JpegMetadataExtractor.php

Go to the documentation of this file.

<?php
class JpegMetadataExtractor {
    const MAX_JPEG_SEGMENTS = 200;
 
    // the max segment is a sanity check.
    // A jpeg file should never even remotely have
    // that many segments. Your average file has about 10.
 
    static function segmentSplitter( $filename ) {
        $showXMP = XMPReader::isSupported();
 
        $segmentCount = 0;
 
        $segments = [
            'XMP_ext' => [],
            'COM' => [],
            'PSIR' => [],
        ];
 
        if ( !$filename ) {
            throw new MWException( "No filename specified for " . __METHOD__ );
        }
        if ( !file_exists( $filename ) || is_dir( $filename ) ) {
            throw new MWException( "Invalid file $filename passed to " . __METHOD__ );
        }
 
        $fh = fopen( $filename, "rb" );
 
        if ( !$fh ) {
            throw new MWException( "Could not open file $filename" );
        }
 
        $buffer = fread( $fh, 2 );
        if ( $buffer !== "\xFF\xD8" ) {
            throw new MWException( "Not a jpeg, no SOI" );
        }
        while ( !feof( $fh ) ) {
            $buffer = fread( $fh, 1 );
            $segmentCount++;
            if ( $segmentCount > self::MAX_JPEG_SEGMENTS ) {
                // this is just a sanity check
                throw new MWException( 'Too many jpeg segments. Aborting' );
            }
            while ( $buffer !== "\xFF" && !feof( $fh ) ) {
                // In theory JPEG files are not allowed to contain anything between the sections,
                // but in practice they sometimes do. It's customary to ignore the garbage data.
                $buffer = fread( $fh, 1 );
            }
 
            $buffer = fread( $fh, 1 );
            while ( $buffer === "\xFF" && !feof( $fh ) ) {
                // Skip through any 0xFF padding bytes.
                $buffer = fread( $fh, 1 );
            }
            if ( $buffer === "\xFE" ) {
                // COM section -- file comment
                // First see if valid utf-8,
                // if not try to convert it to windows-1252.
                $com = $oldCom = trim( self::jpegExtractMarker( $fh ) );
                UtfNormal\Validator::quickIsNFCVerify( $com );
                // turns $com to valid utf-8.
                // thus if no change, its utf-8, otherwise its something else.
                if ( $com !== $oldCom ) {
                    Wikimedia\suppressWarnings();
                    $com = $oldCom = iconv( 'windows-1252', 'UTF-8//IGNORE', $oldCom );
                    Wikimedia\restoreWarnings();
                }
                // Try it again, if its still not a valid string, then probably
                // binary junk or some really weird encoding, so don't extract.
                UtfNormal\Validator::quickIsNFCVerify( $com );
                if ( $com === $oldCom ) {
                    $segments["COM"][] = $oldCom;
                } else {
                    wfDebug( __METHOD__ . " Ignoring JPEG comment as is garbage.\n" );
                }
            } elseif ( $buffer === "\xE1" ) {
                // APP1 section (Exif, XMP, and XMP extended)
                // only extract if XMP is enabled.
                $temp = self::jpegExtractMarker( $fh );
                // check what type of app segment this is.
                if ( substr( $temp, 0, 29 ) === "http://ns.adobe.com/xap/1.0/\x00" && $showXMP ) {
                    $segments["XMP"] = substr( $temp, 29 );
                } elseif ( substr( $temp, 0, 35 ) === "http://ns.adobe.com/xmp/extension/\x00" && $showXMP ) {
                    $segments["XMP_ext"][] = substr( $temp, 35 );
                } elseif ( substr( $temp, 0, 29 ) === "XMP\x00://ns.adobe.com/xap/1.0/\x00" && $showXMP ) {
                    // Some images (especially flickr images) seem to have this.
                    // I really have no idea what the deal is with them, but
                    // whatever...
                    $segments["XMP"] = substr( $temp, 29 );
                    wfDebug( __METHOD__ . ' Found XMP section with wrong app identifier '
                        . "Using anyways.\n" );
                } elseif ( substr( $temp, 0, 6 ) === "Exif\0\0" ) {
                    // Just need to find out what the byte order is.
                    // because php's exif plugin sucks...
                    // This is a II for little Endian, MM for big. Not a unicode BOM.
                    $byteOrderMarker = substr( $temp, 6, 2 );
                    if ( $byteOrderMarker === 'MM' ) {
                        $segments['byteOrder'] = 'BE';
                    } elseif ( $byteOrderMarker === 'II' ) {
                        $segments['byteOrder'] = 'LE';
                    } else {
                        wfDebug( __METHOD__ . " Invalid byte ordering?!\n" );
                    }
                }
            } elseif ( $buffer === "\xED" ) {
                // APP13 - PSIR. IPTC and some photoshop stuff
                $temp = self::jpegExtractMarker( $fh );
                if ( substr( $temp, 0, 14 ) === "Photoshop 3.0\x00" ) {
                    $segments["PSIR"][] = $temp;
                }
            } elseif ( $buffer === "\xD9" || $buffer === "\xDA" ) {
                // EOI - end of image or SOS - start of scan. either way we're past any interesting segments
                return $segments;
            } else {
                // segment we don't care about, so skip
                $size = wfUnpack( "nint", fread( $fh, 2 ), 2 );
                if ( $size['int'] < 2 ) {
                    throw new MWException( "invalid marker size in jpeg" );
                }
                // Note it's possible to seek beyond end of file if truncated.
                // fseek doesn't report a failure in this case.
                fseek( $fh, $size['int'] - 2, SEEK_CUR );
            }
        }
        // shouldn't get here.
        throw new MWException( "Reached end of jpeg file unexpectedly" );
    }
 
