1 =head1 BackupPC Introduction
3 This documentation describes BackupPC version __VERSION__,
4 released on __RELEASEDATE__.
8 BackupPC is a high-performance, enterprise-grade system for backing up
9 Linux and WinXX PCs, desktops and laptops to a server's disk. BackupPC
10 is highly configurable and easy to install and maintain.
12 Given the ever decreasing cost of disks and raid systems, it is now
13 practical and cost effective to backup a large number of machines onto a
14 server's local disk or network storage. For some sites this might be
15 the complete backup solution. For other sites additional permanent
16 archives could be created by periodically backing up the server to tape.
24 A clever pooling scheme minimizes disk storage and disk I/O.
25 Identical files across multiple backups of the same or different PC
26 are stored only once (using hard links), resulting in substantial
27 savings in disk storage and disk writes.
31 Optional compression provides additional reductions in storage
32 (around 40%). The CPU impact of compression is low since only
33 new files (those not already in the pool) need to be compressed.
37 A powerful http/cgi user interface allows administrators to view log
38 files, configuration, current status and allows users to initiate and
39 cancel backups and browse and restore files from backups.
43 No client-side software is needed. On WinXX the standard smb
44 protocol is used to extract backup data. On linux or unix clients,
45 rsync or tar (over ssh/rsh/nfs) is used to extract backup data.
46 Alternatively, rsync can also be used on WinXX (using cygwin),
47 and Samba could be installed on the linux or unix client to
52 Flexible restore options. Single files can be downloaded from
53 any backup directly from the CGI interface. Zip or Tar archives
54 for selected files or directories from any backup can also be
55 downloaded from the CGI interface. Finally, direct restore to
56 the client machine (using smb or tar) for selected files or
57 directories is also supported from the CGI interface.
61 BackupPC supports mobile environments where laptops are only
62 intermittently connected to the network and have dynamic IP addresses
63 (DHCP). Configuration settings allow machines connected via slower WAN
64 connections (eg: dial up, DSL, cable) to not be backed up, even if they
65 use the same fixed or dynamic IP address as when they are connected
70 Flexible configuration parameters allow multiple backups to be performed
71 in parallel, specification of which shares to backup, which directories
72 to backup or not backup, various schedules for full and incremental
73 backups, schedules for email reminders to users and so on. Configuration
74 parameters can be set system-wide or also on a per-PC basis.
78 Users are sent periodic email reminders if their PC has not
79 recently been backed up. Email content, timing and policies
84 BackupPC is Open Source software hosted by SourceForge.
94 A full backup is a complete backup of a share. BackupPC can be configured to
95 do a full backup at a regular interval (often weekly). BackupPC can also
96 be configured to keep a certain number of full backups, and to keep
97 a smaller number of very old full backups.
99 =item Incremental Backup
101 An incremental backup is a backup of files that have changed (based on their
102 modification time) since the last successful full backup. For SMB and
103 tar, BackupPC backups all files that have changed since one hour prior to the
104 start of the last successful full backup. Rsync is more clever: any files
105 who attributes have changed (ie: uid, gid, mtime, modes, size) since the
106 last full are backed up. Deleted and new files are also detected by
107 Rsync incrementals (SMB and tar are not able to detect deleted files or
108 new files whose modification time is prior to the last full dump.
110 BackupPC can also be configured to keep a certain number of incremental
111 backups, and to keep a smaller number of very old incremental backups.
112 (BackupPC does not support multi-level incremental backups, although it
113 would be easy to do so.)
115 BackupPC's CGI interface "fills-in" incremental backups based on the
116 last full backup, giving every backup a "full" appearance. This makes
117 browsing and restoring backups easier.
119 =item Identical Files
121 BackupPC pools identical files using hardlinks. By "identical
122 files" we mean files with identical contents, not necessary the
123 same permissions, ownership or modification time. Two files might
124 have different permissions, ownership, or modification time but
125 will still be pooled whenever the contents are identical. This
126 is possible since BackupPC stores the file meta-data (permissions,
127 ownership, and modification time) separately from the file contents.
131 Based on your site's requirements you need to decide what your backup
132 policy is. BackupPC is not designed to provide exact re-imaging of
133 failed disks. See L<Limitations|limitations> for more information.
134 However, the addition of tar transport for linux/unix clients, plus
135 full support for special file types and unix attributes in v1.4.0
136 likely means an exact image of a linux/unix file system can be made.
138 BackupPC saves backups onto disk. Because of pooling you can relatively
139 economically keep several weeks of old backups. But BackupPC does not
140 provide permanent storage to tape. Other Open Source applications can do
141 this by backing up BackupPC's pool directories to tape.
143 At some sites the disk-based backup will be adequate, without a
144 secondary tape backup. This system is robust to any single failure: if a
145 client disk fails or loses files, the BackupPC server can be used to
146 restore files. If the server disk fails, BackupPC can be restarted on a
147 fresh file system, and create new backups from the clients. The chance
148 of the server disk failing can be made very small by spending more money
149 on increasingly better RAID systems.
151 At other sites a secondary tape backup will be required. This tape
152 backup can be done perhaps weekly from the BackupPC pool file system.
154 One comment: in the US in particular, permanent backups of things like
155 email are becoming strongly discouraged by lawyers because of discovery
156 prior to possible litigation. Using BackupPC without tape backup allows
157 recent file changes or losses to be restored, but without keeping a
158 history more than a month or two old (although this doesn't avoid the
159 problem of old emails languishing in user's email folders forever).
167 =item BackupPC home page
169 The BackupPC Open Source project is hosted on SourceForge. The
170 home page can be found at:
172 http://backuppc.sourceforge.net
174 This page has links to the current documentation, the SourceForge
175 project page and general information.
177 =item SourceForge project
179 The SourceForge project page is at:
181 http://sourceforge.net/projects/backuppc
183 This page has links to the current releases of BackupPC.
187 Two BackupPC mailing lists exist for announcements (backuppc-announce)
188 and reporting information, asking questions, discussing development or
189 any other topic relevant to BackupPC (backuppc-users).
191 You are encouraged to subscribe to either the backuppc-announce
192 or backuppc-users mail list on sourceforge.net at either:
194 http://lists.sourceforge.net/lists/listinfo/backuppc-announce
195 http://lists.sourceforge.net/lists/listinfo/backuppc-users
197 The backuppc-announce list is moderated and is used only for
198 important announcements (eg: new versions). It is low traffic.
199 You only need to subscribe to one list: backuppc-users also
200 receives any messages on backuppc-announce.
202 To post a message to the backuppc-users list, send an email to
204 backuppc-users@lists.sourceforge.net
206 Do not send subscription requests to this address!
208 =item Other Programs of Interest
210 If you want to mirror linux or unix files or directories to a remote server
211 you should consider rsync, L<http://rsync.samba.org>. BackupPC uses
212 rsync as a transport mechanism; if you are already an rsync user you
213 can think of BackupPC as adding efficient storage (compression and
214 pooling) and a convenient user interface to rsync.
216 Unison is a utility that can do two-way, interactive, synchronization.
217 See L<http://www.cis.upenn.edu/~bcpierce/unison>.
219 Two popular open source packages that do tape backup are
220 Amanda (L<http://www.amanda.org>) and
221 afbackup (L<http://sourceforge.net/projects/afbackup>).
222 Amanda can also backup WinXX machines to tape using samba.
223 These packages can be used as back ends to BackupPC to backup the
224 BackupPC server data to tape.
230 Here are some ideas for new features that might appear in future
231 releases of BackupPC:
237 Adding hardlink support to rsync.
241 Adding block and file checksum caching to rsync. This will significantly
242 increase performance.
246 Adding a trip wire feature for notification when files below certain
247 directories change. For example, if you are backing up a DMZ machine,
248 you could request that you get sent email if any files below /bin,
249 /sbin or /usr change.
253 Resuming incomplete full backups. Useful if a machine
254 (eg: laptop) is disconnected from the network during a backup,
255 or if the user manually stops a backup. This would be supported
256 initially for rsync. The partial dump would be kept, and be
257 browsable. When the next dump starts, an incremental against
258 the partial dump would be done to make sure it was up to date,
259 and then the rest of the full dump would be done.
263 Replacing smbclient with the perl module FileSys::SmbClient. This
264 gives much more direct control of the smb transfer, allowing
265 incrementals to depend on any attribute change (eg: exist, mtime,
266 file size, uid, gid), and better support for include and exclude.
267 Currently smbclient incrementals only depend upon mtime, so
268 deleted files or renamed files are not detected. FileSys::SmbClient
269 would also allow resuming of incomplete full backups in the
270 same manner as rsync will.
274 Support --listed-incremental or --incremental for tar,
275 so that incrementals will depend upon any attribute change (eg: exist,
276 mtime, file size, uid, gid), rather than just mtime. This will allow
277 tar to be to as capable as FileSys::SmbClient and rsync.
281 For rysnc (and smb when FileSys::SmbClient is supported, and tar when
282 --listed-incremental is supported) support multi-level incrementals.
283 In fact, since incrementals will now be more "accurate", you could
284 choose to never to full dumps (except the first time), or at a
285 minimum do them infrequently: each incremental would depend upon
286 the last, giving a continuous chain of differential dumps.
290 Add a backup browsing feature that shows backup history by file.
291 So rather than a single directory view, it would be a table showing
292 the files (down) and the backups (across). The internal hardlinks
293 encode which files are identical across backups. You could immediately
294 see which files changed on which backups.
298 More speculative: Storing binary file deltas (in fact, reverse deltas)
299 for files that have the same name as a previous backup, but that aren't
300 already in the pool. This will save storage for things like mailbox
301 files or documents that change slightly between backups. Running some
302 benchmarks on a large pool suggests that the potential savings are
303 around 15-20%, which isn't spectacular, and likely not worth the
304 implementation effort. The program xdelta (v1) on SourceForge (see
305 L<http://sourceforge.net/projects/xdelta>) uses an rsync algorithm for
306 doing efficient binary file deltas. Rather than using an external
307 program, File::RsyncP will eventually get the necessary delta
308 generataion code from rsync.
312 Comments and suggestions are welcome.
316 BackupPC is free. I work on BackupPC because I enjoy doing it and I like
317 to contribute to the open source community.
319 BackupPC already has more than enough features for my own needs. The
320 main compensation for continuing to work on BackupPC is knowing that
321 more and more people find it useful. So feedback is certainly
322 appreciated. Even negative feedback is helpful, for example "We
323 evaluated BackupPC but didn't use it because it doesn't ...".
325 Beyond being a satisfied user and telling other people about it, everyone
326 is encouraged to add links to L<http://backuppc.sourceforge.net> (I'll
327 see then via Google) or otherwise publicize BackupPC. Unlike the
328 commercial products in this space, I have a zero budget (in both
329 time and money) for marketing, PR and advertising, so it's up to
332 Also, everyone is encouraged to contribute patches, bug reports, feature
333 and design suggestions, code, and documentation corrections or
336 =head1 Installing BackupPC
346 A linux, solaris, or unix based server with a substantial amount of free
347 disk space (see the next section for what that means). The CPU and disk
348 performance on this server will determine how many simultaneous backups
349 you can run. You should be able to run 4-8 simultaneous backups on a
350 moderately configured server.
