Thursday, August 03, 2017

Creating a decent Tribblix AMI

Previously, I've described how I created my first Tribblix AMI, then how to do it properly in hvm mode so you can run on modern instances in all regions.

That creates something that will work, but is it actually in a state that's useful?

The first thing is to add an EC2 credential service. That's the thing that will query for metadata and install the keys on the system so you can log after the instance is created. I tried the ec2-credential service from OmniOS, but for some reason it didn't work right on Tribblix. I've tweaked mine a little, forcing it to run after the network comes up, adding retries in case there's a problem, and also disabling it in non-global zones.

Of course, there's more instance metadata that I could query and use, but I haven't yet had a need for anything other than the initial key.

The other thing I've been wondering about is the configuration of Tribblix itself - specifically what the storage should look like and what the default software installation should look like.

My image is built on an 8G "disk" or EBS volume. That might seem a little small, but remember that Tribblix is pretty lean and mean. For a typical server configuration you'll probably be looking at about 1G or so, and that's without any special work. The most annoying thing here is that by default you lose 2G to each of dump and swap, so that's effectively half the disk gone. There's opportunity to modify those, especially as I'm typically using t2.micro instances on the free tier that only have 1G of memory. You might not even want dump at all. As for swap, you do want some (so that anonymous reservations don't eat into actual RAM) but you could cut that down a bit.

As I'm writing this I do wonder whether I could pull some of the instance metadata and shrink the dump and swap volumes appropriately.

The assumption I'm making here, though, is that if you're storing any reasonable amount of data that you're going to attach a separate EBS volume, and you can then size that appropriately to the need at hand. (And you can then move that data around independent of your running instance.) So I think that keeping the root volume fairly small is reasonable. It also keeps my AWS bill down, an important consideration as any charges here come out of my own pocket.

Then, what should the baseline software install look like? Tribblix uses overlays, and there's an assumption that you always start from the base overlay. I'm currently using a dedicated overlay that pulls in cli-tools - essentially you get basic shells, compression tools, basic utilities, but not much else. Many of the normal server utilities don't apply to running in the cloud, as they're aimed at monitoring or managing hardware.

The base set of packages is that installed on the ISO. That includes most storage and network drivers, which are irrelevant - on EC2 you know exactly which drivers you need, so almost all the drivers that are installed are unnecessary. What I need here is a better way of handling installation variants, so it knows the drivers aren't supposed to be there - at the moment I could remove them, but updates and upgrades would simply put them back. In the same vein, I could only ship a 64-bit kernel, as we know there are no 32-bit instance types available.

At the moment I have an LX variant, which is a bit of a hack in terms of the way I've packaged it together, but as the number of interesting variants grows I'm going to have to come up with a better way of handling it, especially as you might want multiple variants together - for instance a 64-bit LX-enabled cloud-optimised image.

Monday, July 31, 2017

Building a Tribblix AMI - hvm mode

After having created a Tribblix AMI to prove that Tribblix basically works on EC2, I then moved on to the next issue - how to create an AMI that will run in hvm mode?

As a reminder, pv mode AMIs are deprecated, aren't supported by all instance types, and don't work in all regions. So you really need something that runs in hvm mode.

The first thought might be to convert the existing pv image to a hvm image. I've tried that and, while you can do the conversion, the image doesn't actually work. The problem here is that ZFS has the physical paths of the devices it's installed on embedded in the pool metadata. Changing from pv to hvm mode changes the emulated hardware, in particular the disk paths, so the ZFS pool isn't where it thought it was and the system panics. If you have a mismatch between the disk layout where the pool was created and where you're running you'll get a panic something like this:



If you had console access and could boot from media you could fix this, but AWS doesn't provide that. (And if you could boot from media you could just do a regular install without all the shenanigans involved in producing an AMI.)

So, you have to create the image on a system that looks like EC2. Which means using xen.

Fortunately, this road has been travelled before. These instructions are exactly what you need. They're for OpenIndiana, but will apply to any illumos distribution. And they're the process used by the OpenZFS project to do their testing. (I'll also mention that the OpenZFS folks have put a number of fixes back into illumos that improve the EC2 experience for us.)

I'm not going to repeat those instruction, that would be boring, so I'll talk about what I had to do or change to make those instructions work for me.

I got one of my spare desktop PCs out and installed Ubuntu 16.04 on it. (I must be spoilt by Tribblix, the Ubuntu install was horrendously slow and very high maintenance.) And then installed xen, rebooted as dom0, and set up the bridge networking.

That was my first pothole. There's this thing called systemd that's come along, and it changes the way network configuration is done. Much cussing and googling, but I got it right first time.

Then I discover that there's a new toolstack here. It's all xl not xm, but otherwise seems the same.

I then tried to start a VM, only to be given a completely meaningless and unhelpful error message. Why tell the user what's wrong when you can just vomit a stack trace?

After a bit of head-scratching I worked out that the system didn't actually support hvm mode. If you run xl info and look for virt_caps, it should mention hvm. That's a bit odd, the sticker on the front of the box looks right.

Manufacturers ship hardware with VT-x disabled in the BIOS, it appears. Into the BIOS we go, to find that the relevant settings are greyed out and you need a BIOS password to get into them. Open the box and start looking for jumpers. Fortunately I found a helpful article - the key here was the bit about the jumper being blue, little details like that make all the difference.

OK, so having wiped the BIOS password, gone into the BIOS and enabled VT-x, I go back to xen. Looking at virt_caps now shows hvm, as it should, and my domain starts.

The idea here is that you connect to the console with VNC. Easy enough, but by the time I had got my ssh tunnel set up and started up my VNC client, my VM had gone. I started it again, it starts booting just fine but then issues a few warnings and then a kernel panic. It's all over pretty quick.

In order to catch what it said, I then used vnc2flv. Someone asked me about screen recorders a while back, and I suggested they did what they wanted to do in a vnc session and use vnc2flv to record it. But it's the same here. Once I had the session recorded I can watch the movie and pause it to see what errors it's spitting out.



