I’ve said it before, in all-caps and bold italics: DO NOT USE AN INSECURE FREAKIN’ CONSUMER OS LIKE WINDOWS ON VITAL CONTROL SYSTEMS!

At least a few of the higher-ups in the military seem to have learned to listen to what their technical people have likely been saying for years. With any luck it’ll filter out to the private sector too, and sooner rather than later.

EDIT, three hours later: in an ironic twist, it turns out that today is the tenth anniversary of the Bill Gates’ “trustworthy computing” memo. While the change in focus has been welcome, it hasn’t really hardened Windows, just elminiated the most blatant insecurities. Windows remains basically a single-user consumer OS, and still tries to be consumer-friendly at the expense of security. So long as Microsoft refuses to require people to learn anything in order to use Windows, it will never be secure.

This may sound strange, but that isn’t necessarily problem. A consumer OS should be easy to use, and shouldn’t require the user to learn any more than he could pick up by sitting at the keyboard and playing with it. At the same time, such an OS should never be used for anything vital — leave it to what it’s good for, which doesn’t include anything that requires security.

It worked, mostly. And over the last couple years, it has gotten better, but I’ve always gotten the impression that it was something of a red-headed stepchild for VMware. They’re very slow to release updates compatible with new versions of operating systems, and with a new version of Ubuntu out every six months now, that’s kind of important. Worse, they have never supported any Linux compositor, so I can only use more recent versions of Ubuntu in “classic” mode… I’d never even seen the new Unity interface in action, let alone used it.

VMware Fusion has other failings too, minor but very annoying:

• Several versions ago, it started telling the VM that the system is almost always plugged in, regardless of whether it is or not.
• When I’m away from my external mouse and have to use the trackpad, trackpad taps have always gotten “stuck” in the Linux VM. The system notices that you’ve tapped only after you move the mouse cursor a bit, which leads to some very annoying problems.
• VMware does not support IPv6 on virtual machines. Not too important right now, but it will be, and I’d like to start playing with it as soon as I can.
• The big one: hard drive accesses are REALLY slow, and I still get a lot of “IOwait” even after making this change. This might not be entirely VMware’s fault, but I have no way of proving or disproving that right now.

My current 20GB virtual hard drive is also getting full, and I wanted to rebuild the virtual machine anyway, so it was time to look at alternatives.

The first thing I noticed yesterday, when I started, was that VMware Fusion v4 had just been released the day before. I eagerly perused its features and changes, only to find that it had only two things that I wanted: full OS X “Lion” compatibility and “Time Machine-compatible snapshots.” Whoopti-do. Sorry, that is not worth upgrading for. Worse, it seemed rushed, because there were lots of reports of problems with it.

Then I checked out Parallels. From various forums I’d already heard that they updated it for new OS versions very quickly, and that it fully supported Ubuntu’s compositor and newer Unity interface, so it started out with an instant bonus. I discovered that they’d just recently released a new version too, v7 (likely why VMware Fusion v4 was rushed into release too early), and there’s a special offer for “cross-grading” from VMware Fusion for only $30 right now. I also discovered that their virtual machines fully support IPv6, and that they provide much better support on their forums than VMware does. Even better, they offer a free two-week trial, which I jumped at.

I haven’t fully finished setting up my first virtual machine yet, but so far I’ve proven that it does fully support Ubuntu’s new Unity interface. (I’m not sure I like the new interface, but it does apparently eliminate the memory leak problem in XOrg that I’d been battling for a while now, by eliminating X itself. ) So far it also seems much faster, I’ve seen very few “IOwait” indications in the System Load Indicator applet.

We’ll see if it remains that way after I’ve used it for a couple weeks. If so, I think we’ve got a winner.

It was inevitable that I would upgrade my current system (a mid-2009 model MacBook Pro) eventually, but I wanted to put it off for a while. I’ve talked before about my love/hate relationship with upgrades; they’re often a big improvement, but at the cost of a lot of unnecessary grief, and I have work to do. But Lion included some features that sounded very nice, and one that I found irresistable: new versions of FileVault (the disk-encryption software) and Time Machine (the automatic backup program) that work properly together, rather than the Snow Leopard versions where Time Machine was crippled if you used FileVault as well.

