UIDGenerator.php

Go to the documentation of this file.

<?php
class UIDGenerator {
    protected static $instance = null;

    protected $nodeIdFile; // string; local file path
    protected $nodeId32; // string; node ID in binary (32 bits)
    protected $nodeId48; // string; node ID in binary (48 bits)

    protected $lockFile88; // string; local file path
    protected $lockFile128; // string; local file path

    protected $fileHandles = array(); // cache file handles

    const QUICK_RAND = 1; // get randomness from fast and insecure sources
    const QUICK_VOLATILE = 2; // use an APC like in-memory counter if available

    protected function __construct() {
        $this->nodeIdFile = wfTempDir() . '/mw-' . __CLASS__ . '-UID-nodeid';
        $nodeId = '';
        if ( is_file( $this->nodeIdFile ) ) {
            $nodeId = file_get_contents( $this->nodeIdFile );
        }
        // Try to get some ID that uniquely identifies this machine (RFC 4122)...
        if ( !preg_match( '/^[0-9a-f]{12}$/i', $nodeId ) ) {
            wfSuppressWarnings();
            if ( wfIsWindows() ) {
                // http://technet.microsoft.com/en-us/library/bb490913.aspx
                $csv = trim( wfShellExec( 'getmac /NH /FO CSV' ) );
                $line = substr( $csv, 0, strcspn( $csv, "\n" ) );
                $info = str_getcsv( $line );
                $nodeId = isset( $info[0] ) ? str_replace( '-', '', $info[0] ) : '';
            } elseif ( is_executable( '/sbin/ifconfig' ) ) { // Linux/BSD/Solaris/OS X
                // See http://linux.die.net/man/8/ifconfig
                $m = array();
                preg_match( '/\s([0-9a-f]{2}(:[0-9a-f]{2}){5})\s/',
                    wfShellExec( '/sbin/ifconfig -a' ), $m );
                $nodeId = isset( $m[1] ) ? str_replace( ':', '', $m[1] ) : '';
            }
            wfRestoreWarnings();
            if ( !preg_match( '/^[0-9a-f]{12}$/i', $nodeId ) ) {
                $nodeId = MWCryptRand::generateHex( 12, true );
                $nodeId[1] = dechex( hexdec( $nodeId[1] ) | 0x1 ); // set multicast bit
            }
            file_put_contents( $this->nodeIdFile, $nodeId ); // cache
        }
        $this->nodeId32 = wfBaseConvert( substr( sha1( $nodeId ), 0, 8 ), 16, 2, 32 );
        $this->nodeId48 = wfBaseConvert( $nodeId, 16, 2, 48 );
        // If different processes run as different users, they may have different temp dirs.
        // This is dealt with by initializing the clock sequence number and counters randomly.
        $this->lockFile88 = wfTempDir() . '/mw-' . __CLASS__ . '-UID-88';
        $this->lockFile128 = wfTempDir() . '/mw-' . __CLASS__ . '-UID-128';
    }

    protected static function singleton() {
        if ( self::$instance === null ) {
            self::$instance = new self();
        }

        return self::$instance;
    }

    public static function newTimestampedUID88( $base = 10 ) {
        if ( !is_integer( $base ) || $base > 36 || $base < 2 ) {
            throw new MWException( "Base must an integer be between 2 and 36" );
        }
        $gen = self::singleton();
        $time = $gen->getTimestampAndDelay( 'lockFile88', 1, 1024 );

        return wfBaseConvert( $gen->getTimestampedID88( $time ), 2, $base );
    }

    protected function getTimestampedID88( array $info ) {
        list( $time, $counter ) = $info;
        // Take the 46 MSBs of "milliseconds since epoch"
        $id_bin = $this->millisecondsSinceEpochBinary( $time );
        // Add a 10 bit counter resulting in 56 bits total
        $id_bin .= str_pad( decbin( $counter ), 10, '0', STR_PAD_LEFT );
        // Add the 32 bit node ID resulting in 88 bits total
        $id_bin .= $this->nodeId32;
        // Convert to a 1-27 digit integer string
        if ( strlen( $id_bin ) !== 88 ) {
            throw new MWException( "Detected overflow for millisecond timestamp." );
        }

        return $id_bin;
    }

    public static function newTimestampedUID128( $base = 10 ) {
        if ( !is_integer( $base ) || $base > 36 || $base < 2 ) {
            throw new MWException( "Base must be an integer between 2 and 36" );
        }
        $gen = self::singleton();
        $time = $gen->getTimestampAndDelay( 'lockFile128', 16384, 1048576 );