352 When BackupPC starts with an empty pool, all the backup data will be
353 written to the pool on disk. After more backups are done, a higher
354 percentage of incoming files will already be in the pool. BackupPC is
355 able to avoid writing to disk new files that are already in the pool.
356 So over time disk writes will reduce significantly (by perhaps a factor
357 of 20 or more), since eventually 95% or more of incoming backup files
358 are typically in the pool. Disk reads from the pool are still needed to
359 do file compares to verify files are an exact match. So, with a mature
360 pool, if a relatively fast client generates data at say 1MB/sec, and you
361 run 4 simultaneous backups, there will be an average server disk load of
362 about 4MB/sec reads and 0.2MB/sec writes (assuming 95% of the incoming
363 files are in the pool). These rates will be perhaps 40% lower if
368 Perl version 5.6.0 or later. BackupPC has been tested with
369 version 5.6.0 and 5.6.1. If you don't have perl, please see
370 L<http://www.cpan.org>.
374 Perl modules Compress::Zlib, Archive::Zip and Rsync. Try "perldoc
375 Compress::Zlib" and "perldoc Archive::Zip" to see if you have these
376 modules. If not, fetch them from L<http://www.cpan.org> and see the
377 instructions below for how to build and install them.
379 The Rsync module is available from L<http://backuppc.sourceforge.net>.
380 You'll need to install the Rsync module if you want to use Rsync as
385 If you are using smb to backup WinXX machines you need smbclient and
386 nmblookup from the samba package. You will also need nmblookup if
387 you are backing up linux/unix DHCP machines. See L<http://www.samba.org>.
388 Version >= 2.2.0 of Samba is required (smbclient's tar feature in
389 2.0.8 and prior has bugs for file path lengths around 100 characters
390 and generates bad output when file lengths change during the backup).
392 See L<http://www.samba.org> for source and binaries. It's pretty easy to
393 fetch and compile samba, and just grab smbclient and nmblookup, without
394 doing the installation. Alternatively, L<http://www.samba.org> has binary
395 distributions for most platforms.
399 If you are using tar to backup linux/unix machines you should have version
400 1.13.7 at a minimum, with version 1.13.20 or higher recommended. Use
401 "tar --version" to check your version. Various GNU mirrors have the newest
402 versions of tar, see for example L<http://www.funet.fi/pub/gnu/alpha/gnu/tar>.
403 As of July 2002 the latest versons is 1.13.25.
407 If you are using rsync to backup linux/unix machines you should have
408 version 2.5.5 on each client machine. See L<http://rsync.samba.org>.
409 Use "rsync --version" to check your version.
411 For BackupPC to use Rsync you will also need to install the perl
412 File::RsyncP module, which is available from
413 L<http://perlrsync.sourceforge.net>.
414 Version 0.20 is required.
418 The Apache web server, see L<http://www.apache.org>, preferably built
419 with mod_perl support.
423 =head2 How much disk space do I need?
425 Here's one real example for an environment that is backing up 65 laptops
426 with compression off. Each full backup averages 3.2GB. Each incremental
427 backup averages about 0.2GB. Storing one full backup and two incremental
428 backups per laptop is around 240GB of raw data. But because of the
429 pooling of identical files, only 87GB is used. This is without
432 Another example, with compression on: backing up 95 laptops, where
433 each backup averages 3.6GB and each incremental averages about 0.3GB.
434 Keeping three weekly full backups, and six incrementals is around
435 1200GB of raw data. Because of pooling and compression, only 150GB
438 Here's a rule of thumb. Add up the C drive usage of all the machines you
439 want to backup (210GB in the first example above). This is a rough
440 minimum space estimate that should allow a couple of full backups and at
441 least half a dozen incremental backups per machine. If compression is on
442 you can reduce the storage requirements by maybe 30-40%. Add some margin
443 in case you add more machines or decide to keep more old backups.
445 Your actual mileage will depend upon the types of clients, operating
446 systems and applications you have. The more uniform the clients and
447 applications the bigger the benefit from pooling common files.
449 For example, the Eudora email tool stores each mail folder in a separate
450 file, and attachments are extracted as separate files. So in the sadly
451 common case of a large attachment emailed to many recipients, Eudora
452 will extract the attachment into a new file. When these machines are
453 backed up, only one copy of the file will be stored on the server, even
454 though the file appears in many different full or incremental backups. In
455 this sense Eudora is a "friendly" application from the point of view of
456 backup storage requirements.
458 An example at the other end of the spectrum is Outlook. Everything
459 (email bodies, attachments, calendar, contact lists) is stored in a
460 single file, which often becomes huge. Any change to this file requires
461 a separate copy of the file to be saved during backup. Outlook is even
462 more troublesome, since it keeps this file locked all the time, so it
463 cannot be read by smbclient whenever Outlook is running. See the
464 L<Limitations|limitations> section for more discussion of this problem.
466 =head2 Step 1: Getting BackupPC
468 Download the latest version from L<http://backuppc.sourceforge.net>.
470 =head2 Step 2: Installing the distribution
472 First off, there are three perl modules you should install.
473 These are all optional, but highly recommended:
479 To enable compression, you will need to install Compress::Zlib
480 from L<http://www.cpan.org>.
481 You can run "perldoc Compress::Zlib" to see if this module is installed.
485 To support restore via Zip archives you will need to install
486 Archive::Zip, also from L<http://www.cpan.org>.
487 You can run "perldoc Archive::Zip" to see if this module is installed.
491 To use rsync and rsyncd with BackupPC you will need to install File::RsyncP.
492 You can run "perldoc File::RsyncP" to see if this module is installed.
493 File::RsyncP is available from L<http://perlrsync.sourceforge.net>.
494 Version 0.20 is required.
498 To build and install these packages, fetch the tar.gz file and
499 then run these commands:
501 tar zxvf Archive-Zip-1.01.tar.gz
508 The same sequence of commands can be used for each module.
510 Now let's move onto BackupPC itself. After fetching
511 BackupPC-__VERSION__.tar.gz, run these commands as root:
513 tar zxf BackupPC-__VERSION__.tar.gz
514 cd BackupPC-__VERSION__
517 You will be prompted for the full paths of various executables, and
518 you will be prompted for the following information:
524 It is best if BackupPC runs as a special user, eg backuppc, that has
525 limited privileges. It is preferred that backuppc belongs to a system
526 administrator group so that sys admin members can browse backuppc files,
527 edit the configuration files and so on. Although configurable, the
528 default settings leave group read permission on pool files, so make
529 sure the BackupPC user's group is chosen restrictively.
531 On this installation, this is __BACKUPPCUSER__.
535 You need to decide where to put the data directory, below which
536 all the BackupPC data is stored. This needs to be a big file system.
538 On this installation, this is __TOPDIR__.
540 =item Install Directory
542 You should decide where the BackupPC scripts, libraries and documentation
543 should be installed, eg: /opt/local/BackupPC.
545 On this installation, this is __INSTALLDIR__.
547 =item CGI bin Directory
549 You should decide where the BackupPC CGI script resides. This will
550 usually below Apache's cgi-bin directory.
552 On this installation, this is __CGIDIR__.
554 =item Apache image directory
556 A directory where BackupPC's images are stored so that Apache can
557 serve them. This should be somewhere under Apache's DocumentRoot
562 =head2 Step 3: Setting up config.pl
564 After running configure.pl, browse through the config file,
565 __INSTALLDIR__/conf/config.pl, and make sure all the default settings
566 are correct. In particular, you will need to decide whether to use
567 smb or tar transport (or whether to set it on a per-PC basis),
568 set the smb share password (if using smb), set the backup policies
569 and modify the email message headers and bodies.
571 BackupPC needs to know the smb share user name and password for each PC
572 that uses smb (ie: all the WinXX clients). The user name is specified
573 in $Conf{SmbShareUserName}. There are four ways to tell BackupPC the smb
580 As an environment variable BPC_SMB_PASSWD set before BackupPC starts.
581 If you start BackupPC manually the BPC_SMB_PASSWD variable must be set
582 manually first. For backward compatability for v1.5.0 and prior, the
583 environment variable PASSWD can be used if BPC_SMB_PASSWD is not set.
584 Warning: on some systems it is possible to see environment variables of
589 Alternatively the BPC_SMB_PASSWD setting can be included in
590 /etc/init.d/backuppc, in which case you must make sure this file
591 is not world (other) readable.
595 As a configuration variable $Conf{SmbSharePasswd} in
596 __TOPDIR__/conf/config.pl. If you put the password
597 here you must make sure this file is not world (other) readable.
601 As a configuration variable $Conf{SmbSharePasswd} in the per-PC
602 configuration file, __TOPDIR__/pc/$host/config.pl. You will have to
603 use this option if the smb share password is different for each host.
604 If you put the password here you must make sure this file is not
605 world (other) readable.
609 Placement and protection of the smb share password is a possible
610 security risk, so please double-check the file and directory
611 permissions. In a future version there might be support for
612 encryption of this password, but a private key will still have to
613 be stored in a protected place. Suggestions are welcome.
615 =head2 Step 4: Setting up the hosts file
617 The file __TOPDIR__/conf/hosts contains the list of clients to backup.
618 BackupPC reads this file in three cases:
628 When BackupPC is sent a HUP (-1) signal. Assuming you installed the
629 init.d script, you can also do this with "/etc/init.d/backuppc reload".
633 When the modification time of the hosts file changes. BackupPC
634 checks the modification time once during each regular wakeup.
638 Whenever you change the hosts file (to add or remove a host) you can
639 either do a kill -HUP BackupPC_pid or simply wait until the next regular
642 Each line in the hosts file contains three fields, separated
649 If this host is a static IP address this must the machine's IP host name
650 (ie: something that can be looked up using nslookup or DNS). If this is
651 a host with a dynamic IP address (ie: DHCP flag is 1) then the host
652 name must be the netbios name of the machine. The host name should
657 Set to 0 if this host has a static IP address (meaning it can be looked
658 up by name in the DNS). If the host's IP address is dynamic (eg, it is
659 assigned by DHCP) then set this flag to 1.
661 The hosts with dhcp = 1 are backed up as follows. If you have
662 configured a DHCP address pool ($Conf{DHCPAddressRanges}) then
663 BackupPC will check the NetBIOS name of each machine in the
664 range. Any hosts that have a valid NetBIOS name (ie: matching
665 an entry in the hosts file) will be backed up.
669 This should be the unix login/email name of the user who "owns" or uses
670 this machine. This is the user who will be sent email about this
671 machine, and this user will have permission to stop/start/browse/restore
672 backups for this host. Leave this blank if no specific person should
673 receive email or be allowed to stop/start/browse/restore backups
674 for this host. Administrators will still have full permissions.
678 The first non-comment line of the hosts file is special: it contains
679 the names of the columns and should not be edited.
681 Here's a simple example of a hosts file:
687 The range of DHCP addresses to search is specified in
688 $Conf{DHCPAddressRanges}.
690 =head2 Step 5: Client Setup
692 Two methods for getting backup data from a client are
693 supported: smb and tar. Smb is the preferred method for WinXX clients
694 and tar is preferred method for linux/unix clients.
696 The transfer method is set using the $Conf{XferMethod} configuration
697 setting. If you have a mixed environment (ie: you will use smb for some
698 clients and tar for others), you will need to pick the most common
699 choice for $Conf{XferMethod} for the main config.pl file, and then
700 override it in the per-PC config file for those hosts that will use
701 the other method. (Or you could run two completely separate instances
702 of BackupPC, with different data directories, one for WinXX and the
703 other for linux/unix, but then common files between the different
704 machine types will duplicated.)
