path: root/doc/build.texinfo

Starting with the Beta 4 distribution, we are using a new configuration
system, which was built using the Free Software Foundation's
@samp{autoconf} program.  This system will hopefully make Kerberos V5
much simpler to build and reduce the amount of effort required in
porting Kerberos V5 to a new platform.

@menu
* Build Requirements::          How much disk space, etc. you need to
                                           build Kerberos.
* Unpacking the Sources::       Preparing the source tree.
* Doing the Build::             Compiling Kerberos.
* Testing the Build::           Making sure Kerberos built correctly.
* Options to Configure::        Command-line options to Configure
* osconf.h::                    Header file-specific configurations
* Shared Library Support::      Building Shared Libraries for Kerberos V5
* OS Incompatibilities::        Special cases to watch for.
* Using Autoconf::              Modifying Kerberos V5's 
                                            configuration scripts.
@end menu

@node Build Requirements, Unpacking the Sources, Building Kerberos V5, Building Kerberos V5
@section Build Requirements

In order to build Kerberos V5, you will need approximately 60-70
megabytes of disk space.  The exact amount will vary depending on the
platform and whether the distribution is compiled with debugging symbol
tables or not.

If you wish to keep a separate @dfn{build tree}, which contains the compiled
@file{*.o} file and executables, separate from your source tree, you
will need a @samp{make} program which supports @samp{VPATH}, or
you will need to use a tool such as @samp{lndir} to produce a symbolic
link tree for your build tree.

@node Unpacking the Sources, Doing the Build, Build Requirements, Building Kerberos V5
@section Unpacking the Sources

The first step in each of these build procedures is to unpack the source
distribution.  The Kerberos V5 distribution comes in two compressed tar
files.  The first file, which is generally named
@file{krb5-@value{RELEASE}.src.tar.gz}, contains the sources for all of Kerberos
except for the crypto library, which is found in the file
@file{krb5-@value{RELEASE}.crypto.tar.gz}.

Both files should be unpacked in the same directory, such as
@file{/u1/krb5-@value{RELEASE}}.  (In the rest of this document, we will assume that
you have chosen to unpack the Kerberos V5 source distribution in this
directory.  Note that the tarfiles will by default all unpack into the
@file{./krb5-@value{RELEASE}} directory, so that if your current directory is
@file{/u1} when you unpack the tarfiles, you will get
@file{/u1/krb5-@value{RELEASE}/src}, etc.)


@node Doing the Build, Testing the Build, Unpacking the Sources, Building Kerberos V5
@section Doing the Build

You have a number of different options in how to build Kerberos.  If you
only need to build Kerberos for one platform, using a single directory
tree which contains both the source files and the object files is the
simplest.  However, if you need to maintain Kerberos for a large number
of platforms, you will probably want to use separate build trees for
each platform. We recommend that you look at @ref{OS
Incompatibilities}, for notes that we have on particular operating
systems. 

@menu
* Building Within a Single Tree::  
* Building with Separate Build Directories::  
* Building using lndir::        
@end menu

@node Building Within a Single Tree, Building with Separate Build Directories, Doing the Build, Doing the Build
@subsection Building Within a Single Tree

If you don't want separate build trees for each architecture, then
use the following abbreviated procedure.

@enumerate
@item
 @code{cd /u1/krb5-@value{RELEASE}/src}
@item
 @code{./configure}
@item
 @code{make}
@end enumerate

That's it!

@node Building with Separate Build Directories, Building using lndir, Building Within a Single Tree, Doing the Build
@subsection Building with Separate Build Directories

If you wish to keep separate build directories for each platform, you
can do so using the following procedure.  (Note, this requires that your
@samp{make} program support @samp{VPATH}.  GNU's make will provide this
functionality, for example.)  If your @samp{make} program does not
support this, see the next section.

