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_REF () { 'R' }
15 sub SIG_NULL () { 'N' }
16 sub SIG_DATA () { 'D' }
17 sub SIG_INDEX () { 'I' }
18 sub SIG_BLIST () { 'B' }
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};
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;
109 $self->open( $obj ) if !defined $obj->_fh;
111 #XXX We have to make sure we don't mess up when autoflush isn't turned on
112 unless ( $obj->_root->{inode} ) {
113 my @stats = stat($obj->_fh);
114 $obj->_root->{inode} = $stats[1];
115 $obj->_root->{end} = $stats[7];
123 # Open a fh to the database, create if nonexistent.
124 # Make sure file signature matches DBM::Deep spec.
129 if (defined($obj->_fh)) { $self->close_fh( $obj ); }
131 # Theoretically, adding O_BINARY should remove the need for the binmode
132 # Of course, testing it is going to be ... interesting.
133 my $flags = O_RDWR | O_CREAT | O_BINARY;
136 my $filename = $obj->_root->{file};
137 sysopen( $fh, $filename, $flags )
138 or $obj->_throw_error("Cannot sysopen file '$filename': $!");
139 $obj->_root->{fh} = $fh;
141 #XXX Can we remove this by using the right sysopen() flags?
142 # Maybe ... q.v. above
143 binmode $fh; # for win32
145 if ($obj->_root->{autoflush}) {
146 my $old = select $fh;
151 seek($fh, 0 + $obj->_root->{file_offset}, SEEK_SET);
153 my $bytes_read = read( $fh, $signature, length(SIG_FILE));
156 # File is empty -- write signature and master index
159 seek($fh, 0 + $obj->_root->{file_offset}, SEEK_SET);
160 print( $fh SIG_FILE);
163 $obj, $obj->_base_offset, $obj->_type,
164 chr(0) x $self->{index_size},
167 # Flush the filehandle
168 my $old_fh = select $fh;
169 my $old_af = $|; $| = 1; $| = $old_af;
176 # Check signature was valid
178 unless ($signature eq SIG_FILE) {
179 $self->close_fh( $obj );
180 $obj->_throw_error("Signature not found -- file is not a Deep DB");
184 # Get our type from master index signature
186 my $tag = $self->load_tag($obj, $obj->_base_offset)
187 or $obj->_throw_error("Corrupted file, no master index record");
189 unless ($obj->{type} eq $tag->{signature}) {
190 $obj->_throw_error("File type mismatch");
193 #XXX We probably also want to store the hash algorithm name and not assume anything
194 #XXX The cool thing would be to allow a different hashing algorithm at every level
203 if ( my $fh = $obj->_root->{fh} ) {
206 $obj->_root->{fh} = undef;
213 # Given offset, signature and content, create tag and write to disk
216 my ($obj, $offset, $sig, $content) = @_;
217 my $size = length($content);
221 seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
222 print( $fh $sig . pack($self->{data_pack}, $size) . $content );
224 if ($offset == $obj->_root->{end}) {
225 $obj->_root->{end} += SIG_SIZE + $self->{data_size} + $size;
231 offset => $offset + SIG_SIZE + $self->{data_size},
238 # Given offset, load single tag and return signature, size and data
241 my ($obj, $offset) = @_;
245 seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
247 #XXX I'm not sure this check will work if autoflush isn't enabled ...
251 read( $fh, $b, SIG_SIZE + $self->{data_size} );
252 my ($sig, $size) = unpack( "A $self->{data_pack}", $b );
255 read( $fh, $buffer, $size);
260 offset => $offset + SIG_SIZE + $self->{data_size},
267 # Adds one key/value pair to bucket list, given offset, MD5 digest of key,
268 # plain (undigested) key and value.
271 my ($obj, $tag, $md5, $plain_key, $value) = @_;
276 my $root = $obj->_root;
278 my $is_dbm_deep = eval {
279 local $SIG{'__DIE__'};
280 $value->isa( 'DBM::Deep' );
283 my $internal_ref = $is_dbm_deep && ($value->_root eq $root);
287 my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 );
289 # Updating a known md5
294 # If value is a hash, array, or raw value with equal or less size, we
295 # can reuse the same content area of the database. Otherwise, we have
296 # to create a new content area at the EOF.