This, I think, is related to illumos bug 7186. It looks like we can't handle the network presented by newer versions of xen.

To get round this I simply disabled the network interface in the VM definition. Then the VM boots just fine and can be installed. You're a little bit limited in that you can't do updates but, as long as nwam is enabled then it will get itself on the network when you do run it on something that does have a compatible network.

For OmniOS, this means you have to manually enable nwam, as they have networking switched off by default. And remember that you must have networking enabled if you're running on EC2 as there's no other way to access your system.

What you'll also need to ensure at this point is that you have a functional user account you can get in to via ssh. With Tribblix and OpenIndiana you have jack, other distros might need to create a user. You wouldn't want that on a production AMI, of course, but you need to be able to log in to the system the first time in order to complete any configuration and add the various bits of AWS integration that you'll need.

Having got my image installed I followed the instructions through and got an AMI that works just fine.

The configuration file I used is:

builder='hvm'
name='ami-template'
vcpus=1
memory=1024
disk=[  'file:/var/tmp/tribblix-0m20.1.iso,hdb:cdrom,r',
        'file:/root/ami-template.img,xvda,w' ]
boot='d'
vnc=1
vnclisten='0.0.0.0'
vncconsole=1
on_crash='preserve'
xen_platform_pci=1
serial='pty'
on_reboot='destroy'


The one crucial thing here, apart from not having a vif line to create a network, is that you must use xvda for the disk. That's what EC2 will present to you, if you use something else you'll get the same panic on boot that I saw when attempting to convert a pv image.

We're almost done. Next time I'll talk about how to go from something that minimally boots up to something that's done well.

Running illumos on AWS - the first Tribblix AMI

I've run Tribblix on all sorts of hardware - desktops, servers, even the occasional laptop. I've had success running it on some of the smaller cloud providers that allow you to install from a custom ISO, or iPXE, such as my adventures with Vultr.

However, running on AWS has eluded me. You might wonder why you would want to, but the reality is that AWS is a huge player, with many people turning to it as their default (and often only) option. So giving everyone who uses AWS access to Tribblix would be a good thing, and would also offer an easy route for people who might want to play with Tribblix to do so.

The first thing to realize is that AWS is not so much a single cloud as a set of independent clouds. Each region is independent, and has a different set of capabilities. For example, EFS is only available in a few regions. These differences can affect us.

On AWS, there are 2 different types of guest. We have pv (the older, paravirtualized) and hvm (the newer, hardware assisted). Any given AMI (Amazon Machine Image) will only run as either a pv or a hvm guest. And some EC2 instance types are pv, others hvm. Newer regions (such as London) are exclusively hvm, so pv isn't an option.

Building an AMI from scratch looked a little daunting, so I looked to see what other illumos distributions might have made AMIs available. If you go to the community AMI page when launching an instances, the only one you'll find is OmniOS. They even have a page explaining how it was done. The snag is that all their images are pv. For my first set of experiments then, I was operating in the Dublin region.

The OmniOS AMI boots up just fine and works pretty much as you would expect. No problems there. How to get Tribblix running though?

The answer lies in the beauty of ZFS and Boot Environments. The basic approach here is to take a running OmniOS image, create a new Boot Environment, install Tribblix into that Boot Environment, and make the Tribblix Boot Environment the one to boot from next time. Once I've successfully booted the Tribblix image, I can clean up and delete the original OmniOS files.

One of the advantages of Tribblix is that I have my own installer. It's quite a bit simpler than some of the other distros, and thus much easier to mangle to do things in new environments. I decided to use the iPXE image as used in my Vultr experiment, because it was easy and I had it to hand. I then wrote a modified installer script (source here) called img_install that was based on my over_install script used to drop Tribblix into an existing ZFS pool. The difference is that the old over_install was run in the context of a Tribblix Live CD; the new img_install is run in the context of an alternative distro. The other thing in that script is that I don't do any boot loader fiddling - the pv instances have a special pv-grub, which I'm careful not to touch.

(By the way, the same trick will work for other illumos distributions. You just need a source archive of some sort and a script to unpack it. For example, I have a script to unpack some of the ISO images in the tribblix-zones repo, which I use to create alien-root zones. It's the same idea of installing an image in a alternate path.)

So all that was involved was to:
  • Start up an OmniOS instance (a micro instance on the free tier works fine)
  • Run the img_install script to create the alternate BE
  • Reboot, so you boot into Tribblix
  • Delete the old OmniOS BE
  • Finish off the install and apply updates
Then you can do the normal create an image trick on the AWS console, and you have a nice shiny Tribblix AMI.

That all worked out just beautifully. Tribblix runs on EC2 just fine.

In the next article, I'll describe how to create a hvm AMI.

Saturday, July 22, 2017

Mucking around with IPv6 and illumos zones

The world is running out of IPv4 addresses, and it's time to move to IPv6.

I remember that story being told over and over at conferences in the mid 1990s. Yet, here we are in 2017 and while there has been progress, we're definitely not there yet.

With zones, illumos (and Solaris) give you virtualized application environments (containers is the trendy term - we tend not to use that in the Solarish context because it got polluted by Sun marketing). Those environments (usually) need to be networked, so why not with IPv6.

So here goes with a few notes on the subject.

Shared-IP zones


With zones, the original networking model was a shared-ip stack, where the zone is given a fully configured network that is just a virtual IP configured on an existing interface. All the setup is done in the global zone, which makes it very easy.

(By the way, this was the cause of the limit of 8192 zones per system, because you can only have 8192 virtual addresses on a single physical interface.)

And configuring an IPv4 address is just a case of adding a net section to the zone configuration:

add net
set physical=aggr1
set address=172.18.1.172/24
end

It's exactly the same for IPv6, the only interesting issue is what the IPv6 address would be. Let's start with the link-local address - the one that starts with fe80:: - as you will generally need that even if you don't have a routed IPv6 network. For a physical interface, the IPv6 address is usually derived from the MAC address. We can't use that one, because we're sharing the interface and the global zone has already grabbed it. So the convention here is to construct something from the IPv4 address. It's then guaranteed to be unique in a broadcast network, which is all that matters for a link-local address. So all we have to do is convert the IPv4 address to hex, for example with printf:

printf "%x%x:%x%x\n" 172 18 1 172

which gives ac12:1ac, so the link-local address would be configured as:

add net
set physical=aggr1
set address=fe80::ac12:1ac/10
end

You're pretty much done here, if you do that your global and non-global zones will be able to communicate using IPv6 on the local subnet.

