test_pie/external/exiftool/lib/Image/ExifTool/PDF.pm

2378 lines
89 KiB
Perl

#------------------------------------------------------------------------------
# File: PDF.pm
#
# Description: Read PDF meta information
#
# Revisions: 07/11/2005 - P. Harvey Created
# 07/25/2005 - P. Harvey Add support for encrypted documents
#
# References: 1) http://www.adobe.com/devnet/pdf/pdf_reference.html
# 2) http://search.cpan.org/dist/Crypt-RC4/
# 3) http://www.adobe.com/devnet/acrobat/pdfs/PDF32000_2008.pdf
# 4) http://www.adobe.com/content/dam/Adobe/en/devnet/pdf/pdfs/adobe_supplement_iso32000.pdf
# 5) http://tools.ietf.org/search/rfc3454
# 6) http://www.armware.dk/RFC/rfc/rfc4013.html
#------------------------------------------------------------------------------
package Image::ExifTool::PDF;
use strict;
use vars qw($VERSION $AUTOLOAD $lastFetched);
use Image::ExifTool qw(:DataAccess :Utils);
require Exporter;
$VERSION = '1.46';
sub FetchObject($$$$);
sub ExtractObject($$;$$);
sub ReadToNested($;$);
sub ProcessDict($$$$;$$);
sub ProcessAcroForm($$$$;$$);
sub ExpandArray($);
sub ReadPDFValue($);
sub CheckPDF($$$);
# $lastFetched - last fetched object reference (used for decryption)
# (undefined if fetched object was already decrypted, eg. object from stream)
my $cryptInfo; # encryption object reference (plus additional information)
my $cryptString; # flag that strings are encrypted
my $cryptStream; # flag that streams are encrypted
my $lastOffset; # last fetched object offset
my %streamObjs; # hash of stream objects
my %fetched; # dicts fetched in verbose mode (to avoid cyclical recursion)
my $pdfVer; # version of PDF file being processed
# filters supported in DecodeStream()
my %supportedFilter = (
'/FlateDecode' => 1,
'/Crypt' => 1,
'/Identity' => 1, # (not filtered)
'/DCTDecode' => 1, # (JPEG image - not filtered)
'/JPXDecode' => 1, # (Jpeg2000 image - not filtered)
'/LZWDecode' => 1, # (usually a bitmapped image)
'/ASCIIHexDecode' => 1,
'/ASCII85Decode' => 1,
# other standard filters that we currently don't support
#'/JBIG2Decode' => 0, # (JBIG2 image format not supported)
#'/CCITTFaxDecode' => 0,
#'/RunLengthDecode' => 0,
);
# tags in main PDF directories
%Image::ExifTool::PDF::Main = (
GROUPS => { 2 => 'Document' },
VARS => { CAPTURE => ['Main','Prev'] },
Info => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Info' },
# Adobe Acrobat 10.1.5 will create a duplicate Info dictionary with
# a different object number when metadata is edited. This flag
# is part of a patch to ignore this duplicate information (unless
# the IgnoreMinorErrors option is used)
IgnoreDuplicates => 1,
},
Root => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Root' },
},
Encrypt => {
NoProcess => 1, # don't process normally (processed in advance)
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Encrypt' },
},
_linearized => {
Name => 'Linearized',
Notes => 'flag set if document is linearized for fast web display; not a real Tag ID',
PrintConv => { 'true' => 'Yes', 'false' => 'No' },
},
);
# tags in PDF Info dictionary
%Image::ExifTool::PDF::Info = (
GROUPS => { 2 => 'Document' },
VARS => { CAPTURE => ['Info'] },
EXTRACT_UNKNOWN => 1, # extract all unknown tags in this directory
WRITE_PROC => \&Image::ExifTool::DummyWriteProc,
CHECK_PROC => \&CheckPDF,
WRITABLE => 'string',
# set PRIORITY to 0 so most recent Info dictionary takes precedence
# (Acrobat Pro bug? doesn't use same object/generation number for
# new Info dictionary when doing incrmental update)
PRIORITY => 0,
NOTES => q{
As well as the tags listed below, the PDF specification allows for
user-defined tags to exist in the Info dictionary. These tags, which should
have corresponding XMP-pdfx entries in the XMP of the PDF XML Metadata
object, are also extracted by ExifTool.
B<Writable> specifies the value format, and may be C<string>, C<date>,
C<integer>, C<real>, C<boolean> or C<name> for PDF tags.
},
Title => { },
Author => { Groups => { 2 => 'Author' } },
Subject => { },
Keywords => { List => 'string' }, # this is a string list
Creator => { },
Producer => { },
CreationDate => {
Name => 'CreateDate',
Writable => 'date',
Groups => { 2 => 'Time' },
Shift => 'Time',
PrintConv => '$self->ConvertDateTime($val)',
PrintConvInv => '$self->InverseDateTime($val)',
},
ModDate => {
Name => 'ModifyDate',
Writable => 'date',
Groups => { 2 => 'Time' },
Shift => 'Time',
PrintConv => '$self->ConvertDateTime($val)',
PrintConvInv => '$self->InverseDateTime($val)',
},
Trapped => {
Protected => 1,
# remove leading '/' from '/True' or '/False'
ValueConv => '$val=~s{^/}{}; $val',
ValueConvInv => '"/$val"',
},
'AAPL:Keywords' => { #PH
Name => 'AppleKeywords',
List => 'array', # this is an array of values
Notes => q{
keywords written by Apple utilities, although they seem to use PDF:Keywords
when reading
},
},
);
# tags in the PDF Root document catalog
%Image::ExifTool::PDF::Root = (
GROUPS => { 2 => 'Document' },
# note: can't capture previous versions of Root since they are not parsed
VARS => { CAPTURE => ['Root'] },
NOTES => 'This is the PDF document catalog.',
MarkInfo => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::MarkInfo' },
},
Metadata => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Metadata' },
},
Pages => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Pages' },
},
Perms => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Perms' },
},
AcroForm => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::AcroForm' },
},
Lang => 'Language',
PageLayout => { },
PageMode => { },
Version => 'PDFVersion',
);
# tags extracted from the PDF Encrypt dictionary
%Image::ExifTool::PDF::Encrypt = (
GROUPS => { 2 => 'Document' },
NOTES => 'Tags extracted from the document Encrypt dictionary.',
Filter => {
Name => 'Encryption',
Notes => q{
extracted value is actually a combination of the Filter, SubFilter, V, R and
Length information from the Encrypt dictionary
},
},
P => {
Name => 'UserAccess',
ValueConv => '$val & 0x0f3c', # ignore reserved bits
PrintConvColumns => 2,
PrintConv => { BITMASK => {
2 => 'Print',
3 => 'Modify',
4 => 'Copy',
5 => 'Annotate',
8 => 'Fill forms',
9 => 'Extract',
10 => 'Assemble',
11 => 'Print high-res',
}},
},
);
# tags in PDF Pages dictionary
%Image::ExifTool::PDF::Pages = (
GROUPS => { 2 => 'Document' },
Count => 'PageCount',
Kids => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Kids' },
},
);
# tags in PDF Perms dictionary
%Image::ExifTool::PDF::Perms = (
NOTES => 'Additional document permissions imposed by digital signatures.',
DocMDP => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Signature' },
},
FieldMDP => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Signature' },
},
UR3 => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Signature' },
},
);
# tags in PDF Perms dictionary
%Image::ExifTool::PDF::AcroForm = (
PROCESS_PROC => \&ProcessAcroForm,
_has_xfa => {
Name => 'HasXFA',
Notes => q{
this tag is defined if a document contains form fields, and is true if it
uses XML Forms Architecture; not a real Tag ID
},
PrintConv => { 'true' => 'Yes', 'false' => 'No' },
},
);
# tags in PDF Kids dictionary
%Image::ExifTool::PDF::Kids = (
Metadata => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Metadata' },
},
PieceInfo => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::PieceInfo' },
},
Resources => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Resources' },
},
Kids => {
Condition => '$self->Options("ExtractEmbedded")',
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Kids' },
},
);
# tags in PDF Resources dictionary
%Image::ExifTool::PDF::Resources = (
ColorSpace => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::ColorSpace' },
},
XObject => {
Condition => '$self->Options("ExtractEmbedded")',
SubDirectory => { TagTable => 'Image::ExifTool::PDF::XObject' },
},
Properties => {
Condition => '$self->Options("ExtractEmbedded")',
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Properties' },
},
);
# tags in PDF ColorSpace dictionary
%Image::ExifTool::PDF::ColorSpace = (
DefaultRGB => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::DefaultRGB' },
},
Cs1 => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Cs1' },
},
);
# tags in PDF DefaultRGB dictionary
%Image::ExifTool::PDF::DefaultRGB = (
ICCBased => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::ICCBased' },
},
);
# tags in PDF Cs1 dictionary
%Image::ExifTool::PDF::Cs1 = (
_stream => {
SubDirectory => { TagTable => 'Image::ExifTool::ICC_Profile::Main' },
},
);
# tags in PDF ICCBased dictionary
%Image::ExifTool::PDF::ICCBased = (
_stream => {
SubDirectory => { TagTable => 'Image::ExifTool::ICC_Profile::Main' },
},
);
# tags in PDF XObject dictionary (parsed only if ExtractEmbedded is enabled)
%Image::ExifTool::PDF::XObject = (
EXTRACT_UNKNOWN => 0, # extract known but numbered tags (Im1, Im2, etc)
Im => {
Notes => q{
the ExtractEmbedded option enables information to be extracted from these
embedded images
},
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Im' },
},
);
# tags in PDF Im# dictionary
%Image::ExifTool::PDF::Im = (
NOTES => q{
Information extracted from embedded images with the ExtractEmbedded option.
The EmbeddedImage and its metadata are extracted only for JPEG and Jpeg2000
image formats.
