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' }
22 # Precalculate index, bucket and bucket list sizes
26 $self->{index_size} = (2**8) * $self->{long_size};
27 $self->{bucket_size} = $self->{hash_size} + $self->{long_size} * 2;
28 $self->{bucket_list_size} = $self->{max_buckets} * $self->{bucket_size};
35 # Set pack/unpack modes (see file header for more)
38 my ($long_s, $long_p, $data_s, $data_p) = @_;
41 # Set to 4 and 'N' for 32-bit offset tags (default). Theoretical limit of 4
43 # (Perl must be compiled with largefile support for files > 2 GB)
45 # Set to 8 and 'Q' for 64-bit offsets. Theoretical limit of 16 XB per file.
46 # (Perl must be compiled with largefile and 64-bit long support)
48 $self->{long_size} = $long_s ? $long_s : 4;
49 $self->{long_pack} = $long_p ? $long_p : 'N';
52 # Set to 4 and 'N' for 32-bit data length prefixes. Limit of 4 GB for each
53 # key/value. Upgrading this is possible (see above) but probably not
54 # necessary. If you need more than 4 GB for a single key or value, this
55 # module is really not for you :-)
57 $self->{data_size} = $data_s ? $data_s : 4;
58 $self->{data_pack} = $data_p ? $data_p : 'N';
60 return $self->precalc_sizes();
65 # Set key digest function (default is MD5)
68 my ($digest_func, $hash_size) = @_;
70 $self->{digest} = $digest_func ? $digest_func : \&Digest::MD5::md5;
71 $self->{hash_size} = $hash_size ? $hash_size : 16;
73 return $self->precalc_sizes();
86 digest => \&Digest::MD5::md5,
90 # Maximum number of buckets per list before another level of indexing is
92 # Increase this value for slightly greater speed, but larger database
93 # files. DO NOT decrease this value below 16, due to risk of recursive
108 $self->open( $obj ) if !defined $obj->_fh;
113 unless ( $obj->{base_offset} ) {
114 seek($fh, 0 + $obj->_root->{file_offset}, SEEK_SET);
116 my $bytes_read = read( $fh, $signature, length(SIG_FILE));
119 # File is empty -- write signature and master index
122 my $loc = $self->_request_space( $obj, length( SIG_FILE ) );
123 seek($fh, $loc + $obj->_root->{file_offset}, SEEK_SET);
124 print( $fh SIG_FILE);
126 $obj->{base_offset} = $self->_request_space(
127 $obj, $self->tag_size( $self->{index_size} ),
131 $obj, $obj->_base_offset, $obj->_type,
132 chr(0)x$self->{index_size},
135 # Flush the filehandle
136 my $old_fh = select $fh;
137 my $old_af = $|; $| = 1; $| = $old_af;
141 $obj->{base_offset} = $bytes_read;
144 # Check signature was valid
146 unless ($signature eq SIG_FILE) {
147 $self->close_fh( $obj );
148 $obj->_throw_error("Signature not found -- file is not a Deep DB");
152 # Get our type from master index signature
154 my $tag = $self->load_tag($obj, $obj->_base_offset)
155 or $obj->_throw_error("Corrupted file, no master index record");
157 unless ($obj->{type} eq $tag->{signature}) {
158 $obj->_throw_error("File type mismatch");
163 #XXX We have to make sure we don't mess up when autoflush isn't turned on
164 unless ( $obj->_root->{inode} ) {
165 my @stats = stat($obj->_fh);
166 $obj->_root->{inode} = $stats[1];
167 $obj->_root->{end} = $stats[7];
177 # Open a fh to the database, create if nonexistent.
178 # Make sure file signature matches DBM::Deep spec.
183 # Theoretically, adding O_BINARY should remove the need for the binmode
184 # Of course, testing it is going to be ... interesting.
