=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 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 * No client-side software is needed. On WinXX the standard smb protocol is used to extract backup data. On linux or unix clients, tar over ssh/rsh/nfs is used to extract backup data (or alternatively Samba can 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 (often weekly). BackupPC can also be configured to keep a certain number of full backups, and to keep a smaller number of very old full backups. =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. To be safe, BackupPC backups all files that have changed since one hour prior to the start of the last successful full backup. 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 would be easy to do so.) 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 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 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. But BackupPC does not provide permanent storage to tape. Other Open Source applications can do this by backing up BackupPC's pool directories to tape. 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. At other sites a secondary tape backup will be required. This tape backup can be done perhaps weekly from the BackupPC pool file system. One comment: in the US in particular, permanent backups of things like email are becoming strongly discouraged by lawyers because of discovery prior to possible litigation. Using BackupPC without tape backup allows recent file changes or losses to be restored, but without keeping a history more than a month or two old (although this doesn't avoid the problem of old emails languishing in user's email folders forever). =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 Mail lists Two BackupPC mailing lists exist for announcements (backuppc-announce) and reporting information, asking questions, discussing development or any other topic relevant to BackupPC (backuppc-users). You are encouraged to subscribe to either the backuppc-announce or backuppc-users mail list on sourceforge.net at either: http://lists.sourceforge.net/lists/listinfo/backuppc-announce http://lists.sourceforge.net/lists/listinfo/backuppc-users 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 list: backuppc-users also receives any messages on backuppc-announce. 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 look at rsync, L. Two popular open source packages that do tape backup are Amanda (L) and afbackup (L). 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. =back =head2 Road map Here are some ideas for new features that might appear in future releases of BackupPC: =over 4 =item * Adding support for rsync as a transport method, in addition to smb and tar. This will give big savings in network traffic for linux/unix clients. I haven't decided whether to save the pool file rsync checksums (that would double the number of files in the pool, but eliminate most server disk reads), or recompute them every time. I expect to use native rsync on the client side. On the server, rsync would need to understand the compressed file format, the file name mangling and the attribute files, so I will either have to add features to rsync or emulate rsync on the server side in perl. =item * Adding a trip wire feature for notification when files below certain directories change. For example, if you are backing up a DMZ machine, you could request that you get sent email if any files below /bin, /sbin or /usr change. =item * Resuming incomplete completed full backups. Useful if a machine (eg: laptop) is disconnected from the network during a backup, or if the user manually stops a backup. This would work by excluding directories that were already complete. =item * More speculative: Storing binary file deltas (in fact, reverse deltas) for files that have the same name as a previous backup, but that aren't already in the pool. This will save storage for things like mailbox files or documents that change slightly between backups. Running some benchmarks on a large pool suggests that the potential savings are around 15-20%, which isn't spectacular, and likely not worth the implementation effort. The program xdelta (v1) on SourceForge (see L) uses an rsync algorithm for doing efficient binary file deltas. =back 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. Even negative feedback is helpful, for example "We evaluated BackupPC but didn't use it because it doesn't ...". Beyond being a satisfied user and telling other people about it, everyone is encouraged to add links to 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! Also, everyone is encouraged to contribute patches, bug reports, feature and design suggestions, code, and documentation corrections or improvements. =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. 