bin/BackupPC

   1 #!/bin/perl
   2 #============================================================= -*-perl-*-
   3 #
   4 # BackupPC: Main program for PC backups.
   5 #
   6 # DESCRIPTION
   7 #
   8 #   BackupPC reads the configuration and status information from
   9 #   $TopDir/conf.  It then runs and manages all the backup activity.
  10 #
  11 #   As specified by $Conf{WakeupSchedule}, BackupPC wakes up periodically
  12 #   to queue backups on all the PCs.  This is a four step process:
  13 #     1) For each host and DHCP address backup requests are queued on the
  14 #        background command queue.
  15 #     2) For each PC, BackupPC_dump is forked.  Several of these may
  16 #        be run in parallel, based on the configuration.
  17 #     3) For each complete, good, backup, BackupPC_link is forked.
  18 #        Only one of these tasks runs at a time.
  19 #     4) In the background BackupPC_trashClean is run to remove any expired
  20 #        backups.  Once each night, BackupPC_nightly is run to complete some
  21 #        additional administrative tasks (pool cleaning etc).
  22 #
  23 #   BackupPC also listens for connections on a unix domain socket and
  24 #   the tcp port $Conf{ServerPort}, which are used by various
  25 #   sub-programs and the CGI script BackupPC_Admin for status reporting
  26 #   and user-initiated backup or backup cancel requests.
  27 #
  28 # AUTHOR
  29 #   Craig Barratt  <cbarratt@users.sourceforge.net>
  30 #
  31 # COPYRIGHT
  32 #   Copyright (C) 2001-2003  Craig Barratt
  33 #
  34 #   This program is free software; you can redistribute it and/or modify
  35 #   it under the terms of the GNU General Public License as published by
  36 #   the Free Software Foundation; either version 2 of the License, or
  37 #   (at your option) any later version.
  38 #
  39 #   This program is distributed in the hope that it will be useful,
  40 #   but WITHOUT ANY WARRANTY; without even the implied warranty of
  41 #   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  42 #   GNU General Public License for more details.
  43 #
  44 #   You should have received a copy of the GNU General Public License
  45 #   along with this program; if not, write to the Free Software
  46 #   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  47 #
  48 #========================================================================
  49 #
  50 # Version 2.1.0beta0, released 20 Mar 2004.
  51 #
  52 # See http://backuppc.sourceforge.net.
  53 #
  54 #========================================================================
  55
  56 use strict;
  57 no  utf8;
  58 use vars qw(%Status %Info $Hosts);
  59 use lib "/usr/local/BackupPC/lib";
  60 use BackupPC::Lib;
  61 use BackupPC::FileZIO;
  62
  63 use File::Path;
  64 use Data::Dumper;
  65 use Getopt::Std;
  66 use Socket;
  67 use Carp;
  68 use Digest::MD5;
  69
  70 ###########################################################################
  71 # Handle command line options
  72 ###########################################################################
  73 my %opts;
  74 if ( !getopts("d", \%opts) || @ARGV != 0 ) {
  75     print("usage: $0 [-d]\n");
  76     exit(1);
  77 }
  78
  79 ###########################################################################
  80 # Initialize major data structures and variables
  81 ###########################################################################
  82
  83 #
  84 # Get an instance of BackupPC::Lib and get some shortcuts.
  85 #
  86 die("BackupPC::Lib->new failed\n") if ( !(my $bpc = BackupPC::Lib->new) );
  87 my $TopDir = $bpc->TopDir();
  88 my $BinDir = $bpc->BinDir();
  89 my %Conf   = $bpc->Conf();
  90
  91 #
  92 # Verify we are running as the correct user
  93 #
  94 if ( $Conf{BackupPCUserVerify}
  95         && $> != (my $uid = (getpwnam($Conf{BackupPCUser}))[2]) ) {
  96     die "Wrong user: my userid is $>, instead of $uid ($Conf{BackupPCUser})\n";
  97 }
  98
  99 #
 100 # %Status maintain status information about each host.
 101 # It is a hash of hashes, whose first index is the host.
 102 #
 103 %Status     = ();
 104
 105 #
 106 # %Info is a hash giving general information about BackupPC status.
 107 #
 108 %Info       = ();
 109
 110 #
 111 # Read old status
 112 #
 113 if ( -f "$TopDir/log/status.pl" && !(my $ret = do "$TopDir/log/status.pl") ) {
 114    die "couldn't parse $TopDir/log/status.pl: $@" if $@;
 115    die "couldn't do $TopDir/log/status.pl: $!"    unless defined $ret;
 116    die "couldn't run $TopDir/log/status.pl";
 117 }
 118
 119 #
 120 # %Jobs maintains information about currently running jobs.
 121 # It is a hash of hashes, whose first index is the host.
 122 #
 123 my %Jobs       = ();
 124
 125 #
 126 # There are three command queues:
 127 #   - @UserQueue is a queue of user initiated backup requests.
 128 #   - @BgQueue is a queue of automatically scheduled backup requests.
 129 #   - @CmdQueue is a queue of administrative jobs, including tasks
 130 #     like BackupPC_link, BackupPC_trashClean, and BackupPC_nightly
 131 # Each queue is an array of hashes.  Each hash stores information
 132 # about the command request.
 133 #
 134 my @UserQueue  = ();
 135 my @CmdQueue   = ();
 136 my @BgQueue    = ();
 137
 138 #
 139 # To quickly lookup if a given host is on a given queue, we keep
 140 # a hash of flags for each queue type.
 141 #
 142 my(%CmdQueueOn, %UserQueueOn, %BgQueueOn);
 143
 144 #
 145 # One or more clients can connect to the server to get status information
 146 # or request/cancel backups etc.  The %Clients hash maintains information
 147 # about each of these socket connections.  The hash key is an incrementing
 148 # number stored in $ClientConnCnt.  Each entry is a hash that contains
 149 # various information about the client connection.
 150 #
 151 my %Clients    = ();
 152 my $ClientConnCnt;
 153
 154 #
 155 # Read file descriptor mask used by select().  Every file descriptor
 156 # on which we expect to read (or accept) has the corresponding bit
 157 # set.
 158 #
 159 my $FDread     = '';
 160
 161 #
 162 # Unix seconds when we next wakeup.  A value of zero forces the scheduler
 163 # to compute the next wakeup time.
 164 #
 165 my $NextWakeup = 0;
 166
 167 #
 168 # Name of signal saved by catch_signal
 169 #
 170 my $SigName = "";
 171
 172 #
 173 # Misc variables
 174 #
 175 my($RunNightlyWhenIdle, $FirstWakeup, $CmdJob, $ServerInetPort);
 176
 177 #
 178 # Complete the rest of the initialization
 179 #
 180 Main_Initialize();
 181
 182 ###########################################################################
 183 # Main loop
 184 ###########################################################################
 185 while ( 1 )
 186 {
 187     #
 188     # Check if we can/should run BackupPC_nightly
 189     #
 190     Main_TryToRun_nightly();
 191
 192     #
 193     # Check if we can run a new command from @CmdQueue.
 194     #
 195     Main_TryToRun_CmdQueue();
 196
 197     #
 198     # Check if we can run a new command from @UserQueue or @BgQueue.
 199     #
 200     Main_TryToRun_Bg_or_User_Queue();
 201
 202     #
 203     # Do a select() to wait for the next interesting thing to happen
 204     # (timeout, signal, someone sends a message, child dies etc).
 205     #
 206     my $fdRead = Main_Select();
 207
 208     #
 209     # Process a signal if we received one.
 210     #
 211     if ( $SigName ) {
 212         Main_Process_Signal();
 213         $fdRead = undef;
 214     }
 215
 216     #
 217     # Check if a timeout has occurred.
 218     #
 219     Main_Check_Timeout();
 220
 221     #
 222     # Check for, and process, any messages (output) from our jobs
 223     #
 224     Main_Check_Job_Messages($fdRead);
 225
 226     #
 227     # Check for, and process, any output from our clients.  Also checks
 228     # for new connections to our SERVER_UNIX and SERVER_INET sockets.
 229     #
 230     Main_Check_Client_Messages($fdRead);
 231 }
 232
 233 ############################################################################
 234 # Main_Initialize()
 235 #
 236 # Main initialization routine.  Called once at statup.
 237 ############################################################################
 238 sub Main_Initialize
 239 {
 240     umask($Conf{UmaskMode});
 241
 242     #
 243     # Check for another running process, check that PASSWD is set and
 244     # verify executables are configured correctly.
 245     #
 246     if ( $Info{pid} ne "" && kill(0, $Info{pid}) ) {
 247         print(STDERR $bpc->timeStamp,
 248                  "Another BackupPC is running (pid $Info{pid}); quitting...\n");
 249         exit(1);
 250     }
 251     foreach my $progName ( qw(SmbClientPath NmbLookupPath PingPath DfPath
 252                               SendmailPath SshPath) ) {
 253         next if ( $Conf{$progName} eq "" || -x $Conf{$progName} );
 254         print(STDERR $bpc->timeStamp,
 255                      "\$Conf{$progName} = '$Conf{$progName}' is not a"
 256                    . " valid executable program\n");
 257         exit(1);
 258     }
 259
 260     if ( $opts{d} ) {
 261         #
 262         # daemonize by forking
 263         #
 264         defined(my $pid = fork) or die "Can't fork: $!";
 265         exit if $pid;   # parent exits
 266     }
 267
 268     #
 269     # Open the LOG file and redirect STDOUT, STDERR etc
 270     #
 271     LogFileOpen();
 272
 273     #
 274     # Read the hosts file (force a read).
