Planet Bozo

August 07, 2020

Worse Than FailureError'd: All Natural Errors

"I'm glad the asdf is vegan. I'm really thinking of going for the asasdfsadf, though. With a name like that, you know it's got to be 2 1/2 times as good for you," writes VJ.

 

Phil G. wrote, "Get games twice as fast with Epic's new multidimensional downloads!"

 

"But...it DOES!" Zed writes.

 

John M. wrote, "I appreciate the helpful suggestion, but I think I'll take a pass."

 

"java.lang.IllegalStateException...must be one of those edgy indie games! I just hope it's not actually illegal," writes Matthijs .

 

"For added flavor, I received this reminder two hours after I'd completed my checkout and purchased that very same item_name," Aaron K. writes.

 

[Advertisement] Keep the plebs out of prod. Restrict NuGet feed privileges with ProGet. Learn more.

XKCDMathematical Symbol Fight

August 06, 2020

Worse Than FailureCodeSOD: A Slow Moving Stream

We’ve talked about Java’s streams in the past. It’s hardly a “new” feature at this point, but its blend of “being really useful” and “based on functional programming techniques” and “different than other APIs” means that we still have developers struggling to figure out how to use it.

Jeff H has a co-worker, Clarence, who is very “anti-stream”. “It creates too many copies of our objects, so it’s terrible for memory, and it’s so much slower. Don’t use streams unless you absolutely have to!” So in many a code review, Jeff submits some very simple, easy to read, and fast-performing bit of stream code, and Clarence objects. “It’s slow. It wastes memory.”

Sometimes, another team member goes to bat for Jeff’s code. Sometimes they don’t. But then, in a recent review, Clarence submitted his own bit of stream code.

schedules.stream().forEach(schedule -> visitors.stream().forEach(scheduleVisitor -> {
    scheduleVisitor.visitSchedule(schedule);

    if (schedule.getDays() != null && !schedule.getDays().isEmpty()) {
        schedule.getDays().stream().forEach(day -> visitors.stream().forEach(dayVisitor -> {
            dayVisitor.visitDay(schedule, day);

            if (day.getSlots() != null && !day.getSlots().isEmpty()) {
                day.getSlots().stream().forEach(slot -> visitors.stream().forEach(slotVisitor -> {
                    slotVisitor.visitSlot(schedule, day, slot);
                }));
            }
        }));
    }
}));

That is six nested “for each” operations, and they’re structured so that we iterate across the same list multiple times. For each schedule, we look at each visitor on that schedule, then we look at each day for that schedule, and then we look at every visitor again, then we look at each day’s slots, and then we look at each visitor again.

Well, if nothing else, we understand why Clarence thinks the Java Streams API is slow. This code did not pass code review.

[Advertisement] ProGet’s got you covered with security and access controls on your NuGet feeds. Learn more.

August 05, 2020

Worse Than FailureCodeSOD: A Private Code Review

Jessica has worked with some cunning developers in the past. To help cope with some of that “cunning”, they’ve recently gone out searching for new developers.

Now, there were some problems with their job description and salary offer, specifically, they were asking for developers who do too much and get paid too little. Which is how Jessica started working with Blair. Jessica was hoping to staff up her team with some mid-level or junior developers with a background in web development. Instead, she got Blair, a 13+ year veteran who had just started doing web development in the past six months.

Now, veteran or not, there is a code review process, so everything Blair does goes through code review. And that catches some… annoying habits, but every once in awhile, something might sneak through. For example, he thinks static is a code smell, and thus removes the keyword any time he sees it. He’ll rewrite most of the code to work around it, except once the method was called from a cshtml template file, so no one discovered that it didn’t work until someone reported the error.

Blair also laments that with all the JavaScript and loosely typed languages, kids these days don’t understand the importance of separation of concerns and putting a barrier between interface and implementation. To prove his point, he submitted his MessageBL class. BL, of course, is to remind you that this class is “business logic”, which is easy to forget because it’s in an assembly called theappname.businesslogic.

Within that class, he implemented a bunch of data access methods, and this pair of methods lays out the pattern he followed.

public async Task<LinkContentUpdateTrackingModel> GetLinkAndContentTrackingModelAndUpdate(int id, Msg msg)
{
    return await GetLinkAndContentTrackingAndUpdate(id, msg);
}

/// <summary>
/// LinkTrackingUpdateLinks
/// returns: HasAnalyticsConfig, LinkTracks, ContentTracks
/// </summary>
/// <param name="id"></param>
/// <param name="msg"></param>
private async Task<LinkContentUpdateTrackingModel> GetLinkAndContentTrackingAndUpdate(int id, Msg msg)
{
  //snip
}

Here, we have one public method, and one private method. Their names, as you can see, are very similar. The public method does nothing but invoke the private method. This public method is, in fact, the only place the private method is invoked. The public method, in turn, is called only twice, from one controller.

This method also doesn’t ever need to be called, because the same block of code which constructs this object also fetches the relevant model objects. So instead of going back to the database with this thing, we could just use the already fetched objects.

But the real magic here is that Blair was veteran enough to know that he should put some “thorough” documentation using Visual Studio’s XML comment features. But he put the comments on the private method.

Jessica was not the one who reviewed this code, but adds:

I won’t blame the code reviewer for letting this through. There’s only so many times you can reject a peer review before you start questioning yourself. And sometimes, because Blair has been here so long, he checks code in without peer review as it’s a purely manual process.

[Advertisement] ProGet’s got you covered with security and access controls on your NuGet feeds. Learn more.