706 Here are some brief client setup notes:
712 The preferred setup for WinXX clients is to set $Conf{XferMethod} to "smb".
714 You need to create shares for the data you want to backup.
715 Open "My Computer", right click on the drive (eg: C), and
716 select "Sharing..." (or select "Properties" and select the
717 "Sharing" tab). In this dialog box you can enable sharing,
718 select the share name and permissions.
720 If this machine uses DHCP you will also need to make sure the
721 NetBios name is set. Go to Control Panel|System|Network Identification
722 (on Win2K) or Control Panel|System|Computer Name (on WinXP).
723 Also, you should go to Control Panel|Network Connections|Local Area
724 Connection|Properties|Internet Protocol (TCP/IP)|Properties|Advanced|WINS
725 and verify that NetBios is not disabled.
727 As an alternative to setting $Conf{XferMethod} to "smb" (using
728 smbclient) for WinXX clients, you can use an smb network filesystem (eg:
729 ksmbfs or similar) on your linux/unix server to mount the share,
730 and then set $Conf{XferMethod} to "tar" (use tar on the network
731 mounted file system).
733 Also, to make sure that file names with 8-bit characters are correctly
734 transferred by smbclient you should add this to samba's smb.conf file:
737 # Accept the windows charset
738 client code page = 850
739 character set = ISO8859-1
741 This setting should work for western europe.
742 See L<http://www.oreilly.com/catalog/samba/chapter/book/ch08_03.html>
743 for more information about settings for other languages.
747 The preferred setup for linux/unix clients is to set $Conf{XferMethod}
750 You can use either smb or tar for linux/unix machines. Smb requires that
751 the Samba server (smbd) be run to provide the shares. Since the smb
752 protocol can't represent special files like symbolic links and fifos,
753 tar is the recommended transport method for linux/unix machines.
754 (In fact, by default samba makes symbolic links look like the file or
755 directory that they point to, so you could get an infinite loop if a
756 symbolic link points to the current or parent directory. If you really
757 need to use Samba shares for linux/unix backups you should turn off the
758 "follow symlinks" samba config setting. See the smb.conf manual page.)
760 The rest of this section describes the tar setup.
762 You must have GNU tar on the client machine. Use "tar --version"
763 or "gtar --version" to verify. The version should be at least
764 1.13.7, and 1.13.20 or greater is recommended.
766 For linux/unix machines you should no backup "/proc". This directory
767 contains a variety of files that look like regular files but they are
768 special files that don't need to be backed up (eg: /proc/kcore is a
769 regular file that contains physical memory). See $Conf{BackupFilesExclude}.
770 It is safe to back up /dev since it contains mostly character-special
771 and block-special files, which are correctly handed by BackupPC
772 (eg: backing up /dev/hda5 just saves the block-special file information,
773 not the contents of the disk).
775 Next you should decide whether to run tar over ssh, rsh or nfs. Ssh is
776 the preferred method. Rsh is not secure and therefore not recommended.
777 Nfs will work, but you need to make sure that the BackupPC user (running
778 on the server) has sufficient permissions to read all the files below
781 Ssh allows BackupPC to run as a privileged user on the client (eg:
782 root), since it needs sufficient permissions to read all the backup
783 files. Ssh is setup so that BackupPC on the server (an otherwise low
784 privileged user) can ssh as root on the client, without being prompted
785 for a password. There are two common versions of ssh: v1 and v2. Here
786 are some instructions for one way to setup ssh. (Check which version
787 of SSH you have by typing "ssh" or "man ssh".)
791 =item OpenSSH Instructions
797 As root on the client machine, use ssh-keygen to generate a
798 public/private key pair, without a pass-phrase:
800 ssh-keygen -t rsa -N ''
802 This will save the public key in ~/.ssh/id_rsa.pub and the private
803 key in ~/.ssh/id_rsa.
807 Repeat the above steps for the BackupPC user (__BACKUPPCUSER__) on the server.
808 Make a copy of the public key to make it recognizable, eg:
810 ssh-keygen -t rsa -N ''
811 cp ~/.ssh/id_rsa.pub ~/.ssh/BackupPC_id_rsa.pub
813 See the ssh and sshd manual pages for extra configuration information.
817 To allow BackupPC to ssh to the client as root, you need to place
818 BackupPC's public key into root's authorized list on the client.
819 Append BackupPC's public key (BackupPC_id_rsa.pub) to root's
820 ~/.ssh/authorized_keys2 file on the client:
822 touch ~/.ssh/authorized_keys2
823 cat BackupPC_id_rsa.pub >> ~/.ssh/authorized_keys2
825 You should edit ~/.ssh/authorized_keys2 and add further specifiers,
826 eg: from, to limit which hosts can login using this key. For example,
827 if your BackupPC host is called backuppc.my.com, there should be
828 one line in ~/.ssh/authorized_keys2 that looks like:
830 from="backuppc.my.com" ssh-rsa [base64 key, eg: ABwBCEAIIALyoqa8....]
832 =item Fix permissions
834 You will probably need to make sure that all the files
835 in ~/.ssh have no group or other read/write permission:
837 chmod -R go-rwx ~/.ssh
839 You should do the same thing for the BackupPC user on the server.
843 As the BackupPC user on the server, verify that this command:
845 ssh -l root clientHostName whoami
851 You might be prompted the first time to accept the client's host key and
852 you might be prompted for root's password on the client. Make sure that
853 this command runs cleanly with no prompts after the first time. You
854 might need to check /etc/hosts.equiv on the client. Look at the
855 man pages for more information. The "-v" option to ssh is a good way
856 to get detailed information about what fails.
860 =item SSH2 Instructions
866 As root on the client machine, use ssh-keygen2 to generate a
867 public/private key pair, without a pass-phrase:
869 ssh-keygen2 -t rsa -P
873 ssh-keygen -t rsa -N ''
875 (This command might just be called ssh-keygen on your machine.)
877 This will save the public key in /.ssh2/id_rsa_1024_a.pub and the private
878 key in /.ssh2/id_rsa_1024_a.
882 Create the identification file /.ssh2/identification:
884 echo "IdKey id_rsa_1024_a" > /.ssh2/identification
888 Repeat the above steps for the BackupPC user (__BACKUPPCUSER__) on the server.
889 Rename the key files to recognizable names, eg:
891 ssh-keygen2 -t rsa -P
892 mv ~/.ssh2/id_rsa_1024_a.pub ~/.ssh2/BackupPC_id_rsa_1024_a.pub
893 mv ~/.ssh2/id_rsa_1024_a ~/.ssh2/BackupPC_id_rsa_1024_a
894 echo "IdKey BackupPC_id_rsa_1024_a" > ~/.ssh2/identification
896 Based on your ssh2 configuration, you might also need to turn off
897 StrictHostKeyChecking and PasswordAuthentication:
899 touch ~/.ssh2/ssh2_config
900 echo "StrictHostKeyChecking ask" >> ~/.ssh2/ssh2_config
901 echo "PasswordAuthentication no" >> ~/.ssh2/ssh2_config
905 To allow BackupPC to ssh to the client as root, you need to place
906 BackupPC's public key into root's authorized list on the client.
907 Copy BackupPC's public key (BackupPC_id_rsa_1024_a.pub) to the
908 /.ssh2 directory on the client. Add the following line to the
909 /.ssh2/authorization file on the client (as root):
911 touch /.ssh2/authorization
912 echo "Key BackupPC_id_rsa_1024_a.pub" >> /.ssh2/authorization
914 =item Fix permissions
916 You will probably need to make sure that all the files
917 in /.ssh2 have no group or other read/write permission:
919 chmod -R go-rwx /.ssh2
921 You should do the same thing for the BackupPC user on the server.
925 As the BackupPC user on the server, verify that this command:
927 ssh2 -l root clientHostName whoami
933 You might be prompted the first time to accept the client's host key and
934 you might be prompted for root's password on the client. Make sure that
935 this command runs cleanly with no prompts after the first time. You
936 might need to check /etc/hosts.equiv on the client. Look at the
937 man pages for more information. The "-v" option to ssh2 is a good way
938 to get detailed information about what fails.
942 =item SSH version 1 Instructions
944 The concept is identical and the steps are similar, but the specific
945 commands and file names are slightly different.
947 First, run ssh-keygen on the client (as root) and server (as the BackupPC
948 user) and simply hit enter when prompted for the pass-phrase:
952 This will save the public key in /.ssh/identity.pub and the private
953 key in /.ssh/identity.
955 Next, append BackupPC's ~/.ssh/identity.pub (from the server) to root's
956 /.ssh/authorized_keys file on the client. It's a single long line that
957 you can cut-and-paste with an editor (make sure it remains a single line).
959 Next, force protocol version 1 by adding:
963 to BackupPC's ~/.ssh/config on the server.
965 Next, run "chmod -R go-rwx ~/.ssh" on the server and "chmod -R go-rwx /.ssh"
970 ssh -l root clientHostName whoami
974 Finally, if this machine uses DHCP you will need to run nmbd (the
975 NetBios name server) from the Samba distribution so that the machine
976 responds to a NetBios name request. See the manual page and Samba
977 documentation for more information.
981 =head2 Step 6: Running BackupPC
983 The installation contains an init.d backuppc script that can be copied
984 to /etc/init.d so that BackupPC can auto-start on boot.
985 See init.d/README for further instructions.
987 BackupPC should be ready to start. If you installed the init.d script,
988 then you should be able to run BackupPC with:
990 /etc/init.d/backuppc start
992 (This script can also be invoked with "stop" to stop BackupPC and "reload"
993 to tell BackupPC to reload config.pl and the hosts file.)
997 __INSTALLDIR__/bin/BackupPC -d
999 as user __BACKUPPCUSER__. The -d option tells BackupPC to run as a daemon
1000 (ie: it does an additional fork).
1002 Any immediate errors will be printed to stderr and BackupPC will quit.
1003 Otherwise, look in __TOPDIR__/log/LOG and verify that BackupPC reports
1004 it has started and all is ok.
1006 =head2 Step 7: Talking to BackupPC
1008 Note: as of version 1.5.0, BackupPC no longer supports telnet
1009 to its TCP port. First off, a unix domain socket is used
1010 instead of a TCP port. (The TCP port can still be re-enabled
1011 if your installation has apache and BackupPC running on different
1012 machines.) Secondly, even if you still use the TCP port, the
1013 messages exchanged over this interface are now protected by
1014 an MD5 digest based on a shared secret (see $Conf{ServerMesgSecret})
1015 as well as sequence numbers and per-session unique keys, preventing
1016 forgery and replay attacks.
1018 You should verify that BackupPC is running by using BackupPC_serverMesg.
1019 This sends a message to BackupPC via the unix (or TCP) socket and prints
1022 You can request status information and start and stop backups using this
1023 interface. This socket interface is mainly provided for the CGI interface
1024 (and some of the BackupPC sub-programs use it too). But right now we just
1025 want to make sure BackupPC is happy. Each of these commands should
1026 produce some status output:
1028 __INSTALLDIR__/bin/BackupPC_serverMesg status info
1029 __INSTALLDIR__/bin/BackupPC_serverMesg status jobs
1030 __INSTALLDIR__/bin/BackupPC_serverMesg status hosts
1032 The output should be some hashes printed with Data::Dumper. If it
1033 looks cryptic and confusing, and doesn't look like an error message,
1036 The jobs status should initially show just BackupPC_trashClean.
1037 The hosts status should produce a list of every host you have listed
1038 in __TOPDIR__/conf/hosts as part of a big cryptic output line.
