Request: after building the programs, please run the `tcpdchk' wrapper configuration checker. See the `tcpdchk.8' manual page (`nroff -man' format) for instructions. `tcpdchk' automatically identifies the most common configuration problems, and will save you and me a lot of time. Changes per release 7.2 (Jan 1995) ================================== - Added a note to the README and manpages on using the IDENT service to detect sequence number spoofing and other host impersonation attacks. - Portability: ConvexOS puts RPC version numbers before the daemon path name (Jukka Ukkonen). - Portability: the AIX compiler disliked the strchr() declaration in socket.c. I should have removed it when I included . - Backwards compatibility: some people relied on the old leading dot or trailing dot magic in daemon process names. - Backwards compatibility: hostname lookup remains enabled when -DPARANOID is turned off. In order to disable hostname lookups you must turn off -DALWAYS_HOSTNAME. - Eliminated false complaints from the tcpdmatch/tcpdchk configuration checking programs about process names not in inetd.conf or about KNOWN username patterns. Changes per release 7.1 (Jan 1995) ================================== - Portability: HP-UX permits you to break inetd.conf entries with backslash-newline. - Portability: EP/IX has no putenv() and some inetd.conf entries are spread out over two lines. - Portability: SCO with NIS support has no *netgrent() routines. Changes per release 7.0 (Jan 1995) ================================== - Added a last-minute workaround for a Solaris 2.4 gethostbyname() foulup with multi-homed hosts in DNS through NIS mode. - Added a last-minute defense against TLI weirdness: address lookups apparently succeed but the result netbuf is empty (ticlts transport). - Dropped several new solutions that were in need of a problem. Beta testers may recognize what new features were kicked out during the last weeks before release 7.0 came out. Such is life. - Got rid of out the environment replacement routines, at least for most architectures. One should not have to replace working system software when all that is needed is a 4.4BSD setenv() emulator. - By popular request I have added an option to send banner messages to clients. There is a Banners.Makefile that gives some aid for sites that are going to use this feature. John C. Wingenbach did some pioneering work here. I used to think that banners are frivolous. Now that I had a personal need for them I know that banners can be useful. - At last: an extensible functional interface to the pattern matching engine. request_init() and request_set() accept a variable-length name-value argument list. The result can be passed to hosts_access(). - When PARANOID mode is disabled (compile time), the wrapper does no hostname lookup or hostname double checks unless required by %letter expansions, or by access control rules that match host names. This is useful for sites that don't care about internet hostnames anyway. Inspired by the authors of the firewalls and internet security book. - When PARANOID mode is disabled (compile time), hosts with a name/name or name/address conflict can be matched with the PARANOID host wildcard pattern, so that you can take some intelligent action instead of just dropping clients. Like showing a banner that explains the problem. - New percent escapes: %A expands to the server address; %H expands to the corresponding hostname (or address if no name is available); %n and %N expand to the client and server hostname (or "unknown"); %s expands to everything we know about the server endpoint (the opposite of the %c sequence for client information). - Symmetry: server and client host information is now treated on equal footing, so that we can reuse a lot of code. - Lazy evaluation of host names, host addresses, usernames, and so on, to avoid doing unnecessary work. - Dropping #ifdefs for some archaic systems made the code simpler. - Dropping the FAIL pattern made the pattern matcher much simpler. Run the "tcpdchk" program to scan your access control files for any uses of this obscure language feature. - Moving host-specific pattern matching from string_match() to the host_match() routine made the code more accurate. Run the "tcpdchk" program to scan your access control files for any dependencies on undocumented or obscure language features that are gone. - daemon@host patterns trigger on clients that connect to a specific internet address. This can be useful for service providers that offer multiple ftp or www archives on different internet addresses, all belonging to one and the same host (www.foo.com, ftp.bar.com, you get the idea). Inspired by a discussion with Rop Gonggrijp, Cor Bosman, and Casper Dik, and earlier discussions with Adrian van Bloois. - The new "tcpdchk" program critcizes all your access control rules and inetd.conf entries. Great for spotting obscure bugs in my own hosts.xxx files. This program also detects hosts with name/address conflicts and with other DNS-related problems. See the "tcpdchk.8" manual page. - The "tcpdmatch" program replaces the poor old "try" command. The new program looks in your inetd.conf file and therefore produces much more accurate predictions. In addition, it