Changes for 2.1.0pl1:

[BackupPC.git] / doc-src / BackupPC.pod

=head1 BackupPC Introduction

This documentation describes BackupPC version __VERSION__,
released on __RELEASEDATE__.

=head2 Overview

BackupPC is a high-performance, enterprise-grade system for backing up
Unix, Linux and WinXX PCs, desktops and laptops to a server's disk.
BackupPC is highly configurable and easy to install and maintain.

Given the ever decreasing cost of disks and raid systems, it is now
practical and cost effective to backup a large number of machines onto
a server's local disk or network storage.  For some sites this might be
the complete backup solution.  For other sites additional permanent
archives could be created by periodically backing up the server to tape.

Features include:

=over 4

=item *

A clever pooling scheme minimizes disk storage and disk I/O.
Identical files across multiple backups of the same or different PC
are stored only once (using hard links), resulting in substantial
savings in disk storage and disk writes.

=item *

Optional compression provides additional reductions in storage
(around 40%).  The CPU impact of compression is low since only
new files (those not already in the pool) need to be compressed.

=item *

A powerful http/cgi user interface allows administrators to view log
files, configuration, current status and allows users to initiate and
cancel backups and browse and restore files from backups.

=item *

The http/cgi user interface has internationalization (i18n) support,
currently providing English, French, German, Spanish, Italian
and Dutch.

=item *

No client-side software is needed. On WinXX the standard smb
protocol is used to extract backup data. On linux or unix clients,
rsync or tar (over ssh/rsh/nfs) is used to extract backup data.
Alternatively, rsync can also be used on WinXX (using cygwin),
and Samba could be installed on the linux or unix client to
provide smb shares).

=item *

Flexible restore options.  Single files can be downloaded from
any backup directly from the CGI interface.  Zip or Tar archives
for selected files or directories from any backup can also be
downloaded from the CGI interface.  Finally, direct restore to
the client machine (using smb or tar) for selected files or
directories is also supported from the CGI interface.

=item *

BackupPC supports mobile environments where laptops are only
intermittently connected to the network and have dynamic IP addresses
(DHCP).  Configuration settings allow machines connected via slower WAN
connections (eg: dial up, DSL, cable) to not be backed up, even if they
use the same fixed or dynamic IP address as when they are connected
directly to the LAN.

=item *

Flexible configuration parameters allow multiple backups to be performed
in parallel, specification of which shares to backup, which directories
to backup or not backup, various schedules for full and incremental
backups, schedules for email reminders to users and so on.  Configuration
parameters can be set system-wide or also on a per-PC basis.

=item *

Users are sent periodic email reminders if their PC has not
recently been backed up.  Email content, timing and policies
are configurable.

=item *

BackupPC is Open Source software hosted by SourceForge.

=back

=head2 Backup basics

=over 4

=item Full Backup

A full backup is a complete backup of a share. BackupPC can be
configured to do a full backup at a regular interval (typically
weekly).  BackupPC can be configured to keep a certain number
of full backups.  Exponential expiry is also supported, allowing
full backups with various vintages to be kept (for example, a
settable number of most recent weekly fulls, plus a settable
number of older fulls that are 2, 4, 8, or 16 weeks apart).

=item Incremental Backup

An incremental backup is a backup of files that have changed (based on
their modification time) since the last successful full backup.  For
SMB and tar, BackupPC backups all files that have changed since one
hour prior to the start of the last successful full backup.  Rsync is
more clever: any files whose attributes have changed (ie: uid, gid,
mtime, modes, size) since the last full are backed up.  Deleted, new
files and renamed files are detected by Rsync incrementals.
In constrast, SMB and tar incrementals are not able to detect deleted
files, renamed files or new files whose modification time is prior to
the last full dump.

BackupPC can also be configured to keep a certain number of incremental
backups, and to keep a smaller number of very old incremental backups.
(BackupPC does not support multi-level incremental backups, although it
will in a future version.)

BackupPC's CGI interface "fills-in" incremental backups based on the 
last full backup, giving every backup a "full" appearance.  This makes
browsing and restoring backups easier.

=item Partial Backup

When a full backup fails or is canceled, and some files have already
been backed up, BackupPC keeps a partial backup containing just the
files that were backed up successfully.  The partial backup is removed
when the next successful backup completes, or if another full backup
fails resulting in a newer partial backup.  A failed full backup
that has not backed up any files, or any failed incremental backup,
is removed; no partial backup is saved in these cases.

The partial backup may be browsed or used to restore files just like
a successful full or incremental backup.

With the rsync transfer method the partial backup is used to resume
the next full backup, avoiding the need to retransfer the file data
already in the partial backup.

=item Identical Files

BackupPC pools identical files using hardlinks.  By "identical
files" we mean files with identical contents, not necessary the
same permissions, ownership or modification time.  Two files might
have different permissions, ownership, or modification time but
will still be pooled whenever the contents are identical.  This
is possible since BackupPC stores the file meta-data (permissions,
ownership, and modification time) separately from the file contents.

=item Backup Policy

Based on your site's requirements you need to decide what your backup
policy is.  BackupPC is not designed to provide exact re-imaging of
failed disks.  See L<Limitations|limitations> for more information.
However, the addition of tar transport for linux/unix clients, plus
full support for special file types and unix attributes in v1.4.0
likely means an exact image of a linux/unix file system can be made.

BackupPC saves backups onto disk. Because of pooling you can relatively
economically keep several weeks of old backups.

At some sites the disk-based backup will be adequate, without a
secondary tape backup. This system is robust to any single failure: if a
client disk fails or loses files, the BackupPC server can be used to
restore files. If the server disk fails, BackupPC can be restarted on a
fresh file system, and create new backups from the clients. The chance
of the server disk failing can be made very small by spending more money
on increasingly better RAID systems.  However, there is still the risk
of catastrophic events like fires or earthquakes that can destroy
both the BackupPC server and the clients it is backing up if they
are physically nearby.

Some sites might choose to do periodic backups to tape or cd/dvd.
This backup can be done perhaps weekly using the archive function of
BackupPC.

Other users have reported success with removable disks to rotate the
BackupPC data drives, or using rsync to mirror the BackupPC data pool
offsite.

=back

=head2 Resources

=over 4

=item BackupPC home page

The BackupPC Open Source project is hosted on SourceForge.  The
home page can be found at:

    http://backuppc.sourceforge.net

This page has links to the current documentation, the SourceForge
project page and general information.

=item SourceForge project

The SourceForge project page is at:

    http://sourceforge.net/projects/backuppc

This page has links to the current releases of BackupPC.

=item BackupPC FAQ

BackupPC has a FAQ at L<http://backuppc.sourceforge.net/faq>.

=item Mail lists

Three BackupPC mailing lists exist for announcements (backuppc-announce),
developers (backuppc-devel), and a general user list for support, asking
questions or any other topic relevant to BackupPC (backuppc-users).

The lists are archived on SourceForge and Gmane.  The SourceForge lists
are not always up to date and the searching is limited, so Gmane is
a good alternative.  See:

    http://news.gmane.org/index.php?prefix=gmane.comp.sysutils.backup.backuppc
    http://sourceforge.net/mailarchive/forum.php?forum_id=503

You can subscribe to these lists by visiting:

    http://lists.sourceforge.net/lists/listinfo/backuppc-announce
    http://lists.sourceforge.net/lists/listinfo/backuppc-users
    http://lists.sourceforge.net/lists/listinfo/backuppc-devel

The backuppc-announce list is moderated and is used only for
important announcements (eg: new versions).  It is low traffic.
You only need to subscribe to one of backuppc-announce and
backuppc-users: backuppc-users also receives any messages on
backuppc-announce.

The backuppc-devel list is only for developers who are working on BackupPC.
Do not post questions or support requests there.  But detailed technical
discussions should happen on this list.

To post a message to the backuppc-users list, send an email to

    backuppc-users@lists.sourceforge.net

Do not send subscription requests to this address!

=item Other Programs of Interest

If you want to mirror linux or unix files or directories to a remote server
you should consider rsync, L<http://rsync.samba.org>.  BackupPC now uses
rsync as a transport mechanism; if you are already an rsync user you
can think of BackupPC as adding efficient storage (compression and
pooling) and a convenient user interface to rsync.

Unison is a utility that can do two-way, interactive, synchronization.
See L<http://www.cis.upenn.edu/~bcpierce/unison>.

Three popular open source packages that do tape backup are
Amanda (L<http://www.amanda.org>),
afbackup (L<http://sourceforge.net/projects/afbackup>), and
Bacula (L<http://www.bacula.org>).
Amanda can also backup WinXX machines to tape using samba.
These packages can be used as back ends to BackupPC to backup the
BackupPC server data to tape.

Various programs and scripts use rsync to provide hardlinked backups.
See, for example, Mike Rubel's site (L<http://www.mikerubel.org/computers/rsync_snapshots>),
JW Schultz's dirvish (L<http://www.dirvish.org>),
Ben Escoto's rdiff-backup (L<http://rdiff-backup.stanford.edu>),
and John Bowman's rlbackup (L<http://www.math.ualberta.ca/imaging/rlbackup>).

BackupPC provides many additional features, such as compressed storage,
hardlinking any matching files (rather than just files with the same name),
and storing special files without root privileges.  But these other scripts
provide simple and effective solutions and are worthy of consideration.

=back

=head2 Road map

The new features planned for future releases of BackupPC
are at L<http://backuppc.sourceforge.net/faq/roadMap.html>.

Comments and suggestions are welcome.

=head2 You can help

BackupPC is free. I work on BackupPC because I enjoy doing it and I like
to contribute to the open source community.

BackupPC already has more than enough features for my own needs.  The
main compensation for continuing to work on BackupPC is knowing that
more and more people find it useful.  So feedback is certainly
appreciated, both positive and negative.

Beyond being a satisfied user and telling other people about it, everyone
is encouraged to add links to L<http://backuppc.sourceforge.net>
(I'll see then via Google) or otherwise publicize BackupPC.  Unlike
the commercial products in this space, I have a zero budget (in both
time and money) for marketing, PR and advertising, so it's up to
all of you!  Feel free to vote for BackupPC at
L<http://freshmeat.net/projects/backuppc>.

Also, everyone is encouraged to contribute patches, bug reports, feature
and design suggestions, new code, FAQs, and documentation corrections or
improvements.  Answering questions on the mail list is a big help too.

=head1 Installing BackupPC

=head2 Requirements

BackupPC requires:

=over 4

=item *

A linux, solaris, or unix based server with a substantial amount of free
disk space (see the next section for what that means). The CPU and disk
performance on this server will determine how many simultaneous backups
you can run. You should be able to run 4-8 simultaneous backups on a
moderately configured server.

Several users have reported significantly better performance using
reiser compared to ext3 for the BackupPC data file system.  It is
also recommended you consider either an LVM or raid setup (either
in HW or SW; eg: 3Ware RAID5) so that you can expand the
file system as necessary.

