Development
All VyOS source code is hosted on GitHub under the VyOS organization which can be found here: https://github.com/vyos
Our code is split into several modules. VyOS is composed of multiple individual packages, some of them are forks of upstream packages and are periodically synced with upstream, so keeping the whole source under a single repository would be very inconvenient and slow. There is now an ongoing effort to consolidate all VyOS-specific framework/config packages into vyos-1x package, but the basic structure is going to stay the same, just with fewer and fewer packages while the base code is rewritten from Perl/BASH into Python using and XML based interface definition for the CLI.
The repository that contains all the ISO build scripts is: https://github.com/vyos/vyos-build
The README.md file will guide you to use the this top level repository.
Submit a Patch
Patches are always more then welcome. To have a clean and easy to maintain repository we have some guidelines when working with Git. A clean repository eases the automatic generation of a changelog file.
A good approach for writing commit messages is actually to have a look at the
file(s) history by invoking git log path/to/file.txt
.
Prepare patch/commit
In a big system, such as VyOS, that is comprised of multiple components, it’s impossible to keep track of all the changes and bugs/feature requests in one’s head. We use a bugtracker known as Phabricator for it (“issue tracker” would be a better term, but this one stuck).
The information is used in three ways:
Keep track of the progress (what we’ve already done in this branch and what we still need to do).
Prepare release notes for upcoming releases
Help future maintainers of VyOS (it could be you!) to find out why certain things have been changed in the codebase or why certain features have been added
To make this approach work, every change must be associated with a task number (prefixed with T) and a component. If there is no bug report/feature request for the changes you are going to make, you have to create a Phabricator task first. Once there is an entry in Phabricator, you should reference its id in your commit message, as shown below:
ddclient: T1030: auto create runtime directories
Jenkins: add current Git commit ID to build description
If there is no Phabricator reference in the commits of your pull request, we have to ask you to amend the commit message. Otherwise we will have to reject it.
Writing good commit messages
The format should be and is inspired by: https://git-scm.com/book/ch5-2.html It is also worth reading https://chris.beams.io/posts/git-commit/
A single, short, summary of the commit (recommended 50 characters or less, not exceeding 80 characters) containing a prefix of the changed component and the corresponding Phabricator reference e.g.
snmp: T1111:
orethernet: T2222:
- multiple components could be concatenated as insnmp: ethernet: T3333
In some contexts, the first line is treated as the subject of an email and the rest of the text as the body. The blank line separating the summary from the body is critical (unless you omit the body entirely); tools like rebase can get confused if you run the two together.
Followed by a message which describes all the details like:
What/why/how something has been changed, makes everyone’s life easier when working with git bisect
All text of the commit message should be wrapped at 72 characters if possible which makes reading commit logs easier with
git log
on a standard terminal (which happens to be 80x25)If applicable a reference to a previous commit should be made linking those commits nicely when browsing the history:
After commit abcd12ef ("snmp: this is a headline") a Python import statement is missing, throwing the following exception: ABCDEF
Always use the
-x
option to thegit cherry-pick
command when back or forward porting an individual commit. This automatically appends the line:(cherry picked from commit <ID>)
to the original authors commit message making it easier when bisecting problems.Every change set must be consistent (self containing)! Do not fix multiple bugs in a single commit. If you already worked on multiple fixes in the same file use git add –patch to only add the parts related to the one issue into your upcoming commit.
Limits:
We only accept bugfixes in packages other than https://github.com/vyos/vyos-1x as no new functionality should use the old style templates (
node.def
and Perl/BASH code. Use the new style XML/Python interface instead.