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 5.6.0 and 5.6.1. If you don't have perl, please see L. =item * Perl modules Compress::Zlib and Archive::Zip. Try "perldoc Compress::Zlib" and "perldoc Archive::Zip" to see if you have these modules. If not, fetch them from L and see the instructions below for how to build and install them. =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. Version >= 2.2.0 of Samba is required (smbclient's tar feature in 2.0.8 and prior has bugs for file path lengths around 100 characters and generates bad output when file lengths change during the backup). See L 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 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. As of July 2002 the latest versons is 1.13.25. =item * The Apache web server, see L, 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 C drive 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 section for more discussion of this problem. =head2 Step 1: Getting BackupPC Download the latest version from L. =head2 Step 2: Installing the distribution First off, to enable compression, you will need to install Compress::Zlib from L. It is optional, but strongly recommended. Also, to support restore via Zip archives you will need to install Archive::Zip, also from L. You can run "perldoc Compress::Zlib" to see if this module is installed. To build and install these packages you should run these commands: tar zxvf Archive-Zip-1.01.tar.gz cd Archive-Zip-1.01 perl Makefile.PL make make test make install 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 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 or tar transport (or whether to set it on a per-PC basis), set the smb share password (if using smb), set the backup policies and modify the email message headers and bodies. BackupPC needs to know the smb share user name and password for each PC that uses smb (ie: all the WinXX clients). 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 PASSWD set before BackupPC starts. If you start BackupPC manually this PASSWD must be set manually first. =item * Alternatively the 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. =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 If this host is a static IP address this must the machine's IP host name (ie: something that can be looked up using nslookup or DNS). If this is a host with a dynamic IP address (ie: DHCP flag is 1) then the host name must be the netbios name of the machine. The host name should be in lower case. =item DHCP flag Set to 0 if this host has a static IP address (meaning it can be looked up by name in the DNS). If the host's IP address is dynamic (eg, it is assigned by DHCP) then set this flag to 1. The hosts with dhcp = 1 are backed up as follows. If you have configured a DHCP address pool ($Conf{DHCPAddressRanges}) then BackupPC will check the NetBIOS name of each machine in the range. Any hosts that have a valid NetBIOS name (ie: matching an entry in the hosts file) will be backed up. =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. =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 farside 0 craig larson 1 gary The range of DHCP addresses to search is specified in $Conf{DHCPAddressRanges}. =head2 Step 5: Client Setup Two methods for getting backup data from a client are supported: smb and tar. Smb is the preferred method for WinXX clients and tar is preferred method 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". 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. 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. 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: [global] # Accept the windows charset client code page = 850 character set = ISO8859-1 This setting should work for western europe. See L for more information about settings for other languages. =item Linux/Unix The preferred setup for linux/unix clients is to set $Conf{XferMethod} to "tar". You can use either 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 is the recommended transport method 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 rest of this section describes the tar setup. 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. For linux/unix machines you should no 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). 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 v2: =over 4 =item Key generation As root on the client machine, use ssh2-keygen to generate a public/private key pair, without a pass-phrase: ssh-keygen2 -t rsa -P This will save the public key in /.ssh2/id_rsa_1024_a.pub and the private key in /.ssh2/id_rsa_1024_a. =item Identification Create the identification file /.ssh2/identification: echo "IdKey id_rsa_1024_a" > /.ssh2/identification =item BackupPC setup Repeat the above steps for the BackupPC user (__BACKUPPCUSER__) on the server. Rename the key files to recognizable names, eg: ssh-keygen2 -t rsa -P mv ~/.ssh2/id_rsa_1024_a.pub ~/.ssh2/BackupPC_id_rsa_1024_a.pub mv ~/.ssh2/id_rsa_1024_a ~/.ssh2/BackupPC_id_rsa_1024_a echo "IdKey BackupPC_id_rsa_1024_a" > ~/.ssh2/identification Based on your ssh2 configuration, you might also need to turn off StrictHostKeyChecking and PasswordAuthentication: touch ~/.ssh2/ssh2_config echo "StrictHostKeyChecking ask" >> ~/.ssh2/ssh2_config echo "PasswordAuthentication no" >> ~/.ssh2/ssh2_config =item Key exchange To allow BackupPC to