1 #============================================================= -*-perl-*-
3 # BackupPC::PoolWrite package
7 # This library defines a BackupPC::PoolWrite class for writing
8 # files to disk that are candidates for pooling. One instance
9 # of this class is used to write each file. The following steps
12 # - As the incoming data arrives, the first 1MB is buffered
13 # in memory so the MD5 digest can be computed.
15 # - A running comparison against all the candidate pool files
16 # (ie: those with the same MD5 digest, usually at most a single
17 # file) is done as new incoming data arrives. Up to $MaxFiles
18 # simultaneous files can be compared in parallel. This
19 # involves reading and uncompressing one or more pool files.
21 # - When a pool file no longer matches it is discarded from
22 # the search. If there are more than $MaxFiles candidates, one of
23 # the new candidates is added to the search, first checking
24 # that it matches up to the current point (this requires
25 # re-reading one of the other pool files).
27 # - When or if no pool files match then the new file is written
28 # to disk. This could occur many MB into the file. We don't
29 # need to buffer all this data in memory since we can copy it
30 # from the last matching pool file, up to the point where it
33 # - When all the new data is complete, if a pool file exactly
34 # matches then the file is simply created as a hardlink to
38 # Craig Barratt <cbarratt@users.sourceforge.net>
41 # Copyright (C) 2001-2007 Craig Barratt
43 # This program is free software; you can redistribute it and/or modify
44 # it under the terms of the GNU General Public License as published by
45 # the Free Software Foundation; either version 2 of the License, or
46 # (at your option) any later version.
48 # This program is distributed in the hope that it will be useful,
49 # but WITHOUT ANY WARRANTY; without even the implied warranty of
50 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
51 # GNU General Public License for more details.
53 # You should have received a copy of the GNU General Public License
54 # along with this program; if not, write to the Free Software
55 # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
57 #========================================================================
59 # Version 3.2.0, released 31 Dec 2008.
61 # See http://backuppc.sourceforge.net.
63 #========================================================================
65 package BackupPC::PoolWrite;
71 use BackupPC::FileZIO;
75 my($class, $bpc, $fileName, $fileSize, $compress) = @_;
78 fileName => $fileName,
79 fileSize => $fileSize,
81 compress => $compress,
92 $self->{hardLinkMax} = $bpc->ConfValue("HardLinkMax");
95 # Always unlink any current file in case it is already linked
97 unlink($fileName) if ( -f $fileName );
98 if ( $fileName =~ m{(.*)/.+} && !-d $1 ) {
99 eval { mkpath($1, 0, 0777) };
101 push(@{$self->{errors}}, "Unable to create directory $1 for $self->{fileName}");
107 my $BufSize = 1048576; # 1MB or 2^20
108 my $MaxFiles = 20; # max number of compare files open at one time
112 my($a, $dataRef) = @_;
114 return if ( $a->{eof} );
115 $a->{data} .= $$dataRef if ( defined($dataRef) );
116 return if ( length($a->{data}) < $BufSize && defined($dataRef) );
119 # Correct the fileSize if it is wrong (rsync might transfer
120 # a file whose length is different to the length sent with the
121 # file list if the file changes between the file list sending
122 # and the file sending). Here we only catch the case where
123 # we haven't computed the digest (ie: we have written no more
124 # than $BufSize). We catch the big file case below.