As mentioned before, I do very little on the Mac OS side of things; the only programs I generally run on it are VMware Fusion, Skype (because it doesn’t work well in virtual machines), and a to-do list program called Things. Most of my work is done in Linux or Windows virtual machines. That turned out to be a major benefit for this upgrade, because there was far less to go wrong; so long as those three programs continued to operate, my work wouldn’t be disrupted, and reports indicated that the latest versions shouldn’t have any problems that would affect me.

Despite that, I waited until Friday (yesterday), so I’d have three days to recover if anything went seriously wrong, and began.

(Of course, Friday morning a friend of mine — who I won’t name, but who comments on this blog regularly — helpfully sent me a link to a bunch of complaints that Lion caused major instability in many systems. And I’d recently seen the same complaint on The Register’s review of it, though that one was fixed by a clean reinstall. The clean reinstall route was my plan B if the system couldn’t manage an upgrade; plan C was to go back to Snow Leopard via my Time Machine backups, but I hoped I wouldn’t have to use it.)

I’d read somewhere that in order to get the FileVault benefits of Lion, you had to turn off the old FileVault and then turn on the new one, so I decided to turn off the old one before I upgraded. That required a great deal of free disk space, so I had to move about 150GB worth of virtual machines to my NAS drive (which took several hours), then decrypt it (which took several more, and I couldn’t use the machine while it worked).

That done, I followed the steps here, including uninstalling VMware Fusion (because VMware recommended reinstalling it after the upgrade anyway, for stability), fired up the Mac App Store (for the first and possibly only time), bought it, and downloaded it. The download took quite a while; it’s nearly 4GB of data, which is twice what we normally use in a full month, but we have a fairly fast connection and plenty of bandwidth allowance that we never use, so I wasn’t worried. I didn’t install it immediately; instead I made a backup of the installer and created a clean-install boot disk on an SD card, as described here. Then disconnected my Time Machine drive (just in case) and let the installer rip.

It took about 40 minutes. I sat through the last three minutes of it, then crossed my fingers while it rebooted and I logged in…

It came up without any problem.

Turning the new FileVault encryption on was a lot easier than turning the old one off — it worked in the background, and apparently encrypted everything in place, so it didn’t need all the room that the old one required. I’ve temporarily lost my third monitor, because the USB video adapter I use for it doesn’t have a stable Lion driver yet, but I knew about that ahead of time. And every hour now, if I’m in the office, Time Machine automatically backs up all the changes that I’ve made since the last backup.

Unless something goes wrong in the next week or so, I’m calling it a success.

Well, I was almost right.

Unity requires 3D graphics support. This MacBook Pro has plenty of that, but VMware Fusion 3 doesn’t support it for Linux guests. When I tried to log in, I got a polite message saying that it couldn’t show the Unity interface, and was reverting to GNOME. Oh well. I’m assuming VMware is working on that problem; until they fix it, I’m stuck with a more stable GNOME system. It could be worse.

Also as usual, this upgrade broke some things, though not too badly:

• GnuCash, the money-manager that I use, would only show a blank window for a couple of my accounts unless I told it to open “New Window with Page”… irritating, but not a show-stopper.
• CodeLite, the programming IDE I’m using, had some very weird scrolling behavior related to the new scroll-bars being a lot smaller than it was expecting. A search immediately turned up a bug-report. It apparently happens with all programs written using the wxWidgets library; presumably the wxWidgets people are already working on a fix. In the mean time, there’s a work-around built into the system to tell it to use the old scroll-bars instead, which works perfectly.
• A couple modules in the VMware Tools package wouldn’t compile, apparently due to API changes. I tried the open-vm-tools package, but it does not replace VMware Tools, no matter what it claims — it doesn’t seem to provide cross-machine copy/paste support or automatic guest screen resizing when switching the host’s resolution (as I must when I switch to the internal monitor to go portable). Again, irritating.

Even better, when I got up this morning there were updates waiting for me, which automatically corrected the problems with both GnuCash and CodeLite.

There’s also a new version of the VMware Tools system that fixes the problems I had, but the only way I found to get it was to extract it manually. A royal pain in the arse, but I did so. On the plus side, it looks like it works perfectly, so far.

I don’t recommend this sort of thing for the general user, or even in many cases the power user. I don’t even recommend it for myself. But if you’re dying to upgrade to Natty, it shouldn’t be too rough.

