1
0
mirror of https://github.com/moparisthebest/SickRage synced 2024-11-17 23:05:11 -05:00
SickRage/lib/hachoir_parser/misc/pdf.py

443 lines
16 KiB
Python
Raw Normal View History

"""
Adobe Portable Document Format (PDF) parser.
Author: Christophe Gisquet <christophe.gisquet@free.fr>
"""
from lib.hachoir_parser import Parser
from lib.hachoir_core.field import (
Field, FieldSet,
ParserError,
GenericVector,
UInt8, UInt16, UInt32,
String,
RawBytes)
from lib.hachoir_core.endian import LITTLE_ENDIAN
from lib.hachoir_core.text_handler import textHandler, hexadecimal
MAGIC = "%PDF-"
ENDMAGIC = "%%EOF"
def getLineEnd(s, pos=None):
if pos == None:
pos = (s.absolute_address+s.current_size)//8
end = s.stream.searchBytesLength("\x0D", False, 8*pos)
other_end = s.stream.searchBytesLength("\x0A", False, 8*pos)
if end == None or (other_end != None and other_end < end):
return other_end
return end
# TODO: rewrite to account for all possible terminations: ' ', '/', '\0XD'
# But this probably requires changing *ALL* of the places they are used,
# as ' ' is swallowed but not the others
def getElementEnd(s, limit=' ', offset=0):
addr = s.absolute_address+s.current_size
addr += 8*offset
pos = s.stream.searchBytesLength(limit, True, addr)
if pos == None:
#s.info("Can't find '%s' starting at %u" % (limit, addr))
return None
return pos
class PDFNumber(Field):
LIMITS = ['[', '/', '\x0D', ']']
"""
sprintf("%i") or sprinf("%.?f")
"""
def __init__(self, parent, name, desc=None):
Field.__init__(self, parent, name, description=desc)
# Get size
size = getElementEnd(parent)
for limit in self.LIMITS:
other_size = getElementEnd(parent, limit)
if other_size != None:
other_size -= 1
if size == None or other_size < size:
size = other_size
self._size = 8*size
# Get value
val = parent.stream.readBytes(self.absolute_address, size)
self.info("Number: size=%u value='%s'" % (size, val))
if val.find('.') != -1:
self.createValue = lambda: float(val)
else:
self.createValue = lambda: int(val)
class PDFString(Field):
"""
A string of the shape:
( This string \
uses 3 lines \
with the CR(LF) inhibited )
"""
def __init__(self, parent, name, desc=None):
Field.__init__(self, parent, name, description=desc)
val = ""
count = 1
off = 1
while not parent.eof:
char = parent.stream.readBytes(self.absolute_address+8*off, 1)
# Non-ASCII
if not char.isalpha() or char == '\\':
off += 1
continue
if char == '(':
count += 1
if char == ')':
count -= 1
# Parenthesis block = 0 => end of string
if count == 0:
off += 1
break
# Add it to the string
val += char
self._size = 8*off
self.createValue = lambda: val
class PDFName(Field):
LIMITS = ['[', '/', '<', ']']
"""
String starting with '/', where characters may be written using their
ASCII code (exemple: '#20' would be ' '
' ', ']' and '\0' are supposed not to be part of the name
"""
def __init__(self, parent, name, desc=None):
Field.__init__(self, parent, name, description=desc)
if parent.stream.readBytes(self.absolute_address, 1) != '/':
raise ParserError("Unknown PDFName '%s'" %
parent.stream.readBytes(self.absolute_address, 10))
size = getElementEnd(parent, offset=1)
#other_size = getElementEnd(parent, '[')-1
#if size == None or (other_size != None and other_size < size):
# size = other_size
for limit in self.LIMITS:
other_size = getElementEnd(parent, limit, 1)
if other_size != None:
other_size -= 1
if size == None or other_size < size:
#self.info("New size: %u" % other_size)
size = other_size
self._size = 8*(size+1)
# Value should be without the initial '/' and final ' '
self.createValue = lambda: parent.stream.readBytes(self.absolute_address+8, size).strip(' ')
class PDFID(Field):
"""
Not described as an object, but let's do as it was.
