#!/usr/bin/perl -w
# fixscript will replace this line with require innshellvars.pl
$ID='$Id: thdexpire.in 4572 2001-02-24 22:31:05Z rra $$';

use POSIX ":fcntl_h";
use SDBM_File;
use Getopt::Std;

# With the -M switch this program installs its own man page.
#-----------------------------------------------------------------------------

=head1 NAME

thdexpire - dynamic expire daemon for timehash and timecaf storage

=head1 SYNOPSIS

B<thdexpire>
[ B<-t> I<minutes> ]
[ B<-f> I<kilobytes> ]
[ B<-i> I<inodes> ]
[ B<-m> I<mindays> ]
[ B<-x> I<minseconds> ]
[ B<-N> ]
[ B<-v> I<level> ]

B<thdexpire -r>

=head1 DESCRIPTION

This is a daemon, to be started along with B<innd>, which periodically
looks if news spool space is getting tight, and frees space by removing
articles until enough is free. It is an adjunct (not a replacement) to
INNs B<expire> program.

=head2 Setting Up

=over 4

=item 1.

Configure your storage classes carefully. Let the default go in class
100 and choose the storage classes as relative (percent) retention
times. E.g. if you want to give C<alt.binaries.*> a fifth of the
default time, put them in class 20. Storage classes above 200 are
ignored by this program. 0 expires immediately. An example is given
in L<"EXAMPLES">.

=item 2.

Set up your F<expire.ctl> in a way that it puts only a maximum cap on
retention times. Run B<expire> from B<news.daily> as usual. However,
it should only expire articles which have an Expires line or are in
classes above 200. See L<"EXAMPLES">.

=item 3.

Ensure to start this daemon along with B<innd>.

=item 4.

To get information and statistics, run B<thdexpire -r> (in parallel to
a running daemon). This will show you the current actual retention
times.

=back

=head2 How It Works

B<thdexpire> works directly on the spool. It assumes the layout
described in the timehash and timecaf sections of L<storage.conf(5)> as of
INN-2.x-CURRENT (Dec. 5, 1998). For every storage class associated
with timehash/timecaf, B<thdexpire> keeps a I<work time> which is the
modification time of the oldest article/CAF file in this class. This
time is chosen so that the difference of the work time of class N to
now (i.e. the I<retention time> for class N) will be N/100 of the
retention time of class 100. The work time of all classes is
continuously adjusted as time goes by. Articles and CAF files which
are older than the work time are deleted.

=head1 OPTIONS

=over 8

=item B<-t> I<minutes>

Check for free space every I<minutes> minutes (default 30).

=item B<-f> I<kilobytes>

Leave I<kilobytes> kilobytes of free disk space on each spool
filesystem (default 50000).

=item B<-i> I<inodes>

Leave I<inodes> inodes free on each spool filesystem (default 5000).

=item B<-m> I<mindays>

Set the minimum normal holding time for class 100 to I<mindays> days
(default 7).

=item B<-x> I<minseconds>

Set the absolute minimum holding time for any article to I<minseconds>
seconds (default 86400, i.e. 1 day).

=item B<-N>

Do not delete any articles, just print what would be done.

=item B<-v> I<level>

Set the verbosity level. Values from 1 to 3 are meaningful, where
higher levels are mostly for debugging.

=item B<-r>

Do not run as a daemon, instead print a report from the database (see
L<FILES>) on the available storage classes, current expire times and
other stuff.

=back

=head1 EXAMPLES

Here is an example F<storage.conf> file:

 # Large postings in binary groups are expired fast:
 # 20% retention time
 method timehash {
   newsgroups: *.binaries.*,*.binaer.*,*.dateien.*,alt.mag.*
   size: 30000
   class: 20
 }

 # Local groups and *.answers groups don't expire at all with
 # thdexpire. These are handled by Expires lines and a cutoff
 # in expire.ctl.
 method timehash {
   newsgroups: *.answers,news.announce.*,local.*
   class: 201
 }

 # Expires lines are honored if they dont exceed 90 days.
 # Exempt those postings from thdexpire handling.
 method timehash {
   newsgroups: *
   expires: 1d,90d
   class: 202
 }

 # Default: should be class 100 because thdexpire bases its
 # calculations thereupon.
 method timecaf {
   newsgroups: *
   class: 100
 }

And here is an F<expire.ctl> which fits:

 # Our local groups are held 6 months
 local.*:A:7:180:180
 # Everything else is handled by thdexpire, or Expires lines
 *:A:7:never:never

Note that B<thdexpire> does not actually use these files, they just
configure other parts of the news system in an appropriate way.

