ktg-plugin-marketplace/plugins/llm-security/scanners/lib/zip-extract.mjs
Kjell Tore Guttormsen fe0193956d feat(llm-security): /security ide-scan <url> — Marketplace/OpenVSX/direct VSIX (v6.4.0)
Pre-installation verification of VS Code extensions via URL — fetch a remote
VSIX, extract it in a hardened sandbox, and run the existing IDE scanner
pipeline against it. No npm dependencies.

Sources:
- VS Code Marketplace (publisher.gallery.vsassets.io direct download)
- OpenVSX (open-vsx.org official API)
- Direct .vsix HTTPS URLs

Defenses:
- HTTPS-only, TLS verified, manual redirect with per-source host whitelist
- 30s total timeout via AbortController
- 50MB compressed cap, 500MB uncompressed, 100x expansion ratio
- Zero-dep ZIP extractor: zip-slip, absolute paths, drive letters, NUL bytes,
  symlinks (Unix mode 0xA000), depth limits, ZIP64 rejected, encrypted rejected
- SHA-256 streamed during fetch, surfaced in meta.source
- Temp dir cleanup in all paths (try/finally)

Files:
- scanners/lib/vsix-fetch.mjs (HTTPS fetcher, host whitelist, streaming SHA-256)
- scanners/lib/zip-extract.mjs (zero-dep parser with hardening caps)
- knowledge/marketplace-api-notes.md (endpoint reference)
- 3 test files (48 tests added: vsix-fetch, zip-extract, ide-extension-url)

Tests: 1296 → 1344 (all green).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-04-17 17:16:26 +02:00

374 lines
13 KiB
JavaScript

// zip-extract.mjs — Zero-dependency ZIP parser and safe extractor for VSIX files.
//
// VSIX is a plain ZIP container. We implement the minimum subset needed to:
// - Parse the End of Central Directory (EOCD)
// - Walk Central Directory headers to enumerate entries
// - Read each Local File Header and inflate its data with node:zlib
// - Validate every entry name BEFORE creating any file (zip-slip, absolute, symlinks)
// - Enforce caps: total entries, total uncompressed bytes, expansion ratio, depth
//
// Compression methods supported:
// 0 STORE (no compression)
// 8 DEFLATE (raw deflate, via zlib.createInflateRaw)
// Anything else is rejected.
//
// Spec: https://pkware.cachefly.net/webdocs/casestudies/APPNOTE.TXT
//
// SECURITY NOTES
// - We resolve every entry path inside `targetDir` and require startsWith(targetDir + sep)
// - We reject absolute paths, drive letters, NUL bytes, and `..` segments after normalization
// - We reject entries whose external_attr indicates a symbolic link (0xA000 in upper word, Unix)
// - We reject ZIP64 entries (we don't claim to support them; >4GB is out of scope for VSIX)
// - We reject entries with general purpose bit 0 (encryption) set
// - Directories are not extracted from entries — created on-demand from file paths
import { createInflateRaw } from 'node:zlib';
import { mkdir, writeFile } from 'node:fs/promises';
import { dirname, join, resolve, sep, normalize } from 'node:path';
const SIG_LFH = 0x04034b50; // Local File Header
const SIG_CD = 0x02014b50; // Central Directory
const SIG_EOCD = 0x06054b50; // End of Central Directory
const SIG_EOCD64_LOC = 0x07064b50; // ZIP64 EOCD locator (presence => reject)
const SIG_EOCD64 = 0x06064b50; // ZIP64 EOCD record
const DEFAULT_CAPS = Object.freeze({
maxEntries: 10_000,
maxUncompressedBytes: 500 * 1024 * 1024, // 500MB
maxExpansionRatio: 100, // sum uncompressed / sum compressed
maxDepth: 20,
});
class ZipError extends Error {
constructor(message, code = 'ZIP_INVALID') {
super(message);
this.code = code;
this.name = 'ZipError';
}
}
function readU16(buf, off) { return buf.readUInt16LE(off); }
function readU32(buf, off) { return buf.readUInt32LE(off); }
/**
* Locate the End of Central Directory (EOCD) by scanning backwards from end of buffer.
* EOCD is 22 bytes minimum; comment can extend it up to 65557 bytes total.