1040 You can also request that all hosts be queued:
1042 __INSTALLDIR__/bin/BackupPC_serverMesg backup all
1044 At this point you should make sure the CGI interface works since
1045 it will be much easier to see what is going on. That's our
1048 =head2 Step 8: CGI interface
1050 The CGI interface script, BackupPC_Admin, is a powerful and flexible
1051 way to see and control what BackupPC is doing. It is written for an
1052 Apache server. If you don't have Apache, see L<http://www.apache.org>.
1054 There are two options for setting up the CGI interface: standard
1055 mode and using mod_perl. Mod_perl provides much higher performance
1056 (around 15x) and is the best choice if your Apache was built with
1057 mod_perl support. To see if your apache was built with mod_perl
1060 httpd -l | egrep mod_perl
1062 If this prints mod_perl.c then your Apache supports mod_perl.
1064 Using mod_perl with BackupPC_Admin requires a dedicated Apache
1065 to be run as the BackupPC user (__BACKUPPCUSER__). This is
1066 because BackupPC_Admin needs permission to access various files
1067 in BackupPC's data directories. In contrast, the standard
1068 installation (without mod_perl) solves this problem by having
1069 BackupPC_Admin installed as setuid to the BackupPC user, so that
1070 BackupPC_Admin runs as the BackuPC user.
1072 Here are some specifics for each setup:
1076 =item Standard Setup
1078 The CGI interface should have been installed by the configure.pl script
1079 in __CGIDIR__/BackupPC_Admin. BackupPC_Admin should have been installed
1080 as setuid to the BackupPC user (__BACKUPPCUSER__), in addition to user
1081 and group execute permission.
1083 You should be very careful about permissions on BackupPC_Admin and
1084 the directory __CGIDIR__: it is important that normal users cannot
1085 directly execute or change BackupPC_Admin, otherwise they can access
1086 backup files for any PC. You might need to change the group ownership
1087 of BackupPC_Admin to a group that Apache belongs to so that Apache
1088 can execute it (don't add "other" execute permission!).
1089 The permissions should look like this:
1091 ls -l __CGIDIR__/BackupPC_Admin
1092 -swxr-x--- 1 __BACKUPPCUSER__ web 82406 Jun 17 22:58 __CGIDIR__/BackupPC_Admin
1094 The setuid script won't work unless perl on your machine was installed
1095 with setuid emulation. This is likely the problem if you get an error
1096 saying such as "Wrong user: my userid is 25, instead of 150", meaning
1097 the script is running as the httpd user, not the BackupPC user.
1098 This is because setuid scripts are disabled by the kernel in most
1099 flavors of unix and linux.
1101 To see if your perl has setuid emulation, see if there is a program
1102 called sperl5.6.0 or sperl5.6.1 in the place where perl is installed.
1103 If you can't find this program, then you have two options: rebuild
1104 and reinstall perl with the setuid emulation turned on (answer "y" to
1105 the question "Do you want to do setuid/setgid emulation?" when you
1106 run perl's configure script), or switch to the mod_perl alternative
1107 for the CGI script (which doesn't need setuid to work).
1109 =item Mod_perl Setup
1111 The advantage of the mod_perl setup is that no setuid script is needed,
1112 and there is a huge performance advantage. Not only does all the perl
1113 code need to be parsed just once, the config.pl and hosts files, plus
1114 the connection to the BackupPC server are cached between requests. The
1115 typical speedup is around 15 times.
1117 To use mod_perl you need to run Apache as user __BACKUPPCUSER__.
1118 If you need to run multiple Apache's for different services then
1119 you need to create multiple top-level Apache directories, each
1120 with their own config file. You can make copies of /etc/init.d/httpd
1121 and use the -d option to httpd to point each http to a different
1122 top-level directory. Or you can use the -f option to explicitly
1123 point to the config file. Multiple Apache's will run on different
1124 Ports (eg: 80 is standard, 8080 is a typical alternative port accessed
1125 via http://yourhost.com:8080).
1127 Inside BackupPC's Apache http.conf file you should check the
1128 settings for ServerRoot, DocumentRoot, User, Group, and Port. See
1129 L<http://httpd.apache.org/docs/server-wide.html> for more details.
1131 For mod_perl, BackupPC_Admin should not have setuid permission, so
1132 you should turn it off:
1134 chmod u-s __CGIDIR__/BackupPC_Admin
1136 To tell Apache to use mod_perl to execute BackupPC_Admin, add this
1137 to Apache's httpd.conf file:
1139 <IfModule mod_perl.c>
1140 PerlModule Apache::Registry
1142 <Location /cgi-bin/BackupPC/BackupPC_Admin> # <--- change path as needed
1143 SetHandler perl-script
1144 PerlHandler Apache::Registry
1149 There are other optimizations and options with mod_perl. For
1150 example, you can tell mod_perl to preload various perl modules,
1151 which saves memory compared to loading separate copies in every
1152 Apache process after they are forked. See Stas's definitive
1153 mod_perl guide at L<http://perl.apache.org/guide>.
1157 BackupPC_Admin requires that users are authenticated by Apache.
1158 Specifically, it expects that Apache sets the REMOTE_USER environment
1159 variable when it runs. There are several ways to do this. One way
1160 is to create a .htaccess file in the cgi-bin directory that looks like:
1162 AuthGroupFile /etc/httpd/conf/group # <--- change path as needed
1163 AuthUserFile /etc/http/conf/passwd # <--- change path as needed
1168 You will also need "AllowOverride Indexes AuthConfig" in the Apache
1169 httpd.conf file to enable the .htaccess file. Alternatively, everything
1170 can go in the Apache httpd.conf file inside a Location directive. The
1171 list of users and password file above can be extracted from the NIS
1174 One alternative is to use LDAP. In Apache's http.conf add these lines:
1176 LoadModule auth_ldap_module modules/auth_ldap.so
1177 AddModule auth_ldap.c
1179 # cgi-bin - auth via LDAP (for BackupPC)
1180 <Location /cgi-binBackupPC/BackupPC_Admin> # <--- change path as needed
1182 AuthName "BackupPC login"
1183 # replace MYDOMAIN, PORT, ORG and CO as needed
1184 AuthLDAPURL ldap://ldap.MYDOMAIN.com:PORT/o=ORG,c=CO?uid?sub?(objectClass=*)
1188 If you want to defeat the user authentication you can force a
1189 particular user name by getting Apache to set REMOTE_USER, eg,
1190 to hardcode the user to www you could add this to httpd.conf:
1192 <Location /cgi-bin/BackupPC/BackupPC_Admin> # <--- change path as needed
1193 Setenv REMOTE_USER www
1196 Finally, you should also edit the config.pl file and adjust, as necessary,
1197 the CGI-specific settings. They're near the end of the config file. In
1198 particular, you should specify which users or groups have administrator
1199 (privileged) access. Also, the configure.pl script placed various
1200 images into $Conf{CgiImageDir} that BackupPC_Admin needs to serve
1201 up. You should make sure that $Conf{CgiImageDirURL} is the correct
1202 URL for the image directory.
1204 =head2 Other installation topics
1208 =item Copying the pool
1210 If the pool disk requirements grow you might need to copy the entire
1211 data directory to a new (bigger) file system. Hopefully you are lucky
1212 enough to avoid this by having the data directory on a RAID file system
1213 or LVM that allows the capacity to be grown in place by adding disks.
1215 The backup data directories contain large numbers of hardlinks. If
1216 you try to copy the pool the target directory will occupy a lot more
1217 space if the hardlinks aren't re-established.
1219 The GNU cp program with the -a option is aware of hardlinks and knows
1220 to re-establish them. So GNU cp -a is the recommended way to copy
1221 the data directory and pool. Don't forget to stop BackupPC while
1224 =item Compressing an existing pool
1226 If you are upgrading BackupPC and want to turn compression on you have
1233 Simply turn on compression. All new backups will be compressed. Both old
1234 (uncompressed) and new (compressed) backups can be browsed and viewed.
1235 Eventually, the old backups will expire and all the pool data will be
1236 compressed. However, until the old backups expire, this approach could
1237 require 60% or more additional pool storage space to store both
1238 uncompressed and compressed versions of the backup files.
1242 Convert all the uncompressed pool files and backups to compressed.
1243 The script __INSTALLDIR__/bin/BackupPC_compressPool does this.
1244 BackupPC must not be running when you run BackupPC_compressPool.
1245 Also, there must be no existing compressed backups when you
1246 run BackupPC_compressPool.
1248 BackupPC_compressPool compresses all the files in the uncompressed pool
1249 (__TOPDIR__/pool) and moves them to the compressed pool
1250 (__TOPDIR__/cpool). It rewrites the files in place, so that the
1251 existing hardlinks are not disturbed.
1255 The rest of this section discusses how to run BackupPC_compressPool.
1257 BackupPC_compressPool takes three command line options:
1263 Test mode: do everything except actually replace the pool files.
1264 Useful for estimating total run time without making any real
1269 Read check: re-read the compressed file and compare it against
1270 the original uncompressed file. Can only be used in test mode.
1274 Number of children to fork. BackupPC_compressPool can take a long time
1275 to run, so to speed things up it spawns four children, each working on a
1276 different part of the pool. You can change the number of children with
1281 Here are the recommended steps for running BackupPC_compressPool:
1287 Stop BackupPC (eg: "/etc/init.d/backuppc stop").
1291 Set $Conf{CompressLevel} to a non-zero number (eg: 3).
1295 Do a dry run of BackupPC_compressPool. Make sure you run this as
1296 the BackupPC user (__BACKUPPCUSER__):
1298 BackupPC_compressPool -t -r
1300 The -t option (test mode) makes BackupPC_compressPool do all the steps,
1301 but not actually change anything. The -r option re-reads the compressed
1302 file and compares it against the original.
1304 BackupPC_compressPool gives a status as it completes each 1% of the job.
1305 It also shows the cumulative compression ratio and estimated completion
1306 time. Once you are comfortable that things look ok, you can kill
1307 BackupPC_compressPool or wait for it to finish.
1311 Now you are ready to run BackupPC_compressPool for real. Once again,
1312 as the BackupPC user (__BACKUPPCUSER__), run:
1314 BackupPC_compressPool
1316 You should put the output into a file and tail this file. (The running
1317 time could be twice as long as the test mode since the test mode file
1318 writes are immediately followed by an unlink, so in test mode it is
1319 likely the file writes never make it to disk.)
1321 It is B<critical> that BackupPC_compressPool runs to completion before
1322 re-starting BackupPC. Before BackupPC_compressPool completes, none of
1323 the existing backups will be in a consistent state. If you must stop
1324 BackupPC_compressPool for some reason, send it an INT or TERM signal
1325 and give it several seconds (or more) to clean up gracefully.
1326 After that, you can re-run BackupPC_compressPool and it will start
1327 again where it left off. Once again, it is critical that it runs
1332 After BackupPC_compressPool completes you should have a complete set
1333 of compressed backups (and your disk usage should be lower). You
1334 can now re-start BackupPC.
