1 package DBM::Deep::Engine;
8 use Fcntl qw( :DEFAULT :flock );
12 # * To add to bucket_size, make sure you modify the following:
15 # - add_bucket() - where the buckets are printed
18 # Setup file and tag signatures. These should never change.
20 sub SIG_FILE () { 'DPDB' }
21 sub SIG_HEADER () { 'h' }
22 sub SIG_INTERNAL () { 'i' }
23 sub SIG_HASH () { 'H' }
24 sub SIG_ARRAY () { 'A' }
25 sub SIG_NULL () { 'N' }
26 sub SIG_DATA () { 'D' }
27 sub SIG_INDEX () { 'I' }
28 sub SIG_BLIST () { 'B' }
29 sub SIG_FREE () { 'F' }
42 digest => \&Digest::MD5::md5,
46 # Maximum number of buckets per list before another level of indexing is
47 # done. Increase this value for slightly greater speed, but larger database
48 # files. DO NOT decrease this value below 16, due to risk of recursive
57 if ( defined $args->{pack_size} ) {
58 if ( lc $args->{pack_size} eq 'small' ) {
59 $args->{long_size} = 2;
60 $args->{long_pack} = 'n';
62 elsif ( lc $args->{pack_size} eq 'medium' ) {
63 $args->{long_size} = 4;
64 $args->{long_pack} = 'N';
66 elsif ( lc $args->{pack_size} eq 'large' ) {
67 $args->{long_size} = 8;
68 $args->{long_pack} = 'Q';
71 die "Unknown pack_size value: '$args->{pack_size}'\n";
75 # Grab the parameters we want to use
76 foreach my $param ( keys %$self ) {
77 next unless exists $args->{$param};
78 $self->{$param} = $args->{$param};
80 Scalar::Util::weaken( $self->{obj} ) if $self->{obj};
82 if ( $self->{max_buckets} < 16 ) {
83 warn "Floor of max_buckets is 16. Setting it to 16 from '$self->{max_buckets}'\n";
84 $self->{max_buckets} = 16;
90 sub _fileobj { return $_[0]{fileobj} }
95 # The 2**8 here indicates the number of different characters in the
96 # current hashing algorithm
97 #XXX Does this need to be updated with different hashing algorithms?
98 $self->{index_size} = (2**8) * $self->{long_size};
99 $self->{bucket_size} = $self->{hash_size} + $self->{long_size} * 3;
100 $self->{bucket_list_size} = $self->{max_buckets} * $self->{bucket_size};
105 sub write_file_header {
108 my $loc = $self->_fileobj->request_space( length( SIG_FILE ) + 21 );
110 $self->_fileobj->print_at( $loc,
113 pack('N', 1), # header version
114 pack('N', 12), # header size
115 pack('N', 0), # currently running transaction IDs
116 pack('n', $self->{long_size}),
117 pack('A', $self->{long_pack}),
118 pack('n', $self->{data_size}),
119 pack('A', $self->{data_pack}),
120 pack('n', $self->{max_buckets}),
123 $self->_fileobj->set_transaction_offset( 13 );
128 sub read_file_header {
131 my $buffer = $self->_fileobj->read_at( 0, length(SIG_FILE) + 9 );
132 return unless length($buffer);
134 my ($file_signature, $sig_header, $header_version, $size) = unpack(
138 unless ( $file_signature eq SIG_FILE ) {
139 $self->_fileobj->close;
140 $self->_throw_error( "Signature not found -- file is not a Deep DB" );
143 unless ( $sig_header eq SIG_HEADER ) {
144 $self->_fileobj->close;
145 $self->_throw_error( "Old file version found." );
148 my $buffer2 = $self->_fileobj->read_at( undef, $size );
149 my ($running_transactions, @values) = unpack( 'N n A n A n', $buffer2 );
151 $self->_fileobj->set_transaction_offset( 13 );
153 if ( @values < 5 || grep { !defined } @values ) {
154 $self->_fileobj->close;
155 $self->_throw_error("Corrupted file - bad header");
158 #XXX Add warnings if values weren't set right
159 @{$self}{qw(long_size long_pack data_size data_pack max_buckets)} = @values;
