${randomUUID()}.tmp`);
try {
await writeFile(tmpPath, plaintext);
await rename(tmpPath, destination);
} catch (err) {
// Best-effort cleanup. A failure to remove the temporary file must
// never replace the error that actually explains what went wrong.
await rm(tmpPath, { force: true }).catch(() => undefined);
throw err;
}
};
export const downloadFile = async (
api: ApiClient,
file: EnteFile,
@@ -72,7 +137,7 @@ export const downloadFile = async (
const stream = await api.getFileStream(file.id);
const header = fromBase64(file.file.decryptionHeader);
const plaintext = await streamDecrypt(stream, header, file.key);
await writeFile(resolvedPath, plaintext);
await writeAtomic(resolvedPath, plaintext);
return { path: resolvedPath, bytesWritten: plaintext.length };
};
@@ -85,6 +150,6 @@ export const downloadThumbnail = async (
const stream = await api.getThumbnailStream(file.id);
const header = fromBase64(file.thumbnail.decryptionHeader);
const plaintext = await streamDecrypt(stream, header, file.key);
await writeFile(resolvedPath, plaintext);
await writeAtomic(resolvedPath, plaintext);
return { path: resolvedPath, bytesWritten: plaintext.length };
};

View File

@@ -15,7 +15,12 @@
* 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 `streamTagFinal()` matches the tag a real final chunk
* carries.
* - `streamTagFinal()` reads libsodium's constant at call time rather than
* at import time, which it must, because the constant does not exist yet
* when this library's modules are evaluated.
* - 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.
@@ -23,13 +28,14 @@
*/
import sodium from "libsodium-wrappers-sumo";
import { beforeAll, describe, expect, it } from "vitest";
import { beforeAll, describe, expect, it, vi } from "vitest";
import {
init,
initStreamPull,
pullStreamChunk,
STREAM_CHUNK_OVERHEAD,
STREAM_CHUNK_SIZE,
streamTagFinal,
} from "../../src/crypto/index.js";
describe("crypto stream constants", () => {
@@ -45,6 +51,85 @@ describe("crypto stream constants", () => {
it("STREAM_CHUNK_OVERHEAD is 17 bytes", () => {
expect(STREAM_CHUNK_OVERHEAD).toBe(17);
});
/**
* `streamTagFinal()` is re-exported so callers can detect a truncated
* stream (a body that ended on a non-final chunk) without importing
* libsodium themselves.
*
* It is a function, not a constant, and that is load-bearing: libsodium
* attaches its own constants to the module object only after
* `sodium.ready` resolves, which is long after this library's modules are
* evaluated. Reading the value at call time yields libsodium's number;
* reading it at module scope would yield `undefined`, and every
* truncation check downstream would then compare against `undefined` and
* reject good downloads. The eagerness itself is guarded by the next test;
* this one pins the value, comparing it against the tag observed on a real
* final chunk pulled back off the wire format.
*/
it("streamTagFinal() is the tag carried by a real final chunk", async () => {
await init();
await sodium.ready;
const key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
const push =
sodium.crypto_secretstream_xchacha20poly1305_init_push(key);
const ciphertext = sodium.crypto_secretstream_xchacha20poly1305_push(
push.state,
new TextEncoder().encode("last chunk"),
null,
sodium.crypto_secretstream_xchacha20poly1305_TAG_FINAL,
);
const state = initStreamPull(push.header, key);
expect(pullStreamChunk(state, ciphertext).tag).toBe(streamTagFinal());
});
/**
* The regression guard for the eagerness property described above.
*
* Under vitest, sodium is already initialised in the worker process by
* the time any source module is evaluated, so an eager module-level read
* would happen to pick up a real value and no ordinary test could tell
* the difference. This test recreates the ordering that a plain Node ESM
* consumer sees: a stand-in sodium module whose `TAG_FINAL` property does
* not exist yet when `src/crypto/stream.ts` is evaluated and only appears
* afterwards, exactly as libsodium attaches its constants inside
* `ready.then(...)`.
