print list of header candidates

[python-delta-tar] / deltatar / crypto.py

#!/usr/bin/env python3

"""
Intra2net 2017

===============================================================================
              crypto -- Encryption Layer for the Deltatar Backup
===============================================================================

Crypto stack:

    - AES-GCM for the symmetric encryption;
    - Scrypt as KDF.

References:

    - NIST Recommendation for Block Cipher Modes of Operation: Galois/Counter
      Mode (GCM) and GMAC
      http://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-38d.pdf

    - AES-GCM v1:
      https://cryptome.org/2014/01/aes-gcm-v1.pdf

    - Authentication weaknesses in GCM
      http://csrc.nist.gov/groups/ST/toolkit/BCM/documents/comments/CWC-GCM/Ferguson2.pdf

Trouble with python-cryptography packages: authentication tags can only be
passed in advance: https://github.com/pyca/cryptography/pull/3421

Errors
-------------------------------------------------------------------------------

Errors fall into roughly three categories:

    - Cryptographical errors or invalid data.

        - ``InvalidGCMTag`` (decryption failed on account of an invalid GCM
          tag),
        - ``InvalidIVFixedPart`` (IV fixed part of object not found in list),
        - ``DuplicateIV`` (the IV of an encrypted object already occurred),
        - ``DecryptionError`` (used in CLI decryption for presenting error
          conditions to the user).

    - Incorrect usage of the library.

        - ``InvalidParameter`` (non-conforming user supplied parameter),
        - ``InvalidHeader`` (data passed for reading not parsable into header),
        - ``FormatError`` (cannot handle header or parameter version),
        - ``RuntimeError``.

    - Bad internal state. If one of these is encountered it means that a state
      was reached that shouldn’t occur during normal processing.

        - ``InternalError``,
        - ``Unreachable``.

Also, ``EndOfFile`` is used as a sentinel to communicate that a stream supplied
for reading is exhausted.

Initialization Vectors
-------------------------------------------------------------------------------

Initialization vectors are checked reuse during the lifetime of a decryptor.
The fixed counters for metadata files cannot be reused and attempts to do so
will cause a DuplicateIV error. This means the length of objects encrypted with
a metadata counter is capped at 63 GB.

For ordinary, non-metadata payload, there is an optional mode with strict IV
checking that causes a crypto context to fail if an IV encountered or created
was already used for decrypting or encrypting, respectively, an earlier object.
Note that this mode can trigger false positives when decrypting non-linearly,
e. g. when traversing the same object multiple times. Since the crypto context
has no notion of a position in a PDT encrypted archive, this condition must be
sorted out downstream.

Command Line Utility
-------------------------------------------------------------------------------

``crypto.py`` may be invoked as a script for decrypting, validating, and
splitting PDT encrypted files. Consult the usage message for details.

Usage examples:

Decrypt from stdin using the password ‘foo’: ::

    $ crypto.py process foo -i - -o - <some-file.tar.gz.pdtcrypt >some-file.tar.gz

Output verbose information about the encrypted objects in the archive: ::

    $ crypto.py process foo -v -i some-file.tar.gz.pdtcrypt -o /dev/null
    PDT: decrypt from some-file.tar.gz.pdtcrypt
    PDT: decrypt to /dev/null
    PDT: source: file some-file.tar.gz.pdtcrypt
    PDT: sink: file /dev/null
    PDT: 0 hdr
    PDT:    · version         = 1                    : 0100
    PDT:    · paramversion    = 1                    : 0100
    PDT:    · nacl                                   : d270 b031 00d1 87e2 c946 610d 7b7f 7e5f
    PDT:    · iv                                     : 02ee 3dd7 a963 1eb1 0100 0000
    PDT:    · ctsize          = 591                  : 4f02 0000 0000 0000
    PDT:    · tag                                    : 5b2d 6d8b 8f82 4842 12fd 0b10 b6e3 369b
    PDT: 64 decrypt obj no. 1,  591 B
    PDT:    · [64] 0% done, read block (591 B of 591 B remaining)
    PDT:    · decrypt ciphertext 591 B
    PDT:    · decrypt plaintext 591 B
    PDT: 655 finalize
    …

Also, the mode *scrypt* allows deriving encryption keys. To calculate the
encryption key from the password ‘foo’ and the salt of the first object in a
PDT encrypted file: ::

    $ crypto.py scrypt foo -i some-file.pdtcrypt
    {"paramversion": 1, "salt": "Cqzbk48e3peEjzWto8D0yA==", "key": "JH9EkMwaM4x9F5aim5gK/Q=="}

The computed 16 byte key is given in hexadecimal notation in the value to
``hash`` and can be fed into Python’s ``binascii.unhexlify()`` to obtain the
corresponding binary representation.

Note that in Scrypt hashing mode, no data integrity checks are being performed.
If the wrong password is given, a wrong key will be derived. Whether the password
was indeed correct can only be determined by decrypting. Note that since PDT
archives essentially consist of a stream of independent objects, the salt and
other parameters may change. Thus a key derived using above method from the
first object doesn’t necessarily apply to any of the subsequent objects.

"""

import base64
import binascii
import bisect
import ctypes
import io
from functools import reduce, partial
import mmap
import os
import struct
import sys
import time
import types
try:
    import enum34
except ImportError as exn:
    pass

if __name__ == "__main__": ## Work around the import mechanism’s lest Python’s
    pwd = os.getcwd()      ## preference for local imports causes a cyclical
    ## import (crypto → pylibscrypt → […] → ./tarfile → crypto).
    sys.path = [ p for p in sys.path if p.find ("deltatar") < 0 ]

import pylibscrypt
from cryptography.hazmat.primitives.ciphers import Cipher, algorithms, modes
from cryptography.hazmat.backends import default_backend
import cryptography


__all__ = [ "hdr_make", "hdr_read", "hdr_fmt", "hdr_fmt_pretty"
          , "scrypt_hashfile"
          , "PDTCRYPT_HDR_SIZE", "AES_GCM_IV_CNT_DATA"
          , "AES_GCM_IV_CNT_INFOFILE", "AES_GCM_IV_CNT_INDEX"
          ]


###############################################################################
## exceptions
###############################################################################

class EndOfFile (Exception):
    """Reached EOF."""
    remainder = 0
    msg       = 0
    def __init__ (self, n=None, msg=None):
        if n is not None:
            self.remainder = n
        self.msg = msg


class InvalidParameter (Exception):
    """Inputs not valid for PDT encryption."""
    pass


class InvalidHeader (Exception):
    """Header not valid."""
    pass


class InvalidGCMTag (Exception):
    """
    The GCM tag calculated during decryption differs from that in the object
    header.
    """
    pass


class InvalidIVFixedPart (Exception):
    """
    IV fixed part not in supplied list: either the backup is corrupt or the
    current object does not belong to it.
    """
    pass


class IVFixedPartError (Exception):
    """
    Error creating a unique IV fixed part: repeated calls to system RNG yielded
    the same sequence of bytes as the last IV used.
    """
    pass


class InvalidFileCounter (Exception):
    """
    When encrypting, an attempted reuse of a dedicated counter (info file,
    index file) was caught.
    """
    pass


class DuplicateIV (Exception):
    """
    During encryption, the current IV fixed part is identical to an already
    existing IV (same prefix and file counter). This indicates tampering or
    programmer error and cannot be recovered from.
    """
    pass


class NonConsecutiveIV (Exception):
    """
    IVs not numbered consecutively. This is a hard error with strict IV
    checking. Precludes random access to the encrypted objects.
    """
    pass


class FormatError (Exception):
    """Unusable parameters in header."""
    pass


class DecryptionError (Exception):
    """Error during decryption with ``crypto.py`` on the command line."""
    pass


class Unreachable (Exception):
    """
    Makeshift __builtin_unreachable(); always a programmer error if
    thrown.
    """
    pass


class InternalError (Exception):
    """Errors not ascribable to bad user inputs or cryptography."""
    pass


###############################################################################
## crypto layer version
###############################################################################

ENCRYPTION_PARAMETERS = \
    { 0: \
        { "kdf": ("dummy", 16)
        , "enc": "passthrough" }
    , 1: \
        { "kdf": ( "scrypt"
                 , { "dkLen"    : 16
                   , "N"        : 1 << 16
                   , "r"        : 8
                   , "p"        : 1
                   , "NaCl_LEN" : 16 })
        , "enc": "aes-gcm" } }