=head1 FILES

=over 4

=item F<E<lt>inn::pathdbE<gt>/thdexpstat.{dir,pag}>

Holds state information like classes, expire times, oldest articles.
When this file is missing, it will be rebuilt the next time the daemon
is started, which basically means scanning the spool directories to
find the oldest articles. With the B<-r> option, the contents of this
file are printed.

=item F<E<lt>inn::innddirE<gt>/thdexpire.pid>

Contains the PID of the running daemon.

=back

=head1 SIGNALS

I<SIGINT> or I<SIGTERM> can be sent to the daemon at any time, causing
it to gracefully exit immediately.

=head1 SEE ALSO

L<expire(8)>, L<news.daily(8)>, L<storage.conf(5)>

=head1 NOTES

This version needs the B<inndf> program supplied with newer releases of INN.

The filenames for timecaf were wrong in older versions of the INN
documentation. This program uses the true filenames, as found by
reading the INN source.

=head1 DIAGNOSTICS

Any error messages are printed on standard error. Normal progress
messages, as specified by the B<-v> option, are printed on standard
output.

=head1 BUGS

Storage classes which are in I<storage.conf> but not on disk (i.e.
which have never been filed into) when the daemon starts are ignored.

The code is ugly and uses too many global variables.
Should probably rewrite it in C.

=head1 RESTRICTIONS

Directories which are left empty are not removed.

The overview database is not affected by B<thdexpire>, it has to be
cleaned up by the daily regular B<news.daily> run. This may need a
patch to B<expire>.

=head1 AUTHOR

Olaf Titz <olaf@bigred.inka.de>. Use and distribution of this work is
permitted under the same terms as the B<INN> package.

=head1 HISTORY

Inspired by the old B<dexpire> program for the traditional spool.

June 1998: wrote the first version for timehash.

November 1998: added code for timecaf, works on multiple spool
filesystems, PODed documentation.

July 1999: bugfixes.

=cut

#-----------------------------------------------------------------------------

chdir $inn::spool || die "chdir $inn::spool: $!";
$opt_r=0;      # make a report
$opt_t=30;     # check interval in minutes
$opt_f=50000;  # required space in kilobytes
$opt_i=5000;   # required space in inodes
$opt_m=7;      # minimum normal (class 100) time in days
$opt_x=86400;  # absolute minimum hold time in seconds
$opt_N=0;      # dont actually delete articles
$opt_v=0;      # verbosity level
$opt_M=0;      # install man page
getopts("rt:f:i:m:x:Nv:M");

$_=$inn::pathdb; $_=$inn::pathnews; # shut up warning
$sfile="$inn::pathdb/thdexpstat";
$ID=~/ ([^,]+,v [^ ]+)/; $ID=$1;

if ($opt_M) {
    print "Installing thdexpire(8) man page\n";
    $0=~m:^(.*)/([^/]+)$:;
    chdir $1 || die "chdir $1";
    exec "pod2man --section=8 --center='Contributed News Software'" .
      " --release='$ID' $2 >$inn::pathnews/man/man8/thdexpire.8";
}

if ($opt_r) {
    tie(%S, SDBM_File, $sfile, O_RDONLY, 0664) || die "open $sfile: $!";
    &report;
    untie %S;
    exit 0;
}

(system "shlock", "-p", $$, "-f", "$inn::innddir/thdexpire.pid")>>8==0
  || die "Already running";
tie(%S, SDBM_File, $sfile, O_RDWR|O_CREAT, 0664) || die "open $sfile: $!";
$SIG{'TERM'}=$SIG{'INT'}='finish';
$|=1;
printf "%s starting at %s\n", $ID, &wtime(time) if ($opt_v>0);

undef @c;
$NOW=time; $ac=$cc=0;
opendir(CD, ".") || &err("opendir $inn::spool: $!");
while ($cd=readdir(CD), defined($cd)) {
    $cd=~/^time(caf)?-([0-9a-f][0-9a-f])$/i || next;
    $c{hex($2)}=1 unless hex($2)>200;
}
closedir CD;
@classes=sort {$a<=>$b} keys %c;
foreach $c (@classes) {
    &initclass($c);
    $S{"work$;$c"}=$S{"oldest$;$c"}&0xFFFFFF00;
}