*
* A call-time read observes the value that appeared after evaluation; a
* module-level read binds `undefined` and this test fails. The stand-in
* uses a sentinel rather than the real tag number so that a read which
* somehow reached the real libsodium would fail too.
*/
it("streamTagFinal() reads the constant at call time, not at import time", async () => {
const SENTINEL = 42;
const late: { tagFinal: number | undefined } = { tagFinal: undefined };
vi.resetModules();
vi.doMock("libsodium-wrappers-sumo", () => ({
default: {
ready: Promise.resolve(),
get crypto_secretstream_xchacha20poly1305_TAG_FINAL() {
return late.tagFinal;
},
},
}));
try {
// Evaluated while the constant is still absent, as it is before
// `sodium.ready` resolves.
const fresh = await import("../../src/crypto/stream.js");
expect(late.tagFinal).toBeUndefined();
// libsodium attaches its constants; a lazy accessor sees them.
late.tagFinal = SENTINEL;
expect(fresh.streamTagFinal()).toBe(SENTINEL);
} finally {
vi.doUnmock("libsodium-wrappers-sumo");
vi.resetModules();
}
});
});
describe("crypto.initStreamPull / pullStreamChunk", () => {

View File

@@ -9,20 +9,57 @@
* 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.
* A transfer that stopped part-way through a chunk is reported the same
* way, since a final chunk that arrived in full always authenticates.
*
* 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. The staging file and the rename are observed directly (see
* the `rename` hook below), not inferred from an empty directory.
*
* 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 { join } from "node:path";
import {
existsSync,
readdirSync,
readFileSync,
rmSync,
mkdtempSync,
writeFileSync,
} from "node:fs";
import { dirname, join } from "node:path";
import { tmpdir } from "node:os";
import { createHash } from "node:crypto";
import sodium from "libsodium-wrappers-sumo";
import { beforeAll, afterAll, describe, expect, it } from "vitest";
import { init, toBase64 } from "../../src/crypto/index.js";
import {
beforeAll,
beforeEach,
afterAll,
describe,
expect,
it,
vi,
} from "vitest";
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";
@@ -31,6 +68,52 @@ import type { EnteFile, FileMetadata } from "../../src/model/types.js";
// Test helpers
// ---------------------------------------------------------------------------
/**
* `rename` is intercepted so that the tests can observe — and fail — the
* final step of the atomic write.
*
* Every other failure in this file is injected inside `streamDecrypt`, which
* runs before anything is written to disk. Those tests therefore cannot tell
* an atomic write from a plain `writeFile`: under both, nothing was ever
* created, so an empty directory proves nothing about cleanup. This hook is
* what closes that gap. It records each rename the downloader performs,
* including whether the source existed at that moment (i.e. that the staged
* temp file really was written), and can be told to fail the rename so the
* cleanup path runs with a temp file genuinely on disk.
*
* `vi.hoisted` is required: `vi.mock` factories are hoisted above the imports,
* so a plain module-level `const` would still be in its temporal dead zone by
* the time the factory runs.
*/
const renameHook = vi.hoisted(() => ({
calls: [] as { from: string; to: string; sourceExisted: boolean }[],
failWith: null as Error | null,
}));
vi.mock("node:fs/promises", async (importOriginal) => {
const actual = await importOriginal<typeof import("node:fs/promises")>();
const { existsSync: sourceExists } = await import("node:fs");
return {
...actual,
rename: async (from: string, to: string): Promise<void> => {
renameHook.calls.push({
from,
to,
sourceExisted: sourceExists(from),
});
if (renameHook.failWith !== null) {
throw renameHook.failWith;
}
await actual.rename(from, to);
},
};
});
beforeEach(() => {
renameHook.calls.length = 0;
renameHook.failWith = null;
});
let testDir: string;
beforeAll(async () => {
@@ -45,6 +128,35 @@ afterAll(() => {
}
});
/**
* Deterministic stand-in for random test payloads.