126 if ( !defined($dataRef) && !defined($a->{digest})
127 && $a->{fileSize} != length($a->{data}) ) {
128 #my $newSize = length($a->{data});
129 #print("Fixing file size from $a->{fileSize} to $newSize\n");
130 $a->{fileSize} = length($a->{data});
133 if ( !defined($a->{digest}) && length($a->{data}) > 0 ) {
135 # build a list of all the candidate matching files
137 my $md5 = Digest::MD5->new;
138 $a->{fileSize} = length($a->{data})
139 if ( $a->{fileSize} < length($a->{data}) );
140 $a->{digest} = $a->{bpc}->Buffer2MD5($md5, $a->{fileSize}, \$a->{data});
141 if ( !defined($a->{base} = $a->{bpc}->MD52Path($a->{digest},
143 push(@{$a->{errors}}, "Unable to get path from '$a->{digest}'"
144 . " for $a->{fileName}");
146 while ( @{$a->{files}} < $MaxFiles ) {
148 my $fileName = $a->{fileCnt} < 0 ? $a->{base}
149 : "$a->{base}_$a->{fileCnt}";
150 last if ( !-f $fileName );
152 # Don't attempt to match pool files that already
153 # have too many hardlinks. Also, don't match pool
154 # files with only one link since starting in
155 # BackupPC v3.0, BackupPC_nightly could be running
156 # in parallel (and removing those files). This doesn't
157 # eliminate all possible race conditions, but just
158 # reduces the odds. Other design steps eliminate
159 # the remaining race conditions of linking vs
162 if ( (stat(_))[3] >= $a->{hardLinkMax}
164 || !defined($fh = BackupPC::FileZIO->open($fileName, 0,
169 push(@{$a->{files}}, {
177 # if there are no candidate files then we must write
178 # the new file to disk
180 if ( !@{$a->{files}} ) {
181 $a->{fhOut} = BackupPC::FileZIO->open($a->{fileName},
183 if ( !defined($a->{fhOut}) ) {
184 push(@{$a->{errors}}, "Unable to open $a->{fileName}"
189 my $dataLen = length($a->{data});
190 if ( !defined($a->{fhOut}) && length($a->{data}) > 0 ) {
192 # See if the new chunk of data continues to match the
195 for ( my $i = 0 ; $i < @{$a->{files}} ; $i++ ) {
197 my $fileName = $a->{fileCnt} < 0 ? $a->{base}
198 : "$a->{base}_$a->{fileCnt}";
199 if ( $dataLen > 0 ) {
200 # verify next $dataLen bytes from candidate file
201 my $n = $a->{files}[$i]->{fh}->read(\$d, $dataLen);
202 next if ( $n == $dataLen && $d eq $a->{data} );
204 # verify candidate file is at EOF
205 my $n = $a->{files}[$i]->{fh}->read(\$d, 100);
208 #print(" File $a->{files}[$i]->{name} doesn't match\n");
210 # this candidate file didn't match. Replace it
211 # with a new candidate file. We have to qualify
212 # any new candidate file by making sure that its
213 # first $a->{nWrite} bytes match, plus the next $dataLen
214 # bytes match $a->{data}.
216 while ( -f $fileName ) {
218 if ( (stat(_))[3] >= $a->{hardLinkMax}
219 || !defined($fh = BackupPC::FileZIO->open($fileName, 0,
222 #print(" Discarding $fileName (open failed)\n");
223 $fileName = "$a->{base}_$a->{fileCnt}";
226 if ( !$a->{files}[$i]->{fh}->rewind() ) {
227 push(@{$a->{errors}},
228 "Unable to rewind $a->{files}[$i]->{name}"
231 $match = $a->filePartialCompare($a->{files}[$i]->{fh}, $fh,
232 $a->{nWrite}, $dataLen, \$a->{data});
234 $a->{files}[$i]->{fh}->close();
235 $a->{files}[$i]->{fh} = $fh,
236 $a->{files}[$i]->{name} = $fileName;
237 #print(" Found new candidate $fileName\n");
241 #print(" Discarding $fileName (no match)\n");
245 $fileName = "$a->{base}_$a->{fileCnt}";
249 # We couldn't find another candidate file
251 if ( @{$a->{files}} == 1 ) {
252 #print(" Exhausted matches, now writing\n");
253 $a->{fhOut} = BackupPC::FileZIO->open($a->{fileName},
255 if ( !defined($a->{fhOut}) ) {
256 push(@{$a->{errors}},
257 "Unable to open $a->{fileName}"
260 if ( !$a->{files}[$i]->{fh}->rewind() ) {
261 push(@{$a->{errors}},
263 . " $a->{files}[$i]->{name} for copy");
265 $a->filePartialCopy($a->{files}[$i]->{fh}, $a->{fhOut},
269 $a->{files}[$i]->{fh}->close();
270 splice(@{$a->{files}}, $i, 1);
275 if ( defined($a->{fhOut}) && $dataLen > 0 ) {
277 # if we are in writing mode then just write the data
279 my $n = $a->{fhOut}->write(\$a->{data});
280 if ( $n != $dataLen ) {
281 push(@{$a->{errors}}, "Unable to write $dataLen bytes to"
282 . " $a->{fileName} (got $n)");
285 $a->{nWrite} += $dataLen;
287 return if ( defined($dataRef) );
290 # We are at EOF, so finish up
295 # Make sure the fileSize was correct. See above for comments about
298 if ( $a->{nWrite} != $a->{fileSize} ) {
300 # Oops, fileSize was wrong, so our MD5 digest was wrong and our
301 # effort to match files likely failed. This is ugly, but our
302 # only choice at this point is to re-write the entire file with
303 # the correct length. We need to rename the file, open it for
304 # reading, and then re-write the file with the correct length.
