Chronology of the project

Fri Oct 31 19:24:57 PDT 2014, Version 1.1:

Client ported to Python 3.x.
Client passes pylint.
Build error in configure/configure.in fixed.

Sun Feb 4 14:44:39 PST 2007, Version 0.9996:

Ignore SIGTERM, because:
- An orderly shutdown sends this signal, and we have nothing to clean up anyway
- Sometimes a system will go into limbo during shutdown, and we want to be around if that happens.

Sun Feb 4 16:46:25 GMT 2007, Version 0.9995:

Conditionalized mlockall() use on _POSIX_MEMLOCK

Added DragonFly BSD support, despite DragonFly's lack of a true mlockall()

Wed Oct 11 10:47:25 PDT 2006
Version 0.9994:

Added slightly more entropy when /dev/random, /dev/urandom and prngd are all unavailable

Mon May 1 18:05:32 PDT 2006

Mon Oct 10 18:07:36 PDT 2005
Version 0.9993

Modified gen-pas to use the random module, instead of the deprecated whrandom module.

Tue May 31 13:40:55 PDT 2005

./configure now will prefer, from most favored to least favored, the following C compilers (if $CC is not set). We're favoring tcc -b because it includes code to check buffer overruns.
1. tcc -b
2. gcc
3. cc

May 2, 2005

Rebooted an RHEL 3 system with FBR that had so much NFS trouble it wouldn't allow ssh logins, and existing ssh sessions were useless.

Mon Apr 11 12:03:30 PDT 2005

Solaris 10: fallback-reboot builds against the version of openssl that is included under /usr/sfw. No SMF support yet, but the rc?.d stuff seems to work fine for now.

Fri Apr 1 15:33:15 PST 2005

The fdisk program in the Darwin CD image doesn't appear to include the (same name for the) type that Darwin wants to see, before it'll install onto the "hard disk"....

Tue Mar 29 15:48:28 PST 2005

"fdisk -a bootufs /dev/rdisk0" under pearpc, to get a reasonable partition table. Had to request a shell...

Tue Mar 29 13:36:57 PST 2005

Trying again with pearpc, and a powerpc .iso. pearpc appears to want its disk file created with a length that's a multiple of 516096, so I'm creating it with:
- dd if=/dev/zero bs=516096 count=$(expr 1024 \* 8 \* 2) | reblock -s 0.5 -e 8257536 516096 300 > Disk.img

Tue Mar 29 09:00:33 PST 2005

Darwin didn't seem to want to boot under QEMU, but OpenDarwin seems to be doing better... But after what seemed like a successful install, I'm getting that dang grey screen with the apple symbol upon reboot.

Mon Mar 28 20:21:48 PST 2005

I got the silly idea that it might be fun to port to Darwin... But QEMU won't boot Darwin for x86... I'm now looking into PearPC and OpenDarwin.

Sun Mar 27 17:45:25 PST 2005

Some autoconf changes to better support compilers other than gcc. Tested with tcc on Fedora Core 3 and Solaris' SUNWspro compilers. Releasing 0.9991.

Fri Mar 25 22:13:13 PST 2005

It's working on FreeBSD 5.4 Beta 1. :) Releasing 0.999.

Fri Mar 25 18:10:00 PST 2005

I've got fallback-reboot.c ./configure'ing and compiling on FreeBSD, and I've taken a stab at getting install-rc-script to contend with *BSD rc systems. Then I discover that FreeBSD 5.4-BETA1 doesn't come with python. :)

Thu Mar 24 21:06:22 PST 2005

Starting over with 5.4-BETA1-i386-disc1.iso. :-S I'm using the same "hard disk" that I used for DragonFly....

Thu Mar 24 20:48:51 PST 2005

I played around with DragonFly some more afterall. Specifically, I got it running under QEMU, and fiddled around a bit. However, I'm now seeing that the latest version of vm_mmap.c has nothing more than stub functions for mlockall() and munlockall(), and comp.unix.programmer doesn't seem to know a means of simulating mlockall() given a functional mlock(). So I really am going to have to move on to FreeBSD. On the bright side for DragonFly, their compiler is predefining __FreeBSD__ to 4, so it shouldn't be too hard to come up with code that'll build and work on both FreeBSD and DragonFly, once DragonFly has a real mlockall() implementation.

Tue Mar 15 16:47:06 PST 2005

Today, I was fiddling with a Sun, starting to do some SVM/RAID 5 experiments. At a moment when the Sun Ultra 1's SCSI subsystem was useless (no commands would run at the shell prompt at all), fallback-reboot still cheerfully gave a prompt....

Sat Mar 12 11:20:14 PST 2005

I put some time into porting fallback-reboot to DragonFly BSD. They sounded like the best match, since:
1. They bill themselves as a more progressive BSD
2. I like Dragonflies a lot
...but unfortunately:
1. The live CD didn't have a ksh or bash, and their /bin/sh doesn't appear to support shell functions
2. As I was hunting around for the right includes and preprocessor symbols, I got popped into a kernel debugger. "continue" just gave an error, and "next" got stuck in an endless loop.
...so, I may try FreeBSD if I find some time for a *BSD port again.