185 my $flags = O_RDWR | O_CREAT | O_BINARY;
188 my $filename = $obj->_root->{file};
189 sysopen( $fh, $filename, $flags )
190 or $obj->_throw_error("Cannot sysopen file '$filename': $!");
191 $obj->_root->{fh} = $fh;
193 #XXX Can we remove this by using the right sysopen() flags?
194 # Maybe ... q.v. above
195 binmode $fh; # for win32
197 if ($obj->_root->{autoflush}) {
198 my $old = select $fh;
210 if ( my $fh = $obj->_root->{fh} ) {
213 $obj->_root->{fh} = undef;
221 return SIG_SIZE + $self->{data_size} + $size;
226 # Given offset, signature and content, create tag and write to disk
229 my ($obj, $offset, $sig, $content) = @_;
230 my $size = length( $content );
234 if ( defined $offset ) {
235 seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
238 print( $fh $sig . pack($self->{data_pack}, $size) . $content );
240 return unless defined $offset;
245 offset => $offset + SIG_SIZE + $self->{data_size},
252 # Given offset, load single tag and return signature, size and data
255 my ($obj, $offset) = @_;
257 # print join(':',map{$_||''}caller(1)), $/;
261 seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
263 #XXX I'm not sure this check will work if autoflush isn't enabled ...
267 read( $fh, $b, SIG_SIZE + $self->{data_size} );
268 my ($sig, $size) = unpack( "A $self->{data_pack}", $b );
271 read( $fh, $buffer, $size);
276 offset => $offset + SIG_SIZE + $self->{data_size},
283 my ($obj, $value, $key) = @_;
285 my $is_dbm_deep = eval {
286 local $SIG{'__DIE__'};
287 $value->isa( 'DBM::Deep' );
290 my $len = SIG_SIZE + $self->{data_size}
291 + $self->{data_size} + length( $key );
293 if ( $is_dbm_deep && $value->_root eq $obj->_root ) {
294 return $len + $self->{long_size};
297 my $r = Scalar::Util::reftype( $value ) || '';
298 if ( $obj->_root->{autobless} ) {
299 # This is for the bit saying whether or not this thing is blessed.
303 unless ( $r eq 'HASH' || $r eq 'ARRAY' ) {
304 if ( defined $value ) {
305 $len += length( $value );
310 $len += $self->{index_size};
312 # if autobless is enabled, must also take into consideration
313 # the class name as it is stored after the key.
314 if ( $obj->_root->{autobless} ) {
315 my $value_class = Scalar::Util::blessed($value);
316 if ( defined $value_class && !$is_dbm_deep ) {
317 $len += $self->{data_size} + length($value_class);
326 # Adds one key/value pair to bucket list, given offset, MD5 digest of key,
327 # plain (undigested) key and value.
330 my ($obj, $tag, $md5, $plain_key, $value) = @_;
332 # This verifies that only supported values will be stored.
334 my $r = Scalar::Util::reftype( $value );
337 last if $r eq 'HASH';
338 last if $r eq 'ARRAY';
341 "Storage of variables of type '$r' is not supported."
348 my $root = $obj->_root;
351 my $actual_length = $self->_length_needed( $obj, $value, $plain_key );
353 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
355 # Updating a known md5
359 if ($actual_length <= $size) {
363 $location = $self->_request_space( $obj, $actual_length );
366 $tag->{offset} + $offset
367 + $self->{hash_size} + $root->{file_offset},
370 print( $fh pack($self->{long_pack}, $location ) );
371 print( $fh pack($self->{long_pack}, $actual_length ) );
375 elsif ( defined $offset ) {
376 $location = $self->_request_space( $obj, $actual_length );
378 seek( $fh, $tag->{offset} + $offset + $root->{file_offset}, SEEK_SET );
379 print( $fh $md5 . pack($self->{long_pack}, $location ) );
380 print( $fh pack($self->{long_pack}, $actual_length ) );
382 # If bucket didn't fit into list, split into a new index level
384 #XXX This is going to be a problem.