},
Width => 'EmbeddedImageWidth',
Height => 'EmbeddedImageHeight',
Filter => { Name => 'EmbeddedImageFilter', List => 1 },
ColorSpace => {
Name => 'EmbeddedImageColorSpace',
List => 1,
RawConv => 'ref $val ? undef : $val', # (ignore color space data)
},
Image_stream => {
Name => 'EmbeddedImage',
Groups => { 2 => 'Preview' },
Binary => 1,
},
);
# tags in PDF Properties dictionary
%Image::ExifTool::PDF::Properties = (
EXTRACT_UNKNOWN => 0, # extract known but numbered tags (MC0, MC1, etc)
MC => {
Notes => q{
the ExtractEmbedded option enables information to be extracted from these
embedded metadata dictionaries
},
SubDirectory => { TagTable => 'Image::ExifTool::PDF::MC' },
}
);
# tags in PDF MC# dictionary
%Image::ExifTool::PDF::MC = (
Metadata => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Metadata' },
}
);
# tags in PDF PieceInfo dictionary
%Image::ExifTool::PDF::PieceInfo = (
AdobePhotoshop => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::AdobePhotoshop' },
},
Illustrator => {
# assume this is an illustrator file if it contains this directory
# and doesn't have a ".PDF" extension
Condition => q{
$self->OverrideFileType("AI") unless $$self{FILE_EXT} and $$self{FILE_EXT} eq 'PDF';
return 1;
},
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Illustrator' },
},
);
# tags in PDF AdobePhotoshop dictionary
%Image::ExifTool::PDF::AdobePhotoshop = (
Private => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Private' },
},
);
# tags in PDF Illustrator dictionary
%Image::ExifTool::PDF::Illustrator = (
Private => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::AIPrivate' },
},
);
# tags in PDF Private dictionary
%Image::ExifTool::PDF::Private = (
ImageResources => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::ImageResources' },
},
);
# tags in PDF AI Private dictionary
%Image::ExifTool::PDF::AIPrivate = (
GROUPS => { 2 => 'Document' },
EXTRACT_UNKNOWN => 0, # extract known but numbered tags
AIMetaData => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::AIMetaData' },
},
AIPrivateData => {
Notes => q{
the ExtractEmbedded option enables information to be extracted from embedded
PostScript documents in the AIPrivateData# and AIPDFPrivateData# streams
},
JoinStreams => 1, # join streams from numbered tags and process as one
SubDirectory => { TagTable => 'Image::ExifTool::PostScript::Main' },
},
AIPDFPrivateData => {
JoinStreams => 1, # join streams from numbered tags and process as one
SubDirectory => { TagTable => 'Image::ExifTool::PostScript::Main' },
},
RoundTripVersion => { },
ContainerVersion => { },
CreatorVersion => { },
);
# tags in PDF AIMetaData dictionary
%Image::ExifTool::PDF::AIMetaData = (
_stream => {
SubDirectory => { TagTable => 'Image::ExifTool::PostScript::Main' },
},
);
# tags in PDF ImageResources dictionary
%Image::ExifTool::PDF::ImageResources = (
_stream => {
SubDirectory => { TagTable => 'Image::ExifTool::Photoshop::Main' },
},
);
# tags in PDF MarkInfo dictionary
%Image::ExifTool::PDF::MarkInfo = (
GROUPS => { 2 => 'Document' },
Marked => {
Name => 'TaggedPDF',
Notes => "not a Tagged PDF if this tag is missing",
PrintConv => { 'true' => 'Yes', 'false' => 'No' },
},
);
# tags in PDF Metadata dictionary
%Image::ExifTool::PDF::Metadata = (
GROUPS => { 2 => 'Document' },
XML_stream => { # this is the stream for a Subtype /XML dictionary (not a real tag)
Name => 'XMP',
SubDirectory => { TagTable => 'Image::ExifTool::XMP::Main' },
},
);
# tags in PDF signature directories (DocMDP, FieldMDP or UR3)
%Image::ExifTool::PDF::Signature = (
GROUPS => { 2 => 'Document' },
ContactInfo => 'SignerContactInfo',
Location => 'SigningLocation',
M => {
Name => 'SigningDate',
Format => 'date',
Groups => { 2 => 'Time' },
PrintConv => '$self->ConvertDateTime($val)',
},
Name => 'SigningAuthority',
Reason => 'SigningReason',
Reference => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Reference' },
},
Prop_AuthTime => {
Name => 'AuthenticationTime',
PrintConv => 'ConvertTimeSpan($val) . " ago"',
},
Prop_AuthType => 'AuthenticationType',
);
# tags in PDF Reference dictionary
%Image::ExifTool::PDF::Reference = (
TransformParams => {
SubDirectory => { TagTable => 'Image::ExifTool::PDF::TransformParams' },
},
);
# tags in PDF TransformParams dictionary
%Image::ExifTool::PDF::TransformParams = (
GROUPS => { 2 => 'Document' },
Annots => {
Name => 'AnnotationUsageRights',
Notes => q{
possible values are Create, Delete, Modify, Copy, Import and Export;
additional values for UR3 signatures are Online and SummaryView
},
List => 1,
},
Document => {
Name => 'DocumentUsageRights',
Notes => 'only possible value is FullSave',
List => 1,
},
Form => {
Name => 'FormUsageRights',
Notes => q{
possible values are FillIn, Import, Export, SubmitStandalone and
SpawnTemplate; additional values for UR3 signatures are BarcodePlaintext and
Online
},
List => 1,
},
FormEX => {
Name => 'FormExtraUsageRights',
Notes => 'UR signatures only; only possible value is BarcodePlaintext',
List => 1,
},
Signature => {
Name => 'SignatureUsageRights',
Notes => 'only possible value is Modify',
List => 1,
},
EF => {
Name => 'EmbeddedFileUsageRights',
Notes => 'possible values are Create, Delete, Modify and Import',
List => 1,
},
Msg => 'UsageRightsMessage',
P => {
Name => 'ModificationPermissions',
Notes => q{
1-3 for DocMDP signatures, default 2; true/false for UR3 signatures, default
false
},
PrintConv => {
1 => 'No changes permitted',
2 => 'Fill forms, Create page templates, Sign',
3 => 'Fill forms, Create page templates, Sign, Create/Delete/Edit annotations',
'true' => 'Restrict all applications to reader permissions',
'false' => 'Do not restrict applications to reader permissions',
},
},
Action => {
Name => 'FieldPermissions',
Notes => 'FieldMDP signatures only',
PrintConv => {
'All' => 'Disallow changes to all form fields',
'Include' => 'Disallow changes to specified form fields',
'Exclude' => 'Allow changes to specified form fields',
},
},
Fields => {
Notes => 'FieldMDP signatures only',
Name => 'FormFields',
List => 1,
},
);
# unknown tags for use in verbose option
%Image::ExifTool::PDF::Unknown = (
GROUPS => { 2 => 'Unknown' },
);
#------------------------------------------------------------------------------
# AutoLoad our writer routines when necessary
#
sub AUTOLOAD
{
return Image::ExifTool::DoAutoLoad($AUTOLOAD, @_);
}
#------------------------------------------------------------------------------
# Convert from PDF to EXIF-style date/time
# Inputs: 0) PDF date/time string (D:YYYYmmddHHMMSS+HH'MM')
# Returns: EXIF date string (YYYY:mm:dd HH:MM:SS+HH:MM)
sub ConvertPDFDate($)
{
my $date = shift;
# remove optional 'D:' prefix
$date =~ s/^D://;
# fill in default values if necessary
# YYYYmmddHHMMSS
my $default = '00000101000000';
if (length $date < length $default) {
$date .= substr($default, length $date);
}
$date =~ /^(\d{4})(\d{2})(\d{2})(\d{2})(\d{2})(\d{2})(.*)/ or return $date;
$date = "$1:$2:$3 $4:$5:$6";
if ($7) {
my $tz = $7;
if ($tz =~ /^\s*Z/i) {
# ignore any "HH'mm'" after the Z (OS X 10.6 does this)
$date .= 'Z';
# tolerate some improper formatting in timezone specification
} elsif ($tz =~ /^\s*([-+])\s*(\d+)[': ]+(\d*)/) {
$date .= $1 . $2 . ':' . ($3 || '00');
}
}
return $date;
}
#------------------------------------------------------------------------------
# Locate any object in the XRef tables (including compressed objects)
# Inputs: 0) XRef reference, 1) object reference string (or free object number)
# Returns: offset to object in file or compressed object reference string,
# 0 if object is free, or undefined on error
sub LocateAnyObject($$)
{
my ($xref, $ref) = @_;
return undef unless $xref;
return $$xref{$ref} if exists $$xref{$ref};
# get the object number
return undef unless $ref =~ /^(\d+)/;
my $objNum = $1;
# return 0 if the object number has been reused (old object is free)
return 0 if defined $$xref{$objNum};
#
# scan our XRef stream dictionaries for this object
#
return undef unless $$xref{dicts};
my $dict;
foreach $dict (@{$$xref{dicts}}) {
# quick check to see if the object is in the range for this xref stream
next if $objNum >= $$dict{Size};
my $index = $$dict{Index};
next if $objNum < $$index[0];
# scan the tables for the specified object
my $size = $$dict{_entry_size};
my $num = scalar(@$index) / 2;
my $tot = 0;
my $i;
for ($i=0; $i<$num; ++$i) {
my $start = $$index[$i*2];
my $count = $$index[$i*2+1];
# table is in ascending order, so quit if we have passed the object
last if $objNum < $start;
if ($objNum < $start + $count) {
my $offset = $size * ($objNum - $start + $tot);
last if $offset + $size > length $$dict{_stream};
my @c = unpack("x$offset C$size", $$dict{_stream});