XKCDExposure Notification

August 04, 2020

Worse Than FailureA Massive Leak

"Memory leaks are impossible in a garbage collected language!" is one of my favorite lies. It feels true, but it isn't. Sure, it's much harder to make them, and they're usually much easier to track down, but you can still create a memory leak. Most times, it's when you create objects, dump them into a data structure, and never empty that data structure. Usually, it's just a matter of finding out what object references are still being held. Usually.

A few months ago, I discovered a new variation on that theme. I was working on a C# application that was leaking memory faster than bad waterway engineering in the Imperial Valley.

A large, glowing, computer-controlled chandelier

I don't exactly work in the "enterprise" space anymore, though I still interact with corporate IT departments and get to see some serious internal WTFs. This is a chandelier we built for the Allegheny Health Network's Cancer Institute which recently opened in Pittsburgh. It's 15 meters tall, weighs about 450kg, and is broken up into 30 segments, each with hundreds of addressable LEDs in a grid. The software we were writing was built to make them blink pretty.

Each of those 30 segments is home to a single-board computer with their GPIO pins wired up to addressable LEDs. Each computer runs a UDP listener, and we blast them with packets containing RGB data, which they dump to the LEDs using a heavily tweaked version of LEDScape.

This is our standard approach to most of our lighting installations. We drop a Beaglebone onto a custom circuit board and let it drive the LEDs, then we have a render-box someplace which generates frame data and chops it up into UDP packets. Depending on the environment, we can drive anything from 30-120 frames per second this way (and probably faster, but that's rarely useful).

Apologies to the networking folks, but this works very well. Yes, we're blasting many megabytes of raw bitmap data across the network, but we're usually on our own dedicated network segment. We use UDP because, well, we don't care about the data that much. A dropped packet or an out of order packet isn't going to make too large a difference in most cases. We don't care if our destination Beaglebone is up or down, we just blast the packets out onto the network, and they get there reliably enough that the system works.

Now, normally, we do this from Python programs on Linux. For this particular installation, though, we have an interactive kiosk which provides details about cancer treatments and patient success stories, and lets the users interact with the chandelier in real time. We wanted to show them a 3D model of the chandelier on the screen, and show them an animation on the UI that was mirrored in the physical object. After considering our options, we decided this was a good case for Unity and C#. After a quick test of doing multitouch interactions, we also decided that we shouldn't deploy to Linux (Unity didn't really have good Linux multitouch support), so we would deploy a Windows kiosk. This meant we were doing most of our development on MacOS, but our final build would be for Windows.

Months go by. We worked on the software while building the physical pieces, which meant the actual testbed hardware wasn't available for most of the development cycle. Custom electronics were being refined and physical designs were changing as we iterated to the best possible outcome. This is normal for us, but it meant that we didn't start getting real end-to-end testing until very late in the process.

Once we started test-hanging chandelier pieces, we started basic developer testing. You know how it is: you push the run button, you test a feature, you push the stop button. Tweak the code, rinse, repeat. Eventually, though, we had about 2/3rds of the chandelier pieces plugged in, and started deploying to the kiosk computer, running Windows.

We left it running, and the next time someone walked by and decided to give the screen a tap… nothing happened. It was hung. Well, that could be anything. We rebooted and checked again, and everything seems fine, until a few minutes later, when it's hung… again. We checked the task manager- which hey, everything is really slow, and sure enough, RAM is full and the computer is so slow because it's constantly thrashing to disk.

We're only a few weeks before we actually have to ship this thing, and we've discovered a massive memory leak, and it's such a sudden discovery that it feels like the draining of Lake Agassiz. No problem, though, we go back to our dev machines, fire it up in the profiler, and start looking for the memory leak.

Which wasn't there. The memory leak only appeared in the Windows build, and never happened in the Mac or Linux builds. Clearly, there must be some different behavior, and it must be around object lifecycles. When you see a memory leak in a GCed language, you assume you're creating objects that the GC ends up thinking are in use. In the case of Unity, your assumption is that you're handing objects off to the game engine, and not telling it you're done with them. So that's what we checked, but we just couldn't find anything that fit the bill.

Well, we needed to create some relatively large arrays to use as framebuffers. Maybe that's where the problem lay? We keep digging through the traces, we added a bunch of profiling code, we spent days trying to dig into this memory leak…

… and then it just went away. Our memory leak just became a Heisenbug, our shipping deadline was even closer, and we officially knew less about what was going wrong than when we started. For bonus points, once this kiosk ships, it's not going to be connected to the Internet, so if we need to patch the software, someone is going to have to go onsite. And we aren't going to have a suitable test environment, because we're not exactly going to build two gigantic chandeliers.

The folks doing assembly had the whole chandelier built up, hanging in three sections (we don't have any 14m tall ceiling spaces), and all connected to the network for a smoke test. There wasn't any smoke, but they needed to do more work. Someone unplugged a third of the chandelier pieces from the network.

And the memory leak came back.

We use UDP because we don't care if our packet sends succeed or not. Frame-by-frame, we just want to dump the data on the network and hope for the best. On MacOS and Linux, our software usually uses a sender thread that just, at the end of the day, wraps around calls to the send system call. It's simple, it's dumb, and it works. We ignore errors.

In C#, though, we didn't do things exactly the same way. Instead, we used the .NET UdpClient object and it's SendAsync method. We assumed that it would do roughly the same thing.

