diff --git a/README.md b/README.md
index 5e3a9a4..44a4fcb 100644
--- a/README.md
+++ b/README.md
@@ -27,14 +27,14 @@ If you want to pack/unpack directories on the file system check out [tar-fs](htt
 To create a pack stream use `tar.pack()` and call `pack.entry(header, [callback])` to add tar entries.
 
 ``` js
-var tar = require('tar-stream')
-var pack = tar.pack() // pack is a stream
+const tar = require('tar-stream')
+const pack = tar.pack() // pack is a stream
 
 // add a file called my-test.txt with the content "Hello World!"
 pack.entry({ name: 'my-test.txt' }, 'Hello World!')
 
 // add a file called my-stream-test.txt from a stream
-var entry = pack.entry({ name: 'my-stream-test.txt', size: 11 }, function(err) {
+const entry = pack.entry({ name: 'my-stream-test.txt', size: 11 }, function(err) {
   // the stream was added
   // no more entries
   pack.finalize()
@@ -54,21 +54,21 @@ pack.pipe(process.stdout)
 To extract a stream use `tar.extract()` and listen for `extract.on('entry', (header, stream, next) )`
 
 ``` js
-var extract = tar.extract()
+const extract = tar.extract()
 
-extract.on('entry', function(header, stream, next) {
+extract.on('entry', function (header, stream, next) {
   // header is the tar header
   // stream is the content body (might be an empty stream)
   // call next when you are done with this entry
 
-  stream.on('end', function() {
+  stream.on('end', function () {
     next() // ready for next entry
   })
 
   stream.resume() // just auto drain the stream
 })
 
-extract.on('finish', function() {
+extract.on('finish', function () {
   // all entries read
 })
 
@@ -77,6 +77,21 @@ pack.pipe(extract)
 
 The tar archive is streamed sequentially, meaning you **must** drain each entry's stream as you get them or else the main extract stream will receive backpressure and stop reading.
 
+## Extracting as an async iterator
+
+The extraction stream in addition to being a writable stream is also an async iterator
+
+``` js
+const extract = tar.extract()
+
+someStream.pipe(extract)
+
+for await (const entry of extract) {
+  entry.header // the tar header
+  entry.resume() // the entry is the stream also
+}
+```
+
 ## Headers
 
 The header object using in `entry` should contain the following properties.
@@ -106,18 +121,18 @@ Most of these values can be found by stat'ing a file.
 Using tar-stream it is easy to rewrite paths / change modes etc in an existing tarball.
 
 ``` js
-var extract = tar.extract()
-var pack = tar.pack()
-var path = require('path')
+const extract = tar.extract()
+const pack = tar.pack()
+const path = require('path')
 
-extract.on('entry', function(header, stream, callback) {
+extract.on('entry', function (header, stream, callback) {
   // let's prefix all names with 'tmp'
   header.name = path.join('tmp', header.name)
   // write the new entry to the pack stream
   stream.pipe(pack.entry(header, callback))
 })
 
-extract.on('finish', function() {
+extract.on('finish', function () {
   // all entries done - lets finalize it
   pack.finalize()
 })
@@ -133,15 +148,15 @@ pack.pipe(newTarballStream)
 
 
 ``` js
-var fs = require('fs')
-var tar = require('tar-stream')
+const fs = require('fs')
+const tar = require('tar-stream')
 
-var pack = tar.pack() // pack is a stream
-var path = 'YourTarBall.tar'
-var yourTarball = fs.createWriteStream(path)
+const pack = tar.pack() // pack is a stream
+const path = 'YourTarBall.tar'
+const yourTarball = fs.createWriteStream(path)
 
 // add a file called YourFile.txt with the content "Hello World!"
-pack.entry({name: 'YourFile.txt'}, 'Hello World!', function (err) {
+pack.entry({ name: 'YourFile.txt' }, 'Hello World!', function (err) {
   if (err) throw err
   pack.finalize()
 })
diff --git a/constants.js b/constants.js
new file mode 100644
index 0000000..15601d2
--- /dev/null
+++ b/constants.js
@@ -0,0 +1,12 @@
+try {
+  module.exports = require('fs').constants
+} catch {
+  module.exports = { // just for envs without fs
+    S_IFMT: 61440,
+    S_IFDIR: 16384,
+    S_IFCHR: 8192,
+    S_IFBLK: 24576,
+    S_IFIFO: 4096,
+    S_IFLNK: 40960
+  }
+}
diff --git a/extract.js b/extract.js
index add94b1..acdb67d 100644
--- a/extract.js
+++ b/extract.js
@@ -1,37 +1,98 @@
-const bl = require('bl')
-const { Writable, PassThrough } = require('streamx')
+const { Writable, Readable, getStreamError } = require('streamx')
+const FIFO = require('fast-fifo')
+const b4a = require('b4a')
 const headers = require('./headers')
 
-const noop = function () {}
+const EMPTY = b4a.alloc(0)
 
-const overflow = function (size) {
-  size &= 511
-  return size && 512 - size
-}
+class BufferList {
+  constructor () {
+    this.buffered = 0
+    this.shifted = 0
+    this.queue = new FIFO()
 
-const emptyStream = function (self, offset) {
-  const s = new Source(self, offset)
-  s.end()
-  return s
-}
+    this._offset = 0
+  }
+
+  push (buffer) {
+    this.buffered += buffer.byteLength
+    this.queue.push(buffer)
+  }
+
+  shiftFirst (size) {
+    return this._buffered === 0 ? null : this._next(size)
+  }
+
+  shift (size) {
+    if (size > this.buffered) return null
+    if (size === 0) return EMPTY
+
+    let chunk = this._next(size)
+
+    if (size === chunk.byteLength) return chunk // likely case
 
-const mixinPax = function (header, pax) {
-  if (pax.path) header.name = pax.path
-  if (pax.linkpath) header.linkname = pax.linkpath
-  if (pax.size) header.size = parseInt(pax.size, 10)
-  header.pax = pax
-  return header
+    const chunks = [chunk]
+
+    while ((size -= chunk.byteLength) > 0) {
+      chunk = this._next(size)
+      chunks.push(chunk)
+    }
+
+    return b4a.concat(chunks)
+  }
+
+  _next (size) {
+    const buf = this.queue.peek()
+    const rem = buf.byteLength - this._offset
+
+    if (size >= rem) {
+      const sub = this._offset ? buf.subarray(this._offset, buf.byteLength) : buf
+      this.queue.shift()
+      this._offset = 0
+      this.buffered -= rem
+      this.shifted += rem
+      return sub
+    }
+
+    this.buffered -= size
+    this.shifted += size
+
+    return buf.subarray(this._offset, (this._offset += size))
+  }
 }
 
-class Source extends PassThrough {
-  constructor (self, offset) {
+class Source extends Readable {
+  constructor (self, header, offset) {
     super()
-    this._parent = self
+
+    this.header = header
     this.offset = offset
+
+    this._parent = self
+  }
+
+  _read (cb) {
+    if (this._parent._stream === this) {
+      this._parent._update()
+    }
+    cb(null)
   }
 