###############################################################################
## constants
###############################################################################

PDTCRYPT_HDR_MAGIC = b"PDTCRYPT"

PDTCRYPT_HDR_SIZE_MAGIC          = 8     #  8
PDTCRYPT_HDR_SIZE_VERSION        = 2     # 10
PDTCRYPT_HDR_SIZE_PARAMVERSION   = 2     # 12
PDTCRYPT_HDR_SIZE_NACL           = 16    # 28
PDTCRYPT_HDR_SIZE_IV             = 12    # 40
PDTCRYPT_HDR_SIZE_CTSIZE         = 8     # 48
PDTCRYPT_HDR_SIZE_TAG            = 16    # 64 GCM auth tag

PDTCRYPT_HDR_SIZE = PDTCRYPT_HDR_SIZE_MAGIC        + PDTCRYPT_HDR_SIZE_VERSION \
                  + PDTCRYPT_HDR_SIZE_PARAMVERSION + PDTCRYPT_HDR_SIZE_NACL    \
                  + PDTCRYPT_HDR_SIZE_IV           + PDTCRYPT_HDR_SIZE_CTSIZE  \
                  + PDTCRYPT_HDR_SIZE_TAG # = 64

# precalculate offsets since Python can’t do constant folding over names
HDR_OFF_VERSION      = PDTCRYPT_HDR_SIZE_MAGIC
HDR_OFF_PARAMVERSION = HDR_OFF_VERSION      + PDTCRYPT_HDR_SIZE_VERSION
HDR_OFF_NACL         = HDR_OFF_PARAMVERSION + PDTCRYPT_HDR_SIZE_PARAMVERSION
HDR_OFF_IV           = HDR_OFF_NACL         + PDTCRYPT_HDR_SIZE_NACL
HDR_OFF_CTSIZE       = HDR_OFF_IV           + PDTCRYPT_HDR_SIZE_IV
HDR_OFF_TAG          = HDR_OFF_CTSIZE       + PDTCRYPT_HDR_SIZE_CTSIZE

FMT_UINT16_LE = "<H"
FMT_UINT64_LE = "<Q"
FMT_I2N_IV    = "<8sL"   # 8 random bytes ‖ 32 bit counter
FMT_I2N_HDR   = ("<"     # host byte order
                 "8s"    # magic
                 "H"     # version
                 "H"     # paramversion
                 "16s"   # sodium chloride
                 "12s"   # iv
                 "Q"     # size
                 "16s")  # GCM tag

# aes+gcm
AES_GCM_MAX_SIZE              = (1 << 36) - (1 << 5) # 2^39 - 2^8 b ≅ 64 GB
PDTCRYPT_MAX_OBJ_SIZE_DEFAULT = 63 * (1 << 30)       #                63 GB
PDTCRYPT_MAX_OBJ_SIZE         = PDTCRYPT_MAX_OBJ_SIZE_DEFAULT

# index and info files are written on-the fly while encrypting so their
# counters must be available inadvance
AES_GCM_IV_CNT_INFOFILE     = 1 # constant
AES_GCM_IV_CNT_INDEX        = AES_GCM_IV_CNT_INFOFILE + 1
AES_GCM_IV_CNT_DATA         = AES_GCM_IV_CNT_INDEX    + 1 # also for multivolume
AES_GCM_IV_CNT_MAX_DEFAULT  = 0xffFFffFF
AES_GCM_IV_CNT_MAX          = AES_GCM_IV_CNT_MAX_DEFAULT

# IV structure and generation
PDTCRYPT_IV_GEN_MAX_RETRIES = 10 # ×
PDTCRYPT_IV_FIXEDPART_SIZE  =  8 # B
PDTCRYPT_IV_COUNTER_SIZE    =  4 # B

###############################################################################
## header, trailer
###############################################################################
#
# Interface:
#
#    struct hdrinfo
#      { version      : u16
#      , paramversion : u16
#      , nacl         : [u8; 16]
#      , iv           : [u8; 12]
#      , ctsize       : usize
#      , tag          : [u8; 16] }
#
#    fn hdr_read (f : handle) -> hdrinfo;
#    fn hdr_make (f : handle, h : hdrinfo) -> IOResult<usize>;
#    fn hdr_fmt (h : hdrinfo) -> String;
#

def hdr_read (data):
    """
    Read bytes as header structure.

    If the input could not be interpreted as a header, fail with
    ``InvalidHeader``.
    """

    try:
        mag, version, paramversion, nacl, iv, ctsize, tag = \
            struct.unpack (FMT_I2N_HDR, data)
    except Exception as exn:
        raise InvalidHeader ("error unpacking header from [%r]: %s"
                             % (binascii.hexlify (data), str (exn)))

    if mag != PDTCRYPT_HDR_MAGIC:
        raise InvalidHeader ("bad magic in header: expected [%s], got [%s]"
                             % (PDTCRYPT_HDR_MAGIC, mag))

    return \
        {      "version" : version
        , "paramversion" : paramversion
        ,         "nacl" : nacl
        ,           "iv" : iv
        ,       "ctsize" : ctsize
        ,          "tag" : tag
        }


def hdr_read_stream (instr):
    """
    Read header from stream at the current position.

    Fail with ``InvalidHeader`` if insufficient bytes were read from the
    stream, or if the content could not be interpreted as a header.
    """
    data = instr.read(PDTCRYPT_HDR_SIZE)
    ldata = len (data)
    if ldata == 0:
        raise EndOfFile
    elif ldata != PDTCRYPT_HDR_SIZE:
        raise InvalidHeader ("hdr_read_stream: expected %d B, received %d B"
                             % (PDTCRYPT_HDR_SIZE, ldata))
    return hdr_read (data)


def hdr_from_params (version, paramversion, nacl, iv, ctsize, tag):
    """
    Assemble the necessary values into a PDTCRYPT header.

    :type      version: int to fit   uint16_t
    :type paramversion: int to fit   uint16_t
    :type         nacl: bytes to fit uint8_t[16]
    :type           iv: bytes to fit uint8_t[12]
    :type         size: int to fit   uint64_t
    :type          tag: bytes to fit uint8_t[16]
    """
    buf  = bytearray (PDTCRYPT_HDR_SIZE)
    bufv = memoryview (buf)

    try:
        struct.pack_into (FMT_I2N_HDR, bufv, 0,
                          PDTCRYPT_HDR_MAGIC,
                          version, paramversion, nacl, iv, ctsize, tag)
    except Exception as exn:
        return False, "error assembling header: %s" % str (exn)

    return True, bytes (buf)


def hdr_make_dummy (s):
    """
    Create a header sized block of bytes initialized to a value derived from a
    string. Used to verify we’ve jumped back correctly to the actual position
    of the object header.
    """
    c = reduce (lambda a, c: a + ord(c), s, 0) % 0xFF
    return bytes (bytearray (struct.pack ("B", c)) * PDTCRYPT_HDR_SIZE)


def hdr_make (hdr):
    """
    Assemble a header from the given header structure.
    """
    return hdr_from_params (version=hdr.get("version"),
                            paramversion=hdr.get("paramversion"),
                            nacl=hdr.get("nacl"), iv=hdr.get("iv"),
                            ctsize=hdr.get("ctsize"), tag=hdr.get("tag"))