We were wrong.

await client.SendAsync(packet, packet.Length, hostip, port);

Async operations in C# use Tasks, which are like promises or futures in other environments. It lets .NET manage background threads without the developer worrying about the details. The await keyword is syntactic sugar which lets .NET know that it can hand off control to another thread while we wait. While we await here, we don't actually await the results of the await, because again: we don't care about the results of the operation. Just send the packet, hope for the best.

We don't care- but Windows does. After a load of investigation, what we discovered is that Windows would first try and resolve the IP address. Which, if a host was down, obviously it couldn't. But Windows was friendly, Windows was smart, and Windows wasn't going to let us down: it kept the Task open and kept trying to resolve the address. It held the task open for 3 seconds before finally deciding that it couldn't reach the host and errored out.

An error which, as I stated before, we were ignoring, because we didn't care.

Still, if you can count and have a vague sense of the linear passage of time, you can see where this is going. We had 30 hosts. We sent each of the 30 packets every second. When one or more of those hosts were down, Windows would keep each of those packets "alive" for 3 seconds. By the time that one expired, 90 more had queued up behind it.

That was the source of our memory leak, and our Heisenbug. If every Beaglebone was up, we didn't have a memory leak. If only one of them was down, the leak was pretty slow. If ten or twenty were out, the leak was a waterfall.

I spent a lot of time reading up on Windows networking after this. Despite digging through the socket APIs, I honestly couldn't figure out how to defeat this behavior. I tried various timeout settings. I tried tracking each task myself and explicitly timing them out if they took longer than a few frames to send. I was never able to tell Windows, "just toss the packet and hope for the best".

Well, my co-worker was building health monitoring on the Beaglebones anyway. While the kiosk wasn't going to be on the Internet via a "real" Internet connection, we did have a cellular modem attached, which we could use to send health info, so getting pings that say "hey, one of the Beaglebones failed" is useful. So my co-worker hooked that into our network sending layer: don't send frames to Beaglebones which are down. Recheck the down Beaglebones every five minutes or so. Continue to hope for the best.

This solution worked. We shipped. The device looks stunning, and as patients and guests come to use it, I hope they find some useful information, a little joy, and maybe some hope while playing with it. And while there may or may not be some ugly little hacks still lurking in that code, this was the one thing which made me say: WTF.

[Advertisement] Utilize BuildMaster to release your software with confidence, at the pace your business demands. Download today!

August 03, 2020

XKCDScientist Tech Help

July 31, 2020

XKCDCosmologist Genres

July 30, 2020

etbeLinks July 2020

iMore has an insightful article about Apple’s transition to the ARM instruction set for new Mac desktops and laptops [1]. I’d still like to see them do something for the server side.

Umair Haque wrote an insightful article about How the American Idiot Made America Unlivable [2]. We are witnessing the destruction of a once great nation.

Chris Lamb wrote an interesting blog post about comedy shows with the laugh tracks edited out [3]. He then compares that to social media with the like count hidden which is an interesting perspective. I’m not going to watch TV shows edited in that way (I’ve enjoyed BBT inspite of all the bad things about it) and I’m not going to try and hide like counts on social media. But it’s interesting to consider these things.

Cory Doctorow wrote an interesting Locus article suggesting that we could have full employment by a transition to renewable energy and methods for cleaning up the climate problems we are too late to prevent [4]. That seems plausible, but I think we should still get a Universal Basic Income.

The Thinking Shop has posters and decks of cards with logical fallacies and cognitive biases [5]. Every company should put some of these in meeting rooms. Also they have free PDFs to download and print your own posters.

gayhomophobe.com [6] is a site that lists powerful homophobic people who hurt GLBT people but then turned out to be gay. It’s presented in an amusing manner, people who hurt others deserve to be mocked.

Wired has an insightful article about the shutdown of Backpage [7]. The owners of Backpage weren’t nice people and they did some stupid things which seem bad (like editing posts to remove terms like “lolita”). But they also worked well with police to find criminals. The opposition to what Backpage were doing conflates sex trafficing, child prostitution, and legal consenting adult sex work. Taking down Backpage seems to be a bad thing for the victims of sex trafficing, for consenting adult sex workers, and for society in general.

Cloudflare has an interesting blog post about short lived certificates for ssh access [8]. Instead of having user’s ssh keys stored on servers each user has to connect to a SSO server to obtain a temporary key before connecting, so revoking an account is easy.

July 17, 2020

Dave HallIf You’re not Using YAML for CloudFormation Templates, You’re Doing it Wrong

In my last blog post, I promised a rant about using YAML for CloudFormation templates. Here it is. If you persevere to the end I’ll also show you have to convert your existing JSON based templates to YAML.

Many of the points I raise below don’t just apply to CloudFormation. They are general comments about why you should use YAML over JSON for configuration when you have a choice.

One criticism of YAML is its reliance on indentation. A lot of the code I write these days is Python, so indentation being significant is normal. Use a decent editor or IDE and this isn’t a problem. It doesn’t matter if you’re using JSON or YAML, you will want to validate and lint your files anyway. How else will you find that trailing comma in your JSON object?

Now we’ve got that out of the way, let me try to convince you to use YAML.

As developers we are regularly told that we need to document our code. CloudFormation is Infrastructure as Code. If it is code, then we need to document it. That starts with the Description property at the top of the file. If you JSON for your templates, that’s it, you have no other opportunity to document your templates. On the other hand, if you use YAML you can add inline comments. Anywhere you need a comment, drop in a hash # and your comment. Your team mates will thank you.