 275     #
 276     exit(1) if ( !HostsUpdate(1) );
 277
 278     #
 279     # Clean up %ENV for taint checking
 280     #
 281     delete @ENV{qw(IFS CDPATH ENV BASH_ENV)};
 282     $ENV{PATH} = $Conf{MyPath};
 283
 284     #
 285     # Initialize server sockets
 286     #
 287     ServerSocketInit();
 288
 289     #
 290     # Catch various signals
 291     #
 292     foreach my $sig ( qw(INT BUS SEGV PIPE TERM ALRM HUP) ) {
 293         $SIG{$sig} = \&catch_signal;
 294     }
 295
 296     #
 297     # Report that we started, and update %Info.
 298     #
 299     print(LOG $bpc->timeStamp, "BackupPC started, pid $$\n");
 300     $Info{ConfigModTime} = $bpc->ConfigMTime();
 301     $Info{pid} = $$;
 302     $Info{startTime} = time;
 303     $Info{ConfigLTime} = time;
 304     $Info{Version} = $bpc->{Version};
 305
 306     #
 307     # Update the status left over form the last time BackupPC ran.
 308     # Requeue any pending links.
 309     #
 310     foreach my $host ( sort(keys(%$Hosts)) ) {
 311         if ( $Status{$host}{state} eq "Status_backup_in_progress" ) {
 312             #
 313             # should we restart it?  skip it for now.
 314             #
 315             $Status{$host}{state} = "Status_idle";
 316         } elsif ( $Status{$host}{state} eq "Status_link_pending"
 317                 || $Status{$host}{state} eq "Status_link_running" ) {
 318             QueueLink($host);
 319         } else {
 320             $Status{$host}{state} = "Status_idle";
 321         }
 322         $Status{$host}{activeJob} = 0;
 323     }
 324     foreach my $host ( sort(keys(%Status)) ) {
 325         next if ( defined($Hosts->{$host}) );
 326         delete($Status{$host});
 327     }
 328
 329     #
 330     # Write out our initial status and save our PID
 331     #
 332     StatusWrite();
 333     if ( open(PID, ">", "$TopDir/log/BackupPC.pid") ) {
 334         print(PID $$);
 335         close(PID);
 336     }
 337
 338     #
 339     # For unknown reasons there is a very infrequent error about not
 340     # being able to coerce GLOBs inside the XS Data::Dumper.  I've
 341     # only seen this on a particular platform and perl version.
 342     # For now the workaround appears to be use the perl version of
 343     # XS Data::Dumper.
 344     #
 345     $Data::Dumper::Useqq = 1;
 346 }
 347
 348 ############################################################################
 349 # Main_TryToRun_nightly()
 350 #
 351 # Checks to see if we can/should run BackupPC_nightly or
 352 # BackupPC_trashClean.  If so we push the appropriate command onto
 353 # @CmdQueue.
 354 ############################################################################
 355 sub Main_TryToRun_nightly
 356 {
 357     #
 358     # Check if we should run BackupPC_nightly or BackupPC_trashClean.
 359     # BackupPC_nightly is run when the current job queue is empty.
 360     # BackupPC_trashClean is run in the background always.
 361     #
 362     my $trashCleanRunning = defined($Jobs{$bpc->trashJob}) ? 1 : 0;
 363     if ( !$trashCleanRunning && !$CmdQueueOn{$bpc->trashJob} ) {
 364         #
 365         # This should only happen once at startup, but just in case this
 366         # code will re-start BackupPC_trashClean if it quits
 367         #
 368         unshift(@CmdQueue, {
 369                 host    => $bpc->trashJob,
 370                 user    => "BackupPC",
 371                 reqTime => time,
 372                 cmd     => ["$BinDir/BackupPC_trashClean"],
 373             });
 374         $CmdQueueOn{$bpc->trashJob} = 1;
 375     }
 376     if ( keys(%Jobs) == $trashCleanRunning && $RunNightlyWhenIdle == 1 ) {
 377         push(@CmdQueue, {
 378                 host    => $bpc->adminJob,
 379                 user    => "BackupPC",
 380                 reqTime => time,
 381                 cmd     => ["$BinDir/BackupPC_nightly"],
 382             });
 383         $CmdQueueOn{$bpc->adminJob} = 1;
 384         $RunNightlyWhenIdle = 2;
 385     }
 386 }
 387
 388 ############################################################################
 389 # Main_TryToRun_CmdQueue()
 390 #
 391 # Decide if we can run a new command from the @CmdQueue.
 392 # We only run one of these at a time.  The @CmdQueue is
 393 # used to run BackupPC_link (for the corresponding host),
 394 # BackupPC_trashClean, and BackupPC_nightly using a fake
 395 # host name of $bpc->adminJob.
 396 ############################################################################
 397 sub Main_TryToRun_CmdQueue
 398 {
 399     my($req, $host);
 400     if ( $CmdJob eq "" && @CmdQueue > 0 && $RunNightlyWhenIdle != 1 ) {
 401         local(*FH);
 402         $req = pop(@CmdQueue);
 403
 404         $host = $req->{host};
 405         if ( defined($Jobs{$host}) ) {
 406             print(LOG $bpc->timeStamp,
 407                        "Botch on admin job for $host: already in use!!\n");
 408             #
 409             # This could happen during normal opertion: a user could
 410             # request a backup while a BackupPC_link is queued from
 411             # a previous backup.  But it is unlikely.  Just put this
 412             # request back on the end of the queue.
 413             #
 414             unshift(@CmdQueue, $req);
 415             return;
 416         }
 417         $CmdQueueOn{$host} = 0;
 418         my $cmd  = $req->{cmd};
 419         my $pid = open(FH, "-|");
 420         if ( !defined($pid) ) {
 421             print(LOG $bpc->timeStamp,
 422                        "can't fork for $host, request by $req->{user}\n");
 423             close(FH);
 424             next;
 425         }
 426         if ( !$pid ) {
 427             setpgrp 0,0;
 428             exec(@$cmd);
 429             print(LOG $bpc->timeStamp, "can't exec @$cmd for $host\n");
 430             exit(0);
 431         }
 432         $Jobs{$host}{pid}       = $pid;
 433         $Jobs{$host}{fh}        = *FH;
 434         $Jobs{$host}{fn}        = fileno(FH);
 435         vec($FDread, $Jobs{$host}{fn}, 1) = 1;
 436         $Jobs{$host}{startTime} = time;
 437         $Jobs{$host}{reqTime}   = $req->{reqTime};
 438         $cmd                    = join(" ", @$cmd);
 439         $Jobs{$host}{cmd}       = $cmd;
 440         $Jobs{$host}{type}      = $Status{$host}{type};
 441         $Status{$host}{state}   = "Status_link_running";
 442         $Status{$host}{activeJob} = 1;
 443         $Status{$host}{endTime} = time;
 444         $CmdJob = $host if ( $host ne $bpc->trashJob );
 445         $cmd =~ s/$BinDir\///g;
 446         print(LOG $bpc->timeStamp, "Running $cmd (pid=$pid)\n");
 447     }
 448 }
 449
 450 ############################################################################
 451 # Main_TryToRun_Bg_or_User_Queue()
 452 #
 453 # Decide if we can run any new backup requests from @BgQueue
 454 # or @UserQueue.  Several of these can be run at the same time
 455 # based on %Conf settings.  Jobs from @UserQueue take priority,
 456 # and at total of $Conf{MaxBackups} + $Conf{MaxUserBackups}
 457 # simultaneous jobs can run from @UserQueue.  After @UserQueue
 458 # is exhausted, up to $Conf{MaxBackups} simultaneous jobs can
 459 # run from @BgQueue.