HDR_FMT = "I2n_header { version: %d, paramversion: %d, nacl: %s[%d]," \
                      " iv: %s[%d], ctsize: %d, tag: %s[%d] }"

def hdr_fmt (h):
    """Format a header structure into readable output."""
    return HDR_FMT % (h["version"], h["paramversion"],
                      binascii.hexlify (h["nacl"]), len(h["nacl"]),
                      binascii.hexlify (h["iv"]), len(h["iv"]),
                      h["ctsize"],
                      binascii.hexlify (h["tag"]), len(h["tag"]))


def hex_spaced_of_bytes (b):
    """Format bytes object, hexdump style."""
    return " ".join ([ "%.2x%.2x" % (c1, c2)
                       for c1, c2 in zip (b[0::2], b[1::2]) ]) \
         + (len (b) | 1 == len (b) and " %.2x" % b[-1] or "") # odd lengths


def hdr_iv_counter (h):
    """Extract the variable part of the IV of the given header."""
    _fixed, cnt = struct.unpack (FMT_I2N_IV, h ["iv"])
    return cnt


def hdr_iv_fixed (h):
    """Extract the fixed part of the IV of the given header."""
    fixed, _cnt = struct.unpack (FMT_I2N_IV, h ["iv"])
    return fixed


hdr_dump = hex_spaced_of_bytes


HDR_FMT_PRETTY = \
"""version         = %-4d                 : %s
paramversion    = %-4d                 : %s
nacl                                   : %s
iv                                     : %s
ctsize          = %-20d : %s
tag                                    : %s
"""

def hdr_fmt_pretty (h):
    """
    Format header structure into multi-line representation of its contents and
    their raw representation. (Omit the implicit “PDTCRYPT” magic bytes that
    precede every header.)
    """
    return HDR_FMT_PRETTY \
                % (h["version"],
                   hex_spaced_of_bytes (struct.pack (FMT_UINT16_LE, h["version"])),
                   h["paramversion"],
                   hex_spaced_of_bytes (struct.pack (FMT_UINT16_LE, h["paramversion"])),
                   hex_spaced_of_bytes (h["nacl"]),
                   hex_spaced_of_bytes (h["iv"]),
                   h["ctsize"],
                   hex_spaced_of_bytes (struct.pack (FMT_UINT64_LE, h["ctsize"])),
                   hex_spaced_of_bytes (h["tag"]))

IV_FMT = "((f %s) (c %d))"

def iv_fmt (iv):
    """Format the two components of an IV in a readable fashion."""
    fixed, cnt = struct.unpack (FMT_I2N_IV, iv)
    return IV_FMT % (binascii.hexlify (fixed), cnt)


###############################################################################
## restoration
###############################################################################

class Location (object):
    n = 0
    offset = 0

def restore_loc_fmt (loc):
    return "%d off:%d" \
        % (loc.n, loc.offset)

def locate_hdr_candidates (fd):
    """
    Walk over instances of the magic string in the payload, collecting their
    positions. If the offset of the first found instance is not zero, the file
    begins with leading garbage.

    :return:    The list of offsets in the file.
    """
    cands = []

    mm = mmap.mmap(fd, 0, mmap.MAP_SHARED, mmap.PROT_READ)
    pos = 0
    while True:
        pos = mm.find (PDTCRYPT_HDR_MAGIC, pos)
        if pos == -1:
            break
        cands.append (pos)
        pos += 1

    return cands


###############################################################################
## passthrough / null encryption
###############################################################################

class PassthroughCipher (object):

    tag = struct.pack ("<QQ", 0, 0)

    def __init__          (self)    : pass

    def update            (self, b) : return b

    def finalize          (self)    : return b""

    def finalize_with_tag (self, _) : return b""

###############################################################################
## convenience wrapper
###############################################################################


def kdf_dummy (klen, password, _nacl):
    """
    Fake KDF for testing purposes that is called when parameter version zero is
    encountered.
    """
    q, r = divmod (klen, len (password))
    if isinstance (password, bytes) is False:
        password = password.encode ()
    return password * q + password [:r], b""


SCRYPT_KEY_MEMO = { } # static because needed for both the info file and the archive


def kdf_scrypt (params, password, nacl):
    """
    Wrapper for the Scrypt KDF, corresponds to parameter version one. The
    computation result is memoized based on the inputs to facilitate spawning
    multiple encryption contexts.
    """
    N = params["N"]
    r = params["r"]
    p = params["p"]
    dkLen = params["dkLen"]

    if nacl is None:
        nacl = os.urandom (params["NaCl_LEN"])

    key_parms = (password, nacl, N, r, p, dkLen)
    global SCRYPT_KEY_MEMO
    if key_parms not in SCRYPT_KEY_MEMO:
        SCRYPT_KEY_MEMO [key_parms] = \
            pylibscrypt.scrypt (password, nacl, N, r, p, dkLen)
    return SCRYPT_KEY_MEMO [key_parms], nacl


def kdf_by_version (paramversion=None, defs=None):
    """
    Pick the KDF handler corresponding to the parameter version or the
    definition set.

    :rtype: function (password : str, nacl : str) -> str
    """
    if paramversion is not None:
        defs = ENCRYPTION_PARAMETERS.get(paramversion, None)
    if defs is None:
        raise InvalidParameter ("no encryption parameters for version %r"
                                % paramversion)
    (kdf, params) = defs["kdf"]
    fn = None
    if kdf == "scrypt" : fn = kdf_scrypt
    if kdf == "dummy"  : fn = kdf_dummy
    if fn is None:
        raise ValueError ("key derivation method %r unknown" % kdf)
    return partial (fn, params)


###############################################################################
## SCRYPT hashing
###############################################################################

def scrypt_hashsource (pw, ins):
    """
    Calculate the SCRYPT hash from the password and the information contained
    in the first header found in ``ins``.

    This does not validate whether the first object is encrypted correctly.
    """
    if isinstance (pw, str) is True:
        pw = str.encode (pw)
    elif isinstance (pw, bytes) is False:
        raise InvalidParameter ("password must be a string, not %s"
                                % type (password))
    if isinstance (ins, io.BufferedReader) is False and \
            isinstance (ins, io.FileIO) is False:
        raise InvalidParameter ("file to hash must be opened in “binary” mode")
    hdr = None
    try:
        hdr = hdr_read_stream (ins)
    except EndOfFile as exn:
        noise ("PDT: malformed input: end of file reading first object header")
        noise ("PDT:")
        return 1

    nacl = hdr ["nacl"]
    pver = hdr ["paramversion"]
    if PDTCRYPT_VERBOSE is True:
        noise ("PDT: salt of first object          : %s" % binascii.hexlify (nacl))
        noise ("PDT: parameter version of archive  : %d" % pver)

    try:
        defs = ENCRYPTION_PARAMETERS.get(pver, None)
        kdfname, params = defs ["kdf"]
        if kdfname != "scrypt":
            noise ("PDT: input is not an SCRYPT archive")
            noise ("")
            return 1
        kdf = kdf_by_version (None, defs)
    except ValueError as exn:
        noise ("PDT: object has unknown parameter version %d" % pver)

    hsh, _void = kdf (pw, nacl)

    return hsh, nacl, hdr ["version"], pver


def scrypt_hashfile (pw, fname):
    """
    Calculate the SCRYPT hash from the password and the information contained
    in the first header found in the given file. The header is read only at
    offset zero.
    """
    with deptdcrypt_mk_stream (PDTCRYPT_SOURCE, fname  or "-") as ins:
        hsh, _void, _void, _void = scrypt_hashsource (pw, ins)
        return hsh


###############################################################################
## AES-GCM context
###############################################################################

class Crypto (object):
    """
    Encryption context to remain alive throughout an entire tarfile pass.
    """
    enc  = None
    nacl = None
    key  = None
    cnt  = None # file counter (uint32_t != 0)
    iv   = None # current IV
    fixed        = None  # accu for 64 bit fixed parts of IV
    used_ivs     = None  # tracks IVs
    strict_ivs   = False # if True, panic on duplicate object IV
    password     = None
    paramversion = None
    stats = { "in"  : 0
            , "out" : 0
            , "obj" : 0 }

    ctsize  = -1
    ptsize  = -1
    info_counter_used  = False
    index_counter_used = False

    def __init__ (self, *al, **akv):
        self.used_ivs = set ()
        self.set_parameters (*al, **akv)


    def next_fixed (self):
        # NOP for decryption
        pass


    def set_object_counter (self, cnt=None):
        """
        Safely set the internal counter of encrypted objects. Numerous
        constraints apply:

        The same counter may not be reused in combination with one IV fixed
        part. This is validated elsewhere in the IV handling.

        Counter zero is invalid. The first two counters are reserved for
        metadata. The implementation does not allow for splitting metadata
        files over multiple encrypted objects. (This would be possible by
        assigning new fixed parts.) Thus in a Deltatar backup there is at most
        one object with a counter value of one and two. On creation of a
        context, the initial counter may be chosen. The globals
        ``AES_GCM_IV_CNT_INFOFILE`` and ``AES_GCM_IV_CNT_INDEX`` can be used to
        request one of the reserved values. If one of these values has been
        used, any further attempt of setting the counter to that value will
        be rejected with an ``InvalidFileCounter`` exception.