161 return length($buffer) + length($buffer2);
168 # Need to remove use of $fh here
169 my $fh = $self->_fileobj->{fh};
172 #XXX The duplication of calculate_sizes needs to go away
173 unless ( $obj->{base_offset} ) {
174 my $bytes_read = $self->read_file_header;
176 $self->calculate_sizes;
179 # File is empty -- write header and master index
182 $self->_fileobj->audit( "# Database created on" );
184 $self->write_file_header;
186 $obj->{base_offset} = $self->_fileobj->request_space(
187 $self->tag_size( $self->{index_size} ),
191 $obj->_base_offset, $obj->_type,
192 chr(0)x$self->{index_size},
195 # Flush the filehandle
196 my $old_fh = select $fh;
197 my $old_af = $|; $| = 1; $| = $old_af;
201 $obj->{base_offset} = $bytes_read;
204 # Get our type from master index header
206 my $tag = $self->load_tag($obj->_base_offset);
209 $self->_throw_error("Corrupted file, no master index record");
212 unless ($obj->_type eq $tag->{signature}) {
214 $self->_throw_error("File type mismatch");
219 $self->calculate_sizes;
222 #XXX We have to make sure we don't mess up when autoflush isn't turned on
223 $self->_fileobj->set_inode;
233 return SIG_SIZE + $self->{data_size} + $size;
238 # Given offset, signature and content, create tag and write to disk
241 my ($offset, $sig, $content) = @_;
242 my $size = length( $content );
244 $self->_fileobj->print_at(
246 $sig, pack($self->{data_pack}, $size), $content,
249 return unless defined $offset;
254 offset => $offset + SIG_SIZE + $self->{data_size},
261 # Given offset, load single tag and return signature, size and data
266 my $fileobj = $self->_fileobj;
268 my $s = SIG_SIZE + $self->{data_size};
269 my $b = $fileobj->read_at( $offset, $s );
270 my ($sig, $size) = unpack( "A $self->{data_pack}", $b );
272 my $buffer = $fileobj->read_at( undef, $size );
277 offset => $offset + SIG_SIZE + $self->{data_size},
282 sub _get_dbm_object {
287 if ($item->isa( 'DBM::Deep' )) {
294 my $r = Scalar::Util::reftype( $item ) || '';
295 if ( $r eq 'HASH' ) {
298 my $obj = tied(%$item);
299 if ($obj->isa( 'DBM::Deep' )) {
306 elsif ( $r eq 'ARRAY' ) {
309 my $obj = tied(@$item);
310 if ($obj->isa( 'DBM::Deep' )) {
323 my ($value, $key) = @_;
325 my $is_dbm_deep = eval {
326 local $SIG{'__DIE__'};
327 $value->isa( 'DBM::Deep' );
331 + $self->{data_size} # size for value
332 + $self->{data_size} # size for key
333 + length( $key ); # length of key
335 if ( $is_dbm_deep && $value->_fileobj eq $self->_fileobj ) {
336 # long_size is for the internal reference
337 return $len + $self->{long_size};
340 if ( $self->_fileobj->{autobless} ) {
341 # This is for the bit saying whether or not this thing is blessed.
345 my $r = Scalar::Util::reftype( $value ) || '';
346 unless ( $r eq 'HASH' || $r eq 'ARRAY' ) {
347 if ( defined $value ) {
348 $len += length( $value );
353 $len += $self->{index_size};
355 # if autobless is enabled, must also take into consideration
356 # the class name as it is stored after the key.
357 if ( $self->_fileobj->{autobless} ) {
358 my $c = Scalar::Util::blessed($value);
359 if ( defined $c && !$is_dbm_deep ) {
360 $len += $self->{data_size} + length($c);
369 # Adds one key/value pair to bucket list, given offset, MD5 digest of key,
370 # plain (undigested) key and value.
373 my ($tag, $md5, $plain_key, $value, $deleted, $orig_key) = @_;
378 # This verifies that only supported values will be stored.