 460 ############################################################################
 461 sub Main_TryToRun_Bg_or_User_Queue
 462 {
 463     my($req, $host);
 464     while ( $RunNightlyWhenIdle == 0 ) {
 465         local(*FH);
 466         my(@args, @deferUserQueue, @deferBgQueue, $progName, $type);
 467         my $nJobs = keys(%Jobs);
 468         #
 469         # CmdJob and trashClean don't count towards MaxBackups / MaxUserBackups
 470         #
 471         $nJobs-- if ( $CmdJob ne "" );
 472         $nJobs-- if ( defined($Jobs{$bpc->trashJob} ) );
 473         if ( $nJobs < $Conf{MaxBackups} + $Conf{MaxUserBackups}
 474                         && @UserQueue > 0 ) {
 475             $req = pop(@UserQueue);
 476             if ( defined($Jobs{$req->{host}}) ) {
 477                 push(@deferUserQueue, $req);
 478                 next;
 479             }
 480             push(@args, $req->{doFull} ? "-f" : "-i")
 481                             if (( !$req->{restore} ) && ( !$req->{archive} ));
 482             $UserQueueOn{$req->{host}} = 0;
 483         } elsif ( $nJobs < $Conf{MaxBackups}
 484                         && (@CmdQueue + $nJobs)
 485                                 <= $Conf{MaxBackups} + $Conf{MaxPendingCmds}
 486                         && @BgQueue > 0 ) {
 487             my $du;
 488             if ( time - $Info{DUlastValueTime} >= 60 ) {
 489                 #
 490                 # Update our notion of disk usage no more than
 491                 # once every minute
 492                 #
 493                 $du = $bpc->CheckFileSystemUsage($TopDir);
 494                 $Info{DUlastValue}     = $du;
 495                 $Info{DUlastValueTime} = time;
 496             } else {
 497                 #
 498                 # if we recently checked it then just use the old value
 499                 #
 500                 $du = $Info{DUlastValue};
 501             }
 502             if ( $Info{DUDailyMaxReset} ) {
 503                 $Info{DUDailyMaxStartTime} = time;
 504                 $Info{DUDailyMaxReset}     = 0;
 505                 $Info{DUDailyMax}          = 0;
 506             }
 507             if ( $du > $Info{DUDailyMax} ) {
 508                 $Info{DUDailyMax}     = $du;
 509                 $Info{DUDailyMaxTime} = time;
 510             }
 511             if ( $du > $Conf{DfMaxUsagePct} ) {
 512                 my $nSkip = @BgQueue + @deferBgQueue;
 513                 print(LOG $bpc->timeStamp,
 514                            "Disk too full ($du%%); skipping $nSkip hosts\n");
 515                 $Info{DUDailySkipHostCnt} += $nSkip;
 516                 @BgQueue = ();
 517                 @deferBgQueue = ();
 518                 %BgQueueOn = ();
 519                 next;
 520             }
 521             $req = pop(@BgQueue);
 522             if ( defined($Jobs{$req->{host}}) ) {
 523                 push(@deferBgQueue, $req);
 524                 next;
 525             }
 526             $BgQueueOn{$req->{host}} = 0;
 527         } else {
 528             while ( @deferBgQueue ) {
 529                 push(@BgQueue, pop(@deferBgQueue));
 530             }
 531             while ( @deferUserQueue ) {
 532                 push(@UserQueue, pop(@deferUserQueue));
 533             }
 534             last;
 535         }
 536         $host = $req->{host};
 537         my $user = $req->{user};
 538         if ( $req->{restore} ) {
 539             $progName = "BackupPC_restore";
 540             $type     = "restore";
 541             push(@args, $req->{hostIP}, $req->{host}, $req->{reqFileName});
 542         } elsif ( $req->{archive} ) {
 543             $progName = "BackupPC_archive";
 544             $type     = "archive";
 545             push(@args, $req->{user}, $req->{host}, $req->{reqFileName});
 546         } else {
 547             $progName = "BackupPC_dump";
 548             $type     = "backup";
 549             push(@args, "-d") if ( $req->{dhcp} );
 550             push(@args, "-e") if ( $req->{dumpExpire} );
 551             push(@args, $host);
 552         }
 553         my $pid = open(FH, "-|");
 554         if ( !defined($pid) ) {
 555             print(LOG $bpc->timeStamp,
 556                    "can't fork to run $progName for $host, request by $user\n");
 557             close(FH);
 558             next;
 559         }
 560         if ( !$pid ) {
 561             setpgrp 0,0;
 562             exec("$BinDir/$progName", @args);
 563             print(LOG $bpc->timeStamp, "can't exec $progName for $host\n");
 564             exit(0);
 565         }
 566         $Jobs{$host}{pid}        = $pid;
 567         $Jobs{$host}{fh}         = *FH;
 568         $Jobs{$host}{fn}         = fileno(FH);
 569         $Jobs{$host}{dhcp}       = $req->{dhcp};
 570         vec($FDread, $Jobs{$host}{fn}, 1) = 1;
 571         $Jobs{$host}{startTime}  = time;
 572         $Jobs{$host}{reqTime}    = $req->{reqTime};
 573         $Jobs{$host}{userReq}    = $req->{userReq};
 574         $Jobs{$host}{cmd}        = join(" ", $progName, @args);
 575         $Jobs{$host}{user}       = $user;
 576         $Jobs{$host}{type}       = $type;
 577         $Status{$host}{userReq}  = $req->{userReq}
 578                                         if ( defined($Hosts->{$host}) );
 579         if ( !$req->{dhcp} ) {
 580             $Status{$host}{state}     = "Status_".$type."_starting";
 581             $Status{$host}{activeJob} = 1;
 582             $Status{$host}{startTime} = time;
 583             $Status{$host}{endTime}   = "";
 584         }
 585     }
 586 }
 587
 588 ############################################################################
 589 # Main_Select()
 590 #
 591 # If necessary, figure out when to next wakeup based on $Conf{WakeupSchedule},
 592 # and then do a select() to wait for the next thing to happen
 593 # (timeout, signal, someone sends a message, child dies etc).
 594 ############################################################################
 595 sub Main_Select
 596 {
 597     if ( $NextWakeup <= 0 ) {
 598         #
 599         # Figure out when to next wakeup based on $Conf{WakeupSchedule}.
 600         #
 601         my($sec,$min,$hour,$mday,$mon,$year,$wday,$yday,$isdst)
 602                                                 = localtime(time);
 603         my($currHours) = $hour + $min / 60 + $sec / 3600;
 604         if ( $bpc->ConfigMTime() != $Info{ConfigModTime} ) {
 605             ServerReload("Re-read config file because mtime changed");
 606         }
 607         my $delta = -1;
 608         foreach my $t ( @{$Conf{WakeupSchedule} || [0..23]} ) {
 609             next if ( $t < 0 || $t > 24 );
 610             my $tomorrow = $t + 24;
 611             if ( $delta < 0
 612                 || ($tomorrow - $currHours > 0
 613                                 && $delta > $tomorrow - $currHours) ) {
 614                 $delta = $tomorrow - $currHours;
 615                 $FirstWakeup = $t == $Conf{WakeupSchedule}[0];
 616             }
 617             if ( $delta < 0
 618                     || ($t - $currHours > 0 && $delta > $t - $currHours) ) {
 619                 $delta = $t - $currHours;
 620                 $FirstWakeup = $t == $Conf{WakeupSchedule}[0];
 621             }
 622         }
 623         $NextWakeup = time + $delta * 3600;
 624         $Info{nextWakeup} = $NextWakeup;
 625         print(LOG $bpc->timeStamp, "Next wakeup is ",
 626                   $bpc->timeStamp($NextWakeup, 1), "\n");
 627     }
 628     #
 629     # Call select(), waiting until either a signal, a timeout,
 630     # any output from our jobs, or any messages from clients
 631     # connected via tcp.
 632     # select() is where we (hopefully) spend most of our time blocked...
 633     #
 634     my $timeout = $NextWakeup - time;
 635     $timeout = 1 if ( $timeout <= 0 );
 636     my $ein = $FDread;
 637     select(my $rout = $FDread, undef, $ein, $timeout);
 638
 639     return $rout;
 640 }
 641
 642 ############################################################################
 643 # Main_Process_Signal()
 644 #
 645 # Signal handler.
 646 ############################################################################
 647 sub Main_Process_Signal
 648 {
 649     #
 650     # Process signals
 651     #
 652     if ( $SigName eq "HUP" ) {
 653         ServerReload("Re-read config file because of a SIG_HUP");
 654     } elsif ( $SigName ) {
 655         ServerShutdown("Got signal $SigName... cleaning up");
 656     }
 657     $SigName = "";
 658 }
 659
 660 ############################################################################
 661 # Main_Check_Timeout()
 662 #
 663 # Check if a timeout has occured, and if so, queue all the PCs for backups.
 664 # Also does log file aging on the first timeout after midnight.
 665 ############################################################################
 666 sub Main_Check_Timeout
 667 {
 668     #
 669     # Process timeouts
 670     #
 671     return if ( time < $NextWakeup || $NextWakeup <= 0 );
 672     $NextWakeup = 0;
 673     if ( $FirstWakeup ) {
 674         #
 675         # This is the first wakeup after midnight.  Do log file aging
 676         # and various house keeping.
 677         #
 678         $FirstWakeup = 0;
 679         printf(LOG "%s24hr disk usage: %d%% max, %d%% recent,"
 680                    . " %d skipped hosts\n",
 681                    $bpc->timeStamp, $Info{DUDailyMax}, $Info{DUlastValue},
 682                    $Info{DUDailySkipHostCnt});
 683         $Info{DUDailyMaxReset}        = 1;
 684         $Info{DUDailyMaxPrev}         = $Info{DUDailyMax};
 685         $Info{DUDailySkipHostCntPrev} = $Info{DUDailySkipHostCnt};
 686         $Info{DUDailySkipHostCnt}     = 0;
 687         my $lastLog = $Conf{MaxOldLogFiles} - 1;
 688         if ( -f "$TopDir/log/LOG.$lastLog" ) {
 689             print(LOG $bpc->timeStamp,
 690                        "Removing $TopDir/log/LOG.$lastLog\n");
 691             unlink("$TopDir/log/LOG.$lastLog");
 692         }
 693         if ( -f "$TopDir/log/LOG.$lastLog.z" ) {
 694             print(LOG $bpc->timeStamp,
 695                        "Removing $TopDir/log/LOG.$lastLog.z\n");
 696             unlink("$TopDir/log/LOG.$lastLog.z");
 697         }
 698         print(LOG $bpc->timeStamp, "Aging LOG files, LOG -> LOG.0 -> "
 699                    . "LOG.1 -> ... -> LOG.$lastLog\n");
 700         close(LOG);
 701         for ( my $i = $lastLog - 1 ; $i >= 0 ; $i-- ) {
 702             my $j = $i + 1;
 703             rename("$TopDir/log/LOG.$i", "$TopDir/log/LOG.$j")
 704                             if ( -f "$TopDir/log/LOG.$i" );
 705             rename("$TopDir/log/LOG.$i.z", "$TopDir/log/LOG.$j.z")
 706                             if ( -f "$TopDir/log/LOG.$i.z" );
 707         }
 708         #
 709         # Compress the log file LOG -> LOG.0.z (if enabled).