When BackupPC starts with an empty pool, all the backup data will be
written to the pool on disk. After more backups are done, a higher
percentage of incoming files will already be in the pool. BackupPC is
able to avoid writing to disk new files that are already in the pool.
So over time disk writes will reduce significantly (by perhaps a factor
of 20 or more), since eventually 95% or more of incoming backup files
are typically in the pool. Disk reads from the pool are still needed to
do file compares to verify files are an exact match. So, with a mature
pool, if a relatively fast client generates data at say 1MB/sec, and you
run 4 simultaneous backups, there will be an average server disk load of
about 4MB/sec reads and 0.2MB/sec writes (assuming 95% of the incoming
files are in the pool). These rates will be perhaps 40% lower if
compression is on.

=item *

Perl version 5.6.0 or later. BackupPC has been tested with
version 5.6.x, and 5.8.x. If you don't have perl, please
see L<http://www.cpan.org>.

=item *

Perl modules Compress::Zlib, Archive::Zip and File::RsyncP.  Try "perldoc
Compress::Zlib" and "perldoc Archive::Zip" to see if you have these
modules.  If not, fetch them from L<http://www.cpan.org> and see the
instructions below for how to build and install them.

The File::RsyncP module is available from L<http://perlrsync.sourceforge.net>
or CPAN.  You'll need to install the File::RsyncP module if you want to use
Rsync as a transport method.

=item *

If you are using smb to backup WinXX machines you need smbclient and
nmblookup from the samba package.  You will also need nmblookup if
you are backing up linux/unix DHCP machines.  See L<http://www.samba.org>.
Version 2.2.0 or later of Samba is required.
Samba versions 3.x are stable and now recommended instead of 2.x.

See L<http://www.samba.org> for source and binaries.  It's pretty easy to
fetch and compile samba, and just grab smbclient and nmblookup, without
doing the installation. Alternatively, L<http://www.samba.org> has binary
distributions for most platforms.

=item *

If you are using tar to backup linux/unix machines you should have version
1.13.7 at a minimum, with version 1.13.20 or higher recommended.  Use
"tar --version" to check your version.  Various GNU mirrors have the newest
versions of tar, see for example L<http://www.funet.fi/pub/gnu/alpha/gnu/tar>.
As of June 2003 the latest version is 1.13.25.

=item *

If you are using rsync to backup linux/unix machines you should have
version 2.5.5 or higher on each client machine.  See
L<http://rsync.samba.org>. Use "rsync --version" to check your version.

For BackupPC to use Rsync you will also need to install the perl
File::RsyncP module, which is available from
L<http://perlrsync.sourceforge.net>.
Version 0.52 or later is required.

=item *

The Apache web server, see L<http://www.apache.org>, preferably built
with mod_perl support.

=back

=head2 How much disk space do I need?

Here's one real example for an environment that is backing up 65 laptops
with compression off. Each full backup averages 3.2GB. Each incremental
backup averages about 0.2GB. Storing one full backup and two incremental
backups per laptop is around 240GB of raw data. But because of the
pooling of identical files, only 87GB is used.  This is without
compression.

Another example, with compression on: backing up 95 laptops, where
each backup averages 3.6GB and each incremental averages about 0.3GB.
Keeping three weekly full backups, and six incrementals is around
1200GB of raw data.  Because of pooling and compression, only 150GB
is needed.

Here's a rule of thumb. Add up the disk usage of all the machines you
want to backup (210GB in the first example above). This is a rough
minimum space estimate that should allow a couple of full backups and at
least half a dozen incremental backups per machine. If compression is on
you can reduce the storage requirements by maybe 30-40%.  Add some margin
in case you add more machines or decide to keep more old backups.

Your actual mileage will depend upon the types of clients, operating
systems and applications you have. The more uniform the clients and
applications the bigger the benefit from pooling common files.

For example, the Eudora email tool stores each mail folder in a separate
file, and attachments are extracted as separate files. So in the sadly
common case of a large attachment emailed to many recipients, Eudora
will extract the attachment into a new file. When these machines are
backed up, only one copy of the file will be stored on the server, even
though the file appears in many different full or incremental backups. In
this sense Eudora is a "friendly" application from the point of view of
backup storage requirements.

An example at the other end of the spectrum is Outlook. Everything
(email bodies, attachments, calendar, contact lists) is stored in a
single file, which often becomes huge. Any change to this file requires
a separate copy of the file to be saved during backup. Outlook is even
more troublesome, since it keeps this file locked all the time, so it
cannot be read by smbclient whenever Outlook is running.  See the
L<Limitations|limitations> section for more discussion of this problem.

In addition to total disk space, you shold make sure you have
plenty of inodes on your BackupPC data partition. Some users have
reported running out of inodes on their BackupPC data partition.
So even if you have plenty of disk space, BackupPC will report
failures when the inodes are exhausted.  This is a particular
problem with ext2/ext3 file systems that have a fixed number of
inodes when the file system is built.  Use "df -i" to see your
inode usage.

=head2 Step 1: Getting BackupPC

Some linux distributions now include BackupPC.  The Debian
distribution, supprted by Ludovic Drolez, can be found at
L<http://packages.debian.org/backuppc>; it should be included
in the next stable Debian release.  On Debian, BackupPC can
be installed with the command:

    apt-get install backuppc

In the future there might be packages for Gentoo and other
linux flavors.  If the packaged version is older than the
released version then you will probably want to install the
lastest version as described below.

Otherwise, manually fetching and installing BackupPC is easy.
Start by downloading the latest version from
L<http://backuppc.sourceforge.net>.  Hit the "Code" button,
then select the "backuppc" or "backuppc-beta" package and
download the latest version.

=head2 Step 2: Installing the distribution

First off, there are three perl modules you should install.
These are all optional, but highly recommended:

=over 4

=item Compress::Zlib

To enable compression, you will need to install Compress::Zlib
from L<http://www.cpan.org>.
You can run "perldoc Compress::Zlib" to see if this module is installed.

=item Archive::Zip

To support restore via Zip archives you will need to install
Archive::Zip, also from L<http://www.cpan.org>.
You can run "perldoc Archive::Zip" to see if this module is installed.

=item File::RsyncP

To use rsync and rsyncd with BackupPC you will need to install File::RsyncP.
You can run "perldoc File::RsyncP" to see if this module is installed.
File::RsyncP is available from L<http://perlrsync.sourceforge.net>.
Version 0.52 or later is required.

=back

To build and install these packages, fetch the tar.gz file and
then run these commands:

    tar zxvf Archive-Zip-1.01.tar.gz
    cd Archive-Zip-1.01
    perl Makefile.PL
    make
    make test
    make install

The same sequence of commands can be used for each module.

Now let's move onto BackupPC itself.  After fetching
BackupPC-__VERSION__.tar.gz, run these commands as root:

    tar zxf BackupPC-__VERSION__.tar.gz
    cd BackupPC-__VERSION__
    perl configure.pl

In the future this release might also have patches available on the
SourceForge site.  These patch files are text files, with a name of
the form

    BackupPC-__VERSION__plN.diff

where N is the patch level, eg: pl5 is patch-level 5.  These
patch files are cumulative: you only need apply the last patch
file, not all the earlier patch files.  If a patch file is
available, eg: BackupPC-__VERSION__pl5.diff, you should apply
the patch after extracting the tar file:

     # fetch BackupPC-__VERSION__.tar.gz
     # fetch BackupPC-__VERSION__pl5.diff
     tar zxf BackupPC-__VERSION__.tar.gz
     cd BackupPC-__VERSION__
     patch -p0 < ../BackupPC-__VERSION__pl5.diff
     perl configure.pl

A patch file includes comments that describe that bug fixes
and changes.  Feel free to review it before you apply the patch.

The configure.pl script also accepts command-line options if you
wish to run it in a non-interactive manner.  It has self-contained
documentation for all the command-line options, which you can
read with perldoc:

    perldoc configure.pl

When you run configure.pl you will be prompted for the full paths
of various executables, and you will be prompted for the following
information:

=over 4

=item BackupPC User

It is best if BackupPC runs as a special user, eg backuppc, that has
limited privileges. It is preferred that backuppc belongs to a system
administrator group so that sys admin members can browse backuppc files,
edit the configuration files and so on. Although configurable, the
default settings leave group read permission on pool files, so make
sure the BackupPC user's group is chosen restrictively.

On this installation, this is __BACKUPPCUSER__.

=item Data Directory

You need to decide where to put the data directory, below which
all the BackupPC data is stored.  This needs to be a big file system.

On this installation, this is __TOPDIR__.

=item Install Directory

You should decide where the BackupPC scripts, libraries and documentation
should be installed, eg: /opt/local/BackupPC.

On this installation, this is __INSTALLDIR__.

=item CGI bin Directory

You should decide where the BackupPC CGI script resides.  This will
usually below Apache's cgi-bin directory.

On this installation, this is __CGIDIR__.

=item Apache image directory

A directory where BackupPC's images are stored so that Apache can
serve them.  This should be somewhere under Apache's DocumentRoot
directory.

=back

=head2 Step 3: Setting up config.pl

After running configure.pl, browse through the config file,
__INSTALLDIR__/conf/config.pl, and make sure all the default settings
are correct. In particular, you will need to decide whether to use
smb, tar or rsync transport (or whether to set it on a per-PC basis)
and set the relevant parameters for that transport method.
See the section L<Client Setup|step 5: client setup> for more details.

=head2 Step 4: Setting up the hosts file

The file __TOPDIR__/conf/hosts contains the list of clients to backup.
BackupPC reads this file in three cases:

=over 4

=item *

Upon startup.

=item *

When BackupPC is sent a HUP (-1) signal.  Assuming you installed the
init.d script, you can also do this with "/etc/init.d/backuppc reload".

=item *

When the modification time of the hosts file changes.  BackupPC
checks the modification time once during each regular wakeup.

=back

Whenever you change the hosts file (to add or remove a host) you can
either do a kill -HUP BackupPC_pid or simply wait until the next regular
wakeup period.

Each line in the hosts file contains three fields, separated
by white space:

=over 4

=item Host name

This is typically the host name or NetBios name of the client machine
and should be in lower case.  The host name can contain spaces (escape
with a backslash), but it is not recommended.

Please read the section L<How BackupPC Finds Hosts|how backuppc finds hosts>.

In certain cases you might want several distinct clients to refer
to the same physical machine.  For example, you might have a database
you want to backup, and you want to bracket the backup of the database
with shutdown/restart using $Conf{DumpPreUserCmd} and $Conf{DumpPostUserCmd}.
But you also want to backup the rest of the machine while the database
is still running.  In the case you can specify two different clients in
the host file, using any mnemonic name (eg: myhost_mysql and myhost), and
use $Conf{ClientNameAlias} in myhost_mysql's config.pl to specify the
real host name of the machine.

=item DHCP flag

Starting with v2.0.0 the way hosts are discovered has changed and now
in most cases you should specify 0 for the DHCP flag, even if the host
has a dynamically assigned IP address.
Please read the section L<How BackupPC Finds Hosts|how backuppc finds hosts>
to understand whether you need to set the DHCP flag.

You only need to set DHCP to 1 if your client machine doesn't
respond to the NetBios multicast request:

    nmblookup myHost

but does respond to a request directed to its IP address:

    nmblookup -A W.X.Y.Z

If you do set DHCP to 1 on any client you will need to specify the range of
DHCP addresses to search is specified in $Conf{DHCPAddressRanges}.

