OPIE Software Distribution, Release 2.4 Important Information
======================================= =====================
Introduction
============
"One-time Passwords In Everything" (OPIE) is a freely distributable
software package originally developed at and for the US Naval Research
Laboratory (NRL). Recent versions are the result of a cooperative effort
between of NRL, several of the original NRL authors, The Inner Net, and many
other contributors from the Internet community.
OPIE is an implementation of the One-Time Password (OTP) System that
is being considered for the Internet standards-track. OPIE provides a one-time
password system. The system should be secure against the passive attacks
now commonplace on the Internet (see RFC 1704 for more details). The system
is vulnerable to active dictionary attacks, though these are not widespread
at present and can be detected through proper use of system audit
software.
OPIE is primarily written for UNIX-like operating systems, but
we are working to make applicable portions portable to other operating systems.
The OPIE software is derived in part from and is fully interoperable with the
Bell Communications Research (Bellcore) S/Key Release 1 software. Because
Bellcore claims "S/Key" as a trademark for their software, NRL was forced to
use a different name (we picked "OPIE") for this software distribution.
OPIE includes the following additions/modifications to the
original Bellcore S/Key(tm) Version 1 software:
* Just about three command installation (unpack the software, run the
configure script, and run make install). While we still recommend that you
follow instructions and test things by hand, the more adventurous can
install OPIE quickly.
* A modified BSD FTP daemon that does OTP.
* A version of su that uses OTP by default.
* MD5 support. MD5 is now the default algorithm, though MD4 is still supported
by changing a parameter in the Makefile. This change was made because MD5 is
widely believed to be cryptographically stronger than MD4 (see RFC 1321).
* A more portable version of MD4 has been substituted for the original MD4.
This should solve the endian problems that were in S/Key.
* Most of the system-dependencies have been moved to a new file "opie_cfg.h".
* Configuration options have been moved to the Makefile.
* Isolated system dependencies (e.g. BSDisms) with appropriate #ifdefs.
* Revised the opiekey(1) program to simultaneously support MD4 and MD5, with
the default algorithm being tunable using the MDX symbol in the Makefile.
* More operating systems are supported by recent versions of OPIE, but older
BSD systems that aren't close to being compliant with the POSIX standard are
no longer supported.
* Transition mechanisms are optional to prevent potential back doors.
* On systems using the /etc/opieaccess transition mechanism, users can choose
to require the use of OPIE to login to their accounts when it would
otherwise be optional.
* Bug fixes
* Cosmetic changes
* Prompts (optionally) identify specifically what kind of entry (system
password, secret pass phrase, or OTP response) is allowed.
* Changes to mostly conform with the draft Internet OTP standard.
A Glance at What's New
======================
2.4 TEST VERSION -- NOT FOR REDISTRIBUTION
Merged in opieauto, which is disabled by default.
Lots of documentation updates.
Portability and bug fixes.
2.32 January 1, 1998.
Indicate support for extended responses in challenges and check for such
indication before generating any extended responses.
Lots of portability and bug fixes.
2.31 March 20, 1997.
Removed active attack protection support due to patent problems.
Removed the supplemental key file; it did more harm than good.
Moved user locks to a separate directory.
Moved user-serviceable configuration options to the configure script.
Lots of portability and bug fixes.
2.3 September 22, 1996
Autoconf is now the only supported configuration method.
Lots of internal functions got re-written in ways that will make some
planned future changes easier.
OTP extended responses, such as automatic re-initialization.
Support for a supplemental key file that stores information that was not
in the original /etc/skeykeys file. This allows OPIE to store extra data needed
for things like the OTP re-initialization extended response without breaking
interoperability with other S/Key derived programs. This file is named
"/etc/opiekeys.ext" by default. Unlike the standard key file, it MUST NOT be
world readable.
OPIE should better support some of the native "features" of drain bamaged
OSs such as AIX, HP-UX, and Solaris.
OPIE's utmp/wtmp handling has been completely re-written. This should solve
many of the utmp/wtmp problems people have been having.
Lots of cleanups.
Bug fixes.
2.22 May 3, 1996.
More minor bug fixes. OPIE once again works on Solaris 2.x.
2.21 April 27, 1996.
Minor bug fixes.
2.2 April 11, 1996.
opiesubr.c, opiesubr2.c, and a few other functions moved into a
subdirectory and split into files with fine granularity. Ditto with missing
function replacements. This subdirectory structure changes a lot of things
around and more splitting like this should be expected in the near future.
Added opiegenerator() library function that should make it very easy to
create OTP clients using the OPIE library (this function is subject to change:
there are a few problems remaining to be solved). Just about re-wrote
opiegetpass() to use raw I/O and got most of the OPIE programs actually using
that function. Autoconf build fixes. Lots of bug fixes. Lots of portability
fixes. Function declarations should be ANSI style for ANSI compilers. Several
fixes to bring OPIE in line with the latest OTP spec. MJR DES key crunch
de-implemented.
