# in the snapshot have already been decompressed, and that objects are
# available simply as files in the filesystem. This simplifies the design.
#
+# Limitations: Since this code is probably using 32-bit arithmetic, files
+# larger than 2-4 GB may not be properly handled.
+#
# Copyright (C) 2007 Michael Vrable
use strict;
use Digest::SHA1;
use File::Basename;
-my $OBJECT_DIR = "."; # Directory where objects are unpacked
+my $OBJECT_DIR; # Where are the unpacked objects available?
+my $DEST_DIR = "."; # Where should restored files should be placed?
my $RECURSION_LIMIT = 3; # Bound on recursive object references
+my $VERBOSE = 0; # Set to 1 to enable debugging messages
+
############################ CHECKSUM VERIFICATION ############################
-# A very simple later for verifying checksums. Checksums may be used on object
+# A very simple layer for verifying checksums. Checksums may be used on object
# references directly, and can also be used to verify entire reconstructed
# files.
#
my $digester = $verifier->{DIGESTER};
my $newhash = $digester->hexdigest();
- if ($verifier->{HASH} ne $newhash) {
+ if ($VERBOSE && $verifier->{HASH} ne $newhash) {
print STDERR "Verification failure: ",
$newhash, " != ", $verifier->{HASH}, "\n";
}
# iterate_objects is a helper function used to iterate over the set of object
# references that contain the file data for a regular file.
+sub parse_int {
+ my $str = shift;
+ if ($str =~ /^0/) {
+ return oct($str);
+ } else {
+ return $str + 0;
+ }
+}
+
+sub uri_decode {
+ my $str = shift;
+ $str =~ s/%([0-9a-f]{2})/chr(hex($1))/ge;
+ return $str;
+}
+
sub iterate_objects {
my $callback = shift; # Function to be called for each reference
my $arg = shift; # Argument passed to callback
next if $obj eq "";
if ($obj =~ /^@(\S+)$/) {
my $indirect = load_ref($1);
- iterate_objects($callback, $arg, $1, $recursion_level + 1);
+ iterate_objects($callback, $arg, $indirect, $recursion_level + 1);
} else {
&$callback($arg, $obj);
}
}
sub obj_callback {
- my $verifier = shift;
+ my $state = shift;
my $obj = shift;
my $data = load_ref($obj);
- print " ", $obj, " (size ", length($data), ")\n";
- verifier_add_bytes($verifier, $data);
+ print FILE $data
+ or die "Error writing file data: $!";
+ verifier_add_bytes($state->{VERIFIER}, $data);
+ $state->{BYTES} += length($data);
+}
+
+# Extract the contents of a regular file by concatenating all the objects that
+# comprise it.
+sub unpack_file {
+ my $name = shift;
+ my %info = @_;
+ my %state = ();
+
+ if (!defined $info{data}) {
+ die "File contents not specified for $name";
+ }
+ if (!defined $info{checksum} || !defined $info{size}) {
+ die "File $name is missing checksum or size";
+ }
+
+ $info{size} = parse_int($info{size});
+
+ # Open the file to be recreated. The data will be written out by the call
+ # to iterate_objects.
+ open FILE, ">", "$DEST_DIR/$name"
+ or die "Cannot write file $name: $!";
+
+ # Set up state so that we can incrementally compute the checksum and length
+ # of the reconstructed data. Then iterate over all objects in the file.
+ $state{VERIFIER} = verifier_create($info{checksum});
+ $state{BYTES} = 0;
+ iterate_objects(\&obj_callback, \%state, $info{data});
+
+ close FILE;
+
+ # Verify that the reconstructed object matches the size/checksum we were
+ # given.
+ if (!verifier_check($state{VERIFIER}) || $state{BYTES} != $info{size}) {
+ die "File reconstruction failed for $name: size or checksum differs";
+ }
}
sub process_file {
my %info = @_;
- # TODO
- print "process_file: ", $info{name}, "\n";
+ if (!defined($info{name})) {
+ die "Filename not specified in metadata block";
+ }
- if (defined $info{data}) {
- my $verifier = verifier_create($info{checksum});
+ my $type = $info{type};
+
+ my $filename = uri_decode($info{name});
+ print "$filename\n" if $VERBOSE;
+
+ # Restore the specified file. How to do so depends upon the file type, so
+ # dispatch based on that.
+ my $dest = "$DEST_DIR/$filename";
+ if ($type eq '-') {
+ # Regular file
+ unpack_file($filename, %info);
+ } elsif ($type eq 'd') {
+ # Directory
+ if ($filename ne '.') {
+ mkdir $dest or die "Cannot create directory $filename: $!";
+ }
+ } elsif ($type eq 'l') {
+ # Symlink
+ if (!defined($info{contents})) {
+ die "Symlink $filename has no value specified";
+ }
+ my $contents = uri_decode($info{contents});
+ symlink $contents, $dest
+ or die "Cannot create symlink $filename: $!";
+
+ # TODO: We can't properly restore all metadata for symbolic links
+ # (attempts to do so below will change metadata for the pointed-to
+ # file). This should be later fixed, but for now we simply return
+ # before getting to the restore metadata step below.
+ return;
+ } elsif ($type eq 'p' || $type eq 's' || $type eq 'c' || $type eq 'b') {
+ # Pipe, socket, character device, block device.
