Contributing to illumos¶
Committing code to illumos can seem very daunting. To get started, you might
consider watching Ryan Zezeski going through fixing a
bug in ZFS in this "illumos Day" talk, which includes submitting a patch for
The Request To Integrate (RTI) process is also documented in the illumos Developer's Guide.
Finding An Area To Contribute To¶
If you're not sure what you want to work on, you can start by looking at our list of bite-size bugs which should be easy for newcomers to pick up. You can also look at the entire list of issues and see if any with status "New" fit your skill set.
If you are adding a new feature or addressing a problem not currently on our list of issues, please create a new issue describing it. You need to create an account in the bug tracker in order to be able to file issues.
Before you start working on the code, it is advisable to first ask on either the developer mailing list or the IRC channel for advice on the particular area you're interested in. For larger bodies of work, it is worth writing an illumos Project Discussion (IPD) document that describes your plan and seek feedback from advocates and others.
Writing The Code¶
Our guide to building illumos covers setting up a build environment and getting the source. To make sure you have a functional build environment, it helps to try at least one build without modifying anything. This also helps you to familiarize yourself with the build system and source tree layout.
An unparalleled resource for understanding the operating system is the book, Solaris Internals: Solaris 10 and OpenSolaris Kernel Architecture.
We have an OpenGrok source browser and search engine which is very useful for development.
We encourage you to "commit early, commit often" as you work, using your personal clone of illumos-gate.git.
The illumos gate is a core technology leveraged by several commercial vendors that rely on it to be a stable, well-tested platform. As such, changes to illumos require qualified code review from community peers. A reasonable effort must also be made to test each change, to ensure that the change is stable enough for others to rely upon.
Your RTI advocate is the ultimate arbiter of whether your testing is sufficient.
The scope of required testing depends on the nature of the change and a discussion with your advocate, but in general advocates adopt a "shrink to fit" mentality. Here are some examples of reasonable testing based on the collective experience of the current RTI advocates. Your advocate may always ask for more testing depending on the nature of the change.
Make sure you test the results of the illumos-gate build with only your changesets applied. To do this, you will need to build the illumos code. It is generally insufficient to test the build results of a downstream fork that contains other non-trivial changes as there may be implicit dependencies that break other consumers of the core illumos gate.
If your code has been running in production as part of a downstream distribution of illumos, or has already gone through other testing, please include that information. Your advocate will determine if any additional testing is required for integration.
Changing a single program¶
If your change affects only a single binary (e.g.,
/usr/bin/ls), it is likely
sufficient to copy the binary to a system running a distribution with a
reasonable similarity to vanilla illumos-gate and running your tests there.
These tests should cover enough functionality to demonstrate the correctness of
your changes, and could involve stress tests in addition to functional tests
depending on the nature of the change. This assumes that the binary has a
trivial set of library dependencies that are reasonably standard across
Changing the implementation of a library¶
If your change affects the implementation of a library, but does not change any externally visible interfaces, you should run tests against a reasonable subset of programs that exercise the code in question. As with a single binary, it is likely sufficient to copy this library to a system running a distribution which has identical interfaces to the vanilla illumos-gate.
Some core libraries (most notably
libc) are challenging to replace on a live
system, so you will likely need to bootstrap a complete new image or boot
environment of your distribution in order to be able to test your changes. How
to do this depends on your distribution; e.g., OpenIndiana or OmniOS can use
onu to update to new packages from a nightly build.
Changing the implementation of a driver or adding a new driver¶
If your change affects a single driver, you may or may not be able to get away with copying that driver to a system running an alternate distribution. On one hand, many of the kernel interfaces (e.g., the DDI) are likely identical, but if the core implementation of that distribution has diverged significantly from illumos-gate, you may be asked to bootstrap a complete environment from your vanilla illumos build. Driver tests should demonstrate a reasonable amount of functional and stress testing on appropriate hardware.
Changing the core kernel or multiple interconnected binaries¶
If you are making a change to the core kernel (e.g.,
have a series of changes across a number of libraries, commands, or kernel
components, they need to be tested as a cohesive whole. You must bootstrap a
usable environment from your illumos-gate build that is sufficient to
demonstrate the correctness of your changes.
Exactly what "bootstrapping" means is dependent on the distribution, but the goal is to run the entirety of the your illumos-gate build (not a combination of downstream and upstream illumos-gate bits) in a testable environment. Not all distros are capable of bootstrapping a vanilla illumos-gate build; check the distribution list and ask on the developer list if you are unsure of how to do this.
For example, on OpenIndiana or OmniOS, you can use
onu to install the
packages produced by a nightly build. On a SmartOS system, you can build
a wholly new "platform image" and boot a system using it.
The scope of testing here will likely be more substantial, and should focus on stability of the system as well as the functionality of the changed component(s).