299 if ( $internal_ref ) {
300 $actual_length = $self->{long_size};
303 my $r = Scalar::Util::reftype( $value ) || '';
304 if ( $r eq 'HASH' || $r eq 'ARRAY' ) {
305 $actual_length = $self->{index_size};
307 # if autobless is enabled, must also take into consideration
308 # the class name, as it is stored along with key/value.
309 if ( $root->{autobless} ) {
310 my $value_class = Scalar::Util::blessed($value);
311 if ( defined $value_class && !$value->isa('DBM::Deep') ) {
312 $actual_length += length($value_class);
316 else { $actual_length = length($value); }
319 seek($fh, $subloc + SIG_SIZE + $root->{file_offset}, SEEK_SET);
321 read( $fh, $size, $self->{data_size});
322 $size = unpack($self->{data_pack}, $size);
324 if ($actual_length <= $size) {
328 $location = $root->{end};
331 $tag->{offset} + $offset + $self->{hash_size} + $root->{file_offset},
334 print( $fh pack($self->{long_pack}, $location) );
338 elsif ( defined $offset ) {
340 $location = $root->{end};
342 seek( $fh, $tag->{offset} + $offset + $root->{file_offset}, SEEK_SET );
343 print( $fh $md5 . pack($self->{long_pack}, $location) );
345 # If bucket didn't fit into list, split into a new index level
347 $self->split_index( $obj, $md5, $tag );
349 $location = $root->{end};
352 $self->write_value( $obj, $location, $plain_key, $value );
359 my ($obj, $location, $key, $value) = @_;
362 my $root = $obj->_root;
364 my $is_dbm_deep = eval {
365 local $SIG{'__DIE__'};
366 $value->isa( 'DBM::Deep' );
369 my $internal_ref = $is_dbm_deep && ($value->_root eq $root);
371 seek($fh, $location + $root->{file_offset}, SEEK_SET);
374 # Write signature based on content type, set content length and write
377 my $r = Scalar::Util::reftype($value) || '';
379 if ( $internal_ref ) {
380 print( $fh SIG_INTERNAL );
381 print( $fh pack($self->{data_pack}, $self->{long_size}) );
382 print( $fh pack($self->{long_pack}, $value->_base_offset) );
383 $content_length = $self->{long_size};
387 print( $fh SIG_HASH );
388 print( $fh pack($self->{data_pack}, $self->{index_size}) . chr(0) x $self->{index_size} );
389 $content_length = $self->{index_size};
391 elsif ($r eq 'ARRAY') {
392 print( $fh SIG_ARRAY );
393 print( $fh pack($self->{data_pack}, $self->{index_size}) . chr(0) x $self->{index_size} );
394 $content_length = $self->{index_size};
396 elsif (!defined($value)) {
397 print( $fh SIG_NULL );
398 print( $fh pack($self->{data_pack}, 0) );
402 print( $fh SIG_DATA );
403 print( $fh pack($self->{data_pack}, length($value)) . $value );
404 $content_length = length($value);
409 # Plain key is stored AFTER value, as keys are typically fetched less often.
411 print( $fh pack($self->{data_pack}, length($key)) . $key );
414 # If value is blessed, preserve class name
416 if ( $root->{autobless} ) {
417 my $value_class = Scalar::Util::blessed($value);
418 if ( defined $value_class && !$value->isa( 'DBM::Deep' ) ) {
420 # Blessed ref -- will restore later
423 print( $fh pack($self->{data_pack}, length($value_class)) . $value_class );
424 $content_length += 1;
425 $content_length += $self->{data_size} + length($value_class);
429 $content_length += 1;
434 # If this is a new content area, advance EOF counter
436 if ($location == $root->{end}) {
437 $root->{end} += SIG_SIZE;
438 $root->{end} += $self->{data_size} + $content_length;
439 $root->{end} += $self->{data_size} + length($key);
443 # If content is a hash or array, create new child DBM::Deep object and
444 # pass each key or element to it.