Note also that the $Conf{ClientNameAlias} feature does not work for
clients with DHCP set to 1.

=item User name

This should be the unix login/email name of the user who "owns" or uses
this machine. This is the user who will be sent email about this
machine, and this user will have permission to stop/start/browse/restore
backups for this host.  Leave this blank if no specific person should
receive email or be allowed to stop/start/browse/restore backups
for this host.  Administrators will still have full permissions.

=item More users

Additional user names, separate by commas and with no white space,
can be specified.  These users will also have full permission in
the CGI interface to stop/start/browse/restore backups for this host.
These users will not be sent email about this host.

=back

The first non-comment line of the hosts file is special: it contains
the names of the columns and should not be edited.

Here's a simple example of a hosts file:

    host        dhcp    user      moreUsers
    farside     0       craig     jim,dave
    larson      1       gary      andy

=head2 Step 5: Client Setup

Two methods for getting backup data from a client are supported: smb and
tar. Smb or rsync are the preferred methods for WinXX clients and rsync or
tar are the preferred methods for linux/unix clients.

The transfer method is set using the $Conf{XferMethod} configuration
setting. If you have a mixed environment (ie: you will use smb for some
clients and tar for others), you will need to pick the most common
choice for $Conf{XferMethod} for the main config.pl file, and then
override it in the per-PC config file for those hosts that will use
the other method.  (Or you could run two completely separate instances
of BackupPC, with different data directories, one for WinXX and the
other for linux/unix, but then common files between the different
machine types will duplicated.)

Here are some brief client setup notes:

=over 4

=item WinXX

The preferred setup for WinXX clients is to set $Conf{XferMethod} to "smb".
(Actually, for v2.0.0, rsyncd is the better method for WinXX if you are
prepared to run rsync/cygwin on your WinXX client.  More information
about this will be provided via the FAQ.)

If you want to use rsyncd for WinXX clients you can find a pre-packaged
zip file on L<http://backuppc.sourceforge.net>. The package is called
cygwin-rsync. It contains rsync.exe, template setup files and the
minimal set of cygwin libraries for everything to run.  The README file
contains instructions for running rsync as a service, so it starts
automatically everytime you boot your machine.

If you build your own rsync, for rsync 2.6.2 it is strongly
recommended you apply the patch in the cygwin-rsync package on
L<http://backuppc.sourceforge.net>.  This patch adds the --checksum-seed
option for checksum caching, and also sends all errors to the client,
which is important so BackupPC can log all file access errors.

Otherwise, to use SMB, you need to create shares for the data you want
to backup. Open "My Computer", right click on the drive (eg: C), and
select "Sharing..." (or select "Properties" and select the "Sharing"
tab). In this dialog box you can enable sharing, select the share name
and permissions.  Many machines will be configured by default to share
the entire C drive as C$ using the administrator password.

If this machine uses DHCP you will also need to make sure the
NetBios name is set.  Go to Control Panel|System|Network Identification
(on Win2K) or Control Panel|System|Computer Name (on WinXP).
Also, you should go to Control Panel|Network Connections|Local Area
Connection|Properties|Internet Protocol (TCP/IP)|Properties|Advanced|WINS
and verify that NetBios is not disabled.

The relevant configuration settings are $Conf{SmbShareName},
$Conf{SmbShareUserName}, $Conf{SmbSharePasswd}, $Conf{SmbClientPath},
$Conf{SmbClientFullCmd}, $Conf{SmbClientIncrCmd} and
$Conf{SmbClientRestoreCmd}.

BackupPC needs to know the smb share user name and password for a
client machine that uses smb.  The user name is specified in
$Conf{SmbShareUserName}. There are four ways to tell BackupPC the
smb share password:

=over 4

=item *

As an environment variable BPC_SMB_PASSWD set before BackupPC starts.
If you start BackupPC manually the BPC_SMB_PASSWD variable must be set
manually first.  For backward compatibility for v1.5.0 and prior, the
environment variable PASSWD can be used if BPC_SMB_PASSWD is not set.
Warning: on some systems it is possible to see environment variables of
running processes.

=item *

Alternatively the BPC_SMB_PASSWD setting can be included in
/etc/init.d/backuppc, in which case you must make sure this file
is not world (other) readable.

=item *

As a configuration variable $Conf{SmbSharePasswd} in
__TOPDIR__/conf/config.pl.  If you put the password
here you must make sure this file is not world (other) readable.

=item *

As a configuration variable $Conf{SmbSharePasswd} in the per-PC
configuration file, __TOPDIR__/pc/$host/config.pl. You will have to
use this option if the smb share password is different for each host.
If you put the password here you must make sure this file is not
world (other) readable.

=back

Placement and protection of the smb share password is a possible
security risk, so please double-check the file and directory
permissions.  In a future version there might be support for
encryption of this password, but a private key will still have to
be stored in a protected place.  Suggestions are welcome.

As an alternative to setting $Conf{XferMethod} to "smb" (using
smbclient) for WinXX clients, you can use an smb network filesystem (eg:
ksmbfs or similar) on your linux/unix server to mount the share,
and then set $Conf{XferMethod} to "tar" (use tar on the network
mounted file system).

Also, to make sure that file names with 8-bit characters are correctly
transferred by smbclient you should add this to samba's smb.conf file
for samba 2.x:

    [global]
        # Accept the windows charset
        client code page = 850
        character set = ISO8859-1

For samba 3.x this should instead be:

    [global]
        unix charset = ISO8859-1

This setting should work for western europe.  
See L<http://www.oreilly.com/catalog/samba/chapter/book/ch08_03.html>
for more information about settings for other languages.

=item Linux/Unix

The preferred setup for linux/unix clients is to set $Conf{XferMethod}
to "rsync", "rsyncd" or "tar".

You can use either rsync, smb, or tar for linux/unix machines. Smb requires
that the Samba server (smbd) be run to provide the shares. Since the smb
protocol can't represent special files like symbolic links and fifos,
tar and rsync are the better transport methods for linux/unix machines.
(In fact, by default samba makes symbolic links look like the file or
directory that they point to, so you could get an infinite loop if a
symbolic link points to the current or parent directory. If you really
need to use Samba shares for linux/unix backups you should turn off the
"follow symlinks" samba config setting. See the smb.conf manual page.)

The requirements for each Xfer Method are:

=over 4

=item tar

You must have GNU tar on the client machine.  Use "tar --version"
or "gtar --version" to verify.  The version should be at least
1.13.7, and 1.13.20 or greater is recommended.  Tar is run on
the client machine via rsh or ssh.

The relevant configuration settings are $Conf{TarClientPath},
$Conf{TarShareName}, $Conf{TarClientCmd}, $Conf{TarFullArgs},
$Conf{TarIncrArgs}, and $Conf{TarClientRestoreCmd}.

=item rsync

You should have at least rsync 2.5.5, and the latest version 2.5.6
is recommended.  Rsync is run on the remote client via rsh or ssh.

The relevant configuration settings are $Conf{RsyncClientPath},
$Conf{RsyncClientCmd}, $Conf{RsyncClientRestoreCmd}, $Conf{RsyncShareName},
$Conf{RsyncArgs}, and $Conf{RsyncRestoreArgs}.

=item rsyncd

You should have at least rsync 2.5.5, and the latest version 2.6.2
is recommended.  In this case the rsync daemon should be running on
the client machine and BackupPC connects directly to it.

The relevant configuration settings are $Conf{RsyncdClientPort},
$Conf{RsyncdUserName}, $Conf{RsyncdPasswd}, $Conf{RsyncdAuthRequired},
$Conf{RsyncShareName}, $Conf{RsyncArgs}, and $Conf{RsyncRestoreArgs}.
$Conf{RsyncShareName} is the name of an rsync module (ie: the thing
in square brackets in rsyncd's conf file -- see rsyncd.conf), not a
file system path.

Be aware that rsyncd will remove the leading '/' from path names in
symbolic links if you specify "use chroot = no" in the rsynd.conf file.
See the rsyncd.conf manual page for more information.

=back

For linux/unix machines you should not backup "/proc".  This directory
contains a variety of files that look like regular files but they are
special files that don't need to be backed up (eg: /proc/kcore is a
regular file that contains physical memory).  See $Conf{BackupFilesExclude}.
It is safe to back up /dev since it contains mostly character-special
and block-special files, which are correctly handed by BackupPC
(eg: backing up /dev/hda5 just saves the block-special file information,
not the contents of the disk).

Alternatively, rather than backup all the file systems as a single
share ("/"), it is easier to restore a single file system if you backup
each file system separately.  To do this you should list each file system
mount point in $Conf{TarShareName} or $Conf{RsyncShareName}, and add the
--one-file-system option to $Conf{TarClientCmd} or add --one-file-system
(note the different punctuation) to $Conf{RsyncArgs}.  In this case there
is no need to exclude /proc explicitly since it looks like a different
file system.

Next you should decide whether to run tar over ssh, rsh or nfs. Ssh is
the preferred method. Rsh is not secure and therefore not recommended.
Nfs will work, but you need to make sure that the BackupPC user (running
on the server) has sufficient permissions to read all the files below
the nfs mount.

Ssh allows BackupPC to run as a privileged user on the client (eg:
root), since it needs sufficient permissions to read all the backup
files. Ssh is setup so that BackupPC on the server (an otherwise low
privileged user) can ssh as root on the client, without being prompted
for a password. There are two common versions of ssh: v1 and v2. Here
are some instructions for one way to setup ssh.  (Check which version
of SSH you have by typing "ssh" or "man ssh".)

=item Mac OS X

In general this should be similar to Linux/Unix machines.
Mark Stosberg reports that you can also use hfstar.
See L<http://fink.sourceforge.net/pdb/package.php/hfstar>.

=item SSH Setup

SSH is a secure way to run tar or rsync on a backup client to extract
the data.  SSH provides strong authentication and encryption of
the network data.

Note that if you run rsyncd (rsync daemon), ssh is not used.
In this case, rsyncd provides its own authentication, but there
is no encryption of network data.  If you want encryption of
network data you can use ssh to create a tunnel, or use a
program like stunnel.  If someone submits instructions I

Setup instructions for ssh are at
L<http://backuppc.sourceforge.net/faq/ssh.html>.

=item Clients that use DHCP

If a client machine uses DHCP BackupPC needs some way to find the
IP address given the host name.  One alternative is to set dhcp
to 1 in the hosts file, and BackupPC will search a pool of IP
addresses looking for hosts.  More efficiently, it is better to
set dhcp = 0 and provide a mechanism for BackupPC to find the
IP address given the host name.

For WinXX machines BackupPC uses the NetBios name server to determine
the IP address given the host name.
For unix machines you can run nmbd (the NetBios name server) from
the Samba distribution so that the machine responds to a NetBios
name request. See the manual page and Samba documentation for more
information.

Alternatively, you can set $Conf{NmbLookupFindHostCmd} to any command
that returns the IP address given the host name.

Please read the section L<How BackupPC Finds Hosts|how backuppc finds hosts>
for more details.

=back

=head2 Step 6: Running BackupPC

The installation contains an init.d backuppc script that can be copied
to /etc/init.d so that BackupPC can auto-start on boot.
See init.d/README for further instructions.