        return wfBaseConvert( $gen->getTimestampedID128( $time ), 2, $base );
    }

    protected function getTimestampedID128( array $info ) {
        list( $time, $counter, $clkSeq ) = $info;
        // Take the 46 MSBs of "milliseconds since epoch"
        $id_bin = $this->millisecondsSinceEpochBinary( $time );
        // Add a 20 bit counter resulting in 66 bits total
        $id_bin .= str_pad( decbin( $counter ), 20, '0', STR_PAD_LEFT );
        // Add a 14 bit clock sequence number resulting in 80 bits total
        $id_bin .= str_pad( decbin( $clkSeq ), 14, '0', STR_PAD_LEFT );
        // Add the 48 bit node ID resulting in 128 bits total
        $id_bin .= $this->nodeId48;
        // Convert to a 1-39 digit integer string
        if ( strlen( $id_bin ) !== 128 ) {
            throw new MWException( "Detected overflow for millisecond timestamp." );
        }

        return $id_bin;
    }

    public static function newUUIDv4( $flags = 0 ) {
        $hex = ( $flags & self::QUICK_RAND )
            ? wfRandomString( 31 )
            : MWCryptRand::generateHex( 31 );

        return sprintf( '%s-%s-%s-%s-%s',
            // "time_low" (32 bits)
            substr( $hex, 0, 8 ),
            // "time_mid" (16 bits)
            substr( $hex, 8, 4 ),
            // "time_hi_and_version" (16 bits)
            '4' . substr( $hex, 12, 3 ),
            // "clk_seq_hi_res (8 bits, variant is binary 10x) and "clk_seq_low" (8 bits)
            dechex( 0x8 | ( hexdec( $hex[15] ) & 0x3 ) ) . $hex[16] . substr( $hex, 17, 2 ),
            // "node" (48 bits)
            substr( $hex, 19, 12 )
        );
    }

    public static function newRawUUIDv4( $flags = 0 ) {
        return str_replace( '-', '', self::newUUIDv4( $flags ) );
    }

    public static function newSequentialPerNodeID( $bucket, $bits = 48, $flags = 0 ) {
        return current( self::newSequentialPerNodeIDs( $bucket, $bits, 1, $flags ) );
    }

    public static function newSequentialPerNodeIDs( $bucket, $bits, $count, $flags = 0 ) {
        $gen = self::singleton();
        return $gen->getSequentialPerNodeIDs( $bucket, $bits, $count, $flags );
    }

    protected function getSequentialPerNodeIDs( $bucket, $bits, $count, $flags ) {
        if ( $count <= 0 ) {
            return array(); // nothing to do
        } elseif ( $count > 10000 ) {
            throw new MWException( "Number of requested IDs ($count) is too high." );
        } elseif ( $bits < 16 || $bits > 48 ) {
            throw new MWException( "Requested bit size ($bits) is out of range." );
        }

        $counter = null; // post-increment persistent counter value

        // Use APC/eAccelerator/xcache if requested, available, and not in CLI mode;
        // Counter values would not survive accross script instances in CLI mode.
        $cache = null;
        if ( ( $flags & self::QUICK_VOLATILE ) && PHP_SAPI !== 'cli' ) {
            try {
                $cache = ObjectCache::newAccelerator( array() );
            } catch ( MWException $e ) {} // not supported
        }
        if ( $cache ) {
            $counter = $cache->incr( $bucket, $count );
            if ( $counter === false ) {
                if ( !$cache->add( $bucket, $count ) ) {
                    throw new MWException( 'Unable to set value to ' . get_class( $cache ) );
                }
                $counter = $count;
            }
        }

        // Note: use of fmod() avoids "division by zero" on 32 bit machines
        if ( $counter === null ) {
            $path = wfTempDir() . '/mw-' . __CLASS__ . '-' . rawurlencode( $bucket ) . '-48';
            // Get the UID lock file handle
            if ( isset( $this->fileHandles[$path] ) ) {
                $handle = $this->fileHandles[$path];
            } else {
                $handle = fopen( $path, 'cb+' );
                $this->fileHandles[$path] = $handle ?: null; // cache
            }
            // Acquire the UID lock file
            if ( $handle === false ) {
                throw new MWException( "Could not open '{$path}'." );
            } elseif ( !flock( $handle, LOCK_EX ) ) {
                fclose( $handle );
                throw new MWException( "Could not acquire '{$path}'." );
            }
            // Fetch the counter value and increment it...
            rewind( $handle );
            $counter = floor( trim( fgets( $handle ) ) ) + $count; // fetch as float
            // Write back the new counter value
            ftruncate( $handle, 0 );
            rewind( $handle );
            fwrite( $handle, fmod( $counter, pow( 2, 48 ) ) ); // warp-around as needed
            fflush( $handle );
            // Release the UID lock file
            flock( $handle, LOCK_UN );
        }

