--- trunk/fuse_dbi.pl 2004/08/07 15:16:50 8 +++ trunk/DBI.pm 2004/11/16 15:34:25 36 @@ -1,93 +1,314 @@ #!/usr/bin/perl +package Fuse::DBI; + +use 5.008; +use strict; +use warnings; + use POSIX qw(ENOENT EISDIR EINVAL ENOSYS O_RDWR); use Fuse; - use DBI; -use strict; +use Carp; +use Data::Dumper; + + +our $VERSION = '0.05'; + +=head1 NAME + +Fuse::DBI - mount your database as filesystem and use it + +=head1 SYNOPSIS + + use Fuse::DBI; + Fuse::DBI->mount( ... ); + +See C below for examples how to set parameters. + +=head1 DESCRIPTION + +This module will use C module, part of C +available at L to mount +your database as file system. + +That will give you possibility to use normal file-system tools (cat, grep, vi) +to manipulate data in database. + +It's actually opposite of Oracle's intention to put everything into database. + + +=head1 METHODS + +=cut + +=head2 mount + +Mount your database as filesystem. + +Let's suppose that your database have table C with following structure: + + id: int + filename: text + size: int + content: text + writable: boolean + +Following is example how to mount table like that to C: + + my $mnt = Fuse::DBI->mount({ + 'filenames' => 'select id,filename,size,writable from files', + 'read' => 'select content from files where id = ?', + 'update' => 'update files set content = ? where id = ?', + 'dsn' => 'DBI:Pg:dbname=test_db', + 'user' => 'database_user', + 'password' => 'database_password', + 'invalidate' => sub { ... }, + }); + +Options: + +=over 5 + +=item filenames + +SQL query which returns C (unique id for that row), C, +C and C boolean flag. + +=item read + +SQL query which returns only one column with content of file and has +placeholder C for C. + +=item update + +SQL query with two pace-holders, one for new content and one for C. + +=item dsn + +C dsn to connect to (contains database driver and name of database). + +=item user + +User with which to connect to database + +=item password + +Password for connecting to database + +=item invalidate + +Optional anonymous code reference which will be executed when data is updated in +database. It can be used as hook to delete cache (for example on-disk-cache) +which is created from data edited through C. + +=item fork + +Optional flag which forks after mount so that executing script will continue +running. Implementation is experimental. + +=back + +=cut + +my $dbh; +my $sth; +my $ctime_start; + +sub read_filenames; +sub fuse_module_loaded; + +# evil, evil way to solve this. It makes this module non-reentrant. But, since +# fuse calls another copy of this script for each mount anyway, this shouldn't +# be a problem. +my $fuse_self; + +sub mount { + my $class = shift; + my $self = {}; + bless($self, $class); + + my $arg = shift; + + print Dumper($arg); + + carp "mount needs 'dsn' to connect to (e.g. dsn => 'DBI:Pg:dbname=test')" unless ($arg->{'dsn'}); + carp "mount needs 'mount' as mountpoint" unless ($arg->{'mount'}); + + # save (some) arguments in self + foreach (qw(mount invalidate)) { + $self->{$_} = $arg->{$_}; + } + + foreach (qw(filenames read update)) { + carp "mount needs '$_' SQL" unless ($arg->{$_}); + } + + $ctime_start = time(); + + my $pid; + if ($arg->{'fork'}) { + $self->{'mounted'} = 1; + $pid = fork(); + die "fork() failed: $!" unless defined $pid; + # child will return to caller + if ($pid) { + return $self; + } + } + + $dbh = DBI->connect($arg->{'dsn'},$arg->{'user'},$arg->{'password'}, {AutoCommit => 0, RaiseError => 1}) || die $DBI::errstr; + + $sth->{'filenames'} = $dbh->prepare($arg->{'filenames'}) || die $dbh->errstr(); + + $sth->{'read'} = $dbh->prepare($arg->{'read'}) || die $dbh->errstr(); + $sth->{'update'} = $dbh->prepare($arg->{'update'}) || die $dbh->errstr(); -my $sql_filenames = q{ - select - oid as id, - namespace||'/'||name||' ['||oid||']' as filename, - length(template) as size, - iseditable as writable - from template ; -}; -my $sql_read = q{ - select template - from template - where oid = ?; + $self->{'sth'} = $sth; + + $self->{'read_filenames'} = sub { $self->read_filenames }; + $self->read_filenames; + + $self->{'mounted'} = 1 unless ($arg->{'fork'}); + + $fuse_self = \$self; + + Fuse::main( + mountpoint=>$arg->{'mount'}, + getattr=>\&e_getattr, + getdir=>\&e_getdir, + open=>\&e_open, + statfs=>\&e_statfs, + read=>\&e_read, + write=>\&e_write, + utime=>\&e_utime, + truncate=>\&e_truncate, + unlink=>\&e_unlink, + rmdir=>\&e_unlink, + debug=>0, + ); + + $self->{'mounted'} = 0; + + exit(0) if ($arg->{'fork'}); + + return 1; + }; -my $sql_update = q{ - update template - set template = ? - where oid = ?; +=head2 umount + +Unmount your database as filesystem. + + $mnt->umount; + +This will also kill background process which is translating +database to filesystem. + +=cut + +sub umount { + my $self = shift; + + if ($self->{'mounted'}) { + system "fusermount -u ".$self->{'mount'} || warn "umount error: $!" && return 0; + } + + return 1; +} + +$SIG{'INT'} = sub { + print STDERR "umount called by SIG INT\n"; + umount; }; +sub DESTROY { + my $self = shift; + return if (! $self->{'mounted'}); + print STDERR "umount called by DESTROY\n"; + $self->umount; +} + +=head2 fuse_module_loaded + +Checks if C module is loaded in kernel. -my $connect = "DBI:Pg:dbname=webgui"; + die "no fuse module loaded in kernel" + unless (Fuse::DBI::fuse_module_loaded); -my $dbh = DBI->connect($connect,"","", { AutoCommit => 0 }) || die $DBI::errstr; +This function in called by C, but might be useful alone also. -print "start transaction\n"; -#$dbh->begin_work || die $dbh->errstr; +=cut -my $sth_filenames = $dbh->prepare($sql_filenames) || die $dbh->errstr(); -$sth_filenames->execute() || die $sth_filenames->errstr(); - -my $sth_read = $dbh->prepare($sql_read) || die $dbh->errstr(); -my $sth_update = $dbh->prepare($sql_update) || die $dbh->errstr(); - -my $ctime_start = time(); - -my (%files) = ( - '.' => { - type => 0040, - mode => 0755, - }, -# a => { -# cont => "File 'a'.\n", -# type => 0100, -# ctime => time()-2000 -# }, -); +sub fuse_module_loaded { + my $lsmod = `lsmod`; + die "can't start lsmod: $!" unless ($lsmod); + if ($lsmod =~ m/fuse/s) { + return 1; + } else { + return 0; + } +} +my %files; my %dirs; -while (my $row = $sth_filenames->fetchrow_hashref() ) { - $files{$row->{'filename'}} = { - size => $row->{'size'}, - mode => $row->{'writable'} ? 0644 : 0444, - id => $row->{'id'} || 99, - }; - - my $d; - foreach (split(m!/!, $row->{'filename'})) { - # first, entry is assumed to be file - if ($d) { - $files{$d} = { - size => $dirs{$d}++, - mode => 0755, - type => 0040 - }; - $files{$d.'/.'} = { - mode => 0755, - type => 0040 - }; - $files{$d.'/..'