307 #print("Doing big file fixup ($a->{fileSize} != $a->{nWrite})\n");
310 $a->{fileSize} = $a->{nWrite};
312 if ( defined($a->{fhOut}) ) {
313 if ( $a->{fileName} =~ /(.*)\// ) {
319 # Find a unique target temporary file name
322 while ( -f "$fileName/t$$.$i" ) {
325 $fileName = "$fileName/t$$.$i";
326 $a->{fhOut}->close();
327 if ( !rename($a->{fileName}, $fileName)
328 || !defined($fh = BackupPC::FileZIO->open($fileName, 0,
330 push(@{$a->{errors}}, "Can't rename $a->{fileName} -> $fileName"
331 . " or open during size fixup");
333 #print("Using temporary name $fileName\n");
334 } elsif ( defined($a->{files}) && defined($a->{files}[0]) ) {
336 # We haven't written anything yet, so just use the
337 # compare file to copy from.
339 $fh = $a->{files}[0]->{fh};
341 #print("Using compare file $a->{files}[0]->{name}\n");
343 if ( defined($fh) ) {
344 my $poolWrite = BackupPC::PoolWrite->new($a->{bpc}, $a->{fileName},
345 $a->{fileSize}, $a->{compress});
348 while ( $nRead < $a->{fileSize} ) {
349 my $thisRead = $a->{fileSize} - $nRead < $BufSize
350 ? $a->{fileSize} - $nRead : $BufSize;
352 my $n = $fh->read(\$data, $thisRead);
353 if ( $n != $thisRead ) {
354 push(@{$a->{errors}},
355 "Unable to read $thisRead bytes during resize"
356 . " from temp $fileName (got $n)");
359 $poolWrite->write(\$data);
363 unlink($fileName) if ( defined($fileName) );
364 if ( @{$a->{errors}} ) {
366 return (0, $a->{digest}, -s $a->{fileName}, $a->{errors});
368 return $poolWrite->close;
373 if ( $a->{fileSize} == 0 ) {
375 # Simply create an empty file
378 if ( !open(OUT, ">", $a->{fileName}) ) {
379 push(@{$a->{errors}}, "Can't open $a->{fileName} for empty"
385 # Close the compare files
387 foreach my $f ( @{$a->{files}} ) {
390 return (1, $a->{digest}, -s $a->{fileName}, $a->{errors});
391 } elsif ( defined($a->{fhOut}) ) {
392 $a->{fhOut}->close();
394 # Close the compare files
396 foreach my $f ( @{$a->{files}} ) {
399 return (0, $a->{digest}, -s $a->{fileName}, $a->{errors});
401 if ( @{$a->{files}} == 0 ) {
402 push(@{$a->{errors}}, "Botch, no matches on $a->{fileName}"
403 . " ($a->{digest})");
404 } elsif ( @{$a->{files}} > 1 ) {
406 # This is no longer a real error because $Conf{HardLinkMax}
407 # could be hit, thereby creating identical pool files
409 #my $str = "Unexpected multiple matches on"
410 # . " $a->{fileName} ($a->{digest})\n";
411 #for ( my $i = 0 ; $i < @{$a->{files}} ; $i++ ) {
412 # $str .= " -> $a->{files}[$i]->{name}\n";
414 #push(@{$a->{errors}}, $str);
416 for ( my $i = 0 ; $i < @{$a->{files}} ; $i++ ) {
417 if ( link($a->{files}[$i]->{name}, $a->{fileName}) ) {
418 #print(" Linked $a->{fileName} to $a->{files}[$i]->{name}\n");
420 # Close the compare files
422 foreach my $f ( @{$a->{files}} ) {
425 return (1, $a->{digest}, -s $a->{fileName}, $a->{errors});
429 # We were unable to link to the pool. Either we're at the
430 # hardlink max, or the pool file got deleted. Recover by
431 # writing the matching file, since we still have an open
434 for ( my $i = 0 ; $i < @{$a->{files}} ; $i++ ) {
435 if ( !$a->{files}[$i]->{fh}->rewind() ) {
436 push(@{$a->{errors}},
437 "Unable to rewind $a->{files}[$i]->{name}"
438 . " for copy after link fail");
441 $a->{fhOut} = BackupPC::FileZIO->open($a->{fileName},
443 if ( !defined($a->{fhOut}) ) {
444 push(@{$a->{errors}},
445 "Unable to open $a->{fileName}"
446 . " for writing after link fail");
448 $a->filePartialCopy($a->{files}[$i]->{fh}, $a->{fhOut},
455 # Close the compare files
457 foreach my $f ( @{$a->{files}} ) {
460 return (0, $a->{digest}, -s $a->{fileName}, $a->{errors});
465 # Finish writing: pass undef dataRef to write so it can do all
466 # the work. Returns a 4 element array:
468 # (existingFlag, digestString, outputFileLength, errorList)
474 return $a->write(undef);
484 if ( defined($a->{fhOut}) ) {
485 $a->{fhOut}->close();
486 unlink($a->{fileName});
488 foreach my $f ( @{$a->{files}} ) {
495 # Copy $nBytes from files $fhIn to $fhOut.