Please submit your patches using the well-known GitHub pull-request against our repositories found in the VyOS GitHub organisation at https://github.com/vyos
Determinine source package
Suppose you want to make a change in the webproxy script but yet you do not know
which of the many VyOS packages ship this file. You can determine the VyOS
package name in question by using Debian’s dpkg -S
command of your running
VyOS installation.
vyos@vyos:~ dpkg -S /opt/vyatta/sbin/vyatta-update-webproxy.pl
vyatta-webproxy: /opt/vyatta/sbin/vyatta-update-webproxy.pl
This means the file in question (/opt/vyatta/sbin/vyatta-update-webproxy.pl
)
is located in the vyatta-webproxy
package which can be found here:
https://github.com/vyos/vyatta-webproxy
Fork Repository and submit Patch
Forking the repository and submitting a GitHub pull-request is the preferred
way of submitting your changes to VyOS. You can fork any VyOS repository to your
very own GitHub account by just appending /fork
to any repository’s URL on
GitHub. To e.g. fork the vyos-1x
repository, open the following URL in your
favourite browser: https://github.com/vyos/vyos-1x/fork
You then can proceed with cloning your fork or add a new remote to your local repository:
Clone:
git clone https://github.com/<user>/vyos-1x.git
Fork:
git remote add myfork https://github.com/<user>/vyos-1x.git
In order to record you as the author of the fix please identify yourself to Git
by setting up your name and email. This can be done local for this one and only
repository git config
or globally using git config --global
.
git config --global user.name "J. Random Hacker"
git config --global user.email "[email protected]"
Make your changes and save them. Do the following for all changes files to record them in your created Git commit:
Add file to Git index using
git add myfile
, or for a whole directory:git add somedir/*
Commit the changes by calling
git commit
. Please use a meaningful commit headline (read above) and don’t forget to reference the Phabricator ID.Submit the patch
git push
and create the GitHub pull-request.
Attach patch to Phabricator task
Follow the above steps on how to “Fork repository to submit a Patch”. Instead of uploading “pushing” your changes to GitHub you can export the patches/ commits and send it to maintainers@vyos.net or attach it directly to the bug (preferred over email)
Export last commit to patch file:
git format-patch
or export the last two commits into its appropriate patch files:git format-patch -2
Coding Guidelines
Like any other project we have some small guidelines about our source code, too. The rules we have are not there to punish you - the rules are in place to help us all. By having a consistent coding style it becomes very easy for new and also longtime contributors to navigate through the sources and all the implied logic of any one source file..
Python 3 shall be used. How long can we keep Python 2 alive anyway? No considerations for Python 2 compatibility should be taken at any time.
Formatting
Python: Tabs shall not be used. Every indentation level should be 4 spaces
XML: Tabs shall not be used. Every indentation level should be 2 spaces
Note
There are extensions to e.g. VIM (xmllint) which will help you to get
your indention levels correct. Add to following to your .vimrc file:
au FileType xml setlocal equalprg=xmllint\ --format\ --recover\ -\
2>/dev/null
now you can call the linter using gg=G
in command mode.
Text generation
Template processor should be used for generating config files. Built-in string formatting may be used for simple line-oriented formats where every line is self-contained, such as iptables rules. Template processor must be used for structured, multi-line formats such as those used by ISC DHCPd.
The default template processor for VyOS code is Jinja2.
Summary
When modifying the source code, remember these rules of the legacy elimination campaign:
No new features in Perl
No old style command definitions
No code incompatible with Python3
Python
The switch to the Python programming language for new code is not merely a change of the language, but a chance to rethink and improve the programming approach.
Let’s face it: VyOS is full of spaghetti code where logic for reading the VyOS config, generating daemon configs, and restarting processes is all mixed up.
Python (or any other language, for that matter) does not provide automatic protection from bad design, so we need to also devise design guidelines and follow them to keep the system extensible and maintainable.
But we are here to assist you and want to guide you through how you can become a good VyOS contributor. The rules we have are not there to punish you - the rules are in place to help us all. What does it mean? By having a consistent coding style it becomes very easy for new contributors and also longtime contributors to navigate through the sources and all the implied logic of the spaghetti code.
Please use the following template as good starting point when developing new modules or even rewrite a whole bunch of code in the new style XML/Pyhon interface.
Configuration Script Structure and Behaviour
Your configuration script or operation mode script which is also written in Python3 should have a line break on 80 characters. This seems to be a bit odd nowadays but as some people also work remotely or program using vi(m) this is a fair good standard which I hope we can rely on.
In addition this also helps when browsing the GitHub codebase on a mobile device if you happen to be a crazy scientist.