   _predestroy () {
-    this._parent.destroy()
+    this._parent.destroy(getStreamError(this))
+  }
+
+  _detach () {
+    if (this._parent._stream === this) {
+      this._parent._stream = null
+      this._parent._missing = overflow(this.header.size)
+      this._parent._update()
+    }
+  }
+
+  _destroy (cb) {
+    this._detach()
+    cb(null)
   }
 }
 
@@ -39,208 +100,306 @@ class Extract extends Writable {
   constructor (opts) {
     super(opts)
 
-    opts = opts || {}
+    if (!opts) opts = {}
 
+    this._buffer = new BufferList()
     this._offset = 0
-    this._buffer = bl()
-    this._missing = 0
-    this._partial = false
-    this._onparse = noop
     this._header = null
     this._stream = null
-    this._overflow = null
-    this._cb = null
+    this._missing = 0
+    this._longHeader = false
+    this._callback = noop
     this._locked = false
+    this._finished = false
     this._pax = null
     this._paxGlobal = null
     this._gnuLongPath = null
     this._gnuLongLinkPath = null
+    this._filenameEncoding = opts.filenameEncoding || 'utf-8'
+    this._allowUnknownFormat = !!opts.allowUnknownFormat
+    this._unlockBound = this._unlock.bind(this)
+  }
 
-    const self = this
-    const b = self._buffer
+  _unlock (err) {
+    this._locked = false
 
-    const oncontinue = function () {
-      self._continue()
+    if (err) {
+      this.destroy(err)
+      this._continueWrite(err)
+      return
     }
 
-    const onunlock = function (err) {
-      self._locked = false
-      if (err) return self.destroy(err)
-      if (!self._stream) oncontinue()
+    this._update()
+  }
+
+  _consumeHeader () {
+    if (this._locked) return false
+
+    this._offset = this._buffer.shifted
+
+    try {
+      this._header = headers.decode(this._buffer.shift(512), this._filenameEncoding, this._allowUnknownFormat)
+    } catch (err) {
+      this._continueWrite(err)
+      return false
     }
 
-    const onstreamend = function () {
-      self._stream = null
-      const drain = overflow(self._header.size)
-      if (drain) self._parse(drain, ondrain)
-      else self._parse(512, onheader)
-      if (!self._locked) oncontinue()
+    if (!this._header) return true
+
+    switch (this._header.type) {
+      case 'gnu-long-path':
+      case 'gnu-long-link-path':
+      case 'pax-global-header':
+      case 'pax-header':
+        this._longHeader = true
+        this._missing = this._header.size
+        return true
     }
 
-    const ondrain = function () {
-      self._buffer.consume(overflow(self._header.size))
-      self._parse(512, onheader)
-      oncontinue()
+    this._locked = true
+    this._applyLongHeaders()
+
+    if (this._header.size === 0 || this._header.type === 'directory') {
+      const stream = this._createStream()
+      stream.push(null)
+      this.emit('entry', this._header, stream, this._unlockBound)
+      return true
     }
 
-    const onpaxglobalheader = function () {
-      const size = self._header.size
-      self._paxGlobal = headers.decodePax(b.slice(0, size))
-      b.consume(size)
-      onstreamend()
+    this._stream = this._createStream()
+    this._missing = this._header.size
+
+    this.emit('entry', this._header, this._stream, this._unlockBound)
+    return true
+  }
+
+  _applyLongHeaders () {
+    if (this._gnuLongPath) {
+      this._header.name = this._gnuLongPath
+      this._gnuLongPath = null
     }
 
-    const onpaxheader = function () {
-      const size = self._header.size
-      self._pax = headers.decodePax(b.slice(0, size))
-      if (self._paxGlobal) self._pax = Object.assign({}, self._paxGlobal, self._pax)
-      b.consume(size)
-      onstreamend()
+    if (this._gnuLongLinkPath) {
+      this._header.linkname = this._gnuLongLinkPath
+      this._gnuLongLinkPath = null
     }
 
-    const ongnulongpath = function () {
-      const size = self._header.size
-      this._gnuLongPath = headers.decodeLongPath(b.slice(0, size), opts.filenameEncoding)
-      b.consume(size)
-      onstreamend()
+    if (this._pax) {
+      if (this._pax.path) this._header.name = this._pax.path
+      if (this._pax.linkpath) this._header.linkname = this._pax.linkpath
+      if (this._pax.size) this._header.size = parseInt(this._pax.size, 10)
+      this._header.pax = this._pax
+      this._pax = null
     }
+  }
 
-    const ongnulonglinkpath = function () {
-      const size = self._header.size
-      this._gnuLongLinkPath = headers.decodeLongPath(b.slice(0, size), opts.filenameEncoding)
-      b.consume(size)
-      onstreamend()
+  _decodeLongHeader (buf) {
+    switch (this._header.type) {
+      case 'gnu-long-path':
+        this._gnuLongPath = headers.decodeLongPath(buf, this._filenameEncoding)
+        break
+      case 'gnu-long-link-path':
+        this._gnuLongLinkPath = headers.decodeLongPath(buf, this._filenameEncoding)
+        break
+      case 'pax-global-header':
+        this._paxGlobal = headers.decodePax(buf)
+        break
+      case 'pax-header':
+        this._pax = this._paxGlobal === null
+          ? headers.decodePax(buf)
+          : Object.assign({}, this._paxGlobal, headers.decodePax(buf))
+        break
     }
+  }
 
-    const onheader = function () {
-      const offset = self._offset
-      let header
-      try {
-        header = self._header = headers.decode(b.slice(0, 512), opts.filenameEncoding, opts.allowUnknownFormat)
-      } catch (err) {
-        self.destroy(err)
-      }
-      b.consume(512)
+  _consumeLongHeader () {
+    this._longHeader = false
+    this._missing = overflow(this._header.size)
 
-      if (!header) {
-        self._parse(512, onheader)
-        oncontinue()
-        return
-      }
+    const buf = this._buffer.shift(this._header.size)
 
-      if (header.type === 'gnu-long-path') {
-        self._parse(header.size, ongnulongpath)
-        oncontinue()
-        return
-      }
+    try {
+      this._decodeLongHeader(buf)
+    } catch (err) {
+      this._continueWrite(err)
+      return false
+    }
 
-      if (header.type === 'gnu-long-link-path') {
-        self._parse(header.size, ongnulonglinkpath)
-        oncontinue()
-        return
-      }
+    return true
+  }
 
-      if (header.type === 'pax-global-header') {
-        self._parse(header.size, onpaxglobalheader)
-        oncontinue()
-        return
-      }
+  _consumeStream () {
+    const buf = this._buffer.shiftFirst(this._missing)
+    if (buf === null) return false
 
-      if (header.type === 'pax-header') {
-        self._parse(header.size, onpaxheader)
-        oncontinue()
-        return
-      }
+    this._missing -= buf.byteLength
+    const drained = this._stream.push(buf)
 
-      if (self._gnuLongPath) {
-        header.name = self._gnuLongPath
-        self._gnuLongPath = null
-      }
+    if (this._missing === 0) {
+      this._stream.push(null)
+      if (drained) this._stream._detach()
+      return drained && this._locked === false
+    }
 