JSON templates don’t support multiline strings. These days many developers have 4K or ultra wide monitors, we don’t want a string that spans the full width of our 34” screen. Text becomes harder to read once you exceed that “90ish” character limit. With JSON your multiline string becomes "[90ish-characters]\n[another-90ish-characters]\n[and-so-on"]. If you opt for YAML, you can use the greater than symbol (>) and then start your multiline comment like so:

Description: >
  This is the first line of my Description
  and it continues on my second line
  and I'll finish it on my third line.

As you can see it much easier to work with multiline string in YAML than JSON.

“Folded blocks” like the one above are created using the > replace new lines with spaces. This allows you to format your text in a more readable format, but allow a machine to use it as intended. If you want to preserve the new line, use the pipe (|) to create a “literal block”. This is great for an inline Lambda functions where the code remains readable and maintainable.

  APIFunction:
    Type: AWS::Lambda::Function
    Properties:
      Code:
        ZipFile: |
          import json
          import random


          def lambda_handler(event, context):
              return {"statusCode": 200, "body": json.dumps({"value": random.random()})}
      FunctionName: "GetRandom"
      Handler: "index.lambda_handler"
      MemorySize: 128
      Role: !GetAtt LambdaServiceRole.Arn
      Runtime: "python3.7"
		Timeout: 5

Both JSON and YAML require you to escape multibyte characters. That’s less of an issue with CloudFormation templates as generally you’re only using the ASCII character set.

In a YAML file generally you don’t need to quote your strings, but in JSON double quotes are used every where, keys, string values and so on. If your string contains a quote you need to escape it. The same goes for tabs, new lines, backslashes and and so on. JSON based CloudFormation templates can be hard to read because of all the escaping. It also makes it harder to handcraft your JSON when your code is a long escaped string on a single line.

Some configuration in CloudFormation can only be expressed as JSON. Step Functions and some of the AppSync objects in CloudFormation only allow inline JSON configuration. You can still use a YAML template and it is easier if you do when working with these objects.

The JSON only configuration needs to be inlined in your template. If you’re using JSON you have to supply this as an escaped string, rather than nested objects. If you’re using YAML you can inline it as a literal block. Both YAML and JSON templates support functions such as Sub being applied to these strings, it is so much more readable with YAML. See this Step Function example lifted from the AWS documentation:

MyStateMachine:
  Type: "AWS::StepFunctions::StateMachine"
  Properties:
    DefinitionString:
      !Sub |
        {
          "Comment": "A simple AWS Step Functions state machine that automates a call center support session.",
          "StartAt": "Open Case",
          "States": {
            "Open Case": {
              "Type": "Task",
              "Resource": "arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:open_case",
              "Next": "Assign Case"
            }, 
            "Assign Case": {
              "Type": "Task",
              "Resource": "arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:assign_case",
              "Next": "Work on Case"
            },
            "Work on Case": {
              "Type": "Task",
              "Resource": "arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:work_on_case",
              "Next": "Is Case Resolved"
            },
            "Is Case Resolved": {
                "Type" : "Choice",
                "Choices": [ 
                  {
                    "Variable": "$.Status",
                    "NumericEquals": 1,
                    "Next": "Close Case"
                  },
                  {
                    "Variable": "$.Status",
                    "NumericEquals": 0,
                    "Next": "Escalate Case"
                  }
              ]
            },
             "Close Case": {
              "Type": "Task",
              "Resource": "arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:close_case",
              "End": true
            },
            "Escalate Case": {
              "Type": "Task",
              "Resource": "arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:escalate_case",
              "Next": "Fail"
            },
            "Fail": {
              "Type": "Fail",
              "Cause": "Engage Tier 2 Support."    }   
          }
        }

If you’re feeling lazy you can use inline JSON for IAM policies that you’ve copied from elsewhere. It’s quicker than converting them to YAML.

YAML templates are smaller and more compact than the same configuration stored in a JSON based template. Smaller yet more readable is winning all round in my book.

If you’re still not convinced that you should use YAML for your CloudFormation templates, go read Amazon’s blog post from 2017 advocating the use of YAML based templates.

Amazon makes it easy to convert your existing templates from JSON to YAML. cfn-flip is aPython based AWS Labs tool for converting CloudFormation templates between JSON and YAML. I will assume you’ve already installed cfn-flip. Once you’ve done that, converting your templates with some automated cleanups is just a command away:

cfn-flip --clean template.json template.yaml

git rm the old json file, git add the new one and git commit and git push your changes. Now you’re all set for your new life using YAML based CloudFormation templates.

If you want to learn more about YAML files in general, I recommend you check our Learn X in Y Minutes’ Guide to YAML. If you want to learn more about YAML based CloudFormation templates, check Amazon’s Guide to CloudFormation Templates.

July 16, 2020

etbeWindows 10 on Debian under KVM

Here are some things that you need to do to get Windows 10 running on a Debian host under KVM.

UEFI Booting

UEFI is big and complex, but most of what it does isn’t needed at all. If all you want to do is boot from an image of a disk with a GPT partition table then you just install the package ovmf and add something like the following to your KVM start script:

UEFI="-drive if=pflash,format=raw,readonly,file=/usr/share/OVMF/OVMF_CODE.fd -drive if=pflash,format=raw,readonly,file=/usr/share/OVMF/OVMF_VARS.fd"

Note that some of the documentation on this doesn’t have the OVMF_VARS.fd file set to readonly. Allowing writes to that file means that the VM boot process (and maybe later) can change EFI variables that affect later boots and other VMs if they all share the same file. For a basic boot you don’t need to change variables so you want it read-only. Also having it read-only is necessary if you want to run KVM as non-root.