BackupPC should be ready to start.  If you installed the init.d script,
then you should be able to run BackupPC with:

    /etc/init.d/backuppc start

(This script can also be invoked with "stop" to stop BackupPC and "reload"
to tell BackupPC to reload config.pl and the hosts file.)

Otherwise, just run

     __INSTALLDIR__/bin/BackupPC -d

as user __BACKUPPCUSER__.  The -d option tells BackupPC to run as a daemon
(ie: it does an additional fork).

Any immediate errors will be printed to stderr and BackupPC will quit.
Otherwise, look in __TOPDIR__/log/LOG and verify that BackupPC reports
it has started and all is ok.

=head2 Step 7: Talking to BackupPC

Note: as of version 1.5.0, BackupPC no longer supports telnet
to its TCP port.  First off, a unix domain socket is used
instead of a TCP port.  (The TCP port can still be re-enabled
if your installation has apache and BackupPC running on different
machines.)  Secondly, even if you still use the TCP port, the
messages exchanged over this interface are now protected by
an MD5 digest based on a shared secret (see $Conf{ServerMesgSecret})
as well as sequence numbers and per-session unique keys, preventing
forgery and replay attacks.

You should verify that BackupPC is running by using BackupPC_serverMesg.
This sends a message to BackupPC via the unix (or TCP) socket and prints
the response.

You can request status information and start and stop backups using this
interface. This socket interface is mainly provided for the CGI interface
(and some of the BackupPC sub-programs use it too).  But right now we just
want to make sure BackupPC is happy.  Each of these commands should
produce some status output:

    __INSTALLDIR__/bin/BackupPC_serverMesg status info
    __INSTALLDIR__/bin/BackupPC_serverMesg status jobs
    __INSTALLDIR__/bin/BackupPC_serverMesg status hosts

The output should be some hashes printed with Data::Dumper.  If it
looks cryptic and confusing, and doesn't look like an error message,
then all is ok.

The jobs status should initially show just BackupPC_trashClean.
The hosts status should produce a list of every host you have listed
in __TOPDIR__/conf/hosts as part of a big cryptic output line.

You can also request that all hosts be queued:

    __INSTALLDIR__/bin/BackupPC_serverMesg backup all

At this point you should make sure the CGI interface works since
it will be much easier to see what is going on.  That's our
next subject.

=head2 Step 8: CGI interface

The CGI interface script, BackupPC_Admin, is a powerful and flexible
way to see and control what BackupPC is doing.  It is written for an
Apache server.  If you don't have Apache, see L<http://www.apache.org>.

There are two options for setting up the CGI interface: standard
mode and using mod_perl.  Mod_perl provides much higher performance
(around 15x) and is the best choice if your Apache was built with
mod_perl support.  To see if your apache was built with mod_perl
run this command:

    httpd -l | egrep mod_perl

If this prints mod_perl.c then your Apache supports mod_perl.

Using mod_perl with BackupPC_Admin requires a dedicated Apache
to be run as the BackupPC user (__BACKUPPCUSER__).  This is
because BackupPC_Admin needs permission to access various files
in BackupPC's data directories.  In contrast, the standard
installation (without mod_perl) solves this problem by having
BackupPC_Admin installed as setuid to the BackupPC user, so that
BackupPC_Admin runs as the BackuPC user.

Here are some specifics for each setup:

=over 4

=item Standard Setup

The CGI interface should have been installed by the configure.pl script
in __CGIDIR__/BackupPC_Admin.  BackupPC_Admin should have been installed
as setuid to the BackupPC user (__BACKUPPCUSER__), in addition to user
and group execute permission.

You should be very careful about permissions on BackupPC_Admin and
the directory __CGIDIR__: it is important that normal users cannot
directly execute or change BackupPC_Admin, otherwise they can access
backup files for any PC. You might need to change the group ownership
of BackupPC_Admin to a group that Apache belongs to so that Apache
can execute it (don't add "other" execute permission!).
The permissions should look like this:

    ls -l __CGIDIR__/BackupPC_Admin
    -swxr-x---    1 __BACKUPPCUSER__   web      82406 Jun 17 22:58 __CGIDIR__/BackupPC_Admin

The setuid script won't work unless perl on your machine was installed
with setuid emulation.  This is likely the problem if you get an error
saying such as "Wrong user: my userid is 25, instead of 150", meaning
the script is running as the httpd user, not the BackupPC user.
This is because setuid scripts are disabled by the kernel in most
flavors of unix and linux.

To see if your perl has setuid emulation, see if there is a program
called sperl5.6.0 (or sperl5.8.2 etc, based on your perl version)
in the place where perl is installed. If you can't find this program,
then you have two options: rebuild and reinstall perl with the setuid
emulation turned on (answer "y" to the question "Do you want to do
setuid/setgid emulation?" when you run perl's configure script), or
switch to the mod_perl alternative for the CGI script (which doesn't
need setuid to work).

=item Mod_perl Setup

The advantage of the mod_perl setup is that no setuid script is needed,
and there is a huge performance advantage.  Not only does all the perl
code need to be parsed just once, the config.pl and hosts files, plus
the connection to the BackupPC server are cached between requests.  The
typical speedup is around 15 times.

To use mod_perl you need to run Apache as user __BACKUPPCUSER__.
If you need to run multiple Apache's for different services then
you need to create multiple top-level Apache directories, each
with their own config file.  You can make copies of /etc/init.d/httpd
and use the -d option to httpd to point each http to a different
top-level directory.  Or you can use the -f option to explicitly
point to the config file.  Multiple Apache's will run on different
Ports (eg: 80 is standard, 8080 is a typical alternative port accessed
via http://yourhost.com:8080).

Inside BackupPC's Apache http.conf file you should check the
settings for ServerRoot, DocumentRoot, User, Group, and Port.  See 
L<http://httpd.apache.org/docs/server-wide.html> for more details.

For mod_perl, BackupPC_Admin should not have setuid permission, so
you should turn it off:

    chmod u-s __CGIDIR__/BackupPC_Admin

To tell Apache to use mod_perl to execute BackupPC_Admin, add this
to Apache's 1.x httpd.conf file:

    <IfModule mod_perl.c>
        PerlModule Apache::Registry
        PerlTaintCheck On
        <Location /cgi-bin/BackupPC/BackupPC_Admin>   # <--- change path as needed
           SetHandler perl-script
           PerlHandler Apache::Registry
           Options ExecCGI
           PerlSendHeader On
        </Location>
    </IfModule>

Apache 2.0.44 with Perl 5.8.0 on RedHat 7.1, Don Silvia reports that
this works (with tweaks from Michael Tuzi):

    LoadModule perl_module modules/mod_perl.so
    PerlModule Apache2

    <Directory /path/to/cgi/>
        SetHandler perl-script
        PerlResponseHandler ModPerl::Registry
        PerlOptions +ParseHeaders
        Options +ExecCGI
        Order deny,allow
        Deny from all
        Allow from 192.168.0  
        AuthName "Backup Admin"
        AuthType Basic
        AuthUserFile /path/to/user_file
        Require valid-user
    </Directory>

There are other optimizations and options with mod_perl.  For
example, you can tell mod_perl to preload various perl modules,
which saves memory compared to loading separate copies in every
Apache process after they are forked.  See Stas's definitive
mod_perl guide at L<http://perl.apache.org/guide>.

=back

BackupPC_Admin requires that users are authenticated by Apache.
Specifically, it expects that Apache sets the REMOTE_USER environment
variable when it runs.  There are several ways to do this.  One way
is to create a .htaccess file in the cgi-bin directory that looks like:

    AuthGroupFile /etc/httpd/conf/group    # <--- change path as needed
    AuthUserFile /etc/http/conf/passwd     # <--- change path as needed
    AuthType basic
    AuthName "access"
    require valid-user

You will also need "AllowOverride Indexes AuthConfig" in the Apache
httpd.conf file to enable the .htaccess file. Alternatively, everything
can go in the Apache httpd.conf file inside a Location directive. The
list of users and password file above can be extracted from the NIS
passwd file.

One alternative is to use LDAP.  In Apache's http.conf add these lines:

    LoadModule auth_ldap_module   modules/auth_ldap.so
    AddModule auth_ldap.c

    # cgi-bin - auth via LDAP (for BackupPC)
    <Location /cgi-binBackupPC/BackupPC_Admin>    # <--- change path as needed
      AuthType Basic
      AuthName "BackupPC login"
      # replace MYDOMAIN, PORT, ORG and CO as needed
      AuthLDAPURL ldap://ldap.MYDOMAIN.com:PORT/o=ORG,c=CO?uid?sub?(objectClass=*)
      require valid-user
    </Location>

If you want to disable the user authentication you can set
$Conf{CgiAdminUsers} to '*', which allows any user to have
full access to all hosts and backups.  In this case the REMOTE_USER
environment variable does not have to be set by Apache.

Alternatively, you can force a particular user name by getting Apache
to set REMOTE_USER, eg, to hardcode the user to www you could add
this to Apache's httpd.conf:

    <Location /cgi-bin/BackupPC/BackupPC_Admin>   # <--- change path as needed
        Setenv REMOTE_USER www
    </Location>

Finally, you should also edit the config.pl file and adjust, as necessary,
the CGI-specific settings.  They're near the end of the config file. In
particular, you should specify which users or groups have administrator
(privileged) access: see the config settings $Conf{CgiAdminUserGroup}
and $Conf{CgiAdminUsers}.  Also, the configure.pl script placed various
images into $Conf{CgiImageDir} that BackupPC_Admin needs to serve
up.  You should make sure that $Conf{CgiImageDirURL} is the correct
URL for the image directory.

See the section L<Fixing installation problems|fixing installation problems> for suggestions on debugging the Apache authentication setup.

=head2 How BackupPC Finds Hosts

Starting with v2.0.0 the way hosts are discovered has changed.  In most
cases you should specify 0 for the DHCP flag in the conf/hosts file,
even if the host has a dynamically assigned IP address.

BackupPC (starting with v2.0.0) looks up hosts with DHCP = 0 in this manner:

=over 4

=item *

First DNS is used to lookup the IP address given the client's name
using perl's gethostbyname() function.  This should succeed for machines
that have fixed IP addresses that are known via DNS.  You can manually
see whether a given host have a DNS entry according to perls'
gethostbyname function with this command:

    perl -e 'print(gethostbyname("myhost") ? "ok\n" : "not found\n");'

=item *

If gethostbyname() fails, BackupPC then attempts a NetBios multicast to
find the host.  Provided your client machine is configured properly,
it should respond to this NetBios multicast request.  Specifically,
BackupPC runs a command of this form:

    nmblookup myhost

If this fails you will see output like:

    querying myhost on 10.10.255.255
    name_query failed to find name myhost

If this success you will see output like:

    querying myhost on 10.10.255.255
    10.10.1.73 myhost<00>

Depending on your netmask you might need to specify the -B option to
nmblookup.  For example:

    nmblookup -B 10.10.1.255 myhost

If necessary, experiment on the nmblookup command that will return the
IP address of the client given its name.  Then update
$Conf{NmbLookupFindHostCmd} with any necessary options to nmblookup.

