1 package DBM::Deep::Engine;
5 use Fcntl qw( :DEFAULT :flock :seek );
8 # Setup file and tag signatures. These should never change.
10 sub SIG_FILE () { 'DPDB' }
11 sub SIG_INTERNAL () { 'i' }
12 sub SIG_HASH () { 'H' }
13 sub SIG_ARRAY () { 'A' }
14 sub SIG_NULL () { 'N' }
15 sub SIG_DATA () { 'D' }
16 sub SIG_INDEX () { 'I' }
17 sub SIG_BLIST () { 'B' }
18 sub SIG_FREE () { 'F' }
23 # Precalculate index, bucket and bucket list sizes
27 $self->{index_size} = (2**8) * $self->{long_size};
28 $self->{bucket_size} = $self->{hash_size} + $self->{long_size} * 2;
29 $self->{bucket_list_size} = $self->{max_buckets} * $self->{bucket_size};
36 # Set pack/unpack modes (see file header for more)
39 my ($long_s, $long_p, $data_s, $data_p) = @_;
42 # Set to 4 and 'N' for 32-bit offset tags (default). Theoretical limit of 4
44 # (Perl must be compiled with largefile support for files > 2 GB)
46 # Set to 8 and 'Q' for 64-bit offsets. Theoretical limit of 16 XB per file.
47 # (Perl must be compiled with largefile and 64-bit long support)
49 $self->{long_size} = $long_s ? $long_s : 4;
50 $self->{long_pack} = $long_p ? $long_p : 'N';
53 # Set to 4 and 'N' for 32-bit data length prefixes. Limit of 4 GB for each
54 # key/value. Upgrading this is possible (see above) but probably not
55 # necessary. If you need more than 4 GB for a single key or value, this
56 # module is really not for you :-)
58 $self->{data_size} = $data_s ? $data_s : 4;
59 $self->{data_pack} = $data_p ? $data_p : 'N';
61 return $self->precalc_sizes();
66 # Set key digest function (default is MD5)
69 my ($digest_func, $hash_size) = @_;
71 $self->{digest} = $digest_func ? $digest_func : \&Digest::MD5::md5;
72 $self->{hash_size} = $hash_size ? $hash_size : 16;
74 return $self->precalc_sizes();
87 digest => \&Digest::MD5::md5,
91 # Maximum number of buckets per list before another level of indexing is
93 # Increase this value for slightly greater speed, but larger database
94 # files. DO NOT decrease this value below 16, due to risk of recursive
100 $self->precalc_sizes;
105 sub write_file_signature {
111 my $loc = $self->_request_space( $obj, length( SIG_FILE ) );
112 seek($fh, $loc + $obj->_root->{file_offset}, SEEK_SET);
113 print( $fh SIG_FILE);
122 $self->open( $obj ) if !defined $obj->_fh;
127 unless ( $obj->{base_offset} ) {
128 seek($fh, 0 + $obj->_root->{file_offset}, SEEK_SET);
130 my $bytes_read = read( $fh, $signature, length(SIG_FILE));
133 # File is empty -- write signature and master index
136 $self->write_file_signature( $obj );
138 $obj->{base_offset} = $self->_request_space(
139 $obj, $self->tag_size( $self->{index_size} ),
143 $obj, $obj->_base_offset, $obj->_type,
144 chr(0)x$self->{index_size},
147 # Flush the filehandle
148 my $old_fh = select $fh;
149 my $old_af = $|; $| = 1; $| = $old_af;
153 $obj->{base_offset} = $bytes_read;
156 # Check signature was valid
158 unless ($signature eq SIG_FILE) {
159 $self->close_fh( $obj );
160 $obj->_throw_error("Signature not found -- file is not a Deep DB");
164 # Get our type from master index signature
166 my $tag = $self->load_tag($obj, $obj->_base_offset)
167 or $obj->_throw_error("Corrupted file, no master index record");
169 unless ($obj->{type} eq $tag->{signature}) {
170 $obj->_throw_error("File type mismatch");
175 #XXX We have to make sure we don't mess up when autoflush isn't turned on
176 unless ( $obj->_root->{inode} ) {
177 my @stats = stat($obj->_fh);
178 $obj->_root->{inode} = $stats[1];
179 $obj->_root->{end} = $stats[7];
189 # Open a fh to the database, create if nonexistent.
190 # Make sure file signature matches DBM::Deep spec.