        Out of bounds values (i. e. below one and more than the maximum of 2³²)
        cause an ``InvalidParameter`` exception to be thrown.
        """
        if cnt is None:
            self.cnt = AES_GCM_IV_CNT_DATA
            return
        if cnt == 0 or cnt > AES_GCM_IV_CNT_MAX + 1:
            raise InvalidParameter ("invalid counter value %d requested: "
                                    "acceptable values are from 1 to %d"
                                    % (cnt, AES_GCM_IV_CNT_MAX))
        if cnt == AES_GCM_IV_CNT_INFOFILE:
            if self.info_counter_used is True:
                raise InvalidFileCounter ("attempted to reuse info file "
                                          "counter %d: must be unique" % cnt)
            self.info_counter_used = True
        elif cnt == AES_GCM_IV_CNT_INDEX:
            if self.index_counter_used is True:
                raise InvalidFileCounter ("attempted to reuse index file "
                                          " counter %d: must be unique" % cnt)
            self.index_counter_used = True
        if cnt <= AES_GCM_IV_CNT_MAX:
            self.cnt = cnt
            return
        # cnt == AES_GCM_IV_CNT_MAX + 1 → wrap
        self.cnt = AES_GCM_IV_CNT_DATA
        self.next_fixed ()


    def set_parameters (self, password=None, key=None, paramversion=None,
                        nacl=None, counter=None, strict_ivs=False):
        """
        Configure the internal state of a crypto context. Not intended for
        external use.
        """
        self.next_fixed ()
        self.set_object_counter (counter)
        self.strict_ivs = strict_ivs

        if paramversion is not None:
            self.paramversion = paramversion

        if key is not None:
            self.key, self.nacl = key, nacl
            return

        if password is not None:
            if isinstance (password, bytes) is False:
                password = str.encode (password)
            self.password = password
            if paramversion is None and nacl is None:
                # postpone key setup until first header is available
                return
            kdf = kdf_by_version (paramversion)
            if kdf is not None:
                self.key, self.nacl = kdf (password, nacl)


    def process (self, buf):
        """
        Encrypt / decrypt a buffer. Invokes the ``.update()`` method on the
        wrapped encryptor or decryptor, respectively.

        The Cryptography exception ``AlreadyFinalized`` is translated to an
        ``InternalError`` at this point. It may occur in sound code when the GC
        closes an encrypting stream after an error. Everywhere else it must be
        treated as a bug.
        """
        if self.enc is None:
            raise RuntimeError ("process: context not initialized")
        self.stats ["in"] += len (buf)
        try:
            out = self.enc.update (buf)
        except cryptography.exceptions.AlreadyFinalized as exn:
            raise InternalError (exn)
        self.stats ["out"] += len (out)
        return out


    def next (self, password, paramversion, nacl, iv):
        """
        Prepare for encrypting another object: Reset the data counters and
        change the configuration in case one of the variable parameters differs
        from the last object. Also check the IV for duplicates and error out
        if strict checking was requested.
        """
        self.ctsize = 0
        self.ptsize = 0
        self.stats ["obj"] += 1

        self.check_duplicate_iv (iv)

        if (   self.paramversion != paramversion
            or self.password     != password
            or self.nacl         != nacl):
            self.set_parameters (password=password, paramversion=paramversion,
                                 nacl=nacl, strict_ivs=self.strict_ivs)


    def check_duplicate_iv (self, iv):
        """
        Add an IV (the 12 byte representation as in the header) to the list. With
        strict checking enabled, this will throw a ``DuplicateIV``. Depending on
        the context, this may indicate a serious error (IV reuse).
        """
        if self.strict_ivs is True and iv in self.used_ivs:
            raise DuplicateIV ("iv %s was reused" % iv_fmt (iv))
        # vi has not been used before; add to collection
        self.used_ivs.add (iv)


    def counters (self):
        """
        Access the data counters.
        """
        return self.stats ["obj"], self.stats ["in"], self.stats ["out"]


    def drop (self):
        """
        Clear the current context regardless of its finalization state. The
        next operation must be ``.next()``.
        """
        self.enc = None


class Encrypt (Crypto):

    lastinfo     = None
    version      = None
    paramenc     = None

    def __init__ (self, version, paramversion, password=None, key=None, nacl=None,
                  counter=AES_GCM_IV_CNT_DATA, strict_ivs=True):
        """
        The ctor will throw immediately if one of the parameters does not conform
        to our expectations.

                  counter=AES_GCM_IV_CNT_DATA, strict_ivs=True):
        :type      version: int to fit   uint16_t
        :type paramversion: int to fit   uint16_t
        :param    password: mutually exclusive with ``key``
        :type     password: bytes
        :param         key: mutually exclusive with ``password``
        :type          key: bytes
        :type         nacl: bytes
        :type      counter: initial object counter the values
                            ``AES_GCM_IV_CNT_INFOFILE`` and
                            ``AES_GCM_IV_CNT_INDEX`` are unique in each backup set
                            and cannot be reused even with different fixed parts.
        :type   strict_ivs: bool
        """
        if         password is     None and key is     None \
                or password is not None and key is not None :
            raise InvalidParameter ("__init__: need either key or password")

        if key is not None:
            if isinstance (key, bytes) is False:
                raise InvalidParameter ("__init__: key must be provided as "
                                        "bytes, not %s" % type (key))
            if nacl is None:
                raise InvalidParameter ("__init__: salt must be provided along "
                                        "with encryption key")
        else: # password, no key
            if isinstance (password, str) is False:
                raise InvalidParameter ("__init__: password must be a string, not %s"
                                        % type (password))
            if len (password) == 0:
                raise InvalidParameter ("__init__: supplied empty password but not "
                                        "permitted for PDT encrypted files")
        # version
        if isinstance (version, int) is False:
            raise InvalidParameter ("__init__: version number must be an "
                                    "integer, not %s" % type (version))
        if version < 0:
            raise InvalidParameter ("__init__: version number must be a "
                                    "nonnegative integer, not %d" % version)
        # paramversion
        if isinstance (paramversion, int) is False:
            raise InvalidParameter ("__init__: crypto parameter version number "
                                    "must be an integer, not %s"
                                    % type (paramversion))
        if paramversion < 0:
            raise InvalidParameter ("__init__: crypto parameter version number "
                                    "must be a nonnegative integer, not %d"
                                    % paramversion)
        # salt
        if nacl is not None:
            if isinstance (nacl, bytes) is False:
                raise InvalidParameter ("__init__: salt given, but of type %s "
                                        "instead of bytes" % type (nacl))
            # salt length would depend on the actual encryption so it can’t be
            # validated at this point
        self.fixed        = [ ]
        self.version      = version
        self.paramenc     = ENCRYPTION_PARAMETERS.get (paramversion) ["enc"]

        super().__init__ (password, key, paramversion, nacl, counter=counter,
                          strict_ivs=strict_ivs)


    def next_fixed (self, retries=PDTCRYPT_IV_GEN_MAX_RETRIES):
        """
        Generate the next IV fixed part by reading eight bytes from
        ``/dev/urandom``. The buffer so obtained is tested against the fixed
        parts used so far to prevent accidental reuse of IVs. After a
        configurable number of attempts to create a unique fixed part, it will
        refuse to continue with an ``IVFixedPartError``. This is unlikely to
        ever happen on a normal system but may detect an issue with the random
        generator.

        The list of fixed parts that were used by the context at hand can be
        accessed through the ``.fixed`` list. Its last element is the fixed
        part currently in use.
        """
        i = 0
        while i < retries:
            fp = os.urandom (PDTCRYPT_IV_FIXEDPART_SIZE)
            if fp not in self.fixed:
                self.fixed.append (fp)
                return
            i += 1
        raise IVFixedPartError ("error obtaining a unique IV fixed part from "
                                "/dev/urandom; giving up after %d tries" % i)


    def iv_make (self):
        """
        Construct a 12-bytes IV from the current fixed part and the object
        counter.
        """
        return struct.pack(FMT_I2N_IV, self.fixed [-1], self.cnt)


    def next (self, filename=None, counter=None):
        """
        Prepare for encrypting the next incoming object. Update the counter
        and put together the IV, possibly changing prefixes. Then create the
        new encryptor.

        The argument ``counter`` can be used to specify a file counter for this
        object. Unless it is one of the reserved values, the counter of
        subsequent objects will be computed from this one.