-      if (self._gnuLongLinkPath) {
-        header.linkname = self._gnuLongLinkPath
-        self._gnuLongLinkPath = null
-      }
+    return drained
+  }
+
+  _createStream () {
+    return new Source(this, this._header, this._offset)
+  }
 
-      if (self._pax) {
-        self._header = header = mixinPax(header, self._pax)
-        self._pax = null
+  _update () {
+    while (this._buffer.buffered > 0 && !this.destroying) {
+      if (this._missing > 0) {
+        if (this._stream !== null) {
+          if (this._consumeStream() === false) return
+          continue
+        }
+
+        if (this._longHeader === true) {
+          if (this._missing > this._buffer.buffered) break
+          if (this._consumeLongHeader() === false) return false
+          continue
+        }
+
+        const ignore = this._buffer.shiftFirst(this._missing)
+        if (ignore !== null) this._missing -= ignore.byteLength
+        continue
       }
 
-      self._locked = true
+      if (this._buffer.buffered < 512) break
+      if (this._stream !== null || this._consumeHeader() === false) return
+    }
 
-      if (!header.size || header.type === 'directory') {
-        self._parse(512, onheader)
-        self.emit('entry', header, emptyStream(self, offset), onunlock)
-        return
-      }
+    this._continueWrite(null)
+  }
 
-      self._stream = new Source(self, offset)
+  _continueWrite (err) {
+    const cb = this._callback
+    this._callback = noop
+    cb(err)
+  }
 
-      self.emit('entry', header, self._stream, onunlock)
-      self._parse(header.size, onstreamend)
-      oncontinue()
-    }
+  _write (data, cb) {
+    this._callback = cb
+    this._buffer.push(data)
+    this._update()
+  }
 
-    this._onheader = onheader
-    this._parse(512, onheader)
+  _final (cb) {
+    this._finished = this._missing === 0 && this._buffer.buffered === 0
+    cb(this._finished ? null : new Error('Unexpected end of data'))
   }
 
-  _parse (size, onparse) {
-    this._offset += size
-    this._missing = size
-    if (onparse === this._onheader) this._partial = false
-    this._onparse = onparse
+  _predestroy () {
+    this._continueWrite(null)
   }
 
-  _continue () {
-    const cb = this._cb
-    this._cb = noop
-    if (this._overflow) this._write(this._overflow, cb)
-    else cb()
+  _destroy (cb) {
+    if (this._stream) this._stream.destroy(getStreamError(this))
+    cb(null)
   }
 
-  _write (data, cb) {
-    const s = this._stream
-    const b = this._buffer
-    const missing = this._missing
-    if (data.byteLength) this._partial = true
-
-    // we do not reach end-of-chunk now. just forward it
-    if (data.byteLength < missing) {
-      this._missing -= data.byteLength
-      this._overflow = null
-      if (s) {
-        if (s.write(data, cb)) cb()
-        else s.once('drain', cb)
-        return
+  [Symbol.asyncIterator] () {
+    let error = null
+
+    let promiseResolve = null
+    let promiseReject = null
+
+    let entryStream = null
+    let entryCallback = null
+
+    const extract = this
+
+    this.on('entry', onentry)
+    this.on('error', (err) => { error = err })
+    this.on('close', onclose)
+
+    return {
+      [Symbol.asyncIterator] () {
+        return this
+      },
+      next () {
+        return new Promise(onnext)
+      },
+      return () {
+        return destroy(null)
+      },
+      throw (err) {
+        return destroy(err)
       }
-      b.append(data)
-      return cb()
     }
 
-    // end-of-chunk. the parser should call cb.
-    this._cb = cb
-    this._missing = 0
+    function consumeCallback (err) {
+      if (!entryCallback) return
+      const cb = entryCallback
+      entryCallback = null
+      cb(err)
+    }
+
+    function onnext (resolve, reject) {
+      if (error) {
+        return reject(error)
+      }
 
-    let overflow = null
-    if (data.byteLength > missing) {
-      overflow = data.subarray(missing)
-      data = data.subarray(0, missing)
+      if (entryStream) {
+        resolve({ value: entryStream, done: false })
+        entryStream = null
+        return
+      }
+
+      promiseResolve = resolve
+      promiseReject = reject
+
+      consumeCallback(null)
+
+      if (extract._finished && promiseResolve) {
+        promiseResolve({ value: undefined, done: true })
+        promiseResolve = promiseReject = null
+      }
     }
 
-    if (s) s.end(data)
-    else b.append(data)
+    function onentry (header, stream, callback) {
+      entryCallback = callback
+      stream.on('error', noop) // no way around this due to tick sillyness
 
-    this._overflow = overflow
-    this._onparse()
-  }
+      if (promiseResolve) {
+        promiseResolve({ value: stream, done: false })
+        promiseResolve = promiseReject = null
+      } else {
+        entryStream = stream
+      }
+    }
 
-  _final (cb) {
-    cb(this._partial ? new Error('Unexpected end of data') : null)
+    function onclose () {
+      consumeCallback(error)
+      if (!promiseResolve) return
+      if (error) promiseReject(error)
+      else promiseResolve({ value: undefined, done: true })
+      promiseResolve = promiseReject = null
+    }
+
+    function destroy (err) {
+      extract.destroy(err)
+      consumeCallback(err)
+      return new Promise((resolve, reject) => {
+        if (extract.destroyed) return resolve({ value: undefined, done: true })
+        extract.once('close', function () {
+          if (err) reject(err)
+          else resolve({ value: undefined, done: true })
+        })
+      })
+    }
   }
 }
 
 module.exports = function extract (opts) {
   return new Extract(opts)
 }
+
+function noop () {}
+
+function overflow (size) {
+  size &= 511
+  return size && 512 - size
+}
diff --git a/headers.js b/headers.js
index a51f0bb..56c3e9b 100644
--- a/headers.js
+++ b/headers.js
@@ -3,15 +3,161 @@ const b4a = require('b4a')
 const ZEROS = '0000000000000000000'
 const SEVENS = '7777777777777777777'
 const ZERO_OFFSET = '0'.charCodeAt(0)
-const USTAR_MAGIC = b4a.from('ustar\x00', 'binary')
-const USTAR_VER = b4a.from('00', 'binary')
-const GNU_MAGIC = b4a.from('ustar\x20', 'binary')
-const GNU_VER = b4a.from('\x20\x00', 'binary')
+const USTAR_MAGIC = b4a.from([0x75, 0x73, 0x74, 0x61, 0x72, 0x00]) // ustar\x00
+const USTAR_VER = b4a.from([ZERO_OFFSET, ZERO_OFFSET])
+const GNU_MAGIC = b4a.from([0x75, 0x73, 0x74, 0x61, 0x72, 0x20]) // ustar\x20
+const GNU_VER = b4a.from([0x20, 0x00])
 const MASK = 0o7777
 const MAGIC_OFFSET = 257
 const VERSION_OFFSET = 263
 