} = { - mode => 0755, - type => 0040 - }; +sub read_filenames { + my $self = shift; + + my $sth = $self->{'sth'} || die "no sth argument"; + + # create empty filesystem + (%files) = ( + '.' => { + type => 0040, + mode => 0755, + }, + # a => { + # cont => "File 'a'.\n", + # type => 0100, + # ctime => time()-2000 + # }, + ); + + # fetch new filename list from database + $sth->{'filenames'}->execute() || die $sth->{'filenames'}->errstr(); + + # read them in with sesible defaults + while (my $row = $sth->{'filenames'}->fetchrow_hashref() ) { + $files{$row->{'filename'}} = { + size => $row->{'size'}, + mode => $row->{'writable'} ? 0644 : 0444, + id => $row->{'id'} || 99, + }; + + my $d; + foreach (split(m!/!, $row->{'filename'})) { + # first, entry is assumed to be file + if ($d) { + $files{$d} = { + size => $dirs{$d}++, + mode => 0755, + type => 0040 + }; + $files{$d.'/.'} = { + mode => 0755, + type => 0040 + }; + $files{$d.'/..'} = { + mode => 0755, + type => 0040 + }; + } + $d .= "/" if ($d); + $d .= "$_"; } - $d .= "/" if ($d); - $d .= "$_"; } + + print "found ",scalar(keys %files)-scalar(keys %dirs)," files, ",scalar(keys %dirs), " dirs\n"; } -print "found ",scalar(keys %files)-scalar(keys %dirs)," files, ",scalar(keys %dirs), " dirs\n"; sub filename_fixup { my ($file) = shift; @@ -120,12 +341,11 @@ # return as many text filenames as you like, followed by the retval. print((scalar keys %files)." files total\n"); my %out; - foreach (keys %files) { - my $f = $_; - $f =~ s/^\E$dirname\Q//; - $f =~ s/^\///; + foreach my $f (sort keys %files) { if ($dirname) { - $out{$f}++ if (/^\E$dirname\Q/ && $f =~ /^[^\/]+$/); + if ($f =~ s/^\Q$dirname\E\///) { + $out{$f}++ if ($f =~ /^[^\/]+$/); + } } else { $out{$f}++ if ($f =~ /^[^\/]+$/); } @@ -134,9 +354,23 @@ $out{'no files? bug?'}++; } print scalar keys %out," files in dir '$dirname'\n"; + print "## ",join(" ",keys %out),"\n"; return (keys %out),0; } +sub read_content { + my ($file,$id) = @_; + + die "read_content needs file and id" unless ($file && $id); + + $sth->{'read'}->execute($id) || die $sth->{'read'}->errstr; + $files{$file}{cont} = $sth->{'read'}->fetchrow_array; + # I should modify ctime only if content in database changed + #$files{$file}{ctime} = time() unless ($files{$file}{ctime}); + print "file '$file' content [",length($files{$file}{cont})," bytes] read in cache\n"; +} + + sub e_open { # VFS sanity check; it keeps all the necessary state, not much to do here. my $file = filename_fixup(shift); @@ -145,11 +379,8 @@ return -ENOENT() unless exists($files{$file}); return -EISDIR() unless exists($files{$file}{id}); - if (!exists($files{$file}{cont})) { - $sth_read->execute($files{$file}{id}) || die $sth_read->errstr; - $files{$file}{cont} = $sth_read->fetchrow_array; - print "file '$file' content read in cache\n"; - } + read_content($file,$files{$file}{id}) unless exists($files{$file}{cont}); + print "open '$file' ",length($files{$file}{cont})," bytes\n"; return 0; } @@ -170,7 +401,7 @@ return -EINVAL() if ($off > $len); return 0 if ($off == $len); - $buf_len = $buf_len-$off if ($off+$buf_len > $len); + $buf_len = $len-$off if ($len - $off < $buf_len); return substr($files{$file}{cont},$off,$buf_len); } @@ -181,9 +412,10 @@ print "invalidate all cached content\n"; foreach my $f (keys %files) { delete $files{$f}{cont}; + delete $files{$f}{ctime}; } print "begin new transaction\n"; - $dbh->begin_work || die $dbh->errstr; + #$dbh->begin_work || die $dbh->errstr; } @@ -192,40 +424,51 @@ $files{$file}{ctime} = time(); - if (!