If you had a routable prefix, then the same scheme can be applied. Just put the fragment onto the end of your prefix.

add net
set physical=aggr1
set address=XXXX:XXXX:XXXX:XXXX::ac12:1ac/64
end


Of course, if you're assigned specific IPv6 addresses then you can use those directly. The above scheme is pretty trivial to script, though (and it actually makes it fairly easy to keep your DNS zone files up to date too).

Exclusive-IP zones


For an exclusive-ip zone, you just hand over a network interface to a zone and let it go figure. So it can assign whatever addresses it likes.

In particular, the zone can use the normal MAC address scheme to generate its IPv6 link-local address.

Originally in older Solaris, you needed to use a genuine physical interface. Which limited you a little bit as there are only so many network cards you can jam into a server. OpenSolaris introduced full network virtualization in the form of Crossbow, so any illumos distribution or Solaris 11 can create fully virtualized network stacks and present those to zones in the same way.

In Tribblix, I use zap to manage zones, and it takes care of creating the appropriate vnics and, if appropriate, etherstubs, and wiring things together. I also poke functional /etc/hostname.* and /etc/defaultrouter files into the zone so the networking at least gets configured when the zone boots. Adding IPv6 to the zone is simply a case of creating matching empty /etc/hostname6.* files (one for each vnic) and the IPv6 addresses will get autoconfigured.

There's one wrinkle with exclusive-ip that deserves a whole section, that of restricting the zone to only using addresses that you've set.

Restricting with allowed-address


Remember that an exclusive-ip zone can manage the network interface. So it could allocate the wrong address and generally cause havoc on the network. To prevent this, set the allowed-address property on the interface. For example:

add net
set physical=vnic1
set allowed-address=172.18.1
.172/24
end

The zone manages the interface, but an attempt to configure an invalid address will be thwarted.

You can see what's happening under the hood by running dladm show-linkprop. You'll see that the protection and allowed-ips properties are set.

(As an aside, it would be fantastic if illumos got the configure-allowed-address feature that Solaris 11 has, which would bypass my trickery in having to poke the network setup files in the zone.)

The same thing works for IPv6. The first problem I discovered is that (unlike Solaris 11) illumos won't accept multiple addresses in a list. Initially I thought this was something about IPv6, but it turns out you need to specify each address you want to add in a separate block.

The next question is going to be - what is the IPv6 address going to be? It's derived from the MAC address, so isn't fixed in advance.

The first step is to get the properties of the vnic. Running dladm show-vnic will give you the properties you need, including the MAC address. If you just want the one field, that's fairly easy too.

# dladm show-vnic -p -o MACADDRESS vnic1
2:8:20:c6:71:d

The IPv6 address is made up from that as a EUI-64 address, which is basically the fe80:: prefix, the first 3 octets, then ff:fe, then the last 3 octets. Oh, and the 7th bit gets flipped. And conventionally the leading zero gets suppressed. An ugly way of scripting this in ksh looks like:

/usr/sbin/dladm show-vnic -p -o MACADDRESS $VNIC | \
    /usr/bin/sed 's=:= =g' | read o1 o2 o3 o4 o5 o6
integer -i2 vi=16#$o1
integer -i2 nvi
nvi=$(($vi ^ 2#00000010))
integer -i16 xv=$nvi
no1=${xv/16#/}
if [ "$no1" = "0" ]; then
    no1=""

fi
if [ "$no3" = "0" ]; then
    no3=""

fi
if [ "$no5" = "0" ]; then
    no5=""

fi
printf "fe80::%s%s:%sff:fe%s:%s%s/10" "$no1" "$o2" "$o3" "$o4" "$o5" "$o6"

So, what you want to do is add something like:

add net
set physical=vnic1
set allowed-address=fe80::8:20ff:fec6:71d/10
end

to your zone configuration. And if you have a routable IPv6 address, you'll need to duplicate the block again for that.

In practice, this didn't quite work for me. If you don't set the allowed-address properties then the addresses get configured correctly, but with the protection set the address doesn't come up properly. If you try it then you get:

# ifconfig vnic1 inet6 up
ifconfig: setifflags: SIOCSLIFFLAGS: vnic1: Invalid argument

However, if you explicitly set the address, running ifconfig by hand:

# ifconfig vnic1 inet6 fe80::8:20ff:fec6:71d/10 up

then it works perfectly.

Thursday, July 13, 2017

What gets into Tribblix?

The software available for Tribblix is a bit of an eclectic mix. How do I choose what software to package?

There are actually a number of different reasons why you get a particular package.

The basics


Some packages are just basic,and you expect to find them. Much of the GNU stack comes in this way. And often things like Perl and Python are a foundational requirement for a lot of other tools.

What I want personally


There are a number of areas where I have specific interests - I'm a bit of a magpie when it comes to X11 window managers, for example. And I need to open office documents, so I had to get LibreOffice working. There a few games or emulators that I like. This also explains why some things might not be present too - I have no real interest in video or multimedia, for example, so that's an area with relatively little coverage.

Oh, that looks cool


I'm often interested in new stuff. (Even if it's actually old stuff that's just new to me.) So if I come across a piece of software and think "that looks cool" I'll often try and build and package it. If it works, fine, it ends up in the repository. Even if I might not end up doing anything with it, I've gone to the effort of making a package so it may as well stay there and somebody else might make use of it.

I have a $DAYJOB


Yes, I have a day job (a very good one, thank you very much), and it involves running applications on illumos. If I'm going to evaluate software I'll do it on Tribblix first. Building stuff on Tribblix is much easier than on, say, OmniOS - I have total control of the environment, and many more tools and prerequisite packages to give me a head start. So I can easily screen out any software that simply isn't going to work, and identify any patches or modifications necessary, before heading into the rather more constrained work environment.