=back

For hosts that have the DHCP flag set to 1, these machines are
discovered as follows:

=over 4

=item *

A DHCP address pool ($Conf{DHCPAddressRanges}) needs to be specified.
BackupPC will check the NetBIOS name of each machine in the range using
a command of the form:

    nmblookup -A W.X.Y.Z

where W.X.Y.Z is each candidate address from $Conf{DHCPAddressRanges}.
Any host that has a valid NetBIOS name returned by this command (ie:
matching an entry in the hosts file) will be backed up.  You can
modify the specific nmblookup command if necessary via $Conf{NmbLookupCmd}.

=item *

You only need to use this DHCP feature if your client machine doesn't
respond to the NetBios multicast request:

    nmblookup myHost

but does respond to a request directed to its IP address:

    nmblookup -A W.X.Y.Z

=back

=head2 Other installation topics

=over 4

=item Removing a client

If there is a machine that no longer needs to be backed up (eg: a retired
machine) you have two choices.  First, you can keep the backups accessible
and browsable, but disable all new backups.  Alternatively, you can
completely remove the client and all its backups.

To disable backups for a client there are two special values for
$Conf{FullPeriod} in that client's per-PC config.pl file:

=over 4

=item -1

Don't do any regular backups on this machine.  Manually
requested backups (via the CGI interface) will still occur.

=item -2

Don't do any backups on this machine.  Manually requested
backups (via the CGI interface) will be ignored.

=back

This will still allow that client's old backups to be browsable
and restorable.

To completely remove a client and all its backups, you should remove its
entry in the conf/hosts file, and then delete the __TOPDIR__/pc/$host
directory.  Whenever you change the hosts file, you should send
BackupPC a HUP (-1) signal so that it re-reads the hosts file.
If you don't do this, BackupPC will automatically re-read the
hosts file at the next regular wakeup.

Note that when you remove a client's backups you won't initially recover
a lot of disk space.  That's because the client's files are still in
the pool.  Overnight, when BackupPC_nightly next runs, all the unused
pool files will be deleted and this will recover the disk space used
by the client's backups.

=item Copying the pool

If the pool disk requirements grow you might need to copy the entire
data directory to a new (bigger) file system.  Hopefully you are lucky
enough to avoid this by having the data directory on a RAID file system
or LVM that allows the capacity to be grown in place by adding disks.

The backup data directories contain large numbers of hardlinks.  If
you try to copy the pool the target directory will occupy a lot more
space if the hardlinks aren't re-established.

The GNU cp program with the -a option is aware of hardlinks and knows
to re-establish them.  So GNU cp -a is the recommended way to copy
the data directory and pool.  Don't forget to stop BackupPC while
the copy runs.

=item Compressing an existing pool

If you are upgrading BackupPC and want to turn compression on you have
two choices:

=over 4

=item *

Simply turn on compression.  All new backups will be compressed. Both old
(uncompressed) and new (compressed) backups can be browsed and viewed.
Eventually, the old backups will expire and all the pool data will be
compressed. However, until the old backups expire, this approach could
require 60% or more additional pool storage space to store both
uncompressed and compressed versions of the backup files.

=item *

Convert all the uncompressed pool files and backups to compressed.
The script __INSTALLDIR__/bin/BackupPC_compressPool does this.
BackupPC must not be running when you run BackupPC_compressPool.
Also, there must be no existing compressed backups when you
run BackupPC_compressPool.

BackupPC_compressPool compresses all the files in the uncompressed pool
(__TOPDIR__/pool) and moves them to the compressed pool
(__TOPDIR__/cpool). It rewrites the files in place, so that the
existing hardlinks are not disturbed.

=back

The rest of this section discusses how to run BackupPC_compressPool.

BackupPC_compressPool takes three command line options:

=over 4

=item -t

Test mode: do everything except actually replace the pool files.
Useful for estimating total run time without making any real
changes.

=item -r

Read check: re-read the compressed file and compare it against
the original uncompressed file.  Can only be used in test mode.

=item -c #

Number of children to fork. BackupPC_compressPool can take a long time
to run, so to speed things up it spawns four children, each working on a
different part of the pool. You can change the number of children with
the -c option.

=back

Here are the recommended steps for running BackupPC_compressPool:

=over 4

=item *

Stop BackupPC (eg: "/etc/init.d/backuppc stop").

=item *

Set $Conf{CompressLevel} to a non-zero number (eg: 3).

=item *

Do a dry run of BackupPC_compressPool.  Make sure you run this as
the BackupPC user (__BACKUPPCUSER__):

    BackupPC_compressPool -t -r

The -t option (test mode) makes BackupPC_compressPool do all the steps,
but not actually change anything. The -r option re-reads the compressed
file and compares it against the original.

BackupPC_compressPool gives a status as it completes each 1% of the job.
It also shows the cumulative compression ratio and estimated completion
time.  Once you are comfortable that things look ok, you can kill
BackupPC_compressPool or wait for it to finish.

=item *

Now you are ready to run BackupPC_compressPool for real.  Once again,
as the BackupPC user (__BACKUPPCUSER__), run:

    BackupPC_compressPool

You should put the output into a file and tail this file.  (The running
time could be twice as long as the test mode since the test mode file
writes are immediately followed by an unlink, so in test mode it is
likely the file writes never make it to disk.)

It is B<critical> that BackupPC_compressPool runs to completion before
re-starting BackupPC.  Before BackupPC_compressPool completes, none of
the existing backups will be in a consistent state.  If you must stop
BackupPC_compressPool for some reason, send it an INT or TERM signal
and give it several seconds (or more) to clean up gracefully.
After that, you can re-run BackupPC_compressPool and it will start
again where it left off.  Once again, it is critical that it runs
to 100% completion.

=back

After BackupPC_compressPool completes you should have a complete set
of compressed backups (and your disk usage should be lower).  You
can now re-start BackupPC.

=back

=head2 Fixing installation problems

Please see the FAQ at L<http://backuppc.sourceforge.net/faq> for
debugging suggestions.

=head1 Restore functions

BackupPC supports several different methods for restoring files. The
most convenient restore options are provided via the CGI interface.
Alternatively, backup files can be restored using manual commands.

=head2 CGI restore options

By selecting a host in the CGI interface, a list of all the backups
for that machine will be displayed.  By selecting the backup number
you can navigate the shares and directory tree for that backup.

BackupPC's CGI interface automatically fills incremental backups
with the corresponding full backup, which means each backup has
a filled appearance.  Therefore, there is no need to do multiple
restores from the incremental and full backups: BackupPC does all
the hard work for you.  You simply select the files and directories
you want from the correct backup vintage in one step.

You can download a single backup file at any time simply by selecting
it.  Your browser should prompt you with the file name and ask you
whether to open the file or save it to disk.

Alternatively, you can select one or more files or directories in
the currently selected directory and select "Restore selected files".
(If you need to restore selected files and directories from several
different parent directories you will need to do that in multiple
steps.)

If you select all the files in a directory, BackupPC will replace
the list of files with the parent directory.  You will be presented
with a screen that has three options:

=over 4

=item Option 1: Direct Restore

With this option the selected files and directories are restored
directly back onto the host, by default in their original location.
Any old files with the same name will be overwritten, so use caution.
You can optionally change the target host name, target share name,
and target path prefix for the restore, allowing you to restore the
files to a different location.

Once you select "Start Restore" you will be prompted one last time
with a summary of the exact source and target files and directories
before you commit.  When you give the final go ahead the restore
operation will be queued like a normal backup job, meaning that it
will be deferred if there is a backup currently running for that host.
When the restore job is run, smbclient, tar, rsync or rsyncd is used
(depending upon $Conf{XferMethod}) to actually restore the files.
Sorry, there is currently no option to cancel a restore that has been
started.

A record of the restore request, including the result and list of
files and directories, is kept.  It can be browsed from the host's
home page.  $Conf{RestoreInfoKeepCnt} specifies how many old restore
status files to keep.

Note that for direct restore to work, the $Conf{XferMethod} must
be able to write to the client.  For example, that means an SMB
share for smbclient needs to be writable, and the rsyncd module
needs "read only" set to "false".  This creates additional security
risks.  If you only create read-only SMB shares (which is a good
idea), then the direct restore will fail.  You can disable the
direct restore option by setting $Conf{SmbClientRestoreCmd},
$Conf{TarClientRestoreCmd} and $Conf{RsyncRestoreArgs} to undef.

=item Option 2: Download Zip archive

With this option a zip file containing the selected files and directories
is downloaded.  The zip file can then be unpacked or individual files
extracted as necessary on the host machine. The compression level can be
specified.  A value of 0 turns off compression.

When you select "Download Zip File" you should be prompted where to
save the restore.zip file.

BackupPC does not consider downloading a zip file as an actual
restore operation, so the details are not saved for later browsing
as in the first case.  However, a mention that a zip file was
downloaded by a particular user, and a list of the files, does
appear in BackupPC's log file.

=item Option 3: Download Tar archive

This is identical to the previous option, except a tar file is downloaded
rather than a zip file (and there is currently no compression option).

=back

=head2 Command-line restore options

Apart from the CGI interface, BackupPC allows you to restore files
and directories from the command line.  The following programs can
be used:

=over 4

=item BackupPC_zcat

For each file name argument it inflates (uncompresses) the file and
writes it to stdout.  To use BackupPC_zcat you could give it the
full file name, eg:

    __INSTALLDIR__/bin/BackupPC_zcat __TOPDIR__/pc/host/5/fc/fcraig/fexample.txt > example.txt

It's your responsibility to make sure the file is really compressed:
BackupPC_zcat doesn't check which backup the requested file is from.
BackupPC_zcat returns a non-zero status if it fails to uncompress
a file.

=item BackupPC_tarCreate

BackupPC_tarCreate creates a tar file for any files or directories in
a particular backup.  Merging of incrementals is done automatically,
so you don't need to worry about whether certain files appear in the
incremental or full backup.

The usage is:

   BackupPC_tarCreate [-t] [-h host] [-n dumpNum] [-s shareName]
                    [-r pathRemove] [-p pathAdd] [-b BLOCKS] [-w writeBufSz]
                    files/directories...

The command-line files and directories are relative to the specified
shareName.  The tar file is written to stdout.

The required options are:

=over 4

=item -h host

host from which the tar archive is created

=item -n dumpNum

dump number from which the tar archive is created

=item -s shareName

share name from which the tar archive is created

=back

Other options are:

=over 4

=item -t

print summary totals

=item -r pathRemove

path prefix that will be replaced with pathAdd

=item -p pathAdd

new path prefix

=item -b BLOCKS

the tar block size, default is 20, meaning tar writes data in 20 * 512
bytes chunks.

=item -w writeBufSz

write buffer size, default 1048576 (1MB).  You can increase this if
you are trying to stream to a fast tape device.

=back

The -h, -n and -s options specify which dump is used to generate
the tar archive.  The -r and -p options can be used to relocate
the paths in the tar archive so extracted files can be placed
in a location different from their original location.

=item BackupPC_zipCreate

BackupPC_zipCreate creates a zip file for any files or directories in
a particular backup.  Merging of incrementals is done automatically,
so you don't need to worry about whether certain files appear in the
incremental or full backup.