#!/usr/bin/env python3
#
# Copyright (C) 2020 VyOS maintainers and contributors
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License version 2 or later as
# published by the Free Software Foundation.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
import sys
from vyos.config import Config
from vyos import ConfigError
def get_config():
if config:
conf = config
else:
conf = Config()
# Base path to CLI nodes
base = ['...', '...']
# Convert the VyOS config to an abstract internal representation
config_data = conf.get_config_dict(base, key_mangling=('-', '_'), get_first_key=True)
return config_data
def verify(config):
# Verify that configuration is valid
if invalid:
raise ConfigError("Descriptive message")
return True
def generate(config):
# Generate daemon configs
pass
def apply(config):
# Apply the generated configs to the live system
pass
try:
c = get_config()
verify(c)
generate(c)
apply(c)
except ConfigError as e:
print(e)
sys.exit(1)
The get_config()
function must convert the VyOS config to an abstract,
internal representation. No other function is allowed to call the vyos.config.
Config
object method directly. The rationale for it is that when config reads
are mixed with other logic, it’s very hard to change the config syntax since
you need to weed out every occurrence of the old syntax. If syntax-specific
code is confined to a single function, the rest of the code can be left
untouched as long as the internal representation remains compatible.
Another advantage is testability of the code. Mocking the entire config subsystem is hard, while constructing an internal representation by hand is way simpler.
The verify()
function takes your internal representation of the config and
checks if it’s valid, otherwise it must raise ConfigError
with an error
message that describes the problem and possibly suggests how to fix it. It must
not make any changes to the system. The rationale for it is again testability
and, in the future when the config backend is ready and every script is
rewritten in this fashion, ability to execute commit dry run (“commit test”
like in JunOS) and abort commit before making any changes to the system if an
error is found in any component.
The generate()
function generates config files for system components.
The apply()
function applies the generated configuration to the live
system. It should use non-disruptive reload whenever possible. It may execute
disruptive operations such as daemon process restart if a particular component
does not support non-disruptive reload, or when the expected service degradation
is minimal (for example, in case of auxiliary services such as LLDPd). In case
of high impact services such as VPN daemon and routing protocols, when non-
disruptive reload is supported for some but not all types of configuration
changes, scripts authors should make effort to determine if a configuration
change can be done in a non-disruptive way and only resort to disruptive restart
if it cannot be avoided.
Unless absolutely necessary, configuration scripts should not modify the active configuration of system components directly. Whenever at all possible, scripts should generate a configuration file or files that can be applied with a single command such as reloading a service through systemd init. Inserting statements one by one is particularly discouraged, for example, when configuring netfilter rules, saving them to a file and loading it with iptables-restore should always be preferred to executing iptables directly.
The apply()
and generate()
functions may raise ConfigError
if, for
example, the daemon failed to start with the updated config. It shouldn’t be a
substitute for proper config checking in the verify()
function. All
reasonable effort should be made to verify that generated configuration is
valid and will be accepted by the daemon, including, when necessary, cross-
checks with other VyOS configuration subtrees.
Exceptions, including VyOSError
(which is raised by vyos.config.Config
on improper config operations, such as trying to use list_nodes()
on a
non-tag node) should not be silenced or caught and re-raised as config error.
Sure this will not look pretty on user’s screen, but it will make way better
bug reports, and help users (and most VyOS users are IT professionals) do their
own debugging as well.
For easy orientation we suggest you take a look on the ntp.py
or
interfaces-bonding.py
(for tag nodes) implementation. Both files can be
found in the vyos-1x repository.
XML (used for CLI definitions)
The bash (or better vbash) completion in VyOS is defined in templates.
Templates are text files (called node.def
) stored in a directory tree. The
directory names define the command names, and template files define the command
behaviour. Before VyOS 1.2 (crux) this files were created by hand. After a
complex redesign process the new style template are automatically generated
from a XML input file.
XML interface definitions for VyOS come with a RelaxNG schema and are located in the vyos-1x module. This schema is a slightly modified schema from VyConf alias VyOS 2.0 So VyOS 1.2.x interface definitions will be reusable in Nextgen VyOS Versions with very minimal changes.