$S{"classes"}=join(",", @classes);
$S{"inittime"}=time;
$S{"ID"}=$ID;
printf "Checked %d articles, %d CAFs in %d seconds\n", $ac, $cc, time-$NOW
  if ($ac+$cc>0 && $opt_v>0);

chdir $inn::spool || die "chdir $inn::spool: $!";
while (1) {
    $S{"lastrun"}=$NOW=time;
    printf "%s\n", &wtime($NOW) if ($opt_v>0);
    $nt=0;
    foreach $c (@classes) {
	$t=($NOW-$S{"work$;$c"})*100/$c;
	$nt=$t if ($nt<$t);
    }
    printf "Normal time (class 100): %s\n", &xtime($NOW-$nt)
        if ($opt_v>0);
    if ($nt<$opt_m*24*60*60) {
	printf " capped at minimum %d days\n", $opt_m
	  if ($opt_v>0);
	$nt=$opt_m*24*60*60;
    }
    if ($nt>180*24*60*60) {
	print " capped at maximum 180 days\n"
	  if ($opt_v>0);
	$nt=180*24*60*60;
    }
    $S{"normaltime"}=$nt;
    $decrement=$opt_t*60;
    $pass=$need=0;
    $x="/";
    undef %needk; undef %needi;
    foreach $c (@classes) {
	$Dart{$c}=$Dcaf{$c}=$Dkb{$c}=$Dino{$c}=0;
	$y=sprintf("time-%02x", $c);
	if (-d $y) {
	    @S=stat(_);
	    if ($#S>=0) {
		$dev{$y}=$S[0];
		unless (defined($needk{$S[0]})) {
		    $x.=" $y";
		    $needk{$S[0]}=$needi{$S[0]}=-1;
		}
	    }
	}
	$y=sprintf("timecaf-%02x", $c);
	if (-d $y) {
	    @S=stat(_);
	    if ($#S>=0) {
		$dev{$y}=$S[0];
		unless (defined($needk{$S[0]})) {
		    $x.=" $y";
		    $needk{$S[0]}=$needi{$S[0]}=-1;
		}
	    }
	}
    }
    if (open(D, "inndf $x |")) {
	while (<D>) {
	    @S=split(/\s+/, $_);
	    $needk{$dev{$S[0]}}=$opt_f-$S[1] unless ($S[0] eq "/");
	}
	close D;
    }
    if (open(D, "inndf -i $x |")) {
	while (<D>) {
	    @S=split(/\s+/, $_);
	    $needi{$dev{$S[0]}}=$opt_i-$S[1] unless ($S[0] eq "/");
	}
	close D;
    }
    foreach $c (keys %needk) {
	printf "Device %d needs to free %d kilobytes, %d inodes\n",
	       $c, $needk{$c}<0?0:$needk{$c}, $needi{$c}<0?0:$needi{$c}
	    if ($opt_v>0 && ($needk{$c}>0 || $needi{$c}>0));
	if ($needk{$c}>0 || $needi{$c}>0) {
	    ++$need;
	}
    }
    if ($opt_v>0 && $need<=0) {
	print "  (nothing to do)\n";
	$tt=0;
    } else {
	$error=0;
	while (!$error && $need>0) {
	    if ($S{"normaltime"}-$decrement<$opt_m*24*60*60) {
		print "  Normal time hit minimum\n" if ($opt_v>0);
		last;
	    }
	    $S{"normaltime"}-=$decrement;
	    printf "  normal time (100) becomes %ld\n", $S{"normaltime"}
	    if ($opt_v>2);
	    ++$pass;
	    $Dart=$Dcaf=$Dkb=$Dino=$need=0;
	    foreach $c (keys %needk) {
		if ($needk{$c}>0 || $needi{$c}>0) {
		    ++$need;
		}
	    }
	    if ($need) {
		foreach $c (@classes) {
		    &worktime($c, $NOW-($S{"normaltime"}*$c/100));
		    $Dart+=$dart; $Dcaf+=$dcaf; $Dkb+=$dbb>>10; $Dino+=$dino;
		    $Dart{$c}+=$dart; $Dcaf{$c}+=$dcaf;
		    $Dkb{$c}+=$dbb>>10; $Dino{$c}+=$dino;
		    last if ($error);
		}
	    }
	    if ($Dart+$Dcaf) {
		printf "  pass %d deleted %d arts, %d CAFs, %d kb\n",
		       $pass, $Dart, $Dcaf, $Dkb if ($opt_v>1);
		$decrement-=$decrement>>2 if ($decrement>10*60);
	    } else {
		$decrement+=$decrement>>1 if ($decrement<4*60*60);
	    }
	}
	$Dkb=$Dart=$Dcaf=$Dino=0;
	foreach $c (@classes) {
	    printf "  class %3d: deleted %6d arts %6d CAFs %10d kb\n",
	           $c, $Dart{$c}, $Dcaf{$c}, $Dkb{$c} if ($opt_v>1);
	    $Dkb+=$Dkb{$c}; $Dart+=$Dart{$c}; $Dcaf+=$Dcaf{$c};
	}
	$tt=time-$NOW;
	printf " deleted %d articles, %d CAFs, %d kb in %d seconds\n",
               $Dart, $Dcaf, $Dkb, time-$NOW if ($opt_v>0);
	if ($tt>$opt_t*60) {
	    printf STDERR "Round needed %d seconds, interval is %d\n",
	    $tt, $opt_t*60;
	    $tt=$opt_t*60;
	}
    }
    sleep $opt_t*60-$tt;
}
&finish(0);