-const clamp = function (index, len, defaultValue) {
+exports.decodeLongPath = function decodeLongPath (buf, encoding) {
+  return decodeStr(buf, 0, buf.length, encoding)
+}
+
+exports.encodePax = function encodePax (opts) { // TODO: encode more stuff in pax
+  let result = ''
+  if (opts.name) result += addLength(' path=' + opts.name + '\n')
+  if (opts.linkname) result += addLength(' linkpath=' + opts.linkname + '\n')
+  const pax = opts.pax
+  if (pax) {
+    for (const key in pax) {
+      result += addLength(' ' + key + '=' + pax[key] + '\n')
+    }
+  }
+  return b4a.from(result)
+}
+
+exports.decodePax = function decodePax (buf) {
+  const result = {}
+
+  while (buf.length) {
+    let i = 0
+    while (i < buf.length && buf[i] !== 32) i++
+    const len = parseInt(buf.subarray(0, i).toString(), 10)
+    if (!len) return result
+
+    const b = b4a.toString(buf.subarray(i + 1, len - 1))
+    const keyIndex = b.indexOf('=')
+    if (keyIndex === -1) return result
+    result[b.slice(0, keyIndex)] = b.slice(keyIndex + 1)
+
+    buf = buf.subarray(len)
+  }
+
+  return result
+}
+
+exports.encode = function encode (opts) {
+  const buf = b4a.alloc(512)
+  let name = opts.name
+  let prefix = ''
+
+  if (opts.typeflag === 5 && name[name.length - 1] !== '/') name += '/'
+  if (b4a.byteLength(name) !== name.length) return null // utf-8
+
+  while (b4a.byteLength(name) > 100) {
+    const i = name.indexOf('/')
+    if (i === -1) return null
+    prefix += prefix ? '/' + name.slice(0, i) : name.slice(0, i)
+    name = name.slice(i + 1)
+  }
+
+  if (b4a.byteLength(name) > 100 || b4a.byteLength(prefix) > 155) return null
+  if (opts.linkname && b4a.byteLength(opts.linkname) > 100) return null
+
+  b4a.write(buf, name)
+  b4a.write(buf, encodeOct(opts.mode & MASK, 6), 100)
+  b4a.write(buf, encodeOct(opts.uid, 6), 108)
+  b4a.write(buf, encodeOct(opts.gid, 6), 116)
+  encodeSize(opts.size, buf, 124)
+  b4a.write(buf, encodeOct((opts.mtime.getTime() / 1000) | 0, 11), 136)
+
+  buf[156] = ZERO_OFFSET + toTypeflag(opts.type)
+
+  if (opts.linkname) b4a.write(buf, opts.linkname, 157)
+
+  b4a.copy(USTAR_MAGIC, buf, MAGIC_OFFSET)
+  b4a.copy(USTAR_VER, buf, VERSION_OFFSET)
+  if (opts.uname) b4a.write(buf, opts.uname, 265)
+  if (opts.gname) b4a.write(buf, opts.gname, 297)
+  b4a.write(buf, encodeOct(opts.devmajor || 0, 6), 329)
+  b4a.write(buf, encodeOct(opts.devminor || 0, 6), 337)
+
+  if (prefix) b4a.write(buf, prefix, 345)
+
+  b4a.write(buf, encodeOct(cksum(buf), 6), 148)
+
+  return buf
+}
+
+exports.decode = function decode (buf, filenameEncoding, allowUnknownFormat) {
+  let typeflag = buf[156] === 0 ? 0 : buf[156] - ZERO_OFFSET
+
+  let name = decodeStr(buf, 0, 100, filenameEncoding)
+  const mode = decodeOct(buf, 100, 8)
+  const uid = decodeOct(buf, 108, 8)
+  const gid = decodeOct(buf, 116, 8)
+  const size = decodeOct(buf, 124, 12)
+  const mtime = decodeOct(buf, 136, 12)
+  const type = toType(typeflag)
+  const linkname = buf[157] === 0 ? null : decodeStr(buf, 157, 100, filenameEncoding)
+  const uname = decodeStr(buf, 265, 32)
+  const gname = decodeStr(buf, 297, 32)
+  const devmajor = decodeOct(buf, 329, 8)
+  const devminor = decodeOct(buf, 337, 8)
+
+  const c = cksum(buf)
+
+  // checksum is still initial value if header was null.
+  if (c === 8 * 32) return null
+
+  // valid checksum
+  if (c !== decodeOct(buf, 148, 8)) throw new Error('Invalid tar header. Maybe the tar is corrupted or it needs to be gunzipped?')
+
+  if (isUSTAR(buf)) {
+    // ustar (posix) format.
+    // prepend prefix, if present.
+    if (buf[345]) name = decodeStr(buf, 345, 155, filenameEncoding) + '/' + name
+  } else if (isGNU(buf)) {
+    // 'gnu'/'oldgnu' format. Similar to ustar, but has support for incremental and
+    // multi-volume tarballs.
+  } else {
+    if (!allowUnknownFormat) {
+      throw new Error('Invalid tar header: unknown format.')
+    }
+  }
+
+  // to support old tar versions that use trailing / to indicate dirs
+  if (typeflag === 0 && name && name[name.length - 1] === '/') typeflag = 5
+
+  return {
+    name,
+    mode,
+    uid,
+    gid,
+    size,
+    mtime: new Date(1000 * mtime),
+    type,
+    linkname,
+    uname,
+    gname,
+    devmajor,
+    devminor,
+    pax: null
+  }
+}
+
+function isUSTAR (buf) {
+  return b4a.equals(USTAR_MAGIC, buf.subarray(MAGIC_OFFSET, MAGIC_OFFSET + 6))
+}
+
+function isGNU (buf) {
+  return b4a.equals(GNU_MAGIC, buf.subarray(MAGIC_OFFSET, MAGIC_OFFSET + 6)) &&
+    b4a.equals(GNU_VER, buf.subarray(VERSION_OFFSET, VERSION_OFFSET + 2))
+}
+
+function clamp (index, len, defaultValue) {
   if (typeof index !== 'number') return defaultValue
   index = ~~index // Coerce to integer.
   if (index >= len) return len
@@ -21,7 +167,7 @@ const clamp = function (index, len, defaultValue) {
   return 0
 }
 
-const toType = function (flag) {
+function toType (flag) {
   switch (flag) {
     case 0:
       return 'file'
@@ -53,7 +199,7 @@ const toType = function (flag) {
   return null
 }
 
-const toTypeflag = function (flag) {
+function toTypeflag (flag) {
   switch (flag) {
     case 'file':
       return 0
@@ -78,24 +224,40 @@ const toTypeflag = function (flag) {
   return 0
 }
 
-const indexOf = function (block, num, offset, end) {
+function indexOf (block, num, offset, end) {
   for (; offset < end; offset++) {
     if (block[offset] === num) return offset
   }
   return end
 }
 
-const cksum = function (block) {
+function cksum (block) {
   let sum = 8 * 32
   for (let i = 0; i < 148; i++) sum += block[i]
   for (let j = 156; j < 512; j++) sum += block[j]
   return sum
 }
 
