????
Current Path : /usr/lib/python3.6/site-packages/pyasn1/codec/ber/ |
Current File : //usr/lib/python3.6/site-packages/pyasn1/codec/ber/encoder.py |
# # This file is part of pyasn1 software. # # Copyright (c) 2005-2017, Ilya Etingof <etingof@gmail.com> # License: http://pyasn1.sf.net/license.html # from pyasn1.type import tag, univ, char, useful from pyasn1.codec.ber import eoo from pyasn1.compat.octets import int2oct, oct2int, ints2octs, null, str2octs from pyasn1.compat.integer import to_bytes from pyasn1 import debug, error __all__ = ['encode'] class AbstractItemEncoder(object): supportIndefLenMode = 1 # An outcome of otherwise legit call `encodeFun(eoo.endOfOctets)` eooIntegerSubstrate = (0, 0) eooOctetsSubstrate = ints2octs(eooIntegerSubstrate) # noinspection PyMethodMayBeStatic def encodeTag(self, singleTag, isConstructed): tagClass, tagFormat, tagId = singleTag encodedTag = tagClass | tagFormat if isConstructed: encodedTag |= tag.tagFormatConstructed if tagId < 31: return encodedTag | tagId, else: substrate = tagId & 0x7f, tagId >>= 7 while tagId: substrate = (0x80 | (tagId & 0x7f),) + substrate tagId >>= 7 return (encodedTag | 0x1F,) + substrate def encodeLength(self, length, defMode): if not defMode and self.supportIndefLenMode: return (0x80,) if length < 0x80: return length, else: substrate = () while length: substrate = (length & 0xff,) + substrate length >>= 8 substrateLen = len(substrate) if substrateLen > 126: raise error.PyAsn1Error('Length octets overflow (%d)' % substrateLen) return (0x80 | substrateLen,) + substrate def encodeValue(self, value, encodeFun, **options): raise error.PyAsn1Error('Not implemented') def encode(self, value, encodeFun, **options): tagSet = value.tagSet # untagged item? if not tagSet: substrate, isConstructed, isOctets = self.encodeValue( value, encodeFun, **options ) return substrate defMode = options.get('defMode', True) for idx, singleTag in enumerate(tagSet.superTags): defModeOverride = defMode # base tag? if not idx: substrate, isConstructed, isOctets = self.encodeValue( value, encodeFun, **options ) if options.get('ifNotEmpty', False) and not substrate: return substrate # primitive form implies definite mode if not isConstructed: defModeOverride = True header = self.encodeTag(singleTag, isConstructed) header += self.encodeLength(len(substrate), defModeOverride) if isOctets: substrate = ints2octs(header) + substrate if not defModeOverride: substrate += self.eooOctetsSubstrate else: substrate = header + substrate if not defModeOverride: substrate += self.eooIntegerSubstrate if not isOctets: substrate = ints2octs(substrate) return substrate class EndOfOctetsEncoder(AbstractItemEncoder): def encodeValue(self, value, encodeFun, **options): return null, False, True class BooleanEncoder(AbstractItemEncoder): supportIndefLenMode = False def encodeValue(self, value, encodeFun, **options): return value and (1,) or (0,), False, False class IntegerEncoder(AbstractItemEncoder): supportIndefLenMode = False supportCompactZero = False def encodeValue(self, value, encodeFun, **options): if value == 0: # de-facto way to encode zero if self.supportCompactZero: return (), False, False else: return (0,), False, False return to_bytes(int(value), signed=True), False, True class BitStringEncoder(AbstractItemEncoder): def encodeValue(self, value, encodeFun, **options): valueLength = len(value) if valueLength % 8: alignedValue = value << (8 - valueLength % 8) else: alignedValue = value maxChunkSize = options.get('maxChunkSize', 0) if not maxChunkSize or len(alignedValue) <= maxChunkSize * 8: substrate = alignedValue.asOctets() return int2oct(len(substrate) * 8 - valueLength) + substrate, False, True # strip off explicit tags alignedValue = alignedValue.clone( tagSet=tag.TagSet(value.tagSet.baseTag, value.tagSet.baseTag) ) stop = 0 substrate = null while stop < valueLength: start = stop stop = min(start + maxChunkSize * 8, valueLength) substrate += encodeFun(alignedValue[start:stop], **options) return substrate, True, True class OctetStringEncoder(AbstractItemEncoder): def encodeValue(self, value, encodeFun, **options): maxChunkSize = options.get('maxChunkSize', 0) if not maxChunkSize or len(value) <= maxChunkSize: return value.asOctets(), False, True else: # will strip off explicit tags baseTagSet = tag.TagSet(value.tagSet.baseTag, value.tagSet.baseTag) pos = 0 substrate = null while True: chunk = value.clone(value[pos:pos + maxChunkSize], tagSet=baseTagSet) if not chunk: break substrate += encodeFun(chunk, **options) pos += maxChunkSize return substrate, True, True class NullEncoder(AbstractItemEncoder): supportIndefLenMode = False def encodeValue(self, value, encodeFun, **options): return null, False, True class ObjectIdentifierEncoder(AbstractItemEncoder): supportIndefLenMode = False def encodeValue(self, value, encodeFun, **options): oid = value.asTuple() # Build the first pair try: first = oid[0] second = oid[1] except IndexError: raise error.PyAsn1Error('Short OID %s' % (value,)) if 0 <= second <= 39: if first == 1: oid = (second + 40,) + oid[2:] elif first == 0: oid = (second,) + oid[2:] elif first == 2: oid = (second + 80,) + oid[2:] else: raise error.PyAsn1Error('Impossible first/second arcs at %s' % (value,)) elif first == 2: oid = (second + 80,) + oid[2:] else: raise error.PyAsn1Error('Impossible first/second arcs at %s' % (value,)) octets = () # Cycle through subIds for subOid in oid: if 0 <= subOid <= 127: # Optimize for the common case octets += (subOid,) elif subOid > 127: # Pack large Sub-Object IDs res = (subOid & 0x7f,) subOid >>= 7 while subOid: res = (0x80 | (subOid & 0x7f),) + res subOid >>= 7 # Add packed Sub-Object ID to resulted Object ID octets += res else: raise error.PyAsn1Error('Negative OID arc %s at %s' % (subOid, value)) return octets, False, False class RealEncoder(AbstractItemEncoder): supportIndefLenMode = 0 binEncBase = 2 # set to None to choose encoding base automatically @staticmethod def _dropFloatingPoint(m, encbase, e): ms, es = 1, 1 if m < 0: ms = -1 # mantissa sign if e < 0: es = -1 # exponenta sign m *= ms if encbase == 8: m *= 2 ** (abs(e) % 3 * es) e = abs(e) // 3 * es elif encbase == 16: m *= 2 ** (abs(e) % 4 * es) e = abs(e) // 4 * es while True: if int(m) != m: m *= encbase e -= 1 continue break return ms, int(m), encbase, e def _chooseEncBase(self, value): m, b, e = value encBase = [2, 8, 16] if value.binEncBase in encBase: return self._dropFloatingPoint(m, value.binEncBase, e) elif self.binEncBase in encBase: return self._dropFloatingPoint(m, self.binEncBase, e) # auto choosing base 2/8/16 mantissa = [m, m, m] exponenta = [e, e, e] sign = 1 encbase = 2 e = float('inf') for i in range(3): (sign, mantissa[i], encBase[i], exponenta[i]) = self._dropFloatingPoint(mantissa[i], encBase[i], exponenta[i]) if abs(exponenta[i]) < abs(e) or (abs(exponenta[i]) == abs(e) and mantissa[i] < m): e = exponenta[i] m = int(mantissa[i]) encbase = encBase[i] return sign, m, encbase, e def encodeValue(self, value, encodeFun, **options): if value.isPlusInf: return (0x40,), False, False if value.isMinusInf: return (0x41,), False, False m, b, e = value if not m: return