3 # Proof-of-concept/reference decoder for LBS-format backup snapshots.
5 # This decoder aims to decompress an LBS snapshot. It is not meant to be
6 # particularly efficient, but should be a small and portable tool for doing so
7 # (important for recovering from data loss). It is also meant to serve as a
8 # check on the snapshot tool and data format itself, and serve as documentation
11 # This decoder does not understand TAR archives; it assumes that all segments
12 # in the snapshot have already been decompressed, and that objects are
13 # available simply as files in the filesystem. This simplifies the design.
15 # Limitations: Since this code is probably using 32-bit arithmetic, files
16 # larger than 2-4 GB may not be properly handled.
18 # Copyright (C) 2007 Michael Vrable
24 my $OBJECT_DIR; # Where are the unpacked objects available?
25 my $DEST_DIR = "."; # Where should restored files should be placed?
26 my $RECURSION_LIMIT = 3; # Bound on recursive object references
28 my $VERBOSE = 0; # Set to 1 to enable debugging messages
30 ############################ CHECKSUM VERIFICATION ############################
31 # A very simple layer for verifying checksums. Checksums may be used on object
32 # references directly, and can also be used to verify entire reconstructed
35 # A checksum to verify is given in the form "algorithm=hexdigest". Given such
36 # a string, we can construct a "verifier" object. Bytes can be incrementally
37 # added to the verifier, and at the end a test can be made to see if the
38 # checksum matches. The caller need not know what algorithm is used. However,
39 # at the moment we only support SHA-1 for computing digest (algorith name
44 if ($checksum !~ m/^(\w+)=([0-9a-f]+)$/) {
45 die "Malformed checksum: $checksum";
47 my ($algorithm, $hash) = ($1, $2);
48 if ($algorithm ne 'sha1') {
49 die "Unsupported checksum algorithm: $algorithm";
53 ALGORITHM => $algorithm,
55 DIGESTER => new Digest::SHA1
61 sub verifier_add_bytes {
63 my $digester = $verifier->{DIGESTER};
66 $digester->add($data);
71 my $digester = $verifier->{DIGESTER};
73 my $newhash = $digester->hexdigest();
74 if ($VERBOSE && $verifier->{HASH} ne $newhash) {
75 print STDERR "Verification failure: ",
76 $newhash, " != ", $verifier->{HASH}, "\n";
78 return ($verifier->{HASH} eq $newhash);
81 ################################ OBJECT ACCESS ################################
82 # The base of the decompressor is the object reference layer. See ref.h for a
83 # description of the format for object references. These functions will parse
84 # an object reference, locate the object data from the filesystem, perform any
85 # necessary integrity checks (if a checksum is included), and return the object
88 # First, try to parse the object reference string into constituent pieces.
89 # The format is segment/object(checksum)[range]. Both the checksum and
93 if ($ref_str !~ m/^([-0-9a-f]+)\/([0-9a-f]+)(\(\S+\))?(\[\S+\])?$/) {
94 die "Malformed object reference: $ref_str";
97 my ($segment, $object, $checksum, $range) = ($1, $2, $3, $4);
99 # Next, use the segment/object components to locate and read the object
100 # contents from disk.
101 open OBJECT, "<", "$OBJECT_DIR/$segment/$object"
102 or die "Unable to open object $OBJECT_DIR/$segment/$object: $!";
103 my $contents = join '', <OBJECT>;
106 # If a checksum was specified in the object reference, verify the object
107 # integrity by computing a checksum of the read data and comparing.
109 $checksum =~ m/^\((\S+)\)$/;
110 my $verifier = verifier_create($1);
111 verifier_add_bytes($verifier, $contents);
112 if (!verifier_check($verifier)) {
113 die "Integrity check for object $ref_str failed";
117 # If a range was specified, then only a subset of the bytes of the object
118 # are desired. Extract just the desired bytes.
120 if ($range !~ m/^\[(\d+)\+(\d+)\]$/) {
121 die "Malformed object range: $range";
124 my $object_size = length $contents;
125 my ($start, $length) = ($1 + 0, $2 + 0);
126 if ($start >= $object_size || $start + $length > $object_size) {
127 die "Object range $range falls outside object bounds "
128 . "(actual size $object_size)";
131 $contents = substr $contents, $start, $length;
137 ############################### FILE PROCESSING ###############################
138 # Process the metadata for a single file. process_file is the main entry
139 # point; it should be given a list of file metadata key/value pairs.