Can you make X or Y or Z available


I get requests from users. I'll pretty much always at least try to add the software asked for - the fact that someone's bothered to ask indicates it might be useful, and I might find it interesting as well. This doesn't always work, of course, and I'll have to punt.

I got bored one day


Sometimes I get a bit of free time (no, this doesn't happen very often), and start looking for packages that might be worth adding. Sometimes this involves looking at other systems to see what they ship.

It's a prerequisite for something else


Dependency hell is a fact of life, so a lot of the time is actually spent building prerequisites. This is one reason for speculatively trying things out - it identifies prerequisites, and they're often going to be needed by other packages too. Although what you'll find is a number of packages with no obvious consumers, because the software I wanted didn't actually work. As I mentioned before, though, I'll keep those packages I built, and they might come in useful later.

Thursday, July 06, 2017

Running LX zones with Tribblix

I mentioned a few months ago a little project I had been working on - nicknamed omnitribblix, it's regular Tribblix with the illumos components coming from illumos-omnios (now via OmniOS Community Edition) rather than vanilla illumos-gate.

One of the changes I made in the recent Milestone 20 update was to split out the release packages to give more flexibility.

Thiis allowed me to release a micro update to Milestone 20 (imaginatively called m20.1 or update 1), which updates the illumos bits but shares the same main package repository as the main Milestone 20 release.

And the other thing I can now do is build variant releases. So Tribblix has an LX variant!

You can download the omnitribblix ISO image from the Tribblix download page. It installs, operates, and is packaged just like regular Tribblix. If you don't use LX zones, you probably wouldn't notice the difference.

(It's versioned as m20lx.1 - the update 1 there means that it's a parallel release to the regular Tribblix Milestone 20 update 1.)

You can also update to the LX variant from either the regular Milestone 20 or Milestone 20 update 1 releases, in the normal way. It's a micro update, or sidegrade perhaps, but uses the same upgrade process as regular upgrades.

And, because of the magic of boot environments, if there's a problem you can roll back.

Anyway, once you have omnitribblix installed, how do you create an LX zone? Very easily, in the same way you create and destroy other zones on Tribblix, using the zap utility.

Before you can do that, though, you need a Linux image of some sort to install.

I've been using the same images I use under Docker. So, for example, if I want Alpine then I would go:

docker run alpine uname -a

and then get the name of the container

docker ps -a

and then export that with

docker export romantic_galileo > alpine.tar

Then copy the alpine.tar file to your omnitribblix system. If you want something a bit richer, then Ubuntu will work. But generally exporting a Docker container like this will work, and the image characteristics will be a good fit for a zone.

And then all you do to create the zone is use zap, specifying that it's an lx brand and telling it where the tarball is:

zap create-zone -z alpine -t lx \
-x 10.0.2.99 -I /tmp/alpine.tar

and just zlogin to it as normal.

There are constraints around networking - you have to be exclusive-ip (the -x flag) and zap will create (and destroy) the vnic for you automatically. But the networking in the zone won't actually be configured. (While you specify the IP address in the command, that just tells zap how to configure the network plumbing and the vnic.) You'll have to log in to the zone and use the native tools to identify and configure the network, like so:

/native/sbin/ifconfig -a
/native/sbin/ifconfig znic0 inet 10.0.2.99 up
/native/usr/sbin/route add net default 10.0.2.2

And off you go. Sitting on an illumos box with all its goodness, with access to the wide variety of the Linux ecosystem at your fingertips.

Sunday, June 18, 2017

Tweaking binaries with elfedit

On Solaris and illumos, you can inspect shared objects (binaries and libraries) with elfdump. In the most common case, you're simply looking for what shared libraries you're linked against, in which case it's elfdump -d (or, for those of us who were doing this years before elfdump came into existence, dump -Lv). For example:

% elfdump -d /bin/true

Dynamic Section:  .dynamic
     index  tag                value
       [0]  NEEDED            0x1d6               libc.so.1
       [1]  INIT              0x8050d20          

and it goes on a bit. But basically you're looking at the NEEDED lines to see which shared libraries you need. (The other field that's generally of interest for a shared library is the SONAME field.)

However, you can go beyond this, and use elfedit to manipulate what's present here. You can essentially replicate the above with:

elfedit -r -e dyn:dump /bin/true

Here the -r flag says read-only (we're just looking), and -e says execute the command that follows, which is dyn:dump - or just show the dynamic section.

If you look around, you'll see that the classic example is to set the runpath (which you might see as RPATH or RUNPATH in the dump output). This was used to fix up binaries that had been built incorrectly, or where you've moved the libraries somewhere other than where the binary normally looks for them. Which might look like:

elfedit -e 'dyn:runpath /my/local/lib' prog

This is the first example in the man page, and the standard example wherever you look. (Note the quotes - that's a single command input to elfedit.)

However, another common case I come across is where libtool has completely mangled the link so the full pathname of the library (at build time, no less) has been embedded in the binary (either in absolute or relative form). In other words, rather than the NEEDED section being

libfoo.so.1

it ends up being

/home/ptribble/build/bar/.libs/libfoo.so.1

With this sort of error, no amount of tinkering with RPATH is going to help the binary find the library. Fortunately, elfedit can help us here too.

First you need to work out which element you want to modify. Back to elfedit again to dump out the structure

% elfedit -r -e dyn:dump /bin/baz
     index  tag                value
       [0]  POSFLAG_1         0x1                 [ LAZY ]
       [1]  NEEDED            0x8e2               /home/.../libfoo.so.1

It might be further down, of course. But the entry we want to edit is index number 1. We can narrow down the output just to this element by using the -dynndx flag to the dyn:dump command, for example

elfedit -r -e 'dyn:dump -dynndx 1' /bin/baz

or, equivalently, using dyn:value

elfedit -r -e 'dyn:value -dynndx 1' /bin/baz

And we can actually set the value as well. This requires the -s flag to set a string, but you end up with:

elfedit -e 'dyn:value -dynndx -s 1 libfoo.so.1' /bin/baz

and then if you use elfdump or elfedit or ldd to look at the binary, it should pick up the library correctly.