-const encodeOct = function (val, n) {
+function encodeOct (val, n) {
   val = val.toString(8)
   if (val.length > n) return SEVENS.slice(0, n) + ' '
-  else return ZEROS.slice(0, n - val.length) + val + ' '
+  return ZEROS.slice(0, n - val.length) + val + ' '
+}
+
+function encodeSizeBin (num, buf, off) {
+  buf[off] = 0x80
+  for (let i = 11; i > 0; i--) {
+    buf[off + i] = num & 0xff
+    num = Math.floor(num / 0x100)
+  }
+}
+
+function encodeSize (num, buf, off) {
+  if (num.toString(8).length > 11) {
+    encodeSizeBin(num, buf, off)
+  } else {
+    b4a.write(buf, encodeOct(num, 11), off)
+  }
 }
 
 /* Copied from the node-tar repo and modified to meet
@@ -129,8 +291,8 @@ function parse256 (buf) {
   return positive ? sum : -1 * sum
 }
 
-const decodeOct = function (val, offset, length) {
-  val = val.slice(offset, offset + length)
+function decodeOct (val, offset, length) {
+  val = val.subarray(offset, offset + length)
   offset = 0
 
   // If prefixed with 0x80 then parse as a base-256 integer
@@ -142,155 +304,18 @@ const decodeOct = function (val, offset, length) {
     const end = clamp(indexOf(val, 32, offset, val.length), val.length, val.length)
     while (offset < end && val[offset] === 0) offset++
     if (end === offset) return 0
-    return parseInt(val.slice(offset, end).toString(), 8)
+    return parseInt(val.subarray(offset, end).toString(), 8)
   }
 }
 
-const decodeStr = function (val, offset, length, encoding) {
-  return val.slice(offset, indexOf(val, 0, offset, offset + length)).toString(encoding)
+function decodeStr (val, offset, length, encoding) {
+  return b4a.toString(val.subarray(offset, indexOf(val, 0, offset, offset + length)), encoding)
 }
 
-const addLength = function (str) {
+function addLength (str) {
   const len = b4a.byteLength(str)
   let digits = Math.floor(Math.log(len) / Math.log(10)) + 1
   if (len + digits >= Math.pow(10, digits)) digits++
 
   return (len + digits) + str
 }
-
-exports.decodeLongPath = function (buf, encoding) {
-  return decodeStr(buf, 0, buf.length, encoding)
-}
-
-exports.encodePax = function (opts) { // TODO: encode more stuff in pax
-  let result = ''
-  if (opts.name) result += addLength(' path=' + opts.name + '\n')
-  if (opts.linkname) result += addLength(' linkpath=' + opts.linkname + '\n')
-  const pax = opts.pax
-  if (pax) {
-    for (const key in pax) {
-      result += addLength(' ' + key + '=' + pax[key] + '\n')
-    }
-  }
-  return b4a.from(result)
-}
-
-exports.decodePax = function (buf) {
-  const result = {}
-
-  while (buf.length) {
-    let i = 0
-    while (i < buf.length && buf[i] !== 32) i++
-    const len = parseInt(buf.slice(0, i).toString(), 10)
-    if (!len) return result
-
-    const b = buf.slice(i + 1, len - 1).toString()
-    const keyIndex = b.indexOf('=')
-    if (keyIndex === -1) return result
-    result[b.slice(0, keyIndex)] = b.slice(keyIndex + 1)
-
-    buf = buf.slice(len)
-  }
-
-  return result
-}
-
-exports.encode = function (opts) {
-  const buf = b4a.alloc(512)
-  let name = opts.name
-  let prefix = ''
-
-  if (opts.typeflag === 5 && name[name.length - 1] !== '/') name += '/'
-  if (b4a.byteLength(name) !== name.length) return null // utf-8
-
-  while (b4a.byteLength(name) > 100) {
-    const i = name.indexOf('/')
-    if (i === -1) return null
-    prefix += prefix ? '/' + name.slice(0, i) : name.slice(0, i)
-    name = name.slice(i + 1)
-  }
-
-  if (b4a.byteLength(name) > 100 || b4a.byteLength(prefix) > 155) return null
-  if (opts.linkname && b4a.byteLength(opts.linkname) > 100) return null
-
-  b4a.write(buf, name)
-  b4a.write(buf, encodeOct(opts.mode & MASK, 6), 100)
-  b4a.write(buf, encodeOct(opts.uid, 6), 108)
-  b4a.write(buf, encodeOct(opts.gid, 6), 116)
-  b4a.write(buf, encodeOct(opts.size, 11), 124)
-  b4a.write(buf, encodeOct((opts.mtime.getTime() / 1000) | 0, 11), 136)
-
-  buf[156] = ZERO_OFFSET + toTypeflag(opts.type)
-
-  if (opts.linkname) b4a.write(buf, opts.linkname, 157)
-
-  b4a.copy(USTAR_MAGIC, buf, MAGIC_OFFSET)
-  b4a.copy(USTAR_VER, buf, VERSION_OFFSET)
-  if (opts.uname) b4a.write(buf, opts.uname, 265)
-  if (opts.gname) b4a.write(buf, opts.gname, 297)
-  b4a.write(buf, encodeOct(opts.devmajor || 0, 6), 329)
-  b4a.write(buf, encodeOct(opts.devminor || 0, 6), 337)
-
-  if (prefix) b4a.write(buf, prefix, 345)
-
-  b4a.write(buf, encodeOct(cksum(buf), 6), 148)
-
-  return buf
-}
-
-exports.decode = function (buf, filenameEncoding, allowUnknownFormat) {
-  let typeflag = buf[156] === 0 ? 0 : buf[156] - ZERO_OFFSET
-
-  let name = decodeStr(buf, 0, 100, filenameEncoding)
-  const mode = decodeOct(buf, 100, 8)
-  const uid = decodeOct(buf, 108, 8)
-  const gid = decodeOct(buf, 116, 8)
-  const size = decodeOct(buf, 124, 12)
-  const mtime = decodeOct(buf, 136, 12)
-  const type = toType(typeflag)
-  const linkname = buf[157] === 0 ? null : decodeStr(buf, 157, 100, filenameEncoding)
-  const uname = decodeStr(buf, 265, 32)
-  const gname = decodeStr(buf, 297, 32)
-  const devmajor = decodeOct(buf, 329, 8)
-  const devminor = decodeOct(buf, 337, 8)
-
-  const c = cksum(buf)
-
-  // checksum is still initial value if header was null.
-  if (c === 8 * 32) return null
-
-  // valid checksum
-  if (c !== decodeOct(buf, 148, 8)) throw new Error('Invalid tar header. Maybe the tar is corrupted or it needs to be gunzipped?')
-
-  if (USTAR_MAGIC.compare(buf, MAGIC_OFFSET, MAGIC_OFFSET + 6) === 0) {
-    // ustar (posix) format.
-    // prepend prefix, if present.
-    if (buf[345]) name = decodeStr(buf, 345, 155, filenameEncoding) + '/' + name
-  } else if (GNU_MAGIC.compare(buf, MAGIC_OFFSET, MAGIC_OFFSET + 6) === 0 &&
-             GNU_VER.compare(buf, VERSION_OFFSET, VERSION_OFFSET + 2) === 0) {
-    // 'gnu'/'oldgnu' format. Similar to ustar, but has support for incremental and
-    // multi-volume tarballs.
-  } else {
-    if (!allowUnknownFormat) {
-      throw new Error('Invalid tar header: unknown format.')
-    }
-  }
-
-  // to support old tar versions that use trailing / to indicate dirs
-  if (typeflag === 0 && name && name[name.length - 1] === '/') typeflag = 5
-
-  return {
-    name,
-    mode,
-    uid,
-    gid,
-    size,
-    mtime: new Date(1000 * mtime),
-    type,
-    linkname,
-    uname,
-    gname,
-    devmajor,
-    devminor
-  }
-}
diff --git a/pack.js b/pack.js
index 9debe0e..b6824c0 100644
--- a/pack.js
+++ b/pack.js
@@ -1,8 +1,7 @@
-const { constants } = require('fs')
-const { Readable, Writable } = require('streamx')
-const { StringDecoder } = require('string_decoder')
+const { Readable, Writable, getStreamError } = require('streamx')
 const b4a = require('b4a')
 