195 # Theoretically, adding O_BINARY should remove the need for the binmode
196 # Of course, testing it is going to be ... interesting.
197 my $flags = O_RDWR | O_CREAT | O_BINARY;
200 my $filename = $obj->_root->{file};
201 sysopen( $fh, $filename, $flags )
202 or $obj->_throw_error("Cannot sysopen file '$filename': $!");
203 $obj->_root->{fh} = $fh;
205 #XXX Can we remove this by using the right sysopen() flags?
206 # Maybe ... q.v. above
207 binmode $fh; # for win32
209 if ($obj->_root->{autoflush}) {
210 my $old = select $fh;
222 if ( my $fh = $obj->_root->{fh} ) {
225 $obj->_root->{fh} = undef;
233 return SIG_SIZE + $self->{data_size} + $size;
238 # Given offset, signature and content, create tag and write to disk
241 my ($obj, $offset, $sig, $content) = @_;
242 my $size = length( $content );
246 if ( defined $offset ) {
247 seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
250 print( $fh $sig . pack($self->{data_pack}, $size) . $content );
252 return unless defined $offset;
257 offset => $offset + SIG_SIZE + $self->{data_size},
264 # Given offset, load single tag and return signature, size and data
267 my ($obj, $offset) = @_;
269 # print join(':',map{$_||''}caller(1)), $/;
273 seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
275 #XXX I'm not sure this check will work if autoflush isn't enabled ...
279 read( $fh, $b, SIG_SIZE + $self->{data_size} );
280 my ($sig, $size) = unpack( "A $self->{data_pack}", $b );
283 read( $fh, $buffer, $size);
288 offset => $offset + SIG_SIZE + $self->{data_size},
295 my ($obj, $value, $key) = @_;
297 my $is_dbm_deep = eval {
298 local $SIG{'__DIE__'};
299 $value->isa( 'DBM::Deep' );
302 my $len = SIG_SIZE + $self->{data_size}
303 + $self->{data_size} + length( $key );
305 if ( $is_dbm_deep && $value->_root eq $obj->_root ) {
306 return $len + $self->{long_size};
309 my $r = Scalar::Util::reftype( $value ) || '';
310 if ( $obj->_root->{autobless} ) {
311 # This is for the bit saying whether or not this thing is blessed.
315 unless ( $r eq 'HASH' || $r eq 'ARRAY' ) {
316 if ( defined $value ) {
317 $len += length( $value );
322 $len += $self->{index_size};
324 # if autobless is enabled, must also take into consideration
325 # the class name as it is stored after the key.
326 if ( $obj->_root->{autobless} ) {
327 my $value_class = Scalar::Util::blessed($value);
328 if ( defined $value_class && !$is_dbm_deep ) {
329 $len += $self->{data_size} + length($value_class);
338 # Adds one key/value pair to bucket list, given offset, MD5 digest of key,
339 # plain (undigested) key and value.
342 my ($obj, $tag, $md5, $plain_key, $value) = @_;
344 # This verifies that only supported values will be stored.
346 my $r = Scalar::Util::reftype( $value );
349 last if $r eq 'HASH';
350 last if $r eq 'ARRAY';
353 "Storage of variables of type '$r' is not supported."