This is really very simple (the hardest part is having to work out what the index of the right entry is). I didn't find anything when searching that actually describes how simple it is, so I thought it worth documenting for the next time I need it.


Friday, June 09, 2017

On Tribblix Milestone 20

Having released a new update for Tribblix, I thought I would add a little commentary on the progress that's being made and the direction things are going in.

This goes beyond the rather dry release notes and list of what's changed.

The big structural change is that the ISO has been built as a single root archive, rather than the old way with a split-off /usr that's lofi-mounted from a compressed image.

The original reason for doing this (and I experimented with it a while ago) was to allow installation on systems without drivers for the device that you're booting from. This might be a system with only USB3 ports, or I've had problems with laptops where illumos doesn't recognize the CD drive. The boot loader (and BIOS) load the initial boot archive, so if you don't need to ever talk to the media device again you're in much better shape.

While we now have USB3 support, this simplified boot is a good thing in any case, and it allows some neat tricks like iPXE boot.

Another logical change is in the release mechanism itself. I've discussed the Tribblix package repositories before. The snag with the traditional repository layout was that the packages that defined a release were in the main Tribblix repository. So, every time I make a new release I end up having to create a whole new Tribblix repository. Every time I update the illumos packages, I needed a new Tribblix repository. Creating a new one isn't too bad; ongoing support for multiple repositories is a lot of unnecessary work.

The way to fix this is to split out the packages (there are 3 of them) that define the properties of a release into their own separate repo. This allows at least 2 new possibilities:

  1. I can release updated illumos packages without spinning a whole new Tribblix release. It would still use the same upgrade mechanism, but the main Tribblix repo is shared and it's a much lighter release process.
  2. I could create variants or spins. For example, I could create a variant that has LX (see omnitribblix). This would just have a different set of illumos packages but shares everything else. Or I could build a 32-bit or 64-bit only distro.
I haven't yet done either of those things, but it's going to happen.

Behind the scenes I've been gradually working to get more packages - especially those that deliver libraries - built as both 32-bit and 64-bit.

Tribblix is fairly clear that it will continue to support 32-bit and 64-bit hardware, at least for a while. (Whereas both OmniOS and OpenIndiana have effectively dropped 32-bit compatibility, mostly by neglect rather than design.) Of course, there is a reasonable amount of software now that's only 64-bit (anything built with go, for example, or OpenJDK 8), but there's a reasonable chance the people using 32-bit hardware aren't necessarily going to want the latest and greatest applications. (This isn't 100% true, by the way - sometime you have to interoperate with other facilities in the environment.) But eventually we're going to have to make a full 64-bit transition, and it would be good to be ready.

That gives a rough idea of the work that's currently underway. Looking ahead, there are a whole long list of packages that need adding or updating (such is a maintainer's life). The one significant place I have been falling behind is that I haven't updated gcc, so that needs work. And, of course, I'm trying to get SPARC into some sort of reasonable shape. But, overall, Tribblix is now pretty solid and a bit more polish and attention to detail would benefit it greatly.

Wednesday, June 07, 2017

Installing Tribblix on Vultr using iPXE

One of the new features in Tribblix 0m20 is that booting and installing using iPXE now works.

Here's an example of using this functionality to install a server running Tribblix in the Vultr cloud. A similar mechanism ought to work for any other provider that allows iPXE boot.

I'm assuming you have signed up and logged in, then go to deploy a server.

First choose where you want to deploy the server. I'm in the UK, so London is a good choice.


Then the critical bit, selecting the Server Type. The bit you want here is in a slightly confusing location, under the "Upload ISO" tab. But then select the "iPXE" radio button and put in the value http://pkgs.tribblix.org/m20/ipxe.txt


The other key option is Server Size. As with many providers, there's a simple scale. For testing, an instance with 1G of memory is more than adequate.


The deploy it. After a few seconds of installing you can then click the link to manage the server, and then view the console, which uses VNC.

If you're reasonably quick you get to see the initial iPXE screen, and can see it downloading the images:


What you can see here is that it's downloaded the original ipxe script we specified. This looks like:

#!ipxe
dhcp
kernel /m20/platform/i86pc/kernel/amd64/unix
initrd /m20/platform/i86pc/boot_archive
boot
 
Which just says to set up the network using dhcp (this might have already been done, but if you're booting off an ipxe iso it may not have been, so we do it anyway), then download the kernel and the boot archive, then boot from what you've just downloaded.

The kernel and the boot archive are on the iso, I've just unpacked them on the server (so the URL given above for the ipxe script will be reasonably permanent for anybody to use). The only slight tweak I've had to make is that the original boot archive is actually gzip compressed and iPXE can't handle that, so it's been uncompressed. The boot archive also now contains the /usr file system as well, rather than it being split off as before. While I'm sure you could mangle the system to download it and sort things out, it's so much easier to put it inside the boot archive.

Then you get into the normal installer, so log in as jack, su to root, and see what disk(s) are available using the new diskinfo tool. Then you can install Tribblix to that disk:



Don't bother adding additional overlays at this point. It won't work - and you'll get an error about not being able to install overlays (you'll get the error anyway because the installer always tries to add some packages that aren't needed in the live environment). This will be fixed in a future update, but it's relatively harmless.

The other thing you should do before the installation is to change the passwords for root and jack. If you change them before running the installer than the change will propagate to the installed system (because all it's doing is a copy). You really don't want the system to boot up wide open to the internet with the default (and well known) passwords.

Once the (pretty quick) install finishes, it'll look like this:


That's just like a normal install, other than the missing overlays. Then just reboot and you'll soon see the new loader, followed by the system booting.

Due to the missing overlays, you'll get an error about the intrd service failing. You'll have to log in (ssh will work at this point) and then add at least the base overlay:

zap install-overlay base

Plus whatever other overlays you might want. Then you can clear the intrd service and you're good to go.