        If this is the first object in a series, ``filename`` is required,
        otherwise it is reused if not present. The value is used to derive a
        header sized placeholder to use until after encryption when all the
        inputs to construct the final header are available. This is then
        matched in ``.done()`` against the value found at the position of the
        header. The motivation for this extra check is primarily to assist
        format debugging: It makes stray headers easy to spot in malformed
        PDTCRYPT files.
        """
        if filename is None:
            if self.lastinfo is None:
                raise InvalidParameter ("next: filename is mandatory for "
                                        "first object")
            filename, _dummy = self.lastinfo
        else:
            if isinstance (filename, str) is False:
                raise InvalidParameter ("next: filename must be a string, no %s"
                                        % type (filename))
        if counter is not None:
            if isinstance (counter, int) is False:
                raise InvalidParameter ("next: the supplied counter is of "
                                        "invalid type %s; please pass an "
                                        "integer instead" % type (counter))
            self.set_object_counter (counter)

        self.iv = self.iv_make ()
        if self.paramenc == "aes-gcm":
            self.enc = Cipher \
                            ( algorithms.AES (self.key)
                            , modes.GCM (self.iv)
                            , backend = default_backend ()) \
                            .encryptor ()
        elif self.paramenc == "passthrough":
            self.enc = PassthroughCipher ()
        else:
            raise InvalidParameter ("next: parameter version %d not known"
                                    % self.paramversion)
        hdrdum = hdr_make_dummy (filename)
        self.lastinfo = (filename, hdrdum)
        super().next (self.password, self.paramversion, self.nacl, self.iv)

        self.set_object_counter (self.cnt + 1)
        return hdrdum


    def done (self, cmpdata):
        """
        Complete encryption of an object. After this has been called, attempts
        of encrypting further data will cause an error until ``.next()`` is
        invoked properly.

        Returns a 64 bytes buffer containing the object header including all
        values including the “late” ones e. g. the ciphertext size and the
        GCM tag.
        """
        if isinstance (cmpdata, bytes) is False:
            raise InvalidParameter ("done: comparison input expected as bytes, "
                                    "not %s" % type (cmpdata))
        if self.lastinfo is None:
            raise RuntimeError ("done: encryption context not initialized")
        filename, hdrdum = self.lastinfo
        if cmpdata != hdrdum:
            raise RuntimeError ("done: bad sync of header for object %d: "
                                "preliminary data does not match; this likely "
                                "indicates a wrongly repositioned stream"
                                % self.cnt)
        data = self.enc.finalize ()
        self.stats ["out"] += len (data)
        self.ctsize += len (data)
        ok, hdr = hdr_from_params (self.version, self.paramversion, self.nacl,
                                   self.iv, self.ctsize, self.enc.tag)
        if ok is False:
            raise InternalError ("error constructing header: %r" % hdr)
        return data, hdr, self.fixed


    def process (self, buf):
        """
        Encrypt a chunk of plaintext with the active encryptor. Returns the
        size of the input consumed. This **must** be checked downstream. If the
        maximum possible object size has been reached, the current context must
        be finalized and a new one established before any further data can be
        encrypted. The second argument is the remainder of the plaintext that
        was not encrypted for the caller to use immediately after the new
        context is ready.
        """
        if isinstance (buf, bytes) is False:
            raise InvalidParameter ("process: expected byte buffer, not %s"
                                    % type (buf))
        bsize = len (buf)
        newptsize = self.ptsize + bsize
        diff = newptsize - PDTCRYPT_MAX_OBJ_SIZE
        if diff > 0:
            bsize -= diff
            newptsize = PDTCRYPT_MAX_OBJ_SIZE
        self.ptsize = newptsize
        data = super().process (buf [:bsize])
        self.ctsize += len (data)
        return bsize, data


class Decrypt (Crypto):

    tag        = None   # GCM tag, part of header
    last_iv    = None   # check consecutive ivs in strict mode

    def __init__ (self, password=None, key=None, counter=None, fixedparts=None,
                  strict_ivs=False):
        """
        Sanitizing ctor for the decryption context. ``fixedparts`` specifies a
        list of IV fixed parts accepted during decryption. If a fixed part is
        encountered that is not in the list, decryption will fail.

        :param    password: mutually exclusive with ``key``
        :type     password: bytes
        :param         key: mutually exclusive with ``password``
        :type          key: bytes
        :type      counter: initial object counter the values
                            ``AES_GCM_IV_CNT_INFOFILE`` and
                            ``AES_GCM_IV_CNT_INDEX`` are unique in each backup set
                            and cannot be reused even with different fixed parts.
        :type   fixedparts: bytes list
        """
        if         password is     None and key is     None \
                or password is not None and key is not None :
            raise InvalidParameter ("__init__: need either key or password")

        if key is not None:
            if isinstance (key, bytes) is False:
                raise InvalidParameter ("__init__: key must be provided as "
                                        "bytes, not %s" % type (key))
        else: # password, no key
            if isinstance (password, str) is False:
                raise InvalidParameter ("__init__: password must be a string, not %s"
                                        % type (password))
            if len (password) == 0:
                raise InvalidParameter ("__init__: supplied empty password but not "
                                        "permitted for PDT encrypted files")
        # fixed parts
        if fixedparts is not None:
            if isinstance (fixedparts, list) is False:
                raise InvalidParameter ("__init__: IV fixed parts must be "
                                        "supplied as list, not %s"
                                        % type (fixedparts))
            self.fixed = fixedparts
            self.fixed.sort ()

        super().__init__ (password=password, key=key, counter=counter,
                          strict_ivs=strict_ivs)


    def valid_fixed_part (self, iv):
        """
        Check if a fixed part was already seen.
        """
        # check if fixed part is known
        fixed, _cnt = struct.unpack (FMT_I2N_IV, iv)
        i = bisect.bisect_left (self.fixed, fixed)
        return i != len (self.fixed) and self.fixed [i] == fixed


    def check_consecutive_iv (self, iv):
        """
        Check whether the counter part of the given IV is indeed the successor
        of the currently present counter. This should always be the case for
        the objects in a well formed PDT archive but should not be enforced
        when decrypting out-of-order.
        """
        fixed, cnt = struct.unpack (FMT_I2N_IV, iv)
        if self.strict_ivs is True \
                and self.last_iv is not None \
                and self.last_iv [0] == fixed \
                and self.last_iv [1] != cnt - 1:
            raise NonConsecutiveIV ("iv %s counter not successor of "
                                    "last object (expected %d, found %d)"
                                    % (iv_fmt (self.last_iv [1]), cnt))
        self.last_iv = (iv, cnt)


    def next (self, hdr):
        """
        Start decrypting the next object. The PDTCRYPT header for the object
        can be given either as already parsed object or as bytes.
        """
        if isinstance (hdr, bytes) is True:
            hdr = hdr_read (hdr)
        elif isinstance (hdr, dict) is False:
            # this won’t catch malformed specs though
            raise InvalidParameter ("next: wrong type of parameter hdr: "
                                    "expected bytes or spec, got %s"
                                    % type (hdr))
        try:
            paramversion = hdr ["paramversion"]
            nacl         = hdr ["nacl"]
            iv           = hdr ["iv"]
            tag          = hdr ["tag"]
        except KeyError:
            raise InvalidHeader ("next: not a header %r" % hdr)

        super().next (self.password, paramversion, nacl, iv)
        if self.fixed is not None and self.valid_fixed_part (iv) is False:
            raise InvalidIVFixedPart ("iv %s has invalid fixed part"
                                      % iv_fmt (iv))
        self.check_consecutive_iv (iv)

        self.tag = tag
        defs = ENCRYPTION_PARAMETERS.get (paramversion, None)
        if defs is None:
            raise FormatError ("header contains unknown parameter version %d; "
                               "maybe the file was created by a more recent "
                               "version of Deltatar" % paramversion)
        enc = defs ["enc"]
        if enc == "aes-gcm":
            self.enc = Cipher \
                            ( algorithms.AES (self.key)
                            , modes.GCM (iv, tag=self.tag)
                            , backend = default_backend ()) \
                            . decryptor ()
        elif enc == "passthrough":
            self.enc = PassthroughCipher ()
        else:
            raise InternalError ("encryption parameter set %d refers to unknown "
                                 "mode %r" % (paramversion, enc))
        self.set_object_counter (self.cnt + 1)


    def done (self, tag=None):
        """
        Stop decryption of the current object and finalize it with the active
        context. This will throw an *InvalidGCMTag* exception to indicate that
        the authentication tag does not match the data. If the tag is correct,
        the rest of the plaintext is returned.
        """
        data = b""
        try:
            if tag is None:
                data = self.enc.finalize ()
            else:
                if isinstance (tag, bytes) is False:
                    raise InvalidParameter ("done: wrong type of parameter "
                                            "tag: expected bytes, got %s"
                                            % type (tag))
                data = self.enc.finalize_with_tag (self.tag)
        except cryptography.exceptions.InvalidTag:
            raise InvalidGCMTag ("done: tag mismatch of object %d: %s "
                                  "rejected by finalize ()"
                                  % (self.cnt, binascii.hexlify (self.tag)))
        self.ctsize += len (data)
        self.stats ["out"] += len (data)
        return data


    def process (self, buf):
        """
        Decrypt the bytes object *buf* with the active decryptor.
        """
        if isinstance (buf, bytes) is False:
            raise InvalidParameter ("process: expected byte buffer, not %s"
                                    % type (buf))
        self.ctsize += len (buf)
        data = super().process (buf)
        self.ptsize += len (data)
        return data