360 my $root = $obj->_root;
363 my $actual_length = $self->_length_needed( $obj, $value, $plain_key );
365 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
367 # $self->_release_space( $obj, $size, $subloc );
368 # Updating a known md5
369 #XXX This needs updating to use _release_space
373 if ($actual_length <= $size) {
377 $location = $self->_request_space( $obj, $actual_length );
380 $tag->{offset} + $offset
381 + $self->{hash_size} + $root->{file_offset},
384 print( $fh pack($self->{long_pack}, $location ) );
385 print( $fh pack($self->{long_pack}, $actual_length ) );
389 elsif ( defined $offset ) {
390 $location = $self->_request_space( $obj, $actual_length );
392 seek( $fh, $tag->{offset} + $offset + $root->{file_offset}, SEEK_SET );
393 print( $fh $md5 . pack($self->{long_pack}, $location ) );
394 print( $fh pack($self->{long_pack}, $actual_length ) );
396 # If bucket didn't fit into list, split into a new index level
397 # split_index() will do the _request_space() call
399 $location = $self->split_index( $obj, $md5, $tag );
402 $self->write_value( $obj, $location, $plain_key, $value );
409 my $r = Scalar::Util::reftype( $item ) || return;
410 if ( $r eq 'HASH' ) {
413 elsif ( $r eq 'ARRAY' ) {
421 sub _get_dbm_object {
426 if ($item->isa( 'DBM::Deep' )) {
433 my $r = Scalar::Util::reftype( $item ) || '';
434 if ( $r eq 'HASH' ) {
437 my $obj = tied(%$item);
438 if ($obj->isa( 'DBM::Deep' )) {
445 elsif ( $r eq 'ARRAY' ) {
448 my $obj = tied(@$item);
449 if ($obj->isa( 'DBM::Deep' )) {
462 my ($obj, $location, $key, $value) = @_;
465 my $root = $obj->_root;
467 my $dbm_deep_obj = _get_dbm_object( $value );
468 if ( $dbm_deep_obj && $dbm_deep_obj->_root ne $obj->_root ) {
469 $obj->_throw_error( "Cannot cross-reference. Use export() instead" );
472 seek($fh, $location + $root->{file_offset}, SEEK_SET);
475 # Write signature based on content type, set content length and write
478 my $r = Scalar::Util::reftype( $value ) || '';
479 if ( $dbm_deep_obj ) {
480 $self->write_tag( $obj, undef, SIG_INTERNAL,pack($self->{long_pack}, $dbm_deep_obj->_base_offset) );
482 elsif ($r eq 'HASH') {
483 if ( !$dbm_deep_obj && tied %{$value} ) {
484 $obj->_throw_error( "Cannot store something that is tied" );
486 $self->write_tag( $obj, undef, SIG_HASH, chr(0)x$self->{index_size} );
488 elsif ($r eq 'ARRAY') {
489 if ( !$dbm_deep_obj && tied @{$value} ) {
490 $obj->_throw_error( "Cannot store something that is tied" );
492 $self->write_tag( $obj, undef, SIG_ARRAY, chr(0)x$self->{index_size} );
494 elsif (!defined($value)) {
495 $self->write_tag( $obj, undef, SIG_NULL, '' );
498 $self->write_tag( $obj, undef, SIG_DATA, $value );
502 # Plain key is stored AFTER value, as keys are typically fetched less often.
504 print( $fh pack($self->{data_pack}, length($key)) . $key );
506 # Internal references don't care about autobless
507 return 1 if $dbm_deep_obj;
510 # If value is blessed, preserve class name
512 if ( $root->{autobless} ) {
513 my $value_class = Scalar::Util::blessed($value);
514 if ( defined $value_class && !$dbm_deep_obj ) {
516 print( $fh pack($self->{data_pack}, length($value_class)) . $value_class );
524 # If content is a hash or array, create new child DBM::Deep object and
525 # pass each key or element to it.
529 tie %$value, 'DBM::Deep', {
530 base_offset => $location,
535 elsif ($r eq 'ARRAY') {
537 tie @$value, 'DBM::Deep', {
538 base_offset => $location,
549 my ($obj, $md5, $tag) = @_;
552 my $root = $obj->_root;
554 my $loc = $self->_request_space(
555 $obj, $self->tag_size( $self->{index_size} ),
558 seek($fh, $tag->{ref_loc} + $root->{file_offset}, SEEK_SET);
559 print( $fh pack($self->{long_pack}, $loc) );
561 my $index_tag = $self->write_tag(
562 $obj, $loc, SIG_INDEX,
563 chr(0)x$self->{index_size},
566 my $newtag_loc = $self->_request_space(
567 $obj, $self->tag_size( $self->{bucket_list_size} ),
570 my $keys = $tag->{content}
571 . $md5 . pack($self->{long_pack}, $newtag_loc)
572 . pack($self->{long_pack}, 0);
576 for (my $i = 0; $i <= $self->{max_buckets}; $i++) {
577 my ($key, $old_subloc, $size) = $self->_get_key_subloc( $keys, $i );
579 die "[INTERNAL ERROR]: No key in split_index()\n" unless $key;
580 die "[INTERNAL ERROR]: No subloc in split_index()\n" unless $old_subloc;
582 my $num = ord(substr($key, $tag->{ch} + 1, 1));