*
* Fixture *content* is never load-bearing here — the assertions turn on
* length, framing, and the secretstream tag — but it must not be a constant
* fill either, or a downloader that reordered or repeated chunks would still
* produce the expected bytes. A seeded linear congruential generator gives
* both: byte patterns that differ across every offset and seed, reproducible
* on any machine, which is what the README asks of fixtures.
*
* It is also the difference between a fast suite and a broken one.
* `sodium.randombytes_buf` goes through the wasm wrapper a byte at a time and
* costs roughly 20 seconds for the 4 MiB chunk below — about two hundred
* times what it costs to encrypt the same buffer, and on its own enough to
* push `make test` past the 30-second cap in `script/test`. This loop fills
* 4 MiB in a few milliseconds.
*/
const patternBytes = (length: number, seed: number): Uint8Array => {
const out = new Uint8Array(length);
let x = seed >>> 0;
for (let i = 0; i < length; i++) {
// Numerical Recipes' LCG constants; the high byte is used because
// the low bits of an LCG have short periods.
x = (Math.imul(x, 1664525) + 1013904223) >>> 0;
out[i] = (x >>> 24) & 0xff;
}
return out;
};
/**
* Encrypt `plaintext` as a secretstream file body (single chunk with
* TAG_FINAL). Returns the key, header, and ciphertext that the mock CDN
@@ -64,6 +176,79 @@ 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 = patternBytes(STREAM_CHUNK_SIZE, i + 1);
plainParts.push(plain);
cipherParts.push(
sodium.crypto_secretstream_xchacha20poly1305_push(
push.state,
plain,
null,
sodium.crypto_secretstream_xchacha20poly1305_TAG_MESSAGE,
),
);
}
const finalPlain = patternBytes(finalChunkPlainSize, leadingChunks + 1);
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 +275,79 @@ const mockFetchForBody = (body: Uint8Array) => {
return fake as typeof globalThis.fetch;
};
/**
* Encrypt a body consisting of one chunk that is *not* tagged TAG_FINAL.
*
* This is the cheap way to present a stream that ended without its final
* chunk: the downloader pulls it, authenticates it, and finds the stream
* over on a TAG_MESSAGE chunk — the same terminal condition as a large file
* whose last chunk was lost, without paying for a 4 MiB fixture. The
* multi-chunk fixture above covers the realistic wire shape; this one is
* used where the test is really about what happens on disk afterwards.
*/
const encryptNonFinalBody = (
plaintext: Uint8Array,
key: Uint8Array,
): { header: Uint8Array; ciphertext: Uint8Array } => {
const push = sodium.crypto_secretstream_xchacha20poly1305_init_push(key);
const ciphertext = sodium.crypto_secretstream_xchacha20poly1305_push(
push.state,
plaintext,
null,
sodium.crypto_secretstream_xchacha20poly1305_TAG_MESSAGE,
);
return { header: push.header, ciphertext };
};
/**
* Compare file contents by digest rather than with `toEqual`. Vitest's deep
* equality walks multi-megabyte buffers byte by byte, which costs seconds on
* the 4 MiB fixtures; a digest comparison is exact and effectively free.
*/
const expectSameBytes = (actual: Uint8Array, expected: Uint8Array): void => {
expect(actual.length).toBe(expected.length);
expect(createHash("sha256").update(actual).digest("hex")).toBe(
createHash("sha256").update(expected).digest("hex"),
);
};
/**
* A multi-chunk fixture shared by the truncation tests: one full 4 MiB
* `TAG_MESSAGE` chunk followed by a small `TAG_FINAL` chunk. It is built once
* and shared because encrypting 4 MiB costs about 100ms. Its plaintext is
* generated rather than drawn from the CSPRNG, which is what keeps that
* encryption the whole cost of the fixture.