ssh to the client as root, you need to place BackupPC's public key into root's authorized list on the client. Copy BackupPC's public key (BackupPC_id_rsa_1024_a.pub) to the /.ssh2 directory on the client. Add the following line to the /.ssh2/authorization file on the client (as root): touch /.ssh2/authorization echo "Key BackupPC_id_rsa_1024_a.pub" >> /.ssh2/authorization =item Fix permissions You will probably need to make sure that all the files in /.ssh2 have no group or other read/write permission: chmod -R go-rwx /.ssh2 You should do the same thing for the BackupPC user on the server. =item Testing As the BackupPC user on the server, verify that this command: ssh2 -l root clientHostName whoami prints root You might be prompted the first time to accept the client's host key and you might be prompted for root's password on the client. Make sure that this command runs cleanly with no prompts after the first time. You might need to check /etc/hosts.equiv on the client. Look at the man pages for more information. The "-v" option to ssh2 is a good way to get detailed information about what fails. =item ssh version 1 instructions The concept is identical and the steps are similar, but the specific commands and file names are slightly different. First, run ssh-keygen on the client (as root) and server (as the BackupPC user) and simply hit enter when prompted for the pass-phrase: ssh-keygen This will save the public key in /.ssh/identity.pub and the private key in /.ssh/identity. Next, append BackupPC's ~/.ssh/identity.pub (from the server) to root's /.ssh/authorized_keys file on the client. It's a single long line that you can cut-and-paste with an editor (make sure it remains a single line). Next, force protocol version 1 by adding: Protocol 1 to BackupPC's ~/.ssh/config on the server. Next, run "chmod -R go-rwx ~/.ssh" on the server and "chmod -R go-rwx /.ssh" on the client. Finally, test using: ssh -l root clientHostName whoami =back Finally, if this machine uses DHCP you will need to 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. =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. 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 sperl5.6.1 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 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 httpd.conf file: PerlModule Apache::Registry PerlTaintCheck On # <--- change path as needed SetHandler perl-script PerlHandler Apache::Registry Options ExecCGI 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. =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) # <--- 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 If you want to defeat the user authentication 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 httpd.conf: # <--- change path as needed Setenv REMOTE_USER www 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. 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. =head2 Other installation topics =over 4 =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 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 Debugging installation problems This section will probably grow based on the types of questions on the BackupPC mail list. Assuming BackupPC can start correctly you should inspect __TOPDIR__/log/LOG for any errors. Assuming backups for a particular host start, you should be able to look in __TOPDIR__/pc/$host/LOG for error messages specific to that host. The most likely problems will relate to connecting to the smb shares on each host. On each failed backup, a file __TOPDIR__/pc/$host/XferERR will be created. This is the stderr output from smbclient. The first line will show the full smbclient command that was run. Based on the error messages you should figure out what is wrong. Possible errors on the server side are invalid host, invalid share name, bad username or password. Possible errors on the client side are misconfiguration of the share, username or password. You should run smbclient manually and verify that you can connect to the host in interactive mode, eg: smbclient '\\hostName\shareName' -U userName shareName should match the $Conf{SmbShareName} setting and userName should match the the $Conf{SmbShareUserName} setting. You will be prompted for the password. You should then see this prompt: smb: \> Verify that "ls" works and then type "quit" to exit. Secondly, you should also verify that nmblookup correctly returns the netbios name. This is essential for DHCP hosts, and depending upon the setting of $Conf{FixedIPNetBiosNameCheck} might also be required for fixed IP address hosts too. Run this command: nmblookup -A hostName Verify that the host name is printed. The output might look like: received 7 names DELLLS13 <00> - P DOMAINNAME <00> - P DELLLS13 <20> - P DOMAINNAME <1e> - P DELLLS13 <03> - P DELLLS13$ <03> - P CRAIG <03> - P The first name, converted to lower case, is used for the host name. =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 or tar 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. =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. =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] 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 =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 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 deleted 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). 