380 my $r = Scalar::Util::reftype( $value );
383 last if $r eq 'HASH';
384 last if $r eq 'ARRAY';
387 "Storage of variables of type '$r' is not supported."
394 my $fileobj = $self->_fileobj;
396 my $actual_length = $self->_length_needed( $value, $plain_key );
398 #ACID - This is a mutation. Must only find the exact transaction
399 my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5, 1 );
402 if ( $fileobj->transaction_id == 0 ) {
403 @transactions = $fileobj->current_transactions;
406 # $self->_release_space( $size, $subloc );
407 # Updating a known md5
408 #XXX This needs updating to use _release_space
412 if ($actual_length <= $size) {
416 $location = $fileobj->request_space( $actual_length );
418 $fileobj->print_at( $tag->{offset} + $offset + $self->{hash_size},
419 pack($self->{long_pack}, $location ),
420 pack($self->{long_pack}, $actual_length ),
421 pack('n n', $fileobj->transaction_id, $deleted ),
426 elsif ( defined $offset ) {
427 $location = $fileobj->request_space( $actual_length );
429 $fileobj->print_at( $tag->{offset} + $offset,
431 pack($self->{long_pack}, $location ),
432 pack($self->{long_pack}, $actual_length ),
433 pack('n n', $fileobj->transaction_id, $deleted ),
436 for ( @transactions ) {
437 my $tag2 = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
438 $fileobj->{transaction_id} = $_;
439 $self->add_bucket( $tag2, $md5, '', '', 1, $orig_key );
440 $fileobj->{transaction_id} = 0;
442 $tag = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
444 # If bucket didn't fit into list, split into a new index level
445 # split_index() will do the _fileobj->request_space() call
447 $location = $self->split_index( $md5, $tag );
450 $self->write_value( $location, $plain_key, $value, $orig_key );
457 my ($location, $key, $value, $orig_key) = @_;
459 my $fileobj = $self->_fileobj;
461 my $dbm_deep_obj = _get_dbm_object( $value );
462 if ( $dbm_deep_obj && $dbm_deep_obj->_fileobj ne $fileobj ) {
463 $self->_throw_error( "Cannot cross-reference. Use export() instead" );
467 # Write signature based on content type, set content length and write
470 my $r = Scalar::Util::reftype( $value ) || '';
471 if ( $dbm_deep_obj ) {
472 $self->write_tag( $location, SIG_INTERNAL,pack($self->{long_pack}, $dbm_deep_obj->_base_offset) );
474 elsif ($r eq 'HASH') {
475 if ( !$dbm_deep_obj && tied %{$value} ) {
476 $self->_throw_error( "Cannot store something that is tied" );
478 $self->write_tag( $location, SIG_HASH, chr(0)x$self->{index_size} );
480 elsif ($r eq 'ARRAY') {
481 if ( !$dbm_deep_obj && tied @{$value} ) {
482 $self->_throw_error( "Cannot store something that is tied" );
484 $self->write_tag( $location, SIG_ARRAY, chr(0)x$self->{index_size} );
486 elsif (!defined($value)) {
487 $self->write_tag( $location, SIG_NULL, '' );
490 $self->write_tag( $location, SIG_DATA, $value );
494 # Plain key is stored AFTER value, as keys are typically fetched less often.
496 $fileobj->print_at( undef, pack($self->{data_pack}, length($key)) . $key );
498 # Internal references don't care about autobless
499 return 1 if $dbm_deep_obj;
502 # If value is blessed, preserve class name
504 if ( $fileobj->{autobless} ) {
505 if ( defined( my $c = Scalar::Util::blessed($value) ) ) {
506 $fileobj->print_at( undef, chr(1), pack($self->{data_pack}, length($c)) . $c );
509 $fileobj->print_at( undef, chr(0) );
514 # Tie the passed in reference so that changes to it are reflected in the
515 # datafile. The use of $location as the base_offset will act as the
516 # the linkage between parent and child.
518 # The overall assignment is a hack around the fact that just tying doesn't
519 # store the values. This may not be the wrong thing to do.