This ID has the shape <hexadecimal ASCII string>
"""
def __init__(self, parent, name, desc=None):
Field.__init__(self, parent, name, description=desc)
self._size = 8*getElementEnd(parent, '>')
self.createValue = lambda: parent.stream.readBytes(self.absolute_address+8, (self._size//8)-1)
class NotABool(Exception): pass
class PDFBool(Field):
"""
"true" or "false" string standing for the boolean value
"""
def __init__(self, parent, name, desc=None):
Field.__init__(self, parent, name, description=desc)
if parent.stream.readBytes(self.absolute_address, 4) == "true":
self._size = 4
self.createValue = lambda: True
elif parent.stream.readBytes(self.absolute_address, 5) == "false":
self._size = 5
self.createValue = lambda: False
raise NotABool
class LineEnd(FieldSet):
"""
Made of 0x0A, 0x0D (we may include several line ends)
"""
def createFields(self):
while not self.eof:
addr = self.absolute_address+self.current_size
char = self.stream.readBytes(addr, 1)
if char == '\x0A':
yield UInt8(self, "lf", "Line feed")
elif char == '\x0D':
yield UInt8(self, "cr", "Line feed")
else:
self.info("Line ends at %u/%u, len %u" %
(addr, self.stream._size, self.current_size))
break
class PDFDictionaryPair(FieldSet):
def createFields(self):
yield PDFName(self, "name", getElementEnd(self))
for field in parsePDFType(self):
yield field
class PDFDictionary(FieldSet):
def createFields(self):
yield String(self, "dict_start", 2)
while not self.eof:
addr = self.absolute_address+self.current_size
if self.stream.readBytes(addr, 2) != '>>':
for field in parsePDFType(self):
yield field
else:
break
yield String(self, "dict_end", 2)
class PDFArray(FieldSet):
"""
Array of possibly non-homogeneous elements, starting with '[' and ending
with ']'
"""
def createFields(self):
yield String(self, "array_start", 1)
while self.stream.readBytes(self.absolute_address+self.current_size, 1) != ']':
for field in parsePDFType(self):
yield field
yield String(self, "array_end", 1)
def parsePDFType(s):
addr = s.absolute_address+s.current_size
char = s.stream.readBytes(addr, 1)
if char == '/':
yield PDFName(s, "type[]", getElementEnd(s))
elif char == '<':
if s.stream.readBytes(addr+8, 1) == '<':
yield PDFDictionary(s, "dict[]")
else:
yield PDFID(s, "id[]")
elif char == '(':
yield PDFString(s, "string[]")
elif char == '[':
yield PDFArray(s, "array[]")
else:
# First parse size
size = getElementEnd(s)
for limit in ['/', '>', '<']:
other_size = getElementEnd(s, limit)
if other_size != None:
other_size -= 1
if size == None or (other_size>0 and other_size < size):
size = other_size
# Get element
name = s.stream.readBytes(addr, size)
char = s.stream.readBytes(addr+8*size+8, 1)
if name.count(' ') > 1 and char == '<':
# Probably a catalog
yield Catalog(s, "catalog[]")
elif name[0] in ('.','-','+', '0', '1', '2', '3', \
'4', '5', '6', '7', '8', '9'):
s.info("Not a catalog: %u spaces and end='%s'" % (name.count(' '), char))
yield PDFNumber(s, "integer[]")
else:
s.info("Trying to parse '%s': %u bytes" % \
(s.stream.readBytes(s.absolute_address+s.current_size, 4), size))
yield String(s, "unknown[]", size)
class Header(FieldSet):
def createFields(self):
yield String(self, "marker", 5, MAGIC)
length = getLineEnd(self, 4)
if length != None:
#self.info("Found at position %08X" % len)
yield String(self, "version", length-1)
yield LineEnd(self, "line_end")
else:
self.warning("Can't determine version!")
def createDescription(self):
return "PDF version %s" % self["version"].display
class Body(FieldSet):
def __init__(self, parent, name, desc=None):
FieldSet.__init__(self, parent, name, desc)
pos = self.stream.searchBytesLength(CrossReferenceTable.MAGIC, False)
if pos == None:
raise ParserError("Can't find xref starting at %u" %
(self.absolute_address//8))
self._size = 8*pos-self.absolute_address
def createFields(self):
while self.stream.readBytes(self.absolute_address+self.current_size, 1) == '%':
size = getLineEnd(self, 4)
if size == 2:
yield textHandler(UInt16(self, "crc32"), hexadecimal)
elif size == 4:
yield textHandler(UInt32(self, "crc32"), hexadecimal)
elif self.stream.readBytes(self.absolute_address+self.current_size, size).isalpha():
yield String(self, "comment[]", size)
else:
RawBytes(self, "unknown_data[]", size)
yield LineEnd(self, "line_end[]")
#abs_offset = self.current_size//8
# TODO: yield objects that read offsets and deduce size from
# "/cross_ref_table/sub_section[]/entries/item[]"
offsets = []
for subsection in self.array("/cross_ref_table/sub_section"):
for obj in subsection.array("entries/item"):
if "byte_offset" in obj:
# Could be inserted already sorted
offsets.append(obj["byte_offset"].value)
offsets.append(self["/cross_ref_table"].absolute_address//8)
offsets.sort()
for index in xrange(len(offsets)-1):
yield Catalog(self, "object[]", size=offsets[index+1]-offsets[index])
class Entry(FieldSet):
static_size = 20*8
def createFields(self):