*/
function findEOCD(buf) {
const minOff = Math.max(0, buf.length - 22 - 0xFFFF);
for (let off = buf.length - 22; off >= minOff; off--) {
if (readU32(buf, off) === SIG_EOCD) {
// Sanity: comment length must fit
const commentLen = readU16(buf, off + 20);
if (off + 22 + commentLen === buf.length) return off;
}
}
throw new ZipError('EOCD signature not found', 'ZIP_NO_EOCD');
}
function parseEOCD(buf, off) {
return {
diskNumber: readU16(buf, off + 4),
cdDisk: readU16(buf, off + 6),
cdEntriesOnDisk: readU16(buf, off + 8),
cdEntriesTotal: readU16(buf, off + 10),
cdSize: readU32(buf, off + 12),
cdOffset: readU32(buf, off + 16),
commentLength: readU16(buf, off + 20),
};
}
/**
* Parse central directory entries and return a structured list.
* @param {Buffer} buf
* @returns {Array}
*/
function parseCentralDirectory(buf) {
// Reject ZIP64 (we do not implement it).
// Look for ZIP64 EOCD locator in the 20 bytes immediately before EOCD.
const eocdOff = findEOCD(buf);
if (eocdOff >= 20 && readU32(buf, eocdOff - 20) === SIG_EOCD64_LOC) {
throw new ZipError('ZIP64 archives are not supported', 'ZIP_ZIP64');
}
const eocd = parseEOCD(buf, eocdOff);
if (eocd.diskNumber !== 0 || eocd.cdDisk !== 0) {
throw new ZipError('multi-disk archives are not supported', 'ZIP_MULTIDISK');
}
if (eocd.cdEntriesOnDisk !== eocd.cdEntriesTotal) {
throw new ZipError('split central directory not supported', 'ZIP_SPLIT_CD');
}
if (eocd.cdOffset === 0xFFFFFFFF || eocd.cdSize === 0xFFFFFFFF) {
throw new ZipError('ZIP64 fields detected', 'ZIP_ZIP64');
}
if (eocd.cdOffset + eocd.cdSize > buf.length) {
throw new ZipError('central directory extends past EOF', 'ZIP_BAD_CD_OFFSET');
}
const entries = [];
let p = eocd.cdOffset;
for (let i = 0; i < eocd.cdEntriesTotal; i++) {
if (p + 46 > buf.length) throw new ZipError('truncated central directory', 'ZIP_TRUNCATED_CD');
if (readU32(buf, p) !== SIG_CD) throw new ZipError('bad central directory signature', 'ZIP_BAD_CD_SIG');
const versionMadeBy = readU16(buf, p + 4);
const generalFlags = readU16(buf, p + 8);
const method = readU16(buf, p + 10);
const crc32 = readU32(buf, p + 16);
const compSize = readU32(buf, p + 20);
const uncompSize = readU32(buf, p + 24);
const nameLen = readU16(buf, p + 28);
const extraLen = readU16(buf, p + 30);
const commentLen = readU16(buf, p + 32);
const externalAttr = readU32(buf, p + 38);
const lfhOffset = readU32(buf, p + 42);
if (compSize === 0xFFFFFFFF || uncompSize === 0xFFFFFFFF || lfhOffset === 0xFFFFFFFF) {
throw new ZipError('ZIP64 fields detected in entry', 'ZIP_ZIP64');
}
const nameStart = p + 46;
if (nameStart + nameLen > buf.length) throw new ZipError('entry name extends past EOF', 'ZIP_BAD_NAME');
const rawName = buf.slice(nameStart, nameStart + nameLen).toString('utf8');
entries.push({
versionMadeBy,
generalFlags,
method,
crc32,
compSize,
uncompSize,
nameLen,
extraLen,
commentLen,
externalAttr,
lfhOffset,
name: rawName,
});
p += 46 + nameLen + extraLen + commentLen;
}
return entries;
}
/**
* Validate an entry name and return a relative path safe to join with targetDir.
* Throws ZipError on any rejected pattern. Returns null for entries that should be skipped (directory entries).