The usage is:

   BackupPC_zipCreate [-t] [-h host] [-n dumpNum] [-s shareName]
                    [-r pathRemove] [-p pathAdd] [-c compressionLevel]
                   files/directories...

The command-line files and directories are relative to the specified
shareName.  The zip file is written to stdout.

The required options are:

=over 4

=item -h host

host from which the zip archive is created

=item -n dumpNum

dump number from which the zip archive is created

=item -s shareName

share name from which the zip archive is created

=back

Other options are:

=over 4

=item -t

print summary totals

=item -r pathRemove

path prefix that will be replaced with pathAdd

=item -p pathAdd

new path prefix

=item -c level

compression level (default is 0, no compression)

=back

The -h, -n and -s options specify which dump is used to generate
the zip archive.  The -r and -p options can be used to relocate
the paths in the zip archive so extracted files can be placed
in a location different from their original location.

=back

Each of these programs reside in __INSTALLDIR__/bin.

=head1 Archive functions

BackupPC supports archiving to removable media. For users that require
offsite backups, BackupPC can create archives that stream to tape
devices, or create files of specified sizes to fit onto cd or dvd media.

Each archive type is specified by a BackupPC host with its XferMethod
set to 'archive'. This allows for multiple configurations at sites where
there might be a combination of tape and cd/dvd backups being made.

BackupPC provides a menu that allows one or more hosts to be archived.
The most recent backup of each host is archived using BackupPC_tarCreate,
and the output is optionally compressed and split into fixed-sized
files (eg: 650MB).

The archive for each host is done by default using
__INSTALLDIR__/BackupPC_archiveHost.  This script can be copied
and customized as needed.

=head2 Configuring an Archive Host

To create an Archive Host, add it to the hosts file just as any other host
and call it a name that best describes the type of archive, e.g. ArchiveDLT

To tell BackupPC that the Host is for Archives, create a config.pl file in 
the Archive Hosts's pc directory, adding the following line:

$Conf{XferMethod} = 'archive';

To further customise the archive's parameters you can adding the changed
parameters in the host's config.pl file. The parameters are explained in
the config.pl file.  Parameters may be fixed or the user can be allowed 
to change them (eg: output device).

The per-host archive command is $Conf{ArchiveClientCmd}.  By default
this invokes

     __INSTALLDIR__/BackupPC_archiveHost

which you can copy and customize as necessary.

=head2 Starting an Archive

In the web interface, click on the Archive Host you wish to use. You will see a
list of previous archives and a summary on each. By clicking the "Start Archive"
button you are presented with the list of hosts and the approximate backup size
(note this is raw size, not projected compressed size) Select the hosts you wish
to archive and press the "Archive Selected Hosts" button.

The next screen allows you to adjust the parameters for this archive run.
Press the "Start the Archive" to start archiving the selected hosts with the
parameters displayed.

=head1 BackupPC Design

=head2 Some design issues

=over 4

=item Pooling common files

To quickly see if a file is already in the pool, an MD5 digest of the
file length and contents is used as the file name in the pool. This
can't guarantee a file is identical: it just reduces the search to
often a single file or handful of files. A complete file comparison
is always done to verify if two files are really the same.

Identical files on multiples backups are represented by hard links.
Hardlinks are used so that identical files all refer to the same
physical file on the server's disk. Also, hard links maintain
reference counts so that BackupPC knows when to delete unused files
from the pool.

For the computer-science majors among you, you can think of the pooling
system used by BackupPC as just a chained hash table stored on a (big)
file system.

=item The hashing function

There is a tradeoff between how much of file is used for the MD5 digest
and the time taken comparing all the files that have the same hash.

Using the file length and just the first 4096 bytes of the file for the
MD5 digest produces some repetitions.  One example: with 900,000 unique
files in the pool, this hash gives about 7,000 repeated files, and in
the worst case 500 files have the same hash.  That's not bad: we only
have to do a single file compare 99.2% of the time.  But in the worst
case we have to compare as many as 500 files checking for a match.

With a modest increase in CPU time, if we use the file length and the
first 256K of the file we now only have 500 repeated files and in the
worst case around 20 files have the same hash. Furthermore, if we
instead use the first and last 128K of the file (more specifically, the
first and eighth 128K chunks for files larger than 1MB) we get only 300
repeated files and in the worst case around 20 files have the same hash.

Based on this experimentation, this is the hash function used by BackupPC.
It is important that you don't change the hash function after files
are already in the pool.  Otherwise your pool will grow to twice the
size until all the old backups (and all the old files with old hashes)
eventually expire.

=item Compression

BackupPC supports compression. It uses the deflate and inflate methods
in the Compress::Zlib module, which is based on the zlib compression
library (see L<http://www.gzip.org/zlib/>).

The $Conf{CompressLevel} setting specifies the compression level to use.
Zero (0) means no compression. Compression levels can be from 1 (least
cpu time, slightly worse compression) to 9 (most cpu time, slightly
better compression). The recommended value is 3. Changing it to 5, for
example, will take maybe 20% more cpu time and will get another 2-3%
additional compression. Diminishing returns set in above 5.  See the zlib
documentation for more information about compression levels.

BackupPC implements compression with minimal CPU load. Rather than
compressing every incoming backup file and then trying to match it
against the pool, BackupPC computes the MD5 digest based on the
uncompressed file, and matches against the candidate pool files by
comparing each uncompressed pool file against the incoming backup file.
Since inflating a file takes roughly a factor of 10 less CPU time than
deflating there is a big saving in CPU time.

The combination of pooling common files and compression can yield
a factor of 8 or more overall saving in backup storage.

=back

=head2 BackupPC operation

BackupPC reads the configuration information from
__TOPDIR__/conf/config.pl. It then runs and manages all the backup
activity. It maintains queues of pending backup requests, user backup
requests and administrative commands. Based on the configuration various
requests will be executed simultaneously.

As specified by $Conf{WakeupSchedule}, BackupPC wakes up periodically
to queue backups on all the PCs.  This is a four step process:

=over 4

=item 1

For each host and DHCP address backup requests are queued on the
background command queue.

=item 2

For each PC, BackupPC_dump is forked. Several of these may be run in
parallel, based on the configuration. First a ping is done to see if
the machine is alive. If this is a DHCP address, nmblookup is run to
get the netbios name, which is used as the host name. If DNS lookup
fails, $Conf{NmbLookupFindHostCmd} is run to find the IP address from
the host name.  The file __TOPDIR__/pc/$host/backups is read to decide
whether a full or incremental backup needs to be run. If no backup is
scheduled, or the ping to $host fails, then BackupPC_dump exits.

The backup is done using the specified XferMethod.  Either samba's smbclient
or tar over ssh/rsh/nfs piped into BackupPC_tarExtract, or rsync over ssh/rsh
is run, or rsyncd is connected to, with the incoming data
extracted to __TOPDIR__/pc/$host/new.  The XferMethod output is put
into __TOPDIR__/pc/$host/XferLOG.

The letter in the XferLOG file shows the type of object, similar to the
first letter of the modes displayed by ls -l:

    d -> directory
    l -> symbolic link
    b -> block special file
    c -> character special file
    p -> pipe file (fifo)
    nothing -> regular file

The words mean:

=over 4

=item create

new for this backup (ie: directory or file not in pool)

=item pool

found a match in the pool

=item same

file is identical to previous backup (contents were
checksummed and verified during full dump).

=item skip

file skipped in incremental because attributes are the
same (only displayed if $Conf{XferLogLevel} >= 2).

=back

As BackupPC_tarExtract extracts the files from smbclient or tar, or as
rsync runs, it checks each file in the backup to see if it is identical
to an existing file from any previous backup of any PC. It does this
without needed to write the file to disk. If the file matches an
existing file, a hardlink is created to the existing file in the pool.
If the file does not match any existing files, the file is written to
disk and the file name is saved in __TOPDIR__/pc/$host/NewFileList for
later processing by BackupPC_link.  BackupPC_tarExtract and rsync can handle
arbitrarily large files and multiple candidate matching files without
needing to write the file to disk in the case of a match.  This
significantly reduces disk writes (and also reads, since the pool file
comparison is done disk to memory, rather than disk to disk).

Based on the configuration settings, BackupPC_dump checks each
old backup to see if any should be removed.  Any expired backups
are moved to __TOPDIR__/trash for later removal by BackupPC_trashClean.

=item 3

For each complete, good, backup, BackupPC_link is run.
To avoid race conditions as new files are linked into the
pool area, only a single BackupPC_link program runs
at a time and the rest are queued.

BackupPC_link reads the NewFileList written by BackupPC_dump and
inspects each new file in the backup. It re-checks if there is a
matching file in the pool (another BackupPC_link
could have added the file since BackupPC_dump checked). If so, the file
is removed and replaced by a hard link to the existing file. If the file
is new, a hard link to the file is made in the pool area, so that this
file is available for checking against each new file and new backup.

Then, if $Conf{IncrFill} is set (note that the default setting is
off), for each incremental backup, hard links are made in the new
backup to all files that were not extracted during the incremental
backups.  The means the incremental backup looks like a complete
image of the PC (with the exception that files that were removed on
the PC since the last full backup will still appear in the backup
directory tree).

The CGI interface knows how to merge unfilled incremental backups will
the most recent prior filled (full) backup, giving the incremental
backups a filled appearance.  The default for $Conf{IncrFill} is off,
since there is no need to fill incremental backups.  This saves
some level of disk activity, since lots of extra hardlinks are no
longer needed (and don't have to be deleted when the backup expires).

=item 4

BackupPC_trashClean is always run in the background to remove any
expired backups. Every 5 minutes it wakes up and removes all the files
in __TOPDIR__/trash.

Also, once each night, BackupPC_nightly is run to complete some additional
administrative tasks, such as cleaning the pool.  This involves removing
any files in the pool that only have a single hard link (meaning no backups
are using that file).  Again, to avoid race conditions, BackupPC_nightly
is only run when there are no BackupPC_dump or BackupPC_link processes
running.  Therefore, when it is time to run BackupPC_nightly, no new
backups are started and BackupPC waits until all backups have finished.
Then BackupPC_nightly is run, and until it finishes no new backups are
started.  If BackupPC_nightly is slow, the settings

=back

BackupPC also listens for TCP connections on $Conf{ServerPort}, which
is used by the CGI script BackupPC_Admin for status reporting and
user-initiated backup or backup cancel requests.

=head2 Storage layout

BackupPC resides in three directories:

=over 4

=item __INSTALLDIR__

Perl scripts comprising BackupPC reside in __INSTALLDIR__/bin,
libraries are in __INSTALLDIR__/lib and documentation
is in __INSTALLDIR__/doc.

=item __CGIDIR__

The CGI script BackupPC_Admin resides in this cgi binary directory.

=item __TOPDIR__

All of BackupPC's data (PC backup images, logs, configuration information)
is stored below this directory.

=back

Below __TOPDIR__ are several directories:

=over 4

=item __TOPDIR__/conf

The directory __TOPDIR__/conf contains:

=over 4

=item config.pl

Configuration file. See L<Configuration file|configuration file>
below for more details.

=item hosts

Hosts file, which lists all the PCs to backup.

=back

=item __TOPDIR__/log

The directory __TOPDIR__/log contains:

=over 4

=item LOG

Current (today's) log file output from BackupPC.