sub initclass
{
    my $C=shift;
    if (!$S{"blocksize$;$C$;CAF"}) {
	# Determine filesystem blocksize
	# unfortunately no way in perl to statfs
	my $x=sprintf("%s/timecaf-%02x/test%d", $inn::spool, $C, $$);
	if (open(A, ">$x")) {
	    print A "X" x 4096;
	    close A;
	    @S=stat $x;
	    $#S>=12 || die "stat: $!";
	    if ($S[12]) {
		$S{"blocksize$;$C$;CAF"}=$S[7]/$S[12];
	    } else {
		$S{"blocksize$;$C$;CAF"}=512;
		warn "hack around broken stat blocksize";
	    }
	    unlink $x;
	}
    }
    return if ($S{"oldest$;$C"});
    my $oldest=time;
    $S{"oldest$;$C"}=$oldest;
    my $base=sprintf("%s/time-%02x", $inn::spool, $C);
    my $count=0;
    if (chdir $base) {
	printf "Finding oldest in class %d (%s)\n", $C, $base if ($opt_v>0);
	opendir(D0, ".");
	while ($d1=readdir(D0), defined($d1)) {
	    $d1=~/^[0-9a-f][0-9a-f]$/ || next;
	    chdir $d1;
	    opendir(D1, ".") || next;
	    while ($d2=readdir(D1), defined($d2)) {
		$d2=~/^[0-9a-f][0-9a-f]$/ || next;
		chdir $d2;
		opendir(D2, ".") || next;
		while ($a=readdir(D2), defined($a)) {
		    $a=~/^\./ && next;
		    @S=stat($a);
		    $oldest=$S[9] if ($S[9]<$oldest);
		    ++$count;
		}
		closedir D2;
		chdir "..";
	    }
	    closedir D1;
	    chdir "..";
	}
	closedir D0;
	$ac+=$count;
    }
    $base=sprintf("%s/timecaf-%02x", $inn::spool, $C);
    if (chdir $base) {
	printf "Finding oldest in class %d (%s)\n", $C, $base if ($opt_v>0);
	opendir(D0, ".");
	while ($d1=readdir(D0), defined($d1)) {
	    $d1=~/^[0-9a-f][0-9a-f]$/ || next;
	    chdir $d1;
	    opendir(D1, ".") || next;
	    while ($a=readdir(D1), defined($a)) {
		$a=~/^\./ && next;
		@S=stat($a);
		$oldest=$S[9] if ($S[9]<$oldest);
		++$count;
	    }
	    closedir D1;
	    chdir "..";
	}
	closedir D0;
	$cc+=$count;
    }
    $S{"count$;$C"}=$count;
    $S{"oldest$;$C"}=$oldest;
}