For example, if you wish to create a build directory for @code{pmax} binaries
you might use the following procedure:

@enumerate
@item 
@code{mkdir /u1/krb5-@value{RELEASE}/pmax}
@item
 @code{cd /u1/krb5-@value{RELEASE}/pmax}
@item
 @code{../src/configure}
@item
 @code{make}
@end enumerate

@node Building using lndir,  , Building with Separate Build Directories, Doing the Build
@subsection Building Using @samp{lndir}

If you wish to keep separate build directories for each platform, and
you do not have access to a @samp{make} program which supports @samp{VPATH},
all is not lost.  You can use the @samp{lndir} program to create
symbolic link trees in your build directory.

For example, if you wish to create a build directory for solaris binaries
you might use the following procedure:

@enumerate
@item
 @code{mkdir /u1/krb5-@value{RELEASE}/solaris}
@item
 @code{cd /u1/krb5-@value{RELEASE}/solaris}
@item
 @code{/u1/krb5-@value{RELEASE}/src/util/lndir `pwd`/../src}
@item
 @code{./configure}
@item
 @code{make}
@end enumerate

You must give an absolute pathname to @samp{lndir} because it has a bug that
makes it fail for relative pathnames. Note that this version differs
from the latest version as distributed and installed by the XConsortium
with X11R6. Either version should be acceptable.

@node Testing the Build, Options to Configure, Doing the Build, Building Kerberos V5
@section Testing the Build

The Kerberos V5 distribution comes with built-in regression tests.  To
run them, simply type the following command while in the top-level build
directory (i.e., the directory where you sent typed @samp{make} to start
building Kerberos; see @ref{Doing the Build}.): 

@example
% make check
@end example

@menu
* The DejaGnu Tests::           
* The KADM5 Tests::     
@end menu

@node The DejaGnu Tests,  The KADM5 Tests, Testing the Build, Testing the Build
@subsection The DejaGnu Tests 

Some of the built-in regression tests are setup to use the DejaGnu
framework for running tests. These tests tend to be more comprehensive
than the normal built-in tests as they setup test servers and test
client/server activities. 

DejaGnu may be found wherever GNU software is archived.  

Most of the tests are setup to run as a non-privledged user.  For some
of the krb-root tests to work properly, either (a) the user running the
tests must not have a .k5login file in the home directory or (b) the
.k5login file must contain an entry for @code{<username>@@KRBTEST.COM}.
There are two series of tests (@samp{rlogind} and @samp{telnetd}) which
require the ability to @samp{rlogin} as root to the local
machine. Admittedly, this does require the use of a @file{.rhosts} file
or some authenticated means. @footnote{If you are fortunate enough to
have a previous version of Kerberos V5 or V4 installed, and the Kerberos
rlogin is first in your path, you can setup @file{.k5login} or
@file{.klogin} respectively to allow you access.}

If you cannot obtain root access to your machine, all the other tests
will still run. Note however, with DejaGnu 1.2, the "untested testcases"
will cause the testsuite to exit with a non-zero exit status which
@samp{make} will consider a failure of the testing process. Do not worry
about this, as these tests are the last run when @samp{make check} is
executed from the top level of the build tree.  This problem does not
exist with DejaGnu 1.3.

@node The KADM5 Tests,  , The DejaGnu Tests, Testing the Build
@subsection The KADM5 Tests

Regression tests for the KADM5 system, including the GSS-RPC, KADM5
client and server libraries, and kpasswd, are also included in this
release.  Each set of KADM5 tests is contained in a sub-directory called
@code{unit-test} directly below the system being tested.  For example,
lib/rpc/unit-test contains the tests for GSS-RPC.  The tests are all
based on DejaGnu (but they are not actually called part of "The DejaGnu
tests," whose naming predates the inclusion of the KADM5 system).  In
addition, they require the Tool Command Language (TCL) header files and
libraries to be available during compilation and some of the tests also
require Perl in order to operate.  If all of these resources are not
available during configuration, the KADM5 tests will not run.  The TCL
installation directory can be specified with the @code{--with-tcl}
configure option.  (See @xref{Options to Configure}.)  The runtest and
perl programs must be in the current execution path.