1338 =head2 Debugging installation problems
1340 This section will probably grow based on the types of questions on
1341 the BackupPC mail list.
1343 Assuming BackupPC can start correctly you should inspect __TOPDIR__/log/LOG
1344 for any errors. Assuming backups for a particular host start, you
1345 should be able to look in __TOPDIR__/pc/$host/LOG for error messages
1346 specific to that host.
1348 The most likely problems will relate to connecting to the smb shares on
1349 each host. On each failed backup, a file __TOPDIR__/pc/$host/XferERR will
1350 be created. This is the stderr output from smbclient. The first line
1351 will show the full smbclient command that was run. Based on the error
1352 messages you should figure out what is wrong. Possible errors on the
1353 server side are invalid host, invalid share name, bad username or password.
1354 Possible errors on the client side are misconfiguration of the share,
1355 username or password.
1357 You should run smbclient manually and verify that you can connect to
1358 the host in interactive mode, eg:
1360 smbclient '\\hostName\shareName' -U userName
1362 shareName should match the $Conf{SmbShareName} setting and userName
1363 should match the the $Conf{SmbShareUserName} setting.
1365 You will be prompted for the password. You should then see this prompt:
1369 Verify that "ls" works and then type "quit" to exit.
1371 Secondly, you should also verify that nmblookup correctly returns
1372 the netbios name. This is essential for DHCP hosts, and depending
1373 upon the setting of $Conf{FixedIPNetBiosNameCheck} might also be
1374 required for fixed IP address hosts too. Run this command:
1376 nmblookup -A hostName
1378 Verify that the host name is printed. The output might look like:
1381 DELLLS13 <00> - P <ACTIVE>
1382 DOMAINNAME <00> - <GROUP> P <ACTIVE>
1383 DELLLS13 <20> - P <ACTIVE>
1384 DOMAINNAME <1e> - <GROUP> P <ACTIVE>
1385 DELLLS13 <03> - P <ACTIVE>
1386 DELLLS13$ <03> - P <ACTIVE>
1387 CRAIG <03> - P <ACTIVE>
1389 The first name, converted to lower case, is used for the host name.
1391 =head1 Restore functions
1393 BackupPC supports several different methods for restoring files. The
1394 most convenient restore options are provided via the CGI interface.
1395 Alternatively, backup files can be restored using manual commands.
1397 =head2 CGI restore options
1399 By selecting a host in the CGI interface, a list of all the backups
1400 for that machine will be displayed. By selecting the backup number
1401 you can navigate the shares and directory tree for that backup.
1403 BackupPC's CGI interface automatically fills incremental backups
1404 with the corresponding full backup, which means each backup has
1405 a filled appearance. Therefore, there is no need to do multiple
1406 restores from the incremental and full backups: BackupPC does all
1407 the hard work for you. You simply select the files and directories
1408 you want from the correct backup vintage in one step.
1410 You can download a single backup file at any time simply by selecting
1411 it. Your browser should prompt you with the file name and ask you
1412 whether to open the file or save it to disk.
1414 Alternatively, you can select one or more files or directories in
1415 the currently selected directory and select "Restore selected files".
1416 (If you need to restore selected files and directories from several
1417 different parent directories you will need to do that in multiple
1420 If you select all the files in a directory, BackupPC will replace
1421 the list of files with the parent directory. You will be presented
1422 with a screen that has three options:
1426 =item Option 1: Direct Restore
1428 With this option the selected files and directories are restored
1429 directly back onto the host, by default in their original location.
1430 Any old files with the same name will be overwritten, so use caution.
1431 You can optionally change the target host name, target share name,
1432 and target path prefix for the restore, allowing you to restore the
1433 files to a different location.
1435 Once you select "Start Restore" you will be prompted one last time
1436 with a summary of the exact source and target files and directories
1437 before you commit. When you give the final go ahead the restore
1438 operation will be queued like a normal backup job, meaning that it
1439 will be deferred if there is a backup currently running for that host.
1440 When the restore job is run, smbclient or tar is used (depending upon
1441 $Conf{XferMethod}) to actually restore the files. Sorry, there is
1442 currently no option to cancel a restore that has been started.
1444 A record of the restore request, including the result and list of
1445 files and directories, is kept. It can be browsed from the host's
1446 home page. $Conf{RestoreInfoKeepCnt} specifies how many old restore
1447 status files to keep.
1449 =item Option 2: Download Zip archive
1451 With this option a zip file containing the selected files and directories
1452 is downloaded. The zip file can then be unpacked or individual files
1453 extracted as necessary on the host machine. The compression level can be
1454 specified. A value of 0 turns off compression.
1456 When you select "Download Zip File" you should be prompted where to
1457 save the restore.zip file.
1459 BackupPC does not consider downloading a zip file as an actual
1460 restore operation, so the details are not saved for later browsing
1461 as in the first case. However, a mention that a zip file was
1462 downloaded by a particular user, and a list of the files, does
1463 appear in BackupPC's log file.
1465 =item Option 3: Download Tar archive
1467 This is identical to the previous option, except a tar file is downloaded
1468 rather than a zip file (and there is currently no compression option).
1472 =head2 Command-line restore options
1474 Apart from the CGI interface, BackupPC allows you to restore files
1475 and directories from the command line. The following programs can
1482 For each file name argument it inflates (uncompresses) the file and
1483 writes it to stdout. To use BackupPC_zcat you could give it the
1486 __INSTALLDIR__/bin/BackupPC_zcat __TOPDIR__/pc/host/5/fc/fcraig/fexample.txt > example.txt
1488 It's your responsibility to make sure the file is really compressed:
1489 BackupPC_zcat doesn't check which backup the requested file is from.
1491 =item BackupPC_tarCreate
1493 BackupPC_tarCreate creates a tar file for any files or directories in
1494 a particular backup. Merging of incrementals is done automatically,
1495 so you don't need to worry about whether certain files appear in the
1496 incremental or full backup.
1500 BackupPC_tarCreate [-t] [-h host] [-n dumpNum] [-s shareName]
1501 [-r pathRemove] [-p pathAdd]
1502 files/directories...
1504 The command-line files and directories are relative to the specified
1505 shareName. The tar file is written to stdout.
1507 The required options are:
1513 host from which the tar archive is created
1517 dump number from which the tar archive is created
1521 share name from which the tar archive is created
1531 print summary totals
1535 path prefix that will be replaced with pathAdd
1543 The -h, -n and -s options specify which dump is used to generate
1544 the tar archive. The -r and -p options can be used to relocate
1545 the paths in the tar archive so extracted files can be placed
1546 in a location different from their original location.
1548 =item BackupPC_zipCreate
1550 BackupPC_zipCreate creates a zip file for any files or directories in
1551 a particular backup. Merging of incrementals is done automatically,
1552 so you don't need to worry about whether certain files appear in the
1553 incremental or full backup.
1557 BackupPC_zipCreate [-t] [-h host] [-n dumpNum] [-s shareName]
1558 [-r pathRemove] [-p pathAdd] [-c compressionLevel]
1559 files/directories...
1561 The command-line files and directories are relative to the specified
1562 shareName. The zip file is written to stdout.
1564 The required options are:
1570 host from which the zip archive is created
1574 dump number from which the zip archive is created
1578 share name from which the zip archive is created
1588 print summary totals
1592 path prefix that will be replaced with pathAdd
1600 compression level (default is 0, no compression)
1604 The -h, -n and -s options specify which dump is used to generate
1605 the zip archive. The -r and -p options can be used to relocate
1606 the paths in the zip archive so extracted files can be placed
1607 in a location different from their original location.
1611 Each of these programs reside in __INSTALLDIR__/bin.
1613 =head1 BackupPC Design
1615 =head2 Some design issues
1619 =item Pooling common files
1621 To quickly see if a file is already in the pool, an MD5 digest of the
1622 file length and contents is used as the file name in the pool. This
1623 can't guarantee a file is identical: it just reduces the search to
1624 often a single file or handful of files. A complete file comparison
1625 is always done to verify if two files are really the same.
1627 Identical files on multiples backups are represented by hard links.
1628 Hardlinks are used so that identical files all refer to the same
1629 physical file on the server's disk. Also, hard links maintain
1630 reference counts so that BackupPC knows when to deleted unused files
1633 For the computer-science majors among you, you can think of the pooling
1634 system used by BackupPC as just a chained hash table stored on a (big)
1637 =item The hashing function
1639 There is a tradeoff between how much of file is used for the MD5 digest
1640 and the time taken comparing all the files that have the same hash.
1642 Using the file length and just the first 4096 bytes of the file for the
1643 MD5 digest produces some repetitions. One example: with 900,000 unique
1644 files in the pool, this hash gives about 7,000 repeated files, and in
1645 the worst case 500 files have the same hash. That's not bad: we only
1646 have to do a single file compare 99.2% of the time. But in the worst
1647 case we have to compare as many as 500 files checking for a match.
1649 With a modest increase in CPU time, if we use the file length and the
1650 first 256K of the file we now only have 500 repeated files and in the
1651 worst case around 20 files have the same hash. Furthermore, if we
1652 instead use the first and last 128K of the file (more specifically, the
1653 first and eighth 128K chunks for files larger than 1MB) we get only 300
1654 repeated files and in the worst case around 20 files have the same hash.
1656 Based on this experimentation, this is the hash function used by BackupPC.
1657 It is important that you don't change the hash function after files
1658 are already in the pool. Otherwise your pool will grow to twice the
1659 size until all the old backups (and all the old files with old hashes)
1664 BackupPC supports compression. It uses the deflate and inflate methods
1665 in the Compress::Zlib module, which is based on the zlib compression
1666 library (see L<http://www.gzip.org/zlib/>).
1668 The $Conf{CompressLevel} setting specifies the compression level to use.
1669 Zero (0) means no compression. Compression levels can be from 1 (least
1670 cpu time, slightly worse compression) to 9 (most cpu time, slightly
1671 better compression). The recommended value is 3. Changing it to 5, for
1672 example, will take maybe 20% more cpu time and will get another 2-3%
1673 additional compression. Diminishing returns set in above 5. See the zlib
1674 documentation for more information about compression levels.
1676 BackupPC implements compression with minimal CPU load. Rather than
1677 compressing every incoming backup file and then trying to match it
1678 against the pool, BackupPC computes the MD5 digest based on the
1679 uncompressed file, and matches against the candidate pool files by
1680 comparing each uncompressed pool file against the incoming backup file.
1681 Since inflating a file takes roughly a factor of 10 less CPU time than
1682 deflating there is a big saving in CPU time.
1684 The combination of pooling common files and compression can yield
1685 a factor of 8 or more overall saving in backup storage.
1689 =head2 BackupPC operation
1691 BackupPC reads the configuration information from
1692 __TOPDIR__/conf/config.pl. It then runs and manages all the backup
1693 activity. It maintains queues of pending backup requests, user backup
1694 requests and administrative commands. Based on the configuration various
1695 requests will be executed simultaneously.
1697 As specified by $Conf{WakeupSchedule}, BackupPC wakes up periodically
1698 to queue backups on all the PCs. This is a four step process:
1704 For each host and DHCP address backup requests are queued on the
1705 background command queue.