###############################################################################
## testing helpers
###############################################################################

def _patch_global (glob, vow, n=None):
    """
    Adapt upper file counter bound for testing IV logic. Completely unsafe.
    """
    assert vow == "I am fully aware that this will void my warranty."
    r = globals () [glob]
    if n is None:
        n = globals () [glob + "_DEFAULT"]
    globals () [glob] = n
    return r

_testing_set_AES_GCM_IV_CNT_MAX = \
        partial (_patch_global, "AES_GCM_IV_CNT_MAX")

_testing_set_PDTCRYPT_MAX_OBJ_SIZE = \
        partial (_patch_global, "PDTCRYPT_MAX_OBJ_SIZE")

###############################################################################
## freestanding invocation
###############################################################################

PDTCRYPT_SUB_PROCESS = 0
PDTCRYPT_SUB_SCRYPT  = 1
PDTCRYPT_SUB_SCAN    = 2

PDTCRYPT_SUB = \
        { "process" : PDTCRYPT_SUB_PROCESS
        , "scrypt"  : PDTCRYPT_SUB_SCRYPT
        , "scan"    : PDTCRYPT_SUB_SCAN }

PDTCRYPT_SECRET_PW   = 0
PDTCRYPT_SECRET_KEY  = 1

PDTCRYPT_DECRYPT   = 1 << 0 # decrypt archive with password
PDTCRYPT_SPLIT     = 1 << 1 # split archive into individual objects
PDTCRYPT_HASH      = 1 << 2 # output scrypt hash for file and given password

PDTCRYPT_SPLITNAME = "pdtcrypt-object-%d.bin"

PDTCRYPT_VERBOSE   = False
PDTCRYPT_STRICTIVS = False
PDTCRYPT_OVERWRITE = False
PDTCRYPT_BLOCKSIZE = 1 << 12
PDTCRYPT_SINK      = 0
PDTCRYPT_SOURCE    = 1
SELF               = None

PDTCRYPT_DEFAULT_VER  = 1
PDTCRYPT_DEFAULT_PVER = 1

# scrypt hashing output control
PDTCRYPT_SCRYPT_INTRANATOR = 0
PDTCRYPT_SCRYPT_PARAMETERS = 1
PDTCRYPT_SCRYPT_DEFAULT    = PDTCRYPT_SCRYPT_INTRANATOR

PDTCRYPT_SCRYPT_FORMAT = \
    { "i2n"    : PDTCRYPT_SCRYPT_INTRANATOR
    , "params" : PDTCRYPT_SCRYPT_PARAMETERS }

PDTCRYPT_TT_COLUMNS = 80 # assume standard terminal

class PDTDecryptionError (Exception):
    """Decryption failed."""

class PDTSplitError (Exception):
    """Decryption failed."""


def noise (*a, **b):
    print (file=sys.stderr, *a, **b)


class PassthroughDecryptor (object):

    curhdr = None # write current header on first data write

    def __init__ (self):
        if PDTCRYPT_VERBOSE is True:
            noise ("PDT: no encryption; data passthrough")

    def next (self, hdr):
        ok, curhdr = hdr_make (hdr)
        if ok is False:
            raise PDTDecryptionError ("bad header %r" % hdr)
        self.curhdr = curhdr

    def done (self):
        if self.curhdr is not None:
            return self.curhdr
        return b""

    def process (self, d):
        if self.curhdr is not None:
            d = self.curhdr + d
            self.curhdr = None
        return d


def depdtcrypt (mode, secret, ins, outs):
    """
    Remove PDTCRYPT layer from all objects encrypted with the secret. Used on a
    Deltatar backup this will yield a (possibly Gzip compressed) tarball.
    """
    ctleft     = -1              # length of ciphertext to consume
    ctcurrent  = 0               # total ciphertext of current object
    total_obj  = 0               # total number of objects read
    total_pt   = 0               # total plaintext bytes
    total_ct   = 0               # total ciphertext bytes
    total_read = 0               # total bytes read
    outfile    = None            # Python file object for output

    if mode & PDTCRYPT_DECRYPT:  # decryptor
        ks = secret [0]
        if ks == PDTCRYPT_SECRET_PW:
            decr = Decrypt (password=secret [1], strict_ivs=PDTCRYPT_STRICTIVS)
        elif ks == PDTCRYPT_SECRET_KEY:
            key = binascii.unhexlify (secret [1])
            decr = Decrypt (key=key, strict_ivs=PDTCRYPT_STRICTIVS)
        else:
            raise InternalError ("‘%d’ does not specify a valid kind of secret"
                                 % ks)
    else:
        decr = PassthroughDecryptor ()

    def nextout (_):
        """Dummy for non-split mode: output file does not vary."""
        return outs

    if mode & PDTCRYPT_SPLIT:
        def nextout (outfile):
            """
            We were passed an fd as outs for accessing the destination
            directory where extracted archive components are supposed
            to end up in.
            """

            if outfile is None:
                if PDTCRYPT_VERBOSE is True:
                    noise ("PDT: no output file to close at this point")
                else:
                    if PDTCRYPT_VERBOSE is True:
                        noise ("PDT: release output file %r" % outfile)
                    # cleanup happens automatically by the GC; the next
                    # line will error out on account of an invalid fd
                    #outfile.close ()

            assert total_obj > 0
            fname = PDTCRYPT_SPLITNAME % total_obj
            try:
                oflags = os.O_CREAT | os.O_WRONLY
                if PDTCRYPT_OVERWRITE is True:
                    oflags |= os.O_TRUNC
                else:
                    oflags |= os.O_EXCL
                outfd = os.open (fname, oflags, 0o600, dir_fd=outs)
                if PDTCRYPT_VERBOSE is True:
                    noise ("PDT: new output file %s → %d" % (fname, outfd))
            except FileExistsError as exn:
                noise ("PDT: refusing to overwrite existing file %s" % fname)
                noise ("")
                raise PDTSplitError ("destination file %s already exists"
                                     % fname)

            return os.fdopen (outfd, "wb", closefd=True)


    def tell (s):
        """ESPIPE is normal on non-seekable stdio stream."""
        try:
            return s.tell ()
        except OSError as exn:
            if exn.errno == os.errno.ESPIPE:
                return -1

    def out (pt, outfile):
        npt = len (pt)
        nonlocal total_pt
        total_pt += npt
        if PDTCRYPT_VERBOSE is True:
            noise ("PDT:\t· decrypt plaintext %d B" % (npt))
        try:
            nn = outfile.write (pt)
        except OSError as exn: # probably ENOSPC
            raise DecryptionError ("error (%s)" % exn)
        if nn != npt:
            raise DecryptionError ("write aborted after %d of %d B" % (nn, npt))

    while True:
        if ctleft <= 0:
            # current object completed; in a valid archive this marks either
            # the start of a new header or the end of the input
            if ctleft == 0: # current object requires finalization
                if PDTCRYPT_VERBOSE is True:
                    noise ("PDT: %d finalize" % tell (ins))
                try:
                    pt = decr.done ()
                except InvalidGCMTag as exn:
                    raise DecryptionError ("error finalizing object %d (%d B): "
                                           "%r" % (total_obj, len (pt), exn)) \
                          from exn
                out (pt, outfile)
                if PDTCRYPT_VERBOSE is True:
                    noise ("PDT:\t· object validated")