*/
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
// ---------------------------------------------------------------------------
@@ -113,35 +371,45 @@ 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));
});
it("uses metadata.title as filename when outPath is omitted", async () => {
// With no `outPath`, the destination is `metadata.title`, used
// verbatim as a path. The title here is therefore given inside the
// test's temporary directory: a bare relative name would resolve
// against the process working directory, i.e. the repo root, and
// `make check` must not create files in the repo — a failure between
// the write and any cleanup would leave one behind.
const plaintext = new Uint8Array([1, 2, 3]);
const key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
const { header, ciphertext } = encryptFileBody(plaintext, key);
const thumbPush =
sodium.crypto_secretstream_xchacha20poly1305_init_push(key);
const file = buildMockEnteFile(key, header, thumbPush.header);
file.metadata.title = "fallback-name.png";
const titlePath = join(testDir, "fallback-name.png");
file.metadata.title = titlePath;
const api = new ApiClient({ fetch: mockFetchForBody(ciphertext) });
const result = await downloadFile(api, file);
expect(result.path).toBe("fallback-name.png");
// Clean up since it writes to cwd
if (existsSync(result.path)) rmSync(result.path);
expect(result.path).toBe(titlePath);
expect(readFileSync(result.path)).toEqual(Buffer.from(plaintext));
});
it("handles a larger single-chunk file (random binary payload)", async () => {
// 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);
// arbitrary binary data (not just ASCII) to verify no encoding bugs.
const plaintext = patternBytes(100_000, 11);
const key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
const { header, ciphertext } = encryptFileBody(plaintext, key);
const thumbPush =
@@ -153,7 +421,27 @@ describe("downloadFile", () => {
const result = await downloadFile(api, file, outPath);
expect(result.bytesWritten).toBe(100_000);
expect(readFileSync(outPath)).toEqual(Buffer.from(plaintext));
expectSameBytes(readFileSync(outPath), 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);
expectSameBytes(readFileSync(outPath), multiChunk.plaintext);
});
});
@@ -173,7 +461,297 @@ 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 a body whose final chunk arrived only in part", async () => {
// The likelier shape of a dropped connection: the transfer stops
// in the middle of a chunk rather than neatly between two. The
// bytes that arrived are a prefix of a complete chunk, so Poly1305
// rejects them — which is, cryptographically, indistinguishable
// from corruption of a whole chunk.
//
// It is still reported as truncation, because that is what it
// almost always is and because this library's entire reason for
// checking TAG_FINAL is to make a short transfer visible. Calling
// a short transfer "authentication failed" would send a user
// hunting for a corrupt file when their network is at fault. The
// underlying authentication failure is kept as the error's
// `cause`, so the real diagnosis is never lost.
const shortBody = multiChunk.body.slice(
0,
multiChunk.body.length - 8,
);
const { api, file } = fixtureFor(
multiChunkKey,
multiChunk.header,
shortBody,
);
const dir = freshDir();
const outPath = join(dir, "partial-final.bin");
const err = await download(api, file, outPath).catch(
(e: unknown) => e,
);
expect(err).toBeInstanceOf(Error);
expect((err as Error).message).toMatch(/truncated/i);
expect((err as Error).cause).toBeInstanceOf(Error);
expect(((err as Error).cause as Error).message).toMatch(
/authentication failed/i,
);
// And, as with every other failure, the destination is untouched
// and no staged temp file survives.
expect(existsSync(outPath)).toBe(false);
expect(readdirSync(dir)).toEqual([]);
});
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.