# extract values from this table entry
# (can be 1, 2, 3, 4, etc.. bytes per value)
my (@t, $j, $k);
my $w = $$dict{W};
for ($j=0; $j<3; ++$j) {
# use default value if W entry is 0 (as per spec)
# - 0th element defaults to 1, others default to 0
$$w[$j] or $t[$j] = ($j ? 0 : 1), next;
$t[$j] = shift(@c);
for ($k=1; $k < $$w[$j]; ++$k) {
$t[$j] = 256 * $t[$j] + shift(@c);
}
}
# by default, use "o g R" as the xref key
# (o = object number, g = generation number)
my $ref2 = "$objNum $t[2] R";
if ($t[0] == 1) {
# normal object reference:
# $t[1]=offset of object from start, $t[2]=generation number
$$xref{$ref2} = $t[1];
} elsif ($t[0] == 2) {
# compressed object reference:
# $t[1]=stream object number, $t[2]=index of object in stream
$ref2 = "$objNum 0 R";
$$xref{$ref2} = "I$t[2] $t[1] 0 R";
} elsif ($t[0] == 0) {
# free object:
# $t[1]=next free object in linked list, $t[2]=generation number
$$xref{$ref2} = 0;
} else {
# treat as a null object
$$xref{$ref2} = undef;
}
$$xref{$objNum} = $t[1]; # remember offsets by object number too
return $$xref{$ref} if $ref eq $ref2;
return 0; # object is free or was reused
}
$tot += $count;
}
}
return undef;
}
#------------------------------------------------------------------------------
# Locate a regular object in the XRef tables (does not include compressed objects)
# Inputs: 0) XRef reference, 1) object reference string (or free object number)
# Returns: offset to object in file, 0 if object is free,
# or undef on error or if object was compressed
sub LocateObject($$)
{
my ($xref, $ref) = @_;
my $offset = LocateAnyObject($xref, $ref);
return undef if $offset and $offset =~ /^I/;
return $offset;
}
#------------------------------------------------------------------------------
# Fetch indirect object from file (from inside a stream if required)
# Inputs: 0) ExifTool object reference, 1) object reference string,
# 2) xref lookup, 3) object name (for warning messages)
# Returns: object data or undefined on error
# Notes: sets $lastFetched to the object reference, or undef if the object
# was extracted from an encrypted stream
sub FetchObject($$$$)
{
my ($et, $ref, $xref, $tag) = @_;
$lastFetched = $ref; # save this for decoding if necessary
my $offset = LocateAnyObject($xref, $ref);
$lastOffset = $offset;
unless ($offset) {
$et->Warn("Bad $tag reference") unless defined $offset;
return undef;
}
my ($data, $obj);
if ($offset =~ s/^I(\d+) //) {
my $index = $1; # object index in stream
my ($objNum) = split ' ', $ref; # save original object number
$ref = $offset; # now a reference to the containing stream object
$obj = $streamObjs{$ref};
unless ($obj) {
# don't try to load the same object stream twice
return undef if defined $obj;
$streamObjs{$ref} = '';
# load the parent object stream
$obj = FetchObject($et, $ref, $xref, $tag);
# make sure it contains everything we need
return undef unless defined $obj and ref($obj) eq 'HASH';
return undef unless $$obj{First} and $$obj{N};
return undef unless DecodeStream($et, $obj);
# add a special '_table' entry to this dictionary which contains
# the list of object number/offset pairs from the stream header
my $num = $$obj{N} * 2;
my @table = split ' ', $$obj{_stream}, $num;
return undef unless @table == $num;
# remove everything before first object in stream
$$obj{_stream} = substr($$obj{_stream}, $$obj{First});
$table[$num-1] =~ s/^(\d+).*/$1/s; # trim excess from last number
$$obj{_table} = \@table;
# save the object stream so we don't have to re-load it later
$streamObjs{$ref} = $obj;
}
# verify that we have the specified object
my $i = 2 * $index;
my $table = $$obj{_table};
unless ($index < $$obj{N} and $$table[$i] == $objNum) {
$et->Warn("Bad index for stream object $tag");
return undef;
}
# extract the object at the specified index in the stream
# (offsets in table are in sequential order, so we can subract from
# the next offset to get the object length)
$offset = $$table[$i + 1];
my $len = ($$table[$i + 3] || length($$obj{_stream})) - $offset;
$data = substr($$obj{_stream}, $offset, $len);
# avoid re-decrypting data in already decrypted streams
undef $lastFetched if $cryptStream;
return ExtractObject($et, \$data);
}
my $raf = $$et{RAF};
$raf->Seek($offset+$$et{PDFBase}, 0) or $et->Warn("Bad $tag offset"), return undef;
# verify that we are reading the expected object
$raf->ReadLine($data) or $et->Warn("Error reading $tag data"), return undef;
($obj = $ref) =~ s/R/obj/;
unless ($data =~ s/^$obj//) {
$et->Warn("$tag object ($obj) not found at $offset");
return undef;
}
# read the first line of data for the object (skipping comments if necessary)
for (;;) {
last if $data =~ /\S/ and $data !~ /^\s*%/;
$raf->ReadLine($data) or $et->Warn("Error reading $tag data"), return undef;
}
return ExtractObject($et, \$data, $raf, $xref);
}
#------------------------------------------------------------------------------
# Convert PDF value to something readable
# Inputs: 0) PDF object data
# Returns: converted object
sub ReadPDFValue($)
{
my $str = shift;
# decode all strings in an array
if (ref $str eq 'ARRAY') {
# create new list to not alter the original data when rewriting
my ($val, @vals);
foreach $val (@$str) {
push @vals, ReadPDFValue($val);
}
return \@vals;
}
length $str or return $str;
my $delim = substr($str, 0, 1);
if ($delim eq '(') { # literal string
$str = $1 if $str =~ /^.*?\((.*)\)/s; # remove brackets
# decode escape sequences in literal strings
while ($str =~ /\\(.)/sg) {
my $n = pos($str) - 2;
my $c = $1;
my $r;
if ($c =~ /[0-7]/) {
# get up to 2 more octal digits
$c .= $1 if $str =~ /\G([0-7]{1,2})/g;
# convert octal escape code
$r = chr(oct($c) & 0xff);
} elsif ($c eq "\x0d") {
# the string is continued if the line ends with '\'
# (also remove "\x0d\x0a")
$c .= $1 if $str =~ /\G(\x0a)/g;
$r = '';
} elsif ($c eq "\x0a") {
$r = '';
} else {
# convert escaped characters
($r = $c) =~ tr/nrtbf/\n\r\t\b\f/;
}
substr($str, $n, length($c)+1) = $r;
# continue search after this character
pos($str) = $n + length($r);
}
Crypt(\$str, $lastFetched) if $cryptString;
} elsif ($delim eq '<') { # hex string
# decode hex data
$str =~ tr/0-9A-Fa-f//dc;
$str .= '0' if length($str) & 0x01; # (by the spec)
$str = pack('H*', $str);
Crypt(\$str, $lastFetched) if $cryptString;
} elsif ($delim eq '/') { # name
$str = substr($str, 1);
# convert escape codes (PDF 1.2 or later)
$str =~ s/#([0-9a-f]{2})/chr(hex($1))/sgei if $pdfVer >= 1.2;
}
return $str;
}
#------------------------------------------------------------------------------
# Extract PDF object from combination of buffered data and file
# Inputs: 0) ExifTool object reference, 1) data reference,
# 2) optional raf reference, 3) optional xref table
# Returns: converted PDF object or undef on error
# a) dictionary object --> hash reference
# b) array object --> array reference
# c) indirect reference --> scalar reference
# d) string, name, integer, boolean, null --> scalar value
# - updates $$dataPt on return to contain unused data
# - creates two bogus entries ('_stream' and '_tags') in dictionaries to represent
# the stream data and a list of the tags (not including '_stream' and '_tags')
# in their original order
sub ExtractObject($$;$$)
{
my ($et, $dataPt, $raf, $xref) = @_;
my (@tags, $data, $objData);
my $dict = { };
my $delim;
for (;;) {
if ($$dataPt =~ /^\s*(<{1,2}|\[|\()/s) {
$delim = $1;
$$dataPt =~ s/^\s+//; # remove leading white space
$objData = ReadToNested($dataPt, $raf);
return undef unless defined $objData;
last;
} elsif ($$dataPt =~ s{^\s*(\S[^[(/<>\s]*)\s*}{}s) {
#
# extract boolean, numerical, string, name, null object or indirect reference
#
$objData = $1;
# look for an indirect reference
if ($objData =~ /^\d+$/ and $$dataPt =~ s/^(\d+)\s+R//s) {
$objData .= "$1 R";
$objData = \$objData; # return scalar reference
}
return $objData; # return simple scalar or scalar reference
}
$raf and $raf->ReadLine($data) or return undef;
$$dataPt .= $data;
}
#
# return literal string or hex string without parsing
#
if ($delim eq '(' or $delim eq '<') {
return $objData;
#
# extract array
#
} elsif ($delim eq '[') {
$objData =~ /^.*?\[(.*)\]/s or return undef;
my $data = $1; # brackets removed
my @list;
for (;;) {
last unless $data =~ m{\s*(\S[^[(/<>\s]*)}sg;
my $val = $1;
if ($val =~ /^(<{1,2}|\[|\()/) {
my $pos = pos($data) - length($val);
# nested dict, array, literal string or hex string
my $buff = substr($data, $pos);
$val = ReadToNested(\$buff);
last unless defined $val;
pos($data) = $pos + length($val);
$val = ExtractObject($et, \$val);
} elsif ($val =~ /^\d/) {
my $pos = pos($data);
if ($data =~ /\G\s+(\d+)\s+R/g) {
$val = \ "$val $1 R"; # make a reference
} else {
pos($data) = $pos;
}
}
push @list, $val;
}
return \@list;