As an experiment I tried booting without the OVMF_VARS.fd file, it didn’t boot and then even after configuring it to use the OVMF_VARS.fd file again Windows gave a boot error about the “boot configuration data file” that required booting from recovery media. Apparently configuration mistakes with EFI can mess up the Windows installation, so be careful and backup the Windows installation regularly!

Linux can boot from EFI but you generally don’t want to unless the boot device is larger than 2TB. It’s relatively easy to convert a Linux installation on a GPT disk to a virtual image on a DOS partition table disk or on block devices without partition tables and that gives a faster boot. If the same person runs the host hardware and the VMs then the best choice for Linux is to have no partition tables just one filesystem per block device (which makes resizing much easier) and have the kernel passed as a parameter to kvm. So booting a VM from EFI is probably only useful for booting Windows VMs and for Linux boot loader development and testing.

As an aside, the Debian Wiki page about Secure Boot on a VM [4] was useful for this. It’s unfortunate that it and so much of the documentation about UEFI is about secure boot which isn’t so useful if you just want to boot a system without regard to the secure boot features.

Emulated IDE Disks

Debian kernels (and probably kernels from many other distributions) are compiled with the paravirtualised storage device drivers. Windows by default doesn’t support such devices so you need to emulate an IDE/SATA disk so you can boot Windows and install the paravirtualised storage driver. The following configuration snippet has a commented line for paravirtualised IO (which is fast) and an uncommented line for a virtual IDE/SATA disk that will allow an unmodified Windows 10 installation to boot.

#DRIVE="-drive format=raw,file=/home/kvm/windows10,if=virtio"
DRIVE="-drive id=disk,format=raw,file=/home/kvm/windows10,if=none -device ahci,id=ahci -device ide-drive,drive=disk,bus=ahci.0"

Spice Video

Spice is an alternative to VNC, Here is the main web site for Spice [1]. Spice has many features that could be really useful for some people, like audio, sharing USB devices from the client, and streaming video support. I don’t have a need for those features right now but it’s handy to have options. My main reason for choosing Spice over VNC is that the mouse cursor in the ssvnc doesn’t follow the actual mouse and can be difficult or impossible to click on items near edges of the screen.

The following configuration will make the QEMU code listen with SSL on port 1234 on all IPv4 addresses. Note that this exposes the Spice password to anyone who can run ps on the KVM server, I’ve filed Debian bug #965061 requesting the option of a password file to address this. Also note that the “qxl” virtual video hardware is VGA compatible and can be expected to work with OS images that haven’t been modified for virtualisation, but that they work better with special video drivers.

KEYDIR=/etc/letsencrypt/live/kvm.example.com-0001
-spice password=xxxxxxxx,x509-cacert-file=$KEYDIR/chain.pem,x509-key-file=$KEYDIR/privkey.pem,x509-cert-file=$KEYDIR/cert.pem,tls-port=1234,tls-channel=main -vga qxl

To connect to the Spice server I installed the spice-client-gtk package in Debian and ran the following command:

spicy -h kvm.example.com -s 1234 -w xxxxxxxx

Note that this exposes the Spice password to anyone who can run ps on the system used as a client for Spice, I’ve filed Debian bug #965060 requesting the option of a password file to address this.

This configuration with an unmodified Windows 10 image only supported 800*600 resolution VGA display.

Networking

To set up bridged networking as non-root you need to do something like the following as root:

chgrp kvm /usr/lib/qemu/qemu-bridge-helper
setcap cap_net_admin+ep /usr/lib/qemu/qemu-bridge-helper
mkdir -p /etc/qemu
echo "allow all" > /etc/qemu/bridge.conf
chgrp kvm /etc/qemu/bridge.conf
chmod 640 /etc/qemu/bridge.conf

Windows 10 supports the emulated Intel E1000 network card. Configuration like the following configures networking on a bridge named br0 with an emulated E1000 card. MAC addresses that have a 1 in the second least significant bit of the first octet are “locally administered” (like IPv4 addresses starting with “10.”), see the Wikipedia page about MAC Address for details.

The following is an example of network configuration where $ID is an ID number for the virtual machine. So far I haven’t come close to 256 VMs on one network so I’ve only needed one octet.

NET="-device e1000,netdev=net0,mac=02:00:00:00:01:$ID -netdev tap,id=net0,helper=/usr/lib/qemu/qemu-bridge-helper,br=br0"

Final KVM Settings

KEYDIR=/etc/letsencrypt/live/kvm.example.com-0001
SPICE="-spice password=xxxxxxxx,x509-cacert-file=$KEYDIR/chain.pem,x509-key-file=$KEYDIR/privkey.pem,x509-cert-file=$KEYDIR/cert.pem,tls-port=1234,tls-channel=main -vga qxl"

UEFI="-drive if=pflash,format=raw,readonly,file=/usr/share/OVMF/OVMF_CODE.fd -drive if=pflash,format=raw,readonly,file=/usr/share/OVMF/OVMF_VARS.fd"

DRIVE="-drive format=raw,file=/home/kvm/windows10,if=virtio"

NET="-device e1000,netdev=net0,mac=02:00:00:00:01:$ID -netdev tap,id=net0,helper=/usr/lib/qemu/qemu-bridge-helper,br=br0"

kvm -m 4000 -smp 2 $SPICE $UEFI $DRIVE $NET

Windows Settings

The Spice Download page has a link for “spice-guest-tools” that has the QNX video driver among other things [2]. This seems to be needed for resolutions greater than 800*600.