detects hosts with name/address conflicts and with other DNS-related problems. See the "tcpdmatch.8" manual page. The inetd.conf lookup was suggested by Everett F Batey. - In the access control tables, the `=' between option name and value is no longer required. - Added 60-second timeout to the safe_finger command, to cover another potential problem. Suggested by Peter Wemm. - Andrew Maffei provided code that works with WIN-TCP on NCR System V.4 UNIX. It reportedly works with versions 02.02.01 and 02.03.00. The code pops off all streams modules above the device driver, pushes the timod module to get at the peer address, and then restores the streams stack to the initial state. Changes per release 6.3 (Mar 1994) ================================== - Keepalives option, to get rid of stuck daemons when people turn off their PC while still connected. Files: options.c, hosts_options.5. - Nice option, to calm down network daemons that take away too much CPU time. Files: options.c, hosts_options.5. - Ultrix perversion: the environ global pointer may be null. The environment replacement routines now check for this. File: environ.c. - Fixed a few places that still assumed the socket is on standard input. Fixed some error messages that did not provide access control file name and line number. File: options.c. - Just when I was going to release 6.2 I received code for Dynix/PTX. That code is specific to PTX 2.x, so I'll keep around my generic PTX code just in case. The difference is in the handling of UDP services. Files: tli_sequent.[hc]. Changes per release 6.2 (Feb 1994) ================================== - Resurrected my year-old code to reduce DNS load by appending a dot to the gethostbyname() argument. This feature is still experimental and it may go away if it causes more problems than it solves. File: socket.c. - Auxiliary code for the Pyramid, BSD universe. Karl Vogel figured out what was missing: yp_get_default_domain() and vfprintf(). Files: workarounds.c, vfprintf.c. - Improved support for Dynix/PTX. The wrapper should now be able to deal with all TLI over IP services. File: ptx.c. - The try command now uses the hostname that gethostbyaddr() would return, instead of the hostname returned by gethostbyname(). This can be significant on systems with NIS that have short host names in the hosts map. For example, gethostbyname("wzv.win.tue.nl") returns "wzv.win.tue.nl"; gethostbyaddr(131.155.210.17) returns "wzv", and that is what we should test with. File: try.c. Changes per release 6.1 (Dec 1993) ================================== - Re-implemented all environment access routines. Most systems have putenv() but no setenv(), some systems have setenv() but no putenv(), and there are even systems that have neither setenv() nor putenv(). The benefit of all this is that more systems can now be treated in the same way. File: environ.c. - Workaround for a weird problem with DG/UX when the wrapper is run as nobody (i.e. fingerd). For some reason the ioctl(fd, I_FIND, "sockmod") call fails even with socket-based applications. The "fix" is to always assume sockets when the ioctl(fd, I_FIND, "timod") call fails. File: fromhost.c. Thanks to Paul de Vries (vries@dutentb.et.tudelft.nl) for helping me to figure out this one. - Implemented a workaround for Dynix/PTX and other systems with TLI that lack some essential support routines. Thanks to Bugs Brouillard (brouill@hsuseq.humboldt.edu) for the hospitality to try things out. The trick is to temporarily switch to the socket API to identify the client, and to switch back to TLI when done. It still does not work right for basic network services such as telnet. File: fromhost.c. - Easy-to-build procedures for SCO UNIX, ConvexOS with UltraNet, EP/IX, Dynix 3.2, Dynix/PTX. File: Makefile. - Variable rfc931 timeout. Files: rfc931.c, options.c, log_tcp.h, try.c. - Further simplification of the rfc931 code. File: rfc931.c. - The fromhost() interface stinks: I cannot change that, but at least the from_sock() and from_tli() functions now accept a file descriptor argument. - Fixed a buglet: fromhost() would pass a garbage file descriptor to the isastream() call. - On some systems the finger client program lives in /usr/bsd. File: safe_finger.c. Changes per release 6.0 (Sept 1993) =================================== - Easy build procedures for common platforms (sun, ultrix, aix, hpux and others). - TLI support, System V.4 style (Solaris, DG/UX). - Username lookup integrated with the access control language. Selective username lookups are now the default (was: no username lookups). - A safer finger command for booby traps. This one solves a host of possible problems with automatic reverse fingers. Thanks, Borja Marcos (borjam@we.lc.ehu.es) for some inspiring discussions. - KNOWN pattern that matches hosts whose name and address are known. - Cleanup of diagnostics. Errors in access-control files are now shown with file name and line number. - With AIX 3.2, hostnames longer than 32 would be truncated. This caused hostname verification failures, so that service would be refused when paranoid mode was enabled. Found by: Adrian van Bloois (A.vanBloois@info.nic.surfnet.nl). - With some IRIX versions, remote username lookups failed because the