585 seek($fh, $newloc[$num] + $root->{file_offset}, SEEK_SET);
587 read( $fh, $subkeys, $self->{bucket_list_size});
589 # This is looking for the first empty spot
590 my ($subloc, $offset, $size) = $self->_find_in_buckets(
591 { content => $subkeys }, '',
594 seek($fh, $newloc[$num] + $offset + $root->{file_offset}, SEEK_SET);
595 print( $fh $key . pack($self->{long_pack}, $old_subloc) );
600 seek($fh, $index_tag->{offset} + ($num * $self->{long_size}) + $root->{file_offset}, SEEK_SET);
602 my $loc = $self->_request_space(
603 $obj, $self->tag_size( $self->{bucket_list_size} ),
606 print( $fh pack($self->{long_pack}, $loc) );
608 my $blist_tag = $self->write_tag(
609 $obj, $loc, SIG_BLIST,
610 chr(0)x$self->{bucket_list_size},
613 seek($fh, $blist_tag->{offset} + $root->{file_offset}, SEEK_SET);
614 print( $fh $key . pack($self->{long_pack}, $old_subloc) );
616 $newloc[$num] = $blist_tag->{offset};
619 $self->_release_space(
620 $obj, $self->tag_size( $self->{bucket_list_size} ),
621 $tag->{offset} - SIG_SIZE - $self->{data_size},
629 my ($obj, $subloc) = @_;
634 # Found match -- seek to offset and read signature
637 seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
638 read( $fh, $signature, SIG_SIZE);
641 # If value is a hash or array, return new DBM::Deep object with correct offset
643 if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) {
644 my $new_obj = DBM::Deep->new({
646 base_offset => $subloc,
650 if ($new_obj->_root->{autobless}) {
652 # Skip over value and plain key to see if object needs
655 seek($fh, $self->{data_size} + $self->{index_size}, SEEK_CUR);
658 read( $fh, $size, $self->{data_size});
659 $size = unpack($self->{data_pack}, $size);
660 if ($size) { seek($fh, $size, SEEK_CUR); }
663 read( $fh, $bless_bit, 1);
664 if (ord($bless_bit)) {
666 # Yes, object needs to be re-blessed
669 read( $fh, $size, $self->{data_size});
670 $size = unpack($self->{data_pack}, $size);
671 if ($size) { read( $fh, $class_name, $size); }
672 if ($class_name) { $new_obj = bless( $new_obj, $class_name ); }
678 elsif ( $signature eq SIG_INTERNAL ) {
680 read( $fh, $size, $self->{data_size});
681 $size = unpack($self->{data_pack}, $size);
685 read( $fh, $new_loc, $size );
686 $new_loc = unpack( $self->{long_pack}, $new_loc );
688 return $self->read_from_loc( $obj, $new_loc );
695 # Otherwise return actual value
697 elsif ($signature eq SIG_DATA) {
699 read( $fh, $size, $self->{data_size});
700 $size = unpack($self->{data_pack}, $size);
703 if ($size) { read( $fh, $value, $size); }
708 # Key exists, but content is null
713 sub get_bucket_value {
715 # Fetch single value given tag and MD5 digested key.
718 my ($obj, $tag, $md5) = @_;
720 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
722 return $self->read_from_loc( $obj, $subloc );
729 # Delete single key/value pair given tag and MD5 digested key.
732 my ($obj, $tag, $md5) = @_;
734 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
735 #XXX This needs _release_space()
738 seek($fh, $tag->{offset} + $offset + $obj->_root->{file_offset}, SEEK_SET);
739 print( $fh substr($tag->{content}, $offset + $self->{bucket_size} ) );
740 print( $fh chr(0) x $self->{bucket_size} );
749 # Check existence of single key given tag and MD5 digested key.
752 my ($obj, $tag, $md5) = @_;
754 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
758 sub find_bucket_list {
760 # Locate offset for bucket list, given digested key
763 my ($obj, $md5, $args) = @_;
764 $args = {} unless $args;
767 # Locate offset for bucket list using digest index system
769 my $tag = $self->load_tag($obj, $obj->_base_offset)
770 or $obj->_throw_error( "INTERNAL ERROR - Cannot find tag" );
773 while ($tag->{signature} ne SIG_BLIST) {
774 my $num = ord substr($md5, $ch, 1);
776 my $ref_loc = $tag->{offset} + ($num * $self->{long_size});
777 $tag = $self->index_lookup( $obj, $tag, $num );
780 return if !$args->{create};
782 my $loc = $self->_request_space(
783 $obj, $self->tag_size( $self->{bucket_list_size} ),
787 seek($fh, $ref_loc + $obj->_root->{file_offset}, SEEK_SET);
788 print( $fh pack($self->{long_pack}, $loc) );
790 $tag = $self->write_tag(
791 $obj, $loc, SIG_BLIST,
792 chr(0)x$self->{bucket_list_size},
795 $tag->{ref_loc} = $ref_loc;
802 $tag->{ref_loc} = $ref_loc;
810 # Given index tag, lookup single entry in index and return .
