howto

一套客户/服务器模式的备份系统代码,跨平台,支持linux,AIX, IRIX, FreeBSD, Digital Unix (OSF1), Solaris and HP-UX.

egrep -i '(eject|offl|unload)' /usr/include/sys/*.h

On Linux also try /usr/include/linux/*.h and /usr/include/asm/*.h.
You should find macros defined in headers with names giving hints
to several kinds of devices. Look into the header, whether the
macros could be used with the ioctl system call. The comments
should tell details. Then you can eject the media with the
following code fragment:

#include <sys/ioctl.h>
#include <your_device_related_header>

{
  int   res, fd;
  char  *devicefile = "/dev/whatever";

  fd = open(devicefile, O_RDONLY);

  if(fd < 0){
    /* catch error */
    ...
  }

  res = ioctl(fd, YOUR_EJECT_MACRO);

  if(res < 0){
    /* catch error */ ...
  }

  close(fd);
}

You might want to extend the utility obtainable via ftp from:
ftp://ftp.zn.ruhr-uni-bochum.de/pub/Linux/eject.c and related
files. Please send me any success news. Thanks !


--------------------------------------------------------------------------

4. How to use several streamer devices on one machine ?

Run an installation of the server side for each streamer device,
install everything into a separate directory and give a different
port number to each installed server. This can be done giving each
server an own service name. For the default installation, the
service is named "afbackup" and has port number 2988. Thus, entries
are provided in files in /etc:

/etc/services:
afbackup  2988/tcp

/etc/inetd:
afbackup stream tcp nowait ...

For a second server, you may add appropriate lines, e.g.:

/etc/services:
afbackup2 2989/tcp

/etc/inetd.conf:
afbackup2 stream tcp nowait ...

Note, that the paths to the configuration files later in the inetd.conf-
lines must be adapted to each installation, respectively. To get the
services active, send a Hangup-Signal to the inetd.
(ps ..., kill -HUP <PID>)

It is important, that every server of several running on the same
host has it's own lock file. So e.g. configure lockfiles, that
are located in each server's var-directories. If they all share
one lockfile, several servers cannot run at the same time, what
is usually not, what you want.

The relations between backup clients and streamer devices on the
server must be unique. Thus the /etc/services on the clients must
contain the appropriate port number for the backup entry, e.g.:

afbackup  2990/tcp

Note, that on the clients the service name must always be "afbackup"
and not "afbackup2" or whatever.

As an alternative, you can supply the individual port number in
the clientside configuration. If you do so, no changes must be
made in any clientside system file, here /etc/services.

Do not use NIS (YP) for maintaining the afbackup-services-entry, i.e.
do not add the entry with "afbackup" above to your NIS-master-services-file.
It is anyway better not to use the files /etc/passwd ... as sources
for your NIS-master-server, but to use a copy of them in a separate
directory (as usually configured on Solaris and other Unixes).


--------------------------------------------------------------------------

5. How to recover from a server crash during backup ?

With some devices there will be the problem, that the end-of-tape mark
is not written on power-down during writing to the tape. Even worse,
when power is up again, the position, where the head is currently placed,
gets corrupt, even if no write access has been applied at power-down.
Some streamers are furthermore unable to start to write at a tape
position, where still records follow, e.g. if there are 5 files on tape,
it is e.g. impossible to go to file 2 and start to write there. An
I/O-error will be reported.

The only way to solve this is to tell the backup system to start to
write at the beginning of the next cartridge. If the next cartridge
has e.g. the label-number 5, log on to the backup server, become root
and type:

  /your/path/to/server/bin/cartis -i 5 1


--------------------------------------------------------------------------

6. How to port to other operating systems ?


* Unix-like systems *

This is not that difficult. The GNU-make is mandatory, but this is
usually no problem. A good way to start is to grep for AIX or sun
over all .c- and .h-files, edit them as needed and run the make.
You might want to run the prosname.sh to find out a specifier for
your operating system. This specifier will be defined as a macro
during compilation (exactly: prepocessing).

An important point is the x_types.h-file. Here the types should be
adapted as described in the comments in this file, lines 28-43.
Insert ifdef-s as needed like for the OSF 1 operating system on alpha
(macros __osf__ and __alpha). Note, that depending on the macro
USE_DEFINE_FOR_X_TYPES the types will be #define-d instead of
typedef-d. This gives you more flexibility, if one of those
possibilities is making problems.

The next point is the behaviour of the C-library concerning the
errno-variable in case the tape comes to it's physical end. In most
cases errno is set to ENOSPC, but not always (e.g. AIX is special).
This can be adapted modifying the definition of the macro
END_OF_TAPE (in budefs.h). This macro is only used in if-s as shown:
  if(END_OF_TAPE) ...
Consult your man-pages for the behaviour of the system calls on
your machine. It might be found under rmt, write or ioctl.

The next is the default name of the tape device. Define the macro
DEFAULT_TAPE_DEVICE (in budefs.h) appropriately for your OS.

A little pathological is the statfs(2) system call. It has a different
number of arguments depending on the system. Consult your man-pages,
how it should be used. statfs is only used in write.c

There may be further patches to be done, but if your system is close
to POSIX this should be easy. The output of the compiler and/or the
linker should give the necessary hints.

Please report porting successes to [e-mail]. Thanks.

Good luck !



* Win-whatever *

This is my point of view:

Porting to Microsoft's Features-and-bugs-accumulations is systematically
made complicated by the Gates-Mafia. They spend a lot of time on taking
care, that it is as difficult as possible to port to/from Win-whatever.
This is one of their monopolization strategies. Developers starting to
write programs shall have to make the basic decision: "Am i gonna hack
for Micro$oft's "operating systems", or for the others ?" Watching the
so-called market this decision is quite easy: Of course they will program
for the "market leader". And as few as possible of what they produce
should be usable on other ("dated") platforms. Companies like Cygnus
are providing cool tools (e. g. a port of the GNU-compiler) to make
things easier but due to the fact, that M$ are not providing so many
internals to the public, in my opinion porting is nonetheless an
annoying job. Thank Bill Gates for his genious strategies.