+const constants = require('./constants')
 const headers = require('./headers')
 
 const DMODE = 0o755
@@ -10,55 +9,97 @@ const FMODE = 0o644
 
 const END_OF_TAR = b4a.alloc(1024)
 
-const noop = function () {}
+class Sink extends Writable {
+  constructor (pack, header, callback) {
+    super({ mapWritable })
 
-const overflow = function (self, size) {
-  size &= 511
-  if (size) self.push(END_OF_TAR.subarray(0, 512 - size))
-}
+    this.written = 0
+    this.header = header
+
+    this._callback = callback
+    this._linkname = null
+    this._isLinkname = header.type === 'symlink' && !header.linkname
+    this._isVoid = header.type !== 'file' && header.type !== 'contiguous-file'
+    this._finished = false
+    this._pack = pack
+    this._openCallback = null
+
+    if (this._pack._stream === null) this._pack._stream = this
+    else this._pack._pending.push(this)
+  }
 
-function modeToType (mode) {
-  switch (mode & constants.S_IFMT) {
-    case constants.S_IFBLK: return 'block-device'
-    case constants.S_IFCHR: return 'character-device'
-    case constants.S_IFDIR: return 'directory'
-    case constants.S_IFIFO: return 'fifo'
-    case constants.S_IFLNK: return 'symlink'
+  _open (cb) {
+    this._openCallback = cb
+    if (this._pack._stream === this) this._continueOpen()
   }
 
-  return 'file'
-}
+  _continueOpen () {
+    if (this._pack._stream === null) this._pack._stream = this
 
-class Sink extends Writable {
-  constructor (to) {
-    super()
-    this.written = 0
-    this._to = to
+    const cb = this._openCallback
+    this._openCallback = null
+    if (cb === null) return
+
+    if (this._pack.destroying) return cb(new Error('pack stream destroyed'))
+    if (this._pack._finalized) return cb(new Error('pack stream is already finalized'))
+
+    this._pack._stream = this
+
+    if (!this._isLinkname) {
+      this._pack._encode(this.header)
+    }
+
+    cb(null)
   }
 
   _write (data, cb) {
+    if (this._isLinkname) {
+      this._linkname = this._linkname ? b4a.concat([this._linkname, data]) : data
+      return cb(null)
+    }
+
+    if (this._isVoid) {
+      return cb(new Error('No body allowed for this entry'))
+    }
+
     this.written += data.byteLength
-    if (this._to.push(data)) return cb()
-    this._to._drain = cb
+    if (this._pack.push(data)) return cb()
+    this._pack._drain = cb
   }
-}
 
-class LinkSink extends Writable {
-  constructor () {
-    super()
-    this.linkname = ''
-    this._decoder = new StringDecoder('utf-8')
+  _final (cb) {
+    if (this._isLinkname) {
+      this.header.linkname = this._linkname ? b4a.toString(this._linkname, 'utf-8') : ''
+      this._pack._encode(this.header)
+    }
+
+    overflow(this._pack, this.header.size)
+
+    if (this.written !== this.header.size) { // corrupting tar
+      return cb(new Error('Size mismatch'))
+    }
+
+    this._pack._done(this)
+    this._finished = true
+
+    cb(null)
   }
 
-  _write (data, cb) {
-    this.linkname += this._decoder.write(data)
-    cb()
+  _getError () {
+    return getStreamError(this) || new Error('tar entry destroyed')
   }
-}
 
-class Void extends Writable {
-  _write (data, cb) {
-    cb(new Error('No body allowed for this entry'))
+  _predestroy () {
+    this._pack.destroy(this._getError())
+  }
+
+  _destroy (cb) {
+    this._pack._done(this)
+
+    if (this._finished) this._callback(null)
+    else this._callback(this._getError())
+
+    cb()
   }
 }
 
@@ -68,12 +109,12 @@ class Pack extends Readable {
     this._drain = noop
     this._finalized = false
     this._finalizing = false
+    this._pending = []
     this._stream = null
   }
 
   entry (header, buffer, callback) {
-    if (this._stream) throw new Error('already piping an entry')
-    if (this._finalized || this.destroyed) return
+    if (this._finalized || this.destroying) throw new Error('already finalized or destroyed')
 
     if (typeof buffer === 'function') {
       callback = buffer
@@ -82,8 +123,6 @@ class Pack extends Readable {
 
     if (!callback) callback = noop
 
-    const self = this
-
     if (!header.size || header.type === 'symlink') header.size = 0
     if (!header.type) header.type = modeToType(header.mode)
     if (!header.mode) header.mode = header.type === 'directory' ? DMODE : FMODE
@@ -92,74 +131,46 @@ class Pack extends Readable {
     if (!header.mtime) header.mtime = new Date()
 
     if (typeof buffer === 'string') buffer = b4a.from(buffer)
+
+    const sink = new Sink(this, header, callback)
+
     if (b4a.isBuffer(buffer)) {
       header.size = buffer.byteLength
-      this._encode(header)
-      const ok = this.push(buffer)
-      overflow(self, header.size)
-      if (ok) process.nextTick(callback)
-      else this._drain = callback
-      return new Void()
-    }
-
-    if (header.type === 'symlink' && !header.linkname) {
-      const linkSink = new LinkSink()
-      linkSink
-        .on('error', function (err) {
-          self.destroy()
-          callback(err)
-        })
-        .on('close', function () {
-          header.linkname = linkSink.linkname
-          self._encode(header)
-          callback()
-        })
-
-      return linkSink
+      sink.write(buffer)
+      sink.end()
+      return sink
     }
 
-    this._encode(header)
-
-    if (header.type !== 'file' && header.type !== 'contiguous-file') {
-      process.nextTick(callback)
-      return new Void()
+    if (sink._isVoid) {
+      sink.end()
+      return sink
     }
 