fgets() library function does not handle partial read()s from sockets. Found by: Daniel O'Callaghan (danny@austin.unimelb.edu.au). - Added a DISCLAIMER document to help you satisfy legal departments. The extension language module has undergone major revisions and extensions. Thanks, John P. Rouillard (rouilj@ra.cs.umb.edu) for discussions, experiments, and for being a good guinea pig. The extensions are documented in hosts_options.5, and are enabled by editing the Makefile STYLE macro definition. - (Extension language) The ":" separator may now occur within options as long as it is protected with a backslash. A warning is issued when a rule ends on ":". - (Extension language) Better verification mode. When the `try' command is run, each option function now explains what it would do. - (Extension language) New "allow" and "deny" keywords so you can now have all rules within a single file. See "nroff -man hosts_options.5" for examples. - (Extension language) "linger" keyword to set the socket linger time (SO_LINGER). From: Marc Boucher . - (Extension language) "severity" keyword to turn the logging noise up or down. Many sites wanted a means to shut up the program; other sites wanted to emphasize specific events. Adapted from code contributed by Dave Mitchell . Changes per release 5.1 (Mar 1993) ================================== - The additional protection against source-routing attacks from hosts that pretend to have someone elses network address has become optional because it causes kernel panics with SunOS <= 4.1.3. Changes per release 5.0 (Mar 1993) ================================== - Additional protection against source-routing attacks from hosts that pretend to have someone elses network address. For example, the address of a trusted host within your own network. - The access control language has been extended with a simple but powerful operator that greatly simplifies the design of rule sets (ALL: .foo.edu EXCEPT dialup.foo.edu). Blank lines are permitted, and long lines can be continued with backslash-newline. - All configurable stuff, including path names, has been moved into the Makefile so that you no longer have to hack source code to just configure the programs. - Ported to Solaris 2. TLI-based applications not yet supported. Several workarounds for System V bugs. - A small loophole in the netgroup lookup code was closed, and the remote username lookup code was made more portable. - Still more documentation. The README file now provides tutorial sections with introductions to client, server, inetd and syslogd. Changes per release 4.3 (Aug 1992) ================================== - Some sites reported that connections would be rejected because localhost != localhost.domain. The host name checking code now special-cases localhost (problem reported by several sites). - The programs now report an error if an existing access control file cannot be opened (e.g. due to lack of privileges). Until now, the programs would just pretend that the access control file does not exist (reported by Darren Reed, avalon@coombs.anu.edu.au). - The timeout period for remote userid lookups was upped to 30 seconds, in order to cope with slow hosts or networks. If this is too long for you, adjust the TIMEOUT definition in file rfc931.c (problem reported by several sites). - On hosts with more than one IP network interface, remote userid lookups could use the IP address of the "wrong" local interface. The problem and its solution were discussed on the rfc931-users mailing list. Scott Schwartz (schwartz@cs.psu.edu) folded the fix into the rfc931.c module. - The result of % expansion (in shell commands) is now checked for stuff that may confuse the shell; it is replaced by underscores (problem reported by Icarus Sparry, I.Sparry@gdr.bath.ac.uk). - A portability problem was fixed that caused compile-time problems on a CRAY (problem reported by Michael Barnett, mikeb@rmit.edu.au). Changes per release 4.0 (Jun 1992) ================================== 1 - network daemons no longer have to live within a common directory 2 - the access control code now uses both the host address and name 3 - an access control pattern that supports netmasks 4 - additional protection against forged host names 5 - a pattern that matches hosts whose name or address lookup fails 6 - an operator that prevents hosts or services from being matched 7 - optional remote username lookup with the RFC 931 protocol 8 - an optional umask to prevent the creation of world-writable files 9 - hooks for access control language extensions 10 - last but not least, thoroughly revised documentation. Changes per release 3.0 (Oct 1991) ================================== Enhancements over the previous release are: support for datagram (UDP and RPC) services, and execution of shell commands when a (remote host, requested service) pair matches a pattern in the access control tables. Changes per release 2.0 (May 1991) ================================== Enhancements over the previous release are: protection against rlogin and rsh attacks through compromised domain name servers, optional netgroup support for systems with NIS (formerly YP), and an extension of the wild card patterns supported by the access control files. Release 1.0 (Jan 1991)