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 $Conf{WakeupSchedule}, BackupPC wakes up periodically to queue backups on all the PCs. This is a four step process: =over 4 =item 1 For each host and DHCP address backup requests are queued on the background command queue. =item 2 For each PC, BackupPC_dump is forked. Several of these may be run in parallel, based on the configuration. First a ping is done to see if the machine is alive. If this is a DHCP address, nmblookup is run to get the netbios name, which is used as the host name. The file __TOPDIR__/pc/$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 $host fails, then BackupPC_dump exits. The backup is done using samba's smbclient or tar over ssh/rsh/nfs piped into BackupPC_tarExtract, extracting the backup into __TOPDIR__/pc/$host/new. The smbclient or tar output is put into __TOPDIR__/pc/$host/XferLOG. As BackupPC_tarExtract extracts the files from smbclient, it checks each file in the backup to see if it is identical to an existing file from any previous backup of any PC. It does this without needed to write the file to disk. If the file matches an existing file, a hardlink is created to the existing file in the pool. If the file does not match any existing files, the file is written to disk and the file name is saved in __TOPDIR__/pc/$host/NewFileList for later processing by BackupPC_link. BackupPC_tarExtract 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, assuming $Conf{IncrFill} is set, 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). As of v1.03, 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 now 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. =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 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 or tar) 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 or tar) 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 or tar) 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 or tar during restore #nnn. (Note that the restore numbers are not related to the backup 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 or tar). =item size Total file size backed up (as reported by smbclient or tar). =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 (zero for tar). =item xferBadFile Number of errors from smbclient that were bad file errors (zero for tar). =item xferBadShare Number of errors from smbclient that were bad share errors (zero for tar). =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. =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 or tar during restore. =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 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 Limitations BackupPC isn't perfect (but it is getting better). Here are some limitations of BackupPC: =over 4 =item Non-unix file attributes not backed up smbclient doesn't extract the WinXX ACLs, so file attributes other than the equivalent (as provided by smbclient) unix attributes are not backed up. =item Locked files are not backed up Under WinXX a locked file cannot be read by smbclient. Such files will not be backed up. This includes the WinXX system registry files. This is especially troublesome for Outlook, which stores all its data in a single large file and keeps it locked whenever it is running. Since many users keep Outlook running all the time their machine is up their Outlook file will not be backed up. Sadly, this file is the most important file to backup. How do commercial backup systems solve this problem? Similarly, all of the data for WinXX services like SQL databases, Exchange etc won't be backed up. So far, the best that BackupPC can do is send warning emails to the user saying that their outlook files haven't been backed up in X days. (X is configurable.) The message invites the user to exit Outlook and gives a URL to manually start a backup. I suspect there is a way of mirroring the outlook.pst file so that at least the mirror copy can be backed up. Or perhaps a manual copy can be started at login. Does some WinXX expert know how to do this? Comment: two users have noted that there are commercial OFM (open file manager) products that are designed to solve this problem, for example from St. Bernard or Columbia Data Products. Apparently Veritas and Legato bundle this product with their commercial products. See for example L. If anyone tries these programs with BackupPC please tell us whether or not they work. =item Don't expect to reconstruct a complete WinXX drive The conclusion from the last few items is that BackupPC is not intended to allow a complete WinXX disk to be re-imaged from the backup. Our approach to system restore in the event of catastrophic failure is to re-image a new disk from a generic master, and then use the BackupPC archive to restore user files. It is likely that linux/unix backups done using tar (rather