If you install DejaGnu, TCL, or Perl after configuring and building
Kerberos and then want to run the KADM5 tests, you will need to
re-configure the tree and run @code{make} at the top level again to make
sure all the proper programs are built.  To save time, you actually only
need to reconfigure and build in the directories src/kadmin/testing,
src/lib/rpc, src/lib/kadm5.

@node Options to Configure, osconf.h, Testing the Build, Building Kerberos V5
@section Options to Configure 

There are a number of options to @samp{configure} which you can use to
control how the Kerberos distribution is built.  The following table
lists the most commonly used options to Kerberos V5's @samp{configure}
program.

@table @code

@item --help

Provides help to configure. This will list the set of commonly used
options for building Kerberos.

@item --prefix=PREFIX

By default, Kerberos will install the package's files rooted at
`/usr/local' as in `/usr/local/bin', `/usr/local/sbin', etc. If you
desire a different location, use this option.

@item --exec-prefix=EXECPREFIX

This option allows one to separate the architecture independent programs
from the configuration files and manual pages. 

@item --localstatedir=LOCALSTATEDIR

This option sets the directory for locally modifiable single-machine
data.  In Kerberos, this mostly is useful for setting a location for the
KDC data files, as they will be installed in
@code{LOCALSTATEDIR/krb5kdc}, which is by default
@code{PREFIX/var/krb5kdc}.

@item --with-cc=COMPILER

Use @code{COMPILER} as the C compiler.

@item --with-ccopts=FLAGS

Use @code{FLAGS} as the default set of C compiler flags.

Note that if you use the native Ultrix compiler on a
DECstation you are likely to lose if you pass no flags to cc; md4.c
takes an estimated 3,469 billion years to compile if you provide neither
the @samp{-g} flag nor the @samp{-O} flag to @samp{cc}.

@item --with-cppopts=CPPOPTS 

Use @code{CPPOPTS} as the default set of C preprocessor flags.  The most
common use of this option is to select certain @code{#define}'s for use
with the operating system's include files.

@item --with-linker=LINKER

Use @code{LINKER} as the default loader if it should be different from C
compiler as specified above.

@item --with-ldopts=LDOPTS

This option allows one to specify optional arguments to be passed to the
linker. This might be used to specify optional library paths.

@item  --with-krb4

This option enables Kerberos V4 backwards compatibility using the
builtin Kerberos V4 library.

@item  --with-krb4=KRB4DIR 

This option enables Kerberos V4 backwards compatibility using a
pre-existing Kerberos V4 installation.  The directory specified by
@code{KRB4DIR} specifies where the V4 header files should be found
(@file{KRB4DIR/include}) as well as where the V4 Kerberos library should
be found (@file{KRB4DIR/lib}).

@item  --without-krb4

Disables Kerberos V4 backwards compatibility. This prevents Kerberos V4
clients from using the V5 services including the KDC. This would be
useful if you know you will never install or need to interact with V4
clients. 

@item --with-netlib[=libs]

Allows for suppression of or replacement of network libraries. By
default, Kerberos V5 configuration will look for @code{-lnsl} and
@code{-lsocket}. If your operating system has a broken resolver library
(see @ref{Solaris versions 2.0 through 2.3}) or fails to pass the tests in
@file{src/tests/resolv} you will need to use this option.

@item --with-vague-errors

If enabled, gives vague and unhelpful error messages to the client... er,
attacker.  (Needed to meet silly government regulations; most other
sites will want to keep this undefined.)

@item --with-kdc-kdb-update

Set this option if you want to allow the KDC to modify the Kerberos
database; this allows the last request information to be updated, as
well as the failure count information.  Note that this doesn't work if
you're using slave servers!!!  It also causes the database to be
modified (and thus needing to be locked) frequently. Please note that
the implementors do not regularly test this feature.