$sth_update->execute($files{$file}{cont},$files{$file}{id})) { - print "update problem: ",$sth_update->errstr; + my ($cont,$id) = ( + $files{$file}{cont}, + $files{$file}{id} + ); + + if (!$sth->{'update'}->execute($cont,$id)) { + print "update problem: ",$sth->{'update'}->errstr; clear_cont; return 0; } else { if (! $dbh->commit) { - print "ERROR: commit problem: ",$sth_update->errstr; + print "ERROR: commit problem: ",$sth->{'update'}->errstr; clear_cont; return 0; } print "updated '$file' [",$files{$file}{id},"]\n"; + + $$fuse_self->{'invalidate'}->() if (ref $$fuse_self->{'invalidate'}); } return 1; } sub e_write { my $file = filename_fixup(shift); - my ($buf_len,$off) = @_; + my ($buffer,$off) = @_; return -ENOENT() unless exists($files{$file}); - my $len = length($files{$file}{cont}); + my $cont = $files{$file}{cont}; + my $len = length($cont); - print "write '$file' [$len bytes] offset $off length\n"; + print "write '$file' [$len bytes] offset $off length ",length($buffer),"\n"; - $files{$file}{cont} = - substr($files{$file}{cont},0,$off) . - $buf_len . - substr($files{$file}{cont},$off+length($buf_len)); + $files{$file}{cont} = ""; + + $files{$file}{cont} .= substr($cont,0,$off) if ($off > 0); + $files{$file}{cont} .= $buffer; + $files{$file}{cont} .= substr($cont,$off+length($buffer),$len-$off-length($buffer)) if ($off+length($buffer) < $len); + + $files{$file}{size} = length($files{$file}{cont}); if (! update_db($file)) { return -ENOSYS(); } else { - return length($buf_len); + return length($buffer); } } @@ -233,7 +476,10 @@ my $file = filename_fixup(shift); my $size = shift; + print "truncate to $size\n"; + $files{$file}{cont} = substr($files{$file}{cont},0,$size); + $files{$file}{size} = $size; return 0 }; @@ -252,19 +498,50 @@ sub e_statfs { return 255, 1, 1, 1, 1, 2 } -# If you run the script directly, it will run fusermount, which will in turn -# re-run this script. Hence the funky semantics. -my ($mountpoint) = ""; -$mountpoint = shift(@ARGV) if @ARGV; -Fuse::main( - mountpoint=>$mountpoint, - getattr=>\&e_getattr, - getdir=>\&e_getdir, - open=>\&e_open, - statfs=>\&e_statfs, - read=>\&e_read, - write=>\&e_write, - utime=>\&e_utime, - truncate=>\&e_truncate, - debug=>0, -); +sub e_unlink { + my $file = filename_fixup(shift); + + if (exists( $dirs{$file} )) { + print "unlink '$file' will re-read template names\n"; + print Dumper($fuse_self); + $$fuse_self->{'read_filenames'}->(); + return 0; + } elsif (exists( $files{$file} )) { + print "unlink '$file' will invalidate cache\n"; + read_content($file,$files{$file}{id}); + return 0; + } + + return -ENOENT(); +} +1; +__END__ + +=head1 EXPORT + +Nothing. + +=head1 SEE ALSO + +C website +L + +Example for WebGUI which comes with this distribution in +directory C. It also contains a lot of documentation +about design of this module, usage and limitations. + +=head1 AUTHOR + +Dobrica Pavlinusic, Edpavlin@rot13.orgE + +=head1 COPYRIGHT AND LICENSE + +Copyright (C) 2004 by Dobrica Pavlinusic + +This library is free software; you can redistribute it and/or modify +it under the same terms as Perl itself, either Perl version 5.8.4 or, +at your option, any later version of Perl 5 you may have available. + + +=cut +