446 if ( ! $internal_ref ) {
448 my $branch = DBM::Deep->new(
449 type => DBM::Deep->TYPE_HASH,
450 base_offset => $location,
453 foreach my $key (keys %{$value}) {
454 $branch->STORE( $key, $value->{$key} );
457 elsif ($r eq 'ARRAY') {
458 my $branch = DBM::Deep->new(
459 type => DBM::Deep->TYPE_ARRAY,
460 base_offset => $location,
464 foreach my $element (@{$value}) {
465 $branch->STORE( $index, $element );
476 my ($obj, $md5, $tag) = @_;
479 my $root = $obj->_root;
480 my $keys = $tag->{content};
482 seek($fh, $tag->{ref_loc} + $root->{file_offset}, SEEK_SET);
483 print( $fh pack($self->{long_pack}, $root->{end}) );
485 my $index_tag = $self->create_tag(
489 chr(0) x $self->{index_size},
494 $keys .= $md5 . pack($self->{long_pack}, 0);
497 for (my $i = 0; $i <= $self->{max_buckets}; $i++) {
498 my ($key, $old_subloc) = $self->_get_key_subloc( $keys, $i );
500 next BUCKET unless $key;
502 my $num = ord(substr($key, $tag->{ch} + 1, 1));
504 if ($offsets[$num]) {
505 my $offset = $offsets[$num] + SIG_SIZE + $self->{data_size};
506 seek($fh, $offset + $root->{file_offset}, SEEK_SET);
508 read( $fh, $subkeys, $self->{bucket_list_size});
510 for (my $k=0; $k<$self->{max_buckets}; $k++) {
511 my ($temp, $subloc) = $self->_get_key_subloc( $subkeys, $k );
514 seek($fh, $offset + ($k * $self->{bucket_size}) + $root->{file_offset}, SEEK_SET);
515 print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) );
521 $offsets[$num] = $root->{end};
522 seek($fh, $index_tag->{offset} + ($num * $self->{long_size}) + $root->{file_offset}, SEEK_SET);
523 print( $fh pack($self->{long_pack}, $root->{end}) );
525 my $blist_tag = $self->create_tag($obj, $root->{end}, SIG_BLIST, chr(0) x $self->{bucket_list_size});
527 seek($fh, $blist_tag->{offset} + $root->{file_offset}, SEEK_SET);
528 print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) );
537 my ($obj, $subloc) = @_;
542 # Found match -- seek to offset and read signature
545 seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
546 read( $fh, $signature, SIG_SIZE);
549 # If value is a hash or array, return new DBM::Deep object with correct offset
551 if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) {
552 my $obj = DBM::Deep->new(
554 base_offset => $subloc,
558 if ($obj->_root->{autobless}) {
560 # Skip over value and plain key to see if object needs
563 seek($fh, $self->{data_size} + $self->{index_size}, SEEK_CUR);
566 read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size);
567 if ($size) { seek($fh, $size, SEEK_CUR); }
570 read( $fh, $bless_bit, 1);
571 if (ord($bless_bit)) {
573 # Yes, object needs to be re-blessed
576 read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size);
577 if ($size) { read( $fh, $class_name, $size); }
578 if ($class_name) { $obj = bless( $obj, $class_name ); }
584 elsif ( $signature eq SIG_INTERNAL ) {
586 read( $fh, $size, $self->{data_size});
587 $size = unpack($self->{data_pack}, $size);
591 read( $fh, $new_loc, $size );
592 $new_loc = unpack( $self->{long_pack}, $new_loc );
594 return $self->read_from_loc( $obj, $new_loc );
601 # Otherwise return actual value
603 elsif ($signature eq SIG_DATA) {
605 read( $fh, $size, $self->{data_size});
606 $size = unpack($self->{data_pack}, $size);
609 if ($size) { read( $fh, $value, $size); }
614 # Key exists, but content is null
619 sub get_bucket_value {
621 # Fetch single value given tag and MD5 digested key.
624 my ($obj, $tag, $md5) = @_;
626 my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 );
628 return $self->read_from_loc( $obj, $subloc );
635 # Delete single key/value pair given tag and MD5 digested key.
638 my ($obj, $tag, $md5) = @_;
640 my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 );
643 seek($fh, $tag->{offset} + $offset + $obj->_root->{file_offset}, SEEK_SET);
644 print( $fh substr($tag->{content}, $offset + $self->{bucket_size} ) );
645 print( $fh chr(0) x $self->{bucket_size} );
654 # Check existence of single key given tag and MD5 digested key.