-    const sink = new Sink(this)
-    sink
-      .on('error', function (err) {
-        self._stream = null
-        self.destroy()
-        callback(err)
-      })
-      .on('close', function () {
-        self._stream = null
-
-        if (sink.written !== header.size) { // corrupting tar
-        }
-
-        overflow(self, header.size)
-        if (self._finalizing) { self.finalize() }
-        callback()
-      })
-
-    this._stream = sink
-
     return sink
   }
 
   finalize () {
-    if (this._stream) {
+    if (this._stream || this._pending.length > 0) {
       this._finalizing = true
       return
     }
 
     if (this._finalized) return
     this._finalized = true
+
     this.push(END_OF_TAR)
     this.push(null)
   }
 
+  _done (stream) {
+    if (stream !== this._stream) return
+
+    this._stream = null
+
+    if (this._finalizing) this.finalize()
+    if (this._pending.length) this._pending.shift()._continueOpen()
+  }
+
   _encode (header) {
     if (!header.pax) {
       const buf = headers.encode(header)
@@ -202,10 +213,28 @@ class Pack extends Readable {
     this.push(headers.encode(newHeader))
   }
 
-  _read (cb) {
+  _doDrain () {
     const drain = this._drain
     this._drain = noop
     drain()
+  }
+
+  _predestroy () {
+    const err = getStreamError(this)
+
+    if (this._stream) this._stream.destroy(err)
+
+    while (this._pending.length) {
+      const stream = this._pending.shift()
+      stream.destroy(err)
+      stream._continueOpen()
+    }
+
+    this._doDrain()
+  }
+
+  _read (cb) {
+    this._doDrain()
     cb()
   }
 }
@@ -213,3 +242,26 @@ class Pack extends Readable {
 module.exports = function pack (opts) {
   return new Pack(opts)
 }
+
+function modeToType (mode) {
+  switch (mode & constants.S_IFMT) {
+    case constants.S_IFBLK: return 'block-device'
+    case constants.S_IFCHR: return 'character-device'
+    case constants.S_IFDIR: return 'directory'
+    case constants.S_IFIFO: return 'fifo'
+    case constants.S_IFLNK: return 'symlink'
+  }
+
+  return 'file'
+}
+
+function noop () {}
+
+function overflow (self, size) {
+  size &= 511
+  if (size) self.push(END_OF_TAR.subarray(0, 512 - size))
+}
+
+function mapWritable (buf) {
+  return b4a.isBuffer(buf) ? buf : b4a.from(buf)
+}
diff --git a/package.json b/package.json
index 9bdcf93..f0fdb80 100644
--- a/package.json
+++ b/package.json
@@ -6,6 +6,9 @@
   "files": [
     "*.js"
   ],
+  "browser": {
+    "fs": false
+  },
   "scripts": {
     "test": "standard && brittle test/*.js"
   },
@@ -20,13 +23,12 @@
   },
   "homepage": "https://github.com/mafintosh/tar-stream",
   "dependencies": {
-    "b4a": "^1.6.1",
-    "bl": "^6.0.0",
-    "streamx": "^2.12.5"
+    "b4a": "^1.6.4",
+    "streamx": "^2.15.0"
   },
   "devDependencies": {
-    "brittle": "^3.1.1",
+    "brittle": "^3.3.2",
     "concat-stream": "^2.0.0",
-    "standard": "^17.0.0"
+    "standard": "^17.0.1"
   }
 }
diff --git a/test/dual.js b/test/dual.js
new file mode 100644
index 0000000..ec4df5f
--- /dev/null
+++ b/test/dual.js
@@ -0,0 +1,48 @@
+const test = require('brittle')
+const { Readable } = require('streamx')
+const tar = require('../')
+
+test('write and read huge archive', function (t) {
+  t.plan(2)
+
+  const pack = tar.pack()
+  const extract = tar.extract()
+
+  extract.on('entry', function (header, stream, next) {
+    let size = 0
+
+    stream.on('data', function (data) {
+      size += data.byteLength
+    })
+
+    stream.on('end', function () {
+      t.is(size, header.size)
+      next()
+    })
+  })
+
+  pack.pipe(extract, function (err) {
+    t.ok(!err, 'pipeline finished')
+  })
+
+  const entry = pack.entry({
+    name: 'huge.txt',
+    size: 10 * 1024 * 1024 * 1024
+  })
+
+  const buf = Buffer.alloc(1024 * 1024)
+
+  let pushed = 0
+
+  const rs = new Readable({
+    read (cb) {
+      this.push(buf)
+      pushed += buf.byteLength
+      if (pushed === entry.header.size) this.push(null)
+      cb(null)
+    }
+  })
+
+  rs.pipe(entry)
+  pack.finalize()
+})
diff --git a/test/extract.js b/test/extract.js
index 7a963be..d4b6604 100644
--- a/test/extract.js
+++ b/test/extract.js
@@ -4,16 +4,6 @@ const fs = require('fs')
 const tar = require('..')
 const fixtures = require('./fixtures')
 
-const clamp = function (index, len, defaultValue) {
-  if (typeof index !== 'number') return defaultValue
-  index = ~~index // Coerce to integer.
-  if (index >= len) return len
-  if (index >= 0) return index
-  index += len
-  if (index >= 0) return index
-  return 0
-}
-
 test('one-file', function (t) {
   t.plan(3)
 
@@ -33,7 +23,8 @@ test('one-file', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -69,7 +60,8 @@ test('chunked-one-file', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -97,7 +89,15 @@ test('multi-file', function (t) {
   const extract = tar.extract()
   let noEntries = false
 
-  const onfile1 = function (header, stream, cb) {
+  extract.once('entry', onfile1)
+
+  extract.on('finish', function () {
+    t.ok(noEntries)
+  })
+
+  extract.end(fs.readFileSync(fixtures.MULTI_FILE_TAR))
+
+  function onfile1 (header, stream, cb) {
     t.alike(header, {
       name: 'file-1.txt',
       mode: 0o644,
@@ -110,7 +110,8 @@ test('multi-file', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     extract.on('entry', onfile2)
@@ -120,7 +121,7 @@ test('multi-file', function (t) {
     }))
   }
 
-  const onfile2 = function (header, stream, cb) {
+  function onfile2 (header, stream, cb) {
     t.alike(header, {
       name: 'file-2.txt',
       mode: 0o644,
@@ -133,7 +134,8 @@ test('multi-file', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -142,14 +144,6 @@ test('multi-file', function (t) {
       cb()
     }))
   }
-
-  extract.once('entry', onfile1)
-
-  extract.on('finish', function () {
-    t.ok(noEntries)
-  })
-
-  extract.end(fs.readFileSync(fixtures.MULTI_FILE_TAR))
 })
 
 test('chunked-multi-file', function (t) {
@@ -158,7 +152,19 @@ test('chunked-multi-file', function (t) {
   const extract = tar.extract()
   let noEntries = false
 
-  const onfile1 = function (header, stream, cb) {
+  extract.once('entry', onfile1)
+
+  extract.on('finish', function () {
+    t.ok(noEntries)
+  })
+
+  const b = fs.readFileSync(fixtures.MULTI_FILE_TAR)
+  for (let i = 0; i < b.length; i += 321) {
+    extract.write(b.subarray(i, clamp(i + 321, b.length, b.length)))
+  }
+  extract.end()
+
+  function onfile1 (header, stream, cb) {
     t.alike(header, {
       name: 'file-1.txt',
       mode: 0o644,
@@ -171,7 +177,8 @@ test('chunked-multi-file', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     extract.on('entry', onfile2)
@@ -181,7 +188,7 @@ test('chunked-multi-file', function (t) {
     }))
   }
 
