Phase 0: skeleton, streaming encryption (spike A)

Repository skeleton (app id ng_backup, AGPL-3.0-or-later), KeyRing with KDF subkeys, keyed blob ids and Argon2id key wrapping, and StreamCipher (secretstream XChaCha20-Poly1305, framed, bounded memory). Spike A: 1 GiB round trip with memory_limit=64M, peak < 2 MiB, ~160/122 MiB/s; tampering, truncation, wrong key/passphrase are rejected.
2026-10-01 18:15:24 +02:00
commit fee189aec3
11 changed files with 1421 additions and 0 deletions
@@ -0,0 +1,11 @@
+<?php
+
+declare(strict_types=1);
+
+// SPDX-FileCopyrightText: 2026 Dillard Blom
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+namespace OCA\NgBackup\Crypto;
+
+class CryptoException extends \RuntimeException {
+}
@@ -0,0 +1,92 @@
+<?php
+
+declare(strict_types=1);
+
+// SPDX-FileCopyrightText: 2026 Dillard Blom
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+namespace OCA\NgBackup\Crypto;
+
+/**
+ * Holds the repository master key and derives the subkeys used for encryption and for the
+ * keyed blob hashes. The master key never leaves this object unencrypted, except through
+ * wrap(), which seals it with a key derived from the admin's passphrase.
+ */
+final class KeyRing {
+	private const KDF_CONTEXT = 'ngbackup';
+	private const SUBKEY_DATA = 1;
+	private const SUBKEY_HASH = 2;
+
+	private string $dataKey;
+	private string $hashKey;
+
+	public function __construct(
+		#[\SensitiveParameter]
+		private string $masterKey,
+	) {
+		if (strlen($masterKey) !== SODIUM_CRYPTO_KDF_KEYBYTES) {
+			throw new \InvalidArgumentException('Master key must be ' . SODIUM_CRYPTO_KDF_KEYBYTES . ' bytes');
+		}
+		$this->dataKey = sodium_crypto_kdf_derive_from_key(
+			SODIUM_CRYPTO_SECRETSTREAM_XCHACHA20POLY1305_KEYBYTES, self::SUBKEY_DATA, self::KDF_CONTEXT, $masterKey);
+		$this->hashKey = sodium_crypto_kdf_derive_from_key(
+			SODIUM_CRYPTO_GENERICHASH_KEYBYTES, self::SUBKEY_HASH, self::KDF_CONTEXT, $masterKey);
+	}
+
+	public static function generate(): self {
+		return new self(sodium_crypto_kdf_keygen());
+	}
+
+	public function dataKey(): string {
+		return $this->dataKey;
+	}
+
+	/** Keyed BLAKE2b of plaintext, so blob ids reveal nothing about the content. */
+	public function blobId(string $plaintext): string {
+		return sodium_bin2hex(sodium_crypto_generichash($plaintext, $this->hashKey, 32));
+	}
+
+	/**
+	 * Seal the master key with a passphrase (Argon2id). The result is what goes into the
+	 * repository config and the recovery kit.
