We recommend building all software inside local zones, so you can set up a clean development environment and not "pollute" your main operating environment with your test packages which are not guaranteed to always work as expected
It may make sense to have several build zones, say one for each major project, or using a "pristine" one for testing of your packaged builds when you think your project is nearing the finish line. In these cases it is recommended to set up one build zone properly, take its datasets' ZFS snapshot, and then clone it and reconfigure the clones as new local zones – this would guarantee that the zones are identical and save you some disk space and internet traffic.
In particular, local zones can be used along with multiple cloned workspaces (see Working on several bugs at once) for preconfigured compilation of the same project code-base with different supported compiler suites (SunStudio, GCC, whatever comes next?)
The Distribution Constructor (the software that produces the final ISO Image) does not work inside a local zone.
OpenSolaris' (and its descendants') local zones can have at least two approaches to networking: a shared IP stack and an exclusive IP stack. There are pros and cons to each of them – they are tools for different jobs somewhat.
Low-level system networking components, such as the
ipfilter firewall and the kernel IP routing tables attach to an "IP stack", and are thus either unique to a zone or shared by all zones with the one shared stack (usually including the global zone). There are also some other nuances, such as that the zones with the shared stack can communicate over IP directly, regardless of their subnetting and, to some extent, default firewall packet filtering (that has to be specially configured), while exclusive-IP zones with addresses in different subnets have to communicate over external routers and are subject to common firewall filtering.
It is the global zone however that defines which physical networks and VLANs the local zone has access to, and hands down the predefined networking interfaces (the local zone can not use or create other interfaces). Also, while shared networking allows to configure and attach (or detach) network interfaces from the global zone to the local zone "on the fly", changes in exclusive networking require reboot of the zone to propagate device delegation.
A local zone with an exclusive IP stack can have most or all the benefits of dedicated hardware networking, including a firewall, access to promiscuous sniffing, routing, configuration of its own IP address (including use of DHCP and static network addressing), etc. This requires a fully dedicated NIC however, which was a limitation in early OpenSolaris, until the CrossBow project came along bringing the ability to create VNICs, which are virtual adapters with their own MAC addresses, that operate as if the VNIC was plugged directly into your local LAN in a particular (or default) VLAN.
Note also that the local zones with an exclusive IP stack are not subject to the host's shared-stack firewall, and so are by default open to the external network with all possible security holes. You're encouraged to revise and minimize the potential breach exposure by disabling unneeded services as well as by configuring at least basic firewall rules in the zone (not covered in this manual).
It is also possible to use "shared networking" where the local zone's interface would be an alias of a NIC available to the global zone (a guide mentioning this is available here: HOW-TO Setup referential build zone for OpenIndiana Addon Consolidations) and the routing tables would be a subset of those available to the global zone.
If your build machine is a VM on a platform which requires the hypervisor host to "know" the VM's MAC addresses – such as VirtualBox running on Solaris with the "bridged networking" mode – your build zone must use a NIC defined by "VM hardware". You still have two options though:
If you plan to use VirtualBox bridged networking on a VM under Solaris-related OSes, see the VirtualBox User Guide for more details on this setup – in particular, the host Solaris machine is also encouraged to use VNICs with explicitly defined MAC addresses in order to attach the VM NICs to them.
The example below creates the zone root under
rpool ZFS pool.
If your machine has other pools, perhaps bigger and/or more performant i.e. by using L2ARC caches, you may want to use a different pool for zone data (by further delegating whole datasets or
lofs-mounting individual paths – such as your building workspaces), or for the whole zone roots altogether. (Note that it may be officially unsupported to hold local zone roots separately from OS roots in some OpenSolaris descendants).
In this example we will host all zones for building inside a dedicated ZFS dataset mounted at
/zones/build (each local zone will have its datasets hanging under this point). On one hand this allows you to group zones with different tasks for administrative visibility, on the other – the common container dataset can be used to inherit specific ZFS properties to the local zones. The container has some explicitly configured ZFS properties, but holds no data on its own (other than automatically created directories as mountpoints for child datasets). As an alternative to loopback-mounting, the NFS client in a local zone can use the global zone's NFS server, but that is likely to be slow (especially if
sync is enabled without a fast ZIL).
Here, we are disabling
sync to speed up builds. Note that setting
sync=disabled may result in data loss in a power loss/system crash scenario, so only enable it for your build environment if you don't mind losing data (from recent writes just before the crash) – and there are few cases where it is not unrecommended to enable this feature:
In the example above, we create the common container for local zones in
rpool/zones, set its mountpoint to
/zones in the host's common filesystem tree, and enable generic lightweight compression (
lzjb by default). Afterwards we create the container for build zones with specific ZFS properties which are not generally recommended for other use-cases. The
rpool/zones/build container inherits other ZFS properties from its ancestors (up to
rpool), including compression and base mountpoint from its parent – so it will be automatically mounted as
/zones/build unless you override that explicitly at some point of the hierarchy.