657 my ($obj, $tag, $md5) = @_;
659 my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 );
663 sub find_bucket_list {
665 # Locate offset for bucket list, given digested key
668 my ($obj, $md5, $args) = @_;
669 $args = {} unless $args;
672 # Locate offset for bucket list using digest index system
674 my $tag = $self->load_tag($obj, $obj->_base_offset)
675 or $obj->_throw_error( "INTERNAL ERROR - Cannot find tag" );
678 while ($tag->{signature} ne SIG_BLIST) {
679 my $num = ord substr($md5, $ch, 1);
681 my $ref_loc = $tag->{offset} + ($num * $self->{long_size});
682 $tag = $self->index_lookup( $obj, $tag, $num );
685 return if ! $args->{create};
688 seek($fh, $ref_loc + $obj->_root->{file_offset}, SEEK_SET);
689 print( $fh pack($self->{long_pack}, $obj->_root->{end}) );
691 $tag = $self->create_tag(
692 $obj, $obj->_root->{end},
694 chr(0) x $self->{bucket_list_size},
697 $tag->{ref_loc} = $ref_loc;
704 $tag->{ref_loc} = $ref_loc;
714 # Given index tag, lookup single entry in index and return .
717 my ($obj, $tag, $index) = @_;
719 my $location = unpack(
723 $index * $self->{long_size},
728 if (!$location) { return; }
730 return $self->load_tag( $obj, $location );
735 # Scan index and recursively step into deeper levels, looking for next key.
738 my ($obj, $offset, $ch, $force_return_next) = @_;
740 my $tag = $self->load_tag($obj, $offset );
744 if ($tag->{signature} ne SIG_BLIST) {
745 my $content = $tag->{content};
746 my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1));
748 for (my $idx = $start; $idx < (2**8); $idx++) {
751 substr($content, $idx * $self->{long_size}, $self->{long_size}),
755 my $result = $self->traverse_index(
756 $obj, $subloc, $ch + 1, $force_return_next,
759 if (defined($result)) { return $result; }
763 $obj->{return_next} = 1;
767 my $keys = $tag->{content};
768 if ($force_return_next) { $obj->{return_next} = 1; }
771 # Iterate through buckets, looking for a key match
773 for (my $i = 0; $i < $self->{max_buckets}; $i++) {
774 my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
776 # End of bucket list -- return to outer loop
778 $obj->{return_next} = 1;
781 # Located previous key -- return next one found
782 elsif ($key eq $obj->{prev_md5}) {
783 $obj->{return_next} = 1;
786 # Seek to bucket location and skip over signature
787 elsif ($obj->{return_next}) {
788 seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
790 # Skip over value to get to plain key
792 read( $fh, $sig, SIG_SIZE );
795 read( $fh, $size, $self->{data_size});
796 $size = unpack($self->{data_pack}, $size);
797 if ($size) { seek($fh, $size, SEEK_CUR); }
799 # Read in plain key and return as scalar
801 read( $fh, $size, $self->{data_size});
802 $size = unpack($self->{data_pack}, $size);
803 if ($size) { read( $fh, $plain_key, $size); }
809 $obj->{return_next} = 1;
810 } # tag is a bucket list
817 # Locate next key, given digested previous one
822 $obj->{prev_md5} = $_[1] ? $_[1] : undef;
823 $obj->{return_next} = 0;
826 # If the previous key was not specifed, start at the top and
827 # return the first one found.
829 if (!$obj->{prev_md5}) {
830 $obj->{prev_md5} = chr(0) x $self->{hash_size};
831 $obj->{return_next} = 1;
834 return $self->traverse_index( $obj, $obj->_base_offset, 0 );
839 sub _get_key_subloc {
841 my ($keys, $idx) = @_;
843 my ($key, $subloc) = unpack(
844 "a$self->{hash_size} $self->{long_pack}",
847 ($idx * $self->{bucket_size}),
848 $self->{bucket_size},
852 return ($key, $subloc);
855 sub _find_in_buckets {
857 my ($tag, $md5) = @_;
860 for ( my $i = 0; $i < $self->{max_buckets}; $i++ ) {
861 my ($key, $subloc) = $self->_get_key_subloc( $tag->{content}, $i );
863 return ($subloc, $i * $self->{bucket_size}) unless $subloc;
865 next BUCKET if $key ne $md5;
867 return ($subloc, $i * $self->{bucket_size});
876 # This will be added in later, after more refactoring is done. This is an early
877 # attempt at refactoring on the physical level instead of the virtual level.
880 my ($obj, $spot, $amount, $unpack) = @_;
883 seek( $fh, $spot + $obj->_root->{file_offset}, SEEK_SET );
886 my $bytes_read = read( $fh, $buffer, $amount );
889 $buffer = unpack( $unpack, $buffer );
893 return ($buffer, $bytes_read);