@item --with-tcl=TCLPATH

Some of the unit-tests in the build tree rely upon using a program in
Tcl. The directory specified by @code{TCLPATH} specifies where the Tcl
header file (@file{TCLPATH/include/tcl.h} as well as where the Tcl
library should be found (@file{TCLPATH/lib}).

@item --enable-shared

This option will turn on the building and use of shared library objects
in the Kerberos build. This option is only supported on certain
platforms. 

@item --enable-dns
@item --enable-dns-for-kdc
@item --enable-dns-for-realm

Enable the use of DNS to look up a host's Kerberos realm, or a realm's
KDCs, if the information is not provided in krb5.conf.  See
@xref{Hostnames for the Master and Slave KDCs}, and @xref{Mapping
Hostnames onto Kerberos Realms}.  By default, DNS lookups are enabled
for the latter but not for the former.

@item --enable-kdc-replay-cache

Enable a cache in the KDC to detect retransmitted messages, and resend
the previous responses to them.  This protects against certain types of
attempts to extract information from the KDC through some of the
hardware preauthentication systems.

@end table

For example, in order to configure Kerberos on a Solaris machine using
the @samp{suncc} compiler with the optimizer turned on, run the configure
script with the following options:

@example
% ./configure --with-cc=suncc --with-ccopts=-O
@end example

@node osconf.h, Shared Library Support, Options to Configure, Building Kerberos V5
@section @file{osconf.h}

There is one configuration file which you may wish to edit to control
various compile-time parameters in the Kerberos distribution:
@file{include/krb5/stock/osconf.h}. The list that follows is by no means
complete, just some of the more interesting variables.

Please note: The former configuration file @file{config.h} no longer
exists as its functionality has been merged into the auto-configuration
process. @xref{Options to Configure}.


@table @code

@item DEFAULT_PROFILE_PATH

The pathname to the file which contains the profiles for the known
realms, their KDCs, etc.

The profile file format is no longer the same format as Kerberos V4's
@file{krb.conf} file. 

@item DEFAULT_KEYTAB_NAME

The type and pathname to the default server keytab file (the equivalent
of Kerberos V4's @file{/etc/srvtab}).  

@item DEFAULT_KDC_ENCTYPE

The default encryption type for the KDC.

@item KDCRCACHE

The name of the replay cache used by the KDC.

@item RCTMPDIR

The directory which stores replay caches.

@item DEFAULT_KDB_FILE

The location of the default database

@end table

@node Shared Library Support, OS Incompatibilities, osconf.h, Building Kerberos V5
@section Shared Library Support

Shared library support is provided for a few operating systems. There
are restrictions as to which compiler to use when using shared
libraries. In all cases, executables linked with the shared libraries in
this build process will have built in the location of the libraries,
therefore obliterating the need for special LD_LIBRARY_PATH, et al environment
variables when using the programs. Except where noted, multiple versions
of the libraries may be installed on the same system and continue to
work.

Currently the supported platforms are
@comment NetBSD 1.0A, AIX 3.2.5, AIX 4.1,
Solaris 2.6 (aka SunOS 5.6) and Irix 6.5.
@comment Alpha OSF/1 >= 2.1, HP-UX >= 9.X.

To enable shared libraries on the above platforms, run the configure
script with the option @samp{--enable-shared}.

@ifset notdef

XXX What does this mean?

One special note is that if the Kerberos V4 compatibility is compiled
in, you @b{must not} specify an alternate Kerberos V4 library from the
one in the tree or you will be missing references.

@end ifset

@node OS Incompatibilities, Using Autoconf, Shared Library Support, Building Kerberos V5
@section Operating System Incompatibilities

This section details operating system incompatibilities with Kerberos V5
which have been reported to the developers at MIT.  If you find
additional incompatibilities, and/or discover work arounds to such
problems, please send a report via the @code{krb5-send-pr} program.
Thanks!