The great thing about schemas is not only that people can know the complete grammar for certain, but also that it can be automatically verified. The scripts/build-command-templates script that converts the XML definitions to old style templates also verifies them against the schema, so a bad definition will cause the package build to fail. I do agree that the format is verbose, but there is no other format now that would allow this. Besides, a specialized XML editor can alleviate the issue with verbosity.
Example:
<?xml version="1.0"?>
<!-- Cron configuration -->
<interfaceDefinition>
<node name="system">
<children>
<node name="task-scheduler">
<properties>
<help>Task scheduler settings</help>
</properties>
<children>
<tagNode name="task" owner="${vyos_conf_scripts_dir}/task_scheduler.py">
<properties>
<help>Scheduled task</help>
<valueHelp>
<format><string></format>
<description>Task name</description>
</valueHelp>
<priority>999</priority>
</properties>
<children>
<leafNode name="crontab-spec">
<properties>
<help>UNIX crontab time specification string</help>
</properties>
</leafNode>
<leafNode name="interval">
<properties>
<help>Execution interval</help>
<valueHelp>
<format><minutes></format>
<description>Execution interval in minutes</description>
</valueHelp>
<valueHelp>
<format><minutes>m</format>
<description>Execution interval in minutes</description>
</valueHelp>
<valueHelp>
<format><hours>h</format>
<description>Execution interval in hours</description>
</valueHelp>
<valueHelp>
<format><days>d</format>
<description>Execution interval in days</description>
</valueHelp>
<constraint>
<regex>[1-9]([0-9]*)([mhd]{0,1})</regex>
</constraint>
</properties>
</leafNode>
<node name="executable">
<properties>
<help>Executable path and arguments</help>
</properties>
<children>
<leafNode name="path">
<properties>
<help>Path to executable</help>
</properties>
</leafNode>
<leafNode name="arguments">
<properties>
<help>Arguments passed to the executable</help>
</properties>
</leafNode>
</children>
</node>
</children>
</tagNode>
</children>
</node>
</children>
</node>
</interfaceDefinition>
Command definitions are purely declarative, and cannot contain any logic. All logic for generating config files for target applications, restarting services and so on is implemented in configuration scripts instead.
GNU Preprocessor
XML interface definition files use the xml.in file extension which was implemented in T1843. XML interface definitions tend to have a lot of duplicated code in areas such as:
VIF (incl. VIF-S/VIF-C)
Address
Description
Enabled/Disabled
Instead of supplying all those XML nodes multiple times there are now include files with predefined features. Brief overview:
IPv4, IPv6 and DHCP(v6) address assignment
IPv4, IPv6 address assignment
VLAN (VIF) definition
MAC address assignment
All interface definition XML input files (.in suffix) will be sent to the GCC preprocess and the output is stored in the build/interface-definitions folder. The previously mentioned scripts/build-command-templates script operates on the build/interface-definitions folder to generate all required CLI nodes.
$ make interface_definitions
install -d -m 0755 build/interface-definitions
install -d -m 0755 build/op-mode-definitions
Generating build/interface-definitions/intel_qat.xml from interface-definitions/intel_qat.xml.in
Generating build/interface-definitions/interfaces-bonding.xml from interface-definitions/interfaces-bonding.xml.in
Generating build/interface-definitions/cron.xml from interface-definitions/cron.xml.in
Generating build/interface-definitions/pppoe-server.xml from interface-definitions/pppoe-server.xml.in
Generating build/interface-definitions/mdns-repeater.xml from interface-definitions/mdns-repeater.xml.in
Generating build/interface-definitions/tftp-server.xml from interface-definitions/tftp-server.xml.in
[...]
Guidelines
Use of numbers
Use of numbers in command names should be avoided unless a number is a
part of a protocol name or similar. Thus, protocols ospfv3
is perfectly
fine, but something like server-1
is questionable at best.
Help String
To ensure uniform look and feel, and improve readability, we should follow a set of guidelines consistently.
Capitalization and punctuation
The first word of every help string must be capitalized. There must not be a period at the end of help strings.
Rationale: this seems to be the unwritten standard in network device CLIs, and a good aesthetic compromise.