 710         # Otherwise, just rename LOG -> LOG.0.
 711         #
 712         BackupPC::FileZIO->compressCopy("$TopDir/log/LOG",
 713                                         "$TopDir/log/LOG.0.z",
 714                                         "$TopDir/log/LOG.0",
 715                                         $Conf{CompressLevel}, 1);
 716         LogFileOpen();
 717         #
 718         # Remember to run nightly script after current jobs are done
 719         #
 720         $RunNightlyWhenIdle = 1;
 721     }
 722     #
 723     # Write out the current status and then queue all the PCs
 724     #
 725     HostsUpdate(0);
 726     StatusWrite();
 727     %BgQueueOn = ()   if ( @BgQueue == 0 );
 728     %UserQueueOn = () if ( @UserQueue == 0 );
 729     %CmdQueueOn = ()  if ( @CmdQueue == 0 );
 730     QueueAllPCs();
 731 }
 732
 733 ############################################################################
 734 # Main_Check_Job_Messages($fdRead)
 735 #
 736 # Check if select() says we have bytes waiting from any of our jobs.
 737 # Handle each of the messages when complete (newline terminated).
 738 ############################################################################
 739 sub Main_Check_Job_Messages
 740 {
 741     my($fdRead) = @_;
 742     foreach my $host ( keys(%Jobs) ) {
 743         next if ( !vec($fdRead, $Jobs{$host}{fn}, 1) );
 744         my $mesg;
 745         #
 746         # do a last check to make sure there is something to read so
 747         # we are absolutely sure we won't block.
 748         #
 749         vec(my $readMask, $Jobs{$host}{fn}, 1) = 1;
 750         if ( !select($readMask, undef, undef, 0.0) ) {
 751             print(LOG $bpc->timeStamp, "Botch in Main_Check_Job_Messages:"
 752                         . " nothing to read from $host.  Debug dump:\n");
 753             my($dump) = Data::Dumper->new(
 754                          [  \%Clients, \%Jobs, \$FDread, \$fdRead],
 755                          [qw(*Clients,  *Jobs   *FDread,  *fdRead)]);
 756             $dump->Indent(1);
 757             print(LOG $dump->Dump);
 758             next;
 759         }
 760         my $nbytes = sysread($Jobs{$host}{fh}, $mesg, 1024);
 761         $Jobs{$host}{mesg} .= $mesg if ( $nbytes > 0 );
 762         #
 763         # Process any complete lines of output from this jobs.
 764         # Any output to STDOUT or STDERR from the children is processed here.
 765         #
 766         while ( $Jobs{$host}{mesg} =~ /(.*?)[\n\r]+(.*)/s ) {
 767             $mesg = $1;
 768             $Jobs{$host}{mesg} = $2;
 769             if ( $Jobs{$host}{dhcp} ) {
 770                 if ( $mesg =~ /^DHCP (\S+) (\S+)/ ) {
 771                     my $newHost = $bpc->uriUnesc($2);
 772                     if ( defined($Jobs{$newHost}) ) {
 773                         print(LOG $bpc->timeStamp,
 774                                 "Backup on $newHost is already running\n");
 775                         kill($bpc->sigName2num("INT"), $Jobs{$host}{pid});
 776                         $nbytes = 0;
 777                         last;
 778                     }
 779                     $Jobs{$host}{dhcpHostIP} = $host;
 780                     $Status{$newHost}{dhcpHostIP} = $host;
 781                     $Jobs{$newHost} = $Jobs{$host};
 782                     delete($Jobs{$host});
 783                     $host = $newHost;
 784                     $Status{$host}{state}      = "Status_backup_starting";
 785                     $Status{$host}{activeJob}  = 1;
 786                     $Status{$host}{startTime}  = $Jobs{$host}{startTime};
 787                     $Status{$host}{endTime}    = "";
 788                     $Jobs{$host}{dhcp}         = 0;
 789                 } else {
 790                     print(LOG $bpc->timeStamp, "dhcp $host: $mesg\n");
 791                 }
 792             } elsif ( $mesg =~ /^started (.*) dump, share=(.*)/ ) {
 793                 $Jobs{$host}{type}      = $1;
 794                 $Jobs{$host}{shareName} = $2;
 795                 print(LOG $bpc->timeStamp,
 796                           "Started $1 backup on $host (pid=$Jobs{$host}{pid}",
 797                           $Jobs{$host}{dhcpHostIP}
 798                                 ? ", dhcp=$Jobs{$host}{dhcpHostIP}" : "",
 799                           ", share=$Jobs{$host}{shareName})\n");
 800                 $Status{$host}{state}     = "Status_backup_in_progress";
 801                 $Status{$host}{reason}    = "";
 802                 $Status{$host}{type}      = $1;
 803                 $Status{$host}{startTime} = time;
 804                 $Status{$host}{deadCnt}   = 0;
 805                 $Status{$host}{aliveCnt}++;
 806                 $Status{$host}{dhcpCheckCnt}--
 807                                 if ( $Status{$host}{dhcpCheckCnt} > 0 );
 808             } elsif ( $mesg =~ /^xferPids (.*)/ ) {
 809                 $Jobs{$host}{xferPid} = $1;
 810             } elsif ( $mesg =~ /^started_restore/ ) {
 811                 $Jobs{$host}{type}    = "restore";
 812                 print(LOG $bpc->timeStamp,
 813                           "Started restore on $host"
 814                           . " (pid=$Jobs{$host}{pid})\n");
 815                 $Status{$host}{state}     = "Status_restore_in_progress";
 816                 $Status{$host}{reason}    = "";
 817                 $Status{$host}{type}      = "restore";
 818                 $Status{$host}{startTime} = time;
 819                 $Status{$host}{deadCnt}   = 0;
 820                 $Status{$host}{aliveCnt}++;
 821             } elsif ( $mesg =~ /^started_archive/ ) {
 822                 $Jobs{$host}{type}    = "archive";
 823                 print(LOG $bpc->timeStamp,
 824                           "Started archive on $host"
 825                           . " (pid=$Jobs{$host}{pid})\n");
 826                 $Status{$host}{state}     = "Status_archive_in_progress";
 827                 $Status{$host}{reason}    = "";
 828                 $Status{$host}{type}      = "archive";
 829                 $Status{$host}{startTime} = time;
 830                 $Status{$host}{deadCnt}   = 0;
 831                 $Status{$host}{aliveCnt}++;
 832             } elsif ( $mesg =~ /^(full|incr) backup complete/ ) {
 833                 print(LOG $bpc->timeStamp, "Finished $1 backup on $host\n");
 834                 $Status{$host}{reason}    = "Reason_backup_done";
 835                 delete($Status{$host}{error});
 836                 delete($Status{$host}{errorTime});
 837                 $Status{$host}{endTime}   = time;
 838             } elsif ( $mesg =~ /^backups disabled/ ) {
 839                 print(LOG $bpc->timeStamp,
 840                             "Ignoring old backup error on $host\n");
 841                 $Status{$host}{reason}    = "Reason_backup_done";
 842                 delete($Status{$host}{error});
 843                 delete($Status{$host}{errorTime});
 844                 $Status{$host}{endTime}   = time;
 845             } elsif ( $mesg =~ /^restore complete/ ) {
 846                 print(LOG $bpc->timeStamp, "Finished restore on $host\n");
 847                 $Status{$host}{reason}    = "Reason_restore_done";
 848                 delete($Status{$host}{error});
 849                 delete($Status{$host}{errorTime});
 850                 $Status{$host}{endTime}   = time;
 851             } elsif ( $mesg =~ /^archive complete/ ) {
 852                 print(LOG $bpc->timeStamp, "Finished archive on $host\n");
 853                 $Status{$host}{reason}    = "Reason_archive_done";
 854                 delete($Status{$host}{error});
 855                 delete($Status{$host}{errorTime});
 856                 $Status{$host}{endTime}   = time;
 857             } elsif ( $mesg =~ /^nothing to do/ ) {
 858                 if ( $Status{$host}{reason} ne "Reason_backup_failed"
 859                         && $Status{$host}{reason} ne "Reason_restore_failed" ) {
 860                     $Status{$host}{state}     = "Status_idle";
 861                     $Status{$host}{reason}    = "Reason_nothing_to_do";
 862                     $Status{$host}{startTime} = time;
 863                 }
 864                 $Status{$host}{dhcpCheckCnt}--
 865                                 if ( $Status{$host}{dhcpCheckCnt} > 0 );
 866             } elsif ( $mesg =~ /^no ping response/