}
#
# extract dictionary
#
# Note: entries are not necessarily separated by whitespace (doh!)
# eg) "/Tag/Name", "/Tag(string)", "/Tag[array]", etc are legal!
# Also, they may be separated by a comment (eg. "/Tag%comment\nValue"),
# but comments have already been removed
while ($objData =~ m{(\s*)/([^/[\]()<>{}\s]+)\s*(\S[^[(/<>\s]*)}sg) {
my $tag = $2;
my $val = $3;
if ($val =~ /^(<{1,2}|\[|\()/) {
# nested dict, array, literal string or hex string
$objData = substr($objData, pos($objData)-length($val));
$val = ReadToNested(\$objData, $raf);
last unless defined $val;
$val = ExtractObject($et, \$val);
pos($objData) = 0;
} elsif ($val =~ /^\d/) {
my $pos = pos($objData);
if ($objData =~ /\G\s+(\d+)\s+R/sg) {
$val = \ "$val $1 R"; # make a reference
} else {
pos($objData) = $pos;
}
}
if ($$dict{$tag}) {
# duplicate dictionary entries are not allowed
$et->Warn("Duplicate '${tag}' entry in dictionary (ignored)");
} else {
# save the entry
push @tags, $tag;
$$dict{$tag} = $val;
}
}
return undef unless @tags;
$$dict{_tags} = \@tags;
return $dict unless $raf; # direct objects can not have streams
#
# extract the stream object
#
# dictionary must specify stream Length
my $length = $$dict{Length} or return $dict;
if (ref $length) {
$length = $$length;
my $oldpos = $raf->Tell();
# get the location of the object specifying the length
# (compressed objects are not allowed)
my $offset = LocateObject($xref, $length) or return $dict;
$offset or $et->Warn('Bad Length object'), return $dict;
$raf->Seek($offset+$$et{PDFBase}, 0) or $et->Warn('Bad Length offset'), return $dict;
# verify that we are reading the expected object
$raf->ReadLine($data) or $et->Warn('Error reading Length data'), return $dict;
$length =~ s/R/obj/;
unless ($data =~ /^$length\s+(\d+)?/) {
$et->Warn("Length object ($length) not found at $offset");
return $dict;
}
if (defined $1) {
$length = $1;
} else {
$raf->ReadLine($data) or $et->Warn('Error reading stream Length'), return $dict;
$data =~ /^\s*(\d+)/ or $et->Warn('Stream length not found'), return $dict;
$length = $1;
}
$raf->Seek($oldpos, 0); # restore position to start of stream
}
# extract the trailing stream data
for (;;) {
# find the stream token
if ($$dataPt =~ /(\S+)/) {
last unless $1 eq 'stream';
# read an extra line because it may contain our \x0a
$$dataPt .= $data if $raf->ReadLine($data);
# remove our stream header
$$dataPt =~ s/^\s*stream(\x0a|\x0d\x0a)//s;
my $more = $length - length($$dataPt);
if ($more > 0) {
unless ($raf->Read($data, $more) == $more) {
$et->Warn('Error reading stream data');
$$dataPt = '';
return $dict;
}
$$dict{_stream} = $$dataPt . $data;
$$dataPt = '';
} elsif ($more < 0) {
$$dict{_stream} = substr($$dataPt, 0, $length);
$$dataPt = substr($$dataPt, $length);
} else {
$$dict{_stream} = $$dataPt;
$$dataPt = '';
}
last;
}
$raf->ReadLine($data) or last;
$$dataPt .= $data;
}
return $dict;
}
#------------------------------------------------------------------------------
# Read to nested delimiter
# Inputs: 0) data reference, 1) optional raf reference
# Returns: data up to and including matching delimiter (or undef on error)
# - updates data reference with trailing data
# - unescapes characters in literal strings
my %closingDelim = ( # lookup for matching delimiter
'(' => ')',
'[' => ']',
'<' => '>',
'<<' => '>>',
);
sub ReadToNested($;$)
{
my ($dataPt, $raf) = @_;
my @delim = (''); # closing delimiter list, most deeply nested first
pos($$dataPt) = 0; # begin at start of data
for (;;) {
unless ($$dataPt =~ /(\\*)(\(|\)|<{1,2}|>{1,2}|\[|\]|%)/g) {
# must read some more data
my $buff;
last unless $raf and $raf->ReadLine($buff);
$$dataPt .= $buff;
pos($$dataPt) = length($$dataPt) - length($buff);
next;
}
# are we in a literal string?
if ($delim[0] eq ')') {
# ignore escaped delimiters (preceded by odd number of \'s)
next if length($1) & 0x01;
# ignore all delimiters but unescaped braces
next unless $2 eq '(' or $2 eq ')';
} elsif ($2 eq '%') {
# ignore the comment
my $pos = pos($$dataPt) - 1;
# remove everything from '%' up to but not including newline
$$dataPt =~ /.*/g;
my $end = pos($$dataPt);
$$dataPt = substr($$dataPt, 0, $pos) . substr($$dataPt, $end);
pos($$dataPt) = $pos;
next;
}
if ($closingDelim{$2}) {
# push the corresponding closing delimiter
unshift @delim, $closingDelim{$2};
next;
}
unless ($2 eq $delim[0]) {
# handle the case where we find a ">>>" and interpret it
# as ">> >" instead of "> >>"
next unless $2 eq '>>' and $delim[0] eq '>';
pos($$dataPt) = pos($$dataPt) - 1;
}
shift @delim; # remove from nesting list
next if $delim[0]; # keep going if we have more nested delimiters
my $pos = pos($$dataPt);
my $buff = substr($$dataPt, 0, $pos);
$$dataPt = substr($$dataPt, $pos);
return $buff; # success!
}
return undef; # didn't find matching delimiter
}
#------------------------------------------------------------------------------
# Decode LZW-encoded data (ref 1)
# Inputs: 0) data reference
# Returns: true on success and data is decoded, or false and data is untouched
sub DecodeLZW($)
{
my $dataPt = shift;
return 0 if length $$dataPt < 4;
my @lzw = (map(chr, 0..255), undef, undef); # LZW code table
my $mask = 0x01ff; # mask for least-significant 9 bits
my @dat = unpack 'n*', $$dataPt . "\0";
my $word = ($dat[0] << 16) | $dat[1];
my ($bit, $pos, $bits, $out) = (0, 2, 9, '');
my $lastVal;
for (;;) {
# bits are packed MSB first in PDF LZW (the PDF spec doesn't mention this)
my $shift = 32 - ($bit + $bits);
if ($shift < 0) {
return 0 if $pos >= @dat; # missing EOD marker
$word = (($word & 0xffff) << 16) | $dat[$pos++]; # read next word
$bit -= 16;
$shift += 16;
};
my $code = ($word >> $shift) & $mask;
$bit += $bits;
my $val = $lzw[$code];
if (defined $val) {
# store new code as previous sequence plus 1st char of new sequence
push @lzw, $lastVal . substr($val, 0, 1) if defined $lastVal;
} elsif ($code == @lzw) { # new code
return 0 unless defined $lastVal;
# we are using the code that we are about to generate, so the last
# character in the new sequence must be the same as the first
# character in the previous sequence (makes sense if you think about it)
$val = $lastVal . substr($lastVal, 0, 1);
push @lzw, $val;
} elsif ($code == 256) { # clear table
splice @lzw, 258;
$bits = 9;
$mask = 0x1ff;
undef $lastVal;
next;
} elsif ($code == 257) { # EOD marker
last; # all done!
} else {
return 0;
}
$out .= $val; # add this byte sequence to the output
# we added a new entry to the LZW table, so we must increase
# the bit width if necessary, up to a maximum of 12
@lzw >= $mask and $bits < 12 and ++$bits, $mask |= $mask << 1;
$lastVal = $val;
}
$$dataPt = $out; # return decompressed data
return 1;
}
#------------------------------------------------------------------------------
# Decode filtered stream
# Inputs: 0) ExifTool object reference, 1) dictionary reference
# Returns: true if stream has been decoded OK
sub DecodeStream($$)
{
local $_;
my ($et, $dict) = @_;
return 0 unless $$dict{_stream}; # no stream to decode
# get list of filters
my (@filters, @decodeParms, $filter);
if (ref $$dict{Filter} eq 'ARRAY') {
@filters = @{$$dict{Filter}};
} elsif (defined $$dict{Filter}) {
@filters = ($$dict{Filter});
}
# be sure we can process all the filters before we take the time to do the decryption
foreach $filter (@filters) {
next if $supportedFilter{$filter};
$et->WarnOnce("Unsupported Filter $filter");
return 0;
}
# apply decryption first if required (and if the default encryption
# has not been overridden by a Crypt filter. Note: the Crypt filter
# must be first in the Filter array: ref 3, page 38)
unless (defined $$dict{_decrypted} or ($filters[0] and $filters[0] eq '/Crypt')) {
CryptStream($dict, $lastFetched);
}
return 1 unless $$dict{Filter}; # Filter entry is mandatory
return 0 if defined $$dict{_filtered}; # avoid double-filtering
$$dict{_filtered} = 1; # set flag to prevent double-filtering
# get array of DecodeParms dictionaries
if (ref $$dict{DecodeParms} eq 'ARRAY') {
@decodeParms = @{$$dict{DecodeParms}};
} else {
@decodeParms = ($$dict{DecodeParms});
}
foreach $filter (@filters) {
my $decodeParms = shift @decodeParms;
if ($filter eq '/FlateDecode') {
# make sure we support the predictor (if used) before decoding
my $pre;
if (ref $decodeParms eq 'HASH') {
$pre = $$decodeParms{Predictor};
if ($pre and $pre ne '1' and $pre ne '12') {
$et->WarnOnce("FlateDecode Predictor $pre currently not supported");
return 0;
}
}
if (eval { require Compress::Zlib }) {
my $inflate = Compress::Zlib::inflateInit();
my ($buff, $stat);
$inflate and ($buff, $stat) = $inflate->inflate($$dict{_stream});
if ($inflate and $stat == Compress::Zlib::Z_STREAM_END()) {
$$dict{_stream} = $buff;
} else {
$et->Warn('Error inflating stream');
return 0;
}
} else {
$et->WarnOnce('Install Compress::Zlib to process filtered streams');
return 0;
}
next unless $pre and $pre eq '12'; # 12 = 'up' prediction
# apply anti-predictor
my $cols = $$decodeParms{Columns};
unless ($cols) {
# currently only support 'up' prediction
$et->WarnOnce('No Columns for decoding stream');
return 0;
}
my @bytes = unpack('C*', $$dict{_stream});
my @pre = (0) x $cols; # initialize predictor array
my $buff = '';
while (@bytes > $cols) {
unless (($_ = shift @bytes) == 2) {
$et->WarnOnce("Unsupported PNG filter $_"); # (yes, PNG)
return 0;
}
foreach (@pre) {
$_ = ($_ + shift(@bytes)) & 0xff;
}
$buff .= pack('C*', @pre);
}
$$dict{_stream} = $buff;