385 $self->split_index( $obj, $md5, $tag );
387 $location = $self->_request_space( $obj, $actual_length );
390 $self->write_value( $obj, $location, $plain_key, $value );
397 my ($obj, $location, $key, $value) = @_;
400 my $root = $obj->_root;
402 my $is_dbm_deep = eval {
403 local $SIG{'__DIE__'};
404 $value->isa( 'DBM::Deep' );
407 my $internal_ref = $is_dbm_deep && ($value->_root eq $root);
409 seek($fh, $location + $root->{file_offset}, SEEK_SET);
412 # Write signature based on content type, set content length and write
415 my $r = Scalar::Util::reftype($value) || '';
416 if ( $internal_ref ) {
417 $self->write_tag( $obj, undef, SIG_INTERNAL,pack($self->{long_pack}, $value->_base_offset) );
419 elsif ($r eq 'HASH') {
420 $self->write_tag( $obj, undef, SIG_HASH, chr(0)x$self->{index_size} );
422 elsif ($r eq 'ARRAY') {
423 $self->write_tag( $obj, undef, SIG_ARRAY, chr(0)x$self->{index_size} );
425 elsif (!defined($value)) {
426 $self->write_tag( $obj, undef, SIG_NULL, '' );
429 $self->write_tag( $obj, undef, SIG_DATA, $value );
433 # Plain key is stored AFTER value, as keys are typically fetched less often.
435 print( $fh pack($self->{data_pack}, length($key)) . $key );
437 # Internal references don't care about autobless
438 return 1 if $internal_ref;
441 # If value is blessed, preserve class name
443 if ( $root->{autobless} ) {
444 my $value_class = Scalar::Util::blessed($value);
445 if ( defined $value_class && !$is_dbm_deep ) {
447 print( $fh pack($self->{data_pack}, length($value_class)) . $value_class );
455 # If content is a hash or array, create new child DBM::Deep object and
456 # pass each key or element to it.
458 if ( !$internal_ref ) {
460 my $branch = DBM::Deep->new(
461 type => DBM::Deep->TYPE_HASH,
462 base_offset => $location,
465 foreach my $key (keys %{$value}) {
466 $branch->STORE( $key, $value->{$key} );
469 elsif ($r eq 'ARRAY') {
470 my $branch = DBM::Deep->new(
471 type => DBM::Deep->TYPE_ARRAY,
472 base_offset => $location,
476 foreach my $element (@{$value}) {
477 $branch->STORE( $index, $element );
488 my ($obj, $md5, $tag) = @_;
491 my $root = $obj->_root;
492 my $keys = $tag->{content};
494 seek($fh, $tag->{ref_loc} + $root->{file_offset}, SEEK_SET);
496 my $loc = $self->_request_space(
497 $obj, $self->tag_size( $self->{index_size} ),
500 print( $fh pack($self->{long_pack}, $loc) );
502 my $index_tag = $self->write_tag(
503 $obj, $loc, SIG_INDEX,
504 chr(0)x$self->{index_size},
509 $keys .= $md5 . (pack($self->{long_pack}, 0) x 2);
512 for (my $i = 0; $i <= $self->{max_buckets}; $i++) {
513 my ($key, $old_subloc, $size) = $self->_get_key_subloc( $keys, $i );
515 next BUCKET unless $key;
517 my $num = ord(substr($key, $tag->{ch} + 1, 1));
519 if ($offsets[$num]) {
520 my $offset = $offsets[$num] + SIG_SIZE + $self->{data_size};
521 seek($fh, $offset + $root->{file_offset}, SEEK_SET);
523 read( $fh, $subkeys, $self->{bucket_list_size});
525 for (my $k=0; $k<$self->{max_buckets}; $k++) {
526 my ($temp, $subloc) = $self->_get_key_subloc( $subkeys, $k );
529 seek($fh, $offset + ($k * $self->{bucket_size}) + $root->{file_offset}, SEEK_SET);
530 print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) );
536 $offsets[$num] = $root->{end};
537 seek($fh, $index_tag->{offset} + ($num * $self->{long_size}) + $root->{file_offset}, SEEK_SET);
539 my $loc = $self->_request_space(
540 $obj, $self->tag_size( $self->{bucket_list_size} ),
543 print( $fh pack($self->{long_pack}, $loc) );