Examples:
Good: “Frobnication algorithm”
Bad: “frobnication algorithm”
Bad: “Frobnication algorithm.”
Horrible: “frobnication algorithm.”
Use of abbreviations and acronyms
Abbreviations and acronyms must be capitalized.
Examples:
Good: “TCP connection timeout”
Bad: “tcp connection timeout”
Horrible: “Tcp connection timeout”
Acronyms also must be capitalized to visually distinguish them from normal words:
Examples:
Good: RADIUS (as in remote authentication for dial-in user services)
Bad: radius (unless it’s about the distance between a center of a circle and any of its points)
Some abbreviations are traditionally written in mixed case. Generally, if it contains words “over” or “version”, the letter should be lowercase. If there’s an accepted spelling (especially if defined by an RFC or another standard), it must be followed.
Examples:
Good: PPPoE, IPsec
Bad: PPPOE, IPSEC
Bad: pppoe, ipsec
Use of verbs
Verbs should be avoided. If a verb can be omitted, omit it.
Examples:
Good: “TCP connection timeout”
Bad: “Set TCP connection timeout”
If a verb is essential, keep it. For example, in the help text of set system
ipv6 disable-forwarding
, “Disable IPv6 forwarding on all interfaces” is a
perfectly justified wording.
Prefer infinitives
Verbs, when they are necessary, should be in their infinitive form.
Examples:
Good: “Disable IPv6 forwarding”
Bad: “Disables IPv6 forwarding”
Migrating old CLI
Old concept/syntax |
New syntax |
Notes |
---|---|---|
mynode/node.def |
<node name=”mynode”> </node> |
Leaf nodes (nodes with values) use <leafNode> tag instead |
mynode/node.tag , tag: |
<tagNode name=”mynode> </node> |
|
help: My node |
<properties> <help>My node</help> |
|
val_help: <format>; some string |
<properties> <valueHelp> <format> format </format> <description> some string </description> |
Do not add angle brackets around the format, they will be inserted automatically |
syntax:expression: pattern |
<properties> <constraint> <regex> … |
<constraintErrorMessage> will be displayed on failure |
syntax:expression: $VAR(@) in “foo”, “bar”, “baz” |
None |
Use regex |
syntax:expression: exec … |
<properties> <constraint> <validator> <name =”foo” argument=”bar”> |
“${vyos_libexecdir}/validators/foo bar $VAR(@)” will be executed, <constraintErrorMessage> will be displayed on failure |
syntax:expression: (arithmetic expression) |
None |
External arithmetic validator may be added if there’s demand, complex validation is better left to commit-time scripts |
priority: 999 |
<properties> <priority>999</priority> |
Please leave a comment explaining why the priority was chosen (e.g. “after interfaces are configured”) |
multi: |
<properties> <multi/> |
Only applicable to leaf nodes |
allowed: echo foo bar |
<properties> <completionHelp> <list> foo bar </list> |
|
allowed: cli-shell-api listNodes vpn ipsec esp-group |
<properties> <completionHelp> <path> vpn ipsec esp-group </path> … |
|
allowed: /path/to/script |
<properties> <completionHelp> <script> /path/to/script </script> … |
|
default: |
None |
Move default values to scripts |
commit:expression: |
None |
All commit time checks should be in the verify() function of the script |
begin:/create:/delete: |
None |
All logic should be in the scripts |
Continuous Integration
VyOS makes use of Jenkins as our Continuous Integration (CI) service. Our VyOS CI server is publicly accessible here: https://ci.vyos.net. You can get a brief overview of all required components shipped in a VyOS ISO.
To build our modules we utilize a CI/CD Pipeline script. Each and every VyOS
component comes with it’s own Jenkinsfile
which is (more or less) a copy.
The Pipeline utilizes the Docker container from the Build ISO section -
but instead of building it from source on every run, we rather always fetch a
fresh copy (if needed) from Dockerhub.
Each module is build on demand if a new commit on the branch in question is found. After a successful run the resulting Debian Package(s) will be deployed to our Debian repository which is used during build time. It is located here: http://dev.packages.vyos.net/repositories/.