 867                             || $mesg =~ /^ping too slow/
 868                             || $mesg =~ /^host not found/ ) {
 869                 $Status{$host}{state}     = "Status_idle";
 870                 if ( $Status{$host}{userReq}
 871                         || $Status{$host}{reason} ne "Reason_backup_failed"
 872                         || $Status{$host}{error} =~ /^aborted by user/ ) {
 873                     $Status{$host}{reason}    = "Reason_no_ping";
 874                     $Status{$host}{error}     = $mesg;
 875                     $Status{$host}{startTime} = time;
 876                 }
 877                 $Status{$host}{deadCnt}++;
 878                 if ( $Status{$host}{deadCnt} >= $Conf{BlackoutBadPingLimit} ) {
 879                     $Status{$host}{aliveCnt} = 0;
 880                 }
 881             } elsif ( $mesg =~ /^dump failed: (.*)/ ) {
 882                 $Status{$host}{state}     = "Status_idle";
 883                 $Status{$host}{error}     = $1;
 884                 $Status{$host}{errorTime} = time;
 885                 $Status{$host}{endTime}   = time;
 886                 if ( $Status{$host}{reason}
 887                         eq "Reason_backup_canceled_by_user" ) {
 888                     print(LOG $bpc->timeStamp,
 889                             "Backup canceled on $host ($1)\n");
 890                 } else {
 891                     $Status{$host}{reason} = "Reason_backup_failed";
 892                     print(LOG $bpc->timeStamp,
 893                             "Backup failed on $host ($1)\n");
 894                 }
 895             } elsif ( $mesg =~ /^restore failed: (.*)/ ) {
 896                 $Status{$host}{state}     = "Status_idle";
 897                 $Status{$host}{error}     = $1;
 898                 $Status{$host}{errorTime} = time;
 899                 $Status{$host}{endTime}   = time;
 900                 if ( $Status{$host}{reason}
 901                          eq "Reason_restore_canceled_by_user" ) {
 902                     print(LOG $bpc->timeStamp,
 903                             "Restore canceled on $host ($1)\n");
 904                 } else {
 905                     $Status{$host}{reason} = "Reason_restore_failed";
 906                     print(LOG $bpc->timeStamp,
 907                             "Restore failed on $host ($1)\n");
 908                 }
 909             } elsif ( $mesg =~ /^archive failed: (.*)/ ) {
 910                 $Status{$host}{state}     = "Status_idle";
 911                 $Status{$host}{error}     = $1;
 912                 $Status{$host}{errorTime} = time;
 913                 $Status{$host}{endTime}   = time;
 914                 if ( $Status{$host}{reason}
 915                          eq "Reason_archive_canceled_by_user" ) {
 916                     print(LOG $bpc->timeStamp,
 917                             "Archive canceled on $host ($1)\n");
 918                 } else {
 919                     $Status{$host}{reason} = "Reason_archive_failed";
 920                     print(LOG $bpc->timeStamp,
 921                             "Archive failed on $host ($1)\n");
 922                 }
 923             } elsif ( $mesg =~ /^log\s+(.*)/ ) {
 924                 print(LOG $bpc->timeStamp, "$1\n");
 925             } elsif ( $mesg =~ /^BackupPC_stats = (.*)/ ) {
 926                 my @f = split(/,/, $1);
 927                 $Info{"$f[0]FileCnt"}       = $f[1];
 928                 $Info{"$f[0]DirCnt"}        = $f[2];
 929                 $Info{"$f[0]Kb"}            = $f[3];
 930                 $Info{"$f[0]Kb2"}           = $f[4];
 931                 $Info{"$f[0]KbRm"}          = $f[5];
 932                 $Info{"$f[0]FileCntRm"}     = $f[6];
 933                 $Info{"$f[0]FileCntRep"}    = $f[7];
 934                 $Info{"$f[0]FileRepMax"}    = $f[8];
 935                 $Info{"$f[0]FileCntRename"} = $f[9];
 936                 $Info{"$f[0]FileLinkMax"}   = $f[10];
 937                 $Info{"$f[0]Time"}          = time;
 938                 printf(LOG "%s%s nightly clean removed %d files of"
 939                            . " size %.2fGB\n",
 940                              $bpc->timeStamp, ucfirst($f[0]),
 941                              $Info{"$f[0]FileCntRm"},
 942                              $Info{"$f[0]KbRm"} / (1000 * 1024));
 943                 printf(LOG "%s%s is %.2fGB, %d files (%d repeated, "
 944                            . "%d max chain, %d max links), %d directories\n",
 945                              $bpc->timeStamp, ucfirst($f[0]),
 946                              $Info{"$f[0]Kb"} / (1000 * 1024),
 947                              $Info{"$f[0]FileCnt"}, $Info{"$f[0]FileCntRep"},
 948                              $Info{"$f[0]FileRepMax"},
 949                              $Info{"$f[0]FileLinkMax"}, $Info{"$f[0]DirCnt"});
 950             } elsif ( $mesg =~ /^BackupPC_nightly lock_off/ ) {
 951                 $RunNightlyWhenIdle = 0;
 952             } elsif ( $mesg =~ /^processState\s+(.+)/ ) {
 953                 $Jobs{$host}{processState} = $1;
 954             } elsif ( $mesg =~ /^link\s+(.+)/ ) {
 955                 my($h) = $1;
 956                 $Status{$h}{needLink} = 1;
 957             } else {
 958                 print(LOG $bpc->timeStamp, "$host: $mesg\n");
 959             }
 960         }
 961         #
 962         # shut down the client connection if we read EOF
 963         #
 964         if ( $nbytes <= 0 ) {
 965             close($Jobs{$host}{fh});
 966             vec($FDread, $Jobs{$host}{fn}, 1) = 0;
 967             if ( $CmdJob eq $host ) {
 968                 my $cmd = $Jobs{$host}{cmd};
 969                 $cmd =~ s/$BinDir\///g;
 970                 print(LOG $bpc->timeStamp, "Finished $host ($cmd)\n");
 971                 $Status{$host}{state}    = "Status_idle";
 972                 $Status{$host}{endTime}  = time;
 973                 $CmdJob = "";
 974                 $RunNightlyWhenIdle = 0 if ( $cmd eq "BackupPC_nightly"
 975                                             && $RunNightlyWhenIdle );
 976             } else {
 977                 #
 978                 # Queue BackupPC_link to complete the backup
 979                 # processing for this host.
 980                 #
 981                 if ( defined($Status{$host})
 982                             && ($Status{$host}{reason} eq "Reason_backup_done"
 983                                 || $Status{$host}{needLink}) ) {
 984                     QueueLink($host);
 985                 } elsif ( defined($Status{$host}) ) {
 986                     $Status{$host}{state} = "Status_idle";
 987                 }
 988             }
 989             delete($Jobs{$host});
 990             $Status{$host}{activeJob} = 0 if ( defined($Status{$host}) );
 991         }
 992     }
 993     #
 994     # When we are idle (empty Jobs, CmdQueue, BgQueue, UserQueue) we
 995     # do a pass over %Status updating the deadCnt and aliveCnt for
 996     # DHCP hosts.  The reason we need to do this later is we can't
 997     # be sure whether a DHCP host is alive or dead until we have passed
 998     # over all the DHCP pool.
 999     #
1000     return if ( @CmdQueue || @BgQueue || @UserQueue || keys(%Jobs) > 1 );
1001     foreach my $host ( keys(%Status) ) {
1002         next if ( $Status{$host}{dhcpCheckCnt} <= 0 );
1003         $Status{$host}{deadCnt} += $Status{$host}{dhcpCheckCnt};
1004         $Status{$host}{dhcpCheckCnt} = 0;
1005         if ( $Status{$host}{deadCnt} >= $Conf{BlackoutBadPingLimit} ) {
1006             $Status{$host}{aliveCnt} = 0;
1007         }
1008     }
1009 }
1010
1011 ############################################################################
1012 # Main_Check_Client_Messages($fdRead)
1013 #
1014 # Check for, and process, any output from our clients.  Also checks
1015 # for new connections to our SERVER_UNIX and SERVER_INET sockets.
1016 ############################################################################
1017 sub Main_Check_Client_Messages
1018 {
1019     my($fdRead) = @_;
1020     foreach my $client ( keys(%Clients) ) {
1021         next if ( !vec($fdRead, $Clients{$client}{fn}, 1) );
1022         my($mesg, $host);
1023         #
1024         # do a last check to make sure there is something to read so
1025         # we are absolutely sure we won't block.