545 my $blist_tag = $self->write_tag(
546 $obj, $loc, SIG_BLIST,
547 chr(0)x$self->{bucket_list_size},
550 seek($fh, $blist_tag->{offset} + $root->{file_offset}, SEEK_SET);
551 print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) );
560 my ($obj, $subloc) = @_;
565 # Found match -- seek to offset and read signature
568 seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
569 read( $fh, $signature, SIG_SIZE);
572 # If value is a hash or array, return new DBM::Deep object with correct offset
574 if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) {
575 my $obj = DBM::Deep->new(
577 base_offset => $subloc,
581 if ($obj->_root->{autobless}) {
583 # Skip over value and plain key to see if object needs
586 seek($fh, $self->{data_size} + $self->{index_size}, SEEK_CUR);
589 read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size);
590 if ($size) { seek($fh, $size, SEEK_CUR); }
593 read( $fh, $bless_bit, 1);
594 if (ord($bless_bit)) {
596 # Yes, object needs to be re-blessed
599 read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size);
600 if ($size) { read( $fh, $class_name, $size); }
601 if ($class_name) { $obj = bless( $obj, $class_name ); }
607 elsif ( $signature eq SIG_INTERNAL ) {
609 read( $fh, $size, $self->{data_size});
610 $size = unpack($self->{data_pack}, $size);
614 read( $fh, $new_loc, $size );
615 $new_loc = unpack( $self->{long_pack}, $new_loc );
617 return $self->read_from_loc( $obj, $new_loc );
624 # Otherwise return actual value
626 elsif ($signature eq SIG_DATA) {
628 read( $fh, $size, $self->{data_size});
629 $size = unpack($self->{data_pack}, $size);
632 if ($size) { read( $fh, $value, $size); }
637 # Key exists, but content is null
642 sub get_bucket_value {
644 # Fetch single value given tag and MD5 digested key.
647 my ($obj, $tag, $md5) = @_;
649 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
651 return $self->read_from_loc( $obj, $subloc );
658 # Delete single key/value pair given tag and MD5 digested key.
661 my ($obj, $tag, $md5) = @_;
663 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
666 seek($fh, $tag->{offset} + $offset + $obj->_root->{file_offset}, SEEK_SET);
667 print( $fh substr($tag->{content}, $offset + $self->{bucket_size} ) );
668 print( $fh chr(0) x $self->{bucket_size} );
677 # Check existence of single key given tag and MD5 digested key.
680 my ($obj, $tag, $md5) = @_;
682 my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5 );
686 sub find_bucket_list {
688 # Locate offset for bucket list, given digested key
691 my ($obj, $md5, $args) = @_;
692 $args = {} unless $args;
695 # Locate offset for bucket list using digest index system
697 my $tag = $self->load_tag($obj, $obj->_base_offset)
698 or $obj->_throw_error( "INTERNAL ERROR - Cannot find tag" );
701 while ($tag->{signature} ne SIG_BLIST) {
702 my $num = ord substr($md5, $ch, 1);
704 my $ref_loc = $tag->{offset} + ($num * $self->{long_size});
705 $tag = $self->index_lookup( $obj, $tag, $num );
708 return if !$args->{create};
711 seek($fh, $ref_loc + $obj->_root->{file_offset}, SEEK_SET);
713 my $loc = $self->_request_space(
714 $obj, $self->tag_size( $self->{bucket_list_size} ),
717 print( $fh pack($self->{long_pack}, $loc) );
719 $tag = $self->write_tag(
720 $obj, $loc, SIG_BLIST,
721 chr(0)x$self->{bucket_list_size},
724 $tag->{ref_loc} = $ref_loc;
731 $tag->{ref_loc} = $ref_loc;
739 # Given index tag, lookup single entry in index and return .