The Virt-Manager Download page has a link for “virt-viewer” which is the Spice client for Windows systems [3], they have MSI files for both i386 and AMD64 Windows.

It’s probably a good idea to set display and system to sleep after never (I haven’t tested what happens if you don’t do that, but there’s no benefit in sleeping). Before uploading an image I disabled the pagefile and set the partition to the minimum size so I had less data to upload.

Problems

Here are some things I haven’t solved yet.

The aSpice Android client for the Spice protocol fails to connect with the QEMU code at the server giving the following message on stderr: “error:14094418:SSL routines:ssl3_read_bytes:tlsv1 alert unknown ca:../ssl/record/rec_layer_s3.c:1544:SSL alert number 48“.

Spice is supposed to support dynamic changes to screen resolution on the VM to match the window size at the client, this doesn’t work for me, not even with the Red Hat QNX drivers installed.

The Windows Spice client doesn’t seem to support TLS, I guess running some sort of proxy for TLS would work but I haven’t tried that yet.

July 14, 2020

etbeDebian PPC64EL Emulation

In my post on Debian S390X Emulation [1] I mentioned having problems booting a Debian PPC64EL kernel under QEMU. Giovanni commented that they had PPC64EL working and gave a link to their site with Debian QEMU images for various architectures [2]. I tried their image which worked then tried mine again which also worked – it seemed that a recent update in Debian/Unstable fixed the bug that made QEMU not work with the PPC64EL kernel.

Here are the instructions on how to do it.

First you need to create a filesystem in an an image file with commands like the following:

truncate -s 4g /vmstore/ppc
mkfs.ext4 /vmstore/ppc
mount -o loop /vmstore/ppc /mnt/tmp

Then visit the Debian Netinst page [3] to download the PPC64EL net install ISO. Then loopback mount it somewhere convenient like /mnt/tmp2.

The package qemu-system-ppc has the program for emulating a PPC64LE system, the qemu-user-static package has the program for emulating PPC64LE for a single program (IE a statically linked program or a chroot environment), you need this to run debootstrap. The following commands should be most of what you need.

apt install qemu-system-ppc qemu-user-static

update-binfmts --display

# qemu ppc64 needs exec stack to solve "Could not allocate dynamic translator buffer"
# so enable that on SE Linux systems
setsebool -P allow_execstack 1

debootstrap --foreign --arch=ppc64el --no-check-gpg buster /mnt/tmp file:///mnt/tmp2
chroot /mnt/tmp /debootstrap/debootstrap --second-stage

cat << END > /mnt/tmp/etc/apt/sources.list
deb http://mirror.internode.on.net/pub/debian/ buster main
deb http://security.debian.org/ buster/updates main
END
echo "APT::Install-Recommends False;" > /mnt/tmp/etc/apt/apt.conf

echo ppc64 > /mnt/tmp/etc/hostname

# /usr/bin/awk: error while loading shared libraries: cannot restore segment prot after reloc: Permission denied
# only needed for chroot
setsebool allow_execmod 1

chroot /mnt/tmp apt update
# why aren't they in the default install?
chroot /mnt/tmp apt install perl dialog
chroot /mnt/tmp apt dist-upgrade
chroot /mnt/tmp apt install bash-completion locales man-db openssh-server build-essential systemd-sysv ifupdown vim ca-certificates gnupg
# install kernel last because systemd install rebuilds initrd
chroot /mnt/tmp apt install linux-image-ppc64el
chroot /mnt/tmp dpkg-reconfigure locales
chroot /mnt/tmp passwd

cat << END > /mnt/tmp/etc/fstab
/dev/vda / ext4 noatime 0 0
#/dev/vdb none swap defaults 0 0
END

mkdir /mnt/tmp/root/.ssh
chmod 700 /mnt/tmp/root/.ssh
cp ~/.ssh/id_rsa.pub /mnt/tmp/root/.ssh/authorized_keys
chmod 600 /mnt/tmp/root/.ssh/authorized_keys

rm /mnt/tmp/vmlinux* /mnt/tmp/initrd*
mkdir /boot/ppc64
cp /mnt/tmp/boot/[vi]* /boot/ppc64

# clean up
umount /mnt/tmp
umount /mnt/tmp2

# setcap binary for starting bridged networking
setcap cap_net_admin+ep /usr/lib/qemu/qemu-bridge-helper

# afterwards set the access on /etc/qemu/bridge.conf so it can only
# be read by the user/group permitted to start qemu/kvm
echo "allow all" > /etc/qemu/bridge.conf

Here is an example script for starting kvm. It can be run by any user that can read /etc/qemu/bridge.conf.

#!/bin/bash
set -e

KERN="kernel /boot/ppc64/vmlinux-4.19.0-9-powerpc64le -initrd /boot/ppc64/initrd.img-4.19.0-9-powerpc64le"

# single network device, can have multiple
NET="-device e1000,netdev=net0,mac=02:02:00:00:01:04 -netdev tap,id=net0,helper=/usr/lib/qemu/qemu-bridge-helper"

# random number generator for fast start of sshd etc
RNG="-object rng-random,filename=/dev/urandom,id=rng0 -device virtio-rng-pci,rng=rng0"

# I have lockdown because it does no harm now and is good for future kernels
# I enable SE Linux everywhere
KERNCMD="net.ifnames=0 noresume security=selinux root=/dev/vda ro lockdown=confidentiality"

kvm -drive format=raw,file=/vmstore/ppc64,if=virtio $RNG -nographic -m 1024 -smp 2 $KERN -curses -append "$KERNCMD" $NET

July 09, 2020

Dave HallLogging Step Functions to CloudWatch

Many AWS Services log to CloudWatch. Some do it out of the box, others need to be configured to log properly. When Amazon released Step Functions, they didn’t include support for logging to CloudWatch. In February 2020, Amazon announced StepFunctions could now log to CloudWatch. Step Functions still support CloudTrail logs, but CloudWatch logging is more useful for many teams.