813 my ($obj, $tag, $index) = @_;
815 my $location = unpack(
819 $index * $self->{long_size},
824 if (!$location) { return; }
826 return $self->load_tag( $obj, $location );
831 # Scan index and recursively step into deeper levels, looking for next key.
834 my ($obj, $offset, $ch, $force_return_next) = @_;
836 my $tag = $self->load_tag($obj, $offset );
840 if ($tag->{signature} ne SIG_BLIST) {
841 my $content = $tag->{content};
842 my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1));
844 for (my $idx = $start; $idx < (2**8); $idx++) {
849 $idx * $self->{long_size},
855 my $result = $self->traverse_index(
856 $obj, $subloc, $ch + 1, $force_return_next,
859 if (defined($result)) { return $result; }
863 $obj->{return_next} = 1;
867 my $keys = $tag->{content};
868 if ($force_return_next) { $obj->{return_next} = 1; }
871 # Iterate through buckets, looking for a key match
873 for (my $i = 0; $i < $self->{max_buckets}; $i++) {
874 my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
876 # End of bucket list -- return to outer loop
878 $obj->{return_next} = 1;
881 # Located previous key -- return next one found
882 elsif ($key eq $obj->{prev_md5}) {
883 $obj->{return_next} = 1;
886 # Seek to bucket location and skip over signature
887 elsif ($obj->{return_next}) {
888 seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
890 # Skip over value to get to plain key
892 read( $fh, $sig, SIG_SIZE );
895 read( $fh, $size, $self->{data_size});
896 $size = unpack($self->{data_pack}, $size);
897 if ($size) { seek($fh, $size, SEEK_CUR); }
899 # Read in plain key and return as scalar
901 read( $fh, $size, $self->{data_size});
902 $size = unpack($self->{data_pack}, $size);
903 if ($size) { read( $fh, $plain_key, $size); }
909 $obj->{return_next} = 1;
910 } # tag is a bucket list
917 # Locate next key, given digested previous one
922 $obj->{prev_md5} = $_[1] ? $_[1] : undef;
923 $obj->{return_next} = 0;
926 # If the previous key was not specifed, start at the top and
927 # return the first one found.
929 if (!$obj->{prev_md5}) {
930 $obj->{prev_md5} = chr(0) x $self->{hash_size};
931 $obj->{return_next} = 1;
934 return $self->traverse_index( $obj, $obj->_base_offset, 0 );
939 sub _get_key_subloc {
941 my ($keys, $idx) = @_;
943 my ($key, $subloc, $size) = unpack(
944 "a$self->{hash_size} $self->{long_pack} $self->{long_pack}",
947 ($idx * $self->{bucket_size}),
948 $self->{bucket_size},
952 return ($key, $subloc, $size);
955 sub _find_in_buckets {
957 my ($tag, $md5) = @_;
960 for ( my $i = 0; $i < $self->{max_buckets}; $i++ ) {
961 my ($key, $subloc, $size) = $self->_get_key_subloc(
965 return ($subloc, $i * $self->{bucket_size}, $size) unless $subloc;
967 next BUCKET if $key ne $md5;
969 return ($subloc, $i * $self->{bucket_size}, $size);
977 # my ($obj, $spot, $data) = @_;
979 # my $fh = $obj->_fh;
980 # seek( $fh, $spot, SEEK_SET );
981 # print( $fh $data );
988 my ($obj, $size) = @_;
990 my $loc = $obj->_root->{end};
991 $obj->_root->{end} += $size;
998 my ($obj, $size, $loc) = @_;
1003 seek( $fh, $loc + $obj->_root->{file_offset}, SEEK_SET );
1005 . pack($self->{long_pack}, $size )
1006 . pack($self->{long_pack}, $next_loc )
1015 # This will be added in later, after more refactoring is done. This is an early
1016 # attempt at refactoring on the physical level instead of the virtual level.
1019 my ($obj, $spot, $amount, $unpack) = @_;
1022 seek( $fh, $spot + $obj->_root->{file_offset}, SEEK_SET );
1025 my $bytes_read = read( $fh, $buffer, $amount );
1028 $buffer = unpack( $unpack, $buffer );
1032 return ($buffer, $bytes_read);