*/
function validateEntryName(rawName, caps) {
if (rawName.length === 0) throw new ZipError('empty entry name', 'ZIP_BAD_NAME');
if (rawName.length > 1024) throw new ZipError('entry name exceeds 1024 chars', 'ZIP_BAD_NAME');
if (rawName.includes('\u0000')) throw new ZipError('NUL byte in entry name', 'ZIP_BAD_NAME');
// Directory entries end with '/' — skip; we'll create dirs as needed for files.
if (rawName.endsWith('/')) return null;
// Reject Windows absolute paths (drive letters, UNC) and POSIX absolute paths.
if (rawName.startsWith('/') || rawName.startsWith('\\')) {
throw new ZipError(`absolute entry path: ${rawName}`, 'ZIP_ABSOLUTE');
}
if (/^[A-Za-z]:[\\/]/.test(rawName)) {
throw new ZipError(`drive-letter entry path: ${rawName}`, 'ZIP_ABSOLUTE');
}
// Normalize: convert backslashes to forward slashes (some zip tools emit \).
const unixName = rawName.replace(/\\/g, '/');
// Reject any path component equal to '..' — even after normalization it must escape.
const parts = unixName.split('/');
for (const part of parts) {
if (part === '..') {
throw new ZipError(`parent traversal in entry: ${rawName}`, 'ZIP_TRAVERSAL');
}
}
// Final path normalization (collapses '.' segments, NFC).
const normalized = normalize(unixName).normalize('NFC');
if (normalized.startsWith('..') || normalized.includes(`${sep}..${sep}`) || normalized === '..') {
throw new ZipError(`parent traversal after normalization: ${rawName}`, 'ZIP_TRAVERSAL');
}
if (normalized.split(sep).length > caps.maxDepth) {
throw new ZipError(`entry path exceeds depth ${caps.maxDepth}: ${rawName}`, 'ZIP_DEEP');
}
return normalized;
}
/**
* Detect symlink entries from the external_attr field.
* For Unix-made entries (versionMadeBy upper byte = 3), the high 16 bits are mode bits.
* Symlink mode is 0xA000.
*/
function isSymlink(entry) {
const madeByOs = (entry.versionMadeBy >>> 8) & 0xFF;
if (madeByOs !== 3) return false; // Only Unix encodes mode bits
const mode = (entry.externalAttr >>> 16) & 0xFFFF;
return (mode & 0xF000) === 0xA000;
}
/**
* Inflate raw deflate bytes via node:zlib, with a hard upper bound on output size.
* Aborts as soon as output exceeds maxBytes (zip-bomb defense).
*/
function inflateBounded(input, expectedSize, maxBytes) {
return new Promise((resolvePromise, reject) => {
const stream = createInflateRaw();
const chunks = [];
let total = 0;
let aborted = false;
stream.on('data', chunk => {
if (aborted) return;
total += chunk.length;
if (total > maxBytes) {
aborted = true;
stream.destroy(new ZipError(`inflate exceeds cap (${maxBytes} bytes)`, 'ZIP_BOMB'));
return;
}
chunks.push(chunk);
});
stream.on('end', () => {
if (aborted) return;
if (total !== expectedSize) {
reject(new ZipError(
`inflated size ${total} does not match expected ${expectedSize}`,
'ZIP_SIZE_MISMATCH',
));
return;
}
resolvePromise(Buffer.concat(chunks, total));
});
stream.on('error', err => {
if (err instanceof ZipError) reject(err);
else reject(new ZipError(`inflate failed: ${err.message}`, 'ZIP_INFLATE'));
});
stream.end(input);
});
}
/**
* Read the data section of one entry given its central directory record.