1709 For each PC, BackupPC_dump is forked. Several of these may be run in
1710 parallel, based on the configuration. First a ping is done to see if the
1711 machine is alive. If this is a DHCP address, nmblookup is run to get
1712 the netbios name, which is used as the host name. The file
1713 __TOPDIR__/pc/$host/backups is read to decide whether a full or
1714 incremental backup needs to be run. If no backup is scheduled, or the ping
1715 to $host fails, then BackupPC_dump exits.
1717 The backup is done using samba's smbclient or tar over ssh/rsh/nfs piped
1718 into BackupPC_tarExtract, extracting the backup into __TOPDIR__/pc/$host/new.
1719 The smbclient or tar output is put into __TOPDIR__/pc/$host/XferLOG.
1721 As BackupPC_tarExtract extracts the files from smbclient, it checks each
1722 file in the backup to see if it is identical to an existing file from
1723 any previous backup of any PC. It does this without needed to write the
1724 file to disk. If the file matches an existing file, a hardlink is
1725 created to the existing file in the pool. If the file does not match any
1726 existing files, the file is written to disk and the file name is saved
1727 in __TOPDIR__/pc/$host/NewFileList for later processing by
1728 BackupPC_link. BackupPC_tarExtract can handle arbitrarily large
1729 files and multiple candidate matching files without needing to
1730 write the file to disk in the case of a match. This significantly
1731 reduces disk writes (and also reads, since the pool file comparison
1732 is done disk to memory, rather than disk to disk).
1734 Based on the configuration settings, BackupPC_dump checks each
1735 old backup to see if any should be removed. Any expired backups
1736 are moved to __TOPDIR__/trash for later removal by BackupPC_trashClean.
1740 For each complete, good, backup, BackupPC_link is run.
1741 To avoid race conditions as new files are linked into the
1742 pool area, only a single BackupPC_link program runs
1743 at a time and the rest are queued.
1745 BackupPC_link reads the NewFileList written by BackupPC_dump and
1746 inspects each new file in the backup. It re-checks if there is a
1747 matching file in the pool (another BackupPC_link
1748 could have added the file since BackupPC_dump checked). If so, the file
1749 is removed and replaced by a hard link to the existing file. If the file
1750 is new, a hard link to the file is made in the pool area, so that this
1751 file is available for checking against each new file and new backup.
1753 Then, assuming $Conf{IncrFill} is set, for each incremental backup,
1754 hard links are made in the new backup to all files that were not extracted
1755 during the incremental backups. The means the incremental backup looks
1756 like a complete image of the PC (with the exception that files
1757 that were removed on the PC since the last full backup will still
1758 appear in the backup directory tree).
1760 As of v1.03, the CGI interface knows how to merge unfilled
1761 incremental backups will the most recent prior filled (full)
1762 backup, giving the incremental backups a filled appearance. The
1763 default for $Conf{IncrFill} is off, since there is now no need to
1764 fill incremental backups. This saves some level of disk activity,
1765 since lots of extra hardlinks are no longer needed (and don't have
1766 to be deleted when the backup expires).
1770 BackupPC_trashClean is always run in the background to remove any
1771 expired backups. Every 5 minutes it wakes up and removes all the files
1772 in __TOPDIR__/trash.
1774 Also, once each night, BackupPC_nightly is run to complete some additional
1775 administrative tasks, such as cleaning the pool. This involves removing
1776 any files in the pool that only have a single hard link (meaning no backups
1777 are using that file). Again, to avoid race conditions, BackupPC_nightly
1778 is only run when there are no BackupPC_dump or BackupPC_link processes
1783 BackupPC also listens for TCP connections on $Conf{ServerPort}, which
1784 is used by the CGI script BackupPC_Admin for status reporting and
1785 user-initiated backup or backup cancel requests.
1787 =head2 Storage layout
1789 BackupPC resides in three directories:
1793 =item __INSTALLDIR__
1795 Perl scripts comprising BackupPC reside in __INSTALLDIR__/bin,
1796 libraries are in __INSTALLDIR__/lib and documentation
1797 is in __INSTALLDIR__/doc.
1801 The CGI script BackupPC_Admin resides in this cgi binary directory.
1805 All of BackupPC's data (PC backup images, logs, configuration information)
1806 is stored below this directory.
1810 Below __TOPDIR__ are several directories:
1814 =item __TOPDIR__/conf
1816 The directory __TOPDIR__/conf contains:
1822 Configuration file. See L<Configuration file|configuration file>
1823 below for more details.
1827 Hosts file, which lists all the PCs to backup.
1831 =item __TOPDIR__/log
1833 The directory __TOPDIR__/log contains:
1839 Current (today's) log file output from BackupPC.
1841 =item LOG.0 or LOG.0.z
1843 Yesterday's log file output. Log files are aged daily and compressed
1844 (if compression is enabled), and old LOG files are deleted.
1848 Contains BackupPC's process id.
1852 A summary of BackupPC's status written periodically by BackupPC so
1853 that certain state information can be maintained if BackupPC is
1854 restarted. Should not be edited.
1856 =item UserEmailInfo.pl
1858 A summary of what email was last sent to each user, and when the
1859 last email was sent. Should not be edited.
1863 =item __TOPDIR__/trash
1865 Any directories and files below this directory are periodically deleted
1866 whenever BackupPC_trashClean checks. When a backup is aborted or when an
1867 old backup expires, BackupPC_dump simply moves the directory to
1868 __TOPDIR__/trash for later removal by BackupPC_trashClean.
1870 =item __TOPDIR__/pool
1872 All uncompressed files from PC backups are stored below __TOPDIR__/pool.
1873 Each file's name is based on the MD5 hex digest of the file contents.
1874 Specifically, for files less than 256K, the file length and the entire
1875 file is used. For files up to 1MB, the file length and the first and
1876 last 128K are used. Finally, for files longer than 1MB, the file length,
1877 and the first and eighth 128K chunks for the file are used.
1879 Each file is stored in a subdirectory X/Y/Z, where X, Y, Z are the
1880 first 3 hex digits of the MD5 digest.
1882 For example, if a file has an MD5 digest of 123456789abcdef0,
1883 the file is stored in __TOPDIR__/pool/1/2/3/123456789abcdef0.
1885 The MD5 digest might not be unique (especially since not all the file's
1886 contents are used for files bigger than 256K). Different files that have
1887 the same MD5 digest are stored with a trailing suffix "_n" where n is
1888 an incrementing number starting at 0. So, for example, if two additional
1889 files were identical to the first, except the last byte was different,
1890 and assuming the file was larger than 1MB (so the MD5 digests are the
1891 same but the files are actually different), the three files would be
1894 __TOPDIR__/pool/1/2/3/123456789abcdef0
1895 __TOPDIR__/pool/1/2/3/123456789abcdef0_0
1896 __TOPDIR__/pool/1/2/3/123456789abcdef0_1
1898 Both BackupPC_dump (actually, BackupPC_tarExtract) and BackupPC_link are
1899 responsible for checking newly backed up files against the pool. For
1900 each file, the MD5 digest is used to generate a file name in the pool
1901 directory. If the file exists in the pool, the contents are compared.
1902 If there is no match, additional files ending in "_n" are checked.
1903 (Actually, BackupPC_tarExtract compares multiple candidate files in
1904 parallel.) If the file contents exactly match, the file is created by
1905 simply making a hard link to the pool file (this is done by
1906 BackupPC_tarExtract as the backup proceeds). Otherwise,
1907 BackupPC_tarExtract writes the new file to disk and a new hard link is
1908 made in the pool to the file (this is done later by BackupPC_link).
1910 Therefore, every file in the pool will have at least 2 hard links
1911 (one for the pool file and one for the backup file below __TOPDIR__/pc).
1912 Identical files from different backups or PCs will all be linked to
1913 the same file. When old backups are deleted, some files in the pool
1914 might only have one link. BackupPC_nightly checks the entire pool
1915 and removes all files that have only a single link, thereby recovering
1916 the storage for that file.
1918 One other issue: zero length files are not pooled, since there are a lot
1919 of these files and on most file systems it doesn't save any disk space
1920 to turn these files into hard links.
1922 =item __TOPDIR__/cpool
1924 All compressed files from PC backups are stored below __TOPDIR__/cpool.
1925 Its layout is the same as __TOPDIR__/pool, and the hashing function
1926 is the same (and, importantly, based on the uncompressed file, not
1927 the compressed file).
1929 =item __TOPDIR__/pc/$host
1931 For each PC $host, all the backups for that PC are stored below
1932 the directory __TOPDIR__/pc/$host. This directory contains the
1939 Current log file for this PC from BackupPC_dump.
1941 =item LOG.0 or LOG.0.z
1943 Last month's log file. Log files are aged monthly and compressed
1944 (if compression is enabled), and old LOG files are deleted.
1946 =item XferERR or XferERR.z
1948 Output from the transport program (ie: smbclient or tar)
1949 for the most recent failed backup.
1953 Subdirectory in which the current backup is stored. This
1954 directory is renamed if the backup succeeds.
1956 =item XferLOG or XferLOG.z
1958 Output from the transport program (ie: smbclient or tar)
1959 for the current backup.
1961 =item nnn (an integer)
1963 Successful backups are in directories numbered sequentially starting at 0.
1965 =item XferLOG.nnn or XferLOG.nnn.z
1967 Output from the transport program (ie: smbclient or tar)
1968 corresponding to backup number nnn.
1970 =item RestoreInfo.nnn
1972 Information about restore request #nnn including who, what, when, and
1973 why. This file is in Data::Dumper format. (Note that the restore
1974 numbers are not related to the backup number.)
1976 =item RestoreLOG.nnn.z
1978 Output from smbclient or tar during restore #nnn. (Note that the restore
1979 numbers are not related to the backup number.)
1983 Optional configuration settings specific to this host. Settings in this
1984 file override the main configuration file.
1988 A tab-delimited ascii table listing information about each successful
1989 backup, one per row. The columns are:
1995 The backup number, an integer that starts at 0 and increments
1996 for each successive backup. The corresponding backup is stored
1997 in the directory num (eg: if this field is 5, then the backup is
1998 stored in __TOPDIR__/pc/$host/5).
2002 Set to "full" or "incr" for full or incremental backup.
2006 Start time of the backup in unix seconds.
2010 Stop time of the backup in unix seconds.
2014 Number of files backed up (as reported by smbclient or tar).
2018 Total file size backed up (as reported by smbclient or tar).
2022 Number of files that were already in the pool
2023 (as determined by BackupPC_dump and BackupPC_link).
2027 Total size of files that were already in the pool
2028 (as determined by BackupPC_dump and BackupPC_link).
2032 Number of files that were not in the pool
2033 (as determined by BackupPC_link).
2037 Total size of files that were not in the pool
2038 (as determined by BackupPC_link).
2042 Number of errors or warnings from smbclient (zero for tar).
2046 Number of errors from smbclient that were bad file errors (zero for tar).
2050 Number of errors from smbclient that were bad share errors (zero for tar).
2054 Number of errors from BackupPC_tarExtract.
2058 The compression level used on this backup. Zero or empty means no
2063 Total compressed size of files that were already in the pool
2064 (as determined by BackupPC_dump and BackupPC_link).
2068 Total compressed size of files that were not in the pool
2069 (as determined by BackupPC_link).
2073 Set if this backup has not been filled in with the most recent
2074 previous filled or full backup. See $Conf{IncrFill}.