Users need to configure Step Functions to log to CloudWatch. This is done on a per State Machine basis. Of course you could click around he console to enable it, but that doesn’t scale. If you use CloudFormation to manage your Step Functions, it is only a few extra lines of configuration to add the logging support.

In my example I will assume you are using YAML for your CloudFormation templates. I’ll save my “if you’re using JSON for CloudFormation you’re doing it wrong” rant for another day. This is a cut down example from one of my services:

---
AWSTemplateFormatVersion: '2010-09-09'
Description: StepFunction with Logging Example.
Parameters:
Resources:
  StepFunctionExecRole:
    Type: AWS::IAM::Role
    Properties:
      AssumeRolePolicyDocument:
        Version: '2012-10-17'
        Statement:
        - Effect: Allow
          Principal:
            Service: !Sub "states.${AWS::Region}.amazonaws.com"
          Action:
          - sts:AssumeRole
      Path: "/"
      Policies:
      - PolicyName: StepFunctionExecRole
        PolicyDocument:
          Version: '2012-10-17'
          Statement:
          - Effect: Allow
            Action:
            - lambda:InvokeFunction
            - lambda:ListFunctions
            Resource: !Sub "arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:my-lambdas-namespace-*"
          - Effect: Allow
            Action:
            - logs:CreateLogDelivery
            - logs:GetLogDelivery
            - logs:UpdateLogDelivery
            - logs:DeleteLogDelivery
            - logs:ListLogDeliveries
            - logs:PutResourcePolicy
            - logs:DescribeResourcePolicies
            - logs:DescribeLogGroups
            Resource: "*"
  MyStateMachineLogGroup:
    Type: AWS::Logs::LogGroup
    Properties:
      LogGroupName: /aws/stepfunction/my-step-function
      RetentionInDays: 14
  DashboardImportStateMachine:
    Type: AWS::StepFunctions::StateMachine
    Properties:
      StateMachineName: my-step-function
      StateMachineType: STANDARD
      LoggingConfiguration:
        Destinations:
          - CloudWatchLogsLogGroup:
             LogGroupArn: !GetAtt MyStateMachineLogGroup.Arn
        IncludeExecutionData: True
        Level: ALL
      DefinitionString:
        !Sub |
        {
          ... JSON Step Function definition goes here
        }
      RoleArn: !GetAtt StepFunctionExecRole.Arn

The key pieces in this example are the second statement in the IAM Role with all the logging permissions, the LogGroup defined by MyStateMachineLogGroup and the LoggingConfiguration section of the Step Function definition.

The IAM role permissions are copied from the example policy in the AWS documentation for using CloudWatch Logging with Step Functions. The CloudWatch IAM permissions model is pretty weak, so we need to grant these broad permissions.

The LogGroup definition creates the log group in CloudWatch. You can use what ever value you want for the LogGroupName. I followed the Amazon convention of prefixing everything with /aws/[service-name]/ and then appended the Step Function name. I recommend using the RetentionInDays configuration. It stops old logs sticking around for ever. In my case I send all my logs to ELK, so I don’t need to retain them in CloudWatch long term.

Finally we use the LoggingConfiguration to tell AWS where we want to send out logs. You can only specify a single Destinations. The IncludeExecutionData determines if the inputs and outputs of each function call is logged. You should not enable this if you are passing sensitive information between your steps. The verbosity of logging is controlled by Level. Amazon has a page on Step Function log levels. For dev you probably want to use ALL to help with debugging but in production you probably only need ERROR level logging.

I removed the Parameters and Output from the template. Use them as you need to.

July 05, 2020

etbeDebian S390X Emulation

I decided to setup some virtual machines for different architectures. One that I decided to try was S390X – the latest 64bit version of the IBM mainframe. Here’s how to do it, I tested on a host running Debian/Unstable but Buster should work in the same way.

First you need to create a filesystem in an an image file with commands like the following:

truncate -s 4g /vmstore/s390x
mkfs.ext4 /vmstore/s390x
mount -o loop /vmstore/s390x /mnt/tmp

Then visit the Debian Netinst page [1] to download the S390X net install ISO. Then loopback mount it somewhere convenient like /mnt/tmp2.

The package qemu-system-misc has the program for emulating a S390X system (among many others), the qemu-user-static package has the program for emulating S390X for a single program (IE a statically linked program or a chroot environment), you need this to run debootstrap. The following commands should be most of what you need.