    private static function jpegExtractMarker( &$fh ) {
        $size = wfUnpack( "nint", fread( $fh, 2 ), 2 );
        if ( $size['int'] < 2 ) {
            throw new MWException( "invalid marker size in jpeg" );
        }
        if ( $size['int'] === 2 ) {
            // fread( ..., 0 ) generates a warning
            return '';
        }
        $segment = fread( $fh, $size['int'] - 2 );
        if ( strlen( $segment ) !== $size['int'] - 2 ) {
            throw new MWException( "Segment shorter than expected" );
        }
 
        return $segment;
    }
 
    public static function doPSIR( $app13 ) {
        if ( !$app13 ) {
            throw new MWException( "No App13 segment given" );
        }
        // First compare hash with real thing
        // 0x404 contains IPTC, 0x425 has hash
        // This is used to determine if the iptc is newer than
        // the xmp data, as xmp programs update the hash,
        // where non-xmp programs don't.
 
        $offset = 14; // skip past PHOTOSHOP 3.0 identifier. should already be checked.
        $appLen = strlen( $app13 );
        $realHash = "";
        $recordedHash = "";
 
        // the +12 is the length of an empty item.
        while ( $offset + 12 <= $appLen ) {
            $valid = true;
            if ( substr( $app13, $offset, 4 ) !== '8BIM' ) {
                // its supposed to be 8BIM
                // but apparently sometimes isn't esp. in
                // really old jpg's
                $valid = false;
            }
            $offset += 4;
            $id = substr( $app13, $offset, 2 );
            // id is a 2 byte id number which identifies
            // the piece of info this record contains.
 
            $offset += 2;
 
            // some record types can contain a name, which
            // is a pascal string 0-padded to be an even
            // number of bytes. Most times (and any time
            // we care) this is empty, making it two null bytes.
 
            $lenName = ord( substr( $app13, $offset, 1 ) ) + 1;
            // we never use the name so skip it. +1 for length byte
            if ( $lenName % 2 == 1 ) {
                $lenName++;
            } // pad to even.
            $offset += $lenName;
 
            // now length of data (unsigned long big endian)
            $lenData = wfUnpack( 'Nlen', substr( $app13, $offset, 4 ), 4 );
            // PHP can take issue with very large unsigned ints and make them negative.
            // Which should never ever happen, as this has to be inside a segment
            // which is limited to a 16 bit number.
            if ( $lenData['len'] < 0 ) {
                throw new MWException( "Too big PSIR (" . $lenData['len'] . ')' );
            }
 
            $offset += 4; // 4bytes length field;
 
            // this should not happen, but check.
            if ( $lenData['len'] + $offset > $appLen ) {
                throw new MWException( "PSIR data too long. (item length=" . $lenData['len']
                    . "; offset=$offset; total length=$appLen)" );
            }
 
            if ( $valid ) {
                switch ( $id ) {
                    case "\x04\x04":
                        // IPTC block
                        $realHash = md5( substr( $app13, $offset, $lenData['len'] ), true );
                        break;
                    case "\x04\x25":
                        $recordedHash = substr( $app13, $offset, $lenData['len'] );
                        break;
                }
            }
 
            // if odd, add 1 to length to account for
            // null pad byte.
            if ( $lenData['len'] % 2 == 1 ) {
                $lenData['len']++;
            }
            $offset += $lenData['len'];
        }
 
        if ( !$realHash || !$recordedHash ) {
            return 'iptc-no-hash';
        } elseif ( $realHash === $recordedHash ) {
            return 'iptc-good-hash';
        } else { /*$realHash !== $recordedHash */
            return 'iptc-bad-hash';
        }
    }
}