My Linux virtual machine was getting some very annoying delays at times that it shouldn’t have. Compiling a fairly small project resulted in a ridiculous amount of “iowait” time. Even switching folders in my current e-mail program could take twenty seconds or more. For months, I’d thought it was just a slow hard drive on the host machine, but I couldn’t figure out why — the data being requested by the programs when the problem happened wasn’t ridiculously large, and the specs of the drive were reasonably fast. I’d looked into upgrading to a solid-state drive, but I couldn’t justify spending the money for the size I needed to replace this one, nearly half the original cost of the definitely-not-cheap machine.

Yesterday morning I’d finally had enough of it. I had to know why it was happening, even if I couldn’t do anything about it immediately. What else could be causing it? Perhaps some interaction between the disk format and VMware Fusion, that could be remedied with a different disk format?

A web search didn’t turn up anything about that, but did produce some interesting information on “iowait” problems with VMware Fusion, including a setting in the VMX file to try (“useNamedFile=false”, with an intriguing explanation here). I tried it, and it worked!

Now comes the exasperating part.

I knew I had made another entry about VMware Fusion settings that helped the speed, so I wanted to add this to it. It didn’t take me long to find it… or to find that this setting is already listed there. I’d had to rebuild my virtual machine since then (wiping out those changes, which I didn’t reapply because the original problem didn’t seem to come up), and I didn’t think to try the settings there because the problem didn’t exhibit any of the other symptoms from that entry.

Oh well, live and learn.

So I’ve started development on a new series of programs. The first is a “better mousetrap” that not only scratches a couple long-time itches of mine, but should be of interest to a lot of other people as well. I’ll refer to that as “Project M” on this blog from here on, and I won’t be giving away many details about it until it’s released.

One of the things that Project M is going to need is good date handling. It’s going to be used with lots of other programs and in lots of different countries worldwide, and it has to understand the dates and times for each country properly. Much of the heavy lifting for date/time handling is taken care of by the Boost.Date_Time library, but one thing it doesn’t do is tell you what time zone the current system is set to.

I was flabbergasted by this omission. Knowing the time zone that you’re in is an absolute requirement for converting times between different zones. But when I dug into it, I saw why: because it’s a hard problem. So the library author punts: you, the developer, have to tell the library what time zone it’s in, and then it will take care of everything else for you. Not necessarily all that difficult, but doing it in a cross-platform way required quite a bit of research, thought, and experimentation (which is why there was no blog entry yesterday, I was too busy working on it).

I can’t really fathom why Linux developers haven’t settled on a good standard for this. The OS knows what time zone the system is in, but it’s all but impossible to get it to divulge that information portably. Different distributions all have different ways of doing it, and most of them are ridiculously indirect.

What I ended up doing was writing code for the system that Ubuntu uses (hopefully shared by other distributions as well), and writing a fall-back method for everything else. The fall-back code uses standard C library calls to fake it. It’ll work fine for any country that ignores the insanity that is Daylight Savings Time (DST), and in any country that shares the politically motivated and apparently randomly-changing dates that DST begins and ends in the US, but will only somewhat work in other places that use DST. Without knowing the actual location of the system, that’s the best I can manage.

Hopefully Linux distributions will decide on a standard method of discovering the system time zone. Until then, there’s inevitably going to be a lot of pain and confusion when it comes to dates and times.

I’d run into a minor compiler bug in my Ubuntu version of GCC. Since a new version of the OS had just been released the day before, promising even faster bootups and other improvements, I thought I’d upgrade the entire distribution instead of trying to jury-rig an updated compiler into the existing one. Unfortunately, for the first time in a long time, the upgrade (as our British cousins put it) went all pear-shaped.

The OS itself upgraded with no problems, but since it installed a new kernel, I lost my VMware Fusion shared folders. No problem, that’s expected, just reinstall VMware Tools to fix it. That’s where the trouble started.

VMware Tools wouldn’t reinstall properly. Two of the modules wouldn’t compile at all (including the one needed for shared folders), and it claimed that there were no drivers for the X-windows subsystem. I tried rebooting it, but that made things a lot worse because the screen came up as a completely unreadable jumble of horizontal lines as soon as I logged in (the log-in screen itself was fine).