} elsif ($filter eq '/Crypt') {
# (we shouldn't have to check the _decrypted flag since we
# already checked the _filtered flag, but what the heck...)
next if defined $$dict{_decrypted};
# assume Identity filter (the default) if DecodeParms are missing
next unless ref $decodeParms eq 'HASH';
my $name = $$decodeParms{Name};
next unless defined $name or $name eq 'Identity';
if ($name ne 'StdCF') {
$et->WarnOnce("Unsupported Crypt Filter $name");
return 0;
}
unless ($cryptInfo) {
$et->WarnOnce('Missing Encrypt StdCF entry');
return 0;
}
# decrypt the stream manually because we want to:
# 1) ignore $cryptStream (StmF) setting
# 2) ignore EncryptMetadata setting (I can't find mention of how to
# reconcile this in the spec., but this would make sense)
# 3) avoid adding the crypt key extension (ref 3, page 58, Algorithm 1b)
# 4) set _decrypted flag so we will recrypt according to StmF when
# writing (since we don't yet write Filter'd streams)
Crypt(\$$dict{_stream}, 'none');
$$dict{_decrypted} = ($cryptStream ? 1 : 0);
} elsif ($filter eq '/LZWDecode') {
# make sure we don't have any unsupported decoding parameters
if (ref $decodeParms eq 'HASH') {
if ($$decodeParms{Predictor}) {
$et->WarnOnce("LZWDecode Predictor $$decodeParms{Predictor} currently not supported");
return 0;
} elsif ($$decodeParms{EarlyChange}) {
$et->WarnOnce("LZWDecode EarlyChange currently not supported");
return 0;
}
}
unless (DecodeLZW(\$$dict{_stream})) {
$et->WarnOnce('LZW decompress error');
return 0;
}
} elsif ($filter eq '/ASCIIHexDecode') {
$$dict{_stream} =~ s/>.*//; # truncate at '>' (end of data mark)
$$dict{_stream} =~ tr/0-9a-zA-Z//d; # remove illegal characters
$$dict{_stream} = pack 'H*', $$dict{_stream};
} elsif ($filter eq '/ASCII85Decode') {
my ($err, @out, $i);
my ($n, $val) = (0, 0);
foreach (split //, $$dict{_stream}) {
if ($_ ge '!' and $_ le 'u') {;
$val = 85 * $val + ord($_) - 33;
next unless ++$n == 5;
} elsif ($_ eq '~') {
$n == 1 and $err = 1; # error to have a single char in the last group of 5
for ($i=$n; $i<5; ++$i) { $val *= 85; }
} elsif ($_ eq 'z') {
$n and $err = 2, last; # error if 'z' isn't the first char
$n = 5;
} else {
next if /^\s$/; # ignore white space
$err = 3, last; # any other character is an error
}
$val = unpack('V', pack('N', $val)); # reverse byte order
while (--$n > 0) {
push @out, $val & 0xff;
$val >>= 8;
}
last if $_ eq '~';
# (both $n and $val are zero again now)
}
$err and $et->WarnOnce("ASCII85Decode error $err");
$$dict{_stream} = pack('C*', @out);
}
}
return 1;
}
#------------------------------------------------------------------------------
# Initialize state for RC4 en/decryption (ref 2)
# Inputs: 0) RC4 key string
# Returns: RC4 key hash reference
sub RC4Init($)
{
my @key = unpack('C*', shift);
my @state = (0 .. 255);
my ($i, $j) = (0, 0);
while ($i < 256) {
my $st = $state[$i];
$j = ($j + $st + $key[$i % scalar(@key)]) & 0xff;
$state[$i++] = $state[$j];
$state[$j] = $st;
}
return { State => \@state, XY => [ 0, 0 ] };
}
#------------------------------------------------------------------------------
# Apply RC4 en/decryption (ref 2)
# Inputs: 0) data reference, 1) RC4 key hash reference or RC4 key string
# - can call this method directly with a key string, or with with the key
# reference returned by RC4Init
# - RC4 is a symmetric algorithm, so encryption is the same as decryption
sub RC4Crypt($$)
{
my ($dataPt, $key) = @_;
$key = RC4Init($key) unless ref $key eq 'HASH';
my $state = $$key{State};
my ($x, $y) = @{$$key{XY}};
my @data = unpack('C*', $$dataPt);
foreach (@data) {
$x = ($x + 1) & 0xff;
my $stx = $$state[$x];
$y = ($stx + $y) & 0xff;
my $sty = $$state[$x] = $$state[$y];
$$state[$y] = $stx;
$_ ^= $$state[($stx + $sty) & 0xff];
}
$$key{XY} = [ $x, $y ];
$$dataPt = pack('C*', @data);
}
#------------------------------------------------------------------------------
# Update AES cipher with a bit of data
# Inputs: 0) data
# Returns: encrypted data
my $cipherMore;
sub CipherUpdate($)
{
my $dat = shift;
my $pos = 0;
$dat = $cipherMore . $dat if length $dat;
while ($pos + 16 <= length($dat)) {
substr($dat,$pos,16) = Image::ExifTool::AES::Cipher(substr($dat,$pos,16));
$pos += 16;
}
if ($pos < length $dat) {
$cipherMore = substr($dat,$pos);
$dat = substr($dat,0,$pos);
} else {
$cipherMore = '';
}
return $dat;
}
#------------------------------------------------------------------------------
# Get encrypted hash
# Inputs: 0) Password, 1) salt, 2) vector, 3) encryption revision
# Returns: hash
sub GetHash($$$$)
{
my ($password, $salt, $vector, $rev) = @_;
# return Rev 5 hash
return Digest::SHA::sha256($password, $salt, $vector) if $rev == 5;
# compute Rev 6 hardened hash
# (ref http://code.google.com/p/origami-pdf/source/browse/lib/origami/encryption.rb)
my $blockSize = 32;
my $input = Digest::SHA::sha256($password, $salt, $vector) . ("\0" x 32);
my $key = substr($input, 0, 16);
my $iv = substr($input, 16, 16);
my $h;
my $x = '';
my $i = 0;
while ($i < 64 or $i < ord(substr($x,-1,1))+32) {
my $block = substr($input, 0, $blockSize);
$x = '';
Image::ExifTool::AES::Crypt(\$x, $key, $iv, 1);
$cipherMore = '';
my ($j, $digest);
for ($j=0; $j<64; ++$j) {
$x = '';
$x .= CipherUpdate($password) if length $password;
$x .= CipherUpdate($block);
$x .= CipherUpdate($vector) if length $vector;
if ($j == 0) {
my @a = unpack('C16', $x);
my $sum = 0;
$sum += $_ foreach @a;
# set SHA block size (32, 48 or 64 bytes = SHA-256, 384 or 512)
$blockSize = 32 + ($sum % 3) * 16;
$digest = Digest::SHA->new($blockSize * 8);
}
$digest->add($x);
}
$h = $digest->digest();
$key = substr($h, 0, 16);
substr($input,0,16) = $h;
$iv = substr($h, 16, 16);
++$i;
}
return substr($h, 0, 32);
}
#------------------------------------------------------------------------------
# Initialize decryption
# Inputs: 0) ExifTool object reference, 1) Encrypt dictionary reference,
# 2) ID from file trailer dictionary
# Returns: error string or undef on success (and sets $cryptInfo)
sub DecryptInit($$$)
{
local $_;
my ($et, $encrypt, $id) = @_;
undef $cryptInfo;
unless ($encrypt and ref $encrypt eq 'HASH') {
return 'Error loading Encrypt object';
}
my $filt = $$encrypt{Filter};
unless ($filt and $filt =~ s/^\///) {
return 'Encrypt dictionary has no Filter!';
}
# extract some interesting tags
my $ver = $$encrypt{V} || 0;
my $rev = $$encrypt{R} || 0;
my $enc = "$filt V$ver";
$enc .= ".$rev" if $filt eq 'Standard';
$enc .= " ($1)" if $$encrypt{SubFilter} and $$encrypt{SubFilter} =~ /^\/(.*)/;
$enc .= ' (' . ($$encrypt{Length} || 40) . '-bit)' if $filt eq 'Standard';
my $tagTablePtr = GetTagTable('Image::ExifTool::PDF::Encrypt');
$et->HandleTag($tagTablePtr, 'Filter', $enc);
if ($filt ne 'Standard') {
return "Encryption filter $filt currently not supported";
} elsif (not defined $$encrypt{R}) {
return 'Standard security handler missing revision';
}
unless ($$encrypt{O} and $$encrypt{P} and $$encrypt{U}) {
return 'Incomplete Encrypt specification';
}
if ("$ver.$rev" >= 5.6) {
# apologize for poor performance (AES is a pure Perl implementation)
$et->Warn('Decryption is very slow for encryption V5.6 or higher', 3);
}
$et->HandleTag($tagTablePtr, 'P', $$encrypt{P});
my %parm; # optional parameters extracted from Encrypt dictionary
if ($ver == 1 or $ver == 2) {
$cryptString = $cryptStream = 1;
} elsif ($ver == 4 or $ver == 5) {
# initialize our $cryptString and $cryptStream flags
foreach ('StrF', 'StmF') {
my $flagPt = $_ eq 'StrF' ? \$cryptString : \$cryptStream;
$$flagPt = $$encrypt{$_};
undef $$flagPt if $$flagPt and $$flagPt eq '/Identity';
return "Unsupported $_ encryption $$flagPt" if $$flagPt and $$flagPt ne '/StdCF';
}
if ($cryptString or $cryptStream) {
return 'Missing or invalid Encrypt StdCF entry' unless ref $$encrypt{CF} eq 'HASH' and
ref $$encrypt{CF}{StdCF} eq 'HASH' and $$encrypt{CF}{StdCF}{CFM};
my $cryptMeth = $$encrypt{CF}{StdCF}{CFM};
unless ($cryptMeth =~ /^\/(V2|AESV2|AESV3)$/) {
return "Unsupported encryption method $cryptMeth";
}
# set "_aesv2" or "_aesv3" flag in %$encrypt hash if AES encryption was used
$$encrypt{'_' . lc($1)} = 1 if $cryptMeth =~ /^\/(AESV2|AESV3)$/;
}
if ($ver == 5) {
# validate OE and UE entries
foreach ('OE', 'UE') {
return "Missing Encrypt $_ entry" unless $$encrypt{$_};
$parm{$_} = ReadPDFValue($$encrypt{$_});
return "Invalid Encrypt $_ entry" unless length $parm{$_} == 32;
}
require Image::ExifTool::AES; # will need this later
}
} else {
return "Encryption version $ver currently not supported";
}
$id or return "Can't decrypt (no document ID)";
# make sure we have the necessary libraries available
if ($ver < 5) {
unless (eval { require Digest::MD5 }) {
return "Install Digest::MD5 to process encrypted PDF";
}
} else {
unless (eval { require Digest::SHA }) {
return "Install Digest::SHA to process AES-256 encrypted PDF";
}
}
# calculate file-level en/decryption key
my $pad = "\x28\xBF\x4E\x5E\x4E\x75\x8A\x41\x64\x00\x4E\x56\xFF\xFA\x01\x08".