            if PDTCRYPT_VERBOSE is True:
                noise ("PDT: %d hdr" % tell (ins))
            try:
                hdr = hdr_read_stream (ins)
                total_read += PDTCRYPT_HDR_SIZE
            except EndOfFile as exn:
                total_read += exn.remainder
                if total_ct + total_obj * PDTCRYPT_HDR_SIZE != total_read:
                    raise PDTDecryptionError ("ciphertext processed (%d B) plus "
                                              "overhead (%d × %d B) does not match "
                                              "the number of bytes read (%d )"
                                              % (total_ct, total_obj, PDTCRYPT_HDR_SIZE,
                                                 total_read))
                # the single good exit
                return total_read, total_obj, total_ct, total_pt
            except InvalidHeader as exn:
                raise PDTDecryptionError ("invalid header at position %d in %r "
                                          "(%s)" % (tell (ins), exn, ins))
            if PDTCRYPT_VERBOSE is True:
                pretty = hdr_fmt_pretty (hdr)
                noise (reduce (lambda a, e: (a + "\n" if a else "") + "PDT:\t· " + e,
                               pretty.splitlines (), ""))
            ctcurrent = ctleft = hdr ["ctsize"]

            decr.next (hdr)

            total_obj += 1 # used in file counter with split mode

            # finalization complete or skipped in case of first object in
            # stream; create a new output file if necessary
            outfile = nextout (outfile)

            if PDTCRYPT_VERBOSE is True:
                noise ("PDT: %d decrypt obj no. %d,  %d B"
                       % (tell (ins), total_obj, ctleft))

        # always allocate a new buffer since python-cryptography doesn’t allow
        # passing a bytearray :/
        nexpect = min (ctleft, PDTCRYPT_BLOCKSIZE)
        if PDTCRYPT_VERBOSE is True:
            noise ("PDT:\t· [%d] %d%% done, read block (%d B of %d B remaining)"
                   % (tell (ins),
                      100 - ctleft * 100 / (ctcurrent > 0 and ctcurrent or 1),
                      nexpect, ctleft))
        ct      = ins.read (nexpect)
        nct     = len (ct)
        if nct < nexpect:
            off = tell (ins)
            raise EndOfFile (nct,
                             "hit EOF after %d of %d B in block [%d:%d); "
                             "%d B ciphertext remaining for object no %d"
                             % (nct, nexpect, off, off + nexpect, ctleft,
                                total_obj))
        ctleft     -= nct
        total_ct   += nct
        total_read += nct

        if PDTCRYPT_VERBOSE is True:
            noise ("PDT:\t· decrypt ciphertext %d B" % (nct))
        pt = decr.process (ct)
        out (pt, outfile)


def deptdcrypt_mk_stream (kind, path):
    """Create stream from file or stdio descriptor."""
    if kind == PDTCRYPT_SINK:
        if path == "-":
            if PDTCRYPT_VERBOSE is True: noise ("PDT: sink: stdout")
            return sys.stdout.buffer
        else:
            if PDTCRYPT_VERBOSE is True: noise ("PDT: sink: file %s" % path)
            return io.FileIO (path, "w")
    if kind == PDTCRYPT_SOURCE:
        if path == "-":
            if PDTCRYPT_VERBOSE is True: noise ("PDT: source: stdin")
            return sys.stdin.buffer
        else:
            if PDTCRYPT_VERBOSE is True: noise ("PDT: source: file %s" % path)
            return io.FileIO (path, "r")

    raise ValueError ("bogus stream “%s” / %s" % (kind, path))


def mode_depdtcrypt (mode, secret, ins, outs):
    try:
        total_read, total_obj, total_ct, total_pt = \
            depdtcrypt (mode, secret, ins, outs)
    except DecryptionError as exn:
        noise ("PDT: Decryption failed:")
        noise ("PDT:")
        noise ("PDT:    “%s”" % exn)
        noise ("PDT:")
        noise ("PDT: Did you specify the correct key / password?")
        noise ("")
        return 1
    except PDTSplitError as exn:
        noise ("PDT: Split operation failed:")
        noise ("PDT:")
        noise ("PDT:    “%s”" % exn)
        noise ("PDT:")
        noise ("PDT: Hint: target directory should be empty.")
        noise ("")
        return 1

    if PDTCRYPT_VERBOSE is True:
        noise ("PDT: decryption successful"                 )
        noise ("PDT:   %.10d bytes read"        % total_read)
        noise ("PDT:   %.10d objects decrypted" % total_obj )
        noise ("PDT:   %.10d bytes ciphertext"  % total_ct  )
        noise ("PDT:   %.10d bytes plaintext"   % total_pt  )
        noise (""                                           )

    return 0


def mode_scrypt (pw, ins=None, nacl=None, fmt=PDTCRYPT_SCRYPT_INTRANATOR):
    hsh = None
    paramversion = PDTCRYPT_DEFAULT_PVER
    if ins is not None:
        hsh, nacl, version, paramversion = scrypt_hashsource (pw, ins)
        defs = ENCRYPTION_PARAMETERS.get(paramversion, None)
    else:
        nacl    = binascii.unhexlify (nacl)
        defs    = ENCRYPTION_PARAMETERS.get(paramversion, None)
        version = PDTCRYPT_DEFAULT_VER

    kdfname, params = defs ["kdf"]
    if hsh is None:
        kdf = kdf_by_version (None, defs)
        hsh, _void = kdf (pw, nacl)

    import json

    if fmt == PDTCRYPT_SCRYPT_INTRANATOR:
        out = json.dumps ({ "salt"          : base64.b64encode (nacl).decode ()
                          , "key"           : base64.b64encode (hsh) .decode ()
                          , "paramversion"  : paramversion })
    elif fmt == PDTCRYPT_SCRYPT_PARAMETERS:
        out = json.dumps ({ "salt"          : binascii.hexlify (nacl).decode ()
                          , "key"           : binascii.hexlify (hsh) .decode ()
                          , "version"       : version
                          , "scrypt_params" : { "N"     : params ["N"]
                                              , "r"     : params ["r"]
                                              , "p"     : params ["p"]
                                              , "dkLen" : params ["dkLen"] } })
    else:
        raise RuntimeError ("bad scrypt output scheme %r" % fmt)

    print (out)


def noise_output_candidates (cands, indent=8, cols=PDTCRYPT_TT_COLUMNS):
    """
    Print a list of offsets without garbling the terminal too much.

    The indent is counted from column zero; if it is wide enough, the “PDT: ”
    marker will be prepended, considered part of the indentation.
    """
    wd   = cols - 1
    nc   = len (cands)
    idt  = " " * indent if indent < 5 else "PDT: " + " " * (indent - 5)
    line = idt
    lpos = indent
    sep  = ","
    lsep = len (sep)
    init = True # prevent leading separator

    if indent >= wd:
        raise ValueError ("the requested indentation exceeds the line "
                          "width by %d" % (indent - wd))

    for n in cands:
        ns  = "%d" % n
        lns = len (ns)
        if init is False:
            line += sep
            lpos += lsep

        lpos += lns
        if lpos > wd: # line break
            noise (line)
            line = idt
            lpos = indent + lns
        elif init is True:
            init = False
        else: # space
            line += ' '
            lpos += 1

        line += ns

    if lpos != indent:
        noise (line)


def mode_scan (pw, fname, nacl=None):
    """
    Dissect a binary file, looking for PDTCRYPT headers and objects.
    """
    try:
        fd = os.open (fname, os.O_RDONLY)
    except FileNotFoundError:
        noise ("PDT: failed to open %s readonly" % fname)
        noise ("")
        usage (err=True)

    try:
        if PDTCRYPT_VERBOSE is True:
            noise ("PDT: scan for potential sync points")
        cands = locate_hdr_candidates (fd)
        if len (cands) == 0:
            noise ("PDT: scan complete: input does not contain potential PDT "
                   "headers; giving up.")
            return -1
        if PDTCRYPT_VERBOSE is True:
            noise ("PDT: scan complete: found %d candidates:" % len (cands))
            noise_output_candidates (cands)
    finally:
        os.close (fd)