The united ZFS properties will be inherited down the road to actual local zone datasets hosted under this container dataset.
This guide below assumes you are on a LAN with DHCP, so we will create a CrossBow VNIC on your primary network interface (assumed to be
e1000g0 here, adapt as necessary). You can skip this step if the local zone would use shared networking or a dedicated (VM-)hardware NIC.
If your primary interface is not
e1000g0, please substitute for the one that is in use. You can normally find this via "
dladm show-phys" or "
If needed, you can also specify binding of the VNIC to a particular VLAN of your external network with
Here the global zone predefines and fixes the IP addressing used by the local zone:
NOTE: It may be possible to save on internet traffic by installing from a locally configured mirror of OpenIndiana IPS repositories set as defaults for your system. This is not explored below, but you're welcome to try that and write a howto chapter here
This step will install the zone by downloading packages from the internet (about 150Mb in this example):
Zone creation as described above can fail for a number of reasons, sometimes obscure (i.e. you followed different steps, or your current OS deviates from one used in making this how-to).
One common problem is inability to create a dataset with the requested
zonepath with error reports similar to this:
In this case an empty dataset with needed mountpoint has to be created manually.
This step can lead to the second common problem: zone root's filesystem access rights must only allow
These two problems can be amended with the following commands (after creation of zones container dataset as detailed above):
Check the commands' output for errors, such as inability to mount the created dataset (before the
This should automatically create and populate the local zone's dataset hierarchy:
You can review the zone configuration by supported commands, as well as by accessing its XML-file descriptor (unsupported although convenient – subject to change in future versions):
You might want to clone another zone later with these basic packages, or roll back to current conditions:
This step answers questions you would otherwise have to answer manually via a console-based wizard upon the first boot. Now the wizard would get its answers from the file (it may still ask questions not covered in the file, i.e. if the future versions of the wizard define new questions).
Mount the zone's ZFS dataset so we can access it by running:
Now create the
Remark: the encrypted root password shown here is:
Note also that in the example above, your zone will try to receive networking settings via DHCP. It is possible to set static IP addressing for an "
ip-type=exclusive" local zone by using traditional Solaris methods with files (relative to zone root):
/etc/netmasks and so on; in this case you might want to disable
dhcp-client in the zone. You may also want to make sure that in the zone's
/etc/hosts file the static IP address would be associated with the zone's network name (short hostname and full FQDN), and that name should be used in
/etc/hostname.vnic0 instead of an IP address directly.
You can also configure individual
ipfilter firewall in the "
ip-type=exclusive" zone (GZ's firewall doesn't apply to non-shared LZ networking).
Now create the
Remark: the encrypted root password shown here is:
After boot you may also want to make sure that in the zone's
/etc/hosts file the static IP address (set by the global zone) would be associated with the zone's network name (short hostname and full FQDN).
You will now want to attach to the zone's console and watch it boot, and answer any questions if prompted:
You can detach from the console by issuing "
~." without quotes (or "
~~." over chained Unix
Once the zone has booted, you can copy the DNS resolution settings from GZ into the local zone (if networking is the same, and if your
sysidcfg or DHCP setups did not take care of that successfully):
You can now
zlogin into the local zone with:
Try pinging some hosts on the internet. Remember to update the root password.
Note that the zone's internetworking is possibly subject to external firewalls on your LAN, and/or access to a proxy server, etc.
You might want to clone another zone from these presets along with working networking:
You can follow the illumos and OpenIndiana subproject guides on setting up the recommended environments (compilers, source code repositories, etc.):
You might want to delegate access to common source-code workspaces (i.e. by
lofs-mounting them from GZ into LZs), to your private package depots, etc.
You might want to clone another zone from these presets now, so as to instantly start working in the clone:
Creating the zone's clones now would involve cloning of the prepared zoneroot's snapshot, copying of the zone configuration (with
zoneadm, or by copying and modifying zone-description XML files and modifying the
index file in
/etc/zones for the time being), and possibly updating the static networking configuration inside the zone root – such as
/etc/nodename files at least (also maybe
/etc/hostname.vnicN for exclusive VNICs and
/etc/motd to describe this zone's purpose).
Note that there are other ways to clone zones, and unlike the hacks in previous paragraph they are "supported", but those would usually clone a zone's current state instead of using a "golden image" as you can do with the snapshots above. Alternately, you can hold the unused preconfigured "dummy" zone as a golden image, and properly clone it with the supported system tools and methods (making the first clone now for your actual development work).
for preconfigured compilation of the same project code-base with different supported compiler suites,