@menu
* AIX::                         
* Alpha OSF/1 V1.3::            
* Alpha OSF/1 (Digital Unix) V2.0++::  
* BSDI::                        
* HPUX::                        
* Solaris versions 2.0 through 2.3::  
* Solaris 2.X::                 
* SGI Irix 5.X::                
* Ultrix 4.2/3::                
@end menu

@node AIX, Alpha OSF/1 V1.3, OS Incompatibilities, OS Incompatibilities
@subsection AIX

The AIX 3.2.5 linker dumps core trying to build a shared
@samp{libkrb5.a} produced with the GNU C compiler.  The native AIX
compiler works fine.  This problem is fixed using the AIX 4.1 linker.

@node Alpha OSF/1 V1.3, Alpha OSF/1 (Digital Unix) V2.0++, AIX, OS Incompatibilities
@subsection Alpha OSF/1 V1.3

Using the native compiler, compiling with the @samp{-O} compiler flag
causes the @code{asn.1} library to be compiled incorrectly.  

Using GCC version 2.6.3 or later instead of the native compiler will also work
fine, both with or without optimization.

@node Alpha OSF/1 (Digital Unix) V2.0++, BSDI, Alpha OSF/1 V1.3, OS Incompatibilities
@subsection Alpha OSF/1 V2.0++

There used to be a bug when using the native compiler in compiling
@file{md4.c} when compiled without either the @samp{-O} or @samp{-g}
compiler options.  We have changed the code and there is no problem
under V2.1, but we do not have access to V2.0 to test and see if the
problem would exist there. (We welcome feedback on this issue). There
was never a problem in using GCC version 2.6.3.

In version 3.2 and beyond of the operating system, we have not seen any
problems with the native compiler. 

@c @node Alpha Tru64 UNIX 5.0
@c @subsection Alpha Tru64 UNIX 5.0
@c ... login.krb5 problems

@node BSDI, HPUX, Alpha OSF/1 (Digital Unix) V2.0++, OS Incompatibilities
@subsection BSDI

BSDI versions 1.0 and 1.1 reportedly has a bad @samp{sed} which causes
it to go into an infinite loop during the build.  The work around is
to use a @samp{sed} from somewhere else, such as GNU.  (This may be
true for some versions of other systems derived from BSD 4.4, such as
NetBSD and FreeBSD.)

@node HPUX, Solaris versions 2.0 through 2.3, BSDI, OS Incompatibilities
@subsection HPUX

The native (bundled) compiler for HPUX currently will not work, because
it is not a full ANSI C compiler.  The optional compiler (c89) should
work as long as you give it the @samp{+Olibcalls -D_HPUX_SOURCE} (this
has only been tested for HPUX 9.0).  At this point, using GCC is
probably your best bet.

@node Solaris versions 2.0 through 2.3, Solaris 2.X, HPUX, OS Incompatibilities
@subsection Solaris versions 2.0 through 2.3

The @code{gethostbyname()} routine is broken; it does not return a fully
qualified domain name, even if you are using the Domain Name Service
routines.  Since Kerberos V5 uses the fully qualified domain name as the
second component of a service principal (i.e,
@samp{host/tsx-11.mit.edu@@ATHENA.MIT.EDU}), this causes problems for servers
who try to figure out their own fully qualified domain name.  

Workarounds:  

@enumerate

@item
   Supply your own resolver library. (such as bind-4.9.3pl1 available
from ftp.vix.com)

@item
   Upgrade to Solaris 2.4

@item
   Make sure your /etc/nsswitch.conf has `files' before `dns' like:

@example
hosts:      files dns
@end example

and then in /etc/hosts, make sure there is a line with your
workstation's IP address and hostname, with the fully qualified domain
name first.  Example:

@example
18.172.1.4      dcl.mit.edu dcl
@end example

Note that making this change may cause other programs in your
environment to break or behave differently.

