zlib,vfs: add ZIP archive support

Add ZIP archive read/write support to zlib, plus a VFS provider that
mounts a ZIP archive as a browsable, read/write file system.

zlib gains three classes and a pair of helper functions:

* ZipEntry represents a single archive member: name, metadata, and
  content, created from a Buffer/stream (`create()`/`createStream()`)
  or read back out of raw entry bytes (`read()`). It is otherwise
  immutable; every accessor and static creator has a synchronous and
  an asynchronous form.
* ZipFile wraps a real ZIP file on disk (`open()`/`openSync()`),
  exposing get/add/delete/stream by name plus `compact()` to reclaim
  space left behind by deleted entries. Opened read-only unless
  `{ writable: true }` is passed.
* ZipBuffer is the equivalent fully in-memory representation,
  serializable back to a Buffer with `toBuffer()`/`toBufferSync()`.
  Always writable, since nothing is written until asked to serialize.
* `createZipArchive()`/`createZipArchiveSync()` build a brand-new
  archive (as a stream or a Buffer) from a list of entries.

Every read path enforces `getMaxZipContentSize()`/
`setMaxZipContentSize()` limits and rejects malformed local/central
directory records, guarding against zip-bomb and corrupt-archive
inputs. New ERR_ZIP_* error codes cover corrupt entries, missing
entries, oversized entries, invalid archives, non-writable archives,
and unsupported ZIP features.

node:vfs gains ArchiveProvider, a VirtualProvider backed by an
already-open ZipBuffer or ZipFile: mounting one exposes the archive's
entries through the same fs-like API any other VFS provider offers.
Directories are inferred from entry-name prefixes rather than stored
explicitly. Because a ZIP member can only be written or read in full,
never edited in place, a file opened for writing is buffered and only
committed as a new archive entry when its handle is closed. The
provider's readonly flag mirrors the underlying archive's own
writable state.

Author: pipobscure

doc/api/errors.md

@@ -3674,6 +3674,50 @@ All attempts at serializing an uncaught exception from a worker thread failed.
 
 The requested functionality is not supported in worker threads.
 
+<a id="ERR_ZIP_ENTRY_CORRUPT"></a>
+
+### `ERR_ZIP_ENTRY_CORRUPT`
+
+A ZIP archive entry failed CRC-32 verification, or produced more or fewer
+bytes than its declared uncompressed size, while being read.
+
+<a id="ERR_ZIP_ENTRY_NOT_FOUND"></a>
+
+### `ERR_ZIP_ENTRY_NOT_FOUND`
+
+A named entry was requested from a [`ZipFile`][] or [`ZipBuffer`][] that does
+not contain an entry with that name.
+
+<a id="ERR_ZIP_ENTRY_TOO_LARGE"></a>
+
+### `ERR_ZIP_ENTRY_TOO_LARGE`
+
+A ZIP archive entry's declared size exceeds the configured limit, or a
+provided entry name or comment exceeds the 65,535-byte encoded length that
+the ZIP format allows.
+
+<a id="ERR_ZIP_INVALID_ARCHIVE"></a>
+
+### `ERR_ZIP_INVALID_ARCHIVE`
+
+Data that was expected to be a ZIP archive, or a structure within one, is
+missing, out of bounds, or otherwise inconsistent with the ZIP format.
+
+<a id="ERR_ZIP_NOT_WRITABLE"></a>
+
+### `ERR_ZIP_NOT_WRITABLE`
+
+A mutating method (such as `zipFile.addEntry()` or `zipFile.delete()`) was
+called on a [`ZipFile`][] that was not opened with `{ writable: true }`.
+
+<a id="ERR_ZIP_UNSUPPORTED_FEATURE"></a>
+
+### `ERR_ZIP_UNSUPPORTED_FEATURE`
+
+A ZIP archive uses a feature outside of what this implementation supports,
+such as entry encryption, an unsupported compression method, or a multi-disk
+archive.
+
 <a id="ERR_ZLIB_INITIALIZATION_FAILED"></a>
 