=item LOG.0 or LOG.0.z

Yesterday's log file output.  Log files are aged daily and compressed
(if compression is enabled), and old LOG files are deleted.

=item BackupPC.pid

Contains BackupPC's process id.

=item status.pl

A summary of BackupPC's status written periodically by BackupPC so
that certain state information can be maintained if BackupPC is
restarted.  Should not be edited.

=item UserEmailInfo.pl

A summary of what email was last sent to each user, and when the
last email was sent.  Should not be edited.

=back

=item __TOPDIR__/trash

Any directories and files below this directory are periodically deleted
whenever BackupPC_trashClean checks. When a backup is aborted or when an
old backup expires, BackupPC_dump simply moves the directory to
__TOPDIR__/trash for later removal by BackupPC_trashClean.

=item __TOPDIR__/pool

All uncompressed files from PC backups are stored below __TOPDIR__/pool.
Each file's name is based on the MD5 hex digest of the file contents.
Specifically, for files less than 256K, the file length and the entire
file is used. For files up to 1MB, the file length and the first and
last 128K are used. Finally, for files longer than 1MB, the file length,
and the first and eighth 128K chunks for the file are used.

Each file is stored in a subdirectory X/Y/Z, where X, Y, Z are the
first 3 hex digits of the MD5 digest.

For example, if a file has an MD5 digest of 123456789abcdef0,
the file is stored in __TOPDIR__/pool/1/2/3/123456789abcdef0.

The MD5 digest might not be unique (especially since not all the file's
contents are used for files bigger than 256K). Different files that have
the same MD5 digest are stored with a trailing suffix "_n" where n is
an incrementing number starting at 0. So, for example, if two additional
files were identical to the first, except the last byte was different,
and assuming the file was larger than 1MB (so the MD5 digests are the
same but the files are actually different), the three files would be
stored as:

        __TOPDIR__/pool/1/2/3/123456789abcdef0
        __TOPDIR__/pool/1/2/3/123456789abcdef0_0
        __TOPDIR__/pool/1/2/3/123456789abcdef0_1

Both BackupPC_dump (actually, BackupPC_tarExtract) and BackupPC_link are
responsible for checking newly backed up files against the pool. For
each file, the MD5 digest is used to generate a file name in the pool
directory. If the file exists in the pool, the contents are compared.
If there is no match, additional files ending in "_n" are checked.
(Actually, BackupPC_tarExtract compares multiple candidate files in
parallel.)  If the file contents exactly match, the file is created by
simply making a hard link to the pool file (this is done by
BackupPC_tarExtract as the backup proceeds). Otherwise,
BackupPC_tarExtract writes the new file to disk and a new hard link is
made in the pool to the file (this is done later by BackupPC_link).

Therefore, every file in the pool will have at least 2 hard links
(one for the pool file and one for the backup file below __TOPDIR__/pc).
Identical files from different backups or PCs will all be linked to
the same file.  When old backups are deleted, some files in the pool
might only have one link.  BackupPC_nightly checks the entire pool
and removes all files that have only a single link, thereby recovering
the storage for that file.

One other issue: zero length files are not pooled, since there are a lot
of these files and on most file systems it doesn't save any disk space
to turn these files into hard links.

=item __TOPDIR__/cpool

All compressed files from PC backups are stored below __TOPDIR__/cpool.
Its layout is the same as __TOPDIR__/pool, and the hashing function
is the same (and, importantly, based on the uncompressed file, not
the compressed file).

=item __TOPDIR__/pc/$host

For each PC $host, all the backups for that PC are stored below
the directory __TOPDIR__/pc/$host.  This directory contains the
following files:

=over 4

=item LOG

Current log file for this PC from BackupPC_dump.

=item LOG.0 or LOG.0.z

Last month's log file.  Log files are aged monthly and compressed
(if compression is enabled), and old LOG files are deleted.

=item XferERR or XferERR.z

Output from the transport program (ie: smbclient, tar or rsync)
for the most recent failed backup.

=item new

Subdirectory in which the current backup is stored.  This
directory is renamed if the backup succeeds.

=item XferLOG or XferLOG.z

Output from the transport program (ie: smbclient, tar or rsync)
for the current backup.

=item nnn (an integer)

Successful backups are in directories numbered sequentially starting at 0.

=item XferLOG.nnn or XferLOG.nnn.z

Output from the transport program (ie: smbclient, tar or rsync)
corresponding to backup number nnn.

=item RestoreInfo.nnn

Information about restore request #nnn including who, what, when, and
why. This file is in Data::Dumper format.  (Note that the restore
numbers are not related to the backup number.)

=item RestoreLOG.nnn.z

Output from smbclient, tar or rsync during restore #nnn.  (Note that the restore
numbers are not related to the backup number.)

=item ArchiveInfo.nnn

Information about archive request #nnn including who, what, when, and
why. This file is in Data::Dumper format.  (Note that the archive
numbers are not related to the restore or backup number.)

=item ArchiveLOG.nnn.z

Output from archive #nnn.  (Note that the archive numbers are not related
to the backup or restore number.)

=item config.pl

Optional configuration settings specific to this host.  Settings in this
file override the main configuration file.

=item backups

A tab-delimited ascii table listing information about each successful
backup, one per row.  The columns are:

=over 4

=item num

The backup number, an integer that starts at 0 and increments
for each successive backup.  The corresponding backup is stored
in the directory num (eg: if this field is 5, then the backup is
stored in __TOPDIR__/pc/$host/5).

=item type

Set to "full" or "incr" for full or incremental backup.

=item startTime

Start time of the backup in unix seconds.

=item endTime

Stop time of the backup in unix seconds.

=item nFiles

Number of files backed up (as reported by smbclient, tar or rsync).

=item size

Total file size backed up (as reported by smbclient, tar or rsync).

=item nFilesExist

Number of files that were already in the pool
(as determined by BackupPC_dump and BackupPC_link).

=item sizeExist

Total size of files that were already in the pool
(as determined by BackupPC_dump and BackupPC_link).

=item nFilesNew

Number of files that were not in the pool
(as determined by BackupPC_link).

=item sizeNew

Total size of files that were not in the pool
(as determined by BackupPC_link).

=item xferErrs

Number of errors or warnings from smbclient, tar or rsync.

=item xferBadFile

Number of errors from smbclient that were bad file errors (zero otherwise).

=item xferBadShare

Number of errors from smbclient that were bad share errors (zero otherwise).

=item tarErrs

Number of errors from BackupPC_tarExtract.

=item compress

The compression level used on this backup.  Zero or empty means no
compression.

=item sizeExistComp

Total compressed size of files that were already in the pool
(as determined by BackupPC_dump and BackupPC_link).

=item sizeNewComp

Total compressed size of files that were not in the pool
(as determined by BackupPC_link).

=item noFill

Set if this backup has not been filled in with the most recent
previous filled or full backup.  See $Conf{IncrFill}.

=item fillFromNum

If this backup was filled (ie: noFill is 0) then this is the
number of the backup that it was filled from

=item mangle

Set if this backup has mangled file names and attributes.  Always
true for backups in v1.4.0 and above.  False for all backups prior
to v1.4.0.

=item xferMethod

Set to the value of $Conf{XferMethod} when this dump was done.

=item level

The level of this dump.  A full dump is level 0.  Currently incrementals
are 1.  But when multi-level incrementals are supported this will reflect
each dump's incremental level.

=back

=item restores

A tab-delimited ascii table listing information about each requested
restore, one per row.  The columns are:

=over 4

=item num

Restore number (matches the suffix of the RestoreInfo.nnn and
RestoreLOG.nnn.z file), unrelated to the backup number.

=item startTime

Start time of the restore in unix seconds.

=item endTime

End time of the restore in unix seconds.

=item result

Result (ok or failed).

=item errorMsg

Error message if restore failed.

=item nFiles

Number of files restored.

=item size

Size in bytes of the restored files.

=item tarCreateErrs

Number of errors from BackupPC_tarCreate during restore.

=item xferErrs

Number of errors from smbclient, tar or rsync during restore.

=back

=item archives

A tab-delimited ascii table listing information about each requested
archive, one per row.  The columns are:

=over 4

=item num

Archive number (matches the suffix of the ArchiveInfo.nnn and
ArchiveLOG.nnn.z file), unrelated to the backup or restore number.

=item startTime

Start time of the restore in unix seconds.

=item endTime

End time of the restore in unix seconds.

=item result

Result (ok or failed).

=item errorMsg

Error message if archive failed.

=back

=back

=back

=head2 Compressed file format

The compressed file format is as generated by Compress::Zlib::deflate
with one minor, but important, tweak. Since Compress::Zlib::inflate
fully inflates its argument in memory, it could take large amounts of
memory if it was inflating a highly compressed file. For example, a
200MB file of 0x0 bytes compresses to around 200K bytes. If
Compress::Zlib::inflate was called with this single 200K buffer, it
would need to allocate 200MB of memory to return the result.

BackupPC watches how efficiently a file is compressing. If a big file
has very high compression (meaning it will use too much memory when it
is inflated), BackupPC calls the flush() method, which gracefully
completes the current compression.  BackupPC then starts another
deflate and simply appends the output file.  So the BackupPC compressed
file format is one or more concatenated deflations/flushes.  The specific
ratios that BackupPC uses is that if a 6MB chunk compresses to less
than 64K then a flush will be done.

Back to the example of the 200MB file of 0x0 bytes.  Adding flushes
every 6MB adds only 200 or so bytes to the 200K output.  So the
storage cost of flushing is negligible.

To easily decompress a BackupPC compressed file, the script
BackupPC_zcat can be found in __INSTALLDIR__/bin.  For each
file name argument it inflates the file and writes it to stdout.

=head2 Rsync checksum caching

An incremental backup with rsync compares attributes on the client
with the last full backup.  Any files with identical attributes
are skipped.  A full backup with rsync sets the --ignore-times
option, which causes every file to be examined independent of
attributes.

Each file is examined by generating block checksums (default 2K
blocks) on the receiving side (that's the BackupPC side), sending
those checksums to the client, where the remote rsync matches those
checksums with the corresponding file.  The matching blocks and new
data is sent back, allowing the client file to be reassembled.
A checksum for the entire file is sent to as an extra check the
the reconstructed file is correct.

This results in significant disk IO and computation for BackupPC:
every file in a full backup, or any file with non-matching attributes
in an incremental backup, needs to be uncompressed, block checksums
computed and sent.  Then the receiving side reassembles the file and
has to verify the whole-file checksum.  Even if the file is identical,
prior to 2.1.0, BackupPC had to read and uncompress the file twice,
once to compute the block checksums and later to verify the whole-file
checksum.

Starting in 2.1.0, BackupPC supports optional checksum caching,
which means the block and file checksums only need to be computed
once for each file.  This results in a significant performance
improvement.  This only works for compressed pool files.
It is enabled by adding

        '--checksum-seed=32761',

to $Conf{RsyncArgs} and $Conf{RsyncRestoreArgs}.

Rsync versions prior to and including rsync-2.6.2 need a small patch to
add support for the --checksum-seed option.  This patch is available in
the cygwin-rsyncd package at L<http://backuppc.sourceforge.net>.
This patch is already included in rsync CVS, so it will be standard
in future versions of rsync.