sub worktime
{
    my $C=shift;
    my $goal=shift;
    $goal&=0xFFFFFF00;
    printf "  goal for class %d becomes %s\n", $C, &xtime($goal)
      if ($opt_v>2);
    if ($goal>$NOW-$opt_x) {
	printf "  goal for class %d cut off\n", $C
	  if ($opt_v>1);
	$error=1;
	return;
    }
    $dart=$dcaf=$dbb=$dino=0;
    $hdir=sprintf("time-%02x", $C);
    $cdir=sprintf("timecaf-%02x", $C);
    while (($_=$S{"work$;$C"})<$goal) {
	printf "  running: %08x\n", $_ if ($opt_v>2);
	($aa,$bb,$cc) = (($_>>24)&0xFF, ($_>>16)&0xFF, ($_>>8)&0xFF);
	$dir=sprintf("%s/%02x/%02x", $hdir, $bb, $cc);
	$pat=sprintf("[0-9a-f]{4}-%02x[0-9a-f]{2}", $aa);
	if (opendir(D, $dir)) {
	    while ($_=readdir(D), defined($_)) {
		/^$pat$/ || next;
		$art="$dir/$_";
		@S=stat($art);
		if ($#S>=7) {
		    if ($opt_N) {
			print "   would delete $art" if ($opt_v>2);
		    } else {
			print "   deleting $art" if ($opt_v>2);
			unlink $art;
		    }
		    ++$dart; ++$dino;
		    printf " %d kb\n", $S[7]>>10 if ($opt_v>2);
		    $dbb+=$S[7];
		    $needk{$dev{$hdir}}-=$S[7]>>10;
		    $needi{$dev{$hdir}}--;
		}
	    }
	} else {
	    printf "  (no dir %s)\n", $dir if ($opt_v>2);
	}
	$caf=sprintf("%s/%02x/%02x%02x.CF", $cdir, $bb, $aa, $cc);
	@S=stat($caf);
	if ($#S>=12) {
	    if ($opt_N) {
		print "   would delete $caf" if ($opt_v>2);
	    } else {
		print "   deleting $caf" if ($opt_v>2);
		unlink $caf;
	    }
	    $y=0;
	    if (open(C, $caf)) {
		# try to find how much there is in the CAF
		sysread(C, $_, 16);
		@C=unpack("a4LLL", $_);
		if ($C[0] eq "CRMT") {
		    $y=$C[3]-$C[1];
		    $dart+=$y;
		}
		close C;
	    }
	    ++$dcaf; ++$dino;
	    if ($S[12]) {
		$x=$S[12]*$S{"blocksize$;$C$;CAF"};
	    } else {
		$x=$S[7];
		warn "hack around broken stat blocksize";
	    }
	    printf " %d arts %d kb\n", $y, $x>>10 if ($opt_v>2);
	    $dbb+=$x;
	    $needk{$dev{$cdir}}-=$x>>10;
	    $needi{$dev{$cdir}}--;
	}
	$S{"work$;$C"}+=0x100;
	$S{"oldest$;$C"}=$S{"work$;$C"} unless ($opt_N);
    }
}

sub report
{
    $NOW=time;
    my $cc=$S{"classes"};
    my $nt=$S{"normaltime"};
    unless ($cc && $nt) {
	print "Not initialized.\n";
	return;
    }
    printf "Version: %s (this: %s)\n", $S{"ID"}, $ID;
    printf "Started at: %s\n", &xtime($S{"inittime"}) if ($S{"inittime"});
    printf "Last run: %s\n", &xtime($S{"lastrun"}) if ($S{"lastrun"});
    printf "Classes: %s\n", $cc;
    foreach $c (split(/,/, $cc)) {
	printf "Class %d:\n", $c;
	#printf "  Initial count %d articles\n", $S{"count$;$c"};
	printf "  Oldest article: %s\n", &xtime($S{"oldest$;$c"});
	printf "  Expiring at:    %s\n", &xtime($S{"work$;$c"});
	printf "  Normal time:    %s\n", &xtime($NOW-$nt*$c/100);
	printf "  Filesystem block size (CAF): %d\n", $S{"blocksize$;$c$;CAF"};
    }
}

sub wtime
{
    my $t=shift;
    my @T=localtime($t);
    sprintf("%04d-%02d-%02d %02d:%02d",
	    $T[5]+1900, $T[4]+1, $T[3], $T[2], $T[1]);
}

sub xtime
{
    my $t=shift;
    if ($NOW-$t<0 || $NOW-$t>350*24*60*60) {
	return &wtime($t);
    }
    my @T=localtime($t);
    my @D=gmtime($NOW-$t);
    sprintf("%04d-%02d-%02d %02d:%02d (%dd %dh %dm)",
	    $T[5]+1900, $T[4]+1, $T[3], $T[2], $T[1],
	    $D[7], $D[2], $D[1]);
}

sub err
{
    printf STDERR "%s\n", shift;
    &finish(0);
}

sub finish
{
    untie(%S);
    unlink "$inn::innddir/thdexpire.pid";
    exit 0;
}
#-----------------------------------------------------------------------------