+ # TODO: Handle these cases.
+ print STDERR "Ignoring special file $filename of type $type\n";
+ return;
+ } else {
+ die "Unknown file type '$type' for file $filename";
+ }
- iterate_objects(\&obj_callback, $verifier, $info{data});
+ # Restore mode, ownership, and any other metadata for the file. This is
+ # split out from the code above since the code is the same regardless of
+ # file type.
+ my $mtime = $info{mtime} || time();
+ utime time(), $mtime, $dest
+ or warn "Unable to update mtime for $dest";
+
+ my $uid = -1;
+ my $gid = -1;
+ if (defined $info{user}) {
+ my @items = split /\s/, $info{user};
+ $uid = parse_int($items[0]) if exists $items[0];
+ }
+ if (defined $info{group}) {
+ my @items = split /\s/, $info{group};
+ $gid = parse_int($items[0]) if exists $items[0];
+ }
+ chown $uid, $gid, $dest
+ or warn "Unable to change ownership for $dest";
- print " checksum: ", (verifier_check($verifier) ? "pass" : "fail"),
- " ", $info{checksum}, "\n";
+ if (defined $info{mode}) {
+ my $mode = parse_int($info{mode});
+ chmod $mode, $dest
+ or warn "Unable to change permissions for $dest";
}
}
# metadata objects.
my %info = ();
my $line;
+ my $last_key;
foreach $line (split /\n/, $metadata) {
# If we find a blank line or a reference to another block, process any
# data for the previous file first.
if ($line eq '' || $line =~ m/^@/) {
process_file(%info) if %info;
%info = ();
+ undef $last_key;
next if $line eq '';
}
# Recursively handle indirect metadata blocks.
if ($line =~ m/^@(\S+)$/) {
- print "Indirect: $1\n";
+ print "Indirect: $1\n" if $VERBOSE;
my $indirect = load_ref($1);
process_metadata($indirect, $recursion_level + 1);
next;
}
- # Try to parse the data as "key: value" pairs of file metadata.
- if ($line =~ m/^(\w+):\s+(.*)\s*$/) {
+ # Try to parse the data as "key: value" pairs of file metadata. Also
+ # handle continuation lines, which start with whitespace and continue
+ # the previous "key: value" pair.
+ if ($line =~ m/^(\w+):\s*(.*)$/) {
$info{$1} = $2;
+ $last_key = $1;
+ } elsif ($line =~/^\s/ && defined $last_key) {
+ $info{$last_key} .= $line;
} else {
print STDERR "Junk in file metadata section: $line\n";
}
# the root object in the snapshot, from which we can reach all other data we
# need.
+# Parse command-line arguments. The first (required) is the name of the
+# snapshot descriptor file. The backup objects are assumed to be stored in the
+# same directory as the descriptor. The second (optional) argument is the
+# directory where the restored files should be written; it defaults to ".";
my $descriptor = $ARGV[0];
unless (defined($descriptor) && -r $descriptor) {
print STDERR "Usage: $0 <snapshot file>\n";
exit 1;
}
+if (defined($ARGV[1])) {
+ $DEST_DIR = $ARGV[1];
+}
+
$OBJECT_DIR = dirname($descriptor);
-print "Source directory: $OBJECT_DIR\n";
+print "Source directory: $OBJECT_DIR\n" if $VERBOSE;
+# Read the snapshot descriptor to find the root object. Parse it to get a set
+# of key/value pairs.
open DESCRIPTOR, "<", $descriptor
or die "Cannot open backup descriptor file $descriptor: $!";
-my $line = <DESCRIPTOR>;
-if ($line !~ m/^root: (\S+)$/) {
- die "Expected 'root:' specification in backup descriptor file";
+my %descriptor = ();
+my ($line, $last_key);
+while (defined($line = <DESCRIPTOR>)) {
+ # Any lines of the form "key: value" should be inserted into the
+ # %descriptor dictionary. Any continuation line (a line starting with
+ # whitespace) will append text to the previous key's value. Ignore other
+ # lines.
+ chomp $line;
+
+ if ($line =~ m/^(\w+):\s*(.*)$/) {
+ $descriptor{$1} = $2;
+ $last_key = $1;
+ } elsif ($line =~/^\s/ && defined $last_key) {
+ $descriptor{$last_key} .= $line;
+ } else {
+ undef $last_key;
+ print STDERR "Ignoring line in backup descriptor: $line\n";
+ }
+}
+
+# A valid backup descriptor should at the very least specify the root metadata
+# object.
+if (!exists $descriptor{Root}) {
+ die "Expected 'Root:' specification in backup descriptor file";
}
-my $root = $1;
+my $root = $descriptor{Root};
close DESCRIPTOR;
-print "Root object: $root\n";
+# Set the umask to something restrictive. As we unpack files, we'll originally
+# write the files/directories without setting the permissions, so be
+# conservative and ensure that they can't be read. Afterwards, we'll properly
+# fix up permissions.
+umask 077;
+# Start processing metadata stored in the root to recreate the files.
+print "Root object: $root\n" if $VERBOSE;
my $contents = load_ref($root);
process_metadata($contents);
projects / cumulus.git / blobdiff