Friday, June 02, 2017

Tribblix memory requirements

Compared to the other illumos distributions, Tribblix has lower memory requirements.

I'm not talking about crazy stunts like running in 48M; here I'm talking about running a fully fledged system.

I've been doing a bit of testing of the upcoming release, which includes running the install under a range of configurations. The test here is to boot the ISO image in VirtualBox with a range of memory sizes and then install the kitchen sink.
  • The live image won't boot at all on a 256M system
  • The live image will boot on a 512M system, but installing to zfs will fail
  • However, installing to ufs works on a 512M system
  • With 768M, installation to zfs is rather slow
  • With 1G or more, you're fine
The upcoming release is going to be built slightly differently, in that it's no longer a split-off /usr configuration. (I discussed how that worked and those strange zlib files some time ago.) The latest OmniOS is a single image; SmartOS likewise. It's just so much easier to construct, and far more reliable.

That change explains the 256M failure - the ramdisk is about 300M, so it simply won't fit. It's likely to have an impact on the 512M case too - in the old scenario you only paged in the bits of the /usr filesystem as and if you needed them, now it's locked into memory.

On a limited memory system there's a way to make things a bit easier. Simply install the base (no additional overlays) from the installer, then add the rest of the overlays and packages later. The point here is that running from disk doesn't lock up anywhere near as much memory as the full OS being resident in RAM does. And some of the packages in the kitchen sink are rather large, which causes problems.

Once you've got Tribblix installed, how well does it cope? Surprisingly well, to be honest. The Xfce desktop runs quite well in either 512M or 768M of memory. I can run firefox on the 768M system without too many problems (given the way it consumes memory, probably not for a long intensive browsing session), while firefox on a 512M system does run, but it's clearly starting to grind. Java applications work, some smaller ones at least. You need to be realistic in your expectations, but the point is that smaller systems do work.

The most limited systems would tend to be older, possibly 32-bit hardware. I could build a 32-bit only image which would be quite a bit smaller - maybe only two-thirds the size. (And if you really wanted to you could get it even smaller - but then you're in the realms of building custom images using mvi or the like.)

However, the aim of keeping Tribblix viable on smallish systems isn't just to allow the use of old hardware, beneficial though that is. If you're running a service on a cloud or hosting provider then being able to use a 1G server instead of a 2G server will halve your costs, and that's a very good thing to be able to do.

Monday, May 29, 2017

Tribblix SPARC progress

Tribblix is one of the relatively few illumos distributions that runs on both SPARC and x86 hardware.

There are valid reasons for the lack of SPARC support in other distributions. For those backed by commercial entities, it makes no sense to support SPARC as they don't have paying customers to foot the bill. Which leaves SPARC support firmly in the hobbyist realm.

Even in Tribblix, SPARC support has lagged the x86 version somewhat. Again, for entirely predictable reasons. While I do have SPARC hardware, it's relatively slow, noisy, power hungry, and heat-producing compared to my regular x86 boxes. And my day to day use is my x86 workstation, so that drives a lot of the desktop work.

But SPARC development of Tribblix hasn't stopped. Far from it, it's just naturally slower.

The current download ISO image at this time is still Milestone 16. Just to clarify the versioning here - that means it was built from exactly the same illumos commit as the corresponding x86 release. Because it took a little longer to get ready, the userland packages (such as they were) tended to be a bit newer.

There have been 3 more Tribblix release on x86 since then. Over the winter (when it was cold and the heat output from the T5140 I use as a build server was a good thing) I tried building updated illumos versions. The T5140 I'm using to do the builds is running a cobbled-together frankendistro of bits of Tribblix, bits of OpenSXCE, some random bits from other people working on SPARC, and a whole lot of elbow grease. I managed to build illumos at the m17 and m18 release points, but m19 was a step too far (some of the native stuff assumes that the host OS isn't terribly antiquated). What this means is that I need to replace that by a current system, and get a properly self-hosting illumos build.

That modernizes the underlying illumos components a bit. What about the rest of the system? The primary effort there was to replace the old core components that had been been borrowed from OpenSXCE while bootstrapping the distribution in the first place with native packages (and that are then up to date and match the x86 build). Some of the components here are pretty crucial - zlib and libxml2, for instance. At one point I messed up libxml2 slightly - not enough to kill SMF (which would be a big worry) but enough to stop zones working (which, apart from indicating that I had broken it, also left me without an imortant test mechanism). Rebuild everything enough times and the problem eventually cleared.

I also had a go at getting my SunBlade 1500 workstation working. It's not terribly quick, but it's quiet enough and sufficiently low power that I can have it running without negatively impacting the home office. That was a bit of a struggle, the bge network driver currently in illumos doesn't work - I assume I'm seeing bug 7746 here, but the solution - to use an older version of the driver - works well enough. With that box available I not only have more testing available but also a lightweight machine that I can use to keep the package backlog under control.

Graphics on SPARC is an interesting problem. OK, so I don't expect this to be a priority, but it would be nice to have something that worked. The first problem I found (a while ago) was that some of the binary graphics drivers wouldn't work at all. For example, the m64 driver (which is what might drive the graphics in my SunBlade 2000) uses hat_getkpfnum which was removed from illumos courtesy of bug 536. Graphics drivers that load often simply don't work, and getting an X server to start is a bit of a nightmare. After far too much manual fiddling I did manage to get a twm desktop running on the aforementioned SunBlade 1500, but don't expect native graphics support to improve any time soon.

Applications is another matter, there's no reason you couldn't run at least some applications on a SPARC system and display them back on your desktop machine. After all, X11 is a network display protocol (despite all the effort to eradicate that and turn it into a local-only display protocol). Or run a VNC server and access that remotely. So I've started (but not finished) building up the components for useful applications.

I haven't yet got an ISO image. That's likely to be a while, but if you have an existing SPARC system running Tribblix m16 then the upgrade to m18 ought to work. Although I would recommend a couple of changes to the procedure if you're going to try this:
  • Refresh and update everything: 'zap refresh ; zap update-overlay -a'
  • Download the current upgrade script from github and run that script in place of 'zap upgrade'
  • After booting into the newly updated BE, refresh and update everything again, just to make sure you're up to date

Saturday, April 29, 2017

OmniTribblix

In Tribblix, it's a basic principle that I ship upstream software unmodified. I don't impose my own views on installation layout, nor do I customize it. Generally, I apply patches only to make stuff compile.