140 # iterate_objects is a helper function used to iterate over the set of object
141 # references that contain the file data for a regular file.
154 $str =~ s/%([0-9a-f]{2})/chr(hex($1))/ge;
158 sub iterate_objects {
159 my $callback = shift; # Function to be called for each reference
160 my $arg = shift; # Argument passed to callback
161 my $text = shift; # Whitespace-separate list of object references
163 # Simple limit to guard against cycles in the object references
164 my $recursion_level = shift || 0;
165 if ($recursion_level >= $RECURSION_LIMIT) {
166 die "Recursion limit reached";
169 # Split the provided text at whitespace boundaries to produce the list of
170 # object references. If any of these start with "@", then we have an
171 # indirect reference, and must look up that object and call iterate_objects
174 foreach $obj (split /\s+/, $text) {
176 if ($obj =~ /^@(\S+)$/) {
177 my $indirect = load_ref($1);
178 iterate_objects($callback, $arg, $indirect, $recursion_level + 1);
180 &$callback($arg, $obj);
188 my $data = load_ref($obj);
190 or die "Error writing file data: $!";
191 verifier_add_bytes($state->{VERIFIER}, $data);
192 $state->{BYTES} += length($data);
195 # Extract the contents of a regular file by concatenating all the objects that
202 if (!defined $info{data}) {
203 die "File contents not specified for $name";
205 if (!defined $info{checksum} || !defined $info{size}) {
206 die "File $name is missing checksum or size";
209 $info{size} = parse_int($info{size});
211 # Open the file to be recreated. The data will be written out by the call
212 # to iterate_objects.
213 open FILE, ">", "$DEST_DIR/$name"
214 or die "Cannot write file $name: $!";
216 # Set up state so that we can incrementally compute the checksum and length
217 # of the reconstructed data. Then iterate over all objects in the file.
218 $state{VERIFIER} = verifier_create($info{checksum});
220 iterate_objects(\&obj_callback, \%state, $info{data});
224 # Verify that the reconstructed object matches the size/checksum we were
226 if (!verifier_check($state{VERIFIER}) || $state{BYTES} != $info{size}) {
227 die "File reconstruction failed for $name: size or checksum differs";
234 if (!defined($info{name})) {
235 die "Filename not specified in metadata block";
238 my $type = $info{type};
240 my $filename = uri_decode($info{name});
241 print "$filename\n" if $VERBOSE;
243 # Restore the specified file. How to do so depends upon the file type, so
244 # dispatch based on that.
245 my $dest = "$DEST_DIR/$filename";
248 unpack_file($filename, %info);
249 } elsif ($type eq 'd') {
251 if ($filename ne '.') {
252 mkdir $dest or die "Cannot create directory $filename: $!";
254 } elsif ($type eq 'l') {
256 if (!defined($info{contents})) {
257 die "Symlink $filename has no value specified";
259 my $contents = uri_decode($info{contents});
260 symlink $contents, $dest
261 or die "Cannot create symlink $filename: $!";
263 # TODO: We can't properly restore all metadata for symbolic links
264 # (attempts to do so below will change metadata for the pointed-to
265 # file). This should be later fixed, but for now we simply return
266 # before getting to the restore metadata step below.
268 } elsif ($type eq 'p' || $type eq 's' || $type eq 'c' || $type eq 'b') {
269 # Pipe, socket, character device, block device.
270 # TODO: Handle these cases.
271 print STDERR "Ignoring special file $filename of type $type\n";
274 die "Unknown file type '$type' for file $filename";
277 # Restore mode, ownership, and any other metadata for the file. This is
278 # split out from the code above since the code is the same regardless of
280 my $mtime = $info{mtime} || time();
281 utime time(), $mtime, $dest
282 or warn "Unable to update mtime for $dest";
286 if (defined $info{user}) {
287 my @items = split /\s/, $info{user};
288 $uid = parse_int($items[0]) if exists $items[0];
290 if (defined $info{group}) {
291 my @items = split /\s/, $info{group};
292 $gid = parse_int($items[0]) if exists $items[0];
294 chown $uid, $gid, $dest
295 or warn "Unable to change ownership for $dest";
297 if (defined $info{mode}) {
298 my $mode = parse_int($info{mode});
300 or warn "Unable to change permissions for $dest";
304 ########################### METADATA LIST PROCESSING ##########################
305 # Process the file metadata listing provided, and as information for each file
306 # is extracted, pass it to process_file. This will recursively follow indirect
307 # references to other metadata objects.