+	 *
+	 * @return array{v:int, salt:string, ops:int, mem:int, nonce:string, key:string}
+	 */
+	public function wrap(#[\SensitiveParameter] string $passphrase): array {
+		$salt = random_bytes(SODIUM_CRYPTO_PWHASH_SALTBYTES);
+		$ops = SODIUM_CRYPTO_PWHASH_OPSLIMIT_MODERATE;
+		$mem = SODIUM_CRYPTO_PWHASH_MEMLIMIT_MODERATE;
+		$kek = sodium_crypto_pwhash(SODIUM_CRYPTO_SECRETBOX_KEYBYTES, $passphrase, $salt, $ops, $mem,
+			SODIUM_CRYPTO_PWHASH_ALG_ARGON2ID13);
+		$nonce = random_bytes(SODIUM_CRYPTO_SECRETBOX_NONCEBYTES);
+		$sealed = sodium_crypto_secretbox($this->masterKey, $nonce, $kek);
+		sodium_memzero($kek);
+		return [
+			'v' => 1,
+			'salt' => sodium_bin2base64($salt, SODIUM_BASE64_VARIANT_ORIGINAL),
+			'ops' => $ops,
+			'mem' => $mem,
+			'nonce' => sodium_bin2base64($nonce, SODIUM_BASE64_VARIANT_ORIGINAL),
+			'key' => sodium_bin2base64($sealed, SODIUM_BASE64_VARIANT_ORIGINAL),
+		];
+	}
+
+	/** @param array{v:int, salt:string, ops:int, mem:int, nonce:string, key:string} $wrapped */
+	public static function unwrap(array $wrapped, #[\SensitiveParameter] string $passphrase): self {
+		$b64 = static fn (string $s): string => sodium_base642bin($s, SODIUM_BASE64_VARIANT_ORIGINAL);
+		$kek = sodium_crypto_pwhash(SODIUM_CRYPTO_SECRETBOX_KEYBYTES, $passphrase, $b64($wrapped['salt']),
+			$wrapped['ops'], $wrapped['mem'], SODIUM_CRYPTO_PWHASH_ALG_ARGON2ID13);
+		$master = sodium_crypto_secretbox_open($b64($wrapped['key']), $b64($wrapped['nonce']), $kek);
+		sodium_memzero($kek);
+		if ($master === false) {
+			throw new CryptoException('Wrong passphrase or damaged key');
+		}
+		return new self($master);
+	}
+
+	public function __destruct() {
+		sodium_memzero($this->masterKey);
+		sodium_memzero($this->dataKey);
+		sodium_memzero($this->hashKey);
+	}
+}
@@ -0,0 +1,154 @@
+<?php
+
+declare(strict_types=1);
+
+// SPDX-FileCopyrightText: 2026 Dillard Blom
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+namespace OCA\NgBackup\Crypto;
+
+/**
+ * Streaming authenticated encryption with libsodium secretstream (XChaCha20-Poly1305).
+ *
+ * Object layout:
+ *   "NGB1" | secretstream header (24 bytes) | frame*
+ *   frame = uint32 big-endian ciphertext length | ciphertext
+ * Each frame holds at most FRAME_SIZE plaintext bytes; the last frame carries TAG_FINAL, so a
+ * truncated object fails to decrypt instead of yielding a shorter plaintext.
+ *
+ * Memory use is bounded by FRAME_SIZE, independent of the object size.
+ */
+final class StreamCipher {
+	public const MAGIC = 'NGB1';
+	public const FRAME_SIZE = 65536;
+	private const HEADER_BYTES = SODIUM_CRYPTO_SECRETSTREAM_XCHACHA20POLY1305_HEADERBYTES;
+	private const ABYTES = SODIUM_CRYPTO_SECRETSTREAM_XCHACHA20POLY1305_ABYTES;
+
+	public function __construct(
+		private KeyRing $keys,
+	) {
+	}
+
+	/**
+	 * @param resource $in readable stream (plaintext)
+	 * @param resource $out writable stream (ciphertext)
+	 * @return int number of ciphertext bytes written
+	 */
+	public function encrypt($in, $out, string $associatedData = ''): int {
+		[$state, $header] = sodium_crypto_secretstream_xchacha20poly1305_init_push($this->keys->dataKey());
+		$written = self::write($out, self::MAGIC . $header);
+
+		// Read one frame ahead so the last frame can be tagged FINAL.