null, False, True if b == 10: return str2octs('\x03%dE%s%d' % (m, e == 0 and '+' or '', e)), False, True elif b == 2: fo = 0x80 # binary encoding ms, m, encbase, e = self._chooseEncBase(value) if ms < 0: # mantissa sign fo |= 0x40 # sign bit # exponenta & mantissa normalization if encbase == 2: while m & 0x1 == 0: m >>= 1 e += 1 elif encbase == 8: while m & 0x7 == 0: m >>= 3 e += 1 fo |= 0x10 else: # encbase = 16 while m & 0xf == 0: m >>= 4 e += 1 fo |= 0x20 sf = 0 # scale factor while m & 0x1 == 0: m >>= 1 sf += 1 if sf > 3: raise error.PyAsn1Error('Scale factor overflow') # bug if raised fo |= sf << 2 eo = null if e == 0 or e == -1: eo = int2oct(e & 0xff) else: while e not in (0, -1): eo = int2oct(e & 0xff) + eo e >>= 8 if e == 0 and eo and oct2int(eo[0]) & 0x80: eo = int2oct(0) + eo if e == -1 and eo and not (oct2int(eo[0]) & 0x80): eo = int2oct(0xff) + eo n = len(eo) if n > 0xff: raise error.PyAsn1Error('Real exponent overflow') if n == 1: pass elif n == 2: fo |= 1 elif n == 3: fo |= 2 else: fo |= 3 eo = int2oct(n & 0xff) + eo po = null while m: po = int2oct(m & 0xff) + po m >>= 8 substrate = int2oct(fo) + eo + po return substrate, False, True else: raise error.PyAsn1Error('Prohibited Real base %s' % b) class SequenceEncoder(AbstractItemEncoder): def encodeValue(self, value, encodeFun, **options): value.verifySizeSpec() namedTypes = value.componentType substrate = null idx = len(value) while idx > 0: idx -= 1 if namedTypes: if namedTypes[idx].isOptional and not value[idx].isValue: continue if namedTypes[idx].isDefaulted and value[idx] == namedTypes[idx].asn1Object: continue substrate = encodeFun(value[idx], **options) + substrate return substrate, True, True class SequenceOfEncoder(AbstractItemEncoder): def encodeValue(self, value, encodeFun, **options): value.verifySizeSpec() substrate = null idx = len(value) while idx > 0: idx -= 1 substrate = encodeFun(value[idx], **options) + substrate return substrate, True, True class ChoiceEncoder(AbstractItemEncoder): def encodeValue(self, value, encodeFun, **options): return encodeFun(value.getComponent(), **options), True, True class AnyEncoder(OctetStringEncoder): def encodeValue(self, value, encodeFun, **options): return value.asOctets(), not options.get('defMode', True), True tagMap = { eoo.endOfOctets.tagSet: EndOfOctetsEncoder(), univ.Boolean.tagSet: BooleanEncoder(), univ.Integer.tagSet: IntegerEncoder(), univ.BitString.tagSet: BitStringEncoder(), univ.OctetString.tagSet: OctetStringEncoder(), univ.Null.tagSet: NullEncoder(), univ.ObjectIdentifier.tagSet: ObjectIdentifierEncoder(), univ.Enumerated.tagSet: IntegerEncoder(), univ.Real.tagSet: RealEncoder(), # Sequence & Set have same tags as SequenceOf & SetOf univ.SequenceOf.tagSet: SequenceOfEncoder(), univ.SetOf.tagSet: SequenceOfEncoder(), univ.Choice.tagSet: ChoiceEncoder(), # character string types char.UTF8String.tagSet: OctetStringEncoder(), char.NumericString.tagSet: OctetStringEncoder(), char.PrintableString.tagSet: OctetStringEncoder(), char.TeletexString.tagSet: OctetStringEncoder(), char.VideotexString.tagSet: OctetStringEncoder(), char.IA5String.tagSet: OctetStringEncoder(), char.GraphicString.tagSet: OctetStringEncoder(), char.VisibleString.tagSet: OctetStringEncoder(), char.GeneralString.tagSet: OctetStringEncoder(), char.UniversalString.tagSet: OctetStringEncoder(), char.BMPString.tagSet: OctetStringEncoder(), # useful types useful.ObjectDescriptor.tagSet: OctetStringEncoder(), useful.GeneralizedTime.tagSet: OctetStringEncoder(), useful.UTCTime.tagSet: OctetStringEncoder() } # Put in ambiguous & non-ambiguous types for faster codec lookup typeMap = { univ.Boolean.typeId: BooleanEncoder(), univ.Integer.typeId: IntegerEncoder(), univ.BitString.typeId: BitStringEncoder(), univ.OctetString.typeId: OctetStringEncoder(), univ.Null.typeId: NullEncoder(), univ.ObjectIdentifier.typeId: ObjectIdentifierEncoder(), univ.Enumerated.typeId: IntegerEncoder(), univ.Real.typeId: RealEncoder(), # Sequence & Set have same tags as SequenceOf & SetOf univ.Set.typeId: SequenceEncoder(), univ.SetOf.typeId: SequenceOfEncoder(), univ.Sequence.typeId: SequenceEncoder(), univ.SequenceOf.typeId: SequenceOfEncoder(), univ.Choice.typeId: ChoiceEncoder(), univ.Any.typeId: AnyEncoder(), # character string types char.UTF8String.typeId: OctetStringEncoder(), char.NumericString.typeId: OctetStringEncoder(), char.PrintableString.typeId: OctetStringEncoder(), char.TeletexString.typeId: OctetStringEncoder(), char.VideotexString.typeId: OctetStringEncoder(), char.IA5String.typeId: OctetStringEncoder(), char.GraphicString.typeId: OctetStringEncoder(), char.VisibleString.typeId: OctetStringEncoder(), char.GeneralString.typeId: OctetStringEncoder(), char.UniversalString.typeId: OctetStringEncoder(), char.BMPString.typeId: OctetStringEncoder(), # useful types useful.ObjectDescriptor.typeId: OctetStringEncoder(), useful.GeneralizedTime.typeId: OctetStringEncoder(), useful.UTCTime.typeId: OctetStringEncoder() } class Encoder(object): fixedDefLengthMode = None fixedChunkSize = None # noinspection PyDefaultArgument def __init__(self, tagMap, typeMap={}): self.__tagMap = tagMap self.__typeMap = typeMap def __call__(self, value, **options): if debug.logger & debug.flagEncoder: logger = debug.logger else: logger = None if logger: logger('encoder called in %sdef mode, chunk size %s for type %s, value:\n%s' % (not options.get('defMode', True) and 'in' or '', options.get('maxChunkSize', 0), value.prettyPrintType(), value.prettyPrint())) if self.fixedDefLengthMode is not None: options.update(defMode=self.fixedDefLengthMode) if self.fixedChunkSize is not None: options.update(maxChunkSize=self.fixedChunkSize) tagSet = value.tagSet try: concreteEncoder = self.__typeMap[value.typeId] except KeyError: # use base type for codec lookup to recover untagged types baseTagSet = tag.TagSet(value.tagSet.baseTag, value.tagSet.baseTag) try: concreteEncoder = self.__tagMap[baseTagSet] except KeyError: raise error.PyAsn1Error('No encoder for %s' % (value,)) if logger: logger('using value codec %s chosen by %s' % (concreteEncoder.__class__.__name__, tagSet)) substrate = concreteEncoder.encode(value, self, **options) if logger: logger('codec %s built %s octets of substrate: %s\nencoder completed' % (concreteEncoder, len(substrate), debug.hexdump(substrate))) return substrate #: Turns ASN.1 object into BER octet stream. #: #: Takes any ASN.1 object (e.g. :py:class:`~pyasn1.type.base.PyAsn1Item` derivative) #: walks all its components recursively and produces a BER octet stream. #: #: Parameters #: ---------- # value: any pyasn1 object (e.g. :py:class:`~pyasn1.type.base.PyAsn1Item` derivative) #: A pyasn1 object to encode #: #: defMode: :py:class:`bool` #: If `False`, produces indefinite length encoding #: #: maxChunkSize: :py:class:`int` #: Maximum chunk size in chunked encoding mode (0 denotes unlimited chunk size) #: #: Returns #: ------- #: : :py:class:`bytes` (Python 3) or :py:class:`str` (Python 2) #: Given ASN.1 object encoded into BER octetstream #: #: Raises #: ------ #: : :py:class:`pyasn1.error.PyAsn1Error` #: On encoding errors encode = Encoder(tagMap, typeMap)