742 my ($obj, $tag, $index) = @_;
744 my $location = unpack(
748 $index * $self->{long_size},
753 if (!$location) { return; }
755 return $self->load_tag( $obj, $location );
760 # Scan index and recursively step into deeper levels, looking for next key.
763 my ($obj, $offset, $ch, $force_return_next) = @_;
765 my $tag = $self->load_tag($obj, $offset );
769 if ($tag->{signature} ne SIG_BLIST) {
770 my $content = $tag->{content};
771 my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1));
773 for (my $idx = $start; $idx < (2**8); $idx++) {
778 $idx * $self->{long_size},
784 my $result = $self->traverse_index(
785 $obj, $subloc, $ch + 1, $force_return_next,
788 if (defined($result)) { return $result; }
792 $obj->{return_next} = 1;
796 my $keys = $tag->{content};
797 if ($force_return_next) { $obj->{return_next} = 1; }
800 # Iterate through buckets, looking for a key match
802 for (my $i = 0; $i < $self->{max_buckets}; $i++) {
803 my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
805 # End of bucket list -- return to outer loop
807 $obj->{return_next} = 1;
810 # Located previous key -- return next one found
811 elsif ($key eq $obj->{prev_md5}) {
812 $obj->{return_next} = 1;
815 # Seek to bucket location and skip over signature
816 elsif ($obj->{return_next}) {
817 seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
819 # Skip over value to get to plain key
821 read( $fh, $sig, SIG_SIZE );
824 read( $fh, $size, $self->{data_size});
825 $size = unpack($self->{data_pack}, $size);
826 if ($size) { seek($fh, $size, SEEK_CUR); }
828 # Read in plain key and return as scalar
830 read( $fh, $size, $self->{data_size});
831 $size = unpack($self->{data_pack}, $size);
832 if ($size) { read( $fh, $plain_key, $size); }
838 $obj->{return_next} = 1;
839 } # tag is a bucket list
846 # Locate next key, given digested previous one
851 $obj->{prev_md5} = $_[1] ? $_[1] : undef;
852 $obj->{return_next} = 0;
855 # If the previous key was not specifed, start at the top and
856 # return the first one found.
858 if (!$obj->{prev_md5}) {
859 $obj->{prev_md5} = chr(0) x $self->{hash_size};
860 $obj->{return_next} = 1;
863 return $self->traverse_index( $obj, $obj->_base_offset, 0 );
868 sub _get_key_subloc {
870 my ($keys, $idx) = @_;
872 my ($key, $subloc, $size) = unpack(
873 "a$self->{hash_size} $self->{long_pack} $self->{long_pack}",
876 ($idx * $self->{bucket_size}),
877 $self->{bucket_size},
881 return ($key, $subloc, $size);
884 sub _find_in_buckets {
886 my ($tag, $md5) = @_;
889 for ( my $i = 0; $i < $self->{max_buckets}; $i++ ) {
890 my ($key, $subloc, $size) = $self->_get_key_subloc(
894 return ($subloc, $i * $self->{bucket_size}, $size) unless $subloc;
896 next BUCKET if $key ne $md5;
898 return ($subloc, $i * $self->{bucket_size}, $size);
906 my ($obj, $size) = @_;
908 my $loc = $obj->_root->{end};
909 $obj->_root->{end} += $size;
916 my ($obj, $size, $loc) = @_;
924 # This will be added in later, after more refactoring is done. This is an early
925 # attempt at refactoring on the physical level instead of the virtual level.
928 my ($obj, $spot, $amount, $unpack) = @_;
931 seek( $fh, $spot + $obj->_root->{file_offset}, SEEK_SET );
934 my $bytes_read = read( $fh, $buffer, $amount );
937 $buffer = unpack( $unpack, $buffer );
941 return ($buffer, $bytes_read);