        $ids = array();
        $divisor = pow( 2, $bits );
        $currentId = floor( $counter - $count ); // pre-increment counter value
        for ( $i = 0; $i < $count; ++$i ) {
            $ids[] = fmod( ++$currentId, $divisor );
        }

        return $ids;
    }

    protected function getTimestampAndDelay( $lockFile, $clockSeqSize, $counterSize ) {
        // Get the UID lock file handle
        $path = $this->$lockFile;
        if ( isset( $this->fileHandles[$path] ) ) {
            $handle = $this->fileHandles[$path];
        } else {
            $handle = fopen( $path, 'cb+' );
            $this->fileHandles[$path] = $handle ?: null; // cache
        }
        // Acquire the UID lock file
        if ( $handle === false ) {
            throw new MWException( "Could not open '{$this->$lockFile}'." );
        } elseif ( !flock( $handle, LOCK_EX ) ) {
            fclose( $handle );
            throw new MWException( "Could not acquire '{$this->$lockFile}'." );
        }
        // Get the current timestamp, clock sequence number, last time, and counter
        rewind( $handle );
        $data = explode( ' ', fgets( $handle ) ); // "<clk seq> <sec> <msec> <counter> <offset>"
        $clockChanged = false; // clock set back significantly?
        if ( count( $data ) == 5 ) { // last UID info already initialized
            $clkSeq = (int)$data[0] % $clockSeqSize;
            $prevTime = array( (int)$data[1], (int)$data[2] );
            $offset = (int)$data[4] % $counterSize; // random counter offset
            $counter = 0; // counter for UIDs with the same timestamp
            // Delay until the clock reaches the time of the last ID.
            // This detects any microtime() drift among processes.
            $time = $this->timeWaitUntil( $prevTime );
            if ( !$time ) { // too long to delay?
                $clockChanged = true; // bump clock sequence number
                $time = self::millitime();
            } elseif ( $time == $prevTime ) {
                // Bump the counter if there are timestamp collisions
                $counter = (int)$data[3] % $counterSize;
                if ( ++$counter >= $counterSize ) { // sanity (starts at 0)
                    flock( $handle, LOCK_UN ); // abort
                    throw new MWException( "Counter overflow for timestamp value." );
                }
            }
        } else { // last UID info not initialized
            $clkSeq = mt_rand( 0, $clockSeqSize - 1 );
            $counter = 0;
            $offset = mt_rand( 0, $counterSize - 1 );
            $time = self::millitime();
        }
        // microtime() and gettimeofday() can drift from time() at least on Windows.
        // The drift is immediate for processes running while the system clock changes.
        // time() does not have this problem. See https://bugs.php.net/bug.php?id=42659.
        if ( abs( time() - $time[0] ) >= 2 ) {
            // We don't want processes using too high or low timestamps to avoid duplicate
            // UIDs and clock sequence number churn. This process should just be restarted.
            flock( $handle, LOCK_UN ); // abort
            throw new MWException( "Process clock is outdated or drifted." );
        }
        // If microtime() is synced and a clock change was detected, then the clock went back
        if ( $clockChanged ) {
            // Bump the clock sequence number and also randomize the counter offset,
            // which is useful for UIDs that do not include the clock sequence number.
            $clkSeq = ( $clkSeq + 1 ) % $clockSeqSize;
            $offset = mt_rand( 0, $counterSize - 1 );
            trigger_error( "Clock was set back; sequence number incremented." );
        }
        // Update the (clock sequence number, timestamp, counter)
        ftruncate( $handle, 0 );
        rewind( $handle );
        fwrite( $handle, "{$clkSeq} {$time[0]} {$time[1]} {$counter} {$offset}" );
        fflush( $handle );
        // Release the UID lock file
        flock( $handle, LOCK_UN );

        return array( $time, ( $counter + $offset ) % $counterSize, $clkSeq );
    }

    protected function timeWaitUntil( array $time ) {
        do {
            $ct = self::millitime();
            if ( $ct >= $time ) { // http://php.net/manual/en/language.operators.comparison.php
                return $ct; // current timestamp is higher than $time
            }
        } while ( ( ( $time[0] - $ct[0] ) * 1000 + ( $time[1] - $ct[1] ) ) <= 10 );

        return false;
    }

    protected function millisecondsSinceEpochBinary( array $time ) {
        list( $sec, $msec ) = $time;
        $ts = 1000 * $sec + $msec;
        if ( $ts > pow( 2, 52 ) ) {
            throw new MWException( __METHOD__ .
                ': sorry, this function doesn\'t work after the year 144680' );
        }

        return substr( wfBaseConvert( $ts, 10, 2, 46 ), -46 );
    }

    protected static function millitime() {
        list( $msec, $sec ) = explode( ' ', microtime() );

        return array( (int)$sec, (int)( $msec * 1000 ) );
    }

    protected function deleteCacheFiles() {
        // Bug: 44850
        foreach ( $this->fileHandles as $path => $handle ) {
            if ( $handle !== null ) {
                fclose( $handle );
            }
            if ( is_file( $path ) ) {
                unlink( $path );
            }
            unset( $this->fileHandles[$path] );
        }
        if ( is_file( $this->nodeIdFile ) ) {
            unlink( $this->nodeIdFile );
        }
    }

    public static function unitTestTearDown() {
        // Bug: 44850
        $gen = self::singleton();
        $gen->deleteCacheFiles();
    }

    function __destruct() {
        array_map( 'fclose', array_filter( $this->fileHandles ) );
    }
}