"\x2E\x2E\x00\xB6\xD0\x68\x3E\x80\x2F\x0C\xA9\xFE\x64\x53\x69\x7A";
my $o = ReadPDFValue($$encrypt{O});
my $u = ReadPDFValue($$encrypt{U});
# set flag indicating whether metadata is encrypted
# (in version 4 and higher, metadata streams may not be encrypted)
if ($ver < 4 or not $$encrypt{EncryptMetadata} or $$encrypt{EncryptMetadata} !~ /false/i) {
$$encrypt{_meta} = 1;
}
# try no password first, then try provided password if available
my ($try, $key);
for ($try=0; ; ++$try) {
my $password;
if ($try == 0) {
$password = '';
} elsif ($try == 1) {
$password = $et->Options('Password');
return 'Document is password protected (use Password option)' unless defined $password;
# make sure there is no UTF-8 flag on the password
if ($] >= 5.006 and (eval { require Encode; Encode::is_utf8($password) } or $@)) {
# repack by hand if Encode isn't available
$password = $@ ? pack('C*',unpack($] < 5.010000 ? 'U0C*' : 'C0C*',$password)) : Encode::encode('utf8',$password);
}
} else {
return 'Incorrect password';
}
if ($ver < 5) {
if (length $password) {
# password must be encoding in PDFDocEncoding (ref iso32000)
$password = $et->Encode($password, 'PDFDoc');
# truncate or pad the password to exactly 32 bytes
if (length($password) > 32) {
$password = substr($password, 0, 32);
} elsif (length($password) < 32) {
$password .= substr($pad, 0, 32-length($password));
}
} else {
$password = $pad;
}
$key = $password . $o . pack('V', $$encrypt{P}) . $id;
my $rep = 1;
if ($rev == 3 or $rev == 4) {
# must add this if metadata not encrypted
$key .= "\xff\xff\xff\xff" unless $$encrypt{_meta};
$rep += 50; # repeat MD5 50 more times if revision is 3 or greater
}
my ($len, $i, $dat);
if ($ver == 1) {
$len = 5;
} else {
$len = $$encrypt{Length} || 40;
$len >= 40 or return 'Bad Encrypt Length';
$len = int($len / 8);
}
for ($i=0; $i<$rep; ++$i) {
$key = substr(Digest::MD5::md5($key), 0, $len);
}
# decrypt U to see if a user password is required
if ($rev >= 3) {
$dat = Digest::MD5::md5($pad . $id);
RC4Crypt(\$dat, $key);
for ($i=1; $i<=19; ++$i) {
my @key = unpack('C*', $key);
foreach (@key) { $_ ^= $i; }
RC4Crypt(\$dat, pack('C*', @key));
}
$dat .= substr($u, 16);
} else {
$dat = $pad;
RC4Crypt(\$dat, $key);
}
last if $dat eq $u; # all done if this was the correct key
} else {
return 'Invalid O or U Encrypt entries' if length($o) < 48 or length($u) < 48;
if (length $password) {
# Note: this should be good for passwords containing reasonable characters,
# but to be bullet-proof we need to apply the SASLprep (IETF RFC 4013) profile
# of stringprep (IETF RFC 3454) to the password before encoding in UTF-8
$password = $et->Encode($password, 'UTF8');
$password = substr($password, 0, 127) if length($password) > 127;
}
# test for the owner password
my $sha = GetHash($password, substr($o,32,8), substr($u,0,48), $rev);
if ($sha eq substr($o, 0, 32)) {
$key = GetHash($password, substr($o,40,8), substr($u,0,48), $rev);
my $dat = ("\0" x 16) . $parm{OE};
# decrypt with no padding
my $err = Image::ExifTool::AES::Crypt(\$dat, $key, 0, 1);
return $err if $err;
$key = $dat; # use this as the file decryption key
last;
}
# test for the user password
$sha = GetHash($password, substr($u,32,8), '', $rev);
if ($sha eq substr($u, 0, 32)) {
$key = GetHash($password, substr($u,40,8), '', $rev);
my $dat = ("\0" x 16) . $parm{UE};
my $err = Image::ExifTool::AES::Crypt(\$dat, $key, 0, 1);
return $err if $err;
$key = $dat; # use this as the file decryption key
last;
}
}
}
$$encrypt{_key} = $key; # save the file-level encryption key
$cryptInfo = $encrypt; # save reference to the file-level Encrypt object
return undef; # success!
}
#------------------------------------------------------------------------------
# Decrypt/Encrypt data
# Inputs: 0) data ref
# 1) PDF object reference to use as crypt key extension (may be 'none' to
# avoid extending the encryption key, as for streams with Crypt Filter)
# 2) encrypt flag (false for decryption)
sub Crypt($$;$)
{
return unless $cryptInfo;
my ($dataPt, $keyExt, $encrypt) = @_;
# do not decrypt if the key extension object is undefined
# (this doubles as a flag to disable decryption/encryption)
return unless defined $keyExt;
my $key = $$cryptInfo{_key};
# apply the necessary crypt key extension
unless ($$cryptInfo{_aesv3}) {
unless ($keyExt eq 'none') {
# extend crypt key using object and generation number
unless ($keyExt =~ /^(I\d+ )?(\d+) (\d+)/) {
$$cryptInfo{_error} = 'Invalid object reference for encryption';
return;
}
$key .= substr(pack('V', $2), 0, 3) . substr(pack('V', $3), 0, 2);
}
# add AES-128 salt if necessary (this little gem is conveniently
# omitted from the Adobe PDF 1.6 documentation, causing me to
# waste 12 hours trying to figure out why this wasn't working --
# it appears in ISO32000 though, so I should have been using that)
$key .= 'sAlT' if $$cryptInfo{_aesv2};
my $len = length($key);
$key = Digest::MD5::md5($key); # get 16-byte MD5 digest
$key = substr($key, 0, $len) if $len < 16; # trim if necessary
}
# perform the decryption/encryption
if ($$cryptInfo{_aesv2} or $$cryptInfo{_aesv3}) {
require Image::ExifTool::AES;
my $err = Image::ExifTool::AES::Crypt($dataPt, $key, $encrypt);
$err and $$cryptInfo{_error} = $err;
} else {
RC4Crypt($dataPt, $key);
}
}
#------------------------------------------------------------------------------
# Decrypt/Encrypt stream data
# Inputs: 0) dictionary ref, 1) PDF object reference to use as crypt key extension
sub CryptStream($$)
{
return unless $cryptStream;
my ($dict, $keyExt) = @_;
my $type = $$dict{Type} || '';
# XRef streams are not encrypted (ref 3, page 50),
# and Metadata may or may not be encrypted
if ($cryptInfo and $type ne '/XRef' and
($$cryptInfo{_meta} or $type ne '/Metadata'))
{
Crypt(\$$dict{_stream}, $keyExt, $$dict{_decrypted});
# toggle _decrypted flag
$$dict{_decrypted} = ($$dict{_decrypted} ? undef : 1);
} else {
$$dict{_decrypted} = 0; # stream should never be encrypted
}
}
#------------------------------------------------------------------------------
# Generate a new PDF tag (based on its ID) and add it to a tag table
# Inputs: 0) tag table ref, 1) tag ID
# Returns: tag info ref
sub NewPDFTag($$)
{
my ($tagTablePtr, $tag) = @_;
my $name = $tag;
# translate URL-like escape sequences
$name =~ s/#([0-9a-f]{2})/chr(hex($1))/ige;
$name =~ s/[^-\w]+/_/g; # translate invalid characters to an underline
$name =~ s/(^|_)([a-z])/\U$2/g; # start words with upper case
my $tagInfo = { Name => $name };
AddTagToTable($tagTablePtr, $tag, $tagInfo);
return $tagInfo;
}
#------------------------------------------------------------------------------
# Process AcroForm dictionary to set HasXMLFormsArchitecture flag
# Inputs: Same as ProcessDict
sub ProcessAcroForm($$$$;$$)
{
my ($et, $tagTablePtr, $dict, $xref, $nesting, $type) = @_;
$et->HandleTag($tagTablePtr, '_has_xfa', $$dict{XFA} ? 'true' : 'false');
return ProcessDict($et, $tagTablePtr, $dict, $xref, $nesting, $type);
}
#------------------------------------------------------------------------------
# Expand array into a string
# Inputs: 0) array ref
# Return: string
sub ExpandArray($)
{
my $val = shift;
my @list = @$val;
foreach (@list) {
ref $_ eq 'SCALAR' and $_ = "ref($$_)", next;
ref $_ eq 'ARRAY' and $_ = ExpandArray($_), next;
defined $_ or $_ = '<undef>', next;
}
return '[' . join(',',@list) . ']';
}
#------------------------------------------------------------------------------
# Process PDF dictionary extract tag values
# Inputs: 0) ExifTool object reference, 1) tag table reference
# 2) dictionary reference, 3) cross-reference table reference,
# 4) nesting depth, 5) dictionary capture type
sub ProcessDict($$$$;$$)
{
my ($et, $tagTablePtr, $dict, $xref, $nesting, $type) = @_;
my $verbose = $et->Options('Verbose');
my $unknown = $$tagTablePtr{EXTRACT_UNKNOWN};
my $embedded = (defined $unknown and not $unknown and $et->Options('ExtractEmbedded'));
my @tags = @{$$dict{_tags}};
my ($next, %join);
my $index = 0;
$nesting = ($nesting || 0) + 1;
if ($nesting > 50) {
$et->WarnOnce('Nesting too deep (directory ignored)');
return;
}
# save entire dictionary for rewriting if specified
if ($$et{PDF_CAPTURE} and $$tagTablePtr{VARS} and
$tagTablePtr->{VARS}->{CAPTURE})
{
my $name;
foreach $name (@{$tagTablePtr->{VARS}->{CAPTURE}}) {
next if $$et{PDF_CAPTURE}{$name};
# make sure we load the right type if indicated
next if $type and $type ne $name;
$$et{PDF_CAPTURE}{$name} = $dict;
last;
}
}
#
# extract information from all tags in the dictionary
#
for (;;) {
my ($tag, $isSubDoc);
if (@tags) {
$tag = shift @tags;
} elsif (defined $next and not $next) {
$tag = 'Next';
$next = 1;
} else {
last;
}
my $val = $$dict{$tag};
my $tagInfo = $et->GetTagInfo($tagTablePtr, $tag);
if ($tagInfo) {
undef $tagInfo if $$tagInfo{NoProcess};
} elsif ($embedded and $tag =~ /^(.*?)(\d+)$/ and