 ### `ERR_ZLIB_INITIALIZATION_FAILED`
@@ -4617,6 +4661,8 @@ An error occurred trying to allocate memory. This should never happen.
 [`ServerResponse`]: http.md#class-httpserverresponse
 [`Temporal`]: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Temporal
 [`Writable`]: stream.md#class-streamwritable
+[`ZipBuffer`]: zlib.md#class-zlibzipbuffer
+[`ZipFile`]: zlib.md#class-zlibzipfile
 [`child_process`]: child_process.md
 [`cipher.getAuthTag()`]: crypto.md#ciphergetauthtag
 [`crypto.getDiffieHellman()`]: crypto.md#cryptogetdiffiehellmangroupname

doc/api/vfs.md

@@ -38,10 +38,12 @@ Mounting a VFS only redirects supported [`node:fs`][] calls whose resolved paths
 are under the mount point. It does not prevent code from using other paths or
 other Node.js APIs to access resources available to the process.
 [`RealFSProvider`][] maps VFS paths under its configured root and rejects paths
-that resolve outside that root, but that check is not a security boundary. Do
-not rely on VFS to run untrusted code; use operating-system-level isolation,
-such as separate users, containers, or platform sandboxes, when a security
-boundary is required.
+that resolve outside that root, but that check is not a security boundary.
+[`ArchiveProvider`][] has no real file-system paths of its own to escape; its
+entries only ever exist within the archive's own namespace. Do not rely on VFS
+to run untrusted code; use operating-system-level isolation, such as separate
+users, containers, or platform sandboxes, when a security boundary is
+required.
 
 ## Basic usage
 
@@ -304,6 +306,55 @@ added: v26.4.0
 
 The resolved absolute path used as the root.
 
+## Class: `ArchiveProvider`
+
+<!-- YAML
+added: REPLACEME
+-->
+
+A provider that exposes the entries of a ZIP archive - either a
+[`zlib.ZipBuffer`][] (in memory) or a [`zlib.ZipFile`][] (on disk) - through
+the VFS API. `provider.readonly` reflects the archive's own
+[`zipFile.writable`][] flag: a `ZipBuffer` is always writable, and a
+`ZipFile` is writable only when opened with `{ writable: true }`.
+
+Directories are recognized both explicitly (an entry whose name ends in `/`)
+and implicitly (any entry name starting with `"<dir>/"`). `readdir()` does
+not support `{ recursive: true }`. Because a ZIP member cannot be edited or
+read in place - only fully written or fully decompressed - a file opened for
+writing only commits its content (as a new archive entry) when the handle is
+closed.
+
+Every method has a synchronous counterpart (`openSync()`, `statSync()`,
+`readdirSync()`, and so on), backed by the equally complete synchronous
+surface [`zlib.ZipBuffer`][]/[`zlib.ZipFile`][] expose. As with those, the
+synchronous methods here block the Node.js event loop and further JavaScript
+execution until the operation - including any deflate/inflate pass -
+completes.
+
+```cjs
+const vfs = require('node:vfs');
+const zlib = require('node:zlib');
+const { readFileSync } = require('node:fs');
+
+async function main() {
+  const zip = new zlib.ZipBuffer(readFileSync('archive.zip'));
+  const archiveVfs = vfs.create(new vfs.ArchiveProvider(zip));
+
+  console.log(await archiveVfs.promises.readdir('/'));
+  await archiveVfs.promises.writeFile('/new.txt', 'hello');
+}
+main();
+```
+
+### `new ArchiveProvider(source)`
+
+<!-- YAML
+added: REPLACEME
+-->
+
+* `source` {zlib.ZipBuffer|zlib.ZipFile} An already-open archive.
+
 ## Implementation details
 
 ### `Stats` objects
@@ -318,10 +369,14 @@ fields use synthetic but stable values:
 * `blocks` is `Math.ceil(size / 512)`.
 * Times default to the moment the entry was created/last modified.
 
+[`ArchiveProvider`]: #class-archiveprovider
 [`MemoryProvider`]: #class-memoryprovider
 [`RealFSProvider`]: #class-realfsprovider
 [`VirtualFileSystem`]: #class-virtualfilesystem
 [`VirtualProvider`]: #class-virtualprovider
 [`fs.BigIntStats`]: fs.md#class-fsbigintstats
 [`fs.Stats`]: fs.md#class-fsstats
 [`node:fs`]: fs.md
+[`zipFile.writable`]: zlib.md#zipfilewritable
+[`zlib.ZipBuffer`]: zlib.md#class-zlibzipbuffer
+[`zlib.ZipFile`]: zlib.md#class-zlibzipfile