1026         #
1027         vec(my $readMask, $Clients{$client}{fn}, 1) = 1;
1028         if ( !select($readMask, undef, undef, 0.0) ) {
1029             print(LOG $bpc->timeStamp, "Botch in Main_Check_Client_Messages:"
1030                         . " nothing to read from $client.  Debug dump:\n");
1031             my($dump) = Data::Dumper->new(
1032                          [  \%Clients, \%Jobs, \$FDread, \$fdRead],
1033                          [qw(*Clients,  *Jobs   *FDread,  *fdRead)]);
1034             $dump->Indent(1);
1035             print(LOG $dump->Dump);
1036             next;
1037         }
1038         my $nbytes = sysread($Clients{$client}{fh}, $mesg, 1024);
1039         $Clients{$client}{mesg} .= $mesg if ( $nbytes > 0 );
1040         #
1041         # Process any complete lines received from this client.
1042         #
1043         while ( $Clients{$client}{mesg} =~ /(.*?)[\n\r]+(.*)/s ) {
1044             my($reply);
1045             my $cmd = $1;
1046             $Clients{$client}{mesg} = $2;
1047             #
1048             # Authenticate the message by checking the MD5 digest
1049             #
1050             my $md5 = Digest::MD5->new;
1051             if ( $cmd !~ /^(.{22}) (.*)/
1052                 || ($md5->add($Clients{$client}{seed}
1053                             . $Clients{$client}{mesgCnt}
1054                             . $Conf{ServerMesgSecret} . $2),
1055                      $md5->b64digest ne $1) ) {
1056                 print(LOG $bpc->timeStamp, "Corrupted message '$cmd' from"
1057                             . " client '$Clients{$client}{clientName}':"
1058                             . " shutting down client connection\n");
1059                 $nbytes = 0;
1060                 last;
1061             }
1062             $Clients{$client}{mesgCnt}++;
1063             $cmd = $2;
1064             if ( $cmd =~ /^stop (\S+)\s+(\S+)\s+(\S*)/ ) {
1065                 $host = $1;
1066                 my $user = $2;
1067                 my $backoff = $3;
1068                 $host = $bpc->uriUnesc($host);
1069                 if ( $CmdJob ne $host && defined($Status{$host})
1070                                       && defined($Jobs{$host}) ) {
1071                     print(LOG $bpc->timeStamp,
1072                                "Stopping current $Jobs{$host}{type} of $host,"
1073                              . " request by $user (backoff=$backoff)\n");
1074                     kill($bpc->sigName2num("INT"), $Jobs{$host}{pid});
1075                     #
1076                     # Don't close the pipe now; wait until the child
1077                     # really exits later.  Otherwise close() will
1078                     # block until the child has exited.
1079                     #  old code:
1080                     ##vec($FDread, $Jobs{$host}{fn}, 1) = 0;
1081                     ##close($Jobs{$host}{fh});
1082                     ##delete($Jobs{$host});
1083
1084                     $Status{$host}{state}    = "Status_idle";
1085                     if ( $Jobs{$host}{type} eq "restore" ) {
1086                         $Status{$host}{reason}
1087                                     = "Reason_restore_canceled_by_user";
1088                     } elsif ( $Jobs{$host}{type} eq "archive" ) {
1089                         $Status{$host}{reason}
1090                                     = "Reason_archive_canceled_by_user";
1091                     } else {
1092                         $Status{$host}{reason}
1093                                     = "Reason_backup_canceled_by_user";
1094                     }
1095                     $Status{$host}{activeJob} = 0;
1096                     $Status{$host}{startTime} = time;
1097                     $reply = "ok: $Jobs{$host}{type} of $host canceled";
1098                 } elsif ( $BgQueueOn{$host} || $UserQueueOn{$host} ) {
1099                     print(LOG $bpc->timeStamp,
1100                                "Stopping pending backup of $host,"
1101                              . " request by $user (backoff=$backoff)\n");
1102                     @BgQueue = grep($_->{host} ne $host, @BgQueue);
1103                     @UserQueue = grep($_->{host} ne $host, @UserQueue);
1104                     $BgQueueOn{$host} = $UserQueueOn{$host} = 0;
1105                     $reply = "ok: pending backup of $host canceled";
1106                 } else {
1107                     print(LOG $bpc->timeStamp,
1108                                "Nothing to do for stop backup of $host,"
1109                              . " request by $user (backoff=$backoff)\n");
1110                     $reply = "ok: no backup was pending or running";
1111                 }
1112                 if ( defined($Status{$host}) && $backoff ne "" ) {
1113                     if ( $backoff > 0 ) {
1114                         $Status{$host}{backoffTime} = time + $backoff * 3600;
1115                     } else {
1116                         delete($Status{$host}{backoffTime});
1117                     }
1118                 }
1119             } elsif ( $cmd =~ /^backup all$/ ) {
1120                 QueueAllPCs();
1121             } elsif ( $cmd =~ /^backup (\S+)\s+(\S+)\s+(\S+)\s+(\S+)/ ) {
1122                 my $hostIP = $1;
1123                 $host      = $2;
1124                 my $user   = $3;
1125                 my $doFull = $4;
1126                 $host      = $bpc->uriUnesc($host);
1127                 $hostIP    = $bpc->uriUnesc($hostIP);
1128                 if ( !defined($Status{$host}) ) {
1129                     print(LOG $bpc->timeStamp,
1130                                "User $user requested backup of unknown host"
1131                              . " $host\n");
1132                     $reply = "error: unknown host $host";
1133                 } elsif ( defined($Jobs{$host})
1134                                 && $Jobs{$host}{type} ne "restore" ) {
1135                     print(LOG $bpc->timeStamp,
1136                                "User $user requested backup of $host,"
1137                              . " but one is currently running\n");
1138                     $reply = "error: backup of $host is already running";
1139                 } else {
1140                     print(LOG $bpc->timeStamp,
1141                                "User $user requested backup of $host"
1142                              . " ($hostIP)\n");
1143                     if ( $BgQueueOn{$hostIP} ) {
1144                         @BgQueue = grep($_->{host} ne $hostIP, @BgQueue);
1145                         $BgQueueOn{$hostIP} = 0;
1146                     }
1147                     if ( $UserQueueOn{$hostIP} ) {
1148                         @UserQueue = grep($_->{host} ne $hostIP, @UserQueue);
1149                         $UserQueueOn{$hostIP} = 0;
1150                     }
1151                     unshift(@UserQueue, {
1152                                 host    => $hostIP,
1153                                 user    => $user,
1154                                 reqTime => time,
1155                                 doFull  => $doFull,
1156                                 userReq => 1,
1157                                 dhcp    => $hostIP eq $host ? 0 : 1,
1158                         });
1159                     $UserQueueOn{$hostIP} = 1;
1160                     $reply = "ok: requested backup of $host";
1161                 }
1162             } elsif ( $cmd =~ /^archive (\S+)\s+(\S+)\s+(\S+)/ ) {
1163                 my $user         = $1;
1164                 my $archivehost  = $2;
1165                 my $reqFileName  = $3;
1166                 $host      = $bpc->uriUnesc($archivehost);
1167                 if ( !defined($Status{$host}) ) {
1168                     print(LOG $bpc->timeStamp,
1169                                "User $user requested archive of unknown archive host"
1170                              . " $host");
1171                     $reply = "archive error: unknown archive host $host";
1172                 } else {
1173                     print(LOG $bpc->timeStamp,
1174                                "User $user requested archive on $host"
1175                              . " ($host)\n");
1176                     if ( defined($Jobs{$host}) ) {
1177                         $reply = "Archive currently running on $host, please try later";
1178                     } else {
1179                         unshift(@UserQueue, {
1180                                 host    => $host,
1181                                 hostIP  => $user,
1182                                 reqFileName => $reqFileName,
1183                                 reqTime => time,
1184                                 dhcp    => 0,
1185                                 archive => 1,
1186                                 userReq => 1,
1187                         });
1188                         $UserQueueOn{$host} = 1;
1189                         $reply = "ok: requested archive on $host";
1190                     }
1191                 }
1192             } elsif ( $cmd =~ /^restore (\S+)\s+(\S+)\s+(\S+)\s+(\S+)/ ) {
1193                 my $hostIP = $1;
1194                 $host      = $2;
1195                 my $user   = $3;
1196                 my $reqFileName = $4;
1197                 $host      = $bpc->uriUnesc($host);
1198                 $hostIP    = $bpc->uriUnesc($hostIP);
1199                 if ( !defined($Status{$host}) ) {
1200                     print(LOG $bpc->timeStamp,
1201                                "User $user requested restore to unknown host"
1202                              . " $host");
1203                     $reply = "restore error: unknown host $host";
1204                 } else {