# Install the basic packages you need
apt install qemu-system-misc qemu-user-static debootstrap

# List the support for different binary formats
update-binfmts --display

# qemu s390x needs exec stack to solve "Could not allocate dynamic translator buffer"
# so you probably need this on SE Linux systems
setsebool allow_execstack 1

# commands to do the main install
debootstrap --foreign --arch=s390x --no-check-gpg buster /mnt/tmp file:///mnt/tmp2
chroot /mnt/tmp /debootstrap/debootstrap --second-stage

# set the apt sources
cat << END > /mnt/tmp/etc/apt/sources.list
deb http://YOURLOCALMIRROR/pub/debian/ buster main
deb http://security.debian.org/ buster/updates main
END
# for minimal install do not want recommended packages
echo "APT::Install-Recommends False;" > /mnt/tmp/etc/apt/apt.conf

# update to latest packages
chroot /mnt/tmp apt update
chroot /mnt/tmp apt dist-upgrade

# install kernel, ssh, and build-essential
chroot /mnt/tmp apt install bash-completion locales linux-image-s390x man-db openssh-server build-essential
chroot /mnt/tmp dpkg-reconfigure locales
echo s390x > /mnt/tmp/etc/hostname
chroot /mnt/tmp passwd

# copy kernel and initrd
mkdir -p /boot/s390x
cp /mnt/tmp/boot/vmlinuz* /mnt/tmp/boot/initrd* /boot/s390x

# setup /etc/fstab
cat << END > /mnt/tmp/etc/fstab
/dev/vda / ext4 noatime 0 0
#/dev/vdb none swap defaults 0 0
END

# clean up
umount /mnt/tmp
umount /mnt/tmp2

# setcap binary for starting bridged networking
setcap cap_net_admin+ep /usr/lib/qemu/qemu-bridge-helper

# afterwards set the access on /etc/qemu/bridge.conf so it can only
# be read by the user/group permitted to start qemu/kvm
echo "allow all" > /etc/qemu/bridge.conf

Some of the above can be considered more as pseudo-code in shell script rather than an exact way of doing things. While you can copy and past all the above into a command line and have a reasonable chance of having it work I think it would be better to look at each command and decide whether it’s right for you and whether you need to alter it slightly for your system.

To run qemu as non-root you need to have a helper program with extra capabilities to setup bridged networking. I’ve included that in the explanation because I think it’s important to have all security options enabled.

The “-object rng-random,filename=/dev/urandom,id=rng0 -device virtio-rng-ccw,rng=rng0” part is to give entropy to the VM from the host, otherwise it will take ages to start sshd. Note that this is slightly but significantly different from the command used for other architectures (the “ccw” is the difference).

I’m not sure if “noresume” on the kernel command line is required, but it doesn’t do any harm. The “net.ifnames=0” stops systemd from renaming Ethernet devices. For the virtual networking the “ccw” again is a difference from other architectures.

Here is a basic command to run a QEMU virtual S390X system. If all goes well it should give you a login: prompt on a curses based text display, you can then login as root and should be able to run “dhclient eth0” and other similar commands to setup networking and allow ssh logins.

qemu-system-s390x -drive format=raw,file=/vmstore/s390x,if=virtio -object rng-random,filename=/dev/urandom,id=rng0 -device virtio-rng-ccw,rng=rng0 -nographic -m 1500 -smp 2 -kernel /boot/s390x/vmlinuz-4.19.0-9-s390x -initrd /boot/s390x/initrd.img-4.19.0-9-s390x -curses -append "net.ifnames=0 noresume root=/dev/vda ro" -device virtio-net-ccw,netdev=net0,mac=02:02:00:00:01:02 -netdev tap,id=net0,helper=/usr/lib/qemu/qemu-bridge-helper

Here is a slightly more complete QEMU command. It has 2 block devices, for root and swap. It has SE Linux enabled for the VM (SE Linux works nicely on S390X). I added the “lockdown=confidentiality” kernel security option even though it’s not supported in 4.19 kernels, it doesn’t do any harm and when I upgrade systems to newer kernels I won’t have to remember to add it.

qemu-system-s390x -drive format=raw,file=/vmstore/s390x,if=virtio -drive format=raw,file=/vmswap/s390x,if=virtio -object rng-random,filename=/dev/urandom,id=rng0 -device virtio-rng-ccw,rng=rng0 -nographic -m 1500 -smp 2 -kernel /boot/s390x/vmlinuz-4.19.0-9-s390x -initrd /boot/s390x/initrd.img-4.19.0-9-s390x -curses -append "net.ifnames=0 noresume security=selinux root=/dev/vda ro lockdown=confidentiality" -device virtio-net-ccw,netdev=net0,mac=02:02:00:00:01:02 -netdev tap,id=net0,helper=/usr/lib/qemu/qemu-bridge-helper

Try It Out

I’ve got a S390X system online for a while, “ssh root@s390x.coker.com.au” with password “SELINUX” to try it out.

PPC64

I’ve tried running a PPC64 virtual machine, I did the same things to set it up and then tried launching it with the following result:

qemu-system-ppc64 -drive format=raw,file=/vmstore/ppc64,if=virtio -nographic -m 1024 -kernel /boot/ppc64/vmlinux-4.19.0-9-powerpc64le -initrd /boot/ppc64/initrd.img-4.19.0-9-powerpc64le -curses -append "root=/dev/vda ro"

Above is the minimal qemu command that I’m using. Below is the result, it stops after the “4.” from “4.19.0-9”. Note that I had originally tried with a more complete and usable set of options, but I trimmed it to the minimal needed to demonstrate the problem.

  Copyright (c) 2004, 2017 IBM Corporation All rights reserved.
  This program and the accompanying materials are made available
  under the terms of the BSD License available at
  http://www.opensource.org/licenses/bsd-license.php

Booting from memory...
Linux ppc64le
#1 SMP Debian 4.

The kernel is from the package linux-image-4.19.0-9-powerpc64le which is a dependency of the package linux-image-ppc64el in Debian/Buster. The program qemu-system-ppc64 is from version 5.0-5 of the qemu-system-ppc package.

Any suggestions on what I should try next would be appreciated.