$$tagTablePtr{$1} and (ref $val ne 'SCALAR' or not $fetched{$$val}))
{
my ($name, $num) = ($1, $2);
$tagInfo = $et->GetTagInfo($tagTablePtr, $name);
if (ref $tagInfo eq 'HASH' and $$tagInfo{JoinStreams}) {
$fetched{$$val} = 1;
my $obj = FetchObject($et, $$val, $xref, $tag);
$join{$name} = [] unless $join{$name};
next unless ref $obj eq 'HASH' and $$obj{_stream};
# save all the stream data to join later
DecodeStream($et, $obj);
$join{$name}->[$num] = $$obj{_stream};
undef $tagInfo; # don't process
} else {
$isSubDoc = 1; # treat as a sub-document
}
}
if ($verbose) {
my ($val2, $extra);
if (ref $val eq 'SCALAR') {
$extra = ", indirect object ($$val)";
if ($fetched{$$val}) {
$val2 = "ref($$val)";
} elsif ($tag eq 'Next' and not $next) {
# handle 'Next' links after all others
$next = 0;
next;
} else {
$fetched{$$val} = 1;
$val = FetchObject($et, $$val, $xref, $tag);
unless (defined $val) {
my $str;
if (defined $lastOffset) {
$val2 = '<free>';
$str = 'Object was freed';
} else {
$val2 = '<err>';
$str = 'Error reading object';
}
$et->VPrint(0, "$$et{INDENT}${str}:\n");
}
}
} elsif (ref $val eq 'HASH') {
$extra = ', direct dictionary';
} elsif (ref $val eq 'ARRAY') {
$extra = ', direct array of ' . scalar(@$val) . ' objects';
} else {
$extra = ', direct object';
}
my $isSubdir;
if (ref $val eq 'HASH') {
$isSubdir = 1;
} elsif (ref $val eq 'ARRAY') {
# recurse into objects in arrays only if they are lists of
# dictionaries or indirect objects which could be dictionaries
$isSubdir = 1 if @$val;
foreach (@$val) {
next if ref $_ eq 'HASH' or ref $_ eq 'SCALAR';
undef $isSubdir;
last;
}
}
if ($isSubdir) {
# create bogus subdirectory to recurse into this dict
$tagInfo or $tagInfo = {
Name => $tag,
SubDirectory => { TagTable => 'Image::ExifTool::PDF::Unknown' },
};
} else {
$val2 = ExpandArray($val) if ref $val eq 'ARRAY';
# generate tag info if we will use it later
if (not $tagInfo and defined $val and $unknown) {
$tagInfo = NewPDFTag($tagTablePtr, $tag);
}
}
$et->VerboseInfo($tag, $tagInfo,
Value => $val2 || $val,
Extra => $extra,
Index => $index++,
);
next unless defined $val;
}
unless ($tagInfo) {
# add any tag found in Info dictionary to table
next unless $unknown;
$tagInfo = NewPDFTag($tagTablePtr, $tag);
}
# increment document number if necessary
my ($oldDocNum, $oldNumTags);
if ($isSubDoc) {
$oldDocNum = $$et{DOC_NUM};
$oldNumTags = $$et{NUM_FOUND};
$$et{DOC_NUM} = ++$$et{DOC_COUNT};
}
if ($$tagInfo{SubDirectory}) {
# process the subdirectory
my @subDicts;
if (ref $val eq 'ARRAY') {
@subDicts = @{$val};
} else {
@subDicts = ( $val );
}
# loop through all values of this tag
for (;;) {
my $subDict = shift @subDicts or last;
# save last fetched object in case we fetch another one here
my $prevFetched = $lastFetched;
if (ref $subDict eq 'SCALAR') {
# only fetch once (other copies are obsolete)
next if $fetched{$$subDict};
if ($$tagInfo{IgnoreDuplicates}) {
my $flag = "ProcessedPDF_$tag";
if ($$et{$flag}) {
next if $et->WarnOnce("Ignored duplicate $tag dictionary", 2);
} else {
$$et{$flag} = 1;
}
}
# load dictionary via an indirect reference
$fetched{$$subDict} = 1;
my $obj = FetchObject($et, $$subDict, $xref, $tag);
unless (defined $obj) {
unless (defined $lastOffset) {
$et->Warn("Error reading $tag object ($$subDict)");
}
next;
}
$subDict = $obj;
}
if (ref $subDict eq 'ARRAY') {
# convert array of key/value pairs to a hash
next if @$subDict < 2;
my %hash = ( _tags => [] );
while (@$subDict >= 2) {
my $key = shift @$subDict;
$key =~ s/^\///;
push @{$hash{_tags}}, $key;
$hash{$key} = shift @$subDict;
}
$subDict = \%hash;
} else {
next unless ref $subDict eq 'HASH';
}
# set flag to re-crypt all strings when rewriting if the dictionary
# came from an encrypted stream
$$subDict{_needCrypt}{'*'} = 1 unless $lastFetched;
my $subTablePtr = GetTagTable($tagInfo->{SubDirectory}->{TagTable});
if (not $verbose) {
my $proc = $$subTablePtr{PROCESS_PROC} || \&ProcessDict;
&$proc($et, $subTablePtr, $subDict, $xref, $nesting);
} elsif ($next) {
# handle 'Next' links at this level to avoid deep recursion
undef $next;
$index = 0;
$tagTablePtr = $subTablePtr;
$dict = $subDict;
@tags = @{$$subDict{_tags}};
$et->VerboseDir($tag, scalar(@tags));
} else {
my $oldIndent = $$et{INDENT};
my $oldDir = $$et{DIR_NAME};
$$et{INDENT} .= '| ';
$$et{DIR_NAME} = $tag;
$et->VerboseDir($tag, scalar(@{$$subDict{_tags}}));
ProcessDict($et, $subTablePtr, $subDict, $xref, $nesting);
$$et{INDENT} = $oldIndent;
$$et{DIR_NAME} = $oldDir;
}
$lastFetched = $prevFetched;
}
} else {
# fetch object if necessary
# (OS X 10.6 writes indirect objects in the Info dictionary!)
if (ref $val eq 'SCALAR') {
my $prevFetched = $lastFetched;
# (note: fetching the same object multiple times is OK here)
$val = FetchObject($et, $$val, $xref, $tag);
if (defined $val) {
$val = ReadPDFValue($val);
# set flag to re-encrypt if necessary if rewritten
$$dict{_needCrypt}{$tag} = ($lastFetched ? 0 : 1) if $cryptString;
$lastFetched = $prevFetched; # restore last fetched object reference
}
} else {
$val = ReadPDFValue($val);
}
# convert from UTF-16 (big endian) to UTF-8 or Latin if necessary
# unless this is binary data (hex-encoded strings would not have been converted)
if (ref $val) {
if (ref $val eq 'ARRAY') {
my $v;
foreach $v (@$val) {
$et->FoundTag($tagInfo, $v);
}
}
} elsif (defined $val) {
my $format = $$tagInfo{Format} || $$tagInfo{Writable} || 'string';
$val = ConvertPDFDate($val) if $format eq 'date';
if (not $$tagInfo{Binary} and $val =~ /[\x18-\x1f\x80-\xff]/) {
# text string is already in Unicode if it starts with "\xfe\xff",
# otherwise we must first convert from PDFDocEncoding
$val = $et->Decode($val, ($val=~s/^\xfe\xff// ? 'UCS2' : 'PDFDoc'), 'MM');
}
if ($$tagInfo{List} and not $$et{OPTIONS}{NoPDFList}) {
# separate tokens in comma or whitespace delimited lists
my @values = ($val =~ /,/) ? split /,+\s*/, $val : split ' ', $val;
foreach $val (@values) {
$et->FoundTag($tagInfo, $val);
}
} else {
# a simple tag value
$et->FoundTag($tagInfo, $val);
}
}
}
if ($isSubDoc) {
# restore original document number
$$et{DOC_NUM} = $oldDocNum;
--$$et{DOC_COUNT} if $oldNumTags == $$et{NUM_FOUND};
}
}
#
# extract information from joined streams if necessary
#
if (%join) {
my ($tag, $i);
foreach $tag (sort keys %join) {
my $list = $join{$tag};
last unless defined $$list[1] and $$list[1] =~ /^%.*?([\x0d\x0a]*)/;
my $buff = "%!PS-Adobe-3.0$1"; # add PS header with same line break
for ($i=1; defined $$list[$i]; ++$i) {
$buff .= $$list[$i];
undef $$list[$i]; # free memory
}
# increment document number for tags extracted from embedded EPS
my $oldDocNum = $$et{DOC_NUM};
my $oldNumTags = $$et{NUM_FOUND};
$$et{DOC_NUM} = ++$$et{DOC_COUNT};
# extract PostScript information
$et->HandleTag($tagTablePtr, $tag, $buff);
$$et{DOC_NUM} = $oldDocNum;
# revert document counter if we didn't add any new tags
--$$et{DOC_COUNT} if $oldNumTags == $$et{NUM_FOUND};
delete $$et{DOC_NUM};
}
}
#
# extract information from stream object if it exists (eg. Metadata stream)
#
for (;;) { # (cheap goto)
last unless $$dict{_stream};
my $tag = '_stream';
# add Subtype (if it exists) to stream name and remove leading '/'
($tag = $$dict{Subtype} . $tag) =~ s/^\/// if $$dict{Subtype};
last unless $$tagTablePtr{$tag};
my $tagInfo = $et->GetTagInfo($tagTablePtr, $tag) or last;
unless ($$tagInfo{SubDirectory}) {
# don't build filter lists across different images
delete $$et{LIST_TAGS}{$$tagTablePtr{Filter}};
# we arrive here only when extracting embedded images
# - only extract known image types and ignore others
my $filter = $$dict{Filter} || '';
$filter = @$filter[-1] if ref $filter eq 'ARRAY'; # (get last Filter type)
my $result;
if ($filter eq '/DCTDecode' or $filter eq '/JPXDecode') {
DecodeStream($et, $dict) or last;
# save the image itself
$et->FoundTag($tagInfo, \$$dict{_stream});
# extract information from embedded image
$result = $et->ExtractInfo(\$$dict{_stream}, { ReEntry => 1 });
}
unless ($result) {
$et->FoundTag('FileType', defined $result ? '(unknown)' : '(unsupported)');
}
last;
}
# decode stream if necessary
DecodeStream($et, $dict) or last;
if ($verbose > 2) {
$et->VPrint(2,"$$et{INDENT}$$et{DIR_NAME} stream data\n");
$et->VerboseDump(\$$dict{_stream});
}
# extract information from stream
my %dirInfo = (
DataPt => \$$dict{_stream},
DataLen => length $$dict{_stream},
DirStart => 0,
DirLen => length $$dict{_stream},
Parent => 'PDF',
);
my $subTablePtr = GetTagTable($tagInfo->{SubDirectory}->{TagTable});
unless ($et->ProcessDirectory(\%dirInfo, $subTablePtr)) {
$et->Warn("Error processing $$tagInfo{Name} information");
}
last;
}
}
#------------------------------------------------------------------------------
# Extract information from PDF file
# Inputs: 0) ExifTool object reference, 1) dirInfo reference
# Returns: 0 if not a PDF file, 1 on success, otherwise a negative error number
sub ReadPDF($$)
{
my ($et, $dirInfo) = @_;