As always, contact your advocate (or the advocates list if you don't yet have someone in mind) if you have more questions about what constitutes sufficient testing for a particular change. You are welcome to ask for help in creating a test plan well in advance of completing the work on you change, so that the advocates can help set expectations and provide feedback throughout the process.
At least one person (other than you, of course) should review your changes. These reviewers should have demonstrated expertise in the area in question or are trusted as knowledgeable by the illumos community at large. If you work for a commercial entity, it is best to include reviewers from outside your company, but this is not required.
Your RTI advocate is the ultimate arbiter of whether your review is sufficient.
If you don't know how to find adequate review, you may post the changes to the developer mailing list to ask for feedback, or ask the RTI advocates if there is anyone in particular they'd like to see review the code.
When seeking review, it is generally best to upload to some web-based review or source hosting system, rather than attach a patch to your list e-mail. The illumos project provides a Gerrit instance which you can use with your illumos bug tracker account credentials. For more information about using Gerrit, see our documentation on Gerrit. Remember to mention the issue ID in the email -- if one does not yet exist, please create one!
Ensuring adequate review¶
It is up to you as a contributor to gather reasonable review to satisfy your RTI advocate that the code is correct and of high quality. Exactly what this means will vary based on the nature and scope of the change, but here are some guidelines that will give the advocates confidence in your changes. As always, trust your advocate and apply "shrink to fit".
- There should be at least one code reviewer (focused on the implementation details) that is different from the advocate (focused on ensuring overall quality of contributions).
- For non-trivial changes to major technology areas (ZFS, DTrace, etc), at least one of the reviewers should be regarded as a subject matter expert by the community.
- Changes with significant impact should participate in a public review when appropriate, or otherwise ensure that alternate opinions are represented.
- When there are disagreements, every attempt should be made to adequately address reviewer's comments. In the event they cannot, it is acceptable to submit an RTI with known dissent. Such disagreements must be noted in the RTI submission with an explanation as to why the review comments were not addressed.
When in doubt, more review is generally better. If anything is unclear, or you're stuck on something, you can always ask the advocates for guidance.
Submitting A Patch¶
When you have adequate code review and test results for your proposed change, you prepare a "request to integrate" (RTI).
First, in the issue(s) you created in the bug tracker, make sure you have:
- Any analysis that lead to the fix; e.g., if you used DTrace or MDB to determine what went wrong it'd be good to include scripts and/or output from your debugging session
- Your testing plan and results from your testing. This could be simple
for simple changes; e.g., "I ran the updated
lsbinary and this is the new output". It could also be more detailed for a complicated bug. Providing the testing notes in the issue allows the advocate to easily see what you did to test, and allows future engineers to better understand how to test their changes in the same area of the system.
You should have a full clean build of the gate with your changes applied. This
currently means a successful build with GCC7 as the primary compiler, and GCC4
and smatch as shadow compilers, with no warnings or errors in the resulting
mail_msg. A clean SPARC build is optional, and only requires GCC4 if you do
happen to do one.
Your RTI e-mail should include:
- The link to the illumos issue(s) you're fixing, e.g., https://illumos.org/issues/10052
- A link to the changes that were reviewed; e.g., a link to your Review Board review
- The full "change set description" (i.e.,
git whatchanged -v origin/master..) including:
- Issue number(s) and description(s)
Reviewed by: First Last <firstname.lastname@example.org>lines
- List of files affected
- Output of
git pbchk(run under
- An attached clean
mail_msgfrom a full nightly build
- Information about how the changes were tested (it's sufficient to mention that the testing notes appear in the bug tracker)
- Your changes attached as as a patch as per
Here is an example change description:
123 Description of the issue in our tracker Reviewed by: Jack <email@example.com> Reviewed by: Ohana Matsumae <firstname.lastname@example.org>
You can use
git commit --amend to edit the commit message.
An Advocate will need to judge whether your code review and testing are adequate for the scope of changes you propose. Note that the advocate's job is not necessarily to review your code, only to judge whether review and testing was adequate.
When you are ready, send an email to the illumos advocates mailing list. Though you are welcome to subscribe to the list, it is not required; you can just send mail to email@example.com. If you're interested in what past RTIs looked like, you can browse past messages.
A reasonably complete example issue is
show-ether" should pick one kstat snapshot and stick with
- Bug #10052 in the bug tracker includes both analysis and testing notes
- The RTI mailing list thread shows the sort of interaction you can expect
Your email needs to be less than 2MB in size. If your change is
large, compress the patch or make it available over HTTP. If your
mail_msg is large, something is wrong.
You should soon get a reply with additional feedback if needed, or a "thank you" for being part of the illumos developer community!
If the change is accepted, the advocate will take care of actually committing to master. Regular contributors may get commit rights: they follow the same system, but may push to master themselves after approval.