Mon Mar 7 10:50:39 PST 2005

Releasing 0.998. In this version:
1. DEBUG cpp symbol renamed to AUTH_NO_REBOOT
2. DEBUG2 cpp symbol renamed to VERBOSE
3. AIX uses plock() instead of mlockall() now
4. install-bufsock knows how to set the $PATH more broadly, to be more likely to pick up a python executable
5. Note: this release not announced on freshmeat.

Sun Mar 6 21:24:02 PST 2005

Releasing 0.997. This version supports AIX 5.1, understands prngd, has a bit better installation system, and allows you to easily change the ports on which fallback-reboot and prngd are on via configure options.

Thu Mar 3 12:16:54 PST 2005

Released 0.996. Includes misc cleanups, especially some safeguards for bad /.fallback-reboot-passwd files. Also improved autoconf portability.

Mon Feb 28 20:32:01 PST 2005

Added a FAQ list.

Wed Feb 23 11:26:25 PST 2005

I tested building the daemon with SUNWspro cc (solaris native cc) today. It built fine, but I had to get rid of the -ansi -pedantic -Wall in the Makefile.

Mon Feb 21 18:00:58 PST 2005

Releasing version 0.995. In this version:
1. The daemon only supports RIPEMD-160, not SHA-1 anymore, since SHA-1 has been broken. This means that old fallback-reboot-client programs (0.99 and before) are incompatible with the new daemon.
2. The new client is compatible with both the new daemon (RIPEMD-160) and the old daemons (SHA-1, Cleartext).
3. The client now attempts to use RIPEMD-160 via the optional, and often not preinstalled, M2Crypto python module. This module comes with Fedora Core 3, but not Redhat Enterprise 3 or Solaris 9. If fallback-reboot-client cannot find this module, it will fall back to a bidirectional pipe to the openssl program, which must be on your $PATH.
4. A minor DOS attack is fixed. The attack was against the daemon, not the host the daemon was running on.
5. This release has been tested on an Ubuntu 4.10 system, using a pipe to openssl, not with M2Crypto.

Thu Feb 3 17:37:16 PST 2005

A serious error in the SIGUSR1 stuff snuck through. I believe it's fixed now - at least on FC3. Releasing 0.99.

Thu Feb 3 10:14:52 PST 2005

Releasing 0.98. Main changes are that the programs go in $prefix/sbin, not $prefix/bin, and the daemon will reread /.fallback-reboot-passwd if you send it a SIGUSR1. Also "make install" does not create /.fallback-reboot-passwd anymore; that's up to S11fallback-reboot now, and even that's only done if the file doesn't already exist.

Wed Feb 2 21:44:21 PST 2005

I have a strong hypothesis as to why bill didn't reboot tonight. I ran my fallback-reboot setup script more than once. The second invocation would have rewritten /.fallback-reboot-passwd, but the daemon memorizes this file's content upon invocation. Also, the second invocation of the daemon would have errored out due to a busy port.

Wed Feb 2 21:38:51 PST 2005

Release 0.96 tonight. Please see the Changelog for changes.
- S11fallback-reboot respects --prefix now, and is generated from S11fallback-reboot.in
- S11fallback-reboot generates /.fallback-reboot-passwd if it does not yet exist
- fallback-rewboot daemon uses SO_REUSEADDR
- "make install" removes the previous binary to avoid Text File Busy errors
- "make clean" also removes S11fallback-reboot

Wed Feb 2 19:01:17 PST 2005

A Redhat Enterprise 3 server got stuck tonight, named bill. We've been collecting fallback-reboot-passwd's using gpg and srsh. I attempted to reboot bill by decrypting bill's fallback-reboot-passwd and entering it into fallback-reboot-client bill, but I consistently got a "Bad password, disconnecting.." error, and no reboot. I plan to check if bill's fallback-reboot-passwd was somehow changed once it's back up.

Thu Jan 27 15:46:16 PST 2005

We've largely rolled out fallback-reboot in UCI/NACS/DCS on the systems to which it is applicable (Recent Solaris, Recent Linux). I currently have a machine that is getting stuck a lot (runs Fedora Core 3), and I'm now finding that although I cannot ssh into the machine, I can fallback-reboot it. It's beginning to look like this FC3 machine is getting bit by the pci=routeirq thing. More later.

Thu Jan 20 15:57:34 PST 2005

I've split out the password into a separate file in the root directory, among other things. Please see the ChangeLog for specifics. Also we've set a rollout of the program in motion in my group, so hopefully we'll have some good data on just how effective this is, within a month or two.

Sun Jan 16, 2005:

I've added cryptography, to strongly deter replay attacks. You can still use it without crypto though - in fact, you'll only get crypto if autoconf can find openssl. The crypto used is your basic challenge-response system based on SHA1.