1205                     print(LOG $bpc->timeStamp,
1206                                "User $user requested restore to $host"
1207                              . " ($hostIP)\n");
1208                     unshift(@UserQueue, {
1209                                 host    => $host,
1210                                 hostIP  => $hostIP,
1211                                 reqFileName => $reqFileName,
1212                                 reqTime => time,
1213                                 dhcp    => 0,
1214                                 restore => 1,
1215                                 userReq => 1,
1216                         });
1217                     $UserQueueOn{$host} = 1;
1218                     if ( defined($Jobs{$host}) ) {
1219                         $reply = "ok: requested restore of $host, but a"
1220                                . " job is currently running,"
1221                                . " so this request will start later";
1222                     } else {
1223                         $reply = "ok: requested restore of $host";
1224                     }
1225                 }
1226             } elsif ( $cmd =~ /^status\s*(.*)/ ) {
1227                 my($args) = $1;
1228                 my($dump, @values, @names);
1229                 foreach my $type ( split(/\s+/, $args) ) {
1230                     if ( $type =~ /^queues/ ) {
1231                         push(@values,  \@BgQueue, \@UserQueue, \@CmdQueue);
1232                         push(@names, qw(*BgQueue   *UserQueue   *CmdQueue));
1233                     } elsif ( $type =~ /^jobs/ ) {
1234                         push(@values,  \%Jobs);
1235                         push(@names, qw(*Jobs));
1236                     } elsif ( $type =~ /^queueLen/ ) {
1237                         push(@values,  {
1238                                 BgQueue   => scalar(@BgQueue),
1239                                 UserQueue => scalar(@UserQueue),
1240                                 CmdQueue  => scalar(@CmdQueue),
1241                             });
1242                         push(@names, qw(*QueueLen));
1243                     } elsif ( $type =~ /^info/ ) {
1244                         push(@values,  \%Info);
1245                         push(@names, qw(*Info));
1246                     } elsif ( $type =~ /^hosts/ ) {
1247                         push(@values,  \%Status);
1248                         push(@names, qw(*Status));
1249                     } elsif ( $type =~ /^host\((.*)\)/ ) {
1250                         my $h = $bpc->uriUnesc($1);
1251                         if ( defined($Status{$h}) ) {
1252                             push(@values,  {
1253                                     %{$Status{$h}},
1254                                     BgQueueOn => $BgQueueOn{$h},
1255                                     UserQueueOn => $UserQueueOn{$h},
1256                                     CmdQueueOn => $CmdQueueOn{$h},
1257                                 });
1258                             push(@names, qw(*StatusHost));
1259                         } else {
1260                             print(LOG $bpc->timeStamp,
1261                                       "Unknown host $h for status request\n");
1262                         }
1263                     } else {
1264                         print(LOG $bpc->timeStamp,
1265                                   "Unknown status request $type\n");
1266                     }
1267                 }
1268                 $dump = Data::Dumper->new(\@values, \@names);
1269                 $dump->Indent(0);
1270                 $reply = $dump->Dump;
1271             } elsif ( $cmd =~ /^link\s+(.+)/ ) {
1272                 my($host) = $1;
1273                 $host = $bpc->uriUnesc($host);
1274                 QueueLink($host);
1275             } elsif ( $cmd =~ /^log\s+(.*)/ ) {
1276                 print(LOG $bpc->timeStamp, "$1\n");
1277             } elsif ( $cmd =~ /^server\s+(\w+)/ ) {
1278                 my($type) = $1;
1279                 if ( $type eq 'reload' ) {
1280                     ServerReload("Reloading config/host files via CGI request");
1281                 } elsif ( $type eq 'shutdown' ) {
1282                     $reply = "Shutting down...\n";
1283                     syswrite($Clients{$client}{fh}, $reply, length($reply));
1284                     ServerShutdown("Server shutting down...");
1285                 }
1286             } elsif ( $cmd =~ /^quit/ || $cmd =~ /^exit/ ) {
1287                 $nbytes = 0;
1288                 last;
1289             } else {
1290                 print(LOG $bpc->timeStamp, "Unknown command $cmd\n");
1291                 $reply = "error: bad command $cmd";
1292             }
1293             #
1294             # send a reply to the client, at a minimum "ok\n".
1295             #
1296             $reply = "ok" if ( $reply eq "" );
1297             $reply .= "\n";
1298             syswrite($Clients{$client}{fh}, $reply, length($reply));
1299         }
1300         #
1301         # Detect possible denial-of-service attack from sending a huge line
1302         # (ie: never terminated).  32K seems to be plenty big enough as
1303         # a limit.
1304         #
1305         if ( length($Clients{$client}{mesg}) > 32 * 1024 ) {
1306             print(LOG $bpc->timeStamp, "Line too long from client"
1307                         . " '$Clients{$client}{clientName}':"
1308                         . " shutting down client connection\n");
1309             $nbytes = 0;
1310         }
1311         #
1312         # Shut down the client connection if we read EOF
1313         #
1314         if ( $nbytes <= 0 ) {
1315             close($Clients{$client}{fh});
1316             vec($FDread, $Clients{$client}{fn}, 1) = 0;
1317             delete($Clients{$client});
1318         }
1319     }
1320     #
1321     # Accept any new connections on each of our listen sockets
1322     #
1323     if ( vec($fdRead, fileno(SERVER_UNIX), 1) ) {
1324         local(*CLIENT);
1325         my $paddr = accept(CLIENT, SERVER_UNIX);
1326         $ClientConnCnt++;
1327         $Clients{$ClientConnCnt}{clientName} = "unix socket";
1328         $Clients{$ClientConnCnt}{mesg} = "";
1329         $Clients{$ClientConnCnt}{fh}   = *CLIENT;
1330         $Clients{$ClientConnCnt}{fn}   = fileno(CLIENT);
1331         vec($FDread, $Clients{$ClientConnCnt}{fn}, 1) = 1;
1332         #
1333         # Generate and send unique seed for MD5 digests to avoid
1334         # replay attacks.  See BackupPC::Lib::ServerMesg().
1335         #
1336         my $seed = time . ",$ClientConnCnt,$$,0\n";
1337         $Clients{$ClientConnCnt}{seed}    = $seed;
1338         $Clients{$ClientConnCnt}{mesgCnt} = 0;
1339         syswrite($Clients{$ClientConnCnt}{fh}, $seed, length($seed));
1340     }
1341     if ( $ServerInetPort > 0 && vec($fdRead, fileno(SERVER_INET), 1) ) {
1342         local(*CLIENT);
1343         my $paddr = accept(CLIENT, SERVER_INET);
1344         my($port,$iaddr) = sockaddr_in($paddr);
1345         my $name = gethostbyaddr($iaddr, AF_INET);
1346         $ClientConnCnt++;
1347         $Clients{$ClientConnCnt}{mesg} = "";
1348         $Clients{$ClientConnCnt}{fh}   = *CLIENT;
1349         $Clients{$ClientConnCnt}{fn}   = fileno(CLIENT);
1350         $Clients{$ClientConnCnt}{clientName} = "$name:$port";
1351         vec($FDread, $Clients{$ClientConnCnt}{fn}, 1) = 1;
1352         #
1353         # Generate and send unique seed for MD5 digests to avoid
1354         # replay attacks.  See BackupPC::Lib::ServerMesg().
1355         #
1356         my $seed = time . ",$ClientConnCnt,$$,$port\n";
1357         $Clients{$ClientConnCnt}{seed}    = $seed;
1358         $Clients{$ClientConnCnt}{mesgCnt} = 0;
1359         syswrite($Clients{$ClientConnCnt}{fh}, $seed, length($seed));
1360     }
1361 }
1362
1363 ###########################################################################
1364 # Miscellaneous subroutines
1365 ###########################################################################
1366
1367 #
1368 # Write the current status to $TopDir/log/status.pl
1369 #
1370 sub StatusWrite
1371 {
1372     my($dump) = Data::Dumper->new(
1373              [  \%Info, \%Status],
1374              [qw(*Info   *Status)]);
1375     $dump->Indent(1);
1376     if ( open(STATUS, ">", "$TopDir/log/status.pl") ) {
1377         print(STATUS $dump->Dump);
1378         close(STATUS);
1379     }
1380 }
1381
1382 #
1383 # Queue all the hosts for backup.  This means queuing all the fixed
1384 # ip hosts and all the dhcp address ranges.  We also additionally
1385 # queue the dhcp hosts with a -e flag to check for expired dumps.
1386 #
1387 sub QueueAllPCs
1388 {
1389     foreach my $host ( sort(keys(%$Hosts)) ) {
1390         delete($Status{$host}{backoffTime})
1391                 if ( defined($Status{$host}{backoffTime})
1392                   && $Status{$host}{backoffTime} < time );
1393         next if ( defined($Jobs{$host})
1394                 || $BgQueueOn{$host}
1395                 || $UserQueueOn{$host}
1396                 || $CmdQueueOn{$host} );
1397         if ( $Hosts->{$host}{dhcp} ) {
1398             $Status{$host}{dhcpCheckCnt}++;
1399             if ( $RunNightlyWhenIdle ) {
1400                 #
1401                 # Once per night queue a check for DHCP hosts that just
1402                 # checks for expired dumps.  We need to do this to handle
1403                 # the case when a DHCP host has not been on the network for
1404                 # a long time, and some of the old dumps need to be expired.
1405                 # Normally expiry checks are done by BackupPC_dump only
1406                 # after the DHCP hosts has been detected on the network.