typ = self.stream.readBytes(self.absolute_address+17*8, 1)
if typ == 'n':
yield PDFNumber(self, "byte_offset")
elif typ == 'f':
yield PDFNumber(self, "next_free_object_number")
else:
yield PDFNumber(self, "unknown_string")
yield PDFNumber(self, "generation_number")
yield UInt8(self, "type")
yield LineEnd(self, "line_end")
def createDescription(self):
if self["type"].value == 'n':
return "In-use entry at offset %u" % int(self["byte_offset"].value)
elif self["type"].value == 'f':
return "Free entry before in-use object %u" % \
int(self["next_free_object_number"].value)
else:
return "unknown %s" % self["unknown_string"].value
class SubSection(FieldSet):
def __init__(self, parent, name, desc=None):
FieldSet.__init__(self, parent, name, desc)
self.info("Got entry count: '%s'" % self["entry_count"].value)
self._size = self.current_size + 8*20*int(self["entry_count"].value) \
+ self["line_end"].size
def createFields(self):
yield PDFNumber(self, "start_number",
"Object number of first entry in subsection")
self.info("start_number = %i" % self["start_number"].value)
yield PDFNumber(self, "entry_count", "Number of entries in subsection")
self.info("entry_count = %i" % self["entry_count"].value)
yield LineEnd(self, "line_end")
yield GenericVector(self, "entries", int(self["entry_count"].value),
Entry)
#yield LineEnd(self, "line_end[]")
def createDescription(self):
return "Subsection with %s elements, starting at %s" % \
(self["entry_count"].value, self["start_number"])
class CrossReferenceTable(FieldSet):
MAGIC = "xref"
def __init__(self, parent, name, desc=None):
FieldSet.__init__(self, parent, name, description=desc)
pos = self.stream.searchBytesLength(Trailer.MAGIC, False)
if pos == None:
raise ParserError("Can't find '%s' starting at %u" \
(Trailer.MAGIC, self.absolute_address//8))
self._size = 8*pos-self.absolute_address
def createFields(self):
yield RawBytes(self, "marker", len(self.MAGIC))
yield LineEnd(self, "line_end[]")
while not self.eof:
yield SubSection(self, "sub_section[]")
class Catalog(FieldSet):
END_NAME = ['<', '/', '[']
def __init__(self, parent, name, size=None, desc=None):
FieldSet.__init__(self, parent, name, description=desc)
if size != None:
self._size = 8*size
# object catalogs are ended with "obj"
elif self["object"].value == "obj":
size = self.stream.searchBytesLength("endobj", False)
if size != None:
self._size = 8*(size+2)
def createFields(self):
yield PDFNumber(self, "index")
yield PDFNumber(self, "unknown[]")
length = getElementEnd(self)
for limit in self.END_NAME:
new_length = getElementEnd(self, limit)-len(limit)
if length == None or (new_length != None and new_length < length):
length = new_length
yield String(self, "object", length, strip=' ')
if self.stream.readBytes(self.absolute_address+self.current_size, 2) == '<<':
yield PDFDictionary(self, "key_list")
# End of catalog: this one has "endobj"
if self["object"].value == "obj":
yield LineEnd(self, "line_end[]")
yield String(self, "end_object", len("endobj"))
yield LineEnd(self, "line_end[]")
class Trailer(FieldSet):
MAGIC = "trailer"
def createFields(self):
yield RawBytes(self, "marker", len(self.MAGIC))
yield LineEnd(self, "line_end[]")
yield String(self, "start_attribute_marker", 2)
addr = self.absolute_address + self.current_size
while self.stream.readBytes(addr, 2) != '>>':
t = PDFName(self, "type[]")
yield t
name = t.value
self.info("Parsing PDFName '%s'" % name)
if name == "Size":
yield PDFNumber(self, "size", "Entries in the file cross-reference section")
elif name == "Prev":
yield PDFNumber(self, "offset")
elif name == "Root":
yield Catalog(self, "object_catalog")
elif name == "Info":
yield Catalog(self, "info")
elif name == "ID":
yield PDFArray(self, "id")
elif name == "Encrypt":
yield PDFDictionary(self, "decrypt")
else:
raise ParserError("Don't know trailer type '%s'" % name)
addr = self.absolute_address + self.current_size
yield String(self, "end_attribute_marker", 2)
yield LineEnd(self, "line_end[]")
yield String(self, "start_xref", 9)
yield LineEnd(self, "line_end[]")
yield PDFNumber(self, "cross_ref_table_start_address")
yield LineEnd(self, "line_end[]")
yield String(self, "end_marker", len(ENDMAGIC))
yield LineEnd(self, "line_end[]")
class PDFDocument(Parser):
endian = LITTLE_ENDIAN
PARSER_TAGS = {
"id": "pdf",
"category": "misc",
"file_ext": ("pdf",),
"mime": (u"application/pdf",),
"min_size": (5+4)*8,
"magic": ((MAGIC, 5),),
"description": "Portable Document Format (PDF) document"
}
def validate(self):
if self.stream.readBytes(0, len(MAGIC)) != MAGIC:
return "Invalid magic string"
return True
# Size is not always determined by position of "%%EOF":
# - updated documents have several of those
# - PDF files should be parsed from *end*
# => TODO: find when a document has been updated
def createFields(self):
yield Header(self, "header")
yield Body(self, "body")
yield CrossReferenceTable(self, "cross_ref_table")
yield Trailer(self, "trailer")