Added sample programs: opiegen (client) and opieserv (server).
Probably broke non-autoconf support along the way :(. I've tried to bring
this back in sync, but it may still be broken.
2.11 December 27, 1995.
Minor bug fixes.
2.10 December 26, 1995.
Optional autoconf support. opieinfo is now a normal program. Bugs fixed --
should work much better on SunOS, HP-UX, and AIX.
2.01 -- 2.04
Bug fix releases.
2.00
Initial release of OPIE 2.0.
System Requirements
===================
In order to build and run properly, OPIE requires:
* A UNIX-like operating system
* An ANSI C compiler and run-time library
* POSIX.1- and X/Open XPG-compliance (including termios)
* The BSD sockets API
* Approximately five megabytes of free disk space
In practice, we believe that many systems who are close to meeting
these requirements but aren't completely there (for example, SunOS with the
native compiler) will also work. Systems who aren't anywhere near close
(for example, DOS) are not likely to work without major adjustments to the
OPIE code.
If OPIE Doesn't Work
====================
Under NO circumstances should you send trouble reports directly to the
authors or contributors. They WILL BE IGNORED.
Make sure you have the latest version of OPIE. The latest version is
available by HTTP at:
http://www.inner.net/pub/opie
(sorry, but anonymous FTP is no longer available)
If you have installed the OPIE software (either through "make test"
in (7) above or "make install" in (14)), you can run "make uninstall" from the
OPIE software distribution directory. This should remove the OPIE software and
restore the original system programs, but it will not work properly (and can
even result in the total loss of the old system programs -- beware!) if the
installation procedure itself did not work properly.
If you are running a release version, try installing the latest public
test version (look around). These frequently have already fixed the problem
you are seeing, but may have new problems of their own (that's why they're
test versions!). Similarly, if you are running a test version, try installing
the latest released version.
OPIE is NOT supported software. We don't promise to support you or
even to acknowledge your mail, but we are interested in bug reports and are
reasonable folks. We also have an interest in seeing OPIE work on as many
systems as we can. However, if your system doesn't meet the basic requirements
for OPIE, this will probably require an unreasonable amount of effort.
The best bug reports include a diagnosis of the problem and a fix.
Your bug report can still be valuable if you can at least diagnose what the
problem is. If you just tell us "it doesn't work," then we won't be able to
do anything to help you.
We've received a number of bug reports from people that look
interesting, only to find when we try to follow up on them that the user
either has an invalid return address or never bothered to respond to our
followup. Please make sure that bug reports you send us have an electronic
mail address that we can reply to somewhere in them (if necessary, just
put it in the message body). If we send you a response and you are unable
to invest the time to work with us to solve the problem, please tell us --
few things are more irritating than when someone sends us information
about a bug that we'd like to fix and then is never heard from again.
We try to respond to all properly submitted bug reports. Improperly
submitted bug reports will be responded to only if we have time left after
responding to properly submitted bug reports. We deliberately ignore bug
"reports" sent to mailing lists or USENET news groups instead of or before
our bug report address. At the least, the latter practice is lacking in
courtesy.
The file BUG-REPORT contains our bug reporting form. Please use it
and follow the submission instructions in that file. We are going to switch
to machine-parsed bug report processing sometime in the near future to make
it easier to coordinate bug hunting.
Gotchas
=======
Solaris 2.x is just a lose. It does a lot of nonstandard and downright
broken things. If you want OPIE to be reliable on your box, upgrade to OpenBSD
or Linux.
While an almost universal "feature", most people remain unaware that
an intruder can log into a system, then log in again by running the "login"
command from a shell. Because the second login is from the local host, the
utmp entry will not show a remote login host anymore. The OPIE replacement
for /bin/login currently carries on this behavior for compatibility reasons.
If you would like to prevent this from happening, you should change the
permissions of /bin/login to 0100, thus preventing unprivileged users from
executing it. This fix should work on non-OPIE /bin/login programs as well.
On 4.3BSDish systems, the supplied /bin/login replacement obtains
the terminal type for the console comes from the console line in the /etc/ttys
file. Several systems contain a default entry in this file that specifies the
console terminal type as "unknown". This is probably not what you want.
The OPIE FTP daemon responds with two 530 error messages if you have
not yet logged in and execute a command that will also do a PORT request. This
is a feature, not a bug, as the FTP client is really sending the server two
commands (for instance, a PORT and a LIST if you tell your BSD FTP client to do
a DIR command) and the server is responding to each of them with an error. The
stock BSD FTP daemon doesn't check the PORT commands to see if you are logged
in, so you would only get one error message. This change should not break any
standards-compliant FTP client, but there are a number of brain-damaged GUI
clients that have a track record for not dealing gracefully with any server
other than the stock BSD one.