When this option is present, BackupPC will add block and file checksums
to the compressed pool file the next time a pool file is used and it
doesn't already have cached checksums.  The first time a new file is
written to the pool, the checksums are not appended.  The next time
checksums are needed for a file, they are computed and added.  So the
full performance benefit of checksum caching won't be noticed until the
third time a pool file is used (eg: the third full backup).

With checksum caching enabled, there is a risk that should a file's contents
in the pool be corrupted due to a disk problem, but the cached checksums
are still correct, the corruption will not be detected by a full backup,
since the file contents are no longer read and compared.  To reduce the
chance that this remains undetected, BackupPC can recheck cached checksums
for a fraction of the files.  This fraction is set with the
$Conf{RsyncCsumCacheVerifyProb} setting.  The default value of 0.01 means
that 1% of the time a file's checksums are read, the checksums are verified.
This reduces performance slightly, but, over time, ensures that files
contents are in sync with the cached checksums.

The format of the cached checksum data can be discovered by looking at
the code.  Basically, the first byte of the compressed file is changed
to denote that checksums are appended.  The block and file checksum
data, plus some other information and magic word, are appended to the
compressed file.  This allows the cache update to be done in-place.

=head2 File name mangling

Backup file names are stored in "mangled" form. Each node of
a path is preceded by "f" (mnemonic: file), and special characters
(\n, \r, % and /) are URI-encoded as "%xx", where xx is the ascii
character's hex value.  So c:/craig/example.txt is now stored as
fc/fcraig/fexample.txt.

This was done mainly so meta-data could be stored alongside the backup
files without name collisions. In particular, the attributes for the
files in a directory are stored in a file called "attrib", and mangling
avoids file name collisions (I discarded the idea of having a duplicate
directory tree for every backup just to store the attributes). Other
meta-data (eg: rsync checksums) could be stored in file names preceded
by, eg, "c". There are two other benefits to mangling: the share name
might contain "/" (eg: "/home/craig" for tar transport), and I wanted
that represented as a single level in the storage tree. Secondly, as
files are written to NewFileList for later processing by BackupPC_link,
embedded newlines in the file's path will cause problems which are
avoided by mangling.

The CGI script undoes the mangling, so it is invisible to the user.
Old (unmangled) backups are still supported by the CGI
interface.

=head2 Special files

Linux/unix file systems support several special file types: symbolic
links, character and block device files, fifos (pipes) and unix-domain
sockets. All except unix-domain sockets are supported by BackupPC
(there's no point in backing up or restoring unix-domain sockets since
they only have meaning after a process creates them). Symbolic links are
stored as a plain file whose contents are the contents of the link (not
the file it points to). This file is compressed and pooled like any
normal file. Character and block device files are also stored as plain
files, whose contents are two integers separated by a comma; the numbers
are the major and minor device number. These files are compressed and
pooled like any normal file. Fifo files are stored as empty plain files
(which are not pooled since they have zero size). In all cases, the
original file type is stored in the attrib file so it can be correctly
restored.

Hardlinks are also supported.  When GNU tar first encounters a file with
more than one link (ie: hardlinks) it dumps it as a regular file.  When
it sees the second and subsequent hardlinks to the same file, it dumps
just the hardlink information.  BackupPC correctly recognizes these
hardlinks and stores them just like symlinks: a regular text file
whose contents is the path of the file linked to.  The CGI script
will download the original file when you click on a hardlink.

Also, BackupPC_tarCreate has enough magic to re-create the hardlinks
dynamically based on whether or not the original file and hardlinks
are both included in the tar file.  For example, imagine a/b/x is a
hardlink to a/c/y.  If you use BackupPC_tarCreate to restore directory
a, then the tar file will include a/b/x as the original file and a/c/y
will be a hardlink to a/b/x.  If, instead you restore a/c, then the
tar file will include a/c/y as the original file, not a hardlink.

=head2 Attribute file format

The unix attributes for the contents of a directory (all the files and
directories in that directory) are stored in a file called attrib.
There is a single attrib file for each directory in a backup.
For example, if c:/craig contains a single file c:/craig/example.txt,
that file would be stored as fc/fcraig/fexample.txt and there would be an
attribute file in fc/fcraig/attrib (and also fc/attrib and ./attrib).
The file fc/fcraig/attrib would contain a single entry containing the
attributes for fc/fcraig/fexample.txt.

The attrib file starts with a magic number, followed by the
concatenation of the following information for each file:

=over 4

=item *

File name length in perl's pack "w" format (variable length base 128).

=item *

File name.

=item *

The unix file type, mode, uid, gid and file size divided by 4GB and
file size modulo 4GB (type mode uid gid sizeDiv4GB sizeMod4GB),
in perl's pack "w" format (variable length base 128).

=item *

The unix mtime (unix seconds) in perl's pack "N" format (32 bit integer).

=back

The attrib file is also compressed if compression is enabled.
See the lib/BackupPC/Attrib.pm module for full details.

Attribute files are pooled just like normal backup files.  This saves
space if all the files in a directory have the same attributes across
multiple backups, which is common.

=head2 Optimizations

BackupPC doesn't care about the access time of files in the pool
since it saves attribute meta-data separate from the files.  Since
BackupPC mostly does reads from disk, maintaining the access time of
files generates a lot of unnecessary disk writes.  So, provided 
BackupPC has a dedicated data disk, you should consider mounting
BackupPC's data directory with the noatime attribute (see mount(1)).

=head2 Limitations

BackupPC isn't perfect (but it is getting better). Please see
L<http://backuppc.sourceforge.net/faq/limitations.html> for a
discussion of some of BackupPC's limitations.

=head2 Security issues

Please see L<http://backuppc.sourceforge.net/faq/security.html> for a
discussion of some of various security issues.

=head1 Configuration File

The BackupPC configuration file resides in __TOPDIR__/conf/config.pl.
Optional per-PC configuration files reside in __TOPDIR__/pc/$host/config.pl.
This file can be used to override settings just for a particular PC.

=head2 Modifying the main configuration file

The configuration file is a perl script that is executed by BackupPC, so
you should be careful to preserve the file syntax (punctuation, quotes
etc) when you edit it. It is recommended that you use CVS, RCS or some
other method of source control for changing config.pl.

BackupPC reads or re-reads the main configuration file and
the hosts file in three cases:

=over 4

=item *

Upon startup.

=item *

When BackupPC is sent a HUP (-1) signal.  Assuming you installed the 
init.d script, you can also do this with "/etc/init.d/backuppc reload".

=item *

When the modification time of config.pl file changes.  BackupPC
checks the modification time once during each regular wakeup.

=back

Whenever you change the configuration file you can either do
a kill -HUP BackupPC_pid or simply wait until the next regular
wakeup period.

Each time the configuration file is re-read a message is reported in the
LOG file, so you can tail it (or view it via the CGI interface) to make
sure your kill -HUP worked. Errors in parsing the configuration file are
also reported in the LOG file.

The optional per-PC configuration file (__TOPDIR__/pc/$host/config.pl)
is read whenever it is needed by BackupPC_dump, BackupPC_link and others.

=head2 Configuration file includes

If you have a heterogeneous set of clients (eg: a variety of WinXX and
linux/unix machines) you will need to create host-specific config.pl files
for some or all of these machines to customize the default settings from
the master config.pl file (at a minimum to set $Conf{XferMethod}).

Since the config.pl file is just regular perl code, you can include
one config file from another.  For example, imagine you had three general
classes of machines: WinXX desktops, linux machines in the DMZ and
linux desktops.  You could create three config files in __TOPDIR__/conf:

    __TOPDIR__/conf/ConfigWinDesktop.pl
    __TOPDIR__/conf/ConfigLinuxDMZ.pl
    __TOPDIR__/conf/ConfigLinuxDesktop.pl

From each client's directory you can either add a symbolic link to
the appropriate config file:

    cd __TOPDIR__/pc/$host
    ln -s ../../conf/ConfigWinDesktop.pl config.pl

or, better yet, create a config.pl file in __TOPDIR__/pc/$host
that includes the default config.pl file using perl's "do"
command:

    do "__TOPDIR__/conf/ConfigWinDesktop.pl";

This alternative allows you to set other configuration options
specific to each host after the "do" command (perhaps even
overriding the settings in the included file).

Note that you could also include snippets of configuration settings
from the main configuration file.  However, be aware that the
modification-time checking that BackupPC does only applies to the
main configuration file: if you change one of the included files,
BackupPC won't notice.  You will need to either touch the main
configuration file too, or send BackupPC a HUP (-1) signal.

=head1 Configuration Parameters

The configuration parameters are divided into five general groups.
The first group (general server configuration) provides general
configuration for BackupPC.  The next two groups describe what to
backup, when to do it, and how long to keep it.  The fourth group
are settings for email reminders, and the final group contains
settings for the CGI interface.

All configuration settings in the second through fifth groups can
be overridden by the per-PC config.pl file.

__CONFIGPOD__

=head1 Version Numbers

Starting with v1.4.0 BackupPC uses a X.Y.Z version numbering system,
instead of X.0Y. The first digit is for major new releases, the middle
digit is for significant feature releases and improvements (most of
the releases have been in this category), and the last digit is for
bug fixes. You should think of the old 1.00, 1.01, 1.02 and 1.03 as
1..0, 1.1.0, 1.2.0 and 1.3.0.

Additionally, patches might be made available.  A patched version
number is of the form X.Y.ZplN (eg: 2.1.0pl2), where N is the
patch level.

=head1 Author

Craig Barratt  <cbarratt@users.sourceforge.net>

See L<http://backuppc.sourceforge.net>.

=head1 Copyright

Copyright (C) 2001-2004 Craig Barratt

=head1 Credits

Ryan Kucera contributed the directory navigation code and images
for v1.5.0.  He contributed the first skeleton of BackupPC_restore.
He also added a significant revision to the CGI interface, including
CSS tags, in v2.1.0, and designed the BackupPC logo.

Xavier Nicollet, with additions from Guillaume Filion, added the
internationalization (i18n) support to the CGI interface for v2.0.0.
Xavier provided the French translation fr.pm, with additions from
Guillaume.

Guillaume Filion wrote BackupPC_zipCreate and added the CGI support
for zip download, in addition to some CGI cleanup, for v1.5.0.
Guillaume continues to support fr.pm updates for each new version.

Josh Marshall implemented the Archive feature in v2.1.0.

Ludovic Drolez supports the BackupPC Debian package.

Javier Gonzalez provided the Spanish translation, es.pm for v2.0.0.

Manfred Herrmann provided the German translation, de.pm for v2.0.0.
Manfred continues to support de.pm updates for each new version,
together with some help frmo Ralph Paßgang.

Lorenzo Cappelletti provided the Italian translation, it.pm for v2.1.0.

Lieven Bridts provided the Dutch translation, nl.pm, for v2.1.0,
with some tweaks from Guus Houtzager.

Many people have reported bugs, made useful suggestions and helped
with testing; see the ChangeLog and the mail lists.

Your name could appear here in the next version!

=head1 License

This program is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2 of the License, or (at your
option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.

You should have received a copy of the GNU General Public License in the
LICENSE file along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.