523 tie %$value, 'DBM::Deep', {
524 base_offset => $location,
526 parent => $self->{obj},
527 parent_key => $orig_key,
531 elsif ($r eq 'ARRAY') {
533 tie @$value, 'DBM::Deep', {
534 base_offset => $location,
536 parent => $self->{obj},
537 parent_key => $orig_key,
547 my ($md5, $tag) = @_;
549 my $fileobj = $self->_fileobj;
551 my $loc = $fileobj->request_space(
552 $self->tag_size( $self->{index_size} ),
555 $fileobj->print_at( $tag->{ref_loc}, pack($self->{long_pack}, $loc) );
557 my $index_tag = $self->write_tag(
559 chr(0)x$self->{index_size},
562 my $newtag_loc = $fileobj->request_space(
563 $self->tag_size( $self->{bucket_list_size} ),
566 my $keys = $tag->{content}
567 . $md5 . pack($self->{long_pack}, $newtag_loc)
568 . pack($self->{long_pack}, 0) # size
569 . pack($self->{long_pack}, 0); # transaction ID
573 # The <= here is deliberate - we have max_buckets+1 keys to iterate
574 # through, unlike every other loop that uses max_buckets as a stop.
575 for (my $i = 0; $i <= $self->{max_buckets}; $i++) {
576 my ($key, $old_subloc, $size) = $self->_get_key_subloc( $keys, $i );
578 die "[INTERNAL ERROR]: No key in split_index()\n" unless $key;
579 die "[INTERNAL ERROR]: No subloc in split_index()\n" unless $old_subloc;
581 my $num = ord(substr($key, $tag->{ch} + 1, 1));
584 my $subkeys = $fileobj->read_at( $newloc[$num], $self->{bucket_list_size} );
586 # This is looking for the first empty spot
587 my ($subloc, $offset, $size) = $self->_find_in_buckets(
588 { content => $subkeys }, '',
592 $newloc[$num] + $offset,
593 $key, pack($self->{long_pack}, $old_subloc),
599 my $loc = $fileobj->request_space(
600 $self->tag_size( $self->{bucket_list_size} ),
604 $index_tag->{offset} + ($num * $self->{long_size}),
605 pack($self->{long_pack}, $loc),
608 my $blist_tag = $self->write_tag(
610 chr(0)x$self->{bucket_list_size},
613 $fileobj->print_at( $blist_tag->{offset}, $key . pack($self->{long_pack}, $old_subloc) );
615 $newloc[$num] = $blist_tag->{offset};
618 $self->_release_space(
619 $self->tag_size( $self->{bucket_list_size} ),
620 $tag->{offset} - SIG_SIZE - $self->{data_size},
628 my ($subloc, $orig_key) = @_;
630 my $fileobj = $self->_fileobj;
632 my $signature = $fileobj->read_at( $subloc, SIG_SIZE );
635 # If value is a hash or array, return new DBM::Deep object with correct offset
637 if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) {
638 my $new_obj = DBM::Deep->new({
640 base_offset => $subloc,
641 fileobj => $self->_fileobj,
642 parent => $self->{obj},
643 parent_key => $orig_key,
646 if ($new_obj->_fileobj->{autobless}) {
648 # Skip over value and plain key to see if object needs
651 $fileobj->increment_pointer( $self->{data_size} + $self->{index_size} );
653 my $size = $fileobj->read_at( undef, $self->{data_size} );
654 $size = unpack($self->{data_pack}, $size);
655 if ($size) { $fileobj->increment_pointer( $size ); }
657 my $bless_bit = $fileobj->read_at( undef, 1 );
658 if (ord($bless_bit)) {
660 # Yes, object needs to be re-blessed
662 my $size = $fileobj->read_at( undef, $self->{data_size} );
663 $size = unpack($self->{data_pack}, $size);
666 if ($size) { $class_name = $fileobj->read_at( undef, $size ); }
667 if (defined $class_name) { $new_obj = bless( $new_obj, $class_name ); }
673 elsif ( $signature eq SIG_INTERNAL ) {
674 my $size = $fileobj->read_at( undef, $self->{data_size} );
675 $size = unpack($self->{data_pack}, $size);
678 my $new_loc = $fileobj->read_at( undef, $size );
679 $new_loc = unpack( $self->{long_pack}, $new_loc );
680 return $self->read_from_loc( $new_loc, $orig_key );
687 # Otherwise return actual value
689 elsif ( $signature eq SIG_DATA ) {
690 my $size = $fileobj->read_at( undef, $self->{data_size} );
691 $size = unpack($self->{data_pack}, $size);
694 if ($size) { $value = $fileobj->read_at( undef, $size ); }
699 # Key exists, but content is null
704 sub get_bucket_value {
706 # Fetch single value given tag and MD5 digested key.