//
// The body here is a single chunk that was never tagged TAG_FINAL,
// which puts the downloader in exactly the terminal state a lost
// last chunk produces, without the cost of a 4 MiB fixture. What
// this test is really about is the state of the filesystem after
// the rejection.
const key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
const { header, ciphertext } = encryptNonFinalBody(
patternBytes(256, 21),
key,
);
const { api, file } = fixtureFor(key, header, ciphertext);
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("reports a corrupt whole chunk as an authentication failure, not truncation", async () => {
// The counterpart to the partial-final-chunk case above, and the
// reason the two are distinguishable at all. A byte is flipped
// inside the first chunk of a multi-chunk body: that chunk arrives
// complete — the stream goes on past it — so its failure cannot be
// a short transfer. It is corruption, and the caller is told so,
// with `pullStreamChunk`'s error propagated unchanged because it is
// the real diagnosis.
//
// The same on-disk guarantee holds for this failure mode as for
// every other: nothing at the destination, nothing left over.
const corrupted = Uint8Array.from(multiChunk.body);
corrupted[10] ^= 0xff;
const { api, file } = fixtureFor(
multiChunkKey,
multiChunk.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 key = sodium.crypto_secretstream_xchacha20poly1305_keygen();
const { header, ciphertext } = encryptNonFinalBody(
patternBytes(256, 22),
key,
);
const { api, file } = fixtureFor(key, header, ciphertext);
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 = patternBytes(512, 23);
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"]);
});
it("stages the plaintext in a sibling temp file and renames it into place", async () => {
// The atomic write, observed directly rather than inferred from an
// empty directory. Every other failure in this file is injected
// inside `streamDecrypt`, which runs before anything is written —
// so under those tests a plain `writeFile` to the destination would
// look identical. This one watches the rename itself.
//
// Two properties are load-bearing. The staging file must exist on
// disk when the rename happens: that is what makes the destination
// appear complete or not at all, instead of filling up as bytes
// land. And it must be a sibling of the destination, because
// `rename` is only atomic within one filesystem — staging in
// `/tmp` and renaming across a mount point would silently become a
// copy, reintroducing the partial-file window this exists to close.
const plaintext = patternBytes(512, 31);
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, "staged.bin");
await download(api, file, outPath);
expect(renameHook.calls).toHaveLength(1);
const staged = renameHook.calls[0]!;
expect(staged.to).toBe(outPath);
expect(staged.from).not.toBe(outPath);
expect(dirname(staged.from)).toBe(dir);
expect(staged.sourceExisted).toBe(true);
// Afterwards the temp file is gone and only the destination is
// left, holding the complete plaintext.
expect(existsSync(staged.from)).toBe(false);
expect(readFileSync(outPath)).toEqual(Buffer.from(plaintext));
expect(readdirSync(dir)).toEqual(["staged.bin"]);
});
it("removes the staged temp file when the rename itself fails", async () => {
// The cleanup path. It can only run when something fails at or
// after the write, which no amount of bad network data can
// produce: by the time anything is written the whole stream has
// already decrypted and verified. Failing the rename is what
// reaches it — a real possibility on a full disk, a read-only
// mount, or a destination that has become a directory.
//
// This is the only case in which the temp file is on disk at the
// moment of failure, so it is the only one that can show it is
// actually removed rather than merely never created. It also pins
// that the caller sees the original failure: a cleanup that threw
// over the top of it would hide why the download failed.
const existing = new TextEncoder().encode("known-good contents");
const plaintext = patternBytes(512, 32);
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, "rename-fails.bin");
writeFileSync(outPath, existing);
renameHook.failWith = new Error("simulated rename failure");
await expect(download(api, file, outPath)).rejects.toThrow(
"simulated rename failure",
);
expect(renameHook.calls).toHaveLength(1);
const staged = renameHook.calls[0]!;
expect(staged.sourceExisted).toBe(true);
expect(existsSync(staged.from)).toBe(false);
// The previous contents are still there, untouched, and the
// directory holds nothing else.
expect(readFileSync(outPath)).toEqual(Buffer.from(existing));
expect(readdirSync(dir)).toEqual(["rename-fails.bin"]);
});
},
);