The /etc/opieaccess transition mechanism is, by definition, a security
hole in the OPIE software because an attacker could use it to circumvent the
requirement for OPIE authentication. You should compile the software with
support for this file disabled unless you absolutely cannot use the software
without it because of your environment. If you do use this support for
transition purposes, you should move people to OTP authentication as quickly
as possible and rebuild and reinstall OPIE with this transition support
disabled so that you won't have a lurking security hole.
If this wasn't already clear, do not let your sequence number fall
below about ten. If your sequence number reaches zero, your OTP sequence
can only be reset by the superuser. System administrators should make this
caveat known to their users.
On Solaris 2.x systems (and possibly others) running NIS+, users
should run keylogin(1) manually after login because opielogin(1) does not
do that automatically like the system login(1) program.
There are reports that some versions of GNU C Compiler (GCC)
(when installed on some systems) use their own termios(4) instead of
the system's termios(4). This can cause problems. If you are having
compilation problems that seem to relate to termios and you are using
GCC, you should probably verify that it is using the system's
termios(4) and not some internal-to-GCC termios(4). One report
indicates that Sun's C compiler works fine with SunOS 4.1.3/4.1.4 on
SPARC, but that some version of GCC on the same system has this
termios(4) problem. We haven't reproduced these problems ourselves
and hence aren't sure what is happening, but we pass this along for
your information. (This may have something to do with the use of GNU
libc)
If a user has a valid entry in the opiekeys database but has an
asterisk in their traditional password entry, they will not be able to
log in via opielogin, but opielogin will decrement their sequence number
if a valid response is received.
On some systems, the OPIE login program does not always display
a "login:" prompt the first time. There is a race condition in many older
telnetds that is probably the cause of this problem. This should be fixed by
replacing your telnetd with the latest version of the stock telnetd
(ftp.cray.com:/src/telnet).
The standard HPUX compiler is severely drain bamaged. One of the
worst parts is that it sometimes won't grok a symbol definition with forward
slashes in them properly and can choke badly on the definition of the key
file's location. If this happens to you, install and use GCC. (This problem
may or may not also come up with the optional HP ANSI C compiler -- we don't
know for sure what compilers have this problem).
As of OPIE 2.2, the seed is converted to lower case and its length is
checked in order to comply with the OTP specification. If any of your users
have seeds that use capital letters or are too long, they need to run the OPIE
2.2 opiepasswd program to re-initialize their sequence to one with a different
seed.
opielogin is a replacement for /bin/login. It is NOT an OPIE "shell."
You can use it as one, but don't be surprised if it doesn't behave the way
you expect -- we've seen various reports of success and failure when used this
way. An OPIE "shell" is on the TODO list.
Clients that use opiegen() will automatically send a re-initialization
extended response if the sequence number falls below ten. If the server does
not support this, the user will need to log in using opiekey and reset his
sequence manually (using opiepasswd).
For reasons that remain very unclear, Solaris passes the login name
from getty/telnetd to login by stuffing it in the terminal input buffer
instead of passing it on the command line like every other *IX. This is just
plain broken. Solaris has other problems with its telnetd and getty; you may
want to consider getting the telnet(d) sources (ftp.cray.com:/src/telnet)
and reasonable getty sources (try sunsite.unc.edu:/pub/Linux/system/Serial, at
least one of agetty, mingetty, and getty_ps should work) and replacing the
Solaris versions with these. OPIE should work *much* more happily with these
programs than the ones that come with Solaris. However, there could be negative
side effects -- this is not a procedure recommended for the faint of heart.
OPIE is a lot more fussy than it used to be about lock files and where
it puts them. The lock file directory must be a directory used only for OPIE
lock files. It must be a directory, owned by the superuser, and must be mode
0700.
opieauto is a potential security hole. It opens a limited window of
exposure by transmitting and storing information that can be used to
generate one or more OTPs earlier than the current sequence number. Every
effort has been made to limit the potential for compromise to the user-
specified window. However, an attacker with superuser priveleges or access to
your account on the client system can still generate OTPs based on the
information cached via opieauto. In practice, there are other ways for such an
an attacker to get your entire secret pass phrase, so this is probably not
creating a significant new security problem. However, because of this
potential for problems and because opieauto uses system features that are not
present on all systems, opieauto support is not compiled in by default and
must be specifically enabled at compile time.
Many users are running OPIE with the key file on a shared NFS volume
in order to use OTP as a single-login system for a cluster of machines. OPIE
was NOT designed to be operated this way, though it does seem to work. If it
fails or if this proves insecure, this is not OPIE's fault. Note that, if you
do this, you probably want to share the OPIE lock files too.