499 my($a, $fhIn, $fhOut, $nBytes) = @_;
502 while ( $nRead < $nBytes ) {
503 my $thisRead = $nBytes - $nRead < $BufSize
504 ? $nBytes - $nRead : $BufSize;
506 my $n = $fhIn->read(\$data, $thisRead);
507 if ( $n != $thisRead ) {
508 push(@{$a->{errors}},
509 "Unable to read $thisRead bytes from "
510 . $fhIn->name . " (got $n)");
513 $n = $fhOut->write(\$data, $thisRead);
514 if ( $n != $thisRead ) {
515 push(@{$a->{errors}},
516 "Unable to write $thisRead bytes to "
517 . $fhOut->name . " (got $n)");
525 # Compare $nBytes from files $fh0 and $fh1, and also compare additional
526 # $extra bytes from $fh1 to $$extraData.
528 sub filePartialCompare
530 my($a, $fh0, $fh1, $nBytes, $extra, $extraData) = @_;
534 while ( $nRead < $nBytes ) {
535 my $thisRead = $nBytes - $nRead < $BufSize
536 ? $nBytes - $nRead : $BufSize;
537 $n = $fh0->read(\$data0, $thisRead);
538 if ( $n != $thisRead ) {
539 push(@{$a->{errors}}, "Unable to read $thisRead bytes from "
540 . $fh0->name . " (got $n)");
543 $n = $fh1->read(\$data1, $thisRead);
544 return 0 if ( $n < $thisRead || $data0 ne $data1 );
548 # verify additional bytes
549 $n = $fh1->read(\$data1, $extra);
550 return 0 if ( $n != $extra || $data1 ne $$extraData );
553 $n = $fh1->read(\$data1, 100);
554 return 0 if ( $n != 0 );
560 # LinkOrCopy() does a hardlink from oldFile to newFile.
562 # If that fails (because there are too many links on oldFile)
563 # then oldFile is copied to newFile, and the pool stats are
564 # returned to be added to the new file list. That allows
565 # BackupPC_link to try again, and to create a new pool file
570 my($bpc, $oldFile, $oldFileComp, $newFile, $newFileComp) = @_;
573 unlink($newFile) if ( -f $newFile );
575 # Try to link if hardlink limit is ok, and compression types
578 return (1, undef) if ( (stat($oldFile))[3] < $bpc->{Conf}{HardLinkMax}
579 && !$oldFileComp == !$newFileComp
580 && link($oldFile, $newFile) );
582 # There are too many links on oldFile, or compression
583 # type if different, so now we have to copy it.
585 # We need to compute the file size, which is expensive
586 # since we need to read the file twice. That's probably
587 # ok since the hardlink limit is rarely hit.
589 my $readFd = BackupPC::FileZIO->open($oldFile, 0, $oldFileComp);
590 if ( !defined($readFd) ) {
591 return (0, undef, undef, undef, ["LinkOrCopy: can't open $oldFile"]);
593 while ( $readFd->read(\$data, $BufSize) > 0 ) {
594 $nRead += length($data);
598 my $poolWrite = BackupPC::PoolWrite->new($bpc, $newFile,
599 $nRead, $newFileComp);
600 while ( $readFd->read(\$data, $BufSize) > 0 ) {
601 $poolWrite->write(\$data);
603 my($exists, $digest, $outSize, $errs) = $poolWrite->close;
605 return ($exists, $digest, $nRead, $outSize, $errs);