709 my ($tag, $md5, $orig_key) = @_;
711 #ACID - This is a read. Can find exact or HEAD
712 my ($subloc, $offset, $size, $is_deleted) = $self->_find_in_buckets( $tag, $md5 );
715 #XXX Need to use real key
716 # $self->add_bucket( $tag, $md5, $orig_key, undef, undef, $orig_key );
719 elsif ( !$is_deleted ) {
720 return $self->read_from_loc( $subloc, $orig_key );
728 # Delete single key/value pair given tag and MD5 digested key.
731 my ($tag, $md5, $orig_key) = @_;
733 #ACID - Although this is a mutation, we must find any transaction.
734 # This is because we need to mark something as deleted that is in the HEAD.
735 my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5 );
739 my $fileobj = $self->_fileobj;
742 if ( $fileobj->transaction_id == 0 ) {
743 @transactions = $fileobj->current_transactions;
746 if ( $fileobj->transaction_id == 0 ) {
747 my $value = $self->read_from_loc( $subloc, $orig_key );
749 for (@transactions) {
750 $fileobj->{transaction_id} = $_;
751 #XXX Need to use real key
752 $self->add_bucket( $tag, $md5, $orig_key, $value, undef, $orig_key );
753 $fileobj->{transaction_id} = 0;
755 $tag = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
757 #XXX This needs _release_space() for the value and anything below
759 $tag->{offset} + $offset,
760 substr( $tag->{content}, $offset + $self->{bucket_size} ),
761 chr(0) x $self->{bucket_size},
765 $self->add_bucket( $tag, $md5, '', '', 1, $orig_key );
773 # Check existence of single key given tag and MD5 digested key.
776 my ($tag, $md5) = @_;
778 #ACID - This is a read. Can find exact or HEAD
779 my ($subloc, $offset, $size, $is_deleted) = $self->_find_in_buckets( $tag, $md5 );
780 return ($subloc && !$is_deleted) && 1;
783 sub find_bucket_list {
785 # Locate offset for bucket list, given digested key
788 my ($offset, $md5, $args) = @_;
789 $args = {} unless $args;
794 # Locate offset for bucket list using digest index system
796 my $tag = $self->load_tag( $offset )
797 or $self->_throw_error( "INTERNAL ERROR - Cannot find tag" );
800 while ($tag->{signature} ne SIG_BLIST) {
801 my $num = ord substr($md5, $ch, 1);
803 my $ref_loc = $tag->{offset} + ($num * $self->{long_size});
804 $tag = $self->index_lookup( $tag, $num );
807 return if !$args->{create};
809 my $loc = $self->_fileobj->request_space(
810 $self->tag_size( $self->{bucket_list_size} ),
813 $self->_fileobj->print_at( $ref_loc, pack($self->{long_pack}, $loc) );
815 $tag = $self->write_tag(
817 chr(0)x$self->{bucket_list_size},
820 $tag->{ref_loc} = $ref_loc;
827 $tag->{ref_loc} = $ref_loc;
835 # Given index tag, lookup single entry in index and return .
838 my ($tag, $index) = @_;
840 my $location = unpack(
844 $index * $self->{long_size},
849 if (!$location) { return; }
851 return $self->load_tag( $location );
856 # Scan index and recursively step into deeper levels, looking for next key.