This means that what you see in Tribblix is exactly what the upstream author intended, and not some distro-specific bastardization of it.

It also makes my life easier, I don't have to maintain patches, and updating software is much easier if it's unmodified.

In particular, I use an absolutely vanilla illumos-gate. (For a long time it differed only in that I had the fix for 5188 applied, relevant because Tribblix actually uses SVR4 packaging, but now that's integrated I don't even need to do that.)

Again, this makes my life easier. (When you're maintaining a distro on your own in your spare time, making decisions that simplify your job is essential.)

But it also has another benefit: because I have no "special" features that I've added, I'm not tied to one particular version or variant or commit of illumos. Any version of illumos-gate will do just fine. When it comes time to make a release, I just clone the gate, build, and go.

What I could do, then, is build an instance of Tribblix atop some other fork of the gate. For example, illumos-omnios.

I did just that, built the gate (it needed a couple of changes to Makefiles because of the way that perl and snmp are slightly different in OmniOS than it is in Tribblix), created packages, built an ISO, booted and installed it in VirtualBox.

As expected, it just works.

But just demonstrating that it works isn't really the reason I wanted to do this. What I'm really after is the LX brand, which has been integrated into current OmniOS.

Installing an LX zone requires a Linux image. The original (Joyent) work was for their own deployment mechanism, using ZFS images. As soon as it was available in OmniOS the first thing I did was use tarballs, which OmniOS now supports. The easiest way to create a Linux image is to create a Docker container the way you like it, and then export it to a tarball. I did that for Alpine and installed a zone based on that.

Then you can do very simple things like:

# zlogin lx1 /bin/uname -a 
Linux lx1 4.4 BrandZ virtual linux x86_64 Linux

It's an attractive idea to simply use this as the base for the next Tribblix release. However, that requires illumos-omnios to be supported in the long term, which is currently at risk.

Wednesday, April 12, 2017

Noisy Tribblix

I've had a couple of Tribblix users ask me why audio doesn't work.

This was something I had noticed myself, and the reason was not that audio was in some way broken, but that the permissions on the audio devices were wrong - owned and only writeable by root.

Now I only wanted to actually get any audio out on fairly rare occasions, so a quick chown wasn't that much of an imposition. But it obviously needed fixing properly.

My assumption here is that most desktop users will be logging in through the SLiM login manager. So all I need to do is fix the permissions just before it calls setuid() to the logged in user. And then reset them back once the user is done.

Now, I could have made up a bunch of chowns myself, or written a helper. There's actually code in SLiM to call ConsoleKit - but I don't have ConsoleKit, and don't really see the need to maintain a port of it just for this.

But illumos already has the capability to do this, and the normal login mechanisms use it. There's code in libdevinfo that sets the permissions according to the rules laid out in the /etc/logindevperm file. So the code is really just a call to di_devperm_login() and di_devperm_logout(), and all is well.

This also fixed another irritating bug - I can now eject memory sticks as myself, without needing to be root.

The next thing that happens, of course, is that it doesn't take very long to realise that Twitter has a lot of videos that play automatically. So I'm sitting there and I can hear either the internal loudspeaker or my headphones warbling away.

So the next thing I need is a way to shut the thing up. Historically, I used the old CDE sdtaudiocontrol, which was pretty good. (In general, I detested CDE as a desktop, the mailer and calendar were decent enough for their time, and the audio control was the only other thing I used much.) I use Xfce as my desktop, it used to have xfce4-mixer but that's now unmaintained and deprecated (and I removed that as part of the migration from gstreamer-0.10 to gstreamer1). Which pretty much leaves the command line audio utilities in illumos, specifically audioctl. I've added the package so users who update will automatically get that as well.

The command

audioctl set-control volume 0

silences things, while

audioctl set-control volume 75

puts the volume back to normal. I've created aliases mute and unmute for those. A more sophisticated approach would be to save the volume and restore it afterwards, but this is enough for now.

Sunday, March 12, 2017

How old are illumos man pages?

I've recently been looking at improving the state of the illumos man pages.

One thing you'll notice is that the date on some of the manpages is old - really old, some of them are dated 1990. That presumably means that they haven't been modified in any meaningful sense for a quarter of a century.

(By date here I mean the date displayed by the man command. Which isn't necessarily the date somebody last touched it, but should correspond to the last meaningful change.)

The distribution of dates looks like this:



As you can see, the dates go all the way back to 1990. There's not just the odd one or two either, there are a decent number of man pages that comes from the 1990s.

There are some obvious features in the chart above.

There's a noticeable spike in 1996, which of course significantly predates illumos or OpenSolaris. It's not entirely obvious why there should be a spike, but 100 of those man pages are related to libcurses.

I suspect the dip in 2005 is a result of the launch of Solaris 10, when everyone had a bit of a breather before development kicked into full gear again.

Then there's a drop in 2010. In fact, there's just a single page for 2010. That's when the OpenSolaris project closed, so there was little work done at that point. Also, our man pages were only integrated into illumos-gate in early 2011 (prior to which they were kept separate), and it's taken a while for man page updates to pick up again.

Of course, one reason the man pages are so old is that the software they're documenting is old. That's not necessarily a bad thing, if it ain't broke don't fix it as they say, but there is a certain amount of total rubbish that we ought to clear out.

Thursday, February 02, 2017

Creating a Tribblix package repository

I've previously described how Tribblix packages are built.

The output of that process should be a zap file, which is an SVR4 package in filesystem format, zipped up. The file naming convention is

PKG_NAME.VERSION.zap

Where PKG_NAME is the SVR4 name of the package (you can define aliases to be more user-friendly), VERSION is obvious - but has to match the installed version as shown, for example, by 'pkginfo -x'.