def usage (err=False):
    out = print
    if err is True:
        out = noise
    indent = ' ' * len (SELF)
    out ("usage: %s SUBCOMMAND { --help" % SELF)
    out ("       %s            | [ -v ] { -p PASSWORD | -k KEY }" % indent)
    out ("       %s              [ { -i | --in }  { - | SOURCE } ]" % indent)
    out ("       %s              [ { -n | --nacl } { SALT } ]" % indent)
    out ("       %s              [ { -o | --out } { - | DESTINATION } ]" % indent)
    out ("       %s              [ -D | --no-decrypt ] [ -S | --split ]" % indent)
    out ("       %s              [ -f | --format ]" % indent)
    out ("")
    out ("\twhere")
    out ("\t\tSUBCOMMAND      main mode: { process | scrypt }")
    out ("\t\t                where:")
    out ("\t\t                   process: extract objects from PDT archive")
    out ("\t\t                   scrypt:  calculate hash from password and first object")
    out ("\t\t-p PASSWORD     password to derive the encryption key from")
    out ("\t\t-k KEY          encryption key as 16 bytes in hexadecimal notation")
    out ("\t\t-s              enforce strict handling of initialization vectors")
    out ("\t\t-i SOURCE       file name to read from")
    out ("\t\t-o DESTINATION  file to write output to")
    out ("\t\t-n SALT         provide salt for scrypt mode in hex encoding")
    out ("\t\t-v              print extra info")
    out ("\t\t-S              split into files at object boundaries; this")
    out ("\t\t                requires DESTINATION to refer to directory")
    out ("\t\t-D              PDT header and ciphertext passthrough")
    out ("\t\t-f              format of SCRYPT hash output (“default” or “parameters”)")
    out ("")
    out ("\tinstead of filenames, “-” may used to specify stdin / stdout")
    out ("")
    sys.exit ((err is True) and 42 or 0)


def bail (msg):
    noise (msg)
    noise ("")
    usage (err=True)
    raise Unreachable


def parse_argv (argv):
    global SELF
    mode          = PDTCRYPT_DECRYPT
    secret        = None
    insspec       = None
    outsspec      = None
    nacl          = None
    scrypt_format = PDTCRYPT_SCRYPT_DEFAULT

    argvi = iter (argv)
    SELF  = os.path.basename (next (argvi))

    try:
        rawsubcmd  = next (argvi)
        subcommand = PDTCRYPT_SUB [rawsubcmd]
    except StopIteration:
        bail ("ERROR: subcommand required")
    except KeyError:
        bail ("ERROR: invalid subcommand “%s” specified" % rawsubcmd)

    def checked_arg ():
        nonlocal argvi
        try:
            return next (argvi)
        except StopIteration:
            bail ("ERROR: argument list incomplete")

    def checked_secret (t, arg):
        nonlocal secret
        if secret is None:
            secret = (t, arg)
        else:
            bail ("ERROR: encountered “%s” but secret already given" % arg)

    for arg in argvi:
        if arg in [ "-h", "--help" ]:
            usage ()
            raise Unreachable
        elif arg in [ "-v", "--verbose", "--wtf" ]:
            global PDTCRYPT_VERBOSE
            PDTCRYPT_VERBOSE = True
        elif arg in [ "-i", "--in", "--source" ]:
            insspec = checked_arg ()
            if PDTCRYPT_VERBOSE is True: noise ("PDT: decrypt from %s" % insspec)
        elif arg in [ "-p", "--password" ]:
            arg = checked_arg ()
            checked_secret (PDTCRYPT_SECRET_PW, arg)
            if PDTCRYPT_VERBOSE is True: noise ("PDT: decrypting with password")
        else:
            if subcommand == PDTCRYPT_SUB_PROCESS:
                if arg in [ "-s", "--strict-ivs" ]:
                    global PDTCRYPT_STRICTIVS
                    PDTCRYPT_STRICTIVS = True
                elif arg in [ "-o", "--out", "--dest", "--sink" ]:
                    outsspec = checked_arg ()
                    if PDTCRYPT_VERBOSE is True: noise ("PDT: decrypt to %s" % outsspec)
                elif arg in [ "-f", "--force" ]:
                    global PDTCRYPT_OVERWRITE
                    PDTCRYPT_OVERWRITE = True
                    if PDTCRYPT_VERBOSE is True: noise ("PDT: overwrite existing files")
                elif arg in [ "-S", "--split" ]:
                    mode |= PDTCRYPT_SPLIT
                    if PDTCRYPT_VERBOSE is True: noise ("PDT: split files")
                elif arg in [ "-D", "--no-decrypt" ]:
                    mode &= ~PDTCRYPT_DECRYPT
                    if PDTCRYPT_VERBOSE is True: noise ("PDT: not decrypting")
                elif arg in [ "-k", "--key" ]:
                    arg = checked_arg ()
                    checked_secret (PDTCRYPT_SECRET_KEY, arg)
                    if PDTCRYPT_VERBOSE is True: noise ("PDT: decrypting with key")
                else:
                    bail ("ERROR: unexpected positional argument “%s”" % arg)
            elif subcommand == PDTCRYPT_SUB_SCRYPT:
                if arg in [ "-n", "--nacl", "--salt" ]:
                    nacl = checked_arg ()
                    if PDTCRYPT_VERBOSE is True: noise ("PDT: salt key with %s" % nacl)
                elif arg in [ "-f", "--format" ]:
                    arg = checked_arg ()
                    try:
                        scrypt_format = PDTCRYPT_SCRYPT_FORMAT [arg]
                    except KeyError:
                        bail ("ERROR: invalid scrypt output format %s" % arg)
                    if PDTCRYPT_VERBOSE is True:
                        noise ("PDT: scrypt output format “%s”" % scrypt_format)
                else:
                    bail ("ERROR: unexpected positional argument “%s”" % arg)
            elif subcommand == PDTCRYPT_SUB_SCAN:
                pass

    if secret is None:
        if PDTCRYPT_VERBOSE is True:
            noise ("ERROR: no password or key specified, trying $PDTCRYPT_PASSWORD")
        epw = os.getenv ("PDTCRYPT_PASSWORD")
        if epw is not None:
            checked_secret (PDTCRYPT_SECRET_PW, epw.strip ())

    if secret is None:
        if PDTCRYPT_VERBOSE is True:
            noise ("ERROR: no password or key specified, trying $PDTCRYPT_KEY")
        ek = os.getenv ("PDTCRYPT_KEY")
        if ek is not None:
            checked_secret (PDTCRYPT_SECRET_KEY, ek.strip ())

    if secret is None:
        if subcommand == PDTCRYPT_SUB_SCRYPT:
            bail ("ERROR: scrypt hash mode requested but no password given")
        elif mode & PDTCRYPT_DECRYPT:
            bail ("ERROR: encryption requested but no password given")

    if subcommand == PDTCRYPT_SUB_SCAN:
        if insspec is None:
            bail ("ERROR: please supply an input file for scanning")
        if insspec == '-':
            bail ("ERROR: input must be seekable; please specify a file")
        return True, partial (mode_scan, secret [1].encode (), insspec, nacl)

    if subcommand == PDTCRYPT_SUB_SCRYPT:
        if secret [0] == PDTCRYPT_SECRET_KEY:
            bail ("ERROR: scrypt mode requires a password")
        if     insspec is not None and nacl is not None \
            or insspec is     None and nacl is     None :
                bail ("ERROR: please supply either an input file or "
                      "the salt")

    # default to stdout
    ins = None
    if insspec is not None or subcommand != PDTCRYPT_SUB_SCRYPT:
        ins = deptdcrypt_mk_stream (PDTCRYPT_SOURCE, insspec or "-")

    if subcommand == PDTCRYPT_SUB_SCRYPT:
        return True, partial (mode_scrypt, secret [1].encode (), ins, nacl,
                              fmt=scrypt_format)

    if mode & PDTCRYPT_SPLIT: # destination must be directory
        if outsspec is None or outsspec == "-":
            bail ("ERROR: split mode is incompatible with stdout sink")

        try:
            try:
                os.makedirs (outsspec, 0o700)
            except FileExistsError:
                # if it’s a directory with appropriate perms, everything is
                # good; otherwise, below invocation of open(2) will fail
                pass
            outs = os.open (outsspec, os.O_DIRECTORY, 0o600)
        except FileNotFoundError as exn:
            bail ("ERROR: cannot create target directory “%s”" % outsspec)
        except NotADirectoryError as exn:
            bail ("ERROR: target path “%s” is not a directory" % outsspec)

    else:
        outs = deptdcrypt_mk_stream (PDTCRYPT_SINK, outsspec or "-")

    return True, partial (mode_depdtcrypt, mode, secret, ins, outs)


def main (argv):
    ok, runner = parse_argv (argv)

    if ok is True: return runner ()

    return 1


if __name__ == "__main__":
    sys.exit (main (sys.argv))