2078 If this backup was filled (ie: noFill is 0) then this is the
2079 number of the backup that it was filled from
2083 Set if this backup has mangled file names and attributes. Always
2084 true for backups in v1.4.0 and above. False for all backups prior
2091 A tab-delimited ascii table listing information about each requested
2092 restore, one per row. The columns are:
2098 Restore number (matches the suffix of the RestoreInfo.nnn and
2099 RestoreLOG.nnn.z file), unrelated to the backup number.
2103 Start time of the restore in unix seconds.
2107 End time of the restore in unix seconds.
2111 Result (ok or failed).
2115 Error message if restore failed.
2119 Number of files restored.
2123 Size in bytes of the restored files.
2127 Number of errors from BackupPC_tarCreate during restore.
2131 Number of errors from smbclient or tar during restore.
2139 =head2 Compressed file format
2141 The compressed file format is as generated by Compress::Zlib::deflate
2142 with one minor, but important, tweak. Since Compress::Zlib::inflate
2143 fully inflates its argument in memory, it could take large amounts of
2144 memory if it was inflating a highly compressed file. For example, a
2145 200MB file of 0x0 bytes compresses to around 200K bytes. If
2146 Compress::Zlib::inflate was called with this single 200K buffer, it
2147 would need to allocate 200MB of memory to return the result.
2149 BackupPC watches how efficiently a file is compressing. If a big file
2150 has very high compression (meaning it will use too much memory when it
2151 is inflated), BackupPC calls the flush() method, which gracefully
2152 completes the current compression. BackupPC then starts another
2153 deflate and simply appends the output file. So the BackupPC compressed
2154 file format is one or more concatenated deflations/flushes. The specific
2155 ratios that BackupPC uses is that if a 6MB chunk compresses to less
2156 than 64K then a flush will be done.
2158 Back to the example of the 200MB file of 0x0 bytes. Adding flushes
2159 every 6MB adds only 200 or so bytes to the 200K output. So the
2160 storage cost of flushing is negligible.
2162 To easily decompress a BackupPC compressed file, the script
2163 BackupPC_zcat can be found in __INSTALLDIR__/bin. For each
2164 file name argument it inflates the file and writes it to stdout.
2166 =head2 File name mangling
2168 Backup file names are stored in "mangled" form. Each node of
2169 a path is preceded by "f" (mnemonic: file), and special characters
2170 (\n, \r, % and /) are URI-encoded as "%xx", where xx is the ascii
2171 character's hex value. So c:/craig/example.txt is now stored as
2172 fc/fcraig/fexample.txt.
2174 This was done mainly so meta-data could be stored alongside the backup
2175 files without name collisions. In particular, the attributes for the
2176 files in a directory are stored in a file called "attrib", and mangling
2177 avoids file name collisions (I discarded the idea of having a duplicate
2178 directory tree for every backup just to store the attributes). Other
2179 meta-data (eg: rsync checksums) could be stored in file names preceded
2180 by, eg, "c". There are two other benefits to mangling: the share name
2181 might contain "/" (eg: "/home/craig" for tar transport), and I wanted
2182 that represented as a single level in the storage tree. Secondly, as
2183 files are written to NewFileList for later processing by BackupPC_link,
2184 embedded newlines in the file's path will cause problems which are
2185 avoided by mangling.
2187 The CGI script undoes the mangling, so it is invisible to the user.
2188 Old (unmangled) backups are still supported by the CGI
2191 =head2 Special files
2193 Linux/unix file systems support several special file types: symbolic
2194 links, character and block device files, fifos (pipes) and unix-domain
2195 sockets. All except unix-domain sockets are supported by BackupPC
2196 (there's no point in backing up or restoring unix-domain sockets since
2197 they only have meaning after a process creates them). Symbolic links are
2198 stored as a plain file whose contents are the contents of the link (not
2199 the file it points to). This file is compressed and pooled like any
2200 normal file. Character and block device files are also stored as plain
2201 files, whose contents are two integers separated by a comma; the numbers
2202 are the major and minor device number. These files are compressed and
2203 pooled like any normal file. Fifo files are stored as empty plain files
2204 (which are not pooled since they have zero size). In all cases, the
2205 original file type is stored in the attrib file so it can be correctly
2208 Hardlinks are also supported. When GNU tar first encounters a file with
2209 more than one link (ie: hardlinks) it dumps it as a regular file. When
2210 it sees the second and subsequent hardlinks to the same file, it dumps
2211 just the hardlink information. BackupPC correctly recognizes these
2212 hardlinks and stores them just like symlinks: a regular text file
2213 whose contents is the path of the file linked to. The CGI script
2214 will download the original file when you click on a hardlink.
2216 Also, BackupPC_tarCreate has enough magic to re-create the hardlinks
2217 dynamically based on whether or not the original file and hardlinks
2218 are both included in the tar file. For example, imagine a/b/x is a
2219 hardlink to a/c/y. If you use BackupPC_tarCreate to restore directory
2220 a, then the tar file will include a/b/x as the original file and a/c/y
2221 will be a hardlink to a/b/x. If, instead you restore a/c, then the
2222 tar file will include a/c/y as the original file, not a hardlink.
2224 =head2 Attribute file format
2226 The unix attributes for the contents of a directory (all the files and
2227 directories in that directory) are stored in a file called attrib.
2228 There is a single attrib file for each directory in a backup.
2229 For example, if c:/craig contains a single file c:/craig/example.txt,
2230 that file would be stored as fc/fcraig/fexample.txt and there would be an
2231 attribute file in fc/fcraig/attrib (and also fc/attrib and ./attrib).
2232 The file fc/fcraig/attrib would contain a single entry containing the
2233 attributes for fc/fcraig/fexample.txt.
2235 The attrib file starts with a magic number, followed by the
2236 concatenation of the following information for each file:
2242 File name length in perl's pack "w" format (variable length base 128).
2250 The unix file type, mode, uid, gid and file size divided by 4GB and
2251 file size modulo 4GB (type mode uid gid sizeDiv4GB sizeMod4GB),
2252 in perl's pack "w" format (variable length base 128).
2256 The unix mtime (unix seconds) in perl's pack "N" format (32 bit integer).
2260 The attrib file is also compressed if compression is enabled.
2261 See the lib/BackupPC/Attrib.pm module for full details.
2263 Attribute files are pooled just like normal backup files. This saves
2264 space if all the files in a directory have the same attributes across
2265 multiple backups, which is common.
2269 BackupPC isn't perfect (but it is getting better). Here are some
2270 limitations of BackupPC:
2274 =item Non-unix file attributes not backed up
2276 smbclient doesn't extract the WinXX ACLs, so file attributes other than
2277 the equivalent (as provided by smbclient) unix attributes are not
2280 =item Locked files are not backed up
2282 Under WinXX a locked file cannot be read by smbclient. Such files will
2283 not be backed up. This includes the WinXX system registry files.
2285 This is especially troublesome for Outlook, which stores all its data
2286 in a single large file and keeps it locked whenever it is running.
2287 Since many users keep Outlook running all the time their machine
2288 is up their Outlook file will not be backed up. Sadly, this file
2289 is the most important file to backup. As one workaround, Microsoft has
2290 a user-level application that periodically asks the user if they want to
2291 make a copy of their outlook.pst file. This copy can then be backed up
2292 by BackupPC. See L<http://office.microsoft.com/downloads/2002/pfbackup.aspx>.
2294 Similarly, all of the data for WinXX services like SQL databases,
2295 Exchange etc won't be backed up. If these applications support
2296 some kind of export or utility to save their data to disk then this
2297 can =used to create files that BackupPC can backup.
2299 So far, the best that BackupPC can do is send warning emails to
2300 the user saying that their outlook files haven't been backed up in
2301 X days. (X is configurable.) The message invites the user to
2302 exit Outlook and gives a URL to manually start a backup.
2304 I suspect there is a way of mirroring the outlook.pst file so
2305 that at least the mirror copy can be backed up. Or perhaps a
2306 manual copy can be started at login. Does some WinXX expert
2307 know how to do this?
2309 Comment: two users have noted that there are commercial OFM (open file
2310 manager) products that are designed to solve this problem, for example
2311 from St. Bernard or Columbia Data Products. Apparently Veritas and
2312 Legato bundle this product with their commercial products. See for
2313 example L<http://www.stbernard.com/products/docs/ofm_whitepaperV8.pdf>.
2314 If anyone tries these programs with BackupPC please tell us whether or
2317 =item Don't expect to reconstruct a complete WinXX drive
2319 The conclusion from the last few items is that BackupPC is not intended
2320 to allow a complete WinXX disk to be re-imaged from the backup. Our
2321 approach to system restore in the event of catastrophic failure is to
2322 re-image a new disk from a generic master, and then use the BackupPC
2323 archive to restore user files.
2325 It is likely that linux/unix backups done using tar (rather than
2326 smb) can be used to reconstruct a complete file system, although
2329 =item Maximum Backup File Sizes
2331 BackupPC can backup and manage very large file sizes, probably as large
2332 as 2^51 bytes (when a double-precision number's mantissa can no longer
2333 represent an integer exactly). In practice, several things outside
2334 BackupPC limit the maximum individual file size. Any one of the
2335 following items will limit the maximum individual file size:
2341 Perl needs to be compiled with uselargefiles defined. Check your
2344 perl -V | egrep largefiles
2346 Without this, the maximum file size will be 2GB.
2350 The BackupPC pool and data directories must be on a file system that
2351 supports large files.
2353 Without this, the maximum file size will be 2GB.
2357 The transport mechanism also limits the maximum individual file size.
2359 GNU tar maximum file size is limited by the tar header format. The tar
2360 header uses 11 octal digits to represent the file size, which is 33 bits
2361 or 8GB. I vaguely recall (but I haven't recently checked) that GNU tar
2362 uses an extra octal digit (replacing a trailing delimiter) if necessary,
2363 allowing 64GB files. So tar transport limits the maximum file size to
2364 8GB or perhaps 64GB. It is possible that files >= 8GB don't work; this
2365 needs to be looked into.
2367 Smbclient is limited to 4GB file sizes. Moreover, a bug in smbclient
2368 (mixing signed and unsigned 32 bit values) causes it to incorrectly
2369 do the tar octal conversion for file sizes from 2GB-4GB. BackupPC_tarExtract
2370 knows about this bug and can recover the correct file size. So smbclient
2371 transport works up to 4GB file sizes.
2375 =item Some tape backup systems aren't smart about hard links
2377 If you backup the BackupPC pool to tape you need to make sure that the
2378 tape backup system is smart about hard links. For example, if you
2379 simply try to tar the BackupPC pool to tape you will backup a lot more
2380 data than is necessary.
2382 Using the example at the start of the installation section, 65 hosts are
2383 backed up with each full backup averaging 3.2GB. Storing one full backup
2384 and two incremental backups per laptop is around 240GB of raw data. But
2385 because of the pooling of identical files, only 87GB is used (with
2386 compression the total is lower). If you run du or tar on the data
2387 directory, there will appear to be 240GB of data, plus the size of the
2388 pool (around 87GB), or 327GB total.
2390 If your tape backup system is not smart about hard links an alternative
2391 is to periodically backup just the last successful backup for each host
2392 to tape. Another alternative is to do a low-level dump of the pool
2393 file system (ie: /dev/hda1 or similar) using dump(1).