859 my ($obj, $offset, $ch, $force_return_next) = @_;
861 my $tag = $self->load_tag( $offset );
863 if ($tag->{signature} ne SIG_BLIST) {
864 my $content = $tag->{content};
865 my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1));
867 for (my $idx = $start; $idx < (2**8); $idx++) {
872 $idx * $self->{long_size},
878 my $result = $self->traverse_index(
879 $obj, $subloc, $ch + 1, $force_return_next,
882 if (defined($result)) { return $result; }
886 $obj->{return_next} = 1;
888 # This is the bucket list
890 my $keys = $tag->{content};
891 if ($force_return_next) { $obj->{return_next} = 1; }
894 # Iterate through buckets, looking for a key match
896 my $transaction_id = $self->_fileobj->transaction_id;
897 for (my $i = 0; $i < $self->{max_buckets}; $i++) {
898 my ($key, $subloc, $size, $trans_id, $is_deleted) = $self->_get_key_subloc( $keys, $i );
901 #XXX Need to find all the copies of this key to find out if $transaction_id has it
902 #XXX marked as deleted, in use, or what.
903 next if $trans_id && $trans_id != $transaction_id;
905 # End of bucket list -- return to outer loop
907 $obj->{return_next} = 1;
910 # Located previous key -- return next one found
911 elsif ($key eq $obj->{prev_md5}) {
912 $obj->{return_next} = 1;
915 # Seek to bucket location and skip over signature
916 elsif ($obj->{return_next}) {
917 my $fileobj = $self->_fileobj;
919 # Skip over value to get to plain key
920 my $sig = $fileobj->read_at( $subloc, SIG_SIZE );
922 my $size = $fileobj->read_at( undef, $self->{data_size} );
923 $size = unpack($self->{data_pack}, $size);
924 if ($size) { $fileobj->increment_pointer( $size ); }
926 # Read in plain key and return as scalar
927 $size = $fileobj->read_at( undef, $self->{data_size} );
928 $size = unpack($self->{data_pack}, $size);
930 if ($size) { $plain_key = $fileobj->read_at( undef, $size); }
936 $obj->{return_next} = 1;
944 # Locate next key, given digested previous one
949 $obj->{prev_md5} = $_[1] ? $_[1] : undef;
950 $obj->{return_next} = 0;
953 # If the previous key was not specifed, start at the top and
954 # return the first one found.
956 if (!$obj->{prev_md5}) {
957 $obj->{prev_md5} = chr(0) x $self->{hash_size};
958 $obj->{return_next} = 1;
961 return $self->traverse_index( $obj, $obj->_base_offset, 0 );
966 sub _get_key_subloc {
968 my ($keys, $idx) = @_;
970 my ($key, $subloc, $size, $transaction_id, $is_deleted) = unpack(
971 # This is 'a', not 'A'. Please read the pack() documentation for the
972 # difference between the two and why it's important.
973 "a$self->{hash_size} $self->{long_pack}2 n2",
976 ($idx * $self->{bucket_size}),
977 $self->{bucket_size},
981 return ($key, $subloc, $size, $transaction_id, $is_deleted);
984 sub _find_in_buckets {
986 my ($tag, $md5, $exact) = @_;
989 my $trans_id = $self->_fileobj->transaction_id;
994 for ( my $i = 0; $i < $self->{max_buckets}; $i++ ) {
995 my ($key, $subloc, $size, $transaction_id, $is_deleted) = $self->_get_key_subloc(
999 my @rv = ($subloc, $i * $self->{bucket_size}, $size, $is_deleted);
1001 unless ( $subloc ) {
1002 if ( !$exact && @zero && $trans_id ) {
1003 @rv = ($zero[2], $zero[0] * $self->{bucket_size},$zero[3],$is_deleted);
1008 next BUCKET if $key ne $md5;
1010 # Save off the HEAD in case we need it.
1011 @zero = ($i,$key,$subloc,$size,$transaction_id,$is_deleted) if $transaction_id == 0;
1013 next BUCKET if $transaction_id != $trans_id;
1021 sub _release_space {
1023 my ($size, $loc) = @_;
1027 $self->_fileobj->print_at( $loc,
1029 pack($self->{long_pack}, $size ),
1030 pack($self->{long_pack}, $next_loc ),
1037 die "DBM::Deep: $_[1]\n";