You can install those packages directly. There's a little helper /usr/lib/zap/instzap that will automate the process of unpacking the zap file and running pkgadd on it, or you can do it by hand.

You don't have to use my tooling. If you've got a scheme for building SVR4 packages already, then you could simply convert those.

However, what you would really want to do is stick the packages in a repository somewhere, so they can be accessed using the normal zap commands.

In Tribblix, a repository is just a web-accessible location that contains the zap files and minimal metadata. The metadata is the catalog and a list of aliases.

The aliases file contains lines with two fields separated by a vertical pipe "|". The first field is the friendly alias, the second the real name of the package. If you want multiple aliases, just add multiple lines to the file.

The catalog file contains lines with 5 fields separated by a vertical pipe "|".The first field is the package name, the second the current version, the third a space-separated list of packages this package depends on, the fourth the size of the file in bytes, the fifth is an MD5 checksum of the file. There's a trailing "|" in this case to terminate the line. The size and checksum are used to verify the download was successful and didn't corrupt the file. If you want to be sure that it's actually the package it claims to be then packages can also be signed.

Here's an example line in the catalog:
TRIBabook|0.5.6.0|TRIBreadline|66923|d98f77cfb3e92ee6495e3902cc46486f|
So the TRIBabook package has a current version of 0.5.6.0, depends on the TRIBreadline package, and has the given size and checksum. It's in the file called TRIBabook.0.5.6.0.zap.

The build repo has scripts that create the catalog and aliases files, which I use for convenience. They do make some assumptions about my package build pipeline, so it might be easier to manage the files by hand.

So, having got a nice place on the web that has your packages and a catalog and some aliases, how does Tribblix know to use it?

Assuming your repository is called myrepo, you need to add a file /etc/zap/repositories/myrepo.repo, containing something like the following (which is the configuration for the main tribblix repo in the milestone 18 release)

NAME=tribblix
DESC=Tribblix packages
URL=http://pkgs.tribblix.org/tribblix-m18/
SIGNED=tribblix

If you aren't signing packages (and how to add the keys to the client is an exercise for the reader) then omit the SIGNED line. The NAME in this case would be myrepo, the DESC is whatever you make it, and the URL points at the directory containing the files.

That's almost it. The other thing you need to do is add the repository to the list of repos that zap will search, which is held in the /etc/zap/repo.list file. By default, this contains

100 tribblix
200 illumos
300 oi

That's a simple list, and the number is a priority, lower numbers have higher priority - they're searched first.

(What the priority scheme means is that if you have a package with the same name in multiple repos, the one in the highest priority repo is the one that will be used. For example, the ssh packages used to come from illumos. As we've moved to openssh, all I had to do was put the new openssh packages into the tribblix repo with the same package names they had before and they were installed instead. Of course, you have to be careful that the replacement package delivers the correct functionality, especially if it's delivering shared libraries.)

If your package names are unique to you (for instance, if you name them MYPKGfoo rather than TRIBfoo) then the priority doesn't matter. If you're deliberately trying to override some of the built-in packages, then your repo has to be the highest priority one so it gets searched first.

If you then utter 'zap refresh' it should go and retrieve the catalog and aliases files and then be set up to use them.

And yes, it would be nice if zap had the facility to manage repos for you - that's planned, I just need to implement it.

Monday, January 16, 2017

Package versions in Tribblix

All packages in Tribblix are versioned. If you look at the pages on the package repository you can see the current version of each package in the repo. On an installed system the pkginfo -x command will give you the package description and version.

As Tribblix is created from different sources, the meaning of the package version can vary.

For illumos packages, the version string matches the Tribblix release. For example, "0.18.0" indicate the Milestone 18 (0m18) release.

For packages inherited from another distro, the version matches in some way the distro release I got the packages from. For example, the OpenIndiana packages were (at this time) cut from the oi151a9 release, and have a version string "0.9o".

For packages I build directly from source, the version string is usually the upstream version, with a build number appended. Initially the build number is 0, then increments. If the upstream version is updated, the build number goes back to zero. So it's reasonably obvious what version of a package is installed.

For example, abiword is version 2.8.6 so the first time it was built the package version was 2.8.6.0. Over time the package has needed to be rebuilt, so it's now up to version 2.8.6.4.

The sharp-eyed will notice that the illumos packages have a build number in them. This hasn't yet been used, it's there just in case.

The scheme is reasonably flexible. For example, OpenSSL has letters in its releases - like 1.0.2j - which I could keep verbatim, but in practice I convert the letter to a numeric sub-version, hence 1.0.2.10.

There are some packages for which I originally forgot to add the build number. That OpenSSL package is an example, but there are others. I've tended not to correct those as it disturbs the flow, I will if it ever becomes convenient.

Some releases have a date, this is just converted to numerical form.

One thing that should be obvious is that the scheme doesn't guarantee that package versions are numeric. They're just strings; it just happens that most packages have version numbers that are numeric or can easily be represented as such.

Also, package versions don't necessarily increase, there is no sense of ordering built into versioning. For example (this does happen) there's an upstream version 1.2, which leads to package versions 1.2.0, 1.2.1, 1.2.2, etc. Then there's an upstream 1.2.1, which is packaged as version 1.2.1.0, which is lower than 1.2.2. And sometimes upstreams try a major version bump, then backtrack.

However, package management in Tribblix ascribes no meaning to the version numbers. It's only test for currency is this - does the version installed match the version in the repository catalog? If they're the same, then you're up to date. If not, then apply the version from the repo.

This then makes it easy to roll back errant packages. All I have to do is put the old version back in the catalog. Anyone who has applied the broken version will get a version mismatch and the older version will get installed whenever they update.

(This simplistic approach only works if I haven't built anything against the newer version of the package I want to roll back. But then, all I have to do is roll all those dependent packages back as well.)

Life's a little more complicated if you might want multiple versions of an application installed. In that case you have to have different packages. For example, I have separate packages for Python 2.7 and 3.6, and there might be 2 corresponding packages for any modules. I used to use multiple packages more extensively, sometimes even for minor version updates, but tend to avoid that now when I can.