2395 Supporting more efficient tape backup is an area for further
2398 =item Incremental backups might included deleted files
2400 To make browsing and restoring backups easier, incremental backups
2401 are "filled-in" from the last complete backup when the backup is
2402 browsed or restored.
2404 However, if a file was deleted by a user after the last full backup, that
2405 file will still appear in the "filled-in" incremental backup. This is not
2406 really a specific problem with BackupPC, rather it is a general issue
2407 with the full/incremental backup paradigm. This minor problem could be
2408 solved by having smbclient list all files when it does the incremental
2409 backup. Volunteers anyone?
2413 Comments or suggestions on these issues are welcome.
2415 =head2 Security issues
2417 Please read this section and consider each of the issues carefully.
2421 =item Smb share password
2423 An important security risk is the manner in which the smb share
2424 passwords are stored. They are in plain text. As described in
2425 L<Step 3: Setting up config.pl|step 3: setting up config.pl> there are four
2426 ways to tell BackupPC the smb share password (manually setting an environment
2427 variable, setting the environment variable in /etc/init.d/backuppc,
2428 putting the password in __TOPDIR__/conf/config.pl, or putting the
2429 password in __TOPDIR__/pc/$host/config.pl). In the latter 3 cases the
2430 smb share password appears in plain text in a file.
2432 If you use any of the latter three methods please make sure that the file's
2433 permission is appropriately restricted. If you also use RCS or CVS, double
2434 check the file permissions of the config.pl,v file.
2436 In future versions there will probably be support for encryption of the
2437 smb share password, but a private key will still have to be stored in a
2438 protected place. Comments and suggestions are welcome.
2440 =item BackupPC socket server
2442 In v1.5.0 the primary method for communication between the CGI program
2443 (BackupPC_Admin) and the server (BackupPC) is via a unix-domain socket.
2444 Since this socket has restricted permissions, no local user should be
2445 able to connect to this port. No backup or restore data passes through
2446 this interface, but an attacker can start or stop backups and get status
2449 If the Apache server and BackupPC_Admin run on a different host to
2450 BackupPC then a TCP port must be enabled by setting $Conf{ServerPort}.
2451 Anyone can connect to this port. To avoid possible attacks via the TCP
2452 socket interface, every client message is protected by an MD5 digest.
2453 The MD5 digest includes four items:
2459 a seed that is sent to the client when the connection opens
2463 a sequence number that increments for each message
2467 a shared secret that is stored in $Conf{ServerMesgSecret}
2475 The message is sent in plain text preceded by the MD5 digest. A
2476 snooper can see the plain-text seed sent by BackupPC and plain-text
2477 message from the client, but cannot construct a valid MD5 digest since
2478 the secret in $Conf{ServerMesgSecret} is unknown. A replay attack is
2479 not possible since the seed changes on a per-connection and
2482 So if you do enable the TCP port, please set $Conf{ServerMesgSecret}
2483 to some hard-to-guess string. A denial-of-service attack is possible
2484 with the TCP port enabled. Someone could simply connect many times
2485 to this port, until BackupPC had exhausted all its file descriptors,
2486 and this would cause new backups and the CGI interface to fail. The
2487 most secure solution is to run BackupPC and Apache on the same machine
2488 and disable the TCP port.
2490 By the way, if you have upgraded from a version of BackupPC prior to
2491 v1.5.0 you should set $Conf{ServerPort} to -1 to disable the TCP port.
2493 =item Installation permissions
2495 It is important to check that the BackupPC scripts in __INSTALLDIR__/bin
2496 and __INSTALLDIR__/lib cannot be edited by normal users. Check the
2497 directory permissions too.
2499 =item Pool permissions
2501 It is important to check that the data files in __TOPDIR__/pool,
2502 __TOPDIR__/pc and __TOPDIR__/trash cannot be read by normal users.
2503 Normal users should not be able to see anything below __TOPDIR__.
2507 Enabling shares on hosts carries security risks. If you are on a private
2508 network and you generally trust your users then there should not be a
2509 problem. But if you have a laptop that is sometimes on public networks
2510 (eg: broadband or even dialup) you should be concerned. A conservative
2511 approach is to use firewall software, and only enable the netbios and
2512 smb ports (137 and 139) on connections from the host running BackupPC.
2514 =item SSH key security
2516 Using ssh for linux/unix clients is quite secure, but the security is
2517 only as good as the protection of ssh's private keys. If an attacker can
2518 devise a way to run a shell as the BackupPC user then they will have
2519 access to BackupPC's private ssh keys. They can then, in turn, ssh to
2520 any client machine as root (or whichever user you have configured
2521 BackupPC to use). This represents a serious compromise of your entire
2522 network. So in vulnerable networks, think carefully about how to protect
2523 the machine running BackupPC and how to prevent attackers from gaining
2524 shell access (as the BackupPC user) to the machine.
2528 The CGI interface, __CGIDIR__/BackupPC_Admin, needs access to the pool
2529 files so it is installed setuid to __BACKUPPCUSER__. The permissions of
2530 this file need to checked carefully. It should be owned by
2531 __BACKUPPCUSER__ and have user and group (but not other) execute
2532 permission. To allow apache/httpd to execute it, the group ownership
2533 should be something that apache/httpd belongs to.
2535 The Apache configuration should be setup for AuthConfig style,
2536 using a .htaccess file so that the user's name is passed into
2537 the script as $ENV{REMOTE_USER}.
2539 If normal users could directly run BackupPC_Admin then there is a serious
2540 security hole: since it is setuid to __BACKUPPCUSER__ any user can
2541 browse and restore any backups. Be aware that anyone who is allowed to
2542 edit or create cgi scripts on your server can execute BackupPC_Admin as
2543 any user! They simply write a cgi script that sets $ENV{REMOTE_USER} and
2544 then execs BackupPC_Admin. The exec succeeds since httpd runs the first
2545 script as user httpd/apache, which in turn has group permission to
2546 execute BackupPC_Admin.
2548 While this setup should be safe, a more conservative approach is to
2549 run a dedicated Apache as user __BACKUPPCUSER__ on a different port.
2550 Then BackupPC_Admin no longer needs to be setuid, and the cgi
2551 directories can be locked down from normal users. Moreover, this
2552 setup is exactly the one used to support mod_perl, so this provides
2553 both the highest performance and the lowest security risk.
2557 Comments and suggestions are welcome.
2559 =head1 Configuration File
2561 The BackupPC configuration file resides in __TOPDIR__/conf/config.pl.
2562 Optional per-PC configuration files reside in __TOPDIR__/pc/$host/config.pl.
2563 This file can be used to override settings just for a particular PC.
2565 =head2 Modifying the main configuration file
2567 The configuration file is a perl script that is executed by BackupPC, so
2568 you should be careful to preserve the file syntax (punctuation, quotes
2569 etc) when you edit it. It is recommended that you use CVS, RCS or some
2570 other method of source control for changing config.pl.
2572 BackupPC reads or re-reads the main configuration file and
2573 the hosts file in three cases:
2583 When BackupPC is sent a HUP (-1) signal. Assuming you installed the
2584 init.d script, you can also do this with "/etc/init.d/backuppc reload".
2588 When the modification time of config.pl file changes. BackupPC
2589 checks the modification time once during each regular wakeup.
2593 Whenever you change the configuration file you can either do
2594 a kill -HUP BackupPC_pid or simply wait until the next regular
2597 Each time the configuration file is re-read a message is reported in the
2598 LOG file, so you can tail it (or view it via the CGI interface) to make
2599 sure your kill -HUP worked. Errors in parsing the configuration file are
2600 also reported in the LOG file.
2602 The optional per-PC configuration file (__TOPDIR__/pc/$host/config.pl)
2603 is read whenever it is needed by BackupPC_dump, BackupPC_link and others.
2605 =head2 Configuration file includes
2607 If you have a heterogeneous set of clients (eg: a variety of WinXX and
2608 linux/unix machines) you will need to create host-specific config.pl files
2609 for some or all of these machines to customize the default settings from
2610 the master config.pl file (at a minimum to set $Conf{XferMethod}).
2612 Since the config.pl file is just regular perl code, you can include
2613 one config file from another. For example, imagine you had three general
2614 classes of machines: WinXX desktops, linux machines in the DMZ and
2615 linux desktops. You could create three config files in __TOPDIR__/conf:
2617 __TOPDIR__/conf/ConfigWinDesktop.pl
2618 __TOPDIR__/conf/ConfigLinuxDMZ.pl
2619 __TOPDIR__/conf/ConfigLinuxDesktop.pl
2621 From each client's directory you can either add a symbolic link to
2622 the appropriate config file:
2624 cd __TOPDIR__/pc/$host
2625 ln -s ../../conf/ConfigWinDesktop.pl config.pl
2627 or, better yet, create a config.pl file in __TOPDIR__/pc/$host
2628 that contains this line:
2630 do "../../conf/ConfigWinDesktop.pl";
2632 This alternative allows you to set other configuration options
2633 specific to each host (perhaps even overriding the settings in
2636 Note that you could also include snippets of configuration settings
2637 from the main configuration file. However, be aware that the
2638 modification-time checking that BackupPC does only applies to the
2639 main configuration file: if you change one of the included files,
2640 BackupPC won't notice. You will need to either touch the main
2641 configuration file too, or send BackupPC a HUP (-1) signal.
2643 =head1 Configuration Parameters
2645 The configuration parameters are divided into five general groups.
2646 The first group (general server configuration) provides general
2647 configuration for BackupPC. The next two groups describe what to
2648 backup, when to do it, and how long to keep it. The fourth group
2649 are settings for email reminders, and the final group contains
2650 settings for the CGI interface.
2652 All configuration settings in the second through fifth groups can
2653 be overridden by the per-PC config.pl file.
2657 =head1 Version Numbers
2659 Starting with v1.4.0 BackupPC switched to a X.Y.Z version numbering
2660 system, instead of X.0Y. The first digit is for major new releases, the
2661 middle digit is for significant feature releases and improvements (most
2662 of the releases have been in this category), and the last digit is for
2663 bug fixes. You should think of the old 1.00, 1.01, 1.02 and 1.03 as
2664 1.0.0, 1.1.0, 1.2.0 and 1.3.0.
2668 Craig Barratt <cbarratt@users.sourceforge.net>
2670 See L<http://backuppc.sourceforge.net>.
2674 Copyright (C) 2001-2002 Craig Barratt
2678 Ryan Kucera contributed the directory navigation code and images
2679 for v1.5.0. He also contributed the first skeleton of BackupPC_restore.
2681 Guillaume Filion wrote BackupPC_zipCreate and added the CGI support
2682 for zip download, in addition to some CGI cleanup, for v1.5.0.
2684 Several people have reported bugs or made useful suggestions; see the
2687 Your name could appear here in the next version!
2691 This program is free software; you can redistribute it and/or modify it
2692 under the terms of the GNU General Public License as published by the
2693 Free Software Foundation; either version 2 of the License, or (at your
2694 option) any later version.
2696 This program is distributed in the hope that it will be useful,
2697 but WITHOUT ANY WARRANTY; without even the implied warranty of
2698 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
2699 General Public License for more details.
2701 You should have received a copy of the GNU General Public License in the
2702 LICENSE file along with this program; if not, write to the Free Software
2703 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.