1407                 #
1408                 unshift(@BgQueue,
1409                     {host => $host, user => "BackupPC", reqTime => time,
1410                      dhcp => 0, dumpExpire => 1});
1411                 $BgQueueOn{$host} = 1;
1412             }
1413         } else {
1414             #
1415             # this is a fixed ip host: queue it
1416             #
1417             unshift(@BgQueue,
1418                 {host => $host, user => "BackupPC", reqTime => time,
1419                  dhcp => $Hosts->{$host}{dhcp}});
1420             $BgQueueOn{$host} = 1;
1421         }
1422     }
1423     foreach my $dhcp ( @{$Conf{DHCPAddressRanges}} ) {
1424         for ( my $i = $dhcp->{first} ; $i <= $dhcp->{last} ; $i++ ) {
1425             my $ipAddr = "$dhcp->{ipAddrBase}.$i";
1426             next if ( defined($Jobs{$ipAddr})
1427                     || $BgQueueOn{$ipAddr}
1428                     || $UserQueueOn{$ipAddr}
1429                     || $CmdQueueOn{$ipAddr} );
1430             #
1431             # this is a potential dhcp ip address (we don't know the
1432             # host name yet): queue it
1433             #
1434             unshift(@BgQueue,
1435                 {host => $ipAddr, user => "BackupPC", reqTime => time,
1436                  dhcp => 1});
1437             $BgQueueOn{$ipAddr} = 1;
1438         }
1439     }
1440 }
1441
1442 #
1443 # Queue a BackupPC_link for the given host
1444 #
1445 sub QueueLink
1446 {
1447     my($host) = @_;
1448
1449     return if ( $CmdQueueOn{$host} );
1450     $Status{$host}{state}    = "Status_link_pending";
1451     $Status{$host}{needLink} = 0;
1452     unshift(@CmdQueue, {
1453             host    => $host,
1454             user    => "BackupPC",
1455             reqTime => time,
1456             cmd     => ["$BinDir/BackupPC_link",  $host],
1457         });
1458     $CmdQueueOn{$host} = 1;
1459 }
1460
1461 #
1462 # Read the hosts file, and update Status if any hosts have been
1463 # added or deleted.  We also track the mtime so the only need to
1464 # update the hosts file on changes.
1465 #
1466 # This function is called at startup, SIGHUP, and on each wakeup.
1467 # It returns 1 on success and undef on failure.
1468 #
1469 sub HostsUpdate
1470 {
1471     my($force) = @_;
1472     my $newHosts;
1473     #
1474     # Nothing to do if we already have the current hosts file
1475     #
1476     return 1 if ( !$force && defined($Hosts)
1477                           && $Info{HostsModTime} == $bpc->HostsMTime() );
1478     if ( !defined($newHosts = $bpc->HostInfoRead()) ) {
1479         print(LOG $bpc->timeStamp, "Can't read hosts file!\n");
1480         return;
1481     }
1482     print(LOG $bpc->timeStamp, "Reading hosts file\n");
1483     $Hosts = $newHosts;
1484     $Info{HostsModTime} = $bpc->HostsMTime();
1485     #
1486     # Now update %Status in case any hosts have been added or deleted
1487     #
1488     foreach my $host ( sort(keys(%$Hosts)) ) {
1489         next if ( defined($Status{$host}) );
1490         $Status{$host}{state} = "Status_idle";
1491         print(LOG $bpc->timeStamp, "Added host $host to backup list\n");
1492     }
1493     foreach my $host ( sort(keys(%Status)) ) {
1494         next if ( $host eq $bpc->trashJob
1495                      || $host eq $bpc->adminJob
1496                      || defined($Hosts->{$host})
1497                      || defined($Jobs{$host})
1498                      || $BgQueueOn{$host}
1499                      || $UserQueueOn{$host}
1500                      || $CmdQueueOn{$host} );
1501         print(LOG $bpc->timeStamp, "Deleted host $host from backup list\n");
1502         delete($Status{$host});
1503     }
1504     return 1;
1505 }
1506
1507 #
1508 # Remember the signal name for later processing
1509 #
1510 sub catch_signal
1511 {
1512     if ( $SigName ) {
1513         $SigName = shift;
1514         foreach my $host ( keys(%Jobs) ) {
1515             kill($bpc->sigName2num("INT"), $Jobs{$host}{pid});
1516         }
1517         #
1518         # In case we are inside the exit handler, reopen the log file
1519         #
1520         close(LOG);
1521         LogFileOpen();
1522         print(LOG "Fatal error: unhandled signal $SigName\n");
1523         unlink("$TopDir/log/BackupPC.pid");
1524         confess("Got new signal $SigName... quitting\n");
1525     } else {
1526         $SigName = shift;
1527     }
1528 }
1529
1530 #
1531 # Open the log file and point STDOUT and STDERR there too
1532 #
1533 sub LogFileOpen
1534 {
1535     mkpath("$TopDir/log", 0, 0777) if ( !-d "$TopDir/log" );
1536     open(LOG, ">>$TopDir/log/LOG")
1537             || die("Can't create LOG file $TopDir/log/LOG");
1538     close(STDOUT);
1539     close(STDERR);
1540     open(STDOUT, ">&LOG");
1541     open(STDERR, ">&LOG");
1542     select(LOG);    $| = 1;
1543     select(STDERR); $| = 1;
1544     select(STDOUT); $| = 1;
1545 }
1546
1547 #
1548 # Initialize the unix-domain and internet-domain sockets that
1549 # we listen to for client connections (from the CGI script and
1550 # some of the BackupPC sub-programs).
1551 #
1552 sub ServerSocketInit
1553 {
1554     if ( !defined(fileno(SERVER_UNIX)) ) {
1555         #
1556         # one-time only: initialize unix-domain socket
1557         #
1558         if ( !socket(SERVER_UNIX, PF_UNIX, SOCK_STREAM, 0) ) {
1559             print(LOG $bpc->timeStamp, "unix socket() failed: $!\n");
1560             exit(1);
1561         }
1562         my $sockFile = "$TopDir/log/BackupPC.sock";
1563         unlink($sockFile);
1564         if ( !bind(SERVER_UNIX, sockaddr_un($sockFile)) ) {
1565             print(LOG $bpc->timeStamp, "unix bind() failed: $!\n");
1566             exit(1);
1567         }
1568         if ( !listen(SERVER_UNIX, SOMAXCONN) ) {
1569             print(LOG $bpc->timeStamp, "unix listen() failed: $!\n");
1570             exit(1);
1571         }
1572         vec($FDread, fileno(SERVER_UNIX), 1) = 1;
1573     }
1574     return if ( $ServerInetPort == $Conf{ServerPort} );
1575     if ( $ServerInetPort > 0 ) {
1576         vec($FDread, fileno(SERVER_INET), 1) = 0;
1577         close(SERVER_INET);
1578         $ServerInetPort = -1;
1579     }
1580     if ( $Conf{ServerPort} > 0 ) {
1581         #
1582         # Setup a socket to listen on $Conf{ServerPort}
1583         #
1584         my $proto = getprotobyname('tcp');
1585         if ( !socket(SERVER_INET, PF_INET, SOCK_STREAM, $proto) ) {
1586             print(LOG $bpc->timeStamp, "inet socket() failed: $!\n");
1587             exit(1);
1588         }
1589         if ( !setsockopt(SERVER_INET, SOL_SOCKET, SO_REUSEADDR, pack("l",1)) ) {
1590             print(LOG $bpc->timeStamp, "setsockopt() failed: $!\n");
1591             exit(1);
1592         }
1593         if ( !bind(SERVER_INET, sockaddr_in($Conf{ServerPort}, INADDR_ANY)) ) {
1594             print(LOG $bpc->timeStamp, "inet bind() failed: $!\n");
1595             exit(1);
1596         }
1597         if ( !listen(SERVER_INET, SOMAXCONN) ) {
1598             print(LOG $bpc->timeStamp, "inet listen() failed: $!\n");
1599             exit(1);
1600         }
1601         vec($FDread, fileno(SERVER_INET), 1) = 1;
1602         $ServerInetPort = $Conf{ServerPort};
1603     }
1604 }
1605
1606 #
1607 # Reload the server.  Used by Main_Process_Signal when $SigName eq "HUP"
1608 # or when the command "server reload" is received.
1609 #
1610 sub ServerReload
1611 {
1612     my($mesg) = @_;
1613     $mesg = $bpc->ConfigRead() || $mesg;
1614     print(LOG $bpc->timeStamp, "$mesg\n");
1615     $Info{ConfigModTime} = $bpc->ConfigMTime();
1616     %Conf = $bpc->Conf();
1617     umask($Conf{UmaskMode});
1618     ServerSocketInit();
1619     HostsUpdate(0);
1620     $NextWakeup = 0;
1621     $Info{ConfigLTime} = time;
1622 }
1623
1624 #
1625 # Gracefully shutdown the server.  Used by Main_Process_Signal when
1626 # $SigName ne "" && $SigName ne "HUP" or when the command
1627 # "server shutdown" is received.
1628 #
1629 sub ServerShutdown
1630 {
1631     my($mesg) = @_;
1632     print(LOG $bpc->timeStamp, "$mesg\n");
1633     if ( keys(%Jobs) ) {
1634         foreach my $host ( keys(%Jobs) ) {
1635             kill($bpc->sigName2num("INT"), $Jobs{$host}{pid});
1636         }
1637         sleep(1);
1638         foreach my $host ( keys(%Jobs) ) {
1639             kill($bpc->sigName2num("KILL"), $Jobs{$host}{pid});
1640         }
1641         %Jobs = ();
1642     }
1643     StatusWrite();
1644     unlink("$TopDir/log/BackupPC.pid");
1645     exit(1);
1646 }
1647