I finally found a way around that (by logging into the recovery terminal and playing with vaguely useful-sounding commands that I no longer recall until I’d set the default log-in resolution to the only one that worked), but it was obvious that something had gone horribly wrong. Several VMware Tools reinstallation attempts gave the same problems, though none of them resulted in the video problems again. The open-vm-tools project (an open-source version of VMware Tools) gave the same results as well.

For a very popular combination like Ubuntu on VMware Fusion, I would expect that such a problem would be widely reported and discussed (as such things always have been in the past), but Google failed to turn up anything remotely similar, no matter how I phrased the search or how far into the results I delved. In fact, I’d found several places on the ‘net that implied that there weren’t any problems.

I figured I just needed an updated version of VMware Tools, since I’ve seen things similar to this before. But I couldn’t find one on the VMware site, or anywhere else trustworthy. One of my searches turned up a page saying explicitly that VMware Player (the free version of the virtualization system) would automatically download an updated version of them when you created a virtual machine with the new Ubuntu, so I spent more than an hour downloading it and the new Ubuntu install disk and installing both onto my wife’s Windows machine so I could extract them. The installation into the virtual machine crashed for reasons unknown, but I was able to extract the new VMware Tools ISO file anyway, and extract the files I needed. To my dismay, they turned out to have the exact same problems that I was already encountering.

To make a long story a little shorter, I eventually created a new virtual machine and installed the new version of Ubuntu from scratch, which worked with no problem. I encountered what seems to be a new bug in the AutoFS system (which cost me several hours of attempted fixes before I gave up and used a less-convenient work-around), grabbed about four hours of sleep, then copied my data to the new virtual machine. I’m about 95% up and running now, and should be up to 99% before the end of the day.

It’s a major consolation that, if I’d still been running Windows instead, it would have taken twice as long to recover from a major problem like this. I had to do it often enough that I had the timing down to a science.

What to do about it? Well, the first thing would be DO NOT USE AN INSECURE FREAKIN’ CONSUMER OS LIKE WINDOWS ON VITAL CONTROL SYSTEMS!

A report released earlier this week says that 99.4% of malware is written for Windows. Why? It’s not just because Windows is the most common OS today, running on an estimated 80% of consumer systems; there are many more Linux servers running on the Internet than Windows ones, yet malware that targets servers (or tries to automatically run on any system) is almost exclusively aimed at Windows systems.

Windows was designed as a single-user system. Security is hard to do and annoys people, so programmers won’t design it into single-user systems. Such systems, the thought runs, don’t need to protect themselves from their user, because if the user does something stupid, he’s the only one that suffers. They don’t need to protect users from each other either, because there’s only one user to deal with.

Malware that targets people is also inevitably targeted at Windows, because Windows users can easily be tricked into letting such programs run. They’re trained to let programs run, not to think about whether they should let them run.

The problem is that when you try to bolt on security as an afterthought, it leaks like a sieve. There’s no defense-in-depth, all a malicious person has to do is get through a single layer of security and he owns the OS, and the machine it runs.

I doubt much malware will ever be written for Linux, and certainly not successful malware. Linux is based on a multi-user OS that was designed from the ground up to protect the system from its users, and protect the users from each other. On a Linux machine, only a system administrator has the power to open the system to that kind of subversion; even if a user does manage to install malware, he can only damage his own files with it, he can’t infect the system itself or the files of other users. The model had survived decades of bored college students before Windows was even a glint in Bill Gates’ eye.

The upshot: there is a secure OS, easily and freely available today, that will prevent disasters if used on industrial control systems. It is called Linux. Windows is not an equally-secure substitute, no matter what Microsoft claims, and it never will be.

Nothing’s going to change because of this post, except that I feel a little better for saying something. Businesses are still going to choose Windows, because a company like Microsoft, and an OS like Windows, is “too big to fail” (sound familiar?). Linux will continue to get the short end of the stick, except in places where technical people are given free rein and an absolute minimum budget (like server rooms). And disasters like Spanair flight 5022 will continue to happen. But some day people will start to put two and two together, and Windows will slowly be banished from control systems everywhere. Let’s hope it starts happening soon.

The description starts out:

Microsoft has a majority market share in the new desktop PC marketplace. This is a bug, which Ubuntu is designed to fix. [...]

I’ve had reason to look at several Linux-style bug reports over the past few years, and this one is a very well-done parody.