my $raf = $$dirInfo{RAF};
my $verbose = $et->Options('Verbose');
my ($buff, $encrypt, $id);
#
# validate PDF file
#
# (linearization dictionary must be in the first 1024 bytes of the file)
$raf->Read($buff, 1024) >= 8 or return 0;
$buff =~ /^(\s*)%PDF-(\d+\.\d+)/ or return 0;
$$et{PDFBase} = length $1 and $et->Warn('PDF header is not at start of file',1);
$pdfVer = $2;
$et->SetFileType(); # set the FileType tag
$et->Warn("May not be able to read a PDF version $pdfVer file") if $pdfVer >= 2.0;
# store PDFVersion tag
my $tagTablePtr = GetTagTable('Image::ExifTool::PDF::Root');
$et->HandleTag($tagTablePtr, 'Version', $pdfVer);
$tagTablePtr = GetTagTable('Image::ExifTool::PDF::Main');
#
# check for a linearized PDF (only if reading)
#
my $capture = $$et{PDF_CAPTURE};
unless ($capture) {
my $lin = 'false';
if ($buff =~ /<</g) {
$buff = substr($buff, pos($buff) - 2);
my $dict = ExtractObject($et, \$buff);
if (ref $dict eq 'HASH' and $$dict{Linearized} and $$dict{L}) {
if (not $$et{VALUE}{FileSize}) {
undef $lin; # can't determine if it is linearized
} elsif ($$dict{L} == $$et{VALUE}{FileSize} - $$et{PDFBase}) {
$lin = 'true';
}
}
}
$et->HandleTag($tagTablePtr, '_linearized', $lin) if $lin;
}
#
# read the xref tables referenced from startxref at the end of the file
#
my @xrefOffsets;
$raf->Seek(0, 2) or return -2;
# the %%EOF must occur within the last 1024 bytes of the file (PDF spec, appendix H)
my $len = $raf->Tell();
$len = 1024 if $len > 1024;
$raf->Seek(-$len, 2) or return -2;
$raf->Read($buff, $len) == $len or return -3;
# find the LAST xref table in the file (may be multiple %%EOF marks,
# and comments between "startxref" and "%%EOF")
$buff =~ /^.*startxref(\s+)(\d+)(\s+)(%[^\x0d\x0a]*\s+)*%%EOF/s or return -4;
my $ws = $1 . $3;
my $xr = $2;
push @xrefOffsets, $xr, 'Main';
# set input record separator
local $/ = $ws =~ /(\x0d\x0a|\x0d|\x0a)/ ? $1 : "\x0a";
my (%xref, @mainDicts, %loaded, $mainFree);
# initialize variables to capture when rewriting
if ($capture) {
$capture->{startxref} = $xr;
$capture->{xref} = \%xref;
$capture->{newline} = $/;
$capture->{mainFree} = $mainFree = { };
}
XRef:
while (@xrefOffsets) {
my $offset = shift @xrefOffsets;
my $type = shift @xrefOffsets;
next if $loaded{$offset}; # avoid infinite recursion
unless ($raf->Seek($offset+$$et{PDFBase}, 0)) {
%loaded or return -5;
$et->Warn('Bad offset for secondary xref table');
next;
}
# Note: care must be taken because ReadLine may read more than we want if
# the newline sequence for this table is different than the rest of the file
for (;;) {
unless ($raf->ReadLine($buff)) {
%loaded or return -6;
$et->Warn('Bad offset for secondary xref table');
next XRef;
}
last if $buff =~/\S/; # skip blank lines
}
my $loadXRefStream;
if ($buff =~ s/^\s*xref\s+//s) {
# load xref table
for (;;) {
# read another line if necessary (skipping blank lines)
$raf->ReadLine($buff) or return -6 until $buff =~ /\S/;
last if $buff =~ s/^\s*trailer([\s<[(])/$1/s;
$buff =~ s/^\s*(\d+)\s+(\d+)\s+//s or return -4;
my ($start, $num) = ($1, $2);
$raf->Seek(-length($buff), 1) or return -4;
my $i;
for ($i=0; $i<$num; ++$i) {
$raf->Read($buff, 20) == 20 or return -6;
$buff =~ /^\s*(\d{10}) (\d{5}) (f|n)/s or return -4;
my $num = $start + $i;
# save offset for newest copy of all objects
# (or next object number for free objects)
unless (defined $xref{$num}) {
my ($offset, $gen) = (int($1), int($2));
$xref{$num} = $offset;
if ($3 eq 'f') {
# save free objects in last xref table for rewriting
$$mainFree{$num} = [ $offset, $gen, 'f' ] if $mainFree;
next;
}
# also save offset keyed by object reference string
$xref{"$num $gen R"} = $offset;
}
}
# (I have a sample from Adobe which has an empty xref table)
# %xref or return -4; # xref table may not be empty
$buff = '';
}
undef $mainFree; # only do this for the last xref table
} elsif ($buff =~ s/^\s*(\d+)\s+(\d+)\s+obj//s) {
# this is a PDF-1.5 cross-reference stream dictionary
$loadXRefStream = 1;
} else {
%loaded or return -4;
$et->Warn('Invalid secondary xref table');
next;
}
my $mainDict = ExtractObject($et, \$buff, $raf, \%xref);
unless (ref $mainDict eq 'HASH') {
%loaded or return -8;
$et->Warn('Error loading secondary dictionary');
next;
}
if ($loadXRefStream) {
# decode and save our XRef stream from PDF-1.5 file
# (but parse it later as required to save time)
# Note: this technique can potentially result in an old object
# being used if the file was incrementally updated and an older
# object from an xref table was replaced by a newer object in an
# xref stream. But doing so isn't a good idea (if allowed at all)
# because a PDF 1.4 consumer would also make this same mistake.
if ($$mainDict{Type} eq '/XRef' and $$mainDict{W} and
@{$$mainDict{W}} > 2 and $$mainDict{Size} and
DecodeStream($et, $mainDict))
{
# create Index entry if it doesn't exist
$$mainDict{Index} or $$mainDict{Index} = [ 0, $$mainDict{Size} ];
# create '_entry_size' entry for internal use
my $w = $$mainDict{W};
my $size = 0;
foreach (@$w) { $size += $_; }
$$mainDict{_entry_size} = $size;
# save this stream dictionary to use later if required
$xref{dicts} = [] unless $xref{dicts};
push @{$xref{dicts}}, $mainDict;
} else {
%loaded or return -9;
$et->Warn('Invalid xref stream in secondary dictionary');
}
}
$loaded{$offset} = 1;
# load XRef stream in hybrid file if it exists
push @xrefOffsets, $$mainDict{XRefStm}, 'XRefStm' if $$mainDict{XRefStm};
$encrypt = $$mainDict{Encrypt} if $$mainDict{Encrypt};
if ($$mainDict{ID} and ref $$mainDict{ID} eq 'ARRAY') {
$id = ReadPDFValue($mainDict->{ID}->[0]);
}
push @mainDicts, $mainDict, $type;
# load previous xref table if it exists
push @xrefOffsets, $$mainDict{Prev}, 'Prev' if $$mainDict{Prev};
}
#
# extract encryption information if necessary
#
if ($encrypt) {
if (ref $encrypt eq 'SCALAR') {
$encrypt = FetchObject($et, $$encrypt, \%xref, 'Encrypt');
}
# generate Encryption tag information
my $err = DecryptInit($et, $encrypt, $id);
if ($err) {
$et->Warn($err);
$$capture{Error} = $err if $capture;
return -1;
}
}
#
# extract the information beginning with each of the main dictionaries
#
my $i = 0;
my $num = (scalar @mainDicts) / 2;
while (@mainDicts) {
my $dict = shift @mainDicts;
my $type = shift @mainDicts;
if ($verbose) {
++$i;
my $n = scalar(@{$$dict{_tags}});
$et->VPrint(0, "PDF dictionary ($i of $num) with $n entries:\n");
}
ProcessDict($et, $tagTablePtr, $dict, \%xref, 0, $type);
}
# handle any decryption errors
if ($encrypt) {
my $err = $$encrypt{_error};
if ($err) {
$et->Warn($err);
$$capture{Error} = $err if $capture;
return -1;
}
}
return 1;
}
#------------------------------------------------------------------------------
# ReadPDF() warning strings for each error return value
my %pdfWarning = (
# -1 is reserved as error return value with no associated warning
-2 => 'Error seeking in file',
-3 => 'Error reading file',
-4 => 'Invalid xref table',
-5 => 'Invalid xref offset',
-6 => 'Error reading xref table',
-7 => 'Error reading trailer',
-8 => 'Error reading main dictionary',
-9 => 'Invalid xref stream in main dictionary',
);
#------------------------------------------------------------------------------
# Extract information from PDF file
# Inputs: 0) ExifTool object reference, 1) dirInfo reference
# Returns: 1 if this was a valid PDF file
sub ProcessPDF($$)
{
my ($et, $dirInfo) = @_;
undef $cryptInfo; # (must not delete after returning so writer can use it)
undef $cryptStream;
undef $cryptString;
my $result = ReadPDF($et, $dirInfo);
if ($result < 0) {
$et->Warn($pdfWarning{$result}) if $pdfWarning{$result};
$result = 1;
}
# clean up and return
undef %streamObjs;
undef %fetched;
return $result;
}
1; # end
__END__
=head1 NAME
Image::ExifTool::PDF - Read PDF meta information
=head1 SYNOPSIS
This module is loaded automatically by Image::ExifTool when required.
=head1 DESCRIPTION
This code reads meta information from PDF (Adobe Portable Document Format)
files. It supports object streams introduced in PDF-1.5 but only with a
limited set of Filter and Predictor algorithms, however all standard
encryption methods through PDF-1.7 extension level 3 are supported,
including AESV2 (AES-128) and AESV3 (AES-256).
=head1 AUTHOR
Copyright 2003-2018, Phil Harvey (phil at owl.phy.queensu.ca)
This library is free software; you can redistribute it and/or modify it
under the same terms as Perl itself.
=head1 REFERENCES
=over 4
=item L<http://partners.adobe.com/public/developer/pdf/index_reference.html>
=item L<Crypt::RC4|Crypt::RC4>
=item L<http://www.adobe.com/devnet/acrobat/pdfs/PDF32000_2008.pdf>
=item L<http://www.adobe.com/content/dam/Adobe/en/devnet/pdf/pdfs/adobe_supplement_iso32000.pdf>
=item L<http://tools.ietf.org/search/rfc3454>
=item L<http://www.armware.dk/RFC/rfc/rfc4013.html>
=back
=head1 SEE ALSO
L<Image::ExifTool::TagNames/PDF Tags>,
L<Image::ExifTool(3pm)|Image::ExifTool>
=cut