Gripes
======
Is it too much to ask that certain OS vendors just do the right thing
and not "fix" what isn't broken? (Look at all the ifdefs in the OPIE code and
the answer is clear)
utmp and wtmp handling in OPIE has been a very, very sore subject.
Every vendor does things differently, and, of course, most of them swear they
are complying to some or other "standard." My (cmetz) conclusion is that the
only thing that is standard about utmp and wtmp handling is that it will be
nonstandard on any given system. I've tried a lot of things and I've wasted
*a lot* of time on trying to make utmp and wtmp handling work for everybody;
my conclusion is that it will never happen. While I am still interested in
hearing about fixes for utmp/wtmp on systems where they don't work, I'm not
likely to go out of my way to fix utmp/wtmp handling. If you want it fixed,
the best way to do it is to fix it yourself and contribute a patch. As long as
the patch is reasonable, it will be included in the next release. If you can't
wait, use the --disable-utmp option.
Credits
=======
First and foremost credit goes to Phil Karn, Neil M. Haller, and John
S. Walden of Bellcore for creating the S/Key Version 1 software distribution
and for making its source code freely available to the public. Without their
work, OPIE would not exist. Neil has also invested a good amount of his time
in the development of a standard for One-Time Passwords so that packages like
OPIE can interoperate.
The first NRL OPIE distribution included modifications made primarily
by Dan McDonald of the U.S. Naval Research Laboratory (NRL) during March 1994.
The 2nd NRL OPIE distribution, which has a number of improvements in areas
such as portability of software and ease of installation, is primarily the
work of Ran Atkinson and Craig Metz. Other NRL contributors include Brian
Adamson, Steve Batsell, Preston Mullen, Bao Phan, Jim Ramsey, and Georg Thomas.
Some of version 2.2 was developed at NRL and released as a work in
progress. Most of the release version was developed by Craig Metz (also of
NRL), others at The Inner Net, and contributors from the Internet community.
Versions beyond 2.2 were developed outside NRL, so don't blame them if they
don't work (But please credit them when it does. Without the NRL effort, there
wouldn't be an OPIE).
We would like to also thank everyone who helped us by by beta testing,
reporting bugs, suggesting improvements, and/or sending us patches. We
appreciate your contributions -- they have helped to make OPIE more of a
community effort. These contributors include:
Mowgli Assor
Lawrie Brown
Andrew Davis
Taso N. Devetzis
Carson Gaspar
Dennis Glatting
Ben Golding
Axel Grewe
"Hobbit"
Kojima Hajime
Darren Hosking
Matt Hucke
Kenji Kamizono
Charles Karney
Jeff Kletsky
Peter Koch
Martijn Koster
Osamu Kurati
Ayamura Kikuchi
Ronald van der Meer
Bret Musser
Hiroshi Nakano
Ikuo Nakagawa
Angelo Neri
C. R. Oldham
Ossama Othman
D. Jason Penney
John Perkins
Steve Price
Jim Simmons
Steve Simmons
Brad Smith
Werner Wiethege
Ken-ichi Yamasaki
Wietse Venema
OPIE development at NRL was sponsored by the Information Security
Program Office (PD 71E), U.S. Space and Naval Warfare Systems Command, Crystal
City, Virginia.
If you have problems with OPIE, please follow the instructions under
"If OPIE Doesn't Work." Under NO circumstances should you send trouble
reports directly to the authors or contributors. They WILL BE IGNORED.
Trademarks
==========
S/Key is a trademark of Bell Communications Research (Bellcore).
UNIX is a trademark of X/Open.
NRL is a trademark of the U. S. Naval Research Laboratory.
All other trademarks are trademarks of their respective owners.
The term "OPIE" is in the public domain and hence cannot be legally
trademarked by anyone. Please do not abuse it.
Copyrights
==========
%%% portions-copyright-cmetz-96
Portions of this software are Copyright 1996-1999 by Craig Metz, All Rights
Reserved. The Inner Net License Version 2 applies to these portions of
the software.
You should have received a copy of the license with this software. If
you didn't get a copy, you may request one from <license@inner.net>.
Portions of this software are Copyright 1995 by Randall Atkinson and Dan
McDonald, All Rights Reserved. All Rights under this copyright are assigned
to the U.S. Naval Research Laboratory (NRL). The NRL Copyright Notice and
License Agreement applies to this software.
Portions of this software are copyright 1980-1990 Regents of the
University of California, all rights reserved. The Berkeley Software
License Agreement specifies the terms and conditions for redistribution.
Portions of this software are copyright 1990 Bell Communications Research
(Bellcore), all rights reserved.