308 sub process_metadata {
309 my ($metadata, $recursion_level) = @_;
311 # Check recursion; this will prevent us from infinitely recursing on an
312 # indirect reference which loops back to itself.
313 $recursion_level ||= 0;
314 if ($recursion_level >= $RECURSION_LIMIT) {
315 die "Recursion limit reached";
318 # Split the metadata into lines, then start processing each line. There
319 # are two primary cases:
320 # - Lines starting with "@" are indirect references to other metadata
321 # objects. Recursively process that object before continuing.
322 # - Other lines should come in groups separated by a blank line; these
323 # contain metadata for a single file that should be passed to
325 # Note that blocks of metadata about a file cannot span a boundary between
330 foreach $line (split /\n/, $metadata) {
331 # If we find a blank line or a reference to another block, process any
332 # data for the previous file first.
333 if ($line eq '' || $line =~ m/^@/) {
334 process_file(%info) if %info;
340 # Recursively handle indirect metadata blocks.
341 if ($line =~ m/^@(\S+)$/) {
342 print "Indirect: $1\n" if $VERBOSE;
343 my $indirect = load_ref($1);
344 process_metadata($indirect, $recursion_level + 1);
348 # Try to parse the data as "key: value" pairs of file metadata. Also
349 # handle continuation lines, which start with whitespace and continue
350 # the previous "key: value" pair.
351 if ($line =~ m/^(\w+):\s*(.*)$/) {
354 } elsif ($line =~/^\s/ && defined $last_key) {
355 $info{$last_key} .= $line;
357 print STDERR "Junk in file metadata section: $line\n";
361 # Process any last file metadata which has not already been processed.
362 process_file(%info) if %info;
365 ############################### MAIN ENTRY POINT ##############################
366 # Program start. We expect to be called with a single argument, which is the
367 # name of the backup descriptor file written by a backup pass. This will name
368 # the root object in the snapshot, from which we can reach all other data we
371 # Parse command-line arguments. The first (required) is the name of the
372 # snapshot descriptor file. The backup objects are assumed to be stored in the
373 # same directory as the descriptor. The second (optional) argument is the
374 # directory where the restored files should be written; it defaults to ".";
375 my $descriptor = $ARGV[0];
376 unless (defined($descriptor) && -r $descriptor) {
377 print STDERR "Usage: $0 <snapshot file>\n";
381 if (defined($ARGV[1])) {
382 $DEST_DIR = $ARGV[1];
385 $OBJECT_DIR = dirname($descriptor);
386 print "Source directory: $OBJECT_DIR\n" if $VERBOSE;
388 # Read the snapshot descriptor to find the root object. Parse it to get a set
389 # of key/value pairs.
390 open DESCRIPTOR, "<", $descriptor
391 or die "Cannot open backup descriptor file $descriptor: $!";
393 my ($line, $last_key);
394 while (defined($line = <DESCRIPTOR>)) {
395 # Any lines of the form "key: value" should be inserted into the
396 # %descriptor dictionary. Any continuation line (a line starting with
397 # whitespace) will append text to the previous key's value. Ignore other
401 if ($line =~ m/^(\w+):\s*(.*)$/) {
402 $descriptor{$1} = $2;
404 } elsif ($line =~/^\s/ && defined $last_key) {
405 $descriptor{$last_key} .= $line;
408 print STDERR "Ignoring line in backup descriptor: $line\n";
412 # A valid backup descriptor should at the very least specify the root metadata
414 if (!exists $descriptor{Root}) {
415 die "Expected 'Root:' specification in backup descriptor file";
417 my $root = $descriptor{Root};
420 # Set the umask to something restrictive. As we unpack files, we'll originally
421 # write the files/directories without setting the permissions, so be
422 # conservative and ensure that they can't be read. Afterwards, we'll properly
423 # fix up permissions.
426 # Start processing metadata stored in the root to recreate the files.
427 print "Root object: $root\n" if $VERBOSE;
428 my $contents = load_ref($root);
429 process_metadata($contents);