-  const onfile2 = function (header, stream, cb) {
+  function onfile2 (header, stream, cb) {
     t.alike(header, {
       name: 'file-2.txt',
       mode: 0o644,
@@ -194,7 +201,8 @@ test('chunked-multi-file', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -203,18 +211,6 @@ test('chunked-multi-file', function (t) {
       cb()
     }))
   }
-
-  extract.once('entry', onfile1)
-
-  extract.on('finish', function () {
-    t.ok(noEntries)
-  })
-
-  const b = fs.readFileSync(fixtures.MULTI_FILE_TAR)
-  for (let i = 0; i < b.length; i += 321) {
-    extract.write(b.subarray(i, clamp(i + 321, b.length, b.length)))
-  }
-  extract.end()
 })
 
 test('pax', function (t) {
@@ -260,7 +256,15 @@ test('types', function (t) {
   const extract = tar.extract()
   let noEntries = false
 
-  const ondir = function (header, stream, cb) {
+  extract.once('entry', ondir)
+
+  extract.on('finish', function () {
+    t.ok(noEntries)
+  })
+
+  extract.end(fs.readFileSync(fixtures.TYPES_TAR))
+
+  function ondir (header, stream, cb) {
     t.alike(header, {
       name: 'directory',
       mode: 0o755,
@@ -273,7 +277,8 @@ test('types', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
     stream.on('data', function () {
       t.ok(false)
@@ -282,7 +287,7 @@ test('types', function (t) {
     cb()
   }
 
-  const onlink = function (header, stream, cb) {
+  function onlink (header, stream, cb) {
     t.alike(header, {
       name: 'directory-link',
       mode: 0o755,
@@ -295,7 +300,8 @@ test('types', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
     stream.on('data', function () {
       t.ok(false)
@@ -303,14 +309,6 @@ test('types', function (t) {
     noEntries = true
     cb()
   }
-
-  extract.once('entry', ondir)
-
-  extract.on('finish', function () {
-    t.ok(noEntries)
-  })
-
-  extract.end(fs.readFileSync(fixtures.TYPES_TAR))
 })
 
 test('long-name', function (t) {
@@ -332,7 +330,8 @@ test('long-name', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -363,13 +362,13 @@ test('unicode-bsd', function (t) { // can unpack a bsdtar unicoded tarball
       gid: 20,
       size: 4,
       mtime: new Date(1387588646000),
-      pax: { 'SCHILY.dev': '16777217', 'SCHILY.ino': '3599143', 'SCHILY.nlink': '1', atime: '1387589077', ctime: '1387588646', path: 'høllø.txt' },
       type: 'file',
       linkname: null,
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: { 'SCHILY.dev': '16777217', 'SCHILY.ino': '3599143', 'SCHILY.nlink': '1', atime: '1387589077', ctime: '1387588646', path: 'høllø.txt' }
     })
 
     stream.pipe(concat(function (data) {
@@ -400,13 +399,13 @@ test('unicode', function (t) { // can unpack a bsdtar unicoded tarball
       gid: 20,
       size: 8,
       mtime: new Date(1387580181000),
-      pax: { path: 'høstål.txt' },
       type: 'file',
       linkname: null,
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: { path: 'høstål.txt' }
     })
 
     stream.pipe(concat(function (data) {
@@ -522,7 +521,8 @@ test('base 256 size', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
     cb()
   })
@@ -554,7 +554,8 @@ test('latin-1', function (t) { // can unpack filenames encoded in latin-1
       uname: 'root',
       gname: 'root',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -610,7 +611,8 @@ test('gnu', function (t) { // can correctly unpack gnu-tar format
       uname: 'myuser',
       gname: 'mygroup',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -650,7 +652,8 @@ test('gnu-incremental', function (t) {
       uname: 'myuser',
       gname: 'mygroup',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -699,7 +702,15 @@ test('unknown format attempts to extract if allowed', function (t) {
   const extract = tar.extract({ allowUnknownFormat: true })
   let noEntries = false
 
-  const onfile1 = function (header, stream, cb) {
+  extract.once('entry', onfile1)
+
+  extract.on('finish', function () {
+    t.ok(noEntries)
+  })
+
+  extract.end(fs.readFileSync(fixtures.UNKNOWN_FORMAT))
+
+  function onfile1 (header, stream, cb) {
     t.alike(header, {
       name: 'file-1.txt',
       mode: 0o644,
@@ -712,7 +723,8 @@ test('unknown format attempts to extract if allowed', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     extract.on('entry', onfile2)
@@ -722,7 +734,7 @@ test('unknown format attempts to extract if allowed', function (t) {
     }))
   }
 
-  const onfile2 = function (header, stream, cb) {
+  function onfile2 (header, stream, cb) {
     t.alike(header, {
       name: 'file-2.txt',
       mode: 0o644,
@@ -735,7 +747,8 @@ test('unknown format attempts to extract if allowed', function (t) {
       uname: 'maf',
       gname: 'staff',
       devmajor: 0,
-      devminor: 0
+      devminor: 0,
+      pax: null
     })
 
     stream.pipe(concat(function (data) {
@@ -744,12 +757,30 @@ test('unknown format attempts to extract if allowed', function (t) {
       cb()
     }))
   }
+})
 
-  extract.once('entry', onfile1)
+test('extract streams are async iterators', async function (t) {
+  const extract = tar.extract()
+  const b = fs.readFileSync(fixtures.MULTI_FILE_TAR)
 
-  extract.on('finish', function () {
-    t.ok(noEntries)
-  })
+  extract.end(b)
 
-  extract.end(fs.readFileSync(fixtures.UNKNOWN_FORMAT))
+  const expected = ['file-1.txt', 'file-2.txt']
+
+  for await (const entry of extract) {
+    t.is(entry.header.name, expected.shift())
+    entry.resume()
+    t.comment('wait a bit...')
+    await new Promise(resolve => setTimeout(resolve, 100))
+  }
 })
+
+function clamp (index, len, defaultValue) {
+  if (typeof index !== 'number') return defaultValue
+  index = ~~index // Coerce to integer.
+  if (index >= len) return len
+  if (index >= 0) return index
+  index += len
+  if (index >= 0) return index
+  return 0
+}
diff --git a/test/slow/huge.js b/test/slow/huge.js
index 4039bfa..3480491 100644
--- a/test/slow/huge.js
+++ b/test/slow/huge.js
@@ -6,7 +6,7 @@ const tar = require('../..')
 const fixtures = require('../fixtures')
 
 test('huge', function (t) {
-  t.plan(1)
+  t.plan(3)
 
   const extract = tar.extract()
   let noEntries = false
@@ -48,6 +48,7 @@ test('huge', function (t) {
 
     noEntries = true
     stream.pipe(countStream)
+    callback()
   })
 
   extract.on('finish', function () {