than smb) can be used to reconstruct a complete file system, although I haven't tried it. =item Maximum Backup File Sizes BackupPC can backup and manage very large file sizes, probably as large as 2^51 bytes (when a double-precision number's mantissa can no longer represent an integer exactly). In practice, several things outside BackupPC limit the maximum individual file size. Any one of the following items will limit the maximum individual file size: =over 4 =item Perl Perl needs to be compiled with uselargefiles defined. Check your installation with: perl -V | egrep largefiles Without this, the maximum file size will be 2GB. =item File system The BackupPC pool and data directories must be on a file system that supports large files. Without this, the maximum file size will be 2GB. =item Transport The transport mechanism also limits the maximum individual file size. GNU tar maximum file size is limited by the tar header format. The tar header uses 11 octal digits to represent the file size, which is 33 bits or 8GB. I vaguely recall (but I haven't recently checked) that GNU tar uses an extra octal digit (replacing a trailing delimiter) if necessary, allowing 64GB files. So tar transport limits the maximum file size to 8GB or perhaps 64GB. It is possible that files >= 8GB don't work; this needs to be looked into. Smbclient is limited to 4GB file sizes. Moreover, a bug in smbclient (mixing signed and unsigned 32 bit values) causes it to incorrectly do the tar octal conversion for file sizes from 2GB-4GB. BackupPC_tarExtract knows about this bug and can recover the correct file size. So smbclient transport works up to 4GB file sizes. =back =item Some tape backup systems aren't smart about hard links If you backup the BackupPC pool to tape you need to make sure that the tape backup system is smart about hard links. For example, if you simply try to tar the BackupPC pool to tape you will backup a lot more data than is necessary. Using the example at the start of the installation section, 65 hosts are backed up with each full backup averaging 3.2GB. 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 (with compression the total is lower). If you run du or tar on the data directory, there will appear to be 240GB of data, plus the size of the pool (around 87GB), or 327GB total. If your tape backup system is not smart about hard links an alternative is to periodically backup just the last successful backup for each host to tape. Another alternative is to do a low-level dump of the pool file system (ie: /dev/hda1 or similar) using dump(1). Supporting more efficient tape backup is an area for further development. =item Incremental backups might included deleted files To make browsing and restoring backups easier, incremental backups are "filled-in" from the last complete backup when the backup is browsed or restored. However, if a file was deleted by a user after the last full backup, that file will still appear in the "filled-in" incremental backup. This is not really a specific problem with BackupPC, rather it is a general issue with the full/incremental backup paradigm. This minor problem could be solved by having smbclient list all files when it does the incremental backup. Volunteers anyone? =back Comments or suggestions on these issues are welcome. =head2 Security issues Please read this section and consider each of the issues carefully. =over 4 =item Smb share password An important security risk is the manner in which the smb share passwords are stored. They are in plain text. As described in L there are four ways to tell BackupPC the smb share password (manually setting an environment variable, setting the environment variable in /etc/init.d/backuppc, putting the password in __TOPDIR__/conf/config.pl, or putting the password in __TOPDIR__/pc/$host/config.pl). In the latter 3 cases the smb share password appears in plain text in a file. If you use any of the latter three methods please make sure that the file's permission is appropriately restricted. If you also use RCS or CVS, double check the file permissions of the config.pl,v file. In future versions there will probably be support for encryption of the smb share password, but a private key will still have to be stored in a protected place. Comments and suggestions are welcome. =item BackupPC socket server In v1.5.0 the primary method for communication between the CGI program (BackupPC_Admin) and the server (BackupPC) is via a unix-domain socket. Since this socket has restricted permissions, no local user should be able to connect to this port. No backup or restore data passes through this interface, but an attacker can start or stop backups and get status through this port. If the Apache server and BackupPC_Admin run on a different host to BackupPC then a TCP port must be enabled by setting $Conf{ServerPort}. Anyone can connect to this port. To avoid possible attacks via the TCP socket interface, every client message is protected by an MD5 digest. The MD5 digest includes four items: =over 4 =item * a seed that is sent to the client when the connection opens =item * a sequence number that increments for each message =item * a shared secret that is stored in $Conf{ServerMesgSecret} =item * the message itself. =back The message is sent in plain text preceded by the MD5 digest. A snooper can see the plain-text seed sent by BackupPC and plain-text message from the client, but cannot construct a valid MD5 digest since the secret in $Conf{ServerMesgSecret} is unknown. A replay attack is not possible since the seed changes on a per-connection and per-message basis. So if you do enable the TCP port, please set $Conf{ServerMesgSecret} to some hard-to-guess string. A denial-of-service attack is possible with the TCP port enabled. Someone could simply connect many times to this port, until BackupPC had exhausted all its file descriptors, and this would cause new backups and the CGI interface to fail. The most secure solution is to run BackupPC and Apache on the same machine and disable the TCP port. By the way, if you have upgraded from a version of BackupPC prior to v1.5.0 you should set $Conf{ServerPort} to -1 to disable the TCP port. =item Installation permissions It is important to check that the BackupPC scripts in __INSTALLDIR__/bin and __INSTALLDIR__/lib cannot be edited by normal users. Check the directory permissions too. =item Pool permissions It is important to check that the data files in __TOPDIR__/pool, __TOPDIR__/pc and __TOPDIR__/trash cannot be read by normal users. Normal users should not be able to see anything below __TOPDIR__. =item Host shares Enabling shares on hosts carries security risks. If you are on a private network and you generally trust your users then there should not be a problem. But if you have a laptop that is sometimes on public networks (eg: broadband or even dialup) you should be concerned. A conservative approach is to use firewall software, and only enable the netbios and smb ports (137 and 139) on connections from the host running BackupPC. =item SSH key security Using ssh for linux/unix clients is quite secure, but the security is only as good as the protection of ssh's private keys. If an attacker can devise a way to run a shell as the BackupPC user then they will have access to BackupPC's private ssh keys. They can then, in turn, ssh to any client machine as root (or whichever user you have configured BackupPC to use). This represents a serious compromise of your entire network. So in vulnerable networks, think carefully about how to protect the machine running BackupPC and how to prevent attackers from gaining shell access (as the BackupPC user) to the machine. =item CGI interface The CGI interface, __CGIDIR__/BackupPC_Admin, needs access to the pool files so it is installed setuid to __BACKUPPCUSER__. The permissions of this file need to checked carefully. It should be owned by __BACKUPPCUSER__ and have user and group (but not other) execute permission. To allow apache/httpd to execute it, the group ownership should be something that apache/httpd belongs to. The Apache configuration should be setup for AuthConfig style, using a .htaccess file so that the user's name is passed into the script as $ENV{REMOTE_USER}. If normal users could directly run BackupPC_Admin then there is a serious security hole: since it is setuid to __BACKUPPCUSER__ any user can browse and restore any backups. Be aware that anyone who is allowed to edit or create cgi scripts on your server can execute BackupPC_Admin as any user! They simply write a cgi script that sets $ENV{REMOTE_USER} and then execs BackupPC_Admin. The exec succeeds since httpd runs the first script as user httpd/apache, which in turn has group permission to execute BackupPC_Admin. While this setup should be safe, a more conservative approach is to run a dedicated Apache as user __BACKUPPCUSER__ on a different port. Then BackupPC_Admin no longer needs to be setuid, and the cgi directories can be locked down from normal users. Moreover, this setup is exactly the one used to support mod_perl, so this provides both the highest performance and the lowest security risk. =back Comments and suggestions are welcome. =head1 Configuration File The BackupPC configuration file resides in __TOPDIR__/conf/config.pl. Optional per-PC configuration files reside in __TOPDIR__/pc/$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 contains this line: do "../../conf/ConfigWinDesktop.pl"; This alternative allows you to set other configuration options specific to each host (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 switched to 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 1.0.0, 1.1.0, 1.2.0 and 1.3.0. =head1 Author Craig Barratt See L. =head1 Copyright Copyright (C) 2001-2002 Craig Barratt =head1 Credits Ryan Kucera contributed the directory navigation code and images for v1.5.0. He also contributed the first skeleton of BackupPC_restore. Guillaume Filion wrote BackupPC_zipCreate and added the CGI support for zip download, in addition to some CGI cleanup, for v1.5.0. Several people have reported bugs or made useful suggestions; see the ChangeLog. 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.