In short, at the moment i'm not gonna provide information how to port
to Micro$oft-platforms. If somebody will do a port, i don't hinder him
but will not provide any support for it. As this software (like the most
on Unix) heavily relies on POSIX-conformance and Mafia$oft has announced,
that the "POSIX-subsystem for NT" will not be shipped anymore in the near
future (BTW they discourage to use it at all "cause of security problems"
(Bullshit) - see the Microsoft web pages), the porting job will either
substitute all POSIX-calls by Win32-stuff (super-heavy efforts), or bring
only temporary fun (see above).


--------------------------------------------------------------------------

7. How to provide recovery from hard crashes (disk crash, ...) ?

A key to this is the clientside StartupInfoProgram parameter. This
command should read the standard input and write it to some place
outside of the local machine - to be more precise - not to a disk
undergoing backups or containing the clientside backup log files.
The information written to the standard input of this program is
the minimum information required to restore everything after a
complete loss of the saved filesystems and the client side of the
backup system. Recovery can be achieved using the restore-utility
with the -e flag (See: PROGRAMS) and supplying the minimum recovery
information to the standard input of restore. Several options exist:

- Write this information to a mail-program (assumed the mail folders
  are outside of the filesystems undergoing backup) and sending this
  information to a backup-user. Later the mailfile can be piped into
  the restore-utility (mail-related protocol lines and other unneeded
  stuff will be ignored). For each machine, that is a backup client,
  an individual mail user should be configured, cause the minimum
  restore information does NOT contain the hostname (to be able to
  restore to a different machine, what might make perfect sense in
  some situations)

- Write the information into a file (of course: always append),
  that resides on an NFS-mounted filesystem, eventually for security
  reasons exported especially to this machine only. To be even more
  secure, the exported directory might be owned by a non-root-user,
  who is the only one, who may write to this directory. This way it
  can be avoided to export a directory with root-access. Then the
  StartupInfoProgram should be something like:
   su myuser -c "touch /path/to/mininfo; cat >> /path/to/mininfo"
  The mininfo-file should have a name, that allows to deduce the
  name of the backup-client, that wrote it. E.g. simply use the
  hostname for this file.

- Write the information to a file on floppy disk. Then the floppy
  disk must always be in the drive, whenever a backup runs. The
  floppy could be mounted using the amd automounter as explained in
  ftp://ftp.zn.ruhr-uni-bochum.de/pub/linux/README.amd.floppy.cdrom
  or using the mtools usually installed for convenience. In the
  former case the command should contain a final sync. In the
  latter case the file must be first copied from floppy, then
  appended the information, finally copied back to floppy e.g. like
  this:
   mcopy -n a:mininfo /tmp/mininfo.$$; touch /tmp/mininfo.$$; \
       cat >> /tmp/mininfo.$$; mcopy -o /tmp/mininfo.$$ a:mininfo; \
       /bin/rm -f /tmp/mininfo.$$; exit 0
  Note, that the whole command must be entered in one line using
  the (x)afclientconfig command. In the configuration file lineend
  escaping is allowed, but not recognized by (x)afclientconfig. An
  alternative is to put everything into one script, that is started
  as StartupInfoProgram (Don't forget to provide a good exit code
  on successful completion)

My personal favourite is the second option, but individual preferences
or requirements might lead to different solutions. There are more
options here. If someone thinks, i have forgotten an important one,
feel free to email me about it.

It might be a good idea to compile afbackup linked statically with
all required libraries (building afbackup e.g. using the command
make EXTRA_LD_FLAGS=-static when using gcc), install it, run the
configuration program(s), if not yet done, tar everything and
put it to a floppy disk (if enough space is available).

To recover from a heavy system crash perform the following steps:
- Replace bad disk(s) as required
- Boot from floppy or cdrom (the booted kernel must be network-able)
- Add the backup server to /etc/hosts and the following line to
  /etc/services: afbackup 2988/tcp
- Mount your new disk filesystem(s) e.g. in /tmp/a and in a way, that
  this directory reflects your original directory hierarchy below
  / (like usually most system setup tools do)
- Untar your packed and statically linked afbackup-distribution, but
  NOT to the place, where it originally lived (e.g. /tmp/a/usr/backup),
  cause it will be overwritten, if you also saved the clientside
  afbackup-installation, what i strongly recommend.
- Run the restore-command with -e providing the minimum restore
  information saved outside of the machine to stdin:
  /path/to/staticlinked/afrestore -C /tmp/a -e < /path/to/mininfo-file

Bootsector stuff is NOT restored in this procedure. For Linux
you will have to reinstall lilo, but this is usually no problem.


--------------------------------------------------------------------------

8. How to make differential backups ?

A differential backup means for me: Save all filesystem entries modified
since the previous full backup, not only those modified since the last
incremental backup.

This task can be accomplished using the -a option of the incr_backup
command. It tells incr_backup to keep the timestamp. If -a is omitted
one time, another differential backup is no longer possible since the
timestamp is modified without -a. So if differential backups are required,
you have to do without incremental backups.


--------------------------------------------------------------------------

9. How to use several servers for one client ?

Several storage units can be configured for one client. A storage unit
is a combination of a hostname, a port number and a cartridge set number.
Several servers can be configured on one machine, each operating an own
streamer device or directory for storing the data.

The storage units are configured by the first three parameters of the
client side. These are hostnames, port numbers and cartridge set numbers,
respectively. Several entries can be made for each of these parameters.