* @param {Buffer} buf
* @param {object} entry
* @param {number} remainingBudget - max bytes we may still inflate this archive
* @returns {Promise<Buffer>}
*/
async function readEntryData(buf, entry, remainingBudget) {
const lfhOff = entry.lfhOffset;
if (lfhOff + 30 > buf.length) throw new ZipError('LFH past EOF', 'ZIP_BAD_LFH');
if (readU32(buf, lfhOff) !== SIG_LFH) throw new ZipError('bad LFH signature', 'ZIP_BAD_LFH');
const lfhNameLen = readU16(buf, lfhOff + 26);
const lfhExtraLen = readU16(buf, lfhOff + 28);
const dataStart = lfhOff + 30 + lfhNameLen + lfhExtraLen;
const dataEnd = dataStart + entry.compSize;
if (dataEnd > buf.length) throw new ZipError('entry data past EOF', 'ZIP_BAD_DATA');
const compressed = buf.slice(dataStart, dataEnd);
const cap = Math.min(entry.uncompSize, remainingBudget);
if (entry.method === 0) {
if (entry.compSize !== entry.uncompSize) {
throw new ZipError('STORED entry compSize != uncompSize', 'ZIP_BAD_STORED');
}
if (entry.uncompSize > remainingBudget) {
throw new ZipError('STORED entry exceeds budget', 'ZIP_BOMB');
}
return compressed;
}
if (entry.method === 8) {
return inflateBounded(compressed, entry.uncompSize, cap);
}
throw new ZipError(`unsupported compression method ${entry.method}`, 'ZIP_BAD_METHOD');
}
/**
* Extract a VSIX/ZIP buffer to targetDir with all caps and validations applied.
* targetDir must already exist or be creatable; caller owns cleanup.
*
* @param {Buffer} buf - The full ZIP buffer
* @param {string} targetDir - Absolute path to extract into
* @param {object} [opts]
* @param {object} [opts.caps] - Override default caps
* @returns {Promise<{ entries: number, bytes: number, files: string[] }>}
*/
export async function extractToDir(buf, targetDir, opts = {}) {
const caps = { ...DEFAULT_CAPS, ...(opts.caps || {}) };
const absTarget = resolve(targetDir);
await mkdir(absTarget, { recursive: true });
const entries = parseCentralDirectory(buf);
if (entries.length > caps.maxEntries) {
throw new ZipError(`too many entries (${entries.length} > ${caps.maxEntries})`, 'ZIP_TOO_MANY_ENTRIES');
}
let totalUncomp = 0;
let totalComp = 0;
const files = [];
for (const entry of entries) {
if (entry.generalFlags & 0x0001) {
throw new ZipError(`encrypted entry not allowed: ${entry.name}`, 'ZIP_ENCRYPTED');
}
if (isSymlink(entry)) {
throw new ZipError(`symlink entry not allowed: ${entry.name}`, 'ZIP_SYMLINK');
}
const safeName = validateEntryName(entry.name, caps);
if (safeName === null) continue; // directory entry
const fullPath = join(absTarget, safeName);
const resolved = resolve(fullPath);
if (resolved !== absTarget && !resolved.startsWith(absTarget + sep)) {
throw new ZipError(`zip-slip: ${entry.name} resolves outside target`, 'ZIP_TRAVERSAL');
}
if (entry.uncompSize > caps.maxUncompressedBytes) {
throw new ZipError(`entry ${entry.name} exceeds maxUncompressedBytes`, 'ZIP_BOMB');
}
const remaining = caps.maxUncompressedBytes - totalUncomp;
const data = await readEntryData(buf, entry, remaining);
totalUncomp += data.length;
totalComp += Math.max(1, entry.compSize); // avoid div-by-zero in ratio
if (totalUncomp > caps.maxUncompressedBytes) {
throw new ZipError(`total uncompressed exceeds cap`, 'ZIP_BOMB');
}
if (totalUncomp / totalComp > caps.maxExpansionRatio) {
throw new ZipError(
`expansion ratio exceeds ${caps.maxExpansionRatio}x (${totalUncomp}/${totalComp})`,
'ZIP_BOMB',
);
}
await mkdir(dirname(resolved), { recursive: true });
await writeFile(resolved, data);
files.push(safeName);
}
return { entries: files.length, bytes: totalUncomp, files };
}
/**
* Parse central directory only — no extraction. Useful for inspection / tests.
*/
export function listEntries(buf) {
return parseCentralDirectory(buf).map(e => ({
name: e.name,
method: e.method,
compSize: e.compSize,
uncompSize: e.uncompSize,
isDir: e.name.endsWith('/'),
isSymlink: isSymlink(e),
}));
}
export { ZipError };
export const __testing = {
DEFAULT_CAPS,
validateEntryName,
isSymlink,
parseCentralDirectory,
inflateBounded,
};