@end enumerate

@node Solaris 2.X, SGI Irix 5.X, Solaris versions 2.0 through 2.3, OS Incompatibilities
@subsection Solaris 2.X

You @b{must} compile Kerberos V5 without the UCB compatibility
libraries.  This means that @file{/usr/ucblib} must not be in the
LD_LIBRARY_PATH environment variable when you compile it.  Alternatively
you can use the @code{-i} option to @samp{cc}, by using the specifying
@code{--with-ccopts=-i} option to @samp{configure}.

@node SGI Irix 5.X, Ultrix 4.2/3, Solaris 2.X, OS Incompatibilities
@subsection SGI Irix 5.X

If you are building in a tree separate from the source tree, the vendors
version of make does not work properly with regards to
@samp{VPATH}. It also has problems with standard inference rules in 5.2
(not tested yet in 5.3) so one needs to use GNU's make.

Under 5.2, there is a bug in the optional System V @code{-lsocket}
library in which the routine @code{gethostbyname()} is broken. The
system supplied version in @code{-lc} appears to work though so one may
simply specify @code{--with-netlib} option to @samp{configure}. 

In 5.3, @code{gethostbyname()} is no longer present in @code{-lsocket} and
is no longer an issue. 

@node Ultrix 4.2/3,  , SGI Irix 5.X, OS Incompatibilities
@subsection Ultrix 4.2/3

The DEC MIPS platform currently will not support the native compiler,
since the Ultrix compiler is not a full ANSI C compiler.  You should use
GCC instead.

@ifset notdef

On the DEC MIPS platform, using the native compiler, @file{md4.c} and
@file{md5.c} can not be compiled with the optimizer set at level 1.
That is, you must specify either @samp{--with-ccopts=-O} and
@samp{--with-ccopts=-g} to configure.  If you don't specify either, the
compile will never complete.

The optimizer isn't hung; it just takes an exponentially long time.
Compiling 6 out of the 48 algorithmic steps takes 3 seconds; compiling 7
steps takes 9 seconds; compiling 8 steps takes 27 seconds, and so on.
Calculations estimate it will finish in approximately 3,469 billion
years....

Using GCC instead of the native compiler will also work fine, both with
or without optimization.

@end ifset

@node Using Autoconf,  , OS Incompatibilities, Building Kerberos V5
@section Using @samp{Autoconf}

(If you are not a developer, you can skip this section.)

In most of the Kerberos V5 source directories, there is a
@file{configure} script which automatically determines the compilation
environment and creates the proper Makefiles for a particular platform.
These @file{configure} files are generated using @samp{autoconf} version
2.4, which can be found in the @file{src/util/autoconf} directory in the
distribution.

Normal users will not need to worry about running @samp{autoconf}; the
distribution comes with the @file{configure} files already prebuilt.
Developers who wish to modify the @file{configure.in} files should see
@ref{Top, , Overview, autoconf, The Autoconf Manual}.  

Note that in order to run @samp{autoconf}, you must have GNU @samp{m4}
in your path.  Before you use the @samp{autoconf} in the Kerberos V5
source tree, you may also need to run @samp{configure}, and then run
@samp{make} in the @file{src/util/autoconf} directory in order to
properly set up @samp{autoconf}.

One tool which is provided for the convenience of developers can be
found in @file{src/util/reconf}.  This program should be run while the
current directory is the top source directory.  It will automatically
rebuild any @file{configure} files which need rebuilding.  If you know
that you have made a change that will require that all the
@file{configure} files need to be rebuilt from scratch, specify the
@code{--force} option:

@example
% cd /u1/krb5-@value{RELEASE}/src
% ./util/reconf --force
@end example

The developmental sources are a raw source tree (before it's been packaged
for public release), without the pre-built @file{configure} files.
In order to build from such a source tree, you must do:

@example
% cd krb5/util/autoconf
% ./configure
% make
% cd ../..
% util/reconf
@end example

Then follow the instructions for building packaged source trees (above).
To install the binaries into a binary tree, do:

@example
% cd /u1/krb5-@value{RELEASE}/src
% make all
% make install DESTDIR=somewhere-else
@end example