+		$current = self::readFull($in, self::FRAME_SIZE);
+		do {
+			$next = $current === '' ? '' : self::readFull($in, self::FRAME_SIZE);
+			$tag = $next === ''
+				? SODIUM_CRYPTO_SECRETSTREAM_XCHACHA20POLY1305_TAG_FINAL
+				: SODIUM_CRYPTO_SECRETSTREAM_XCHACHA20POLY1305_TAG_MESSAGE;
+			$cipher = sodium_crypto_secretstream_xchacha20poly1305_push($state, $current, $associatedData, $tag);
+			$written += self::write($out, pack('N', strlen($cipher)) . $cipher);
+			$current = $next;
+		} while ($tag !== SODIUM_CRYPTO_SECRETSTREAM_XCHACHA20POLY1305_TAG_FINAL);
+
+		sodium_memzero($state);
+		return $written;
+	}
+
+	/**
+	 * @param resource $in readable stream (ciphertext)
+	 * @param resource $out writable stream (plaintext)
+	 * @return int number of plaintext bytes written
+	 * @throws CryptoException on a wrong key, tampering or truncation
+	 */
+	public function decrypt($in, $out, string $associatedData = ''): int {
+		$head = self::readFull($in, strlen(self::MAGIC) + self::HEADER_BYTES);
+		if (strlen($head) !== strlen(self::MAGIC) + self::HEADER_BYTES || !str_starts_with($head, self::MAGIC)) {
+			throw new CryptoException('Not an NG Backup object');
+		}
+		$state = sodium_crypto_secretstream_xchacha20poly1305_init_pull(substr($head, strlen(self::MAGIC)), $this->keys->dataKey());
+
+		$written = 0;
+		while (true) {
+			$len = self::readFull($in, 4);
+			if (strlen($len) !== 4) {
+				throw new CryptoException('Object is truncated');
+			}
+			$size = unpack('N', $len)[1];
+			if ($size < self::ABYTES || $size > self::FRAME_SIZE + self::ABYTES) {
+				throw new CryptoException('Invalid frame size');
+			}
+			$cipher = self::readFull($in, $size);
+			if (strlen($cipher) !== $size) {
+				throw new CryptoException('Object is truncated');
+			}
+			$result = sodium_crypto_secretstream_xchacha20poly1305_pull($state, $cipher, $associatedData);
+			if ($result === false) {
+				throw new CryptoException('Decryption failed (wrong key or damaged data)');
+			}
+			[$plain, $tag] = $result;
+			$written += self::write($out, $plain);
+			if ($tag === SODIUM_CRYPTO_SECRETSTREAM_XCHACHA20POLY1305_TAG_FINAL) {
+				break;
+			}
+		}
+		if (self::readFull($in, 1) !== '') {
+			throw new CryptoException('Unexpected data after the final frame');
+		}
+		sodium_memzero($state);
+		return $written;
+	}
+
+	/** One-shot helpers for small objects (manifests, index). */
+	public function encryptString(string $plain, string $associatedData = ''): string {
+		$in = fopen('php://memory', 'w+b');
+		fwrite($in, $plain);
+		rewind($in);
+		$out = fopen('php://memory', 'w+b');
+		$this->encrypt($in, $out, $associatedData);
+		rewind($out);
+		return (string)stream_get_contents($out);
+	}
+
+	public function decryptString(string $cipher, string $associatedData = ''): string {
+		$in = fopen('php://memory', 'w+b');
+		fwrite($in, $cipher);
+		rewind($in);
+		$out = fopen('php://memory', 'w+b');
+		$this->decrypt($in, $out, $associatedData);
+		rewind($out);
+		return (string)stream_get_contents($out);
+	}
+
+	/** fread() may return short reads on network streams; keep reading until $length or EOF. */
+	private static function readFull($in, int $length): string {
+		$buffer = '';
+		while (strlen($buffer) < $length && !feof($in)) {
+			$chunk = fread($in, $length - strlen($buffer));
+			if ($chunk === false) {
+				throw new CryptoException('Read error');
+			}
+			if ($chunk === '') {
+				// Non-blocking stream without data yet, or EOF not yet flagged.
+				if (feof($in)) {
+					break;
+				}
+				continue;
+			}
+			$buffer .= $chunk;
+		}
+		return $buffer;
+	}
+
+	private static function write($out, string $data): int {
+		$total = 0;
+		$length = strlen($data);
+		while ($total < $length) {
+			$n = fwrite($out, $total === 0 ? $data : substr($data, $total));
+			if ($n === false || $n === 0) {
+				throw new CryptoException('Write error');
+			}
+			$total += $n;
+		}
+		return $total;
+	}
+}